1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Linux Socket Filter - Kernel level socket filtering
5 * Based on the design of the Berkeley Packet Filter. The new
6 * internal format has been designed by PLUMgrid:
8 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
12 * Jay Schulist <jschlst@samba.org>
13 * Alexei Starovoitov <ast@plumgrid.com>
14 * Daniel Borkmann <dborkman@redhat.com>
16 * Andi Kleen - Fix a few bad bugs and races.
17 * Kris Katterjohn - Added many additional checks in bpf_check_classic()
20 #include <linux/atomic.h>
21 #include <linux/bpf_verifier.h>
22 #include <linux/module.h>
23 #include <linux/types.h>
25 #include <linux/fcntl.h>
26 #include <linux/socket.h>
27 #include <linux/sock_diag.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>
36 #include <net/protocol.h>
37 #include <net/netlink.h>
38 #include <linux/skbuff.h>
39 #include <linux/skmsg.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>
56 #include <net/sock_reuseport.h>
57 #include <net/busy_poll.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>
71 #include <net/net_namespace.h>
72 #include <linux/seg6_local.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>
82 #include <net/mptcp.h>
83 #include <net/netfilter/nf_conntrack_bpf.h>
85 static const struct bpf_func_proto *
86 bpf_sk_base_func_proto(enum bpf_func_id func_id);
88 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len)
90 if (in_compat_syscall()) {
91 struct compat_sock_fprog f32;
93 if (len != sizeof(f32))
95 if (copy_from_sockptr(&f32, src, sizeof(f32)))
97 memset(dst, 0, sizeof(*dst));
99 dst->filter = compat_ptr(f32.filter);
101 if (len != sizeof(*dst))
103 if (copy_from_sockptr(dst, src, sizeof(*dst)))
109 EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user);
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
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.
124 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
127 struct sk_filter *filter;
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
134 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
135 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
138 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
142 err = security_sock_rcv_skb(sk, skb);
147 filter = rcu_dereference(sk->sk_filter);
149 struct sock *save_sk = skb->sk;
150 unsigned int pkt_len;
153 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
155 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
161 EXPORT_SYMBOL(sk_filter_trim_cap);
163 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
165 return skb_get_poff(skb);
168 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
172 if (skb_is_nonlinear(skb))
175 if (skb->len < sizeof(struct nlattr))
178 if (a > skb->len - sizeof(struct nlattr))
181 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
183 return (void *) nla - (void *) skb->data;
188 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
192 if (skb_is_nonlinear(skb))
195 if (skb->len < sizeof(struct nlattr))
198 if (a > skb->len - sizeof(struct nlattr))
201 nla = (struct nlattr *) &skb->data[a];
202 if (nla->nla_len > skb->len - a)
205 nla = nla_find_nested(nla, x);
207 return (void *) nla - (void *) skb->data;
212 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
213 data, int, headlen, int, offset)
216 const int len = sizeof(tmp);
219 if (headlen - offset >= len)
220 return *(u8 *)(data + offset);
221 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
224 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
232 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
235 return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
239 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
240 data, int, headlen, int, offset)
243 const int len = sizeof(tmp);
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);
251 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
253 return get_unaligned_be16(ptr);
259 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
262 return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
266 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
267 data, int, headlen, int, offset)
270 const int len = sizeof(tmp);
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);
278 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
280 return get_unaligned_be32(ptr);
286 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
289 return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
293 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
294 struct bpf_insn *insn_buf)
296 struct bpf_insn *insn = insn_buf;
300 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4);
302 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
303 offsetof(struct sk_buff, mark));
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);
315 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2);
317 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
318 offsetof(struct sk_buff, queue_mapping));
321 case SKF_AD_VLAN_TAG:
322 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2);
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));
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);
337 return insn - insn_buf;
340 static bool convert_bpf_extensions(struct sock_filter *fp,
341 struct bpf_insn **insnp)
343 struct bpf_insn *insn = *insnp;
347 case SKF_AD_OFF + SKF_AD_PROTOCOL:
348 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2);
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);
357 case SKF_AD_OFF + SKF_AD_PKTTYPE:
358 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
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);
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));
377 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
378 offsetof(struct net_device, type));
381 case SKF_AD_OFF + SKF_AD_MARK:
382 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
386 case SKF_AD_OFF + SKF_AD_RXHASH:
387 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4);
389 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
390 offsetof(struct sk_buff, hash));
393 case SKF_AD_OFF + SKF_AD_QUEUE:
394 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
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);
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);
410 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
411 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2);
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);
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:
426 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
428 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
430 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
431 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
433 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
434 *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
436 case SKF_AD_OFF + SKF_AD_NLATTR:
437 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
439 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
440 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
442 case SKF_AD_OFF + SKF_AD_CPU:
443 *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
445 case SKF_AD_OFF + SKF_AD_RANDOM:
446 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
447 bpf_user_rnd_init_once();
452 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
454 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
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.
462 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
470 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
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;
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;
488 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
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));
494 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
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,
503 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
504 *insn++ = BPF_JMP_A(8);
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);
511 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
513 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
515 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
518 switch (BPF_SIZE(fp->code)) {
520 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
523 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
526 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
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();
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
548 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
549 * style extended BPF (eBPF).
550 * Conversion workflow:
552 * 1) First pass for calculating the new program length:
553 * bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
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)
559 static int bpf_convert_filter(struct sock_filter *prog, int len,
560 struct bpf_prog *new_prog, int *new_len,
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;
569 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
570 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
572 if (len <= 0 || len > BPF_MAXINSNS)
576 first_insn = new_prog->insnsi;
577 addrs = kcalloc(len, sizeof(*addrs),
578 GFP_KERNEL | __GFP_NOWARN);
584 new_insn = first_insn;
587 /* Classic BPF related prologue emission. */
589 /* Classic BPF expects A and X to be reset first. These need
590 * to be guaranteed to be the first two instructions.
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);
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.
599 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
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.
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);
619 for (i = 0; i < len; fp++, i++) {
620 struct bpf_insn tmp_insns[32] = { };
621 struct bpf_insn *insn = tmp_insns;
624 addrs[i] = new_insn - first_insn;
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.
659 if (BPF_CLASS(fp->code) == BPF_LD &&
660 BPF_MODE(fp->code) == BPF_ABS &&
661 convert_bpf_extensions(fp, &insn))
663 if (BPF_CLASS(fp->code) == BPF_LD &&
664 convert_bpf_ld_abs(fp, &insn)) {
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.
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();
680 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
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.
689 #define BPF_EMIT_JMP \
691 const s32 off_min = S16_MIN, off_max = S16_MAX; \
694 if (target >= len || target < 0) \
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) \
705 case BPF_JMP | BPF_JA:
706 target = i + fp->k + 1;
707 insn->code = fp->code;
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
724 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
726 insn->dst_reg = BPF_REG_A;
727 insn->src_reg = BPF_REG_TMP;
730 insn->dst_reg = BPF_REG_A;
732 bpf_src = BPF_SRC(fp->code);
733 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
736 /* Common case where 'jump_false' is next insn. */
738 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
739 target = i + fp->jt + 1;
744 /* Convert some jumps when 'jump_true' is next insn. */
746 switch (BPF_OP(fp->code)) {
748 insn->code = BPF_JMP | BPF_JNE | bpf_src;
751 insn->code = BPF_JMP | BPF_JLE | bpf_src;
754 insn->code = BPF_JMP | BPF_JLT | bpf_src;
760 target = i + fp->jf + 1;
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;
771 insn->code = BPF_JMP | BPF_JA;
772 target = i + fp->jf + 1;
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,
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);
791 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
793 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
795 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
797 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
799 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
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.
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,
810 *insn = BPF_EXIT_INSN();
813 /* Store to stack. */
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,
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
824 if (new_prog && new_prog->aux->stack_depth < stack_off)
825 new_prog->aux->stack_depth = stack_off;
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,
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);
845 case BPF_MISC | BPF_TAX:
846 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
850 case BPF_MISC | BPF_TXA:
851 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
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));
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);
868 /* Unknown instruction. */
875 memcpy(new_insn, tmp_insns,
876 sizeof(*insn) * (insn - tmp_insns));
877 new_insn += insn - tmp_insns;
881 /* Only calculating new length. */
882 *new_len = new_insn - first_insn;
884 *new_len += 4; /* Prologue bits. */
889 if (new_flen != new_insn - first_insn) {
890 new_flen = new_insn - first_insn;
897 BUG_ON(*new_len != new_flen);
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.
911 static int check_load_and_stores(const struct sock_filter *filter, int flen)
913 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
916 BUILD_BUG_ON(BPF_MEMWORDS > 16);
918 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
922 memset(masks, 0xff, flen * sizeof(*masks));
924 for (pc = 0; pc < flen; pc++) {
925 memvalid &= masks[pc];
927 switch (filter[pc].code) {
930 memvalid |= (1 << filter[pc].k);
932 case BPF_LD | BPF_MEM:
933 case BPF_LDX | BPF_MEM:
934 if (!(memvalid & (1 << filter[pc].k))) {
939 case BPF_JMP | BPF_JA:
940 /* A jump must set masks on target */
941 masks[pc + 1 + filter[pc].k] &= memvalid;
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;
964 static bool chk_code_allowed(u16 code_to_probe)
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 */
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,
1024 if (code_to_probe >= ARRAY_SIZE(codes))
1027 return codes[code_to_probe];
1030 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1035 if (flen == 0 || flen > BPF_MAXINSNS)
1042 * bpf_check_classic - verify socket filter code
1043 * @filter: filter to verify
1044 * @flen: length of filter
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.
1051 * All jumps are forward as they are not signed.
1053 * Returns 0 if the rule set is legal or -EINVAL if not.
1055 static int bpf_check_classic(const struct sock_filter *filter,
1061 /* Check the filter code now */
1062 for (pc = 0; pc < flen; pc++) {
1063 const struct sock_filter *ftest = &filter[pc];
1065 /* May we actually operate on this code? */
1066 if (!chk_code_allowed(ftest->code))
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 */
1077 case BPF_ALU | BPF_LSH | BPF_K:
1078 case BPF_ALU | BPF_RSH | BPF_K:
1082 case BPF_LD | BPF_MEM:
1083 case BPF_LDX | BPF_MEM:
1086 /* Check for invalid memory addresses */
1087 if (ftest->k >= BPF_MEMWORDS)
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)
1095 if (ftest->k >= (unsigned int)(flen - pc - 1))
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)
1111 case BPF_LD | BPF_W | BPF_ABS:
1112 case BPF_LD | BPF_H | BPF_ABS:
1113 case BPF_LD | BPF_B | BPF_ABS:
1115 if (bpf_anc_helper(ftest) & BPF_ANC)
1117 /* Ancillary operation unknown or unsupported */
1118 if (anc_found == false && ftest->k >= SKF_AD_OFF)
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);
1133 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1134 const struct sock_fprog *fprog)
1136 unsigned int fsize = bpf_classic_proglen(fprog);
1137 struct sock_fprog_kern *fkprog;
1139 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1143 fkprog = fp->orig_prog;
1144 fkprog->len = fprog->len;
1146 fkprog->filter = kmemdup(fp->insns, fsize,
1147 GFP_KERNEL | __GFP_NOWARN);
1148 if (!fkprog->filter) {
1149 kfree(fp->orig_prog);
1156 static void bpf_release_orig_filter(struct bpf_prog *fp)
1158 struct sock_fprog_kern *fprog = fp->orig_prog;
1161 kfree(fprog->filter);
1166 static void __bpf_prog_release(struct bpf_prog *prog)
1168 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1171 bpf_release_orig_filter(prog);
1172 bpf_prog_free(prog);
1176 static void __sk_filter_release(struct sk_filter *fp)
1178 __bpf_prog_release(fp->prog);
1183 * sk_filter_release_rcu - Release a socket filter by rcu_head
1184 * @rcu: rcu_head that contains the sk_filter to free
1186 static void sk_filter_release_rcu(struct rcu_head *rcu)
1188 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1190 __sk_filter_release(fp);
1194 * sk_filter_release - release a socket filter
1195 * @fp: filter to remove
1197 * Remove a filter from a socket and release its resources.
1199 static void sk_filter_release(struct sk_filter *fp)
1201 if (refcount_dec_and_test(&fp->refcnt))
1202 call_rcu(&fp->rcu, sk_filter_release_rcu);
1205 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1207 u32 filter_size = bpf_prog_size(fp->prog->len);
1209 atomic_sub(filter_size, &sk->sk_omem_alloc);
1210 sk_filter_release(fp);
1213 /* try to charge the socket memory if there is space available
1214 * return true on success
1216 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1218 u32 filter_size = bpf_prog_size(fp->prog->len);
1219 int optmem_max = READ_ONCE(sysctl_optmem_max);
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);
1230 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1232 if (!refcount_inc_not_zero(&fp->refcnt))
1235 if (!__sk_filter_charge(sk, fp)) {
1236 sk_filter_release(fp);
1242 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
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;
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.
1253 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1254 sizeof(struct bpf_insn));
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.
1260 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1261 GFP_KERNEL | __GFP_NOWARN);
1267 /* 1st pass: calculate the new program length. */
1268 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1273 /* Expand fp for appending the new filter representation. */
1275 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1277 /* The old_fp is still around in case we couldn't
1278 * allocate new memory, so uncharge on that one.
1287 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1288 err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
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
1298 fp = bpf_prog_select_runtime(fp, &err);
1308 __bpf_prog_release(fp);
1309 return ERR_PTR(err);
1312 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1313 bpf_aux_classic_check_t trans)
1317 fp->bpf_func = NULL;
1320 err = bpf_check_classic(fp->insns, fp->len);
1322 __bpf_prog_release(fp);
1323 return ERR_PTR(err);
1326 /* There might be additional checks and transformations
1327 * needed on classic filters, f.e. in case of seccomp.
1330 err = trans(fp->insns, fp->len);
1332 __bpf_prog_release(fp);
1333 return ERR_PTR(err);
1337 /* Probe if we can JIT compile the filter and if so, do
1338 * the compilation of the filter.
1340 bpf_jit_compile(fp);
1342 /* JIT compiler couldn't process this filter, so do the eBPF translation
1343 * for the optimized interpreter.
1346 fp = bpf_migrate_filter(fp);
1352 * bpf_prog_create - create an unattached filter
1353 * @pfp: the unattached filter that is created
1354 * @fprog: the filter program
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.
1361 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1363 unsigned int fsize = bpf_classic_proglen(fprog);
1364 struct bpf_prog *fp;
1366 /* Make sure new filter is there and in the right amounts. */
1367 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1370 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1374 memcpy(fp->insns, fprog->filter, fsize);
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.
1381 fp->orig_prog = NULL;
1383 /* bpf_prepare_filter() already takes care of freeing
1384 * memory in case something goes wrong.
1386 fp = bpf_prepare_filter(fp, NULL);
1393 EXPORT_SYMBOL_GPL(bpf_prog_create);
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
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.
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)
1409 unsigned int fsize = bpf_classic_proglen(fprog);
1410 struct bpf_prog *fp;
1413 /* Make sure new filter is there and in the right amounts. */
1414 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1417 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1421 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1422 __bpf_prog_free(fp);
1426 fp->len = fprog->len;
1427 fp->orig_prog = NULL;
1430 err = bpf_prog_store_orig_filter(fp, fprog);
1432 __bpf_prog_free(fp);
1437 /* bpf_prepare_filter() already takes care of freeing
1438 * memory in case something goes wrong.
1440 fp = bpf_prepare_filter(fp, trans);
1447 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1449 void bpf_prog_destroy(struct bpf_prog *fp)
1451 __bpf_prog_release(fp);
1453 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1455 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1457 struct sk_filter *fp, *old_fp;
1459 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1465 if (!__sk_filter_charge(sk, fp)) {
1469 refcount_set(&fp->refcnt, 1);
1471 old_fp = rcu_dereference_protected(sk->sk_filter,
1472 lockdep_sock_is_held(sk));
1473 rcu_assign_pointer(sk->sk_filter, fp);
1476 sk_filter_uncharge(sk, old_fp);
1482 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1484 unsigned int fsize = bpf_classic_proglen(fprog);
1485 struct bpf_prog *prog;
1488 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1489 return ERR_PTR(-EPERM);
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);
1495 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1497 return ERR_PTR(-ENOMEM);
1499 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1500 __bpf_prog_free(prog);
1501 return ERR_PTR(-EFAULT);
1504 prog->len = fprog->len;
1506 err = bpf_prog_store_orig_filter(prog, fprog);
1508 __bpf_prog_free(prog);
1509 return ERR_PTR(-ENOMEM);
1512 /* bpf_prepare_filter() already takes care of freeing
1513 * memory in case something goes wrong.
1515 return bpf_prepare_filter(prog, NULL);
1519 * sk_attach_filter - attach a socket filter
1520 * @fprog: the filter program
1521 * @sk: the socket to use
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.
1528 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1530 struct bpf_prog *prog = __get_filter(fprog, sk);
1534 return PTR_ERR(prog);
1536 err = __sk_attach_prog(prog, sk);
1538 __bpf_prog_release(prog);
1544 EXPORT_SYMBOL_GPL(sk_attach_filter);
1546 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1548 struct bpf_prog *prog = __get_filter(fprog, sk);
1552 return PTR_ERR(prog);
1554 if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max))
1557 err = reuseport_attach_prog(sk, prog);
1560 __bpf_prog_release(prog);
1565 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1567 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1568 return ERR_PTR(-EPERM);
1570 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1573 int sk_attach_bpf(u32 ufd, struct sock *sk)
1575 struct bpf_prog *prog = __get_bpf(ufd, sk);
1579 return PTR_ERR(prog);
1581 err = __sk_attach_prog(prog, sk);
1590 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1592 struct bpf_prog *prog;
1595 if (sock_flag(sk, SOCK_FILTER_LOCKED))
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);
1602 return PTR_ERR(prog);
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.
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)) {
1620 /* BPF_PROG_TYPE_SOCKET_FILTER */
1621 if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max)) {
1627 err = reuseport_attach_prog(sk, prog);
1635 void sk_reuseport_prog_free(struct bpf_prog *prog)
1640 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1643 bpf_prog_destroy(prog);
1646 struct bpf_scratchpad {
1648 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1649 u8 buff[MAX_BPF_STACK];
1653 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1655 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1656 unsigned int write_len)
1658 return skb_ensure_writable(skb, write_len);
1661 static inline int bpf_try_make_writable(struct sk_buff *skb,
1662 unsigned int write_len)
1664 int err = __bpf_try_make_writable(skb, write_len);
1666 bpf_compute_data_pointers(skb);
1670 static int bpf_try_make_head_writable(struct sk_buff *skb)
1672 return bpf_try_make_writable(skb, skb_headlen(skb));
1675 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1677 if (skb_at_tc_ingress(skb))
1678 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1681 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1683 if (skb_at_tc_ingress(skb))
1684 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1687 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1688 const void *, from, u32, len, u64, flags)
1692 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1694 if (unlikely(offset > INT_MAX))
1696 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1699 ptr = skb->data + offset;
1700 if (flags & BPF_F_RECOMPUTE_CSUM)
1701 __skb_postpull_rcsum(skb, ptr, len, offset);
1703 memcpy(ptr, from, len);
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);
1713 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1714 .func = bpf_skb_store_bytes,
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,
1724 int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, const void *from,
1727 return ____bpf_skb_store_bytes(skb, offset, from, len, flags);
1730 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1731 void *, to, u32, len)
1735 if (unlikely(offset > INT_MAX))
1738 ptr = skb_header_pointer(skb, offset, len, to);
1742 memcpy(to, ptr, len);
1750 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1751 .func = bpf_skb_load_bytes,
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,
1760 int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, void *to, u32 len)
1762 return ____bpf_skb_load_bytes(skb, offset, to, len);
1765 BPF_CALL_4(bpf_flow_dissector_load_bytes,
1766 const struct bpf_flow_dissector *, ctx, u32, offset,
1767 void *, to, u32, len)
1771 if (unlikely(offset > 0xffff))
1774 if (unlikely(!ctx->skb))
1777 ptr = skb_header_pointer(ctx->skb, offset, len, to);
1781 memcpy(to, ptr, len);
1789 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1790 .func = bpf_flow_dissector_load_bytes,
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,
1799 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1800 u32, offset, void *, to, u32, len, u32, start_header)
1802 u8 *end = skb_tail_pointer(skb);
1805 if (unlikely(offset > 0xffff))
1808 switch (start_header) {
1809 case BPF_HDR_START_MAC:
1810 if (unlikely(!skb_mac_header_was_set(skb)))
1812 start = skb_mac_header(skb);
1814 case BPF_HDR_START_NET:
1815 start = skb_network_header(skb);
1821 ptr = start + offset;
1823 if (likely(ptr + len <= end)) {
1824 memcpy(to, ptr, len);
1833 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1834 .func = bpf_skb_load_bytes_relative,
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,
1844 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
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.
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.
1855 return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1858 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1859 .func = bpf_skb_pull_data,
1861 .ret_type = RET_INTEGER,
1862 .arg1_type = ARG_PTR_TO_CTX,
1863 .arg2_type = ARG_ANYTHING,
1866 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1868 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1871 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1872 .func = bpf_sk_fullsock,
1874 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
1875 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
1878 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1879 unsigned int write_len)
1881 return __bpf_try_make_writable(skb, write_len);
1884 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
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.
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.
1895 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1898 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1899 .func = sk_skb_pull_data,
1901 .ret_type = RET_INTEGER,
1902 .arg1_type = ARG_PTR_TO_CTX,
1903 .arg2_type = ARG_ANYTHING,
1906 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1907 u64, from, u64, to, u64, flags)
1911 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1913 if (unlikely(offset > 0xffff || offset & 1))
1915 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1918 ptr = (__sum16 *)(skb->data + offset);
1919 switch (flags & BPF_F_HDR_FIELD_MASK) {
1921 if (unlikely(from != 0))
1924 csum_replace_by_diff(ptr, to);
1927 csum_replace2(ptr, from, to);
1930 csum_replace4(ptr, from, to);
1939 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1940 .func = bpf_l3_csum_replace,
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,
1950 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1951 u64, from, u64, to, u64, flags)
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;
1958 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1959 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1961 if (unlikely(offset > 0xffff || offset & 1))
1963 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1966 ptr = (__sum16 *)(skb->data + offset);
1967 if (is_mmzero && !do_mforce && !*ptr)
1970 switch (flags & BPF_F_HDR_FIELD_MASK) {
1972 if (unlikely(from != 0))
1975 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1978 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1981 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1987 if (is_mmzero && !*ptr)
1988 *ptr = CSUM_MANGLED_0;
1992 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1993 .func = bpf_l4_csum_replace,
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,
2003 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
2004 __be32 *, to, u32, to_size, __wsum, seed)
2006 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
2007 u32 diff_size = from_size + to_size;
2010 /* This is quite flexible, some examples:
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
2016 * Even for diffing, from_size and to_size don't need to be equal.
2018 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
2019 diff_size > sizeof(sp->diff)))
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];
2027 return csum_partial(sp->diff, diff_size, seed);
2030 static const struct bpf_func_proto bpf_csum_diff_proto = {
2031 .func = bpf_csum_diff,
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,
2042 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
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.
2048 if (skb->ip_summed == CHECKSUM_COMPLETE)
2049 return (skb->csum = csum_add(skb->csum, csum));
2054 static const struct bpf_func_proto bpf_csum_update_proto = {
2055 .func = bpf_csum_update,
2057 .ret_type = RET_INTEGER,
2058 .arg1_type = ARG_PTR_TO_CTX,
2059 .arg2_type = ARG_ANYTHING,
2062 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
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.
2069 case BPF_CSUM_LEVEL_INC:
2070 __skb_incr_checksum_unnecessary(skb);
2072 case BPF_CSUM_LEVEL_DEC:
2073 __skb_decr_checksum_unnecessary(skb);
2075 case BPF_CSUM_LEVEL_RESET:
2076 __skb_reset_checksum_unnecessary(skb);
2078 case BPF_CSUM_LEVEL_QUERY:
2079 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2080 skb->csum_level : -EACCES;
2088 static const struct bpf_func_proto bpf_csum_level_proto = {
2089 .func = bpf_csum_level,
2091 .ret_type = RET_INTEGER,
2092 .arg1_type = ARG_PTR_TO_CTX,
2093 .arg2_type = ARG_ANYTHING,
2096 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2098 return dev_forward_skb_nomtu(dev, skb);
2101 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2102 struct sk_buff *skb)
2104 int ret = ____dev_forward_skb(dev, skb, false);
2108 ret = netif_rx(skb);
2114 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2118 if (dev_xmit_recursion()) {
2119 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2125 skb_set_redirected_noclear(skb, skb_at_tc_ingress(skb));
2126 skb_clear_tstamp(skb);
2128 dev_xmit_recursion_inc();
2129 ret = dev_queue_xmit(skb);
2130 dev_xmit_recursion_dec();
2135 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2138 unsigned int mlen = skb_network_offset(skb);
2140 if (unlikely(skb->len <= mlen)) {
2146 __skb_pull(skb, mlen);
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.
2153 if (!skb_at_tc_ingress(skb))
2154 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
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);
2162 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2165 /* Verify that a link layer header is carried */
2166 if (unlikely(skb->mac_header >= skb->network_header || skb->len == 0)) {
2171 bpf_push_mac_rcsum(skb);
2172 return flags & BPF_F_INGRESS ?
2173 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2176 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2179 if (dev_is_mac_header_xmit(dev))
2180 return __bpf_redirect_common(skb, dev, flags);
2182 return __bpf_redirect_no_mac(skb, dev, flags);
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)
2189 u32 hh_len = LL_RESERVED_SPACE(dev);
2190 const struct in6_addr *nexthop;
2191 struct dst_entry *dst = NULL;
2192 struct neighbour *neigh;
2194 if (dev_xmit_recursion()) {
2195 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2200 skb_clear_tstamp(skb);
2202 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2203 skb = skb_expand_head(skb, hh_len);
2211 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2212 &ipv6_hdr(skb)->daddr);
2214 nexthop = &nh->ipv6_nh;
2216 neigh = ip_neigh_gw6(dev, nexthop);
2217 if (likely(!IS_ERR(neigh))) {
2220 sock_confirm_neigh(skb, neigh);
2222 dev_xmit_recursion_inc();
2223 ret = neigh_output(neigh, skb, false);
2224 dev_xmit_recursion_dec();
2229 rcu_read_unlock_bh();
2231 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2237 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2238 struct bpf_nh_params *nh)
2240 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2241 struct net *net = dev_net(dev);
2242 int err, ret = NET_XMIT_DROP;
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,
2256 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2260 skb_dst_set(skb, dst);
2261 } else if (nh->nh_family != AF_INET6) {
2265 err = bpf_out_neigh_v6(net, skb, dev, nh);
2266 if (unlikely(net_xmit_eval(err)))
2267 dev->stats.tx_errors++;
2269 ret = NET_XMIT_SUCCESS;
2272 dev->stats.tx_errors++;
2278 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2279 struct bpf_nh_params *nh)
2282 return NET_XMIT_DROP;
2284 #endif /* CONFIG_IPV6 */
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)
2290 u32 hh_len = LL_RESERVED_SPACE(dev);
2291 struct neighbour *neigh;
2292 bool is_v6gw = false;
2294 if (dev_xmit_recursion()) {
2295 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2300 skb_clear_tstamp(skb);
2302 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2303 skb = skb_expand_head(skb, hh_len);
2310 struct dst_entry *dst = skb_dst(skb);
2311 struct rtable *rt = container_of(dst, struct rtable, dst);
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);
2317 } else if (nh->nh_family == AF_INET) {
2318 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2324 if (likely(!IS_ERR(neigh))) {
2327 sock_confirm_neigh(skb, neigh);
2329 dev_xmit_recursion_inc();
2330 ret = neigh_output(neigh, skb, is_v6gw);
2331 dev_xmit_recursion_dec();
2342 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2343 struct bpf_nh_params *nh)
2345 const struct iphdr *ip4h = ip_hdr(skb);
2346 struct net *net = dev_net(dev);
2347 int err, ret = NET_XMIT_DROP;
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,
2361 rt = ip_route_output_flow(net, &fl4, NULL);
2364 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2369 skb_dst_set(skb, &rt->dst);
2372 err = bpf_out_neigh_v4(net, skb, dev, nh);
2373 if (unlikely(net_xmit_eval(err)))
2374 dev->stats.tx_errors++;
2376 ret = NET_XMIT_SUCCESS;
2379 dev->stats.tx_errors++;
2385 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2386 struct bpf_nh_params *nh)
2389 return NET_XMIT_DROP;
2391 #endif /* CONFIG_INET */
2393 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2394 struct bpf_nh_params *nh)
2396 struct ethhdr *ethh = eth_hdr(skb);
2398 if (unlikely(skb->mac_header >= skb->network_header))
2400 bpf_push_mac_rcsum(skb);
2401 if (is_multicast_ether_addr(ethh->h_dest))
2404 skb_pull(skb, sizeof(*ethh));
2405 skb_unset_mac_header(skb);
2406 skb_reset_network_header(skb);
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);
2417 /* Internal, non-exposed redirect flags. */
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)
2425 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2427 struct net_device *dev;
2428 struct sk_buff *clone;
2431 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2434 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2438 clone = skb_clone(skb, GFP_ATOMIC);
2439 if (unlikely(!clone))
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
2447 ret = bpf_try_make_head_writable(skb);
2448 if (unlikely(ret)) {
2453 return __bpf_redirect(clone, dev, flags);
2456 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2457 .func = bpf_clone_redirect,
2459 .ret_type = RET_INTEGER,
2460 .arg1_type = ARG_PTR_TO_CTX,
2461 .arg2_type = ARG_ANYTHING,
2462 .arg3_type = ARG_ANYTHING,
2465 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2466 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2468 int skb_do_redirect(struct sk_buff *skb)
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;
2475 dev = dev_get_by_index_rcu(net, ri->tgt_index);
2480 if (flags & BPF_F_PEER) {
2481 const struct net_device_ops *ops = dev->netdev_ops;
2483 if (unlikely(!ops->ndo_get_peer_dev ||
2484 !skb_at_tc_ingress(skb)))
2486 dev = ops->ndo_get_peer_dev(dev);
2487 if (unlikely(!dev ||
2488 !(dev->flags & IFF_UP) ||
2489 net_eq(net, dev_net(dev))))
2494 return flags & BPF_F_NEIGH ?
2495 __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2497 __bpf_redirect(skb, dev, flags);
2503 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2505 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2507 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2511 ri->tgt_index = ifindex;
2513 return TC_ACT_REDIRECT;
2516 static const struct bpf_func_proto bpf_redirect_proto = {
2517 .func = bpf_redirect,
2519 .ret_type = RET_INTEGER,
2520 .arg1_type = ARG_ANYTHING,
2521 .arg2_type = ARG_ANYTHING,
2524 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2526 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2528 if (unlikely(flags))
2531 ri->flags = BPF_F_PEER;
2532 ri->tgt_index = ifindex;
2534 return TC_ACT_REDIRECT;
2537 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2538 .func = bpf_redirect_peer,
2540 .ret_type = RET_INTEGER,
2541 .arg1_type = ARG_ANYTHING,
2542 .arg2_type = ARG_ANYTHING,
2545 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2546 int, plen, u64, flags)
2548 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2550 if (unlikely((plen && plen < sizeof(*params)) || flags))
2553 ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2554 ri->tgt_index = ifindex;
2556 BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2558 memcpy(&ri->nh, params, sizeof(ri->nh));
2560 return TC_ACT_REDIRECT;
2563 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2564 .func = bpf_redirect_neigh,
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,
2573 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2575 msg->apply_bytes = bytes;
2579 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2580 .func = bpf_msg_apply_bytes,
2582 .ret_type = RET_INTEGER,
2583 .arg1_type = ARG_PTR_TO_CTX,
2584 .arg2_type = ARG_ANYTHING,
2587 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2589 msg->cork_bytes = bytes;
2593 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2594 .func = bpf_msg_cork_bytes,
2596 .ret_type = RET_INTEGER,
2597 .arg1_type = ARG_PTR_TO_CTX,
2598 .arg2_type = ARG_ANYTHING,
2601 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2602 u32, end, u64, flags)
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;
2610 if (unlikely(flags || end <= start))
2613 /* First find the starting scatterlist element */
2617 len = sk_msg_elem(msg, i)->length;
2618 if (start < offset + len)
2620 sk_msg_iter_var_next(i);
2621 } while (i != msg->sg.end);
2623 if (unlikely(start >= offset + len))
2627 /* The start may point into the sg element so we need to also
2628 * account for the headroom.
