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                 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4344
4345         return __xdp_do_redirect_frame(ri, dev, xdp_convert_buff_to_frame(xdp),
4346                                        xdp_prog);
4347 }
4348 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4349
4350 int xdp_do_redirect_frame(struct net_device *dev, struct xdp_buff *xdp,
4351                           struct xdp_frame *xdpf, struct bpf_prog *xdp_prog)
4352 {
4353         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4354         enum bpf_map_type map_type = ri->map_type;
4355
4356         if (map_type == BPF_MAP_TYPE_XSKMAP)
4357                 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4358
4359         return __xdp_do_redirect_frame(ri, dev, xdpf, xdp_prog);
4360 }
4361 EXPORT_SYMBOL_GPL(xdp_do_redirect_frame);
4362
4363 static int xdp_do_generic_redirect_map(struct net_device *dev,
4364                                        struct sk_buff *skb,
4365                                        struct xdp_buff *xdp,
4366                                        struct bpf_prog *xdp_prog,
4367                                        void *fwd,
4368                                        enum bpf_map_type map_type, u32 map_id)
4369 {
4370         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4371         struct bpf_map *map;
4372         int err;
4373
4374         switch (map_type) {
4375         case BPF_MAP_TYPE_DEVMAP:
4376                 fallthrough;
4377         case BPF_MAP_TYPE_DEVMAP_HASH:
4378                 map = READ_ONCE(ri->map);
4379                 if (unlikely(map)) {
4380                         WRITE_ONCE(ri->map, NULL);
4381                         err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4382                                                      ri->flags & BPF_F_EXCLUDE_INGRESS);
4383                 } else {
4384                         err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4385                 }
4386                 if (unlikely(err))
4387                         goto err;
4388                 break;
4389         case BPF_MAP_TYPE_XSKMAP:
4390                 err = xsk_generic_rcv(fwd, xdp);
4391                 if (err)
4392                         goto err;
4393                 consume_skb(skb);
4394                 break;
4395         case BPF_MAP_TYPE_CPUMAP:
4396                 err = cpu_map_generic_redirect(fwd, skb);
4397                 if (unlikely(err))
4398                         goto err;
4399                 break;
4400         default:
4401                 err = -EBADRQC;
4402                 goto err;
4403         }
4404
4405         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4406         return 0;
4407 err:
4408         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4409         return err;
4410 }
4411
4412 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4413                             struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4414 {
4415         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4416         enum bpf_map_type map_type = ri->map_type;
4417         void *fwd = ri->tgt_value;
4418         u32 map_id = ri->map_id;
4419         int err;
4420
4421         ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4422         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4423
4424         if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4425                 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4426                 if (unlikely(!fwd)) {
4427                         err = -EINVAL;
4428                         goto err;
4429                 }
4430
4431                 err = xdp_ok_fwd_dev(fwd, skb->len);
4432                 if (unlikely(err))
4433                         goto err;
4434
4435                 skb->dev = fwd;
4436                 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4437                 generic_xdp_tx(skb, xdp_prog);
4438                 return 0;
4439         }
4440
4441         return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id);
4442 err:
4443         _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4444         return err;
4445 }
4446
4447 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4448 {
4449         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4450
4451         if (unlikely(flags))
4452                 return XDP_ABORTED;
4453
4454         /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4455          * by map_idr) is used for ifindex based XDP redirect.
4456          */
4457         ri->tgt_index = ifindex;
4458         ri->map_id = INT_MAX;
4459         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4460
4461         return XDP_REDIRECT;
4462 }
4463
4464 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4465         .func           = bpf_xdp_redirect,
4466         .gpl_only       = false,
4467         .ret_type       = RET_INTEGER,
4468         .arg1_type      = ARG_ANYTHING,
4469         .arg2_type      = ARG_ANYTHING,
4470 };
4471
4472 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u64, key,
4473            u64, flags)
4474 {
4475         return map->ops->map_redirect(map, key, flags);
4476 }
4477
4478 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4479         .func           = bpf_xdp_redirect_map,
4480         .gpl_only       = false,
4481         .ret_type       = RET_INTEGER,
4482         .arg1_type      = ARG_CONST_MAP_PTR,
4483         .arg2_type      = ARG_ANYTHING,
4484         .arg3_type      = ARG_ANYTHING,
4485 };
4486
4487 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4488                                   unsigned long off, unsigned long len)
4489 {
4490         void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4491
4492         if (unlikely(!ptr))
4493                 return len;
4494         if (ptr != dst_buff)
4495                 memcpy(dst_buff, ptr, len);
4496
4497         return 0;
4498 }
4499
4500 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4501            u64, flags, void *, meta, u64, meta_size)
4502 {
4503         u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4504
4505         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4506                 return -EINVAL;
4507         if (unlikely(!skb || skb_size > skb->len))
4508                 return -EFAULT;
4509
4510         return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4511                                 bpf_skb_copy);
4512 }
4513
4514 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4515         .func           = bpf_skb_event_output,
4516         .gpl_only       = true,
4517         .ret_type       = RET_INTEGER,
4518         .arg1_type      = ARG_PTR_TO_CTX,
4519         .arg2_type      = ARG_CONST_MAP_PTR,
4520         .arg3_type      = ARG_ANYTHING,
4521         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4522         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4523 };
4524
4525 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4526
4527 const struct bpf_func_proto bpf_skb_output_proto = {
4528         .func           = bpf_skb_event_output,
4529         .gpl_only       = true,
4530         .ret_type       = RET_INTEGER,
4531         .arg1_type      = ARG_PTR_TO_BTF_ID,
4532         .arg1_btf_id    = &bpf_skb_output_btf_ids[0],
4533         .arg2_type      = ARG_CONST_MAP_PTR,
4534         .arg3_type      = ARG_ANYTHING,
4535         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4536         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4537 };
4538
4539 static unsigned short bpf_tunnel_key_af(u64 flags)
4540 {
4541         return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4542 }
4543
4544 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4545            u32, size, u64, flags)
4546 {
4547         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4548         u8 compat[sizeof(struct bpf_tunnel_key)];
4549         void *to_orig = to;
4550         int err;
4551
4552         if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6 |
4553                                          BPF_F_TUNINFO_FLAGS)))) {
4554                 err = -EINVAL;
4555                 goto err_clear;
4556         }
4557         if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4558                 err = -EPROTO;
4559                 goto err_clear;
4560         }
4561         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4562                 err = -EINVAL;
4563                 switch (size) {
4564                 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4565                 case offsetof(struct bpf_tunnel_key, tunnel_label):
4566                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4567                         goto set_compat;
4568                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4569                         /* Fixup deprecated structure layouts here, so we have
4570                          * a common path later on.
4571                          */
4572                         if (ip_tunnel_info_af(info) != AF_INET)
4573                                 goto err_clear;
4574 set_compat:
4575                         to = (struct bpf_tunnel_key *)compat;
4576                         break;
4577                 default:
4578                         goto err_clear;
4579                 }
4580         }
4581
4582         to->tunnel_id = be64_to_cpu(info->key.tun_id);
4583         to->tunnel_tos = info->key.tos;
4584         to->tunnel_ttl = info->key.ttl;
4585         if (flags & BPF_F_TUNINFO_FLAGS)
4586                 to->tunnel_flags = info->key.tun_flags;
4587         else
4588                 to->tunnel_ext = 0;
4589
4590         if (flags & BPF_F_TUNINFO_IPV6) {
4591                 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4592                        sizeof(to->remote_ipv6));
4593                 memcpy(to->local_ipv6, &info->key.u.ipv6.dst,
4594                        sizeof(to->local_ipv6));
4595                 to->tunnel_label = be32_to_cpu(info->key.label);
4596         } else {
4597                 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4598                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4599                 to->local_ipv4 = be32_to_cpu(info->key.u.ipv4.dst);
4600                 memset(&to->local_ipv6[1], 0, sizeof(__u32) * 3);
4601                 to->tunnel_label = 0;
4602         }
4603
4604         if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4605                 memcpy(to_orig, to, size);
4606
4607         return 0;
4608 err_clear:
4609         memset(to_orig, 0, size);
4610         return err;
4611 }
4612
4613 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4614         .func           = bpf_skb_get_tunnel_key,
4615         .gpl_only       = false,
4616         .ret_type       = RET_INTEGER,
4617         .arg1_type      = ARG_PTR_TO_CTX,
4618         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
4619         .arg3_type      = ARG_CONST_SIZE,
4620         .arg4_type      = ARG_ANYTHING,
4621 };
4622
4623 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4624 {
4625         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4626         int err;
4627
4628         if (unlikely(!info ||
4629                      !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4630                 err = -ENOENT;
4631                 goto err_clear;
4632         }
4633         if (unlikely(size < info->options_len)) {
4634                 err = -ENOMEM;
4635                 goto err_clear;
4636         }
4637
4638         ip_tunnel_info_opts_get(to, info);
4639         if (size > info->options_len)
4640                 memset(to + info->options_len, 0, size - info->options_len);
4641
4642         return info->options_len;
4643 err_clear:
4644         memset(to, 0, size);
4645         return err;
4646 }
4647
4648 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4649         .func           = bpf_skb_get_tunnel_opt,
4650         .gpl_only       = false,
4651         .ret_type       = RET_INTEGER,
4652         .arg1_type      = ARG_PTR_TO_CTX,
4653         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
4654         .arg3_type      = ARG_CONST_SIZE,
4655 };
4656
4657 static struct metadata_dst __percpu *md_dst;
4658
4659 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4660            const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4661 {
4662         struct metadata_dst *md = this_cpu_ptr(md_dst);
4663         u8 compat[sizeof(struct bpf_tunnel_key)];
4664         struct ip_tunnel_info *info;
4665
4666         if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4667                                BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER |
4668                                BPF_F_NO_TUNNEL_KEY)))
4669                 return -EINVAL;
4670         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4671                 switch (size) {
4672                 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4673                 case offsetof(struct bpf_tunnel_key, tunnel_label):
4674                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4675                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4676                         /* Fixup deprecated structure layouts here, so we have
4677                          * a common path later on.
4678                          */
4679                         memcpy(compat, from, size);
4680                         memset(compat + size, 0, sizeof(compat) - size);
4681                         from = (const struct bpf_tunnel_key *) compat;
4682                         break;
4683                 default:
4684                         return -EINVAL;
4685                 }
4686         }
4687         if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4688                      from->tunnel_ext))
4689                 return -EINVAL;
4690
4691         skb_dst_drop(skb);
4692         dst_hold((struct dst_entry *) md);
4693         skb_dst_set(skb, (struct dst_entry *) md);
4694
4695         info = &md->u.tun_info;
4696         memset(info, 0, sizeof(*info));
4697         info->mode = IP_TUNNEL_INFO_TX;
4698
4699         info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4700         if (flags & BPF_F_DONT_FRAGMENT)
4701                 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4702         if (flags & BPF_F_ZERO_CSUM_TX)
4703                 info->key.tun_flags &= ~TUNNEL_CSUM;
4704         if (flags & BPF_F_SEQ_NUMBER)
4705                 info->key.tun_flags |= TUNNEL_SEQ;
4706         if (flags & BPF_F_NO_TUNNEL_KEY)
4707                 info->key.tun_flags &= ~TUNNEL_KEY;
4708
4709         info->key.tun_id = cpu_to_be64(from->tunnel_id);
4710         info->key.tos = from->tunnel_tos;
4711         info->key.ttl = from->tunnel_ttl;
4712
4713         if (flags & BPF_F_TUNINFO_IPV6) {
4714                 info->mode |= IP_TUNNEL_INFO_IPV6;
4715                 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4716                        sizeof(from->remote_ipv6));
4717                 memcpy(&info->key.u.ipv6.src, from->local_ipv6,
4718                        sizeof(from->local_ipv6));
4719                 info->key.label = cpu_to_be32(from->tunnel_label) &
4720                                   IPV6_FLOWLABEL_MASK;
4721         } else {
4722                 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4723                 info->key.u.ipv4.src = cpu_to_be32(from->local_ipv4);
4724                 info->key.flow_flags = FLOWI_FLAG_ANYSRC;
4725         }
4726
4727         return 0;
4728 }
4729
4730 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4731         .func           = bpf_skb_set_tunnel_key,
4732         .gpl_only       = false,
4733         .ret_type       = RET_INTEGER,
4734         .arg1_type      = ARG_PTR_TO_CTX,
4735         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4736         .arg3_type      = ARG_CONST_SIZE,
4737         .arg4_type      = ARG_ANYTHING,
4738 };
4739
4740 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4741            const u8 *, from, u32, size)
4742 {
4743         struct ip_tunnel_info *info = skb_tunnel_info(skb);
4744         const struct metadata_dst *md = this_cpu_ptr(md_dst);
4745
4746         if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4747                 return -EINVAL;
4748         if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4749                 return -ENOMEM;
4750
4751         ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4752
4753         return 0;
4754 }
4755
4756 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4757         .func           = bpf_skb_set_tunnel_opt,
4758         .gpl_only       = false,
4759         .ret_type       = RET_INTEGER,
4760         .arg1_type      = ARG_PTR_TO_CTX,
4761         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4762         .arg3_type      = ARG_CONST_SIZE,
4763 };
4764
4765 static const struct bpf_func_proto *
4766 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4767 {
4768         if (!md_dst) {
4769                 struct metadata_dst __percpu *tmp;
4770
4771                 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4772                                                 METADATA_IP_TUNNEL,
4773                                                 GFP_KERNEL);
4774                 if (!tmp)
4775                         return NULL;
4776                 if (cmpxchg(&md_dst, NULL, tmp))
4777                         metadata_dst_free_percpu(tmp);
4778         }
4779
4780         switch (which) {
4781         case BPF_FUNC_skb_set_tunnel_key:
4782                 return &bpf_skb_set_tunnel_key_proto;
4783         case BPF_FUNC_skb_set_tunnel_opt:
4784                 return &bpf_skb_set_tunnel_opt_proto;
4785         default:
4786                 return NULL;
4787         }
4788 }
4789
4790 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4791            u32, idx)
4792 {
4793         struct bpf_array *array = container_of(map, struct bpf_array, map);
4794         struct cgroup *cgrp;
4795         struct sock *sk;
4796
4797         sk = skb_to_full_sk(skb);
4798         if (!sk || !sk_fullsock(sk))
4799                 return -ENOENT;
4800         if (unlikely(idx >= array->map.max_entries))
4801                 return -E2BIG;
4802
4803         cgrp = READ_ONCE(array->ptrs[idx]);
4804         if (unlikely(!cgrp))
4805                 return -EAGAIN;
4806
4807         return sk_under_cgroup_hierarchy(sk, cgrp);
4808 }
4809
4810 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4811         .func           = bpf_skb_under_cgroup,
4812         .gpl_only       = false,
4813         .ret_type       = RET_INTEGER,
4814         .arg1_type      = ARG_PTR_TO_CTX,
4815         .arg2_type      = ARG_CONST_MAP_PTR,
4816         .arg3_type      = ARG_ANYTHING,
4817 };
4818
4819 #ifdef CONFIG_SOCK_CGROUP_DATA
4820 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4821 {
4822         struct cgroup *cgrp;
4823
4824         sk = sk_to_full_sk(sk);
4825         if (!sk || !sk_fullsock(sk))
4826                 return 0;
4827
4828         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4829         return cgroup_id(cgrp);
4830 }
4831
4832 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4833 {
4834         return __bpf_sk_cgroup_id(skb->sk);
4835 }
4836
4837 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4838         .func           = bpf_skb_cgroup_id,
4839         .gpl_only       = false,
4840         .ret_type       = RET_INTEGER,
4841         .arg1_type      = ARG_PTR_TO_CTX,
4842 };
4843
4844 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4845                                               int ancestor_level)
4846 {
4847         struct cgroup *ancestor;
4848         struct cgroup *cgrp;
4849
4850         sk = sk_to_full_sk(sk);
4851         if (!sk || !sk_fullsock(sk))
4852                 return 0;
4853
4854         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4855         ancestor = cgroup_ancestor(cgrp, ancestor_level);
4856         if (!ancestor)
4857                 return 0;
4858
4859         return cgroup_id(ancestor);
4860 }
4861
4862 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4863            ancestor_level)
4864 {
4865         return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4866 }
4867
4868 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4869         .func           = bpf_skb_ancestor_cgroup_id,
4870         .gpl_only       = false,
4871         .ret_type       = RET_INTEGER,
4872         .arg1_type      = ARG_PTR_TO_CTX,
4873         .arg2_type      = ARG_ANYTHING,
4874 };
4875
4876 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4877 {
4878         return __bpf_sk_cgroup_id(sk);
4879 }
4880
4881 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4882         .func           = bpf_sk_cgroup_id,
4883         .gpl_only       = false,
4884         .ret_type       = RET_INTEGER,
4885         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4886 };
4887
4888 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4889 {
4890         return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4891 }
4892
4893 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4894         .func           = bpf_sk_ancestor_cgroup_id,
4895         .gpl_only       = false,
4896         .ret_type       = RET_INTEGER,
4897         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4898         .arg2_type      = ARG_ANYTHING,
4899 };
4900 #endif
4901
4902 static unsigned long bpf_xdp_copy(void *dst, const void *ctx,
4903                                   unsigned long off, unsigned long len)
4904 {
4905         struct xdp_buff *xdp = (struct xdp_buff *)ctx;
4906
4907         bpf_xdp_copy_buf(xdp, off, dst, len, false);
4908         return 0;
4909 }
4910
4911 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4912            u64, flags, void *, meta, u64, meta_size)
4913 {
4914         u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4915
4916         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4917                 return -EINVAL;
4918
4919         if (unlikely(!xdp || xdp_size > xdp_get_buff_len(xdp)))
4920                 return -EFAULT;
4921
4922         return bpf_event_output(map, flags, meta, meta_size, xdp,
4923                                 xdp_size, bpf_xdp_copy);
4924 }
4925
4926 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4927         .func           = bpf_xdp_event_output,
4928         .gpl_only       = true,
4929         .ret_type       = RET_INTEGER,
4930         .arg1_type      = ARG_PTR_TO_CTX,
4931         .arg2_type      = ARG_CONST_MAP_PTR,
4932         .arg3_type      = ARG_ANYTHING,
4933         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4934         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4935 };
4936
4937 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4938
4939 const struct bpf_func_proto bpf_xdp_output_proto = {
4940         .func           = bpf_xdp_event_output,
4941         .gpl_only       = true,
4942         .ret_type       = RET_INTEGER,
4943         .arg1_type      = ARG_PTR_TO_BTF_ID,
4944         .arg1_btf_id    = &bpf_xdp_output_btf_ids[0],
4945         .arg2_type      = ARG_CONST_MAP_PTR,
4946         .arg3_type      = ARG_ANYTHING,
4947         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4948         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4949 };
4950
4951 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4952 {
4953         return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4954 }
4955
4956 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4957         .func           = bpf_get_socket_cookie,
4958         .gpl_only       = false,
4959         .ret_type       = RET_INTEGER,
4960         .arg1_type      = ARG_PTR_TO_CTX,
4961 };
4962
4963 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4964 {
4965         return __sock_gen_cookie(ctx->sk);
4966 }
4967
4968 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4969         .func           = bpf_get_socket_cookie_sock_addr,
4970         .gpl_only       = false,
4971         .ret_type       = RET_INTEGER,
4972         .arg1_type      = ARG_PTR_TO_CTX,
4973 };
4974
4975 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4976 {
4977         return __sock_gen_cookie(ctx);
4978 }
4979
4980 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4981         .func           = bpf_get_socket_cookie_sock,
4982         .gpl_only       = false,
4983         .ret_type       = RET_INTEGER,
4984         .arg1_type      = ARG_PTR_TO_CTX,
4985 };
4986
4987 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
4988 {
4989         return sk ? sock_gen_cookie(sk) : 0;
4990 }
4991
4992 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
4993         .func           = bpf_get_socket_ptr_cookie,
4994         .gpl_only       = false,
4995         .ret_type       = RET_INTEGER,
4996         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON | PTR_MAYBE_NULL,
4997 };
4998
4999 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
5000 {
5001         return __sock_gen_cookie(ctx->sk);
5002 }
5003
5004 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
5005         .func           = bpf_get_socket_cookie_sock_ops,
5006         .gpl_only       = false,
5007         .ret_type       = RET_INTEGER,
5008         .arg1_type      = ARG_PTR_TO_CTX,
5009 };
5010
5011 static u64 __bpf_get_netns_cookie(struct sock *sk)
5012 {
5013         const struct net *net = sk ? sock_net(sk) : &init_net;
5014
5015         return net->net_cookie;
5016 }
5017
5018 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
5019 {
5020         return __bpf_get_netns_cookie(ctx);
5021 }
5022
5023 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
5024         .func           = bpf_get_netns_cookie_sock,
5025         .gpl_only       = false,
5026         .ret_type       = RET_INTEGER,
5027         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
5028 };
5029
5030 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
5031 {
5032         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5033 }
5034
5035 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
5036         .func           = bpf_get_netns_cookie_sock_addr,
5037         .gpl_only       = false,
5038         .ret_type       = RET_INTEGER,
5039         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
5040 };
5041
5042 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
5043 {
5044         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5045 }
5046
5047 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = {
5048         .func           = bpf_get_netns_cookie_sock_ops,
5049         .gpl_only       = false,
5050         .ret_type       = RET_INTEGER,
5051         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
5052 };
5053
5054 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx)
5055 {
5056         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5057 }
5058
5059 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = {
5060         .func           = bpf_get_netns_cookie_sk_msg,
5061         .gpl_only       = false,
5062         .ret_type       = RET_INTEGER,
5063         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
5064 };
5065
5066 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
5067 {
5068         struct sock *sk = sk_to_full_sk(skb->sk);
5069         kuid_t kuid;
5070
5071         if (!sk || !sk_fullsock(sk))
5072                 return overflowuid;
5073         kuid = sock_net_uid(sock_net(sk), sk);
5074         return from_kuid_munged(sock_net(sk)->user_ns, kuid);
5075 }
5076
5077 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
5078         .func           = bpf_get_socket_uid,
5079         .gpl_only       = false,
5080         .ret_type       = RET_INTEGER,
5081         .arg1_type      = ARG_PTR_TO_CTX,
5082 };
5083
5084 static int sol_socket_sockopt(struct sock *sk, int optname,
5085                               char *optval, int *optlen,
5086                               bool getopt)
5087 {
5088         switch (optname) {
5089         case SO_REUSEADDR:
5090         case SO_SNDBUF:
5091         case SO_RCVBUF:
5092         case SO_KEEPALIVE:
5093         case SO_PRIORITY:
5094         case SO_REUSEPORT:
5095         case SO_RCVLOWAT:
5096         case SO_MARK:
5097         case SO_MAX_PACING_RATE:
5098         case SO_BINDTOIFINDEX:
5099         case SO_TXREHASH:
5100                 if (*optlen != sizeof(int))
5101                         return -EINVAL;
5102                 break;
5103         case SO_BINDTODEVICE:
5104                 break;
5105         default:
5106                 return -EINVAL;
5107         }
5108
5109         if (getopt) {
5110                 if (optname == SO_BINDTODEVICE)
5111                         return -EINVAL;
5112                 return sk_getsockopt(sk, SOL_SOCKET, optname,
5113                                      KERNEL_SOCKPTR(optval),
5114                                      KERNEL_SOCKPTR(optlen));
5115         }
5116
5117         return sk_setsockopt(sk, SOL_SOCKET, optname,
5118                              KERNEL_SOCKPTR(optval), *optlen);
5119 }
5120
5121 static int bpf_sol_tcp_setsockopt(struct sock *sk, int optname,
5122                                   char *optval, int optlen)
5123 {
5124         struct tcp_sock *tp = tcp_sk(sk);
5125         unsigned long timeout;
5126         int val;
5127
5128         if (optlen != sizeof(int))
5129                 return -EINVAL;
5130
5131         val = *(int *)optval;
5132
5133         /* Only some options are supported */
5134         switch (optname) {
5135         case TCP_BPF_IW:
5136                 if (val <= 0 || tp->data_segs_out > tp->syn_data)
5137                         return -EINVAL;
5138                 tcp_snd_cwnd_set(tp, val);
5139                 break;
5140         case TCP_BPF_SNDCWND_CLAMP:
5141                 if (val <= 0)
5142                         return -EINVAL;
5143                 tp->snd_cwnd_clamp = val;
5144                 tp->snd_ssthresh = val;
5145                 break;
5146         case TCP_BPF_DELACK_MAX:
5147                 timeout = usecs_to_jiffies(val);
5148                 if (timeout > TCP_DELACK_MAX ||
5149                     timeout < TCP_TIMEOUT_MIN)
5150                         return -EINVAL;
5151                 inet_csk(sk)->icsk_delack_max = timeout;
5152                 break;
5153         case TCP_BPF_RTO_MIN:
5154                 timeout = usecs_to_jiffies(val);
5155                 if (timeout > TCP_RTO_MIN ||
5156                     timeout < TCP_TIMEOUT_MIN)
5157                         return -EINVAL;
5158                 inet_csk(sk)->icsk_rto_min = timeout;
5159                 break;
5160         default:
5161                 return -EINVAL;
5162         }
5163
5164         return 0;
5165 }
5166
5167 static int sol_tcp_sockopt_congestion(struct sock *sk, char *optval,
5168                                       int *optlen, bool getopt)
5169 {
5170         struct tcp_sock *tp;
5171         int ret;
5172
5173         if (*optlen < 2)
5174                 return -EINVAL;
5175
5176         if (getopt) {
5177                 if (!inet_csk(sk)->icsk_ca_ops)
5178                         return -EINVAL;
5179                 /* BPF expects NULL-terminated tcp-cc string */
5180                 optval[--(*optlen)] = '\0';
5181                 return do_tcp_getsockopt(sk, SOL_TCP, TCP_CONGESTION,
5182                                          KERNEL_SOCKPTR(optval),
5183                                          KERNEL_SOCKPTR(optlen));
5184         }
5185
5186         /* "cdg" is the only cc that alloc a ptr
5187          * in inet_csk_ca area.  The bpf-tcp-cc may
5188          * overwrite this ptr after switching to cdg.
5189          */
5190         if (*optlen >= sizeof("cdg") - 1 && !strncmp("cdg", optval, *optlen))
5191                 return -ENOTSUPP;
5192
5193         /* It stops this looping
5194          *
5195          * .init => bpf_setsockopt(tcp_cc) => .init =>
5196          * bpf_setsockopt(tcp_cc)" => .init => ....
5197          *
5198          * The second bpf_setsockopt(tcp_cc) is not allowed
5199          * in order to break the loop when both .init
5200          * are the same bpf prog.
5201          *
5202          * This applies even the second bpf_setsockopt(tcp_cc)
5203          * does not cause a loop.  This limits only the first
5204          * '.init' can call bpf_setsockopt(TCP_CONGESTION) to
5205          * pick a fallback cc (eg. peer does not support ECN)
5206          * and the second '.init' cannot fallback to
5207          * another.
5208          */
5209         tp = tcp_sk(sk);
5210         if (tp->bpf_chg_cc_inprogress)
5211                 return -EBUSY;
5212
5213         tp->bpf_chg_cc_inprogress = 1;
5214         ret = do_tcp_setsockopt(sk, SOL_TCP, TCP_CONGESTION,
5215                                 KERNEL_SOCKPTR(optval), *optlen);
5216         tp->bpf_chg_cc_inprogress = 0;
5217         return ret;
5218 }
5219
5220 static int sol_tcp_sockopt(struct sock *sk, int optname,
5221                            char *optval, int *optlen,
5222                            bool getopt)
5223 {
5224         if (sk->sk_protocol != IPPROTO_TCP)
5225                 return -EINVAL;
5226
5227         switch (optname) {
5228         case TCP_NODELAY:
5229         case TCP_MAXSEG:
5230         case TCP_KEEPIDLE:
5231         case TCP_KEEPINTVL:
5232         case TCP_KEEPCNT:
5233         case TCP_SYNCNT:
5234         case TCP_WINDOW_CLAMP:
5235         case TCP_THIN_LINEAR_TIMEOUTS:
5236         case TCP_USER_TIMEOUT:
5237         case TCP_NOTSENT_LOWAT:
5238         case TCP_SAVE_SYN:
5239                 if (*optlen != sizeof(int))
5240                         return -EINVAL;
5241                 break;
5242         case TCP_CONGESTION:
5243                 return sol_tcp_sockopt_congestion(sk, optval, optlen, getopt);
5244         case TCP_SAVED_SYN:
5245                 if (*optlen < 1)
5246                         return -EINVAL;
5247                 break;
5248         default:
5249                 if (getopt)
5250                         return -EINVAL;
5251                 return bpf_sol_tcp_setsockopt(sk, optname, optval, *optlen);
5252         }
5253
5254         if (getopt) {
5255                 if (optname == TCP_SAVED_SYN) {
5256                         struct tcp_sock *tp = tcp_sk(sk);
5257
5258                         if (!tp->saved_syn ||
5259                             *optlen > tcp_saved_syn_len(tp->saved_syn))
5260                                 return -EINVAL;
5261                         memcpy(optval, tp->saved_syn->data, *optlen);
5262                         /* It cannot free tp->saved_syn here because it
5263                          * does not know if the user space still needs it.
