net/core/filter.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * Linux Socket Filter - Kernel level socket filtering
   4  *
   5  * Based on the design of the Berkeley Packet Filter. The new
   6  * internal format has been designed by PLUMgrid:
   7  *
   8  *      Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
   9  *
  10  * Authors:
  11  *
  12  *      Jay Schulist <jschlst@samba.org>
  13  *      Alexei Starovoitov <ast@plumgrid.com>
  14  *      Daniel Borkmann <dborkman@redhat.com>
  15  *
  16  * Andi Kleen - Fix a few bad bugs and races.
  17  * Kris Katterjohn - Added many additional checks in bpf_check_classic()
  18  */
  19
  20 #include <linux/atomic.h>
  21 #include <linux/bpf_verifier.h>
  22 #include <linux/module.h>
  23 #include <linux/types.h>
  24 #include <linux/mm.h>
  25 #include <linux/fcntl.h>
  26 #include <linux/socket.h>
  27 #include <linux/sock_diag.h>
  28 #include <linux/in.h>
  29 #include <linux/inet.h>
  30 #include <linux/netdevice.h>
  31 #include <linux/if_packet.h>
  32 #include <linux/if_arp.h>
  33 #include <linux/gfp.h>
  34 #include <net/inet_common.h>
  35 #include <net/ip.h>
  36 #include <net/protocol.h>
  37 #include <net/netlink.h>
  38 #include <linux/skbuff.h>
  39 #include <linux/skmsg.h>
  40 #include <net/sock.h>
  41 #include <net/flow_dissector.h>
  42 #include <linux/errno.h>
  43 #include <linux/timer.h>
  44 #include <linux/uaccess.h>
  45 #include <asm/unaligned.h>
  46 #include <linux/filter.h>
  47 #include <linux/ratelimit.h>
  48 #include <linux/seccomp.h>
  49 #include <linux/if_vlan.h>
  50 #include <linux/bpf.h>
  51 #include <linux/btf.h>
  52 #include <net/sch_generic.h>
  53 #include <net/cls_cgroup.h>
  54 #include <net/dst_metadata.h>
  55 #include <net/dst.h>
  56 #include <net/sock_reuseport.h>
  57 #include <net/busy_poll.h>
  58 #include <net/tcp.h>
  59 #include <net/xfrm.h>
  60 #include <net/udp.h>
  61 #include <linux/bpf_trace.h>
  62 #include <net/xdp_sock.h>
  63 #include <linux/inetdevice.h>
  64 #include <net/inet_hashtables.h>
  65 #include <net/inet6_hashtables.h>
  66 #include <net/ip_fib.h>
  67 #include <net/nexthop.h>
  68 #include <net/flow.h>
  69 #include <net/arp.h>
  70 #include <net/ipv6.h>
  71 #include <net/net_namespace.h>
  72 #include <linux/seg6_local.h>
  73 #include <net/seg6.h>
  74 #include <net/seg6_local.h>
  75 #include <net/lwtunnel.h>
  76 #include <net/ipv6_stubs.h>
  77 #include <net/bpf_sk_storage.h>
  78 #include <net/transp_v6.h>
  79 #include <linux/btf_ids.h>
  80 #include <net/tls.h>
  81 #include <net/xdp.h>
  82 #include <net/mptcp.h>
  83 #include <net/netfilter/nf_conntrack_bpf.h>
  84
  85 static const struct bpf_func_proto *
  86 bpf_sk_base_func_proto(enum bpf_func_id func_id);
  87
  88 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len)
  89 {
  90         if (in_compat_syscall()) {
  91                 struct compat_sock_fprog f32;
  92
  93                 if (len != sizeof(f32))
  94                         return -EINVAL;
  95                 if (copy_from_sockptr(&f32, src, sizeof(f32)))
  96                         return -EFAULT;
  97                 memset(dst, 0, sizeof(*dst));
  98                 dst->len = f32.len;
  99                 dst->filter = compat_ptr(f32.filter);
 100         } else {
 101                 if (len != sizeof(*dst))
 102                         return -EINVAL;
 103                 if (copy_from_sockptr(dst, src, sizeof(*dst)))
 104                         return -EFAULT;
 105         }
 106
 107         return 0;
 108 }
 109 EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user);
 110
 111 /**
 112  *      sk_filter_trim_cap - run a packet through a socket filter
 113  *      @sk: sock associated with &sk_buff
 114  *      @skb: buffer to filter
 115  *      @cap: limit on how short the eBPF program may trim the packet
 116  *
 117  * Run the eBPF program and then cut skb->data to correct size returned by
 118  * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
 119  * than pkt_len we keep whole skb->data. This is the socket level
 120  * wrapper to bpf_prog_run. It returns 0 if the packet should
 121  * be accepted or -EPERM if the packet should be tossed.
 122  *
 123  */
 124 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
 125 {
 126         int err;
 127         struct sk_filter *filter;
 128
 129         /*
 130          * If the skb was allocated from pfmemalloc reserves, only
 131          * allow SOCK_MEMALLOC sockets to use it as this socket is
 132          * helping free memory
 133          */
 134         if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
 135                 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
 136                 return -ENOMEM;
 137         }
 138         err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
 139         if (err)
 140                 return err;
 141
 142         err = security_sock_rcv_skb(sk, skb);
 143         if (err)
 144                 return err;
 145
 146         rcu_read_lock();
 147         filter = rcu_dereference(sk->sk_filter);
 148         if (filter) {
 149                 struct sock *save_sk = skb->sk;
 150                 unsigned int pkt_len;
 151
 152                 skb->sk = sk;
 153                 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
 154                 skb->sk = save_sk;
 155                 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
 156         }
 157         rcu_read_unlock();
 158
 159         return err;
 160 }
 161 EXPORT_SYMBOL(sk_filter_trim_cap);
 162
 163 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
 164 {
 165         return skb_get_poff(skb);
 166 }
 167
 168 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
 169 {
 170         struct nlattr *nla;
 171
 172         if (skb_is_nonlinear(skb))
 173                 return 0;
 174
 175         if (skb->len < sizeof(struct nlattr))
 176                 return 0;
 177
 178         if (a > skb->len - sizeof(struct nlattr))
 179                 return 0;
 180
 181         nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
 182         if (nla)
 183                 return (void *) nla - (void *) skb->data;
 184
 185         return 0;
 186 }
 187
 188 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
 189 {
 190         struct nlattr *nla;
 191
 192         if (skb_is_nonlinear(skb))
 193                 return 0;
 194
 195         if (skb->len < sizeof(struct nlattr))
 196                 return 0;
 197
 198         if (a > skb->len - sizeof(struct nlattr))
 199                 return 0;
 200
 201         nla = (struct nlattr *) &skb->data[a];
 202         if (nla->nla_len > skb->len - a)
 203                 return 0;
 204
 205         nla = nla_find_nested(nla, x);
 206         if (nla)
 207                 return (void *) nla - (void *) skb->data;
 208
 209         return 0;
 210 }
 211
 212 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
 213            data, int, headlen, int, offset)
 214 {
 215         u8 tmp, *ptr;
 216         const int len = sizeof(tmp);
 217
 218         if (offset >= 0) {
 219                 if (headlen - offset >= len)
 220                         return *(u8 *)(data + offset);
 221                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
 222                         return tmp;
 223         } else {
 224                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
 225                 if (likely(ptr))
 226                         return *(u8 *)ptr;
 227         }
 228
 229         return -EFAULT;
 230 }
 231
 232 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
 233            int, offset)
 234 {
 235         return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
 236                                          offset);
 237 }
 238
 239 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
 240            data, int, headlen, int, offset)
 241 {
 242         __be16 tmp, *ptr;
 243         const int len = sizeof(tmp);
 244
 245         if (offset >= 0) {
 246                 if (headlen - offset >= len)
 247                         return get_unaligned_be16(data + offset);
 248                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
 249                         return be16_to_cpu(tmp);
 250         } else {
 251                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
 252                 if (likely(ptr))
 253                         return get_unaligned_be16(ptr);
 254         }
 255
 256         return -EFAULT;
 257 }
 258
 259 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
 260            int, offset)
 261 {
 262         return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
 263                                           offset);
 264 }
 265
 266 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
 267            data, int, headlen, int, offset)
 268 {
 269         __be32 tmp, *ptr;
 270         const int len = sizeof(tmp);
 271
 272         if (likely(offset >= 0)) {
 273                 if (headlen - offset >= len)
 274                         return get_unaligned_be32(data + offset);
 275                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
 276                         return be32_to_cpu(tmp);
 277         } else {
 278                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
 279                 if (likely(ptr))
 280                         return get_unaligned_be32(ptr);
 281         }
 282
 283         return -EFAULT;
 284 }
 285
 286 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
 287            int, offset)
 288 {
 289         return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
 290                                           offset);
 291 }
 292
 293 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
 294                               struct bpf_insn *insn_buf)
 295 {
 296         struct bpf_insn *insn = insn_buf;
 297
 298         switch (skb_field) {
 299         case SKF_AD_MARK:
 300                 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4);
 301
 302                 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
 303                                       offsetof(struct sk_buff, mark));
 304                 break;
 305
 306         case SKF_AD_PKTTYPE:
 307                 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET);
 308                 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
 309 #ifdef __BIG_ENDIAN_BITFIELD
 310                 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
 311 #endif
 312                 break;
 313
 314         case SKF_AD_QUEUE:
 315                 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2);
 316
 317                 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
 318                                       offsetof(struct sk_buff, queue_mapping));
 319                 break;
 320
 321         case SKF_AD_VLAN_TAG:
 322                 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2);
 323
 324                 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
 325                 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
 326                                       offsetof(struct sk_buff, vlan_tci));
 327                 break;
 328         case SKF_AD_VLAN_TAG_PRESENT:
 329                 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_all) != 4);
 330                 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
 331                                       offsetof(struct sk_buff, vlan_all));
 332                 *insn++ = BPF_JMP_IMM(BPF_JEQ, dst_reg, 0, 1);
 333                 *insn++ = BPF_ALU32_IMM(BPF_MOV, dst_reg, 1);
 334                 break;
 335         }
 336
 337         return insn - insn_buf;
 338 }
 339
 340 static bool convert_bpf_extensions(struct sock_filter *fp,
 341                                    struct bpf_insn **insnp)
 342 {
 343         struct bpf_insn *insn = *insnp;
 344         u32 cnt;
 345
 346         switch (fp->k) {
 347         case SKF_AD_OFF + SKF_AD_PROTOCOL:
 348                 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2);
 349
 350                 /* A = *(u16 *) (CTX + offsetof(protocol)) */
 351                 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
 352                                       offsetof(struct sk_buff, protocol));
 353                 /* A = ntohs(A) [emitting a nop or swap16] */
 354                 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
 355                 break;
 356
 357         case SKF_AD_OFF + SKF_AD_PKTTYPE:
 358                 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
 359                 insn += cnt - 1;
 360                 break;
 361
 362         case SKF_AD_OFF + SKF_AD_IFINDEX:
 363         case SKF_AD_OFF + SKF_AD_HATYPE:
 364                 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4);
 365                 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2);
 366
 367                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
 368                                       BPF_REG_TMP, BPF_REG_CTX,
 369                                       offsetof(struct sk_buff, dev));
 370                 /* if (tmp != 0) goto pc + 1 */
 371                 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
 372                 *insn++ = BPF_EXIT_INSN();
 373                 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
 374                         *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
 375                                             offsetof(struct net_device, ifindex));
 376                 else
 377                         *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
 378                                             offsetof(struct net_device, type));
 379                 break;
 380
 381         case SKF_AD_OFF + SKF_AD_MARK:
 382                 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
 383                 insn += cnt - 1;
 384                 break;
 385
 386         case SKF_AD_OFF + SKF_AD_RXHASH:
 387                 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4);
 388
 389                 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
 390                                     offsetof(struct sk_buff, hash));
 391                 break;
 392
 393         case SKF_AD_OFF + SKF_AD_QUEUE:
 394                 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
 395                 insn += cnt - 1;
 396                 break;
 397
 398         case SKF_AD_OFF + SKF_AD_VLAN_TAG:
 399                 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
 400                                          BPF_REG_A, BPF_REG_CTX, insn);
 401                 insn += cnt - 1;
 402                 break;
 403
 404         case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
 405                 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
 406                                          BPF_REG_A, BPF_REG_CTX, insn);
 407                 insn += cnt - 1;
 408                 break;
 409
 410         case SKF_AD_OFF + SKF_AD_VLAN_TPID:
 411                 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2);
 412
 413                 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
 414                 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
 415                                       offsetof(struct sk_buff, vlan_proto));
 416                 /* A = ntohs(A) [emitting a nop or swap16] */
 417                 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
 418                 break;
 419
 420         case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
 421         case SKF_AD_OFF + SKF_AD_NLATTR:
 422         case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
 423         case SKF_AD_OFF + SKF_AD_CPU:
 424         case SKF_AD_OFF + SKF_AD_RANDOM:
 425                 /* arg1 = CTX */
 426                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
 427                 /* arg2 = A */
 428                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
 429                 /* arg3 = X */
 430                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
 431                 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
 432                 switch (fp->k) {
 433                 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
 434                         *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
 435                         break;
 436                 case SKF_AD_OFF + SKF_AD_NLATTR:
 437                         *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
 438                         break;
 439                 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
 440                         *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
 441                         break;
 442                 case SKF_AD_OFF + SKF_AD_CPU:
 443                         *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
 444                         break;
 445                 case SKF_AD_OFF + SKF_AD_RANDOM:
 446                         *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
 447                         bpf_user_rnd_init_once();
 448                         break;
 449                 }
 450                 break;
 451
 452         case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
 453                 /* A ^= X */
 454                 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
 455                 break;
 456
 457         default:
 458                 /* This is just a dummy call to avoid letting the compiler
 459                  * evict __bpf_call_base() as an optimization. Placed here
 460                  * where no-one bothers.
 461                  */
 462                 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
 463                 return false;
 464         }
 465
 466         *insnp = insn;
 467         return true;
 468 }
 469
 470 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
 471 {
 472         const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
 473         int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
 474         bool endian = BPF_SIZE(fp->code) == BPF_H ||
 475                       BPF_SIZE(fp->code) == BPF_W;
 476         bool indirect = BPF_MODE(fp->code) == BPF_IND;
 477         const int ip_align = NET_IP_ALIGN;
 478         struct bpf_insn *insn = *insnp;
 479         int offset = fp->k;
 480
 481         if (!indirect &&
 482             ((unaligned_ok && offset >= 0) ||
 483              (!unaligned_ok && offset >= 0 &&
 484               offset + ip_align >= 0 &&
 485               offset + ip_align % size == 0))) {
 486                 bool ldx_off_ok = offset <= S16_MAX;
 487
 488                 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
 489                 if (offset)
 490                         *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
 491                 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
 492                                       size, 2 + endian + (!ldx_off_ok * 2));
 493                 if (ldx_off_ok) {
 494                         *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
 495                                               BPF_REG_D, offset);
 496                 } else {
 497                         *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
 498                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
 499                         *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
 500                                               BPF_REG_TMP, 0);
 501                 }
 502                 if (endian)
 503                         *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
 504                 *insn++ = BPF_JMP_A(8);
 505         }
 506
 507         *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
 508         *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
 509         *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
 510         if (!indirect) {
 511                 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
 512         } else {
 513                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
 514                 if (fp->k)
 515                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
 516         }
 517
 518         switch (BPF_SIZE(fp->code)) {
 519         case BPF_B:
 520                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
 521                 break;
 522         case BPF_H:
 523                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
 524                 break;
 525         case BPF_W:
 526                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
 527                 break;
 528         default:
 529                 return false;
 530         }
 531
 532         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
 533         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 534         *insn   = BPF_EXIT_INSN();
 535
 536         *insnp = insn;
 537         return true;
 538 }
 539
 540 /**
 541  *      bpf_convert_filter - convert filter program
 542  *      @prog: the user passed filter program
 543  *      @len: the length of the user passed filter program
 544  *      @new_prog: allocated 'struct bpf_prog' or NULL
 545  *      @new_len: pointer to store length of converted program
 546  *      @seen_ld_abs: bool whether we've seen ld_abs/ind
 547  *
 548  * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
 549  * style extended BPF (eBPF).
 550  * Conversion workflow:
 551  *
 552  * 1) First pass for calculating the new program length:
 553  *   bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
 554  *
 555  * 2) 2nd pass to remap in two passes: 1st pass finds new
 556  *    jump offsets, 2nd pass remapping:
 557  *   bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
 558  */
 559 static int bpf_convert_filter(struct sock_filter *prog, int len,
 560                               struct bpf_prog *new_prog, int *new_len,
 561                               bool *seen_ld_abs)
 562 {
 563         int new_flen = 0, pass = 0, target, i, stack_off;
 564         struct bpf_insn *new_insn, *first_insn = NULL;
 565         struct sock_filter *fp;
 566         int *addrs = NULL;
 567         u8 bpf_src;
 568
 569         BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
 570         BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
 571
 572         if (len <= 0 || len > BPF_MAXINSNS)
 573                 return -EINVAL;
 574
 575         if (new_prog) {
 576                 first_insn = new_prog->insnsi;
 577                 addrs = kcalloc(len, sizeof(*addrs),
 578                                 GFP_KERNEL | __GFP_NOWARN);
 579                 if (!addrs)
 580                         return -ENOMEM;
 581         }
 582
 583 do_pass:
 584         new_insn = first_insn;
 585         fp = prog;
 586
 587         /* Classic BPF related prologue emission. */
 588         if (new_prog) {
 589                 /* Classic BPF expects A and X to be reset first. These need
 590                  * to be guaranteed to be the first two instructions.
 591                  */
 592                 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 593                 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
 594
 595                 /* All programs must keep CTX in callee saved BPF_REG_CTX.
 596                  * In eBPF case it's done by the compiler, here we need to
 597                  * do this ourself. Initial CTX is present in BPF_REG_ARG1.
 598                  */
 599                 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
 600                 if (*seen_ld_abs) {
 601                         /* For packet access in classic BPF, cache skb->data
 602                          * in callee-saved BPF R8 and skb->len - skb->data_len
 603                          * (headlen) in BPF R9. Since classic BPF is read-only
 604                          * on CTX, we only need to cache it once.
 605                          */
 606                         *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
 607                                                   BPF_REG_D, BPF_REG_CTX,
 608                                                   offsetof(struct sk_buff, data));
 609                         *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
 610                                                   offsetof(struct sk_buff, len));
 611                         *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
 612                                                   offsetof(struct sk_buff, data_len));
 613                         *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
 614                 }
 615         } else {
 616                 new_insn += 3;
 617         }
 618
 619         for (i = 0; i < len; fp++, i++) {
 620                 struct bpf_insn tmp_insns[32] = { };
 621                 struct bpf_insn *insn = tmp_insns;
 622
 623                 if (addrs)
 624                         addrs[i] = new_insn - first_insn;
 625
 626                 switch (fp->code) {
 627                 /* All arithmetic insns and skb loads map as-is. */
 628                 case BPF_ALU | BPF_ADD | BPF_X:
 629                 case BPF_ALU | BPF_ADD | BPF_K:
 630                 case BPF_ALU | BPF_SUB | BPF_X:
 631                 case BPF_ALU | BPF_SUB | BPF_K:
 632                 case BPF_ALU | BPF_AND | BPF_X:
 633                 case BPF_ALU | BPF_AND | BPF_K:
 634                 case BPF_ALU | BPF_OR | BPF_X:
 635                 case BPF_ALU | BPF_OR | BPF_K:
 636                 case BPF_ALU | BPF_LSH | BPF_X:
 637                 case BPF_ALU | BPF_LSH | BPF_K:
 638                 case BPF_ALU | BPF_RSH | BPF_X:
 639                 case BPF_ALU | BPF_RSH | BPF_K:
 640                 case BPF_ALU | BPF_XOR | BPF_X:
 641                 case BPF_ALU | BPF_XOR | BPF_K:
 642                 case BPF_ALU | BPF_MUL | BPF_X:
 643                 case BPF_ALU | BPF_MUL | BPF_K:
 644                 case BPF_ALU | BPF_DIV | BPF_X:
 645                 case BPF_ALU | BPF_DIV | BPF_K:
 646                 case BPF_ALU | BPF_MOD | BPF_X:
 647                 case BPF_ALU | BPF_MOD | BPF_K:
 648                 case BPF_ALU | BPF_NEG:
 649                 case BPF_LD | BPF_ABS | BPF_W:
 650                 case BPF_LD | BPF_ABS | BPF_H:
 651                 case BPF_LD | BPF_ABS | BPF_B:
 652                 case BPF_LD | BPF_IND | BPF_W:
 653                 case BPF_LD | BPF_IND | BPF_H:
 654                 case BPF_LD | BPF_IND | BPF_B:
 655                         /* Check for overloaded BPF extension and
 656                          * directly convert it if found, otherwise
 657                          * just move on with mapping.
 658                          */
 659                         if (BPF_CLASS(fp->code) == BPF_LD &&
 660                             BPF_MODE(fp->code) == BPF_ABS &&
 661                             convert_bpf_extensions(fp, &insn))
 662                                 break;
 663                         if (BPF_CLASS(fp->code) == BPF_LD &&
 664                             convert_bpf_ld_abs(fp, &insn)) {
 665                                 *seen_ld_abs = true;
 666                                 break;
 667                         }
 668
 669                         if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
 670                             fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
 671                                 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
 672                                 /* Error with exception code on div/mod by 0.
 673                                  * For cBPF programs, this was always return 0.
 674                                  */
 675                                 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
 676                                 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 677                                 *insn++ = BPF_EXIT_INSN();
 678                         }
 679
 680                         *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
 681                         break;
 682
 683                 /* Jump transformation cannot use BPF block macros
 684                  * everywhere as offset calculation and target updates
 685                  * require a bit more work than the rest, i.e. jump
 686                  * opcodes map as-is, but offsets need adjustment.
 687                  */
 688
 689 #define BPF_EMIT_JMP                                                    \
 690         do {                                                            \
 691                 const s32 off_min = S16_MIN, off_max = S16_MAX;         \
 692                 s32 off;                                                \
 693                                                                         \
 694                 if (target >= len || target < 0)                        \
 695                         goto err;                                       \
 696                 off = addrs ? addrs[target] - addrs[i] - 1 : 0;         \
 697                 /* Adjust pc relative offset for 2nd or 3rd insn. */    \
 698                 off -= insn - tmp_insns;                                \
 699                 /* Reject anything not fitting into insn->off. */       \
 700                 if (off < off_min || off > off_max)                     \
 701                         goto err;                                       \
 702                 insn->off = off;                                        \
 703         } while (0)
 704
 705                 case BPF_JMP | BPF_JA:
 706                         target = i + fp->k + 1;
 707                         insn->code = fp->code;
 708                         BPF_EMIT_JMP;
 709                         break;
 710
 711                 case BPF_JMP | BPF_JEQ | BPF_K:
 712                 case BPF_JMP | BPF_JEQ | BPF_X:
 713                 case BPF_JMP | BPF_JSET | BPF_K:
 714                 case BPF_JMP | BPF_JSET | BPF_X:
 715                 case BPF_JMP | BPF_JGT | BPF_K:
 716                 case BPF_JMP | BPF_JGT | BPF_X:
 717                 case BPF_JMP | BPF_JGE | BPF_K:
 718                 case BPF_JMP | BPF_JGE | BPF_X:
 719                         if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
 720                                 /* BPF immediates are signed, zero extend
 721                                  * immediate into tmp register and use it
 722                                  * in compare insn.
 723                                  */
 724                                 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
 725
 726                                 insn->dst_reg = BPF_REG_A;
 727                                 insn->src_reg = BPF_REG_TMP;
 728                                 bpf_src = BPF_X;
 729                         } else {
 730                                 insn->dst_reg = BPF_REG_A;
 731                                 insn->imm = fp->k;
 732                                 bpf_src = BPF_SRC(fp->code);
 733                                 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
 734                         }
 735
 736                         /* Common case where 'jump_false' is next insn. */
 737                         if (fp->jf == 0) {
 738                                 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
 739                                 target = i + fp->jt + 1;
 740                                 BPF_EMIT_JMP;
 741                                 break;
 742                         }
 743
 744                         /* Convert some jumps when 'jump_true' is next insn. */
 745                         if (fp->jt == 0) {
 746                                 switch (BPF_OP(fp->code)) {
 747                                 case BPF_JEQ:
 748                                         insn->code = BPF_JMP | BPF_JNE | bpf_src;
 749                                         break;
 750                                 case BPF_JGT:
 751                                         insn->code = BPF_JMP | BPF_JLE | bpf_src;
 752                                         break;
 753                                 case BPF_JGE:
 754                                         insn->code = BPF_JMP | BPF_JLT | bpf_src;
 755                                         break;
 756                                 default:
 757                                         goto jmp_rest;
 758                                 }
 759
 760                                 target = i + fp->jf + 1;
 761                                 BPF_EMIT_JMP;
 762                                 break;
 763                         }
 764 jmp_rest:
 765                         /* Other jumps are mapped into two insns: Jxx and JA. */
 766                         target = i + fp->jt + 1;
 767                         insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
 768                         BPF_EMIT_JMP;
 769                         insn++;
 770
 771                         insn->code = BPF_JMP | BPF_JA;
 772                         target = i + fp->jf + 1;
 773                         BPF_EMIT_JMP;
 774                         break;
 775
 776                 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
 777                 case BPF_LDX | BPF_MSH | BPF_B: {
 778                         struct sock_filter tmp = {
 779                                 .code   = BPF_LD | BPF_ABS | BPF_B,
 780                                 .k      = fp->k,
 781                         };
 782
 783                         *seen_ld_abs = true;
 784
 785                         /* X = A */
 786                         *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 787                         /* A = BPF_R0 = *(u8 *) (skb->data + K) */
 788                         convert_bpf_ld_abs(&tmp, &insn);
 789                         insn++;
 790                         /* A &= 0xf */
 791                         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
 792                         /* A <<= 2 */
 793                         *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
 794                         /* tmp = X */
 795                         *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
 796                         /* X = A */
 797                         *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 798                         /* A = tmp */
 799                         *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
 800                         break;
 801                 }
 802                 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
 803                  * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
 804                  */
 805                 case BPF_RET | BPF_A:
 806                 case BPF_RET | BPF_K:
 807                         if (BPF_RVAL(fp->code) == BPF_K)
 808                                 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
 809                                                         0, fp->k);
 810                         *insn = BPF_EXIT_INSN();
 811                         break;
 812
 813                 /* Store to stack. */
 814                 case BPF_ST:
 815                 case BPF_STX:
 816                         stack_off = fp->k * 4  + 4;
 817                         *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
 818                                             BPF_ST ? BPF_REG_A : BPF_REG_X,
 819                                             -stack_off);
 820                         /* check_load_and_stores() verifies that classic BPF can
 821                          * load from stack only after write, so tracking
 822                          * stack_depth for ST|STX insns is enough
 823                          */
 824                         if (new_prog && new_prog->aux->stack_depth < stack_off)
 825                                 new_prog->aux->stack_depth = stack_off;
 826                         break;
 827
 828                 /* Load from stack. */
 829                 case BPF_LD | BPF_MEM:
 830                 case BPF_LDX | BPF_MEM:
 831                         stack_off = fp->k * 4  + 4;
 832                         *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD  ?
 833                                             BPF_REG_A : BPF_REG_X, BPF_REG_FP,
 834                                             -stack_off);
 835                         break;
 836
 837                 /* A = K or X = K */
 838                 case BPF_LD | BPF_IMM:
 839                 case BPF_LDX | BPF_IMM:
 840                         *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
 841                                               BPF_REG_A : BPF_REG_X, fp->k);
 842                         break;
 843
 844                 /* X = A */
 845                 case BPF_MISC | BPF_TAX:
 846                         *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 847                         break;
 848
 849                 /* A = X */
 850                 case BPF_MISC | BPF_TXA:
 851                         *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
 852                         break;
 853
 854                 /* A = skb->len or X = skb->len */
 855                 case BPF_LD | BPF_W | BPF_LEN:
 856                 case BPF_LDX | BPF_W | BPF_LEN:
 857                         *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
 858                                             BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
 859                                             offsetof(struct sk_buff, len));
 860                         break;
 861
 862                 /* Access seccomp_data fields. */
 863                 case BPF_LDX | BPF_ABS | BPF_W:
 864                         /* A = *(u32 *) (ctx + K) */
 865                         *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
 866                         break;
 867
 868                 /* Unknown instruction. */
 869                 default:
 870                         goto err;
 871                 }
 872
 873                 insn++;
 874                 if (new_prog)
 875                         memcpy(new_insn, tmp_insns,
 876                                sizeof(*insn) * (insn - tmp_insns));
 877                 new_insn += insn - tmp_insns;
 878         }
 879
 880         if (!new_prog) {
 881                 /* Only calculating new length. */
 882                 *new_len = new_insn - first_insn;
 883                 if (*seen_ld_abs)
 884                         *new_len += 4; /* Prologue bits. */
 885                 return 0;
 886         }
 887
 888         pass++;
 889         if (new_flen != new_insn - first_insn) {
 890                 new_flen = new_insn - first_insn;
 891                 if (pass > 2)
 892                         goto err;
 893                 goto do_pass;
 894         }
 895
 896         kfree(addrs);
 897         BUG_ON(*new_len != new_flen);
 898         return 0;
 899 err:
 900         kfree(addrs);
 901         return -EINVAL;
 902 }
 903
 904 /* Security:
 905  *
 906  * As we dont want to clear mem[] array for each packet going through
 907  * __bpf_prog_run(), we check that filter loaded by user never try to read
 908  * a cell if not previously written, and we check all branches to be sure
 909  * a malicious user doesn't try to abuse us.
 910  */
 911 static int check_load_and_stores(const struct sock_filter *filter, int flen)
 912 {
 913         u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
 914         int pc, ret = 0;
 915
 916         BUILD_BUG_ON(BPF_MEMWORDS > 16);
 917
 918         masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
 919         if (!masks)
 920                 return -ENOMEM;
 921
 922         memset(masks, 0xff, flen * sizeof(*masks));
 923
 924         for (pc = 0; pc < flen; pc++) {
 925                 memvalid &= masks[pc];
 926
 927                 switch (filter[pc].code) {
 928                 case BPF_ST:
 929                 case BPF_STX:
 930                         memvalid |= (1 << filter[pc].k);
 931                         break;
 932                 case BPF_LD | BPF_MEM:
 933                 case BPF_LDX | BPF_MEM:
 934                         if (!(memvalid & (1 << filter[pc].k))) {
 935                                 ret = -EINVAL;
 936                                 goto error;
 937                         }
 938                         break;
 939                 case BPF_JMP | BPF_JA:
 940                         /* A jump must set masks on target */
 941                         masks[pc + 1 + filter[pc].k] &= memvalid;
 942                         memvalid = ~0;
 943                         break;
 944                 case BPF_JMP | BPF_JEQ | BPF_K:
 945                 case BPF_JMP | BPF_JEQ | BPF_X:
 946                 case BPF_JMP | BPF_JGE | BPF_K:
 947                 case BPF_JMP | BPF_JGE | BPF_X:
 948                 case BPF_JMP | BPF_JGT | BPF_K:
 949                 case BPF_JMP | BPF_JGT | BPF_X:
 950                 case BPF_JMP | BPF_JSET | BPF_K:
 951                 case BPF_JMP | BPF_JSET | BPF_X:
 952                         /* A jump must set masks on targets */
 953                         masks[pc + 1 + filter[pc].jt] &= memvalid;
 954                         masks[pc + 1 + filter[pc].jf] &= memvalid;
 955                         memvalid = ~0;
 956                         break;
 957                 }
 958         }
 959 error:
 960         kfree(masks);
 961         return ret;
 962 }
 963
 964 static bool chk_code_allowed(u16 code_to_probe)
 965 {
 966         static const bool codes[] = {
 967                 /* 32 bit ALU operations */
 968                 [BPF_ALU | BPF_ADD | BPF_K] = true,
 969                 [BPF_ALU | BPF_ADD | BPF_X] = true,
 970                 [BPF_ALU | BPF_SUB | BPF_K] = true,
 971                 [BPF_ALU | BPF_SUB | BPF_X] = true,
 972                 [BPF_ALU | BPF_MUL | BPF_K] = true,
 973                 [BPF_ALU | BPF_MUL | BPF_X] = true,
 974                 [BPF_ALU | BPF_DIV | BPF_K] = true,
 975                 [BPF_ALU | BPF_DIV | BPF_X] = true,
 976                 [BPF_ALU | BPF_MOD | BPF_K] = true,
 977                 [BPF_ALU | BPF_MOD | BPF_X] = true,
 978                 [BPF_ALU | BPF_AND | BPF_K] = true,
 979                 [BPF_ALU | BPF_AND | BPF_X] = true,
 980                 [BPF_ALU | BPF_OR | BPF_K] = true,
 981                 [BPF_ALU | BPF_OR | BPF_X] = true,
 982                 [BPF_ALU | BPF_XOR | BPF_K] = true,
 983                 [BPF_ALU | BPF_XOR | BPF_X] = true,
 984                 [BPF_ALU | BPF_LSH | BPF_K] = true,
 985                 [BPF_ALU | BPF_LSH | BPF_X] = true,
 986                 [BPF_ALU | BPF_RSH | BPF_K] = true,
 987                 [BPF_ALU | BPF_RSH | BPF_X] = true,
 988                 [BPF_ALU | BPF_NEG] = true,
 989                 /* Load instructions */
 990                 [BPF_LD | BPF_W | BPF_ABS] = true,
 991                 [BPF_LD | BPF_H | BPF_ABS] = true,
 992                 [BPF_LD | BPF_B | BPF_ABS] = true,
 993                 [BPF_LD | BPF_W | BPF_LEN] = true,
 994                 [BPF_LD | BPF_W | BPF_IND] = true,
 995                 [BPF_LD | BPF_H | BPF_IND] = true,
 996                 [BPF_LD | BPF_B | BPF_IND] = true,
 997                 [BPF_LD | BPF_IMM] = true,
 998                 [BPF_LD | BPF_MEM] = true,
 999                 [BPF_LDX | BPF_W | BPF_LEN] = true,
1000                 [BPF_LDX | BPF_B | BPF_MSH] = true,
1001                 [BPF_LDX | BPF_IMM] = true,
1002                 [BPF_LDX | BPF_MEM] = true,
1003                 /* Store instructions */
1004                 [BPF_ST] = true,
1005                 [BPF_STX] = true,
1006                 /* Misc instructions */
1007                 [BPF_MISC | BPF_TAX] = true,
1008                 [BPF_MISC | BPF_TXA] = true,
1009                 /* Return instructions */
1010                 [BPF_RET | BPF_K] = true,
1011                 [BPF_RET | BPF_A] = true,
1012                 /* Jump instructions */
1013                 [BPF_JMP | BPF_JA] = true,
1014                 [BPF_JMP | BPF_JEQ | BPF_K] = true,
1015                 [BPF_JMP | BPF_JEQ | BPF_X] = true,
1016                 [BPF_JMP | BPF_JGE | BPF_K] = true,
1017                 [BPF_JMP | BPF_JGE | BPF_X] = true,
1018                 [BPF_JMP | BPF_JGT | BPF_K] = true,
1019                 [BPF_JMP | BPF_JGT | BPF_X] = true,
1020                 [BPF_JMP | BPF_JSET | BPF_K] = true,
1021                 [BPF_JMP | BPF_JSET | BPF_X] = true,
1022         };
1023
1024         if (code_to_probe >= ARRAY_SIZE(codes))
1025                 return false;
1026
1027         return codes[code_to_probe];
1028 }
1029
1030 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1031                                 unsigned int flen)
1032 {
1033         if (filter == NULL)
1034                 return false;
1035         if (flen == 0 || flen > BPF_MAXINSNS)
1036                 return false;
1037
1038         return true;
1039 }
1040
1041 /**
1042  *      bpf_check_classic - verify socket filter code
1043  *      @filter: filter to verify
1044  *      @flen: length of filter
1045  *
1046  * Check the user's filter code. If we let some ugly
1047  * filter code slip through kaboom! The filter must contain
1048  * no references or jumps that are out of range, no illegal
1049  * instructions, and must end with a RET instruction.
1050  *
1051  * All jumps are forward as they are not signed.
1052  *
1053  * Returns 0 if the rule set is legal or -EINVAL if not.
1054  */
1055 static int bpf_check_classic(const struct sock_filter *filter,
1056                              unsigned int flen)
1057 {
1058         bool anc_found;
1059         int pc;
1060
1061         /* Check the filter code now */
1062         for (pc = 0; pc < flen; pc++) {
1063                 const struct sock_filter *ftest = &filter[pc];
1064
1065                 /* May we actually operate on this code? */
1066                 if (!chk_code_allowed(ftest->code))
1067                         return -EINVAL;
1068
1069                 /* Some instructions need special checks */
1070                 switch (ftest->code) {
1071                 case BPF_ALU | BPF_DIV | BPF_K:
1072                 case BPF_ALU | BPF_MOD | BPF_K:
1073                         /* Check for division by zero */
1074                         if (ftest->k == 0)
1075                                 return -EINVAL;
1076                         break;
1077                 case BPF_ALU | BPF_LSH | BPF_K:
1078                 case BPF_ALU | BPF_RSH | BPF_K:
1079                         if (ftest->k >= 32)
1080                                 return -EINVAL;
1081                         break;
1082                 case BPF_LD | BPF_MEM:
1083                 case BPF_LDX | BPF_MEM:
1084                 case BPF_ST:
1085                 case BPF_STX:
1086                         /* Check for invalid memory addresses */
1087                         if (ftest->k >= BPF_MEMWORDS)
1088                                 return -EINVAL;
1089                         break;
1090                 case BPF_JMP | BPF_JA:
1091                         /* Note, the large ftest->k might cause loops.
1092                          * Compare this with conditional jumps below,
1093                          * where offsets are limited. --ANK (981016)
1094                          */
1095                         if (ftest->k >= (unsigned int)(flen - pc - 1))
1096                                 return -EINVAL;
1097                         break;
1098                 case BPF_JMP | BPF_JEQ | BPF_K:
1099                 case BPF_JMP | BPF_JEQ | BPF_X:
1100                 case BPF_JMP | BPF_JGE | BPF_K:
1101                 case BPF_JMP | BPF_JGE | BPF_X:
1102                 case BPF_JMP | BPF_JGT | BPF_K:
1103                 case BPF_JMP | BPF_JGT | BPF_X:
1104                 case BPF_JMP | BPF_JSET | BPF_K:
1105                 case BPF_JMP | BPF_JSET | BPF_X:
1106                         /* Both conditionals must be safe */
1107                         if (pc + ftest->jt + 1 >= flen ||
1108                             pc + ftest->jf + 1 >= flen)
1109                                 return -EINVAL;
1110                         break;
1111                 case BPF_LD | BPF_W | BPF_ABS:
1112                 case BPF_LD | BPF_H | BPF_ABS:
1113                 case BPF_LD | BPF_B | BPF_ABS:
1114                         anc_found = false;
1115                         if (bpf_anc_helper(ftest) & BPF_ANC)
1116                                 anc_found = true;
1117                         /* Ancillary operation unknown or unsupported */
1118                         if (anc_found == false && ftest->k >= SKF_AD_OFF)
1119                                 return -EINVAL;
1120                 }
1121         }
1122
1123         /* Last instruction must be a RET code */
1124         switch (filter[flen - 1].code) {
1125         case BPF_RET | BPF_K:
1126         case BPF_RET | BPF_A:
1127                 return check_load_and_stores(filter, flen);
1128         }
1129
1130         return -EINVAL;
1131 }
1132
1133 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1134                                       const struct sock_fprog *fprog)
1135 {
1136         unsigned int fsize = bpf_classic_proglen(fprog);
1137         struct sock_fprog_kern *fkprog;
1138
1139         fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1140         if (!fp->orig_prog)
1141                 return -ENOMEM;
1142
1143         fkprog = fp->orig_prog;
1144         fkprog->len = fprog->len;
1145
1146         fkprog->filter = kmemdup(fp->insns, fsize,
1147                                  GFP_KERNEL | __GFP_NOWARN);
1148         if (!fkprog->filter) {
1149                 kfree(fp->orig_prog);
1150                 return -ENOMEM;
1151         }
1152
1153         return 0;
1154 }
1155
1156 static void bpf_release_orig_filter(struct bpf_prog *fp)
1157 {
1158         struct sock_fprog_kern *fprog = fp->orig_prog;
1159
1160         if (fprog) {
1161                 kfree(fprog->filter);
1162                 kfree(fprog);
1163         }
1164 }
1165
1166 static void __bpf_prog_release(struct bpf_prog *prog)
1167 {
1168         if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1169                 bpf_prog_put(prog);
1170         } else {
1171                 bpf_release_orig_filter(prog);
1172                 bpf_prog_free(prog);
1173         }
1174 }
1175
1176 static void __sk_filter_release(struct sk_filter *fp)
1177 {
1178         __bpf_prog_release(fp->prog);
1179         kfree(fp);
1180 }
1181
1182 /**
1183  *      sk_filter_release_rcu - Release a socket filter by rcu_head
1184  *      @rcu: rcu_head that contains the sk_filter to free
1185  */
1186 static void sk_filter_release_rcu(struct rcu_head *rcu)
1187 {
1188         struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1189
1190         __sk_filter_release(fp);
1191 }
1192
1193 /**
1194  *      sk_filter_release - release a socket filter
1195  *      @fp: filter to remove
1196  *
1197  *      Remove a filter from a socket and release its resources.
1198  */
1199 static void sk_filter_release(struct sk_filter *fp)
1200 {
1201         if (refcount_dec_and_test(&fp->refcnt))
1202                 call_rcu(&fp->rcu, sk_filter_release_rcu);
1203 }
1204
1205 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1206 {
1207         u32 filter_size = bpf_prog_size(fp->prog->len);
1208
1209         atomic_sub(filter_size, &sk->sk_omem_alloc);
1210         sk_filter_release(fp);
1211 }
1212
1213 /* try to charge the socket memory if there is space available
1214  * return true on success
1215  */
1216 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1217 {
1218         u32 filter_size = bpf_prog_size(fp->prog->len);
1219         int optmem_max = READ_ONCE(sysctl_optmem_max);
1220
1221         /* same check as in sock_kmalloc() */
1222         if (filter_size <= optmem_max &&
1223             atomic_read(&sk->sk_omem_alloc) + filter_size < optmem_max) {
1224                 atomic_add(filter_size, &sk->sk_omem_alloc);
1225                 return true;
1226         }
1227         return false;
1228 }
1229
1230 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1231 {
1232         if (!refcount_inc_not_zero(&fp->refcnt))
1233                 return false;
1234
1235         if (!__sk_filter_charge(sk, fp)) {
1236                 sk_filter_release(fp);
1237                 return false;
1238         }
1239         return true;
1240 }
1241
1242 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1243 {
1244         struct sock_filter *old_prog;
1245         struct bpf_prog *old_fp;
1246         int err, new_len, old_len = fp->len;
1247         bool seen_ld_abs = false;
1248
1249         /* We are free to overwrite insns et al right here as it won't be used at
1250          * this point in time anymore internally after the migration to the eBPF
1251          * instruction representation.
1252          */
1253         BUILD_BUG_ON(sizeof(struct sock_filter) !=
1254                      sizeof(struct bpf_insn));
1255
1256         /* Conversion cannot happen on overlapping memory areas,
1257          * so we need to keep the user BPF around until the 2nd
1258          * pass. At this time, the user BPF is stored in fp->insns.
1259          */
1260         old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1261                            GFP_KERNEL | __GFP_NOWARN);
1262         if (!old_prog) {
1263                 err = -ENOMEM;
1264                 goto out_err;
1265         }
1266
1267         /* 1st pass: calculate the new program length. */
1268         err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1269                                  &seen_ld_abs);
1270         if (err)
1271                 goto out_err_free;
1272
1273         /* Expand fp for appending the new filter representation. */
1274         old_fp = fp;
1275         fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1276         if (!fp) {
1277                 /* The old_fp is still around in case we couldn't
1278                  * allocate new memory, so uncharge on that one.
1279                  */
1280                 fp = old_fp;
1281                 err = -ENOMEM;
1282                 goto out_err_free;
1283         }
1284
1285         fp->len = new_len;
1286
1287         /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1288         err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1289                                  &seen_ld_abs);
1290         if (err)
1291                 /* 2nd bpf_convert_filter() can fail only if it fails
1292                  * to allocate memory, remapping must succeed. Note,
1293                  * that at this time old_fp has already been released
1294                  * by krealloc().
1295                  */
1296                 goto out_err_free;
1297
1298         fp = bpf_prog_select_runtime(fp, &err);
1299         if (err)
1300                 goto out_err_free;
1301
1302         kfree(old_prog);
1303         return fp;
1304
1305 out_err_free:
1306         kfree(old_prog);
1307 out_err:
1308         __bpf_prog_release(fp);
1309         return ERR_PTR(err);
1310 }
1311
1312 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1313                                            bpf_aux_classic_check_t trans)
1314 {
1315         int err;
1316
1317         fp->bpf_func = NULL;
1318         fp->jited = 0;
1319
1320         err = bpf_check_classic(fp->insns, fp->len);
1321         if (err) {
1322                 __bpf_prog_release(fp);
1323                 return ERR_PTR(err);
1324         }
1325
1326         /* There might be additional checks and transformations
1327          * needed on classic filters, f.e. in case of seccomp.
1328          */
1329         if (trans) {
1330                 err = trans(fp->insns, fp->len);
1331                 if (err) {
1332                         __bpf_prog_release(fp);
1333                         return ERR_PTR(err);
1334                 }
1335         }
1336
1337         /* Probe if we can JIT compile the filter and if so, do
1338          * the compilation of the filter.
1339          */
1340         bpf_jit_compile(fp);
1341
1342         /* JIT compiler couldn't process this filter, so do the eBPF translation
1343          * for the optimized interpreter.
1344          */
1345         if (!fp->jited)
1346                 fp = bpf_migrate_filter(fp);
1347
1348         return fp;
1349 }
1350
1351 /**
1352  *      bpf_prog_create - create an unattached filter
1353  *      @pfp: the unattached filter that is created
1354  *      @fprog: the filter program
1355  *
1356  * Create a filter independent of any socket. We first run some
1357  * sanity checks on it to make sure it does not explode on us later.
1358  * If an error occurs or there is insufficient memory for the filter
1359  * a negative errno code is returned. On success the return is zero.
1360  */
1361 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1362 {
1363         unsigned int fsize = bpf_classic_proglen(fprog);
1364         struct bpf_prog *fp;
1365
1366         /* Make sure new filter is there and in the right amounts. */
1367         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1368                 return -EINVAL;
1369
1370         fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1371         if (!fp)
1372                 return -ENOMEM;
1373
1374         memcpy(fp->insns, fprog->filter, fsize);
1375
1376         fp->len = fprog->len;
1377         /* Since unattached filters are not copied back to user
1378          * space through sk_get_filter(), we do not need to hold
1379          * a copy here, and can spare us the work.
1380          */
1381         fp->orig_prog = NULL;
1382
1383         /* bpf_prepare_filter() already takes care of freeing
1384          * memory in case something goes wrong.
1385          */
1386         fp = bpf_prepare_filter(fp, NULL);
1387         if (IS_ERR(fp))
1388                 return PTR_ERR(fp);
1389
1390         *pfp = fp;
1391         return 0;
1392 }
1393 EXPORT_SYMBOL_GPL(bpf_prog_create);
1394
1395 /**
1396  *      bpf_prog_create_from_user - create an unattached filter from user buffer
1397  *      @pfp: the unattached filter that is created
1398  *      @fprog: the filter program
1399  *      @trans: post-classic verifier transformation handler
1400  *      @save_orig: save classic BPF program
1401  *
1402  * This function effectively does the same as bpf_prog_create(), only
1403  * that it builds up its insns buffer from user space provided buffer.
1404  * It also allows for passing a bpf_aux_classic_check_t handler.
1405  */
1406 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1407                               bpf_aux_classic_check_t trans, bool save_orig)
1408 {
1409         unsigned int fsize = bpf_classic_proglen(fprog);
1410         struct bpf_prog *fp;
1411         int err;
1412
1413         /* Make sure new filter is there and in the right amounts. */
1414         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1415                 return -EINVAL;
1416
1417         fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1418         if (!fp)
1419                 return -ENOMEM;
1420
1421         if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1422                 __bpf_prog_free(fp);
1423                 return -EFAULT;
1424         }
1425
1426         fp->len = fprog->len;
1427         fp->orig_prog = NULL;
1428
1429         if (save_orig) {
1430                 err = bpf_prog_store_orig_filter(fp, fprog);
1431                 if (err) {
1432                         __bpf_prog_free(fp);
1433                         return -ENOMEM;
1434                 }
1435         }
1436
1437         /* bpf_prepare_filter() already takes care of freeing
1438          * memory in case something goes wrong.
1439          */
1440         fp = bpf_prepare_filter(fp, trans);
1441         if (IS_ERR(fp))
1442                 return PTR_ERR(fp);
1443
1444         *pfp = fp;
1445         return 0;
1446 }
1447 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1448
1449 void bpf_prog_destroy(struct bpf_prog *fp)
1450 {
1451         __bpf_prog_release(fp);
1452 }
1453 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1454
1455 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1456 {
1457         struct sk_filter *fp, *old_fp;
1458
1459         fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1460         if (!fp)
1461                 return -ENOMEM;
1462
1463         fp->prog = prog;
1464
1465         if (!__sk_filter_charge(sk, fp)) {
1466                 kfree(fp);
1467                 return -ENOMEM;
1468         }
1469         refcount_set(&fp->refcnt, 1);
1470
1471         old_fp = rcu_dereference_protected(sk->sk_filter,
1472                                            lockdep_sock_is_held(sk));
1473         rcu_assign_pointer(sk->sk_filter, fp);
1474
1475         if (old_fp)
1476                 sk_filter_uncharge(sk, old_fp);
1477
1478         return 0;
1479 }
1480
1481 static
1482 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1483 {
1484         unsigned int fsize = bpf_classic_proglen(fprog);
1485         struct bpf_prog *prog;
1486         int err;
1487
1488         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1489                 return ERR_PTR(-EPERM);
1490
1491         /* Make sure new filter is there and in the right amounts. */
1492         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1493                 return ERR_PTR(-EINVAL);
1494
1495         prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1496         if (!prog)
1497                 return ERR_PTR(-ENOMEM);
1498
1499         if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1500                 __bpf_prog_free(prog);
1501                 return ERR_PTR(-EFAULT);
1502         }
1503
1504         prog->len = fprog->len;
1505
1506         err = bpf_prog_store_orig_filter(prog, fprog);
1507         if (err) {
1508                 __bpf_prog_free(prog);
1509                 return ERR_PTR(-ENOMEM);
1510         }
1511
1512         /* bpf_prepare_filter() already takes care of freeing
1513          * memory in case something goes wrong.