2630 bytes_sg_total = start - offset + bytes;
2631 if (!test_bit(i, msg->sg.copy) && bytes_sg_total <= len)
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.
2645 copy += sk_msg_elem(msg, i)->length;
2646 sk_msg_iter_var_next(i);
2647 if (bytes_sg_total <= copy)
2649 } while (i != msg->sg.end);
2652 if (unlikely(bytes_sg_total > copy))
2655 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2657 if (unlikely(!page))
2660 raw = page_address(page);
2663 sge = sk_msg_elem(msg, i);
2664 from = sg_virt(sge);
2668 memcpy(to, from, len);
2671 put_page(sg_page(sge));
2673 sk_msg_iter_var_next(i);
2674 } while (i != last_sge);
2676 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
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.
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;
2690 sk_msg_iter_var_next(i);
2694 if (i + shift >= NR_MSG_FRAG_IDS)
2695 move_from = i + shift - NR_MSG_FRAG_IDS;
2697 move_from = i + shift;
2698 if (move_from == msg->sg.end)
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);
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;
2712 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2713 msg->data_end = msg->data + bytes;
2717 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2718 .func = bpf_msg_pull_data,
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,
2727 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2728 u32, len, u64, flags)
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;
2735 if (unlikely(flags))
2738 if (unlikely(len == 0))
2741 /* First find the starting scatterlist element */
2745 l = sk_msg_elem(msg, i)->length;
2747 if (start < offset + l)
2749 sk_msg_iter_var_next(i);
2750 } while (i != msg->sg.end);
2752 if (start >= offset + l)
2755 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
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
2764 if (!space || (space == 1 && start != offset))
2765 copy = msg->sg.data[i].length;
2767 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2768 get_order(copy + len));
2769 if (unlikely(!page))
2775 raw = page_address(page);
2777 psge = sk_msg_elem(msg, i);
2778 front = start - offset;
2779 back = psge->length - front;
2780 from = sg_virt(psge);
2783 memcpy(raw, from, front);
2787 to = raw + front + len;
2789 memcpy(to, from, back);
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);
2797 psge->length = start - offset;
2798 rsge.length -= psge->length;
2799 rsge.offset += start;
2801 sk_msg_iter_var_next(i);
2802 sg_unmark_end(psge);
2803 sg_unmark_end(&rsge);
2804 sk_msg_iter_next(msg, end);
2807 /* Slot(s) to place newly allocated data */
2810 /* Shift one or two slots as needed */
2812 sge = sk_msg_elem_cpy(msg, i);
2814 sk_msg_iter_var_next(i);
2815 sg_unmark_end(&sge);
2816 sk_msg_iter_next(msg, end);
2818 nsge = sk_msg_elem_cpy(msg, i);
2820 sk_msg_iter_var_next(i);
2821 nnsge = sk_msg_elem_cpy(msg, i);
2824 while (i != msg->sg.end) {
2825 msg->sg.data[i] = sge;
2827 sk_msg_iter_var_next(i);
2830 nnsge = sk_msg_elem_cpy(msg, i);
2832 nsge = sk_msg_elem_cpy(msg, i);
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);
2843 get_page(sg_page(&rsge));
2844 sk_msg_iter_var_next(new);
2845 msg->sg.data[new] = rsge;
2848 sk_msg_compute_data_pointers(msg);
2852 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2853 .func = bpf_msg_push_data,
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,
2862 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2868 sk_msg_iter_var_next(i);
2869 msg->sg.data[prev] = msg->sg.data[i];
2870 } while (i != msg->sg.end);
2872 sk_msg_iter_prev(msg, end);
2875 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2877 struct scatterlist tmp, sge;
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);
2884 while (i != msg->sg.end) {
2885 msg->sg.data[i] = sge;
2886 sk_msg_iter_var_next(i);
2888 tmp = sk_msg_elem_cpy(msg, i);
2892 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2893 u32, len, u64, flags)
2895 u32 i = 0, l = 0, space, offset = 0;
2896 u64 last = start + len;
2899 if (unlikely(flags))
2902 /* First find the starting scatterlist element */
2906 l = sk_msg_elem(msg, i)->length;
2908 if (start < offset + l)
2910 sk_msg_iter_var_next(i);
2911 } while (i != msg->sg.end);
2913 /* Bounds checks: start and pop must be inside message */
2914 if (start >= offset + l || last >= msg->sg.size)
2917 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2920 /* --------------| offset
2921 * -| start |-------- len -------|
2923 * |----- a ----|-------- pop -------|----- b ----|
2924 * |______________________________________________| length
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.
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
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.
2941 if (start != offset) {
2942 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2944 int b = sge->length - pop - a;
2946 sk_msg_iter_var_next(i);
2948 if (pop < sge->length - a) {
2951 sk_msg_shift_right(msg, i);
2952 nsge = sk_msg_elem(msg, i);
2953 get_page(sg_page(sge));
2956 b, sge->offset + pop + a);
2958 struct page *page, *orig;
2961 page = alloc_pages(__GFP_NOWARN |
2962 __GFP_COMP | GFP_ATOMIC,
2964 if (unlikely(!page))
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);
2977 } else if (pop >= sge->length - a) {
2978 pop -= (sge->length - a);
2983 /* From above the current layout _must_ be as follows,
2988 * |---- pop ---|---------------- b ------------|
2989 * |____________________________________________| length
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.
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.
3001 struct scatterlist *sge = sk_msg_elem(msg, i);
3003 if (pop < sge->length) {
3009 sk_msg_shift_left(msg, i);
3011 sk_msg_iter_var_next(i);
3014 sk_mem_uncharge(msg->sk, len - pop);
3015 msg->sg.size -= (len - pop);
3016 sk_msg_compute_data_pointers(msg);
3020 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
3021 .func = bpf_msg_pop_data,
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,
3030 #ifdef CONFIG_CGROUP_NET_CLASSID
3031 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3033 return __task_get_classid(current);
3036 const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3037 .func = bpf_get_cgroup_classid_curr,
3039 .ret_type = RET_INTEGER,
3042 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3044 struct sock *sk = skb_to_full_sk(skb);
3046 if (!sk || !sk_fullsock(sk))
3049 return sock_cgroup_classid(&sk->sk_cgrp_data);
3052 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3053 .func = bpf_skb_cgroup_classid,
3055 .ret_type = RET_INTEGER,
3056 .arg1_type = ARG_PTR_TO_CTX,
3060 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3062 return task_get_classid(skb);
3065 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3066 .func = bpf_get_cgroup_classid,
3068 .ret_type = RET_INTEGER,
3069 .arg1_type = ARG_PTR_TO_CTX,
3072 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3074 return dst_tclassid(skb);
3077 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3078 .func = bpf_get_route_realm,
3080 .ret_type = RET_INTEGER,
3081 .arg1_type = ARG_PTR_TO_CTX,
3084 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
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.
3091 return skb_get_hash(skb);
3094 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3095 .func = bpf_get_hash_recalc,
3097 .ret_type = RET_INTEGER,
3098 .arg1_type = ARG_PTR_TO_CTX,
3101 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3103 /* After all direct packet write, this can be used once for
3104 * triggering a lazy recalc on next skb_get_hash() invocation.
3106 skb_clear_hash(skb);
3110 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3111 .func = bpf_set_hash_invalid,
3113 .ret_type = RET_INTEGER,
3114 .arg1_type = ARG_PTR_TO_CTX,
3117 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
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
3123 __skb_set_sw_hash(skb, hash, true);
3127 static const struct bpf_func_proto bpf_set_hash_proto = {
3128 .func = bpf_set_hash,
3130 .ret_type = RET_INTEGER,
3131 .arg1_type = ARG_PTR_TO_CTX,
3132 .arg2_type = ARG_ANYTHING,
3135 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3140 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3141 vlan_proto != htons(ETH_P_8021AD)))
3142 vlan_proto = htons(ETH_P_8021Q);
3144 bpf_push_mac_rcsum(skb);
3145 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3146 bpf_pull_mac_rcsum(skb);
3148 bpf_compute_data_pointers(skb);
3152 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3153 .func = bpf_skb_vlan_push,
3155 .ret_type = RET_INTEGER,
3156 .arg1_type = ARG_PTR_TO_CTX,
3157 .arg2_type = ARG_ANYTHING,
3158 .arg3_type = ARG_ANYTHING,
3161 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3165 bpf_push_mac_rcsum(skb);
3166 ret = skb_vlan_pop(skb);
3167 bpf_pull_mac_rcsum(skb);
3169 bpf_compute_data_pointers(skb);
3173 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3174 .func = bpf_skb_vlan_pop,
3176 .ret_type = RET_INTEGER,
3177 .arg1_type = ARG_PTR_TO_CTX,
3180 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3182 /* Caller already did skb_cow() with len as headroom,
3183 * so no need to do it here.
3186 memmove(skb->data, skb->data + len, off);
3187 memset(skb->data + off, 0, len);
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
3197 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3201 /* skb_ensure_writable() is not needed here, as we're
3202 * already working on an uncloned skb.
3204 if (unlikely(!pskb_may_pull(skb, off + len)))
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);
3215 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3217 bool trans_same = skb->transport_header == skb->network_header;
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.
3224 ret = bpf_skb_generic_push(skb, off, len);
3226 skb->mac_header -= len;
3227 skb->network_header -= len;
3229 skb->transport_header = skb->network_header;
3235 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3237 bool trans_same = skb->transport_header == skb->network_header;
3240 /* Same here, __skb_push()/__skb_pull() pair not needed. */
3241 ret = bpf_skb_generic_pop(skb, off, len);
3243 skb->mac_header += len;
3244 skb->network_header += len;
3246 skb->transport_header = skb->network_header;
3252 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3254 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3255 u32 off = skb_mac_header_len(skb);
3258 ret = skb_cow(skb, len_diff);
3259 if (unlikely(ret < 0))
3262 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3263 if (unlikely(ret < 0))
3266 if (skb_is_gso(skb)) {
3267 struct skb_shared_info *shinfo = skb_shinfo(skb);
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;
3276 skb->protocol = htons(ETH_P_IPV6);
3277 skb_clear_hash(skb);
3282 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3284 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3285 u32 off = skb_mac_header_len(skb);
3288 ret = skb_unclone(skb, GFP_ATOMIC);
3289 if (unlikely(ret < 0))
3292 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3293 if (unlikely(ret < 0))
3296 if (skb_is_gso(skb)) {
3297 struct skb_shared_info *shinfo = skb_shinfo(skb);
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;
3306 skb->protocol = htons(ETH_P_IP);
3307 skb_clear_hash(skb);
3312 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3314 __be16 from_proto = skb->protocol;
3316 if (from_proto == htons(ETH_P_IP) &&
3317 to_proto == htons(ETH_P_IPV6))
3318 return bpf_skb_proto_4_to_6(skb);
3320 if (from_proto == htons(ETH_P_IPV6) &&
3321 to_proto == htons(ETH_P_IP))
3322 return bpf_skb_proto_6_to_4(skb);
3327 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3332 if (unlikely(flags))
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.
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
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.
3352 ret = bpf_skb_proto_xlat(skb, proto);
3353 bpf_compute_data_pointers(skb);
3357 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3358 .func = bpf_skb_change_proto,
3360 .ret_type = RET_INTEGER,
3361 .arg1_type = ARG_PTR_TO_CTX,
3362 .arg2_type = ARG_ANYTHING,
3363 .arg3_type = ARG_ANYTHING,
3366 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
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)))
3373 skb->pkt_type = pkt_type;
3377 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3378 .func = bpf_skb_change_type,
3380 .ret_type = RET_INTEGER,
3381 .arg1_type = ARG_PTR_TO_CTX,
3382 .arg2_type = ARG_ANYTHING,
3385 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
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);
3397 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3398 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3400 #define BPF_F_ADJ_ROOM_DECAP_L3_MASK (BPF_F_ADJ_ROOM_DECAP_L3_IPV4 | \
3401 BPF_F_ADJ_ROOM_DECAP_L3_IPV6)
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)
3412 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
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;
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))
3428 ret = skb_cow_head(skb, len_diff);
3429 if (unlikely(ret < 0))
3433 if (skb->protocol != htons(ETH_P_IP) &&
3434 skb->protocol != htons(ETH_P_IPV6))
3437 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3438 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3441 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3442 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3445 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH &&
3446 inner_mac_len < ETH_HLEN)
3449 if (skb->encapsulation)
3452 mac_len = skb->network_header - skb->mac_header;
3453 inner_net = skb->network_header;
3454 if (inner_mac_len > len_diff)
3456 inner_trans = skb->transport_header;
3459 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3460 if (unlikely(ret < 0))
3464 skb->inner_mac_header = inner_net - inner_mac_len;
3465 skb->inner_network_header = inner_net;
3466 skb->inner_transport_header = inner_trans;
3468 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH)
3469 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
3471 skb_set_inner_protocol(skb, skb->protocol);
3473 skb->encapsulation = 1;
3474 skb_set_network_header(skb, mac_len);
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;
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);
3491 skb_set_transport_header(skb, mac_len + nh_len);
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);
3503 if (skb_is_gso(skb)) {
3504 struct skb_shared_info *shinfo = skb_shinfo(skb);
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);
3510 /* Header must be checked, and gso_segs recomputed. */
3511 shinfo->gso_type |= gso_type;
3512 shinfo->gso_segs = 0;
3518 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
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)))
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))
3535 ret = skb_unclone(skb, GFP_ATOMIC);
3536 if (unlikely(ret < 0))
3539 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3540 if (unlikely(ret < 0))
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);
3551 if (skb_is_gso(skb)) {
3552 struct skb_shared_info *shinfo = skb_shinfo(skb);
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);
3558 /* Header must be checked, and gso_segs recomputed. */
3559 shinfo->gso_type |= SKB_GSO_DODGY;
3560 shinfo->gso_segs = 0;
3566 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3568 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3569 u32, mode, u64, flags)
3571 u32 len_diff_abs = abs(len_diff);
3572 bool shrink = len_diff < 0;
3575 if (unlikely(flags || mode))
3577 if (unlikely(len_diff_abs > 0xfffU))
3581 ret = skb_cow(skb, len_diff);
3582 if (unlikely(ret < 0))
3584 __skb_push(skb, len_diff_abs);
3585 memset(skb->data, 0, len_diff_abs);
3587 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3589 __skb_pull(skb, len_diff_abs);
3591 if (tls_sw_has_ctx_rx(skb->sk)) {
3592 struct strp_msg *rxm = strp_msg(skb);
3594 rxm->full_len += len_diff;
3599 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3600 .func = sk_skb_adjust_room,
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,
3609 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3610 u32, mode, u64, flags)
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;
3620 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3621 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3623 if (unlikely(len_diff_abs > 0xfffU))
3625 if (unlikely(proto != htons(ETH_P_IP) &&
3626 proto != htons(ETH_P_IPV6)))
3629 off = skb_mac_header_len(skb);
3631 case BPF_ADJ_ROOM_NET:
3632 off += bpf_skb_net_base_len(skb);
3634 case BPF_ADJ_ROOM_MAC:
3640 if (flags & BPF_F_ADJ_ROOM_DECAP_L3_MASK) {
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);
3648 case BPF_F_ADJ_ROOM_DECAP_L3_IPV6:
3649 len_min = sizeof(struct ipv6hdr);
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 &&
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);
3668 bpf_compute_data_pointers(skb);
3672 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3673 .func = bpf_skb_adjust_room,
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,
3682 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3684 u32 min_len = skb_network_offset(skb);
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);
3694 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3696 unsigned int old_len = skb->len;
3699 ret = __skb_grow_rcsum(skb, new_len);
3701 memset(skb->data + old_len, 0, new_len - old_len);
3705 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3707 return __skb_trim_rcsum(skb, new_len);
3710 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3713 u32 max_len = BPF_SKB_MAX_LEN;
3714 u32 min_len = __bpf_skb_min_len(skb);
3717 if (unlikely(flags || new_len > max_len || new_len < min_len))
3719 if (skb->encapsulation)
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.
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.
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.
3738 ret = __bpf_try_make_writable(skb, skb->len);
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))
3750 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3753 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3755 bpf_compute_data_pointers(skb);
3759 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3760 .func = bpf_skb_change_tail,
3762 .ret_type = RET_INTEGER,
3763 .arg1_type = ARG_PTR_TO_CTX,
3764 .arg2_type = ARG_ANYTHING,
3765 .arg3_type = ARG_ANYTHING,
3768 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3771 return __bpf_skb_change_tail(skb, new_len, flags);
3774 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3775 .func = sk_skb_change_tail,
3777 .ret_type = RET_INTEGER,
3778 .arg1_type = ARG_PTR_TO_CTX,
3779 .arg2_type = ARG_ANYTHING,
3780 .arg3_type = ARG_ANYTHING,
3783 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3786 u32 max_len = BPF_SKB_MAX_LEN;
3787 u32 new_len = skb->len + head_room;
3790 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3791 new_len < skb->len))
3794 ret = skb_cow(skb, head_room);
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.
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);
3814 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3817 int ret = __bpf_skb_change_head(skb, head_room, flags);
3819 bpf_compute_data_pointers(skb);
3823 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3824 .func = bpf_skb_change_head,
3826 .ret_type = RET_INTEGER,
3827 .arg1_type = ARG_PTR_TO_CTX,
3828 .arg2_type = ARG_ANYTHING,
3829 .arg3_type = ARG_ANYTHING,
3832 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3835 return __bpf_skb_change_head(skb, head_room, flags);
3838 static const struct bpf_func_proto sk_skb_change_head_proto = {
3839 .func = sk_skb_change_head,
3841 .ret_type = RET_INTEGER,
3842 .arg1_type = ARG_PTR_TO_CTX,
3843 .arg2_type = ARG_ANYTHING,
3844 .arg3_type = ARG_ANYTHING,
3847 BPF_CALL_1(bpf_xdp_get_buff_len, struct xdp_buff*, xdp)
3849 return xdp_get_buff_len(xdp);
3852 static const struct bpf_func_proto bpf_xdp_get_buff_len_proto = {
3853 .func = bpf_xdp_get_buff_len,
3855 .ret_type = RET_INTEGER,
3856 .arg1_type = ARG_PTR_TO_CTX,
3859 BTF_ID_LIST_SINGLE(bpf_xdp_get_buff_len_bpf_ids, struct, xdp_buff)
3861 const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto = {
3862 .func = bpf_xdp_get_buff_len,
3864 .arg1_type = ARG_PTR_TO_BTF_ID,
3865 .arg1_btf_id = &bpf_xdp_get_buff_len_bpf_ids[0],
3868 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3870 return xdp_data_meta_unsupported(xdp) ? 0 :
3871 xdp->data - xdp->data_meta;
3874 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
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;
3881 if (unlikely(data < data_start ||
3882 data > xdp->data_end - ETH_HLEN))
3886 memmove(xdp->data_meta + offset,
3887 xdp->data_meta, metalen);
3888 xdp->data_meta += offset;
3894 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3895 .func = bpf_xdp_adjust_head,
3897 .ret_type = RET_INTEGER,
3898 .arg1_type = ARG_PTR_TO_CTX,
3899 .arg2_type = ARG_ANYTHING,
3902 void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off,
3903 void *buf, unsigned long len, bool flush)
3905 unsigned long ptr_len, ptr_off = 0;
3906 skb_frag_t *next_frag, *end_frag;
3907 struct skb_shared_info *sinfo;
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);
3918 sinfo = xdp_get_shared_info_from_buff(xdp);
3919 end_frag = &sinfo->frags[sinfo->nr_frags];
3920 next_frag = &sinfo->frags[0];
3922 ptr_len = xdp->data_end - xdp->data;
3923 ptr_buf = xdp->data;
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);
3930 src = flush ? buf : ptr_buf + copy_off;
3931 dst = flush ? ptr_buf + copy_off : buf;
3932 memcpy(dst, src, copy_len);
3939 if (!len || next_frag == end_frag)
3943 ptr_buf = skb_frag_address(next_frag);
3944 ptr_len = skb_frag_size(next_frag);
3949 void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len)
3951 u32 size = xdp->data_end - xdp->data;
3952 struct skb_shared_info *sinfo;
3953 void *addr = xdp->data;
3956 if (unlikely(offset > 0xffff || len > 0xffff))
3957 return ERR_PTR(-EFAULT);
3959 if (unlikely(offset + len > xdp_get_buff_len(xdp)))
3960 return ERR_PTR(-EINVAL);
3962 if (likely(offset < size)) /* linear area */
3965 sinfo = xdp_get_shared_info_from_buff(xdp);
3967 for (i = 0; i < sinfo->nr_frags; i++) { /* paged area */
3968 u32 frag_size = skb_frag_size(&sinfo->frags[i]);
3970 if (offset < frag_size) {
3971 addr = skb_frag_address(&sinfo->frags[i]);
3975 offset -= frag_size;
3978 return offset + len <= size ? addr + offset : NULL;
3981 BPF_CALL_4(bpf_xdp_load_bytes, struct xdp_buff *, xdp, u32, offset,
3982 void *, buf, u32, len)
3986 ptr = bpf_xdp_pointer(xdp, offset, len);
3988 return PTR_ERR(ptr);
3991 bpf_xdp_copy_buf(xdp, offset, buf, len, false);
3993 memcpy(buf, ptr, len);
3998 static const struct bpf_func_proto bpf_xdp_load_bytes_proto = {
3999 .func = bpf_xdp_load_bytes,
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,
4008 int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len)
4010 return ____bpf_xdp_load_bytes(xdp, offset, buf, len);
4013 BPF_CALL_4(bpf_xdp_store_bytes, struct xdp_buff *, xdp, u32, offset,
4014 void *, buf, u32, len)
4018 ptr = bpf_xdp_pointer(xdp, offset, len);
4020 return PTR_ERR(ptr);
4023 bpf_xdp_copy_buf(xdp, offset, buf, len, true);
4025 memcpy(ptr, buf, len);
4030 static const struct bpf_func_proto bpf_xdp_store_bytes_proto = {
4031 .func = bpf_xdp_store_bytes,
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,
4040 int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len)
4042 return ____bpf_xdp_store_bytes(xdp, offset, buf, len);
4045 static int bpf_xdp_frags_increase_tail(struct xdp_buff *xdp, int offset)
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;
4052 if (!rxq->frag_size || rxq->frag_size > xdp->frame_sz)
4055 tailroom = rxq->frag_size - skb_frag_size(frag) - skb_frag_off(frag);
4056 if (unlikely(offset > tailroom))
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;
4066 static int bpf_xdp_frags_shrink_tail(struct xdp_buff *xdp, int offset)
4068 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
4069 int i, n_frags_free = 0, len_free = 0;
4071 if (unlikely(offset > (int)xdp_get_buff_len(xdp) - ETH_HLEN))
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));
4081 if (skb_frag_size(frag) == shrink) {
4082 struct page *page = skb_frag_page(frag);
4084 __xdp_return(page_address(page), &xdp->rxq->mem,
4088 skb_frag_size_sub(frag, shrink);
4092 sinfo->nr_frags -= n_frags_free;
4093 sinfo->xdp_frags_size -= len_free;
4095 if (unlikely(!sinfo->nr_frags)) {
4096 xdp_buff_clear_frags_flag(xdp);
4097 xdp->data_end -= offset;
4103 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
4105 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
4106 void *data_end = xdp->data_end + offset;
4108 if (unlikely(xdp_buff_has_frags(xdp))) { /* non-linear xdp buff */
4110 return bpf_xdp_frags_shrink_tail(xdp, -offset);
4112 return bpf_xdp_frags_increase_tail(xdp, offset);
4115 /* Notice that xdp_data_hard_end have reserved some tailroom */
4116 if (unlikely(data_end > data_hard_end))
4119 if (unlikely(data_end < xdp->data + ETH_HLEN))
4122 /* Clear memory area on grow, can contain uninit kernel memory */
4124 memset(xdp->data_end, 0, offset);
4126 xdp->data_end = data_end;
4131 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
4132 .func = bpf_xdp_adjust_tail,
4134 .ret_type = RET_INTEGER,
4135 .arg1_type = ARG_PTR_TO_CTX,
4136 .arg2_type = ARG_ANYTHING,
4139 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
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;
4145 if (xdp_data_meta_unsupported(xdp))
4147 if (unlikely(meta < xdp_frame_end ||
4150 if (unlikely(xdp_metalen_invalid(metalen)))
4153 xdp->data_meta = meta;
4158 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
4159 .func = bpf_xdp_adjust_meta,
4161 .ret_type = RET_INTEGER,
4162 .arg1_type = ARG_PTR_TO_CTX,
4163 .arg2_type = ARG_ANYTHING,
4169 * XDP_REDIRECT works by a three-step process, implemented in the functions
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.
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.
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.
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().
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.
4202 void xdp_do_flush(void)
4208 EXPORT_SYMBOL_GPL(xdp_do_flush);
4210 void bpf_clear_redirect_map(struct bpf_map *map)
4212 struct bpf_redirect_info *ri;
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.
4222 if (unlikely(READ_ONCE(ri->map) == map))
4223 cmpxchg(&ri->map, map, NULL);
4227 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
4228 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key);
4230 u32 xdp_master_redirect(struct xdp_buff *xdp)
4232 struct net_device *master, *slave;
4233 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
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.
4243 ri->tgt_index = slave->ifindex;
4244 ri->map_id = INT_MAX;
4245 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4246 return XDP_REDIRECT;
4250 EXPORT_SYMBOL_GPL(xdp_master_redirect);
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)
4257 enum bpf_map_type map_type = ri->map_type;
4258 void *fwd = ri->tgt_value;
4259 u32 map_id = ri->map_id;
4262 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4263 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4265 err = __xsk_map_redirect(fwd, xdp);
4269 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4272 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
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)
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;
4287 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4288 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4290 if (unlikely(!xdpf)) {
4296 case BPF_MAP_TYPE_DEVMAP:
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);
4305 err = dev_map_enqueue(fwd, xdpf, dev);
4308 case BPF_MAP_TYPE_CPUMAP:
4309 err = cpu_map_enqueue(fwd, xdpf, dev);
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)) {
4318 err = dev_xdp_enqueue(fwd, xdpf, dev);
4329 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4332 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4336 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
4337 struct bpf_prog *xdp_prog)
4339 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4340 enum bpf_map_type map_type = ri->map_type;
4342 if (map_type == BPF_MAP_TYPE_XSKMAP) {
4343 /* XDP_REDIRECT is not supported AF_XDP yet. */
4344 if (unlikely(xdp_buff_has_frags(xdp)))
4347 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4350 return __xdp_do_redirect_frame(ri, dev, xdp_convert_buff_to_frame(xdp),
4353 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4355 int xdp_do_redirect_frame(struct net_device *dev, struct xdp_buff *xdp,
4356 struct xdp_frame *xdpf, struct bpf_prog *xdp_prog)
4358 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4359 enum bpf_map_type map_type = ri->map_type;
4361 if (map_type == BPF_MAP_TYPE_XSKMAP)
4362 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4364 return __xdp_do_redirect_frame(ri, dev, xdpf, xdp_prog);
4366 EXPORT_SYMBOL_GPL(xdp_do_redirect_frame);
4368 static int xdp_do_generic_redirect_map(struct net_device *dev,
4369 struct sk_buff *skb,
4370 struct xdp_buff *xdp,
4371 struct bpf_prog *xdp_prog,
4373 enum bpf_map_type map_type, u32 map_id)
4375 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4376 struct bpf_map *map;
4380 case BPF_MAP_TYPE_DEVMAP:
4382 case BPF_MAP_TYPE_DEVMAP_HASH:
4383 map = READ_ONCE(ri->map);
4384 if (unlikely(map)) {
4385 WRITE_ONCE(ri->map, NULL);
4386 err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4387 ri->flags & BPF_F_EXCLUDE_INGRESS);
4389 err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4394 case BPF_MAP_TYPE_XSKMAP:
4395 err = xsk_generic_rcv(fwd, xdp);
4400 case BPF_MAP_TYPE_CPUMAP:
4401 err = cpu_map_generic_redirect(fwd, skb);
4410 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4413 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4417 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4418 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4420 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4421 enum bpf_map_type map_type = ri->map_type;
4422 void *fwd = ri->tgt_value;
4423 u32 map_id = ri->map_id;
4426 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4427 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4429 if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4430 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4431 if (unlikely(!fwd)) {
4436 err = xdp_ok_fwd_dev(fwd, skb->len);
4441 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4442 generic_xdp_tx(skb, xdp_prog);
4446 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id);
4448 _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4452 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4454 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4456 if (unlikely(flags))
4459 /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4460 * by map_idr) is used for ifindex based XDP redirect.
4462 ri->tgt_index = ifindex;
4463 ri->map_id = INT_MAX;
4464 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4466 return XDP_REDIRECT;
4469 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4470 .func = bpf_xdp_redirect,
4472 .ret_type = RET_INTEGER,
4473 .arg1_type = ARG_ANYTHING,
4474 .arg2_type = ARG_ANYTHING,
4477 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u64, key,
4480 return map->ops->map_redirect(map, key, flags);
4483 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4484 .func = bpf_xdp_redirect_map,
4486 .ret_type = RET_INTEGER,
4487 .arg1_type = ARG_CONST_MAP_PTR,
4488 .arg2_type = ARG_ANYTHING,
4489 .arg3_type = ARG_ANYTHING,
4492 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4493 unsigned long off, unsigned long len)
4495 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4499 if (ptr != dst_buff)
4500 memcpy(dst_buff, ptr, len);
4505 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4506 u64, flags, void *, meta, u64, meta_size)
4508 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4510 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4512 if (unlikely(!skb || skb_size > skb->len))
4515 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4519 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4520 .func = bpf_skb_event_output,
4522 .ret_type = RET_INTEGER,
4523 .arg1_type = ARG_PTR_TO_CTX,
4524 .arg2_type = ARG_CONST_MAP_PTR,
4525 .arg3_type = ARG_ANYTHING,
4526 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4527 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4530 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4532 const struct bpf_func_proto bpf_skb_output_proto = {
4533 .func = bpf_skb_event_output,
4535 .ret_type = RET_INTEGER,
4536 .arg1_type = ARG_PTR_TO_BTF_ID,
4537 .arg1_btf_id = &bpf_skb_output_btf_ids[0],
4538 .arg2_type = ARG_CONST_MAP_PTR,
4539 .arg3_type = ARG_ANYTHING,
4540 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4541 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4544 static unsigned short bpf_tunnel_key_af(u64 flags)
4546 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4549 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4550 u32, size, u64, flags)
4552 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4553 u8 compat[sizeof(struct bpf_tunnel_key)];
4557 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6 |
4558 BPF_F_TUNINFO_FLAGS)))) {
4562 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4566 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4569 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4570 case offsetof(struct bpf_tunnel_key, tunnel_label):
4571 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4573 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4574 /* Fixup deprecated structure layouts here, so we have
4575 * a common path later on.