5264                          */
5265                         return 0;
5266                 }
5267
5268                 return do_tcp_getsockopt(sk, SOL_TCP, optname,
5269                                          KERNEL_SOCKPTR(optval),
5270                                          KERNEL_SOCKPTR(optlen));
5271         }
5272
5273         return do_tcp_setsockopt(sk, SOL_TCP, optname,
5274                                  KERNEL_SOCKPTR(optval), *optlen);
5275 }
5276
5277 static int sol_ip_sockopt(struct sock *sk, int optname,
5278                           char *optval, int *optlen,
5279                           bool getopt)
5280 {
5281         if (sk->sk_family != AF_INET)
5282                 return -EINVAL;
5283
5284         switch (optname) {
5285         case IP_TOS:
5286                 if (*optlen != sizeof(int))
5287                         return -EINVAL;
5288                 break;
5289         default:
5290                 return -EINVAL;
5291         }
5292
5293         if (getopt)
5294                 return do_ip_getsockopt(sk, SOL_IP, optname,
5295                                         KERNEL_SOCKPTR(optval),
5296                                         KERNEL_SOCKPTR(optlen));
5297
5298         return do_ip_setsockopt(sk, SOL_IP, optname,
5299                                 KERNEL_SOCKPTR(optval), *optlen);
5300 }
5301
5302 static int sol_ipv6_sockopt(struct sock *sk, int optname,
5303                             char *optval, int *optlen,
5304                             bool getopt)
5305 {
5306         if (sk->sk_family != AF_INET6)
5307                 return -EINVAL;
5308
5309         switch (optname) {
5310         case IPV6_TCLASS:
5311         case IPV6_AUTOFLOWLABEL:
5312                 if (*optlen != sizeof(int))
5313                         return -EINVAL;
5314                 break;
5315         default:
5316                 return -EINVAL;
5317         }
5318
5319         if (getopt)
5320                 return ipv6_bpf_stub->ipv6_getsockopt(sk, SOL_IPV6, optname,
5321                                                       KERNEL_SOCKPTR(optval),
5322                                                       KERNEL_SOCKPTR(optlen));
5323
5324         return ipv6_bpf_stub->ipv6_setsockopt(sk, SOL_IPV6, optname,
5325                                               KERNEL_SOCKPTR(optval), *optlen);
5326 }
5327
5328 static int __bpf_setsockopt(struct sock *sk, int level, int optname,
5329                             char *optval, int optlen)
5330 {
5331         if (!sk_fullsock(sk))
5332                 return -EINVAL;
5333
5334         if (level == SOL_SOCKET)
5335                 return sol_socket_sockopt(sk, optname, optval, &optlen, false);
5336         else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5337                 return sol_ip_sockopt(sk, optname, optval, &optlen, false);
5338         else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5339                 return sol_ipv6_sockopt(sk, optname, optval, &optlen, false);
5340         else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5341                 return sol_tcp_sockopt(sk, optname, optval, &optlen, false);
5342
5343         return -EINVAL;
5344 }
5345
5346 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
5347                            char *optval, int optlen)
5348 {
5349         if (sk_fullsock(sk))
5350                 sock_owned_by_me(sk);
5351         return __bpf_setsockopt(sk, level, optname, optval, optlen);
5352 }
5353
5354 static int __bpf_getsockopt(struct sock *sk, int level, int optname,
5355                             char *optval, int optlen)
5356 {
5357         int err, saved_optlen = optlen;
5358
5359         if (!sk_fullsock(sk)) {
5360                 err = -EINVAL;
5361                 goto done;
5362         }
5363
5364         if (level == SOL_SOCKET)
5365                 err = sol_socket_sockopt(sk, optname, optval, &optlen, true);
5366         else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5367                 err = sol_tcp_sockopt(sk, optname, optval, &optlen, true);
5368         else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5369                 err = sol_ip_sockopt(sk, optname, optval, &optlen, true);
5370         else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5371                 err = sol_ipv6_sockopt(sk, optname, optval, &optlen, true);
5372         else
5373                 err = -EINVAL;
5374
5375 done:
5376         if (err)
5377                 optlen = 0;
5378         if (optlen < saved_optlen)
5379                 memset(optval + optlen, 0, saved_optlen - optlen);
5380         return err;
5381 }
5382
5383 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
5384                            char *optval, int optlen)
5385 {
5386         if (sk_fullsock(sk))
5387                 sock_owned_by_me(sk);
5388         return __bpf_getsockopt(sk, level, optname, optval, optlen);
5389 }
5390
5391 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level,
5392            int, optname, char *, optval, int, optlen)
5393 {
5394         return _bpf_setsockopt(sk, level, optname, optval, optlen);
5395 }
5396
5397 const struct bpf_func_proto bpf_sk_setsockopt_proto = {
5398         .func           = bpf_sk_setsockopt,
5399         .gpl_only       = false,
5400         .ret_type       = RET_INTEGER,
5401         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5402         .arg2_type      = ARG_ANYTHING,
5403         .arg3_type      = ARG_ANYTHING,
5404         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5405         .arg5_type      = ARG_CONST_SIZE,
5406 };
5407
5408 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level,
5409            int, optname, char *, optval, int, optlen)
5410 {
5411         return _bpf_getsockopt(sk, level, optname, optval, optlen);
5412 }
5413
5414 const struct bpf_func_proto bpf_sk_getsockopt_proto = {
5415         .func           = bpf_sk_getsockopt,
5416         .gpl_only       = false,
5417         .ret_type       = RET_INTEGER,
5418         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5419         .arg2_type      = ARG_ANYTHING,
5420         .arg3_type      = ARG_ANYTHING,
5421         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5422         .arg5_type      = ARG_CONST_SIZE,
5423 };
5424
5425 BPF_CALL_5(bpf_unlocked_sk_setsockopt, struct sock *, sk, int, level,
5426            int, optname, char *, optval, int, optlen)
5427 {
5428         return __bpf_setsockopt(sk, level, optname, optval, optlen);
5429 }
5430
5431 const struct bpf_func_proto bpf_unlocked_sk_setsockopt_proto = {
5432         .func           = bpf_unlocked_sk_setsockopt,
5433         .gpl_only       = false,
5434         .ret_type       = RET_INTEGER,
5435         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5436         .arg2_type      = ARG_ANYTHING,
5437         .arg3_type      = ARG_ANYTHING,
5438         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5439         .arg5_type      = ARG_CONST_SIZE,
5440 };
5441
5442 BPF_CALL_5(bpf_unlocked_sk_getsockopt, struct sock *, sk, int, level,
5443            int, optname, char *, optval, int, optlen)
5444 {
5445         return __bpf_getsockopt(sk, level, optname, optval, optlen);
5446 }
5447
5448 const struct bpf_func_proto bpf_unlocked_sk_getsockopt_proto = {
5449         .func           = bpf_unlocked_sk_getsockopt,
5450         .gpl_only       = false,
5451         .ret_type       = RET_INTEGER,
5452         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5453         .arg2_type      = ARG_ANYTHING,
5454         .arg3_type      = ARG_ANYTHING,
5455         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5456         .arg5_type      = ARG_CONST_SIZE,
5457 };
5458
5459 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5460            int, level, int, optname, char *, optval, int, optlen)
5461 {
5462         return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5463 }
5464
5465 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5466         .func           = bpf_sock_addr_setsockopt,
5467         .gpl_only       = false,
5468         .ret_type       = RET_INTEGER,
5469         .arg1_type      = ARG_PTR_TO_CTX,
5470         .arg2_type      = ARG_ANYTHING,
5471         .arg3_type      = ARG_ANYTHING,
5472         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5473         .arg5_type      = ARG_CONST_SIZE,
5474 };
5475
5476 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5477            int, level, int, optname, char *, optval, int, optlen)
5478 {
5479         return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5480 }
5481
5482 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5483         .func           = bpf_sock_addr_getsockopt,
5484         .gpl_only       = false,
5485         .ret_type       = RET_INTEGER,
5486         .arg1_type      = ARG_PTR_TO_CTX,
5487         .arg2_type      = ARG_ANYTHING,
5488         .arg3_type      = ARG_ANYTHING,
5489         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5490         .arg5_type      = ARG_CONST_SIZE,
5491 };
5492
5493 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5494            int, level, int, optname, char *, optval, int, optlen)
5495 {
5496         return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5497 }
5498
5499 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5500         .func           = bpf_sock_ops_setsockopt,
5501         .gpl_only       = false,
5502         .ret_type       = RET_INTEGER,
5503         .arg1_type      = ARG_PTR_TO_CTX,
5504         .arg2_type      = ARG_ANYTHING,
5505         .arg3_type      = ARG_ANYTHING,
5506         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5507         .arg5_type      = ARG_CONST_SIZE,
5508 };
5509
5510 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5511                                 int optname, const u8 **start)
5512 {
5513         struct sk_buff *syn_skb = bpf_sock->syn_skb;
5514         const u8 *hdr_start;
5515         int ret;
5516
5517         if (syn_skb) {
5518                 /* sk is a request_sock here */
5519
5520                 if (optname == TCP_BPF_SYN) {
5521                         hdr_start = syn_skb->data;
5522                         ret = tcp_hdrlen(syn_skb);
5523                 } else if (optname == TCP_BPF_SYN_IP) {
5524                         hdr_start = skb_network_header(syn_skb);
5525                         ret = skb_network_header_len(syn_skb) +
5526                                 tcp_hdrlen(syn_skb);
5527                 } else {
5528                         /* optname == TCP_BPF_SYN_MAC */
5529                         hdr_start = skb_mac_header(syn_skb);
5530                         ret = skb_mac_header_len(syn_skb) +
5531                                 skb_network_header_len(syn_skb) +
5532                                 tcp_hdrlen(syn_skb);
5533                 }
5534         } else {
5535                 struct sock *sk = bpf_sock->sk;
5536                 struct saved_syn *saved_syn;
5537
5538                 if (sk->sk_state == TCP_NEW_SYN_RECV)
5539                         /* synack retransmit. bpf_sock->syn_skb will
5540                          * not be available.  It has to resort to
5541                          * saved_syn (if it is saved).
5542                          */
5543                         saved_syn = inet_reqsk(sk)->saved_syn;
5544                 else
5545                         saved_syn = tcp_sk(sk)->saved_syn;
5546
5547                 if (!saved_syn)
5548                         return -ENOENT;
5549
5550                 if (optname == TCP_BPF_SYN) {
5551                         hdr_start = saved_syn->data +
5552                                 saved_syn->mac_hdrlen +
5553                                 saved_syn->network_hdrlen;
5554                         ret = saved_syn->tcp_hdrlen;
5555                 } else if (optname == TCP_BPF_SYN_IP) {
5556                         hdr_start = saved_syn->data +
5557                                 saved_syn->mac_hdrlen;
5558                         ret = saved_syn->network_hdrlen +
5559                                 saved_syn->tcp_hdrlen;
5560                 } else {
5561                         /* optname == TCP_BPF_SYN_MAC */
5562
5563                         /* TCP_SAVE_SYN may not have saved the mac hdr */
5564                         if (!saved_syn->mac_hdrlen)
5565                                 return -ENOENT;
5566
5567                         hdr_start = saved_syn->data;
5568                         ret = saved_syn->mac_hdrlen +
5569                                 saved_syn->network_hdrlen +
5570                                 saved_syn->tcp_hdrlen;
5571                 }
5572         }
5573
5574         *start = hdr_start;
5575         return ret;
5576 }
5577
5578 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5579            int, level, int, optname, char *, optval, int, optlen)
5580 {
5581         if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5582             optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5583                 int ret, copy_len = 0;
5584                 const u8 *start;
5585
5586                 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5587                 if (ret > 0) {
5588                         copy_len = ret;
5589                         if (optlen < copy_len) {
5590                                 copy_len = optlen;
5591                                 ret = -ENOSPC;
5592                         }
5593
5594                         memcpy(optval, start, copy_len);
5595                 }
5596
5597                 /* Zero out unused buffer at the end */
5598                 memset(optval + copy_len, 0, optlen - copy_len);
5599
5600                 return ret;
5601         }
5602
5603         return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5604 }
5605
5606 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5607         .func           = bpf_sock_ops_getsockopt,
5608         .gpl_only       = false,
5609         .ret_type       = RET_INTEGER,
5610         .arg1_type      = ARG_PTR_TO_CTX,
5611         .arg2_type      = ARG_ANYTHING,
5612         .arg3_type      = ARG_ANYTHING,
5613         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5614         .arg5_type      = ARG_CONST_SIZE,
5615 };
5616
5617 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5618            int, argval)
5619 {
5620         struct sock *sk = bpf_sock->sk;
5621         int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5622
5623         if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5624                 return -EINVAL;
5625
5626         tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5627
5628         return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5629 }
5630
5631 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5632         .func           = bpf_sock_ops_cb_flags_set,
5633         .gpl_only       = false,
5634         .ret_type       = RET_INTEGER,
5635         .arg1_type      = ARG_PTR_TO_CTX,
5636         .arg2_type      = ARG_ANYTHING,
5637 };
5638
5639 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5640 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5641
5642 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5643            int, addr_len)
5644 {
5645 #ifdef CONFIG_INET
5646         struct sock *sk = ctx->sk;
5647         u32 flags = BIND_FROM_BPF;
5648         int err;
5649
5650         err = -EINVAL;
5651         if (addr_len < offsetofend(struct sockaddr, sa_family))
5652                 return err;
5653         if (addr->sa_family == AF_INET) {
5654                 if (addr_len < sizeof(struct sockaddr_in))
5655                         return err;
5656                 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5657                         flags |= BIND_FORCE_ADDRESS_NO_PORT;
5658                 return __inet_bind(sk, addr, addr_len, flags);
5659 #if IS_ENABLED(CONFIG_IPV6)
5660         } else if (addr->sa_family == AF_INET6) {
5661                 if (addr_len < SIN6_LEN_RFC2133)
5662                         return err;
5663                 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5664                         flags |= BIND_FORCE_ADDRESS_NO_PORT;
5665                 /* ipv6_bpf_stub cannot be NULL, since it's called from
5666                  * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5667                  */
5668                 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5669 #endif /* CONFIG_IPV6 */
5670         }
5671 #endif /* CONFIG_INET */
5672
5673         return -EAFNOSUPPORT;
5674 }
5675
5676 static const struct bpf_func_proto bpf_bind_proto = {
5677         .func           = bpf_bind,
5678         .gpl_only       = false,
5679         .ret_type       = RET_INTEGER,
5680         .arg1_type      = ARG_PTR_TO_CTX,
5681         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5682         .arg3_type      = ARG_CONST_SIZE,
5683 };
5684
5685 #ifdef CONFIG_XFRM
5686
5687 #if (IS_BUILTIN(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF)) || \
5688     (IS_MODULE(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES))
5689
5690 struct metadata_dst __percpu *xfrm_bpf_md_dst;
5691 EXPORT_SYMBOL_GPL(xfrm_bpf_md_dst);
5692
5693 #endif
5694
5695 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5696            struct bpf_xfrm_state *, to, u32, size, u64, flags)
5697 {
5698         const struct sec_path *sp = skb_sec_path(skb);
5699         const struct xfrm_state *x;
5700
5701         if (!sp || unlikely(index >= sp->len || flags))
5702                 goto err_clear;
5703
5704         x = sp->xvec[index];
5705
5706         if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5707                 goto err_clear;
5708
5709         to->reqid = x->props.reqid;
5710         to->spi = x->id.spi;
5711         to->family = x->props.family;
5712         to->ext = 0;
5713
5714         if (to->family == AF_INET6) {
5715                 memcpy(to->remote_ipv6, x->props.saddr.a6,
5716                        sizeof(to->remote_ipv6));
5717         } else {
5718                 to->remote_ipv4 = x->props.saddr.a4;
5719                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5720         }
5721
5722         return 0;
5723 err_clear:
5724         memset(to, 0, size);
5725         return -EINVAL;
5726 }
5727
5728 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5729         .func           = bpf_skb_get_xfrm_state,
5730         .gpl_only       = false,
5731         .ret_type       = RET_INTEGER,
5732         .arg1_type      = ARG_PTR_TO_CTX,
5733         .arg2_type      = ARG_ANYTHING,
5734         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
5735         .arg4_type      = ARG_CONST_SIZE,
5736         .arg5_type      = ARG_ANYTHING,
5737 };
5738 #endif
5739
5740 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5741 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params, u32 mtu)
5742 {
5743         params->h_vlan_TCI = 0;
5744         params->h_vlan_proto = 0;
5745         if (mtu)
5746                 params->mtu_result = mtu; /* union with tot_len */
5747
5748         return 0;
5749 }
5750 #endif
5751
5752 #if IS_ENABLED(CONFIG_INET)
5753 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5754                                u32 flags, bool check_mtu)
5755 {
5756         struct fib_nh_common *nhc;
5757         struct in_device *in_dev;
5758         struct neighbour *neigh;
5759         struct net_device *dev;
5760         struct fib_result res;
5761         struct flowi4 fl4;
5762         u32 mtu = 0;
5763         int err;
5764
5765         dev = dev_get_by_index_rcu(net, params->ifindex);
5766         if (unlikely(!dev))
5767                 return -ENODEV;
5768
5769         /* verify forwarding is enabled on this interface */
5770         in_dev = __in_dev_get_rcu(dev);
5771         if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5772                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5773
5774         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5775                 fl4.flowi4_iif = 1;
5776                 fl4.flowi4_oif = params->ifindex;
5777         } else {
5778                 fl4.flowi4_iif = params->ifindex;
5779                 fl4.flowi4_oif = 0;
5780         }
5781         fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5782         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5783         fl4.flowi4_flags = 0;
5784
5785         fl4.flowi4_proto = params->l4_protocol;
5786         fl4.daddr = params->ipv4_dst;
5787         fl4.saddr = params->ipv4_src;
5788         fl4.fl4_sport = params->sport;
5789         fl4.fl4_dport = params->dport;
5790         fl4.flowi4_multipath_hash = 0;
5791
5792         if (flags & BPF_FIB_LOOKUP_DIRECT) {
5793                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5794                 struct fib_table *tb;
5795
5796                 if (flags & BPF_FIB_LOOKUP_TBID) {
5797                         tbid = params->tbid;
5798                         /* zero out for vlan output */
5799                         params->tbid = 0;
5800                 }
5801
5802                 tb = fib_get_table(net, tbid);
5803                 if (unlikely(!tb))
5804                         return BPF_FIB_LKUP_RET_NOT_FWDED;
5805
5806                 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5807         } else {
5808                 fl4.flowi4_mark = 0;
5809                 fl4.flowi4_secid = 0;
5810                 fl4.flowi4_tun_key.tun_id = 0;
5811                 fl4.flowi4_uid = sock_net_uid(net, NULL);
5812
5813                 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5814         }
5815
5816         if (err) {
5817                 /* map fib lookup errors to RTN_ type */
5818                 if (err == -EINVAL)
5819                         return BPF_FIB_LKUP_RET_BLACKHOLE;
5820                 if (err == -EHOSTUNREACH)
5821                         return BPF_FIB_LKUP_RET_UNREACHABLE;
5822                 if (err == -EACCES)
5823                         return BPF_FIB_LKUP_RET_PROHIBIT;
5824
5825                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5826         }
5827
5828         if (res.type != RTN_UNICAST)
5829                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5830
5831         if (fib_info_num_path(res.fi) > 1)
5832                 fib_select_path(net, &res, &fl4, NULL);
5833
5834         if (check_mtu) {
5835                 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5836                 if (params->tot_len > mtu) {
5837                         params->mtu_result = mtu; /* union with tot_len */
5838                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5839                 }
5840         }
5841
5842         nhc = res.nhc;
5843
5844         /* do not handle lwt encaps right now */
5845         if (nhc->nhc_lwtstate)
5846                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5847
5848         dev = nhc->nhc_dev;
5849
5850         params->rt_metric = res.fi->fib_priority;
5851         params->ifindex = dev->ifindex;
5852
5853         /* xdp and cls_bpf programs are run in RCU-bh so
5854          * rcu_read_lock_bh is not needed here
5855          */
5856         if (likely(nhc->nhc_gw_family != AF_INET6)) {
5857                 if (nhc->nhc_gw_family)
5858                         params->ipv4_dst = nhc->nhc_gw.ipv4;
5859         } else {
5860                 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5861
5862                 params->family = AF_INET6;
5863                 *dst = nhc->nhc_gw.ipv6;
5864         }
5865
5866         if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH)
5867                 goto set_fwd_params;
5868
5869         if (likely(nhc->nhc_gw_family != AF_INET6))
5870                 neigh = __ipv4_neigh_lookup_noref(dev,
5871                                                   (__force u32)params->ipv4_dst);
5872         else
5873                 neigh = __ipv6_neigh_lookup_noref_stub(dev, params->ipv6_dst);
5874
5875         if (!neigh || !(READ_ONCE(neigh->nud_state) & NUD_VALID))
5876                 return BPF_FIB_LKUP_RET_NO_NEIGH;
5877         memcpy(params->dmac, neigh->ha, ETH_ALEN);
5878         memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5879
5880 set_fwd_params:
5881         return bpf_fib_set_fwd_params(params, mtu);
5882 }
5883 #endif
5884
5885 #if IS_ENABLED(CONFIG_IPV6)
5886 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5887                                u32 flags, bool check_mtu)
5888 {
5889         struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5890         struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5891         struct fib6_result res = {};
5892         struct neighbour *neigh;
5893         struct net_device *dev;
5894         struct inet6_dev *idev;
5895         struct flowi6 fl6;
5896         int strict = 0;
5897         int oif, err;
5898         u32 mtu = 0;
5899
5900         /* link local addresses are never forwarded */
5901         if (rt6_need_strict(dst) || rt6_need_strict(src))
5902                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5903
5904         dev = dev_get_by_index_rcu(net, params->ifindex);
5905         if (unlikely(!dev))
5906                 return -ENODEV;
5907
5908         idev = __in6_dev_get_safely(dev);
5909         if (unlikely(!idev || !idev->cnf.forwarding))
5910                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5911
5912         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5913                 fl6.flowi6_iif = 1;
5914                 oif = fl6.flowi6_oif = params->ifindex;
5915         } else {
5916                 oif = fl6.flowi6_iif = params->ifindex;
5917                 fl6.flowi6_oif = 0;
5918                 strict = RT6_LOOKUP_F_HAS_SADDR;
5919         }
5920         fl6.flowlabel = params->flowinfo;
5921         fl6.flowi6_scope = 0;
5922         fl6.flowi6_flags = 0;
5923         fl6.mp_hash = 0;
5924
5925         fl6.flowi6_proto = params->l4_protocol;
5926         fl6.daddr = *dst;
5927         fl6.saddr = *src;
5928         fl6.fl6_sport = params->sport;
5929         fl6.fl6_dport = params->dport;
5930
5931         if (flags & BPF_FIB_LOOKUP_DIRECT) {
5932                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5933                 struct fib6_table *tb;
5934
5935                 if (flags & BPF_FIB_LOOKUP_TBID) {
5936                         tbid = params->tbid;
5937                         /* zero out for vlan output */
5938                         params->tbid = 0;
5939                 }
5940
5941                 tb = ipv6_stub->fib6_get_table(net, tbid);
5942                 if (unlikely(!tb))
5943                         return BPF_FIB_LKUP_RET_NOT_FWDED;
5944
5945                 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5946                                                    strict);
5947         } else {
5948                 fl6.flowi6_mark = 0;
5949                 fl6.flowi6_secid = 0;
5950                 fl6.flowi6_tun_key.tun_id = 0;
5951                 fl6.flowi6_uid = sock_net_uid(net, NULL);
5952
5953                 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5954         }
5955
5956         if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5957                      res.f6i == net->ipv6.fib6_null_entry))
5958                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5959
5960         switch (res.fib6_type) {
5961         /* only unicast is forwarded */
5962         case RTN_UNICAST:
5963                 break;
5964         case RTN_BLACKHOLE:
5965                 return BPF_FIB_LKUP_RET_BLACKHOLE;
5966         case RTN_UNREACHABLE:
5967                 return BPF_FIB_LKUP_RET_UNREACHABLE;
5968         case RTN_PROHIBIT:
5969                 return BPF_FIB_LKUP_RET_PROHIBIT;
5970         default:
5971                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5972         }
5973
5974         ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5975                                     fl6.flowi6_oif != 0, NULL, strict);
5976
5977         if (check_mtu) {
5978                 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5979                 if (params->tot_len > mtu) {
5980                         params->mtu_result = mtu; /* union with tot_len */
5981                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5982                 }
5983         }
5984
5985         if (res.nh->fib_nh_lws)
5986                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5987
5988         if (res.nh->fib_nh_gw_family)
5989                 *dst = res.nh->fib_nh_gw6;
5990
5991         dev = res.nh->fib_nh_dev;
5992         params->rt_metric = res.f6i->fib6_metric;
5993         params->ifindex = dev->ifindex;
5994
5995         if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH)
5996                 goto set_fwd_params;
5997
5998         /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
5999          * not needed here.
6000          */
6001         neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
6002         if (!neigh || !(READ_ONCE(neigh->nud_state) & NUD_VALID))
6003                 return BPF_FIB_LKUP_RET_NO_NEIGH;
6004         memcpy(params->dmac, neigh->ha, ETH_ALEN);
6005         memcpy(params->smac, dev->dev_addr, ETH_ALEN);
6006
6007 set_fwd_params:
6008         return bpf_fib_set_fwd_params(params, mtu);
6009 }
6010 #endif
6011
6012 #define BPF_FIB_LOOKUP_MASK (BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT | \
6013                              BPF_FIB_LOOKUP_SKIP_NEIGH | BPF_FIB_LOOKUP_TBID)
6014
6015 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
6016            struct bpf_fib_lookup *, params, int, plen, u32, flags)
6017 {
6018         if (plen < sizeof(*params))
6019                 return -EINVAL;
6020
6021         if (flags & ~BPF_FIB_LOOKUP_MASK)
6022                 return -EINVAL;
6023
6024         switch (params->family) {
6025 #if IS_ENABLED(CONFIG_INET)
6026         case AF_INET:
6027                 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
6028                                            flags, true);
6029 #endif
6030 #if IS_ENABLED(CONFIG_IPV6)
6031         case AF_INET6:
6032                 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
6033                                            flags, true);
6034 #endif
6035         }
6036         return -EAFNOSUPPORT;
6037 }
6038
6039 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
6040         .func           = bpf_xdp_fib_lookup,
6041         .gpl_only       = true,
6042         .ret_type       = RET_INTEGER,
6043         .arg1_type      = ARG_PTR_TO_CTX,
6044         .arg2_type      = ARG_PTR_TO_MEM,
6045         .arg3_type      = ARG_CONST_SIZE,
6046         .arg4_type      = ARG_ANYTHING,
6047 };
6048
6049 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
6050            struct bpf_fib_lookup *, params, int, plen, u32, flags)
6051 {
6052         struct net *net = dev_net(skb->dev);
6053         int rc = -EAFNOSUPPORT;
6054         bool check_mtu = false;
6055
6056         if (plen < sizeof(*params))
6057                 return -EINVAL;
6058
6059         if (flags & ~BPF_FIB_LOOKUP_MASK)
6060                 return -EINVAL;
6061
6062         if (params->tot_len)
6063                 check_mtu = true;
6064
6065         switch (params->family) {
6066 #if IS_ENABLED(CONFIG_INET)
6067         case AF_INET:
6068                 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
6069                 break;
6070 #endif
6071 #if IS_ENABLED(CONFIG_IPV6)
6072         case AF_INET6:
6073                 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
6074                 break;
6075 #endif
6076         }
6077
6078         if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
6079                 struct net_device *dev;
6080
6081                 /* When tot_len isn't provided by user, check skb
6082                  * against MTU of FIB lookup resulting net_device
6083                  */
6084                 dev = dev_get_by_index_rcu(net, params->ifindex);
6085                 if (!is_skb_forwardable(dev, skb))
6086                         rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
6087
6088                 params->mtu_result = dev->mtu; /* union with tot_len */
6089         }
6090
6091         return rc;
6092 }
6093
6094 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
6095         .func           = bpf_skb_fib_lookup,
6096         .gpl_only       = true,
6097         .ret_type       = RET_INTEGER,
6098         .arg1_type      = ARG_PTR_TO_CTX,
6099         .arg2_type      = ARG_PTR_TO_MEM,
6100         .arg3_type      = ARG_CONST_SIZE,
6101         .arg4_type      = ARG_ANYTHING,
6102 };
6103
6104 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
6105                                             u32 ifindex)
6106 {
6107         struct net *netns = dev_net(dev_curr);
6108
6109         /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
6110         if (ifindex == 0)
6111                 return dev_curr;
6112
6113         return dev_get_by_index_rcu(netns, ifindex);
6114 }
6115
6116 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
6117            u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6118 {
6119         int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6120         struct net_device *dev = skb->dev;
6121         int skb_len, dev_len;
6122         int mtu;
6123
6124         if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
6125                 return -EINVAL;
6126
6127         if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
6128                 return -EINVAL;
6129
6130         dev = __dev_via_ifindex(dev, ifindex);
6131         if (unlikely(!dev))
6132                 return -ENODEV;
6133
6134         mtu = READ_ONCE(dev->mtu);
6135
6136         dev_len = mtu + dev->hard_header_len;
6137
6138         /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6139         skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
6140
6141         skb_len += len_diff; /* minus result pass check */
6142         if (skb_len <= dev_len) {
6143                 ret = BPF_MTU_CHK_RET_SUCCESS;
6144                 goto out;
6145         }
6146         /* At this point, skb->len exceed MTU, but as it include length of all
6147          * segments, it can still be below MTU.  The SKB can possibly get
6148          * re-segmented in transmit path (see validate_xmit_skb).  Thus, user
6149          * must choose if segs are to be MTU checked.