1514          */
1515         return bpf_prepare_filter(prog, NULL);
1516 }
1517
1518 /**
1519  *      sk_attach_filter - attach a socket filter
1520  *      @fprog: the filter program
1521  *      @sk: the socket to use
1522  *
1523  * Attach the user's filter code. We first run some sanity checks on
1524  * it to make sure it does not explode on us later. If an error
1525  * occurs or there is insufficient memory for the filter a negative
1526  * errno code is returned. On success the return is zero.
1527  */
1528 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1529 {
1530         struct bpf_prog *prog = __get_filter(fprog, sk);
1531         int err;
1532
1533         if (IS_ERR(prog))
1534                 return PTR_ERR(prog);
1535
1536         err = __sk_attach_prog(prog, sk);
1537         if (err < 0) {
1538                 __bpf_prog_release(prog);
1539                 return err;
1540         }
1541
1542         return 0;
1543 }
1544 EXPORT_SYMBOL_GPL(sk_attach_filter);
1545
1546 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1547 {
1548         struct bpf_prog *prog = __get_filter(fprog, sk);
1549         int err;
1550
1551         if (IS_ERR(prog))
1552                 return PTR_ERR(prog);
1553
1554         if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max))
1555                 err = -ENOMEM;
1556         else
1557                 err = reuseport_attach_prog(sk, prog);
1558
1559         if (err)
1560                 __bpf_prog_release(prog);
1561
1562         return err;
1563 }
1564
1565 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1566 {
1567         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1568                 return ERR_PTR(-EPERM);
1569
1570         return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1571 }
1572
1573 int sk_attach_bpf(u32 ufd, struct sock *sk)
1574 {
1575         struct bpf_prog *prog = __get_bpf(ufd, sk);
1576         int err;
1577
1578         if (IS_ERR(prog))
1579                 return PTR_ERR(prog);
1580
1581         err = __sk_attach_prog(prog, sk);
1582         if (err < 0) {
1583                 bpf_prog_put(prog);
1584                 return err;
1585         }
1586
1587         return 0;
1588 }
1589
1590 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1591 {
1592         struct bpf_prog *prog;
1593         int err;
1594
1595         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1596                 return -EPERM;
1597
1598         prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1599         if (PTR_ERR(prog) == -EINVAL)
1600                 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1601         if (IS_ERR(prog))
1602                 return PTR_ERR(prog);
1603
1604         if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1605                 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1606                  * bpf prog (e.g. sockmap).  It depends on the
1607                  * limitation imposed by bpf_prog_load().
1608                  * Hence, sysctl_optmem_max is not checked.
1609                  */
1610                 if ((sk->sk_type != SOCK_STREAM &&
1611                      sk->sk_type != SOCK_DGRAM) ||
1612                     (sk->sk_protocol != IPPROTO_UDP &&
1613                      sk->sk_protocol != IPPROTO_TCP) ||
1614                     (sk->sk_family != AF_INET &&
1615                      sk->sk_family != AF_INET6)) {
1616                         err = -ENOTSUPP;
1617                         goto err_prog_put;
1618                 }
1619         } else {
1620                 /* BPF_PROG_TYPE_SOCKET_FILTER */
1621                 if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max)) {
1622                         err = -ENOMEM;
1623                         goto err_prog_put;
1624                 }
1625         }
1626
1627         err = reuseport_attach_prog(sk, prog);
1628 err_prog_put:
1629         if (err)
1630                 bpf_prog_put(prog);
1631
1632         return err;
1633 }
1634
1635 void sk_reuseport_prog_free(struct bpf_prog *prog)
1636 {
1637         if (!prog)
1638                 return;
1639
1640         if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1641                 bpf_prog_put(prog);
1642         else
1643                 bpf_prog_destroy(prog);
1644 }
1645
1646 struct bpf_scratchpad {
1647         union {
1648                 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1649                 u8     buff[MAX_BPF_STACK];
1650         };
1651 };
1652
1653 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1654
1655 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1656                                           unsigned int write_len)
1657 {
1658         return skb_ensure_writable(skb, write_len);
1659 }
1660
1661 static inline int bpf_try_make_writable(struct sk_buff *skb,
1662                                         unsigned int write_len)
1663 {
1664         int err = __bpf_try_make_writable(skb, write_len);
1665
1666         bpf_compute_data_pointers(skb);
1667         return err;
1668 }
1669
1670 static int bpf_try_make_head_writable(struct sk_buff *skb)
1671 {
1672         return bpf_try_make_writable(skb, skb_headlen(skb));
1673 }
1674
1675 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1676 {
1677         if (skb_at_tc_ingress(skb))
1678                 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1679 }
1680
1681 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1682 {
1683         if (skb_at_tc_ingress(skb))
1684                 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1685 }
1686
1687 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1688            const void *, from, u32, len, u64, flags)
1689 {
1690         void *ptr;
1691
1692         if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1693                 return -EINVAL;
1694         if (unlikely(offset > INT_MAX))
1695                 return -EFAULT;
1696         if (unlikely(bpf_try_make_writable(skb, offset + len)))
1697                 return -EFAULT;
1698
1699         ptr = skb->data + offset;
1700         if (flags & BPF_F_RECOMPUTE_CSUM)
1701                 __skb_postpull_rcsum(skb, ptr, len, offset);
1702
1703         memcpy(ptr, from, len);
1704
1705         if (flags & BPF_F_RECOMPUTE_CSUM)
1706                 __skb_postpush_rcsum(skb, ptr, len, offset);
1707         if (flags & BPF_F_INVALIDATE_HASH)
1708                 skb_clear_hash(skb);
1709
1710         return 0;
1711 }
1712
1713 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1714         .func           = bpf_skb_store_bytes,
1715         .gpl_only       = false,
1716         .ret_type       = RET_INTEGER,
1717         .arg1_type      = ARG_PTR_TO_CTX,
1718         .arg2_type      = ARG_ANYTHING,
1719         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
1720         .arg4_type      = ARG_CONST_SIZE,
1721         .arg5_type      = ARG_ANYTHING,
1722 };
1723
1724 int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, const void *from,
1725                           u32 len, u64 flags)
1726 {
1727         return ____bpf_skb_store_bytes(skb, offset, from, len, flags);
1728 }
1729
1730 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1731            void *, to, u32, len)
1732 {
1733         void *ptr;
1734
1735         if (unlikely(offset > INT_MAX))
1736                 goto err_clear;
1737
1738         ptr = skb_header_pointer(skb, offset, len, to);
1739         if (unlikely(!ptr))
1740                 goto err_clear;
1741         if (ptr != to)
1742                 memcpy(to, ptr, len);
1743
1744         return 0;
1745 err_clear:
1746         memset(to, 0, len);
1747         return -EFAULT;
1748 }
1749
1750 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1751         .func           = bpf_skb_load_bytes,
1752         .gpl_only       = false,
1753         .ret_type       = RET_INTEGER,
1754         .arg1_type      = ARG_PTR_TO_CTX,
1755         .arg2_type      = ARG_ANYTHING,
1756         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1757         .arg4_type      = ARG_CONST_SIZE,
1758 };
1759
1760 int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, void *to, u32 len)
1761 {
1762         return ____bpf_skb_load_bytes(skb, offset, to, len);
1763 }
1764
1765 BPF_CALL_4(bpf_flow_dissector_load_bytes,
1766            const struct bpf_flow_dissector *, ctx, u32, offset,
1767            void *, to, u32, len)
1768 {
1769         void *ptr;
1770
1771         if (unlikely(offset > 0xffff))
1772                 goto err_clear;
1773
1774         if (unlikely(!ctx->skb))
1775                 goto err_clear;
1776
1777         ptr = skb_header_pointer(ctx->skb, offset, len, to);
1778         if (unlikely(!ptr))
1779                 goto err_clear;
1780         if (ptr != to)
1781                 memcpy(to, ptr, len);
1782
1783         return 0;
1784 err_clear:
1785         memset(to, 0, len);
1786         return -EFAULT;
1787 }
1788
1789 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1790         .func           = bpf_flow_dissector_load_bytes,
1791         .gpl_only       = false,
1792         .ret_type       = RET_INTEGER,
1793         .arg1_type      = ARG_PTR_TO_CTX,
1794         .arg2_type      = ARG_ANYTHING,
1795         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1796         .arg4_type      = ARG_CONST_SIZE,
1797 };
1798
1799 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1800            u32, offset, void *, to, u32, len, u32, start_header)
1801 {
1802         u8 *end = skb_tail_pointer(skb);
1803         u8 *start, *ptr;
1804
1805         if (unlikely(offset > 0xffff))
1806                 goto err_clear;
1807
1808         switch (start_header) {
1809         case BPF_HDR_START_MAC:
1810                 if (unlikely(!skb_mac_header_was_set(skb)))
1811                         goto err_clear;
1812                 start = skb_mac_header(skb);
1813                 break;
1814         case BPF_HDR_START_NET:
1815                 start = skb_network_header(skb);
1816                 break;
1817         default:
1818                 goto err_clear;
1819         }
1820
1821         ptr = start + offset;
1822
1823         if (likely(ptr + len <= end)) {
1824                 memcpy(to, ptr, len);
1825                 return 0;
1826         }
1827
1828 err_clear:
1829         memset(to, 0, len);
1830         return -EFAULT;
1831 }
1832
1833 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1834         .func           = bpf_skb_load_bytes_relative,
1835         .gpl_only       = false,
1836         .ret_type       = RET_INTEGER,
1837         .arg1_type      = ARG_PTR_TO_CTX,
1838         .arg2_type      = ARG_ANYTHING,
1839         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1840         .arg4_type      = ARG_CONST_SIZE,
1841         .arg5_type      = ARG_ANYTHING,
1842 };
1843
1844 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1845 {
1846         /* Idea is the following: should the needed direct read/write
1847          * test fail during runtime, we can pull in more data and redo
1848          * again, since implicitly, we invalidate previous checks here.
1849          *
1850          * Or, since we know how much we need to make read/writeable,
1851          * this can be done once at the program beginning for direct
1852          * access case. By this we overcome limitations of only current
1853          * headroom being accessible.
1854          */
1855         return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1856 }
1857
1858 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1859         .func           = bpf_skb_pull_data,
1860         .gpl_only       = false,
1861         .ret_type       = RET_INTEGER,
1862         .arg1_type      = ARG_PTR_TO_CTX,
1863         .arg2_type      = ARG_ANYTHING,
1864 };
1865
1866 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1867 {
1868         return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1869 }
1870
1871 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1872         .func           = bpf_sk_fullsock,
1873         .gpl_only       = false,
1874         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
1875         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
1876 };
1877
1878 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1879                                            unsigned int write_len)
1880 {
1881         return __bpf_try_make_writable(skb, write_len);
1882 }
1883
1884 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1885 {
1886         /* Idea is the following: should the needed direct read/write
1887          * test fail during runtime, we can pull in more data and redo
1888          * again, since implicitly, we invalidate previous checks here.
1889          *
1890          * Or, since we know how much we need to make read/writeable,
1891          * this can be done once at the program beginning for direct
1892          * access case. By this we overcome limitations of only current
1893          * headroom being accessible.
1894          */
1895         return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1896 }
1897
1898 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1899         .func           = sk_skb_pull_data,
1900         .gpl_only       = false,
1901         .ret_type       = RET_INTEGER,
1902         .arg1_type      = ARG_PTR_TO_CTX,
1903         .arg2_type      = ARG_ANYTHING,
1904 };
1905
1906 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1907            u64, from, u64, to, u64, flags)
1908 {
1909         __sum16 *ptr;
1910
1911         if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1912                 return -EINVAL;
1913         if (unlikely(offset > 0xffff || offset & 1))
1914                 return -EFAULT;
1915         if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1916                 return -EFAULT;
1917
1918         ptr = (__sum16 *)(skb->data + offset);
1919         switch (flags & BPF_F_HDR_FIELD_MASK) {
1920         case 0:
1921                 if (unlikely(from != 0))
1922                         return -EINVAL;
1923
1924                 csum_replace_by_diff(ptr, to);
1925                 break;
1926         case 2:
1927                 csum_replace2(ptr, from, to);
1928                 break;
1929         case 4:
1930                 csum_replace4(ptr, from, to);
1931                 break;
1932         default:
1933                 return -EINVAL;
1934         }
1935
1936         return 0;
1937 }
1938
1939 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1940         .func           = bpf_l3_csum_replace,
1941         .gpl_only       = false,
1942         .ret_type       = RET_INTEGER,
1943         .arg1_type      = ARG_PTR_TO_CTX,
1944         .arg2_type      = ARG_ANYTHING,
1945         .arg3_type      = ARG_ANYTHING,
1946         .arg4_type      = ARG_ANYTHING,
1947         .arg5_type      = ARG_ANYTHING,
1948 };
1949
1950 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1951            u64, from, u64, to, u64, flags)
1952 {
1953         bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1954         bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1955         bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1956         __sum16 *ptr;
1957
1958         if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1959                                BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1960                 return -EINVAL;
1961         if (unlikely(offset > 0xffff || offset & 1))
1962                 return -EFAULT;
1963         if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1964                 return -EFAULT;
1965
1966         ptr = (__sum16 *)(skb->data + offset);
1967         if (is_mmzero && !do_mforce && !*ptr)
1968                 return 0;
1969
1970         switch (flags & BPF_F_HDR_FIELD_MASK) {
1971         case 0:
1972                 if (unlikely(from != 0))
1973                         return -EINVAL;
1974
1975                 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1976                 break;
1977         case 2:
1978                 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1979                 break;
1980         case 4:
1981                 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1982                 break;
1983         default:
1984                 return -EINVAL;
1985         }
1986
1987         if (is_mmzero && !*ptr)
1988                 *ptr = CSUM_MANGLED_0;
1989         return 0;
1990 }
1991
1992 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1993         .func           = bpf_l4_csum_replace,
1994         .gpl_only       = false,
1995         .ret_type       = RET_INTEGER,
1996         .arg1_type      = ARG_PTR_TO_CTX,
1997         .arg2_type      = ARG_ANYTHING,
1998         .arg3_type      = ARG_ANYTHING,
1999         .arg4_type      = ARG_ANYTHING,
2000         .arg5_type      = ARG_ANYTHING,
2001 };
2002
2003 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
2004            __be32 *, to, u32, to_size, __wsum, seed)
2005 {
2006         struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
2007         u32 diff_size = from_size + to_size;
2008         int i, j = 0;
2009
2010         /* This is quite flexible, some examples:
2011          *
2012          * from_size == 0, to_size > 0,  seed := csum --> pushing data
2013          * from_size > 0,  to_size == 0, seed := csum --> pulling data
2014          * from_size > 0,  to_size > 0,  seed := 0    --> diffing data
2015          *
2016          * Even for diffing, from_size and to_size don't need to be equal.
2017          */
2018         if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
2019                      diff_size > sizeof(sp->diff)))
2020                 return -EINVAL;
2021
2022         for (i = 0; i < from_size / sizeof(__be32); i++, j++)
2023                 sp->diff[j] = ~from[i];
2024         for (i = 0; i <   to_size / sizeof(__be32); i++, j++)
2025                 sp->diff[j] = to[i];
2026
2027         return csum_partial(sp->diff, diff_size, seed);
2028 }
2029
2030 static const struct bpf_func_proto bpf_csum_diff_proto = {
2031         .func           = bpf_csum_diff,
2032         .gpl_only       = false,
2033         .pkt_access     = true,
2034         .ret_type       = RET_INTEGER,
2035         .arg1_type      = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2036         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
2037         .arg3_type      = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2038         .arg4_type      = ARG_CONST_SIZE_OR_ZERO,
2039         .arg5_type      = ARG_ANYTHING,
2040 };
2041
2042 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2043 {
2044         /* The interface is to be used in combination with bpf_csum_diff()
2045          * for direct packet writes. csum rotation for alignment as well
2046          * as emulating csum_sub() can be done from the eBPF program.
2047          */
2048         if (skb->ip_summed == CHECKSUM_COMPLETE)
2049                 return (skb->csum = csum_add(skb->csum, csum));
2050
2051         return -ENOTSUPP;
2052 }
2053
2054 static const struct bpf_func_proto bpf_csum_update_proto = {
2055         .func           = bpf_csum_update,
2056         .gpl_only       = false,
2057         .ret_type       = RET_INTEGER,
2058         .arg1_type      = ARG_PTR_TO_CTX,
2059         .arg2_type      = ARG_ANYTHING,
2060 };
2061
2062 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
2063 {
2064         /* The interface is to be used in combination with bpf_skb_adjust_room()
2065          * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET
2066          * is passed as flags, for example.
2067          */
2068         switch (level) {
2069         case BPF_CSUM_LEVEL_INC:
2070                 __skb_incr_checksum_unnecessary(skb);
2071                 break;
2072         case BPF_CSUM_LEVEL_DEC:
2073                 __skb_decr_checksum_unnecessary(skb);
2074                 break;
2075         case BPF_CSUM_LEVEL_RESET:
2076                 __skb_reset_checksum_unnecessary(skb);
2077                 break;
2078         case BPF_CSUM_LEVEL_QUERY:
2079                 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2080                        skb->csum_level : -EACCES;
2081         default:
2082                 return -EINVAL;
2083         }
2084
2085         return 0;
2086 }
2087
2088 static const struct bpf_func_proto bpf_csum_level_proto = {
2089         .func           = bpf_csum_level,
2090         .gpl_only       = false,
2091         .ret_type       = RET_INTEGER,
2092         .arg1_type      = ARG_PTR_TO_CTX,
2093         .arg2_type      = ARG_ANYTHING,
2094 };
2095
2096 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2097 {
2098         return dev_forward_skb_nomtu(dev, skb);
2099 }
2100
2101 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2102                                       struct sk_buff *skb)
2103 {
2104         int ret = ____dev_forward_skb(dev, skb, false);
2105
2106         if (likely(!ret)) {
2107                 skb->dev = dev;
2108                 ret = netif_rx(skb);
2109         }
2110
2111         return ret;
2112 }
2113
2114 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2115 {
2116         int ret;
2117
2118         if (dev_xmit_recursion()) {
2119                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2120                 kfree_skb(skb);
2121                 return -ENETDOWN;
2122         }
2123
2124         skb->dev = dev;
2125         skb_set_redirected_noclear(skb, skb_at_tc_ingress(skb));
2126         skb_clear_tstamp(skb);
2127
2128         dev_xmit_recursion_inc();
2129         ret = dev_queue_xmit(skb);
2130         dev_xmit_recursion_dec();
2131
2132         return ret;
2133 }
2134
2135 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2136                                  u32 flags)
2137 {
2138         unsigned int mlen = skb_network_offset(skb);
2139
2140         if (unlikely(skb->len <= mlen)) {
2141                 kfree_skb(skb);
2142                 return -ERANGE;
2143         }
2144
2145         if (mlen) {
2146                 __skb_pull(skb, mlen);
2147
2148                 /* At ingress, the mac header has already been pulled once.
2149                  * At egress, skb_pospull_rcsum has to be done in case that
2150                  * the skb is originated from ingress (i.e. a forwarded skb)
2151                  * to ensure that rcsum starts at net header.
2152                  */
2153                 if (!skb_at_tc_ingress(skb))
2154                         skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2155         }
2156         skb_pop_mac_header(skb);
2157         skb_reset_mac_len(skb);
2158         return flags & BPF_F_INGRESS ?
2159                __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2160 }
2161
2162 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2163                                  u32 flags)
2164 {
2165         /* Verify that a link layer header is carried */
2166         if (unlikely(skb->mac_header >= skb->network_header || skb->len == 0)) {
2167                 kfree_skb(skb);
2168                 return -ERANGE;
2169         }
2170
2171         bpf_push_mac_rcsum(skb);
2172         return flags & BPF_F_INGRESS ?
2173                __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2174 }
2175
2176 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2177                           u32 flags)
2178 {
2179         if (dev_is_mac_header_xmit(dev))
2180                 return __bpf_redirect_common(skb, dev, flags);
2181         else
2182                 return __bpf_redirect_no_mac(skb, dev, flags);
2183 }
2184
2185 #if IS_ENABLED(CONFIG_IPV6)
2186 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2187                             struct net_device *dev, struct bpf_nh_params *nh)
2188 {
2189         u32 hh_len = LL_RESERVED_SPACE(dev);
2190         const struct in6_addr *nexthop;
2191         struct dst_entry *dst = NULL;
2192         struct neighbour *neigh;
2193
2194         if (dev_xmit_recursion()) {
2195                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2196                 goto out_drop;
2197         }
2198
2199         skb->dev = dev;
2200         skb_clear_tstamp(skb);
2201
2202         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2203                 skb = skb_expand_head(skb, hh_len);
2204                 if (!skb)
2205                         return -ENOMEM;
2206         }
2207
2208         rcu_read_lock();
2209         if (!nh) {
2210                 dst = skb_dst(skb);
2211                 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2212                                       &ipv6_hdr(skb)->daddr);
2213         } else {
2214                 nexthop = &nh->ipv6_nh;
2215         }
2216         neigh = ip_neigh_gw6(dev, nexthop);
2217         if (likely(!IS_ERR(neigh))) {
2218                 int ret;
2219
2220                 sock_confirm_neigh(skb, neigh);
2221                 local_bh_disable();
2222                 dev_xmit_recursion_inc();
2223                 ret = neigh_output(neigh, skb, false);
2224                 dev_xmit_recursion_dec();
2225                 local_bh_enable();
2226                 rcu_read_unlock();
2227                 return ret;
2228         }
2229         rcu_read_unlock_bh();
2230         if (dst)
2231                 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2232 out_drop:
2233         kfree_skb(skb);
2234         return -ENETDOWN;
2235 }
2236
2237 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2238                                    struct bpf_nh_params *nh)
2239 {
2240         const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2241         struct net *net = dev_net(dev);
2242         int err, ret = NET_XMIT_DROP;
2243
2244         if (!nh) {
2245                 struct dst_entry *dst;
2246                 struct flowi6 fl6 = {
2247                         .flowi6_flags = FLOWI_FLAG_ANYSRC,
2248                         .flowi6_mark  = skb->mark,
2249                         .flowlabel    = ip6_flowinfo(ip6h),
2250                         .flowi6_oif   = dev->ifindex,
2251                         .flowi6_proto = ip6h->nexthdr,
2252                         .daddr        = ip6h->daddr,
2253                         .saddr        = ip6h->saddr,
2254                 };
2255
2256                 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2257                 if (IS_ERR(dst))
2258                         goto out_drop;
2259
2260                 skb_dst_set(skb, dst);
2261         } else if (nh->nh_family != AF_INET6) {
2262                 goto out_drop;
2263         }
2264
2265         err = bpf_out_neigh_v6(net, skb, dev, nh);
2266         if (unlikely(net_xmit_eval(err)))
2267                 dev->stats.tx_errors++;
2268         else
2269                 ret = NET_XMIT_SUCCESS;
2270         goto out_xmit;
2271 out_drop:
2272         dev->stats.tx_errors++;
2273         kfree_skb(skb);
2274 out_xmit:
2275         return ret;
2276 }
2277 #else
2278 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2279                                    struct bpf_nh_params *nh)
2280 {
2281         kfree_skb(skb);
2282         return NET_XMIT_DROP;
2283 }
2284 #endif /* CONFIG_IPV6 */
2285
2286 #if IS_ENABLED(CONFIG_INET)
2287 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2288                             struct net_device *dev, struct bpf_nh_params *nh)
2289 {
2290         u32 hh_len = LL_RESERVED_SPACE(dev);
2291         struct neighbour *neigh;
2292         bool is_v6gw = false;
2293
2294         if (dev_xmit_recursion()) {
2295                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2296                 goto out_drop;
2297         }
2298
2299         skb->dev = dev;
2300         skb_clear_tstamp(skb);
2301
2302         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2303                 skb = skb_expand_head(skb, hh_len);
2304                 if (!skb)
2305                         return -ENOMEM;
2306         }
2307
2308         rcu_read_lock();
2309         if (!nh) {
2310                 struct dst_entry *dst = skb_dst(skb);
2311                 struct rtable *rt = container_of(dst, struct rtable, dst);
2312
2313                 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2314         } else if (nh->nh_family == AF_INET6) {
2315                 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2316                 is_v6gw = true;
2317         } else if (nh->nh_family == AF_INET) {
2318                 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2319         } else {
2320                 rcu_read_unlock();
2321                 goto out_drop;
2322         }
2323
2324         if (likely(!IS_ERR(neigh))) {
2325                 int ret;
2326
2327                 sock_confirm_neigh(skb, neigh);
2328                 local_bh_disable();
2329                 dev_xmit_recursion_inc();
2330                 ret = neigh_output(neigh, skb, is_v6gw);
2331                 dev_xmit_recursion_dec();
2332                 local_bh_enable();
2333                 rcu_read_unlock();
2334                 return ret;
2335         }
2336         rcu_read_unlock();
2337 out_drop:
2338         kfree_skb(skb);
2339         return -ENETDOWN;
2340 }
2341
2342 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2343                                    struct bpf_nh_params *nh)
2344 {
2345         const struct iphdr *ip4h = ip_hdr(skb);
2346         struct net *net = dev_net(dev);
2347         int err, ret = NET_XMIT_DROP;
2348
2349         if (!nh) {
2350                 struct flowi4 fl4 = {
2351                         .flowi4_flags = FLOWI_FLAG_ANYSRC,
2352                         .flowi4_mark  = skb->mark,
2353                         .flowi4_tos   = RT_TOS(ip4h->tos),
2354                         .flowi4_oif   = dev->ifindex,
2355                         .flowi4_proto = ip4h->protocol,
2356                         .daddr        = ip4h->daddr,
2357                         .saddr        = ip4h->saddr,
2358                 };
2359                 struct rtable *rt;
2360
2361                 rt = ip_route_output_flow(net, &fl4, NULL);
2362                 if (IS_ERR(rt))
2363                         goto out_drop;
2364                 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2365                         ip_rt_put(rt);
2366                         goto out_drop;
2367                 }
2368
2369                 skb_dst_set(skb, &rt->dst);
2370         }
2371
2372         err = bpf_out_neigh_v4(net, skb, dev, nh);
2373         if (unlikely(net_xmit_eval(err)))
2374                 dev->stats.tx_errors++;
2375         else
2376                 ret = NET_XMIT_SUCCESS;
2377         goto out_xmit;
2378 out_drop:
2379         dev->stats.tx_errors++;
2380         kfree_skb(skb);
2381 out_xmit:
2382         return ret;
2383 }
2384 #else
2385 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2386                                    struct bpf_nh_params *nh)
2387 {
2388         kfree_skb(skb);
2389         return NET_XMIT_DROP;
2390 }
2391 #endif /* CONFIG_INET */
2392
2393 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2394                                 struct bpf_nh_params *nh)
2395 {
2396         struct ethhdr *ethh = eth_hdr(skb);
2397
2398         if (unlikely(skb->mac_header >= skb->network_header))
2399                 goto out;
2400         bpf_push_mac_rcsum(skb);
2401         if (is_multicast_ether_addr(ethh->h_dest))
2402                 goto out;
2403
2404         skb_pull(skb, sizeof(*ethh));
2405         skb_unset_mac_header(skb);
2406         skb_reset_network_header(skb);
2407
2408         if (skb->protocol == htons(ETH_P_IP))
2409                 return __bpf_redirect_neigh_v4(skb, dev, nh);
2410         else if (skb->protocol == htons(ETH_P_IPV6))
2411                 return __bpf_redirect_neigh_v6(skb, dev, nh);
2412 out:
2413         kfree_skb(skb);
2414         return -ENOTSUPP;
2415 }
2416
2417 /* Internal, non-exposed redirect flags. */
2418 enum {
2419         BPF_F_NEIGH     = (1ULL << 1),
2420         BPF_F_PEER      = (1ULL << 2),
2421         BPF_F_NEXTHOP   = (1ULL << 3),
2422 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2423 };
2424
2425 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2426 {
2427         struct net_device *dev;
2428         struct sk_buff *clone;
2429         int ret;
2430
2431         if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2432                 return -EINVAL;
2433
2434         dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2435         if (unlikely(!dev))
2436                 return -EINVAL;
2437
2438         clone = skb_clone(skb, GFP_ATOMIC);
2439         if (unlikely(!clone))
2440                 return -ENOMEM;
2441
2442         /* For direct write, we need to keep the invariant that the skbs
2443          * we're dealing with need to be uncloned. Should uncloning fail
2444          * here, we need to free the just generated clone to unclone once
2445          * again.
2446          */
2447         ret = bpf_try_make_head_writable(skb);
2448         if (unlikely(ret)) {
2449                 kfree_skb(clone);
2450                 return -ENOMEM;
2451         }
2452
2453         return __bpf_redirect(clone, dev, flags);
2454 }
2455
2456 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2457         .func           = bpf_clone_redirect,
2458         .gpl_only       = false,
2459         .ret_type       = RET_INTEGER,
2460         .arg1_type      = ARG_PTR_TO_CTX,
2461         .arg2_type      = ARG_ANYTHING,
2462         .arg3_type      = ARG_ANYTHING,
2463 };
2464
2465 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2466 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2467
2468 int skb_do_redirect(struct sk_buff *skb)
2469 {
2470         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2471         struct net *net = dev_net(skb->dev);
2472         struct net_device *dev;
2473         u32 flags = ri->flags;
2474
2475         dev = dev_get_by_index_rcu(net, ri->tgt_index);
2476         ri->tgt_index = 0;
2477         ri->flags = 0;
2478         if (unlikely(!dev))
2479                 goto out_drop;
2480         if (flags & BPF_F_PEER) {
2481                 const struct net_device_ops *ops = dev->netdev_ops;
2482
2483                 if (unlikely(!ops->ndo_get_peer_dev ||
2484                              !skb_at_tc_ingress(skb)))
2485                         goto out_drop;
2486                 dev = ops->ndo_get_peer_dev(dev);
2487                 if (unlikely(!dev ||
2488                              !(dev->flags & IFF_UP) ||
2489                              net_eq(net, dev_net(dev))))
2490                         goto out_drop;
2491                 skb->dev = dev;
2492                 return -EAGAIN;
2493         }
2494         return flags & BPF_F_NEIGH ?
2495                __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2496                                     &ri->nh : NULL) :
2497                __bpf_redirect(skb, dev, flags);
2498 out_drop:
2499         kfree_skb(skb);
2500         return -EINVAL;
2501 }
2502
2503 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2504 {
2505         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2506
2507         if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2508                 return TC_ACT_SHOT;
2509
2510         ri->flags = flags;
2511         ri->tgt_index = ifindex;
2512
2513         return TC_ACT_REDIRECT;
2514 }
2515
2516 static const struct bpf_func_proto bpf_redirect_proto = {
2517         .func           = bpf_redirect,
2518         .gpl_only       = false,
2519         .ret_type       = RET_INTEGER,
2520         .arg1_type      = ARG_ANYTHING,
2521         .arg2_type      = ARG_ANYTHING,
2522 };
2523
2524 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2525 {
2526         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2527
2528         if (unlikely(flags))
2529                 return TC_ACT_SHOT;
2530
2531         ri->flags = BPF_F_PEER;
2532         ri->tgt_index = ifindex;
2533
2534         return TC_ACT_REDIRECT;
2535 }
2536
2537 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2538         .func           = bpf_redirect_peer,
2539         .gpl_only       = false,
2540         .ret_type       = RET_INTEGER,
2541         .arg1_type      = ARG_ANYTHING,
2542         .arg2_type      = ARG_ANYTHING,
2543 };
2544
2545 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2546            int, plen, u64, flags)
2547 {
2548         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2549
2550         if (unlikely((plen && plen < sizeof(*params)) || flags))
2551                 return TC_ACT_SHOT;
2552
2553         ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2554         ri->tgt_index = ifindex;
2555
2556         BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2557         if (plen)
2558                 memcpy(&ri->nh, params, sizeof(ri->nh));
2559
2560         return TC_ACT_REDIRECT;
2561 }
2562
2563 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2564         .func           = bpf_redirect_neigh,
2565         .gpl_only       = false,
2566         .ret_type       = RET_INTEGER,
2567         .arg1_type      = ARG_ANYTHING,
2568         .arg2_type      = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2569         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
2570         .arg4_type      = ARG_ANYTHING,
2571 };
2572
2573 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2574 {
2575         msg->apply_bytes = bytes;
2576         return 0;
2577 }
2578
2579 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2580         .func           = bpf_msg_apply_bytes,
2581         .gpl_only       = false,
2582         .ret_type       = RET_INTEGER,
2583         .arg1_type      = ARG_PTR_TO_CTX,
2584         .arg2_type      = ARG_ANYTHING,
2585 };
2586
2587 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2588 {
2589         msg->cork_bytes = bytes;
2590         return 0;
2591 }
2592
2593 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2594         .func           = bpf_msg_cork_bytes,
2595         .gpl_only       = false,
2596         .ret_type       = RET_INTEGER,
2597         .arg1_type      = ARG_PTR_TO_CTX,
2598         .arg2_type      = ARG_ANYTHING,
2599 };
2600
2601 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2602            u32, end, u64, flags)
2603 {
2604         u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2605         u32 first_sge, last_sge, i, shift, bytes_sg_total;
2606         struct scatterlist *sge;
2607         u8 *raw, *to, *from;
2608         struct page *page;
2609
2610         if (unlikely(flags || end <= start))
2611                 return -EINVAL;
2612
2613         /* First find the starting scatterlist element */
2614         i = msg->sg.start;
2615         do {
2616                 offset += len;
2617                 len = sk_msg_elem(msg, i)->length;
2618                 if (start < offset + len)
2619                         break;
2620                 sk_msg_iter_var_next(i);
2621         } while (i != msg->sg.end);
2622
2623         if (unlikely(start >= offset + len))
2624                 return -EINVAL;
2625
2626         first_sge = i;
2627         /* The start may point into the sg element so we need to also
2628          * account for the headroom.
2629          */
2630         bytes_sg_total = start - offset + bytes;
2631         if (!test_bit(i, msg->sg.copy) && bytes_sg_total <= len)
2632                 goto out;
2633
2634         /* At this point we need to linearize multiple scatterlist
2635          * elements or a single shared page. Either way we need to
2636          * copy into a linear buffer exclusively owned by BPF. Then
2637          * place the buffer in the scatterlist and fixup the original
2638          * entries by removing the entries now in the linear buffer
2639          * and shifting the remaining entries. For now we do not try
2640          * to copy partial entries to avoid complexity of running out
2641          * of sg_entry slots. The downside is reading a single byte
2642          * will copy the entire sg entry.
2643          */
2644         do {
2645                 copy += sk_msg_elem(msg, i)->length;
2646                 sk_msg_iter_var_next(i);
2647                 if (bytes_sg_total <= copy)
2648                         break;
2649         } while (i != msg->sg.end);
2650         last_sge = i;
2651
2652         if (unlikely(bytes_sg_total > copy))
2653                 return -EINVAL;
2654
2655         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2656                            get_order(copy));
2657         if (unlikely(!page))
2658                 return -ENOMEM;
2659
2660         raw = page_address(page);
2661         i = first_sge;
2662         do {
2663                 sge = sk_msg_elem(msg, i);
2664                 from = sg_virt(sge);
2665                 len = sge->length;
2666                 to = raw + poffset;
2667
2668                 memcpy(to, from, len);
2669                 poffset += len;
2670                 sge->length = 0;
2671                 put_page(sg_page(sge));
2672
2673                 sk_msg_iter_var_next(i);
2674         } while (i != last_sge);
2675
2676         sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2677
2678         /* To repair sg ring we need to shift entries. If we only
2679          * had a single entry though we can just replace it and
2680          * be done. Otherwise walk the ring and shift the entries.
2681          */
2682         WARN_ON_ONCE(last_sge == first_sge);
2683         shift = last_sge > first_sge ?
2684                 last_sge - first_sge - 1 :
2685                 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2686         if (!shift)
2687                 goto out;
2688
2689         i = first_sge;
2690         sk_msg_iter_var_next(i);
2691         do {
2692                 u32 move_from;
2693
2694                 if (i + shift >= NR_MSG_FRAG_IDS)
2695                         move_from = i + shift - NR_MSG_FRAG_IDS;
2696                 else
2697                         move_from = i + shift;
2698                 if (move_from == msg->sg.end)
2699                         break;
2700
2701                 msg->sg.data[i] = msg->sg.data[move_from];
2702                 msg->sg.data[move_from].length = 0;
2703                 msg->sg.data[move_from].page_link = 0;
2704                 msg->sg.data[move_from].offset = 0;
2705                 sk_msg_iter_var_next(i);
2706         } while (1);
2707
2708         msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2709                       msg->sg.end - shift + NR_MSG_FRAG_IDS :
2710                       msg->sg.end - shift;
2711 out:
2712         msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2713         msg->data_end = msg->data + bytes;
2714         return 0;
2715 }
2716
2717 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2718         .func           = bpf_msg_pull_data,
2719         .gpl_only       = false,
2720         .ret_type       = RET_INTEGER,
2721         .arg1_type      = ARG_PTR_TO_CTX,
2722         .arg2_type      = ARG_ANYTHING,
2723         .arg3_type      = ARG_ANYTHING,
2724         .arg4_type      = ARG_ANYTHING,
2725 };
2726
2727 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2728            u32, len, u64, flags)
2729 {
2730         struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2731         u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2732         u8 *raw, *to, *from;
2733         struct page *page;
2734
2735         if (unlikely(flags))
2736                 return -EINVAL;
2737
2738         if (unlikely(len == 0))
2739                 return 0;
2740
2741         /* First find the starting scatterlist element */
2742         i = msg->sg.start;
2743         do {
2744                 offset += l;
2745                 l = sk_msg_elem(msg, i)->length;
2746
2747                 if (start < offset + l)
2748                         break;
2749                 sk_msg_iter_var_next(i);
2750         } while (i != msg->sg.end);
2751
2752         if (start >= offset + l)
2753                 return -EINVAL;
2754
2755         space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2756
2757         /* If no space available will fallback to copy, we need at
2758          * least one scatterlist elem available to push data into
2759          * when start aligns to the beginning of an element or two
2760          * when it falls inside an element. We handle the start equals
2761          * offset case because its the common case for inserting a
2762          * header.
2763          */
2764         if (!space || (space == 1 && start != offset))
2765                 copy = msg->sg.data[i].length;
2766
2767         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2768                            get_order(copy + len));
2769         if (unlikely(!page))
2770                 return -ENOMEM;
2771
2772         if (copy) {
2773                 int front, back;
2774
2775                 raw = page_address(page);
2776
2777                 psge = sk_msg_elem(msg, i);
2778                 front = start - offset;
2779                 back = psge->length - front;
2780                 from = sg_virt(psge);
2781
2782                 if (front)
2783                         memcpy(raw, from, front);
2784
2785                 if (back) {
2786                         from += front;
2787                         to = raw + front + len;
2788
2789                         memcpy(to, from, back);
2790                 }
2791
2792                 put_page(sg_page(psge));
2793         } else if (start - offset) {
2794                 psge = sk_msg_elem(msg, i);
2795                 rsge = sk_msg_elem_cpy(msg, i);
2796
2797                 psge->length = start - offset;
2798                 rsge.length -= psge->length;
2799                 rsge.offset += start;
2800
2801                 sk_msg_iter_var_next(i);
2802                 sg_unmark_end(psge);
2803                 sg_unmark_end(&rsge);
2804                 sk_msg_iter_next(msg, end);
2805         }
2806
2807         /* Slot(s) to place newly allocated data */
2808         new = i;
2809
2810         /* Shift one or two slots as needed */
2811         if (!copy) {
2812                 sge = sk_msg_elem_cpy(msg, i);
2813
2814                 sk_msg_iter_var_next(i);
2815                 sg_unmark_end(&sge);
2816                 sk_msg_iter_next(msg, end);
2817
2818                 nsge = sk_msg_elem_cpy(msg, i);
2819                 if (rsge.length) {
2820                         sk_msg_iter_var_next(i);
2821                         nnsge = sk_msg_elem_cpy(msg, i);
2822                 }
2823
2824                 while (i != msg->sg.end) {
2825                         msg->sg.data[i] = sge;
2826                         sge = nsge;
2827                         sk_msg_iter_var_next(i);
2828                         if (rsge.length) {
2829                                 nsge = nnsge;
2830                                 nnsge = sk_msg_elem_cpy(msg, i);
2831                         } else {
2832                                 nsge = sk_msg_elem_cpy(msg, i);
2833                         }
2834                 }
2835         }
2836
2837         /* Place newly allocated data buffer */
2838         sk_mem_charge(msg->sk, len);
2839         msg->sg.size += len;
2840         __clear_bit(new, msg->sg.copy);
2841         sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2842         if (rsge.length) {
2843                 get_page(sg_page(&rsge));
2844                 sk_msg_iter_var_next(new);
2845                 msg->sg.data[new] = rsge;
2846         }
2847
2848         sk_msg_compute_data_pointers(msg);
2849         return 0;
2850 }
2851
2852 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2853         .func           = bpf_msg_push_data,
2854         .gpl_only       = false,
2855         .ret_type       = RET_INTEGER,
2856         .arg1_type      = ARG_PTR_TO_CTX,
2857         .arg2_type      = ARG_ANYTHING,
2858         .arg3_type      = ARG_ANYTHING,
2859         .arg4_type      = ARG_ANYTHING,
2860 };
2861
2862 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2863 {
2864         int prev;
2865
2866         do {
2867                 prev = i;
2868                 sk_msg_iter_var_next(i);
2869                 msg->sg.data[prev] = msg->sg.data[i];
2870         } while (i != msg->sg.end);
2871
2872         sk_msg_iter_prev(msg, end);
2873 }
2874
2875 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2876 {
2877         struct scatterlist tmp, sge;
2878
2879         sk_msg_iter_next(msg, end);
2880         sge = sk_msg_elem_cpy(msg, i);
2881         sk_msg_iter_var_next(i);
2882         tmp = sk_msg_elem_cpy(msg, i);
2883
2884         while (i != msg->sg.end) {
2885                 msg->sg.data[i] = sge;
2886                 sk_msg_iter_var_next(i);
2887                 sge = tmp;
2888                 tmp = sk_msg_elem_cpy(msg, i);
2889         }
2890 }
2891
2892 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2893            u32, len, u64, flags)
2894 {
2895         u32 i = 0, l = 0, space, offset = 0;
2896         u64 last = start + len;
2897         int pop;
2898
2899         if (unlikely(flags))
2900                 return -EINVAL;
2901
2902         /* First find the starting scatterlist element */
2903         i = msg->sg.start;
2904         do {
2905                 offset += l;
2906                 l = sk_msg_elem(msg, i)->length;
2907
2908                 if (start < offset + l)
2909                         break;
2910                 sk_msg_iter_var_next(i);
2911         } while (i != msg->sg.end);
2912
2913         /* Bounds checks: start and pop must be inside message */
2914         if (start >= offset + l || last >= msg->sg.size)
2915                 return -EINVAL;
2916
2917         space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2918
2919         pop = len;
2920         /* --------------| offset
2921          * -| start      |-------- len -------|
2922          *
2923          *  |----- a ----|-------- pop -------|----- b ----|
2924          *  |______________________________________________| length
2925          *
2926          *
2927          * a:   region at front of scatter element to save
2928          * b:   region at back of scatter element to save when length > A + pop
2929          * pop: region to pop from element, same as input 'pop' here will be
2930          *      decremented below per iteration.
2931          *
2932          * Two top-level cases to handle when start != offset, first B is non
2933          * zero and second B is zero corresponding to when a pop includes more
2934          * than one element.
2935          *
2936          * Then if B is non-zero AND there is no space allocate space and
2937          * compact A, B regions into page. If there is space shift ring to
2938          * the rigth free'ing the next element in ring to place B, leaving
2939          * A untouched except to reduce length.
2940          */
2941         if (start != offset) {
2942                 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2943                 int a = start;
2944                 int b = sge->length - pop - a;
2945
2946                 sk_msg_iter_var_next(i);
2947
2948                 if (pop < sge->length - a) {
2949                         if (space) {
2950                                 sge->length = a;
2951                                 sk_msg_shift_right(msg, i);
2952                                 nsge = sk_msg_elem(msg, i);
2953                                 get_page(sg_page(sge));
2954                                 sg_set_page(nsge,
2955                                             sg_page(sge),
2956                                             b, sge->offset + pop + a);
2957                         } else {
2958                                 struct page *page, *orig;
2959                                 u8 *to, *from;
2960
2961                                 page = alloc_pages(__GFP_NOWARN |
2962                                                    __GFP_COMP   | GFP_ATOMIC,
2963                                                    get_order(a + b));
2964                                 if (unlikely(!page))
2965                                         return -ENOMEM;
2966
2967                                 sge->length = a;
2968                                 orig = sg_page(sge);
2969                                 from = sg_virt(sge);
2970                                 to = page_address(page);
2971                                 memcpy(to, from, a);
2972                                 memcpy(to + a, from + a + pop, b);
2973                                 sg_set_page(sge, page, a + b, 0);
2974                                 put_page(orig);
2975                         }
2976                         pop = 0;
2977                 } else if (pop >= sge->length - a) {
2978                         pop -= (sge->length - a);
2979                         sge->length = a;
2980                 }
2981         }
2982
2983         /* From above the current layout _must_ be as follows,
2984          *
2985          * -| offset
2986          * -| start
2987          *
2988          *  |---- pop ---|---------------- b ------------|
2989          *  |____________________________________________| length
2990          *
2991          * Offset and start of the current msg elem are equal because in the
2992          * previous case we handled offset != start and either consumed the
2993          * entire element and advanced to the next element OR pop == 0.
2994          *
2995          * Two cases to handle here are first pop is less than the length
2996          * leaving some remainder b above. Simply adjust the element's layout
2997          * in this case. Or pop >= length of the element so that b = 0. In this
2998          * case advance to next element decrementing pop.
2999          */
3000         while (pop) {
3001                 struct scatterlist *sge = sk_msg_elem(msg, i);
3002
3003                 if (pop < sge->length) {
3004                         sge->length -= pop;
3005                         sge->offset += pop;
3006                         pop = 0;
3007                 } else {
3008                         pop -= sge->length;
3009                         sk_msg_shift_left(msg, i);
3010                 }
3011                 sk_msg_iter_var_next(i);
3012         }
3013
3014         sk_mem_uncharge(msg->sk, len - pop);
3015         msg->sg.size -= (len - pop);
3016         sk_msg_compute_data_pointers(msg);
3017         return 0;
3018 }
3019
3020 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
3021         .func           = bpf_msg_pop_data,
3022         .gpl_only       = false,
3023         .ret_type       = RET_INTEGER,
3024         .arg1_type      = ARG_PTR_TO_CTX,
3025         .arg2_type      = ARG_ANYTHING,
3026         .arg3_type      = ARG_ANYTHING,
3027         .arg4_type      = ARG_ANYTHING,
3028 };
3029
3030 #ifdef CONFIG_CGROUP_NET_CLASSID
3031 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3032 {
3033         return __task_get_classid(current);
3034 }
3035
3036 const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3037         .func           = bpf_get_cgroup_classid_curr,
3038         .gpl_only       = false,
3039         .ret_type       = RET_INTEGER,
3040 };
3041
3042 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3043 {
3044         struct sock *sk = skb_to_full_sk(skb);
3045
3046         if (!sk || !sk_fullsock(sk))
3047                 return 0;
3048
3049         return sock_cgroup_classid(&sk->sk_cgrp_data);
3050 }
3051
3052 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3053         .func           = bpf_skb_cgroup_classid,
3054         .gpl_only       = false,
3055         .ret_type       = RET_INTEGER,
3056         .arg1_type      = ARG_PTR_TO_CTX,
3057 };
3058 #endif
3059
3060 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3061 {
3062         return task_get_classid(skb);
3063 }
3064
3065 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3066         .func           = bpf_get_cgroup_classid,
3067         .gpl_only       = false,
3068         .ret_type       = RET_INTEGER,
3069         .arg1_type      = ARG_PTR_TO_CTX,
3070 };
3071
3072 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3073 {
3074         return dst_tclassid(skb);
3075 }
3076
3077 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3078         .func           = bpf_get_route_realm,
3079         .gpl_only       = false,
3080         .ret_type       = RET_INTEGER,
3081         .arg1_type      = ARG_PTR_TO_CTX,
3082 };
3083
3084 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3085 {
3086         /* If skb_clear_hash() was called due to mangling, we can
3087          * trigger SW recalculation here. Later access to hash
3088          * can then use the inline skb->hash via context directly
3089          * instead of calling this helper again.
3090          */
3091         return skb_get_hash(skb);
3092 }
3093
3094 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3095         .func           = bpf_get_hash_recalc,
3096         .gpl_only       = false,
3097         .ret_type       = RET_INTEGER,
3098         .arg1_type      = ARG_PTR_TO_CTX,
3099 };
3100
3101 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3102 {
3103         /* After all direct packet write, this can be used once for
3104          * triggering a lazy recalc on next skb_get_hash() invocation.