4577 if (ip_tunnel_info_af(info) != AF_INET)
4580 to = (struct bpf_tunnel_key *)compat;
4587 to->tunnel_id = be64_to_cpu(info->key.tun_id);
4588 to->tunnel_tos = info->key.tos;
4589 to->tunnel_ttl = info->key.ttl;
4590 if (flags & BPF_F_TUNINFO_FLAGS)
4591 to->tunnel_flags = info->key.tun_flags;
4595 if (flags & BPF_F_TUNINFO_IPV6) {
4596 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4597 sizeof(to->remote_ipv6));
4598 memcpy(to->local_ipv6, &info->key.u.ipv6.dst,
4599 sizeof(to->local_ipv6));
4600 to->tunnel_label = be32_to_cpu(info->key.label);
4602 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4603 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4604 to->local_ipv4 = be32_to_cpu(info->key.u.ipv4.dst);
4605 memset(&to->local_ipv6[1], 0, sizeof(__u32) * 3);
4606 to->tunnel_label = 0;
4609 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4610 memcpy(to_orig, to, size);
4614 memset(to_orig, 0, size);
4618 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4619 .func = bpf_skb_get_tunnel_key,
4621 .ret_type = RET_INTEGER,
4622 .arg1_type = ARG_PTR_TO_CTX,
4623 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4624 .arg3_type = ARG_CONST_SIZE,
4625 .arg4_type = ARG_ANYTHING,
4628 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4630 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4633 if (unlikely(!info ||
4634 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4638 if (unlikely(size < info->options_len)) {
4643 ip_tunnel_info_opts_get(to, info);
4644 if (size > info->options_len)
4645 memset(to + info->options_len, 0, size - info->options_len);
4647 return info->options_len;
4649 memset(to, 0, size);
4653 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4654 .func = bpf_skb_get_tunnel_opt,
4656 .ret_type = RET_INTEGER,
4657 .arg1_type = ARG_PTR_TO_CTX,
4658 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4659 .arg3_type = ARG_CONST_SIZE,
4662 static struct metadata_dst __percpu *md_dst;
4664 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4665 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4667 struct metadata_dst *md = this_cpu_ptr(md_dst);
4668 u8 compat[sizeof(struct bpf_tunnel_key)];
4669 struct ip_tunnel_info *info;
4671 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4672 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER |
4673 BPF_F_NO_TUNNEL_KEY)))
4675 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4677 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4678 case offsetof(struct bpf_tunnel_key, tunnel_label):
4679 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4680 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4681 /* Fixup deprecated structure layouts here, so we have
4682 * a common path later on.
4684 memcpy(compat, from, size);
4685 memset(compat + size, 0, sizeof(compat) - size);
4686 from = (const struct bpf_tunnel_key *) compat;
4692 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4697 dst_hold((struct dst_entry *) md);
4698 skb_dst_set(skb, (struct dst_entry *) md);
4700 info = &md->u.tun_info;
4701 memset(info, 0, sizeof(*info));
4702 info->mode = IP_TUNNEL_INFO_TX;
4704 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4705 if (flags & BPF_F_DONT_FRAGMENT)
4706 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4707 if (flags & BPF_F_ZERO_CSUM_TX)
4708 info->key.tun_flags &= ~TUNNEL_CSUM;
4709 if (flags & BPF_F_SEQ_NUMBER)
4710 info->key.tun_flags |= TUNNEL_SEQ;
4711 if (flags & BPF_F_NO_TUNNEL_KEY)
4712 info->key.tun_flags &= ~TUNNEL_KEY;
4714 info->key.tun_id = cpu_to_be64(from->tunnel_id);
4715 info->key.tos = from->tunnel_tos;
4716 info->key.ttl = from->tunnel_ttl;
4718 if (flags & BPF_F_TUNINFO_IPV6) {
4719 info->mode |= IP_TUNNEL_INFO_IPV6;
4720 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4721 sizeof(from->remote_ipv6));
4722 memcpy(&info->key.u.ipv6.src, from->local_ipv6,
4723 sizeof(from->local_ipv6));
4724 info->key.label = cpu_to_be32(from->tunnel_label) &
4725 IPV6_FLOWLABEL_MASK;
4727 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4728 info->key.u.ipv4.src = cpu_to_be32(from->local_ipv4);
4729 info->key.flow_flags = FLOWI_FLAG_ANYSRC;
4735 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4736 .func = bpf_skb_set_tunnel_key,
4738 .ret_type = RET_INTEGER,
4739 .arg1_type = ARG_PTR_TO_CTX,
4740 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4741 .arg3_type = ARG_CONST_SIZE,
4742 .arg4_type = ARG_ANYTHING,
4745 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4746 const u8 *, from, u32, size)
4748 struct ip_tunnel_info *info = skb_tunnel_info(skb);
4749 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4751 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4753 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4756 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4761 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4762 .func = bpf_skb_set_tunnel_opt,
4764 .ret_type = RET_INTEGER,
4765 .arg1_type = ARG_PTR_TO_CTX,
4766 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4767 .arg3_type = ARG_CONST_SIZE,
4770 static const struct bpf_func_proto *
4771 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4774 struct metadata_dst __percpu *tmp;
4776 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4781 if (cmpxchg(&md_dst, NULL, tmp))
4782 metadata_dst_free_percpu(tmp);
4786 case BPF_FUNC_skb_set_tunnel_key:
4787 return &bpf_skb_set_tunnel_key_proto;
4788 case BPF_FUNC_skb_set_tunnel_opt:
4789 return &bpf_skb_set_tunnel_opt_proto;
4795 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4798 struct bpf_array *array = container_of(map, struct bpf_array, map);
4799 struct cgroup *cgrp;
4802 sk = skb_to_full_sk(skb);
4803 if (!sk || !sk_fullsock(sk))
4805 if (unlikely(idx >= array->map.max_entries))
4808 cgrp = READ_ONCE(array->ptrs[idx]);
4809 if (unlikely(!cgrp))
4812 return sk_under_cgroup_hierarchy(sk, cgrp);
4815 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4816 .func = bpf_skb_under_cgroup,
4818 .ret_type = RET_INTEGER,
4819 .arg1_type = ARG_PTR_TO_CTX,
4820 .arg2_type = ARG_CONST_MAP_PTR,
4821 .arg3_type = ARG_ANYTHING,
4824 #ifdef CONFIG_SOCK_CGROUP_DATA
4825 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4827 struct cgroup *cgrp;
4829 sk = sk_to_full_sk(sk);
4830 if (!sk || !sk_fullsock(sk))
4833 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4834 return cgroup_id(cgrp);
4837 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4839 return __bpf_sk_cgroup_id(skb->sk);
4842 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4843 .func = bpf_skb_cgroup_id,
4845 .ret_type = RET_INTEGER,
4846 .arg1_type = ARG_PTR_TO_CTX,
4849 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4852 struct cgroup *ancestor;
4853 struct cgroup *cgrp;
4855 sk = sk_to_full_sk(sk);
4856 if (!sk || !sk_fullsock(sk))
4859 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4860 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4864 return cgroup_id(ancestor);
4867 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4870 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4873 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4874 .func = bpf_skb_ancestor_cgroup_id,
4876 .ret_type = RET_INTEGER,
4877 .arg1_type = ARG_PTR_TO_CTX,
4878 .arg2_type = ARG_ANYTHING,
4881 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4883 return __bpf_sk_cgroup_id(sk);
4886 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4887 .func = bpf_sk_cgroup_id,
4889 .ret_type = RET_INTEGER,
4890 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4893 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4895 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4898 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4899 .func = bpf_sk_ancestor_cgroup_id,
4901 .ret_type = RET_INTEGER,
4902 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4903 .arg2_type = ARG_ANYTHING,
4907 static unsigned long bpf_xdp_copy(void *dst, const void *ctx,
4908 unsigned long off, unsigned long len)
4910 struct xdp_buff *xdp = (struct xdp_buff *)ctx;
4912 bpf_xdp_copy_buf(xdp, off, dst, len, false);
4916 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4917 u64, flags, void *, meta, u64, meta_size)
4919 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4921 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4924 if (unlikely(!xdp || xdp_size > xdp_get_buff_len(xdp)))
4927 return bpf_event_output(map, flags, meta, meta_size, xdp,
4928 xdp_size, bpf_xdp_copy);
4931 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4932 .func = bpf_xdp_event_output,
4934 .ret_type = RET_INTEGER,
4935 .arg1_type = ARG_PTR_TO_CTX,
4936 .arg2_type = ARG_CONST_MAP_PTR,
4937 .arg3_type = ARG_ANYTHING,
4938 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4939 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4942 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4944 const struct bpf_func_proto bpf_xdp_output_proto = {
4945 .func = bpf_xdp_event_output,
4947 .ret_type = RET_INTEGER,
4948 .arg1_type = ARG_PTR_TO_BTF_ID,
4949 .arg1_btf_id = &bpf_xdp_output_btf_ids[0],
4950 .arg2_type = ARG_CONST_MAP_PTR,
4951 .arg3_type = ARG_ANYTHING,
4952 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4953 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4956 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4958 return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4961 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4962 .func = bpf_get_socket_cookie,
4964 .ret_type = RET_INTEGER,
4965 .arg1_type = ARG_PTR_TO_CTX,
4968 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4970 return __sock_gen_cookie(ctx->sk);
4973 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4974 .func = bpf_get_socket_cookie_sock_addr,
4976 .ret_type = RET_INTEGER,
4977 .arg1_type = ARG_PTR_TO_CTX,
4980 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4982 return __sock_gen_cookie(ctx);
4985 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4986 .func = bpf_get_socket_cookie_sock,
4988 .ret_type = RET_INTEGER,
4989 .arg1_type = ARG_PTR_TO_CTX,
4992 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
4994 return sk ? sock_gen_cookie(sk) : 0;
4997 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
4998 .func = bpf_get_socket_ptr_cookie,
5000 .ret_type = RET_INTEGER,
5001 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON | PTR_MAYBE_NULL,
5004 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
5006 return __sock_gen_cookie(ctx->sk);
5009 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
5010 .func = bpf_get_socket_cookie_sock_ops,
5012 .ret_type = RET_INTEGER,
5013 .arg1_type = ARG_PTR_TO_CTX,
5016 static u64 __bpf_get_netns_cookie(struct sock *sk)
5018 const struct net *net = sk ? sock_net(sk) : &init_net;
5020 return net->net_cookie;
5023 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
5025 return __bpf_get_netns_cookie(ctx);
5028 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
5029 .func = bpf_get_netns_cookie_sock,
5031 .ret_type = RET_INTEGER,
5032 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5035 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
5037 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5040 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
5041 .func = bpf_get_netns_cookie_sock_addr,
5043 .ret_type = RET_INTEGER,
5044 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5047 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
5049 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5052 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = {
5053 .func = bpf_get_netns_cookie_sock_ops,
5055 .ret_type = RET_INTEGER,
5056 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5059 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx)
5061 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5064 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = {
5065 .func = bpf_get_netns_cookie_sk_msg,
5067 .ret_type = RET_INTEGER,
5068 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5071 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
5073 struct sock *sk = sk_to_full_sk(skb->sk);
5076 if (!sk || !sk_fullsock(sk))
5078 kuid = sock_net_uid(sock_net(sk), sk);
5079 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
5082 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
5083 .func = bpf_get_socket_uid,
5085 .ret_type = RET_INTEGER,
5086 .arg1_type = ARG_PTR_TO_CTX,
5089 static int sol_socket_sockopt(struct sock *sk, int optname,
5090 char *optval, int *optlen,
5102 case SO_MAX_PACING_RATE:
5103 case SO_BINDTOIFINDEX:
5105 if (*optlen != sizeof(int))
5108 case SO_BINDTODEVICE:
5115 if (optname == SO_BINDTODEVICE)
5117 return sk_getsockopt(sk, SOL_SOCKET, optname,
5118 KERNEL_SOCKPTR(optval),
5119 KERNEL_SOCKPTR(optlen));
5122 return sk_setsockopt(sk, SOL_SOCKET, optname,
5123 KERNEL_SOCKPTR(optval), *optlen);
5126 static int bpf_sol_tcp_setsockopt(struct sock *sk, int optname,
5127 char *optval, int optlen)
5129 struct tcp_sock *tp = tcp_sk(sk);
5130 unsigned long timeout;
5133 if (optlen != sizeof(int))
5136 val = *(int *)optval;
5138 /* Only some options are supported */
5141 if (val <= 0 || tp->data_segs_out > tp->syn_data)
5143 tcp_snd_cwnd_set(tp, val);
5145 case TCP_BPF_SNDCWND_CLAMP:
5148 tp->snd_cwnd_clamp = val;
5149 tp->snd_ssthresh = val;
5151 case TCP_BPF_DELACK_MAX:
5152 timeout = usecs_to_jiffies(val);
5153 if (timeout > TCP_DELACK_MAX ||
5154 timeout < TCP_TIMEOUT_MIN)
5156 inet_csk(sk)->icsk_delack_max = timeout;
5158 case TCP_BPF_RTO_MIN:
5159 timeout = usecs_to_jiffies(val);
5160 if (timeout > TCP_RTO_MIN ||
5161 timeout < TCP_TIMEOUT_MIN)
5163 inet_csk(sk)->icsk_rto_min = timeout;
5172 static int sol_tcp_sockopt_congestion(struct sock *sk, char *optval,
5173 int *optlen, bool getopt)
5175 struct tcp_sock *tp;
5182 if (!inet_csk(sk)->icsk_ca_ops)
5184 /* BPF expects NULL-terminated tcp-cc string */
5185 optval[--(*optlen)] = '\0';
5186 return do_tcp_getsockopt(sk, SOL_TCP, TCP_CONGESTION,
5187 KERNEL_SOCKPTR(optval),
5188 KERNEL_SOCKPTR(optlen));
5191 /* "cdg" is the only cc that alloc a ptr
5192 * in inet_csk_ca area. The bpf-tcp-cc may
5193 * overwrite this ptr after switching to cdg.
5195 if (*optlen >= sizeof("cdg") - 1 && !strncmp("cdg", optval, *optlen))
5198 /* It stops this looping
5200 * .init => bpf_setsockopt(tcp_cc) => .init =>
5201 * bpf_setsockopt(tcp_cc)" => .init => ....
5203 * The second bpf_setsockopt(tcp_cc) is not allowed
5204 * in order to break the loop when both .init
5205 * are the same bpf prog.
5207 * This applies even the second bpf_setsockopt(tcp_cc)
5208 * does not cause a loop. This limits only the first
5209 * '.init' can call bpf_setsockopt(TCP_CONGESTION) to
5210 * pick a fallback cc (eg. peer does not support ECN)
5211 * and the second '.init' cannot fallback to
5215 if (tp->bpf_chg_cc_inprogress)
5218 tp->bpf_chg_cc_inprogress = 1;
5219 ret = do_tcp_setsockopt(sk, SOL_TCP, TCP_CONGESTION,
5220 KERNEL_SOCKPTR(optval), *optlen);
5221 tp->bpf_chg_cc_inprogress = 0;
5225 static int sol_tcp_sockopt(struct sock *sk, int optname,
5226 char *optval, int *optlen,
5229 if (sk->sk_protocol != IPPROTO_TCP)
5239 case TCP_WINDOW_CLAMP:
5240 case TCP_THIN_LINEAR_TIMEOUTS:
5241 case TCP_USER_TIMEOUT:
5242 case TCP_NOTSENT_LOWAT:
5244 if (*optlen != sizeof(int))
5247 case TCP_CONGESTION:
5248 return sol_tcp_sockopt_congestion(sk, optval, optlen, getopt);
5256 return bpf_sol_tcp_setsockopt(sk, optname, optval, *optlen);
5260 if (optname == TCP_SAVED_SYN) {
5261 struct tcp_sock *tp = tcp_sk(sk);
5263 if (!tp->saved_syn ||
5264 *optlen > tcp_saved_syn_len(tp->saved_syn))
5266 memcpy(optval, tp->saved_syn->data, *optlen);
5267 /* It cannot free tp->saved_syn here because it
5268 * does not know if the user space still needs it.
5273 return do_tcp_getsockopt(sk, SOL_TCP, optname,
5274 KERNEL_SOCKPTR(optval),
5275 KERNEL_SOCKPTR(optlen));
5278 return do_tcp_setsockopt(sk, SOL_TCP, optname,
5279 KERNEL_SOCKPTR(optval), *optlen);
5282 static int sol_ip_sockopt(struct sock *sk, int optname,
5283 char *optval, int *optlen,
5286 if (sk->sk_family != AF_INET)
5291 if (*optlen != sizeof(int))
5299 return do_ip_getsockopt(sk, SOL_IP, optname,
5300 KERNEL_SOCKPTR(optval),
5301 KERNEL_SOCKPTR(optlen));
5303 return do_ip_setsockopt(sk, SOL_IP, optname,
5304 KERNEL_SOCKPTR(optval), *optlen);
5307 static int sol_ipv6_sockopt(struct sock *sk, int optname,
5308 char *optval, int *optlen,
5311 if (sk->sk_family != AF_INET6)
5316 case IPV6_AUTOFLOWLABEL:
5317 if (*optlen != sizeof(int))
5325 return ipv6_bpf_stub->ipv6_getsockopt(sk, SOL_IPV6, optname,
5326 KERNEL_SOCKPTR(optval),
5327 KERNEL_SOCKPTR(optlen));
5329 return ipv6_bpf_stub->ipv6_setsockopt(sk, SOL_IPV6, optname,
5330 KERNEL_SOCKPTR(optval), *optlen);
5333 static int __bpf_setsockopt(struct sock *sk, int level, int optname,
5334 char *optval, int optlen)
5336 if (!sk_fullsock(sk))
5339 if (level == SOL_SOCKET)
5340 return sol_socket_sockopt(sk, optname, optval, &optlen, false);
5341 else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5342 return sol_ip_sockopt(sk, optname, optval, &optlen, false);
5343 else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5344 return sol_ipv6_sockopt(sk, optname, optval, &optlen, false);
5345 else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5346 return sol_tcp_sockopt(sk, optname, optval, &optlen, false);
5351 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
5352 char *optval, int optlen)
5354 if (sk_fullsock(sk))
5355 sock_owned_by_me(sk);
5356 return __bpf_setsockopt(sk, level, optname, optval, optlen);
5359 static int __bpf_getsockopt(struct sock *sk, int level, int optname,
5360 char *optval, int optlen)
5362 int err, saved_optlen = optlen;
5364 if (!sk_fullsock(sk)) {
5369 if (level == SOL_SOCKET)
5370 err = sol_socket_sockopt(sk, optname, optval, &optlen, true);
5371 else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5372 err = sol_tcp_sockopt(sk, optname, optval, &optlen, true);
5373 else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5374 err = sol_ip_sockopt(sk, optname, optval, &optlen, true);
5375 else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5376 err = sol_ipv6_sockopt(sk, optname, optval, &optlen, true);
5383 if (optlen < saved_optlen)
5384 memset(optval + optlen, 0, saved_optlen - optlen);
5388 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
5389 char *optval, int optlen)
5391 if (sk_fullsock(sk))
5392 sock_owned_by_me(sk);
5393 return __bpf_getsockopt(sk, level, optname, optval, optlen);
5396 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level,
5397 int, optname, char *, optval, int, optlen)
5399 return _bpf_setsockopt(sk, level, optname, optval, optlen);
5402 const struct bpf_func_proto bpf_sk_setsockopt_proto = {
5403 .func = bpf_sk_setsockopt,
5405 .ret_type = RET_INTEGER,
5406 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5407 .arg2_type = ARG_ANYTHING,
5408 .arg3_type = ARG_ANYTHING,
5409 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5410 .arg5_type = ARG_CONST_SIZE,
5413 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level,
5414 int, optname, char *, optval, int, optlen)
5416 return _bpf_getsockopt(sk, level, optname, optval, optlen);
5419 const struct bpf_func_proto bpf_sk_getsockopt_proto = {
5420 .func = bpf_sk_getsockopt,
5422 .ret_type = RET_INTEGER,
5423 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5424 .arg2_type = ARG_ANYTHING,
5425 .arg3_type = ARG_ANYTHING,
5426 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5427 .arg5_type = ARG_CONST_SIZE,
5430 BPF_CALL_5(bpf_unlocked_sk_setsockopt, struct sock *, sk, int, level,
5431 int, optname, char *, optval, int, optlen)
5433 return __bpf_setsockopt(sk, level, optname, optval, optlen);
5436 const struct bpf_func_proto bpf_unlocked_sk_setsockopt_proto = {
5437 .func = bpf_unlocked_sk_setsockopt,
5439 .ret_type = RET_INTEGER,
5440 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5441 .arg2_type = ARG_ANYTHING,
5442 .arg3_type = ARG_ANYTHING,
5443 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5444 .arg5_type = ARG_CONST_SIZE,
5447 BPF_CALL_5(bpf_unlocked_sk_getsockopt, struct sock *, sk, int, level,
5448 int, optname, char *, optval, int, optlen)
5450 return __bpf_getsockopt(sk, level, optname, optval, optlen);
5453 const struct bpf_func_proto bpf_unlocked_sk_getsockopt_proto = {
5454 .func = bpf_unlocked_sk_getsockopt,
5456 .ret_type = RET_INTEGER,
5457 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5458 .arg2_type = ARG_ANYTHING,
5459 .arg3_type = ARG_ANYTHING,
5460 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5461 .arg5_type = ARG_CONST_SIZE,
5464 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5465 int, level, int, optname, char *, optval, int, optlen)
5467 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5470 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5471 .func = bpf_sock_addr_setsockopt,
5473 .ret_type = RET_INTEGER,
5474 .arg1_type = ARG_PTR_TO_CTX,
5475 .arg2_type = ARG_ANYTHING,
5476 .arg3_type = ARG_ANYTHING,
5477 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5478 .arg5_type = ARG_CONST_SIZE,
5481 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5482 int, level, int, optname, char *, optval, int, optlen)
5484 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5487 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5488 .func = bpf_sock_addr_getsockopt,
5490 .ret_type = RET_INTEGER,
5491 .arg1_type = ARG_PTR_TO_CTX,
5492 .arg2_type = ARG_ANYTHING,
5493 .arg3_type = ARG_ANYTHING,
5494 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5495 .arg5_type = ARG_CONST_SIZE,
5498 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5499 int, level, int, optname, char *, optval, int, optlen)
5501 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5504 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5505 .func = bpf_sock_ops_setsockopt,
5507 .ret_type = RET_INTEGER,
5508 .arg1_type = ARG_PTR_TO_CTX,
5509 .arg2_type = ARG_ANYTHING,
5510 .arg3_type = ARG_ANYTHING,
5511 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5512 .arg5_type = ARG_CONST_SIZE,
5515 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5516 int optname, const u8 **start)
5518 struct sk_buff *syn_skb = bpf_sock->syn_skb;
5519 const u8 *hdr_start;
5523 /* sk is a request_sock here */
5525 if (optname == TCP_BPF_SYN) {
5526 hdr_start = syn_skb->data;
5527 ret = tcp_hdrlen(syn_skb);
5528 } else if (optname == TCP_BPF_SYN_IP) {
5529 hdr_start = skb_network_header(syn_skb);
5530 ret = skb_network_header_len(syn_skb) +
5531 tcp_hdrlen(syn_skb);
5533 /* optname == TCP_BPF_SYN_MAC */
5534 hdr_start = skb_mac_header(syn_skb);
5535 ret = skb_mac_header_len(syn_skb) +
5536 skb_network_header_len(syn_skb) +
5537 tcp_hdrlen(syn_skb);
5540 struct sock *sk = bpf_sock->sk;
5541 struct saved_syn *saved_syn;
5543 if (sk->sk_state == TCP_NEW_SYN_RECV)
5544 /* synack retransmit. bpf_sock->syn_skb will
5545 * not be available. It has to resort to
5546 * saved_syn (if it is saved).
5548 saved_syn = inet_reqsk(sk)->saved_syn;
5550 saved_syn = tcp_sk(sk)->saved_syn;
5555 if (optname == TCP_BPF_SYN) {
5556 hdr_start = saved_syn->data +
5557 saved_syn->mac_hdrlen +
5558 saved_syn->network_hdrlen;
5559 ret = saved_syn->tcp_hdrlen;
5560 } else if (optname == TCP_BPF_SYN_IP) {
5561 hdr_start = saved_syn->data +
5562 saved_syn->mac_hdrlen;
5563 ret = saved_syn->network_hdrlen +
5564 saved_syn->tcp_hdrlen;
5566 /* optname == TCP_BPF_SYN_MAC */
5568 /* TCP_SAVE_SYN may not have saved the mac hdr */
5569 if (!saved_syn->mac_hdrlen)
5572 hdr_start = saved_syn->data;
5573 ret = saved_syn->mac_hdrlen +
5574 saved_syn->network_hdrlen +
5575 saved_syn->tcp_hdrlen;
5583 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5584 int, level, int, optname, char *, optval, int, optlen)
5586 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5587 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5588 int ret, copy_len = 0;
5591 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5594 if (optlen < copy_len) {
5599 memcpy(optval, start, copy_len);
5602 /* Zero out unused buffer at the end */
5603 memset(optval + copy_len, 0, optlen - copy_len);
5608 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5611 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5612 .func = bpf_sock_ops_getsockopt,
5614 .ret_type = RET_INTEGER,
5615 .arg1_type = ARG_PTR_TO_CTX,
5616 .arg2_type = ARG_ANYTHING,
5617 .arg3_type = ARG_ANYTHING,
5618 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5619 .arg5_type = ARG_CONST_SIZE,
5622 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5625 struct sock *sk = bpf_sock->sk;
5626 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5628 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5631 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5633 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5636 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5637 .func = bpf_sock_ops_cb_flags_set,
5639 .ret_type = RET_INTEGER,
5640 .arg1_type = ARG_PTR_TO_CTX,
5641 .arg2_type = ARG_ANYTHING,
5644 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5645 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5647 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5651 struct sock *sk = ctx->sk;
5652 u32 flags = BIND_FROM_BPF;
5656 if (addr_len < offsetofend(struct sockaddr, sa_family))
5658 if (addr->sa_family == AF_INET) {
5659 if (addr_len < sizeof(struct sockaddr_in))
5661 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5662 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5663 return __inet_bind(sk, addr, addr_len, flags);
5664 #if IS_ENABLED(CONFIG_IPV6)
5665 } else if (addr->sa_family == AF_INET6) {
5666 if (addr_len < SIN6_LEN_RFC2133)
5668 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5669 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5670 /* ipv6_bpf_stub cannot be NULL, since it's called from
5671 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5673 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5674 #endif /* CONFIG_IPV6 */
5676 #endif /* CONFIG_INET */
5678 return -EAFNOSUPPORT;
5681 static const struct bpf_func_proto bpf_bind_proto = {
5684 .ret_type = RET_INTEGER,
5685 .arg1_type = ARG_PTR_TO_CTX,
5686 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5687 .arg3_type = ARG_CONST_SIZE,
5692 #if (IS_BUILTIN(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF)) || \
5693 (IS_MODULE(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES))
5695 struct metadata_dst __percpu *xfrm_bpf_md_dst;
5696 EXPORT_SYMBOL_GPL(xfrm_bpf_md_dst);
5700 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5701 struct bpf_xfrm_state *, to, u32, size, u64, flags)
5703 const struct sec_path *sp = skb_sec_path(skb);
5704 const struct xfrm_state *x;
5706 if (!sp || unlikely(index >= sp->len || flags))
5709 x = sp->xvec[index];
5711 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5714 to->reqid = x->props.reqid;
5715 to->spi = x->id.spi;
5716 to->family = x->props.family;
5719 if (to->family == AF_INET6) {
5720 memcpy(to->remote_ipv6, x->props.saddr.a6,
5721 sizeof(to->remote_ipv6));
5723 to->remote_ipv4 = x->props.saddr.a4;
5724 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5729 memset(to, 0, size);
5733 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5734 .func = bpf_skb_get_xfrm_state,
5736 .ret_type = RET_INTEGER,
5737 .arg1_type = ARG_PTR_TO_CTX,
5738 .arg2_type = ARG_ANYTHING,
5739 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
5740 .arg4_type = ARG_CONST_SIZE,
5741 .arg5_type = ARG_ANYTHING,
5745 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5746 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params, u32 mtu)
5748 params->h_vlan_TCI = 0;
5749 params->h_vlan_proto = 0;
5751 params->mtu_result = mtu; /* union with tot_len */
5757 #if IS_ENABLED(CONFIG_INET)
5758 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5759 u32 flags, bool check_mtu)
5761 struct fib_nh_common *nhc;
5762 struct in_device *in_dev;
5763 struct neighbour *neigh;
5764 struct net_device *dev;
5765 struct fib_result res;
5770 dev = dev_get_by_index_rcu(net, params->ifindex);
5774 /* verify forwarding is enabled on this interface */
5775 in_dev = __in_dev_get_rcu(dev);
5776 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5777 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5779 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5781 fl4.flowi4_oif = params->ifindex;
5783 fl4.flowi4_iif = params->ifindex;
5786 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5787 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5788 fl4.flowi4_flags = 0;
5790 fl4.flowi4_proto = params->l4_protocol;
5791 fl4.daddr = params->ipv4_dst;
5792 fl4.saddr = params->ipv4_src;
5793 fl4.fl4_sport = params->sport;
5794 fl4.fl4_dport = params->dport;
5795 fl4.flowi4_multipath_hash = 0;
5797 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5798 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5799 struct fib_table *tb;
5801 if (flags & BPF_FIB_LOOKUP_TBID) {
5802 tbid = params->tbid;
5803 /* zero out for vlan output */
5807 tb = fib_get_table(net, tbid);
5809 return BPF_FIB_LKUP_RET_NOT_FWDED;
5811 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5813 fl4.flowi4_mark = 0;
5814 fl4.flowi4_secid = 0;
5815 fl4.flowi4_tun_key.tun_id = 0;
5816 fl4.flowi4_uid = sock_net_uid(net, NULL);
5818 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5822 /* map fib lookup errors to RTN_ type */
5824 return BPF_FIB_LKUP_RET_BLACKHOLE;
5825 if (err == -EHOSTUNREACH)
5826 return BPF_FIB_LKUP_RET_UNREACHABLE;
5828 return BPF_FIB_LKUP_RET_PROHIBIT;
5830 return BPF_FIB_LKUP_RET_NOT_FWDED;
5833 if (res.type != RTN_UNICAST)
5834 return BPF_FIB_LKUP_RET_NOT_FWDED;
5836 if (fib_info_num_path(res.fi) > 1)
5837 fib_select_path(net, &res, &fl4, NULL);
5840 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5841 if (params->tot_len > mtu) {
5842 params->mtu_result = mtu; /* union with tot_len */
5843 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5849 /* do not handle lwt encaps right now */
5850 if (nhc->nhc_lwtstate)
5851 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5855 params->rt_metric = res.fi->fib_priority;
5856 params->ifindex = dev->ifindex;
5858 /* xdp and cls_bpf programs are run in RCU-bh so
5859 * rcu_read_lock_bh is not needed here
5861 if (likely(nhc->nhc_gw_family != AF_INET6)) {
5862 if (nhc->nhc_gw_family)
5863 params->ipv4_dst = nhc->nhc_gw.ipv4;
5865 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5867 params->family = AF_INET6;
5868 *dst = nhc->nhc_gw.ipv6;
5871 if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH)
5872 goto set_fwd_params;
5874 if (likely(nhc->nhc_gw_family != AF_INET6))
5875 neigh = __ipv4_neigh_lookup_noref(dev,
5876 (__force u32)params->ipv4_dst);
5878 neigh = __ipv6_neigh_lookup_noref_stub(dev, params->ipv6_dst);
5880 if (!neigh || !(READ_ONCE(neigh->nud_state) & NUD_VALID))
5881 return BPF_FIB_LKUP_RET_NO_NEIGH;
5882 memcpy(params->dmac, neigh->ha, ETH_ALEN);
5883 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5886 return bpf_fib_set_fwd_params(params, mtu);
5890 #if IS_ENABLED(CONFIG_IPV6)
5891 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5892 u32 flags, bool check_mtu)
5894 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5895 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5896 struct fib6_result res = {};
5897 struct neighbour *neigh;
5898 struct net_device *dev;
5899 struct inet6_dev *idev;
5905 /* link local addresses are never forwarded */
5906 if (rt6_need_strict(dst) || rt6_need_strict(src))
5907 return BPF_FIB_LKUP_RET_NOT_FWDED;
5909 dev = dev_get_by_index_rcu(net, params->ifindex);
5913 idev = __in6_dev_get_safely(dev);
5914 if (unlikely(!idev || !idev->cnf.forwarding))
5915 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5917 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5919 oif = fl6.flowi6_oif = params->ifindex;
5921 oif = fl6.flowi6_iif = params->ifindex;
5923 strict = RT6_LOOKUP_F_HAS_SADDR;
5925 fl6.flowlabel = params->flowinfo;
5926 fl6.flowi6_scope = 0;
5927 fl6.flowi6_flags = 0;
5930 fl6.flowi6_proto = params->l4_protocol;
5933 fl6.fl6_sport = params->sport;
5934 fl6.fl6_dport = params->dport;
5936 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5937 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5938 struct fib6_table *tb;
5940 if (flags & BPF_FIB_LOOKUP_TBID) {
5941 tbid = params->tbid;
5942 /* zero out for vlan output */
5946 tb = ipv6_stub->fib6_get_table(net, tbid);
5948 return BPF_FIB_LKUP_RET_NOT_FWDED;
5950 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5953 fl6.flowi6_mark = 0;
5954 fl6.flowi6_secid = 0;
5955 fl6.flowi6_tun_key.tun_id = 0;
5956 fl6.flowi6_uid = sock_net_uid(net, NULL);
5958 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5961 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5962 res.f6i == net->ipv6.fib6_null_entry))
5963 return BPF_FIB_LKUP_RET_NOT_FWDED;
5965 switch (res.fib6_type) {
5966 /* only unicast is forwarded */
5970 return BPF_FIB_LKUP_RET_BLACKHOLE;
5971 case RTN_UNREACHABLE:
5972 return BPF_FIB_LKUP_RET_UNREACHABLE;
5974 return BPF_FIB_LKUP_RET_PROHIBIT;
5976 return BPF_FIB_LKUP_RET_NOT_FWDED;
5979 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5980 fl6.flowi6_oif != 0, NULL, strict);
5983 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5984 if (params->tot_len > mtu) {
5985 params->mtu_result = mtu; /* union with tot_len */
5986 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5990 if (res.nh->fib_nh_lws)
5991 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5993 if (res.nh->fib_nh_gw_family)
5994 *dst = res.nh->fib_nh_gw6;
5996 dev = res.nh->fib_nh_dev;
5997 params->rt_metric = res.f6i->fib6_metric;
5998 params->ifindex = dev->ifindex;
6000 if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH)
6001 goto set_fwd_params;
6003 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
6006 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
6007 if (!neigh || !(READ_ONCE(neigh->nud_state) & NUD_VALID))
6008 return BPF_FIB_LKUP_RET_NO_NEIGH;
6009 memcpy(params->dmac, neigh->ha, ETH_ALEN);
6010 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
6013 return bpf_fib_set_fwd_params(params, mtu);
6017 #define BPF_FIB_LOOKUP_MASK (BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT | \
6018 BPF_FIB_LOOKUP_SKIP_NEIGH | BPF_FIB_LOOKUP_TBID)
6020 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
6021 struct bpf_fib_lookup *, params, int, plen, u32, flags)
6023 if (plen < sizeof(*params))
6026 if (flags & ~BPF_FIB_LOOKUP_MASK)
6029 switch (params->family) {
6030 #if IS_ENABLED(CONFIG_INET)
6032 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
6035 #if IS_ENABLED(CONFIG_IPV6)
6037 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
6041 return -EAFNOSUPPORT;
6044 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
6045 .func = bpf_xdp_fib_lookup,
6047 .ret_type = RET_INTEGER,
6048 .arg1_type = ARG_PTR_TO_CTX,
6049 .arg2_type = ARG_PTR_TO_MEM,
6050 .arg3_type = ARG_CONST_SIZE,
6051 .arg4_type = ARG_ANYTHING,
6054 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
6055 struct bpf_fib_lookup *, params, int, plen, u32, flags)
6057 struct net *net = dev_net(skb->dev);
6058 int rc = -EAFNOSUPPORT;
6059 bool check_mtu = false;
6061 if (plen < sizeof(*params))
6064 if (flags & ~BPF_FIB_LOOKUP_MASK)
6067 if (params->tot_len)
6070 switch (params->family) {
6071 #if IS_ENABLED(CONFIG_INET)
6073 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
6076 #if IS_ENABLED(CONFIG_IPV6)
6078 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
6083 if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
6084 struct net_device *dev;
6086 /* When tot_len isn't provided by user, check skb
6087 * against MTU of FIB lookup resulting net_device
6089 dev = dev_get_by_index_rcu(net, params->ifindex);
6090 if (!is_skb_forwardable(dev, skb))
6091 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
6093 params->mtu_result = dev->mtu; /* union with tot_len */
6099 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
6100 .func = bpf_skb_fib_lookup,
6102 .ret_type = RET_INTEGER,
6103 .arg1_type = ARG_PTR_TO_CTX,
6104 .arg2_type = ARG_PTR_TO_MEM,
6105 .arg3_type = ARG_CONST_SIZE,
6106 .arg4_type = ARG_ANYTHING,
6109 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
6112 struct net *netns = dev_net(dev_curr);
6114 /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
6118 return dev_get_by_index_rcu(netns, ifindex);
6121 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
6122 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6124 int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6125 struct net_device *dev = skb->dev;
6126 int skb_len, dev_len;
6129 if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
6132 if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
6135 dev = __dev_via_ifindex(dev, ifindex);
6139 mtu = READ_ONCE(dev->mtu);
6141 dev_len = mtu + dev->hard_header_len;
6143 /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6144 skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
6146 skb_len += len_diff; /* minus result pass check */
6147 if (skb_len <= dev_len) {
6148 ret = BPF_MTU_CHK_RET_SUCCESS;
6151 /* At this point, skb->len exceed MTU, but as it include length of all
6152 * segments, it can still be below MTU. The SKB can possibly get
6153 * re-segmented in transmit path (see validate_xmit_skb). Thus, user
6154 * must choose if segs are to be MTU checked.