6150          */
6151         if (skb_is_gso(skb)) {
6152                 ret = BPF_MTU_CHK_RET_SUCCESS;
6153
6154                 if (flags & BPF_MTU_CHK_SEGS &&
6155                     !skb_gso_validate_network_len(skb, mtu))
6156                         ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
6157         }
6158 out:
6159         /* BPF verifier guarantees valid pointer */
6160         *mtu_len = mtu;
6161
6162         return ret;
6163 }
6164
6165 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
6166            u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6167 {
6168         struct net_device *dev = xdp->rxq->dev;
6169         int xdp_len = xdp->data_end - xdp->data;
6170         int ret = BPF_MTU_CHK_RET_SUCCESS;
6171         int mtu, dev_len;
6172
6173         /* XDP variant doesn't support multi-buffer segment check (yet) */
6174         if (unlikely(flags))
6175                 return -EINVAL;
6176
6177         dev = __dev_via_ifindex(dev, ifindex);
6178         if (unlikely(!dev))
6179                 return -ENODEV;
6180
6181         mtu = READ_ONCE(dev->mtu);
6182
6183         /* Add L2-header as dev MTU is L3 size */
6184         dev_len = mtu + dev->hard_header_len;
6185
6186         /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6187         if (*mtu_len)
6188                 xdp_len = *mtu_len + dev->hard_header_len;
6189
6190         xdp_len += len_diff; /* minus result pass check */
6191         if (xdp_len > dev_len)
6192                 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6193
6194         /* BPF verifier guarantees valid pointer */
6195         *mtu_len = mtu;
6196
6197         return ret;
6198 }
6199
6200 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
6201         .func           = bpf_skb_check_mtu,
6202         .gpl_only       = true,
6203         .ret_type       = RET_INTEGER,
6204         .arg1_type      = ARG_PTR_TO_CTX,
6205         .arg2_type      = ARG_ANYTHING,
6206         .arg3_type      = ARG_PTR_TO_INT,
6207         .arg4_type      = ARG_ANYTHING,
6208         .arg5_type      = ARG_ANYTHING,
6209 };
6210
6211 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
6212         .func           = bpf_xdp_check_mtu,
6213         .gpl_only       = true,
6214         .ret_type       = RET_INTEGER,
6215         .arg1_type      = ARG_PTR_TO_CTX,
6216         .arg2_type      = ARG_ANYTHING,
6217         .arg3_type      = ARG_PTR_TO_INT,
6218         .arg4_type      = ARG_ANYTHING,
6219         .arg5_type      = ARG_ANYTHING,
6220 };
6221
6222 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6223 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
6224 {
6225         int err;
6226         struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
6227
6228         if (!seg6_validate_srh(srh, len, false))
6229                 return -EINVAL;
6230
6231         switch (type) {
6232         case BPF_LWT_ENCAP_SEG6_INLINE:
6233                 if (skb->protocol != htons(ETH_P_IPV6))
6234                         return -EBADMSG;
6235
6236                 err = seg6_do_srh_inline(skb, srh);
6237                 break;
6238         case BPF_LWT_ENCAP_SEG6:
6239                 skb_reset_inner_headers(skb);
6240                 skb->encapsulation = 1;
6241                 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
6242                 break;
6243         default:
6244                 return -EINVAL;
6245         }
6246
6247         bpf_compute_data_pointers(skb);
6248         if (err)
6249                 return err;
6250
6251         skb_set_transport_header(skb, sizeof(struct ipv6hdr));
6252
6253         return seg6_lookup_nexthop(skb, NULL, 0);
6254 }
6255 #endif /* CONFIG_IPV6_SEG6_BPF */
6256
6257 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6258 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
6259                              bool ingress)
6260 {
6261         return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
6262 }
6263 #endif
6264
6265 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
6266            u32, len)
6267 {
6268         switch (type) {
6269 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6270         case BPF_LWT_ENCAP_SEG6:
6271         case BPF_LWT_ENCAP_SEG6_INLINE:
6272                 return bpf_push_seg6_encap(skb, type, hdr, len);
6273 #endif
6274 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6275         case BPF_LWT_ENCAP_IP:
6276                 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
6277 #endif
6278         default:
6279                 return -EINVAL;
6280         }
6281 }
6282
6283 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
6284            void *, hdr, u32, len)
6285 {
6286         switch (type) {
6287 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6288         case BPF_LWT_ENCAP_IP:
6289                 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
6290 #endif
6291         default:
6292                 return -EINVAL;
6293         }
6294 }
6295
6296 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
6297         .func           = bpf_lwt_in_push_encap,
6298         .gpl_only       = false,
6299         .ret_type       = RET_INTEGER,
6300         .arg1_type      = ARG_PTR_TO_CTX,
6301         .arg2_type      = ARG_ANYTHING,
6302         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6303         .arg4_type      = ARG_CONST_SIZE
6304 };
6305
6306 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
6307         .func           = bpf_lwt_xmit_push_encap,
6308         .gpl_only       = false,
6309         .ret_type       = RET_INTEGER,
6310         .arg1_type      = ARG_PTR_TO_CTX,
6311         .arg2_type      = ARG_ANYTHING,
6312         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6313         .arg4_type      = ARG_CONST_SIZE
6314 };
6315
6316 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6317 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
6318            const void *, from, u32, len)
6319 {
6320         struct seg6_bpf_srh_state *srh_state =
6321                 this_cpu_ptr(&seg6_bpf_srh_states);
6322         struct ipv6_sr_hdr *srh = srh_state->srh;
6323         void *srh_tlvs, *srh_end, *ptr;
6324         int srhoff = 0;
6325
6326         if (srh == NULL)
6327                 return -EINVAL;
6328
6329         srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
6330         srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
6331
6332         ptr = skb->data + offset;
6333         if (ptr >= srh_tlvs && ptr + len <= srh_end)
6334                 srh_state->valid = false;
6335         else if (ptr < (void *)&srh->flags ||
6336                  ptr + len > (void *)&srh->segments)
6337                 return -EFAULT;
6338
6339         if (unlikely(bpf_try_make_writable(skb, offset + len)))
6340                 return -EFAULT;
6341         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6342                 return -EINVAL;
6343         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6344
6345         memcpy(skb->data + offset, from, len);
6346         return 0;
6347 }
6348
6349 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
6350         .func           = bpf_lwt_seg6_store_bytes,
6351         .gpl_only       = false,
6352         .ret_type       = RET_INTEGER,
6353         .arg1_type      = ARG_PTR_TO_CTX,
6354         .arg2_type      = ARG_ANYTHING,
6355         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6356         .arg4_type      = ARG_CONST_SIZE
6357 };
6358
6359 static void bpf_update_srh_state(struct sk_buff *skb)
6360 {
6361         struct seg6_bpf_srh_state *srh_state =
6362                 this_cpu_ptr(&seg6_bpf_srh_states);
6363         int srhoff = 0;
6364
6365         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
6366                 srh_state->srh = NULL;
6367         } else {
6368                 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6369                 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
6370                 srh_state->valid = true;
6371         }
6372 }
6373
6374 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
6375            u32, action, void *, param, u32, param_len)
6376 {
6377         struct seg6_bpf_srh_state *srh_state =
6378                 this_cpu_ptr(&seg6_bpf_srh_states);
6379         int hdroff = 0;
6380         int err;
6381
6382         switch (action) {
6383         case SEG6_LOCAL_ACTION_END_X:
6384                 if (!seg6_bpf_has_valid_srh(skb))
6385                         return -EBADMSG;
6386                 if (param_len != sizeof(struct in6_addr))
6387                         return -EINVAL;
6388                 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
6389         case SEG6_LOCAL_ACTION_END_T:
6390                 if (!seg6_bpf_has_valid_srh(skb))
6391                         return -EBADMSG;
6392                 if (param_len != sizeof(int))
6393                         return -EINVAL;
6394                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6395         case SEG6_LOCAL_ACTION_END_DT6:
6396                 if (!seg6_bpf_has_valid_srh(skb))
6397                         return -EBADMSG;
6398                 if (param_len != sizeof(int))
6399                         return -EINVAL;
6400
6401                 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
6402                         return -EBADMSG;
6403                 if (!pskb_pull(skb, hdroff))
6404                         return -EBADMSG;
6405
6406                 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
6407                 skb_reset_network_header(skb);
6408                 skb_reset_transport_header(skb);
6409                 skb->encapsulation = 0;
6410
6411                 bpf_compute_data_pointers(skb);
6412                 bpf_update_srh_state(skb);
6413                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6414         case SEG6_LOCAL_ACTION_END_B6:
6415                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6416                         return -EBADMSG;
6417                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
6418                                           param, param_len);
6419                 if (!err)
6420                         bpf_update_srh_state(skb);
6421
6422                 return err;
6423         case SEG6_LOCAL_ACTION_END_B6_ENCAP:
6424                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6425                         return -EBADMSG;
6426                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
6427                                           param, param_len);
6428                 if (!err)
6429                         bpf_update_srh_state(skb);
6430
6431                 return err;
6432         default:
6433                 return -EINVAL;
6434         }
6435 }
6436
6437 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
6438         .func           = bpf_lwt_seg6_action,
6439         .gpl_only       = false,
6440         .ret_type       = RET_INTEGER,
6441         .arg1_type      = ARG_PTR_TO_CTX,
6442         .arg2_type      = ARG_ANYTHING,
6443         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6444         .arg4_type      = ARG_CONST_SIZE
6445 };
6446
6447 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
6448            s32, len)
6449 {
6450         struct seg6_bpf_srh_state *srh_state =
6451                 this_cpu_ptr(&seg6_bpf_srh_states);
6452         struct ipv6_sr_hdr *srh = srh_state->srh;
6453         void *srh_end, *srh_tlvs, *ptr;
6454         struct ipv6hdr *hdr;
6455         int srhoff = 0;
6456         int ret;
6457
6458         if (unlikely(srh == NULL))
6459                 return -EINVAL;
6460
6461         srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
6462                         ((srh->first_segment + 1) << 4));
6463         srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
6464                         srh_state->hdrlen);
6465         ptr = skb->data + offset;
6466
6467         if (unlikely(ptr < srh_tlvs || ptr > srh_end))
6468                 return -EFAULT;
6469         if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
6470                 return -EFAULT;
6471
6472         if (len > 0) {
6473                 ret = skb_cow_head(skb, len);
6474                 if (unlikely(ret < 0))
6475                         return ret;
6476
6477                 ret = bpf_skb_net_hdr_push(skb, offset, len);
6478         } else {
6479                 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6480         }
6481
6482         bpf_compute_data_pointers(skb);
6483         if (unlikely(ret < 0))
6484                 return ret;
6485
6486         hdr = (struct ipv6hdr *)skb->data;
6487         hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6488
6489         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6490                 return -EINVAL;
6491         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6492         srh_state->hdrlen += len;
6493         srh_state->valid = false;
6494         return 0;
6495 }
6496
6497 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6498         .func           = bpf_lwt_seg6_adjust_srh,
6499         .gpl_only       = false,
6500         .ret_type       = RET_INTEGER,
6501         .arg1_type      = ARG_PTR_TO_CTX,
6502         .arg2_type      = ARG_ANYTHING,
6503         .arg3_type      = ARG_ANYTHING,
6504 };
6505 #endif /* CONFIG_IPV6_SEG6_BPF */
6506
6507 #ifdef CONFIG_INET
6508 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6509                               int dif, int sdif, u8 family, u8 proto)
6510 {
6511         struct inet_hashinfo *hinfo = net->ipv4.tcp_death_row.hashinfo;
6512         bool refcounted = false;
6513         struct sock *sk = NULL;
6514
6515         if (family == AF_INET) {
6516                 __be32 src4 = tuple->ipv4.saddr;
6517                 __be32 dst4 = tuple->ipv4.daddr;
6518
6519                 if (proto == IPPROTO_TCP)
6520                         sk = __inet_lookup(net, hinfo, NULL, 0,
6521                                            src4, tuple->ipv4.sport,
6522                                            dst4, tuple->ipv4.dport,
6523                                            dif, sdif, &refcounted);
6524                 else
6525                         sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6526                                                dst4, tuple->ipv4.dport,
6527                                                dif, sdif, net->ipv4.udp_table, NULL);
6528 #if IS_ENABLED(CONFIG_IPV6)
6529         } else {
6530                 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6531                 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6532
6533                 if (proto == IPPROTO_TCP)
6534                         sk = __inet6_lookup(net, hinfo, NULL, 0,
6535                                             src6, tuple->ipv6.sport,
6536                                             dst6, ntohs(tuple->ipv6.dport),
6537                                             dif, sdif, &refcounted);
6538                 else if (likely(ipv6_bpf_stub))
6539                         sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6540                                                             src6, tuple->ipv6.sport,
6541                                                             dst6, tuple->ipv6.dport,
6542                                                             dif, sdif,
6543                                                             net->ipv4.udp_table, NULL);
6544 #endif
6545         }
6546
6547         if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6548                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6549                 sk = NULL;
6550         }
6551         return sk;
6552 }
6553
6554 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6555  * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6556  */
6557 static struct sock *
6558 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6559                  struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6560                  u64 flags, int sdif)
6561 {
6562         struct sock *sk = NULL;
6563         struct net *net;
6564         u8 family;
6565
6566         if (len == sizeof(tuple->ipv4))
6567                 family = AF_INET;
6568         else if (len == sizeof(tuple->ipv6))
6569                 family = AF_INET6;
6570         else
6571                 return NULL;
6572
6573         if (unlikely(flags || !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6574                 goto out;
6575
6576         if (sdif < 0) {
6577                 if (family == AF_INET)
6578                         sdif = inet_sdif(skb);
6579                 else
6580                         sdif = inet6_sdif(skb);
6581         }
6582
6583         if ((s32)netns_id < 0) {
6584                 net = caller_net;
6585                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6586         } else {
6587                 net = get_net_ns_by_id(caller_net, netns_id);
6588                 if (unlikely(!net))
6589                         goto out;
6590                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6591                 put_net(net);
6592         }
6593
6594 out:
6595         return sk;
6596 }
6597
6598 static struct sock *
6599 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6600                 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6601                 u64 flags, int sdif)
6602 {
6603         struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6604                                            ifindex, proto, netns_id, flags,
6605                                            sdif);
6606
6607         if (sk) {
6608                 struct sock *sk2 = sk_to_full_sk(sk);
6609
6610                 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6611                  * sock refcnt is decremented to prevent a request_sock leak.
6612                  */
6613                 if (!sk_fullsock(sk2))
6614                         sk2 = NULL;
6615                 if (sk2 != sk) {
6616                         sock_gen_put(sk);
6617                         /* Ensure there is no need to bump sk2 refcnt */
6618                         if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6619                                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6620                                 return NULL;
6621                         }
6622                         sk = sk2;
6623                 }
6624         }
6625
6626         return sk;
6627 }
6628
6629 static struct sock *
6630 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6631                u8 proto, u64 netns_id, u64 flags)
6632 {
6633         struct net *caller_net;
6634         int ifindex;
6635
6636         if (skb->dev) {
6637                 caller_net = dev_net(skb->dev);
6638                 ifindex = skb->dev->ifindex;
6639         } else {
6640                 caller_net = sock_net(skb->sk);
6641                 ifindex = 0;
6642         }
6643
6644         return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6645                                 netns_id, flags, -1);
6646 }
6647
6648 static struct sock *
6649 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6650               u8 proto, u64 netns_id, u64 flags)
6651 {
6652         struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6653                                          flags);
6654
6655         if (sk) {
6656                 struct sock *sk2 = sk_to_full_sk(sk);
6657
6658                 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6659                  * sock refcnt is decremented to prevent a request_sock leak.
6660                  */
6661                 if (!sk_fullsock(sk2))
6662                         sk2 = NULL;
6663                 if (sk2 != sk) {
6664                         sock_gen_put(sk);
6665                         /* Ensure there is no need to bump sk2 refcnt */
6666                         if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6667                                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6668                                 return NULL;
6669                         }
6670                         sk = sk2;
6671                 }
6672         }
6673
6674         return sk;
6675 }
6676
6677 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6678            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6679 {
6680         return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6681                                              netns_id, flags);
6682 }
6683
6684 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6685         .func           = bpf_skc_lookup_tcp,
6686         .gpl_only       = false,
6687         .pkt_access     = true,
6688         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6689         .arg1_type      = ARG_PTR_TO_CTX,
6690         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6691         .arg3_type      = ARG_CONST_SIZE,
6692         .arg4_type      = ARG_ANYTHING,
6693         .arg5_type      = ARG_ANYTHING,
6694 };
6695
6696 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6697            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6698 {
6699         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6700                                             netns_id, flags);
6701 }
6702
6703 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6704         .func           = bpf_sk_lookup_tcp,
6705         .gpl_only       = false,
6706         .pkt_access     = true,
6707         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6708         .arg1_type      = ARG_PTR_TO_CTX,
6709         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6710         .arg3_type      = ARG_CONST_SIZE,
6711         .arg4_type      = ARG_ANYTHING,
6712         .arg5_type      = ARG_ANYTHING,
6713 };
6714
6715 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6716            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6717 {
6718         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6719                                             netns_id, flags);
6720 }
6721
6722 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6723         .func           = bpf_sk_lookup_udp,
6724         .gpl_only       = false,
6725         .pkt_access     = true,
6726         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6727         .arg1_type      = ARG_PTR_TO_CTX,
6728         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6729         .arg3_type      = ARG_CONST_SIZE,
6730         .arg4_type      = ARG_ANYTHING,
6731         .arg5_type      = ARG_ANYTHING,
6732 };
6733
6734 BPF_CALL_5(bpf_tc_skc_lookup_tcp, struct sk_buff *, skb,
6735            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6736 {
6737         struct net_device *dev = skb->dev;
6738         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6739         struct net *caller_net = dev_net(dev);
6740
6741         return (unsigned long)__bpf_skc_lookup(skb, tuple, len, caller_net,
6742                                                ifindex, IPPROTO_TCP, netns_id,
6743                                                flags, sdif);
6744 }
6745
6746 static const struct bpf_func_proto bpf_tc_skc_lookup_tcp_proto = {
6747         .func           = bpf_tc_skc_lookup_tcp,
6748         .gpl_only       = false,
6749         .pkt_access     = true,
6750         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6751         .arg1_type      = ARG_PTR_TO_CTX,
6752         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6753         .arg3_type      = ARG_CONST_SIZE,
6754         .arg4_type      = ARG_ANYTHING,
6755         .arg5_type      = ARG_ANYTHING,
6756 };
6757
6758 BPF_CALL_5(bpf_tc_sk_lookup_tcp, struct sk_buff *, skb,
6759            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6760 {
6761         struct net_device *dev = skb->dev;
6762         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6763         struct net *caller_net = dev_net(dev);
6764
6765         return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6766                                               ifindex, IPPROTO_TCP, netns_id,
6767                                               flags, sdif);
6768 }
6769
6770 static const struct bpf_func_proto bpf_tc_sk_lookup_tcp_proto = {
6771         .func           = bpf_tc_sk_lookup_tcp,
6772         .gpl_only       = false,
6773         .pkt_access     = true,
6774         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6775         .arg1_type      = ARG_PTR_TO_CTX,
6776         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6777         .arg3_type      = ARG_CONST_SIZE,
6778         .arg4_type      = ARG_ANYTHING,
6779         .arg5_type      = ARG_ANYTHING,
6780 };
6781
6782 BPF_CALL_5(bpf_tc_sk_lookup_udp, struct sk_buff *, skb,
6783            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6784 {
6785         struct net_device *dev = skb->dev;
6786         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6787         struct net *caller_net = dev_net(dev);
6788
6789         return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6790                                               ifindex, IPPROTO_UDP, netns_id,
6791                                               flags, sdif);
6792 }
6793
6794 static const struct bpf_func_proto bpf_tc_sk_lookup_udp_proto = {
6795         .func           = bpf_tc_sk_lookup_udp,
6796         .gpl_only       = false,
6797         .pkt_access     = true,
6798         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6799         .arg1_type      = ARG_PTR_TO_CTX,
6800         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6801         .arg3_type      = ARG_CONST_SIZE,
6802         .arg4_type      = ARG_ANYTHING,
6803         .arg5_type      = ARG_ANYTHING,
6804 };
6805
6806 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6807 {
6808         if (sk && sk_is_refcounted(sk))
6809                 sock_gen_put(sk);
6810         return 0;
6811 }
6812
6813 static const struct bpf_func_proto bpf_sk_release_proto = {
6814         .func           = bpf_sk_release,
6815         .gpl_only       = false,
6816         .ret_type       = RET_INTEGER,
6817         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON | OBJ_RELEASE,
6818 };
6819
6820 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6821            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6822 {
6823         struct net_device *dev = ctx->rxq->dev;
6824         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6825         struct net *caller_net = dev_net(dev);
6826
6827         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6828                                               ifindex, IPPROTO_UDP, netns_id,
6829                                               flags, sdif);
6830 }
6831
6832 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6833         .func           = bpf_xdp_sk_lookup_udp,
6834         .gpl_only       = false,
6835         .pkt_access     = true,
6836         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6837         .arg1_type      = ARG_PTR_TO_CTX,
6838         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6839         .arg3_type      = ARG_CONST_SIZE,
6840         .arg4_type      = ARG_ANYTHING,
6841         .arg5_type      = ARG_ANYTHING,
6842 };
6843
6844 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6845            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6846 {
6847         struct net_device *dev = ctx->rxq->dev;
6848         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6849         struct net *caller_net = dev_net(dev);
6850
6851         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6852                                                ifindex, IPPROTO_TCP, netns_id,
6853                                                flags, sdif);
6854 }
6855
6856 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6857         .func           = bpf_xdp_skc_lookup_tcp,
6858         .gpl_only       = false,
6859         .pkt_access     = true,
6860         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6861         .arg1_type      = ARG_PTR_TO_CTX,
6862         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6863         .arg3_type      = ARG_CONST_SIZE,
6864         .arg4_type      = ARG_ANYTHING,
6865         .arg5_type      = ARG_ANYTHING,
6866 };
6867
6868 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6869            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6870 {
6871         struct net_device *dev = ctx->rxq->dev;
6872         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6873         struct net *caller_net = dev_net(dev);
6874
6875         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6876                                               ifindex, IPPROTO_TCP, netns_id,
6877                                               flags, sdif);
6878 }
6879
6880 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6881         .func           = bpf_xdp_sk_lookup_tcp,
6882         .gpl_only       = false,
6883         .pkt_access     = true,
6884         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6885         .arg1_type      = ARG_PTR_TO_CTX,
6886         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6887         .arg3_type      = ARG_CONST_SIZE,
6888         .arg4_type      = ARG_ANYTHING,
6889         .arg5_type      = ARG_ANYTHING,
6890 };
6891
6892 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6893            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6894 {
6895         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6896                                                sock_net(ctx->sk), 0,
6897                                                IPPROTO_TCP, netns_id, flags,
6898                                                -1);
6899 }
6900
6901 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6902         .func           = bpf_sock_addr_skc_lookup_tcp,
6903         .gpl_only       = false,
6904         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6905         .arg1_type      = ARG_PTR_TO_CTX,
6906         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6907         .arg3_type      = ARG_CONST_SIZE,
6908         .arg4_type      = ARG_ANYTHING,
6909         .arg5_type      = ARG_ANYTHING,
6910 };
6911
6912 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6913            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6914 {
6915         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6916                                               sock_net(ctx->sk), 0, IPPROTO_TCP,
6917                                               netns_id, flags, -1);
6918 }
6919
6920 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6921         .func           = bpf_sock_addr_sk_lookup_tcp,
6922         .gpl_only       = false,
6923         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6924         .arg1_type      = ARG_PTR_TO_CTX,
6925         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6926         .arg3_type      = ARG_CONST_SIZE,
6927         .arg4_type      = ARG_ANYTHING,
6928         .arg5_type      = ARG_ANYTHING,
6929 };
6930
6931 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6932            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6933 {
6934         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6935                                               sock_net(ctx->sk), 0, IPPROTO_UDP,
6936                                               netns_id, flags, -1);
6937 }
6938
6939 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6940         .func           = bpf_sock_addr_sk_lookup_udp,
6941         .gpl_only       = false,
6942         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6943         .arg1_type      = ARG_PTR_TO_CTX,
6944         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6945         .arg3_type      = ARG_CONST_SIZE,
6946         .arg4_type      = ARG_ANYTHING,
6947         .arg5_type      = ARG_ANYTHING,
6948 };
6949
6950 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6951                                   struct bpf_insn_access_aux *info)
6952 {
6953         if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6954                                           icsk_retransmits))
6955                 return false;
6956
6957         if (off % size != 0)
6958                 return false;
6959
6960         switch (off) {
6961         case offsetof(struct bpf_tcp_sock, bytes_received):
6962         case offsetof(struct bpf_tcp_sock, bytes_acked):
6963                 return size == sizeof(__u64);
6964         default:
6965                 return size == sizeof(__u32);
6966         }
6967 }
6968
6969 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6970                                     const struct bpf_insn *si,
6971                                     struct bpf_insn *insn_buf,
6972                                     struct bpf_prog *prog, u32 *target_size)
6973 {
6974         struct bpf_insn *insn = insn_buf;
6975
6976 #define BPF_TCP_SOCK_GET_COMMON(FIELD)                                  \
6977         do {                                                            \
6978                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) >     \
6979                              sizeof_field(struct bpf_tcp_sock, FIELD)); \
6980                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6981                                       si->dst_reg, si->src_reg,         \
6982                                       offsetof(struct tcp_sock, FIELD)); \
6983         } while (0)
6984
6985 #define BPF_INET_SOCK_GET_COMMON(FIELD)                                 \
6986         do {                                                            \
6987                 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock,  \
6988                                           FIELD) >                      \
6989                              sizeof_field(struct bpf_tcp_sock, FIELD)); \
6990                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                 \
6991                                         struct inet_connection_sock,    \
6992                                         FIELD),                         \
6993                                       si->dst_reg, si->src_reg,         \
6994                                       offsetof(                         \
6995                                         struct inet_connection_sock,    \
6996                                         FIELD));                        \
6997         } while (0)
6998
6999         BTF_TYPE_EMIT(struct bpf_tcp_sock);
7000
7001         switch (si->off) {
7002         case offsetof(struct bpf_tcp_sock, rtt_min):
7003                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
7004                              sizeof(struct minmax));
7005                 BUILD_BUG_ON(sizeof(struct minmax) <
7006                              sizeof(struct minmax_sample));
7007
7008                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7009                                       offsetof(struct tcp_sock, rtt_min) +
7010                                       offsetof(struct minmax_sample, v));
7011                 break;
7012         case offsetof(struct bpf_tcp_sock, snd_cwnd):
7013                 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
7014                 break;
7015         case offsetof(struct bpf_tcp_sock, srtt_us):
7016                 BPF_TCP_SOCK_GET_COMMON(srtt_us);
7017                 break;
7018         case offsetof(struct bpf_tcp_sock, snd_ssthresh):
7019                 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
7020                 break;
7021         case offsetof(struct bpf_tcp_sock, rcv_nxt):
7022                 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
7023                 break;
7024         case offsetof(struct bpf_tcp_sock, snd_nxt):
7025                 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
7026                 break;
7027         case offsetof(struct bpf_tcp_sock, snd_una):
7028                 BPF_TCP_SOCK_GET_COMMON(snd_una);
7029                 break;
7030         case offsetof(struct bpf_tcp_sock, mss_cache):
7031                 BPF_TCP_SOCK_GET_COMMON(mss_cache);
7032                 break;
7033         case offsetof(struct bpf_tcp_sock, ecn_flags):
7034                 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
7035                 break;
7036         case offsetof(struct bpf_tcp_sock, rate_delivered):
7037                 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
7038                 break;
7039         case offsetof(struct bpf_tcp_sock, rate_interval_us):
7040                 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
7041                 break;
7042         case offsetof(struct bpf_tcp_sock, packets_out):
7043                 BPF_TCP_SOCK_GET_COMMON(packets_out);
7044                 break;
7045         case offsetof(struct bpf_tcp_sock, retrans_out):
7046                 BPF_TCP_SOCK_GET_COMMON(retrans_out);
7047                 break;
7048         case offsetof(struct bpf_tcp_sock, total_retrans):
7049                 BPF_TCP_SOCK_GET_COMMON(total_retrans);
7050                 break;
7051         case offsetof(struct bpf_tcp_sock, segs_in):
7052                 BPF_TCP_SOCK_GET_COMMON(segs_in);
7053                 break;
7054         case offsetof(struct bpf_tcp_sock, data_segs_in):
7055                 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
7056                 break;
7057         case offsetof(struct bpf_tcp_sock, segs_out):
7058                 BPF_TCP_SOCK_GET_COMMON(segs_out);
7059                 break;
7060         case offsetof(struct bpf_tcp_sock, data_segs_out):
7061                 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
7062                 break;
7063         case offsetof(struct bpf_tcp_sock, lost_out):
7064                 BPF_TCP_SOCK_GET_COMMON(lost_out);
7065                 break;
7066         case offsetof(struct bpf_tcp_sock, sacked_out):
7067                 BPF_TCP_SOCK_GET_COMMON(sacked_out);
7068                 break;
7069         case offsetof(struct bpf_tcp_sock, bytes_received):
7070                 BPF_TCP_SOCK_GET_COMMON(bytes_received);
7071                 break;
7072         case offsetof(struct bpf_tcp_sock, bytes_acked):
7073                 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
7074                 break;
7075         case offsetof(struct bpf_tcp_sock, dsack_dups):
7076                 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
7077                 break;
7078         case offsetof(struct bpf_tcp_sock, delivered):
7079                 BPF_TCP_SOCK_GET_COMMON(delivered);
7080                 break;
7081         case offsetof(struct bpf_tcp_sock, delivered_ce):
7082                 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
7083                 break;
7084         case offsetof(struct bpf_tcp_sock, icsk_retransmits):
7085                 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
7086                 break;
7087         }
7088
7089         return insn - insn_buf;
7090 }
7091
7092 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
7093 {
7094         if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
7095                 return (unsigned long)sk;
7096
7097         return (unsigned long)NULL;
7098 }
7099
7100 const struct bpf_func_proto bpf_tcp_sock_proto = {
7101         .func           = bpf_tcp_sock,
7102         .gpl_only       = false,
7103         .ret_type       = RET_PTR_TO_TCP_SOCK_OR_NULL,
7104         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
7105 };
7106
7107 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
7108 {
7109         sk = sk_to_full_sk(sk);
7110
7111         if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
7112                 return (unsigned long)sk;
7113
7114         return (unsigned long)NULL;
7115 }
7116
7117 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
7118         .func           = bpf_get_listener_sock,
7119         .gpl_only       = false,
7120         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
7121         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
7122 };
7123
7124 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
7125 {
7126         unsigned int iphdr_len;
7127
7128         switch (skb_protocol(skb, true)) {
7129         case cpu_to_be16(ETH_P_IP):
7130                 iphdr_len = sizeof(struct iphdr);
7131                 break;
7132         case cpu_to_be16(ETH_P_IPV6):
7133                 iphdr_len = sizeof(struct ipv6hdr);
7134                 break;
7135         default:
7136                 return 0;
7137         }
7138
7139         if (skb_headlen(skb) < iphdr_len)
7140                 return 0;
7141
7142         if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
7143                 return 0;
7144
7145         return INET_ECN_set_ce(skb);
7146 }
7147
7148 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7149                                   struct bpf_insn_access_aux *info)
7150 {
7151         if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
7152                 return false;
7153
7154         if (off % size != 0)
7155                 return false;
7156
7157         switch (off) {
7158         default:
7159                 return size == sizeof(__u32);
7160         }
7161 }
7162
7163 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
7164                                     const struct bpf_insn *si,
7165                                     struct bpf_insn *insn_buf,
7166                                     struct bpf_prog *prog, u32 *target_size)
7167 {
7168         struct bpf_insn *insn = insn_buf;
7169
7170 #define BPF_XDP_SOCK_GET(FIELD)                                         \
7171         do {                                                            \
7172                 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) >     \
7173                              sizeof_field(struct bpf_xdp_sock, FIELD)); \
7174                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
7175                                       si->dst_reg, si->src_reg,         \
7176                                       offsetof(struct xdp_sock, FIELD)); \
7177         } while (0)
7178
7179         switch (si->off) {
7180         case offsetof(struct bpf_xdp_sock, queue_id):
7181                 BPF_XDP_SOCK_GET(queue_id);
7182                 break;
7183         }
7184
7185         return insn - insn_buf;
7186 }
7187
7188 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
7189         .func           = bpf_skb_ecn_set_ce,
7190         .gpl_only       = false,
7191         .ret_type       = RET_INTEGER,
7192         .arg1_type      = ARG_PTR_TO_CTX,
7193 };
7194
7195 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7196            struct tcphdr *, th, u32, th_len)
7197 {
7198 #ifdef CONFIG_SYN_COOKIES
7199         u32 cookie;
7200         int ret;
7201
7202         if (unlikely(!sk || th_len < sizeof(*th)))
7203                 return -EINVAL;
7204
7205         /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
7206         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7207                 return -EINVAL;
7208
7209         if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7210                 return -EINVAL;
7211
7212         if (!th->ack || th->rst || th->syn)
7213                 return -ENOENT;
7214
7215         if (unlikely(iph_len < sizeof(struct iphdr)))
7216                 return -EINVAL;
7217
7218         if (tcp_synq_no_recent_overflow(sk))
7219                 return -ENOENT;
7220
7221         cookie = ntohl(th->ack_seq) - 1;
7222
7223         /* Both struct iphdr and struct ipv6hdr have the version field at the
7224          * same offset so we can cast to the shorter header (struct iphdr).
7225          */
7226         switch (((struct iphdr *)iph)->version) {
7227         case 4:
7228                 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7229                         return -EINVAL;
7230
7231                 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
7232                 break;
7233
7234 #if IS_BUILTIN(CONFIG_IPV6)
7235         case 6:
7236                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7237                         return -EINVAL;
7238
7239                 if (sk->sk_family != AF_INET6)
7240                         return -EINVAL;
7241
7242                 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
7243                 break;
7244 #endif /* CONFIG_IPV6 */
7245
7246         default:
7247                 return -EPROTONOSUPPORT;
7248         }
7249
7250         if (ret > 0)
7251                 return 0;
7252
7253         return -ENOENT;
7254 #else
7255         return -ENOTSUPP;
7256 #endif
7257 }
7258
7259 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
7260         .func           = bpf_tcp_check_syncookie,
7261         .gpl_only       = true,
7262         .pkt_access     = true,
7263         .ret_type       = RET_INTEGER,
7264         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7265         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7266         .arg3_type      = ARG_CONST_SIZE,
7267         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7268         .arg5_type      = ARG_CONST_SIZE,
7269 };
7270
7271 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7272            struct tcphdr *, th, u32, th_len)
7273 {
7274 #ifdef CONFIG_SYN_COOKIES
7275         u32 cookie;
7276         u16 mss;
7277
7278         if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
7279                 return -EINVAL;
7280
7281         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7282                 return -EINVAL;
7283
7284         if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7285                 return -ENOENT;
7286
7287         if (!th->syn || th->ack || th->fin || th->rst)
7288                 return -EINVAL;
7289
7290         if (unlikely(iph_len < sizeof(struct iphdr)))
7291                 return -EINVAL;
7292
7293         /* Both struct iphdr and struct ipv6hdr have the version field at the
7294          * same offset so we can cast to the shorter header (struct iphdr).
7295          */
7296         switch (((struct iphdr *)iph)->version) {
7297         case 4:
7298                 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7299                         return -EINVAL;
7300
7301                 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
7302                 break;
7303
7304 #if IS_BUILTIN(CONFIG_IPV6)
7305         case 6:
7306                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7307                         return -EINVAL;
7308
7309                 if (sk->sk_family != AF_INET6)
7310                         return -EINVAL;
7311
7312                 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
7313                 break;
7314 #endif /* CONFIG_IPV6 */
7315
7316         default:
7317                 return -EPROTONOSUPPORT;
7318         }
7319         if (mss == 0)
7320                 return -ENOENT;
7321
7322         return cookie | ((u64)mss << 32);
7323 #else
7324         return -EOPNOTSUPP;
7325 #endif /* CONFIG_SYN_COOKIES */
7326 }
7327
7328 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
7329         .func           = bpf_tcp_gen_syncookie,
7330         .gpl_only       = true, /* __cookie_v*_init_sequence() is GPL */
7331         .pkt_access     = true,
7332         .ret_type       = RET_INTEGER,
7333         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7334         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7335         .arg3_type      = ARG_CONST_SIZE,
7336         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7337         .arg5_type      = ARG_CONST_SIZE,
7338 };
7339
7340 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
7341 {
7342         if (!sk || flags != 0)
7343                 return -EINVAL;
7344         if (!skb_at_tc_ingress(skb))
7345                 return -EOPNOTSUPP;
7346         if (unlikely(dev_net(skb->dev) != sock_net(sk)))
7347                 return -ENETUNREACH;
7348         if (sk_unhashed(sk))
7349                 return -EOPNOTSUPP;
7350         if (sk_is_refcounted(sk) &&
7351             unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
7352                 return -ENOENT;
7353
7354         skb_orphan(skb);
7355         skb->sk = sk;
7356         skb->destructor = sock_pfree;
7357
7358         return 0;
7359 }
7360
7361 static const struct bpf_func_proto bpf_sk_assign_proto = {
7362         .func           = bpf_sk_assign,
7363         .gpl_only       = false,
7364         .ret_type       = RET_INTEGER,
7365         .arg1_type      = ARG_PTR_TO_CTX,
7366         .arg2_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7367         .arg3_type      = ARG_ANYTHING,
7368 };
7369
7370 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
7371                                     u8 search_kind, const u8 *magic,
7372                                     u8 magic_len, bool *eol)
7373 {
7374         u8 kind, kind_len;
7375
7376         *eol = false;
7377
7378         while (op < opend) {
7379                 kind = op[0];
7380
7381                 if (kind == TCPOPT_EOL) {
7382                         *eol = true;
7383                         return ERR_PTR(-ENOMSG);
7384                 } else if (kind == TCPOPT_NOP) {
7385                         op++;
7386                         continue;
7387                 }
7388
7389                 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
7390                         /* Something is wrong in the received header.