3105          */
3106         skb_clear_hash(skb);
3107         return 0;
3108 }
3109
3110 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3111         .func           = bpf_set_hash_invalid,
3112         .gpl_only       = false,
3113         .ret_type       = RET_INTEGER,
3114         .arg1_type      = ARG_PTR_TO_CTX,
3115 };
3116
3117 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3118 {
3119         /* Set user specified hash as L4(+), so that it gets returned
3120          * on skb_get_hash() call unless BPF prog later on triggers a
3121          * skb_clear_hash().
3122          */
3123         __skb_set_sw_hash(skb, hash, true);
3124         return 0;
3125 }
3126
3127 static const struct bpf_func_proto bpf_set_hash_proto = {
3128         .func           = bpf_set_hash,
3129         .gpl_only       = false,
3130         .ret_type       = RET_INTEGER,
3131         .arg1_type      = ARG_PTR_TO_CTX,
3132         .arg2_type      = ARG_ANYTHING,
3133 };
3134
3135 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3136            u16, vlan_tci)
3137 {
3138         int ret;
3139
3140         if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3141                      vlan_proto != htons(ETH_P_8021AD)))
3142                 vlan_proto = htons(ETH_P_8021Q);
3143
3144         bpf_push_mac_rcsum(skb);
3145         ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3146         bpf_pull_mac_rcsum(skb);
3147
3148         bpf_compute_data_pointers(skb);
3149         return ret;
3150 }
3151
3152 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3153         .func           = bpf_skb_vlan_push,
3154         .gpl_only       = false,
3155         .ret_type       = RET_INTEGER,
3156         .arg1_type      = ARG_PTR_TO_CTX,
3157         .arg2_type      = ARG_ANYTHING,
3158         .arg3_type      = ARG_ANYTHING,
3159 };
3160
3161 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3162 {
3163         int ret;
3164
3165         bpf_push_mac_rcsum(skb);
3166         ret = skb_vlan_pop(skb);
3167         bpf_pull_mac_rcsum(skb);
3168
3169         bpf_compute_data_pointers(skb);
3170         return ret;
3171 }
3172
3173 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3174         .func           = bpf_skb_vlan_pop,
3175         .gpl_only       = false,
3176         .ret_type       = RET_INTEGER,
3177         .arg1_type      = ARG_PTR_TO_CTX,
3178 };
3179
3180 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3181 {
3182         /* Caller already did skb_cow() with len as headroom,
3183          * so no need to do it here.
3184          */
3185         skb_push(skb, len);
3186         memmove(skb->data, skb->data + len, off);
3187         memset(skb->data + off, 0, len);
3188
3189         /* No skb_postpush_rcsum(skb, skb->data + off, len)
3190          * needed here as it does not change the skb->csum
3191          * result for checksum complete when summing over
3192          * zeroed blocks.
3193          */
3194         return 0;
3195 }
3196
3197 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3198 {
3199         void *old_data;
3200
3201         /* skb_ensure_writable() is not needed here, as we're
3202          * already working on an uncloned skb.
3203          */
3204         if (unlikely(!pskb_may_pull(skb, off + len)))
3205                 return -ENOMEM;
3206
3207         old_data = skb->data;
3208         __skb_pull(skb, len);
3209         skb_postpull_rcsum(skb, old_data + off, len);
3210         memmove(skb->data, old_data, off);
3211
3212         return 0;
3213 }
3214
3215 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3216 {
3217         bool trans_same = skb->transport_header == skb->network_header;
3218         int ret;
3219
3220         /* There's no need for __skb_push()/__skb_pull() pair to
3221          * get to the start of the mac header as we're guaranteed
3222          * to always start from here under eBPF.
3223          */
3224         ret = bpf_skb_generic_push(skb, off, len);
3225         if (likely(!ret)) {
3226                 skb->mac_header -= len;
3227                 skb->network_header -= len;
3228                 if (trans_same)
3229                         skb->transport_header = skb->network_header;
3230         }
3231
3232         return ret;
3233 }
3234
3235 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3236 {
3237         bool trans_same = skb->transport_header == skb->network_header;
3238         int ret;
3239
3240         /* Same here, __skb_push()/__skb_pull() pair not needed. */
3241         ret = bpf_skb_generic_pop(skb, off, len);
3242         if (likely(!ret)) {
3243                 skb->mac_header += len;
3244                 skb->network_header += len;
3245                 if (trans_same)
3246                         skb->transport_header = skb->network_header;
3247         }
3248
3249         return ret;
3250 }
3251
3252 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3253 {
3254         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3255         u32 off = skb_mac_header_len(skb);
3256         int ret;
3257
3258         ret = skb_cow(skb, len_diff);
3259         if (unlikely(ret < 0))
3260                 return ret;
3261
3262         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3263         if (unlikely(ret < 0))
3264                 return ret;
3265
3266         if (skb_is_gso(skb)) {
3267                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3268
3269                 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */
3270                 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3271                         shinfo->gso_type &= ~SKB_GSO_TCPV4;
3272                         shinfo->gso_type |=  SKB_GSO_TCPV6;
3273                 }
3274         }
3275
3276         skb->protocol = htons(ETH_P_IPV6);
3277         skb_clear_hash(skb);
3278
3279         return 0;
3280 }
3281
3282 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3283 {
3284         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3285         u32 off = skb_mac_header_len(skb);
3286         int ret;
3287
3288         ret = skb_unclone(skb, GFP_ATOMIC);
3289         if (unlikely(ret < 0))
3290                 return ret;
3291
3292         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3293         if (unlikely(ret < 0))
3294                 return ret;
3295
3296         if (skb_is_gso(skb)) {
3297                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3298
3299                 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */
3300                 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3301                         shinfo->gso_type &= ~SKB_GSO_TCPV6;
3302                         shinfo->gso_type |=  SKB_GSO_TCPV4;
3303                 }
3304         }
3305
3306         skb->protocol = htons(ETH_P_IP);
3307         skb_clear_hash(skb);
3308
3309         return 0;
3310 }
3311
3312 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3313 {
3314         __be16 from_proto = skb->protocol;
3315
3316         if (from_proto == htons(ETH_P_IP) &&
3317               to_proto == htons(ETH_P_IPV6))
3318                 return bpf_skb_proto_4_to_6(skb);
3319
3320         if (from_proto == htons(ETH_P_IPV6) &&
3321               to_proto == htons(ETH_P_IP))
3322                 return bpf_skb_proto_6_to_4(skb);
3323
3324         return -ENOTSUPP;
3325 }
3326
3327 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3328            u64, flags)
3329 {
3330         int ret;
3331
3332         if (unlikely(flags))
3333                 return -EINVAL;
3334
3335         /* General idea is that this helper does the basic groundwork
3336          * needed for changing the protocol, and eBPF program fills the
3337          * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3338          * and other helpers, rather than passing a raw buffer here.
3339          *
3340          * The rationale is to keep this minimal and without a need to
3341          * deal with raw packet data. F.e. even if we would pass buffers
3342          * here, the program still needs to call the bpf_lX_csum_replace()
3343          * helpers anyway. Plus, this way we keep also separation of
3344          * concerns, since f.e. bpf_skb_store_bytes() should only take
3345          * care of stores.
3346          *
3347          * Currently, additional options and extension header space are
3348          * not supported, but flags register is reserved so we can adapt
3349          * that. For offloads, we mark packet as dodgy, so that headers
3350          * need to be verified first.
3351          */
3352         ret = bpf_skb_proto_xlat(skb, proto);
3353         bpf_compute_data_pointers(skb);
3354         return ret;
3355 }
3356
3357 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3358         .func           = bpf_skb_change_proto,
3359         .gpl_only       = false,
3360         .ret_type       = RET_INTEGER,
3361         .arg1_type      = ARG_PTR_TO_CTX,
3362         .arg2_type      = ARG_ANYTHING,
3363         .arg3_type      = ARG_ANYTHING,
3364 };
3365
3366 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3367 {
3368         /* We only allow a restricted subset to be changed for now. */
3369         if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3370                      !skb_pkt_type_ok(pkt_type)))
3371                 return -EINVAL;
3372
3373         skb->pkt_type = pkt_type;
3374         return 0;
3375 }
3376
3377 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3378         .func           = bpf_skb_change_type,
3379         .gpl_only       = false,
3380         .ret_type       = RET_INTEGER,
3381         .arg1_type      = ARG_PTR_TO_CTX,
3382         .arg2_type      = ARG_ANYTHING,
3383 };
3384
3385 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3386 {
3387         switch (skb->protocol) {
3388         case htons(ETH_P_IP):
3389                 return sizeof(struct iphdr);
3390         case htons(ETH_P_IPV6):
3391                 return sizeof(struct ipv6hdr);
3392         default:
3393                 return ~0U;
3394         }
3395 }
3396
3397 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK    (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3398                                          BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3399
3400 #define BPF_F_ADJ_ROOM_DECAP_L3_MASK    (BPF_F_ADJ_ROOM_DECAP_L3_IPV4 | \
3401                                          BPF_F_ADJ_ROOM_DECAP_L3_IPV6)
3402
3403 #define BPF_F_ADJ_ROOM_MASK             (BPF_F_ADJ_ROOM_FIXED_GSO | \
3404                                          BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3405                                          BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3406                                          BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3407                                          BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \
3408                                          BPF_F_ADJ_ROOM_ENCAP_L2( \
3409                                           BPF_ADJ_ROOM_ENCAP_L2_MASK) | \
3410                                          BPF_F_ADJ_ROOM_DECAP_L3_MASK)
3411
3412 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3413                             u64 flags)
3414 {
3415         u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3416         bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3417         u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3418         unsigned int gso_type = SKB_GSO_DODGY;
3419         int ret;
3420
3421         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3422                 /* udp gso_size delineates datagrams, only allow if fixed */
3423                 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3424                     !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3425                         return -ENOTSUPP;
3426         }
3427
3428         ret = skb_cow_head(skb, len_diff);
3429         if (unlikely(ret < 0))
3430                 return ret;
3431
3432         if (encap) {
3433                 if (skb->protocol != htons(ETH_P_IP) &&
3434                     skb->protocol != htons(ETH_P_IPV6))
3435                         return -ENOTSUPP;
3436
3437                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3438                     flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3439                         return -EINVAL;
3440
3441                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3442                     flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3443                         return -EINVAL;
3444
3445                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH &&
3446                     inner_mac_len < ETH_HLEN)
3447                         return -EINVAL;
3448
3449                 if (skb->encapsulation)
3450                         return -EALREADY;
3451
3452                 mac_len = skb->network_header - skb->mac_header;
3453                 inner_net = skb->network_header;
3454                 if (inner_mac_len > len_diff)
3455                         return -EINVAL;
3456                 inner_trans = skb->transport_header;
3457         }
3458
3459         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3460         if (unlikely(ret < 0))
3461                 return ret;
3462
3463         if (encap) {
3464                 skb->inner_mac_header = inner_net - inner_mac_len;
3465                 skb->inner_network_header = inner_net;
3466                 skb->inner_transport_header = inner_trans;
3467
3468                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH)
3469                         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
3470                 else
3471                         skb_set_inner_protocol(skb, skb->protocol);
3472
3473                 skb->encapsulation = 1;
3474                 skb_set_network_header(skb, mac_len);
3475
3476                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3477                         gso_type |= SKB_GSO_UDP_TUNNEL;
3478                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3479                         gso_type |= SKB_GSO_GRE;
3480                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3481                         gso_type |= SKB_GSO_IPXIP6;
3482                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3483                         gso_type |= SKB_GSO_IPXIP4;
3484
3485                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3486                     flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3487                         int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3488                                         sizeof(struct ipv6hdr) :
3489                                         sizeof(struct iphdr);
3490
3491                         skb_set_transport_header(skb, mac_len + nh_len);
3492                 }
3493
3494                 /* Match skb->protocol to new outer l3 protocol */
3495                 if (skb->protocol == htons(ETH_P_IP) &&
3496                     flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3497                         skb->protocol = htons(ETH_P_IPV6);
3498                 else if (skb->protocol == htons(ETH_P_IPV6) &&
3499                          flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3500                         skb->protocol = htons(ETH_P_IP);
3501         }
3502
3503         if (skb_is_gso(skb)) {
3504                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3505
3506                 /* Due to header grow, MSS needs to be downgraded. */
3507                 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3508                         skb_decrease_gso_size(shinfo, len_diff);
3509
3510                 /* Header must be checked, and gso_segs recomputed. */
3511                 shinfo->gso_type |= gso_type;
3512                 shinfo->gso_segs = 0;
3513         }
3514
3515         return 0;
3516 }
3517
3518 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3519                               u64 flags)
3520 {
3521         int ret;
3522
3523         if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3524                                BPF_F_ADJ_ROOM_DECAP_L3_MASK |
3525                                BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3526                 return -EINVAL;
3527
3528         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3529                 /* udp gso_size delineates datagrams, only allow if fixed */
3530                 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3531                     !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3532                         return -ENOTSUPP;
3533         }
3534
3535         ret = skb_unclone(skb, GFP_ATOMIC);
3536         if (unlikely(ret < 0))
3537                 return ret;
3538
3539         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3540         if (unlikely(ret < 0))
3541                 return ret;
3542
3543         /* Match skb->protocol to new outer l3 protocol */
3544         if (skb->protocol == htons(ETH_P_IP) &&
3545             flags & BPF_F_ADJ_ROOM_DECAP_L3_IPV6)
3546                 skb->protocol = htons(ETH_P_IPV6);
3547         else if (skb->protocol == htons(ETH_P_IPV6) &&
3548                  flags & BPF_F_ADJ_ROOM_DECAP_L3_IPV4)
3549                 skb->protocol = htons(ETH_P_IP);
3550
3551         if (skb_is_gso(skb)) {
3552                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3553
3554                 /* Due to header shrink, MSS can be upgraded. */
3555                 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3556                         skb_increase_gso_size(shinfo, len_diff);
3557
3558                 /* Header must be checked, and gso_segs recomputed. */
3559                 shinfo->gso_type |= SKB_GSO_DODGY;
3560                 shinfo->gso_segs = 0;
3561         }
3562
3563         return 0;
3564 }
3565
3566 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3567
3568 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3569            u32, mode, u64, flags)
3570 {
3571         u32 len_diff_abs = abs(len_diff);
3572         bool shrink = len_diff < 0;
3573         int ret = 0;
3574
3575         if (unlikely(flags || mode))
3576                 return -EINVAL;
3577         if (unlikely(len_diff_abs > 0xfffU))
3578                 return -EFAULT;
3579
3580         if (!shrink) {
3581                 ret = skb_cow(skb, len_diff);
3582                 if (unlikely(ret < 0))
3583                         return ret;
3584                 __skb_push(skb, len_diff_abs);
3585                 memset(skb->data, 0, len_diff_abs);
3586         } else {
3587                 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3588                         return -ENOMEM;
3589                 __skb_pull(skb, len_diff_abs);
3590         }
3591         if (tls_sw_has_ctx_rx(skb->sk)) {
3592                 struct strp_msg *rxm = strp_msg(skb);
3593
3594                 rxm->full_len += len_diff;
3595         }
3596         return ret;
3597 }
3598
3599 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3600         .func           = sk_skb_adjust_room,
3601         .gpl_only       = false,
3602         .ret_type       = RET_INTEGER,
3603         .arg1_type      = ARG_PTR_TO_CTX,
3604         .arg2_type      = ARG_ANYTHING,
3605         .arg3_type      = ARG_ANYTHING,
3606         .arg4_type      = ARG_ANYTHING,
3607 };
3608
3609 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3610            u32, mode, u64, flags)
3611 {
3612         u32 len_cur, len_diff_abs = abs(len_diff);
3613         u32 len_min = bpf_skb_net_base_len(skb);
3614         u32 len_max = BPF_SKB_MAX_LEN;
3615         __be16 proto = skb->protocol;
3616         bool shrink = len_diff < 0;
3617         u32 off;
3618         int ret;
3619
3620         if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3621                                BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3622                 return -EINVAL;
3623         if (unlikely(len_diff_abs > 0xfffU))
3624                 return -EFAULT;
3625         if (unlikely(proto != htons(ETH_P_IP) &&
3626                      proto != htons(ETH_P_IPV6)))
3627                 return -ENOTSUPP;
3628
3629         off = skb_mac_header_len(skb);
3630         switch (mode) {
3631         case BPF_ADJ_ROOM_NET:
3632                 off += bpf_skb_net_base_len(skb);
3633                 break;
3634         case BPF_ADJ_ROOM_MAC:
3635                 break;
3636         default:
3637                 return -ENOTSUPP;
3638         }
3639
3640         if (flags & BPF_F_ADJ_ROOM_DECAP_L3_MASK) {
3641                 if (!shrink)
3642                         return -EINVAL;
3643
3644                 switch (flags & BPF_F_ADJ_ROOM_DECAP_L3_MASK) {
3645                 case BPF_F_ADJ_ROOM_DECAP_L3_IPV4:
3646                         len_min = sizeof(struct iphdr);
3647                         break;
3648                 case BPF_F_ADJ_ROOM_DECAP_L3_IPV6:
3649                         len_min = sizeof(struct ipv6hdr);
3650                         break;
3651                 default:
3652                         return -EINVAL;
3653                 }
3654         }
3655
3656         len_cur = skb->len - skb_network_offset(skb);
3657         if ((shrink && (len_diff_abs >= len_cur ||
3658                         len_cur - len_diff_abs < len_min)) ||
3659             (!shrink && (skb->len + len_diff_abs > len_max &&
3660                          !skb_is_gso(skb))))
3661                 return -ENOTSUPP;
3662
3663         ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3664                        bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3665         if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3666                 __skb_reset_checksum_unnecessary(skb);
3667
3668         bpf_compute_data_pointers(skb);
3669         return ret;
3670 }
3671
3672 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3673         .func           = bpf_skb_adjust_room,
3674         .gpl_only       = false,
3675         .ret_type       = RET_INTEGER,
3676         .arg1_type      = ARG_PTR_TO_CTX,
3677         .arg2_type      = ARG_ANYTHING,
3678         .arg3_type      = ARG_ANYTHING,
3679         .arg4_type      = ARG_ANYTHING,
3680 };
3681
3682 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3683 {
3684         u32 min_len = skb_network_offset(skb);
3685
3686         if (skb_transport_header_was_set(skb))
3687                 min_len = skb_transport_offset(skb);
3688         if (skb->ip_summed == CHECKSUM_PARTIAL)
3689                 min_len = skb_checksum_start_offset(skb) +
3690                           skb->csum_offset + sizeof(__sum16);
3691         return min_len;
3692 }
3693
3694 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3695 {
3696         unsigned int old_len = skb->len;
3697         int ret;
3698
3699         ret = __skb_grow_rcsum(skb, new_len);
3700         if (!ret)
3701                 memset(skb->data + old_len, 0, new_len - old_len);
3702         return ret;
3703 }
3704
3705 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3706 {
3707         return __skb_trim_rcsum(skb, new_len);
3708 }
3709
3710 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3711                                         u64 flags)
3712 {
3713         u32 max_len = BPF_SKB_MAX_LEN;
3714         u32 min_len = __bpf_skb_min_len(skb);
3715         int ret;
3716
3717         if (unlikely(flags || new_len > max_len || new_len < min_len))
3718                 return -EINVAL;
3719         if (skb->encapsulation)
3720                 return -ENOTSUPP;
3721
3722         /* The basic idea of this helper is that it's performing the
3723          * needed work to either grow or trim an skb, and eBPF program
3724          * rewrites the rest via helpers like bpf_skb_store_bytes(),
3725          * bpf_lX_csum_replace() and others rather than passing a raw
3726          * buffer here. This one is a slow path helper and intended
3727          * for replies with control messages.
3728          *
3729          * Like in bpf_skb_change_proto(), we want to keep this rather
3730          * minimal and without protocol specifics so that we are able
3731          * to separate concerns as in bpf_skb_store_bytes() should only
3732          * be the one responsible for writing buffers.
3733          *
3734          * It's really expected to be a slow path operation here for
3735          * control message replies, so we're implicitly linearizing,
3736          * uncloning and drop offloads from the skb by this.
3737          */
3738         ret = __bpf_try_make_writable(skb, skb->len);
3739         if (!ret) {
3740                 if (new_len > skb->len)
3741                         ret = bpf_skb_grow_rcsum(skb, new_len);
3742                 else if (new_len < skb->len)
3743                         ret = bpf_skb_trim_rcsum(skb, new_len);
3744                 if (!ret && skb_is_gso(skb))
3745                         skb_gso_reset(skb);
3746         }
3747         return ret;
3748 }
3749
3750 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3751            u64, flags)
3752 {
3753         int ret = __bpf_skb_change_tail(skb, new_len, flags);
3754
3755         bpf_compute_data_pointers(skb);
3756         return ret;
3757 }
3758
3759 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3760         .func           = bpf_skb_change_tail,
3761         .gpl_only       = false,
3762         .ret_type       = RET_INTEGER,
3763         .arg1_type      = ARG_PTR_TO_CTX,
3764         .arg2_type      = ARG_ANYTHING,
3765         .arg3_type      = ARG_ANYTHING,
3766 };
3767
3768 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3769            u64, flags)
3770 {
3771         return __bpf_skb_change_tail(skb, new_len, flags);
3772 }
3773
3774 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3775         .func           = sk_skb_change_tail,
3776         .gpl_only       = false,
3777         .ret_type       = RET_INTEGER,
3778         .arg1_type      = ARG_PTR_TO_CTX,
3779         .arg2_type      = ARG_ANYTHING,
3780         .arg3_type      = ARG_ANYTHING,
3781 };
3782
3783 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3784                                         u64 flags)
3785 {
3786         u32 max_len = BPF_SKB_MAX_LEN;
3787         u32 new_len = skb->len + head_room;
3788         int ret;
3789
3790         if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3791                      new_len < skb->len))
3792                 return -EINVAL;
3793
3794         ret = skb_cow(skb, head_room);
3795         if (likely(!ret)) {
3796                 /* Idea for this helper is that we currently only
3797                  * allow to expand on mac header. This means that
3798                  * skb->protocol network header, etc, stay as is.
3799                  * Compared to bpf_skb_change_tail(), we're more
3800                  * flexible due to not needing to linearize or
3801                  * reset GSO. Intention for this helper is to be
3802                  * used by an L3 skb that needs to push mac header
3803                  * for redirection into L2 device.
3804                  */
3805                 __skb_push(skb, head_room);
3806                 memset(skb->data, 0, head_room);
3807                 skb_reset_mac_header(skb);
3808                 skb_reset_mac_len(skb);
3809         }
3810
3811         return ret;
3812 }
3813
3814 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3815            u64, flags)
3816 {
3817         int ret = __bpf_skb_change_head(skb, head_room, flags);
3818
3819         bpf_compute_data_pointers(skb);
3820         return ret;
3821 }
3822
3823 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3824         .func           = bpf_skb_change_head,
3825         .gpl_only       = false,
3826         .ret_type       = RET_INTEGER,
3827         .arg1_type      = ARG_PTR_TO_CTX,
3828         .arg2_type      = ARG_ANYTHING,
3829         .arg3_type      = ARG_ANYTHING,
3830 };
3831
3832 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3833            u64, flags)
3834 {
3835         return __bpf_skb_change_head(skb, head_room, flags);
3836 }
3837
3838 static const struct bpf_func_proto sk_skb_change_head_proto = {
3839         .func           = sk_skb_change_head,
3840         .gpl_only       = false,
3841         .ret_type       = RET_INTEGER,
3842         .arg1_type      = ARG_PTR_TO_CTX,
3843         .arg2_type      = ARG_ANYTHING,
3844         .arg3_type      = ARG_ANYTHING,
3845 };
3846
3847 BPF_CALL_1(bpf_xdp_get_buff_len, struct xdp_buff*, xdp)
3848 {
3849         return xdp_get_buff_len(xdp);
3850 }
3851
3852 static const struct bpf_func_proto bpf_xdp_get_buff_len_proto = {
3853         .func           = bpf_xdp_get_buff_len,
3854         .gpl_only       = false,
3855         .ret_type       = RET_INTEGER,
3856         .arg1_type      = ARG_PTR_TO_CTX,
3857 };
3858
3859 BTF_ID_LIST_SINGLE(bpf_xdp_get_buff_len_bpf_ids, struct, xdp_buff)
3860
3861 const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto = {
3862         .func           = bpf_xdp_get_buff_len,
3863         .gpl_only       = false,
3864         .arg1_type      = ARG_PTR_TO_BTF_ID,
3865         .arg1_btf_id    = &bpf_xdp_get_buff_len_bpf_ids[0],
3866 };
3867
3868 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3869 {
3870         return xdp_data_meta_unsupported(xdp) ? 0 :
3871                xdp->data - xdp->data_meta;
3872 }
3873
3874 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3875 {
3876         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3877         unsigned long metalen = xdp_get_metalen(xdp);
3878         void *data_start = xdp_frame_end + metalen;
3879         void *data = xdp->data + offset;
3880
3881         if (unlikely(data < data_start ||
3882                      data > xdp->data_end - ETH_HLEN))
3883                 return -EINVAL;
3884
3885         if (metalen)
3886                 memmove(xdp->data_meta + offset,
3887                         xdp->data_meta, metalen);
3888         xdp->data_meta += offset;
3889         xdp->data = data;
3890
3891         return 0;
3892 }
3893
3894 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3895         .func           = bpf_xdp_adjust_head,
3896         .gpl_only       = false,
3897         .ret_type       = RET_INTEGER,
3898         .arg1_type      = ARG_PTR_TO_CTX,
3899         .arg2_type      = ARG_ANYTHING,
3900 };
3901
3902 void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off,
3903                       void *buf, unsigned long len, bool flush)
3904 {
3905         unsigned long ptr_len, ptr_off = 0;
3906         skb_frag_t *next_frag, *end_frag;
3907         struct skb_shared_info *sinfo;
3908         void *src, *dst;
3909         u8 *ptr_buf;
3910
3911         if (likely(xdp->data_end - xdp->data >= off + len)) {
3912                 src = flush ? buf : xdp->data + off;
3913                 dst = flush ? xdp->data + off : buf;
3914                 memcpy(dst, src, len);
3915                 return;
3916         }
3917
3918         sinfo = xdp_get_shared_info_from_buff(xdp);
3919         end_frag = &sinfo->frags[sinfo->nr_frags];
3920         next_frag = &sinfo->frags[0];
3921
3922         ptr_len = xdp->data_end - xdp->data;
3923         ptr_buf = xdp->data;
3924
3925         while (true) {
3926                 if (off < ptr_off + ptr_len) {
3927                         unsigned long copy_off = off - ptr_off;
3928                         unsigned long copy_len = min(len, ptr_len - copy_off);
3929
3930                         src = flush ? buf : ptr_buf + copy_off;
3931                         dst = flush ? ptr_buf + copy_off : buf;
3932                         memcpy(dst, src, copy_len);
3933
3934                         off += copy_len;
3935                         len -= copy_len;
3936                         buf += copy_len;
3937                 }
3938
3939                 if (!len || next_frag == end_frag)
3940                         break;
3941
3942                 ptr_off += ptr_len;
3943                 ptr_buf = skb_frag_address(next_frag);
3944                 ptr_len = skb_frag_size(next_frag);
3945                 next_frag++;
3946         }
3947 }
3948
3949 void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len)
3950 {
3951         u32 size = xdp->data_end - xdp->data;
3952         struct skb_shared_info *sinfo;
3953         void *addr = xdp->data;
3954         int i;
3955
3956         if (unlikely(offset > 0xffff || len > 0xffff))
3957                 return ERR_PTR(-EFAULT);
3958
3959         if (unlikely(offset + len > xdp_get_buff_len(xdp)))
3960                 return ERR_PTR(-EINVAL);
3961
3962         if (likely(offset < size)) /* linear area */
3963                 goto out;
3964
3965         sinfo = xdp_get_shared_info_from_buff(xdp);
3966         offset -= size;
3967         for (i = 0; i < sinfo->nr_frags; i++) { /* paged area */
3968                 u32 frag_size = skb_frag_size(&sinfo->frags[i]);
3969
3970                 if  (offset < frag_size) {
3971                         addr = skb_frag_address(&sinfo->frags[i]);
3972                         size = frag_size;
3973                         break;
3974                 }
3975                 offset -= frag_size;
3976         }
3977 out:
3978         return offset + len <= size ? addr + offset : NULL;
3979 }
3980
3981 BPF_CALL_4(bpf_xdp_load_bytes, struct xdp_buff *, xdp, u32, offset,
3982            void *, buf, u32, len)
3983 {
3984         void *ptr;
3985
3986         ptr = bpf_xdp_pointer(xdp, offset, len);
3987         if (IS_ERR(ptr))
3988                 return PTR_ERR(ptr);
3989
3990         if (!ptr)
3991                 bpf_xdp_copy_buf(xdp, offset, buf, len, false);
3992         else
3993                 memcpy(buf, ptr, len);
3994
3995         return 0;
3996 }
3997
3998 static const struct bpf_func_proto bpf_xdp_load_bytes_proto = {
3999         .func           = bpf_xdp_load_bytes,
4000         .gpl_only       = false,
4001         .ret_type       = RET_INTEGER,
4002         .arg1_type      = ARG_PTR_TO_CTX,
4003         .arg2_type      = ARG_ANYTHING,
4004         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
4005         .arg4_type      = ARG_CONST_SIZE,
4006 };
4007
4008 int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len)
4009 {
4010         return ____bpf_xdp_load_bytes(xdp, offset, buf, len);
4011 }
4012
4013 BPF_CALL_4(bpf_xdp_store_bytes, struct xdp_buff *, xdp, u32, offset,
4014            void *, buf, u32, len)
4015 {
4016         void *ptr;
4017
4018         ptr = bpf_xdp_pointer(xdp, offset, len);
4019         if (IS_ERR(ptr))
4020                 return PTR_ERR(ptr);
4021
4022         if (!ptr)
4023                 bpf_xdp_copy_buf(xdp, offset, buf, len, true);
4024         else
4025                 memcpy(ptr, buf, len);
4026
4027         return 0;
4028 }
4029
4030 static const struct bpf_func_proto bpf_xdp_store_bytes_proto = {
4031         .func           = bpf_xdp_store_bytes,
4032         .gpl_only       = false,
4033         .ret_type       = RET_INTEGER,
4034         .arg1_type      = ARG_PTR_TO_CTX,
4035         .arg2_type      = ARG_ANYTHING,
4036         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
4037         .arg4_type      = ARG_CONST_SIZE,
4038 };
4039
4040 int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len)
4041 {
4042         return ____bpf_xdp_store_bytes(xdp, offset, buf, len);
4043 }
4044
4045 static int bpf_xdp_frags_increase_tail(struct xdp_buff *xdp, int offset)
4046 {
4047         struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
4048         skb_frag_t *frag = &sinfo->frags[sinfo->nr_frags - 1];
4049         struct xdp_rxq_info *rxq = xdp->rxq;
4050         unsigned int tailroom;
4051
4052         if (!rxq->frag_size || rxq->frag_size > xdp->frame_sz)
4053                 return -EOPNOTSUPP;
4054
4055         tailroom = rxq->frag_size - skb_frag_size(frag) - skb_frag_off(frag);
4056         if (unlikely(offset > tailroom))
4057                 return -EINVAL;
4058
4059         memset(skb_frag_address(frag) + skb_frag_size(frag), 0, offset);
4060         skb_frag_size_add(frag, offset);
4061         sinfo->xdp_frags_size += offset;
4062
4063         return 0;
4064 }
4065
4066 static int bpf_xdp_frags_shrink_tail(struct xdp_buff *xdp, int offset)
4067 {
4068         struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
4069         int i, n_frags_free = 0, len_free = 0;
4070
4071         if (unlikely(offset > (int)xdp_get_buff_len(xdp) - ETH_HLEN))
4072                 return -EINVAL;
4073
4074         for (i = sinfo->nr_frags - 1; i >= 0 && offset > 0; i--) {
4075                 skb_frag_t *frag = &sinfo->frags[i];
4076                 int shrink = min_t(int, offset, skb_frag_size(frag));
4077
4078                 len_free += shrink;
4079                 offset -= shrink;
4080
4081                 if (skb_frag_size(frag) == shrink) {
4082                         struct page *page = skb_frag_page(frag);
4083
4084                         __xdp_return(page_address(page), &xdp->rxq->mem,
4085                                      false, NULL);
4086                         n_frags_free++;
4087                 } else {
4088                         skb_frag_size_sub(frag, shrink);
4089                         break;
4090                 }
4091         }
4092         sinfo->nr_frags -= n_frags_free;
4093         sinfo->xdp_frags_size -= len_free;
4094
4095         if (unlikely(!sinfo->nr_frags)) {
4096                 xdp_buff_clear_frags_flag(xdp);
4097                 xdp->data_end -= offset;
4098         }
4099
4100         return 0;
4101 }
4102
4103 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
4104 {
4105         void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
4106         void *data_end = xdp->data_end + offset;
4107
4108         if (unlikely(xdp_buff_has_frags(xdp))) { /* non-linear xdp buff */
4109                 if (offset < 0)
4110                         return bpf_xdp_frags_shrink_tail(xdp, -offset);
4111
4112                 return bpf_xdp_frags_increase_tail(xdp, offset);
4113         }
4114
4115         /* Notice that xdp_data_hard_end have reserved some tailroom */
4116         if (unlikely(data_end > data_hard_end))
4117                 return -EINVAL;
4118
4119         if (unlikely(data_end < xdp->data + ETH_HLEN))
4120                 return -EINVAL;
4121
4122         /* Clear memory area on grow, can contain uninit kernel memory */
4123         if (offset > 0)
4124                 memset(xdp->data_end, 0, offset);
4125
4126         xdp->data_end = data_end;
4127
4128         return 0;
4129 }
4130
4131 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
4132         .func           = bpf_xdp_adjust_tail,
4133         .gpl_only       = false,
4134         .ret_type       = RET_INTEGER,
4135         .arg1_type      = ARG_PTR_TO_CTX,
4136         .arg2_type      = ARG_ANYTHING,
4137 };
4138
4139 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
4140 {
4141         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
4142         void *meta = xdp->data_meta + offset;
4143         unsigned long metalen = xdp->data - meta;
4144
4145         if (xdp_data_meta_unsupported(xdp))
4146                 return -ENOTSUPP;
4147         if (unlikely(meta < xdp_frame_end ||
4148                      meta > xdp->data))
4149                 return -EINVAL;
4150         if (unlikely(xdp_metalen_invalid(metalen)))
4151                 return -EACCES;
4152
4153         xdp->data_meta = meta;
4154
4155         return 0;
4156 }
4157
4158 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
4159         .func           = bpf_xdp_adjust_meta,
4160         .gpl_only       = false,
4161         .ret_type       = RET_INTEGER,
4162         .arg1_type      = ARG_PTR_TO_CTX,
4163         .arg2_type      = ARG_ANYTHING,
4164 };
4165
4166 /**
4167  * DOC: xdp redirect
4168  *
4169  * XDP_REDIRECT works by a three-step process, implemented in the functions
4170  * below:
4171  *
4172  * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target
4173  *    of the redirect and store it (along with some other metadata) in a per-CPU
4174  *    struct bpf_redirect_info.
4175  *
4176  * 2. When the program returns the XDP_REDIRECT return code, the driver will
4177  *    call xdp_do_redirect() which will use the information in struct
4178  *    bpf_redirect_info to actually enqueue the frame into a map type-specific
4179  *    bulk queue structure.
4180  *
4181  * 3. Before exiting its NAPI poll loop, the driver will call
4182  *    xdp_do_flush(), which will flush all the different bulk queues,
4183  *    thus completing the redirect. Note that xdp_do_flush() must be
4184  *    called before napi_complete_done() in the driver, as the
4185  *    XDP_REDIRECT logic relies on being inside a single NAPI instance
4186  *    through to the xdp_do_flush() call for RCU protection of all
4187  *    in-kernel data structures.
4188  */
4189 /*
4190  * Pointers to the map entries will be kept around for this whole sequence of
4191  * steps, protected by RCU. However, there is no top-level rcu_read_lock() in
4192  * the core code; instead, the RCU protection relies on everything happening
4193  * inside a single NAPI poll sequence, which means it's between a pair of calls
4194  * to local_bh_disable()/local_bh_enable().
4195  *
4196  * The map entries are marked as __rcu and the map code makes sure to
4197  * dereference those pointers with rcu_dereference_check() in a way that works
4198  * for both sections that to hold an rcu_read_lock() and sections that are
4199  * called from NAPI without a separate rcu_read_lock(). The code below does not
4200  * use RCU annotations, but relies on those in the map code.
4201  */
4202 void xdp_do_flush(void)
4203 {
4204         __dev_flush();
4205         __cpu_map_flush();
4206         __xsk_map_flush();
4207 }
4208 EXPORT_SYMBOL_GPL(xdp_do_flush);
4209
4210 void bpf_clear_redirect_map(struct bpf_map *map)
4211 {
4212         struct bpf_redirect_info *ri;
4213         int cpu;
4214
4215         for_each_possible_cpu(cpu) {
4216                 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
4217                 /* Avoid polluting remote cacheline due to writes if
4218                  * not needed. Once we pass this test, we need the
4219                  * cmpxchg() to make sure it hasn't been changed in
4220                  * the meantime by remote CPU.
4221                  */
4222                 if (unlikely(READ_ONCE(ri->map) == map))
4223                         cmpxchg(&ri->map, map, NULL);
4224         }
4225 }
4226
4227 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
4228 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key);
4229
4230 u32 xdp_master_redirect(struct xdp_buff *xdp)
4231 {
4232         struct net_device *master, *slave;
4233         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4234
4235         master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev);
4236         slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp);
4237         if (slave && slave != xdp->rxq->dev) {
4238                 /* The target device is different from the receiving device, so
4239                  * redirect it to the new device.
4240                  * Using XDP_REDIRECT gets the correct behaviour from XDP enabled
4241                  * drivers to unmap the packet from their rx ring.
4242                  */
4243                 ri->tgt_index = slave->ifindex;
4244                 ri->map_id = INT_MAX;
4245                 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4246                 return XDP_REDIRECT;
4247         }
4248         return XDP_TX;
4249 }
4250 EXPORT_SYMBOL_GPL(xdp_master_redirect);
4251
4252 static inline int __xdp_do_redirect_xsk(struct bpf_redirect_info *ri,
4253                                         struct net_device *dev,
4254                                         struct xdp_buff *xdp,
4255                                         struct bpf_prog *xdp_prog)
4256 {
4257         enum bpf_map_type map_type = ri->map_type;
4258         void *fwd = ri->tgt_value;
4259         u32 map_id = ri->map_id;
4260         int err;
4261
4262         ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4263         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4264
4265         err = __xsk_map_redirect(fwd, xdp);
4266         if (unlikely(err))
4267                 goto err;
4268
4269         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4270         return 0;
4271 err:
4272         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4273         return err;
4274 }
4275
4276 static __always_inline int __xdp_do_redirect_frame(struct bpf_redirect_info *ri,
4277                                                    struct net_device *dev,
4278                                                    struct xdp_frame *xdpf,
4279                                                    struct bpf_prog *xdp_prog)
4280 {
4281         enum bpf_map_type map_type = ri->map_type;
4282         void *fwd = ri->tgt_value;
4283         u32 map_id = ri->map_id;
4284         struct bpf_map *map;
4285         int err;
4286
4287         ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4288         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4289
4290         if (unlikely(!xdpf)) {
4291                 err = -EOVERFLOW;
4292                 goto err;
4293         }
4294
4295         switch (map_type) {
4296         case BPF_MAP_TYPE_DEVMAP:
4297                 fallthrough;
4298         case BPF_MAP_TYPE_DEVMAP_HASH:
4299                 map = READ_ONCE(ri->map);
4300                 if (unlikely(map)) {
4301                         WRITE_ONCE(ri->map, NULL);
4302                         err = dev_map_enqueue_multi(xdpf, dev, map,
4303                                                     ri->flags & BPF_F_EXCLUDE_INGRESS);
4304                 } else {
4305                         err = dev_map_enqueue(fwd, xdpf, dev);
4306                 }
4307                 break;
4308         case BPF_MAP_TYPE_CPUMAP:
4309                 err = cpu_map_enqueue(fwd, xdpf, dev);
4310                 break;
4311         case BPF_MAP_TYPE_UNSPEC:
4312                 if (map_id == INT_MAX) {
4313                         fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4314                         if (unlikely(!fwd)) {
4315                                 err = -EINVAL;
4316                                 break;
4317                         }
4318                         err = dev_xdp_enqueue(fwd, xdpf, dev);
4319                         break;
4320                 }
4321                 fallthrough;
4322         default:
4323                 err = -EBADRQC;
4324         }
4325
4326         if (unlikely(err))
4327                 goto err;
4328
4329         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4330         return 0;
4331 err:
4332         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4333         return err;
4334 }
4335
4336 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
4337                     struct bpf_prog *xdp_prog)
4338 {
4339         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4340         enum bpf_map_type map_type = ri->map_type;
4341
4342         if (map_type == BPF_MAP_TYPE_XSKMAP) {
4343                 /* XDP_REDIRECT is not supported AF_XDP yet. */
4344                 if (unlikely(xdp_buff_has_frags(xdp)))
4345                         return -EOPNOTSUPP;
4346
4347                 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4348         }
4349
4350         return __xdp_do_redirect_frame(ri, dev, xdp_convert_buff_to_frame(xdp),
4351                                        xdp_prog);
4352 }
4353 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4354
4355 int xdp_do_redirect_frame(struct net_device *dev, struct xdp_buff *xdp,
4356                           struct xdp_frame *xdpf, struct bpf_prog *xdp_prog)
4357 {
4358         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4359         enum bpf_map_type map_type = ri->map_type;
4360
4361         if (map_type == BPF_MAP_TYPE_XSKMAP)
4362                 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4363
4364         return __xdp_do_redirect_frame(ri, dev, xdpf, xdp_prog);
4365 }
4366 EXPORT_SYMBOL_GPL(xdp_do_redirect_frame);
4367
4368 static int xdp_do_generic_redirect_map(struct net_device *dev,
4369                                        struct sk_buff *skb,
4370                                        struct xdp_buff *xdp,
4371                                        struct bpf_prog *xdp_prog,
4372                                        void *fwd,
4373                                        enum bpf_map_type map_type, u32 map_id)
4374 {
4375         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4376         struct bpf_map *map;
4377         int err;
4378
4379         switch (map_type) {
4380         case BPF_MAP_TYPE_DEVMAP:
4381                 fallthrough;
4382         case BPF_MAP_TYPE_DEVMAP_HASH:
4383                 map = READ_ONCE(ri->map);
4384                 if (unlikely(map)) {
4385                         WRITE_ONCE(ri->map, NULL);
4386                         err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4387                                                      ri->flags & BPF_F_EXCLUDE_INGRESS);
4388                 } else {
4389                         err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4390                 }
4391                 if (unlikely(err))
4392                         goto err;
4393                 break;
4394         case BPF_MAP_TYPE_XSKMAP:
4395                 err = xsk_generic_rcv(fwd, xdp);
4396                 if (err)
4397                         goto err;
4398                 consume_skb(skb);
4399                 break;
4400         case BPF_MAP_TYPE_CPUMAP:
4401                 err = cpu_map_generic_redirect(fwd, skb);
4402                 if (unlikely(err))
4403                         goto err;
4404                 break;
4405         default:
4406                 err = -EBADRQC;
4407                 goto err;
4408         }
4409
4410         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4411         return 0;
4412 err:
4413         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4414         return err;
4415 }
4416
4417 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4418                             struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4419 {
4420         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4421         enum bpf_map_type map_type = ri->map_type;
4422         void *fwd = ri->tgt_value;
4423         u32 map_id = ri->map_id;
4424         int err;
4425
4426         ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4427         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4428
4429         if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4430                 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4431                 if (unlikely(!fwd)) {
4432                         err = -EINVAL;
4433                         goto err;
4434                 }
4435
4436                 err = xdp_ok_fwd_dev(fwd, skb->len);
4437                 if (unlikely(err))
4438                         goto err;
4439
4440                 skb->dev = fwd;
4441                 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4442                 generic_xdp_tx(skb, xdp_prog);
4443                 return 0;
4444         }
4445
4446         return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id);
4447 err:
4448         _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4449         return err;
4450 }
4451
4452 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4453 {
4454         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4455
4456         if (unlikely(flags))
4457                 return XDP_ABORTED;
4458
4459         /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4460          * by map_idr) is used for ifindex based XDP redirect.
4461          */
4462         ri->tgt_index = ifindex;
4463         ri->map_id = INT_MAX;
4464         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4465
4466         return XDP_REDIRECT;
4467 }
4468
4469 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4470         .func           = bpf_xdp_redirect,
4471         .gpl_only       = false,
4472         .ret_type       = RET_INTEGER,
4473         .arg1_type      = ARG_ANYTHING,
4474         .arg2_type      = ARG_ANYTHING,
4475 };
4476
4477 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u64, key,
4478            u64, flags)
4479 {
4480         return map->ops->map_redirect(map, key, flags);
4481 }
4482
4483 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4484         .func           = bpf_xdp_redirect_map,
4485         .gpl_only       = false,
4486         .ret_type       = RET_INTEGER,
4487         .arg1_type      = ARG_CONST_MAP_PTR,
4488         .arg2_type      = ARG_ANYTHING,
4489         .arg3_type      = ARG_ANYTHING,
4490 };
4491
4492 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4493                                   unsigned long off, unsigned long len)
4494 {
4495         void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4496
4497         if (unlikely(!ptr))
4498                 return len;
4499         if (ptr != dst_buff)
4500                 memcpy(dst_buff, ptr, len);
4501
4502         return 0;
4503 }
4504
4505 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4506            u64, flags, void *, meta, u64, meta_size)
4507 {
4508         u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4509
4510         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4511                 return -EINVAL;
4512         if (unlikely(!skb || skb_size > skb->len))
4513                 return -EFAULT;
4514
4515         return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4516                                 bpf_skb_copy);
4517 }
4518
4519 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4520         .func           = bpf_skb_event_output,
4521         .gpl_only       = true,
4522         .ret_type       = RET_INTEGER,
4523         .arg1_type      = ARG_PTR_TO_CTX,
4524         .arg2_type      = ARG_CONST_MAP_PTR,
4525         .arg3_type      = ARG_ANYTHING,
4526         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4527         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4528 };
4529
4530 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4531
4532 const struct bpf_func_proto bpf_skb_output_proto = {
4533         .func           = bpf_skb_event_output,
4534         .gpl_only       = true,
4535         .ret_type       = RET_INTEGER,
4536         .arg1_type      = ARG_PTR_TO_BTF_ID,
4537         .arg1_btf_id    = &bpf_skb_output_btf_ids[0],
4538         .arg2_type      = ARG_CONST_MAP_PTR,
4539         .arg3_type      = ARG_ANYTHING,
4540         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4541         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4542 };
4543
4544 static unsigned short bpf_tunnel_key_af(u64 flags)
4545 {
4546         return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4547 }
4548
4549 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4550            u32, size, u64, flags)
4551 {
4552         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4553         u8 compat[sizeof(struct bpf_tunnel_key)];
4554         void *to_orig = to;
4555         int err;
4556
4557         if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6 |
4558                                          BPF_F_TUNINFO_FLAGS)))) {
4559                 err = -EINVAL;
4560                 goto err_clear;
4561         }
4562         if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4563                 err = -EPROTO;
4564                 goto err_clear;
4565         }
4566         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4567                 err = -EINVAL;
4568                 switch (size) {
4569                 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4570                 case offsetof(struct bpf_tunnel_key, tunnel_label):
4571                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4572                         goto set_compat;
4573                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4574                         /* Fixup deprecated structure layouts here, so we have
4575                          * a common path later on.
4576                          */
4577                         if (ip_tunnel_info_af(info) != AF_INET)
4578                                 goto err_clear;
4579 set_compat:
4580                         to = (struct bpf_tunnel_key *)compat;
4581                         break;
4582                 default:
4583                         goto err_clear;
4584                 }
4585         }
4586
4587         to->tunnel_id = be64_to_cpu(info->key.tun_id);
4588         to->tunnel_tos = info->key.tos;
4589         to->tunnel_ttl = info->key.ttl;
4590         if (flags & BPF_F_TUNINFO_FLAGS)
4591                 to->tunnel_flags = info->key.tun_flags;
4592         else
4593                 to->tunnel_ext = 0;
4594
4595         if (flags & BPF_F_TUNINFO_IPV6) {
4596                 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4597                        sizeof(to->remote_ipv6));
4598                 memcpy(to->local_ipv6, &info->key.u.ipv6.dst,
4599                        sizeof(to->local_ipv6));
4600                 to->tunnel_label = be32_to_cpu(info->key.label);
4601         } else {
4602                 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4603                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4604                 to->local_ipv4 = be32_to_cpu(info->key.u.ipv4.dst);
4605                 memset(&to->local_ipv6[1], 0, sizeof(__u32) * 3);
4606                 to->tunnel_label = 0;
4607         }
4608
4609         if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4610                 memcpy(to_orig, to, size);
4611
4612         return 0;
4613 err_clear:
4614         memset(to_orig, 0, size);
4615         return err;
4616 }
4617
4618 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4619         .func           = bpf_skb_get_tunnel_key,
4620         .gpl_only       = false,
4621         .ret_type       = RET_INTEGER,
4622         .arg1_type      = ARG_PTR_TO_CTX,
4623         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
4624         .arg3_type      = ARG_CONST_SIZE,
4625         .arg4_type      = ARG_ANYTHING,
4626 };
4627
4628 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4629 {
4630         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4631         int err;
4632
4633         if (unlikely(!info ||
4634                      !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4635                 err = -ENOENT;
4636                 goto err_clear;
4637         }
4638         if (unlikely(size < info->options_len)) {
4639                 err = -ENOMEM;
4640                 goto err_clear;
4641         }
4642
4643         ip_tunnel_info_opts_get(to, info);
4644         if (size > info->options_len)
4645                 memset(to + info->options_len, 0, size - info->options_len);
4646
4647         return info->options_len;
4648 err_clear:
4649         memset(to, 0, size);
4650         return err;
4651 }
4652
4653 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4654         .func           = bpf_skb_get_tunnel_opt,
4655         .gpl_only       = false,
4656         .ret_type       = RET_INTEGER,
4657         .arg1_type      = ARG_PTR_TO_CTX,
4658         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
4659         .arg3_type      = ARG_CONST_SIZE,
4660 };
4661
4662 static struct metadata_dst __percpu *md_dst;
4663
4664 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4665            const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4666 {
4667         struct metadata_dst *md = this_cpu_ptr(md_dst);
4668         u8 compat[sizeof(struct bpf_tunnel_key)];
4669         struct ip_tunnel_info *info;
4670
4671         if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4672                                BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER |
4673                                BPF_F_NO_TUNNEL_KEY)))
4674                 return -EINVAL;
4675         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4676                 switch (size) {
4677                 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4678                 case offsetof(struct bpf_tunnel_key, tunnel_label):
4679                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4680                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4681                         /* Fixup deprecated structure layouts here, so we have
4682                          * a common path later on.
