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/module.h>
22 #include <linux/types.h>
24 #include <linux/fcntl.h>
25 #include <linux/socket.h>
26 #include <linux/sock_diag.h>
28 #include <linux/inet.h>
29 #include <linux/netdevice.h>
30 #include <linux/if_packet.h>
31 #include <linux/if_arp.h>
32 #include <linux/gfp.h>
33 #include <net/inet_common.h>
35 #include <net/protocol.h>
36 #include <net/netlink.h>
37 #include <linux/skbuff.h>
38 #include <linux/skmsg.h>
40 #include <net/flow_dissector.h>
41 #include <linux/errno.h>
42 #include <linux/timer.h>
43 #include <linux/uaccess.h>
44 #include <asm/unaligned.h>
45 #include <linux/filter.h>
46 #include <linux/ratelimit.h>
47 #include <linux/seccomp.h>
48 #include <linux/if_vlan.h>
49 #include <linux/bpf.h>
50 #include <linux/btf.h>
51 #include <net/sch_generic.h>
52 #include <net/cls_cgroup.h>
53 #include <net/dst_metadata.h>
55 #include <net/sock_reuseport.h>
56 #include <net/busy_poll.h>
60 #include <linux/bpf_trace.h>
61 #include <net/xdp_sock.h>
62 #include <linux/inetdevice.h>
63 #include <net/inet_hashtables.h>
64 #include <net/inet6_hashtables.h>
65 #include <net/ip_fib.h>
66 #include <net/nexthop.h>
70 #include <net/net_namespace.h>
71 #include <linux/seg6_local.h>
73 #include <net/seg6_local.h>
74 #include <net/lwtunnel.h>
75 #include <net/ipv6_stubs.h>
76 #include <net/bpf_sk_storage.h>
77 #include <net/transp_v6.h>
78 #include <linux/btf_ids.h>
82 static const struct bpf_func_proto *
83 bpf_sk_base_func_proto(enum bpf_func_id func_id);
85 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len)
87 if (in_compat_syscall()) {
88 struct compat_sock_fprog f32;
90 if (len != sizeof(f32))
92 if (copy_from_sockptr(&f32, src, sizeof(f32)))
94 memset(dst, 0, sizeof(*dst));
96 dst->filter = compat_ptr(f32.filter);
98 if (len != sizeof(*dst))
100 if (copy_from_sockptr(dst, src, sizeof(*dst)))
106 EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user);
109 * sk_filter_trim_cap - run a packet through a socket filter
110 * @sk: sock associated with &sk_buff
111 * @skb: buffer to filter
112 * @cap: limit on how short the eBPF program may trim the packet
114 * Run the eBPF program and then cut skb->data to correct size returned by
115 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
116 * than pkt_len we keep whole skb->data. This is the socket level
117 * wrapper to BPF_PROG_RUN. It returns 0 if the packet should
118 * be accepted or -EPERM if the packet should be tossed.