6156 if (skb_is_gso(skb)) {
6157 ret = BPF_MTU_CHK_RET_SUCCESS;
6159 if (flags & BPF_MTU_CHK_SEGS &&
6160 !skb_gso_validate_network_len(skb, mtu))
6161 ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
6164 /* BPF verifier guarantees valid pointer */
6170 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
6171 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6173 struct net_device *dev = xdp->rxq->dev;
6174 int xdp_len = xdp->data_end - xdp->data;
6175 int ret = BPF_MTU_CHK_RET_SUCCESS;
6178 /* XDP variant doesn't support multi-buffer segment check (yet) */
6179 if (unlikely(flags))
6182 dev = __dev_via_ifindex(dev, ifindex);
6186 mtu = READ_ONCE(dev->mtu);
6188 /* Add L2-header as dev MTU is L3 size */
6189 dev_len = mtu + dev->hard_header_len;
6191 /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6193 xdp_len = *mtu_len + dev->hard_header_len;
6195 xdp_len += len_diff; /* minus result pass check */
6196 if (xdp_len > dev_len)
6197 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6199 /* BPF verifier guarantees valid pointer */
6205 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
6206 .func = bpf_skb_check_mtu,
6208 .ret_type = RET_INTEGER,
6209 .arg1_type = ARG_PTR_TO_CTX,
6210 .arg2_type = ARG_ANYTHING,
6211 .arg3_type = ARG_PTR_TO_INT,
6212 .arg4_type = ARG_ANYTHING,
6213 .arg5_type = ARG_ANYTHING,
6216 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
6217 .func = bpf_xdp_check_mtu,
6219 .ret_type = RET_INTEGER,
6220 .arg1_type = ARG_PTR_TO_CTX,
6221 .arg2_type = ARG_ANYTHING,
6222 .arg3_type = ARG_PTR_TO_INT,
6223 .arg4_type = ARG_ANYTHING,
6224 .arg5_type = ARG_ANYTHING,
6227 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6228 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
6231 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
6233 if (!seg6_validate_srh(srh, len, false))
6237 case BPF_LWT_ENCAP_SEG6_INLINE:
6238 if (skb->protocol != htons(ETH_P_IPV6))
6241 err = seg6_do_srh_inline(skb, srh);
6243 case BPF_LWT_ENCAP_SEG6:
6244 skb_reset_inner_headers(skb);
6245 skb->encapsulation = 1;
6246 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
6252 bpf_compute_data_pointers(skb);
6256 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
6258 return seg6_lookup_nexthop(skb, NULL, 0);
6260 #endif /* CONFIG_IPV6_SEG6_BPF */
6262 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6263 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
6266 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
6270 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
6274 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6275 case BPF_LWT_ENCAP_SEG6:
6276 case BPF_LWT_ENCAP_SEG6_INLINE:
6277 return bpf_push_seg6_encap(skb, type, hdr, len);
6279 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6280 case BPF_LWT_ENCAP_IP:
6281 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
6288 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
6289 void *, hdr, u32, len)
6292 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6293 case BPF_LWT_ENCAP_IP:
6294 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
6301 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
6302 .func = bpf_lwt_in_push_encap,
6304 .ret_type = RET_INTEGER,
6305 .arg1_type = ARG_PTR_TO_CTX,
6306 .arg2_type = ARG_ANYTHING,
6307 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6308 .arg4_type = ARG_CONST_SIZE
6311 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
6312 .func = bpf_lwt_xmit_push_encap,
6314 .ret_type = RET_INTEGER,
6315 .arg1_type = ARG_PTR_TO_CTX,
6316 .arg2_type = ARG_ANYTHING,
6317 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6318 .arg4_type = ARG_CONST_SIZE
6321 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6322 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
6323 const void *, from, u32, len)
6325 struct seg6_bpf_srh_state *srh_state =
6326 this_cpu_ptr(&seg6_bpf_srh_states);
6327 struct ipv6_sr_hdr *srh = srh_state->srh;
6328 void *srh_tlvs, *srh_end, *ptr;
6334 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
6335 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
6337 ptr = skb->data + offset;
6338 if (ptr >= srh_tlvs && ptr + len <= srh_end)
6339 srh_state->valid = false;
6340 else if (ptr < (void *)&srh->flags ||
6341 ptr + len > (void *)&srh->segments)
6344 if (unlikely(bpf_try_make_writable(skb, offset + len)))
6346 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6348 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6350 memcpy(skb->data + offset, from, len);
6354 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
6355 .func = bpf_lwt_seg6_store_bytes,
6357 .ret_type = RET_INTEGER,
6358 .arg1_type = ARG_PTR_TO_CTX,
6359 .arg2_type = ARG_ANYTHING,
6360 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6361 .arg4_type = ARG_CONST_SIZE
6364 static void bpf_update_srh_state(struct sk_buff *skb)
6366 struct seg6_bpf_srh_state *srh_state =
6367 this_cpu_ptr(&seg6_bpf_srh_states);
6370 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
6371 srh_state->srh = NULL;
6373 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6374 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
6375 srh_state->valid = true;
6379 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
6380 u32, action, void *, param, u32, param_len)
6382 struct seg6_bpf_srh_state *srh_state =
6383 this_cpu_ptr(&seg6_bpf_srh_states);
6388 case SEG6_LOCAL_ACTION_END_X:
6389 if (!seg6_bpf_has_valid_srh(skb))
6391 if (param_len != sizeof(struct in6_addr))
6393 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
6394 case SEG6_LOCAL_ACTION_END_T:
6395 if (!seg6_bpf_has_valid_srh(skb))
6397 if (param_len != sizeof(int))
6399 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6400 case SEG6_LOCAL_ACTION_END_DT6:
6401 if (!seg6_bpf_has_valid_srh(skb))
6403 if (param_len != sizeof(int))
6406 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
6408 if (!pskb_pull(skb, hdroff))
6411 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
6412 skb_reset_network_header(skb);
6413 skb_reset_transport_header(skb);
6414 skb->encapsulation = 0;
6416 bpf_compute_data_pointers(skb);
6417 bpf_update_srh_state(skb);
6418 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6419 case SEG6_LOCAL_ACTION_END_B6:
6420 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6422 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
6425 bpf_update_srh_state(skb);
6428 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
6429 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6431 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
6434 bpf_update_srh_state(skb);
6442 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
6443 .func = bpf_lwt_seg6_action,
6445 .ret_type = RET_INTEGER,
6446 .arg1_type = ARG_PTR_TO_CTX,
6447 .arg2_type = ARG_ANYTHING,
6448 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6449 .arg4_type = ARG_CONST_SIZE
6452 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
6455 struct seg6_bpf_srh_state *srh_state =
6456 this_cpu_ptr(&seg6_bpf_srh_states);
6457 struct ipv6_sr_hdr *srh = srh_state->srh;
6458 void *srh_end, *srh_tlvs, *ptr;
6459 struct ipv6hdr *hdr;
6463 if (unlikely(srh == NULL))
6466 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
6467 ((srh->first_segment + 1) << 4));
6468 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
6470 ptr = skb->data + offset;
6472 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
6474 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
6478 ret = skb_cow_head(skb, len);
6479 if (unlikely(ret < 0))
6482 ret = bpf_skb_net_hdr_push(skb, offset, len);
6484 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6487 bpf_compute_data_pointers(skb);
6488 if (unlikely(ret < 0))
6491 hdr = (struct ipv6hdr *)skb->data;
6492 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6494 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6496 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6497 srh_state->hdrlen += len;
6498 srh_state->valid = false;
6502 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6503 .func = bpf_lwt_seg6_adjust_srh,
6505 .ret_type = RET_INTEGER,
6506 .arg1_type = ARG_PTR_TO_CTX,
6507 .arg2_type = ARG_ANYTHING,
6508 .arg3_type = ARG_ANYTHING,
6510 #endif /* CONFIG_IPV6_SEG6_BPF */
6513 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6514 int dif, int sdif, u8 family, u8 proto)
6516 struct inet_hashinfo *hinfo = net->ipv4.tcp_death_row.hashinfo;
6517 bool refcounted = false;
6518 struct sock *sk = NULL;
6520 if (family == AF_INET) {
6521 __be32 src4 = tuple->ipv4.saddr;
6522 __be32 dst4 = tuple->ipv4.daddr;
6524 if (proto == IPPROTO_TCP)
6525 sk = __inet_lookup(net, hinfo, NULL, 0,
6526 src4, tuple->ipv4.sport,
6527 dst4, tuple->ipv4.dport,
6528 dif, sdif, &refcounted);
6530 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6531 dst4, tuple->ipv4.dport,
6532 dif, sdif, net->ipv4.udp_table, NULL);
6533 #if IS_ENABLED(CONFIG_IPV6)
6535 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6536 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6538 if (proto == IPPROTO_TCP)
6539 sk = __inet6_lookup(net, hinfo, NULL, 0,
6540 src6, tuple->ipv6.sport,
6541 dst6, ntohs(tuple->ipv6.dport),
6542 dif, sdif, &refcounted);
6543 else if (likely(ipv6_bpf_stub))
6544 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6545 src6, tuple->ipv6.sport,
6546 dst6, tuple->ipv6.dport,
6548 net->ipv4.udp_table, NULL);
6552 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6553 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6559 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6560 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6562 static struct sock *
6563 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6564 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6565 u64 flags, int sdif)
6567 struct sock *sk = NULL;
6571 if (len == sizeof(tuple->ipv4))
6573 else if (len == sizeof(tuple->ipv6))
6578 if (unlikely(flags || !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6582 if (family == AF_INET)
6583 sdif = inet_sdif(skb);
6585 sdif = inet6_sdif(skb);
6588 if ((s32)netns_id < 0) {
6590 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6592 net = get_net_ns_by_id(caller_net, netns_id);
6595 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6603 static struct sock *
6604 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6605 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6606 u64 flags, int sdif)
6608 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6609 ifindex, proto, netns_id, flags,
6613 struct sock *sk2 = sk_to_full_sk(sk);
6615 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6616 * sock refcnt is decremented to prevent a request_sock leak.
6618 if (!sk_fullsock(sk2))
6622 /* Ensure there is no need to bump sk2 refcnt */
6623 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6624 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6634 static struct sock *
6635 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6636 u8 proto, u64 netns_id, u64 flags)
6638 struct net *caller_net;
6642 caller_net = dev_net(skb->dev);
6643 ifindex = skb->dev->ifindex;
6645 caller_net = sock_net(skb->sk);
6649 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6650 netns_id, flags, -1);
6653 static struct sock *
6654 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6655 u8 proto, u64 netns_id, u64 flags)
6657 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6661 struct sock *sk2 = sk_to_full_sk(sk);
6663 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6664 * sock refcnt is decremented to prevent a request_sock leak.
6666 if (!sk_fullsock(sk2))
6670 /* Ensure there is no need to bump sk2 refcnt */
6671 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6672 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6682 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6683 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6685 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6689 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6690 .func = bpf_skc_lookup_tcp,
6693 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6694 .arg1_type = ARG_PTR_TO_CTX,
6695 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6696 .arg3_type = ARG_CONST_SIZE,
6697 .arg4_type = ARG_ANYTHING,
6698 .arg5_type = ARG_ANYTHING,
6701 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6702 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6704 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6708 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6709 .func = bpf_sk_lookup_tcp,
6712 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6713 .arg1_type = ARG_PTR_TO_CTX,
6714 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6715 .arg3_type = ARG_CONST_SIZE,
6716 .arg4_type = ARG_ANYTHING,
6717 .arg5_type = ARG_ANYTHING,
6720 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6721 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6723 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6727 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6728 .func = bpf_sk_lookup_udp,
6731 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6732 .arg1_type = ARG_PTR_TO_CTX,
6733 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6734 .arg3_type = ARG_CONST_SIZE,
6735 .arg4_type = ARG_ANYTHING,
6736 .arg5_type = ARG_ANYTHING,
6739 BPF_CALL_5(bpf_tc_skc_lookup_tcp, struct sk_buff *, skb,
6740 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6742 struct net_device *dev = skb->dev;
6743 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6744 struct net *caller_net = dev_net(dev);
6746 return (unsigned long)__bpf_skc_lookup(skb, tuple, len, caller_net,
6747 ifindex, IPPROTO_TCP, netns_id,
6751 static const struct bpf_func_proto bpf_tc_skc_lookup_tcp_proto = {
6752 .func = bpf_tc_skc_lookup_tcp,
6755 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6756 .arg1_type = ARG_PTR_TO_CTX,
6757 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6758 .arg3_type = ARG_CONST_SIZE,
6759 .arg4_type = ARG_ANYTHING,
6760 .arg5_type = ARG_ANYTHING,
6763 BPF_CALL_5(bpf_tc_sk_lookup_tcp, struct sk_buff *, skb,
6764 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6766 struct net_device *dev = skb->dev;
6767 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6768 struct net *caller_net = dev_net(dev);
6770 return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6771 ifindex, IPPROTO_TCP, netns_id,
6775 static const struct bpf_func_proto bpf_tc_sk_lookup_tcp_proto = {
6776 .func = bpf_tc_sk_lookup_tcp,
6779 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6780 .arg1_type = ARG_PTR_TO_CTX,
6781 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6782 .arg3_type = ARG_CONST_SIZE,
6783 .arg4_type = ARG_ANYTHING,
6784 .arg5_type = ARG_ANYTHING,
6787 BPF_CALL_5(bpf_tc_sk_lookup_udp, struct sk_buff *, skb,
6788 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6790 struct net_device *dev = skb->dev;
6791 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6792 struct net *caller_net = dev_net(dev);
6794 return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6795 ifindex, IPPROTO_UDP, netns_id,
6799 static const struct bpf_func_proto bpf_tc_sk_lookup_udp_proto = {
6800 .func = bpf_tc_sk_lookup_udp,
6803 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6804 .arg1_type = ARG_PTR_TO_CTX,
6805 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6806 .arg3_type = ARG_CONST_SIZE,
6807 .arg4_type = ARG_ANYTHING,
6808 .arg5_type = ARG_ANYTHING,
6811 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6813 if (sk && sk_is_refcounted(sk))
6818 static const struct bpf_func_proto bpf_sk_release_proto = {
6819 .func = bpf_sk_release,
6821 .ret_type = RET_INTEGER,
6822 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON | OBJ_RELEASE,
6825 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6826 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6828 struct net_device *dev = ctx->rxq->dev;
6829 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6830 struct net *caller_net = dev_net(dev);
6832 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6833 ifindex, IPPROTO_UDP, netns_id,
6837 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6838 .func = bpf_xdp_sk_lookup_udp,
6841 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6842 .arg1_type = ARG_PTR_TO_CTX,
6843 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6844 .arg3_type = ARG_CONST_SIZE,
6845 .arg4_type = ARG_ANYTHING,
6846 .arg5_type = ARG_ANYTHING,
6849 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6850 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6852 struct net_device *dev = ctx->rxq->dev;
6853 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6854 struct net *caller_net = dev_net(dev);
6856 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6857 ifindex, IPPROTO_TCP, netns_id,
6861 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6862 .func = bpf_xdp_skc_lookup_tcp,
6865 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6866 .arg1_type = ARG_PTR_TO_CTX,
6867 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6868 .arg3_type = ARG_CONST_SIZE,
6869 .arg4_type = ARG_ANYTHING,
6870 .arg5_type = ARG_ANYTHING,
6873 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6874 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6876 struct net_device *dev = ctx->rxq->dev;
6877 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6878 struct net *caller_net = dev_net(dev);
6880 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6881 ifindex, IPPROTO_TCP, netns_id,
6885 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6886 .func = bpf_xdp_sk_lookup_tcp,
6889 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6890 .arg1_type = ARG_PTR_TO_CTX,
6891 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6892 .arg3_type = ARG_CONST_SIZE,
6893 .arg4_type = ARG_ANYTHING,
6894 .arg5_type = ARG_ANYTHING,
6897 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6898 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6900 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6901 sock_net(ctx->sk), 0,
6902 IPPROTO_TCP, netns_id, flags,
6906 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6907 .func = bpf_sock_addr_skc_lookup_tcp,
6909 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6910 .arg1_type = ARG_PTR_TO_CTX,
6911 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6912 .arg3_type = ARG_CONST_SIZE,
6913 .arg4_type = ARG_ANYTHING,
6914 .arg5_type = ARG_ANYTHING,
6917 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6918 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6920 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6921 sock_net(ctx->sk), 0, IPPROTO_TCP,
6922 netns_id, flags, -1);
6925 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6926 .func = bpf_sock_addr_sk_lookup_tcp,
6928 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6929 .arg1_type = ARG_PTR_TO_CTX,
6930 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6931 .arg3_type = ARG_CONST_SIZE,
6932 .arg4_type = ARG_ANYTHING,
6933 .arg5_type = ARG_ANYTHING,
6936 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6937 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6939 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6940 sock_net(ctx->sk), 0, IPPROTO_UDP,
6941 netns_id, flags, -1);
6944 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6945 .func = bpf_sock_addr_sk_lookup_udp,
6947 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6948 .arg1_type = ARG_PTR_TO_CTX,
6949 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6950 .arg3_type = ARG_CONST_SIZE,
6951 .arg4_type = ARG_ANYTHING,
6952 .arg5_type = ARG_ANYTHING,
6955 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6956 struct bpf_insn_access_aux *info)
6958 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6962 if (off % size != 0)
6966 case offsetof(struct bpf_tcp_sock, bytes_received):
6967 case offsetof(struct bpf_tcp_sock, bytes_acked):
6968 return size == sizeof(__u64);
6970 return size == sizeof(__u32);
6974 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6975 const struct bpf_insn *si,
6976 struct bpf_insn *insn_buf,
6977 struct bpf_prog *prog, u32 *target_size)
6979 struct bpf_insn *insn = insn_buf;
6981 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \
6983 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \
6984 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6985 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6986 si->dst_reg, si->src_reg, \
6987 offsetof(struct tcp_sock, FIELD)); \
6990 #define BPF_INET_SOCK_GET_COMMON(FIELD) \
6992 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \
6994 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6995 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6996 struct inet_connection_sock, \
6998 si->dst_reg, si->src_reg, \
7000 struct inet_connection_sock, \
7004 BTF_TYPE_EMIT(struct bpf_tcp_sock);
7007 case offsetof(struct bpf_tcp_sock, rtt_min):
7008 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
7009 sizeof(struct minmax));
7010 BUILD_BUG_ON(sizeof(struct minmax) <
7011 sizeof(struct minmax_sample));
7013 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7014 offsetof(struct tcp_sock, rtt_min) +
7015 offsetof(struct minmax_sample, v));
7017 case offsetof(struct bpf_tcp_sock, snd_cwnd):
7018 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
7020 case offsetof(struct bpf_tcp_sock, srtt_us):
7021 BPF_TCP_SOCK_GET_COMMON(srtt_us);
7023 case offsetof(struct bpf_tcp_sock, snd_ssthresh):
7024 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
7026 case offsetof(struct bpf_tcp_sock, rcv_nxt):
7027 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
7029 case offsetof(struct bpf_tcp_sock, snd_nxt):
7030 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
7032 case offsetof(struct bpf_tcp_sock, snd_una):
7033 BPF_TCP_SOCK_GET_COMMON(snd_una);
7035 case offsetof(struct bpf_tcp_sock, mss_cache):
7036 BPF_TCP_SOCK_GET_COMMON(mss_cache);
7038 case offsetof(struct bpf_tcp_sock, ecn_flags):
7039 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
7041 case offsetof(struct bpf_tcp_sock, rate_delivered):
7042 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
7044 case offsetof(struct bpf_tcp_sock, rate_interval_us):
7045 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
7047 case offsetof(struct bpf_tcp_sock, packets_out):
7048 BPF_TCP_SOCK_GET_COMMON(packets_out);
7050 case offsetof(struct bpf_tcp_sock, retrans_out):
7051 BPF_TCP_SOCK_GET_COMMON(retrans_out);
7053 case offsetof(struct bpf_tcp_sock, total_retrans):
7054 BPF_TCP_SOCK_GET_COMMON(total_retrans);
7056 case offsetof(struct bpf_tcp_sock, segs_in):
7057 BPF_TCP_SOCK_GET_COMMON(segs_in);
7059 case offsetof(struct bpf_tcp_sock, data_segs_in):
7060 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
7062 case offsetof(struct bpf_tcp_sock, segs_out):
7063 BPF_TCP_SOCK_GET_COMMON(segs_out);
7065 case offsetof(struct bpf_tcp_sock, data_segs_out):
7066 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
7068 case offsetof(struct bpf_tcp_sock, lost_out):
7069 BPF_TCP_SOCK_GET_COMMON(lost_out);
7071 case offsetof(struct bpf_tcp_sock, sacked_out):
7072 BPF_TCP_SOCK_GET_COMMON(sacked_out);
7074 case offsetof(struct bpf_tcp_sock, bytes_received):
7075 BPF_TCP_SOCK_GET_COMMON(bytes_received);
7077 case offsetof(struct bpf_tcp_sock, bytes_acked):
7078 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
7080 case offsetof(struct bpf_tcp_sock, dsack_dups):
7081 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
7083 case offsetof(struct bpf_tcp_sock, delivered):
7084 BPF_TCP_SOCK_GET_COMMON(delivered);
7086 case offsetof(struct bpf_tcp_sock, delivered_ce):
7087 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
7089 case offsetof(struct bpf_tcp_sock, icsk_retransmits):
7090 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
7094 return insn - insn_buf;
7097 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
7099 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
7100 return (unsigned long)sk;
7102 return (unsigned long)NULL;
7105 const struct bpf_func_proto bpf_tcp_sock_proto = {
7106 .func = bpf_tcp_sock,
7108 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
7109 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
7112 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
7114 sk = sk_to_full_sk(sk);
7116 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
7117 return (unsigned long)sk;
7119 return (unsigned long)NULL;
7122 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
7123 .func = bpf_get_listener_sock,
7125 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
7126 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
7129 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
7131 unsigned int iphdr_len;
7133 switch (skb_protocol(skb, true)) {
7134 case cpu_to_be16(ETH_P_IP):
7135 iphdr_len = sizeof(struct iphdr);
7137 case cpu_to_be16(ETH_P_IPV6):
7138 iphdr_len = sizeof(struct ipv6hdr);
7144 if (skb_headlen(skb) < iphdr_len)
7147 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
7150 return INET_ECN_set_ce(skb);
7153 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7154 struct bpf_insn_access_aux *info)
7156 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
7159 if (off % size != 0)
7164 return size == sizeof(__u32);
7168 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
7169 const struct bpf_insn *si,
7170 struct bpf_insn *insn_buf,
7171 struct bpf_prog *prog, u32 *target_size)
7173 struct bpf_insn *insn = insn_buf;
7175 #define BPF_XDP_SOCK_GET(FIELD) \
7177 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \
7178 sizeof_field(struct bpf_xdp_sock, FIELD)); \
7179 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
7180 si->dst_reg, si->src_reg, \
7181 offsetof(struct xdp_sock, FIELD)); \
7185 case offsetof(struct bpf_xdp_sock, queue_id):
7186 BPF_XDP_SOCK_GET(queue_id);
7190 return insn - insn_buf;
7193 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
7194 .func = bpf_skb_ecn_set_ce,
7196 .ret_type = RET_INTEGER,
7197 .arg1_type = ARG_PTR_TO_CTX,
7200 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7201 struct tcphdr *, th, u32, th_len)
7203 #ifdef CONFIG_SYN_COOKIES
7207 if (unlikely(!sk || th_len < sizeof(*th)))
7210 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
7211 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7214 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7217 if (!th->ack || th->rst || th->syn)
7220 if (unlikely(iph_len < sizeof(struct iphdr)))
7223 if (tcp_synq_no_recent_overflow(sk))
7226 cookie = ntohl(th->ack_seq) - 1;
7228 /* Both struct iphdr and struct ipv6hdr have the version field at the
7229 * same offset so we can cast to the shorter header (struct iphdr).
7231 switch (((struct iphdr *)iph)->version) {
7233 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7236 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
7239 #if IS_BUILTIN(CONFIG_IPV6)
7241 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7244 if (sk->sk_family != AF_INET6)
7247 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
7249 #endif /* CONFIG_IPV6 */
7252 return -EPROTONOSUPPORT;
7264 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
7265 .func = bpf_tcp_check_syncookie,
7268 .ret_type = RET_INTEGER,
7269 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7270 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7271 .arg3_type = ARG_CONST_SIZE,
7272 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7273 .arg5_type = ARG_CONST_SIZE,
7276 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7277 struct tcphdr *, th, u32, th_len)
7279 #ifdef CONFIG_SYN_COOKIES
7283 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
7286 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7289 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7292 if (!th->syn || th->ack || th->fin || th->rst)
7295 if (unlikely(iph_len < sizeof(struct iphdr)))
7298 /* Both struct iphdr and struct ipv6hdr have the version field at the
7299 * same offset so we can cast to the shorter header (struct iphdr).
7301 switch (((struct iphdr *)iph)->version) {
7303 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7306 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
7309 #if IS_BUILTIN(CONFIG_IPV6)
7311 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7314 if (sk->sk_family != AF_INET6)
7317 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
7319 #endif /* CONFIG_IPV6 */
7322 return -EPROTONOSUPPORT;
7327 return cookie | ((u64)mss << 32);
7330 #endif /* CONFIG_SYN_COOKIES */
7333 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
7334 .func = bpf_tcp_gen_syncookie,
7335 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */
7337 .ret_type = RET_INTEGER,
7338 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7339 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7340 .arg3_type = ARG_CONST_SIZE,
7341 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7342 .arg5_type = ARG_CONST_SIZE,
7345 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
7347 if (!sk || flags != 0)
7349 if (!skb_at_tc_ingress(skb))
7351 if (unlikely(dev_net(skb->dev) != sock_net(sk)))
7352 return -ENETUNREACH;
7353 if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
7354 return -ESOCKTNOSUPPORT;
7355 if (sk_is_refcounted(sk) &&
7356 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
7361 skb->destructor = sock_pfree;
7366 static const struct bpf_func_proto bpf_sk_assign_proto = {
7367 .func = bpf_sk_assign,
7369 .ret_type = RET_INTEGER,
7370 .arg1_type = ARG_PTR_TO_CTX,
7371 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7372 .arg3_type = ARG_ANYTHING,
7375 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
7376 u8 search_kind, const u8 *magic,
7377 u8 magic_len, bool *eol)
7383 while (op < opend) {
7386 if (kind == TCPOPT_EOL) {
7388 return ERR_PTR(-ENOMSG);
7389 } else if (kind == TCPOPT_NOP) {
7394 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
7395 /* Something is wrong in the received header.