4683                          */
4684                         memcpy(compat, from, size);
4685                         memset(compat + size, 0, sizeof(compat) - size);
4686                         from = (const struct bpf_tunnel_key *) compat;
4687                         break;
4688                 default:
4689                         return -EINVAL;
4690                 }
4691         }
4692         if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4693                      from->tunnel_ext))
4694                 return -EINVAL;
4695
4696         skb_dst_drop(skb);
4697         dst_hold((struct dst_entry *) md);
4698         skb_dst_set(skb, (struct dst_entry *) md);
4699
4700         info = &md->u.tun_info;
4701         memset(info, 0, sizeof(*info));
4702         info->mode = IP_TUNNEL_INFO_TX;
4703
4704         info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4705         if (flags & BPF_F_DONT_FRAGMENT)
4706                 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4707         if (flags & BPF_F_ZERO_CSUM_TX)
4708                 info->key.tun_flags &= ~TUNNEL_CSUM;
4709         if (flags & BPF_F_SEQ_NUMBER)
4710                 info->key.tun_flags |= TUNNEL_SEQ;
4711         if (flags & BPF_F_NO_TUNNEL_KEY)
4712                 info->key.tun_flags &= ~TUNNEL_KEY;
4713
4714         info->key.tun_id = cpu_to_be64(from->tunnel_id);
4715         info->key.tos = from->tunnel_tos;
4716         info->key.ttl = from->tunnel_ttl;
4717
4718         if (flags & BPF_F_TUNINFO_IPV6) {
4719                 info->mode |= IP_TUNNEL_INFO_IPV6;
4720                 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4721                        sizeof(from->remote_ipv6));
4722                 memcpy(&info->key.u.ipv6.src, from->local_ipv6,
4723                        sizeof(from->local_ipv6));
4724                 info->key.label = cpu_to_be32(from->tunnel_label) &
4725                                   IPV6_FLOWLABEL_MASK;
4726         } else {
4727                 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4728                 info->key.u.ipv4.src = cpu_to_be32(from->local_ipv4);
4729                 info->key.flow_flags = FLOWI_FLAG_ANYSRC;
4730         }
4731
4732         return 0;
4733 }
4734
4735 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4736         .func           = bpf_skb_set_tunnel_key,
4737         .gpl_only       = false,
4738         .ret_type       = RET_INTEGER,
4739         .arg1_type      = ARG_PTR_TO_CTX,
4740         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4741         .arg3_type      = ARG_CONST_SIZE,
4742         .arg4_type      = ARG_ANYTHING,
4743 };
4744
4745 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4746            const u8 *, from, u32, size)
4747 {
4748         struct ip_tunnel_info *info = skb_tunnel_info(skb);
4749         const struct metadata_dst *md = this_cpu_ptr(md_dst);
4750
4751         if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4752                 return -EINVAL;
4753         if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4754                 return -ENOMEM;
4755
4756         ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4757
4758         return 0;
4759 }
4760
4761 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4762         .func           = bpf_skb_set_tunnel_opt,
4763         .gpl_only       = false,
4764         .ret_type       = RET_INTEGER,
4765         .arg1_type      = ARG_PTR_TO_CTX,
4766         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4767         .arg3_type      = ARG_CONST_SIZE,
4768 };
4769
4770 static const struct bpf_func_proto *
4771 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4772 {
4773         if (!md_dst) {
4774                 struct metadata_dst __percpu *tmp;
4775
4776                 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4777                                                 METADATA_IP_TUNNEL,
4778                                                 GFP_KERNEL);
4779                 if (!tmp)
4780                         return NULL;
4781                 if (cmpxchg(&md_dst, NULL, tmp))
4782                         metadata_dst_free_percpu(tmp);
4783         }
4784
4785         switch (which) {
4786         case BPF_FUNC_skb_set_tunnel_key:
4787                 return &bpf_skb_set_tunnel_key_proto;
4788         case BPF_FUNC_skb_set_tunnel_opt:
4789                 return &bpf_skb_set_tunnel_opt_proto;
4790         default:
4791                 return NULL;
4792         }
4793 }
4794
4795 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4796            u32, idx)
4797 {
4798         struct bpf_array *array = container_of(map, struct bpf_array, map);
4799         struct cgroup *cgrp;
4800         struct sock *sk;
4801
4802         sk = skb_to_full_sk(skb);
4803         if (!sk || !sk_fullsock(sk))
4804                 return -ENOENT;
4805         if (unlikely(idx >= array->map.max_entries))
4806                 return -E2BIG;
4807
4808         cgrp = READ_ONCE(array->ptrs[idx]);
4809         if (unlikely(!cgrp))
4810                 return -EAGAIN;
4811
4812         return sk_under_cgroup_hierarchy(sk, cgrp);
4813 }
4814
4815 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4816         .func           = bpf_skb_under_cgroup,
4817         .gpl_only       = false,
4818         .ret_type       = RET_INTEGER,
4819         .arg1_type      = ARG_PTR_TO_CTX,
4820         .arg2_type      = ARG_CONST_MAP_PTR,
4821         .arg3_type      = ARG_ANYTHING,
4822 };
4823
4824 #ifdef CONFIG_SOCK_CGROUP_DATA
4825 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4826 {
4827         struct cgroup *cgrp;
4828
4829         sk = sk_to_full_sk(sk);
4830         if (!sk || !sk_fullsock(sk))
4831                 return 0;
4832
4833         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4834         return cgroup_id(cgrp);
4835 }
4836
4837 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4838 {
4839         return __bpf_sk_cgroup_id(skb->sk);
4840 }
4841
4842 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4843         .func           = bpf_skb_cgroup_id,
4844         .gpl_only       = false,
4845         .ret_type       = RET_INTEGER,
4846         .arg1_type      = ARG_PTR_TO_CTX,
4847 };
4848
4849 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4850                                               int ancestor_level)
4851 {
4852         struct cgroup *ancestor;
4853         struct cgroup *cgrp;
4854
4855         sk = sk_to_full_sk(sk);
4856         if (!sk || !sk_fullsock(sk))
4857                 return 0;
4858
4859         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4860         ancestor = cgroup_ancestor(cgrp, ancestor_level);
4861         if (!ancestor)
4862                 return 0;
4863
4864         return cgroup_id(ancestor);
4865 }
4866
4867 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4868            ancestor_level)
4869 {
4870         return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4871 }
4872
4873 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4874         .func           = bpf_skb_ancestor_cgroup_id,
4875         .gpl_only       = false,
4876         .ret_type       = RET_INTEGER,
4877         .arg1_type      = ARG_PTR_TO_CTX,
4878         .arg2_type      = ARG_ANYTHING,
4879 };
4880
4881 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4882 {
4883         return __bpf_sk_cgroup_id(sk);
4884 }
4885
4886 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4887         .func           = bpf_sk_cgroup_id,
4888         .gpl_only       = false,
4889         .ret_type       = RET_INTEGER,
4890         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4891 };
4892
4893 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4894 {
4895         return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4896 }
4897
4898 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4899         .func           = bpf_sk_ancestor_cgroup_id,
4900         .gpl_only       = false,
4901         .ret_type       = RET_INTEGER,
4902         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4903         .arg2_type      = ARG_ANYTHING,
4904 };
4905 #endif
4906
4907 static unsigned long bpf_xdp_copy(void *dst, const void *ctx,
4908                                   unsigned long off, unsigned long len)
4909 {
4910         struct xdp_buff *xdp = (struct xdp_buff *)ctx;
4911
4912         bpf_xdp_copy_buf(xdp, off, dst, len, false);
4913         return 0;
4914 }
4915
4916 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4917            u64, flags, void *, meta, u64, meta_size)
4918 {
4919         u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4920
4921         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4922                 return -EINVAL;
4923
4924         if (unlikely(!xdp || xdp_size > xdp_get_buff_len(xdp)))
4925                 return -EFAULT;
4926
4927         return bpf_event_output(map, flags, meta, meta_size, xdp,
4928                                 xdp_size, bpf_xdp_copy);
4929 }
4930
4931 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4932         .func           = bpf_xdp_event_output,
4933         .gpl_only       = true,
4934         .ret_type       = RET_INTEGER,
4935         .arg1_type      = ARG_PTR_TO_CTX,
4936         .arg2_type      = ARG_CONST_MAP_PTR,
4937         .arg3_type      = ARG_ANYTHING,
4938         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4939         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4940 };
4941
4942 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4943
4944 const struct bpf_func_proto bpf_xdp_output_proto = {
4945         .func           = bpf_xdp_event_output,
4946         .gpl_only       = true,
4947         .ret_type       = RET_INTEGER,
4948         .arg1_type      = ARG_PTR_TO_BTF_ID,
4949         .arg1_btf_id    = &bpf_xdp_output_btf_ids[0],
4950         .arg2_type      = ARG_CONST_MAP_PTR,
4951         .arg3_type      = ARG_ANYTHING,
4952         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4953         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4954 };
4955
4956 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4957 {
4958         return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4959 }
4960
4961 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4962         .func           = bpf_get_socket_cookie,
4963         .gpl_only       = false,
4964         .ret_type       = RET_INTEGER,
4965         .arg1_type      = ARG_PTR_TO_CTX,
4966 };
4967
4968 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4969 {
4970         return __sock_gen_cookie(ctx->sk);
4971 }
4972
4973 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4974         .func           = bpf_get_socket_cookie_sock_addr,
4975         .gpl_only       = false,
4976         .ret_type       = RET_INTEGER,
4977         .arg1_type      = ARG_PTR_TO_CTX,
4978 };
4979
4980 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4981 {
4982         return __sock_gen_cookie(ctx);
4983 }
4984
4985 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4986         .func           = bpf_get_socket_cookie_sock,
4987         .gpl_only       = false,
4988         .ret_type       = RET_INTEGER,
4989         .arg1_type      = ARG_PTR_TO_CTX,
4990 };
4991
4992 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
4993 {
4994         return sk ? sock_gen_cookie(sk) : 0;
4995 }
4996
4997 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
4998         .func           = bpf_get_socket_ptr_cookie,
4999         .gpl_only       = false,
5000         .ret_type       = RET_INTEGER,
5001         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON | PTR_MAYBE_NULL,
5002 };
5003
5004 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
5005 {
5006         return __sock_gen_cookie(ctx->sk);
5007 }
5008
5009 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
5010         .func           = bpf_get_socket_cookie_sock_ops,
5011         .gpl_only       = false,
5012         .ret_type       = RET_INTEGER,
5013         .arg1_type      = ARG_PTR_TO_CTX,
5014 };
5015
5016 static u64 __bpf_get_netns_cookie(struct sock *sk)
5017 {
5018         const struct net *net = sk ? sock_net(sk) : &init_net;
5019
5020         return net->net_cookie;
5021 }
5022
5023 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
5024 {
5025         return __bpf_get_netns_cookie(ctx);
5026 }
5027
5028 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
5029         .func           = bpf_get_netns_cookie_sock,
5030         .gpl_only       = false,
5031         .ret_type       = RET_INTEGER,
5032         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
5033 };
5034
5035 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
5036 {
5037         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5038 }
5039
5040 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
5041         .func           = bpf_get_netns_cookie_sock_addr,
5042         .gpl_only       = false,
5043         .ret_type       = RET_INTEGER,
5044         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
5045 };
5046
5047 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
5048 {
5049         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5050 }
5051
5052 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = {
5053         .func           = bpf_get_netns_cookie_sock_ops,
5054         .gpl_only       = false,
5055         .ret_type       = RET_INTEGER,
5056         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
5057 };
5058
5059 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx)
5060 {
5061         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5062 }
5063
5064 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = {
5065         .func           = bpf_get_netns_cookie_sk_msg,
5066         .gpl_only       = false,
5067         .ret_type       = RET_INTEGER,
5068         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
5069 };
5070
5071 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
5072 {
5073         struct sock *sk = sk_to_full_sk(skb->sk);
5074         kuid_t kuid;
5075
5076         if (!sk || !sk_fullsock(sk))
5077                 return overflowuid;
5078         kuid = sock_net_uid(sock_net(sk), sk);
5079         return from_kuid_munged(sock_net(sk)->user_ns, kuid);
5080 }
5081
5082 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
5083         .func           = bpf_get_socket_uid,
5084         .gpl_only       = false,
5085         .ret_type       = RET_INTEGER,
5086         .arg1_type      = ARG_PTR_TO_CTX,
5087 };
5088
5089 static int sol_socket_sockopt(struct sock *sk, int optname,
5090                               char *optval, int *optlen,
5091                               bool getopt)
5092 {
5093         switch (optname) {
5094         case SO_REUSEADDR:
5095         case SO_SNDBUF:
5096         case SO_RCVBUF:
5097         case SO_KEEPALIVE:
5098         case SO_PRIORITY:
5099         case SO_REUSEPORT:
5100         case SO_RCVLOWAT:
5101         case SO_MARK:
5102         case SO_MAX_PACING_RATE:
5103         case SO_BINDTOIFINDEX:
5104         case SO_TXREHASH:
5105                 if (*optlen != sizeof(int))
5106                         return -EINVAL;
5107                 break;
5108         case SO_BINDTODEVICE:
5109                 break;
5110         default:
5111                 return -EINVAL;
5112         }
5113
5114         if (getopt) {
5115                 if (optname == SO_BINDTODEVICE)
5116                         return -EINVAL;
5117                 return sk_getsockopt(sk, SOL_SOCKET, optname,
5118                                      KERNEL_SOCKPTR(optval),
5119                                      KERNEL_SOCKPTR(optlen));
5120         }
5121
5122         return sk_setsockopt(sk, SOL_SOCKET, optname,
5123                              KERNEL_SOCKPTR(optval), *optlen);
5124 }
5125
5126 static int bpf_sol_tcp_setsockopt(struct sock *sk, int optname,
5127                                   char *optval, int optlen)
5128 {
5129         struct tcp_sock *tp = tcp_sk(sk);
5130         unsigned long timeout;
5131         int val;
5132
5133         if (optlen != sizeof(int))
5134                 return -EINVAL;
5135
5136         val = *(int *)optval;
5137
5138         /* Only some options are supported */
5139         switch (optname) {
5140         case TCP_BPF_IW:
5141                 if (val <= 0 || tp->data_segs_out > tp->syn_data)
5142                         return -EINVAL;
5143                 tcp_snd_cwnd_set(tp, val);
5144                 break;
5145         case TCP_BPF_SNDCWND_CLAMP:
5146                 if (val <= 0)
5147                         return -EINVAL;
5148                 tp->snd_cwnd_clamp = val;
5149                 tp->snd_ssthresh = val;
5150                 break;
5151         case TCP_BPF_DELACK_MAX:
5152                 timeout = usecs_to_jiffies(val);
5153                 if (timeout > TCP_DELACK_MAX ||
5154                     timeout < TCP_TIMEOUT_MIN)
5155                         return -EINVAL;
5156                 inet_csk(sk)->icsk_delack_max = timeout;
5157                 break;
5158         case TCP_BPF_RTO_MIN:
5159                 timeout = usecs_to_jiffies(val);
5160                 if (timeout > TCP_RTO_MIN ||
5161                     timeout < TCP_TIMEOUT_MIN)
5162                         return -EINVAL;
5163                 inet_csk(sk)->icsk_rto_min = timeout;
5164                 break;
5165         default:
5166                 return -EINVAL;
5167         }
5168
5169         return 0;
5170 }
5171
5172 static int sol_tcp_sockopt_congestion(struct sock *sk, char *optval,
5173                                       int *optlen, bool getopt)
5174 {
5175         struct tcp_sock *tp;
5176         int ret;
5177
5178         if (*optlen < 2)
5179                 return -EINVAL;
5180
5181         if (getopt) {
5182                 if (!inet_csk(sk)->icsk_ca_ops)
5183                         return -EINVAL;
5184                 /* BPF expects NULL-terminated tcp-cc string */
5185                 optval[--(*optlen)] = '\0';
5186                 return do_tcp_getsockopt(sk, SOL_TCP, TCP_CONGESTION,
5187                                          KERNEL_SOCKPTR(optval),
5188                                          KERNEL_SOCKPTR(optlen));
5189         }
5190
5191         /* "cdg" is the only cc that alloc a ptr
5192          * in inet_csk_ca area.  The bpf-tcp-cc may
5193          * overwrite this ptr after switching to cdg.
5194          */
5195         if (*optlen >= sizeof("cdg") - 1 && !strncmp("cdg", optval, *optlen))
5196                 return -ENOTSUPP;
5197
5198         /* It stops this looping
5199          *
5200          * .init => bpf_setsockopt(tcp_cc) => .init =>
5201          * bpf_setsockopt(tcp_cc)" => .init => ....
5202          *
5203          * The second bpf_setsockopt(tcp_cc) is not allowed
5204          * in order to break the loop when both .init
5205          * are the same bpf prog.
5206          *
5207          * This applies even the second bpf_setsockopt(tcp_cc)
5208          * does not cause a loop.  This limits only the first
5209          * '.init' can call bpf_setsockopt(TCP_CONGESTION) to
5210          * pick a fallback cc (eg. peer does not support ECN)
5211          * and the second '.init' cannot fallback to
5212          * another.
5213          */
5214         tp = tcp_sk(sk);
5215         if (tp->bpf_chg_cc_inprogress)
5216                 return -EBUSY;
5217
5218         tp->bpf_chg_cc_inprogress = 1;
5219         ret = do_tcp_setsockopt(sk, SOL_TCP, TCP_CONGESTION,
5220                                 KERNEL_SOCKPTR(optval), *optlen);
5221         tp->bpf_chg_cc_inprogress = 0;
5222         return ret;
5223 }
5224
5225 static int sol_tcp_sockopt(struct sock *sk, int optname,
5226                            char *optval, int *optlen,
5227                            bool getopt)
5228 {
5229         if (sk->sk_protocol != IPPROTO_TCP)
5230                 return -EINVAL;
5231
5232         switch (optname) {
5233         case TCP_NODELAY:
5234         case TCP_MAXSEG:
5235         case TCP_KEEPIDLE:
5236         case TCP_KEEPINTVL:
5237         case TCP_KEEPCNT:
5238         case TCP_SYNCNT:
5239         case TCP_WINDOW_CLAMP:
5240         case TCP_THIN_LINEAR_TIMEOUTS:
5241         case TCP_USER_TIMEOUT:
5242         case TCP_NOTSENT_LOWAT:
5243         case TCP_SAVE_SYN:
5244                 if (*optlen != sizeof(int))
5245                         return -EINVAL;
5246                 break;
5247         case TCP_CONGESTION:
5248                 return sol_tcp_sockopt_congestion(sk, optval, optlen, getopt);
5249         case TCP_SAVED_SYN:
5250                 if (*optlen < 1)
5251                         return -EINVAL;
5252                 break;
5253         default:
5254                 if (getopt)
5255                         return -EINVAL;
5256                 return bpf_sol_tcp_setsockopt(sk, optname, optval, *optlen);
5257         }
5258
5259         if (getopt) {
5260                 if (optname == TCP_SAVED_SYN) {
5261                         struct tcp_sock *tp = tcp_sk(sk);
5262
5263                         if (!tp->saved_syn ||
5264                             *optlen > tcp_saved_syn_len(tp->saved_syn))
5265                                 return -EINVAL;
5266                         memcpy(optval, tp->saved_syn->data, *optlen);
5267                         /* It cannot free tp->saved_syn here because it
5268                          * does not know if the user space still needs it.
5269                          */
5270                         return 0;
5271                 }
5272
5273                 return do_tcp_getsockopt(sk, SOL_TCP, optname,
5274                                          KERNEL_SOCKPTR(optval),
5275                                          KERNEL_SOCKPTR(optlen));
5276         }
5277
5278         return do_tcp_setsockopt(sk, SOL_TCP, optname,
5279                                  KERNEL_SOCKPTR(optval), *optlen);
5280 }
5281
5282 static int sol_ip_sockopt(struct sock *sk, int optname,
5283                           char *optval, int *optlen,
5284                           bool getopt)
5285 {
5286         if (sk->sk_family != AF_INET)
5287                 return -EINVAL;
5288
5289         switch (optname) {
5290         case IP_TOS:
5291                 if (*optlen != sizeof(int))
5292                         return -EINVAL;
5293                 break;
5294         default:
5295                 return -EINVAL;
5296         }
5297
5298         if (getopt)
5299                 return do_ip_getsockopt(sk, SOL_IP, optname,
5300                                         KERNEL_SOCKPTR(optval),
5301                                         KERNEL_SOCKPTR(optlen));
5302
5303         return do_ip_setsockopt(sk, SOL_IP, optname,
5304                                 KERNEL_SOCKPTR(optval), *optlen);
5305 }
5306
5307 static int sol_ipv6_sockopt(struct sock *sk, int optname,
5308                             char *optval, int *optlen,
5309                             bool getopt)
5310 {
5311         if (sk->sk_family != AF_INET6)
5312                 return -EINVAL;
5313
5314         switch (optname) {
5315         case IPV6_TCLASS:
5316         case IPV6_AUTOFLOWLABEL:
5317                 if (*optlen != sizeof(int))
5318                         return -EINVAL;
5319                 break;
5320         default:
5321                 return -EINVAL;
5322         }
5323
5324         if (getopt)
5325                 return ipv6_bpf_stub->ipv6_getsockopt(sk, SOL_IPV6, optname,
5326                                                       KERNEL_SOCKPTR(optval),
5327                                                       KERNEL_SOCKPTR(optlen));
5328
5329         return ipv6_bpf_stub->ipv6_setsockopt(sk, SOL_IPV6, optname,
5330                                               KERNEL_SOCKPTR(optval), *optlen);
5331 }
5332
5333 static int __bpf_setsockopt(struct sock *sk, int level, int optname,
5334                             char *optval, int optlen)
5335 {
5336         if (!sk_fullsock(sk))
5337                 return -EINVAL;
5338
5339         if (level == SOL_SOCKET)
5340                 return sol_socket_sockopt(sk, optname, optval, &optlen, false);
5341         else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5342                 return sol_ip_sockopt(sk, optname, optval, &optlen, false);
5343         else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5344                 return sol_ipv6_sockopt(sk, optname, optval, &optlen, false);
5345         else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5346                 return sol_tcp_sockopt(sk, optname, optval, &optlen, false);
5347
5348         return -EINVAL;
5349 }
5350
5351 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
5352                            char *optval, int optlen)
5353 {
5354         if (sk_fullsock(sk))
5355                 sock_owned_by_me(sk);
5356         return __bpf_setsockopt(sk, level, optname, optval, optlen);
5357 }
5358
5359 static int __bpf_getsockopt(struct sock *sk, int level, int optname,
5360                             char *optval, int optlen)
5361 {
5362         int err, saved_optlen = optlen;
5363
5364         if (!sk_fullsock(sk)) {
5365                 err = -EINVAL;
5366                 goto done;
5367         }
5368
5369         if (level == SOL_SOCKET)
5370                 err = sol_socket_sockopt(sk, optname, optval, &optlen, true);
5371         else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5372                 err = sol_tcp_sockopt(sk, optname, optval, &optlen, true);
5373         else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5374                 err = sol_ip_sockopt(sk, optname, optval, &optlen, true);
5375         else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5376                 err = sol_ipv6_sockopt(sk, optname, optval, &optlen, true);
5377         else
5378                 err = -EINVAL;
5379
5380 done:
5381         if (err)
5382                 optlen = 0;
5383         if (optlen < saved_optlen)
5384                 memset(optval + optlen, 0, saved_optlen - optlen);
5385         return err;
5386 }
5387
5388 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
5389                            char *optval, int optlen)
5390 {
5391         if (sk_fullsock(sk))
5392                 sock_owned_by_me(sk);
5393         return __bpf_getsockopt(sk, level, optname, optval, optlen);
5394 }
5395
5396 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level,
5397            int, optname, char *, optval, int, optlen)
5398 {
5399         return _bpf_setsockopt(sk, level, optname, optval, optlen);
5400 }
5401
5402 const struct bpf_func_proto bpf_sk_setsockopt_proto = {
5403         .func           = bpf_sk_setsockopt,
5404         .gpl_only       = false,
5405         .ret_type       = RET_INTEGER,
5406         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5407         .arg2_type      = ARG_ANYTHING,
5408         .arg3_type      = ARG_ANYTHING,
5409         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5410         .arg5_type      = ARG_CONST_SIZE,
5411 };
5412
5413 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level,
5414            int, optname, char *, optval, int, optlen)
5415 {
5416         return _bpf_getsockopt(sk, level, optname, optval, optlen);
5417 }
5418
5419 const struct bpf_func_proto bpf_sk_getsockopt_proto = {
5420         .func           = bpf_sk_getsockopt,
5421         .gpl_only       = false,
5422         .ret_type       = RET_INTEGER,
5423         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5424         .arg2_type      = ARG_ANYTHING,
5425         .arg3_type      = ARG_ANYTHING,
5426         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5427         .arg5_type      = ARG_CONST_SIZE,
5428 };
5429
5430 BPF_CALL_5(bpf_unlocked_sk_setsockopt, struct sock *, sk, int, level,
5431            int, optname, char *, optval, int, optlen)
5432 {
5433         return __bpf_setsockopt(sk, level, optname, optval, optlen);
5434 }
5435
5436 const struct bpf_func_proto bpf_unlocked_sk_setsockopt_proto = {
5437         .func           = bpf_unlocked_sk_setsockopt,
5438         .gpl_only       = false,
5439         .ret_type       = RET_INTEGER,
5440         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5441         .arg2_type      = ARG_ANYTHING,
5442         .arg3_type      = ARG_ANYTHING,
5443         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5444         .arg5_type      = ARG_CONST_SIZE,
5445 };
5446
5447 BPF_CALL_5(bpf_unlocked_sk_getsockopt, struct sock *, sk, int, level,
5448            int, optname, char *, optval, int, optlen)
5449 {
5450         return __bpf_getsockopt(sk, level, optname, optval, optlen);
5451 }
5452
5453 const struct bpf_func_proto bpf_unlocked_sk_getsockopt_proto = {
5454         .func           = bpf_unlocked_sk_getsockopt,
5455         .gpl_only       = false,
5456         .ret_type       = RET_INTEGER,
5457         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5458         .arg2_type      = ARG_ANYTHING,
5459         .arg3_type      = ARG_ANYTHING,
5460         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5461         .arg5_type      = ARG_CONST_SIZE,
5462 };
5463
5464 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5465            int, level, int, optname, char *, optval, int, optlen)
5466 {
5467         return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5468 }
5469
5470 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5471         .func           = bpf_sock_addr_setsockopt,
5472         .gpl_only       = false,
5473         .ret_type       = RET_INTEGER,
5474         .arg1_type      = ARG_PTR_TO_CTX,
5475         .arg2_type      = ARG_ANYTHING,
5476         .arg3_type      = ARG_ANYTHING,
5477         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5478         .arg5_type      = ARG_CONST_SIZE,
5479 };
5480
5481 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5482            int, level, int, optname, char *, optval, int, optlen)
5483 {
5484         return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5485 }
5486
5487 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5488         .func           = bpf_sock_addr_getsockopt,
5489         .gpl_only       = false,
5490         .ret_type       = RET_INTEGER,
5491         .arg1_type      = ARG_PTR_TO_CTX,
5492         .arg2_type      = ARG_ANYTHING,
5493         .arg3_type      = ARG_ANYTHING,
5494         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5495         .arg5_type      = ARG_CONST_SIZE,
5496 };
5497
5498 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5499            int, level, int, optname, char *, optval, int, optlen)
5500 {
5501         return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5502 }
5503
5504 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5505         .func           = bpf_sock_ops_setsockopt,
5506         .gpl_only       = false,
5507         .ret_type       = RET_INTEGER,
5508         .arg1_type      = ARG_PTR_TO_CTX,
5509         .arg2_type      = ARG_ANYTHING,
5510         .arg3_type      = ARG_ANYTHING,
5511         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5512         .arg5_type      = ARG_CONST_SIZE,
5513 };
5514
5515 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5516                                 int optname, const u8 **start)
5517 {
5518         struct sk_buff *syn_skb = bpf_sock->syn_skb;
5519         const u8 *hdr_start;
5520         int ret;
5521
5522         if (syn_skb) {
5523                 /* sk is a request_sock here */
5524
5525                 if (optname == TCP_BPF_SYN) {
5526                         hdr_start = syn_skb->data;
5527                         ret = tcp_hdrlen(syn_skb);
5528                 } else if (optname == TCP_BPF_SYN_IP) {
5529                         hdr_start = skb_network_header(syn_skb);
5530                         ret = skb_network_header_len(syn_skb) +
5531                                 tcp_hdrlen(syn_skb);
5532                 } else {
5533                         /* optname == TCP_BPF_SYN_MAC */
5534                         hdr_start = skb_mac_header(syn_skb);
5535                         ret = skb_mac_header_len(syn_skb) +
5536                                 skb_network_header_len(syn_skb) +
5537                                 tcp_hdrlen(syn_skb);
5538                 }
5539         } else {
5540                 struct sock *sk = bpf_sock->sk;
5541                 struct saved_syn *saved_syn;
5542
5543                 if (sk->sk_state == TCP_NEW_SYN_RECV)
5544                         /* synack retransmit. bpf_sock->syn_skb will
5545                          * not be available.  It has to resort to
5546                          * saved_syn (if it is saved).
5547                          */
5548                         saved_syn = inet_reqsk(sk)->saved_syn;
5549                 else
5550                         saved_syn = tcp_sk(sk)->saved_syn;
5551
5552                 if (!saved_syn)
5553                         return -ENOENT;
5554
5555                 if (optname == TCP_BPF_SYN) {
5556                         hdr_start = saved_syn->data +
5557                                 saved_syn->mac_hdrlen +
5558                                 saved_syn->network_hdrlen;
5559                         ret = saved_syn->tcp_hdrlen;
5560                 } else if (optname == TCP_BPF_SYN_IP) {
5561                         hdr_start = saved_syn->data +
5562                                 saved_syn->mac_hdrlen;
5563                         ret = saved_syn->network_hdrlen +
5564                                 saved_syn->tcp_hdrlen;
5565                 } else {
5566                         /* optname == TCP_BPF_SYN_MAC */
5567
5568                         /* TCP_SAVE_SYN may not have saved the mac hdr */
5569                         if (!saved_syn->mac_hdrlen)
5570                                 return -ENOENT;
5571
5572                         hdr_start = saved_syn->data;
5573                         ret = saved_syn->mac_hdrlen +
5574                                 saved_syn->network_hdrlen +
5575                                 saved_syn->tcp_hdrlen;
5576                 }
5577         }
5578
5579         *start = hdr_start;
5580         return ret;
5581 }
5582
5583 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5584            int, level, int, optname, char *, optval, int, optlen)
5585 {
5586         if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5587             optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5588                 int ret, copy_len = 0;
5589                 const u8 *start;
5590
5591                 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5592                 if (ret > 0) {
5593                         copy_len = ret;
5594                         if (optlen < copy_len) {
5595                                 copy_len = optlen;
5596                                 ret = -ENOSPC;
5597                         }
5598
5599                         memcpy(optval, start, copy_len);
5600                 }
5601
5602                 /* Zero out unused buffer at the end */
5603                 memset(optval + copy_len, 0, optlen - copy_len);
5604
5605                 return ret;
5606         }
5607
5608         return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5609 }
5610
5611 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5612         .func           = bpf_sock_ops_getsockopt,
5613         .gpl_only       = false,
5614         .ret_type       = RET_INTEGER,
5615         .arg1_type      = ARG_PTR_TO_CTX,
5616         .arg2_type      = ARG_ANYTHING,
5617         .arg3_type      = ARG_ANYTHING,
5618         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5619         .arg5_type      = ARG_CONST_SIZE,
5620 };
5621
5622 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5623            int, argval)
5624 {
5625         struct sock *sk = bpf_sock->sk;
5626         int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5627
5628         if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5629                 return -EINVAL;
5630
5631         tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5632
5633         return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5634 }
5635
5636 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5637         .func           = bpf_sock_ops_cb_flags_set,
5638         .gpl_only       = false,
5639         .ret_type       = RET_INTEGER,
5640         .arg1_type      = ARG_PTR_TO_CTX,
5641         .arg2_type      = ARG_ANYTHING,
5642 };
5643
5644 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5645 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5646
5647 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5648            int, addr_len)
5649 {
5650 #ifdef CONFIG_INET
5651         struct sock *sk = ctx->sk;
5652         u32 flags = BIND_FROM_BPF;
5653         int err;
5654
5655         err = -EINVAL;
5656         if (addr_len < offsetofend(struct sockaddr, sa_family))
5657                 return err;
5658         if (addr->sa_family == AF_INET) {
5659                 if (addr_len < sizeof(struct sockaddr_in))
5660                         return err;
5661                 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5662                         flags |= BIND_FORCE_ADDRESS_NO_PORT;
5663                 return __inet_bind(sk, addr, addr_len, flags);
5664 #if IS_ENABLED(CONFIG_IPV6)
5665         } else if (addr->sa_family == AF_INET6) {
5666                 if (addr_len < SIN6_LEN_RFC2133)
5667                         return err;
5668                 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5669                         flags |= BIND_FORCE_ADDRESS_NO_PORT;
5670                 /* ipv6_bpf_stub cannot be NULL, since it's called from
5671                  * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5672                  */
5673                 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5674 #endif /* CONFIG_IPV6 */
5675         }
5676 #endif /* CONFIG_INET */
5677
5678         return -EAFNOSUPPORT;
5679 }
5680
5681 static const struct bpf_func_proto bpf_bind_proto = {
5682         .func           = bpf_bind,
5683         .gpl_only       = false,
5684         .ret_type       = RET_INTEGER,
5685         .arg1_type      = ARG_PTR_TO_CTX,
5686         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5687         .arg3_type      = ARG_CONST_SIZE,
5688 };
5689
5690 #ifdef CONFIG_XFRM
5691
5692 #if (IS_BUILTIN(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF)) || \
5693     (IS_MODULE(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES))
5694
5695 struct metadata_dst __percpu *xfrm_bpf_md_dst;
5696 EXPORT_SYMBOL_GPL(xfrm_bpf_md_dst);
5697
5698 #endif
5699
5700 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5701            struct bpf_xfrm_state *, to, u32, size, u64, flags)
5702 {
5703         const struct sec_path *sp = skb_sec_path(skb);
5704         const struct xfrm_state *x;
5705
5706         if (!sp || unlikely(index >= sp->len || flags))
5707                 goto err_clear;
5708
5709         x = sp->xvec[index];
5710
5711         if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5712                 goto err_clear;
5713
5714         to->reqid = x->props.reqid;
5715         to->spi = x->id.spi;
5716         to->family = x->props.family;
5717         to->ext = 0;
5718
5719         if (to->family == AF_INET6) {
5720                 memcpy(to->remote_ipv6, x->props.saddr.a6,
5721                        sizeof(to->remote_ipv6));
5722         } else {
5723                 to->remote_ipv4 = x->props.saddr.a4;
5724                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5725         }
5726
5727         return 0;
5728 err_clear:
5729         memset(to, 0, size);
5730         return -EINVAL;
5731 }
5732
5733 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5734         .func           = bpf_skb_get_xfrm_state,
5735         .gpl_only       = false,
5736         .ret_type       = RET_INTEGER,
5737         .arg1_type      = ARG_PTR_TO_CTX,
5738         .arg2_type      = ARG_ANYTHING,
5739         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
5740         .arg4_type      = ARG_CONST_SIZE,
5741         .arg5_type      = ARG_ANYTHING,
5742 };
5743 #endif
5744
5745 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5746 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params, u32 mtu)
5747 {
5748         params->h_vlan_TCI = 0;
5749         params->h_vlan_proto = 0;
5750         if (mtu)
5751                 params->mtu_result = mtu; /* union with tot_len */
5752
5753         return 0;
5754 }
5755 #endif
5756
5757 #if IS_ENABLED(CONFIG_INET)
5758 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5759                                u32 flags, bool check_mtu)
5760 {
5761         struct fib_nh_common *nhc;
5762         struct in_device *in_dev;
5763         struct neighbour *neigh;
5764         struct net_device *dev;
5765         struct fib_result res;
5766         struct flowi4 fl4;
5767         u32 mtu = 0;
5768         int err;
5769
5770         dev = dev_get_by_index_rcu(net, params->ifindex);
5771         if (unlikely(!dev))
5772                 return -ENODEV;
5773
5774         /* verify forwarding is enabled on this interface */
5775         in_dev = __in_dev_get_rcu(dev);
5776         if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5777                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5778
5779         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5780                 fl4.flowi4_iif = 1;
5781                 fl4.flowi4_oif = params->ifindex;
5782         } else {
5783                 fl4.flowi4_iif = params->ifindex;
5784                 fl4.flowi4_oif = 0;
5785         }
5786         fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5787         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5788         fl4.flowi4_flags = 0;
5789
5790         fl4.flowi4_proto = params->l4_protocol;
5791         fl4.daddr = params->ipv4_dst;
5792         fl4.saddr = params->ipv4_src;
5793         fl4.fl4_sport = params->sport;
5794         fl4.fl4_dport = params->dport;
5795         fl4.flowi4_multipath_hash = 0;
5796
5797         if (flags & BPF_FIB_LOOKUP_DIRECT) {
5798                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5799                 struct fib_table *tb;
5800
5801                 if (flags & BPF_FIB_LOOKUP_TBID) {
5802                         tbid = params->tbid;
5803                         /* zero out for vlan output */
5804                         params->tbid = 0;
5805                 }
5806
5807                 tb = fib_get_table(net, tbid);
5808                 if (unlikely(!tb))
5809                         return BPF_FIB_LKUP_RET_NOT_FWDED;
5810
5811                 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5812         } else {
5813                 fl4.flowi4_mark = 0;
5814                 fl4.flowi4_secid = 0;
5815                 fl4.flowi4_tun_key.tun_id = 0;
5816                 fl4.flowi4_uid = sock_net_uid(net, NULL);
5817
5818                 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5819         }
5820
5821         if (err) {
5822                 /* map fib lookup errors to RTN_ type */
5823                 if (err == -EINVAL)
5824                         return BPF_FIB_LKUP_RET_BLACKHOLE;
5825                 if (err == -EHOSTUNREACH)
5826                         return BPF_FIB_LKUP_RET_UNREACHABLE;
5827                 if (err == -EACCES)
5828                         return BPF_FIB_LKUP_RET_PROHIBIT;
5829
5830                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5831         }
5832
5833         if (res.type != RTN_UNICAST)
5834                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5835
5836         if (fib_info_num_path(res.fi) > 1)
5837                 fib_select_path(net, &res, &fl4, NULL);
5838
5839         if (check_mtu) {
5840                 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5841                 if (params->tot_len > mtu) {
5842                         params->mtu_result = mtu; /* union with tot_len */
5843                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5844                 }
5845         }
5846
5847         nhc = res.nhc;
5848
5849         /* do not handle lwt encaps right now */
5850         if (nhc->nhc_lwtstate)
5851                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5852
5853         dev = nhc->nhc_dev;
5854
5855         params->rt_metric = res.fi->fib_priority;
5856         params->ifindex = dev->ifindex;
5857
5858         /* xdp and cls_bpf programs are run in RCU-bh so
5859          * rcu_read_lock_bh is not needed here
5860          */
5861         if (likely(nhc->nhc_gw_family != AF_INET6)) {
5862                 if (nhc->nhc_gw_family)
5863                         params->ipv4_dst = nhc->nhc_gw.ipv4;
5864         } else {
5865                 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5866
5867                 params->family = AF_INET6;
5868                 *dst = nhc->nhc_gw.ipv6;
5869         }
5870
5871         if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH)
5872                 goto set_fwd_params;
5873
5874         if (likely(nhc->nhc_gw_family != AF_INET6))
5875                 neigh = __ipv4_neigh_lookup_noref(dev,
5876                                                   (__force u32)params->ipv4_dst);
5877         else
5878                 neigh = __ipv6_neigh_lookup_noref_stub(dev, params->ipv6_dst);
5879
5880         if (!neigh || !(READ_ONCE(neigh->nud_state) & NUD_VALID))
5881                 return BPF_FIB_LKUP_RET_NO_NEIGH;
5882         memcpy(params->dmac, neigh->ha, ETH_ALEN);
5883         memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5884
5885 set_fwd_params:
5886         return bpf_fib_set_fwd_params(params, mtu);
5887 }
5888 #endif
5889
5890 #if IS_ENABLED(CONFIG_IPV6)
5891 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5892                                u32 flags, bool check_mtu)
5893 {
5894         struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5895         struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5896         struct fib6_result res = {};
5897         struct neighbour *neigh;
5898         struct net_device *dev;
5899         struct inet6_dev *idev;
5900         struct flowi6 fl6;
5901         int strict = 0;
5902         int oif, err;
5903         u32 mtu = 0;
5904
5905         /* link local addresses are never forwarded */
5906         if (rt6_need_strict(dst) || rt6_need_strict(src))
5907                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5908
5909         dev = dev_get_by_index_rcu(net, params->ifindex);
5910         if (unlikely(!dev))
5911                 return -ENODEV;
5912
5913         idev = __in6_dev_get_safely(dev);
5914         if (unlikely(!idev || !idev->cnf.forwarding))
5915                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5916
5917         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5918                 fl6.flowi6_iif = 1;
5919                 oif = fl6.flowi6_oif = params->ifindex;
5920         } else {
5921                 oif = fl6.flowi6_iif = params->ifindex;
5922                 fl6.flowi6_oif = 0;
5923                 strict = RT6_LOOKUP_F_HAS_SADDR;
5924         }
5925         fl6.flowlabel = params->flowinfo;
5926         fl6.flowi6_scope = 0;
5927         fl6.flowi6_flags = 0;
5928         fl6.mp_hash = 0;
5929
5930         fl6.flowi6_proto = params->l4_protocol;
5931         fl6.daddr = *dst;
5932         fl6.saddr = *src;
5933         fl6.fl6_sport = params->sport;
5934         fl6.fl6_dport = params->dport;
5935
5936         if (flags & BPF_FIB_LOOKUP_DIRECT) {
5937                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5938                 struct fib6_table *tb;
5939
5940                 if (flags & BPF_FIB_LOOKUP_TBID) {
5941                         tbid = params->tbid;
5942                         /* zero out for vlan output */
5943                         params->tbid = 0;
5944                 }
5945
5946                 tb = ipv6_stub->fib6_get_table(net, tbid);
5947                 if (unlikely(!tb))
5948                         return BPF_FIB_LKUP_RET_NOT_FWDED;
5949
5950                 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5951                                                    strict);
5952         } else {
5953                 fl6.flowi6_mark = 0;
5954                 fl6.flowi6_secid = 0;
5955                 fl6.flowi6_tun_key.tun_id = 0;
5956                 fl6.flowi6_uid = sock_net_uid(net, NULL);
5957
5958                 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5959         }
5960
5961         if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5962                      res.f6i == net->ipv6.fib6_null_entry))
5963                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5964
5965         switch (res.fib6_type) {
5966         /* only unicast is forwarded */
5967         case RTN_UNICAST:
5968                 break;
5969         case RTN_BLACKHOLE:
5970                 return BPF_FIB_LKUP_RET_BLACKHOLE;
5971         case RTN_UNREACHABLE:
5972                 return BPF_FIB_LKUP_RET_UNREACHABLE;
5973         case RTN_PROHIBIT:
5974                 return BPF_FIB_LKUP_RET_PROHIBIT;
5975         default:
5976                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5977         }
5978
5979         ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5980                                     fl6.flowi6_oif != 0, NULL, strict);
5981
5982         if (check_mtu) {
5983                 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5984                 if (params->tot_len > mtu) {
5985                         params->mtu_result = mtu; /* union with tot_len */
5986                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5987                 }
5988         }
5989
5990         if (res.nh->fib_nh_lws)
5991                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5992
5993         if (res.nh->fib_nh_gw_family)
5994                 *dst = res.nh->fib_nh_gw6;
5995
5996         dev = res.nh->fib_nh_dev;
5997         params->rt_metric = res.f6i->fib6_metric;
5998         params->ifindex = dev->ifindex;
5999
6000         if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH)
6001                 goto set_fwd_params;
6002
6003         /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
6004          * not needed here.
6005          */
6006         neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
6007         if (!neigh || !(READ_ONCE(neigh->nud_state) & NUD_VALID))
6008                 return BPF_FIB_LKUP_RET_NO_NEIGH;
6009         memcpy(params->dmac, neigh->ha, ETH_ALEN);
6010         memcpy(params->smac, dev->dev_addr, ETH_ALEN);
6011
6012 set_fwd_params:
6013         return bpf_fib_set_fwd_params(params, mtu);
6014 }
6015 #endif
6016
6017 #define BPF_FIB_LOOKUP_MASK (BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT | \
6018                              BPF_FIB_LOOKUP_SKIP_NEIGH | BPF_FIB_LOOKUP_TBID)
6019
6020 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
6021            struct bpf_fib_lookup *, params, int, plen, u32, flags)
6022 {
6023         if (plen < sizeof(*params))
6024                 return -EINVAL;
6025
6026         if (flags & ~BPF_FIB_LOOKUP_MASK)
6027                 return -EINVAL;
6028
6029         switch (params->family) {
6030 #if IS_ENABLED(CONFIG_INET)
6031         case AF_INET:
6032                 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
6033                                            flags, true);
6034 #endif
6035 #if IS_ENABLED(CONFIG_IPV6)
6036         case AF_INET6:
6037                 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
6038                                            flags, true);
6039 #endif
6040         }
6041         return -EAFNOSUPPORT;
6042 }
6043
6044 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
6045         .func           = bpf_xdp_fib_lookup,
6046         .gpl_only       = true,
6047         .ret_type       = RET_INTEGER,
6048         .arg1_type      = ARG_PTR_TO_CTX,
6049         .arg2_type      = ARG_PTR_TO_MEM,
6050         .arg3_type      = ARG_CONST_SIZE,
6051         .arg4_type      = ARG_ANYTHING,
6052 };
6053
6054 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
6055            struct bpf_fib_lookup *, params, int, plen, u32, flags)
6056 {
6057         struct net *net = dev_net(skb->dev);
6058         int rc = -EAFNOSUPPORT;
6059         bool check_mtu = false;
6060
6061         if (plen < sizeof(*params))
6062                 return -EINVAL;
6063
6064         if (flags & ~BPF_FIB_LOOKUP_MASK)
6065                 return -EINVAL;
6066
6067         if (params->tot_len)
6068                 check_mtu = true;
6069
6070         switch (params->family) {
6071 #if IS_ENABLED(CONFIG_INET)
6072         case AF_INET:
6073                 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
6074                 break;
6075 #endif
6076 #if IS_ENABLED(CONFIG_IPV6)
6077         case AF_INET6:
6078                 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
6079                 break;
6080 #endif
6081         }
6082
6083         if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
6084                 struct net_device *dev;
6085
6086                 /* When tot_len isn't provided by user, check skb
6087                  * against MTU of FIB lookup resulting net_device
6088                  */
6089                 dev = dev_get_by_index_rcu(net, params->ifindex);
6090                 if (!is_skb_forwardable(dev, skb))
6091                         rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
6092
6093                 params->mtu_result = dev->mtu; /* union with tot_len */
6094         }
6095
6096         return rc;
6097 }
6098
6099 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
6100         .func           = bpf_skb_fib_lookup,
6101         .gpl_only       = true,
6102         .ret_type       = RET_INTEGER,
6103         .arg1_type      = ARG_PTR_TO_CTX,
6104         .arg2_type      = ARG_PTR_TO_MEM,
6105         .arg3_type      = ARG_CONST_SIZE,
6106         .arg4_type      = ARG_ANYTHING,
6107 };
6108
6109 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
6110                                             u32 ifindex)
6111 {
6112         struct net *netns = dev_net(dev_curr);
6113
6114         /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
6115         if (ifindex == 0)
6116                 return dev_curr;
6117
6118         return dev_get_by_index_rcu(netns, ifindex);
6119 }
6120
6121 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
6122            u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6123 {
6124         int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6125         struct net_device *dev = skb->dev;
6126         int skb_len, dev_len;
6127         int mtu;
6128
6129         if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
6130                 return -EINVAL;
6131
6132         if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
6133                 return -EINVAL;
6134
6135         dev = __dev_via_ifindex(dev, ifindex);
6136         if (unlikely(!dev))
6137                 return -ENODEV;
6138
6139         mtu = READ_ONCE(dev->mtu);
6140
6141         dev_len = mtu + dev->hard_header_len;
6142
6143         /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6144         skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
6145
6146         skb_len += len_diff; /* minus result pass check */
6147         if (skb_len <= dev_len) {
6148                 ret = BPF_MTU_CHK_RET_SUCCESS;
6149                 goto out;
6150         }
6151         /* At this point, skb->len exceed MTU, but as it include length of all
6152          * segments, it can still be below MTU.  The SKB can possibly get
6153          * re-segmented in transmit path (see validate_xmit_skb).  Thus, user
6154          * must choose if segs are to be MTU checked.