7391                          * Follow the TCP stack's tcp_parse_options()
7392                          * and just bail here.
7393                          */
7394                         return ERR_PTR(-EFAULT);
7395
7396                 kind_len = op[1];
7397                 if (search_kind == kind) {
7398                         if (!magic_len)
7399                                 return op;
7400
7401                         if (magic_len > kind_len - 2)
7402                                 return ERR_PTR(-ENOMSG);
7403
7404                         if (!memcmp(&op[2], magic, magic_len))
7405                                 return op;
7406                 }
7407
7408                 op += kind_len;
7409         }
7410
7411         return ERR_PTR(-ENOMSG);
7412 }
7413
7414 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7415            void *, search_res, u32, len, u64, flags)
7416 {
7417         bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
7418         const u8 *op, *opend, *magic, *search = search_res;
7419         u8 search_kind, search_len, copy_len, magic_len;
7420         int ret;
7421
7422         /* 2 byte is the minimal option len except TCPOPT_NOP and
7423          * TCPOPT_EOL which are useless for the bpf prog to learn
7424          * and this helper disallow loading them also.
7425          */
7426         if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
7427                 return -EINVAL;
7428
7429         search_kind = search[0];
7430         search_len = search[1];
7431
7432         if (search_len > len || search_kind == TCPOPT_NOP ||
7433             search_kind == TCPOPT_EOL)
7434                 return -EINVAL;
7435
7436         if (search_kind == TCPOPT_EXP || search_kind == 253) {
7437                 /* 16 or 32 bit magic.  +2 for kind and kind length */
7438                 if (search_len != 4 && search_len != 6)
7439                         return -EINVAL;
7440                 magic = &search[2];
7441                 magic_len = search_len - 2;
7442         } else {
7443                 if (search_len)
7444                         return -EINVAL;
7445                 magic = NULL;
7446                 magic_len = 0;
7447         }
7448
7449         if (load_syn) {
7450                 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
7451                 if (ret < 0)
7452                         return ret;
7453
7454                 opend = op + ret;
7455                 op += sizeof(struct tcphdr);
7456         } else {
7457                 if (!bpf_sock->skb ||
7458                     bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7459                         /* This bpf_sock->op cannot call this helper */
7460                         return -EPERM;
7461
7462                 opend = bpf_sock->skb_data_end;
7463                 op = bpf_sock->skb->data + sizeof(struct tcphdr);
7464         }
7465
7466         op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
7467                                 &eol);
7468         if (IS_ERR(op))
7469                 return PTR_ERR(op);
7470
7471         copy_len = op[1];
7472         ret = copy_len;
7473         if (copy_len > len) {
7474                 ret = -ENOSPC;
7475                 copy_len = len;
7476         }
7477
7478         memcpy(search_res, op, copy_len);
7479         return ret;
7480 }
7481
7482 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
7483         .func           = bpf_sock_ops_load_hdr_opt,
7484         .gpl_only       = false,
7485         .ret_type       = RET_INTEGER,
7486         .arg1_type      = ARG_PTR_TO_CTX,
7487         .arg2_type      = ARG_PTR_TO_MEM,
7488         .arg3_type      = ARG_CONST_SIZE,
7489         .arg4_type      = ARG_ANYTHING,
7490 };
7491
7492 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7493            const void *, from, u32, len, u64, flags)
7494 {
7495         u8 new_kind, new_kind_len, magic_len = 0, *opend;
7496         const u8 *op, *new_op, *magic = NULL;
7497         struct sk_buff *skb;
7498         bool eol;
7499
7500         if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
7501                 return -EPERM;
7502
7503         if (len < 2 || flags)
7504                 return -EINVAL;
7505
7506         new_op = from;
7507         new_kind = new_op[0];
7508         new_kind_len = new_op[1];
7509
7510         if (new_kind_len > len || new_kind == TCPOPT_NOP ||
7511             new_kind == TCPOPT_EOL)
7512                 return -EINVAL;
7513
7514         if (new_kind_len > bpf_sock->remaining_opt_len)
7515                 return -ENOSPC;
7516
7517         /* 253 is another experimental kind */
7518         if (new_kind == TCPOPT_EXP || new_kind == 253)  {
7519                 if (new_kind_len < 4)
7520                         return -EINVAL;
7521                 /* Match for the 2 byte magic also.
7522                  * RFC 6994: the magic could be 2 or 4 bytes.
7523                  * Hence, matching by 2 byte only is on the
7524                  * conservative side but it is the right
7525                  * thing to do for the 'search-for-duplication'
7526                  * purpose.
7527                  */
7528                 magic = &new_op[2];
7529                 magic_len = 2;
7530         }
7531
7532         /* Check for duplication */
7533         skb = bpf_sock->skb;
7534         op = skb->data + sizeof(struct tcphdr);
7535         opend = bpf_sock->skb_data_end;
7536
7537         op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
7538                                 &eol);
7539         if (!IS_ERR(op))
7540                 return -EEXIST;
7541
7542         if (PTR_ERR(op) != -ENOMSG)
7543                 return PTR_ERR(op);
7544
7545         if (eol)
7546                 /* The option has been ended.  Treat it as no more
7547                  * header option can be written.
7548                  */
7549                 return -ENOSPC;
7550
7551         /* No duplication found.  Store the header option. */
7552         memcpy(opend, from, new_kind_len);
7553
7554         bpf_sock->remaining_opt_len -= new_kind_len;
7555         bpf_sock->skb_data_end += new_kind_len;
7556
7557         return 0;
7558 }
7559
7560 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
7561         .func           = bpf_sock_ops_store_hdr_opt,
7562         .gpl_only       = false,
7563         .ret_type       = RET_INTEGER,
7564         .arg1_type      = ARG_PTR_TO_CTX,
7565         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7566         .arg3_type      = ARG_CONST_SIZE,
7567         .arg4_type      = ARG_ANYTHING,
7568 };
7569
7570 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7571            u32, len, u64, flags)
7572 {
7573         if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7574                 return -EPERM;
7575
7576         if (flags || len < 2)
7577                 return -EINVAL;
7578
7579         if (len > bpf_sock->remaining_opt_len)
7580                 return -ENOSPC;
7581
7582         bpf_sock->remaining_opt_len -= len;
7583
7584         return 0;
7585 }
7586
7587 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7588         .func           = bpf_sock_ops_reserve_hdr_opt,
7589         .gpl_only       = false,
7590         .ret_type       = RET_INTEGER,
7591         .arg1_type      = ARG_PTR_TO_CTX,
7592         .arg2_type      = ARG_ANYTHING,
7593         .arg3_type      = ARG_ANYTHING,
7594 };
7595
7596 BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb,
7597            u64, tstamp, u32, tstamp_type)
7598 {
7599         /* skb_clear_delivery_time() is done for inet protocol */
7600         if (skb->protocol != htons(ETH_P_IP) &&
7601             skb->protocol != htons(ETH_P_IPV6))
7602                 return -EOPNOTSUPP;
7603
7604         switch (tstamp_type) {
7605         case BPF_SKB_TSTAMP_DELIVERY_MONO:
7606                 if (!tstamp)
7607                         return -EINVAL;
7608                 skb->tstamp = tstamp;
7609                 skb->mono_delivery_time = 1;
7610                 break;
7611         case BPF_SKB_TSTAMP_UNSPEC:
7612                 if (tstamp)
7613                         return -EINVAL;
7614                 skb->tstamp = 0;
7615                 skb->mono_delivery_time = 0;
7616                 break;
7617         default:
7618                 return -EINVAL;
7619         }
7620
7621         return 0;
7622 }
7623
7624 static const struct bpf_func_proto bpf_skb_set_tstamp_proto = {
7625         .func           = bpf_skb_set_tstamp,
7626         .gpl_only       = false,
7627         .ret_type       = RET_INTEGER,
7628         .arg1_type      = ARG_PTR_TO_CTX,
7629         .arg2_type      = ARG_ANYTHING,
7630         .arg3_type      = ARG_ANYTHING,
7631 };
7632
7633 #ifdef CONFIG_SYN_COOKIES
7634 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv4, struct iphdr *, iph,
7635            struct tcphdr *, th, u32, th_len)
7636 {
7637         u32 cookie;
7638         u16 mss;
7639
7640         if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7641                 return -EINVAL;
7642
7643         mss = tcp_parse_mss_option(th, 0) ?: TCP_MSS_DEFAULT;
7644         cookie = __cookie_v4_init_sequence(iph, th, &mss);
7645
7646         return cookie | ((u64)mss << 32);
7647 }
7648
7649 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv4_proto = {
7650         .func           = bpf_tcp_raw_gen_syncookie_ipv4,
7651         .gpl_only       = true, /* __cookie_v4_init_sequence() is GPL */
7652         .pkt_access     = true,
7653         .ret_type       = RET_INTEGER,
7654         .arg1_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7655         .arg1_size      = sizeof(struct iphdr),
7656         .arg2_type      = ARG_PTR_TO_MEM,
7657         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
7658 };
7659
7660 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv6, struct ipv6hdr *, iph,
7661            struct tcphdr *, th, u32, th_len)
7662 {
7663 #if IS_BUILTIN(CONFIG_IPV6)
7664         const u16 mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
7665                 sizeof(struct ipv6hdr);
7666         u32 cookie;
7667         u16 mss;
7668
7669         if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7670                 return -EINVAL;
7671
7672         mss = tcp_parse_mss_option(th, 0) ?: mss_clamp;
7673         cookie = __cookie_v6_init_sequence(iph, th, &mss);
7674
7675         return cookie | ((u64)mss << 32);
7676 #else
7677         return -EPROTONOSUPPORT;
7678 #endif
7679 }
7680
7681 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv6_proto = {
7682         .func           = bpf_tcp_raw_gen_syncookie_ipv6,
7683         .gpl_only       = true, /* __cookie_v6_init_sequence() is GPL */
7684         .pkt_access     = true,
7685         .ret_type       = RET_INTEGER,
7686         .arg1_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7687         .arg1_size      = sizeof(struct ipv6hdr),
7688         .arg2_type      = ARG_PTR_TO_MEM,
7689         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
7690 };
7691
7692 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv4, struct iphdr *, iph,
7693            struct tcphdr *, th)
7694 {
7695         u32 cookie = ntohl(th->ack_seq) - 1;
7696
7697         if (__cookie_v4_check(iph, th, cookie) > 0)
7698                 return 0;
7699
7700         return -EACCES;
7701 }
7702
7703 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv4_proto = {
7704         .func           = bpf_tcp_raw_check_syncookie_ipv4,
7705         .gpl_only       = true, /* __cookie_v4_check is GPL */
7706         .pkt_access     = true,
7707         .ret_type       = RET_INTEGER,
7708         .arg1_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7709         .arg1_size      = sizeof(struct iphdr),
7710         .arg2_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7711         .arg2_size      = sizeof(struct tcphdr),
7712 };
7713
7714 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv6, struct ipv6hdr *, iph,
7715            struct tcphdr *, th)
7716 {
7717 #if IS_BUILTIN(CONFIG_IPV6)
7718         u32 cookie = ntohl(th->ack_seq) - 1;
7719
7720         if (__cookie_v6_check(iph, th, cookie) > 0)
7721                 return 0;
7722
7723         return -EACCES;
7724 #else
7725         return -EPROTONOSUPPORT;
7726 #endif
7727 }
7728
7729 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv6_proto = {
7730         .func           = bpf_tcp_raw_check_syncookie_ipv6,
7731         .gpl_only       = true, /* __cookie_v6_check is GPL */
7732         .pkt_access     = true,
7733         .ret_type       = RET_INTEGER,
7734         .arg1_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7735         .arg1_size      = sizeof(struct ipv6hdr),
7736         .arg2_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7737         .arg2_size      = sizeof(struct tcphdr),
7738 };
7739 #endif /* CONFIG_SYN_COOKIES */
7740
7741 #endif /* CONFIG_INET */
7742
7743 bool bpf_helper_changes_pkt_data(void *func)
7744 {
7745         if (func == bpf_skb_vlan_push ||
7746             func == bpf_skb_vlan_pop ||
7747             func == bpf_skb_store_bytes ||
7748             func == bpf_skb_change_proto ||
7749             func == bpf_skb_change_head ||
7750             func == sk_skb_change_head ||
7751             func == bpf_skb_change_tail ||
7752             func == sk_skb_change_tail ||
7753             func == bpf_skb_adjust_room ||
7754             func == sk_skb_adjust_room ||
7755             func == bpf_skb_pull_data ||
7756             func == sk_skb_pull_data ||
7757             func == bpf_clone_redirect ||
7758             func == bpf_l3_csum_replace ||
7759             func == bpf_l4_csum_replace ||
7760             func == bpf_xdp_adjust_head ||
7761             func == bpf_xdp_adjust_meta ||
7762             func == bpf_msg_pull_data ||
7763             func == bpf_msg_push_data ||
7764             func == bpf_msg_pop_data ||
7765             func == bpf_xdp_adjust_tail ||
7766 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7767             func == bpf_lwt_seg6_store_bytes ||
7768             func == bpf_lwt_seg6_adjust_srh ||
7769             func == bpf_lwt_seg6_action ||
7770 #endif
7771 #ifdef CONFIG_INET
7772             func == bpf_sock_ops_store_hdr_opt ||
7773 #endif
7774             func == bpf_lwt_in_push_encap ||
7775             func == bpf_lwt_xmit_push_encap)
7776                 return true;
7777
7778         return false;
7779 }
7780
7781 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7782 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7783
7784 static const struct bpf_func_proto *
7785 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7786 {
7787         const struct bpf_func_proto *func_proto;
7788
7789         func_proto = cgroup_common_func_proto(func_id, prog);
7790         if (func_proto)
7791                 return func_proto;
7792
7793         func_proto = cgroup_current_func_proto(func_id, prog);
7794         if (func_proto)
7795                 return func_proto;
7796
7797         switch (func_id) {
7798         case BPF_FUNC_get_socket_cookie:
7799                 return &bpf_get_socket_cookie_sock_proto;
7800         case BPF_FUNC_get_netns_cookie:
7801                 return &bpf_get_netns_cookie_sock_proto;
7802         case BPF_FUNC_perf_event_output:
7803                 return &bpf_event_output_data_proto;
7804         case BPF_FUNC_sk_storage_get:
7805                 return &bpf_sk_storage_get_cg_sock_proto;
7806         case BPF_FUNC_ktime_get_coarse_ns:
7807                 return &bpf_ktime_get_coarse_ns_proto;
7808         default:
7809                 return bpf_base_func_proto(func_id);
7810         }
7811 }
7812
7813 static const struct bpf_func_proto *
7814 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7815 {
7816         const struct bpf_func_proto *func_proto;
7817
7818         func_proto = cgroup_common_func_proto(func_id, prog);
7819         if (func_proto)
7820                 return func_proto;
7821
7822         func_proto = cgroup_current_func_proto(func_id, prog);
7823         if (func_proto)
7824                 return func_proto;
7825
7826         switch (func_id) {
7827         case BPF_FUNC_bind:
7828                 switch (prog->expected_attach_type) {
7829                 case BPF_CGROUP_INET4_CONNECT:
7830                 case BPF_CGROUP_INET6_CONNECT:
7831                         return &bpf_bind_proto;
7832                 default:
7833                         return NULL;
7834                 }
7835         case BPF_FUNC_get_socket_cookie:
7836                 return &bpf_get_socket_cookie_sock_addr_proto;
7837         case BPF_FUNC_get_netns_cookie:
7838                 return &bpf_get_netns_cookie_sock_addr_proto;
7839         case BPF_FUNC_perf_event_output:
7840                 return &bpf_event_output_data_proto;
7841 #ifdef CONFIG_INET
7842         case BPF_FUNC_sk_lookup_tcp:
7843                 return &bpf_sock_addr_sk_lookup_tcp_proto;
7844         case BPF_FUNC_sk_lookup_udp:
7845                 return &bpf_sock_addr_sk_lookup_udp_proto;
7846         case BPF_FUNC_sk_release:
7847                 return &bpf_sk_release_proto;
7848         case BPF_FUNC_skc_lookup_tcp:
7849                 return &bpf_sock_addr_skc_lookup_tcp_proto;
7850 #endif /* CONFIG_INET */
7851         case BPF_FUNC_sk_storage_get:
7852                 return &bpf_sk_storage_get_proto;
7853         case BPF_FUNC_sk_storage_delete:
7854                 return &bpf_sk_storage_delete_proto;
7855         case BPF_FUNC_setsockopt:
7856                 switch (prog->expected_attach_type) {
7857                 case BPF_CGROUP_INET4_BIND:
7858                 case BPF_CGROUP_INET6_BIND:
7859                 case BPF_CGROUP_INET4_CONNECT:
7860                 case BPF_CGROUP_INET6_CONNECT:
7861                 case BPF_CGROUP_UDP4_RECVMSG:
7862                 case BPF_CGROUP_UDP6_RECVMSG:
7863                 case BPF_CGROUP_UDP4_SENDMSG:
7864                 case BPF_CGROUP_UDP6_SENDMSG:
7865                 case BPF_CGROUP_INET4_GETPEERNAME:
7866                 case BPF_CGROUP_INET6_GETPEERNAME:
7867                 case BPF_CGROUP_INET4_GETSOCKNAME:
7868                 case BPF_CGROUP_INET6_GETSOCKNAME:
7869                         return &bpf_sock_addr_setsockopt_proto;
7870                 default:
7871                         return NULL;
7872                 }
7873         case BPF_FUNC_getsockopt:
7874                 switch (prog->expected_attach_type) {
7875                 case BPF_CGROUP_INET4_BIND:
7876                 case BPF_CGROUP_INET6_BIND:
7877                 case BPF_CGROUP_INET4_CONNECT:
7878                 case BPF_CGROUP_INET6_CONNECT:
7879                 case BPF_CGROUP_UDP4_RECVMSG:
7880                 case BPF_CGROUP_UDP6_RECVMSG:
7881                 case BPF_CGROUP_UDP4_SENDMSG:
7882                 case BPF_CGROUP_UDP6_SENDMSG:
7883                 case BPF_CGROUP_INET4_GETPEERNAME:
7884                 case BPF_CGROUP_INET6_GETPEERNAME:
7885                 case BPF_CGROUP_INET4_GETSOCKNAME:
7886                 case BPF_CGROUP_INET6_GETSOCKNAME:
7887                         return &bpf_sock_addr_getsockopt_proto;
7888                 default:
7889                         return NULL;
7890                 }
7891         default:
7892                 return bpf_sk_base_func_proto(func_id);
7893         }
7894 }
7895
7896 static const struct bpf_func_proto *
7897 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7898 {
7899         switch (func_id) {
7900         case BPF_FUNC_skb_load_bytes:
7901                 return &bpf_skb_load_bytes_proto;
7902         case BPF_FUNC_skb_load_bytes_relative:
7903                 return &bpf_skb_load_bytes_relative_proto;
7904         case BPF_FUNC_get_socket_cookie:
7905                 return &bpf_get_socket_cookie_proto;
7906         case BPF_FUNC_get_socket_uid:
7907                 return &bpf_get_socket_uid_proto;
7908         case BPF_FUNC_perf_event_output:
7909                 return &bpf_skb_event_output_proto;
7910         default:
7911                 return bpf_sk_base_func_proto(func_id);
7912         }
7913 }
7914
7915 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7916 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7917
7918 static const struct bpf_func_proto *
7919 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7920 {
7921         const struct bpf_func_proto *func_proto;
7922
7923         func_proto = cgroup_common_func_proto(func_id, prog);
7924         if (func_proto)
7925                 return func_proto;
7926
7927         switch (func_id) {
7928         case BPF_FUNC_sk_fullsock:
7929                 return &bpf_sk_fullsock_proto;
7930         case BPF_FUNC_sk_storage_get:
7931                 return &bpf_sk_storage_get_proto;
7932         case BPF_FUNC_sk_storage_delete:
7933                 return &bpf_sk_storage_delete_proto;
7934         case BPF_FUNC_perf_event_output:
7935                 return &bpf_skb_event_output_proto;
7936 #ifdef CONFIG_SOCK_CGROUP_DATA
7937         case BPF_FUNC_skb_cgroup_id:
7938                 return &bpf_skb_cgroup_id_proto;
7939         case BPF_FUNC_skb_ancestor_cgroup_id:
7940                 return &bpf_skb_ancestor_cgroup_id_proto;
7941         case BPF_FUNC_sk_cgroup_id:
7942                 return &bpf_sk_cgroup_id_proto;
7943         case BPF_FUNC_sk_ancestor_cgroup_id:
7944                 return &bpf_sk_ancestor_cgroup_id_proto;
7945 #endif
7946 #ifdef CONFIG_INET
7947         case BPF_FUNC_sk_lookup_tcp:
7948                 return &bpf_sk_lookup_tcp_proto;
7949         case BPF_FUNC_sk_lookup_udp:
7950                 return &bpf_sk_lookup_udp_proto;
7951         case BPF_FUNC_sk_release:
7952                 return &bpf_sk_release_proto;
7953         case BPF_FUNC_skc_lookup_tcp:
7954                 return &bpf_skc_lookup_tcp_proto;
7955         case BPF_FUNC_tcp_sock:
7956                 return &bpf_tcp_sock_proto;
7957         case BPF_FUNC_get_listener_sock:
7958                 return &bpf_get_listener_sock_proto;
7959         case BPF_FUNC_skb_ecn_set_ce:
7960                 return &bpf_skb_ecn_set_ce_proto;
7961 #endif
7962         default:
7963                 return sk_filter_func_proto(func_id, prog);
7964         }
7965 }
7966
7967 static const struct bpf_func_proto *
7968 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7969 {
7970         switch (func_id) {
7971         case BPF_FUNC_skb_store_bytes:
7972                 return &bpf_skb_store_bytes_proto;
7973         case BPF_FUNC_skb_load_bytes:
7974                 return &bpf_skb_load_bytes_proto;
7975         case BPF_FUNC_skb_load_bytes_relative:
7976                 return &bpf_skb_load_bytes_relative_proto;
7977         case BPF_FUNC_skb_pull_data:
7978                 return &bpf_skb_pull_data_proto;
7979         case BPF_FUNC_csum_diff:
7980                 return &bpf_csum_diff_proto;
7981         case BPF_FUNC_csum_update:
7982                 return &bpf_csum_update_proto;
7983         case BPF_FUNC_csum_level:
7984                 return &bpf_csum_level_proto;
7985         case BPF_FUNC_l3_csum_replace:
7986                 return &bpf_l3_csum_replace_proto;
7987         case BPF_FUNC_l4_csum_replace:
7988                 return &bpf_l4_csum_replace_proto;
7989         case BPF_FUNC_clone_redirect:
7990                 return &bpf_clone_redirect_proto;
7991         case BPF_FUNC_get_cgroup_classid:
7992                 return &bpf_get_cgroup_classid_proto;
7993         case BPF_FUNC_skb_vlan_push:
7994                 return &bpf_skb_vlan_push_proto;
7995         case BPF_FUNC_skb_vlan_pop:
7996                 return &bpf_skb_vlan_pop_proto;
7997         case BPF_FUNC_skb_change_proto:
7998                 return &bpf_skb_change_proto_proto;
7999         case BPF_FUNC_skb_change_type:
8000                 return &bpf_skb_change_type_proto;
8001         case BPF_FUNC_skb_adjust_room:
8002                 return &bpf_skb_adjust_room_proto;
8003         case BPF_FUNC_skb_change_tail:
8004                 return &bpf_skb_change_tail_proto;
8005         case BPF_FUNC_skb_change_head:
8006                 return &bpf_skb_change_head_proto;
8007         case BPF_FUNC_skb_get_tunnel_key:
8008                 return &bpf_skb_get_tunnel_key_proto;
8009         case BPF_FUNC_skb_set_tunnel_key:
8010                 return bpf_get_skb_set_tunnel_proto(func_id);
8011         case BPF_FUNC_skb_get_tunnel_opt:
8012                 return &bpf_skb_get_tunnel_opt_proto;
8013         case BPF_FUNC_skb_set_tunnel_opt:
8014                 return bpf_get_skb_set_tunnel_proto(func_id);
8015         case BPF_FUNC_redirect:
8016                 return &bpf_redirect_proto;
8017         case BPF_FUNC_redirect_neigh:
8018                 return &bpf_redirect_neigh_proto;
8019         case BPF_FUNC_redirect_peer:
8020                 return &bpf_redirect_peer_proto;
8021         case BPF_FUNC_get_route_realm:
8022                 return &bpf_get_route_realm_proto;
8023         case BPF_FUNC_get_hash_recalc:
8024                 return &bpf_get_hash_recalc_proto;
8025         case BPF_FUNC_set_hash_invalid:
8026                 return &bpf_set_hash_invalid_proto;
8027         case BPF_FUNC_set_hash:
8028                 return &bpf_set_hash_proto;
8029         case BPF_FUNC_perf_event_output:
8030                 return &bpf_skb_event_output_proto;
8031         case BPF_FUNC_get_smp_processor_id:
8032                 return &bpf_get_smp_processor_id_proto;
8033         case BPF_FUNC_skb_under_cgroup:
8034                 return &bpf_skb_under_cgroup_proto;
8035         case BPF_FUNC_get_socket_cookie:
8036                 return &bpf_get_socket_cookie_proto;
8037         case BPF_FUNC_get_socket_uid:
8038                 return &bpf_get_socket_uid_proto;
8039         case BPF_FUNC_fib_lookup:
8040                 return &bpf_skb_fib_lookup_proto;
8041         case BPF_FUNC_check_mtu:
8042                 return &bpf_skb_check_mtu_proto;
8043         case BPF_FUNC_sk_fullsock:
8044                 return &bpf_sk_fullsock_proto;
8045         case BPF_FUNC_sk_storage_get:
8046                 return &bpf_sk_storage_get_proto;
8047         case BPF_FUNC_sk_storage_delete:
8048                 return &bpf_sk_storage_delete_proto;
8049 #ifdef CONFIG_XFRM
8050         case BPF_FUNC_skb_get_xfrm_state:
8051                 return &bpf_skb_get_xfrm_state_proto;
8052 #endif
8053 #ifdef CONFIG_CGROUP_NET_CLASSID
8054         case BPF_FUNC_skb_cgroup_classid:
8055                 return &bpf_skb_cgroup_classid_proto;
8056 #endif
8057 #ifdef CONFIG_SOCK_CGROUP_DATA
8058         case BPF_FUNC_skb_cgroup_id:
8059                 return &bpf_skb_cgroup_id_proto;
8060         case BPF_FUNC_skb_ancestor_cgroup_id:
8061                 return &bpf_skb_ancestor_cgroup_id_proto;
8062 #endif
8063 #ifdef CONFIG_INET
8064         case BPF_FUNC_sk_lookup_tcp:
8065                 return &bpf_tc_sk_lookup_tcp_proto;
8066         case BPF_FUNC_sk_lookup_udp:
8067                 return &bpf_tc_sk_lookup_udp_proto;
8068         case BPF_FUNC_sk_release:
8069                 return &bpf_sk_release_proto;
8070         case BPF_FUNC_tcp_sock:
8071                 return &bpf_tcp_sock_proto;
8072         case BPF_FUNC_get_listener_sock:
8073                 return &bpf_get_listener_sock_proto;
8074         case BPF_FUNC_skc_lookup_tcp:
8075                 return &bpf_tc_skc_lookup_tcp_proto;
8076         case BPF_FUNC_tcp_check_syncookie:
8077                 return &bpf_tcp_check_syncookie_proto;
8078         case BPF_FUNC_skb_ecn_set_ce:
8079                 return &bpf_skb_ecn_set_ce_proto;
8080         case BPF_FUNC_tcp_gen_syncookie:
8081                 return &bpf_tcp_gen_syncookie_proto;
8082         case BPF_FUNC_sk_assign:
8083                 return &bpf_sk_assign_proto;
8084         case BPF_FUNC_skb_set_tstamp:
8085                 return &bpf_skb_set_tstamp_proto;
8086 #ifdef CONFIG_SYN_COOKIES
8087         case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8088                 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8089         case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8090                 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8091         case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8092                 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8093         case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8094                 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8095 #endif
8096 #endif
8097         default:
8098                 return bpf_sk_base_func_proto(func_id);
8099         }
8100 }
8101
8102 static const struct bpf_func_proto *
8103 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8104 {
8105         switch (func_id) {
8106         case BPF_FUNC_perf_event_output:
8107                 return &bpf_xdp_event_output_proto;
8108         case BPF_FUNC_get_smp_processor_id:
8109                 return &bpf_get_smp_processor_id_proto;
8110         case BPF_FUNC_csum_diff:
8111                 return &bpf_csum_diff_proto;
8112         case BPF_FUNC_xdp_adjust_head:
8113                 return &bpf_xdp_adjust_head_proto;
8114         case BPF_FUNC_xdp_adjust_meta:
8115                 return &bpf_xdp_adjust_meta_proto;
8116         case BPF_FUNC_redirect:
8117                 return &bpf_xdp_redirect_proto;
8118         case BPF_FUNC_redirect_map:
8119                 return &bpf_xdp_redirect_map_proto;
8120         case BPF_FUNC_xdp_adjust_tail:
8121                 return &bpf_xdp_adjust_tail_proto;
8122         case BPF_FUNC_xdp_get_buff_len:
8123                 return &bpf_xdp_get_buff_len_proto;
8124         case BPF_FUNC_xdp_load_bytes:
8125                 return &bpf_xdp_load_bytes_proto;
8126         case BPF_FUNC_xdp_store_bytes:
8127                 return &bpf_xdp_store_bytes_proto;
8128         case BPF_FUNC_fib_lookup:
8129                 return &bpf_xdp_fib_lookup_proto;
8130         case BPF_FUNC_check_mtu:
8131                 return &bpf_xdp_check_mtu_proto;
8132 #ifdef CONFIG_INET
8133         case BPF_FUNC_sk_lookup_udp:
8134                 return &bpf_xdp_sk_lookup_udp_proto;
8135         case BPF_FUNC_sk_lookup_tcp:
8136                 return &bpf_xdp_sk_lookup_tcp_proto;
8137         case BPF_FUNC_sk_release:
8138                 return &bpf_sk_release_proto;
8139         case BPF_FUNC_skc_lookup_tcp:
8140                 return &bpf_xdp_skc_lookup_tcp_proto;
8141         case BPF_FUNC_tcp_check_syncookie:
8142                 return &bpf_tcp_check_syncookie_proto;
8143         case BPF_FUNC_tcp_gen_syncookie:
8144                 return &bpf_tcp_gen_syncookie_proto;
8145 #ifdef CONFIG_SYN_COOKIES
8146         case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8147                 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8148         case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8149                 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8150         case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8151                 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8152         case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8153                 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8154 #endif
8155 #endif
8156         default:
8157                 return bpf_sk_base_func_proto(func_id);
8158         }
8159
8160 #if IS_MODULE(CONFIG_NF_CONNTRACK) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES)
8161         /* The nf_conn___init type is used in the NF_CONNTRACK kfuncs. The
8162          * kfuncs are defined in two different modules, and we want to be able
8163          * to use them interchangably with the same BTF type ID. Because modules
8164          * can't de-duplicate BTF IDs between each other, we need the type to be
8165          * referenced in the vmlinux BTF or the verifier will get confused about
8166          * the different types. So we add this dummy type reference which will
8167          * be included in vmlinux BTF, allowing both modules to refer to the
8168          * same type ID.