7396 * Follow the TCP stack's tcp_parse_options()
7397 * and just bail here.
7399 return ERR_PTR(-EFAULT);
7402 if (search_kind == kind) {
7406 if (magic_len > kind_len - 2)
7407 return ERR_PTR(-ENOMSG);
7409 if (!memcmp(&op[2], magic, magic_len))
7416 return ERR_PTR(-ENOMSG);
7419 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7420 void *, search_res, u32, len, u64, flags)
7422 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
7423 const u8 *op, *opend, *magic, *search = search_res;
7424 u8 search_kind, search_len, copy_len, magic_len;
7427 /* 2 byte is the minimal option len except TCPOPT_NOP and
7428 * TCPOPT_EOL which are useless for the bpf prog to learn
7429 * and this helper disallow loading them also.
7431 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
7434 search_kind = search[0];
7435 search_len = search[1];
7437 if (search_len > len || search_kind == TCPOPT_NOP ||
7438 search_kind == TCPOPT_EOL)
7441 if (search_kind == TCPOPT_EXP || search_kind == 253) {
7442 /* 16 or 32 bit magic. +2 for kind and kind length */
7443 if (search_len != 4 && search_len != 6)
7446 magic_len = search_len - 2;
7455 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
7460 op += sizeof(struct tcphdr);
7462 if (!bpf_sock->skb ||
7463 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7464 /* This bpf_sock->op cannot call this helper */
7467 opend = bpf_sock->skb_data_end;
7468 op = bpf_sock->skb->data + sizeof(struct tcphdr);
7471 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
7478 if (copy_len > len) {
7483 memcpy(search_res, op, copy_len);
7487 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
7488 .func = bpf_sock_ops_load_hdr_opt,
7490 .ret_type = RET_INTEGER,
7491 .arg1_type = ARG_PTR_TO_CTX,
7492 .arg2_type = ARG_PTR_TO_MEM,
7493 .arg3_type = ARG_CONST_SIZE,
7494 .arg4_type = ARG_ANYTHING,
7497 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7498 const void *, from, u32, len, u64, flags)
7500 u8 new_kind, new_kind_len, magic_len = 0, *opend;
7501 const u8 *op, *new_op, *magic = NULL;
7502 struct sk_buff *skb;
7505 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
7508 if (len < 2 || flags)
7512 new_kind = new_op[0];
7513 new_kind_len = new_op[1];
7515 if (new_kind_len > len || new_kind == TCPOPT_NOP ||
7516 new_kind == TCPOPT_EOL)
7519 if (new_kind_len > bpf_sock->remaining_opt_len)
7522 /* 253 is another experimental kind */
7523 if (new_kind == TCPOPT_EXP || new_kind == 253) {
7524 if (new_kind_len < 4)
7526 /* Match for the 2 byte magic also.
7527 * RFC 6994: the magic could be 2 or 4 bytes.
7528 * Hence, matching by 2 byte only is on the
7529 * conservative side but it is the right
7530 * thing to do for the 'search-for-duplication'
7537 /* Check for duplication */
7538 skb = bpf_sock->skb;
7539 op = skb->data + sizeof(struct tcphdr);
7540 opend = bpf_sock->skb_data_end;
7542 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
7547 if (PTR_ERR(op) != -ENOMSG)
7551 /* The option has been ended. Treat it as no more
7552 * header option can be written.
7556 /* No duplication found. Store the header option. */
7557 memcpy(opend, from, new_kind_len);
7559 bpf_sock->remaining_opt_len -= new_kind_len;
7560 bpf_sock->skb_data_end += new_kind_len;
7565 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
7566 .func = bpf_sock_ops_store_hdr_opt,
7568 .ret_type = RET_INTEGER,
7569 .arg1_type = ARG_PTR_TO_CTX,
7570 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7571 .arg3_type = ARG_CONST_SIZE,
7572 .arg4_type = ARG_ANYTHING,
7575 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7576 u32, len, u64, flags)
7578 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7581 if (flags || len < 2)
7584 if (len > bpf_sock->remaining_opt_len)
7587 bpf_sock->remaining_opt_len -= len;
7592 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7593 .func = bpf_sock_ops_reserve_hdr_opt,
7595 .ret_type = RET_INTEGER,
7596 .arg1_type = ARG_PTR_TO_CTX,
7597 .arg2_type = ARG_ANYTHING,
7598 .arg3_type = ARG_ANYTHING,
7601 BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb,
7602 u64, tstamp, u32, tstamp_type)
7604 /* skb_clear_delivery_time() is done for inet protocol */
7605 if (skb->protocol != htons(ETH_P_IP) &&
7606 skb->protocol != htons(ETH_P_IPV6))
7609 switch (tstamp_type) {
7610 case BPF_SKB_TSTAMP_DELIVERY_MONO:
7613 skb->tstamp = tstamp;
7614 skb->mono_delivery_time = 1;
7616 case BPF_SKB_TSTAMP_UNSPEC:
7620 skb->mono_delivery_time = 0;
7629 static const struct bpf_func_proto bpf_skb_set_tstamp_proto = {
7630 .func = bpf_skb_set_tstamp,
7632 .ret_type = RET_INTEGER,
7633 .arg1_type = ARG_PTR_TO_CTX,
7634 .arg2_type = ARG_ANYTHING,
7635 .arg3_type = ARG_ANYTHING,
7638 #ifdef CONFIG_SYN_COOKIES
7639 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv4, struct iphdr *, iph,
7640 struct tcphdr *, th, u32, th_len)
7645 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7648 mss = tcp_parse_mss_option(th, 0) ?: TCP_MSS_DEFAULT;
7649 cookie = __cookie_v4_init_sequence(iph, th, &mss);
7651 return cookie | ((u64)mss << 32);
7654 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv4_proto = {
7655 .func = bpf_tcp_raw_gen_syncookie_ipv4,
7656 .gpl_only = true, /* __cookie_v4_init_sequence() is GPL */
7658 .ret_type = RET_INTEGER,
7659 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7660 .arg1_size = sizeof(struct iphdr),
7661 .arg2_type = ARG_PTR_TO_MEM,
7662 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7665 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv6, struct ipv6hdr *, iph,
7666 struct tcphdr *, th, u32, th_len)
7668 #if IS_BUILTIN(CONFIG_IPV6)
7669 const u16 mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
7670 sizeof(struct ipv6hdr);
7674 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7677 mss = tcp_parse_mss_option(th, 0) ?: mss_clamp;
7678 cookie = __cookie_v6_init_sequence(iph, th, &mss);
7680 return cookie | ((u64)mss << 32);
7682 return -EPROTONOSUPPORT;
7686 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv6_proto = {
7687 .func = bpf_tcp_raw_gen_syncookie_ipv6,
7688 .gpl_only = true, /* __cookie_v6_init_sequence() is GPL */
7690 .ret_type = RET_INTEGER,
7691 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7692 .arg1_size = sizeof(struct ipv6hdr),
7693 .arg2_type = ARG_PTR_TO_MEM,
7694 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7697 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv4, struct iphdr *, iph,
7698 struct tcphdr *, th)
7700 u32 cookie = ntohl(th->ack_seq) - 1;
7702 if (__cookie_v4_check(iph, th, cookie) > 0)
7708 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv4_proto = {
7709 .func = bpf_tcp_raw_check_syncookie_ipv4,
7710 .gpl_only = true, /* __cookie_v4_check is GPL */
7712 .ret_type = RET_INTEGER,
7713 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7714 .arg1_size = sizeof(struct iphdr),
7715 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7716 .arg2_size = sizeof(struct tcphdr),
7719 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv6, struct ipv6hdr *, iph,
7720 struct tcphdr *, th)
7722 #if IS_BUILTIN(CONFIG_IPV6)
7723 u32 cookie = ntohl(th->ack_seq) - 1;
7725 if (__cookie_v6_check(iph, th, cookie) > 0)
7730 return -EPROTONOSUPPORT;
7734 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv6_proto = {
7735 .func = bpf_tcp_raw_check_syncookie_ipv6,
7736 .gpl_only = true, /* __cookie_v6_check is GPL */
7738 .ret_type = RET_INTEGER,
7739 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7740 .arg1_size = sizeof(struct ipv6hdr),
7741 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7742 .arg2_size = sizeof(struct tcphdr),
7744 #endif /* CONFIG_SYN_COOKIES */
7746 #endif /* CONFIG_INET */
7748 bool bpf_helper_changes_pkt_data(void *func)
7750 if (func == bpf_skb_vlan_push ||
7751 func == bpf_skb_vlan_pop ||
7752 func == bpf_skb_store_bytes ||
7753 func == bpf_skb_change_proto ||
7754 func == bpf_skb_change_head ||
7755 func == sk_skb_change_head ||
7756 func == bpf_skb_change_tail ||
7757 func == sk_skb_change_tail ||
7758 func == bpf_skb_adjust_room ||
7759 func == sk_skb_adjust_room ||
7760 func == bpf_skb_pull_data ||
7761 func == sk_skb_pull_data ||
7762 func == bpf_clone_redirect ||
7763 func == bpf_l3_csum_replace ||
7764 func == bpf_l4_csum_replace ||
7765 func == bpf_xdp_adjust_head ||
7766 func == bpf_xdp_adjust_meta ||
7767 func == bpf_msg_pull_data ||
7768 func == bpf_msg_push_data ||
7769 func == bpf_msg_pop_data ||
7770 func == bpf_xdp_adjust_tail ||
7771 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7772 func == bpf_lwt_seg6_store_bytes ||
7773 func == bpf_lwt_seg6_adjust_srh ||
7774 func == bpf_lwt_seg6_action ||
7777 func == bpf_sock_ops_store_hdr_opt ||
7779 func == bpf_lwt_in_push_encap ||
7780 func == bpf_lwt_xmit_push_encap)
7786 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7787 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7789 static const struct bpf_func_proto *
7790 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7792 const struct bpf_func_proto *func_proto;
7794 func_proto = cgroup_common_func_proto(func_id, prog);
7798 func_proto = cgroup_current_func_proto(func_id, prog);
7803 case BPF_FUNC_get_socket_cookie:
7804 return &bpf_get_socket_cookie_sock_proto;
7805 case BPF_FUNC_get_netns_cookie:
7806 return &bpf_get_netns_cookie_sock_proto;
7807 case BPF_FUNC_perf_event_output:
7808 return &bpf_event_output_data_proto;
7809 case BPF_FUNC_sk_storage_get:
7810 return &bpf_sk_storage_get_cg_sock_proto;
7811 case BPF_FUNC_ktime_get_coarse_ns:
7812 return &bpf_ktime_get_coarse_ns_proto;
7814 return bpf_base_func_proto(func_id);
7818 static const struct bpf_func_proto *
7819 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7821 const struct bpf_func_proto *func_proto;
7823 func_proto = cgroup_common_func_proto(func_id, prog);
7827 func_proto = cgroup_current_func_proto(func_id, prog);
7833 switch (prog->expected_attach_type) {
7834 case BPF_CGROUP_INET4_CONNECT:
7835 case BPF_CGROUP_INET6_CONNECT:
7836 return &bpf_bind_proto;
7840 case BPF_FUNC_get_socket_cookie:
7841 return &bpf_get_socket_cookie_sock_addr_proto;
7842 case BPF_FUNC_get_netns_cookie:
7843 return &bpf_get_netns_cookie_sock_addr_proto;
7844 case BPF_FUNC_perf_event_output:
7845 return &bpf_event_output_data_proto;
7847 case BPF_FUNC_sk_lookup_tcp:
7848 return &bpf_sock_addr_sk_lookup_tcp_proto;
7849 case BPF_FUNC_sk_lookup_udp:
7850 return &bpf_sock_addr_sk_lookup_udp_proto;
7851 case BPF_FUNC_sk_release:
7852 return &bpf_sk_release_proto;
7853 case BPF_FUNC_skc_lookup_tcp:
7854 return &bpf_sock_addr_skc_lookup_tcp_proto;
7855 #endif /* CONFIG_INET */
7856 case BPF_FUNC_sk_storage_get:
7857 return &bpf_sk_storage_get_proto;
7858 case BPF_FUNC_sk_storage_delete:
7859 return &bpf_sk_storage_delete_proto;
7860 case BPF_FUNC_setsockopt:
7861 switch (prog->expected_attach_type) {
7862 case BPF_CGROUP_INET4_BIND:
7863 case BPF_CGROUP_INET6_BIND:
7864 case BPF_CGROUP_INET4_CONNECT:
7865 case BPF_CGROUP_INET6_CONNECT:
7866 case BPF_CGROUP_UDP4_RECVMSG:
7867 case BPF_CGROUP_UDP6_RECVMSG:
7868 case BPF_CGROUP_UDP4_SENDMSG:
7869 case BPF_CGROUP_UDP6_SENDMSG:
7870 case BPF_CGROUP_INET4_GETPEERNAME:
7871 case BPF_CGROUP_INET6_GETPEERNAME:
7872 case BPF_CGROUP_INET4_GETSOCKNAME:
7873 case BPF_CGROUP_INET6_GETSOCKNAME:
7874 return &bpf_sock_addr_setsockopt_proto;
7878 case BPF_FUNC_getsockopt:
7879 switch (prog->expected_attach_type) {
7880 case BPF_CGROUP_INET4_BIND:
7881 case BPF_CGROUP_INET6_BIND:
7882 case BPF_CGROUP_INET4_CONNECT:
7883 case BPF_CGROUP_INET6_CONNECT:
7884 case BPF_CGROUP_UDP4_RECVMSG:
7885 case BPF_CGROUP_UDP6_RECVMSG:
7886 case BPF_CGROUP_UDP4_SENDMSG:
7887 case BPF_CGROUP_UDP6_SENDMSG:
7888 case BPF_CGROUP_INET4_GETPEERNAME:
7889 case BPF_CGROUP_INET6_GETPEERNAME:
7890 case BPF_CGROUP_INET4_GETSOCKNAME:
7891 case BPF_CGROUP_INET6_GETSOCKNAME:
7892 return &bpf_sock_addr_getsockopt_proto;
7897 return bpf_sk_base_func_proto(func_id);
7901 static const struct bpf_func_proto *
7902 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7905 case BPF_FUNC_skb_load_bytes:
7906 return &bpf_skb_load_bytes_proto;
7907 case BPF_FUNC_skb_load_bytes_relative:
7908 return &bpf_skb_load_bytes_relative_proto;
7909 case BPF_FUNC_get_socket_cookie:
7910 return &bpf_get_socket_cookie_proto;
7911 case BPF_FUNC_get_socket_uid:
7912 return &bpf_get_socket_uid_proto;
7913 case BPF_FUNC_perf_event_output:
7914 return &bpf_skb_event_output_proto;
7916 return bpf_sk_base_func_proto(func_id);
7920 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7921 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7923 static const struct bpf_func_proto *
7924 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7926 const struct bpf_func_proto *func_proto;
7928 func_proto = cgroup_common_func_proto(func_id, prog);
7933 case BPF_FUNC_sk_fullsock:
7934 return &bpf_sk_fullsock_proto;
7935 case BPF_FUNC_sk_storage_get:
7936 return &bpf_sk_storage_get_proto;
7937 case BPF_FUNC_sk_storage_delete:
7938 return &bpf_sk_storage_delete_proto;
7939 case BPF_FUNC_perf_event_output:
7940 return &bpf_skb_event_output_proto;
7941 #ifdef CONFIG_SOCK_CGROUP_DATA
7942 case BPF_FUNC_skb_cgroup_id:
7943 return &bpf_skb_cgroup_id_proto;
7944 case BPF_FUNC_skb_ancestor_cgroup_id:
7945 return &bpf_skb_ancestor_cgroup_id_proto;
7946 case BPF_FUNC_sk_cgroup_id:
7947 return &bpf_sk_cgroup_id_proto;
7948 case BPF_FUNC_sk_ancestor_cgroup_id:
7949 return &bpf_sk_ancestor_cgroup_id_proto;
7952 case BPF_FUNC_sk_lookup_tcp:
7953 return &bpf_sk_lookup_tcp_proto;
7954 case BPF_FUNC_sk_lookup_udp:
7955 return &bpf_sk_lookup_udp_proto;
7956 case BPF_FUNC_sk_release:
7957 return &bpf_sk_release_proto;
7958 case BPF_FUNC_skc_lookup_tcp:
7959 return &bpf_skc_lookup_tcp_proto;
7960 case BPF_FUNC_tcp_sock:
7961 return &bpf_tcp_sock_proto;
7962 case BPF_FUNC_get_listener_sock:
7963 return &bpf_get_listener_sock_proto;
7964 case BPF_FUNC_skb_ecn_set_ce:
7965 return &bpf_skb_ecn_set_ce_proto;
7968 return sk_filter_func_proto(func_id, prog);
7972 static const struct bpf_func_proto *
7973 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7976 case BPF_FUNC_skb_store_bytes:
7977 return &bpf_skb_store_bytes_proto;
7978 case BPF_FUNC_skb_load_bytes:
7979 return &bpf_skb_load_bytes_proto;
7980 case BPF_FUNC_skb_load_bytes_relative:
7981 return &bpf_skb_load_bytes_relative_proto;
7982 case BPF_FUNC_skb_pull_data:
7983 return &bpf_skb_pull_data_proto;
7984 case BPF_FUNC_csum_diff:
7985 return &bpf_csum_diff_proto;
7986 case BPF_FUNC_csum_update:
7987 return &bpf_csum_update_proto;
7988 case BPF_FUNC_csum_level:
7989 return &bpf_csum_level_proto;
7990 case BPF_FUNC_l3_csum_replace:
7991 return &bpf_l3_csum_replace_proto;
7992 case BPF_FUNC_l4_csum_replace:
7993 return &bpf_l4_csum_replace_proto;
7994 case BPF_FUNC_clone_redirect:
7995 return &bpf_clone_redirect_proto;
7996 case BPF_FUNC_get_cgroup_classid:
7997 return &bpf_get_cgroup_classid_proto;
7998 case BPF_FUNC_skb_vlan_push:
7999 return &bpf_skb_vlan_push_proto;
8000 case BPF_FUNC_skb_vlan_pop:
8001 return &bpf_skb_vlan_pop_proto;
8002 case BPF_FUNC_skb_change_proto:
8003 return &bpf_skb_change_proto_proto;
8004 case BPF_FUNC_skb_change_type:
8005 return &bpf_skb_change_type_proto;
8006 case BPF_FUNC_skb_adjust_room:
8007 return &bpf_skb_adjust_room_proto;
8008 case BPF_FUNC_skb_change_tail:
8009 return &bpf_skb_change_tail_proto;
8010 case BPF_FUNC_skb_change_head:
8011 return &bpf_skb_change_head_proto;
8012 case BPF_FUNC_skb_get_tunnel_key:
8013 return &bpf_skb_get_tunnel_key_proto;
8014 case BPF_FUNC_skb_set_tunnel_key:
8015 return bpf_get_skb_set_tunnel_proto(func_id);
8016 case BPF_FUNC_skb_get_tunnel_opt:
8017 return &bpf_skb_get_tunnel_opt_proto;
8018 case BPF_FUNC_skb_set_tunnel_opt:
8019 return bpf_get_skb_set_tunnel_proto(func_id);
8020 case BPF_FUNC_redirect:
8021 return &bpf_redirect_proto;
8022 case BPF_FUNC_redirect_neigh:
8023 return &bpf_redirect_neigh_proto;
8024 case BPF_FUNC_redirect_peer:
8025 return &bpf_redirect_peer_proto;
8026 case BPF_FUNC_get_route_realm:
8027 return &bpf_get_route_realm_proto;
8028 case BPF_FUNC_get_hash_recalc:
8029 return &bpf_get_hash_recalc_proto;
8030 case BPF_FUNC_set_hash_invalid:
8031 return &bpf_set_hash_invalid_proto;
8032 case BPF_FUNC_set_hash:
8033 return &bpf_set_hash_proto;
8034 case BPF_FUNC_perf_event_output:
8035 return &bpf_skb_event_output_proto;
8036 case BPF_FUNC_get_smp_processor_id:
8037 return &bpf_get_smp_processor_id_proto;
8038 case BPF_FUNC_skb_under_cgroup:
8039 return &bpf_skb_under_cgroup_proto;
8040 case BPF_FUNC_get_socket_cookie:
8041 return &bpf_get_socket_cookie_proto;
8042 case BPF_FUNC_get_socket_uid:
8043 return &bpf_get_socket_uid_proto;
8044 case BPF_FUNC_fib_lookup:
8045 return &bpf_skb_fib_lookup_proto;
8046 case BPF_FUNC_check_mtu:
8047 return &bpf_skb_check_mtu_proto;
8048 case BPF_FUNC_sk_fullsock:
8049 return &bpf_sk_fullsock_proto;
8050 case BPF_FUNC_sk_storage_get:
8051 return &bpf_sk_storage_get_proto;
8052 case BPF_FUNC_sk_storage_delete:
8053 return &bpf_sk_storage_delete_proto;
8055 case BPF_FUNC_skb_get_xfrm_state:
8056 return &bpf_skb_get_xfrm_state_proto;
8058 #ifdef CONFIG_CGROUP_NET_CLASSID
8059 case BPF_FUNC_skb_cgroup_classid:
8060 return &bpf_skb_cgroup_classid_proto;
8062 #ifdef CONFIG_SOCK_CGROUP_DATA
8063 case BPF_FUNC_skb_cgroup_id:
8064 return &bpf_skb_cgroup_id_proto;
8065 case BPF_FUNC_skb_ancestor_cgroup_id:
8066 return &bpf_skb_ancestor_cgroup_id_proto;
8069 case BPF_FUNC_sk_lookup_tcp:
8070 return &bpf_tc_sk_lookup_tcp_proto;
8071 case BPF_FUNC_sk_lookup_udp:
8072 return &bpf_tc_sk_lookup_udp_proto;
8073 case BPF_FUNC_sk_release:
8074 return &bpf_sk_release_proto;
8075 case BPF_FUNC_tcp_sock:
8076 return &bpf_tcp_sock_proto;
8077 case BPF_FUNC_get_listener_sock:
8078 return &bpf_get_listener_sock_proto;
8079 case BPF_FUNC_skc_lookup_tcp:
8080 return &bpf_tc_skc_lookup_tcp_proto;
8081 case BPF_FUNC_tcp_check_syncookie:
8082 return &bpf_tcp_check_syncookie_proto;
8083 case BPF_FUNC_skb_ecn_set_ce:
8084 return &bpf_skb_ecn_set_ce_proto;
8085 case BPF_FUNC_tcp_gen_syncookie:
8086 return &bpf_tcp_gen_syncookie_proto;
8087 case BPF_FUNC_sk_assign:
8088 return &bpf_sk_assign_proto;
8089 case BPF_FUNC_skb_set_tstamp:
8090 return &bpf_skb_set_tstamp_proto;
8091 #ifdef CONFIG_SYN_COOKIES
8092 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8093 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8094 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8095 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8096 case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8097 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8098 case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8099 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8103 return bpf_sk_base_func_proto(func_id);
8107 static const struct bpf_func_proto *
8108 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8111 case BPF_FUNC_perf_event_output:
8112 return &bpf_xdp_event_output_proto;
8113 case BPF_FUNC_get_smp_processor_id:
8114 return &bpf_get_smp_processor_id_proto;
8115 case BPF_FUNC_csum_diff:
8116 return &bpf_csum_diff_proto;
8117 case BPF_FUNC_xdp_adjust_head:
8118 return &bpf_xdp_adjust_head_proto;
8119 case BPF_FUNC_xdp_adjust_meta:
8120 return &bpf_xdp_adjust_meta_proto;
8121 case BPF_FUNC_redirect:
8122 return &bpf_xdp_redirect_proto;
8123 case BPF_FUNC_redirect_map:
8124 return &bpf_xdp_redirect_map_proto;
8125 case BPF_FUNC_xdp_adjust_tail:
8126 return &bpf_xdp_adjust_tail_proto;
8127 case BPF_FUNC_xdp_get_buff_len:
8128 return &bpf_xdp_get_buff_len_proto;
8129 case BPF_FUNC_xdp_load_bytes:
8130 return &bpf_xdp_load_bytes_proto;
8131 case BPF_FUNC_xdp_store_bytes:
8132 return &bpf_xdp_store_bytes_proto;
8133 case BPF_FUNC_fib_lookup:
8134 return &bpf_xdp_fib_lookup_proto;
8135 case BPF_FUNC_check_mtu:
8136 return &bpf_xdp_check_mtu_proto;
8138 case BPF_FUNC_sk_lookup_udp:
8139 return &bpf_xdp_sk_lookup_udp_proto;
8140 case BPF_FUNC_sk_lookup_tcp:
8141 return &bpf_xdp_sk_lookup_tcp_proto;
8142 case BPF_FUNC_sk_release:
8143 return &bpf_sk_release_proto;
8144 case BPF_FUNC_skc_lookup_tcp:
8145 return &bpf_xdp_skc_lookup_tcp_proto;
8146 case BPF_FUNC_tcp_check_syncookie:
8147 return &bpf_tcp_check_syncookie_proto;
8148 case BPF_FUNC_tcp_gen_syncookie:
8149 return &bpf_tcp_gen_syncookie_proto;
8150 #ifdef CONFIG_SYN_COOKIES
8151 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8152 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8153 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8154 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8155 case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8156 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8157 case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8158 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8162 return bpf_sk_base_func_proto(func_id);
8165 #if IS_MODULE(CONFIG_NF_CONNTRACK) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES)
8166 /* The nf_conn___init type is used in the NF_CONNTRACK kfuncs. The
8167 * kfuncs are defined in two different modules, and we want to be able
8168 * to use them interchangably with the same BTF type ID. Because modules
8169 * can't de-duplicate BTF IDs between each other, we need the type to be
8170 * referenced in the vmlinux BTF or the verifier will get confused about
8171 * the different types. So we add this dummy type reference which will
8172 * be included in vmlinux BTF, allowing both modules to refer to the
8175 BTF_TYPE_EMIT(struct nf_conn___init);
8179 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
8180 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
8182 static const struct bpf_func_proto *
8183 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8185 const struct bpf_func_proto *func_proto;
8187 func_proto = cgroup_common_func_proto(func_id, prog);
8192 case BPF_FUNC_setsockopt:
8193 return &bpf_sock_ops_setsockopt_proto;
8194 case BPF_FUNC_getsockopt:
8195 return &bpf_sock_ops_getsockopt_proto;
8196 case BPF_FUNC_sock_ops_cb_flags_set:
8197 return &bpf_sock_ops_cb_flags_set_proto;
8198 case BPF_FUNC_sock_map_update:
8199 return &bpf_sock_map_update_proto;
8200 case BPF_FUNC_sock_hash_update:
8201 return &bpf_sock_hash_update_proto;
8202 case BPF_FUNC_get_socket_cookie:
8203 return &bpf_get_socket_cookie_sock_ops_proto;
8204 case BPF_FUNC_perf_event_output:
8205 return &bpf_event_output_data_proto;
8206 case BPF_FUNC_sk_storage_get:
8207 return &bpf_sk_storage_get_proto;
8208 case BPF_FUNC_sk_storage_delete:
8209 return &bpf_sk_storage_delete_proto;
8210 case BPF_FUNC_get_netns_cookie:
8211 return &bpf_get_netns_cookie_sock_ops_proto;
8213 case BPF_FUNC_load_hdr_opt:
8214 return &bpf_sock_ops_load_hdr_opt_proto;
8215 case BPF_FUNC_store_hdr_opt:
8216 return &bpf_sock_ops_store_hdr_opt_proto;
8217 case BPF_FUNC_reserve_hdr_opt:
8218 return &bpf_sock_ops_reserve_hdr_opt_proto;
8219 case BPF_FUNC_tcp_sock:
8220 return &bpf_tcp_sock_proto;
8221 #endif /* CONFIG_INET */
8223 return bpf_sk_base_func_proto(func_id);
8227 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
8228 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
8230 static const struct bpf_func_proto *
8231 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8234 case BPF_FUNC_msg_redirect_map:
8235 return &bpf_msg_redirect_map_proto;
8236 case BPF_FUNC_msg_redirect_hash:
8237 return &bpf_msg_redirect_hash_proto;
8238 case BPF_FUNC_msg_apply_bytes:
8239 return &bpf_msg_apply_bytes_proto;
8240 case BPF_FUNC_msg_cork_bytes:
8241 return &bpf_msg_cork_bytes_proto;
8242 case BPF_FUNC_msg_pull_data:
8243 return &bpf_msg_pull_data_proto;
8244 case BPF_FUNC_msg_push_data:
8245 return &bpf_msg_push_data_proto;
8246 case BPF_FUNC_msg_pop_data:
8247 return &bpf_msg_pop_data_proto;
8248 case BPF_FUNC_perf_event_output:
8249 return &bpf_event_output_data_proto;
8250 case BPF_FUNC_get_current_uid_gid:
8251 return &bpf_get_current_uid_gid_proto;
8252 case BPF_FUNC_get_current_pid_tgid:
8253 return &bpf_get_current_pid_tgid_proto;
8254 case BPF_FUNC_sk_storage_get:
8255 return &bpf_sk_storage_get_proto;
8256 case BPF_FUNC_sk_storage_delete:
8257 return &bpf_sk_storage_delete_proto;
8258 case BPF_FUNC_get_netns_cookie:
8259 return &bpf_get_netns_cookie_sk_msg_proto;
8260 #ifdef CONFIG_CGROUP_NET_CLASSID
8261 case BPF_FUNC_get_cgroup_classid:
8262 return &bpf_get_cgroup_classid_curr_proto;
8265 return bpf_sk_base_func_proto(func_id);
8269 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
8270 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
8272 static const struct bpf_func_proto *
8273 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8276 case BPF_FUNC_skb_store_bytes:
8277 return &bpf_skb_store_bytes_proto;
8278 case BPF_FUNC_skb_load_bytes:
8279 return &bpf_skb_load_bytes_proto;
8280 case BPF_FUNC_skb_pull_data:
8281 return &sk_skb_pull_data_proto;
8282 case BPF_FUNC_skb_change_tail:
8283 return &sk_skb_change_tail_proto;
8284 case BPF_FUNC_skb_change_head:
8285 return &sk_skb_change_head_proto;
8286 case BPF_FUNC_skb_adjust_room:
8287 return &sk_skb_adjust_room_proto;
8288 case BPF_FUNC_get_socket_cookie:
8289 return &bpf_get_socket_cookie_proto;
8290 case BPF_FUNC_get_socket_uid:
8291 return &bpf_get_socket_uid_proto;
8292 case BPF_FUNC_sk_redirect_map:
8293 return &bpf_sk_redirect_map_proto;
8294 case BPF_FUNC_sk_redirect_hash:
8295 return &bpf_sk_redirect_hash_proto;
8296 case BPF_FUNC_perf_event_output:
8297 return &bpf_skb_event_output_proto;
8299 case BPF_FUNC_sk_lookup_tcp:
8300 return &bpf_sk_lookup_tcp_proto;
8301 case BPF_FUNC_sk_lookup_udp:
8302 return &bpf_sk_lookup_udp_proto;
8303 case BPF_FUNC_sk_release:
8304 return &bpf_sk_release_proto;
8305 case BPF_FUNC_skc_lookup_tcp:
8306 return &bpf_skc_lookup_tcp_proto;
8309 return bpf_sk_base_func_proto(func_id);
8313 static const struct bpf_func_proto *
8314 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8317 case BPF_FUNC_skb_load_bytes:
8318 return &bpf_flow_dissector_load_bytes_proto;
8320 return bpf_sk_base_func_proto(func_id);
8324 static const struct bpf_func_proto *
8325 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8328 case BPF_FUNC_skb_load_bytes:
8329 return &bpf_skb_load_bytes_proto;
8330 case BPF_FUNC_skb_pull_data:
8331 return &bpf_skb_pull_data_proto;
8332 case BPF_FUNC_csum_diff:
8333 return &bpf_csum_diff_proto;
8334 case BPF_FUNC_get_cgroup_classid:
8335 return &bpf_get_cgroup_classid_proto;
8336 case BPF_FUNC_get_route_realm:
8337 return &bpf_get_route_realm_proto;
8338 case BPF_FUNC_get_hash_recalc:
8339 return &bpf_get_hash_recalc_proto;
8340 case BPF_FUNC_perf_event_output:
8341 return &bpf_skb_event_output_proto;
8342 case BPF_FUNC_get_smp_processor_id:
8343 return &bpf_get_smp_processor_id_proto;
8344 case BPF_FUNC_skb_under_cgroup:
8345 return &bpf_skb_under_cgroup_proto;
8347 return bpf_sk_base_func_proto(func_id);
8351 static const struct bpf_func_proto *
8352 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8355 case BPF_FUNC_lwt_push_encap:
8356 return &bpf_lwt_in_push_encap_proto;
8358 return lwt_out_func_proto(func_id, prog);
8362 static const struct bpf_func_proto *
8363 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8366 case BPF_FUNC_skb_get_tunnel_key:
8367 return &bpf_skb_get_tunnel_key_proto;
8368 case BPF_FUNC_skb_set_tunnel_key:
8369 return bpf_get_skb_set_tunnel_proto(func_id);
8370 case BPF_FUNC_skb_get_tunnel_opt:
8371 return &bpf_skb_get_tunnel_opt_proto;
8372 case BPF_FUNC_skb_set_tunnel_opt:
8373 return bpf_get_skb_set_tunnel_proto(func_id);
8374 case BPF_FUNC_redirect:
8375 return &bpf_redirect_proto;