121 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
124 struct sk_filter *filter;
127 * If the skb was allocated from pfmemalloc reserves, only
128 * allow SOCK_MEMALLOC sockets to use it as this socket is
129 * helping free memory
131 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
132 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
135 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
139 err = security_sock_rcv_skb(sk, skb);
144 filter = rcu_dereference(sk->sk_filter);
146 struct sock *save_sk = skb->sk;
147 unsigned int pkt_len;
150 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
152 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
158 EXPORT_SYMBOL(sk_filter_trim_cap);
160 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
162 return skb_get_poff(skb);
165 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
169 if (skb_is_nonlinear(skb))
172 if (skb->len < sizeof(struct nlattr))
175 if (a > skb->len - sizeof(struct nlattr))
178 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
180 return (void *) nla - (void *) skb->data;
185 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
189 if (skb_is_nonlinear(skb))
192 if (skb->len < sizeof(struct nlattr))
195 if (a > skb->len - sizeof(struct nlattr))
198 nla = (struct nlattr *) &skb->data[a];
199 if (nla->nla_len > skb->len - a)
202 nla = nla_find_nested(nla, x);
204 return (void *) nla - (void *) skb->data;
209 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
210 data, int, headlen, int, offset)
213 const int len = sizeof(tmp);
216 if (headlen - offset >= len)
217 return *(u8 *)(data + offset);
218 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
221 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
229 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
232 return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
236 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
237 data, int, headlen, int, offset)
240 const int len = sizeof(tmp);
243 if (headlen - offset >= len)
244 return get_unaligned_be16(data + offset);
245 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
246 return be16_to_cpu(tmp);
248 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
250 return get_unaligned_be16(ptr);
256 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
259 return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
263 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
264 data, int, headlen, int, offset)
267 const int len = sizeof(tmp);
269 if (likely(offset >= 0)) {
270 if (headlen - offset >= len)
271 return get_unaligned_be32(data + offset);
272 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
273 return be32_to_cpu(tmp);
275 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
277 return get_unaligned_be32(ptr);
283 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
286 return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
290 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
291 struct bpf_insn *insn_buf)
293 struct bpf_insn *insn = insn_buf;
297 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4);
299 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
300 offsetof(struct sk_buff, mark));
304 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
305 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
306 #ifdef __BIG_ENDIAN_BITFIELD
307 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
312 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2);
314 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
315 offsetof(struct sk_buff, queue_mapping));
318 case SKF_AD_VLAN_TAG:
319 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2);
321 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
322 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
323 offsetof(struct sk_buff, vlan_tci));
325 case SKF_AD_VLAN_TAG_PRESENT:
326 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_VLAN_PRESENT_OFFSET());
327 if (PKT_VLAN_PRESENT_BIT)
328 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, PKT_VLAN_PRESENT_BIT);
329 if (PKT_VLAN_PRESENT_BIT < 7)
330 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
334 return insn - insn_buf;
337 static bool convert_bpf_extensions(struct sock_filter *fp,
338 struct bpf_insn **insnp)
340 struct bpf_insn *insn = *insnp;
344 case SKF_AD_OFF + SKF_AD_PROTOCOL:
345 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2);
347 /* A = *(u16 *) (CTX + offsetof(protocol)) */
348 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
349 offsetof(struct sk_buff, protocol));
350 /* A = ntohs(A) [emitting a nop or swap16] */
351 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
354 case SKF_AD_OFF + SKF_AD_PKTTYPE:
355 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
359 case SKF_AD_OFF + SKF_AD_IFINDEX:
360 case SKF_AD_OFF + SKF_AD_HATYPE:
361 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4);
362 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2);
364 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
365 BPF_REG_TMP, BPF_REG_CTX,
366 offsetof(struct sk_buff, dev));
367 /* if (tmp != 0) goto pc + 1 */
368 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
369 *insn++ = BPF_EXIT_INSN();
370 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
371 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
372 offsetof(struct net_device, ifindex));
374 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
375 offsetof(struct net_device, type));
378 case SKF_AD_OFF + SKF_AD_MARK:
379 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
383 case SKF_AD_OFF + SKF_AD_RXHASH:
384 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4);
386 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
387 offsetof(struct sk_buff, hash));
390 case SKF_AD_OFF + SKF_AD_QUEUE:
391 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
395 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
396 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
397 BPF_REG_A, BPF_REG_CTX, insn);
401 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
402 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
403 BPF_REG_A, BPF_REG_CTX, insn);
407 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
408 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2);
410 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
411 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
412 offsetof(struct sk_buff, vlan_proto));
413 /* A = ntohs(A) [emitting a nop or swap16] */
414 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
417 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
418 case SKF_AD_OFF + SKF_AD_NLATTR:
419 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
420 case SKF_AD_OFF + SKF_AD_CPU:
421 case SKF_AD_OFF + SKF_AD_RANDOM:
423 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
425 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
427 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
428 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
430 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
431 *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
433 case SKF_AD_OFF + SKF_AD_NLATTR:
434 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
436 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
437 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
439 case SKF_AD_OFF + SKF_AD_CPU:
440 *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
442 case SKF_AD_OFF + SKF_AD_RANDOM:
443 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
444 bpf_user_rnd_init_once();
449 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
451 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
455 /* This is just a dummy call to avoid letting the compiler
456 * evict __bpf_call_base() as an optimization. Placed here
457 * where no-one bothers.