8169          */
8170         BTF_TYPE_EMIT(struct nf_conn___init);
8171 #endif
8172 }
8173
8174 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
8175 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
8176
8177 static const struct bpf_func_proto *
8178 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8179 {
8180         const struct bpf_func_proto *func_proto;
8181
8182         func_proto = cgroup_common_func_proto(func_id, prog);
8183         if (func_proto)
8184                 return func_proto;
8185
8186         switch (func_id) {
8187         case BPF_FUNC_setsockopt:
8188                 return &bpf_sock_ops_setsockopt_proto;
8189         case BPF_FUNC_getsockopt:
8190                 return &bpf_sock_ops_getsockopt_proto;
8191         case BPF_FUNC_sock_ops_cb_flags_set:
8192                 return &bpf_sock_ops_cb_flags_set_proto;
8193         case BPF_FUNC_sock_map_update:
8194                 return &bpf_sock_map_update_proto;
8195         case BPF_FUNC_sock_hash_update:
8196                 return &bpf_sock_hash_update_proto;
8197         case BPF_FUNC_get_socket_cookie:
8198                 return &bpf_get_socket_cookie_sock_ops_proto;
8199         case BPF_FUNC_perf_event_output:
8200                 return &bpf_event_output_data_proto;
8201         case BPF_FUNC_sk_storage_get:
8202                 return &bpf_sk_storage_get_proto;
8203         case BPF_FUNC_sk_storage_delete:
8204                 return &bpf_sk_storage_delete_proto;
8205         case BPF_FUNC_get_netns_cookie:
8206                 return &bpf_get_netns_cookie_sock_ops_proto;
8207 #ifdef CONFIG_INET
8208         case BPF_FUNC_load_hdr_opt:
8209                 return &bpf_sock_ops_load_hdr_opt_proto;
8210         case BPF_FUNC_store_hdr_opt:
8211                 return &bpf_sock_ops_store_hdr_opt_proto;
8212         case BPF_FUNC_reserve_hdr_opt:
8213                 return &bpf_sock_ops_reserve_hdr_opt_proto;
8214         case BPF_FUNC_tcp_sock:
8215                 return &bpf_tcp_sock_proto;
8216 #endif /* CONFIG_INET */
8217         default:
8218                 return bpf_sk_base_func_proto(func_id);
8219         }
8220 }
8221
8222 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
8223 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
8224
8225 static const struct bpf_func_proto *
8226 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8227 {
8228         switch (func_id) {
8229         case BPF_FUNC_msg_redirect_map:
8230                 return &bpf_msg_redirect_map_proto;
8231         case BPF_FUNC_msg_redirect_hash:
8232                 return &bpf_msg_redirect_hash_proto;
8233         case BPF_FUNC_msg_apply_bytes:
8234                 return &bpf_msg_apply_bytes_proto;
8235         case BPF_FUNC_msg_cork_bytes:
8236                 return &bpf_msg_cork_bytes_proto;
8237         case BPF_FUNC_msg_pull_data:
8238                 return &bpf_msg_pull_data_proto;
8239         case BPF_FUNC_msg_push_data:
8240                 return &bpf_msg_push_data_proto;
8241         case BPF_FUNC_msg_pop_data:
8242                 return &bpf_msg_pop_data_proto;
8243         case BPF_FUNC_perf_event_output:
8244                 return &bpf_event_output_data_proto;
8245         case BPF_FUNC_get_current_uid_gid:
8246                 return &bpf_get_current_uid_gid_proto;
8247         case BPF_FUNC_get_current_pid_tgid:
8248                 return &bpf_get_current_pid_tgid_proto;
8249         case BPF_FUNC_sk_storage_get:
8250                 return &bpf_sk_storage_get_proto;
8251         case BPF_FUNC_sk_storage_delete:
8252                 return &bpf_sk_storage_delete_proto;
8253         case BPF_FUNC_get_netns_cookie:
8254                 return &bpf_get_netns_cookie_sk_msg_proto;
8255 #ifdef CONFIG_CGROUP_NET_CLASSID
8256         case BPF_FUNC_get_cgroup_classid:
8257                 return &bpf_get_cgroup_classid_curr_proto;
8258 #endif
8259         default:
8260                 return bpf_sk_base_func_proto(func_id);
8261         }
8262 }
8263
8264 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
8265 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
8266
8267 static const struct bpf_func_proto *
8268 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8269 {
8270         switch (func_id) {
8271         case BPF_FUNC_skb_store_bytes:
8272                 return &bpf_skb_store_bytes_proto;
8273         case BPF_FUNC_skb_load_bytes:
8274                 return &bpf_skb_load_bytes_proto;
8275         case BPF_FUNC_skb_pull_data:
8276                 return &sk_skb_pull_data_proto;
8277         case BPF_FUNC_skb_change_tail:
8278                 return &sk_skb_change_tail_proto;
8279         case BPF_FUNC_skb_change_head:
8280                 return &sk_skb_change_head_proto;
8281         case BPF_FUNC_skb_adjust_room:
8282                 return &sk_skb_adjust_room_proto;
8283         case BPF_FUNC_get_socket_cookie:
8284                 return &bpf_get_socket_cookie_proto;
8285         case BPF_FUNC_get_socket_uid:
8286                 return &bpf_get_socket_uid_proto;
8287         case BPF_FUNC_sk_redirect_map:
8288                 return &bpf_sk_redirect_map_proto;
8289         case BPF_FUNC_sk_redirect_hash:
8290                 return &bpf_sk_redirect_hash_proto;
8291         case BPF_FUNC_perf_event_output:
8292                 return &bpf_skb_event_output_proto;
8293 #ifdef CONFIG_INET
8294         case BPF_FUNC_sk_lookup_tcp:
8295                 return &bpf_sk_lookup_tcp_proto;
8296         case BPF_FUNC_sk_lookup_udp:
8297                 return &bpf_sk_lookup_udp_proto;
8298         case BPF_FUNC_sk_release:
8299                 return &bpf_sk_release_proto;
8300         case BPF_FUNC_skc_lookup_tcp:
8301                 return &bpf_skc_lookup_tcp_proto;
8302 #endif
8303         default:
8304                 return bpf_sk_base_func_proto(func_id);
8305         }
8306 }
8307
8308 static const struct bpf_func_proto *
8309 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8310 {
8311         switch (func_id) {
8312         case BPF_FUNC_skb_load_bytes:
8313                 return &bpf_flow_dissector_load_bytes_proto;
8314         default:
8315                 return bpf_sk_base_func_proto(func_id);
8316         }
8317 }
8318
8319 static const struct bpf_func_proto *
8320 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8321 {
8322         switch (func_id) {
8323         case BPF_FUNC_skb_load_bytes:
8324                 return &bpf_skb_load_bytes_proto;
8325         case BPF_FUNC_skb_pull_data:
8326                 return &bpf_skb_pull_data_proto;
8327         case BPF_FUNC_csum_diff:
8328                 return &bpf_csum_diff_proto;
8329         case BPF_FUNC_get_cgroup_classid:
8330                 return &bpf_get_cgroup_classid_proto;
8331         case BPF_FUNC_get_route_realm:
8332                 return &bpf_get_route_realm_proto;
8333         case BPF_FUNC_get_hash_recalc:
8334                 return &bpf_get_hash_recalc_proto;
8335         case BPF_FUNC_perf_event_output:
8336                 return &bpf_skb_event_output_proto;
8337         case BPF_FUNC_get_smp_processor_id:
8338                 return &bpf_get_smp_processor_id_proto;
8339         case BPF_FUNC_skb_under_cgroup:
8340                 return &bpf_skb_under_cgroup_proto;
8341         default:
8342                 return bpf_sk_base_func_proto(func_id);
8343         }
8344 }
8345
8346 static const struct bpf_func_proto *
8347 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8348 {
8349         switch (func_id) {
8350         case BPF_FUNC_lwt_push_encap:
8351                 return &bpf_lwt_in_push_encap_proto;
8352         default:
8353                 return lwt_out_func_proto(func_id, prog);
8354         }
8355 }
8356
8357 static const struct bpf_func_proto *
8358 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8359 {
8360         switch (func_id) {
8361         case BPF_FUNC_skb_get_tunnel_key:
8362                 return &bpf_skb_get_tunnel_key_proto;
8363         case BPF_FUNC_skb_set_tunnel_key:
8364                 return bpf_get_skb_set_tunnel_proto(func_id);
8365         case BPF_FUNC_skb_get_tunnel_opt:
8366                 return &bpf_skb_get_tunnel_opt_proto;
8367         case BPF_FUNC_skb_set_tunnel_opt:
8368                 return bpf_get_skb_set_tunnel_proto(func_id);
8369         case BPF_FUNC_redirect:
8370                 return &bpf_redirect_proto;
8371         case BPF_FUNC_clone_redirect:
8372                 return &bpf_clone_redirect_proto;
8373         case BPF_FUNC_skb_change_tail:
8374                 return &bpf_skb_change_tail_proto;
8375         case BPF_FUNC_skb_change_head:
8376                 return &bpf_skb_change_head_proto;
8377         case BPF_FUNC_skb_store_bytes:
8378                 return &bpf_skb_store_bytes_proto;
8379         case BPF_FUNC_csum_update:
8380                 return &bpf_csum_update_proto;
8381         case BPF_FUNC_csum_level:
8382                 return &bpf_csum_level_proto;
8383         case BPF_FUNC_l3_csum_replace:
8384                 return &bpf_l3_csum_replace_proto;
8385         case BPF_FUNC_l4_csum_replace:
8386                 return &bpf_l4_csum_replace_proto;
8387         case BPF_FUNC_set_hash_invalid:
8388                 return &bpf_set_hash_invalid_proto;
8389         case BPF_FUNC_lwt_push_encap:
8390                 return &bpf_lwt_xmit_push_encap_proto;
8391         default:
8392                 return lwt_out_func_proto(func_id, prog);
8393         }
8394 }
8395
8396 static const struct bpf_func_proto *
8397 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8398 {
8399         switch (func_id) {
8400 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
8401         case BPF_FUNC_lwt_seg6_store_bytes:
8402                 return &bpf_lwt_seg6_store_bytes_proto;
8403         case BPF_FUNC_lwt_seg6_action:
8404                 return &bpf_lwt_seg6_action_proto;
8405         case BPF_FUNC_lwt_seg6_adjust_srh:
8406                 return &bpf_lwt_seg6_adjust_srh_proto;
8407 #endif
8408         default:
8409                 return lwt_out_func_proto(func_id, prog);
8410         }
8411 }
8412
8413 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
8414                                     const struct bpf_prog *prog,
8415                                     struct bpf_insn_access_aux *info)
8416 {
8417         const int size_default = sizeof(__u32);
8418
8419         if (off < 0 || off >= sizeof(struct __sk_buff))
8420                 return false;
8421
8422         /* The verifier guarantees that size > 0. */
8423         if (off % size != 0)
8424                 return false;
8425
8426         switch (off) {
8427         case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8428                 if (off + size > offsetofend(struct __sk_buff, cb[4]))
8429                         return false;
8430                 break;
8431         case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
8432         case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
8433         case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
8434         case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
8435         case bpf_ctx_range(struct __sk_buff, data):
8436         case bpf_ctx_range(struct __sk_buff, data_meta):
8437         case bpf_ctx_range(struct __sk_buff, data_end):
8438                 if (size != size_default)
8439                         return false;
8440                 break;
8441         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8442                 return false;
8443         case bpf_ctx_range(struct __sk_buff, hwtstamp):
8444                 if (type == BPF_WRITE || size != sizeof(__u64))
8445                         return false;
8446                 break;
8447         case bpf_ctx_range(struct __sk_buff, tstamp):
8448                 if (size != sizeof(__u64))
8449                         return false;
8450                 break;
8451         case offsetof(struct __sk_buff, sk):
8452                 if (type == BPF_WRITE || size != sizeof(__u64))
8453                         return false;
8454                 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
8455                 break;
8456         case offsetof(struct __sk_buff, tstamp_type):
8457                 return false;
8458         case offsetofend(struct __sk_buff, tstamp_type) ... offsetof(struct __sk_buff, hwtstamp) - 1:
8459                 /* Explicitly prohibit access to padding in __sk_buff. */
8460                 return false;
8461         default:
8462                 /* Only narrow read access allowed for now. */
8463                 if (type == BPF_WRITE) {
8464                         if (size != size_default)
8465                                 return false;
8466                 } else {
8467                         bpf_ctx_record_field_size(info, size_default);
8468                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8469                                 return false;
8470                 }
8471         }
8472
8473         return true;
8474 }
8475
8476 static bool sk_filter_is_valid_access(int off, int size,
8477                                       enum bpf_access_type type,
8478                                       const struct bpf_prog *prog,
8479                                       struct bpf_insn_access_aux *info)
8480 {
8481         switch (off) {
8482         case bpf_ctx_range(struct __sk_buff, tc_classid):
8483         case bpf_ctx_range(struct __sk_buff, data):
8484         case bpf_ctx_range(struct __sk_buff, data_meta):
8485         case bpf_ctx_range(struct __sk_buff, data_end):
8486         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8487         case bpf_ctx_range(struct __sk_buff, tstamp):
8488         case bpf_ctx_range(struct __sk_buff, wire_len):
8489         case bpf_ctx_range(struct __sk_buff, hwtstamp):
8490                 return false;
8491         }
8492
8493         if (type == BPF_WRITE) {
8494                 switch (off) {
8495                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8496                         break;
8497                 default:
8498                         return false;
8499                 }
8500         }
8501
8502         return bpf_skb_is_valid_access(off, size, type, prog, info);
8503 }
8504
8505 static bool cg_skb_is_valid_access(int off, int size,
8506                                    enum bpf_access_type type,
8507                                    const struct bpf_prog *prog,
8508                                    struct bpf_insn_access_aux *info)
8509 {
8510         switch (off) {
8511         case bpf_ctx_range(struct __sk_buff, tc_classid):
8512         case bpf_ctx_range(struct __sk_buff, data_meta):
8513         case bpf_ctx_range(struct __sk_buff, wire_len):
8514                 return false;
8515         case bpf_ctx_range(struct __sk_buff, data):
8516         case bpf_ctx_range(struct __sk_buff, data_end):
8517                 if (!bpf_capable())
8518                         return false;
8519                 break;
8520         }
8521
8522         if (type == BPF_WRITE) {
8523                 switch (off) {
8524                 case bpf_ctx_range(struct __sk_buff, mark):
8525                 case bpf_ctx_range(struct __sk_buff, priority):
8526                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8527                         break;
8528                 case bpf_ctx_range(struct __sk_buff, tstamp):
8529                         if (!bpf_capable())
8530                                 return false;
8531                         break;
8532                 default:
8533                         return false;
8534                 }
8535         }
8536
8537         switch (off) {
8538         case bpf_ctx_range(struct __sk_buff, data):
8539                 info->reg_type = PTR_TO_PACKET;
8540                 break;
8541         case bpf_ctx_range(struct __sk_buff, data_end):
8542                 info->reg_type = PTR_TO_PACKET_END;
8543                 break;
8544         }
8545
8546         return bpf_skb_is_valid_access(off, size, type, prog, info);
8547 }
8548
8549 static bool lwt_is_valid_access(int off, int size,
8550                                 enum bpf_access_type type,
8551                                 const struct bpf_prog *prog,
8552                                 struct bpf_insn_access_aux *info)
8553 {
8554         switch (off) {
8555         case bpf_ctx_range(struct __sk_buff, tc_classid):
8556         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8557         case bpf_ctx_range(struct __sk_buff, data_meta):
8558         case bpf_ctx_range(struct __sk_buff, tstamp):
8559         case bpf_ctx_range(struct __sk_buff, wire_len):
8560         case bpf_ctx_range(struct __sk_buff, hwtstamp):
8561                 return false;
8562         }
8563
8564         if (type == BPF_WRITE) {
8565                 switch (off) {
8566                 case bpf_ctx_range(struct __sk_buff, mark):
8567                 case bpf_ctx_range(struct __sk_buff, priority):
8568                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8569                         break;
8570                 default:
8571                         return false;
8572                 }
8573         }
8574
8575         switch (off) {
8576         case bpf_ctx_range(struct __sk_buff, data):
8577                 info->reg_type = PTR_TO_PACKET;
8578                 break;
8579         case bpf_ctx_range(struct __sk_buff, data_end):
8580                 info->reg_type = PTR_TO_PACKET_END;
8581                 break;
8582         }
8583
8584         return bpf_skb_is_valid_access(off, size, type, prog, info);
8585 }
8586
8587 /* Attach type specific accesses */
8588 static bool __sock_filter_check_attach_type(int off,
8589                                             enum bpf_access_type access_type,
8590                                             enum bpf_attach_type attach_type)
8591 {
8592         switch (off) {
8593         case offsetof(struct bpf_sock, bound_dev_if):
8594         case offsetof(struct bpf_sock, mark):
8595         case offsetof(struct bpf_sock, priority):
8596                 switch (attach_type) {
8597                 case BPF_CGROUP_INET_SOCK_CREATE:
8598                 case BPF_CGROUP_INET_SOCK_RELEASE:
8599                         goto full_access;
8600                 default:
8601                         return false;
8602                 }
8603         case bpf_ctx_range(struct bpf_sock, src_ip4):
8604                 switch (attach_type) {
8605                 case BPF_CGROUP_INET4_POST_BIND:
8606                         goto read_only;
8607                 default:
8608                         return false;
8609                 }
8610         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8611                 switch (attach_type) {
8612                 case BPF_CGROUP_INET6_POST_BIND:
8613                         goto read_only;
8614                 default:
8615                         return false;
8616                 }
8617         case bpf_ctx_range(struct bpf_sock, src_port):
8618                 switch (attach_type) {
8619                 case BPF_CGROUP_INET4_POST_BIND:
8620                 case BPF_CGROUP_INET6_POST_BIND:
8621                         goto read_only;
8622                 default:
8623                         return false;
8624                 }
8625         }
8626 read_only:
8627         return access_type == BPF_READ;
8628 full_access:
8629         return true;
8630 }
8631
8632 bool bpf_sock_common_is_valid_access(int off, int size,
8633                                      enum bpf_access_type type,
8634                                      struct bpf_insn_access_aux *info)
8635 {
8636         switch (off) {
8637         case bpf_ctx_range_till(struct bpf_sock, type, priority):
8638                 return false;
8639         default:
8640                 return bpf_sock_is_valid_access(off, size, type, info);
8641         }
8642 }
8643
8644 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
8645                               struct bpf_insn_access_aux *info)
8646 {
8647         const int size_default = sizeof(__u32);
8648         int field_size;
8649
8650         if (off < 0 || off >= sizeof(struct bpf_sock))
8651                 return false;
8652         if (off % size != 0)
8653                 return false;
8654
8655         switch (off) {
8656         case offsetof(struct bpf_sock, state):
8657         case offsetof(struct bpf_sock, family):
8658         case offsetof(struct bpf_sock, type):
8659         case offsetof(struct bpf_sock, protocol):
8660         case offsetof(struct bpf_sock, src_port):
8661         case offsetof(struct bpf_sock, rx_queue_mapping):
8662         case bpf_ctx_range(struct bpf_sock, src_ip4):
8663         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8664         case bpf_ctx_range(struct bpf_sock, dst_ip4):
8665         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8666                 bpf_ctx_record_field_size(info, size_default);
8667                 return bpf_ctx_narrow_access_ok(off, size, size_default);
8668         case bpf_ctx_range(struct bpf_sock, dst_port):
8669                 field_size = size == size_default ?
8670                         size_default : sizeof_field(struct bpf_sock, dst_port);
8671                 bpf_ctx_record_field_size(info, field_size);
8672                 return bpf_ctx_narrow_access_ok(off, size, field_size);
8673         case offsetofend(struct bpf_sock, dst_port) ...
8674              offsetof(struct bpf_sock, dst_ip4) - 1:
8675                 return false;
8676         }
8677
8678         return size == size_default;
8679 }
8680
8681 static bool sock_filter_is_valid_access(int off, int size,
8682                                         enum bpf_access_type type,
8683                                         const struct bpf_prog *prog,
8684                                         struct bpf_insn_access_aux *info)
8685 {
8686         if (!bpf_sock_is_valid_access(off, size, type, info))
8687                 return false;
8688         return __sock_filter_check_attach_type(off, type,
8689                                                prog->expected_attach_type);
8690 }
8691
8692 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
8693                              const struct bpf_prog *prog)
8694 {
8695         /* Neither direct read nor direct write requires any preliminary
8696          * action.
8697          */
8698         return 0;
8699 }
8700
8701 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
8702                                 const struct bpf_prog *prog, int drop_verdict)
8703 {
8704         struct bpf_insn *insn = insn_buf;
8705
8706         if (!direct_write)
8707                 return 0;
8708
8709         /* if (!skb->cloned)
8710          *       goto start;
8711          *
8712          * (Fast-path, otherwise approximation that we might be
8713          *  a clone, do the rest in helper.)
8714          */
8715         *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET);
8716         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
8717         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
8718
8719         /* ret = bpf_skb_pull_data(skb, 0); */
8720         *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
8721         *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
8722         *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
8723                                BPF_FUNC_skb_pull_data);
8724         /* if (!ret)
8725          *      goto restore;
8726          * return TC_ACT_SHOT;
8727          */
8728         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
8729         *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
8730         *insn++ = BPF_EXIT_INSN();
8731
8732         /* restore: */
8733         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
8734         /* start: */
8735         *insn++ = prog->insnsi[0];
8736
8737         return insn - insn_buf;
8738 }
8739
8740 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
8741                           struct bpf_insn *insn_buf)
8742 {
8743         bool indirect = BPF_MODE(orig->code) == BPF_IND;
8744         struct bpf_insn *insn = insn_buf;
8745
8746         if (!indirect) {
8747                 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
8748         } else {
8749                 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
8750                 if (orig->imm)
8751                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
8752         }
8753         /* We're guaranteed here that CTX is in R6. */
8754         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
8755
8756         switch (BPF_SIZE(orig->code)) {
8757         case BPF_B:
8758                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
8759                 break;
8760         case BPF_H:
8761                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
8762                 break;
8763         case BPF_W:
8764                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
8765                 break;
8766         }
8767
8768         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
8769         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
8770         *insn++ = BPF_EXIT_INSN();
8771
8772         return insn - insn_buf;
8773 }
8774
8775 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8776                                const struct bpf_prog *prog)
8777 {
8778         return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8779 }
8780
8781 static bool tc_cls_act_is_valid_access(int off, int size,
8782                                        enum bpf_access_type type,
8783                                        const struct bpf_prog *prog,
8784                                        struct bpf_insn_access_aux *info)
8785 {
8786         if (type == BPF_WRITE) {
8787                 switch (off) {
8788                 case bpf_ctx_range(struct __sk_buff, mark):
8789                 case bpf_ctx_range(struct __sk_buff, tc_index):
8790                 case bpf_ctx_range(struct __sk_buff, priority):
8791                 case bpf_ctx_range(struct __sk_buff, tc_classid):
8792                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8793                 case bpf_ctx_range(struct __sk_buff, tstamp):
8794                 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8795                         break;
8796                 default:
8797                         return false;
8798                 }
8799         }
8800
8801         switch (off) {
8802         case bpf_ctx_range(struct __sk_buff, data):
8803                 info->reg_type = PTR_TO_PACKET;
8804                 break;
8805         case bpf_ctx_range(struct __sk_buff, data_meta):
8806                 info->reg_type = PTR_TO_PACKET_META;
8807                 break;
8808         case bpf_ctx_range(struct __sk_buff, data_end):
8809                 info->reg_type = PTR_TO_PACKET_END;
8810                 break;
8811         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8812                 return false;
8813         case offsetof(struct __sk_buff, tstamp_type):
8814                 /* The convert_ctx_access() on reading and writing
8815                  * __sk_buff->tstamp depends on whether the bpf prog
8816                  * has used __sk_buff->tstamp_type or not.
8817                  * Thus, we need to set prog->tstamp_type_access
8818                  * earlier during is_valid_access() here.
8819                  */
8820                 ((struct bpf_prog *)prog)->tstamp_type_access = 1;
8821                 return size == sizeof(__u8);
8822         }
8823
8824         return bpf_skb_is_valid_access(off, size, type, prog, info);
8825 }
8826
8827 DEFINE_MUTEX(nf_conn_btf_access_lock);
8828 EXPORT_SYMBOL_GPL(nf_conn_btf_access_lock);
8829
8830 int (*nfct_btf_struct_access)(struct bpf_verifier_log *log,
8831                               const struct bpf_reg_state *reg,
8832                               int off, int size);
8833 EXPORT_SYMBOL_GPL(nfct_btf_struct_access);
8834
8835 static int tc_cls_act_btf_struct_access(struct bpf_verifier_log *log,
8836                                         const struct bpf_reg_state *reg,
8837                                         int off, int size)
8838 {
8839         int ret = -EACCES;
8840
8841         mutex_lock(&nf_conn_btf_access_lock);
8842         if (nfct_btf_struct_access)
8843                 ret = nfct_btf_struct_access(log, reg, off, size);
8844         mutex_unlock(&nf_conn_btf_access_lock);
8845
8846         return ret;
8847 }
8848
8849 static bool __is_valid_xdp_access(int off, int size)
8850 {
8851         if (off < 0 || off >= sizeof(struct xdp_md))
8852                 return false;
8853         if (off % size != 0)
8854                 return false;
8855         if (size != sizeof(__u32))
8856                 return false;
8857
8858         return true;
8859 }
8860
8861 static bool xdp_is_valid_access(int off, int size,
8862                                 enum bpf_access_type type,
8863                                 const struct bpf_prog *prog,
8864                                 struct bpf_insn_access_aux *info)
8865 {
8866         if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8867                 switch (off) {
8868                 case offsetof(struct xdp_md, egress_ifindex):
8869                         return false;
8870                 }
8871         }
8872
8873         if (type == BPF_WRITE) {
8874                 if (bpf_prog_is_offloaded(prog->aux)) {
8875                         switch (off) {
8876                         case offsetof(struct xdp_md, rx_queue_index):
8877                                 return __is_valid_xdp_access(off, size);
8878                         }
8879                 }
8880                 return false;
8881         }
8882
8883         switch (off) {
8884         case offsetof(struct xdp_md, data):
8885                 info->reg_type = PTR_TO_PACKET;
8886                 break;
8887         case offsetof(struct xdp_md, data_meta):
8888                 info->reg_type = PTR_TO_PACKET_META;
8889                 break;
8890         case offsetof(struct xdp_md, data_end):
8891                 info->reg_type = PTR_TO_PACKET_END;
8892                 break;
8893         }
8894
8895         return __is_valid_xdp_access(off, size);
8896 }
8897
8898 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act)
8899 {
8900         const u32 act_max = XDP_REDIRECT;
8901
8902         pr_warn_once("%s XDP return value %u on prog %s (id %d) dev %s, expect packet loss!\n",
8903                      act > act_max ? "Illegal" : "Driver unsupported",
8904                      act, prog->aux->name, prog->aux->id, dev ? dev->name : "N/A");
8905 }
8906 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
8907
8908 static int xdp_btf_struct_access(struct bpf_verifier_log *log,
8909                                  const struct bpf_reg_state *reg,
8910                                  int off, int size)
8911 {
8912         int ret = -EACCES;
8913
8914         mutex_lock(&nf_conn_btf_access_lock);
8915         if (nfct_btf_struct_access)
8916                 ret = nfct_btf_struct_access(log, reg, off, size);
8917         mutex_unlock(&nf_conn_btf_access_lock);
8918
8919         return ret;
8920 }
8921
8922 static bool sock_addr_is_valid_access(int off, int size,
8923                                       enum bpf_access_type type,
8924                                       const struct bpf_prog *prog,
8925                                       struct bpf_insn_access_aux *info)
8926 {
8927         const int size_default = sizeof(__u32);
8928
8929         if (off < 0 || off >= sizeof(struct bpf_sock_addr))
8930                 return false;
8931         if (off % size != 0)
8932                 return false;
8933
8934         /* Disallow access to IPv6 fields from IPv4 contex and vise
8935          * versa.