6155          */
6156         if (skb_is_gso(skb)) {
6157                 ret = BPF_MTU_CHK_RET_SUCCESS;
6158
6159                 if (flags & BPF_MTU_CHK_SEGS &&
6160                     !skb_gso_validate_network_len(skb, mtu))
6161                         ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
6162         }
6163 out:
6164         /* BPF verifier guarantees valid pointer */
6165         *mtu_len = mtu;
6166
6167         return ret;
6168 }
6169
6170 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
6171            u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6172 {
6173         struct net_device *dev = xdp->rxq->dev;
6174         int xdp_len = xdp->data_end - xdp->data;
6175         int ret = BPF_MTU_CHK_RET_SUCCESS;
6176         int mtu, dev_len;
6177
6178         /* XDP variant doesn't support multi-buffer segment check (yet) */
6179         if (unlikely(flags))
6180                 return -EINVAL;
6181
6182         dev = __dev_via_ifindex(dev, ifindex);
6183         if (unlikely(!dev))
6184                 return -ENODEV;
6185
6186         mtu = READ_ONCE(dev->mtu);
6187
6188         /* Add L2-header as dev MTU is L3 size */
6189         dev_len = mtu + dev->hard_header_len;
6190
6191         /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6192         if (*mtu_len)
6193                 xdp_len = *mtu_len + dev->hard_header_len;
6194
6195         xdp_len += len_diff; /* minus result pass check */
6196         if (xdp_len > dev_len)
6197                 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6198
6199         /* BPF verifier guarantees valid pointer */
6200         *mtu_len = mtu;
6201
6202         return ret;
6203 }
6204
6205 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
6206         .func           = bpf_skb_check_mtu,
6207         .gpl_only       = true,
6208         .ret_type       = RET_INTEGER,
6209         .arg1_type      = ARG_PTR_TO_CTX,
6210         .arg2_type      = ARG_ANYTHING,
6211         .arg3_type      = ARG_PTR_TO_INT,
6212         .arg4_type      = ARG_ANYTHING,
6213         .arg5_type      = ARG_ANYTHING,
6214 };
6215
6216 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
6217         .func           = bpf_xdp_check_mtu,
6218         .gpl_only       = true,
6219         .ret_type       = RET_INTEGER,
6220         .arg1_type      = ARG_PTR_TO_CTX,
6221         .arg2_type      = ARG_ANYTHING,
6222         .arg3_type      = ARG_PTR_TO_INT,
6223         .arg4_type      = ARG_ANYTHING,
6224         .arg5_type      = ARG_ANYTHING,
6225 };
6226
6227 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6228 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
6229 {
6230         int err;
6231         struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
6232
6233         if (!seg6_validate_srh(srh, len, false))
6234                 return -EINVAL;
6235
6236         switch (type) {
6237         case BPF_LWT_ENCAP_SEG6_INLINE:
6238                 if (skb->protocol != htons(ETH_P_IPV6))
6239                         return -EBADMSG;
6240
6241                 err = seg6_do_srh_inline(skb, srh);
6242                 break;
6243         case BPF_LWT_ENCAP_SEG6:
6244                 skb_reset_inner_headers(skb);
6245                 skb->encapsulation = 1;
6246                 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
6247                 break;
6248         default:
6249                 return -EINVAL;
6250         }
6251
6252         bpf_compute_data_pointers(skb);
6253         if (err)
6254                 return err;
6255
6256         skb_set_transport_header(skb, sizeof(struct ipv6hdr));
6257
6258         return seg6_lookup_nexthop(skb, NULL, 0);
6259 }
6260 #endif /* CONFIG_IPV6_SEG6_BPF */
6261
6262 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6263 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
6264                              bool ingress)
6265 {
6266         return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
6267 }
6268 #endif
6269
6270 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
6271            u32, len)
6272 {
6273         switch (type) {
6274 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6275         case BPF_LWT_ENCAP_SEG6:
6276         case BPF_LWT_ENCAP_SEG6_INLINE:
6277                 return bpf_push_seg6_encap(skb, type, hdr, len);
6278 #endif
6279 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6280         case BPF_LWT_ENCAP_IP:
6281                 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
6282 #endif
6283         default:
6284                 return -EINVAL;
6285         }
6286 }
6287
6288 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
6289            void *, hdr, u32, len)
6290 {
6291         switch (type) {
6292 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6293         case BPF_LWT_ENCAP_IP:
6294                 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
6295 #endif
6296         default:
6297                 return -EINVAL;
6298         }
6299 }
6300
6301 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
6302         .func           = bpf_lwt_in_push_encap,
6303         .gpl_only       = false,
6304         .ret_type       = RET_INTEGER,
6305         .arg1_type      = ARG_PTR_TO_CTX,
6306         .arg2_type      = ARG_ANYTHING,
6307         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6308         .arg4_type      = ARG_CONST_SIZE
6309 };
6310
6311 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
6312         .func           = bpf_lwt_xmit_push_encap,
6313         .gpl_only       = false,
6314         .ret_type       = RET_INTEGER,
6315         .arg1_type      = ARG_PTR_TO_CTX,
6316         .arg2_type      = ARG_ANYTHING,
6317         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6318         .arg4_type      = ARG_CONST_SIZE
6319 };
6320
6321 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6322 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
6323            const void *, from, u32, len)
6324 {
6325         struct seg6_bpf_srh_state *srh_state =
6326                 this_cpu_ptr(&seg6_bpf_srh_states);
6327         struct ipv6_sr_hdr *srh = srh_state->srh;
6328         void *srh_tlvs, *srh_end, *ptr;
6329         int srhoff = 0;
6330
6331         if (srh == NULL)
6332                 return -EINVAL;
6333
6334         srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
6335         srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
6336
6337         ptr = skb->data + offset;
6338         if (ptr >= srh_tlvs && ptr + len <= srh_end)
6339                 srh_state->valid = false;
6340         else if (ptr < (void *)&srh->flags ||
6341                  ptr + len > (void *)&srh->segments)
6342                 return -EFAULT;
6343
6344         if (unlikely(bpf_try_make_writable(skb, offset + len)))
6345                 return -EFAULT;
6346         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6347                 return -EINVAL;
6348         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6349
6350         memcpy(skb->data + offset, from, len);
6351         return 0;
6352 }
6353
6354 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
6355         .func           = bpf_lwt_seg6_store_bytes,
6356         .gpl_only       = false,
6357         .ret_type       = RET_INTEGER,
6358         .arg1_type      = ARG_PTR_TO_CTX,
6359         .arg2_type      = ARG_ANYTHING,
6360         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6361         .arg4_type      = ARG_CONST_SIZE
6362 };
6363
6364 static void bpf_update_srh_state(struct sk_buff *skb)
6365 {
6366         struct seg6_bpf_srh_state *srh_state =
6367                 this_cpu_ptr(&seg6_bpf_srh_states);
6368         int srhoff = 0;
6369
6370         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
6371                 srh_state->srh = NULL;
6372         } else {
6373                 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6374                 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
6375                 srh_state->valid = true;
6376         }
6377 }
6378
6379 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
6380            u32, action, void *, param, u32, param_len)
6381 {
6382         struct seg6_bpf_srh_state *srh_state =
6383                 this_cpu_ptr(&seg6_bpf_srh_states);
6384         int hdroff = 0;
6385         int err;
6386
6387         switch (action) {
6388         case SEG6_LOCAL_ACTION_END_X:
6389                 if (!seg6_bpf_has_valid_srh(skb))
6390                         return -EBADMSG;
6391                 if (param_len != sizeof(struct in6_addr))
6392                         return -EINVAL;
6393                 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
6394         case SEG6_LOCAL_ACTION_END_T:
6395                 if (!seg6_bpf_has_valid_srh(skb))
6396                         return -EBADMSG;
6397                 if (param_len != sizeof(int))
6398                         return -EINVAL;
6399                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6400         case SEG6_LOCAL_ACTION_END_DT6:
6401                 if (!seg6_bpf_has_valid_srh(skb))
6402                         return -EBADMSG;
6403                 if (param_len != sizeof(int))
6404                         return -EINVAL;
6405
6406                 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
6407                         return -EBADMSG;
6408                 if (!pskb_pull(skb, hdroff))
6409                         return -EBADMSG;
6410
6411                 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
6412                 skb_reset_network_header(skb);
6413                 skb_reset_transport_header(skb);
6414                 skb->encapsulation = 0;
6415
6416                 bpf_compute_data_pointers(skb);
6417                 bpf_update_srh_state(skb);
6418                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6419         case SEG6_LOCAL_ACTION_END_B6:
6420                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6421                         return -EBADMSG;
6422                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
6423                                           param, param_len);
6424                 if (!err)
6425                         bpf_update_srh_state(skb);
6426
6427                 return err;
6428         case SEG6_LOCAL_ACTION_END_B6_ENCAP:
6429                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6430                         return -EBADMSG;
6431                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
6432                                           param, param_len);
6433                 if (!err)
6434                         bpf_update_srh_state(skb);
6435
6436                 return err;
6437         default:
6438                 return -EINVAL;
6439         }
6440 }
6441
6442 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
6443         .func           = bpf_lwt_seg6_action,
6444         .gpl_only       = false,
6445         .ret_type       = RET_INTEGER,
6446         .arg1_type      = ARG_PTR_TO_CTX,
6447         .arg2_type      = ARG_ANYTHING,
6448         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6449         .arg4_type      = ARG_CONST_SIZE
6450 };
6451
6452 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
6453            s32, len)
6454 {
6455         struct seg6_bpf_srh_state *srh_state =
6456                 this_cpu_ptr(&seg6_bpf_srh_states);
6457         struct ipv6_sr_hdr *srh = srh_state->srh;
6458         void *srh_end, *srh_tlvs, *ptr;
6459         struct ipv6hdr *hdr;
6460         int srhoff = 0;
6461         int ret;
6462
6463         if (unlikely(srh == NULL))
6464                 return -EINVAL;
6465
6466         srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
6467                         ((srh->first_segment + 1) << 4));
6468         srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
6469                         srh_state->hdrlen);
6470         ptr = skb->data + offset;
6471
6472         if (unlikely(ptr < srh_tlvs || ptr > srh_end))
6473                 return -EFAULT;
6474         if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
6475                 return -EFAULT;
6476
6477         if (len > 0) {
6478                 ret = skb_cow_head(skb, len);
6479                 if (unlikely(ret < 0))
6480                         return ret;
6481
6482                 ret = bpf_skb_net_hdr_push(skb, offset, len);
6483         } else {
6484                 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6485         }
6486
6487         bpf_compute_data_pointers(skb);
6488         if (unlikely(ret < 0))
6489                 return ret;
6490
6491         hdr = (struct ipv6hdr *)skb->data;
6492         hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6493
6494         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6495                 return -EINVAL;
6496         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6497         srh_state->hdrlen += len;
6498         srh_state->valid = false;
6499         return 0;
6500 }
6501
6502 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6503         .func           = bpf_lwt_seg6_adjust_srh,
6504         .gpl_only       = false,
6505         .ret_type       = RET_INTEGER,
6506         .arg1_type      = ARG_PTR_TO_CTX,
6507         .arg2_type      = ARG_ANYTHING,
6508         .arg3_type      = ARG_ANYTHING,
6509 };
6510 #endif /* CONFIG_IPV6_SEG6_BPF */
6511
6512 #ifdef CONFIG_INET
6513 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6514                               int dif, int sdif, u8 family, u8 proto)
6515 {
6516         struct inet_hashinfo *hinfo = net->ipv4.tcp_death_row.hashinfo;
6517         bool refcounted = false;
6518         struct sock *sk = NULL;
6519
6520         if (family == AF_INET) {
6521                 __be32 src4 = tuple->ipv4.saddr;
6522                 __be32 dst4 = tuple->ipv4.daddr;
6523
6524                 if (proto == IPPROTO_TCP)
6525                         sk = __inet_lookup(net, hinfo, NULL, 0,
6526                                            src4, tuple->ipv4.sport,
6527                                            dst4, tuple->ipv4.dport,
6528                                            dif, sdif, &refcounted);
6529                 else
6530                         sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6531                                                dst4, tuple->ipv4.dport,
6532                                                dif, sdif, net->ipv4.udp_table, NULL);
6533 #if IS_ENABLED(CONFIG_IPV6)
6534         } else {
6535                 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6536                 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6537
6538                 if (proto == IPPROTO_TCP)
6539                         sk = __inet6_lookup(net, hinfo, NULL, 0,
6540                                             src6, tuple->ipv6.sport,
6541                                             dst6, ntohs(tuple->ipv6.dport),
6542                                             dif, sdif, &refcounted);
6543                 else if (likely(ipv6_bpf_stub))
6544                         sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6545                                                             src6, tuple->ipv6.sport,
6546                                                             dst6, tuple->ipv6.dport,
6547                                                             dif, sdif,
6548                                                             net->ipv4.udp_table, NULL);
6549 #endif
6550         }
6551
6552         if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6553                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6554                 sk = NULL;
6555         }
6556         return sk;
6557 }
6558
6559 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6560  * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6561  */
6562 static struct sock *
6563 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6564                  struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6565                  u64 flags, int sdif)
6566 {
6567         struct sock *sk = NULL;
6568         struct net *net;
6569         u8 family;
6570
6571         if (len == sizeof(tuple->ipv4))
6572                 family = AF_INET;
6573         else if (len == sizeof(tuple->ipv6))
6574                 family = AF_INET6;
6575         else
6576                 return NULL;
6577
6578         if (unlikely(flags || !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6579                 goto out;
6580
6581         if (sdif < 0) {
6582                 if (family == AF_INET)
6583                         sdif = inet_sdif(skb);
6584                 else
6585                         sdif = inet6_sdif(skb);
6586         }
6587
6588         if ((s32)netns_id < 0) {
6589                 net = caller_net;
6590                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6591         } else {
6592                 net = get_net_ns_by_id(caller_net, netns_id);
6593                 if (unlikely(!net))
6594                         goto out;
6595                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6596                 put_net(net);
6597         }
6598
6599 out:
6600         return sk;
6601 }
6602
6603 static struct sock *
6604 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6605                 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6606                 u64 flags, int sdif)
6607 {
6608         struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6609                                            ifindex, proto, netns_id, flags,
6610                                            sdif);
6611
6612         if (sk) {
6613                 struct sock *sk2 = sk_to_full_sk(sk);
6614
6615                 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6616                  * sock refcnt is decremented to prevent a request_sock leak.
6617                  */
6618                 if (!sk_fullsock(sk2))
6619                         sk2 = NULL;
6620                 if (sk2 != sk) {
6621                         sock_gen_put(sk);
6622                         /* Ensure there is no need to bump sk2 refcnt */
6623                         if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6624                                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6625                                 return NULL;
6626                         }
6627                         sk = sk2;
6628                 }
6629         }
6630
6631         return sk;
6632 }
6633
6634 static struct sock *
6635 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6636                u8 proto, u64 netns_id, u64 flags)
6637 {
6638         struct net *caller_net;
6639         int ifindex;
6640
6641         if (skb->dev) {
6642                 caller_net = dev_net(skb->dev);
6643                 ifindex = skb->dev->ifindex;
6644         } else {
6645                 caller_net = sock_net(skb->sk);
6646                 ifindex = 0;
6647         }
6648
6649         return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6650                                 netns_id, flags, -1);
6651 }
6652
6653 static struct sock *
6654 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6655               u8 proto, u64 netns_id, u64 flags)
6656 {
6657         struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6658                                          flags);
6659
6660         if (sk) {
6661                 struct sock *sk2 = sk_to_full_sk(sk);
6662
6663                 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6664                  * sock refcnt is decremented to prevent a request_sock leak.
6665                  */
6666                 if (!sk_fullsock(sk2))
6667                         sk2 = NULL;
6668                 if (sk2 != sk) {
6669                         sock_gen_put(sk);
6670                         /* Ensure there is no need to bump sk2 refcnt */
6671                         if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6672                                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6673                                 return NULL;
6674                         }
6675                         sk = sk2;
6676                 }
6677         }
6678
6679         return sk;
6680 }
6681
6682 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6683            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6684 {
6685         return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6686                                              netns_id, flags);
6687 }
6688
6689 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6690         .func           = bpf_skc_lookup_tcp,
6691         .gpl_only       = false,
6692         .pkt_access     = true,
6693         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6694         .arg1_type      = ARG_PTR_TO_CTX,
6695         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6696         .arg3_type      = ARG_CONST_SIZE,
6697         .arg4_type      = ARG_ANYTHING,
6698         .arg5_type      = ARG_ANYTHING,
6699 };
6700
6701 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6702            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6703 {
6704         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6705                                             netns_id, flags);
6706 }
6707
6708 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6709         .func           = bpf_sk_lookup_tcp,
6710         .gpl_only       = false,
6711         .pkt_access     = true,
6712         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6713         .arg1_type      = ARG_PTR_TO_CTX,
6714         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6715         .arg3_type      = ARG_CONST_SIZE,
6716         .arg4_type      = ARG_ANYTHING,
6717         .arg5_type      = ARG_ANYTHING,
6718 };
6719
6720 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6721            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6722 {
6723         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6724                                             netns_id, flags);
6725 }
6726
6727 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6728         .func           = bpf_sk_lookup_udp,
6729         .gpl_only       = false,
6730         .pkt_access     = true,
6731         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6732         .arg1_type      = ARG_PTR_TO_CTX,
6733         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6734         .arg3_type      = ARG_CONST_SIZE,
6735         .arg4_type      = ARG_ANYTHING,
6736         .arg5_type      = ARG_ANYTHING,
6737 };
6738
6739 BPF_CALL_5(bpf_tc_skc_lookup_tcp, struct sk_buff *, skb,
6740            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6741 {
6742         struct net_device *dev = skb->dev;
6743         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6744         struct net *caller_net = dev_net(dev);
6745
6746         return (unsigned long)__bpf_skc_lookup(skb, tuple, len, caller_net,
6747                                                ifindex, IPPROTO_TCP, netns_id,
6748                                                flags, sdif);
6749 }
6750
6751 static const struct bpf_func_proto bpf_tc_skc_lookup_tcp_proto = {
6752         .func           = bpf_tc_skc_lookup_tcp,
6753         .gpl_only       = false,
6754         .pkt_access     = true,
6755         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6756         .arg1_type      = ARG_PTR_TO_CTX,
6757         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6758         .arg3_type      = ARG_CONST_SIZE,
6759         .arg4_type      = ARG_ANYTHING,
6760         .arg5_type      = ARG_ANYTHING,
6761 };
6762
6763 BPF_CALL_5(bpf_tc_sk_lookup_tcp, struct sk_buff *, skb,
6764            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6765 {
6766         struct net_device *dev = skb->dev;
6767         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6768         struct net *caller_net = dev_net(dev);
6769
6770         return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6771                                               ifindex, IPPROTO_TCP, netns_id,
6772                                               flags, sdif);
6773 }
6774
6775 static const struct bpf_func_proto bpf_tc_sk_lookup_tcp_proto = {
6776         .func           = bpf_tc_sk_lookup_tcp,
6777         .gpl_only       = false,
6778         .pkt_access     = true,
6779         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6780         .arg1_type      = ARG_PTR_TO_CTX,
6781         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6782         .arg3_type      = ARG_CONST_SIZE,
6783         .arg4_type      = ARG_ANYTHING,
6784         .arg5_type      = ARG_ANYTHING,
6785 };
6786
6787 BPF_CALL_5(bpf_tc_sk_lookup_udp, struct sk_buff *, skb,
6788            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6789 {
6790         struct net_device *dev = skb->dev;
6791         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6792         struct net *caller_net = dev_net(dev);
6793
6794         return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6795                                               ifindex, IPPROTO_UDP, netns_id,
6796                                               flags, sdif);
6797 }
6798
6799 static const struct bpf_func_proto bpf_tc_sk_lookup_udp_proto = {
6800         .func           = bpf_tc_sk_lookup_udp,
6801         .gpl_only       = false,
6802         .pkt_access     = true,
6803         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6804         .arg1_type      = ARG_PTR_TO_CTX,
6805         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6806         .arg3_type      = ARG_CONST_SIZE,
6807         .arg4_type      = ARG_ANYTHING,
6808         .arg5_type      = ARG_ANYTHING,
6809 };
6810
6811 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6812 {
6813         if (sk && sk_is_refcounted(sk))
6814                 sock_gen_put(sk);
6815         return 0;
6816 }
6817
6818 static const struct bpf_func_proto bpf_sk_release_proto = {
6819         .func           = bpf_sk_release,
6820         .gpl_only       = false,
6821         .ret_type       = RET_INTEGER,
6822         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON | OBJ_RELEASE,
6823 };
6824
6825 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6826            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6827 {
6828         struct net_device *dev = ctx->rxq->dev;
6829         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6830         struct net *caller_net = dev_net(dev);
6831
6832         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6833                                               ifindex, IPPROTO_UDP, netns_id,
6834                                               flags, sdif);
6835 }
6836
6837 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6838         .func           = bpf_xdp_sk_lookup_udp,
6839         .gpl_only       = false,
6840         .pkt_access     = true,
6841         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6842         .arg1_type      = ARG_PTR_TO_CTX,
6843         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6844         .arg3_type      = ARG_CONST_SIZE,
6845         .arg4_type      = ARG_ANYTHING,
6846         .arg5_type      = ARG_ANYTHING,
6847 };
6848
6849 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6850            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6851 {
6852         struct net_device *dev = ctx->rxq->dev;
6853         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6854         struct net *caller_net = dev_net(dev);
6855
6856         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6857                                                ifindex, IPPROTO_TCP, netns_id,
6858                                                flags, sdif);
6859 }
6860
6861 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6862         .func           = bpf_xdp_skc_lookup_tcp,
6863         .gpl_only       = false,
6864         .pkt_access     = true,
6865         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6866         .arg1_type      = ARG_PTR_TO_CTX,
6867         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6868         .arg3_type      = ARG_CONST_SIZE,
6869         .arg4_type      = ARG_ANYTHING,
6870         .arg5_type      = ARG_ANYTHING,
6871 };
6872
6873 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6874            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6875 {
6876         struct net_device *dev = ctx->rxq->dev;
6877         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6878         struct net *caller_net = dev_net(dev);
6879
6880         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6881                                               ifindex, IPPROTO_TCP, netns_id,
6882                                               flags, sdif);
6883 }
6884
6885 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6886         .func           = bpf_xdp_sk_lookup_tcp,
6887         .gpl_only       = false,
6888         .pkt_access     = true,
6889         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6890         .arg1_type      = ARG_PTR_TO_CTX,
6891         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6892         .arg3_type      = ARG_CONST_SIZE,
6893         .arg4_type      = ARG_ANYTHING,
6894         .arg5_type      = ARG_ANYTHING,
6895 };
6896
6897 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6898            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6899 {
6900         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6901                                                sock_net(ctx->sk), 0,
6902                                                IPPROTO_TCP, netns_id, flags,
6903                                                -1);
6904 }
6905
6906 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6907         .func           = bpf_sock_addr_skc_lookup_tcp,
6908         .gpl_only       = false,
6909         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6910         .arg1_type      = ARG_PTR_TO_CTX,
6911         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6912         .arg3_type      = ARG_CONST_SIZE,
6913         .arg4_type      = ARG_ANYTHING,
6914         .arg5_type      = ARG_ANYTHING,
6915 };
6916
6917 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6918            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6919 {
6920         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6921                                               sock_net(ctx->sk), 0, IPPROTO_TCP,
6922                                               netns_id, flags, -1);
6923 }
6924
6925 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6926         .func           = bpf_sock_addr_sk_lookup_tcp,
6927         .gpl_only       = false,
6928         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6929         .arg1_type      = ARG_PTR_TO_CTX,
6930         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6931         .arg3_type      = ARG_CONST_SIZE,
6932         .arg4_type      = ARG_ANYTHING,
6933         .arg5_type      = ARG_ANYTHING,
6934 };
6935
6936 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6937            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6938 {
6939         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6940                                               sock_net(ctx->sk), 0, IPPROTO_UDP,
6941                                               netns_id, flags, -1);
6942 }
6943
6944 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6945         .func           = bpf_sock_addr_sk_lookup_udp,
6946         .gpl_only       = false,
6947         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6948         .arg1_type      = ARG_PTR_TO_CTX,
6949         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6950         .arg3_type      = ARG_CONST_SIZE,
6951         .arg4_type      = ARG_ANYTHING,
6952         .arg5_type      = ARG_ANYTHING,
6953 };
6954
6955 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6956                                   struct bpf_insn_access_aux *info)
6957 {
6958         if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6959                                           icsk_retransmits))
6960                 return false;
6961
6962         if (off % size != 0)
6963                 return false;
6964
6965         switch (off) {
6966         case offsetof(struct bpf_tcp_sock, bytes_received):
6967         case offsetof(struct bpf_tcp_sock, bytes_acked):
6968                 return size == sizeof(__u64);
6969         default:
6970                 return size == sizeof(__u32);
6971         }
6972 }
6973
6974 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6975                                     const struct bpf_insn *si,
6976                                     struct bpf_insn *insn_buf,
6977                                     struct bpf_prog *prog, u32 *target_size)
6978 {
6979         struct bpf_insn *insn = insn_buf;
6980
6981 #define BPF_TCP_SOCK_GET_COMMON(FIELD)                                  \
6982         do {                                                            \
6983                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) >     \
6984                              sizeof_field(struct bpf_tcp_sock, FIELD)); \
6985                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6986                                       si->dst_reg, si->src_reg,         \
6987                                       offsetof(struct tcp_sock, FIELD)); \
6988         } while (0)
6989
6990 #define BPF_INET_SOCK_GET_COMMON(FIELD)                                 \
6991         do {                                                            \
6992                 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock,  \
6993                                           FIELD) >                      \
6994                              sizeof_field(struct bpf_tcp_sock, FIELD)); \
6995                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                 \
6996                                         struct inet_connection_sock,    \
6997                                         FIELD),                         \
6998                                       si->dst_reg, si->src_reg,         \
6999                                       offsetof(                         \
7000                                         struct inet_connection_sock,    \
7001                                         FIELD));                        \
7002         } while (0)
7003
7004         BTF_TYPE_EMIT(struct bpf_tcp_sock);
7005
7006         switch (si->off) {
7007         case offsetof(struct bpf_tcp_sock, rtt_min):
7008                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
7009                              sizeof(struct minmax));
7010                 BUILD_BUG_ON(sizeof(struct minmax) <
7011                              sizeof(struct minmax_sample));
7012
7013                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7014                                       offsetof(struct tcp_sock, rtt_min) +
7015                                       offsetof(struct minmax_sample, v));
7016                 break;
7017         case offsetof(struct bpf_tcp_sock, snd_cwnd):
7018                 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
7019                 break;
7020         case offsetof(struct bpf_tcp_sock, srtt_us):
7021                 BPF_TCP_SOCK_GET_COMMON(srtt_us);
7022                 break;
7023         case offsetof(struct bpf_tcp_sock, snd_ssthresh):
7024                 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
7025                 break;
7026         case offsetof(struct bpf_tcp_sock, rcv_nxt):
7027                 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
7028                 break;
7029         case offsetof(struct bpf_tcp_sock, snd_nxt):
7030                 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
7031                 break;
7032         case offsetof(struct bpf_tcp_sock, snd_una):
7033                 BPF_TCP_SOCK_GET_COMMON(snd_una);
7034                 break;
7035         case offsetof(struct bpf_tcp_sock, mss_cache):
7036                 BPF_TCP_SOCK_GET_COMMON(mss_cache);
7037                 break;
7038         case offsetof(struct bpf_tcp_sock, ecn_flags):
7039                 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
7040                 break;
7041         case offsetof(struct bpf_tcp_sock, rate_delivered):
7042                 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
7043                 break;
7044         case offsetof(struct bpf_tcp_sock, rate_interval_us):
7045                 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
7046                 break;
7047         case offsetof(struct bpf_tcp_sock, packets_out):
7048                 BPF_TCP_SOCK_GET_COMMON(packets_out);
7049                 break;
7050         case offsetof(struct bpf_tcp_sock, retrans_out):
7051                 BPF_TCP_SOCK_GET_COMMON(retrans_out);
7052                 break;
7053         case offsetof(struct bpf_tcp_sock, total_retrans):
7054                 BPF_TCP_SOCK_GET_COMMON(total_retrans);
7055                 break;
7056         case offsetof(struct bpf_tcp_sock, segs_in):
7057                 BPF_TCP_SOCK_GET_COMMON(segs_in);
7058                 break;
7059         case offsetof(struct bpf_tcp_sock, data_segs_in):
7060                 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
7061                 break;
7062         case offsetof(struct bpf_tcp_sock, segs_out):
7063                 BPF_TCP_SOCK_GET_COMMON(segs_out);
7064                 break;
7065         case offsetof(struct bpf_tcp_sock, data_segs_out):
7066                 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
7067                 break;
7068         case offsetof(struct bpf_tcp_sock, lost_out):
7069                 BPF_TCP_SOCK_GET_COMMON(lost_out);
7070                 break;
7071         case offsetof(struct bpf_tcp_sock, sacked_out):
7072                 BPF_TCP_SOCK_GET_COMMON(sacked_out);
7073                 break;
7074         case offsetof(struct bpf_tcp_sock, bytes_received):
7075                 BPF_TCP_SOCK_GET_COMMON(bytes_received);
7076                 break;
7077         case offsetof(struct bpf_tcp_sock, bytes_acked):
7078                 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
7079                 break;
7080         case offsetof(struct bpf_tcp_sock, dsack_dups):
7081                 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
7082                 break;
7083         case offsetof(struct bpf_tcp_sock, delivered):
7084                 BPF_TCP_SOCK_GET_COMMON(delivered);
7085                 break;
7086         case offsetof(struct bpf_tcp_sock, delivered_ce):
7087                 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
7088                 break;
7089         case offsetof(struct bpf_tcp_sock, icsk_retransmits):
7090                 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
7091                 break;
7092         }
7093
7094         return insn - insn_buf;
7095 }
7096
7097 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
7098 {
7099         if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
7100                 return (unsigned long)sk;
7101
7102         return (unsigned long)NULL;
7103 }
7104
7105 const struct bpf_func_proto bpf_tcp_sock_proto = {
7106         .func           = bpf_tcp_sock,
7107         .gpl_only       = false,
7108         .ret_type       = RET_PTR_TO_TCP_SOCK_OR_NULL,
7109         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
7110 };
7111
7112 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
7113 {
7114         sk = sk_to_full_sk(sk);
7115
7116         if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
7117                 return (unsigned long)sk;
7118
7119         return (unsigned long)NULL;
7120 }
7121
7122 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
7123         .func           = bpf_get_listener_sock,
7124         .gpl_only       = false,
7125         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
7126         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
7127 };
7128
7129 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
7130 {
7131         unsigned int iphdr_len;
7132
7133         switch (skb_protocol(skb, true)) {
7134         case cpu_to_be16(ETH_P_IP):
7135                 iphdr_len = sizeof(struct iphdr);
7136                 break;
7137         case cpu_to_be16(ETH_P_IPV6):
7138                 iphdr_len = sizeof(struct ipv6hdr);
7139                 break;
7140         default:
7141                 return 0;
7142         }
7143
7144         if (skb_headlen(skb) < iphdr_len)
7145                 return 0;
7146
7147         if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
7148                 return 0;
7149
7150         return INET_ECN_set_ce(skb);
7151 }
7152
7153 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7154                                   struct bpf_insn_access_aux *info)
7155 {
7156         if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
7157                 return false;
7158
7159         if (off % size != 0)
7160                 return false;
7161
7162         switch (off) {
7163         default:
7164                 return size == sizeof(__u32);
7165         }
7166 }
7167
7168 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
7169                                     const struct bpf_insn *si,
7170                                     struct bpf_insn *insn_buf,
7171                                     struct bpf_prog *prog, u32 *target_size)
7172 {
7173         struct bpf_insn *insn = insn_buf;
7174
7175 #define BPF_XDP_SOCK_GET(FIELD)                                         \
7176         do {                                                            \
7177                 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) >     \
7178                              sizeof_field(struct bpf_xdp_sock, FIELD)); \
7179                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
7180                                       si->dst_reg, si->src_reg,         \
7181                                       offsetof(struct xdp_sock, FIELD)); \
7182         } while (0)
7183
7184         switch (si->off) {
7185         case offsetof(struct bpf_xdp_sock, queue_id):
7186                 BPF_XDP_SOCK_GET(queue_id);
7187                 break;
7188         }
7189
7190         return insn - insn_buf;
7191 }
7192
7193 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
7194         .func           = bpf_skb_ecn_set_ce,
7195         .gpl_only       = false,
7196         .ret_type       = RET_INTEGER,
7197         .arg1_type      = ARG_PTR_TO_CTX,
7198 };
7199
7200 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7201            struct tcphdr *, th, u32, th_len)
7202 {
7203 #ifdef CONFIG_SYN_COOKIES
7204         u32 cookie;
7205         int ret;
7206
7207         if (unlikely(!sk || th_len < sizeof(*th)))
7208                 return -EINVAL;
7209
7210         /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
7211         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7212                 return -EINVAL;
7213
7214         if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7215                 return -EINVAL;
7216
7217         if (!th->ack || th->rst || th->syn)
7218                 return -ENOENT;
7219
7220         if (unlikely(iph_len < sizeof(struct iphdr)))
7221                 return -EINVAL;
7222
7223         if (tcp_synq_no_recent_overflow(sk))
7224                 return -ENOENT;
7225
7226         cookie = ntohl(th->ack_seq) - 1;
7227
7228         /* Both struct iphdr and struct ipv6hdr have the version field at the
7229          * same offset so we can cast to the shorter header (struct iphdr).
7230          */
7231         switch (((struct iphdr *)iph)->version) {
7232         case 4:
7233                 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7234                         return -EINVAL;
7235
7236                 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
7237                 break;
7238
7239 #if IS_BUILTIN(CONFIG_IPV6)
7240         case 6:
7241                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7242                         return -EINVAL;
7243
7244                 if (sk->sk_family != AF_INET6)
7245                         return -EINVAL;
7246
7247                 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
7248                 break;
7249 #endif /* CONFIG_IPV6 */
7250
7251         default:
7252                 return -EPROTONOSUPPORT;
7253         }
7254
7255         if (ret > 0)
7256                 return 0;
7257
7258         return -ENOENT;
7259 #else
7260         return -ENOTSUPP;
7261 #endif
7262 }
7263
7264 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
7265         .func           = bpf_tcp_check_syncookie,
7266         .gpl_only       = true,
7267         .pkt_access     = true,
7268         .ret_type       = RET_INTEGER,
7269         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7270         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7271         .arg3_type      = ARG_CONST_SIZE,
7272         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7273         .arg5_type      = ARG_CONST_SIZE,
7274 };
7275
7276 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7277            struct tcphdr *, th, u32, th_len)
7278 {
7279 #ifdef CONFIG_SYN_COOKIES
7280         u32 cookie;
7281         u16 mss;
7282
7283         if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
7284                 return -EINVAL;
7285
7286         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7287                 return -EINVAL;
7288
7289         if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7290                 return -ENOENT;
7291
7292         if (!th->syn || th->ack || th->fin || th->rst)
7293                 return -EINVAL;
7294
7295         if (unlikely(iph_len < sizeof(struct iphdr)))
7296                 return -EINVAL;
7297
7298         /* Both struct iphdr and struct ipv6hdr have the version field at the
7299          * same offset so we can cast to the shorter header (struct iphdr).
7300          */
7301         switch (((struct iphdr *)iph)->version) {
7302         case 4:
7303                 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7304                         return -EINVAL;
7305
7306                 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
7307                 break;
7308
7309 #if IS_BUILTIN(CONFIG_IPV6)
7310         case 6:
7311                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7312                         return -EINVAL;
7313
7314                 if (sk->sk_family != AF_INET6)
7315                         return -EINVAL;
7316
7317                 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
7318                 break;
7319 #endif /* CONFIG_IPV6 */
7320
7321         default:
7322                 return -EPROTONOSUPPORT;
7323         }
7324         if (mss == 0)
7325                 return -ENOENT;
7326
7327         return cookie | ((u64)mss << 32);
7328 #else
7329         return -EOPNOTSUPP;
7330 #endif /* CONFIG_SYN_COOKIES */
7331 }
7332
7333 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
7334         .func           = bpf_tcp_gen_syncookie,
7335         .gpl_only       = true, /* __cookie_v*_init_sequence() is GPL */
7336         .pkt_access     = true,
7337         .ret_type       = RET_INTEGER,
7338         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7339         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7340         .arg3_type      = ARG_CONST_SIZE,
7341         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7342         .arg5_type      = ARG_CONST_SIZE,
7343 };
7344
7345 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
7346 {
7347         if (!sk || flags != 0)
7348                 return -EINVAL;
7349         if (!skb_at_tc_ingress(skb))
7350                 return -EOPNOTSUPP;
7351         if (unlikely(dev_net(skb->dev) != sock_net(sk)))
7352                 return -ENETUNREACH;
7353         if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
7354                 return -ESOCKTNOSUPPORT;
7355         if (sk_is_refcounted(sk) &&
7356             unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
7357                 return -ENOENT;
7358
7359         skb_orphan(skb);
7360         skb->sk = sk;
7361         skb->destructor = sock_pfree;
7362
7363         return 0;
7364 }
7365
7366 static const struct bpf_func_proto bpf_sk_assign_proto = {
7367         .func           = bpf_sk_assign,
7368         .gpl_only       = false,
7369         .ret_type       = RET_INTEGER,
7370         .arg1_type      = ARG_PTR_TO_CTX,
7371         .arg2_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7372         .arg3_type      = ARG_ANYTHING,
7373 };
7374
7375 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
7376                                     u8 search_kind, const u8 *magic,
7377                                     u8 magic_len, bool *eol)
7378 {
7379         u8 kind, kind_len;
7380
7381         *eol = false;
7382
7383         while (op < opend) {
7384                 kind = op[0];
7385
7386                 if (kind == TCPOPT_EOL) {
7387                         *eol = true;
7388                         return ERR_PTR(-ENOMSG);
7389                 } else if (kind == TCPOPT_NOP) {
7390                         op++;
7391                         continue;
7392                 }
7393
7394                 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
7395                         /* Something is wrong in the received header.
7396                          * Follow the TCP stack's tcp_parse_options()
7397                          * and just bail here.
7398                          */
7399                         return ERR_PTR(-EFAULT);
7400
7401                 kind_len = op[1];
7402                 if (search_kind == kind) {
7403                         if (!magic_len)
7404                                 return op;
7405
7406                         if (magic_len > kind_len - 2)
7407                                 return ERR_PTR(-ENOMSG);
7408
7409                         if (!memcmp(&op[2], magic, magic_len))
7410                                 return op;
7411                 }
7412
7413                 op += kind_len;
7414         }
7415
7416         return ERR_PTR(-ENOMSG);
7417 }
7418
7419 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7420            void *, search_res, u32, len, u64, flags)
7421 {
7422         bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
7423         const u8 *op, *opend, *magic, *search = search_res;
7424         u8 search_kind, search_len, copy_len, magic_len;
7425         int ret;
7426
7427         /* 2 byte is the minimal option len except TCPOPT_NOP and
7428          * TCPOPT_EOL which are useless for the bpf prog to learn
7429          * and this helper disallow loading them also.
7430          */
7431         if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
7432                 return -EINVAL;
7433
7434         search_kind = search[0];
7435         search_len = search[1];
7436
7437         if (search_len > len || search_kind == TCPOPT_NOP ||
7438             search_kind == TCPOPT_EOL)
7439                 return -EINVAL;
7440
7441         if (search_kind == TCPOPT_EXP || search_kind == 253) {
7442                 /* 16 or 32 bit magic.  +2 for kind and kind length */
7443                 if (search_len != 4 && search_len != 6)
7444                         return -EINVAL;
7445                 magic = &search[2];
7446                 magic_len = search_len - 2;
7447         } else {
7448                 if (search_len)
7449                         return -EINVAL;
7450                 magic = NULL;
7451                 magic_len = 0;
7452         }
7453
7454         if (load_syn) {
7455                 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
7456                 if (ret < 0)
7457                         return ret;
7458
7459                 opend = op + ret;
7460                 op += sizeof(struct tcphdr);
7461         } else {
7462                 if (!bpf_sock->skb ||
7463                     bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7464                         /* This bpf_sock->op cannot call this helper */
7465                         return -EPERM;
7466
7467                 opend = bpf_sock->skb_data_end;
7468                 op = bpf_sock->skb->data + sizeof(struct tcphdr);
7469         }
7470
7471         op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
7472                                 &eol);
7473         if (IS_ERR(op))
7474                 return PTR_ERR(op);
7475
7476         copy_len = op[1];
7477         ret = copy_len;
7478         if (copy_len > len) {
7479                 ret = -ENOSPC;
7480                 copy_len = len;
7481         }
7482
7483         memcpy(search_res, op, copy_len);
7484         return ret;
7485 }
7486
7487 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
7488         .func           = bpf_sock_ops_load_hdr_opt,
7489         .gpl_only       = false,
7490         .ret_type       = RET_INTEGER,
7491         .arg1_type      = ARG_PTR_TO_CTX,
7492         .arg2_type      = ARG_PTR_TO_MEM,
7493         .arg3_type      = ARG_CONST_SIZE,
7494         .arg4_type      = ARG_ANYTHING,
7495 };
7496
7497 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7498            const void *, from, u32, len, u64, flags)
7499 {
7500         u8 new_kind, new_kind_len, magic_len = 0, *opend;
7501         const u8 *op, *new_op, *magic = NULL;
7502         struct sk_buff *skb;
7503         bool eol;
7504
7505         if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
7506                 return -EPERM;
7507
7508         if (len < 2 || flags)
7509                 return -EINVAL;
7510
7511         new_op = from;
7512         new_kind = new_op[0];
7513         new_kind_len = new_op[1];
7514
7515         if (new_kind_len > len || new_kind == TCPOPT_NOP ||
7516             new_kind == TCPOPT_EOL)
7517                 return -EINVAL;
7518
7519         if (new_kind_len > bpf_sock->remaining_opt_len)
7520                 return -ENOSPC;
7521
7522         /* 253 is another experimental kind */
7523         if (new_kind == TCPOPT_EXP || new_kind == 253)  {
7524                 if (new_kind_len < 4)
7525                         return -EINVAL;
7526                 /* Match for the 2 byte magic also.
7527                  * RFC 6994: the magic could be 2 or 4 bytes.
7528                  * Hence, matching by 2 byte only is on the
7529                  * conservative side but it is the right
7530                  * thing to do for the 'search-for-duplication'
7531                  * purpose.
7532                  */
7533                 magic = &new_op[2];
7534                 magic_len = 2;
7535         }
7536
7537         /* Check for duplication */
7538         skb = bpf_sock->skb;
7539         op = skb->data + sizeof(struct tcphdr);
7540         opend = bpf_sock->skb_data_end;
7541
7542         op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
7543                                 &eol);
7544         if (!IS_ERR(op))
7545                 return -EEXIST;
7546
7547         if (PTR_ERR(op) != -ENOMSG)
7548                 return PTR_ERR(op);
7549
7550         if (eol)
7551                 /* The option has been ended.  Treat it as no more
7552                  * header option can be written.