459 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
467 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
469 const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
470 int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
471 bool endian = BPF_SIZE(fp->code) == BPF_H ||
472 BPF_SIZE(fp->code) == BPF_W;
473 bool indirect = BPF_MODE(fp->code) == BPF_IND;
474 const int ip_align = NET_IP_ALIGN;
475 struct bpf_insn *insn = *insnp;
479 ((unaligned_ok && offset >= 0) ||
480 (!unaligned_ok && offset >= 0 &&
481 offset + ip_align >= 0 &&
482 offset + ip_align % size == 0))) {
483 bool ldx_off_ok = offset <= S16_MAX;
485 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
487 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
488 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
489 size, 2 + endian + (!ldx_off_ok * 2));
491 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
494 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
495 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
496 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
500 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
501 *insn++ = BPF_JMP_A(8);
504 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
505 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
506 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
508 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
510 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
512 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
515 switch (BPF_SIZE(fp->code)) {
517 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
520 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
523 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
529 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
530 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
531 *insn = BPF_EXIT_INSN();
538 * bpf_convert_filter - convert filter program
539 * @prog: the user passed filter program
540 * @len: the length of the user passed filter program
541 * @new_prog: allocated 'struct bpf_prog' or NULL
542 * @new_len: pointer to store length of converted program
543 * @seen_ld_abs: bool whether we've seen ld_abs/ind
545 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
546 * style extended BPF (eBPF).
547 * Conversion workflow:
549 * 1) First pass for calculating the new program length:
550 * bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
552 * 2) 2nd pass to remap in two passes: 1st pass finds new
553 * jump offsets, 2nd pass remapping:
554 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
556 static int bpf_convert_filter(struct sock_filter *prog, int len,
557 struct bpf_prog *new_prog, int *new_len,
560 int new_flen = 0, pass = 0, target, i, stack_off;
561 struct bpf_insn *new_insn, *first_insn = NULL;
562 struct sock_filter *fp;
566 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
567 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
569 if (len <= 0 || len > BPF_MAXINSNS)
573 first_insn = new_prog->insnsi;
574 addrs = kcalloc(len, sizeof(*addrs),
575 GFP_KERNEL | __GFP_NOWARN);
581 new_insn = first_insn;
584 /* Classic BPF related prologue emission. */
586 /* Classic BPF expects A and X to be reset first. These need
587 * to be guaranteed to be the first two instructions.
589 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
590 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
592 /* All programs must keep CTX in callee saved BPF_REG_CTX.
593 * In eBPF case it's done by the compiler, here we need to
594 * do this ourself. Initial CTX is present in BPF_REG_ARG1.
596 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
598 /* For packet access in classic BPF, cache skb->data
599 * in callee-saved BPF R8 and skb->len - skb->data_len
600 * (headlen) in BPF R9. Since classic BPF is read-only
601 * on CTX, we only need to cache it once.