8376 case BPF_FUNC_clone_redirect:
8377 return &bpf_clone_redirect_proto;
8378 case BPF_FUNC_skb_change_tail:
8379 return &bpf_skb_change_tail_proto;
8380 case BPF_FUNC_skb_change_head:
8381 return &bpf_skb_change_head_proto;
8382 case BPF_FUNC_skb_store_bytes:
8383 return &bpf_skb_store_bytes_proto;
8384 case BPF_FUNC_csum_update:
8385 return &bpf_csum_update_proto;
8386 case BPF_FUNC_csum_level:
8387 return &bpf_csum_level_proto;
8388 case BPF_FUNC_l3_csum_replace:
8389 return &bpf_l3_csum_replace_proto;
8390 case BPF_FUNC_l4_csum_replace:
8391 return &bpf_l4_csum_replace_proto;
8392 case BPF_FUNC_set_hash_invalid:
8393 return &bpf_set_hash_invalid_proto;
8394 case BPF_FUNC_lwt_push_encap:
8395 return &bpf_lwt_xmit_push_encap_proto;
8397 return lwt_out_func_proto(func_id, prog);
8401 static const struct bpf_func_proto *
8402 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8405 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
8406 case BPF_FUNC_lwt_seg6_store_bytes:
8407 return &bpf_lwt_seg6_store_bytes_proto;
8408 case BPF_FUNC_lwt_seg6_action:
8409 return &bpf_lwt_seg6_action_proto;
8410 case BPF_FUNC_lwt_seg6_adjust_srh:
8411 return &bpf_lwt_seg6_adjust_srh_proto;
8414 return lwt_out_func_proto(func_id, prog);
8418 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
8419 const struct bpf_prog *prog,
8420 struct bpf_insn_access_aux *info)
8422 const int size_default = sizeof(__u32);
8424 if (off < 0 || off >= sizeof(struct __sk_buff))
8427 /* The verifier guarantees that size > 0. */
8428 if (off % size != 0)
8432 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8433 if (off + size > offsetofend(struct __sk_buff, cb[4]))
8436 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
8437 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
8438 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
8439 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
8440 case bpf_ctx_range(struct __sk_buff, data):
8441 case bpf_ctx_range(struct __sk_buff, data_meta):
8442 case bpf_ctx_range(struct __sk_buff, data_end):
8443 if (size != size_default)
8446 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8448 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8449 if (type == BPF_WRITE || size != sizeof(__u64))
8452 case bpf_ctx_range(struct __sk_buff, tstamp):
8453 if (size != sizeof(__u64))
8456 case offsetof(struct __sk_buff, sk):
8457 if (type == BPF_WRITE || size != sizeof(__u64))
8459 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
8461 case offsetof(struct __sk_buff, tstamp_type):
8463 case offsetofend(struct __sk_buff, tstamp_type) ... offsetof(struct __sk_buff, hwtstamp) - 1:
8464 /* Explicitly prohibit access to padding in __sk_buff. */
8467 /* Only narrow read access allowed for now. */
8468 if (type == BPF_WRITE) {
8469 if (size != size_default)
8472 bpf_ctx_record_field_size(info, size_default);
8473 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8481 static bool sk_filter_is_valid_access(int off, int size,
8482 enum bpf_access_type type,
8483 const struct bpf_prog *prog,
8484 struct bpf_insn_access_aux *info)
8487 case bpf_ctx_range(struct __sk_buff, tc_classid):
8488 case bpf_ctx_range(struct __sk_buff, data):
8489 case bpf_ctx_range(struct __sk_buff, data_meta):
8490 case bpf_ctx_range(struct __sk_buff, data_end):
8491 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8492 case bpf_ctx_range(struct __sk_buff, tstamp):
8493 case bpf_ctx_range(struct __sk_buff, wire_len):
8494 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8498 if (type == BPF_WRITE) {
8500 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8507 return bpf_skb_is_valid_access(off, size, type, prog, info);
8510 static bool cg_skb_is_valid_access(int off, int size,
8511 enum bpf_access_type type,
8512 const struct bpf_prog *prog,
8513 struct bpf_insn_access_aux *info)
8516 case bpf_ctx_range(struct __sk_buff, tc_classid):
8517 case bpf_ctx_range(struct __sk_buff, data_meta):
8518 case bpf_ctx_range(struct __sk_buff, wire_len):
8520 case bpf_ctx_range(struct __sk_buff, data):
8521 case bpf_ctx_range(struct __sk_buff, data_end):
8527 if (type == BPF_WRITE) {
8529 case bpf_ctx_range(struct __sk_buff, mark):
8530 case bpf_ctx_range(struct __sk_buff, priority):
8531 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8533 case bpf_ctx_range(struct __sk_buff, tstamp):
8543 case bpf_ctx_range(struct __sk_buff, data):
8544 info->reg_type = PTR_TO_PACKET;
8546 case bpf_ctx_range(struct __sk_buff, data_end):
8547 info->reg_type = PTR_TO_PACKET_END;
8551 return bpf_skb_is_valid_access(off, size, type, prog, info);
8554 static bool lwt_is_valid_access(int off, int size,
8555 enum bpf_access_type type,
8556 const struct bpf_prog *prog,
8557 struct bpf_insn_access_aux *info)
8560 case bpf_ctx_range(struct __sk_buff, tc_classid):
8561 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8562 case bpf_ctx_range(struct __sk_buff, data_meta):
8563 case bpf_ctx_range(struct __sk_buff, tstamp):
8564 case bpf_ctx_range(struct __sk_buff, wire_len):
8565 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8569 if (type == BPF_WRITE) {
8571 case bpf_ctx_range(struct __sk_buff, mark):
8572 case bpf_ctx_range(struct __sk_buff, priority):
8573 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8581 case bpf_ctx_range(struct __sk_buff, data):
8582 info->reg_type = PTR_TO_PACKET;
8584 case bpf_ctx_range(struct __sk_buff, data_end):
8585 info->reg_type = PTR_TO_PACKET_END;
8589 return bpf_skb_is_valid_access(off, size, type, prog, info);
8592 /* Attach type specific accesses */
8593 static bool __sock_filter_check_attach_type(int off,
8594 enum bpf_access_type access_type,
8595 enum bpf_attach_type attach_type)
8598 case offsetof(struct bpf_sock, bound_dev_if):
8599 case offsetof(struct bpf_sock, mark):
8600 case offsetof(struct bpf_sock, priority):
8601 switch (attach_type) {
8602 case BPF_CGROUP_INET_SOCK_CREATE:
8603 case BPF_CGROUP_INET_SOCK_RELEASE:
8608 case bpf_ctx_range(struct bpf_sock, src_ip4):
8609 switch (attach_type) {
8610 case BPF_CGROUP_INET4_POST_BIND:
8615 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8616 switch (attach_type) {
8617 case BPF_CGROUP_INET6_POST_BIND:
8622 case bpf_ctx_range(struct bpf_sock, src_port):
8623 switch (attach_type) {
8624 case BPF_CGROUP_INET4_POST_BIND:
8625 case BPF_CGROUP_INET6_POST_BIND:
8632 return access_type == BPF_READ;
8637 bool bpf_sock_common_is_valid_access(int off, int size,
8638 enum bpf_access_type type,
8639 struct bpf_insn_access_aux *info)
8642 case bpf_ctx_range_till(struct bpf_sock, type, priority):
8645 return bpf_sock_is_valid_access(off, size, type, info);
8649 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
8650 struct bpf_insn_access_aux *info)
8652 const int size_default = sizeof(__u32);
8655 if (off < 0 || off >= sizeof(struct bpf_sock))
8657 if (off % size != 0)
8661 case offsetof(struct bpf_sock, state):
8662 case offsetof(struct bpf_sock, family):
8663 case offsetof(struct bpf_sock, type):
8664 case offsetof(struct bpf_sock, protocol):
8665 case offsetof(struct bpf_sock, src_port):
8666 case offsetof(struct bpf_sock, rx_queue_mapping):
8667 case bpf_ctx_range(struct bpf_sock, src_ip4):
8668 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8669 case bpf_ctx_range(struct bpf_sock, dst_ip4):
8670 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8671 bpf_ctx_record_field_size(info, size_default);
8672 return bpf_ctx_narrow_access_ok(off, size, size_default);
8673 case bpf_ctx_range(struct bpf_sock, dst_port):
8674 field_size = size == size_default ?
8675 size_default : sizeof_field(struct bpf_sock, dst_port);
8676 bpf_ctx_record_field_size(info, field_size);
8677 return bpf_ctx_narrow_access_ok(off, size, field_size);
8678 case offsetofend(struct bpf_sock, dst_port) ...
8679 offsetof(struct bpf_sock, dst_ip4) - 1:
8683 return size == size_default;
8686 static bool sock_filter_is_valid_access(int off, int size,
8687 enum bpf_access_type type,
8688 const struct bpf_prog *prog,
8689 struct bpf_insn_access_aux *info)
8691 if (!bpf_sock_is_valid_access(off, size, type, info))
8693 return __sock_filter_check_attach_type(off, type,
8694 prog->expected_attach_type);
8697 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
8698 const struct bpf_prog *prog)
8700 /* Neither direct read nor direct write requires any preliminary
8706 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
8707 const struct bpf_prog *prog, int drop_verdict)
8709 struct bpf_insn *insn = insn_buf;
8714 /* if (!skb->cloned)
8717 * (Fast-path, otherwise approximation that we might be
8718 * a clone, do the rest in helper.)
8720 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET);
8721 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
8722 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
8724 /* ret = bpf_skb_pull_data(skb, 0); */
8725 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
8726 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
8727 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
8728 BPF_FUNC_skb_pull_data);
8731 * return TC_ACT_SHOT;
8733 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
8734 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
8735 *insn++ = BPF_EXIT_INSN();
8738 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
8740 *insn++ = prog->insnsi[0];
8742 return insn - insn_buf;
8745 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
8746 struct bpf_insn *insn_buf)
8748 bool indirect = BPF_MODE(orig->code) == BPF_IND;
8749 struct bpf_insn *insn = insn_buf;
8752 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
8754 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
8756 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
8758 /* We're guaranteed here that CTX is in R6. */
8759 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
8761 switch (BPF_SIZE(orig->code)) {
8763 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
8766 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
8769 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
8773 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
8774 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
8775 *insn++ = BPF_EXIT_INSN();
8777 return insn - insn_buf;
8780 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8781 const struct bpf_prog *prog)
8783 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8786 static bool tc_cls_act_is_valid_access(int off, int size,
8787 enum bpf_access_type type,
8788 const struct bpf_prog *prog,
8789 struct bpf_insn_access_aux *info)
8791 if (type == BPF_WRITE) {
8793 case bpf_ctx_range(struct __sk_buff, mark):
8794 case bpf_ctx_range(struct __sk_buff, tc_index):
8795 case bpf_ctx_range(struct __sk_buff, priority):
8796 case bpf_ctx_range(struct __sk_buff, tc_classid):
8797 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8798 case bpf_ctx_range(struct __sk_buff, tstamp):
8799 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8807 case bpf_ctx_range(struct __sk_buff, data):
8808 info->reg_type = PTR_TO_PACKET;
8810 case bpf_ctx_range(struct __sk_buff, data_meta):
8811 info->reg_type = PTR_TO_PACKET_META;
8813 case bpf_ctx_range(struct __sk_buff, data_end):
8814 info->reg_type = PTR_TO_PACKET_END;
8816 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8818 case offsetof(struct __sk_buff, tstamp_type):
8819 /* The convert_ctx_access() on reading and writing
8820 * __sk_buff->tstamp depends on whether the bpf prog
8821 * has used __sk_buff->tstamp_type or not.
8822 * Thus, we need to set prog->tstamp_type_access
8823 * earlier during is_valid_access() here.
8825 ((struct bpf_prog *)prog)->tstamp_type_access = 1;
8826 return size == sizeof(__u8);
8829 return bpf_skb_is_valid_access(off, size, type, prog, info);
8832 DEFINE_MUTEX(nf_conn_btf_access_lock);
8833 EXPORT_SYMBOL_GPL(nf_conn_btf_access_lock);
8835 int (*nfct_btf_struct_access)(struct bpf_verifier_log *log,
8836 const struct bpf_reg_state *reg,
8838 EXPORT_SYMBOL_GPL(nfct_btf_struct_access);
8840 static int tc_cls_act_btf_struct_access(struct bpf_verifier_log *log,
8841 const struct bpf_reg_state *reg,
8846 mutex_lock(&nf_conn_btf_access_lock);
8847 if (nfct_btf_struct_access)
8848 ret = nfct_btf_struct_access(log, reg, off, size);
8849 mutex_unlock(&nf_conn_btf_access_lock);
8854 static bool __is_valid_xdp_access(int off, int size)
8856 if (off < 0 || off >= sizeof(struct xdp_md))
8858 if (off % size != 0)
8860 if (size != sizeof(__u32))
8866 static bool xdp_is_valid_access(int off, int size,
8867 enum bpf_access_type type,
8868 const struct bpf_prog *prog,
8869 struct bpf_insn_access_aux *info)
8871 if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8873 case offsetof(struct xdp_md, egress_ifindex):
8878 if (type == BPF_WRITE) {
8879 if (bpf_prog_is_offloaded(prog->aux)) {
8881 case offsetof(struct xdp_md, rx_queue_index):
8882 return __is_valid_xdp_access(off, size);
8889 case offsetof(struct xdp_md, data):
8890 info->reg_type = PTR_TO_PACKET;
8892 case offsetof(struct xdp_md, data_meta):
8893 info->reg_type = PTR_TO_PACKET_META;
8895 case offsetof(struct xdp_md, data_end):
8896 info->reg_type = PTR_TO_PACKET_END;
8900 return __is_valid_xdp_access(off, size);
8903 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act)
8905 const u32 act_max = XDP_REDIRECT;
8907 pr_warn_once("%s XDP return value %u on prog %s (id %d) dev %s, expect packet loss!\n",
8908 act > act_max ? "Illegal" : "Driver unsupported",
8909 act, prog->aux->name, prog->aux->id, dev ? dev->name : "N/A");
8911 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
8913 static int xdp_btf_struct_access(struct bpf_verifier_log *log,
8914 const struct bpf_reg_state *reg,
8919 mutex_lock(&nf_conn_btf_access_lock);
8920 if (nfct_btf_struct_access)
8921 ret = nfct_btf_struct_access(log, reg, off, size);
8922 mutex_unlock(&nf_conn_btf_access_lock);
8927 static bool sock_addr_is_valid_access(int off, int size,
8928 enum bpf_access_type type,
8929 const struct bpf_prog *prog,
8930 struct bpf_insn_access_aux *info)
8932 const int size_default = sizeof(__u32);
8934 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
8936 if (off % size != 0)
8939 /* Disallow access to IPv6 fields from IPv4 contex and vise
8943 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8944 switch (prog->expected_attach_type) {
8945 case BPF_CGROUP_INET4_BIND:
8946 case BPF_CGROUP_INET4_CONNECT:
8947 case BPF_CGROUP_INET4_GETPEERNAME:
8948 case BPF_CGROUP_INET4_GETSOCKNAME:
8949 case BPF_CGROUP_UDP4_SENDMSG:
8950 case BPF_CGROUP_UDP4_RECVMSG:
8956 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8957 switch (prog->expected_attach_type) {
8958 case BPF_CGROUP_INET6_BIND:
8959 case BPF_CGROUP_INET6_CONNECT:
8960 case BPF_CGROUP_INET6_GETPEERNAME:
8961 case BPF_CGROUP_INET6_GETSOCKNAME:
8962 case BPF_CGROUP_UDP6_SENDMSG:
8963 case BPF_CGROUP_UDP6_RECVMSG:
8969 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8970 switch (prog->expected_attach_type) {
8971 case BPF_CGROUP_UDP4_SENDMSG:
8977 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8979 switch (prog->expected_attach_type) {
8980 case BPF_CGROUP_UDP6_SENDMSG:
8989 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8990 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8991 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8992 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8994 case bpf_ctx_range(struct bpf_sock_addr, user_port):
8995 if (type == BPF_READ) {
8996 bpf_ctx_record_field_size(info, size_default);
8998 if (bpf_ctx_wide_access_ok(off, size,
8999 struct bpf_sock_addr,
9003 if (bpf_ctx_wide_access_ok(off, size,
9004 struct bpf_sock_addr,
9008 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
9011 if (bpf_ctx_wide_access_ok(off, size,
9012 struct bpf_sock_addr,
9016 if (bpf_ctx_wide_access_ok(off, size,
9017 struct bpf_sock_addr,
9021 if (size != size_default)
9025 case offsetof(struct bpf_sock_addr, sk):
9026 if (type != BPF_READ)
9028 if (size != sizeof(__u64))
9030 info->reg_type = PTR_TO_SOCKET;
9033 if (type == BPF_READ) {
9034 if (size != size_default)
9044 static bool sock_ops_is_valid_access(int off, int size,
9045 enum bpf_access_type type,
9046 const struct bpf_prog *prog,
9047 struct bpf_insn_access_aux *info)
9049 const int size_default = sizeof(__u32);
9051 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
9054 /* The verifier guarantees that size > 0. */
9055 if (off % size != 0)
9058 if (type == BPF_WRITE) {
9060 case offsetof(struct bpf_sock_ops, reply):
9061 case offsetof(struct bpf_sock_ops, sk_txhash):
9062 if (size != size_default)
9070 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
9072 if (size != sizeof(__u64))
9075 case offsetof(struct bpf_sock_ops, sk):
9076 if (size != sizeof(__u64))
9078 info->reg_type = PTR_TO_SOCKET_OR_NULL;
9080 case offsetof(struct bpf_sock_ops, skb_data):
9081 if (size != sizeof(__u64))
9083 info->reg_type = PTR_TO_PACKET;
9085 case offsetof(struct bpf_sock_ops, skb_data_end):
9086 if (size != sizeof(__u64))
9088 info->reg_type = PTR_TO_PACKET_END;
9090 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9091 bpf_ctx_record_field_size(info, size_default);
9092 return bpf_ctx_narrow_access_ok(off, size,
9094 case offsetof(struct bpf_sock_ops, skb_hwtstamp):
9095 if (size != sizeof(__u64))
9099 if (size != size_default)
9108 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
9109 const struct bpf_prog *prog)
9111 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
9114 static bool sk_skb_is_valid_access(int off, int size,
9115 enum bpf_access_type type,
9116 const struct bpf_prog *prog,
9117 struct bpf_insn_access_aux *info)
9120 case bpf_ctx_range(struct __sk_buff, tc_classid):
9121 case bpf_ctx_range(struct __sk_buff, data_meta):
9122 case bpf_ctx_range(struct __sk_buff, tstamp):
9123 case bpf_ctx_range(struct __sk_buff, wire_len):
9124 case bpf_ctx_range(struct __sk_buff, hwtstamp):
9128 if (type == BPF_WRITE) {
9130 case bpf_ctx_range(struct __sk_buff, tc_index):
9131 case bpf_ctx_range(struct __sk_buff, priority):
9139 case bpf_ctx_range(struct __sk_buff, mark):
9141 case bpf_ctx_range(struct __sk_buff, data):
9142 info->reg_type = PTR_TO_PACKET;
9144 case bpf_ctx_range(struct __sk_buff, data_end):
9145 info->reg_type = PTR_TO_PACKET_END;
9149 return bpf_skb_is_valid_access(off, size, type, prog, info);
9152 static bool sk_msg_is_valid_access(int off, int size,
9153 enum bpf_access_type type,
9154 const struct bpf_prog *prog,
9155 struct bpf_insn_access_aux *info)
9157 if (type == BPF_WRITE)
9160 if (off % size != 0)
9164 case offsetof(struct sk_msg_md, data):
9165 info->reg_type = PTR_TO_PACKET;
9166 if (size != sizeof(__u64))
9169 case offsetof(struct sk_msg_md, data_end):
9170 info->reg_type = PTR_TO_PACKET_END;
9171 if (size != sizeof(__u64))
9174 case offsetof(struct sk_msg_md, sk):
9175 if (size != sizeof(__u64))
9177 info->reg_type = PTR_TO_SOCKET;
9179 case bpf_ctx_range(struct sk_msg_md, family):
9180 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
9181 case bpf_ctx_range(struct sk_msg_md, local_ip4):
9182 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
9183 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
9184 case bpf_ctx_range(struct sk_msg_md, remote_port):
9185 case bpf_ctx_range(struct sk_msg_md, local_port):
9186 case bpf_ctx_range(struct sk_msg_md, size):
9187 if (size != sizeof(__u32))
9196 static bool flow_dissector_is_valid_access(int off, int size,
9197 enum bpf_access_type type,
9198 const struct bpf_prog *prog,
9199 struct bpf_insn_access_aux *info)
9201 const int size_default = sizeof(__u32);
9203 if (off < 0 || off >= sizeof(struct __sk_buff))
9206 if (type == BPF_WRITE)
9210 case bpf_ctx_range(struct __sk_buff, data):
9211 if (size != size_default)
9213 info->reg_type = PTR_TO_PACKET;
9215 case bpf_ctx_range(struct __sk_buff, data_end):
9216 if (size != size_default)
9218 info->reg_type = PTR_TO_PACKET_END;
9220 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
9221 if (size != sizeof(__u64))
9223 info->reg_type = PTR_TO_FLOW_KEYS;
9230 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
9231 const struct bpf_insn *si,
9232 struct bpf_insn *insn_buf,
9233 struct bpf_prog *prog,
9237 struct bpf_insn *insn = insn_buf;
9240 case offsetof(struct __sk_buff, data):
9241 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
9242 si->dst_reg, si->src_reg,
9243 offsetof(struct bpf_flow_dissector, data));
9246 case offsetof(struct __sk_buff, data_end):
9247 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
9248 si->dst_reg, si->src_reg,
9249 offsetof(struct bpf_flow_dissector, data_end));
9252 case offsetof(struct __sk_buff, flow_keys):
9253 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
9254 si->dst_reg, si->src_reg,
9255 offsetof(struct bpf_flow_dissector, flow_keys));
9259 return insn - insn_buf;
9262 static struct bpf_insn *bpf_convert_tstamp_type_read(const struct bpf_insn *si,
9263 struct bpf_insn *insn)
9265 __u8 value_reg = si->dst_reg;
9266 __u8 skb_reg = si->src_reg;
9267 /* AX is needed because src_reg and dst_reg could be the same */
9268 __u8 tmp_reg = BPF_REG_AX;
9270 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg,
9271 SKB_BF_MONO_TC_OFFSET);
9272 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg,
9273 SKB_MONO_DELIVERY_TIME_MASK, 2);
9274 *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_UNSPEC);
9275 *insn++ = BPF_JMP_A(1);
9276 *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_DELIVERY_MONO);
9281 static struct bpf_insn *bpf_convert_shinfo_access(__u8 dst_reg, __u8 skb_reg,
9282 struct bpf_insn *insn)
9284 /* si->dst_reg = skb_shinfo(SKB); */
9285 #ifdef NET_SKBUFF_DATA_USES_OFFSET
9286 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9287 BPF_REG_AX, skb_reg,
9288 offsetof(struct sk_buff, end));
9289 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
9291 offsetof(struct sk_buff, head));
9292 *insn++ = BPF_ALU64_REG(BPF_ADD, dst_reg, BPF_REG_AX);
9294 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9296 offsetof(struct sk_buff, end));
9302 static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog,
9303 const struct bpf_insn *si,
9304 struct bpf_insn *insn)
9306 __u8 value_reg = si->dst_reg;
9307 __u8 skb_reg = si->src_reg;
9309 #ifdef CONFIG_NET_CLS_ACT
9310 /* If the tstamp_type is read,
9311 * the bpf prog is aware the tstamp could have delivery time.
9312 * Thus, read skb->tstamp as is if tstamp_type_access is true.
9314 if (!prog->tstamp_type_access) {
9315 /* AX is needed because src_reg and dst_reg could be the same */
9316 __u8 tmp_reg = BPF_REG_AX;
9318 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, SKB_BF_MONO_TC_OFFSET);
9319 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg,
9320 TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK);
9321 *insn++ = BPF_JMP32_IMM(BPF_JNE, tmp_reg,
9322 TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK, 2);
9323 /* skb->tc_at_ingress && skb->mono_delivery_time,
9324 * read 0 as the (rcv) timestamp.
9326 *insn++ = BPF_MOV64_IMM(value_reg, 0);
9327 *insn++ = BPF_JMP_A(1);
9331 *insn++ = BPF_LDX_MEM(BPF_DW, value_reg, skb_reg,
9332 offsetof(struct sk_buff, tstamp));
9336 static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog,
9337 const struct bpf_insn *si,
9338 struct bpf_insn *insn)
9340 __u8 value_reg = si->src_reg;
9341 __u8 skb_reg = si->dst_reg;
9343 #ifdef CONFIG_NET_CLS_ACT
9344 /* If the tstamp_type is read,
9345 * the bpf prog is aware the tstamp could have delivery time.
9346 * Thus, write skb->tstamp as is if tstamp_type_access is true.
9347 * Otherwise, writing at ingress will have to clear the
9348 * mono_delivery_time bit also.
9350 if (!prog->tstamp_type_access) {
9351 __u8 tmp_reg = BPF_REG_AX;
9353 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, SKB_BF_MONO_TC_OFFSET);
9354 /* Writing __sk_buff->tstamp as ingress, goto <clear> */
9355 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1);
9357 *insn++ = BPF_JMP_A(2);
9358 /* <clear>: mono_delivery_time */
9359 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_MONO_DELIVERY_TIME_MASK);
9360 *insn++ = BPF_STX_MEM(BPF_B, skb_reg, tmp_reg, SKB_BF_MONO_TC_OFFSET);
9364 /* <store>: skb->tstamp = tstamp */
9365 *insn++ = BPF_RAW_INSN(BPF_CLASS(si->code) | BPF_DW | BPF_MEM,
9366 skb_reg, value_reg, offsetof(struct sk_buff, tstamp), si->imm);
9370 #define BPF_EMIT_STORE(size, si, off) \
9371 BPF_RAW_INSN(BPF_CLASS((si)->code) | (size) | BPF_MEM, \
9372 (si)->dst_reg, (si)->src_reg, (off), (si)->imm)
9374 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
9375 const struct bpf_insn *si,
9376 struct bpf_insn *insn_buf,
9377 struct bpf_prog *prog, u32 *target_size)
9379 struct bpf_insn *insn = insn_buf;
9383 case offsetof(struct __sk_buff, len):
9384 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9385 bpf_target_off(struct sk_buff, len, 4,
9389 case offsetof(struct __sk_buff, protocol):
9390 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9391 bpf_target_off(struct sk_buff, protocol, 2,
9395 case offsetof(struct __sk_buff, vlan_proto):
9396 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9397 bpf_target_off(struct sk_buff, vlan_proto, 2,
9401 case offsetof(struct __sk_buff, priority):
9402 if (type == BPF_WRITE)
9403 *insn++ = BPF_EMIT_STORE(BPF_W, si,
9404 bpf_target_off(struct sk_buff, priority, 4,
9407 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9408 bpf_target_off(struct sk_buff, priority, 4,
9412 case offsetof(struct __sk_buff, ingress_ifindex):
9413 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9414 bpf_target_off(struct sk_buff, skb_iif, 4,
9418 case offsetof(struct __sk_buff, ifindex):
9419 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9420 si->dst_reg, si->src_reg,
9421 offsetof(struct sk_buff, dev));
9422 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9423 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9424 bpf_target_off(struct net_device, ifindex, 4,
9428 case offsetof(struct __sk_buff, hash):
9429 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9430 bpf_target_off(struct sk_buff, hash, 4,
9434 case offsetof(struct __sk_buff, mark):
9435 if (type == BPF_WRITE)
9436 *insn++ = BPF_EMIT_STORE(BPF_W, si,
9437 bpf_target_off(struct sk_buff, mark, 4,
9440 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9441 bpf_target_off(struct sk_buff, mark, 4,
9445 case offsetof(struct __sk_buff, pkt_type):
9447 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
9449 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
9450 #ifdef __BIG_ENDIAN_BITFIELD
9451 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
9455 case offsetof(struct __sk_buff, queue_mapping):
9456 if (type == BPF_WRITE) {
9457 u32 off = bpf_target_off(struct sk_buff, queue_mapping, 2, target_size);
9459 if (BPF_CLASS(si->code) == BPF_ST && si->imm >= NO_QUEUE_MAPPING) {
9460 *insn++ = BPF_JMP_A(0); /* noop */
9464 if (BPF_CLASS(si->code) == BPF_STX)
9465 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
9466 *insn++ = BPF_EMIT_STORE(BPF_H, si, off);
9468 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9469 bpf_target_off(struct sk_buff,
9475 case offsetof(struct __sk_buff, vlan_present):
9476 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9477 bpf_target_off(struct sk_buff,
9478 vlan_all, 4, target_size));
9479 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9480 *insn++ = BPF_ALU32_IMM(BPF_MOV, si->dst_reg, 1);
9483 case offsetof(struct __sk_buff, vlan_tci):
9484 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9485 bpf_target_off(struct sk_buff, vlan_tci, 2,
9489 case offsetof(struct __sk_buff, cb[0]) ...