7553                  */
7554                 return -ENOSPC;
7555
7556         /* No duplication found.  Store the header option. */
7557         memcpy(opend, from, new_kind_len);
7558
7559         bpf_sock->remaining_opt_len -= new_kind_len;
7560         bpf_sock->skb_data_end += new_kind_len;
7561
7562         return 0;
7563 }
7564
7565 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
7566         .func           = bpf_sock_ops_store_hdr_opt,
7567         .gpl_only       = false,
7568         .ret_type       = RET_INTEGER,
7569         .arg1_type      = ARG_PTR_TO_CTX,
7570         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7571         .arg3_type      = ARG_CONST_SIZE,
7572         .arg4_type      = ARG_ANYTHING,
7573 };
7574
7575 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7576            u32, len, u64, flags)
7577 {
7578         if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7579                 return -EPERM;
7580
7581         if (flags || len < 2)
7582                 return -EINVAL;
7583
7584         if (len > bpf_sock->remaining_opt_len)
7585                 return -ENOSPC;
7586
7587         bpf_sock->remaining_opt_len -= len;
7588
7589         return 0;
7590 }
7591
7592 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7593         .func           = bpf_sock_ops_reserve_hdr_opt,
7594         .gpl_only       = false,
7595         .ret_type       = RET_INTEGER,
7596         .arg1_type      = ARG_PTR_TO_CTX,
7597         .arg2_type      = ARG_ANYTHING,
7598         .arg3_type      = ARG_ANYTHING,
7599 };
7600
7601 BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb,
7602            u64, tstamp, u32, tstamp_type)
7603 {
7604         /* skb_clear_delivery_time() is done for inet protocol */
7605         if (skb->protocol != htons(ETH_P_IP) &&
7606             skb->protocol != htons(ETH_P_IPV6))
7607                 return -EOPNOTSUPP;
7608
7609         switch (tstamp_type) {
7610         case BPF_SKB_TSTAMP_DELIVERY_MONO:
7611                 if (!tstamp)
7612                         return -EINVAL;
7613                 skb->tstamp = tstamp;
7614                 skb->mono_delivery_time = 1;
7615                 break;
7616         case BPF_SKB_TSTAMP_UNSPEC:
7617                 if (tstamp)
7618                         return -EINVAL;
7619                 skb->tstamp = 0;
7620                 skb->mono_delivery_time = 0;
7621                 break;
7622         default:
7623                 return -EINVAL;
7624         }
7625
7626         return 0;
7627 }
7628
7629 static const struct bpf_func_proto bpf_skb_set_tstamp_proto = {
7630         .func           = bpf_skb_set_tstamp,
7631         .gpl_only       = false,
7632         .ret_type       = RET_INTEGER,
7633         .arg1_type      = ARG_PTR_TO_CTX,
7634         .arg2_type      = ARG_ANYTHING,
7635         .arg3_type      = ARG_ANYTHING,
7636 };
7637
7638 #ifdef CONFIG_SYN_COOKIES
7639 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv4, struct iphdr *, iph,
7640            struct tcphdr *, th, u32, th_len)
7641 {
7642         u32 cookie;
7643         u16 mss;
7644
7645         if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7646                 return -EINVAL;
7647
7648         mss = tcp_parse_mss_option(th, 0) ?: TCP_MSS_DEFAULT;
7649         cookie = __cookie_v4_init_sequence(iph, th, &mss);
7650
7651         return cookie | ((u64)mss << 32);
7652 }
7653
7654 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv4_proto = {
7655         .func           = bpf_tcp_raw_gen_syncookie_ipv4,
7656         .gpl_only       = true, /* __cookie_v4_init_sequence() is GPL */
7657         .pkt_access     = true,
7658         .ret_type       = RET_INTEGER,
7659         .arg1_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7660         .arg1_size      = sizeof(struct iphdr),
7661         .arg2_type      = ARG_PTR_TO_MEM,
7662         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
7663 };
7664
7665 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv6, struct ipv6hdr *, iph,
7666            struct tcphdr *, th, u32, th_len)
7667 {
7668 #if IS_BUILTIN(CONFIG_IPV6)
7669         const u16 mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
7670                 sizeof(struct ipv6hdr);
7671         u32 cookie;
7672         u16 mss;
7673
7674         if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7675                 return -EINVAL;
7676
7677         mss = tcp_parse_mss_option(th, 0) ?: mss_clamp;
7678         cookie = __cookie_v6_init_sequence(iph, th, &mss);
7679
7680         return cookie | ((u64)mss << 32);
7681 #else
7682         return -EPROTONOSUPPORT;
7683 #endif
7684 }
7685
7686 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv6_proto = {
7687         .func           = bpf_tcp_raw_gen_syncookie_ipv6,
7688         .gpl_only       = true, /* __cookie_v6_init_sequence() is GPL */
7689         .pkt_access     = true,
7690         .ret_type       = RET_INTEGER,
7691         .arg1_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7692         .arg1_size      = sizeof(struct ipv6hdr),
7693         .arg2_type      = ARG_PTR_TO_MEM,
7694         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
7695 };
7696
7697 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv4, struct iphdr *, iph,
7698            struct tcphdr *, th)
7699 {
7700         u32 cookie = ntohl(th->ack_seq) - 1;
7701
7702         if (__cookie_v4_check(iph, th, cookie) > 0)
7703                 return 0;
7704
7705         return -EACCES;
7706 }
7707
7708 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv4_proto = {
7709         .func           = bpf_tcp_raw_check_syncookie_ipv4,
7710         .gpl_only       = true, /* __cookie_v4_check is GPL */
7711         .pkt_access     = true,
7712         .ret_type       = RET_INTEGER,
7713         .arg1_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7714         .arg1_size      = sizeof(struct iphdr),
7715         .arg2_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7716         .arg2_size      = sizeof(struct tcphdr),
7717 };
7718
7719 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv6, struct ipv6hdr *, iph,
7720            struct tcphdr *, th)
7721 {
7722 #if IS_BUILTIN(CONFIG_IPV6)
7723         u32 cookie = ntohl(th->ack_seq) - 1;
7724
7725         if (__cookie_v6_check(iph, th, cookie) > 0)
7726                 return 0;
7727
7728         return -EACCES;
7729 #else
7730         return -EPROTONOSUPPORT;
7731 #endif
7732 }
7733
7734 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv6_proto = {
7735         .func           = bpf_tcp_raw_check_syncookie_ipv6,
7736         .gpl_only       = true, /* __cookie_v6_check is GPL */
7737         .pkt_access     = true,
7738         .ret_type       = RET_INTEGER,
7739         .arg1_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7740         .arg1_size      = sizeof(struct ipv6hdr),
7741         .arg2_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7742         .arg2_size      = sizeof(struct tcphdr),
7743 };
7744 #endif /* CONFIG_SYN_COOKIES */
7745
7746 #endif /* CONFIG_INET */
7747
7748 bool bpf_helper_changes_pkt_data(void *func)
7749 {
7750         if (func == bpf_skb_vlan_push ||
7751             func == bpf_skb_vlan_pop ||
7752             func == bpf_skb_store_bytes ||
7753             func == bpf_skb_change_proto ||
7754             func == bpf_skb_change_head ||
7755             func == sk_skb_change_head ||
7756             func == bpf_skb_change_tail ||
7757             func == sk_skb_change_tail ||
7758             func == bpf_skb_adjust_room ||
7759             func == sk_skb_adjust_room ||
7760             func == bpf_skb_pull_data ||
7761             func == sk_skb_pull_data ||
7762             func == bpf_clone_redirect ||
7763             func == bpf_l3_csum_replace ||
7764             func == bpf_l4_csum_replace ||
7765             func == bpf_xdp_adjust_head ||
7766             func == bpf_xdp_adjust_meta ||
7767             func == bpf_msg_pull_data ||
7768             func == bpf_msg_push_data ||
7769             func == bpf_msg_pop_data ||
7770             func == bpf_xdp_adjust_tail ||
7771 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7772             func == bpf_lwt_seg6_store_bytes ||
7773             func == bpf_lwt_seg6_adjust_srh ||
7774             func == bpf_lwt_seg6_action ||
7775 #endif
7776 #ifdef CONFIG_INET
7777             func == bpf_sock_ops_store_hdr_opt ||
7778 #endif
7779             func == bpf_lwt_in_push_encap ||
7780             func == bpf_lwt_xmit_push_encap)
7781                 return true;
7782
7783         return false;
7784 }
7785
7786 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7787 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7788
7789 static const struct bpf_func_proto *
7790 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7791 {
7792         const struct bpf_func_proto *func_proto;
7793
7794         func_proto = cgroup_common_func_proto(func_id, prog);
7795         if (func_proto)
7796                 return func_proto;
7797
7798         func_proto = cgroup_current_func_proto(func_id, prog);
7799         if (func_proto)
7800                 return func_proto;
7801
7802         switch (func_id) {
7803         case BPF_FUNC_get_socket_cookie:
7804                 return &bpf_get_socket_cookie_sock_proto;
7805         case BPF_FUNC_get_netns_cookie:
7806                 return &bpf_get_netns_cookie_sock_proto;
7807         case BPF_FUNC_perf_event_output:
7808                 return &bpf_event_output_data_proto;
7809         case BPF_FUNC_sk_storage_get:
7810                 return &bpf_sk_storage_get_cg_sock_proto;
7811         case BPF_FUNC_ktime_get_coarse_ns:
7812                 return &bpf_ktime_get_coarse_ns_proto;
7813         default:
7814                 return bpf_base_func_proto(func_id);
7815         }
7816 }
7817
7818 static const struct bpf_func_proto *
7819 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7820 {
7821         const struct bpf_func_proto *func_proto;
7822
7823         func_proto = cgroup_common_func_proto(func_id, prog);
7824         if (func_proto)
7825                 return func_proto;
7826
7827         func_proto = cgroup_current_func_proto(func_id, prog);
7828         if (func_proto)
7829                 return func_proto;
7830
7831         switch (func_id) {
7832         case BPF_FUNC_bind:
7833                 switch (prog->expected_attach_type) {
7834                 case BPF_CGROUP_INET4_CONNECT:
7835                 case BPF_CGROUP_INET6_CONNECT:
7836                         return &bpf_bind_proto;
7837                 default:
7838                         return NULL;
7839                 }
7840         case BPF_FUNC_get_socket_cookie:
7841                 return &bpf_get_socket_cookie_sock_addr_proto;
7842         case BPF_FUNC_get_netns_cookie:
7843                 return &bpf_get_netns_cookie_sock_addr_proto;
7844         case BPF_FUNC_perf_event_output:
7845                 return &bpf_event_output_data_proto;
7846 #ifdef CONFIG_INET
7847         case BPF_FUNC_sk_lookup_tcp:
7848                 return &bpf_sock_addr_sk_lookup_tcp_proto;
7849         case BPF_FUNC_sk_lookup_udp:
7850                 return &bpf_sock_addr_sk_lookup_udp_proto;
7851         case BPF_FUNC_sk_release:
7852                 return &bpf_sk_release_proto;
7853         case BPF_FUNC_skc_lookup_tcp:
7854                 return &bpf_sock_addr_skc_lookup_tcp_proto;
7855 #endif /* CONFIG_INET */
7856         case BPF_FUNC_sk_storage_get:
7857                 return &bpf_sk_storage_get_proto;
7858         case BPF_FUNC_sk_storage_delete:
7859                 return &bpf_sk_storage_delete_proto;
7860         case BPF_FUNC_setsockopt:
7861                 switch (prog->expected_attach_type) {
7862                 case BPF_CGROUP_INET4_BIND:
7863                 case BPF_CGROUP_INET6_BIND:
7864                 case BPF_CGROUP_INET4_CONNECT:
7865                 case BPF_CGROUP_INET6_CONNECT:
7866                 case BPF_CGROUP_UDP4_RECVMSG:
7867                 case BPF_CGROUP_UDP6_RECVMSG:
7868                 case BPF_CGROUP_UDP4_SENDMSG:
7869                 case BPF_CGROUP_UDP6_SENDMSG:
7870                 case BPF_CGROUP_INET4_GETPEERNAME:
7871                 case BPF_CGROUP_INET6_GETPEERNAME:
7872                 case BPF_CGROUP_INET4_GETSOCKNAME:
7873                 case BPF_CGROUP_INET6_GETSOCKNAME:
7874                         return &bpf_sock_addr_setsockopt_proto;
7875                 default:
7876                         return NULL;
7877                 }
7878         case BPF_FUNC_getsockopt:
7879                 switch (prog->expected_attach_type) {
7880                 case BPF_CGROUP_INET4_BIND:
7881                 case BPF_CGROUP_INET6_BIND:
7882                 case BPF_CGROUP_INET4_CONNECT:
7883                 case BPF_CGROUP_INET6_CONNECT:
7884                 case BPF_CGROUP_UDP4_RECVMSG:
7885                 case BPF_CGROUP_UDP6_RECVMSG:
7886                 case BPF_CGROUP_UDP4_SENDMSG:
7887                 case BPF_CGROUP_UDP6_SENDMSG:
7888                 case BPF_CGROUP_INET4_GETPEERNAME:
7889                 case BPF_CGROUP_INET6_GETPEERNAME:
7890                 case BPF_CGROUP_INET4_GETSOCKNAME:
7891                 case BPF_CGROUP_INET6_GETSOCKNAME:
7892                         return &bpf_sock_addr_getsockopt_proto;
7893                 default:
7894                         return NULL;
7895                 }
7896         default:
7897                 return bpf_sk_base_func_proto(func_id);
7898         }
7899 }
7900
7901 static const struct bpf_func_proto *
7902 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7903 {
7904         switch (func_id) {
7905         case BPF_FUNC_skb_load_bytes:
7906                 return &bpf_skb_load_bytes_proto;
7907         case BPF_FUNC_skb_load_bytes_relative:
7908                 return &bpf_skb_load_bytes_relative_proto;
7909         case BPF_FUNC_get_socket_cookie:
7910                 return &bpf_get_socket_cookie_proto;
7911         case BPF_FUNC_get_socket_uid:
7912                 return &bpf_get_socket_uid_proto;
7913         case BPF_FUNC_perf_event_output:
7914                 return &bpf_skb_event_output_proto;
7915         default:
7916                 return bpf_sk_base_func_proto(func_id);
7917         }
7918 }
7919
7920 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7921 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7922
7923 static const struct bpf_func_proto *
7924 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7925 {
7926         const struct bpf_func_proto *func_proto;
7927
7928         func_proto = cgroup_common_func_proto(func_id, prog);
7929         if (func_proto)
7930                 return func_proto;
7931
7932         switch (func_id) {
7933         case BPF_FUNC_sk_fullsock:
7934                 return &bpf_sk_fullsock_proto;
7935         case BPF_FUNC_sk_storage_get:
7936                 return &bpf_sk_storage_get_proto;
7937         case BPF_FUNC_sk_storage_delete:
7938                 return &bpf_sk_storage_delete_proto;
7939         case BPF_FUNC_perf_event_output:
7940                 return &bpf_skb_event_output_proto;
7941 #ifdef CONFIG_SOCK_CGROUP_DATA
7942         case BPF_FUNC_skb_cgroup_id:
7943                 return &bpf_skb_cgroup_id_proto;
7944         case BPF_FUNC_skb_ancestor_cgroup_id:
7945                 return &bpf_skb_ancestor_cgroup_id_proto;
7946         case BPF_FUNC_sk_cgroup_id:
7947                 return &bpf_sk_cgroup_id_proto;
7948         case BPF_FUNC_sk_ancestor_cgroup_id:
7949                 return &bpf_sk_ancestor_cgroup_id_proto;
7950 #endif
7951 #ifdef CONFIG_INET
7952         case BPF_FUNC_sk_lookup_tcp:
7953                 return &bpf_sk_lookup_tcp_proto;
7954         case BPF_FUNC_sk_lookup_udp:
7955                 return &bpf_sk_lookup_udp_proto;
7956         case BPF_FUNC_sk_release:
7957                 return &bpf_sk_release_proto;
7958         case BPF_FUNC_skc_lookup_tcp:
7959                 return &bpf_skc_lookup_tcp_proto;
7960         case BPF_FUNC_tcp_sock:
7961                 return &bpf_tcp_sock_proto;
7962         case BPF_FUNC_get_listener_sock:
7963                 return &bpf_get_listener_sock_proto;
7964         case BPF_FUNC_skb_ecn_set_ce:
7965                 return &bpf_skb_ecn_set_ce_proto;
7966 #endif
7967         default:
7968                 return sk_filter_func_proto(func_id, prog);
7969         }
7970 }
7971
7972 static const struct bpf_func_proto *
7973 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7974 {
7975         switch (func_id) {
7976         case BPF_FUNC_skb_store_bytes:
7977                 return &bpf_skb_store_bytes_proto;
7978         case BPF_FUNC_skb_load_bytes:
7979                 return &bpf_skb_load_bytes_proto;
7980         case BPF_FUNC_skb_load_bytes_relative:
7981                 return &bpf_skb_load_bytes_relative_proto;
7982         case BPF_FUNC_skb_pull_data:
7983                 return &bpf_skb_pull_data_proto;
7984         case BPF_FUNC_csum_diff:
7985                 return &bpf_csum_diff_proto;
7986         case BPF_FUNC_csum_update:
7987                 return &bpf_csum_update_proto;
7988         case BPF_FUNC_csum_level:
7989                 return &bpf_csum_level_proto;
7990         case BPF_FUNC_l3_csum_replace:
7991                 return &bpf_l3_csum_replace_proto;
7992         case BPF_FUNC_l4_csum_replace:
7993                 return &bpf_l4_csum_replace_proto;
7994         case BPF_FUNC_clone_redirect:
7995                 return &bpf_clone_redirect_proto;
7996         case BPF_FUNC_get_cgroup_classid:
7997                 return &bpf_get_cgroup_classid_proto;
7998         case BPF_FUNC_skb_vlan_push:
7999                 return &bpf_skb_vlan_push_proto;
8000         case BPF_FUNC_skb_vlan_pop:
8001                 return &bpf_skb_vlan_pop_proto;
8002         case BPF_FUNC_skb_change_proto:
8003                 return &bpf_skb_change_proto_proto;
8004         case BPF_FUNC_skb_change_type:
8005                 return &bpf_skb_change_type_proto;
8006         case BPF_FUNC_skb_adjust_room:
8007                 return &bpf_skb_adjust_room_proto;
8008         case BPF_FUNC_skb_change_tail:
8009                 return &bpf_skb_change_tail_proto;
8010         case BPF_FUNC_skb_change_head:
8011                 return &bpf_skb_change_head_proto;
8012         case BPF_FUNC_skb_get_tunnel_key:
8013                 return &bpf_skb_get_tunnel_key_proto;
8014         case BPF_FUNC_skb_set_tunnel_key:
8015                 return bpf_get_skb_set_tunnel_proto(func_id);
8016         case BPF_FUNC_skb_get_tunnel_opt:
8017                 return &bpf_skb_get_tunnel_opt_proto;
8018         case BPF_FUNC_skb_set_tunnel_opt:
8019                 return bpf_get_skb_set_tunnel_proto(func_id);
8020         case BPF_FUNC_redirect:
8021                 return &bpf_redirect_proto;
8022         case BPF_FUNC_redirect_neigh:
8023                 return &bpf_redirect_neigh_proto;
8024         case BPF_FUNC_redirect_peer:
8025                 return &bpf_redirect_peer_proto;
8026         case BPF_FUNC_get_route_realm:
8027                 return &bpf_get_route_realm_proto;
8028         case BPF_FUNC_get_hash_recalc:
8029                 return &bpf_get_hash_recalc_proto;
8030         case BPF_FUNC_set_hash_invalid:
8031                 return &bpf_set_hash_invalid_proto;
8032         case BPF_FUNC_set_hash:
8033                 return &bpf_set_hash_proto;
8034         case BPF_FUNC_perf_event_output:
8035                 return &bpf_skb_event_output_proto;
8036         case BPF_FUNC_get_smp_processor_id:
8037                 return &bpf_get_smp_processor_id_proto;
8038         case BPF_FUNC_skb_under_cgroup:
8039                 return &bpf_skb_under_cgroup_proto;
8040         case BPF_FUNC_get_socket_cookie:
8041                 return &bpf_get_socket_cookie_proto;
8042         case BPF_FUNC_get_socket_uid:
8043                 return &bpf_get_socket_uid_proto;
8044         case BPF_FUNC_fib_lookup:
8045                 return &bpf_skb_fib_lookup_proto;
8046         case BPF_FUNC_check_mtu:
8047                 return &bpf_skb_check_mtu_proto;
8048         case BPF_FUNC_sk_fullsock:
8049                 return &bpf_sk_fullsock_proto;
8050         case BPF_FUNC_sk_storage_get:
8051                 return &bpf_sk_storage_get_proto;
8052         case BPF_FUNC_sk_storage_delete:
8053                 return &bpf_sk_storage_delete_proto;
8054 #ifdef CONFIG_XFRM
8055         case BPF_FUNC_skb_get_xfrm_state:
8056                 return &bpf_skb_get_xfrm_state_proto;
8057 #endif
8058 #ifdef CONFIG_CGROUP_NET_CLASSID
8059         case BPF_FUNC_skb_cgroup_classid:
8060                 return &bpf_skb_cgroup_classid_proto;
8061 #endif
8062 #ifdef CONFIG_SOCK_CGROUP_DATA
8063         case BPF_FUNC_skb_cgroup_id:
8064                 return &bpf_skb_cgroup_id_proto;
8065         case BPF_FUNC_skb_ancestor_cgroup_id:
8066                 return &bpf_skb_ancestor_cgroup_id_proto;
8067 #endif
8068 #ifdef CONFIG_INET
8069         case BPF_FUNC_sk_lookup_tcp:
8070                 return &bpf_tc_sk_lookup_tcp_proto;
8071         case BPF_FUNC_sk_lookup_udp:
8072                 return &bpf_tc_sk_lookup_udp_proto;
8073         case BPF_FUNC_sk_release:
8074                 return &bpf_sk_release_proto;
8075         case BPF_FUNC_tcp_sock:
8076                 return &bpf_tcp_sock_proto;
8077         case BPF_FUNC_get_listener_sock:
8078                 return &bpf_get_listener_sock_proto;
8079         case BPF_FUNC_skc_lookup_tcp:
8080                 return &bpf_tc_skc_lookup_tcp_proto;
8081         case BPF_FUNC_tcp_check_syncookie:
8082                 return &bpf_tcp_check_syncookie_proto;
8083         case BPF_FUNC_skb_ecn_set_ce:
8084                 return &bpf_skb_ecn_set_ce_proto;
8085         case BPF_FUNC_tcp_gen_syncookie:
8086                 return &bpf_tcp_gen_syncookie_proto;
8087         case BPF_FUNC_sk_assign:
8088                 return &bpf_sk_assign_proto;
8089         case BPF_FUNC_skb_set_tstamp:
8090                 return &bpf_skb_set_tstamp_proto;
8091 #ifdef CONFIG_SYN_COOKIES
8092         case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8093                 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8094         case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8095                 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8096         case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8097                 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8098         case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8099                 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8100 #endif
8101 #endif
8102         default:
8103                 return bpf_sk_base_func_proto(func_id);
8104         }
8105 }
8106
8107 static const struct bpf_func_proto *
8108 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8109 {
8110         switch (func_id) {
8111         case BPF_FUNC_perf_event_output:
8112                 return &bpf_xdp_event_output_proto;
8113         case BPF_FUNC_get_smp_processor_id:
8114                 return &bpf_get_smp_processor_id_proto;
8115         case BPF_FUNC_csum_diff:
8116                 return &bpf_csum_diff_proto;
8117         case BPF_FUNC_xdp_adjust_head:
8118                 return &bpf_xdp_adjust_head_proto;
8119         case BPF_FUNC_xdp_adjust_meta:
8120                 return &bpf_xdp_adjust_meta_proto;
8121         case BPF_FUNC_redirect:
8122                 return &bpf_xdp_redirect_proto;
8123         case BPF_FUNC_redirect_map:
8124                 return &bpf_xdp_redirect_map_proto;
8125         case BPF_FUNC_xdp_adjust_tail:
8126                 return &bpf_xdp_adjust_tail_proto;
8127         case BPF_FUNC_xdp_get_buff_len:
8128                 return &bpf_xdp_get_buff_len_proto;
8129         case BPF_FUNC_xdp_load_bytes:
8130                 return &bpf_xdp_load_bytes_proto;
8131         case BPF_FUNC_xdp_store_bytes:
8132                 return &bpf_xdp_store_bytes_proto;
8133         case BPF_FUNC_fib_lookup:
8134                 return &bpf_xdp_fib_lookup_proto;
8135         case BPF_FUNC_check_mtu:
8136                 return &bpf_xdp_check_mtu_proto;
8137 #ifdef CONFIG_INET
8138         case BPF_FUNC_sk_lookup_udp:
8139                 return &bpf_xdp_sk_lookup_udp_proto;
8140         case BPF_FUNC_sk_lookup_tcp:
8141                 return &bpf_xdp_sk_lookup_tcp_proto;
8142         case BPF_FUNC_sk_release:
8143                 return &bpf_sk_release_proto;
8144         case BPF_FUNC_skc_lookup_tcp:
8145                 return &bpf_xdp_skc_lookup_tcp_proto;
8146         case BPF_FUNC_tcp_check_syncookie:
8147                 return &bpf_tcp_check_syncookie_proto;
8148         case BPF_FUNC_tcp_gen_syncookie:
8149                 return &bpf_tcp_gen_syncookie_proto;
8150 #ifdef CONFIG_SYN_COOKIES
8151         case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8152                 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8153         case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8154                 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8155         case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8156                 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8157         case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8158                 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8159 #endif
8160 #endif
8161         default:
8162                 return bpf_sk_base_func_proto(func_id);
8163         }
8164
8165 #if IS_MODULE(CONFIG_NF_CONNTRACK) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES)
8166         /* The nf_conn___init type is used in the NF_CONNTRACK kfuncs. The
8167          * kfuncs are defined in two different modules, and we want to be able
8168          * to use them interchangably with the same BTF type ID. Because modules
8169          * can't de-duplicate BTF IDs between each other, we need the type to be
8170          * referenced in the vmlinux BTF or the verifier will get confused about
8171          * the different types. So we add this dummy type reference which will
8172          * be included in vmlinux BTF, allowing both modules to refer to the
8173          * same type ID.
8174          */
8175         BTF_TYPE_EMIT(struct nf_conn___init);
8176 #endif
8177 }
8178
8179 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
8180 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
8181
8182 static const struct bpf_func_proto *
8183 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8184 {
8185         const struct bpf_func_proto *func_proto;
8186
8187         func_proto = cgroup_common_func_proto(func_id, prog);
8188         if (func_proto)
8189                 return func_proto;
8190
8191         switch (func_id) {
8192         case BPF_FUNC_setsockopt:
8193                 return &bpf_sock_ops_setsockopt_proto;
8194         case BPF_FUNC_getsockopt:
8195                 return &bpf_sock_ops_getsockopt_proto;
8196         case BPF_FUNC_sock_ops_cb_flags_set:
8197                 return &bpf_sock_ops_cb_flags_set_proto;
8198         case BPF_FUNC_sock_map_update:
8199                 return &bpf_sock_map_update_proto;
8200         case BPF_FUNC_sock_hash_update:
8201                 return &bpf_sock_hash_update_proto;
8202         case BPF_FUNC_get_socket_cookie:
8203                 return &bpf_get_socket_cookie_sock_ops_proto;
8204         case BPF_FUNC_perf_event_output:
8205                 return &bpf_event_output_data_proto;
8206         case BPF_FUNC_sk_storage_get:
8207                 return &bpf_sk_storage_get_proto;
8208         case BPF_FUNC_sk_storage_delete:
8209                 return &bpf_sk_storage_delete_proto;
8210         case BPF_FUNC_get_netns_cookie:
8211                 return &bpf_get_netns_cookie_sock_ops_proto;
8212 #ifdef CONFIG_INET
8213         case BPF_FUNC_load_hdr_opt:
8214                 return &bpf_sock_ops_load_hdr_opt_proto;
8215         case BPF_FUNC_store_hdr_opt:
8216                 return &bpf_sock_ops_store_hdr_opt_proto;
8217         case BPF_FUNC_reserve_hdr_opt:
8218                 return &bpf_sock_ops_reserve_hdr_opt_proto;
8219         case BPF_FUNC_tcp_sock:
8220                 return &bpf_tcp_sock_proto;
8221 #endif /* CONFIG_INET */
8222         default:
8223                 return bpf_sk_base_func_proto(func_id);
8224         }
8225 }
8226
8227 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
8228 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
8229
8230 static const struct bpf_func_proto *
8231 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8232 {
8233         switch (func_id) {
8234         case BPF_FUNC_msg_redirect_map:
8235                 return &bpf_msg_redirect_map_proto;
8236         case BPF_FUNC_msg_redirect_hash:
8237                 return &bpf_msg_redirect_hash_proto;
8238         case BPF_FUNC_msg_apply_bytes:
8239                 return &bpf_msg_apply_bytes_proto;
8240         case BPF_FUNC_msg_cork_bytes:
8241                 return &bpf_msg_cork_bytes_proto;
8242         case BPF_FUNC_msg_pull_data:
8243                 return &bpf_msg_pull_data_proto;
8244         case BPF_FUNC_msg_push_data:
8245                 return &bpf_msg_push_data_proto;
8246         case BPF_FUNC_msg_pop_data:
8247                 return &bpf_msg_pop_data_proto;
8248         case BPF_FUNC_perf_event_output:
8249                 return &bpf_event_output_data_proto;
8250         case BPF_FUNC_get_current_uid_gid:
8251                 return &bpf_get_current_uid_gid_proto;
8252         case BPF_FUNC_get_current_pid_tgid:
8253                 return &bpf_get_current_pid_tgid_proto;
8254         case BPF_FUNC_sk_storage_get:
8255                 return &bpf_sk_storage_get_proto;
8256         case BPF_FUNC_sk_storage_delete:
8257                 return &bpf_sk_storage_delete_proto;
8258         case BPF_FUNC_get_netns_cookie:
8259                 return &bpf_get_netns_cookie_sk_msg_proto;
8260 #ifdef CONFIG_CGROUP_NET_CLASSID
8261         case BPF_FUNC_get_cgroup_classid:
8262                 return &bpf_get_cgroup_classid_curr_proto;
8263 #endif
8264         default:
8265                 return bpf_sk_base_func_proto(func_id);
8266         }
8267 }
8268
8269 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
8270 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
8271
8272 static const struct bpf_func_proto *
8273 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8274 {
8275         switch (func_id) {
8276         case BPF_FUNC_skb_store_bytes:
8277                 return &bpf_skb_store_bytes_proto;
8278         case BPF_FUNC_skb_load_bytes:
8279                 return &bpf_skb_load_bytes_proto;
8280         case BPF_FUNC_skb_pull_data:
8281                 return &sk_skb_pull_data_proto;
8282         case BPF_FUNC_skb_change_tail:
8283                 return &sk_skb_change_tail_proto;
8284         case BPF_FUNC_skb_change_head:
8285                 return &sk_skb_change_head_proto;
8286         case BPF_FUNC_skb_adjust_room:
8287                 return &sk_skb_adjust_room_proto;
8288         case BPF_FUNC_get_socket_cookie:
8289                 return &bpf_get_socket_cookie_proto;
8290         case BPF_FUNC_get_socket_uid:
8291                 return &bpf_get_socket_uid_proto;
8292         case BPF_FUNC_sk_redirect_map:
8293                 return &bpf_sk_redirect_map_proto;
8294         case BPF_FUNC_sk_redirect_hash:
8295                 return &bpf_sk_redirect_hash_proto;
8296         case BPF_FUNC_perf_event_output:
8297                 return &bpf_skb_event_output_proto;
8298 #ifdef CONFIG_INET
8299         case BPF_FUNC_sk_lookup_tcp:
8300                 return &bpf_sk_lookup_tcp_proto;
8301         case BPF_FUNC_sk_lookup_udp:
8302                 return &bpf_sk_lookup_udp_proto;
8303         case BPF_FUNC_sk_release:
8304                 return &bpf_sk_release_proto;
8305         case BPF_FUNC_skc_lookup_tcp:
8306                 return &bpf_skc_lookup_tcp_proto;
8307 #endif
8308         default:
8309                 return bpf_sk_base_func_proto(func_id);
8310         }
8311 }
8312
8313 static const struct bpf_func_proto *
8314 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8315 {
8316         switch (func_id) {
8317         case BPF_FUNC_skb_load_bytes:
8318                 return &bpf_flow_dissector_load_bytes_proto;
8319         default:
8320                 return bpf_sk_base_func_proto(func_id);
8321         }
8322 }
8323
8324 static const struct bpf_func_proto *
8325 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8326 {
8327         switch (func_id) {
8328         case BPF_FUNC_skb_load_bytes:
8329                 return &bpf_skb_load_bytes_proto;
8330         case BPF_FUNC_skb_pull_data:
8331                 return &bpf_skb_pull_data_proto;
8332         case BPF_FUNC_csum_diff:
8333                 return &bpf_csum_diff_proto;
8334         case BPF_FUNC_get_cgroup_classid:
8335                 return &bpf_get_cgroup_classid_proto;
8336         case BPF_FUNC_get_route_realm:
8337                 return &bpf_get_route_realm_proto;
8338         case BPF_FUNC_get_hash_recalc:
8339                 return &bpf_get_hash_recalc_proto;
8340         case BPF_FUNC_perf_event_output:
8341                 return &bpf_skb_event_output_proto;
8342         case BPF_FUNC_get_smp_processor_id:
8343                 return &bpf_get_smp_processor_id_proto;
8344         case BPF_FUNC_skb_under_cgroup:
8345                 return &bpf_skb_under_cgroup_proto;
8346         default:
8347                 return bpf_sk_base_func_proto(func_id);
8348         }
8349 }
8350
8351 static const struct bpf_func_proto *
8352 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8353 {
8354         switch (func_id) {
8355         case BPF_FUNC_lwt_push_encap:
8356                 return &bpf_lwt_in_push_encap_proto;
8357         default:
8358                 return lwt_out_func_proto(func_id, prog);
8359         }
8360 }
8361
8362 static const struct bpf_func_proto *
8363 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8364 {
8365         switch (func_id) {
8366         case BPF_FUNC_skb_get_tunnel_key:
8367                 return &bpf_skb_get_tunnel_key_proto;
8368         case BPF_FUNC_skb_set_tunnel_key:
8369                 return bpf_get_skb_set_tunnel_proto(func_id);
8370         case BPF_FUNC_skb_get_tunnel_opt:
8371                 return &bpf_skb_get_tunnel_opt_proto;
8372         case BPF_FUNC_skb_set_tunnel_opt:
8373                 return bpf_get_skb_set_tunnel_proto(func_id);
8374         case BPF_FUNC_redirect:
8375                 return &bpf_redirect_proto;
8376         case BPF_FUNC_clone_redirect:
8377                 return &bpf_clone_redirect_proto;
8378         case BPF_FUNC_skb_change_tail:
8379                 return &bpf_skb_change_tail_proto;
8380         case BPF_FUNC_skb_change_head:
8381                 return &bpf_skb_change_head_proto;
8382         case BPF_FUNC_skb_store_bytes:
8383                 return &bpf_skb_store_bytes_proto;
8384         case BPF_FUNC_csum_update:
8385                 return &bpf_csum_update_proto;
8386         case BPF_FUNC_csum_level:
8387                 return &bpf_csum_level_proto;
8388         case BPF_FUNC_l3_csum_replace:
8389                 return &bpf_l3_csum_replace_proto;
8390         case BPF_FUNC_l4_csum_replace:
8391                 return &bpf_l4_csum_replace_proto;
8392         case BPF_FUNC_set_hash_invalid:
8393                 return &bpf_set_hash_invalid_proto;
8394         case BPF_FUNC_lwt_push_encap:
8395                 return &bpf_lwt_xmit_push_encap_proto;
8396         default:
8397                 return lwt_out_func_proto(func_id, prog);
8398         }
8399 }
8400
8401 static const struct bpf_func_proto *
8402 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8403 {
8404         switch (func_id) {
8405 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
8406         case BPF_FUNC_lwt_seg6_store_bytes:
8407                 return &bpf_lwt_seg6_store_bytes_proto;
8408         case BPF_FUNC_lwt_seg6_action:
8409                 return &bpf_lwt_seg6_action_proto;
8410         case BPF_FUNC_lwt_seg6_adjust_srh:
8411                 return &bpf_lwt_seg6_adjust_srh_proto;
8412 #endif
8413         default:
8414                 return lwt_out_func_proto(func_id, prog);
8415         }
8416 }
8417
8418 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
8419                                     const struct bpf_prog *prog,
8420                                     struct bpf_insn_access_aux *info)
8421 {
8422         const int size_default = sizeof(__u32);
8423
8424         if (off < 0 || off >= sizeof(struct __sk_buff))
8425                 return false;
8426
8427         /* The verifier guarantees that size > 0. */
8428         if (off % size != 0)
8429                 return false;
8430
8431         switch (off) {
8432         case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8433                 if (off + size > offsetofend(struct __sk_buff, cb[4]))
8434                         return false;
8435                 break;
8436         case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
8437         case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
8438         case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
8439         case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
8440         case bpf_ctx_range(struct __sk_buff, data):
8441         case bpf_ctx_range(struct __sk_buff, data_meta):
8442         case bpf_ctx_range(struct __sk_buff, data_end):
8443                 if (size != size_default)
8444                         return false;
8445                 break;
8446         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8447                 return false;
8448         case bpf_ctx_range(struct __sk_buff, hwtstamp):
8449                 if (type == BPF_WRITE || size != sizeof(__u64))
8450                         return false;
8451                 break;
8452         case bpf_ctx_range(struct __sk_buff, tstamp):
8453                 if (size != sizeof(__u64))
8454                         return false;
8455                 break;
8456         case offsetof(struct __sk_buff, sk):
8457                 if (type == BPF_WRITE || size != sizeof(__u64))
8458                         return false;
8459                 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
8460                 break;
8461         case offsetof(struct __sk_buff, tstamp_type):
8462                 return false;
8463         case offsetofend(struct __sk_buff, tstamp_type) ... offsetof(struct __sk_buff, hwtstamp) - 1:
8464                 /* Explicitly prohibit access to padding in __sk_buff. */
8465                 return false;
8466         default:
8467                 /* Only narrow read access allowed for now. */
8468                 if (type == BPF_WRITE) {
8469                         if (size != size_default)
8470                                 return false;
8471                 } else {
8472                         bpf_ctx_record_field_size(info, size_default);
8473                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8474                                 return false;
8475                 }
8476         }
8477
8478         return true;
8479 }
8480
8481 static bool sk_filter_is_valid_access(int off, int size,
8482                                       enum bpf_access_type type,
8483                                       const struct bpf_prog *prog,
8484                                       struct bpf_insn_access_aux *info)
8485 {
8486         switch (off) {
8487         case bpf_ctx_range(struct __sk_buff, tc_classid):
8488         case bpf_ctx_range(struct __sk_buff, data):
8489         case bpf_ctx_range(struct __sk_buff, data_meta):
8490         case bpf_ctx_range(struct __sk_buff, data_end):
8491         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8492         case bpf_ctx_range(struct __sk_buff, tstamp):
8493         case bpf_ctx_range(struct __sk_buff, wire_len):
8494         case bpf_ctx_range(struct __sk_buff, hwtstamp):
8495                 return false;
8496         }
8497
8498         if (type == BPF_WRITE) {
8499                 switch (off) {
8500                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8501                         break;
8502                 default:
8503                         return false;
8504                 }
8505         }
8506
8507         return bpf_skb_is_valid_access(off, size, type, prog, info);
8508 }
8509
8510 static bool cg_skb_is_valid_access(int off, int size,
8511                                    enum bpf_access_type type,
8512                                    const struct bpf_prog *prog,
8513                                    struct bpf_insn_access_aux *info)
8514 {
8515         switch (off) {
8516         case bpf_ctx_range(struct __sk_buff, tc_classid):
8517         case bpf_ctx_range(struct __sk_buff, data_meta):
8518         case bpf_ctx_range(struct __sk_buff, wire_len):
8519                 return false;
8520         case bpf_ctx_range(struct __sk_buff, data):
8521         case bpf_ctx_range(struct __sk_buff, data_end):
8522                 if (!bpf_capable())
8523                         return false;
8524                 break;
8525         }
8526
8527         if (type == BPF_WRITE) {
8528                 switch (off) {
8529                 case bpf_ctx_range(struct __sk_buff, mark):
8530                 case bpf_ctx_range(struct __sk_buff, priority):
8531                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8532                         break;
8533                 case bpf_ctx_range(struct __sk_buff, tstamp):
8534                         if (!bpf_capable())
8535                                 return false;
8536                         break;
8537                 default:
8538                         return false;
8539                 }
8540         }
8541
8542         switch (off) {
8543         case bpf_ctx_range(struct __sk_buff, data):
8544                 info->reg_type = PTR_TO_PACKET;
8545                 break;
8546         case bpf_ctx_range(struct __sk_buff, data_end):
8547                 info->reg_type = PTR_TO_PACKET_END;
8548                 break;
8549         }
8550
8551         return bpf_skb_is_valid_access(off, size, type, prog, info);
8552 }
8553
8554 static bool lwt_is_valid_access(int off, int size,
8555                                 enum bpf_access_type type,
8556                                 const struct bpf_prog *prog,
8557                                 struct bpf_insn_access_aux *info)
8558 {
8559         switch (off) {
8560         case bpf_ctx_range(struct __sk_buff, tc_classid):
8561         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8562         case bpf_ctx_range(struct __sk_buff, data_meta):
8563         case bpf_ctx_range(struct __sk_buff, tstamp):
8564         case bpf_ctx_range(struct __sk_buff, wire_len):
8565         case bpf_ctx_range(struct __sk_buff, hwtstamp):
8566                 return false;
8567         }
8568
8569         if (type == BPF_WRITE) {
8570                 switch (off) {
8571                 case bpf_ctx_range(struct __sk_buff, mark):
8572                 case bpf_ctx_range(struct __sk_buff, priority):
8573                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8574                         break;
8575                 default:
8576                         return false;
8577                 }
8578         }
8579
8580         switch (off) {
8581         case bpf_ctx_range(struct __sk_buff, data):
8582                 info->reg_type = PTR_TO_PACKET;
8583                 break;
8584         case bpf_ctx_range(struct __sk_buff, data_end):
8585                 info->reg_type = PTR_TO_PACKET_END;
8586                 break;
8587         }
8588
8589         return bpf_skb_is_valid_access(off, size, type, prog, info);
8590 }
8591
8592 /* Attach type specific accesses */
8593 static bool __sock_filter_check_attach_type(int off,
8594                                             enum bpf_access_type access_type,
8595                                             enum bpf_attach_type attach_type)
8596 {
8597         switch (off) {
8598         case offsetof(struct bpf_sock, bound_dev_if):
8599         case offsetof(struct bpf_sock, mark):
8600         case offsetof(struct bpf_sock, priority):
8601                 switch (attach_type) {
8602                 case BPF_CGROUP_INET_SOCK_CREATE:
8603                 case BPF_CGROUP_INET_SOCK_RELEASE:
8604                         goto full_access;
8605                 default:
8606                         return false;
8607                 }
8608         case bpf_ctx_range(struct bpf_sock, src_ip4):
8609                 switch (attach_type) {
8610                 case BPF_CGROUP_INET4_POST_BIND:
8611                         goto read_only;
8612                 default:
8613                         return false;
8614                 }
8615         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8616                 switch (attach_type) {
8617                 case BPF_CGROUP_INET6_POST_BIND:
8618                         goto read_only;
8619                 default:
8620                         return false;
8621                 }
8622         case bpf_ctx_range(struct bpf_sock, src_port):
8623                 switch (attach_type) {
8624                 case BPF_CGROUP_INET4_POST_BIND:
8625                 case BPF_CGROUP_INET6_POST_BIND:
8626                         goto read_only;
8627                 default:
8628                         return false;
8629                 }
8630         }
8631 read_only:
8632         return access_type == BPF_READ;
8633 full_access:
8634         return true;
8635 }
8636
8637 bool bpf_sock_common_is_valid_access(int off, int size,
8638                                      enum bpf_access_type type,
8639                                      struct bpf_insn_access_aux *info)
8640 {
8641         switch (off) {
8642         case bpf_ctx_range_till(struct bpf_sock, type, priority):
8643                 return false;
8644         default:
8645                 return bpf_sock_is_valid_access(off, size, type, info);
8646         }
8647 }
8648
8649 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
8650                               struct bpf_insn_access_aux *info)
8651 {
8652         const int size_default = sizeof(__u32);
8653         int field_size;
8654
8655         if (off < 0 || off >= sizeof(struct bpf_sock))
8656                 return false;
8657         if (off % size != 0)
8658                 return false;
8659
8660         switch (off) {
8661         case offsetof(struct bpf_sock, state):
8662         case offsetof(struct bpf_sock, family):
8663         case offsetof(struct bpf_sock, type):
8664         case offsetof(struct bpf_sock, protocol):
8665         case offsetof(struct bpf_sock, src_port):
8666         case offsetof(struct bpf_sock, rx_queue_mapping):
8667         case bpf_ctx_range(struct bpf_sock, src_ip4):
8668         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8669         case bpf_ctx_range(struct bpf_sock, dst_ip4):
8670         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8671                 bpf_ctx_record_field_size(info, size_default);
8672                 return bpf_ctx_narrow_access_ok(off, size, size_default);
8673         case bpf_ctx_range(struct bpf_sock, dst_port):
8674                 field_size = size == size_default ?
8675                         size_default : sizeof_field(struct bpf_sock, dst_port);
8676                 bpf_ctx_record_field_size(info, field_size);
8677                 return bpf_ctx_narrow_access_ok(off, size, field_size);
8678         case offsetofend(struct bpf_sock, dst_port) ...
8679              offsetof(struct bpf_sock, dst_ip4) - 1:
8680                 return false;
8681         }
8682
8683         return size == size_default;
8684 }
8685
8686 static bool sock_filter_is_valid_access(int off, int size,
8687                                         enum bpf_access_type type,
8688                                         const struct bpf_prog *prog,
8689                                         struct bpf_insn_access_aux *info)
8690 {
8691         if (!bpf_sock_is_valid_access(off, size, type, info))
8692                 return false;
8693         return __sock_filter_check_attach_type(off, type,
8694                                                prog->expected_attach_type);
8695 }
8696
8697 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
8698                              const struct bpf_prog *prog)
8699 {
8700         /* Neither direct read nor direct write requires any preliminary
8701          * action.
8702          */
8703         return 0;
8704 }
8705
8706 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
8707                                 const struct bpf_prog *prog, int drop_verdict)
8708 {
8709         struct bpf_insn *insn = insn_buf;
8710
8711         if (!direct_write)
8712                 return 0;
8713
8714         /* if (!skb->cloned)
8715          *       goto start;
8716          *
8717          * (Fast-path, otherwise approximation that we might be
8718          *  a clone, do the rest in helper.)
8719          */
8720         *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET);
8721         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
8722         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
8723
8724         /* ret = bpf_skb_pull_data(skb, 0); */
8725         *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
8726         *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
8727         *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
8728                                BPF_FUNC_skb_pull_data);
8729         /* if (!ret)
8730          *      goto restore;
8731          * return TC_ACT_SHOT;
8732          */
8733         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
8734         *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
8735         *insn++ = BPF_EXIT_INSN();
8736
8737         /* restore: */
8738         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
8739         /* start: */
8740         *insn++ = prog->insnsi[0];
8741
8742         return insn - insn_buf;
8743 }
8744
8745 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
8746                           struct bpf_insn *insn_buf)
8747 {
8748         bool indirect = BPF_MODE(orig->code) == BPF_IND;
8749         struct bpf_insn *insn = insn_buf;
8750
8751         if (!indirect) {
8752                 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
8753         } else {
8754                 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
8755                 if (orig->imm)
8756                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
8757         }
8758         /* We're guaranteed here that CTX is in R6. */
8759         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
8760
8761         switch (BPF_SIZE(orig->code)) {
8762         case BPF_B:
8763                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
8764                 break;
8765         case BPF_H:
8766                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
8767                 break;
8768         case BPF_W:
8769                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
8770                 break;
8771         }
8772
8773         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
8774         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
8775         *insn++ = BPF_EXIT_INSN();
8776
8777         return insn - insn_buf;
8778 }
8779
8780 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8781                                const struct bpf_prog *prog)
8782 {
8783         return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8784 }
8785
8786 static bool tc_cls_act_is_valid_access(int off, int size,
8787                                        enum bpf_access_type type,
8788                                        const struct bpf_prog *prog,
8789                                        struct bpf_insn_access_aux *info)
8790 {
8791         if (type == BPF_WRITE) {
8792                 switch (off) {
8793                 case bpf_ctx_range(struct __sk_buff, mark):
8794                 case bpf_ctx_range(struct __sk_buff, tc_index):
8795                 case bpf_ctx_range(struct __sk_buff, priority):
8796                 case bpf_ctx_range(struct __sk_buff, tc_classid):
8797                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8798                 case bpf_ctx_range(struct __sk_buff, tstamp):
8799                 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8800                         break;
8801                 default:
8802                         return false;
8803                 }
8804         }
8805
8806         switch (off) {
8807         case bpf_ctx_range(struct __sk_buff, data):
8808                 info->reg_type = PTR_TO_PACKET;
8809                 break;
8810         case bpf_ctx_range(struct __sk_buff, data_meta):
8811                 info->reg_type = PTR_TO_PACKET_META;
8812                 break;
8813         case bpf_ctx_range(struct __sk_buff, data_end):
8814                 info->reg_type = PTR_TO_PACKET_END;
8815                 break;
8816         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8817                 return false;
8818         case offsetof(struct __sk_buff, tstamp_type):
8819                 /* The convert_ctx_access() on reading and writing
8820                  * __sk_buff->tstamp depends on whether the bpf prog
8821                  * has used __sk_buff->tstamp_type or not.
8822                  * Thus, we need to set prog->tstamp_type_access
8823                  * earlier during is_valid_access() here.
8824                  */
8825                 ((struct bpf_prog *)prog)->tstamp_type_access = 1;
8826                 return size == sizeof(__u8);
8827         }
8828
8829         return bpf_skb_is_valid_access(off, size, type, prog, info);
8830 }
8831
8832 DEFINE_MUTEX(nf_conn_btf_access_lock);
8833 EXPORT_SYMBOL_GPL(nf_conn_btf_access_lock);
8834
8835 int (*nfct_btf_struct_access)(struct bpf_verifier_log *log,
8836                               const struct bpf_reg_state *reg,
8837                               int off, int size);
8838 EXPORT_SYMBOL_GPL(nfct_btf_struct_access);
8839
8840 static int tc_cls_act_btf_struct_access(struct bpf_verifier_log *log,
8841                                         const struct bpf_reg_state *reg,
8842                                         int off, int size)
8843 {
8844         int ret = -EACCES;
8845
8846         mutex_lock(&nf_conn_btf_access_lock);
8847         if (nfct_btf_struct_access)
8848                 ret = nfct_btf_struct_access(log, reg, off, size);
8849         mutex_unlock(&nf_conn_btf_access_lock);
8850
8851         return ret;
8852 }
8853
8854 static bool __is_valid_xdp_access(int off, int size)
8855 {
8856         if (off < 0 || off >= sizeof(struct xdp_md))
8857                 return false;
8858         if (off % size != 0)
8859                 return false;
8860         if (size != sizeof(__u32))
8861                 return false;
8862
8863         return true;
8864 }
8865
8866 static bool xdp_is_valid_access(int off, int size,
8867                                 enum bpf_access_type type,
8868                                 const struct bpf_prog *prog,
8869                                 struct bpf_insn_access_aux *info)
8870 {
8871         if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8872                 switch (off) {
8873                 case offsetof(struct xdp_md, egress_ifindex):
8874                         return false;
8875                 }
8876         }
8877
8878         if (type == BPF_WRITE) {
8879                 if (bpf_prog_is_offloaded(prog->aux)) {
8880                         switch (off) {
8881                         case offsetof(struct xdp_md, rx_queue_index):
8882                                 return __is_valid_xdp_access(off, size);
8883                         }
8884                 }
8885                 return false;
8886         }
8887
8888         switch (off) {
8889         case offsetof(struct xdp_md, data):
8890                 info->reg_type = PTR_TO_PACKET;
8891                 break;
8892         case offsetof(struct xdp_md, data_meta):
8893                 info->reg_type = PTR_TO_PACKET_META;
8894                 break;
8895         case offsetof(struct xdp_md, data_end):
8896                 info->reg_type = PTR_TO_PACKET_END;
8897                 break;
8898         }
8899
8900         return __is_valid_xdp_access(off, size);
8901 }
8902
8903 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act)
8904 {
8905         const u32 act_max = XDP_REDIRECT;
8906
8907         pr_warn_once("%s XDP return value %u on prog %s (id %d) dev %s, expect packet loss!\n",
8908                      act > act_max ? "Illegal" : "Driver unsupported",
8909                      act, prog->aux->name, prog->aux->id, dev ? dev->name : "N/A");
8910 }
8911 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
8912
8913 static int xdp_btf_struct_access(struct bpf_verifier_log *log,
8914                                  const struct bpf_reg_state *reg,
8915                                  int off, int size)
8916 {
8917         int ret = -EACCES;
8918
8919         mutex_lock(&nf_conn_btf_access_lock);
8920         if (nfct_btf_struct_access)
8921                 ret = nfct_btf_struct_access(log, reg, off, size);
8922         mutex_unlock(&nf_conn_btf_access_lock);
8923
8924         return ret;
8925 }
8926
8927 static bool sock_addr_is_valid_access(int off, int size,
8928                                       enum bpf_access_type type,
8929                                       const struct bpf_prog *prog,
8930                                       struct bpf_insn_access_aux *info)
8931 {
8932         const int size_default = sizeof(__u32);
8933
8934         if (off < 0 || off >= sizeof(struct bpf_sock_addr))
8935                 return false;
8936         if (off % size != 0)
8937                 return false;
8938
8939         /* Disallow access to IPv6 fields from IPv4 contex and vise
8940          * versa.