603 *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
604 BPF_REG_D, BPF_REG_CTX,
605 offsetof(struct sk_buff, data));
606 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
607 offsetof(struct sk_buff, len));
608 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
609 offsetof(struct sk_buff, data_len));
610 *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
616 for (i = 0; i < len; fp++, i++) {
617 struct bpf_insn tmp_insns[32] = { };
618 struct bpf_insn *insn = tmp_insns;
621 addrs[i] = new_insn - first_insn;
624 /* All arithmetic insns and skb loads map as-is. */
625 case BPF_ALU | BPF_ADD | BPF_X:
626 case BPF_ALU | BPF_ADD | BPF_K:
627 case BPF_ALU | BPF_SUB | BPF_X:
628 case BPF_ALU | BPF_SUB | BPF_K:
629 case BPF_ALU | BPF_AND | BPF_X:
630 case BPF_ALU | BPF_AND | BPF_K:
631 case BPF_ALU | BPF_OR | BPF_X:
632 case BPF_ALU | BPF_OR | BPF_K:
633 case BPF_ALU | BPF_LSH | BPF_X:
634 case BPF_ALU | BPF_LSH | BPF_K:
635 case BPF_ALU | BPF_RSH | BPF_X:
636 case BPF_ALU | BPF_RSH | BPF_K:
637 case BPF_ALU | BPF_XOR | BPF_X:
638 case BPF_ALU | BPF_XOR | BPF_K:
639 case BPF_ALU | BPF_MUL | BPF_X:
640 case BPF_ALU | BPF_MUL | BPF_K:
641 case BPF_ALU | BPF_DIV | BPF_X:
642 case BPF_ALU | BPF_DIV | BPF_K:
643 case BPF_ALU | BPF_MOD | BPF_X:
644 case BPF_ALU | BPF_MOD | BPF_K:
645 case BPF_ALU | BPF_NEG:
646 case BPF_LD | BPF_ABS | BPF_W:
647 case BPF_LD | BPF_ABS | BPF_H:
648 case BPF_LD | BPF_ABS | BPF_B:
649 case BPF_LD | BPF_IND | BPF_W:
650 case BPF_LD | BPF_IND | BPF_H:
651 case BPF_LD | BPF_IND | BPF_B:
652 /* Check for overloaded BPF extension and
653 * directly convert it if found, otherwise
654 * just move on with mapping.
656 if (BPF_CLASS(fp->code) == BPF_LD &&
657 BPF_MODE(fp->code) == BPF_ABS &&
658 convert_bpf_extensions(fp, &insn))
660 if (BPF_CLASS(fp->code) == BPF_LD &&
661 convert_bpf_ld_abs(fp, &insn)) {
666 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
667 fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
668 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
669 /* Error with exception code on div/mod by 0.
670 * For cBPF programs, this was always return 0.
672 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
673 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
674 *insn++ = BPF_EXIT_INSN();
677 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
680 /* Jump transformation cannot use BPF block macros
681 * everywhere as offset calculation and target updates
682 * require a bit more work than the rest, i.e. jump
683 * opcodes map as-is, but offsets need adjustment.
686 #define BPF_EMIT_JMP \
688 const s32 off_min = S16_MIN, off_max = S16_MAX; \
691 if (target >= len || target < 0) \
693 off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
694 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
695 off -= insn - tmp_insns; \
696 /* Reject anything not fitting into insn->off. */ \
697 if (off < off_min || off > off_max) \
702 case BPF_JMP | BPF_JA:
703 target = i + fp->k + 1;
704 insn->code = fp->code;
708 case BPF_JMP | BPF_JEQ | BPF_K:
709 case BPF_JMP | BPF_JEQ | BPF_X:
710 case BPF_JMP | BPF_JSET | BPF_K:
711 case BPF_JMP | BPF_JSET | BPF_X:
712 case BPF_JMP | BPF_JGT | BPF_K:
713 case BPF_JMP | BPF_JGT | BPF_X:
714 case BPF_JMP | BPF_JGE | BPF_K:
715 case BPF_JMP | BPF_JGE | BPF_X:
716 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
717 /* BPF immediates are signed, zero extend
718 * immediate into tmp register and use it
721 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
723 insn->dst_reg = BPF_REG_A;
724 insn->src_reg = BPF_REG_TMP;
727 insn->dst_reg = BPF_REG_A;
729 bpf_src = BPF_SRC(fp->code);