8936          */
8937         switch (off) {
8938         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8939                 switch (prog->expected_attach_type) {
8940                 case BPF_CGROUP_INET4_BIND:
8941                 case BPF_CGROUP_INET4_CONNECT:
8942                 case BPF_CGROUP_INET4_GETPEERNAME:
8943                 case BPF_CGROUP_INET4_GETSOCKNAME:
8944                 case BPF_CGROUP_UDP4_SENDMSG:
8945                 case BPF_CGROUP_UDP4_RECVMSG:
8946                         break;
8947                 default:
8948                         return false;
8949                 }
8950                 break;
8951         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8952                 switch (prog->expected_attach_type) {
8953                 case BPF_CGROUP_INET6_BIND:
8954                 case BPF_CGROUP_INET6_CONNECT:
8955                 case BPF_CGROUP_INET6_GETPEERNAME:
8956                 case BPF_CGROUP_INET6_GETSOCKNAME:
8957                 case BPF_CGROUP_UDP6_SENDMSG:
8958                 case BPF_CGROUP_UDP6_RECVMSG:
8959                         break;
8960                 default:
8961                         return false;
8962                 }
8963                 break;
8964         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8965                 switch (prog->expected_attach_type) {
8966                 case BPF_CGROUP_UDP4_SENDMSG:
8967                         break;
8968                 default:
8969                         return false;
8970                 }
8971                 break;
8972         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8973                                 msg_src_ip6[3]):
8974                 switch (prog->expected_attach_type) {
8975                 case BPF_CGROUP_UDP6_SENDMSG:
8976                         break;
8977                 default:
8978                         return false;
8979                 }
8980                 break;
8981         }
8982
8983         switch (off) {
8984         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8985         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8986         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8987         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8988                                 msg_src_ip6[3]):
8989         case bpf_ctx_range(struct bpf_sock_addr, user_port):
8990                 if (type == BPF_READ) {
8991                         bpf_ctx_record_field_size(info, size_default);
8992
8993                         if (bpf_ctx_wide_access_ok(off, size,
8994                                                    struct bpf_sock_addr,
8995                                                    user_ip6))
8996                                 return true;
8997
8998                         if (bpf_ctx_wide_access_ok(off, size,
8999                                                    struct bpf_sock_addr,
9000                                                    msg_src_ip6))
9001                                 return true;
9002
9003                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
9004                                 return false;
9005                 } else {
9006                         if (bpf_ctx_wide_access_ok(off, size,
9007                                                    struct bpf_sock_addr,
9008                                                    user_ip6))
9009                                 return true;
9010
9011                         if (bpf_ctx_wide_access_ok(off, size,
9012                                                    struct bpf_sock_addr,
9013                                                    msg_src_ip6))
9014                                 return true;
9015
9016                         if (size != size_default)
9017                                 return false;
9018                 }
9019                 break;
9020         case offsetof(struct bpf_sock_addr, sk):
9021                 if (type != BPF_READ)
9022                         return false;
9023                 if (size != sizeof(__u64))
9024                         return false;
9025                 info->reg_type = PTR_TO_SOCKET;
9026                 break;
9027         default:
9028                 if (type == BPF_READ) {
9029                         if (size != size_default)
9030                                 return false;
9031                 } else {
9032                         return false;
9033                 }
9034         }
9035
9036         return true;
9037 }
9038
9039 static bool sock_ops_is_valid_access(int off, int size,
9040                                      enum bpf_access_type type,
9041                                      const struct bpf_prog *prog,
9042                                      struct bpf_insn_access_aux *info)
9043 {
9044         const int size_default = sizeof(__u32);
9045
9046         if (off < 0 || off >= sizeof(struct bpf_sock_ops))
9047                 return false;
9048
9049         /* The verifier guarantees that size > 0. */
9050         if (off % size != 0)
9051                 return false;
9052
9053         if (type == BPF_WRITE) {
9054                 switch (off) {
9055                 case offsetof(struct bpf_sock_ops, reply):
9056                 case offsetof(struct bpf_sock_ops, sk_txhash):
9057                         if (size != size_default)
9058                                 return false;
9059                         break;
9060                 default:
9061                         return false;
9062                 }
9063         } else {
9064                 switch (off) {
9065                 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
9066                                         bytes_acked):
9067                         if (size != sizeof(__u64))
9068                                 return false;
9069                         break;
9070                 case offsetof(struct bpf_sock_ops, sk):
9071                         if (size != sizeof(__u64))
9072                                 return false;
9073                         info->reg_type = PTR_TO_SOCKET_OR_NULL;
9074                         break;
9075                 case offsetof(struct bpf_sock_ops, skb_data):
9076                         if (size != sizeof(__u64))
9077                                 return false;
9078                         info->reg_type = PTR_TO_PACKET;
9079                         break;
9080                 case offsetof(struct bpf_sock_ops, skb_data_end):
9081                         if (size != sizeof(__u64))
9082                                 return false;
9083                         info->reg_type = PTR_TO_PACKET_END;
9084                         break;
9085                 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9086                         bpf_ctx_record_field_size(info, size_default);
9087                         return bpf_ctx_narrow_access_ok(off, size,
9088                                                         size_default);
9089                 case offsetof(struct bpf_sock_ops, skb_hwtstamp):
9090                         if (size != sizeof(__u64))
9091                                 return false;
9092                         break;
9093                 default:
9094                         if (size != size_default)
9095                                 return false;
9096                         break;
9097                 }
9098         }
9099
9100         return true;
9101 }
9102
9103 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
9104                            const struct bpf_prog *prog)
9105 {
9106         return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
9107 }
9108
9109 static bool sk_skb_is_valid_access(int off, int size,
9110                                    enum bpf_access_type type,
9111                                    const struct bpf_prog *prog,
9112                                    struct bpf_insn_access_aux *info)
9113 {
9114         switch (off) {
9115         case bpf_ctx_range(struct __sk_buff, tc_classid):
9116         case bpf_ctx_range(struct __sk_buff, data_meta):
9117         case bpf_ctx_range(struct __sk_buff, tstamp):
9118         case bpf_ctx_range(struct __sk_buff, wire_len):
9119         case bpf_ctx_range(struct __sk_buff, hwtstamp):
9120                 return false;
9121         }
9122
9123         if (type == BPF_WRITE) {
9124                 switch (off) {
9125                 case bpf_ctx_range(struct __sk_buff, tc_index):
9126                 case bpf_ctx_range(struct __sk_buff, priority):
9127                         break;
9128                 default:
9129                         return false;
9130                 }
9131         }
9132
9133         switch (off) {
9134         case bpf_ctx_range(struct __sk_buff, mark):
9135                 return false;
9136         case bpf_ctx_range(struct __sk_buff, data):
9137                 info->reg_type = PTR_TO_PACKET;
9138                 break;
9139         case bpf_ctx_range(struct __sk_buff, data_end):
9140                 info->reg_type = PTR_TO_PACKET_END;
9141                 break;
9142         }
9143
9144         return bpf_skb_is_valid_access(off, size, type, prog, info);
9145 }
9146
9147 static bool sk_msg_is_valid_access(int off, int size,
9148                                    enum bpf_access_type type,
9149                                    const struct bpf_prog *prog,
9150                                    struct bpf_insn_access_aux *info)
9151 {
9152         if (type == BPF_WRITE)
9153                 return false;
9154
9155         if (off % size != 0)
9156                 return false;
9157
9158         switch (off) {
9159         case offsetof(struct sk_msg_md, data):
9160                 info->reg_type = PTR_TO_PACKET;
9161                 if (size != sizeof(__u64))
9162                         return false;
9163                 break;
9164         case offsetof(struct sk_msg_md, data_end):
9165                 info->reg_type = PTR_TO_PACKET_END;
9166                 if (size != sizeof(__u64))
9167                         return false;
9168                 break;
9169         case offsetof(struct sk_msg_md, sk):
9170                 if (size != sizeof(__u64))
9171                         return false;
9172                 info->reg_type = PTR_TO_SOCKET;
9173                 break;
9174         case bpf_ctx_range(struct sk_msg_md, family):
9175         case bpf_ctx_range(struct sk_msg_md, remote_ip4):
9176         case bpf_ctx_range(struct sk_msg_md, local_ip4):
9177         case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
9178         case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
9179         case bpf_ctx_range(struct sk_msg_md, remote_port):
9180         case bpf_ctx_range(struct sk_msg_md, local_port):
9181         case bpf_ctx_range(struct sk_msg_md, size):
9182                 if (size != sizeof(__u32))
9183                         return false;
9184                 break;
9185         default:
9186                 return false;
9187         }
9188         return true;
9189 }
9190
9191 static bool flow_dissector_is_valid_access(int off, int size,
9192                                            enum bpf_access_type type,
9193                                            const struct bpf_prog *prog,
9194                                            struct bpf_insn_access_aux *info)
9195 {
9196         const int size_default = sizeof(__u32);
9197
9198         if (off < 0 || off >= sizeof(struct __sk_buff))
9199                 return false;
9200
9201         if (type == BPF_WRITE)
9202                 return false;
9203
9204         switch (off) {
9205         case bpf_ctx_range(struct __sk_buff, data):
9206                 if (size != size_default)
9207                         return false;
9208                 info->reg_type = PTR_TO_PACKET;
9209                 return true;
9210         case bpf_ctx_range(struct __sk_buff, data_end):
9211                 if (size != size_default)
9212                         return false;
9213                 info->reg_type = PTR_TO_PACKET_END;
9214                 return true;
9215         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
9216                 if (size != sizeof(__u64))
9217                         return false;
9218                 info->reg_type = PTR_TO_FLOW_KEYS;
9219                 return true;
9220         default:
9221                 return false;
9222         }
9223 }
9224
9225 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
9226                                              const struct bpf_insn *si,
9227                                              struct bpf_insn *insn_buf,
9228                                              struct bpf_prog *prog,
9229                                              u32 *target_size)
9230
9231 {
9232         struct bpf_insn *insn = insn_buf;
9233
9234         switch (si->off) {
9235         case offsetof(struct __sk_buff, data):
9236                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
9237                                       si->dst_reg, si->src_reg,
9238                                       offsetof(struct bpf_flow_dissector, data));
9239                 break;
9240
9241         case offsetof(struct __sk_buff, data_end):
9242                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
9243                                       si->dst_reg, si->src_reg,
9244                                       offsetof(struct bpf_flow_dissector, data_end));
9245                 break;
9246
9247         case offsetof(struct __sk_buff, flow_keys):
9248                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
9249                                       si->dst_reg, si->src_reg,
9250                                       offsetof(struct bpf_flow_dissector, flow_keys));
9251                 break;
9252         }
9253
9254         return insn - insn_buf;
9255 }
9256
9257 static struct bpf_insn *bpf_convert_tstamp_type_read(const struct bpf_insn *si,
9258                                                      struct bpf_insn *insn)
9259 {
9260         __u8 value_reg = si->dst_reg;
9261         __u8 skb_reg = si->src_reg;
9262         /* AX is needed because src_reg and dst_reg could be the same */
9263         __u8 tmp_reg = BPF_REG_AX;
9264
9265         *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg,
9266                               SKB_BF_MONO_TC_OFFSET);
9267         *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg,
9268                                 SKB_MONO_DELIVERY_TIME_MASK, 2);
9269         *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_UNSPEC);
9270         *insn++ = BPF_JMP_A(1);
9271         *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_DELIVERY_MONO);
9272
9273         return insn;
9274 }
9275
9276 static struct bpf_insn *bpf_convert_shinfo_access(__u8 dst_reg, __u8 skb_reg,
9277                                                   struct bpf_insn *insn)
9278 {
9279         /* si->dst_reg = skb_shinfo(SKB); */
9280 #ifdef NET_SKBUFF_DATA_USES_OFFSET
9281         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9282                               BPF_REG_AX, skb_reg,
9283                               offsetof(struct sk_buff, end));
9284         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
9285                               dst_reg, skb_reg,
9286                               offsetof(struct sk_buff, head));
9287         *insn++ = BPF_ALU64_REG(BPF_ADD, dst_reg, BPF_REG_AX);
9288 #else
9289         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9290                               dst_reg, skb_reg,
9291                               offsetof(struct sk_buff, end));
9292 #endif
9293
9294         return insn;
9295 }
9296
9297 static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog,
9298                                                 const struct bpf_insn *si,
9299                                                 struct bpf_insn *insn)
9300 {
9301         __u8 value_reg = si->dst_reg;
9302         __u8 skb_reg = si->src_reg;
9303
9304 #ifdef CONFIG_NET_XGRESS
9305         /* If the tstamp_type is read,
9306          * the bpf prog is aware the tstamp could have delivery time.
9307          * Thus, read skb->tstamp as is if tstamp_type_access is true.
9308          */
9309         if (!prog->tstamp_type_access) {
9310                 /* AX is needed because src_reg and dst_reg could be the same */
9311                 __u8 tmp_reg = BPF_REG_AX;
9312
9313                 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, SKB_BF_MONO_TC_OFFSET);
9314                 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg,
9315                                         TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK);
9316                 *insn++ = BPF_JMP32_IMM(BPF_JNE, tmp_reg,
9317                                         TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK, 2);
9318                 /* skb->tc_at_ingress && skb->mono_delivery_time,
9319                  * read 0 as the (rcv) timestamp.
9320                  */
9321                 *insn++ = BPF_MOV64_IMM(value_reg, 0);
9322                 *insn++ = BPF_JMP_A(1);
9323         }
9324 #endif
9325
9326         *insn++ = BPF_LDX_MEM(BPF_DW, value_reg, skb_reg,
9327                               offsetof(struct sk_buff, tstamp));
9328         return insn;
9329 }
9330
9331 static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog,
9332                                                  const struct bpf_insn *si,
9333                                                  struct bpf_insn *insn)
9334 {
9335         __u8 value_reg = si->src_reg;
9336         __u8 skb_reg = si->dst_reg;
9337
9338 #ifdef CONFIG_NET_XGRESS
9339         /* If the tstamp_type is read,
9340          * the bpf prog is aware the tstamp could have delivery time.
9341          * Thus, write skb->tstamp as is if tstamp_type_access is true.
9342          * Otherwise, writing at ingress will have to clear the
9343          * mono_delivery_time bit also.
9344          */
9345         if (!prog->tstamp_type_access) {
9346                 __u8 tmp_reg = BPF_REG_AX;
9347
9348                 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, SKB_BF_MONO_TC_OFFSET);
9349                 /* Writing __sk_buff->tstamp as ingress, goto <clear> */
9350                 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1);
9351                 /* goto <store> */
9352                 *insn++ = BPF_JMP_A(2);
9353                 /* <clear>: mono_delivery_time */
9354                 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_MONO_DELIVERY_TIME_MASK);
9355                 *insn++ = BPF_STX_MEM(BPF_B, skb_reg, tmp_reg, SKB_BF_MONO_TC_OFFSET);
9356         }
9357 #endif
9358
9359         /* <store>: skb->tstamp = tstamp */
9360         *insn++ = BPF_RAW_INSN(BPF_CLASS(si->code) | BPF_DW | BPF_MEM,
9361                                skb_reg, value_reg, offsetof(struct sk_buff, tstamp), si->imm);
9362         return insn;
9363 }
9364
9365 #define BPF_EMIT_STORE(size, si, off)                                   \
9366         BPF_RAW_INSN(BPF_CLASS((si)->code) | (size) | BPF_MEM,          \
9367                      (si)->dst_reg, (si)->src_reg, (off), (si)->imm)
9368
9369 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
9370                                   const struct bpf_insn *si,
9371                                   struct bpf_insn *insn_buf,
9372                                   struct bpf_prog *prog, u32 *target_size)
9373 {
9374         struct bpf_insn *insn = insn_buf;
9375         int off;
9376
9377         switch (si->off) {
9378         case offsetof(struct __sk_buff, len):
9379                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9380                                       bpf_target_off(struct sk_buff, len, 4,
9381                                                      target_size));
9382                 break;
9383
9384         case offsetof(struct __sk_buff, protocol):
9385                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9386                                       bpf_target_off(struct sk_buff, protocol, 2,
9387                                                      target_size));
9388                 break;
9389
9390         case offsetof(struct __sk_buff, vlan_proto):
9391                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9392                                       bpf_target_off(struct sk_buff, vlan_proto, 2,
9393                                                      target_size));
9394                 break;
9395
9396         case offsetof(struct __sk_buff, priority):
9397                 if (type == BPF_WRITE)
9398                         *insn++ = BPF_EMIT_STORE(BPF_W, si,
9399                                                  bpf_target_off(struct sk_buff, priority, 4,
9400                                                                 target_size));
9401                 else
9402                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9403                                               bpf_target_off(struct sk_buff, priority, 4,
9404                                                              target_size));
9405                 break;
9406
9407         case offsetof(struct __sk_buff, ingress_ifindex):
9408                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9409                                       bpf_target_off(struct sk_buff, skb_iif, 4,
9410                                                      target_size));
9411                 break;
9412
9413         case offsetof(struct __sk_buff, ifindex):
9414                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9415                                       si->dst_reg, si->src_reg,
9416                                       offsetof(struct sk_buff, dev));
9417                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9418                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9419                                       bpf_target_off(struct net_device, ifindex, 4,
9420                                                      target_size));
9421                 break;
9422
9423         case offsetof(struct __sk_buff, hash):
9424                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9425                                       bpf_target_off(struct sk_buff, hash, 4,
9426                                                      target_size));
9427                 break;
9428
9429         case offsetof(struct __sk_buff, mark):
9430                 if (type == BPF_WRITE)
9431                         *insn++ = BPF_EMIT_STORE(BPF_W, si,
9432                                                  bpf_target_off(struct sk_buff, mark, 4,
9433                                                                 target_size));
9434                 else
9435                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9436                                               bpf_target_off(struct sk_buff, mark, 4,
9437                                                              target_size));
9438                 break;
9439
9440         case offsetof(struct __sk_buff, pkt_type):
9441                 *target_size = 1;
9442                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
9443                                       PKT_TYPE_OFFSET);
9444                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
9445 #ifdef __BIG_ENDIAN_BITFIELD
9446                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
9447 #endif
9448                 break;
9449
9450         case offsetof(struct __sk_buff, queue_mapping):
9451                 if (type == BPF_WRITE) {
9452                         u32 off = bpf_target_off(struct sk_buff, queue_mapping, 2, target_size);
9453
9454                         if (BPF_CLASS(si->code) == BPF_ST && si->imm >= NO_QUEUE_MAPPING) {
9455                                 *insn++ = BPF_JMP_A(0); /* noop */
9456                                 break;
9457                         }
9458
9459                         if (BPF_CLASS(si->code) == BPF_STX)
9460                                 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
9461                         *insn++ = BPF_EMIT_STORE(BPF_H, si, off);
9462                 } else {
9463                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9464                                               bpf_target_off(struct sk_buff,
9465                                                              queue_mapping,
9466                                                              2, target_size));
9467                 }
9468                 break;
9469
9470         case offsetof(struct __sk_buff, vlan_present):
9471                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9472                                       bpf_target_off(struct sk_buff,
9473                                                      vlan_all, 4, target_size));
9474                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9475                 *insn++ = BPF_ALU32_IMM(BPF_MOV, si->dst_reg, 1);
9476                 break;
9477
9478         case offsetof(struct __sk_buff, vlan_tci):
9479                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9480                                       bpf_target_off(struct sk_buff, vlan_tci, 2,
9481                                                      target_size));
9482                 break;
9483
9484         case offsetof(struct __sk_buff, cb[0]) ...
9485              offsetofend(struct __sk_buff, cb[4]) - 1:
9486                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
9487                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
9488                               offsetof(struct qdisc_skb_cb, data)) %
9489                              sizeof(__u64));
9490
9491                 prog->cb_access = 1;
9492                 off  = si->off;
9493                 off -= offsetof(struct __sk_buff, cb[0]);
9494                 off += offsetof(struct sk_buff, cb);
9495                 off += offsetof(struct qdisc_skb_cb, data);
9496                 if (type == BPF_WRITE)
9497                         *insn++ = BPF_EMIT_STORE(BPF_SIZE(si->code), si, off);
9498                 else
9499                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
9500                                               si->src_reg, off);
9501                 break;
9502
9503         case offsetof(struct __sk_buff, tc_classid):
9504                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
9505
9506                 off  = si->off;
9507                 off -= offsetof(struct __sk_buff, tc_classid);
9508                 off += offsetof(struct sk_buff, cb);
9509                 off += offsetof(struct qdisc_skb_cb, tc_classid);
9510                 *target_size = 2;
9511                 if (type == BPF_WRITE)
9512                         *insn++ = BPF_EMIT_STORE(BPF_H, si, off);
9513                 else
9514                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
9515                                               si->src_reg, off);
9516                 break;
9517
9518         case offsetof(struct __sk_buff, data):
9519                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9520                                       si->dst_reg, si->src_reg,
9521                                       offsetof(struct sk_buff, data));
9522                 break;
9523
9524         case offsetof(struct __sk_buff, data_meta):
9525                 off  = si->off;
9526                 off -= offsetof(struct __sk_buff, data_meta);
9527                 off += offsetof(struct sk_buff, cb);
9528                 off += offsetof(struct bpf_skb_data_end, data_meta);
9529                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9530                                       si->src_reg, off);
9531                 break;
9532
9533         case offsetof(struct __sk_buff, data_end):
9534                 off  = si->off;
9535                 off -= offsetof(struct __sk_buff, data_end);
9536                 off += offsetof(struct sk_buff, cb);
9537                 off += offsetof(struct bpf_skb_data_end, data_end);
9538                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9539                                       si->src_reg, off);
9540                 break;
9541
9542         case offsetof(struct __sk_buff, tc_index):
9543 #ifdef CONFIG_NET_SCHED
9544                 if (type == BPF_WRITE)
9545                         *insn++ = BPF_EMIT_STORE(BPF_H, si,
9546                                                  bpf_target_off(struct sk_buff, tc_index, 2,
9547                                                                 target_size));
9548                 else
9549                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9550                                               bpf_target_off(struct sk_buff, tc_index, 2,
9551                                                              target_size));
9552 #else
9553                 *target_size = 2;
9554                 if (type == BPF_WRITE)
9555                         *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
9556                 else
9557                         *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9558 #endif
9559                 break;
9560
9561         case offsetof(struct __sk_buff, napi_id):
9562 #if defined(CONFIG_NET_RX_BUSY_POLL)
9563                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9564                                       bpf_target_off(struct sk_buff, napi_id, 4,
9565                                                      target_size));
9566                 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
9567                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9568 #else
9569                 *target_size = 4;
9570                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9571 #endif
9572                 break;
9573         case offsetof(struct __sk_buff, family):
9574                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9575
9576                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9577                                       si->dst_reg, si->src_reg,
9578                                       offsetof(struct sk_buff, sk));
9579                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9580                                       bpf_target_off(struct sock_common,
9581                                                      skc_family,
9582                                                      2, target_size));
9583                 break;
9584         case offsetof(struct __sk_buff, remote_ip4):
9585                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9586
9587                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9588                                       si->dst_reg, si->src_reg,
9589                                       offsetof(struct sk_buff, sk));
9590                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9591                                       bpf_target_off(struct sock_common,
9592                                                      skc_daddr,
9593                                                      4, target_size));
9594                 break;
9595         case offsetof(struct __sk_buff, local_ip4):
9596                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9597                                           skc_rcv_saddr) != 4);
9598
9599                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9600                                       si->dst_reg, si->src_reg,
9601                                       offsetof(struct sk_buff, sk));
9602                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9603                                       bpf_target_off(struct sock_common,
9604                                                      skc_rcv_saddr,
9605                                                      4, target_size));
9606                 break;
9607         case offsetof(struct __sk_buff, remote_ip6[0]) ...
9608              offsetof(struct __sk_buff, remote_ip6[3]):
9609 #if IS_ENABLED(CONFIG_IPV6)
9610                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9611                                           skc_v6_daddr.s6_addr32[0]) != 4);
9612
9613                 off = si->off;
9614                 off -= offsetof(struct __sk_buff, remote_ip6[0]);
9615
9616                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9617                                       si->dst_reg, si->src_reg,
9618                                       offsetof(struct sk_buff, sk));
9619                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9620                                       offsetof(struct sock_common,
9621                                                skc_v6_daddr.s6_addr32[0]) +
9622                                       off);
9623 #else
9624                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9625 #endif
9626                 break;
9627         case offsetof(struct __sk_buff, local_ip6[0]) ...
9628              offsetof(struct __sk_buff, local_ip6[3]):
9629 #if IS_ENABLED(CONFIG_IPV6)
9630                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9631                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9632
9633                 off = si->off;
9634                 off -= offsetof(struct __sk_buff, local_ip6[0]);
9635
9636                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9637                                       si->dst_reg, si->src_reg,
9638                                       offsetof(struct sk_buff, sk));
9639                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9640                                       offsetof(struct sock_common,
9641                                                skc_v6_rcv_saddr.s6_addr32[0]) +
9642                                       off);
9643 #else
9644                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9645 #endif
9646                 break;
9647
9648         case offsetof(struct __sk_buff, remote_port):
9649                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9650
9651                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9652                                       si->dst_reg, si->src_reg,
9653                                       offsetof(struct sk_buff, sk));
9654                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9655                                       bpf_target_off(struct sock_common,
9656                                                      skc_dport,
9657                                                      2, target_size));
9658 #ifndef __BIG_ENDIAN_BITFIELD
9659                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9660 #endif
9661                 break;
9662
9663         case offsetof(struct __sk_buff, local_port):
9664                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9665
9666                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9667                                       si->dst_reg, si->src_reg,
9668                                       offsetof(struct sk_buff, sk));
9669                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9670                                       bpf_target_off(struct sock_common,
9671                                                      skc_num, 2, target_size));
9672                 break;
9673
9674         case offsetof(struct __sk_buff, tstamp):
9675                 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
9676
9677                 if (type == BPF_WRITE)
9678                         insn = bpf_convert_tstamp_write(prog, si, insn);
9679                 else
9680                         insn = bpf_convert_tstamp_read(prog, si, insn);
9681                 break;
9682
9683         case offsetof(struct __sk_buff, tstamp_type):
9684                 insn = bpf_convert_tstamp_type_read(si, insn);
9685                 break;
9686
9687         case offsetof(struct __sk_buff, gso_segs):
9688                 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9689                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
9690                                       si->dst_reg, si->dst_reg,
9691                                       bpf_target_off(struct skb_shared_info,
9692                                                      gso_segs, 2,
9693                                                      target_size));
9694                 break;
9695         case offsetof(struct __sk_buff, gso_size):
9696                 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9697                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
9698                                       si->dst_reg, si->dst_reg,
9699                                       bpf_target_off(struct skb_shared_info,
9700                                                      gso_size, 2,
9701                                                      target_size));
9702                 break;
9703         case offsetof(struct __sk_buff, wire_len):
9704                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
9705
9706                 off = si->off;
9707                 off -= offsetof(struct __sk_buff, wire_len);
9708                 off += offsetof(struct sk_buff, cb);
9709                 off += offsetof(struct qdisc_skb_cb, pkt_len);
9710                 *target_size = 4;
9711                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
9712                 break;
9713
9714         case offsetof(struct __sk_buff, sk):
9715                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9716                                       si->dst_reg, si->src_reg,
9717                                       offsetof(struct sk_buff, sk));
9718                 break;
9719         case offsetof(struct __sk_buff, hwtstamp):
9720                 BUILD_BUG_ON(sizeof_field(struct skb_shared_hwtstamps, hwtstamp) != 8);
9721                 BUILD_BUG_ON(offsetof(struct skb_shared_hwtstamps, hwtstamp) != 0);
9722
9723                 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9724                 *insn++ = BPF_LDX_MEM(BPF_DW,
9725                                       si->dst_reg, si->dst_reg,
9726                                       bpf_target_off(struct skb_shared_info,
9727                                                      hwtstamps, 8,
9728                                                      target_size));
9729                 break;
9730         }
9731
9732         return insn - insn_buf;
9733 }
9734
9735 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
9736                                 const struct bpf_insn *si,
9737                                 struct bpf_insn *insn_buf,
9738                                 struct bpf_prog *prog, u32 *target_size)
9739 {
9740         struct bpf_insn *insn = insn_buf;
9741         int off;
9742
9743         switch (si->off) {
9744         case offsetof(struct bpf_sock, bound_dev_if):
9745                 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
9746
9747                 if (type == BPF_WRITE)
9748                         *insn++ = BPF_EMIT_STORE(BPF_W, si,
9749                                                  offsetof(struct sock, sk_bound_dev_if));
9750                 else
9751                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9752                                       offsetof(struct sock, sk_bound_dev_if));
9753                 break;
9754
9755         case offsetof(struct bpf_sock, mark):
9756                 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
9757
9758                 if (type == BPF_WRITE)
9759                         *insn++ = BPF_EMIT_STORE(BPF_W, si,
9760                                                  offsetof(struct sock, sk_mark));
9761                 else
9762                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9763                                       offsetof(struct sock, sk_mark));
9764                 break;
9765
9766         case offsetof(struct bpf_sock, priority):
9767                 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
9768
9769                 if (type == BPF_WRITE)
9770                         *insn++ = BPF_EMIT_STORE(BPF_W, si,
9771                                                  offsetof(struct sock, sk_priority));
9772                 else
9773                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9774                                       offsetof(struct sock, sk_priority));
9775                 break;
9776
9777         case offsetof(struct bpf_sock, family):
9778                 *insn++ = BPF_LDX_MEM(
9779                         BPF_FIELD_SIZEOF(struct sock_common, skc_family),
9780                         si->dst_reg, si->src_reg,
9781                         bpf_target_off(struct sock_common,
9782                                        skc_family,
9783                                        sizeof_field(struct sock_common,
9784                                                     skc_family),
9785                                        target_size));
9786                 break;
9787
9788         case offsetof(struct bpf_sock, type):
9789                 *insn++ = BPF_LDX_MEM(
9790                         BPF_FIELD_SIZEOF(struct sock, sk_type),
9791                         si->dst_reg, si->src_reg,
9792                         bpf_target_off(struct sock, sk_type,
9793                                        sizeof_field(struct sock, sk_type),
9794                                        target_size));
9795                 break;
9796
9797         case offsetof(struct bpf_sock, protocol):
9798                 *insn++ = BPF_LDX_MEM(
9799                         BPF_FIELD_SIZEOF(struct sock, sk_protocol),
9800                         si->dst_reg, si->src_reg,
9801                         bpf_target_off(struct sock, sk_protocol,
9802                                        sizeof_field(struct sock, sk_protocol),
9803                                        target_size));
9804                 break;
9805
9806         case offsetof(struct bpf_sock, src_ip4):
9807                 *insn++ = BPF_LDX_MEM(
9808                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9809                         bpf_target_off(struct sock_common, skc_rcv_saddr,
9810                                        sizeof_field(struct sock_common,
9811                                                     skc_rcv_saddr),
9812                                        target_size));
9813                 break;
9814
9815         case offsetof(struct bpf_sock, dst_ip4):
9816                 *insn++ = BPF_LDX_MEM(
9817                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9818                         bpf_target_off(struct sock_common, skc_daddr,
9819                                        sizeof_field(struct sock_common,
9820                                                     skc_daddr),
9821                                        target_size));
9822                 break;
9823
9824         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
9825 #if IS_ENABLED(CONFIG_IPV6)
9826                 off = si->off;
9827                 off -= offsetof(struct bpf_sock, src_ip6[0]);
9828                 *insn++ = BPF_LDX_MEM(
9829                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9830                         bpf_target_off(
9831                                 struct sock_common,
9832                                 skc_v6_rcv_saddr.s6_addr32[0],
9833                                 sizeof_field(struct sock_common,
9834                                              skc_v6_rcv_saddr.s6_addr32[0]),
9835                                 target_size) + off);
9836 #else
9837                 (void)off;
9838                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9839 #endif
9840                 break;
9841
9842         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
9843 #if IS_ENABLED(CONFIG_IPV6)
9844                 off = si->off;
9845                 off -= offsetof(struct bpf_sock, dst_ip6[0]);
9846                 *insn++ = BPF_LDX_MEM(
9847                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9848                         bpf_target_off(struct sock_common,
9849                                        skc_v6_daddr.s6_addr32[0],
9850                                        sizeof_field(struct sock_common,
9851                                                     skc_v6_daddr.s6_addr32[0]),
9852                                        target_size) + off);
9853 #else
9854                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9855                 *target_size = 4;
9856 #endif
9857                 break;
9858
9859         case offsetof(struct bpf_sock, src_port):
9860                 *insn++ = BPF_LDX_MEM(
9861                         BPF_FIELD_SIZEOF(struct sock_common, skc_num),
9862                         si->dst_reg, si->src_reg,
9863                         bpf_target_off(struct sock_common, skc_num,
9864                                        sizeof_field(struct sock_common,
9865                                                     skc_num),
9866                                        target_size));
9867                 break;
9868
9869         case offsetof(struct bpf_sock, dst_port):
9870                 *insn++ = BPF_LDX_MEM(
9871                         BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
9872                         si->dst_reg, si->src_reg,
9873                         bpf_target_off(struct sock_common, skc_dport,
9874                                        sizeof_field(struct sock_common,
9875                                                     skc_dport),
9876                                        target_size));
9877                 break;
9878
9879         case offsetof(struct bpf_sock, state):
9880                 *insn++ = BPF_LDX_MEM(
9881                         BPF_FIELD_SIZEOF(struct sock_common, skc_state),
9882                         si->dst_reg, si->src_reg,
9883                         bpf_target_off(struct sock_common, skc_state,
9884                                        sizeof_field(struct sock_common,
9885                                                     skc_state),
9886                                        target_size));
9887                 break;
9888         case offsetof(struct bpf_sock, rx_queue_mapping):
9889 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
9890                 *insn++ = BPF_LDX_MEM(
9891                         BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
9892                         si->dst_reg, si->src_reg,
9893                         bpf_target_off(struct sock, sk_rx_queue_mapping,
9894                                        sizeof_field(struct sock,
9895                                                     sk_rx_queue_mapping),
9896                                        target_size));
9897                 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
9898                                       1);
9899                 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9900 #else
9901                 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9902                 *target_size = 2;
9903 #endif
9904                 break;
9905         }
9906
9907         return insn - insn_buf;
9908 }
9909
9910 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
9911                                          const struct bpf_insn *si,
9912                                          struct bpf_insn *insn_buf,
9913                                          struct bpf_prog *prog, u32 *target_size)
9914 {
9915         struct bpf_insn *insn = insn_buf;
9916
9917         switch (si->off) {
9918         case offsetof(struct __sk_buff, ifindex):
9919                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9920                                       si->dst_reg, si->src_reg,
9921                                       offsetof(struct sk_buff, dev));
9922                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9923                                       bpf_target_off(struct net_device, ifindex, 4,
9924                                                      target_size));
9925                 break;
9926         default:
9927                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9928                                               target_size);
9929         }
9930
9931         return insn - insn_buf;
9932 }
9933
9934 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
9935                                   const struct bpf_insn *si,
9936                                   struct bpf_insn *insn_buf,
9937                                   struct bpf_prog *prog, u32 *target_size)
9938 {
9939         struct bpf_insn *insn = insn_buf;
9940
9941         switch (si->off) {
9942         case offsetof(struct xdp_md, data):
9943                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
9944                                       si->dst_reg, si->src_reg,
9945                                       offsetof(struct xdp_buff, data));
9946                 break;
9947         case offsetof(struct xdp_md, data_meta):
9948                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
9949                                       si->dst_reg, si->src_reg,
9950                                       offsetof(struct xdp_buff, data_meta));
9951                 break;
9952         case offsetof(struct xdp_md, data_end):
9953                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
9954                                       si->dst_reg, si->src_reg,
9955                                       offsetof(struct xdp_buff, data_end));
9956                 break;
9957         case offsetof(struct xdp_md, ingress_ifindex):
9958                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9959                                       si->dst_reg, si->src_reg,
9960                                       offsetof(struct xdp_buff, rxq));
9961                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
9962                                       si->dst_reg, si->dst_reg,
9963                                       offsetof(struct xdp_rxq_info, dev));
9964                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9965                                       offsetof(struct net_device, ifindex));
9966                 break;
9967         case offsetof(struct xdp_md, rx_queue_index):
9968                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9969                                       si->dst_reg, si->src_reg,
9970                                       offsetof(struct xdp_buff, rxq));
9971                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9972                                       offsetof(struct xdp_rxq_info,
9973                                                queue_index));
9974                 break;
9975         case offsetof(struct xdp_md, egress_ifindex):
9976                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
9977                                       si->dst_reg, si->src_reg,
9978                                       offsetof(struct xdp_buff, txq));
9979                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
9980                                       si->dst_reg, si->dst_reg,
9981                                       offsetof(struct xdp_txq_info, dev));
9982                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9983                                       offsetof(struct net_device, ifindex));
9984                 break;
9985         }
9986
9987         return insn - insn_buf;
9988 }
9989
9990 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
9991  * context Structure, F is Field in context structure that contains a pointer
9992  * to Nested Structure of type NS that has the field NF.
9993  *
9994  * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
9995  * sure that SIZE is not greater than actual size of S.F.NF.
9996  *
9997  * If offset OFF is provided, the load happens from that offset relative to
9998  * offset of NF.
9999  */
10000 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF)          \
10001         do {                                                                   \
10002                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg,     \
10003                                       si->src_reg, offsetof(S, F));            \
10004                 *insn++ = BPF_LDX_MEM(                                         \
10005                         SIZE, si->dst_reg, si->dst_reg,                        \
10006                         bpf_target_off(NS, NF, sizeof_field(NS, NF),           \
10007                                        target_size)                            \
10008                                 + OFF);                                        \
10009         } while (0)
10010
10011 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF)                              \
10012         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF,                     \
10013                                              BPF_FIELD_SIZEOF(NS, NF), 0)
10014
10015 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
10016  * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
10017  *
10018  * In addition it uses Temporary Field TF (member of struct S) as the 3rd
10019  * "register" since two registers available in convert_ctx_access are not
10020  * enough: we can't override neither SRC, since it contains value to store, nor
10021  * DST since it contains pointer to context that may be used by later
10022  * instructions. But we need a temporary place to save pointer to nested
10023  * structure whose field we want to store to.