9490 offsetofend(struct __sk_buff, cb[4]) - 1:
9491 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
9492 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
9493 offsetof(struct qdisc_skb_cb, data)) %
9496 prog->cb_access = 1;
9498 off -= offsetof(struct __sk_buff, cb[0]);
9499 off += offsetof(struct sk_buff, cb);
9500 off += offsetof(struct qdisc_skb_cb, data);
9501 if (type == BPF_WRITE)
9502 *insn++ = BPF_EMIT_STORE(BPF_SIZE(si->code), si, off);
9504 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
9508 case offsetof(struct __sk_buff, tc_classid):
9509 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
9512 off -= offsetof(struct __sk_buff, tc_classid);
9513 off += offsetof(struct sk_buff, cb);
9514 off += offsetof(struct qdisc_skb_cb, tc_classid);
9516 if (type == BPF_WRITE)
9517 *insn++ = BPF_EMIT_STORE(BPF_H, si, off);
9519 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
9523 case offsetof(struct __sk_buff, data):
9524 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9525 si->dst_reg, si->src_reg,
9526 offsetof(struct sk_buff, data));
9529 case offsetof(struct __sk_buff, data_meta):
9531 off -= offsetof(struct __sk_buff, data_meta);
9532 off += offsetof(struct sk_buff, cb);
9533 off += offsetof(struct bpf_skb_data_end, data_meta);
9534 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9538 case offsetof(struct __sk_buff, data_end):
9540 off -= offsetof(struct __sk_buff, data_end);
9541 off += offsetof(struct sk_buff, cb);
9542 off += offsetof(struct bpf_skb_data_end, data_end);
9543 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9547 case offsetof(struct __sk_buff, tc_index):
9548 #ifdef CONFIG_NET_SCHED
9549 if (type == BPF_WRITE)
9550 *insn++ = BPF_EMIT_STORE(BPF_H, si,
9551 bpf_target_off(struct sk_buff, tc_index, 2,
9554 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9555 bpf_target_off(struct sk_buff, tc_index, 2,
9559 if (type == BPF_WRITE)
9560 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
9562 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9566 case offsetof(struct __sk_buff, napi_id):
9567 #if defined(CONFIG_NET_RX_BUSY_POLL)
9568 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9569 bpf_target_off(struct sk_buff, napi_id, 4,
9571 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
9572 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9575 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9578 case offsetof(struct __sk_buff, family):
9579 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9581 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9582 si->dst_reg, si->src_reg,
9583 offsetof(struct sk_buff, sk));
9584 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9585 bpf_target_off(struct sock_common,
9589 case offsetof(struct __sk_buff, remote_ip4):
9590 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9592 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9593 si->dst_reg, si->src_reg,
9594 offsetof(struct sk_buff, sk));
9595 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9596 bpf_target_off(struct sock_common,
9600 case offsetof(struct __sk_buff, local_ip4):
9601 BUILD_BUG_ON(sizeof_field(struct sock_common,
9602 skc_rcv_saddr) != 4);
9604 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9605 si->dst_reg, si->src_reg,
9606 offsetof(struct sk_buff, sk));
9607 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9608 bpf_target_off(struct sock_common,
9612 case offsetof(struct __sk_buff, remote_ip6[0]) ...
9613 offsetof(struct __sk_buff, remote_ip6[3]):
9614 #if IS_ENABLED(CONFIG_IPV6)
9615 BUILD_BUG_ON(sizeof_field(struct sock_common,
9616 skc_v6_daddr.s6_addr32[0]) != 4);
9619 off -= offsetof(struct __sk_buff, remote_ip6[0]);
9621 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9622 si->dst_reg, si->src_reg,
9623 offsetof(struct sk_buff, sk));
9624 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9625 offsetof(struct sock_common,
9626 skc_v6_daddr.s6_addr32[0]) +
9629 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9632 case offsetof(struct __sk_buff, local_ip6[0]) ...
9633 offsetof(struct __sk_buff, local_ip6[3]):
9634 #if IS_ENABLED(CONFIG_IPV6)
9635 BUILD_BUG_ON(sizeof_field(struct sock_common,
9636 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9639 off -= offsetof(struct __sk_buff, local_ip6[0]);
9641 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9642 si->dst_reg, si->src_reg,
9643 offsetof(struct sk_buff, sk));
9644 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9645 offsetof(struct sock_common,
9646 skc_v6_rcv_saddr.s6_addr32[0]) +
9649 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9653 case offsetof(struct __sk_buff, remote_port):
9654 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9656 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9657 si->dst_reg, si->src_reg,
9658 offsetof(struct sk_buff, sk));
9659 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9660 bpf_target_off(struct sock_common,
9663 #ifndef __BIG_ENDIAN_BITFIELD
9664 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9668 case offsetof(struct __sk_buff, local_port):
9669 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9671 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9672 si->dst_reg, si->src_reg,
9673 offsetof(struct sk_buff, sk));
9674 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9675 bpf_target_off(struct sock_common,
9676 skc_num, 2, target_size));
9679 case offsetof(struct __sk_buff, tstamp):
9680 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
9682 if (type == BPF_WRITE)
9683 insn = bpf_convert_tstamp_write(prog, si, insn);
9685 insn = bpf_convert_tstamp_read(prog, si, insn);
9688 case offsetof(struct __sk_buff, tstamp_type):
9689 insn = bpf_convert_tstamp_type_read(si, insn);
9692 case offsetof(struct __sk_buff, gso_segs):
9693 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9694 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
9695 si->dst_reg, si->dst_reg,
9696 bpf_target_off(struct skb_shared_info,
9700 case offsetof(struct __sk_buff, gso_size):
9701 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9702 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
9703 si->dst_reg, si->dst_reg,
9704 bpf_target_off(struct skb_shared_info,
9708 case offsetof(struct __sk_buff, wire_len):
9709 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
9712 off -= offsetof(struct __sk_buff, wire_len);
9713 off += offsetof(struct sk_buff, cb);
9714 off += offsetof(struct qdisc_skb_cb, pkt_len);
9716 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
9719 case offsetof(struct __sk_buff, sk):
9720 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9721 si->dst_reg, si->src_reg,
9722 offsetof(struct sk_buff, sk));
9724 case offsetof(struct __sk_buff, hwtstamp):
9725 BUILD_BUG_ON(sizeof_field(struct skb_shared_hwtstamps, hwtstamp) != 8);
9726 BUILD_BUG_ON(offsetof(struct skb_shared_hwtstamps, hwtstamp) != 0);
9728 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9729 *insn++ = BPF_LDX_MEM(BPF_DW,
9730 si->dst_reg, si->dst_reg,
9731 bpf_target_off(struct skb_shared_info,
9737 return insn - insn_buf;
9740 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
9741 const struct bpf_insn *si,
9742 struct bpf_insn *insn_buf,
9743 struct bpf_prog *prog, u32 *target_size)
9745 struct bpf_insn *insn = insn_buf;
9749 case offsetof(struct bpf_sock, bound_dev_if):
9750 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
9752 if (type == BPF_WRITE)
9753 *insn++ = BPF_EMIT_STORE(BPF_W, si,
9754 offsetof(struct sock, sk_bound_dev_if));
9756 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9757 offsetof(struct sock, sk_bound_dev_if));
9760 case offsetof(struct bpf_sock, mark):
9761 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
9763 if (type == BPF_WRITE)
9764 *insn++ = BPF_EMIT_STORE(BPF_W, si,
9765 offsetof(struct sock, sk_mark));
9767 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9768 offsetof(struct sock, sk_mark));
9771 case offsetof(struct bpf_sock, priority):
9772 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
9774 if (type == BPF_WRITE)
9775 *insn++ = BPF_EMIT_STORE(BPF_W, si,
9776 offsetof(struct sock, sk_priority));
9778 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9779 offsetof(struct sock, sk_priority));
9782 case offsetof(struct bpf_sock, family):
9783 *insn++ = BPF_LDX_MEM(
9784 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
9785 si->dst_reg, si->src_reg,
9786 bpf_target_off(struct sock_common,
9788 sizeof_field(struct sock_common,
9793 case offsetof(struct bpf_sock, type):
9794 *insn++ = BPF_LDX_MEM(
9795 BPF_FIELD_SIZEOF(struct sock, sk_type),
9796 si->dst_reg, si->src_reg,
9797 bpf_target_off(struct sock, sk_type,
9798 sizeof_field(struct sock, sk_type),
9802 case offsetof(struct bpf_sock, protocol):
9803 *insn++ = BPF_LDX_MEM(
9804 BPF_FIELD_SIZEOF(struct sock, sk_protocol),
9805 si->dst_reg, si->src_reg,
9806 bpf_target_off(struct sock, sk_protocol,
9807 sizeof_field(struct sock, sk_protocol),
9811 case offsetof(struct bpf_sock, src_ip4):
9812 *insn++ = BPF_LDX_MEM(
9813 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9814 bpf_target_off(struct sock_common, skc_rcv_saddr,
9815 sizeof_field(struct sock_common,
9820 case offsetof(struct bpf_sock, dst_ip4):
9821 *insn++ = BPF_LDX_MEM(
9822 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9823 bpf_target_off(struct sock_common, skc_daddr,
9824 sizeof_field(struct sock_common,
9829 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
9830 #if IS_ENABLED(CONFIG_IPV6)
9832 off -= offsetof(struct bpf_sock, src_ip6[0]);
9833 *insn++ = BPF_LDX_MEM(
9834 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9837 skc_v6_rcv_saddr.s6_addr32[0],
9838 sizeof_field(struct sock_common,
9839 skc_v6_rcv_saddr.s6_addr32[0]),
9840 target_size) + off);
9843 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9847 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
9848 #if IS_ENABLED(CONFIG_IPV6)
9850 off -= offsetof(struct bpf_sock, dst_ip6[0]);
9851 *insn++ = BPF_LDX_MEM(
9852 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9853 bpf_target_off(struct sock_common,
9854 skc_v6_daddr.s6_addr32[0],
9855 sizeof_field(struct sock_common,
9856 skc_v6_daddr.s6_addr32[0]),
9857 target_size) + off);
9859 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9864 case offsetof(struct bpf_sock, src_port):
9865 *insn++ = BPF_LDX_MEM(
9866 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
9867 si->dst_reg, si->src_reg,
9868 bpf_target_off(struct sock_common, skc_num,
9869 sizeof_field(struct sock_common,
9874 case offsetof(struct bpf_sock, dst_port):
9875 *insn++ = BPF_LDX_MEM(
9876 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
9877 si->dst_reg, si->src_reg,
9878 bpf_target_off(struct sock_common, skc_dport,
9879 sizeof_field(struct sock_common,
9884 case offsetof(struct bpf_sock, state):
9885 *insn++ = BPF_LDX_MEM(
9886 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
9887 si->dst_reg, si->src_reg,
9888 bpf_target_off(struct sock_common, skc_state,
9889 sizeof_field(struct sock_common,
9893 case offsetof(struct bpf_sock, rx_queue_mapping):
9894 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
9895 *insn++ = BPF_LDX_MEM(
9896 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
9897 si->dst_reg, si->src_reg,
9898 bpf_target_off(struct sock, sk_rx_queue_mapping,
9899 sizeof_field(struct sock,
9900 sk_rx_queue_mapping),
9902 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
9904 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9906 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9912 return insn - insn_buf;
9915 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
9916 const struct bpf_insn *si,
9917 struct bpf_insn *insn_buf,
9918 struct bpf_prog *prog, u32 *target_size)
9920 struct bpf_insn *insn = insn_buf;
9923 case offsetof(struct __sk_buff, ifindex):
9924 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9925 si->dst_reg, si->src_reg,
9926 offsetof(struct sk_buff, dev));
9927 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9928 bpf_target_off(struct net_device, ifindex, 4,
9932 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9936 return insn - insn_buf;
9939 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
9940 const struct bpf_insn *si,
9941 struct bpf_insn *insn_buf,
9942 struct bpf_prog *prog, u32 *target_size)
9944 struct bpf_insn *insn = insn_buf;
9947 case offsetof(struct xdp_md, data):
9948 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
9949 si->dst_reg, si->src_reg,
9950 offsetof(struct xdp_buff, data));
9952 case offsetof(struct xdp_md, data_meta):
9953 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
9954 si->dst_reg, si->src_reg,
9955 offsetof(struct xdp_buff, data_meta));
9957 case offsetof(struct xdp_md, data_end):
9958 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
9959 si->dst_reg, si->src_reg,
9960 offsetof(struct xdp_buff, data_end));
9962 case offsetof(struct xdp_md, ingress_ifindex):
9963 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9964 si->dst_reg, si->src_reg,
9965 offsetof(struct xdp_buff, rxq));
9966 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
9967 si->dst_reg, si->dst_reg,
9968 offsetof(struct xdp_rxq_info, dev));
9969 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9970 offsetof(struct net_device, ifindex));
9972 case offsetof(struct xdp_md, rx_queue_index):
9973 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9974 si->dst_reg, si->src_reg,
9975 offsetof(struct xdp_buff, rxq));
9976 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9977 offsetof(struct xdp_rxq_info,
9980 case offsetof(struct xdp_md, egress_ifindex):
9981 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
9982 si->dst_reg, si->src_reg,
9983 offsetof(struct xdp_buff, txq));
9984 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
9985 si->dst_reg, si->dst_reg,
9986 offsetof(struct xdp_txq_info, dev));
9987 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9988 offsetof(struct net_device, ifindex));
9992 return insn - insn_buf;
9995 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
9996 * context Structure, F is Field in context structure that contains a pointer
9997 * to Nested Structure of type NS that has the field NF.
9999 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
10000 * sure that SIZE is not greater than actual size of S.F.NF.
10002 * If offset OFF is provided, the load happens from that offset relative to
10005 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
10007 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
10008 si->src_reg, offsetof(S, F)); \
10009 *insn++ = BPF_LDX_MEM( \
10010 SIZE, si->dst_reg, si->dst_reg, \
10011 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
10016 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
10017 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
10018 BPF_FIELD_SIZEOF(NS, NF), 0)
10020 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
10021 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
10023 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
10024 * "register" since two registers available in convert_ctx_access are not
10025 * enough: we can't override neither SRC, since it contains value to store, nor
10026 * DST since it contains pointer to context that may be used by later
10027 * instructions. But we need a temporary place to save pointer to nested
10028 * structure whose field we want to store to.
10030 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
10032 int tmp_reg = BPF_REG_9; \
10033 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
10035 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
10037 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
10038 offsetof(S, TF)); \
10039 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
10040 si->dst_reg, offsetof(S, F)); \
10041 *insn++ = BPF_RAW_INSN(SIZE | BPF_MEM | BPF_CLASS(si->code), \
10042 tmp_reg, si->src_reg, \
10043 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
10047 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
10048 offsetof(S, TF)); \
10051 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
10054 if (type == BPF_WRITE) { \
10055 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
10058 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
10059 S, NS, F, NF, SIZE, OFF); \
10063 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
10064 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
10065 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
10067 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
10068 const struct bpf_insn *si,
10069 struct bpf_insn *insn_buf,
10070 struct bpf_prog *prog, u32 *target_size)
10072 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
10073 struct bpf_insn *insn = insn_buf;
10076 case offsetof(struct bpf_sock_addr, user_family):
10077 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10078 struct sockaddr, uaddr, sa_family);
10081 case offsetof(struct bpf_sock_addr, user_ip4):
10082 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10083 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
10084 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
10087 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
10089 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
10090 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10091 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
10092 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
10096 case offsetof(struct bpf_sock_addr, user_port):
10097 /* To get port we need to know sa_family first and then treat
10098 * sockaddr as either sockaddr_in or sockaddr_in6.
10099 * Though we can simplify since port field has same offset and
10100 * size in both structures.
10101 * Here we check this invariant and use just one of the
10102 * structures if it's true.
10104 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
10105 offsetof(struct sockaddr_in6, sin6_port));
10106 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
10107 sizeof_field(struct sockaddr_in6, sin6_port));
10108 /* Account for sin6_port being smaller than user_port. */
10109 port_size = min(port_size, BPF_LDST_BYTES(si));
10110 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10111 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
10112 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
10115 case offsetof(struct bpf_sock_addr, family):
10116 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10117 struct sock, sk, sk_family);
10120 case offsetof(struct bpf_sock_addr, type):
10121 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10122 struct sock, sk, sk_type);
10125 case offsetof(struct bpf_sock_addr, protocol):
10126 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10127 struct sock, sk, sk_protocol);
10130 case offsetof(struct bpf_sock_addr, msg_src_ip4):
10131 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
10132 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10133 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
10134 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
10137 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
10140 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
10141 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
10142 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10143 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
10144 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
10146 case offsetof(struct bpf_sock_addr, sk):
10147 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
10148 si->dst_reg, si->src_reg,
10149 offsetof(struct bpf_sock_addr_kern, sk));
10153 return insn - insn_buf;
10156 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
10157 const struct bpf_insn *si,
10158 struct bpf_insn *insn_buf,
10159 struct bpf_prog *prog,
10162 struct bpf_insn *insn = insn_buf;
10165 /* Helper macro for adding read access to tcp_sock or sock fields. */
10166 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
10168 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \
10169 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
10170 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
10171 if (si->dst_reg == reg || si->src_reg == reg) \
10173 if (si->dst_reg == reg || si->src_reg == reg) \
10175 if (si->dst_reg == si->src_reg) { \
10176 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
10177 offsetof(struct bpf_sock_ops_kern, \
10179 fullsock_reg = reg; \
10182 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10183 struct bpf_sock_ops_kern, \
10185 fullsock_reg, si->src_reg, \
10186 offsetof(struct bpf_sock_ops_kern, \
10188 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
10189 if (si->dst_reg == si->src_reg) \
10190 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10191 offsetof(struct bpf_sock_ops_kern, \
10193 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10194 struct bpf_sock_ops_kern, sk),\
10195 si->dst_reg, si->src_reg, \
10196 offsetof(struct bpf_sock_ops_kern, sk));\
10197 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
10199 si->dst_reg, si->dst_reg, \
10200 offsetof(OBJ, OBJ_FIELD)); \
10201 if (si->dst_reg == si->src_reg) { \
10202 *insn++ = BPF_JMP_A(1); \
10203 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10204 offsetof(struct bpf_sock_ops_kern, \
10209 #define SOCK_OPS_GET_SK() \
10211 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \
10212 if (si->dst_reg == reg || si->src_reg == reg) \
10214 if (si->dst_reg == reg || si->src_reg == reg) \
10216 if (si->dst_reg == si->src_reg) { \
10217 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
10218 offsetof(struct bpf_sock_ops_kern, \
10220 fullsock_reg = reg; \
10223 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10224 struct bpf_sock_ops_kern, \
10226 fullsock_reg, si->src_reg, \
10227 offsetof(struct bpf_sock_ops_kern, \
10229 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
10230 if (si->dst_reg == si->src_reg) \
10231 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10232 offsetof(struct bpf_sock_ops_kern, \
10234 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10235 struct bpf_sock_ops_kern, sk),\
10236 si->dst_reg, si->src_reg, \
10237 offsetof(struct bpf_sock_ops_kern, sk));\
10238 if (si->dst_reg == si->src_reg) { \
10239 *insn++ = BPF_JMP_A(1); \
10240 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10241 offsetof(struct bpf_sock_ops_kern, \
10246 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
10247 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
10249 /* Helper macro for adding write access to tcp_sock or sock fields.
10250 * The macro is called with two registers, dst_reg which contains a pointer
10251 * to ctx (context) and src_reg which contains the value that should be
10252 * stored. However, we need an additional register since we cannot overwrite
10253 * dst_reg because it may be used later in the program.
10254 * Instead we "borrow" one of the other register. We first save its value
10255 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
10256 * it at the end of the macro.
10258 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
10260 int reg = BPF_REG_9; \
10261 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
10262 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
10263 if (si->dst_reg == reg || si->src_reg == reg) \
10265 if (si->dst_reg == reg || si->src_reg == reg) \
10267 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
10268 offsetof(struct bpf_sock_ops_kern, \
10270 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10271 struct bpf_sock_ops_kern, \
10273 reg, si->dst_reg, \
10274 offsetof(struct bpf_sock_ops_kern, \
10276 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
10277 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10278 struct bpf_sock_ops_kern, sk),\
10279 reg, si->dst_reg, \
10280 offsetof(struct bpf_sock_ops_kern, sk));\
10281 *insn++ = BPF_RAW_INSN(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD) | \
10282 BPF_MEM | BPF_CLASS(si->code), \
10283 reg, si->src_reg, \
10284 offsetof(OBJ, OBJ_FIELD), \
10286 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
10287 offsetof(struct bpf_sock_ops_kern, \
10291 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
10293 if (TYPE == BPF_WRITE) \
10294 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
10296 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
10300 case offsetof(struct bpf_sock_ops, op):
10301 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10303 si->dst_reg, si->src_reg,
10304 offsetof(struct bpf_sock_ops_kern, op));
10307 case offsetof(struct bpf_sock_ops, replylong[0]) ...
10308 offsetof(struct bpf_sock_ops, replylong[3]):
10309 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
10310 sizeof_field(struct bpf_sock_ops_kern, reply));
10311 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
10312 sizeof_field(struct bpf_sock_ops_kern, replylong));
10314 off -= offsetof(struct bpf_sock_ops, replylong[0]);
10315 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
10316 if (type == BPF_WRITE)
10317 *insn++ = BPF_EMIT_STORE(BPF_W, si, off);
10319 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10323 case offsetof(struct bpf_sock_ops, family):
10324 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10326 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10327 struct bpf_sock_ops_kern, sk),
10328 si->dst_reg, si->src_reg,
10329 offsetof(struct bpf_sock_ops_kern, sk));
10330 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10331 offsetof(struct sock_common, skc_family));
10334 case offsetof(struct bpf_sock_ops, remote_ip4):
10335 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10337 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10338 struct bpf_sock_ops_kern, sk),
10339 si->dst_reg, si->src_reg,
10340 offsetof(struct bpf_sock_ops_kern, sk));
10341 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10342 offsetof(struct sock_common, skc_daddr));
10345 case offsetof(struct bpf_sock_ops, local_ip4):
10346 BUILD_BUG_ON(sizeof_field(struct sock_common,
10347 skc_rcv_saddr) != 4);
10349 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10350 struct bpf_sock_ops_kern, sk),
10351 si->dst_reg, si->src_reg,
10352 offsetof(struct bpf_sock_ops_kern, sk));
10353 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10354 offsetof(struct sock_common,
10358 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
10359 offsetof(struct bpf_sock_ops, remote_ip6[3]):
10360 #if IS_ENABLED(CONFIG_IPV6)
10361 BUILD_BUG_ON(sizeof_field(struct sock_common,
10362 skc_v6_daddr.s6_addr32[0]) != 4);
10365 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
10366 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10367 struct bpf_sock_ops_kern, sk),
10368 si->dst_reg, si->src_reg,
10369 offsetof(struct bpf_sock_ops_kern, sk));
10370 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10371 offsetof(struct sock_common,
10372 skc_v6_daddr.s6_addr32[0]) +
10375 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10379 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
10380 offsetof(struct bpf_sock_ops, local_ip6[3]):
10381 #if IS_ENABLED(CONFIG_IPV6)
10382 BUILD_BUG_ON(sizeof_field(struct sock_common,
10383 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10386 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
10387 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10388 struct bpf_sock_ops_kern, sk),
10389 si->dst_reg, si->src_reg,
10390 offsetof(struct bpf_sock_ops_kern, sk));
10391 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10392 offsetof(struct sock_common,
10393 skc_v6_rcv_saddr.s6_addr32[0]) +
10396 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10400 case offsetof(struct bpf_sock_ops, remote_port):
10401 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10403 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10404 struct bpf_sock_ops_kern, sk),
10405 si->dst_reg, si->src_reg,
10406 offsetof(struct bpf_sock_ops_kern, sk));
10407 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10408 offsetof(struct sock_common, skc_dport));
10409 #ifndef __BIG_ENDIAN_BITFIELD
10410 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10414 case offsetof(struct bpf_sock_ops, local_port):
10415 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10417 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10418 struct bpf_sock_ops_kern, sk),
10419 si->dst_reg, si->src_reg,
10420 offsetof(struct bpf_sock_ops_kern, sk));
10421 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10422 offsetof(struct sock_common, skc_num));
10425 case offsetof(struct bpf_sock_ops, is_fullsock):
10426 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10427 struct bpf_sock_ops_kern,
10429 si->dst_reg, si->src_reg,
10430 offsetof(struct bpf_sock_ops_kern,
10434 case offsetof(struct bpf_sock_ops, state):
10435 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
10437 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10438 struct bpf_sock_ops_kern, sk),
10439 si->dst_reg, si->src_reg,
10440 offsetof(struct bpf_sock_ops_kern, sk));
10441 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
10442 offsetof(struct sock_common, skc_state));
10445 case offsetof(struct bpf_sock_ops, rtt_min):
10446 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
10447 sizeof(struct minmax));
10448 BUILD_BUG_ON(sizeof(struct minmax) <
10449 sizeof(struct minmax_sample));
10451 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10452 struct bpf_sock_ops_kern, sk),
10453 si->dst_reg, si->src_reg,
10454 offsetof(struct bpf_sock_ops_kern, sk));
10455 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10456 offsetof(struct tcp_sock, rtt_min) +
10457 sizeof_field(struct minmax_sample, t));
10460 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
10461 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
10465 case offsetof(struct bpf_sock_ops, sk_txhash):
10466 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
10467 struct sock, type);
10469 case offsetof(struct bpf_sock_ops, snd_cwnd):
10470 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
10472 case offsetof(struct bpf_sock_ops, srtt_us):
10473 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
10475 case offsetof(struct bpf_sock_ops, snd_ssthresh):
10476 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
10478 case offsetof(struct bpf_sock_ops, rcv_nxt):
10479 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
10481 case offsetof(struct bpf_sock_ops, snd_nxt):
10482 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
10484 case offsetof(struct bpf_sock_ops, snd_una):
10485 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
10487 case offsetof(struct bpf_sock_ops, mss_cache):
10488 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
10490 case offsetof(struct bpf_sock_ops, ecn_flags):
10491 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
10493 case offsetof(struct bpf_sock_ops, rate_delivered):
10494 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
10496 case offsetof(struct bpf_sock_ops, rate_interval_us):
10497 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
10499 case offsetof(struct bpf_sock_ops, packets_out):
10500 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
10502 case offsetof(struct bpf_sock_ops, retrans_out):
10503 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
10505 case offsetof(struct bpf_sock_ops, total_retrans):
10506 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
10508 case offsetof(struct bpf_sock_ops, segs_in):
10509 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
10511 case offsetof(struct bpf_sock_ops, data_segs_in):
10512 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
10514 case offsetof(struct bpf_sock_ops, segs_out):
10515 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
10517 case offsetof(struct bpf_sock_ops, data_segs_out):
10518 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
10520 case offsetof(struct bpf_sock_ops, lost_out):
10521 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
10523 case offsetof(struct bpf_sock_ops, sacked_out):
10524 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
10526 case offsetof(struct bpf_sock_ops, bytes_received):
10527 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
10529 case offsetof(struct bpf_sock_ops, bytes_acked):
10530 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
10532 case offsetof(struct bpf_sock_ops, sk):
10535 case offsetof(struct bpf_sock_ops, skb_data_end):
10536 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10538 si->dst_reg, si->src_reg,
10539 offsetof(struct bpf_sock_ops_kern,
10542 case offsetof(struct bpf_sock_ops, skb_data):
10543 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10545 si->dst_reg, si->src_reg,
10546 offsetof(struct bpf_sock_ops_kern,
10548 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10549 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10550 si->dst_reg, si->dst_reg,
10551 offsetof(struct sk_buff, data));
10553 case offsetof(struct bpf_sock_ops, skb_len):
10554 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10556 si->dst_reg, si->src_reg,
10557 offsetof(struct bpf_sock_ops_kern,
10559 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10560 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10561 si->dst_reg, si->dst_reg,
10562 offsetof(struct sk_buff, len));
10564 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
10565 off = offsetof(struct sk_buff, cb);
10566 off += offsetof(struct tcp_skb_cb, tcp_flags);
10567 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
10568 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10570 si->dst_reg, si->src_reg,
10571 offsetof(struct bpf_sock_ops_kern,
10573 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10574 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
10576 si->dst_reg, si->dst_reg, off);
10578 case offsetof(struct bpf_sock_ops, skb_hwtstamp): {
10579 struct bpf_insn *jmp_on_null_skb;
10581 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10583 si->dst_reg, si->src_reg,
10584 offsetof(struct bpf_sock_ops_kern,
10586 /* Reserve one insn to test skb == NULL */
10587 jmp_on_null_skb = insn++;
10588 insn = bpf_convert_shinfo_access(si->dst_reg, si->dst_reg, insn);
10589 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
10590 bpf_target_off(struct skb_shared_info,
10593 *jmp_on_null_skb = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0,
10594 insn - jmp_on_null_skb - 1);
10598 return insn - insn_buf;
10601 /* data_end = skb->data + skb_headlen() */
10602 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
10603 struct bpf_insn *insn)
10606 int temp_reg_off = offsetof(struct sk_buff, cb) +
10607 offsetof(struct sk_skb_cb, temp_reg);
10609 if (si->src_reg == si->dst_reg) {
10610 /* We need an extra register, choose and save a register. */
10612 if (si->src_reg == reg || si->dst_reg == reg)
10614 if (si->src_reg == reg || si->dst_reg == reg)
10616 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off);
10621 /* reg = skb->data */
10622 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10624 offsetof(struct sk_buff, data));
10625 /* AX = skb->len */
10626 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10627 BPF_REG_AX, si->src_reg,
10628 offsetof(struct sk_buff, len));
10629 /* reg = skb->data + skb->len */
10630 *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX);
10631 /* AX = skb->data_len */
10632 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
10633 BPF_REG_AX, si->src_reg,
10634 offsetof(struct sk_buff, data_len));
10636 /* reg = skb->data + skb->len - skb->data_len */
10637 *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX);
10639 if (si->src_reg == si->dst_reg) {
10640 /* Restore the saved register */
10641 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg);
10642 *insn++ = BPF_MOV64_REG(si->dst_reg, reg);
10643 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off);
10649 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
10650 const struct bpf_insn *si,
10651 struct bpf_insn *insn_buf,
10652 struct bpf_prog *prog, u32 *target_size)
10654 struct bpf_insn *insn = insn_buf;
10658 case offsetof(struct __sk_buff, data_end):
10659 insn = bpf_convert_data_end_access(si, insn);
10661 case offsetof(struct __sk_buff, cb[0]) ...
10662 offsetofend(struct __sk_buff, cb[4]) - 1:
10663 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20);
10664 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
10665 offsetof(struct sk_skb_cb, data)) %
10668 prog->cb_access = 1;
10670 off -= offsetof(struct __sk_buff, cb[0]);
10671 off += offsetof(struct sk_buff, cb);
10672 off += offsetof(struct sk_skb_cb, data);
10673 if (type == BPF_WRITE)
10674 *insn++ = BPF_EMIT_STORE(BPF_SIZE(si->code), si, off);
10676 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
10682 return bpf_convert_ctx_access(type, si, insn_buf, prog,
10686 return insn - insn_buf;
10689 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
10690 const struct bpf_insn *si,
10691 struct bpf_insn *insn_buf,
10692 struct bpf_prog *prog, u32 *target_size)
10694 struct bpf_insn *insn = insn_buf;
10695 #if IS_ENABLED(CONFIG_IPV6)
10699 /* convert ctx uses the fact sg element is first in struct */
10700 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
10703 case offsetof(struct sk_msg_md, data):
10704 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
10705 si->dst_reg, si->src_reg,
10706 offsetof(struct sk_msg, data));
10708 case offsetof(struct sk_msg_md, data_end):
10709 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
10710 si->dst_reg, si->src_reg,
10711 offsetof(struct sk_msg, data_end));
10713 case offsetof(struct sk_msg_md, family):
10714 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10716 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10717 struct sk_msg, sk),
10718 si->dst_reg, si->src_reg,
10719 offsetof(struct sk_msg, sk));
10720 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10721 offsetof(struct sock_common, skc_family));
10724 case offsetof(struct sk_msg_md, remote_ip4):
10725 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10727 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10728 struct sk_msg, sk),
10729 si->dst_reg, si->src_reg,
10730 offsetof(struct sk_msg, sk));
10731 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10732 offsetof(struct sock_common, skc_daddr));
10735 case offsetof(struct sk_msg_md, local_ip4):
10736 BUILD_BUG_ON(sizeof_field(struct sock_common,
10737 skc_rcv_saddr) != 4);
10739 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10740 struct sk_msg, sk),
10741 si->dst_reg, si->src_reg,
10742 offsetof(struct sk_msg, sk));
10743 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10744 offsetof(struct sock_common,
10748 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
10749 offsetof(struct sk_msg_md, remote_ip6[3]):
10750 #if IS_ENABLED(CONFIG_IPV6)
10751 BUILD_BUG_ON(sizeof_field(struct sock_common,
10752 skc_v6_daddr.s6_addr32[0]) != 4);
10755 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
10756 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10757 struct sk_msg, sk),
10758 si->dst_reg, si->src_reg,
10759 offsetof(struct sk_msg, sk));
10760 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10761 offsetof(struct sock_common,
10762 skc_v6_daddr.s6_addr32[0]) +
10765 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10769 case offsetof(struct sk_msg_md, local_ip6[0]) ...