730 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
733 /* Common case where 'jump_false' is next insn. */
735 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
736 target = i + fp->jt + 1;
741 /* Convert some jumps when 'jump_true' is next insn. */
743 switch (BPF_OP(fp->code)) {
745 insn->code = BPF_JMP | BPF_JNE | bpf_src;
748 insn->code = BPF_JMP | BPF_JLE | bpf_src;
751 insn->code = BPF_JMP | BPF_JLT | bpf_src;
757 target = i + fp->jf + 1;
762 /* Other jumps are mapped into two insns: Jxx and JA. */
763 target = i + fp->jt + 1;
764 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
768 insn->code = BPF_JMP | BPF_JA;
769 target = i + fp->jf + 1;
773 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
774 case BPF_LDX | BPF_MSH | BPF_B: {
775 struct sock_filter tmp = {
776 .code = BPF_LD | BPF_ABS | BPF_B,
783 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
784 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
785 convert_bpf_ld_abs(&tmp, &insn);
788 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
790 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
792 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
794 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
796 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
799 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
800 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
802 case BPF_RET | BPF_A:
803 case BPF_RET | BPF_K:
804 if (BPF_RVAL(fp->code) == BPF_K)
805 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
807 *insn = BPF_EXIT_INSN();
810 /* Store to stack. */
813 stack_off = fp->k * 4 + 4;
814 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
815 BPF_ST ? BPF_REG_A : BPF_REG_X,
817 /* check_load_and_stores() verifies that classic BPF can
818 * load from stack only after write, so tracking
819 * stack_depth for ST|STX insns is enough
821 if (new_prog && new_prog->aux->stack_depth < stack_off)
822 new_prog->aux->stack_depth = stack_off;
825 /* Load from stack. */
826 case BPF_LD | BPF_MEM:
827 case BPF_LDX | BPF_MEM:
828 stack_off = fp->k * 4 + 4;
829 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
830 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
835 case BPF_LD | BPF_IMM:
836 case BPF_LDX | BPF_IMM:
837 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
838 BPF_REG_A : BPF_REG_X, fp->k);
842 case BPF_MISC | BPF_TAX:
843 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
847 case BPF_MISC | BPF_TXA:
848 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
851 /* A = skb->len or X = skb->len */
852 case BPF_LD | BPF_W | BPF_LEN:
853 case BPF_LDX | BPF_W | BPF_LEN:
854 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
855 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
856 offsetof(struct sk_buff, len));
859 /* Access seccomp_data fields. */
860 case BPF_LDX | BPF_ABS | BPF_W:
861 /* A = *(u32 *) (ctx + K) */
862 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
865 /* Unknown instruction. */
872 memcpy(new_insn, tmp_insns,
873 sizeof(*insn) * (insn - tmp_insns));
874 new_insn += insn - tmp_insns;
878 /* Only calculating new length. */
879 *new_len = new_insn - first_insn;
881 *new_len += 4; /* Prologue bits. */
886 if (new_flen != new_insn - first_insn) {
887 new_flen = new_insn - first_insn;
894 BUG_ON(*new_len != new_flen);
903 * As we dont want to clear mem[] array for each packet going through
904 * __bpf_prog_run(), we check that filter loaded by user never try to read
905 * a cell if not previously written, and we check all branches to be sure
906 * a malicious user doesn't try to abuse us.