10024  */
10025 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF)          \
10026         do {                                                                   \
10027                 int tmp_reg = BPF_REG_9;                                       \
10028                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
10029                         --tmp_reg;                                             \
10030                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
10031                         --tmp_reg;                                             \
10032                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg,            \
10033                                       offsetof(S, TF));                        \
10034                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg,         \
10035                                       si->dst_reg, offsetof(S, F));            \
10036                 *insn++ = BPF_RAW_INSN(SIZE | BPF_MEM | BPF_CLASS(si->code),   \
10037                                        tmp_reg, si->src_reg,                   \
10038                         bpf_target_off(NS, NF, sizeof_field(NS, NF),           \
10039                                        target_size)                            \
10040                                        + OFF,                                  \
10041                                        si->imm);                               \
10042                 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg,            \
10043                                       offsetof(S, TF));                        \
10044         } while (0)
10045
10046 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
10047                                                       TF)                      \
10048         do {                                                                   \
10049                 if (type == BPF_WRITE) {                                       \
10050                         SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE,   \
10051                                                          OFF, TF);             \
10052                 } else {                                                       \
10053                         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(                  \
10054                                 S, NS, F, NF, SIZE, OFF);  \
10055                 }                                                              \
10056         } while (0)
10057
10058 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF)                 \
10059         SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(                         \
10060                 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
10061
10062 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
10063                                         const struct bpf_insn *si,
10064                                         struct bpf_insn *insn_buf,
10065                                         struct bpf_prog *prog, u32 *target_size)
10066 {
10067         int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
10068         struct bpf_insn *insn = insn_buf;
10069
10070         switch (si->off) {
10071         case offsetof(struct bpf_sock_addr, user_family):
10072                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10073                                             struct sockaddr, uaddr, sa_family);
10074                 break;
10075
10076         case offsetof(struct bpf_sock_addr, user_ip4):
10077                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10078                         struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
10079                         sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
10080                 break;
10081
10082         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
10083                 off = si->off;
10084                 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
10085                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10086                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
10087                         sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
10088                         tmp_reg);
10089                 break;
10090
10091         case offsetof(struct bpf_sock_addr, user_port):
10092                 /* To get port we need to know sa_family first and then treat
10093                  * sockaddr as either sockaddr_in or sockaddr_in6.
10094                  * Though we can simplify since port field has same offset and
10095                  * size in both structures.
10096                  * Here we check this invariant and use just one of the
10097                  * structures if it's true.
10098                  */
10099                 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
10100                              offsetof(struct sockaddr_in6, sin6_port));
10101                 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
10102                              sizeof_field(struct sockaddr_in6, sin6_port));
10103                 /* Account for sin6_port being smaller than user_port. */
10104                 port_size = min(port_size, BPF_LDST_BYTES(si));
10105                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10106                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
10107                         sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
10108                 break;
10109
10110         case offsetof(struct bpf_sock_addr, family):
10111                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10112                                             struct sock, sk, sk_family);
10113                 break;
10114
10115         case offsetof(struct bpf_sock_addr, type):
10116                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10117                                             struct sock, sk, sk_type);
10118                 break;
10119
10120         case offsetof(struct bpf_sock_addr, protocol):
10121                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10122                                             struct sock, sk, sk_protocol);
10123                 break;
10124
10125         case offsetof(struct bpf_sock_addr, msg_src_ip4):
10126                 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
10127                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10128                         struct bpf_sock_addr_kern, struct in_addr, t_ctx,
10129                         s_addr, BPF_SIZE(si->code), 0, tmp_reg);
10130                 break;
10131
10132         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
10133                                 msg_src_ip6[3]):
10134                 off = si->off;
10135                 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
10136                 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
10137                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10138                         struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
10139                         s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
10140                 break;
10141         case offsetof(struct bpf_sock_addr, sk):
10142                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
10143                                       si->dst_reg, si->src_reg,
10144                                       offsetof(struct bpf_sock_addr_kern, sk));
10145                 break;
10146         }
10147
10148         return insn - insn_buf;
10149 }
10150
10151 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
10152                                        const struct bpf_insn *si,
10153                                        struct bpf_insn *insn_buf,
10154                                        struct bpf_prog *prog,
10155                                        u32 *target_size)
10156 {
10157         struct bpf_insn *insn = insn_buf;
10158         int off;
10159
10160 /* Helper macro for adding read access to tcp_sock or sock fields. */
10161 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
10162         do {                                                                  \
10163                 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2;     \
10164                 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) >                   \
10165                              sizeof_field(struct bpf_sock_ops, BPF_FIELD));   \
10166                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10167                         reg--;                                                \
10168                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10169                         reg--;                                                \
10170                 if (si->dst_reg == si->src_reg) {                             \
10171                         *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg,       \
10172                                           offsetof(struct bpf_sock_ops_kern,  \
10173                                           temp));                             \
10174                         fullsock_reg = reg;                                   \
10175                         jmp += 2;                                             \
10176                 }                                                             \
10177                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10178                                                 struct bpf_sock_ops_kern,     \
10179                                                 is_fullsock),                 \
10180                                       fullsock_reg, si->src_reg,              \
10181                                       offsetof(struct bpf_sock_ops_kern,      \
10182                                                is_fullsock));                 \
10183                 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp);         \
10184                 if (si->dst_reg == si->src_reg)                               \
10185                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
10186                                       offsetof(struct bpf_sock_ops_kern,      \
10187                                       temp));                                 \
10188                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10189                                                 struct bpf_sock_ops_kern, sk),\
10190                                       si->dst_reg, si->src_reg,               \
10191                                       offsetof(struct bpf_sock_ops_kern, sk));\
10192                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ,                   \
10193                                                        OBJ_FIELD),            \
10194                                       si->dst_reg, si->dst_reg,               \
10195                                       offsetof(OBJ, OBJ_FIELD));              \
10196                 if (si->dst_reg == si->src_reg) {                             \
10197                         *insn++ = BPF_JMP_A(1);                               \
10198                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
10199                                       offsetof(struct bpf_sock_ops_kern,      \
10200                                       temp));                                 \
10201                 }                                                             \
10202         } while (0)
10203
10204 #define SOCK_OPS_GET_SK()                                                             \
10205         do {                                                                  \
10206                 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1;     \
10207                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10208                         reg--;                                                \
10209                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10210                         reg--;                                                \
10211                 if (si->dst_reg == si->src_reg) {                             \
10212                         *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg,       \
10213                                           offsetof(struct bpf_sock_ops_kern,  \
10214                                           temp));                             \
10215                         fullsock_reg = reg;                                   \
10216                         jmp += 2;                                             \
10217                 }                                                             \
10218                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10219                                                 struct bpf_sock_ops_kern,     \
10220                                                 is_fullsock),                 \
10221                                       fullsock_reg, si->src_reg,              \
10222                                       offsetof(struct bpf_sock_ops_kern,      \
10223                                                is_fullsock));                 \
10224                 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp);         \
10225                 if (si->dst_reg == si->src_reg)                               \
10226                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
10227                                       offsetof(struct bpf_sock_ops_kern,      \
10228                                       temp));                                 \
10229                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10230                                                 struct bpf_sock_ops_kern, sk),\
10231                                       si->dst_reg, si->src_reg,               \
10232                                       offsetof(struct bpf_sock_ops_kern, sk));\
10233                 if (si->dst_reg == si->src_reg) {                             \
10234                         *insn++ = BPF_JMP_A(1);                               \
10235                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
10236                                       offsetof(struct bpf_sock_ops_kern,      \
10237                                       temp));                                 \
10238                 }                                                             \
10239         } while (0)
10240
10241 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
10242                 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
10243
10244 /* Helper macro for adding write access to tcp_sock or sock fields.
10245  * The macro is called with two registers, dst_reg which contains a pointer
10246  * to ctx (context) and src_reg which contains the value that should be
10247  * stored. However, we need an additional register since we cannot overwrite
10248  * dst_reg because it may be used later in the program.
10249  * Instead we "borrow" one of the other register. We first save its value
10250  * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
10251  * it at the end of the macro.
10252  */
10253 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
10254         do {                                                                  \
10255                 int reg = BPF_REG_9;                                          \
10256                 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) >                   \
10257                              sizeof_field(struct bpf_sock_ops, BPF_FIELD));   \
10258                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10259                         reg--;                                                \
10260                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10261                         reg--;                                                \
10262                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg,               \
10263                                       offsetof(struct bpf_sock_ops_kern,      \
10264                                                temp));                        \
10265                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10266                                                 struct bpf_sock_ops_kern,     \
10267                                                 is_fullsock),                 \
10268                                       reg, si->dst_reg,                       \
10269                                       offsetof(struct bpf_sock_ops_kern,      \
10270                                                is_fullsock));                 \
10271                 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2);                    \
10272                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10273                                                 struct bpf_sock_ops_kern, sk),\
10274                                       reg, si->dst_reg,                       \
10275                                       offsetof(struct bpf_sock_ops_kern, sk));\
10276                 *insn++ = BPF_RAW_INSN(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD) |     \
10277                                        BPF_MEM | BPF_CLASS(si->code),         \
10278                                        reg, si->src_reg,                      \
10279                                        offsetof(OBJ, OBJ_FIELD),              \
10280                                        si->imm);                              \
10281                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg,               \
10282                                       offsetof(struct bpf_sock_ops_kern,      \
10283                                                temp));                        \
10284         } while (0)
10285
10286 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE)            \
10287         do {                                                                  \
10288                 if (TYPE == BPF_WRITE)                                        \
10289                         SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
10290                 else                                                          \
10291                         SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
10292         } while (0)
10293
10294         switch (si->off) {
10295         case offsetof(struct bpf_sock_ops, op):
10296                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10297                                                        op),
10298                                       si->dst_reg, si->src_reg,
10299                                       offsetof(struct bpf_sock_ops_kern, op));
10300                 break;
10301
10302         case offsetof(struct bpf_sock_ops, replylong[0]) ...
10303              offsetof(struct bpf_sock_ops, replylong[3]):
10304                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
10305                              sizeof_field(struct bpf_sock_ops_kern, reply));
10306                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
10307                              sizeof_field(struct bpf_sock_ops_kern, replylong));
10308                 off = si->off;
10309                 off -= offsetof(struct bpf_sock_ops, replylong[0]);
10310                 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
10311                 if (type == BPF_WRITE)
10312                         *insn++ = BPF_EMIT_STORE(BPF_W, si, off);
10313                 else
10314                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10315                                               off);
10316                 break;
10317
10318         case offsetof(struct bpf_sock_ops, family):
10319                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10320
10321                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10322                                               struct bpf_sock_ops_kern, sk),
10323                                       si->dst_reg, si->src_reg,
10324                                       offsetof(struct bpf_sock_ops_kern, sk));
10325                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10326                                       offsetof(struct sock_common, skc_family));
10327                 break;
10328
10329         case offsetof(struct bpf_sock_ops, remote_ip4):
10330                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10331
10332                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10333                                                 struct bpf_sock_ops_kern, sk),
10334                                       si->dst_reg, si->src_reg,
10335                                       offsetof(struct bpf_sock_ops_kern, sk));
10336                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10337                                       offsetof(struct sock_common, skc_daddr));
10338                 break;
10339
10340         case offsetof(struct bpf_sock_ops, local_ip4):
10341                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10342                                           skc_rcv_saddr) != 4);
10343
10344                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10345                                               struct bpf_sock_ops_kern, sk),
10346                                       si->dst_reg, si->src_reg,
10347                                       offsetof(struct bpf_sock_ops_kern, sk));
10348                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10349                                       offsetof(struct sock_common,
10350                                                skc_rcv_saddr));
10351                 break;
10352
10353         case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
10354              offsetof(struct bpf_sock_ops, remote_ip6[3]):
10355 #if IS_ENABLED(CONFIG_IPV6)
10356                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10357                                           skc_v6_daddr.s6_addr32[0]) != 4);
10358
10359                 off = si->off;
10360                 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
10361                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10362                                                 struct bpf_sock_ops_kern, sk),
10363                                       si->dst_reg, si->src_reg,
10364                                       offsetof(struct bpf_sock_ops_kern, sk));
10365                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10366                                       offsetof(struct sock_common,
10367                                                skc_v6_daddr.s6_addr32[0]) +
10368                                       off);
10369 #else
10370                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10371 #endif
10372                 break;
10373
10374         case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
10375              offsetof(struct bpf_sock_ops, local_ip6[3]):
10376 #if IS_ENABLED(CONFIG_IPV6)
10377                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10378                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10379
10380                 off = si->off;
10381                 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
10382                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10383                                                 struct bpf_sock_ops_kern, sk),
10384                                       si->dst_reg, si->src_reg,
10385                                       offsetof(struct bpf_sock_ops_kern, sk));
10386                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10387                                       offsetof(struct sock_common,
10388                                                skc_v6_rcv_saddr.s6_addr32[0]) +
10389                                       off);
10390 #else
10391                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10392 #endif
10393                 break;
10394
10395         case offsetof(struct bpf_sock_ops, remote_port):
10396                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10397
10398                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10399                                                 struct bpf_sock_ops_kern, sk),
10400                                       si->dst_reg, si->src_reg,
10401                                       offsetof(struct bpf_sock_ops_kern, sk));
10402                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10403                                       offsetof(struct sock_common, skc_dport));
10404 #ifndef __BIG_ENDIAN_BITFIELD
10405                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10406 #endif
10407                 break;
10408
10409         case offsetof(struct bpf_sock_ops, local_port):
10410                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10411
10412                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10413                                                 struct bpf_sock_ops_kern, sk),
10414                                       si->dst_reg, si->src_reg,
10415                                       offsetof(struct bpf_sock_ops_kern, sk));
10416                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10417                                       offsetof(struct sock_common, skc_num));
10418                 break;
10419
10420         case offsetof(struct bpf_sock_ops, is_fullsock):
10421                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10422                                                 struct bpf_sock_ops_kern,
10423                                                 is_fullsock),
10424                                       si->dst_reg, si->src_reg,
10425                                       offsetof(struct bpf_sock_ops_kern,
10426                                                is_fullsock));
10427                 break;
10428
10429         case offsetof(struct bpf_sock_ops, state):
10430                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
10431
10432                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10433                                                 struct bpf_sock_ops_kern, sk),
10434                                       si->dst_reg, si->src_reg,
10435                                       offsetof(struct bpf_sock_ops_kern, sk));
10436                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
10437                                       offsetof(struct sock_common, skc_state));
10438                 break;
10439
10440         case offsetof(struct bpf_sock_ops, rtt_min):
10441                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
10442                              sizeof(struct minmax));
10443                 BUILD_BUG_ON(sizeof(struct minmax) <
10444                              sizeof(struct minmax_sample));
10445
10446                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10447                                                 struct bpf_sock_ops_kern, sk),
10448                                       si->dst_reg, si->src_reg,
10449                                       offsetof(struct bpf_sock_ops_kern, sk));
10450                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10451                                       offsetof(struct tcp_sock, rtt_min) +
10452                                       sizeof_field(struct minmax_sample, t));
10453                 break;
10454
10455         case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
10456                 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
10457                                    struct tcp_sock);
10458                 break;
10459
10460         case offsetof(struct bpf_sock_ops, sk_txhash):
10461                 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
10462                                           struct sock, type);
10463                 break;
10464         case offsetof(struct bpf_sock_ops, snd_cwnd):
10465                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
10466                 break;
10467         case offsetof(struct bpf_sock_ops, srtt_us):
10468                 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
10469                 break;
10470         case offsetof(struct bpf_sock_ops, snd_ssthresh):
10471                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
10472                 break;
10473         case offsetof(struct bpf_sock_ops, rcv_nxt):
10474                 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
10475                 break;
10476         case offsetof(struct bpf_sock_ops, snd_nxt):
10477                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
10478                 break;
10479         case offsetof(struct bpf_sock_ops, snd_una):
10480                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
10481                 break;
10482         case offsetof(struct bpf_sock_ops, mss_cache):
10483                 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
10484                 break;
10485         case offsetof(struct bpf_sock_ops, ecn_flags):
10486                 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
10487                 break;
10488         case offsetof(struct bpf_sock_ops, rate_delivered):
10489                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
10490                 break;
10491         case offsetof(struct bpf_sock_ops, rate_interval_us):
10492                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
10493                 break;
10494         case offsetof(struct bpf_sock_ops, packets_out):
10495                 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
10496                 break;
10497         case offsetof(struct bpf_sock_ops, retrans_out):
10498                 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
10499                 break;
10500         case offsetof(struct bpf_sock_ops, total_retrans):
10501                 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
10502                 break;
10503         case offsetof(struct bpf_sock_ops, segs_in):
10504                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
10505                 break;
10506         case offsetof(struct bpf_sock_ops, data_segs_in):
10507                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
10508                 break;
10509         case offsetof(struct bpf_sock_ops, segs_out):
10510                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
10511                 break;
10512         case offsetof(struct bpf_sock_ops, data_segs_out):
10513                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
10514                 break;
10515         case offsetof(struct bpf_sock_ops, lost_out):
10516                 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
10517                 break;
10518         case offsetof(struct bpf_sock_ops, sacked_out):
10519                 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
10520                 break;
10521         case offsetof(struct bpf_sock_ops, bytes_received):
10522                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
10523                 break;
10524         case offsetof(struct bpf_sock_ops, bytes_acked):
10525                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
10526                 break;
10527         case offsetof(struct bpf_sock_ops, sk):
10528                 SOCK_OPS_GET_SK();
10529                 break;
10530         case offsetof(struct bpf_sock_ops, skb_data_end):
10531                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10532                                                        skb_data_end),
10533                                       si->dst_reg, si->src_reg,
10534                                       offsetof(struct bpf_sock_ops_kern,
10535                                                skb_data_end));
10536                 break;
10537         case offsetof(struct bpf_sock_ops, skb_data):
10538                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10539                                                        skb),
10540                                       si->dst_reg, si->src_reg,
10541                                       offsetof(struct bpf_sock_ops_kern,
10542                                                skb));
10543                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10544                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10545                                       si->dst_reg, si->dst_reg,
10546                                       offsetof(struct sk_buff, data));
10547                 break;
10548         case offsetof(struct bpf_sock_ops, skb_len):
10549                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10550                                                        skb),
10551                                       si->dst_reg, si->src_reg,
10552                                       offsetof(struct bpf_sock_ops_kern,
10553                                                skb));
10554                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10555                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10556                                       si->dst_reg, si->dst_reg,
10557                                       offsetof(struct sk_buff, len));
10558                 break;
10559         case offsetof(struct bpf_sock_ops, skb_tcp_flags):
10560                 off = offsetof(struct sk_buff, cb);
10561                 off += offsetof(struct tcp_skb_cb, tcp_flags);
10562                 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
10563                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10564                                                        skb),
10565                                       si->dst_reg, si->src_reg,
10566                                       offsetof(struct bpf_sock_ops_kern,
10567                                                skb));
10568                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10569                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
10570                                                        tcp_flags),
10571                                       si->dst_reg, si->dst_reg, off);
10572                 break;
10573         case offsetof(struct bpf_sock_ops, skb_hwtstamp): {
10574                 struct bpf_insn *jmp_on_null_skb;
10575
10576                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10577                                                        skb),
10578                                       si->dst_reg, si->src_reg,
10579                                       offsetof(struct bpf_sock_ops_kern,
10580                                                skb));
10581                 /* Reserve one insn to test skb == NULL */
10582                 jmp_on_null_skb = insn++;
10583                 insn = bpf_convert_shinfo_access(si->dst_reg, si->dst_reg, insn);
10584                 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
10585                                       bpf_target_off(struct skb_shared_info,
10586                                                      hwtstamps, 8,
10587                                                      target_size));
10588                 *jmp_on_null_skb = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0,
10589                                                insn - jmp_on_null_skb - 1);
10590                 break;
10591         }
10592         }
10593         return insn - insn_buf;
10594 }
10595
10596 /* data_end = skb->data + skb_headlen() */
10597 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
10598                                                     struct bpf_insn *insn)
10599 {
10600         int reg;
10601         int temp_reg_off = offsetof(struct sk_buff, cb) +
10602                            offsetof(struct sk_skb_cb, temp_reg);
10603
10604         if (si->src_reg == si->dst_reg) {
10605                 /* We need an extra register, choose and save a register. */
10606                 reg = BPF_REG_9;
10607                 if (si->src_reg == reg || si->dst_reg == reg)
10608                         reg--;
10609                 if (si->src_reg == reg || si->dst_reg == reg)
10610                         reg--;
10611                 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off);
10612         } else {
10613                 reg = si->dst_reg;
10614         }
10615
10616         /* reg = skb->data */
10617         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10618                               reg, si->src_reg,
10619                               offsetof(struct sk_buff, data));
10620         /* AX = skb->len */
10621         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10622                               BPF_REG_AX, si->src_reg,
10623                               offsetof(struct sk_buff, len));
10624         /* reg = skb->data + skb->len */
10625         *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX);
10626         /* AX = skb->data_len */
10627         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
10628                               BPF_REG_AX, si->src_reg,
10629                               offsetof(struct sk_buff, data_len));
10630
10631         /* reg = skb->data + skb->len - skb->data_len */
10632         *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX);
10633
10634         if (si->src_reg == si->dst_reg) {
10635                 /* Restore the saved register */
10636                 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg);
10637                 *insn++ = BPF_MOV64_REG(si->dst_reg, reg);
10638                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off);
10639         }
10640
10641         return insn;
10642 }
10643
10644 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
10645                                      const struct bpf_insn *si,
10646                                      struct bpf_insn *insn_buf,
10647                                      struct bpf_prog *prog, u32 *target_size)
10648 {
10649         struct bpf_insn *insn = insn_buf;
10650         int off;
10651
10652         switch (si->off) {
10653         case offsetof(struct __sk_buff, data_end):
10654                 insn = bpf_convert_data_end_access(si, insn);
10655                 break;
10656         case offsetof(struct __sk_buff, cb[0]) ...
10657              offsetofend(struct __sk_buff, cb[4]) - 1:
10658                 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20);
10659                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
10660                               offsetof(struct sk_skb_cb, data)) %
10661                              sizeof(__u64));
10662
10663                 prog->cb_access = 1;
10664                 off  = si->off;
10665                 off -= offsetof(struct __sk_buff, cb[0]);
10666                 off += offsetof(struct sk_buff, cb);
10667                 off += offsetof(struct sk_skb_cb, data);
10668                 if (type == BPF_WRITE)
10669                         *insn++ = BPF_EMIT_STORE(BPF_SIZE(si->code), si, off);
10670                 else
10671                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
10672                                               si->src_reg, off);
10673                 break;
10674
10675
10676         default:
10677                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
10678                                               target_size);
10679         }
10680
10681         return insn - insn_buf;
10682 }
10683
10684 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
10685                                      const struct bpf_insn *si,
10686                                      struct bpf_insn *insn_buf,
10687                                      struct bpf_prog *prog, u32 *target_size)
10688 {
10689         struct bpf_insn *insn = insn_buf;
10690 #if IS_ENABLED(CONFIG_IPV6)
10691         int off;
10692 #endif
10693
10694         /* convert ctx uses the fact sg element is first in struct */
10695         BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
10696
10697         switch (si->off) {
10698         case offsetof(struct sk_msg_md, data):
10699                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
10700                                       si->dst_reg, si->src_reg,
10701                                       offsetof(struct sk_msg, data));
10702                 break;
10703         case offsetof(struct sk_msg_md, data_end):
10704                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
10705                                       si->dst_reg, si->src_reg,
10706                                       offsetof(struct sk_msg, data_end));
10707                 break;
10708         case offsetof(struct sk_msg_md, family):
10709                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10710
10711                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10712                                               struct sk_msg, sk),
10713                                       si->dst_reg, si->src_reg,
10714                                       offsetof(struct sk_msg, sk));
10715                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10716                                       offsetof(struct sock_common, skc_family));
10717                 break;
10718
10719         case offsetof(struct sk_msg_md, remote_ip4):
10720                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10721
10722                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10723                                                 struct sk_msg, sk),
10724                                       si->dst_reg, si->src_reg,
10725                                       offsetof(struct sk_msg, sk));
10726                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10727                                       offsetof(struct sock_common, skc_daddr));
10728                 break;
10729
10730         case offsetof(struct sk_msg_md, local_ip4):
10731                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10732                                           skc_rcv_saddr) != 4);
10733
10734                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10735                                               struct sk_msg, sk),
10736                                       si->dst_reg, si->src_reg,
10737                                       offsetof(struct sk_msg, sk));
10738                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10739                                       offsetof(struct sock_common,
10740                                                skc_rcv_saddr));
10741                 break;
10742
10743         case offsetof(struct sk_msg_md, remote_ip6[0]) ...
10744              offsetof(struct sk_msg_md, remote_ip6[3]):
10745 #if IS_ENABLED(CONFIG_IPV6)
10746                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10747                                           skc_v6_daddr.s6_addr32[0]) != 4);
10748
10749                 off = si->off;
10750                 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
10751                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10752                                                 struct sk_msg, sk),
10753                                       si->dst_reg, si->src_reg,
10754                                       offsetof(struct sk_msg, sk));
10755                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10756                                       offsetof(struct sock_common,
10757                                                skc_v6_daddr.s6_addr32[0]) +
10758                                       off);
10759 #else
10760                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10761 #endif
10762                 break;
10763
10764         case offsetof(struct sk_msg_md, local_ip6[0]) ...
10765              offsetof(struct sk_msg_md, local_ip6[3]):
10766 #if IS_ENABLED(CONFIG_IPV6)
10767                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10768                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10769
10770                 off = si->off;
10771                 off -= offsetof(struct sk_msg_md, local_ip6[0]);
10772                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10773                                                 struct sk_msg, sk),
10774                                       si->dst_reg, si->src_reg,
10775                                       offsetof(struct sk_msg, sk));
10776                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10777                                       offsetof(struct sock_common,
10778                                                skc_v6_rcv_saddr.s6_addr32[0]) +
10779                                       off);
10780 #else
10781                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10782 #endif
10783                 break;
10784
10785         case offsetof(struct sk_msg_md, remote_port):
10786                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10787
10788                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10789                                                 struct sk_msg, sk),
10790                                       si->dst_reg, si->src_reg,
10791                                       offsetof(struct sk_msg, sk));
10792                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10793                                       offsetof(struct sock_common, skc_dport));
10794 #ifndef __BIG_ENDIAN_BITFIELD
10795                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10796 #endif
10797                 break;
10798
10799         case offsetof(struct sk_msg_md, local_port):
10800                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10801
10802                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10803                                                 struct sk_msg, sk),
10804                                       si->dst_reg, si->src_reg,
10805                                       offsetof(struct sk_msg, sk));
10806                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10807                                       offsetof(struct sock_common, skc_num));
10808                 break;
10809
10810         case offsetof(struct sk_msg_md, size):
10811                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
10812                                       si->dst_reg, si->src_reg,
10813                                       offsetof(struct sk_msg_sg, size));
10814                 break;
10815
10816         case offsetof(struct sk_msg_md, sk):
10817                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
10818                                       si->dst_reg, si->src_reg,
10819                                       offsetof(struct sk_msg, sk));
10820                 break;
10821         }
10822
10823         return insn - insn_buf;
10824 }
10825
10826 const struct bpf_verifier_ops sk_filter_verifier_ops = {
10827         .get_func_proto         = sk_filter_func_proto,
10828         .is_valid_access        = sk_filter_is_valid_access,
10829         .convert_ctx_access     = bpf_convert_ctx_access,
10830         .gen_ld_abs             = bpf_gen_ld_abs,
10831 };
10832
10833 const struct bpf_prog_ops sk_filter_prog_ops = {
10834         .test_run               = bpf_prog_test_run_skb,
10835 };
10836
10837 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
10838         .get_func_proto         = tc_cls_act_func_proto,
10839         .is_valid_access        = tc_cls_act_is_valid_access,
10840         .convert_ctx_access     = tc_cls_act_convert_ctx_access,
10841         .gen_prologue           = tc_cls_act_prologue,
10842         .gen_ld_abs             = bpf_gen_ld_abs,
10843         .btf_struct_access      = tc_cls_act_btf_struct_access,
10844 };
10845
10846 const struct bpf_prog_ops tc_cls_act_prog_ops = {
10847         .test_run               = bpf_prog_test_run_skb,
10848 };
10849
10850 const struct bpf_verifier_ops xdp_verifier_ops = {
10851         .get_func_proto         = xdp_func_proto,
10852         .is_valid_access        = xdp_is_valid_access,
10853         .convert_ctx_access     = xdp_convert_ctx_access,
10854         .gen_prologue           = bpf_noop_prologue,
10855         .btf_struct_access      = xdp_btf_struct_access,
10856 };
10857
10858 const struct bpf_prog_ops xdp_prog_ops = {
10859         .test_run               = bpf_prog_test_run_xdp,
10860 };
10861
10862 const struct bpf_verifier_ops cg_skb_verifier_ops = {
10863         .get_func_proto         = cg_skb_func_proto,
10864         .is_valid_access        = cg_skb_is_valid_access,
10865         .convert_ctx_access     = bpf_convert_ctx_access,
10866 };
10867
10868 const struct bpf_prog_ops cg_skb_prog_ops = {
10869         .test_run               = bpf_prog_test_run_skb,
10870 };
10871
10872 const struct bpf_verifier_ops lwt_in_verifier_ops = {
10873         .get_func_proto         = lwt_in_func_proto,
10874         .is_valid_access        = lwt_is_valid_access,
10875         .convert_ctx_access     = bpf_convert_ctx_access,
10876 };
10877
10878 const struct bpf_prog_ops lwt_in_prog_ops = {
10879         .test_run               = bpf_prog_test_run_skb,
10880 };
10881
10882 const struct bpf_verifier_ops lwt_out_verifier_ops = {
10883         .get_func_proto         = lwt_out_func_proto,
10884         .is_valid_access        = lwt_is_valid_access,
10885         .convert_ctx_access     = bpf_convert_ctx_access,
10886 };
10887
10888 const struct bpf_prog_ops lwt_out_prog_ops = {
10889         .test_run               = bpf_prog_test_run_skb,
10890 };
10891
10892 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
10893         .get_func_proto         = lwt_xmit_func_proto,
10894         .is_valid_access        = lwt_is_valid_access,
10895         .convert_ctx_access     = bpf_convert_ctx_access,
10896         .gen_prologue           = tc_cls_act_prologue,
10897 };
10898
10899 const struct bpf_prog_ops lwt_xmit_prog_ops = {
10900         .test_run               = bpf_prog_test_run_skb,
10901 };
10902
10903 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
10904         .get_func_proto         = lwt_seg6local_func_proto,
10905         .is_valid_access        = lwt_is_valid_access,
10906         .convert_ctx_access     = bpf_convert_ctx_access,
10907 };
10908
10909 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
10910         .test_run               = bpf_prog_test_run_skb,
10911 };
10912
10913 const struct bpf_verifier_ops cg_sock_verifier_ops = {
10914         .get_func_proto         = sock_filter_func_proto,
10915         .is_valid_access        = sock_filter_is_valid_access,
10916         .convert_ctx_access     = bpf_sock_convert_ctx_access,
10917 };
10918
10919 const struct bpf_prog_ops cg_sock_prog_ops = {
10920 };
10921
10922 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
10923         .get_func_proto         = sock_addr_func_proto,
10924         .is_valid_access        = sock_addr_is_valid_access,
10925         .convert_ctx_access     = sock_addr_convert_ctx_access,
10926 };
10927
10928 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
10929 };
10930
10931 const struct bpf_verifier_ops sock_ops_verifier_ops = {
10932         .get_func_proto         = sock_ops_func_proto,
10933         .is_valid_access        = sock_ops_is_valid_access,
10934         .convert_ctx_access     = sock_ops_convert_ctx_access,
10935 };
10936
10937 const struct bpf_prog_ops sock_ops_prog_ops = {
10938 };
10939
10940 const struct bpf_verifier_ops sk_skb_verifier_ops = {
10941         .get_func_proto         = sk_skb_func_proto,
10942         .is_valid_access        = sk_skb_is_valid_access,
10943         .convert_ctx_access     = sk_skb_convert_ctx_access,
10944         .gen_prologue           = sk_skb_prologue,
10945 };
10946
10947 const struct bpf_prog_ops sk_skb_prog_ops = {
10948 };
10949
10950 const struct bpf_verifier_ops sk_msg_verifier_ops = {
10951         .get_func_proto         = sk_msg_func_proto,
10952         .is_valid_access        = sk_msg_is_valid_access,
10953         .convert_ctx_access     = sk_msg_convert_ctx_access,
10954         .gen_prologue           = bpf_noop_prologue,
10955 };
10956
10957 const struct bpf_prog_ops sk_msg_prog_ops = {
10958 };
10959
10960 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
10961         .get_func_proto         = flow_dissector_func_proto,
10962         .is_valid_access        = flow_dissector_is_valid_access,
10963         .convert_ctx_access     = flow_dissector_convert_ctx_access,
10964 };
10965
10966 const struct bpf_prog_ops flow_dissector_prog_ops = {
10967         .test_run               = bpf_prog_test_run_flow_dissector,
10968 };
10969
10970 int sk_detach_filter(struct sock *sk)
10971 {
10972         int ret = -ENOENT;
10973         struct sk_filter *filter;
10974
10975         if (sock_flag(sk, SOCK_FILTER_LOCKED))
10976                 return -EPERM;
10977
10978         filter = rcu_dereference_protected(sk->sk_filter,
10979                                            lockdep_sock_is_held(sk));
10980         if (filter) {
10981                 RCU_INIT_POINTER(sk->sk_filter, NULL);
10982                 sk_filter_uncharge(sk, filter);
10983                 ret = 0;
10984         }
10985
10986         return ret;
10987 }
10988 EXPORT_SYMBOL_GPL(sk_detach_filter);
10989
10990 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len)
10991 {
10992         struct sock_fprog_kern *fprog;
10993         struct sk_filter *filter;
10994         int ret = 0;
10995
10996         sockopt_lock_sock(sk);
10997         filter = rcu_dereference_protected(sk->sk_filter,
10998                                            lockdep_sock_is_held(sk));
10999         if (!filter)
11000                 goto out;
11001
11002         /* We're copying the filter that has been originally attached,
11003          * so no conversion/decode needed anymore. eBPF programs that
11004          * have no original program cannot be dumped through this.