8941          */
8942         switch (off) {
8943         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8944                 switch (prog->expected_attach_type) {
8945                 case BPF_CGROUP_INET4_BIND:
8946                 case BPF_CGROUP_INET4_CONNECT:
8947                 case BPF_CGROUP_INET4_GETPEERNAME:
8948                 case BPF_CGROUP_INET4_GETSOCKNAME:
8949                 case BPF_CGROUP_UDP4_SENDMSG:
8950                 case BPF_CGROUP_UDP4_RECVMSG:
8951                         break;
8952                 default:
8953                         return false;
8954                 }
8955                 break;
8956         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8957                 switch (prog->expected_attach_type) {
8958                 case BPF_CGROUP_INET6_BIND:
8959                 case BPF_CGROUP_INET6_CONNECT:
8960                 case BPF_CGROUP_INET6_GETPEERNAME:
8961                 case BPF_CGROUP_INET6_GETSOCKNAME:
8962                 case BPF_CGROUP_UDP6_SENDMSG:
8963                 case BPF_CGROUP_UDP6_RECVMSG:
8964                         break;
8965                 default:
8966                         return false;
8967                 }
8968                 break;
8969         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8970                 switch (prog->expected_attach_type) {
8971                 case BPF_CGROUP_UDP4_SENDMSG:
8972                         break;
8973                 default:
8974                         return false;
8975                 }
8976                 break;
8977         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8978                                 msg_src_ip6[3]):
8979                 switch (prog->expected_attach_type) {
8980                 case BPF_CGROUP_UDP6_SENDMSG:
8981                         break;
8982                 default:
8983                         return false;
8984                 }
8985                 break;
8986         }
8987
8988         switch (off) {
8989         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8990         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8991         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8992         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8993                                 msg_src_ip6[3]):
8994         case bpf_ctx_range(struct bpf_sock_addr, user_port):
8995                 if (type == BPF_READ) {
8996                         bpf_ctx_record_field_size(info, size_default);
8997
8998                         if (bpf_ctx_wide_access_ok(off, size,
8999                                                    struct bpf_sock_addr,
9000                                                    user_ip6))
9001                                 return true;
9002
9003                         if (bpf_ctx_wide_access_ok(off, size,
9004                                                    struct bpf_sock_addr,
9005                                                    msg_src_ip6))
9006                                 return true;
9007
9008                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
9009                                 return false;
9010                 } else {
9011                         if (bpf_ctx_wide_access_ok(off, size,
9012                                                    struct bpf_sock_addr,
9013                                                    user_ip6))
9014                                 return true;
9015
9016                         if (bpf_ctx_wide_access_ok(off, size,
9017                                                    struct bpf_sock_addr,
9018                                                    msg_src_ip6))
9019                                 return true;
9020
9021                         if (size != size_default)
9022                                 return false;
9023                 }
9024                 break;
9025         case offsetof(struct bpf_sock_addr, sk):
9026                 if (type != BPF_READ)
9027                         return false;
9028                 if (size != sizeof(__u64))
9029                         return false;
9030                 info->reg_type = PTR_TO_SOCKET;
9031                 break;
9032         default:
9033                 if (type == BPF_READ) {
9034                         if (size != size_default)
9035                                 return false;
9036                 } else {
9037                         return false;
9038                 }
9039         }
9040
9041         return true;
9042 }
9043
9044 static bool sock_ops_is_valid_access(int off, int size,
9045                                      enum bpf_access_type type,
9046                                      const struct bpf_prog *prog,
9047                                      struct bpf_insn_access_aux *info)
9048 {
9049         const int size_default = sizeof(__u32);
9050
9051         if (off < 0 || off >= sizeof(struct bpf_sock_ops))
9052                 return false;
9053
9054         /* The verifier guarantees that size > 0. */
9055         if (off % size != 0)
9056                 return false;
9057
9058         if (type == BPF_WRITE) {
9059                 switch (off) {
9060                 case offsetof(struct bpf_sock_ops, reply):
9061                 case offsetof(struct bpf_sock_ops, sk_txhash):
9062                         if (size != size_default)
9063                                 return false;
9064                         break;
9065                 default:
9066                         return false;
9067                 }
9068         } else {
9069                 switch (off) {
9070                 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
9071                                         bytes_acked):
9072                         if (size != sizeof(__u64))
9073                                 return false;
9074                         break;
9075                 case offsetof(struct bpf_sock_ops, sk):
9076                         if (size != sizeof(__u64))
9077                                 return false;
9078                         info->reg_type = PTR_TO_SOCKET_OR_NULL;
9079                         break;
9080                 case offsetof(struct bpf_sock_ops, skb_data):
9081                         if (size != sizeof(__u64))
9082                                 return false;
9083                         info->reg_type = PTR_TO_PACKET;
9084                         break;
9085                 case offsetof(struct bpf_sock_ops, skb_data_end):
9086                         if (size != sizeof(__u64))
9087                                 return false;
9088                         info->reg_type = PTR_TO_PACKET_END;
9089                         break;
9090                 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9091                         bpf_ctx_record_field_size(info, size_default);
9092                         return bpf_ctx_narrow_access_ok(off, size,
9093                                                         size_default);
9094                 case offsetof(struct bpf_sock_ops, skb_hwtstamp):
9095                         if (size != sizeof(__u64))
9096                                 return false;
9097                         break;
9098                 default:
9099                         if (size != size_default)
9100                                 return false;
9101                         break;
9102                 }
9103         }
9104
9105         return true;
9106 }
9107
9108 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
9109                            const struct bpf_prog *prog)
9110 {
9111         return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
9112 }
9113
9114 static bool sk_skb_is_valid_access(int off, int size,
9115                                    enum bpf_access_type type,
9116                                    const struct bpf_prog *prog,
9117                                    struct bpf_insn_access_aux *info)
9118 {
9119         switch (off) {
9120         case bpf_ctx_range(struct __sk_buff, tc_classid):
9121         case bpf_ctx_range(struct __sk_buff, data_meta):
9122         case bpf_ctx_range(struct __sk_buff, tstamp):
9123         case bpf_ctx_range(struct __sk_buff, wire_len):
9124         case bpf_ctx_range(struct __sk_buff, hwtstamp):
9125                 return false;
9126         }
9127
9128         if (type == BPF_WRITE) {
9129                 switch (off) {
9130                 case bpf_ctx_range(struct __sk_buff, tc_index):
9131                 case bpf_ctx_range(struct __sk_buff, priority):
9132                         break;
9133                 default:
9134                         return false;
9135                 }
9136         }
9137
9138         switch (off) {
9139         case bpf_ctx_range(struct __sk_buff, mark):
9140                 return false;
9141         case bpf_ctx_range(struct __sk_buff, data):
9142                 info->reg_type = PTR_TO_PACKET;
9143                 break;
9144         case bpf_ctx_range(struct __sk_buff, data_end):
9145                 info->reg_type = PTR_TO_PACKET_END;
9146                 break;
9147         }
9148
9149         return bpf_skb_is_valid_access(off, size, type, prog, info);
9150 }
9151
9152 static bool sk_msg_is_valid_access(int off, int size,
9153                                    enum bpf_access_type type,
9154                                    const struct bpf_prog *prog,
9155                                    struct bpf_insn_access_aux *info)
9156 {
9157         if (type == BPF_WRITE)
9158                 return false;
9159
9160         if (off % size != 0)
9161                 return false;
9162
9163         switch (off) {
9164         case offsetof(struct sk_msg_md, data):
9165                 info->reg_type = PTR_TO_PACKET;
9166                 if (size != sizeof(__u64))
9167                         return false;
9168                 break;
9169         case offsetof(struct sk_msg_md, data_end):
9170                 info->reg_type = PTR_TO_PACKET_END;
9171                 if (size != sizeof(__u64))
9172                         return false;
9173                 break;
9174         case offsetof(struct sk_msg_md, sk):
9175                 if (size != sizeof(__u64))
9176                         return false;
9177                 info->reg_type = PTR_TO_SOCKET;
9178                 break;
9179         case bpf_ctx_range(struct sk_msg_md, family):
9180         case bpf_ctx_range(struct sk_msg_md, remote_ip4):
9181         case bpf_ctx_range(struct sk_msg_md, local_ip4):
9182         case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
9183         case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
9184         case bpf_ctx_range(struct sk_msg_md, remote_port):
9185         case bpf_ctx_range(struct sk_msg_md, local_port):
9186         case bpf_ctx_range(struct sk_msg_md, size):
9187                 if (size != sizeof(__u32))
9188                         return false;
9189                 break;
9190         default:
9191                 return false;
9192         }
9193         return true;
9194 }
9195
9196 static bool flow_dissector_is_valid_access(int off, int size,
9197                                            enum bpf_access_type type,
9198                                            const struct bpf_prog *prog,
9199                                            struct bpf_insn_access_aux *info)
9200 {
9201         const int size_default = sizeof(__u32);
9202
9203         if (off < 0 || off >= sizeof(struct __sk_buff))
9204                 return false;
9205
9206         if (type == BPF_WRITE)
9207                 return false;
9208
9209         switch (off) {
9210         case bpf_ctx_range(struct __sk_buff, data):
9211                 if (size != size_default)
9212                         return false;
9213                 info->reg_type = PTR_TO_PACKET;
9214                 return true;
9215         case bpf_ctx_range(struct __sk_buff, data_end):
9216                 if (size != size_default)
9217                         return false;
9218                 info->reg_type = PTR_TO_PACKET_END;
9219                 return true;
9220         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
9221                 if (size != sizeof(__u64))
9222                         return false;
9223                 info->reg_type = PTR_TO_FLOW_KEYS;
9224                 return true;
9225         default:
9226                 return false;
9227         }
9228 }
9229
9230 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
9231                                              const struct bpf_insn *si,
9232                                              struct bpf_insn *insn_buf,
9233                                              struct bpf_prog *prog,
9234                                              u32 *target_size)
9235
9236 {
9237         struct bpf_insn *insn = insn_buf;
9238
9239         switch (si->off) {
9240         case offsetof(struct __sk_buff, data):
9241                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
9242                                       si->dst_reg, si->src_reg,
9243                                       offsetof(struct bpf_flow_dissector, data));
9244                 break;
9245
9246         case offsetof(struct __sk_buff, data_end):
9247                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
9248                                       si->dst_reg, si->src_reg,
9249                                       offsetof(struct bpf_flow_dissector, data_end));
9250                 break;
9251
9252         case offsetof(struct __sk_buff, flow_keys):
9253                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
9254                                       si->dst_reg, si->src_reg,
9255                                       offsetof(struct bpf_flow_dissector, flow_keys));
9256                 break;
9257         }
9258
9259         return insn - insn_buf;
9260 }
9261
9262 static struct bpf_insn *bpf_convert_tstamp_type_read(const struct bpf_insn *si,
9263                                                      struct bpf_insn *insn)
9264 {
9265         __u8 value_reg = si->dst_reg;
9266         __u8 skb_reg = si->src_reg;
9267         /* AX is needed because src_reg and dst_reg could be the same */
9268         __u8 tmp_reg = BPF_REG_AX;
9269
9270         *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg,
9271                               SKB_BF_MONO_TC_OFFSET);
9272         *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg,
9273                                 SKB_MONO_DELIVERY_TIME_MASK, 2);
9274         *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_UNSPEC);
9275         *insn++ = BPF_JMP_A(1);
9276         *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_DELIVERY_MONO);
9277
9278         return insn;
9279 }
9280
9281 static struct bpf_insn *bpf_convert_shinfo_access(__u8 dst_reg, __u8 skb_reg,
9282                                                   struct bpf_insn *insn)
9283 {
9284         /* si->dst_reg = skb_shinfo(SKB); */
9285 #ifdef NET_SKBUFF_DATA_USES_OFFSET
9286         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9287                               BPF_REG_AX, skb_reg,
9288                               offsetof(struct sk_buff, end));
9289         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
9290                               dst_reg, skb_reg,
9291                               offsetof(struct sk_buff, head));
9292         *insn++ = BPF_ALU64_REG(BPF_ADD, dst_reg, BPF_REG_AX);
9293 #else
9294         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9295                               dst_reg, skb_reg,
9296                               offsetof(struct sk_buff, end));
9297 #endif
9298
9299         return insn;
9300 }
9301
9302 static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog,
9303                                                 const struct bpf_insn *si,
9304                                                 struct bpf_insn *insn)
9305 {
9306         __u8 value_reg = si->dst_reg;
9307         __u8 skb_reg = si->src_reg;
9308
9309 #ifdef CONFIG_NET_CLS_ACT
9310         /* If the tstamp_type is read,
9311          * the bpf prog is aware the tstamp could have delivery time.
9312          * Thus, read skb->tstamp as is if tstamp_type_access is true.
9313          */
9314         if (!prog->tstamp_type_access) {
9315                 /* AX is needed because src_reg and dst_reg could be the same */
9316                 __u8 tmp_reg = BPF_REG_AX;
9317
9318                 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, SKB_BF_MONO_TC_OFFSET);
9319                 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg,
9320                                         TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK);
9321                 *insn++ = BPF_JMP32_IMM(BPF_JNE, tmp_reg,
9322                                         TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK, 2);
9323                 /* skb->tc_at_ingress && skb->mono_delivery_time,
9324                  * read 0 as the (rcv) timestamp.
9325                  */
9326                 *insn++ = BPF_MOV64_IMM(value_reg, 0);
9327                 *insn++ = BPF_JMP_A(1);
9328         }
9329 #endif
9330
9331         *insn++ = BPF_LDX_MEM(BPF_DW, value_reg, skb_reg,
9332                               offsetof(struct sk_buff, tstamp));
9333         return insn;
9334 }
9335
9336 static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog,
9337                                                  const struct bpf_insn *si,
9338                                                  struct bpf_insn *insn)
9339 {
9340         __u8 value_reg = si->src_reg;
9341         __u8 skb_reg = si->dst_reg;
9342
9343 #ifdef CONFIG_NET_CLS_ACT
9344         /* If the tstamp_type is read,
9345          * the bpf prog is aware the tstamp could have delivery time.
9346          * Thus, write skb->tstamp as is if tstamp_type_access is true.
9347          * Otherwise, writing at ingress will have to clear the
9348          * mono_delivery_time bit also.
9349          */
9350         if (!prog->tstamp_type_access) {
9351                 __u8 tmp_reg = BPF_REG_AX;
9352
9353                 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, SKB_BF_MONO_TC_OFFSET);
9354                 /* Writing __sk_buff->tstamp as ingress, goto <clear> */
9355                 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1);
9356                 /* goto <store> */
9357                 *insn++ = BPF_JMP_A(2);
9358                 /* <clear>: mono_delivery_time */
9359                 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_MONO_DELIVERY_TIME_MASK);
9360                 *insn++ = BPF_STX_MEM(BPF_B, skb_reg, tmp_reg, SKB_BF_MONO_TC_OFFSET);
9361         }
9362 #endif
9363
9364         /* <store>: skb->tstamp = tstamp */
9365         *insn++ = BPF_RAW_INSN(BPF_CLASS(si->code) | BPF_DW | BPF_MEM,
9366                                skb_reg, value_reg, offsetof(struct sk_buff, tstamp), si->imm);
9367         return insn;
9368 }
9369
9370 #define BPF_EMIT_STORE(size, si, off)                                   \
9371         BPF_RAW_INSN(BPF_CLASS((si)->code) | (size) | BPF_MEM,          \
9372                      (si)->dst_reg, (si)->src_reg, (off), (si)->imm)
9373
9374 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
9375                                   const struct bpf_insn *si,
9376                                   struct bpf_insn *insn_buf,
9377                                   struct bpf_prog *prog, u32 *target_size)
9378 {
9379         struct bpf_insn *insn = insn_buf;
9380         int off;
9381
9382         switch (si->off) {
9383         case offsetof(struct __sk_buff, len):
9384                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9385                                       bpf_target_off(struct sk_buff, len, 4,
9386                                                      target_size));
9387                 break;
9388
9389         case offsetof(struct __sk_buff, protocol):
9390                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9391                                       bpf_target_off(struct sk_buff, protocol, 2,
9392                                                      target_size));
9393                 break;
9394
9395         case offsetof(struct __sk_buff, vlan_proto):
9396                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9397                                       bpf_target_off(struct sk_buff, vlan_proto, 2,
9398                                                      target_size));
9399                 break;
9400
9401         case offsetof(struct __sk_buff, priority):
9402                 if (type == BPF_WRITE)
9403                         *insn++ = BPF_EMIT_STORE(BPF_W, si,
9404                                                  bpf_target_off(struct sk_buff, priority, 4,
9405                                                                 target_size));
9406                 else
9407                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9408                                               bpf_target_off(struct sk_buff, priority, 4,
9409                                                              target_size));
9410                 break;
9411
9412         case offsetof(struct __sk_buff, ingress_ifindex):
9413                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9414                                       bpf_target_off(struct sk_buff, skb_iif, 4,
9415                                                      target_size));
9416                 break;
9417
9418         case offsetof(struct __sk_buff, ifindex):
9419                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9420                                       si->dst_reg, si->src_reg,
9421                                       offsetof(struct sk_buff, dev));
9422                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9423                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9424                                       bpf_target_off(struct net_device, ifindex, 4,
9425                                                      target_size));
9426                 break;
9427
9428         case offsetof(struct __sk_buff, hash):
9429                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9430                                       bpf_target_off(struct sk_buff, hash, 4,
9431                                                      target_size));
9432                 break;
9433
9434         case offsetof(struct __sk_buff, mark):
9435                 if (type == BPF_WRITE)
9436                         *insn++ = BPF_EMIT_STORE(BPF_W, si,
9437                                                  bpf_target_off(struct sk_buff, mark, 4,
9438                                                                 target_size));
9439                 else
9440                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9441                                               bpf_target_off(struct sk_buff, mark, 4,
9442                                                              target_size));
9443                 break;
9444
9445         case offsetof(struct __sk_buff, pkt_type):
9446                 *target_size = 1;
9447                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
9448                                       PKT_TYPE_OFFSET);
9449                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
9450 #ifdef __BIG_ENDIAN_BITFIELD
9451                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
9452 #endif
9453                 break;
9454
9455         case offsetof(struct __sk_buff, queue_mapping):
9456                 if (type == BPF_WRITE) {
9457                         u32 off = bpf_target_off(struct sk_buff, queue_mapping, 2, target_size);
9458
9459                         if (BPF_CLASS(si->code) == BPF_ST && si->imm >= NO_QUEUE_MAPPING) {
9460                                 *insn++ = BPF_JMP_A(0); /* noop */
9461                                 break;
9462                         }
9463
9464                         if (BPF_CLASS(si->code) == BPF_STX)
9465                                 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
9466                         *insn++ = BPF_EMIT_STORE(BPF_H, si, off);
9467                 } else {
9468                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9469                                               bpf_target_off(struct sk_buff,
9470                                                              queue_mapping,
9471                                                              2, target_size));
9472                 }
9473                 break;
9474
9475         case offsetof(struct __sk_buff, vlan_present):
9476                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9477                                       bpf_target_off(struct sk_buff,
9478                                                      vlan_all, 4, target_size));
9479                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9480                 *insn++ = BPF_ALU32_IMM(BPF_MOV, si->dst_reg, 1);
9481                 break;
9482
9483         case offsetof(struct __sk_buff, vlan_tci):
9484                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9485                                       bpf_target_off(struct sk_buff, vlan_tci, 2,
9486                                                      target_size));
9487                 break;
9488
9489         case offsetof(struct __sk_buff, cb[0]) ...
9490              offsetofend(struct __sk_buff, cb[4]) - 1:
9491                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
9492                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
9493                               offsetof(struct qdisc_skb_cb, data)) %
9494                              sizeof(__u64));
9495
9496                 prog->cb_access = 1;
9497                 off  = si->off;
9498                 off -= offsetof(struct __sk_buff, cb[0]);
9499                 off += offsetof(struct sk_buff, cb);
9500                 off += offsetof(struct qdisc_skb_cb, data);
9501                 if (type == BPF_WRITE)
9502                         *insn++ = BPF_EMIT_STORE(BPF_SIZE(si->code), si, off);
9503                 else
9504                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
9505                                               si->src_reg, off);
9506                 break;
9507
9508         case offsetof(struct __sk_buff, tc_classid):
9509                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
9510
9511                 off  = si->off;
9512                 off -= offsetof(struct __sk_buff, tc_classid);
9513                 off += offsetof(struct sk_buff, cb);
9514                 off += offsetof(struct qdisc_skb_cb, tc_classid);
9515                 *target_size = 2;
9516                 if (type == BPF_WRITE)
9517                         *insn++ = BPF_EMIT_STORE(BPF_H, si, off);
9518                 else
9519                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
9520                                               si->src_reg, off);
9521                 break;
9522
9523         case offsetof(struct __sk_buff, data):
9524                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9525                                       si->dst_reg, si->src_reg,
9526                                       offsetof(struct sk_buff, data));
9527                 break;
9528
9529         case offsetof(struct __sk_buff, data_meta):
9530                 off  = si->off;
9531                 off -= offsetof(struct __sk_buff, data_meta);
9532                 off += offsetof(struct sk_buff, cb);
9533                 off += offsetof(struct bpf_skb_data_end, data_meta);
9534                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9535                                       si->src_reg, off);
9536                 break;
9537
9538         case offsetof(struct __sk_buff, data_end):
9539                 off  = si->off;
9540                 off -= offsetof(struct __sk_buff, data_end);
9541                 off += offsetof(struct sk_buff, cb);
9542                 off += offsetof(struct bpf_skb_data_end, data_end);
9543                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9544                                       si->src_reg, off);
9545                 break;
9546
9547         case offsetof(struct __sk_buff, tc_index):
9548 #ifdef CONFIG_NET_SCHED
9549                 if (type == BPF_WRITE)
9550                         *insn++ = BPF_EMIT_STORE(BPF_H, si,
9551                                                  bpf_target_off(struct sk_buff, tc_index, 2,
9552                                                                 target_size));
9553                 else
9554                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9555                                               bpf_target_off(struct sk_buff, tc_index, 2,
9556                                                              target_size));
9557 #else
9558                 *target_size = 2;
9559                 if (type == BPF_WRITE)
9560                         *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
9561                 else
9562                         *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9563 #endif
9564                 break;
9565
9566         case offsetof(struct __sk_buff, napi_id):
9567 #if defined(CONFIG_NET_RX_BUSY_POLL)
9568                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9569                                       bpf_target_off(struct sk_buff, napi_id, 4,
9570                                                      target_size));
9571                 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
9572                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9573 #else
9574                 *target_size = 4;
9575                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9576 #endif
9577                 break;
9578         case offsetof(struct __sk_buff, family):
9579                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9580
9581                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9582                                       si->dst_reg, si->src_reg,
9583                                       offsetof(struct sk_buff, sk));
9584                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9585                                       bpf_target_off(struct sock_common,
9586                                                      skc_family,
9587                                                      2, target_size));
9588                 break;
9589         case offsetof(struct __sk_buff, remote_ip4):
9590                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9591
9592                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9593                                       si->dst_reg, si->src_reg,
9594                                       offsetof(struct sk_buff, sk));
9595                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9596                                       bpf_target_off(struct sock_common,
9597                                                      skc_daddr,
9598                                                      4, target_size));
9599                 break;
9600         case offsetof(struct __sk_buff, local_ip4):
9601                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9602                                           skc_rcv_saddr) != 4);
9603
9604                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9605                                       si->dst_reg, si->src_reg,
9606                                       offsetof(struct sk_buff, sk));
9607                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9608                                       bpf_target_off(struct sock_common,
9609                                                      skc_rcv_saddr,
9610                                                      4, target_size));
9611                 break;
9612         case offsetof(struct __sk_buff, remote_ip6[0]) ...
9613              offsetof(struct __sk_buff, remote_ip6[3]):
9614 #if IS_ENABLED(CONFIG_IPV6)
9615                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9616                                           skc_v6_daddr.s6_addr32[0]) != 4);
9617
9618                 off = si->off;
9619                 off -= offsetof(struct __sk_buff, remote_ip6[0]);
9620
9621                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9622                                       si->dst_reg, si->src_reg,
9623                                       offsetof(struct sk_buff, sk));
9624                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9625                                       offsetof(struct sock_common,
9626                                                skc_v6_daddr.s6_addr32[0]) +
9627                                       off);
9628 #else
9629                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9630 #endif
9631                 break;
9632         case offsetof(struct __sk_buff, local_ip6[0]) ...
9633              offsetof(struct __sk_buff, local_ip6[3]):
9634 #if IS_ENABLED(CONFIG_IPV6)
9635                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9636                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9637
9638                 off = si->off;
9639                 off -= offsetof(struct __sk_buff, local_ip6[0]);
9640
9641                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9642                                       si->dst_reg, si->src_reg,
9643                                       offsetof(struct sk_buff, sk));
9644                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9645                                       offsetof(struct sock_common,
9646                                                skc_v6_rcv_saddr.s6_addr32[0]) +
9647                                       off);
9648 #else
9649                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9650 #endif
9651                 break;
9652
9653         case offsetof(struct __sk_buff, remote_port):
9654                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9655
9656                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9657                                       si->dst_reg, si->src_reg,
9658                                       offsetof(struct sk_buff, sk));
9659                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9660                                       bpf_target_off(struct sock_common,
9661                                                      skc_dport,
9662                                                      2, target_size));
9663 #ifndef __BIG_ENDIAN_BITFIELD
9664                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9665 #endif
9666                 break;
9667
9668         case offsetof(struct __sk_buff, local_port):
9669                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9670
9671                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9672                                       si->dst_reg, si->src_reg,
9673                                       offsetof(struct sk_buff, sk));
9674                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9675                                       bpf_target_off(struct sock_common,
9676                                                      skc_num, 2, target_size));
9677                 break;
9678
9679         case offsetof(struct __sk_buff, tstamp):
9680                 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
9681
9682                 if (type == BPF_WRITE)
9683                         insn = bpf_convert_tstamp_write(prog, si, insn);
9684                 else
9685                         insn = bpf_convert_tstamp_read(prog, si, insn);
9686                 break;
9687
9688         case offsetof(struct __sk_buff, tstamp_type):
9689                 insn = bpf_convert_tstamp_type_read(si, insn);
9690                 break;
9691
9692         case offsetof(struct __sk_buff, gso_segs):
9693                 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9694                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
9695                                       si->dst_reg, si->dst_reg,
9696                                       bpf_target_off(struct skb_shared_info,
9697                                                      gso_segs, 2,
9698                                                      target_size));
9699                 break;
9700         case offsetof(struct __sk_buff, gso_size):
9701                 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9702                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
9703                                       si->dst_reg, si->dst_reg,
9704                                       bpf_target_off(struct skb_shared_info,
9705                                                      gso_size, 2,
9706                                                      target_size));
9707                 break;
9708         case offsetof(struct __sk_buff, wire_len):
9709                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
9710
9711                 off = si->off;
9712                 off -= offsetof(struct __sk_buff, wire_len);
9713                 off += offsetof(struct sk_buff, cb);
9714                 off += offsetof(struct qdisc_skb_cb, pkt_len);
9715                 *target_size = 4;
9716                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
9717                 break;
9718
9719         case offsetof(struct __sk_buff, sk):
9720                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9721                                       si->dst_reg, si->src_reg,
9722                                       offsetof(struct sk_buff, sk));
9723                 break;
9724         case offsetof(struct __sk_buff, hwtstamp):
9725                 BUILD_BUG_ON(sizeof_field(struct skb_shared_hwtstamps, hwtstamp) != 8);
9726                 BUILD_BUG_ON(offsetof(struct skb_shared_hwtstamps, hwtstamp) != 0);
9727
9728                 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9729                 *insn++ = BPF_LDX_MEM(BPF_DW,
9730                                       si->dst_reg, si->dst_reg,
9731                                       bpf_target_off(struct skb_shared_info,
9732                                                      hwtstamps, 8,
9733                                                      target_size));
9734                 break;
9735         }
9736
9737         return insn - insn_buf;
9738 }
9739
9740 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
9741                                 const struct bpf_insn *si,
9742                                 struct bpf_insn *insn_buf,
9743                                 struct bpf_prog *prog, u32 *target_size)
9744 {
9745         struct bpf_insn *insn = insn_buf;
9746         int off;
9747
9748         switch (si->off) {
9749         case offsetof(struct bpf_sock, bound_dev_if):
9750                 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
9751
9752                 if (type == BPF_WRITE)
9753                         *insn++ = BPF_EMIT_STORE(BPF_W, si,
9754                                                  offsetof(struct sock, sk_bound_dev_if));
9755                 else
9756                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9757                                       offsetof(struct sock, sk_bound_dev_if));
9758                 break;
9759
9760         case offsetof(struct bpf_sock, mark):
9761                 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
9762
9763                 if (type == BPF_WRITE)
9764                         *insn++ = BPF_EMIT_STORE(BPF_W, si,
9765                                                  offsetof(struct sock, sk_mark));
9766                 else
9767                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9768                                       offsetof(struct sock, sk_mark));
9769                 break;
9770
9771         case offsetof(struct bpf_sock, priority):
9772                 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
9773
9774                 if (type == BPF_WRITE)
9775                         *insn++ = BPF_EMIT_STORE(BPF_W, si,
9776                                                  offsetof(struct sock, sk_priority));
9777                 else
9778                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9779                                       offsetof(struct sock, sk_priority));
9780                 break;
9781
9782         case offsetof(struct bpf_sock, family):
9783                 *insn++ = BPF_LDX_MEM(
9784                         BPF_FIELD_SIZEOF(struct sock_common, skc_family),
9785                         si->dst_reg, si->src_reg,
9786                         bpf_target_off(struct sock_common,
9787                                        skc_family,
9788                                        sizeof_field(struct sock_common,
9789                                                     skc_family),
9790                                        target_size));
9791                 break;
9792
9793         case offsetof(struct bpf_sock, type):
9794                 *insn++ = BPF_LDX_MEM(
9795                         BPF_FIELD_SIZEOF(struct sock, sk_type),
9796                         si->dst_reg, si->src_reg,
9797                         bpf_target_off(struct sock, sk_type,
9798                                        sizeof_field(struct sock, sk_type),
9799                                        target_size));
9800                 break;
9801
9802         case offsetof(struct bpf_sock, protocol):
9803                 *insn++ = BPF_LDX_MEM(
9804                         BPF_FIELD_SIZEOF(struct sock, sk_protocol),
9805                         si->dst_reg, si->src_reg,
9806                         bpf_target_off(struct sock, sk_protocol,
9807                                        sizeof_field(struct sock, sk_protocol),
9808                                        target_size));
9809                 break;
9810
9811         case offsetof(struct bpf_sock, src_ip4):
9812                 *insn++ = BPF_LDX_MEM(
9813                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9814                         bpf_target_off(struct sock_common, skc_rcv_saddr,
9815                                        sizeof_field(struct sock_common,
9816                                                     skc_rcv_saddr),
9817                                        target_size));
9818                 break;
9819
9820         case offsetof(struct bpf_sock, dst_ip4):
9821                 *insn++ = BPF_LDX_MEM(
9822                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9823                         bpf_target_off(struct sock_common, skc_daddr,
9824                                        sizeof_field(struct sock_common,
9825                                                     skc_daddr),
9826                                        target_size));
9827                 break;
9828
9829         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
9830 #if IS_ENABLED(CONFIG_IPV6)
9831                 off = si->off;
9832                 off -= offsetof(struct bpf_sock, src_ip6[0]);
9833                 *insn++ = BPF_LDX_MEM(
9834                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9835                         bpf_target_off(
9836                                 struct sock_common,
9837                                 skc_v6_rcv_saddr.s6_addr32[0],
9838                                 sizeof_field(struct sock_common,
9839                                              skc_v6_rcv_saddr.s6_addr32[0]),
9840                                 target_size) + off);
9841 #else
9842                 (void)off;
9843                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9844 #endif
9845                 break;
9846
9847         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
9848 #if IS_ENABLED(CONFIG_IPV6)
9849                 off = si->off;
9850                 off -= offsetof(struct bpf_sock, dst_ip6[0]);
9851                 *insn++ = BPF_LDX_MEM(
9852                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9853                         bpf_target_off(struct sock_common,
9854                                        skc_v6_daddr.s6_addr32[0],
9855                                        sizeof_field(struct sock_common,
9856                                                     skc_v6_daddr.s6_addr32[0]),
9857                                        target_size) + off);
9858 #else
9859                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9860                 *target_size = 4;
9861 #endif
9862                 break;
9863
9864         case offsetof(struct bpf_sock, src_port):
9865                 *insn++ = BPF_LDX_MEM(
9866                         BPF_FIELD_SIZEOF(struct sock_common, skc_num),
9867                         si->dst_reg, si->src_reg,
9868                         bpf_target_off(struct sock_common, skc_num,
9869                                        sizeof_field(struct sock_common,
9870                                                     skc_num),
9871                                        target_size));
9872                 break;
9873
9874         case offsetof(struct bpf_sock, dst_port):
9875                 *insn++ = BPF_LDX_MEM(
9876                         BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
9877                         si->dst_reg, si->src_reg,
9878                         bpf_target_off(struct sock_common, skc_dport,
9879                                        sizeof_field(struct sock_common,
9880                                                     skc_dport),
9881                                        target_size));
9882                 break;
9883
9884         case offsetof(struct bpf_sock, state):
9885                 *insn++ = BPF_LDX_MEM(
9886                         BPF_FIELD_SIZEOF(struct sock_common, skc_state),
9887                         si->dst_reg, si->src_reg,
9888                         bpf_target_off(struct sock_common, skc_state,
9889                                        sizeof_field(struct sock_common,
9890                                                     skc_state),
9891                                        target_size));
9892                 break;
9893         case offsetof(struct bpf_sock, rx_queue_mapping):
9894 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
9895                 *insn++ = BPF_LDX_MEM(
9896                         BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
9897                         si->dst_reg, si->src_reg,
9898                         bpf_target_off(struct sock, sk_rx_queue_mapping,
9899                                        sizeof_field(struct sock,
9900                                                     sk_rx_queue_mapping),
9901                                        target_size));
9902                 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
9903                                       1);
9904                 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9905 #else
9906                 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9907                 *target_size = 2;
9908 #endif
9909                 break;
9910         }
9911
9912         return insn - insn_buf;
9913 }
9914
9915 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
9916                                          const struct bpf_insn *si,
9917                                          struct bpf_insn *insn_buf,
9918                                          struct bpf_prog *prog, u32 *target_size)
9919 {
9920         struct bpf_insn *insn = insn_buf;
9921
9922         switch (si->off) {
9923         case offsetof(struct __sk_buff, ifindex):
9924                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9925                                       si->dst_reg, si->src_reg,
9926                                       offsetof(struct sk_buff, dev));
9927                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9928                                       bpf_target_off(struct net_device, ifindex, 4,
9929                                                      target_size));
9930                 break;
9931         default:
9932                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9933                                               target_size);
9934         }
9935
9936         return insn - insn_buf;
9937 }
9938
9939 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
9940                                   const struct bpf_insn *si,
9941                                   struct bpf_insn *insn_buf,
9942                                   struct bpf_prog *prog, u32 *target_size)
9943 {
9944         struct bpf_insn *insn = insn_buf;
9945
9946         switch (si->off) {
9947         case offsetof(struct xdp_md, data):
9948                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
9949                                       si->dst_reg, si->src_reg,
9950                                       offsetof(struct xdp_buff, data));
9951                 break;
9952         case offsetof(struct xdp_md, data_meta):
9953                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
9954                                       si->dst_reg, si->src_reg,
9955                                       offsetof(struct xdp_buff, data_meta));
9956                 break;
9957         case offsetof(struct xdp_md, data_end):
9958                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
9959                                       si->dst_reg, si->src_reg,
9960                                       offsetof(struct xdp_buff, data_end));
9961                 break;
9962         case offsetof(struct xdp_md, ingress_ifindex):
9963                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9964                                       si->dst_reg, si->src_reg,
9965                                       offsetof(struct xdp_buff, rxq));
9966                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
9967                                       si->dst_reg, si->dst_reg,
9968                                       offsetof(struct xdp_rxq_info, dev));
9969                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9970                                       offsetof(struct net_device, ifindex));
9971                 break;
9972         case offsetof(struct xdp_md, rx_queue_index):
9973                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9974                                       si->dst_reg, si->src_reg,
9975                                       offsetof(struct xdp_buff, rxq));
9976                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9977                                       offsetof(struct xdp_rxq_info,
9978                                                queue_index));
9979                 break;
9980         case offsetof(struct xdp_md, egress_ifindex):
9981                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
9982                                       si->dst_reg, si->src_reg,
9983                                       offsetof(struct xdp_buff, txq));
9984                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
9985                                       si->dst_reg, si->dst_reg,
9986                                       offsetof(struct xdp_txq_info, dev));
9987                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9988                                       offsetof(struct net_device, ifindex));
9989                 break;
9990         }
9991
9992         return insn - insn_buf;
9993 }
9994
9995 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
9996  * context Structure, F is Field in context structure that contains a pointer
9997  * to Nested Structure of type NS that has the field NF.
9998  *
9999  * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
10000  * sure that SIZE is not greater than actual size of S.F.NF.
10001  *
10002  * If offset OFF is provided, the load happens from that offset relative to
10003  * offset of NF.
10004  */
10005 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF)          \
10006         do {                                                                   \
10007                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg,     \
10008                                       si->src_reg, offsetof(S, F));            \
10009                 *insn++ = BPF_LDX_MEM(                                         \
10010                         SIZE, si->dst_reg, si->dst_reg,                        \
10011                         bpf_target_off(NS, NF, sizeof_field(NS, NF),           \
10012                                        target_size)                            \
10013                                 + OFF);                                        \
10014         } while (0)
10015
10016 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF)                              \
10017         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF,                     \
10018                                              BPF_FIELD_SIZEOF(NS, NF), 0)
10019
10020 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
10021  * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
10022  *
10023  * In addition it uses Temporary Field TF (member of struct S) as the 3rd
10024  * "register" since two registers available in convert_ctx_access are not
10025  * enough: we can't override neither SRC, since it contains value to store, nor
10026  * DST since it contains pointer to context that may be used by later
10027  * instructions. But we need a temporary place to save pointer to nested
10028  * structure whose field we want to store to.
10029  */
10030 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF)          \
10031         do {                                                                   \
10032                 int tmp_reg = BPF_REG_9;                                       \
10033                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
10034                         --tmp_reg;                                             \
10035                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
10036                         --tmp_reg;                                             \
10037                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg,            \
10038                                       offsetof(S, TF));                        \
10039                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg,         \
10040                                       si->dst_reg, offsetof(S, F));            \
10041                 *insn++ = BPF_RAW_INSN(SIZE | BPF_MEM | BPF_CLASS(si->code),   \
10042                                        tmp_reg, si->src_reg,                   \
10043                         bpf_target_off(NS, NF, sizeof_field(NS, NF),           \
10044                                        target_size)                            \
10045                                        + OFF,                                  \
10046                                        si->imm);                               \
10047                 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg,            \
10048                                       offsetof(S, TF));                        \
10049         } while (0)
10050
10051 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
10052                                                       TF)                      \
10053         do {                                                                   \
10054                 if (type == BPF_WRITE) {                                       \
10055                         SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE,   \
10056                                                          OFF, TF);             \
10057                 } else {                                                       \
10058                         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(                  \
10059                                 S, NS, F, NF, SIZE, OFF);  \
10060                 }                                                              \
10061         } while (0)
10062
10063 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF)                 \
10064         SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(                         \
10065                 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
10066
10067 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
10068                                         const struct bpf_insn *si,
10069                                         struct bpf_insn *insn_buf,
10070                                         struct bpf_prog *prog, u32 *target_size)
10071 {
10072         int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
10073         struct bpf_insn *insn = insn_buf;
10074
10075         switch (si->off) {
10076         case offsetof(struct bpf_sock_addr, user_family):
10077                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10078                                             struct sockaddr, uaddr, sa_family);
10079                 break;
10080
10081         case offsetof(struct bpf_sock_addr, user_ip4):
10082                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10083                         struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
10084                         sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
10085                 break;
10086
10087         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
10088                 off = si->off;
10089                 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
10090                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10091                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
10092                         sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
10093                         tmp_reg);
10094                 break;
10095
10096         case offsetof(struct bpf_sock_addr, user_port):
10097                 /* To get port we need to know sa_family first and then treat
10098                  * sockaddr as either sockaddr_in or sockaddr_in6.
10099                  * Though we can simplify since port field has same offset and
10100                  * size in both structures.
10101                  * Here we check this invariant and use just one of the
10102                  * structures if it's true.
10103                  */
10104                 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
10105                              offsetof(struct sockaddr_in6, sin6_port));
10106                 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
10107                              sizeof_field(struct sockaddr_in6, sin6_port));
10108                 /* Account for sin6_port being smaller than user_port. */
10109                 port_size = min(port_size, BPF_LDST_BYTES(si));
10110                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10111                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
10112                         sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
10113                 break;
10114
10115         case offsetof(struct bpf_sock_addr, family):
10116                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10117                                             struct sock, sk, sk_family);
10118                 break;
10119
10120         case offsetof(struct bpf_sock_addr, type):
10121                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10122                                             struct sock, sk, sk_type);
10123                 break;
10124
10125         case offsetof(struct bpf_sock_addr, protocol):
10126                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10127                                             struct sock, sk, sk_protocol);
10128                 break;
10129
10130         case offsetof(struct bpf_sock_addr, msg_src_ip4):
10131                 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
10132                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10133                         struct bpf_sock_addr_kern, struct in_addr, t_ctx,
10134                         s_addr, BPF_SIZE(si->code), 0, tmp_reg);
10135                 break;
10136
10137         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
10138                                 msg_src_ip6[3]):
10139                 off = si->off;
10140                 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
10141                 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
10142                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10143                         struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
10144                         s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
10145                 break;
10146         case offsetof(struct bpf_sock_addr, sk):
10147                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
10148                                       si->dst_reg, si->src_reg,
10149                                       offsetof(struct bpf_sock_addr_kern, sk));
10150                 break;
10151         }
10152
10153         return insn - insn_buf;
10154 }
10155
10156 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
10157                                        const struct bpf_insn *si,
10158                                        struct bpf_insn *insn_buf,
10159                                        struct bpf_prog *prog,
10160                                        u32 *target_size)
10161 {
10162         struct bpf_insn *insn = insn_buf;
10163         int off;
10164
10165 /* Helper macro for adding read access to tcp_sock or sock fields. */
10166 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
10167         do {                                                                  \
10168                 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2;     \
10169                 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) >                   \
10170                              sizeof_field(struct bpf_sock_ops, BPF_FIELD));   \
10171                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10172                         reg--;                                                \
10173                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10174                         reg--;                                                \
10175                 if (si->dst_reg == si->src_reg) {                             \
10176                         *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg,       \
10177                                           offsetof(struct bpf_sock_ops_kern,  \
10178                                           temp));                             \
10179                         fullsock_reg = reg;                                   \
10180                         jmp += 2;                                             \
10181                 }                                                             \
10182                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10183                                                 struct bpf_sock_ops_kern,     \
10184                                                 is_fullsock),                 \
10185                                       fullsock_reg, si->src_reg,              \
10186                                       offsetof(struct bpf_sock_ops_kern,      \
10187                                                is_fullsock));                 \
10188                 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp);         \
10189                 if (si->dst_reg == si->src_reg)                               \
10190                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
10191                                       offsetof(struct bpf_sock_ops_kern,      \
10192                                       temp));                                 \
10193                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10194                                                 struct bpf_sock_ops_kern, sk),\
10195                                       si->dst_reg, si->src_reg,               \
10196                                       offsetof(struct bpf_sock_ops_kern, sk));\
10197                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ,                   \
10198                                                        OBJ_FIELD),            \
10199                                       si->dst_reg, si->dst_reg,               \
10200                                       offsetof(OBJ, OBJ_FIELD));              \
10201                 if (si->dst_reg == si->src_reg) {                             \
10202                         *insn++ = BPF_JMP_A(1);                               \
10203                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
10204                                       offsetof(struct bpf_sock_ops_kern,      \
10205                                       temp));                                 \
10206                 }                                                             \
10207         } while (0)
10208
10209 #define SOCK_OPS_GET_SK()                                                             \
10210         do {                                                                  \
10211                 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1;     \
10212                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10213                         reg--;                                                \
10214                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10215                         reg--;                                                \
10216                 if (si->dst_reg == si->src_reg) {                             \
10217                         *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg,       \
10218                                           offsetof(struct bpf_sock_ops_kern,  \
10219                                           temp));                             \
10220                         fullsock_reg = reg;                                   \
10221                         jmp += 2;                                             \
10222                 }                                                             \
10223                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10224                                                 struct bpf_sock_ops_kern,     \
10225                                                 is_fullsock),                 \
10226                                       fullsock_reg, si->src_reg,              \
10227                                       offsetof(struct bpf_sock_ops_kern,      \
10228                                                is_fullsock));                 \
10229                 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp);         \
10230                 if (si->dst_reg == si->src_reg)                               \
10231                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
10232                                       offsetof(struct bpf_sock_ops_kern,      \
10233                                       temp));                                 \
10234                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10235                                                 struct bpf_sock_ops_kern, sk),\
10236                                       si->dst_reg, si->src_reg,               \
10237                                       offsetof(struct bpf_sock_ops_kern, sk));\
10238                 if (si->dst_reg == si->src_reg) {                             \
10239                         *insn++ = BPF_JMP_A(1);                               \
10240                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
10241                                       offsetof(struct bpf_sock_ops_kern,      \
10242                                       temp));                                 \
10243                 }                                                             \
10244         } while (0)
10245
10246 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
10247                 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
10248
10249 /* Helper macro for adding write access to tcp_sock or sock fields.
10250  * The macro is called with two registers, dst_reg which contains a pointer
10251  * to ctx (context) and src_reg which contains the value that should be
10252  * stored. However, we need an additional register since we cannot overwrite
10253  * dst_reg because it may be used later in the program.
10254  * Instead we "borrow" one of the other register. We first save its value
10255  * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
10256  * it at the end of the macro.
10257  */
10258 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
10259         do {                                                                  \
10260                 int reg = BPF_REG_9;                                          \
10261                 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) >                   \
10262                              sizeof_field(struct bpf_sock_ops, BPF_FIELD));   \
10263                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10264                         reg--;                                                \
10265                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10266                         reg--;                                                \
10267                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg,               \
10268                                       offsetof(struct bpf_sock_ops_kern,      \
10269                                                temp));                        \
10270                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10271                                                 struct bpf_sock_ops_kern,     \
10272                                                 is_fullsock),                 \
10273                                       reg, si->dst_reg,                       \
10274                                       offsetof(struct bpf_sock_ops_kern,      \
10275                                                is_fullsock));                 \
10276                 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2);                    \
10277                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10278                                                 struct bpf_sock_ops_kern, sk),\
10279                                       reg, si->dst_reg,                       \
10280                                       offsetof(struct bpf_sock_ops_kern, sk));\
10281                 *insn++ = BPF_RAW_INSN(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD) |     \
10282                                        BPF_MEM | BPF_CLASS(si->code),         \
10283                                        reg, si->src_reg,                      \
10284                                        offsetof(OBJ, OBJ_FIELD),              \
10285                                        si->imm);                              \
10286                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg,               \
10287                                       offsetof(struct bpf_sock_ops_kern,      \
10288                                                temp));                        \
10289         } while (0)
10290
10291 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE)            \
10292         do {                                                                  \
10293                 if (TYPE == BPF_WRITE)                                        \
10294                         SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
10295                 else                                                          \
10296                         SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
10297         } while (0)
10298
10299         switch (si->off) {
10300         case offsetof(struct bpf_sock_ops, op):
10301                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10302                                                        op),
10303                                       si->dst_reg, si->src_reg,
10304                                       offsetof(struct bpf_sock_ops_kern, op));
10305                 break;
10306
10307         case offsetof(struct bpf_sock_ops, replylong[0]) ...