10770 offsetof(struct sk_msg_md, local_ip6[3]):
10771 #if IS_ENABLED(CONFIG_IPV6)
10772 BUILD_BUG_ON(sizeof_field(struct sock_common,
10773 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10776 off -= offsetof(struct sk_msg_md, local_ip6[0]);
10777 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10778 struct sk_msg, sk),
10779 si->dst_reg, si->src_reg,
10780 offsetof(struct sk_msg, sk));
10781 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10782 offsetof(struct sock_common,
10783 skc_v6_rcv_saddr.s6_addr32[0]) +
10786 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10790 case offsetof(struct sk_msg_md, remote_port):
10791 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10793 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10794 struct sk_msg, sk),
10795 si->dst_reg, si->src_reg,
10796 offsetof(struct sk_msg, sk));
10797 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10798 offsetof(struct sock_common, skc_dport));
10799 #ifndef __BIG_ENDIAN_BITFIELD
10800 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10804 case offsetof(struct sk_msg_md, local_port):
10805 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10807 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10808 struct sk_msg, sk),
10809 si->dst_reg, si->src_reg,
10810 offsetof(struct sk_msg, sk));
10811 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10812 offsetof(struct sock_common, skc_num));
10815 case offsetof(struct sk_msg_md, size):
10816 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
10817 si->dst_reg, si->src_reg,
10818 offsetof(struct sk_msg_sg, size));
10821 case offsetof(struct sk_msg_md, sk):
10822 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
10823 si->dst_reg, si->src_reg,
10824 offsetof(struct sk_msg, sk));
10828 return insn - insn_buf;
10831 const struct bpf_verifier_ops sk_filter_verifier_ops = {
10832 .get_func_proto = sk_filter_func_proto,
10833 .is_valid_access = sk_filter_is_valid_access,
10834 .convert_ctx_access = bpf_convert_ctx_access,
10835 .gen_ld_abs = bpf_gen_ld_abs,
10838 const struct bpf_prog_ops sk_filter_prog_ops = {
10839 .test_run = bpf_prog_test_run_skb,
10842 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
10843 .get_func_proto = tc_cls_act_func_proto,
10844 .is_valid_access = tc_cls_act_is_valid_access,
10845 .convert_ctx_access = tc_cls_act_convert_ctx_access,
10846 .gen_prologue = tc_cls_act_prologue,
10847 .gen_ld_abs = bpf_gen_ld_abs,
10848 .btf_struct_access = tc_cls_act_btf_struct_access,
10851 const struct bpf_prog_ops tc_cls_act_prog_ops = {
10852 .test_run = bpf_prog_test_run_skb,
10855 const struct bpf_verifier_ops xdp_verifier_ops = {
10856 .get_func_proto = xdp_func_proto,
10857 .is_valid_access = xdp_is_valid_access,
10858 .convert_ctx_access = xdp_convert_ctx_access,
10859 .gen_prologue = bpf_noop_prologue,
10860 .btf_struct_access = xdp_btf_struct_access,
10863 const struct bpf_prog_ops xdp_prog_ops = {
10864 .test_run = bpf_prog_test_run_xdp,
10867 const struct bpf_verifier_ops cg_skb_verifier_ops = {
10868 .get_func_proto = cg_skb_func_proto,
10869 .is_valid_access = cg_skb_is_valid_access,
10870 .convert_ctx_access = bpf_convert_ctx_access,
10873 const struct bpf_prog_ops cg_skb_prog_ops = {
10874 .test_run = bpf_prog_test_run_skb,
10877 const struct bpf_verifier_ops lwt_in_verifier_ops = {
10878 .get_func_proto = lwt_in_func_proto,
10879 .is_valid_access = lwt_is_valid_access,
10880 .convert_ctx_access = bpf_convert_ctx_access,
10883 const struct bpf_prog_ops lwt_in_prog_ops = {
10884 .test_run = bpf_prog_test_run_skb,
10887 const struct bpf_verifier_ops lwt_out_verifier_ops = {
10888 .get_func_proto = lwt_out_func_proto,
10889 .is_valid_access = lwt_is_valid_access,
10890 .convert_ctx_access = bpf_convert_ctx_access,
10893 const struct bpf_prog_ops lwt_out_prog_ops = {
10894 .test_run = bpf_prog_test_run_skb,
10897 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
10898 .get_func_proto = lwt_xmit_func_proto,
10899 .is_valid_access = lwt_is_valid_access,
10900 .convert_ctx_access = bpf_convert_ctx_access,
10901 .gen_prologue = tc_cls_act_prologue,
10904 const struct bpf_prog_ops lwt_xmit_prog_ops = {
10905 .test_run = bpf_prog_test_run_skb,
10908 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
10909 .get_func_proto = lwt_seg6local_func_proto,
10910 .is_valid_access = lwt_is_valid_access,
10911 .convert_ctx_access = bpf_convert_ctx_access,
10914 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
10915 .test_run = bpf_prog_test_run_skb,
10918 const struct bpf_verifier_ops cg_sock_verifier_ops = {
10919 .get_func_proto = sock_filter_func_proto,
10920 .is_valid_access = sock_filter_is_valid_access,
10921 .convert_ctx_access = bpf_sock_convert_ctx_access,
10924 const struct bpf_prog_ops cg_sock_prog_ops = {
10927 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
10928 .get_func_proto = sock_addr_func_proto,
10929 .is_valid_access = sock_addr_is_valid_access,
10930 .convert_ctx_access = sock_addr_convert_ctx_access,
10933 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
10936 const struct bpf_verifier_ops sock_ops_verifier_ops = {
10937 .get_func_proto = sock_ops_func_proto,
10938 .is_valid_access = sock_ops_is_valid_access,
10939 .convert_ctx_access = sock_ops_convert_ctx_access,
10942 const struct bpf_prog_ops sock_ops_prog_ops = {
10945 const struct bpf_verifier_ops sk_skb_verifier_ops = {
10946 .get_func_proto = sk_skb_func_proto,
10947 .is_valid_access = sk_skb_is_valid_access,
10948 .convert_ctx_access = sk_skb_convert_ctx_access,
10949 .gen_prologue = sk_skb_prologue,
10952 const struct bpf_prog_ops sk_skb_prog_ops = {
10955 const struct bpf_verifier_ops sk_msg_verifier_ops = {
10956 .get_func_proto = sk_msg_func_proto,
10957 .is_valid_access = sk_msg_is_valid_access,
10958 .convert_ctx_access = sk_msg_convert_ctx_access,
10959 .gen_prologue = bpf_noop_prologue,
10962 const struct bpf_prog_ops sk_msg_prog_ops = {
10965 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
10966 .get_func_proto = flow_dissector_func_proto,
10967 .is_valid_access = flow_dissector_is_valid_access,
10968 .convert_ctx_access = flow_dissector_convert_ctx_access,
10971 const struct bpf_prog_ops flow_dissector_prog_ops = {
10972 .test_run = bpf_prog_test_run_flow_dissector,
10975 int sk_detach_filter(struct sock *sk)
10978 struct sk_filter *filter;
10980 if (sock_flag(sk, SOCK_FILTER_LOCKED))
10983 filter = rcu_dereference_protected(sk->sk_filter,
10984 lockdep_sock_is_held(sk));
10986 RCU_INIT_POINTER(sk->sk_filter, NULL);
10987 sk_filter_uncharge(sk, filter);
10993 EXPORT_SYMBOL_GPL(sk_detach_filter);
10995 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len)
10997 struct sock_fprog_kern *fprog;
10998 struct sk_filter *filter;
11001 sockopt_lock_sock(sk);
11002 filter = rcu_dereference_protected(sk->sk_filter,
11003 lockdep_sock_is_held(sk));
11007 /* We're copying the filter that has been originally attached,
11008 * so no conversion/decode needed anymore. eBPF programs that
11009 * have no original program cannot be dumped through this.
11012 fprog = filter->prog->orig_prog;
11018 /* User space only enquires number of filter blocks. */
11022 if (len < fprog->len)
11026 if (copy_to_sockptr(optval, fprog->filter, bpf_classic_proglen(fprog)))
11029 /* Instead of bytes, the API requests to return the number
11030 * of filter blocks.
11034 sockopt_release_sock(sk);
11039 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
11040 struct sock_reuseport *reuse,
11041 struct sock *sk, struct sk_buff *skb,
11042 struct sock *migrating_sk,
11045 reuse_kern->skb = skb;
11046 reuse_kern->sk = sk;
11047 reuse_kern->selected_sk = NULL;
11048 reuse_kern->migrating_sk = migrating_sk;
11049 reuse_kern->data_end = skb->data + skb_headlen(skb);
11050 reuse_kern->hash = hash;
11051 reuse_kern->reuseport_id = reuse->reuseport_id;
11052 reuse_kern->bind_inany = reuse->bind_inany;
11055 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
11056 struct bpf_prog *prog, struct sk_buff *skb,
11057 struct sock *migrating_sk,
11060 struct sk_reuseport_kern reuse_kern;
11061 enum sk_action action;
11063 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
11064 action = bpf_prog_run(prog, &reuse_kern);
11066 if (action == SK_PASS)
11067 return reuse_kern.selected_sk;
11069 return ERR_PTR(-ECONNREFUSED);
11072 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
11073 struct bpf_map *, map, void *, key, u32, flags)
11075 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
11076 struct sock_reuseport *reuse;
11077 struct sock *selected_sk;
11079 selected_sk = map->ops->map_lookup_elem(map, key);
11083 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
11085 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
11086 if (sk_is_refcounted(selected_sk))
11087 sock_put(selected_sk);
11089 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
11090 * The only (!reuse) case here is - the sk has already been
11091 * unhashed (e.g. by close()), so treat it as -ENOENT.
11093 * Other maps (e.g. sock_map) do not provide this guarantee and
11094 * the sk may never be in the reuseport group to begin with.
11096 return is_sockarray ? -ENOENT : -EINVAL;
11099 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
11100 struct sock *sk = reuse_kern->sk;
11102 if (sk->sk_protocol != selected_sk->sk_protocol)
11103 return -EPROTOTYPE;
11104 else if (sk->sk_family != selected_sk->sk_family)
11105 return -EAFNOSUPPORT;
11107 /* Catch all. Likely bound to a different sockaddr. */
11111 reuse_kern->selected_sk = selected_sk;
11116 static const struct bpf_func_proto sk_select_reuseport_proto = {
11117 .func = sk_select_reuseport,
11119 .ret_type = RET_INTEGER,
11120 .arg1_type = ARG_PTR_TO_CTX,
11121 .arg2_type = ARG_CONST_MAP_PTR,
11122 .arg3_type = ARG_PTR_TO_MAP_KEY,
11123 .arg4_type = ARG_ANYTHING,
11126 BPF_CALL_4(sk_reuseport_load_bytes,
11127 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11128 void *, to, u32, len)
11130 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
11133 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
11134 .func = sk_reuseport_load_bytes,
11136 .ret_type = RET_INTEGER,
11137 .arg1_type = ARG_PTR_TO_CTX,
11138 .arg2_type = ARG_ANYTHING,
11139 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
11140 .arg4_type = ARG_CONST_SIZE,
11143 BPF_CALL_5(sk_reuseport_load_bytes_relative,
11144 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11145 void *, to, u32, len, u32, start_header)
11147 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
11148 len, start_header);
11151 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
11152 .func = sk_reuseport_load_bytes_relative,
11154 .ret_type = RET_INTEGER,
11155 .arg1_type = ARG_PTR_TO_CTX,
11156 .arg2_type = ARG_ANYTHING,
11157 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
11158 .arg4_type = ARG_CONST_SIZE,
11159 .arg5_type = ARG_ANYTHING,
11162 static const struct bpf_func_proto *
11163 sk_reuseport_func_proto(enum bpf_func_id func_id,
11164 const struct bpf_prog *prog)
11167 case BPF_FUNC_sk_select_reuseport:
11168 return &sk_select_reuseport_proto;
11169 case BPF_FUNC_skb_load_bytes:
11170 return &sk_reuseport_load_bytes_proto;
11171 case BPF_FUNC_skb_load_bytes_relative:
11172 return &sk_reuseport_load_bytes_relative_proto;
11173 case BPF_FUNC_get_socket_cookie:
11174 return &bpf_get_socket_ptr_cookie_proto;
11175 case BPF_FUNC_ktime_get_coarse_ns:
11176 return &bpf_ktime_get_coarse_ns_proto;
11178 return bpf_base_func_proto(func_id);
11183 sk_reuseport_is_valid_access(int off, int size,
11184 enum bpf_access_type type,
11185 const struct bpf_prog *prog,
11186 struct bpf_insn_access_aux *info)
11188 const u32 size_default = sizeof(__u32);
11190 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
11191 off % size || type != BPF_READ)
11195 case offsetof(struct sk_reuseport_md, data):
11196 info->reg_type = PTR_TO_PACKET;
11197 return size == sizeof(__u64);
11199 case offsetof(struct sk_reuseport_md, data_end):
11200 info->reg_type = PTR_TO_PACKET_END;
11201 return size == sizeof(__u64);
11203 case offsetof(struct sk_reuseport_md, hash):
11204 return size == size_default;
11206 case offsetof(struct sk_reuseport_md, sk):
11207 info->reg_type = PTR_TO_SOCKET;
11208 return size == sizeof(__u64);
11210 case offsetof(struct sk_reuseport_md, migrating_sk):
11211 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
11212 return size == sizeof(__u64);
11214 /* Fields that allow narrowing */
11215 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
11216 if (size < sizeof_field(struct sk_buff, protocol))
11219 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
11220 case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
11221 case bpf_ctx_range(struct sk_reuseport_md, len):
11222 bpf_ctx_record_field_size(info, size_default);
11223 return bpf_ctx_narrow_access_ok(off, size, size_default);
11230 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
11231 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
11232 si->dst_reg, si->src_reg, \
11233 bpf_target_off(struct sk_reuseport_kern, F, \
11234 sizeof_field(struct sk_reuseport_kern, F), \
11238 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
11239 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
11244 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \
11245 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
11250 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
11251 const struct bpf_insn *si,
11252 struct bpf_insn *insn_buf,
11253 struct bpf_prog *prog,
11256 struct bpf_insn *insn = insn_buf;
11259 case offsetof(struct sk_reuseport_md, data):
11260 SK_REUSEPORT_LOAD_SKB_FIELD(data);
11263 case offsetof(struct sk_reuseport_md, len):
11264 SK_REUSEPORT_LOAD_SKB_FIELD(len);
11267 case offsetof(struct sk_reuseport_md, eth_protocol):
11268 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
11271 case offsetof(struct sk_reuseport_md, ip_protocol):
11272 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
11275 case offsetof(struct sk_reuseport_md, data_end):
11276 SK_REUSEPORT_LOAD_FIELD(data_end);
11279 case offsetof(struct sk_reuseport_md, hash):
11280 SK_REUSEPORT_LOAD_FIELD(hash);
11283 case offsetof(struct sk_reuseport_md, bind_inany):
11284 SK_REUSEPORT_LOAD_FIELD(bind_inany);
11287 case offsetof(struct sk_reuseport_md, sk):
11288 SK_REUSEPORT_LOAD_FIELD(sk);
11291 case offsetof(struct sk_reuseport_md, migrating_sk):
11292 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
11296 return insn - insn_buf;
11299 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
11300 .get_func_proto = sk_reuseport_func_proto,
11301 .is_valid_access = sk_reuseport_is_valid_access,
11302 .convert_ctx_access = sk_reuseport_convert_ctx_access,
11305 const struct bpf_prog_ops sk_reuseport_prog_ops = {
11308 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
11309 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
11311 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
11312 struct sock *, sk, u64, flags)
11314 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
11315 BPF_SK_LOOKUP_F_NO_REUSEPORT)))
11317 if (unlikely(sk && sk_is_refcounted(sk)))
11318 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
11319 if (unlikely(sk && sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN))
11320 return -ESOCKTNOSUPPORT; /* only accept TCP socket in LISTEN */
11321 if (unlikely(sk && sk_is_udp(sk) && sk->sk_state != TCP_CLOSE))
11322 return -ESOCKTNOSUPPORT; /* only accept UDP socket in CLOSE */
11324 /* Check if socket is suitable for packet L3/L4 protocol */
11325 if (sk && sk->sk_protocol != ctx->protocol)
11326 return -EPROTOTYPE;
11327 if (sk && sk->sk_family != ctx->family &&
11328 (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
11329 return -EAFNOSUPPORT;
11331 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
11334 /* Select socket as lookup result */
11335 ctx->selected_sk = sk;
11336 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
11340 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
11341 .func = bpf_sk_lookup_assign,
11343 .ret_type = RET_INTEGER,
11344 .arg1_type = ARG_PTR_TO_CTX,
11345 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL,
11346 .arg3_type = ARG_ANYTHING,
11349 static const struct bpf_func_proto *
11350 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
11353 case BPF_FUNC_perf_event_output:
11354 return &bpf_event_output_data_proto;
11355 case BPF_FUNC_sk_assign:
11356 return &bpf_sk_lookup_assign_proto;
11357 case BPF_FUNC_sk_release:
11358 return &bpf_sk_release_proto;
11360 return bpf_sk_base_func_proto(func_id);
11364 static bool sk_lookup_is_valid_access(int off, int size,
11365 enum bpf_access_type type,
11366 const struct bpf_prog *prog,
11367 struct bpf_insn_access_aux *info)
11369 if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
11371 if (off % size != 0)
11373 if (type != BPF_READ)
11377 case offsetof(struct bpf_sk_lookup, sk):
11378 info->reg_type = PTR_TO_SOCKET_OR_NULL;
11379 return size == sizeof(__u64);
11381 case bpf_ctx_range(struct bpf_sk_lookup, family):
11382 case bpf_ctx_range(struct bpf_sk_lookup, protocol):
11383 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
11384 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
11385 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
11386 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
11387 case bpf_ctx_range(struct bpf_sk_lookup, local_port):
11388 case bpf_ctx_range(struct bpf_sk_lookup, ingress_ifindex):
11389 bpf_ctx_record_field_size(info, sizeof(__u32));
11390 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
11392 case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
11393 /* Allow 4-byte access to 2-byte field for backward compatibility */
11394 if (size == sizeof(__u32))
11396 bpf_ctx_record_field_size(info, sizeof(__be16));
11397 return bpf_ctx_narrow_access_ok(off, size, sizeof(__be16));
11399 case offsetofend(struct bpf_sk_lookup, remote_port) ...
11400 offsetof(struct bpf_sk_lookup, local_ip4) - 1:
11401 /* Allow access to zero padding for backward compatibility */
11402 bpf_ctx_record_field_size(info, sizeof(__u16));
11403 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u16));
11410 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
11411 const struct bpf_insn *si,
11412 struct bpf_insn *insn_buf,
11413 struct bpf_prog *prog,
11416 struct bpf_insn *insn = insn_buf;
11419 case offsetof(struct bpf_sk_lookup, sk):
11420 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11421 offsetof(struct bpf_sk_lookup_kern, selected_sk));
11424 case offsetof(struct bpf_sk_lookup, family):
11425 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11426 bpf_target_off(struct bpf_sk_lookup_kern,
11427 family, 2, target_size));
11430 case offsetof(struct bpf_sk_lookup, protocol):
11431 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11432 bpf_target_off(struct bpf_sk_lookup_kern,
11433 protocol, 2, target_size));
11436 case offsetof(struct bpf_sk_lookup, remote_ip4):
11437 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11438 bpf_target_off(struct bpf_sk_lookup_kern,
11439 v4.saddr, 4, target_size));
11442 case offsetof(struct bpf_sk_lookup, local_ip4):
11443 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11444 bpf_target_off(struct bpf_sk_lookup_kern,
11445 v4.daddr, 4, target_size));
11448 case bpf_ctx_range_till(struct bpf_sk_lookup,
11449 remote_ip6[0], remote_ip6[3]): {
11450 #if IS_ENABLED(CONFIG_IPV6)
11453 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
11454 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11455 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11456 offsetof(struct bpf_sk_lookup_kern, v6.saddr));
11457 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11458 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11460 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11464 case bpf_ctx_range_till(struct bpf_sk_lookup,
11465 local_ip6[0], local_ip6[3]): {
11466 #if IS_ENABLED(CONFIG_IPV6)
11469 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
11470 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11471 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11472 offsetof(struct bpf_sk_lookup_kern, v6.daddr));
11473 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11474 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11476 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11480 case offsetof(struct bpf_sk_lookup, remote_port):
11481 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11482 bpf_target_off(struct bpf_sk_lookup_kern,
11483 sport, 2, target_size));
11486 case offsetofend(struct bpf_sk_lookup, remote_port):
11488 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11491 case offsetof(struct bpf_sk_lookup, local_port):
11492 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11493 bpf_target_off(struct bpf_sk_lookup_kern,
11494 dport, 2, target_size));
11497 case offsetof(struct bpf_sk_lookup, ingress_ifindex):
11498 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11499 bpf_target_off(struct bpf_sk_lookup_kern,
11500 ingress_ifindex, 4, target_size));
11504 return insn - insn_buf;
11507 const struct bpf_prog_ops sk_lookup_prog_ops = {
11508 .test_run = bpf_prog_test_run_sk_lookup,
11511 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
11512 .get_func_proto = sk_lookup_func_proto,
11513 .is_valid_access = sk_lookup_is_valid_access,
11514 .convert_ctx_access = sk_lookup_convert_ctx_access,
11517 #endif /* CONFIG_INET */
11519 DEFINE_BPF_DISPATCHER(xdp)
11521 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
11523 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
11526 BTF_ID_LIST_GLOBAL(btf_sock_ids, MAX_BTF_SOCK_TYPE)
11527 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
11529 #undef BTF_SOCK_TYPE
11531 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
11533 /* tcp6_sock type is not generated in dwarf and hence btf,
11534 * trigger an explicit type generation here.
11536 BTF_TYPE_EMIT(struct tcp6_sock);
11537 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
11538 sk->sk_family == AF_INET6)
11539 return (unsigned long)sk;
11541 return (unsigned long)NULL;
11544 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
11545 .func = bpf_skc_to_tcp6_sock,
11547 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11548 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11549 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
11552 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
11554 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
11555 return (unsigned long)sk;
11557 return (unsigned long)NULL;
11560 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
11561 .func = bpf_skc_to_tcp_sock,
11563 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11564 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11565 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
11568 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
11570 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
11571 * generated if CONFIG_INET=n. Trigger an explicit generation here.
11573 BTF_TYPE_EMIT(struct inet_timewait_sock);
11574 BTF_TYPE_EMIT(struct tcp_timewait_sock);
11577 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
11578 return (unsigned long)sk;
11581 #if IS_BUILTIN(CONFIG_IPV6)
11582 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
11583 return (unsigned long)sk;
11586 return (unsigned long)NULL;
11589 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
11590 .func = bpf_skc_to_tcp_timewait_sock,
11592 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11593 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11594 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
11597 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
11600 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11601 return (unsigned long)sk;
11604 #if IS_BUILTIN(CONFIG_IPV6)
11605 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11606 return (unsigned long)sk;
11609 return (unsigned long)NULL;
11612 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
11613 .func = bpf_skc_to_tcp_request_sock,
11615 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11616 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11617 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
11620 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
11622 /* udp6_sock type is not generated in dwarf and hence btf,
11623 * trigger an explicit type generation here.
11625 BTF_TYPE_EMIT(struct udp6_sock);
11626 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
11627 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
11628 return (unsigned long)sk;
11630 return (unsigned long)NULL;
11633 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
11634 .func = bpf_skc_to_udp6_sock,
11636 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11637 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11638 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
11641 BPF_CALL_1(bpf_skc_to_unix_sock, struct sock *, sk)
11643 /* unix_sock type is not generated in dwarf and hence btf,
11644 * trigger an explicit type generation here.
11646 BTF_TYPE_EMIT(struct unix_sock);
11647 if (sk && sk_fullsock(sk) && sk->sk_family == AF_UNIX)
11648 return (unsigned long)sk;
11650 return (unsigned long)NULL;
11653 const struct bpf_func_proto bpf_skc_to_unix_sock_proto = {
11654 .func = bpf_skc_to_unix_sock,
11656 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11657 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11658 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UNIX],
11661 BPF_CALL_1(bpf_skc_to_mptcp_sock, struct sock *, sk)
11663 BTF_TYPE_EMIT(struct mptcp_sock);
11664 return (unsigned long)bpf_mptcp_sock_from_subflow(sk);
11667 const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto = {
11668 .func = bpf_skc_to_mptcp_sock,
11670 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11671 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
11672 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_MPTCP],
11675 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
11677 return (unsigned long)sock_from_file(file);
11680 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
11681 BTF_ID(struct, socket)
11682 BTF_ID(struct, file)
11684 const struct bpf_func_proto bpf_sock_from_file_proto = {
11685 .func = bpf_sock_from_file,
11687 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11688 .ret_btf_id = &bpf_sock_from_file_btf_ids[0],
11689 .arg1_type = ARG_PTR_TO_BTF_ID,
11690 .arg1_btf_id = &bpf_sock_from_file_btf_ids[1],
11693 static const struct bpf_func_proto *
11694 bpf_sk_base_func_proto(enum bpf_func_id func_id)
11696 const struct bpf_func_proto *func;
11699 case BPF_FUNC_skc_to_tcp6_sock:
11700 func = &bpf_skc_to_tcp6_sock_proto;
11702 case BPF_FUNC_skc_to_tcp_sock:
11703 func = &bpf_skc_to_tcp_sock_proto;
11705 case BPF_FUNC_skc_to_tcp_timewait_sock:
11706 func = &bpf_skc_to_tcp_timewait_sock_proto;
11708 case BPF_FUNC_skc_to_tcp_request_sock:
11709 func = &bpf_skc_to_tcp_request_sock_proto;
11711 case BPF_FUNC_skc_to_udp6_sock:
11712 func = &bpf_skc_to_udp6_sock_proto;
11714 case BPF_FUNC_skc_to_unix_sock:
11715 func = &bpf_skc_to_unix_sock_proto;
11717 case BPF_FUNC_skc_to_mptcp_sock:
11718 func = &bpf_skc_to_mptcp_sock_proto;
11720 case BPF_FUNC_ktime_get_coarse_ns:
11721 return &bpf_ktime_get_coarse_ns_proto;
11723 return bpf_base_func_proto(func_id);
11726 if (!perfmon_capable())
11733 __diag_ignore_all("-Wmissing-prototypes",
11734 "Global functions as their definitions will be in vmlinux BTF");
11735 __bpf_kfunc int bpf_dynptr_from_skb(struct sk_buff *skb, u64 flags,
11736 struct bpf_dynptr_kern *ptr__uninit)
11739 bpf_dynptr_set_null(ptr__uninit);
11743 bpf_dynptr_init(ptr__uninit, skb, BPF_DYNPTR_TYPE_SKB, 0, skb->len);
11748 __bpf_kfunc int bpf_dynptr_from_xdp(struct xdp_buff *xdp, u64 flags,
11749 struct bpf_dynptr_kern *ptr__uninit)
11752 bpf_dynptr_set_null(ptr__uninit);
11756 bpf_dynptr_init(ptr__uninit, xdp, BPF_DYNPTR_TYPE_XDP, 0, xdp_get_buff_len(xdp));
11762 int bpf_dynptr_from_skb_rdonly(struct sk_buff *skb, u64 flags,
11763 struct bpf_dynptr_kern *ptr__uninit)
11767 err = bpf_dynptr_from_skb(skb, flags, ptr__uninit);
11771 bpf_dynptr_set_rdonly(ptr__uninit);
11776 BTF_SET8_START(bpf_kfunc_check_set_skb)
11777 BTF_ID_FLAGS(func, bpf_dynptr_from_skb)
11778 BTF_SET8_END(bpf_kfunc_check_set_skb)
11780 BTF_SET8_START(bpf_kfunc_check_set_xdp)
11781 BTF_ID_FLAGS(func, bpf_dynptr_from_xdp)
11782 BTF_SET8_END(bpf_kfunc_check_set_xdp)
11784 static const struct btf_kfunc_id_set bpf_kfunc_set_skb = {
11785 .owner = THIS_MODULE,
11786 .set = &bpf_kfunc_check_set_skb,
11789 static const struct btf_kfunc_id_set bpf_kfunc_set_xdp = {
11790 .owner = THIS_MODULE,
11791 .set = &bpf_kfunc_check_set_xdp,
11794 static int __init bpf_kfunc_init(void)
11798 ret = register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_kfunc_set_skb);
11799 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_ACT, &bpf_kfunc_set_skb);
11800 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SK_SKB, &bpf_kfunc_set_skb);
11801 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SOCKET_FILTER, &bpf_kfunc_set_skb);
11802 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_CGROUP_SKB, &bpf_kfunc_set_skb);
11803 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_OUT, &bpf_kfunc_set_skb);
11804 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_IN, &bpf_kfunc_set_skb);
11805 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_XMIT, &bpf_kfunc_set_skb);
11806 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_SEG6LOCAL, &bpf_kfunc_set_skb);
11807 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_NETFILTER, &bpf_kfunc_set_skb);
11808 return ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_XDP, &bpf_kfunc_set_xdp);
11810 late_initcall(bpf_kfunc_init);
11812 /* Disables missing prototype warnings */
11814 __diag_ignore_all("-Wmissing-prototypes",
11815 "Global functions as their definitions will be in vmlinux BTF");
11817 /* bpf_sock_destroy: Destroy the given socket with ECONNABORTED error code.
11819 * The function expects a non-NULL pointer to a socket, and invokes the
11820 * protocol specific socket destroy handlers.
11822 * The helper can only be called from BPF contexts that have acquired the socket
11826 * @sock: Pointer to socket to be destroyed
11829 * On error, may return EPROTONOSUPPORT, EINVAL.
11830 * EPROTONOSUPPORT if protocol specific destroy handler is not supported.
11833 __bpf_kfunc int bpf_sock_destroy(struct sock_common *sock)
11835 struct sock *sk = (struct sock *)sock;
11837 /* The locking semantics that allow for synchronous execution of the
11838 * destroy handlers are only supported for TCP and UDP.
11839 * Supporting protocols will need to acquire sock lock in the BPF context
11840 * prior to invoking this kfunc.
11842 if (!sk->sk_prot->diag_destroy || (sk->sk_protocol != IPPROTO_TCP &&
11843 sk->sk_protocol != IPPROTO_UDP))
11844 return -EOPNOTSUPP;
11846 return sk->sk_prot->diag_destroy(sk, ECONNABORTED);
11851 BTF_SET8_START(bpf_sk_iter_kfunc_ids)
11852 BTF_ID_FLAGS(func, bpf_sock_destroy, KF_TRUSTED_ARGS)
11853 BTF_SET8_END(bpf_sk_iter_kfunc_ids)
11855 static int tracing_iter_filter(const struct bpf_prog *prog, u32 kfunc_id)
11857 if (btf_id_set8_contains(&bpf_sk_iter_kfunc_ids, kfunc_id) &&
11858 prog->expected_attach_type != BPF_TRACE_ITER)
11863 static const struct btf_kfunc_id_set bpf_sk_iter_kfunc_set = {
11864 .owner = THIS_MODULE,
11865 .set = &bpf_sk_iter_kfunc_ids,
11866 .filter = tracing_iter_filter,
11869 static int init_subsystem(void)
11871 return register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_sk_iter_kfunc_set);
11873 late_initcall(init_subsystem);