11005          */
11006         ret = -EACCES;
11007         fprog = filter->prog->orig_prog;
11008         if (!fprog)
11009                 goto out;
11010
11011         ret = fprog->len;
11012         if (!len)
11013                 /* User space only enquires number of filter blocks. */
11014                 goto out;
11015
11016         ret = -EINVAL;
11017         if (len < fprog->len)
11018                 goto out;
11019
11020         ret = -EFAULT;
11021         if (copy_to_sockptr(optval, fprog->filter, bpf_classic_proglen(fprog)))
11022                 goto out;
11023
11024         /* Instead of bytes, the API requests to return the number
11025          * of filter blocks.
11026          */
11027         ret = fprog->len;
11028 out:
11029         sockopt_release_sock(sk);
11030         return ret;
11031 }
11032
11033 #ifdef CONFIG_INET
11034 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
11035                                     struct sock_reuseport *reuse,
11036                                     struct sock *sk, struct sk_buff *skb,
11037                                     struct sock *migrating_sk,
11038                                     u32 hash)
11039 {
11040         reuse_kern->skb = skb;
11041         reuse_kern->sk = sk;
11042         reuse_kern->selected_sk = NULL;
11043         reuse_kern->migrating_sk = migrating_sk;
11044         reuse_kern->data_end = skb->data + skb_headlen(skb);
11045         reuse_kern->hash = hash;
11046         reuse_kern->reuseport_id = reuse->reuseport_id;
11047         reuse_kern->bind_inany = reuse->bind_inany;
11048 }
11049
11050 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
11051                                   struct bpf_prog *prog, struct sk_buff *skb,
11052                                   struct sock *migrating_sk,
11053                                   u32 hash)
11054 {
11055         struct sk_reuseport_kern reuse_kern;
11056         enum sk_action action;
11057
11058         bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
11059         action = bpf_prog_run(prog, &reuse_kern);
11060
11061         if (action == SK_PASS)
11062                 return reuse_kern.selected_sk;
11063         else
11064                 return ERR_PTR(-ECONNREFUSED);
11065 }
11066
11067 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
11068            struct bpf_map *, map, void *, key, u32, flags)
11069 {
11070         bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
11071         struct sock_reuseport *reuse;
11072         struct sock *selected_sk;
11073
11074         selected_sk = map->ops->map_lookup_elem(map, key);
11075         if (!selected_sk)
11076                 return -ENOENT;
11077
11078         reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
11079         if (!reuse) {
11080                 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
11081                 if (sk_is_refcounted(selected_sk))
11082                         sock_put(selected_sk);
11083
11084                 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
11085                  * The only (!reuse) case here is - the sk has already been
11086                  * unhashed (e.g. by close()), so treat it as -ENOENT.
11087                  *
11088                  * Other maps (e.g. sock_map) do not provide this guarantee and
11089                  * the sk may never be in the reuseport group to begin with.
11090                  */
11091                 return is_sockarray ? -ENOENT : -EINVAL;
11092         }
11093
11094         if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
11095                 struct sock *sk = reuse_kern->sk;
11096
11097                 if (sk->sk_protocol != selected_sk->sk_protocol)
11098                         return -EPROTOTYPE;
11099                 else if (sk->sk_family != selected_sk->sk_family)
11100                         return -EAFNOSUPPORT;
11101
11102                 /* Catch all. Likely bound to a different sockaddr. */
11103                 return -EBADFD;
11104         }
11105
11106         reuse_kern->selected_sk = selected_sk;
11107
11108         return 0;
11109 }
11110
11111 static const struct bpf_func_proto sk_select_reuseport_proto = {
11112         .func           = sk_select_reuseport,
11113         .gpl_only       = false,
11114         .ret_type       = RET_INTEGER,
11115         .arg1_type      = ARG_PTR_TO_CTX,
11116         .arg2_type      = ARG_CONST_MAP_PTR,
11117         .arg3_type      = ARG_PTR_TO_MAP_KEY,
11118         .arg4_type      = ARG_ANYTHING,
11119 };
11120
11121 BPF_CALL_4(sk_reuseport_load_bytes,
11122            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11123            void *, to, u32, len)
11124 {
11125         return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
11126 }
11127
11128 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
11129         .func           = sk_reuseport_load_bytes,
11130         .gpl_only       = false,
11131         .ret_type       = RET_INTEGER,
11132         .arg1_type      = ARG_PTR_TO_CTX,
11133         .arg2_type      = ARG_ANYTHING,
11134         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
11135         .arg4_type      = ARG_CONST_SIZE,
11136 };
11137
11138 BPF_CALL_5(sk_reuseport_load_bytes_relative,
11139            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11140            void *, to, u32, len, u32, start_header)
11141 {
11142         return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
11143                                                len, start_header);
11144 }
11145
11146 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
11147         .func           = sk_reuseport_load_bytes_relative,
11148         .gpl_only       = false,
11149         .ret_type       = RET_INTEGER,
11150         .arg1_type      = ARG_PTR_TO_CTX,
11151         .arg2_type      = ARG_ANYTHING,
11152         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
11153         .arg4_type      = ARG_CONST_SIZE,
11154         .arg5_type      = ARG_ANYTHING,
11155 };
11156
11157 static const struct bpf_func_proto *
11158 sk_reuseport_func_proto(enum bpf_func_id func_id,
11159                         const struct bpf_prog *prog)
11160 {
11161         switch (func_id) {
11162         case BPF_FUNC_sk_select_reuseport:
11163                 return &sk_select_reuseport_proto;
11164         case BPF_FUNC_skb_load_bytes:
11165                 return &sk_reuseport_load_bytes_proto;
11166         case BPF_FUNC_skb_load_bytes_relative:
11167                 return &sk_reuseport_load_bytes_relative_proto;
11168         case BPF_FUNC_get_socket_cookie:
11169                 return &bpf_get_socket_ptr_cookie_proto;
11170         case BPF_FUNC_ktime_get_coarse_ns:
11171                 return &bpf_ktime_get_coarse_ns_proto;
11172         default:
11173                 return bpf_base_func_proto(func_id);
11174         }
11175 }
11176
11177 static bool
11178 sk_reuseport_is_valid_access(int off, int size,
11179                              enum bpf_access_type type,
11180                              const struct bpf_prog *prog,
11181                              struct bpf_insn_access_aux *info)
11182 {
11183         const u32 size_default = sizeof(__u32);
11184
11185         if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
11186             off % size || type != BPF_READ)
11187                 return false;
11188
11189         switch (off) {
11190         case offsetof(struct sk_reuseport_md, data):
11191                 info->reg_type = PTR_TO_PACKET;
11192                 return size == sizeof(__u64);
11193
11194         case offsetof(struct sk_reuseport_md, data_end):
11195                 info->reg_type = PTR_TO_PACKET_END;
11196                 return size == sizeof(__u64);
11197
11198         case offsetof(struct sk_reuseport_md, hash):
11199                 return size == size_default;
11200
11201         case offsetof(struct sk_reuseport_md, sk):
11202                 info->reg_type = PTR_TO_SOCKET;
11203                 return size == sizeof(__u64);
11204
11205         case offsetof(struct sk_reuseport_md, migrating_sk):
11206                 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
11207                 return size == sizeof(__u64);
11208
11209         /* Fields that allow narrowing */
11210         case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
11211                 if (size < sizeof_field(struct sk_buff, protocol))
11212                         return false;
11213                 fallthrough;
11214         case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
11215         case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
11216         case bpf_ctx_range(struct sk_reuseport_md, len):
11217                 bpf_ctx_record_field_size(info, size_default);
11218                 return bpf_ctx_narrow_access_ok(off, size, size_default);
11219
11220         default:
11221                 return false;
11222         }
11223 }
11224
11225 #define SK_REUSEPORT_LOAD_FIELD(F) ({                                   \
11226         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
11227                               si->dst_reg, si->src_reg,                 \
11228                               bpf_target_off(struct sk_reuseport_kern, F, \
11229                                              sizeof_field(struct sk_reuseport_kern, F), \
11230                                              target_size));             \
11231         })
11232
11233 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD)                          \
11234         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
11235                                     struct sk_buff,                     \
11236                                     skb,                                \
11237                                     SKB_FIELD)
11238
11239 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD)                            \
11240         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
11241                                     struct sock,                        \
11242                                     sk,                                 \
11243                                     SK_FIELD)
11244
11245 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
11246                                            const struct bpf_insn *si,
11247                                            struct bpf_insn *insn_buf,
11248                                            struct bpf_prog *prog,
11249                                            u32 *target_size)
11250 {
11251         struct bpf_insn *insn = insn_buf;
11252
11253         switch (si->off) {
11254         case offsetof(struct sk_reuseport_md, data):
11255                 SK_REUSEPORT_LOAD_SKB_FIELD(data);
11256                 break;
11257
11258         case offsetof(struct sk_reuseport_md, len):
11259                 SK_REUSEPORT_LOAD_SKB_FIELD(len);
11260                 break;
11261
11262         case offsetof(struct sk_reuseport_md, eth_protocol):
11263                 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
11264                 break;
11265
11266         case offsetof(struct sk_reuseport_md, ip_protocol):
11267                 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
11268                 break;
11269
11270         case offsetof(struct sk_reuseport_md, data_end):
11271                 SK_REUSEPORT_LOAD_FIELD(data_end);
11272                 break;
11273
11274         case offsetof(struct sk_reuseport_md, hash):
11275                 SK_REUSEPORT_LOAD_FIELD(hash);
11276                 break;
11277
11278         case offsetof(struct sk_reuseport_md, bind_inany):
11279                 SK_REUSEPORT_LOAD_FIELD(bind_inany);
11280                 break;
11281
11282         case offsetof(struct sk_reuseport_md, sk):
11283                 SK_REUSEPORT_LOAD_FIELD(sk);
11284                 break;
11285
11286         case offsetof(struct sk_reuseport_md, migrating_sk):
11287                 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
11288                 break;
11289         }
11290
11291         return insn - insn_buf;
11292 }
11293
11294 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
11295         .get_func_proto         = sk_reuseport_func_proto,
11296         .is_valid_access        = sk_reuseport_is_valid_access,
11297         .convert_ctx_access     = sk_reuseport_convert_ctx_access,
11298 };
11299
11300 const struct bpf_prog_ops sk_reuseport_prog_ops = {
11301 };
11302
11303 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
11304 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
11305
11306 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
11307            struct sock *, sk, u64, flags)
11308 {
11309         if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
11310                                BPF_SK_LOOKUP_F_NO_REUSEPORT)))
11311                 return -EINVAL;
11312         if (unlikely(sk && sk_is_refcounted(sk)))
11313                 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
11314         if (unlikely(sk && sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN))
11315                 return -ESOCKTNOSUPPORT; /* only accept TCP socket in LISTEN */
11316         if (unlikely(sk && sk_is_udp(sk) && sk->sk_state != TCP_CLOSE))
11317                 return -ESOCKTNOSUPPORT; /* only accept UDP socket in CLOSE */
11318
11319         /* Check if socket is suitable for packet L3/L4 protocol */
11320         if (sk && sk->sk_protocol != ctx->protocol)
11321                 return -EPROTOTYPE;
11322         if (sk && sk->sk_family != ctx->family &&
11323             (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
11324                 return -EAFNOSUPPORT;
11325
11326         if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
11327                 return -EEXIST;
11328
11329         /* Select socket as lookup result */
11330         ctx->selected_sk = sk;
11331         ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
11332         return 0;
11333 }
11334
11335 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
11336         .func           = bpf_sk_lookup_assign,
11337         .gpl_only       = false,
11338         .ret_type       = RET_INTEGER,
11339         .arg1_type      = ARG_PTR_TO_CTX,
11340         .arg2_type      = ARG_PTR_TO_SOCKET_OR_NULL,
11341         .arg3_type      = ARG_ANYTHING,
11342 };
11343
11344 static const struct bpf_func_proto *
11345 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
11346 {
11347         switch (func_id) {
11348         case BPF_FUNC_perf_event_output:
11349                 return &bpf_event_output_data_proto;
11350         case BPF_FUNC_sk_assign:
11351                 return &bpf_sk_lookup_assign_proto;
11352         case BPF_FUNC_sk_release:
11353                 return &bpf_sk_release_proto;
11354         default:
11355                 return bpf_sk_base_func_proto(func_id);
11356         }
11357 }
11358
11359 static bool sk_lookup_is_valid_access(int off, int size,
11360                                       enum bpf_access_type type,
11361                                       const struct bpf_prog *prog,
11362                                       struct bpf_insn_access_aux *info)
11363 {
11364         if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
11365                 return false;
11366         if (off % size != 0)
11367                 return false;
11368         if (type != BPF_READ)
11369                 return false;
11370
11371         switch (off) {
11372         case offsetof(struct bpf_sk_lookup, sk):
11373                 info->reg_type = PTR_TO_SOCKET_OR_NULL;
11374                 return size == sizeof(__u64);
11375
11376         case bpf_ctx_range(struct bpf_sk_lookup, family):
11377         case bpf_ctx_range(struct bpf_sk_lookup, protocol):
11378         case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
11379         case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
11380         case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
11381         case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
11382         case bpf_ctx_range(struct bpf_sk_lookup, local_port):
11383         case bpf_ctx_range(struct bpf_sk_lookup, ingress_ifindex):
11384                 bpf_ctx_record_field_size(info, sizeof(__u32));
11385                 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
11386
11387         case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
11388                 /* Allow 4-byte access to 2-byte field for backward compatibility */
11389                 if (size == sizeof(__u32))
11390                         return true;
11391                 bpf_ctx_record_field_size(info, sizeof(__be16));
11392                 return bpf_ctx_narrow_access_ok(off, size, sizeof(__be16));
11393
11394         case offsetofend(struct bpf_sk_lookup, remote_port) ...
11395              offsetof(struct bpf_sk_lookup, local_ip4) - 1:
11396                 /* Allow access to zero padding for backward compatibility */
11397                 bpf_ctx_record_field_size(info, sizeof(__u16));
11398                 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u16));
11399
11400         default:
11401                 return false;
11402         }
11403 }
11404
11405 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
11406                                         const struct bpf_insn *si,
11407                                         struct bpf_insn *insn_buf,
11408                                         struct bpf_prog *prog,
11409                                         u32 *target_size)
11410 {
11411         struct bpf_insn *insn = insn_buf;
11412
11413         switch (si->off) {
11414         case offsetof(struct bpf_sk_lookup, sk):
11415                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11416                                       offsetof(struct bpf_sk_lookup_kern, selected_sk));
11417                 break;
11418
11419         case offsetof(struct bpf_sk_lookup, family):
11420                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11421                                       bpf_target_off(struct bpf_sk_lookup_kern,
11422                                                      family, 2, target_size));
11423                 break;
11424
11425         case offsetof(struct bpf_sk_lookup, protocol):
11426                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11427                                       bpf_target_off(struct bpf_sk_lookup_kern,
11428                                                      protocol, 2, target_size));
11429                 break;
11430
11431         case offsetof(struct bpf_sk_lookup, remote_ip4):
11432                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11433                                       bpf_target_off(struct bpf_sk_lookup_kern,
11434                                                      v4.saddr, 4, target_size));
11435                 break;
11436
11437         case offsetof(struct bpf_sk_lookup, local_ip4):
11438                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11439                                       bpf_target_off(struct bpf_sk_lookup_kern,
11440                                                      v4.daddr, 4, target_size));
11441                 break;
11442
11443         case bpf_ctx_range_till(struct bpf_sk_lookup,
11444                                 remote_ip6[0], remote_ip6[3]): {
11445 #if IS_ENABLED(CONFIG_IPV6)
11446                 int off = si->off;
11447
11448                 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
11449                 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11450                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11451                                       offsetof(struct bpf_sk_lookup_kern, v6.saddr));
11452                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11453                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11454 #else
11455                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11456 #endif
11457                 break;
11458         }
11459         case bpf_ctx_range_till(struct bpf_sk_lookup,
11460                                 local_ip6[0], local_ip6[3]): {
11461 #if IS_ENABLED(CONFIG_IPV6)
11462                 int off = si->off;
11463
11464                 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
11465                 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11466                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11467                                       offsetof(struct bpf_sk_lookup_kern, v6.daddr));
11468                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11469                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11470 #else
11471                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11472 #endif
11473                 break;
11474         }
11475         case offsetof(struct bpf_sk_lookup, remote_port):
11476                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11477                                       bpf_target_off(struct bpf_sk_lookup_kern,
11478                                                      sport, 2, target_size));
11479                 break;
11480
11481         case offsetofend(struct bpf_sk_lookup, remote_port):
11482                 *target_size = 2;
11483                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11484                 break;
11485
11486         case offsetof(struct bpf_sk_lookup, local_port):
11487                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11488                                       bpf_target_off(struct bpf_sk_lookup_kern,
11489                                                      dport, 2, target_size));
11490                 break;
11491
11492         case offsetof(struct bpf_sk_lookup, ingress_ifindex):
11493                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11494                                       bpf_target_off(struct bpf_sk_lookup_kern,
11495                                                      ingress_ifindex, 4, target_size));
11496                 break;
11497         }
11498
11499         return insn - insn_buf;
11500 }
11501
11502 const struct bpf_prog_ops sk_lookup_prog_ops = {
11503         .test_run = bpf_prog_test_run_sk_lookup,
11504 };
11505
11506 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
11507         .get_func_proto         = sk_lookup_func_proto,
11508         .is_valid_access        = sk_lookup_is_valid_access,
11509         .convert_ctx_access     = sk_lookup_convert_ctx_access,
11510 };
11511
11512 #endif /* CONFIG_INET */
11513
11514 DEFINE_BPF_DISPATCHER(xdp)
11515
11516 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
11517 {
11518         bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
11519 }
11520
11521 BTF_ID_LIST_GLOBAL(btf_sock_ids, MAX_BTF_SOCK_TYPE)
11522 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
11523 BTF_SOCK_TYPE_xxx
11524 #undef BTF_SOCK_TYPE
11525
11526 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
11527 {
11528         /* tcp6_sock type is not generated in dwarf and hence btf,
11529          * trigger an explicit type generation here.
11530          */
11531         BTF_TYPE_EMIT(struct tcp6_sock);
11532         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
11533             sk->sk_family == AF_INET6)
11534                 return (unsigned long)sk;
11535
11536         return (unsigned long)NULL;
11537 }
11538
11539 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
11540         .func                   = bpf_skc_to_tcp6_sock,
11541         .gpl_only               = false,
11542         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11543         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11544         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
11545 };
11546
11547 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
11548 {
11549         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
11550                 return (unsigned long)sk;
11551
11552         return (unsigned long)NULL;
11553 }
11554
11555 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
11556         .func                   = bpf_skc_to_tcp_sock,
11557         .gpl_only               = false,
11558         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11559         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11560         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
11561 };
11562
11563 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
11564 {
11565         /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
11566          * generated if CONFIG_INET=n. Trigger an explicit generation here.
11567          */
11568         BTF_TYPE_EMIT(struct inet_timewait_sock);
11569         BTF_TYPE_EMIT(struct tcp_timewait_sock);
11570
11571 #ifdef CONFIG_INET
11572         if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
11573                 return (unsigned long)sk;
11574 #endif
11575
11576 #if IS_BUILTIN(CONFIG_IPV6)
11577         if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
11578                 return (unsigned long)sk;
11579 #endif
11580
11581         return (unsigned long)NULL;
11582 }
11583
11584 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
11585         .func                   = bpf_skc_to_tcp_timewait_sock,
11586         .gpl_only               = false,
11587         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11588         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11589         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
11590 };
11591
11592 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
11593 {
11594 #ifdef CONFIG_INET
11595         if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11596                 return (unsigned long)sk;
11597 #endif
11598
11599 #if IS_BUILTIN(CONFIG_IPV6)
11600         if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11601                 return (unsigned long)sk;
11602 #endif
11603
11604         return (unsigned long)NULL;
11605 }
11606
11607 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
11608         .func                   = bpf_skc_to_tcp_request_sock,
11609         .gpl_only               = false,
11610         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11611         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11612         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
11613 };
11614
11615 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
11616 {
11617         /* udp6_sock type is not generated in dwarf and hence btf,
11618          * trigger an explicit type generation here.
11619          */
11620         BTF_TYPE_EMIT(struct udp6_sock);
11621         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
11622             sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
11623                 return (unsigned long)sk;
11624
11625         return (unsigned long)NULL;
11626 }
11627
11628 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
11629         .func                   = bpf_skc_to_udp6_sock,
11630         .gpl_only               = false,
11631         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11632         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11633         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
11634 };
11635
11636 BPF_CALL_1(bpf_skc_to_unix_sock, struct sock *, sk)
11637 {
11638         /* unix_sock type is not generated in dwarf and hence btf,
11639          * trigger an explicit type generation here.
11640          */
11641         BTF_TYPE_EMIT(struct unix_sock);
11642         if (sk && sk_fullsock(sk) && sk->sk_family == AF_UNIX)
11643                 return (unsigned long)sk;
11644
11645         return (unsigned long)NULL;
11646 }
11647
11648 const struct bpf_func_proto bpf_skc_to_unix_sock_proto = {
11649         .func                   = bpf_skc_to_unix_sock,
11650         .gpl_only               = false,
11651         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11652         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11653         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_UNIX],
11654 };
11655
11656 BPF_CALL_1(bpf_skc_to_mptcp_sock, struct sock *, sk)
11657 {
11658         BTF_TYPE_EMIT(struct mptcp_sock);
11659         return (unsigned long)bpf_mptcp_sock_from_subflow(sk);
11660 }
11661
11662 const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto = {
11663         .func           = bpf_skc_to_mptcp_sock,
11664         .gpl_only       = false,
11665         .ret_type       = RET_PTR_TO_BTF_ID_OR_NULL,
11666         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
11667         .ret_btf_id     = &btf_sock_ids[BTF_SOCK_TYPE_MPTCP],
11668 };
11669
11670 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
11671 {
11672         return (unsigned long)sock_from_file(file);
11673 }
11674
11675 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
11676 BTF_ID(struct, socket)
11677 BTF_ID(struct, file)
11678
11679 const struct bpf_func_proto bpf_sock_from_file_proto = {
11680         .func           = bpf_sock_from_file,
11681         .gpl_only       = false,
11682         .ret_type       = RET_PTR_TO_BTF_ID_OR_NULL,
11683         .ret_btf_id     = &bpf_sock_from_file_btf_ids[0],
11684         .arg1_type      = ARG_PTR_TO_BTF_ID,
11685         .arg1_btf_id    = &bpf_sock_from_file_btf_ids[1],
11686 };
11687
11688 static const struct bpf_func_proto *
11689 bpf_sk_base_func_proto(enum bpf_func_id func_id)
11690 {
11691         const struct bpf_func_proto *func;
11692
11693         switch (func_id) {
11694         case BPF_FUNC_skc_to_tcp6_sock:
11695                 func = &bpf_skc_to_tcp6_sock_proto;
11696                 break;
11697         case BPF_FUNC_skc_to_tcp_sock:
11698                 func = &bpf_skc_to_tcp_sock_proto;
11699                 break;
11700         case BPF_FUNC_skc_to_tcp_timewait_sock:
11701                 func = &bpf_skc_to_tcp_timewait_sock_proto;
11702                 break;
11703         case BPF_FUNC_skc_to_tcp_request_sock:
11704                 func = &bpf_skc_to_tcp_request_sock_proto;
11705                 break;
11706         case BPF_FUNC_skc_to_udp6_sock:
11707                 func = &bpf_skc_to_udp6_sock_proto;
11708                 break;
11709         case BPF_FUNC_skc_to_unix_sock:
11710                 func = &bpf_skc_to_unix_sock_proto;
11711                 break;
11712         case BPF_FUNC_skc_to_mptcp_sock:
11713                 func = &bpf_skc_to_mptcp_sock_proto;
11714                 break;
11715         case BPF_FUNC_ktime_get_coarse_ns:
11716                 return &bpf_ktime_get_coarse_ns_proto;
11717         default:
11718                 return bpf_base_func_proto(func_id);
11719         }
11720
11721         if (!perfmon_capable())
11722                 return NULL;
11723
11724         return func;
11725 }
11726
11727 __diag_push();
11728 __diag_ignore_all("-Wmissing-prototypes",
11729                   "Global functions as their definitions will be in vmlinux BTF");
11730 __bpf_kfunc int bpf_dynptr_from_skb(struct sk_buff *skb, u64 flags,
11731                                     struct bpf_dynptr_kern *ptr__uninit)
11732 {
11733         if (flags) {
11734                 bpf_dynptr_set_null(ptr__uninit);
11735                 return -EINVAL;
11736         }
11737
11738         bpf_dynptr_init(ptr__uninit, skb, BPF_DYNPTR_TYPE_SKB, 0, skb->len);
11739
11740         return 0;
11741 }
11742
11743 __bpf_kfunc int bpf_dynptr_from_xdp(struct xdp_buff *xdp, u64 flags,
11744                                     struct bpf_dynptr_kern *ptr__uninit)
11745 {
11746         if (flags) {
11747                 bpf_dynptr_set_null(ptr__uninit);
11748                 return -EINVAL;
11749         }
11750
11751         bpf_dynptr_init(ptr__uninit, xdp, BPF_DYNPTR_TYPE_XDP, 0, xdp_get_buff_len(xdp));
11752
11753         return 0;
11754 }
11755 __diag_pop();
11756
11757 int bpf_dynptr_from_skb_rdonly(struct sk_buff *skb, u64 flags,
11758                                struct bpf_dynptr_kern *ptr__uninit)
11759 {
11760         int err;
11761
11762         err = bpf_dynptr_from_skb(skb, flags, ptr__uninit);
11763         if (err)
11764                 return err;
11765
11766         bpf_dynptr_set_rdonly(ptr__uninit);
11767
11768         return 0;
11769 }
11770
11771 BTF_SET8_START(bpf_kfunc_check_set_skb)
11772 BTF_ID_FLAGS(func, bpf_dynptr_from_skb)
11773 BTF_SET8_END(bpf_kfunc_check_set_skb)
11774
11775 BTF_SET8_START(bpf_kfunc_check_set_xdp)
11776 BTF_ID_FLAGS(func, bpf_dynptr_from_xdp)
11777 BTF_SET8_END(bpf_kfunc_check_set_xdp)
11778
11779 static const struct btf_kfunc_id_set bpf_kfunc_set_skb = {
11780         .owner = THIS_MODULE,
11781         .set = &bpf_kfunc_check_set_skb,
11782 };
11783
11784 static const struct btf_kfunc_id_set bpf_kfunc_set_xdp = {
11785         .owner = THIS_MODULE,
11786         .set = &bpf_kfunc_check_set_xdp,
11787 };
11788
11789 static int __init bpf_kfunc_init(void)
11790 {
11791         int ret;
11792
11793         ret = register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_kfunc_set_skb);
11794         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_ACT, &bpf_kfunc_set_skb);
11795         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SK_SKB, &bpf_kfunc_set_skb);
11796         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SOCKET_FILTER, &bpf_kfunc_set_skb);
11797         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_CGROUP_SKB, &bpf_kfunc_set_skb);
11798         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_OUT, &bpf_kfunc_set_skb);
11799         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_IN, &bpf_kfunc_set_skb);
11800         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_XMIT, &bpf_kfunc_set_skb);
11801         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_SEG6LOCAL, &bpf_kfunc_set_skb);
11802         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_NETFILTER, &bpf_kfunc_set_skb);
11803         return ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_XDP, &bpf_kfunc_set_xdp);
11804 }
11805 late_initcall(bpf_kfunc_init);
11806
11807 /* Disables missing prototype warnings */
11808 __diag_push();
11809 __diag_ignore_all("-Wmissing-prototypes",
11810                   "Global functions as their definitions will be in vmlinux BTF");
11811
11812 /* bpf_sock_destroy: Destroy the given socket with ECONNABORTED error code.
11813  *
11814  * The function expects a non-NULL pointer to a socket, and invokes the
11815  * protocol specific socket destroy handlers.
11816  *
11817  * The helper can only be called from BPF contexts that have acquired the socket
11818  * locks.
11819  *
11820  * Parameters:
11821  * @sock: Pointer to socket to be destroyed
11822  *
11823  * Return:
11824  * On error, may return EPROTONOSUPPORT, EINVAL.
11825  * EPROTONOSUPPORT if protocol specific destroy handler is not supported.
11826  * 0 otherwise
11827  */
11828 __bpf_kfunc int bpf_sock_destroy(struct sock_common *sock)
11829 {
11830         struct sock *sk = (struct sock *)sock;
11831
11832         /* The locking semantics that allow for synchronous execution of the
11833          * destroy handlers are only supported for TCP and UDP.
11834          * Supporting protocols will need to acquire sock lock in the BPF context
11835          * prior to invoking this kfunc.
11836          */
11837         if (!sk->sk_prot->diag_destroy || (sk->sk_protocol != IPPROTO_TCP &&
11838                                            sk->sk_protocol != IPPROTO_UDP))
11839                 return -EOPNOTSUPP;
11840
11841         return sk->sk_prot->diag_destroy(sk, ECONNABORTED);
11842 }
11843
11844 __diag_pop()
11845
11846 BTF_SET8_START(bpf_sk_iter_kfunc_ids)
11847 BTF_ID_FLAGS(func, bpf_sock_destroy, KF_TRUSTED_ARGS)
11848 BTF_SET8_END(bpf_sk_iter_kfunc_ids)
11849
11850 static int tracing_iter_filter(const struct bpf_prog *prog, u32 kfunc_id)
11851 {
11852         if (btf_id_set8_contains(&bpf_sk_iter_kfunc_ids, kfunc_id) &&
11853             prog->expected_attach_type != BPF_TRACE_ITER)
11854                 return -EACCES;
11855         return 0;
11856 }
11857
11858 static const struct btf_kfunc_id_set bpf_sk_iter_kfunc_set = {
11859         .owner = THIS_MODULE,
11860         .set   = &bpf_sk_iter_kfunc_ids,
11861         .filter = tracing_iter_filter,
11862 };
11863
11864 static int init_subsystem(void)
11865 {
11866         return register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_sk_iter_kfunc_set);
11867 }
11868 late_initcall(init_subsystem);