10308              offsetof(struct bpf_sock_ops, replylong[3]):
10309                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
10310                              sizeof_field(struct bpf_sock_ops_kern, reply));
10311                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
10312                              sizeof_field(struct bpf_sock_ops_kern, replylong));
10313                 off = si->off;
10314                 off -= offsetof(struct bpf_sock_ops, replylong[0]);
10315                 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
10316                 if (type == BPF_WRITE)
10317                         *insn++ = BPF_EMIT_STORE(BPF_W, si, off);
10318                 else
10319                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10320                                               off);
10321                 break;
10322
10323         case offsetof(struct bpf_sock_ops, family):
10324                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10325
10326                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10327                                               struct bpf_sock_ops_kern, sk),
10328                                       si->dst_reg, si->src_reg,
10329                                       offsetof(struct bpf_sock_ops_kern, sk));
10330                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10331                                       offsetof(struct sock_common, skc_family));
10332                 break;
10333
10334         case offsetof(struct bpf_sock_ops, remote_ip4):
10335                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10336
10337                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10338                                                 struct bpf_sock_ops_kern, sk),
10339                                       si->dst_reg, si->src_reg,
10340                                       offsetof(struct bpf_sock_ops_kern, sk));
10341                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10342                                       offsetof(struct sock_common, skc_daddr));
10343                 break;
10344
10345         case offsetof(struct bpf_sock_ops, local_ip4):
10346                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10347                                           skc_rcv_saddr) != 4);
10348
10349                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10350                                               struct bpf_sock_ops_kern, sk),
10351                                       si->dst_reg, si->src_reg,
10352                                       offsetof(struct bpf_sock_ops_kern, sk));
10353                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10354                                       offsetof(struct sock_common,
10355                                                skc_rcv_saddr));
10356                 break;
10357
10358         case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
10359              offsetof(struct bpf_sock_ops, remote_ip6[3]):
10360 #if IS_ENABLED(CONFIG_IPV6)
10361                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10362                                           skc_v6_daddr.s6_addr32[0]) != 4);
10363
10364                 off = si->off;
10365                 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
10366                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10367                                                 struct bpf_sock_ops_kern, sk),
10368                                       si->dst_reg, si->src_reg,
10369                                       offsetof(struct bpf_sock_ops_kern, sk));
10370                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10371                                       offsetof(struct sock_common,
10372                                                skc_v6_daddr.s6_addr32[0]) +
10373                                       off);
10374 #else
10375                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10376 #endif
10377                 break;
10378
10379         case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
10380              offsetof(struct bpf_sock_ops, local_ip6[3]):
10381 #if IS_ENABLED(CONFIG_IPV6)
10382                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10383                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10384
10385                 off = si->off;
10386                 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
10387                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10388                                                 struct bpf_sock_ops_kern, sk),
10389                                       si->dst_reg, si->src_reg,
10390                                       offsetof(struct bpf_sock_ops_kern, sk));
10391                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10392                                       offsetof(struct sock_common,
10393                                                skc_v6_rcv_saddr.s6_addr32[0]) +
10394                                       off);
10395 #else
10396                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10397 #endif
10398                 break;
10399
10400         case offsetof(struct bpf_sock_ops, remote_port):
10401                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10402
10403                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10404                                                 struct bpf_sock_ops_kern, sk),
10405                                       si->dst_reg, si->src_reg,
10406                                       offsetof(struct bpf_sock_ops_kern, sk));
10407                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10408                                       offsetof(struct sock_common, skc_dport));
10409 #ifndef __BIG_ENDIAN_BITFIELD
10410                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10411 #endif
10412                 break;
10413
10414         case offsetof(struct bpf_sock_ops, local_port):
10415                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10416
10417                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10418                                                 struct bpf_sock_ops_kern, sk),
10419                                       si->dst_reg, si->src_reg,
10420                                       offsetof(struct bpf_sock_ops_kern, sk));
10421                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10422                                       offsetof(struct sock_common, skc_num));
10423                 break;
10424
10425         case offsetof(struct bpf_sock_ops, is_fullsock):
10426                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10427                                                 struct bpf_sock_ops_kern,
10428                                                 is_fullsock),
10429                                       si->dst_reg, si->src_reg,
10430                                       offsetof(struct bpf_sock_ops_kern,
10431                                                is_fullsock));
10432                 break;
10433
10434         case offsetof(struct bpf_sock_ops, state):
10435                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
10436
10437                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10438                                                 struct bpf_sock_ops_kern, sk),
10439                                       si->dst_reg, si->src_reg,
10440                                       offsetof(struct bpf_sock_ops_kern, sk));
10441                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
10442                                       offsetof(struct sock_common, skc_state));
10443                 break;
10444
10445         case offsetof(struct bpf_sock_ops, rtt_min):
10446                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
10447                              sizeof(struct minmax));
10448                 BUILD_BUG_ON(sizeof(struct minmax) <
10449                              sizeof(struct minmax_sample));
10450
10451                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10452                                                 struct bpf_sock_ops_kern, sk),
10453                                       si->dst_reg, si->src_reg,
10454                                       offsetof(struct bpf_sock_ops_kern, sk));
10455                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10456                                       offsetof(struct tcp_sock, rtt_min) +
10457                                       sizeof_field(struct minmax_sample, t));
10458                 break;
10459
10460         case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
10461                 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
10462                                    struct tcp_sock);
10463                 break;
10464
10465         case offsetof(struct bpf_sock_ops, sk_txhash):
10466                 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
10467                                           struct sock, type);
10468                 break;
10469         case offsetof(struct bpf_sock_ops, snd_cwnd):
10470                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
10471                 break;
10472         case offsetof(struct bpf_sock_ops, srtt_us):
10473                 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
10474                 break;
10475         case offsetof(struct bpf_sock_ops, snd_ssthresh):
10476                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
10477                 break;
10478         case offsetof(struct bpf_sock_ops, rcv_nxt):
10479                 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
10480                 break;
10481         case offsetof(struct bpf_sock_ops, snd_nxt):
10482                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
10483                 break;
10484         case offsetof(struct bpf_sock_ops, snd_una):
10485                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
10486                 break;
10487         case offsetof(struct bpf_sock_ops, mss_cache):
10488                 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
10489                 break;
10490         case offsetof(struct bpf_sock_ops, ecn_flags):
10491                 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
10492                 break;
10493         case offsetof(struct bpf_sock_ops, rate_delivered):
10494                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
10495                 break;
10496         case offsetof(struct bpf_sock_ops, rate_interval_us):
10497                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
10498                 break;
10499         case offsetof(struct bpf_sock_ops, packets_out):
10500                 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
10501                 break;
10502         case offsetof(struct bpf_sock_ops, retrans_out):
10503                 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
10504                 break;
10505         case offsetof(struct bpf_sock_ops, total_retrans):
10506                 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
10507                 break;
10508         case offsetof(struct bpf_sock_ops, segs_in):
10509                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
10510                 break;
10511         case offsetof(struct bpf_sock_ops, data_segs_in):
10512                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
10513                 break;
10514         case offsetof(struct bpf_sock_ops, segs_out):
10515                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
10516                 break;
10517         case offsetof(struct bpf_sock_ops, data_segs_out):
10518                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
10519                 break;
10520         case offsetof(struct bpf_sock_ops, lost_out):
10521                 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
10522                 break;
10523         case offsetof(struct bpf_sock_ops, sacked_out):
10524                 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
10525                 break;
10526         case offsetof(struct bpf_sock_ops, bytes_received):
10527                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
10528                 break;
10529         case offsetof(struct bpf_sock_ops, bytes_acked):
10530                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
10531                 break;
10532         case offsetof(struct bpf_sock_ops, sk):
10533                 SOCK_OPS_GET_SK();
10534                 break;
10535         case offsetof(struct bpf_sock_ops, skb_data_end):
10536                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10537                                                        skb_data_end),
10538                                       si->dst_reg, si->src_reg,
10539                                       offsetof(struct bpf_sock_ops_kern,
10540                                                skb_data_end));
10541                 break;
10542         case offsetof(struct bpf_sock_ops, skb_data):
10543                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10544                                                        skb),
10545                                       si->dst_reg, si->src_reg,
10546                                       offsetof(struct bpf_sock_ops_kern,
10547                                                skb));
10548                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10549                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10550                                       si->dst_reg, si->dst_reg,
10551                                       offsetof(struct sk_buff, data));
10552                 break;
10553         case offsetof(struct bpf_sock_ops, skb_len):
10554                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10555                                                        skb),
10556                                       si->dst_reg, si->src_reg,
10557                                       offsetof(struct bpf_sock_ops_kern,
10558                                                skb));
10559                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10560                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10561                                       si->dst_reg, si->dst_reg,
10562                                       offsetof(struct sk_buff, len));
10563                 break;
10564         case offsetof(struct bpf_sock_ops, skb_tcp_flags):
10565                 off = offsetof(struct sk_buff, cb);
10566                 off += offsetof(struct tcp_skb_cb, tcp_flags);
10567                 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
10568                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10569                                                        skb),
10570                                       si->dst_reg, si->src_reg,
10571                                       offsetof(struct bpf_sock_ops_kern,
10572                                                skb));
10573                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10574                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
10575                                                        tcp_flags),
10576                                       si->dst_reg, si->dst_reg, off);
10577                 break;
10578         case offsetof(struct bpf_sock_ops, skb_hwtstamp): {
10579                 struct bpf_insn *jmp_on_null_skb;
10580
10581                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10582                                                        skb),
10583                                       si->dst_reg, si->src_reg,
10584                                       offsetof(struct bpf_sock_ops_kern,
10585                                                skb));
10586                 /* Reserve one insn to test skb == NULL */
10587                 jmp_on_null_skb = insn++;
10588                 insn = bpf_convert_shinfo_access(si->dst_reg, si->dst_reg, insn);
10589                 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
10590                                       bpf_target_off(struct skb_shared_info,
10591                                                      hwtstamps, 8,
10592                                                      target_size));
10593                 *jmp_on_null_skb = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0,
10594                                                insn - jmp_on_null_skb - 1);
10595                 break;
10596         }
10597         }
10598         return insn - insn_buf;
10599 }
10600
10601 /* data_end = skb->data + skb_headlen() */
10602 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
10603                                                     struct bpf_insn *insn)
10604 {
10605         int reg;
10606         int temp_reg_off = offsetof(struct sk_buff, cb) +
10607                            offsetof(struct sk_skb_cb, temp_reg);
10608
10609         if (si->src_reg == si->dst_reg) {
10610                 /* We need an extra register, choose and save a register. */
10611                 reg = BPF_REG_9;
10612                 if (si->src_reg == reg || si->dst_reg == reg)
10613                         reg--;
10614                 if (si->src_reg == reg || si->dst_reg == reg)
10615                         reg--;
10616                 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off);
10617         } else {
10618                 reg = si->dst_reg;
10619         }
10620
10621         /* reg = skb->data */
10622         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10623                               reg, si->src_reg,
10624                               offsetof(struct sk_buff, data));
10625         /* AX = skb->len */
10626         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10627                               BPF_REG_AX, si->src_reg,
10628                               offsetof(struct sk_buff, len));
10629         /* reg = skb->data + skb->len */
10630         *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX);
10631         /* AX = skb->data_len */
10632         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
10633                               BPF_REG_AX, si->src_reg,
10634                               offsetof(struct sk_buff, data_len));
10635
10636         /* reg = skb->data + skb->len - skb->data_len */
10637         *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX);
10638
10639         if (si->src_reg == si->dst_reg) {
10640                 /* Restore the saved register */
10641                 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg);
10642                 *insn++ = BPF_MOV64_REG(si->dst_reg, reg);
10643                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off);
10644         }
10645
10646         return insn;
10647 }
10648
10649 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
10650                                      const struct bpf_insn *si,
10651                                      struct bpf_insn *insn_buf,
10652                                      struct bpf_prog *prog, u32 *target_size)
10653 {
10654         struct bpf_insn *insn = insn_buf;
10655         int off;
10656
10657         switch (si->off) {
10658         case offsetof(struct __sk_buff, data_end):
10659                 insn = bpf_convert_data_end_access(si, insn);
10660                 break;
10661         case offsetof(struct __sk_buff, cb[0]) ...
10662              offsetofend(struct __sk_buff, cb[4]) - 1:
10663                 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20);
10664                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
10665                               offsetof(struct sk_skb_cb, data)) %
10666                              sizeof(__u64));
10667
10668                 prog->cb_access = 1;
10669                 off  = si->off;
10670                 off -= offsetof(struct __sk_buff, cb[0]);
10671                 off += offsetof(struct sk_buff, cb);
10672                 off += offsetof(struct sk_skb_cb, data);
10673                 if (type == BPF_WRITE)
10674                         *insn++ = BPF_EMIT_STORE(BPF_SIZE(si->code), si, off);
10675                 else
10676                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
10677                                               si->src_reg, off);
10678                 break;
10679
10680
10681         default:
10682                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
10683                                               target_size);
10684         }
10685
10686         return insn - insn_buf;
10687 }
10688
10689 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
10690                                      const struct bpf_insn *si,
10691                                      struct bpf_insn *insn_buf,
10692                                      struct bpf_prog *prog, u32 *target_size)
10693 {
10694         struct bpf_insn *insn = insn_buf;
10695 #if IS_ENABLED(CONFIG_IPV6)
10696         int off;
10697 #endif
10698
10699         /* convert ctx uses the fact sg element is first in struct */
10700         BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
10701
10702         switch (si->off) {
10703         case offsetof(struct sk_msg_md, data):
10704                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
10705                                       si->dst_reg, si->src_reg,
10706                                       offsetof(struct sk_msg, data));
10707                 break;
10708         case offsetof(struct sk_msg_md, data_end):
10709                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
10710                                       si->dst_reg, si->src_reg,
10711                                       offsetof(struct sk_msg, data_end));
10712                 break;
10713         case offsetof(struct sk_msg_md, family):
10714                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10715
10716                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10717                                               struct sk_msg, sk),
10718                                       si->dst_reg, si->src_reg,
10719                                       offsetof(struct sk_msg, sk));
10720                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10721                                       offsetof(struct sock_common, skc_family));
10722                 break;
10723
10724         case offsetof(struct sk_msg_md, remote_ip4):
10725                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10726
10727                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10728                                                 struct sk_msg, sk),
10729                                       si->dst_reg, si->src_reg,
10730                                       offsetof(struct sk_msg, sk));
10731                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10732                                       offsetof(struct sock_common, skc_daddr));
10733                 break;
10734
10735         case offsetof(struct sk_msg_md, local_ip4):
10736                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10737                                           skc_rcv_saddr) != 4);
10738
10739                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10740                                               struct sk_msg, sk),
10741                                       si->dst_reg, si->src_reg,
10742                                       offsetof(struct sk_msg, sk));
10743                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10744                                       offsetof(struct sock_common,
10745                                                skc_rcv_saddr));
10746                 break;
10747
10748         case offsetof(struct sk_msg_md, remote_ip6[0]) ...
10749              offsetof(struct sk_msg_md, remote_ip6[3]):
10750 #if IS_ENABLED(CONFIG_IPV6)
10751                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10752                                           skc_v6_daddr.s6_addr32[0]) != 4);
10753
10754                 off = si->off;
10755                 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
10756                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10757                                                 struct sk_msg, sk),
10758                                       si->dst_reg, si->src_reg,
10759                                       offsetof(struct sk_msg, sk));
10760                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10761                                       offsetof(struct sock_common,
10762                                                skc_v6_daddr.s6_addr32[0]) +
10763                                       off);
10764 #else
10765                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10766 #endif
10767                 break;
10768
10769         case offsetof(struct sk_msg_md, local_ip6[0]) ...
10770              offsetof(struct sk_msg_md, local_ip6[3]):
10771 #if IS_ENABLED(CONFIG_IPV6)
10772                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10773                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10774
10775                 off = si->off;
10776                 off -= offsetof(struct sk_msg_md, local_ip6[0]);
10777                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10778                                                 struct sk_msg, sk),
10779                                       si->dst_reg, si->src_reg,
10780                                       offsetof(struct sk_msg, sk));
10781                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10782                                       offsetof(struct sock_common,
10783                                                skc_v6_rcv_saddr.s6_addr32[0]) +
10784                                       off);
10785 #else
10786                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10787 #endif
10788                 break;
10789
10790         case offsetof(struct sk_msg_md, remote_port):
10791                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10792
10793                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10794                                                 struct sk_msg, sk),
10795                                       si->dst_reg, si->src_reg,
10796                                       offsetof(struct sk_msg, sk));
10797                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10798                                       offsetof(struct sock_common, skc_dport));
10799 #ifndef __BIG_ENDIAN_BITFIELD
10800                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10801 #endif
10802                 break;
10803
10804         case offsetof(struct sk_msg_md, local_port):
10805                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10806
10807                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10808                                                 struct sk_msg, sk),
10809                                       si->dst_reg, si->src_reg,
10810                                       offsetof(struct sk_msg, sk));
10811                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10812                                       offsetof(struct sock_common, skc_num));
10813                 break;
10814
10815         case offsetof(struct sk_msg_md, size):
10816                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
10817                                       si->dst_reg, si->src_reg,
10818                                       offsetof(struct sk_msg_sg, size));
10819                 break;
10820
10821         case offsetof(struct sk_msg_md, sk):
10822                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
10823                                       si->dst_reg, si->src_reg,
10824                                       offsetof(struct sk_msg, sk));
10825                 break;
10826         }
10827
10828         return insn - insn_buf;
10829 }
10830
10831 const struct bpf_verifier_ops sk_filter_verifier_ops = {
10832         .get_func_proto         = sk_filter_func_proto,
10833         .is_valid_access        = sk_filter_is_valid_access,
10834         .convert_ctx_access     = bpf_convert_ctx_access,
10835         .gen_ld_abs             = bpf_gen_ld_abs,
10836 };
10837
10838 const struct bpf_prog_ops sk_filter_prog_ops = {
10839         .test_run               = bpf_prog_test_run_skb,
10840 };
10841
10842 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
10843         .get_func_proto         = tc_cls_act_func_proto,
10844         .is_valid_access        = tc_cls_act_is_valid_access,
10845         .convert_ctx_access     = tc_cls_act_convert_ctx_access,
10846         .gen_prologue           = tc_cls_act_prologue,
10847         .gen_ld_abs             = bpf_gen_ld_abs,
10848         .btf_struct_access      = tc_cls_act_btf_struct_access,
10849 };
10850
10851 const struct bpf_prog_ops tc_cls_act_prog_ops = {
10852         .test_run               = bpf_prog_test_run_skb,
10853 };
10854
10855 const struct bpf_verifier_ops xdp_verifier_ops = {
10856         .get_func_proto         = xdp_func_proto,
10857         .is_valid_access        = xdp_is_valid_access,
10858         .convert_ctx_access     = xdp_convert_ctx_access,
10859         .gen_prologue           = bpf_noop_prologue,
10860         .btf_struct_access      = xdp_btf_struct_access,
10861 };
10862
10863 const struct bpf_prog_ops xdp_prog_ops = {
10864         .test_run               = bpf_prog_test_run_xdp,
10865 };
10866
10867 const struct bpf_verifier_ops cg_skb_verifier_ops = {
10868         .get_func_proto         = cg_skb_func_proto,
10869         .is_valid_access        = cg_skb_is_valid_access,
10870         .convert_ctx_access     = bpf_convert_ctx_access,
10871 };
10872
10873 const struct bpf_prog_ops cg_skb_prog_ops = {
10874         .test_run               = bpf_prog_test_run_skb,
10875 };
10876
10877 const struct bpf_verifier_ops lwt_in_verifier_ops = {
10878         .get_func_proto         = lwt_in_func_proto,
10879         .is_valid_access        = lwt_is_valid_access,
10880         .convert_ctx_access     = bpf_convert_ctx_access,
10881 };
10882
10883 const struct bpf_prog_ops lwt_in_prog_ops = {
10884         .test_run               = bpf_prog_test_run_skb,
10885 };
10886
10887 const struct bpf_verifier_ops lwt_out_verifier_ops = {
10888         .get_func_proto         = lwt_out_func_proto,
10889         .is_valid_access        = lwt_is_valid_access,
10890         .convert_ctx_access     = bpf_convert_ctx_access,
10891 };
10892
10893 const struct bpf_prog_ops lwt_out_prog_ops = {
10894         .test_run               = bpf_prog_test_run_skb,
10895 };
10896
10897 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
10898         .get_func_proto         = lwt_xmit_func_proto,
10899         .is_valid_access        = lwt_is_valid_access,
10900         .convert_ctx_access     = bpf_convert_ctx_access,
10901         .gen_prologue           = tc_cls_act_prologue,
10902 };
10903
10904 const struct bpf_prog_ops lwt_xmit_prog_ops = {
10905         .test_run               = bpf_prog_test_run_skb,
10906 };
10907
10908 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
10909         .get_func_proto         = lwt_seg6local_func_proto,
10910         .is_valid_access        = lwt_is_valid_access,
10911         .convert_ctx_access     = bpf_convert_ctx_access,
10912 };
10913
10914 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
10915         .test_run               = bpf_prog_test_run_skb,
10916 };
10917
10918 const struct bpf_verifier_ops cg_sock_verifier_ops = {
10919         .get_func_proto         = sock_filter_func_proto,
10920         .is_valid_access        = sock_filter_is_valid_access,
10921         .convert_ctx_access     = bpf_sock_convert_ctx_access,
10922 };
10923
10924 const struct bpf_prog_ops cg_sock_prog_ops = {
10925 };
10926
10927 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
10928         .get_func_proto         = sock_addr_func_proto,
10929         .is_valid_access        = sock_addr_is_valid_access,
10930         .convert_ctx_access     = sock_addr_convert_ctx_access,
10931 };
10932
10933 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
10934 };
10935
10936 const struct bpf_verifier_ops sock_ops_verifier_ops = {
10937         .get_func_proto         = sock_ops_func_proto,
10938         .is_valid_access        = sock_ops_is_valid_access,
10939         .convert_ctx_access     = sock_ops_convert_ctx_access,
10940 };
10941
10942 const struct bpf_prog_ops sock_ops_prog_ops = {
10943 };
10944
10945 const struct bpf_verifier_ops sk_skb_verifier_ops = {
10946         .get_func_proto         = sk_skb_func_proto,
10947         .is_valid_access        = sk_skb_is_valid_access,
10948         .convert_ctx_access     = sk_skb_convert_ctx_access,
10949         .gen_prologue           = sk_skb_prologue,
10950 };
10951
10952 const struct bpf_prog_ops sk_skb_prog_ops = {
10953 };
10954
10955 const struct bpf_verifier_ops sk_msg_verifier_ops = {
10956         .get_func_proto         = sk_msg_func_proto,
10957         .is_valid_access        = sk_msg_is_valid_access,
10958         .convert_ctx_access     = sk_msg_convert_ctx_access,
10959         .gen_prologue           = bpf_noop_prologue,
10960 };
10961
10962 const struct bpf_prog_ops sk_msg_prog_ops = {
10963 };
10964
10965 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
10966         .get_func_proto         = flow_dissector_func_proto,
10967         .is_valid_access        = flow_dissector_is_valid_access,
10968         .convert_ctx_access     = flow_dissector_convert_ctx_access,
10969 };
10970
10971 const struct bpf_prog_ops flow_dissector_prog_ops = {
10972         .test_run               = bpf_prog_test_run_flow_dissector,
10973 };
10974
10975 int sk_detach_filter(struct sock *sk)
10976 {
10977         int ret = -ENOENT;
10978         struct sk_filter *filter;
10979
10980         if (sock_flag(sk, SOCK_FILTER_LOCKED))
10981                 return -EPERM;
10982
10983         filter = rcu_dereference_protected(sk->sk_filter,
10984                                            lockdep_sock_is_held(sk));
10985         if (filter) {
10986                 RCU_INIT_POINTER(sk->sk_filter, NULL);
10987                 sk_filter_uncharge(sk, filter);
10988                 ret = 0;
10989         }
10990
10991         return ret;
10992 }
10993 EXPORT_SYMBOL_GPL(sk_detach_filter);
10994
10995 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len)
10996 {
10997         struct sock_fprog_kern *fprog;
10998         struct sk_filter *filter;
10999         int ret = 0;
11000
11001         sockopt_lock_sock(sk);
11002         filter = rcu_dereference_protected(sk->sk_filter,
11003                                            lockdep_sock_is_held(sk));
11004         if (!filter)
11005                 goto out;
11006
11007         /* We're copying the filter that has been originally attached,
11008          * so no conversion/decode needed anymore. eBPF programs that
11009          * have no original program cannot be dumped through this.
11010          */
11011         ret = -EACCES;
11012         fprog = filter->prog->orig_prog;
11013         if (!fprog)
11014                 goto out;
11015
11016         ret = fprog->len;
11017         if (!len)
11018                 /* User space only enquires number of filter blocks. */
11019                 goto out;
11020
11021         ret = -EINVAL;
11022         if (len < fprog->len)
11023                 goto out;
11024
11025         ret = -EFAULT;
11026         if (copy_to_sockptr(optval, fprog->filter, bpf_classic_proglen(fprog)))
11027                 goto out;
11028
11029         /* Instead of bytes, the API requests to return the number
11030          * of filter blocks.
11031          */
11032         ret = fprog->len;
11033 out:
11034         sockopt_release_sock(sk);
11035         return ret;
11036 }
11037
11038 #ifdef CONFIG_INET
11039 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
11040                                     struct sock_reuseport *reuse,
11041                                     struct sock *sk, struct sk_buff *skb,
11042                                     struct sock *migrating_sk,
11043                                     u32 hash)
11044 {
11045         reuse_kern->skb = skb;
11046         reuse_kern->sk = sk;
11047         reuse_kern->selected_sk = NULL;
11048         reuse_kern->migrating_sk = migrating_sk;
11049         reuse_kern->data_end = skb->data + skb_headlen(skb);
11050         reuse_kern->hash = hash;
11051         reuse_kern->reuseport_id = reuse->reuseport_id;
11052         reuse_kern->bind_inany = reuse->bind_inany;
11053 }
11054
11055 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
11056                                   struct bpf_prog *prog, struct sk_buff *skb,
11057                                   struct sock *migrating_sk,
11058                                   u32 hash)
11059 {
11060         struct sk_reuseport_kern reuse_kern;
11061         enum sk_action action;
11062
11063         bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
11064         action = bpf_prog_run(prog, &reuse_kern);
11065
11066         if (action == SK_PASS)
11067                 return reuse_kern.selected_sk;
11068         else
11069                 return ERR_PTR(-ECONNREFUSED);
11070 }
11071
11072 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
11073            struct bpf_map *, map, void *, key, u32, flags)
11074 {
11075         bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
11076         struct sock_reuseport *reuse;
11077         struct sock *selected_sk;
11078
11079         selected_sk = map->ops->map_lookup_elem(map, key);
11080         if (!selected_sk)
11081                 return -ENOENT;
11082
11083         reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
11084         if (!reuse) {
11085                 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
11086                 if (sk_is_refcounted(selected_sk))
11087                         sock_put(selected_sk);
11088
11089                 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
11090                  * The only (!reuse) case here is - the sk has already been
11091                  * unhashed (e.g. by close()), so treat it as -ENOENT.
11092                  *
11093                  * Other maps (e.g. sock_map) do not provide this guarantee and
11094                  * the sk may never be in the reuseport group to begin with.
11095                  */
11096                 return is_sockarray ? -ENOENT : -EINVAL;
11097         }
11098
11099         if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
11100                 struct sock *sk = reuse_kern->sk;
11101
11102                 if (sk->sk_protocol != selected_sk->sk_protocol)
11103                         return -EPROTOTYPE;
11104                 else if (sk->sk_family != selected_sk->sk_family)
11105                         return -EAFNOSUPPORT;
11106
11107                 /* Catch all. Likely bound to a different sockaddr. */
11108                 return -EBADFD;
11109         }
11110
11111         reuse_kern->selected_sk = selected_sk;
11112
11113         return 0;
11114 }
11115
11116 static const struct bpf_func_proto sk_select_reuseport_proto = {
11117         .func           = sk_select_reuseport,
11118         .gpl_only       = false,
11119         .ret_type       = RET_INTEGER,
11120         .arg1_type      = ARG_PTR_TO_CTX,
11121         .arg2_type      = ARG_CONST_MAP_PTR,
11122         .arg3_type      = ARG_PTR_TO_MAP_KEY,
11123         .arg4_type      = ARG_ANYTHING,
11124 };
11125
11126 BPF_CALL_4(sk_reuseport_load_bytes,
11127            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11128            void *, to, u32, len)
11129 {
11130         return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
11131 }
11132
11133 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
11134         .func           = sk_reuseport_load_bytes,
11135         .gpl_only       = false,
11136         .ret_type       = RET_INTEGER,
11137         .arg1_type      = ARG_PTR_TO_CTX,
11138         .arg2_type      = ARG_ANYTHING,
11139         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
11140         .arg4_type      = ARG_CONST_SIZE,
11141 };
11142
11143 BPF_CALL_5(sk_reuseport_load_bytes_relative,
11144            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11145            void *, to, u32, len, u32, start_header)
11146 {
11147         return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
11148                                                len, start_header);
11149 }
11150
11151 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
11152         .func           = sk_reuseport_load_bytes_relative,
11153         .gpl_only       = false,
11154         .ret_type       = RET_INTEGER,
11155         .arg1_type      = ARG_PTR_TO_CTX,
11156         .arg2_type      = ARG_ANYTHING,
11157         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
11158         .arg4_type      = ARG_CONST_SIZE,
11159         .arg5_type      = ARG_ANYTHING,
11160 };
11161
11162 static const struct bpf_func_proto *
11163 sk_reuseport_func_proto(enum bpf_func_id func_id,
11164                         const struct bpf_prog *prog)
11165 {
11166         switch (func_id) {
11167         case BPF_FUNC_sk_select_reuseport:
11168                 return &sk_select_reuseport_proto;
11169         case BPF_FUNC_skb_load_bytes:
11170                 return &sk_reuseport_load_bytes_proto;
11171         case BPF_FUNC_skb_load_bytes_relative:
11172                 return &sk_reuseport_load_bytes_relative_proto;
11173         case BPF_FUNC_get_socket_cookie:
11174                 return &bpf_get_socket_ptr_cookie_proto;
11175         case BPF_FUNC_ktime_get_coarse_ns:
11176                 return &bpf_ktime_get_coarse_ns_proto;
11177         default:
11178                 return bpf_base_func_proto(func_id);
11179         }
11180 }
11181
11182 static bool
11183 sk_reuseport_is_valid_access(int off, int size,
11184                              enum bpf_access_type type,
11185                              const struct bpf_prog *prog,
11186                              struct bpf_insn_access_aux *info)
11187 {
11188         const u32 size_default = sizeof(__u32);
11189
11190         if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
11191             off % size || type != BPF_READ)
11192                 return false;
11193
11194         switch (off) {
11195         case offsetof(struct sk_reuseport_md, data):
11196                 info->reg_type = PTR_TO_PACKET;
11197                 return size == sizeof(__u64);
11198
11199         case offsetof(struct sk_reuseport_md, data_end):
11200                 info->reg_type = PTR_TO_PACKET_END;
11201                 return size == sizeof(__u64);
11202
11203         case offsetof(struct sk_reuseport_md, hash):
11204                 return size == size_default;
11205
11206         case offsetof(struct sk_reuseport_md, sk):
11207                 info->reg_type = PTR_TO_SOCKET;
11208                 return size == sizeof(__u64);
11209
11210         case offsetof(struct sk_reuseport_md, migrating_sk):
11211                 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
11212                 return size == sizeof(__u64);
11213
11214         /* Fields that allow narrowing */
11215         case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
11216                 if (size < sizeof_field(struct sk_buff, protocol))
11217                         return false;
11218                 fallthrough;
11219         case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
11220         case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
11221         case bpf_ctx_range(struct sk_reuseport_md, len):
11222                 bpf_ctx_record_field_size(info, size_default);
11223                 return bpf_ctx_narrow_access_ok(off, size, size_default);
11224
11225         default:
11226                 return false;
11227         }
11228 }
11229
11230 #define SK_REUSEPORT_LOAD_FIELD(F) ({                                   \
11231         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
11232                               si->dst_reg, si->src_reg,                 \
11233                               bpf_target_off(struct sk_reuseport_kern, F, \
11234                                              sizeof_field(struct sk_reuseport_kern, F), \
11235                                              target_size));             \
11236         })
11237
11238 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD)                          \
11239         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
11240                                     struct sk_buff,                     \
11241                                     skb,                                \
11242                                     SKB_FIELD)
11243
11244 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD)                            \
11245         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
11246                                     struct sock,                        \
11247                                     sk,                                 \
11248                                     SK_FIELD)
11249
11250 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
11251                                            const struct bpf_insn *si,
11252                                            struct bpf_insn *insn_buf,
11253                                            struct bpf_prog *prog,
11254                                            u32 *target_size)
11255 {
11256         struct bpf_insn *insn = insn_buf;
11257
11258         switch (si->off) {
11259         case offsetof(struct sk_reuseport_md, data):
11260                 SK_REUSEPORT_LOAD_SKB_FIELD(data);
11261                 break;
11262
11263         case offsetof(struct sk_reuseport_md, len):
11264                 SK_REUSEPORT_LOAD_SKB_FIELD(len);
11265                 break;
11266
11267         case offsetof(struct sk_reuseport_md, eth_protocol):
11268                 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
11269                 break;
11270
11271         case offsetof(struct sk_reuseport_md, ip_protocol):
11272                 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
11273                 break;
11274
11275         case offsetof(struct sk_reuseport_md, data_end):
11276                 SK_REUSEPORT_LOAD_FIELD(data_end);
11277                 break;
11278
11279         case offsetof(struct sk_reuseport_md, hash):
11280                 SK_REUSEPORT_LOAD_FIELD(hash);
11281                 break;
11282
11283         case offsetof(struct sk_reuseport_md, bind_inany):
11284                 SK_REUSEPORT_LOAD_FIELD(bind_inany);
11285                 break;
11286
11287         case offsetof(struct sk_reuseport_md, sk):
11288                 SK_REUSEPORT_LOAD_FIELD(sk);
11289                 break;
11290
11291         case offsetof(struct sk_reuseport_md, migrating_sk):
11292                 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
11293                 break;
11294         }
11295
11296         return insn - insn_buf;
11297 }
11298
11299 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
11300         .get_func_proto         = sk_reuseport_func_proto,
11301         .is_valid_access        = sk_reuseport_is_valid_access,
11302         .convert_ctx_access     = sk_reuseport_convert_ctx_access,
11303 };
11304
11305 const struct bpf_prog_ops sk_reuseport_prog_ops = {
11306 };
11307
11308 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
11309 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
11310
11311 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
11312            struct sock *, sk, u64, flags)
11313 {
11314         if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
11315                                BPF_SK_LOOKUP_F_NO_REUSEPORT)))
11316                 return -EINVAL;
11317         if (unlikely(sk && sk_is_refcounted(sk)))
11318                 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
11319         if (unlikely(sk && sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN))
11320                 return -ESOCKTNOSUPPORT; /* only accept TCP socket in LISTEN */
11321         if (unlikely(sk && sk_is_udp(sk) && sk->sk_state != TCP_CLOSE))
11322                 return -ESOCKTNOSUPPORT; /* only accept UDP socket in CLOSE */
11323
11324         /* Check if socket is suitable for packet L3/L4 protocol */
11325         if (sk && sk->sk_protocol != ctx->protocol)
11326                 return -EPROTOTYPE;
11327         if (sk && sk->sk_family != ctx->family &&
11328             (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
11329                 return -EAFNOSUPPORT;
11330
11331         if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
11332                 return -EEXIST;
11333
11334         /* Select socket as lookup result */
11335         ctx->selected_sk = sk;
11336         ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
11337         return 0;
11338 }
11339
11340 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
11341         .func           = bpf_sk_lookup_assign,
11342         .gpl_only       = false,
11343         .ret_type       = RET_INTEGER,
11344         .arg1_type      = ARG_PTR_TO_CTX,
11345         .arg2_type      = ARG_PTR_TO_SOCKET_OR_NULL,
11346         .arg3_type      = ARG_ANYTHING,
11347 };
11348
11349 static const struct bpf_func_proto *
11350 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
11351 {
11352         switch (func_id) {
11353         case BPF_FUNC_perf_event_output:
11354                 return &bpf_event_output_data_proto;
11355         case BPF_FUNC_sk_assign:
11356                 return &bpf_sk_lookup_assign_proto;
11357         case BPF_FUNC_sk_release:
11358                 return &bpf_sk_release_proto;
11359         default:
11360                 return bpf_sk_base_func_proto(func_id);
11361         }
11362 }
11363
11364 static bool sk_lookup_is_valid_access(int off, int size,
11365                                       enum bpf_access_type type,
11366                                       const struct bpf_prog *prog,
11367                                       struct bpf_insn_access_aux *info)
11368 {
11369         if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
11370                 return false;
11371         if (off % size != 0)
11372                 return false;
11373         if (type != BPF_READ)
11374                 return false;
11375
11376         switch (off) {
11377         case offsetof(struct bpf_sk_lookup, sk):
11378                 info->reg_type = PTR_TO_SOCKET_OR_NULL;
11379                 return size == sizeof(__u64);
11380
11381         case bpf_ctx_range(struct bpf_sk_lookup, family):
11382         case bpf_ctx_range(struct bpf_sk_lookup, protocol):
11383         case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
11384         case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
11385         case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
11386         case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
11387         case bpf_ctx_range(struct bpf_sk_lookup, local_port):
11388         case bpf_ctx_range(struct bpf_sk_lookup, ingress_ifindex):
11389                 bpf_ctx_record_field_size(info, sizeof(__u32));
11390                 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
11391
11392         case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
11393                 /* Allow 4-byte access to 2-byte field for backward compatibility */
11394                 if (size == sizeof(__u32))
11395                         return true;
11396                 bpf_ctx_record_field_size(info, sizeof(__be16));
11397                 return bpf_ctx_narrow_access_ok(off, size, sizeof(__be16));
11398
11399         case offsetofend(struct bpf_sk_lookup, remote_port) ...
11400              offsetof(struct bpf_sk_lookup, local_ip4) - 1:
11401                 /* Allow access to zero padding for backward compatibility */
11402                 bpf_ctx_record_field_size(info, sizeof(__u16));
11403                 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u16));
11404
11405         default:
11406                 return false;
11407         }
11408 }
11409
11410 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
11411                                         const struct bpf_insn *si,
11412                                         struct bpf_insn *insn_buf,
11413                                         struct bpf_prog *prog,
11414                                         u32 *target_size)
11415 {
11416         struct bpf_insn *insn = insn_buf;
11417
11418         switch (si->off) {
11419         case offsetof(struct bpf_sk_lookup, sk):
11420                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11421                                       offsetof(struct bpf_sk_lookup_kern, selected_sk));
11422                 break;
11423
11424         case offsetof(struct bpf_sk_lookup, family):
11425                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11426                                       bpf_target_off(struct bpf_sk_lookup_kern,
11427                                                      family, 2, target_size));
11428                 break;
11429
11430         case offsetof(struct bpf_sk_lookup, protocol):
11431                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11432                                       bpf_target_off(struct bpf_sk_lookup_kern,
11433                                                      protocol, 2, target_size));
11434                 break;
11435
11436         case offsetof(struct bpf_sk_lookup, remote_ip4):
11437                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11438                                       bpf_target_off(struct bpf_sk_lookup_kern,
11439                                                      v4.saddr, 4, target_size));
11440                 break;
11441
11442         case offsetof(struct bpf_sk_lookup, local_ip4):
11443                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11444                                       bpf_target_off(struct bpf_sk_lookup_kern,
11445                                                      v4.daddr, 4, target_size));
11446                 break;
11447
11448         case bpf_ctx_range_till(struct bpf_sk_lookup,
11449                                 remote_ip6[0], remote_ip6[3]): {
11450 #if IS_ENABLED(CONFIG_IPV6)
11451                 int off = si->off;
11452
11453                 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
11454                 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11455                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11456                                       offsetof(struct bpf_sk_lookup_kern, v6.saddr));
11457                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11458                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11459 #else
11460                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11461 #endif
11462                 break;
11463         }
11464         case bpf_ctx_range_till(struct bpf_sk_lookup,
11465                                 local_ip6[0], local_ip6[3]): {
11466 #if IS_ENABLED(CONFIG_IPV6)
11467                 int off = si->off;
11468
11469                 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
11470                 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11471                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11472                                       offsetof(struct bpf_sk_lookup_kern, v6.daddr));
11473                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11474                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11475 #else
11476                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11477 #endif
11478                 break;
11479         }
11480         case offsetof(struct bpf_sk_lookup, remote_port):
11481                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11482                                       bpf_target_off(struct bpf_sk_lookup_kern,
11483                                                      sport, 2, target_size));
11484                 break;
11485
11486         case offsetofend(struct bpf_sk_lookup, remote_port):
11487                 *target_size = 2;
11488                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11489                 break;
11490
11491         case offsetof(struct bpf_sk_lookup, local_port):
11492                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11493                                       bpf_target_off(struct bpf_sk_lookup_kern,
11494                                                      dport, 2, target_size));
11495                 break;
11496
11497         case offsetof(struct bpf_sk_lookup, ingress_ifindex):
11498                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11499                                       bpf_target_off(struct bpf_sk_lookup_kern,
11500                                                      ingress_ifindex, 4, target_size));
11501                 break;
11502         }
11503
11504         return insn - insn_buf;
11505 }
11506
11507 const struct bpf_prog_ops sk_lookup_prog_ops = {
11508         .test_run = bpf_prog_test_run_sk_lookup,
11509 };
11510
11511 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
11512         .get_func_proto         = sk_lookup_func_proto,
11513         .is_valid_access        = sk_lookup_is_valid_access,
11514         .convert_ctx_access     = sk_lookup_convert_ctx_access,
11515 };
11516
11517 #endif /* CONFIG_INET */
11518
11519 DEFINE_BPF_DISPATCHER(xdp)
11520
11521 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
11522 {
11523         bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
11524 }
11525
11526 BTF_ID_LIST_GLOBAL(btf_sock_ids, MAX_BTF_SOCK_TYPE)
11527 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
11528 BTF_SOCK_TYPE_xxx
11529 #undef BTF_SOCK_TYPE
11530
11531 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
11532 {
11533         /* tcp6_sock type is not generated in dwarf and hence btf,
11534          * trigger an explicit type generation here.
11535          */
11536         BTF_TYPE_EMIT(struct tcp6_sock);
11537         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
11538             sk->sk_family == AF_INET6)
11539                 return (unsigned long)sk;
11540
11541         return (unsigned long)NULL;
11542 }
11543
11544 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
11545         .func                   = bpf_skc_to_tcp6_sock,
11546         .gpl_only               = false,
11547         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11548         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11549         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
11550 };
11551
11552 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
11553 {
11554         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
11555                 return (unsigned long)sk;
11556
11557         return (unsigned long)NULL;
11558 }
11559
11560 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
11561         .func                   = bpf_skc_to_tcp_sock,
11562         .gpl_only               = false,
11563         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11564         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11565         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
11566 };
11567
11568 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
11569 {
11570         /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
11571          * generated if CONFIG_INET=n. Trigger an explicit generation here.
11572          */
11573         BTF_TYPE_EMIT(struct inet_timewait_sock);
11574         BTF_TYPE_EMIT(struct tcp_timewait_sock);
11575
11576 #ifdef CONFIG_INET
11577         if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
11578                 return (unsigned long)sk;
11579 #endif
11580
11581 #if IS_BUILTIN(CONFIG_IPV6)
11582         if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
11583                 return (unsigned long)sk;
11584 #endif
11585
11586         return (unsigned long)NULL;
11587 }
11588
11589 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
11590         .func                   = bpf_skc_to_tcp_timewait_sock,
11591         .gpl_only               = false,
11592         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11593         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11594         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
11595 };
11596
11597 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
11598 {
11599 #ifdef CONFIG_INET
11600         if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11601                 return (unsigned long)sk;
11602 #endif
11603
11604 #if IS_BUILTIN(CONFIG_IPV6)
11605         if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11606                 return (unsigned long)sk;
11607 #endif
11608
11609         return (unsigned long)NULL;
11610 }
11611
11612 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
11613         .func                   = bpf_skc_to_tcp_request_sock,
11614         .gpl_only               = false,
11615         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11616         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11617         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
11618 };
11619
11620 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
11621 {
11622         /* udp6_sock type is not generated in dwarf and hence btf,
11623          * trigger an explicit type generation here.
11624          */
11625         BTF_TYPE_EMIT(struct udp6_sock);
11626         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
11627             sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
11628                 return (unsigned long)sk;
11629
11630         return (unsigned long)NULL;
11631 }
11632
11633 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
11634         .func                   = bpf_skc_to_udp6_sock,
11635         .gpl_only               = false,
11636         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11637         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11638         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
11639 };
11640
11641 BPF_CALL_1(bpf_skc_to_unix_sock, struct sock *, sk)
11642 {
11643         /* unix_sock type is not generated in dwarf and hence btf,
11644          * trigger an explicit type generation here.
11645          */
11646         BTF_TYPE_EMIT(struct unix_sock);
11647         if (sk && sk_fullsock(sk) && sk->sk_family == AF_UNIX)
11648                 return (unsigned long)sk;
11649
11650         return (unsigned long)NULL;
11651 }
11652
11653 const struct bpf_func_proto bpf_skc_to_unix_sock_proto = {
11654         .func                   = bpf_skc_to_unix_sock,
11655         .gpl_only               = false,
11656         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11657         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11658         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_UNIX],
11659 };
11660
11661 BPF_CALL_1(bpf_skc_to_mptcp_sock, struct sock *, sk)
11662 {
11663         BTF_TYPE_EMIT(struct mptcp_sock);
11664         return (unsigned long)bpf_mptcp_sock_from_subflow(sk);
11665 }
11666
11667 const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto = {
11668         .func           = bpf_skc_to_mptcp_sock,
11669         .gpl_only       = false,
11670         .ret_type       = RET_PTR_TO_BTF_ID_OR_NULL,
11671         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
11672         .ret_btf_id     = &btf_sock_ids[BTF_SOCK_TYPE_MPTCP],
11673 };
11674
11675 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
11676 {
11677         return (unsigned long)sock_from_file(file);
11678 }
11679
11680 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
11681 BTF_ID(struct, socket)
11682 BTF_ID(struct, file)
11683
11684 const struct bpf_func_proto bpf_sock_from_file_proto = {
11685         .func           = bpf_sock_from_file,
11686         .gpl_only       = false,
11687         .ret_type       = RET_PTR_TO_BTF_ID_OR_NULL,
11688         .ret_btf_id     = &bpf_sock_from_file_btf_ids[0],
11689         .arg1_type      = ARG_PTR_TO_BTF_ID,
11690         .arg1_btf_id    = &bpf_sock_from_file_btf_ids[1],
11691 };
11692
11693 static const struct bpf_func_proto *
11694 bpf_sk_base_func_proto(enum bpf_func_id func_id)
11695 {
11696         const struct bpf_func_proto *func;
11697
11698         switch (func_id) {
11699         case BPF_FUNC_skc_to_tcp6_sock:
11700                 func = &bpf_skc_to_tcp6_sock_proto;
11701                 break;
11702         case BPF_FUNC_skc_to_tcp_sock:
11703                 func = &bpf_skc_to_tcp_sock_proto;
11704                 break;
11705         case BPF_FUNC_skc_to_tcp_timewait_sock:
11706                 func = &bpf_skc_to_tcp_timewait_sock_proto;
11707                 break;
11708         case BPF_FUNC_skc_to_tcp_request_sock:
11709                 func = &bpf_skc_to_tcp_request_sock_proto;
11710                 break;
11711         case BPF_FUNC_skc_to_udp6_sock:
11712                 func = &bpf_skc_to_udp6_sock_proto;
11713                 break;
11714         case BPF_FUNC_skc_to_unix_sock:
11715                 func = &bpf_skc_to_unix_sock_proto;
11716                 break;
11717         case BPF_FUNC_skc_to_mptcp_sock:
11718                 func = &bpf_skc_to_mptcp_sock_proto;
11719                 break;
11720         case BPF_FUNC_ktime_get_coarse_ns:
11721                 return &bpf_ktime_get_coarse_ns_proto;
11722         default:
11723                 return bpf_base_func_proto(func_id);
11724         }
11725
11726         if (!perfmon_capable())
11727                 return NULL;
11728
11729         return func;
11730 }
11731
11732 __diag_push();
11733 __diag_ignore_all("-Wmissing-prototypes",
11734                   "Global functions as their definitions will be in vmlinux BTF");
11735 __bpf_kfunc int bpf_dynptr_from_skb(struct sk_buff *skb, u64 flags,
11736                                     struct bpf_dynptr_kern *ptr__uninit)
11737 {
11738         if (flags) {
11739                 bpf_dynptr_set_null(ptr__uninit);
11740                 return -EINVAL;
11741         }
11742
11743         bpf_dynptr_init(ptr__uninit, skb, BPF_DYNPTR_TYPE_SKB, 0, skb->len);
11744
11745         return 0;
11746 }
11747
11748 __bpf_kfunc int bpf_dynptr_from_xdp(struct xdp_buff *xdp, u64 flags,
11749                                     struct bpf_dynptr_kern *ptr__uninit)
11750 {
11751         if (flags) {
11752                 bpf_dynptr_set_null(ptr__uninit);
11753                 return -EINVAL;
11754         }
11755
11756         bpf_dynptr_init(ptr__uninit, xdp, BPF_DYNPTR_TYPE_XDP, 0, xdp_get_buff_len(xdp));
11757
11758         return 0;
11759 }
11760 __diag_pop();
11761
11762 int bpf_dynptr_from_skb_rdonly(struct sk_buff *skb, u64 flags,
11763                                struct bpf_dynptr_kern *ptr__uninit)
11764 {
11765         int err;
11766
11767         err = bpf_dynptr_from_skb(skb, flags, ptr__uninit);
11768         if (err)
11769                 return err;
11770
11771         bpf_dynptr_set_rdonly(ptr__uninit);
11772
11773         return 0;
11774 }
11775
11776 BTF_SET8_START(bpf_kfunc_check_set_skb)
11777 BTF_ID_FLAGS(func, bpf_dynptr_from_skb)
11778 BTF_SET8_END(bpf_kfunc_check_set_skb)
11779
11780 BTF_SET8_START(bpf_kfunc_check_set_xdp)
11781 BTF_ID_FLAGS(func, bpf_dynptr_from_xdp)
11782 BTF_SET8_END(bpf_kfunc_check_set_xdp)
11783
11784 static const struct btf_kfunc_id_set bpf_kfunc_set_skb = {
11785         .owner = THIS_MODULE,
11786         .set = &bpf_kfunc_check_set_skb,
11787 };
11788
11789 static const struct btf_kfunc_id_set bpf_kfunc_set_xdp = {
11790         .owner = THIS_MODULE,
11791         .set = &bpf_kfunc_check_set_xdp,
11792 };
11793
11794 static int __init bpf_kfunc_init(void)
11795 {
11796         int ret;
11797
11798         ret = register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_kfunc_set_skb);
11799         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_ACT, &bpf_kfunc_set_skb);
11800         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SK_SKB, &bpf_kfunc_set_skb);
11801         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SOCKET_FILTER, &bpf_kfunc_set_skb);
11802         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_CGROUP_SKB, &bpf_kfunc_set_skb);
11803         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_OUT, &bpf_kfunc_set_skb);
11804         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_IN, &bpf_kfunc_set_skb);
11805         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_XMIT, &bpf_kfunc_set_skb);
11806         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_SEG6LOCAL, &bpf_kfunc_set_skb);
11807         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_NETFILTER, &bpf_kfunc_set_skb);
11808         return ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_XDP, &bpf_kfunc_set_xdp);
11809 }
11810 late_initcall(bpf_kfunc_init);
11811
11812 /* Disables missing prototype warnings */
11813 __diag_push();
11814 __diag_ignore_all("-Wmissing-prototypes",
11815                   "Global functions as their definitions will be in vmlinux BTF");
11816
11817 /* bpf_sock_destroy: Destroy the given socket with ECONNABORTED error code.
11818  *
11819  * The function expects a non-NULL pointer to a socket, and invokes the
11820  * protocol specific socket destroy handlers.
11821  *
11822  * The helper can only be called from BPF contexts that have acquired the socket
11823  * locks.
11824  *
11825  * Parameters:
11826  * @sock: Pointer to socket to be destroyed
11827  *
11828  * Return:
11829  * On error, may return EPROTONOSUPPORT, EINVAL.
11830  * EPROTONOSUPPORT if protocol specific destroy handler is not supported.
11831  * 0 otherwise
11832  */
11833 __bpf_kfunc int bpf_sock_destroy(struct sock_common *sock)
11834 {
11835         struct sock *sk = (struct sock *)sock;
11836
11837         /* The locking semantics that allow for synchronous execution of the
11838          * destroy handlers are only supported for TCP and UDP.
11839          * Supporting protocols will need to acquire sock lock in the BPF context
11840          * prior to invoking this kfunc.
11841          */
11842         if (!sk->sk_prot->diag_destroy || (sk->sk_protocol != IPPROTO_TCP &&
11843                                            sk->sk_protocol != IPPROTO_UDP))
11844                 return -EOPNOTSUPP;
11845
11846         return sk->sk_prot->diag_destroy(sk, ECONNABORTED);
11847 }
11848
11849 __diag_pop()
11850
11851 BTF_SET8_START(bpf_sk_iter_kfunc_ids)
11852 BTF_ID_FLAGS(func, bpf_sock_destroy, KF_TRUSTED_ARGS)
11853 BTF_SET8_END(bpf_sk_iter_kfunc_ids)
11854
11855 static int tracing_iter_filter(const struct bpf_prog *prog, u32 kfunc_id)
11856 {
11857         if (btf_id_set8_contains(&bpf_sk_iter_kfunc_ids, kfunc_id) &&
11858             prog->expected_attach_type != BPF_TRACE_ITER)
11859                 return -EACCES;
11860         return 0;
11861 }
11862
11863 static const struct btf_kfunc_id_set bpf_sk_iter_kfunc_set = {
11864         .owner = THIS_MODULE,
11865         .set   = &bpf_sk_iter_kfunc_ids,
11866         .filter = tracing_iter_filter,
11867 };
11868
11869 static int init_subsystem(void)
11870 {
11871         return register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_sk_iter_kfunc_set);
11872 }
11873 late_initcall(init_subsystem);