908 static int check_load_and_stores(const struct sock_filter *filter, int flen)
910 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
913 BUILD_BUG_ON(BPF_MEMWORDS > 16);
915 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
919 memset(masks, 0xff, flen * sizeof(*masks));
921 for (pc = 0; pc < flen; pc++) {
922 memvalid &= masks[pc];
924 switch (filter[pc].code) {
927 memvalid |= (1 << filter[pc].k);
929 case BPF_LD | BPF_MEM:
930 case BPF_LDX | BPF_MEM:
931 if (!(memvalid & (1 << filter[pc].k))) {
936 case BPF_JMP | BPF_JA:
937 /* A jump must set masks on target */
938 masks[pc + 1 + filter[pc].k] &= memvalid;
941 case BPF_JMP | BPF_JEQ | BPF_K:
942 case BPF_JMP | BPF_JEQ | BPF_X:
943 case BPF_JMP | BPF_JGE | BPF_K:
944 case BPF_JMP | BPF_JGE | BPF_X:
945 case BPF_JMP | BPF_JGT | BPF_K:
946 case BPF_JMP | BPF_JGT | BPF_X:
947 case BPF_JMP | BPF_JSET | BPF_K:
948 case BPF_JMP | BPF_JSET | BPF_X:
949 /* A jump must set masks on targets */
950 masks[pc + 1 + filter[pc].jt] &= memvalid;
951 masks[pc + 1 + filter[pc].jf] &= memvalid;
961 static bool chk_code_allowed(u16 code_to_probe)
963 static const bool codes[] = {
964 /* 32 bit ALU operations */
965 [BPF_ALU | BPF_ADD | BPF_K] = true,
966 [BPF_ALU | BPF_ADD | BPF_X] = true,
967 [BPF_ALU | BPF_SUB | BPF_K] = true,
968 [BPF_ALU | BPF_SUB | BPF_X] = true,
969 [BPF_ALU | BPF_MUL | BPF_K] = true,
970 [BPF_ALU | BPF_MUL | BPF_X] = true,
971 [BPF_ALU | BPF_DIV | BPF_K] = true,
972 [BPF_ALU | BPF_DIV | BPF_X] = true,
973 [BPF_ALU | BPF_MOD | BPF_K] = true,
974 [BPF_ALU | BPF_MOD | BPF_X] = true,
975 [BPF_ALU | BPF_AND | BPF_K] = true,
976 [BPF_ALU | BPF_AND | BPF_X] = true,
977 [BPF_ALU | BPF_OR | BPF_K] = true,
978 [BPF_ALU | BPF_OR | BPF_X] = true,
979 [BPF_ALU | BPF_XOR | BPF_K] = true,
980 [BPF_ALU | BPF_XOR | BPF_X] = true,
981 [BPF_ALU | BPF_LSH | BPF_K] = true,
982 [BPF_ALU | BPF_LSH | BPF_X] = true,
983 [BPF_ALU | BPF_RSH | BPF_K] = true,
984 [BPF_ALU | BPF_RSH | BPF_X] = true,
985 [BPF_ALU | BPF_NEG] = true,
986 /* Load instructions */
987 [BPF_LD | BPF_W | BPF_ABS] = true,
988 [BPF_LD | BPF_H | BPF_ABS] = true,
989 [BPF_LD | BPF_B | BPF_ABS] = true,
990 [BPF_LD | BPF_W | BPF_LEN] = true,
991 [BPF_LD | BPF_W | BPF_IND] = true,
992 [BPF_LD | BPF_H | BPF_IND] = true,
993 [BPF_LD | BPF_B | BPF_IND] = true,
994 [BPF_LD | BPF_IMM] = true,
995 [BPF_LD | BPF_MEM] = true,
996 [BPF_LDX | BPF_W | BPF_LEN] = true,
997 [BPF_LDX | BPF_B | BPF_MSH] = true,
998 [BPF_LDX | BPF_IMM] = true,
999 [BPF_LDX | BPF_MEM] = true,
1000 /* Store instructions */
1003 /* Misc instructions */
1004 [BPF_MISC | BPF_TAX] = true,
1005 [BPF_MISC | BPF_TXA] = true,
1006 /* Return instructions */
1007 [BPF_RET | BPF_K] = true,
1008 [BPF_RET | BPF_A] = true,
1009 /* Jump instructions */
1010 [BPF_JMP | BPF_JA] = true,
1011 [BPF_JMP | BPF_JEQ | BPF_K] = true,
1012 [BPF_JMP | BPF_JEQ | BPF_X] = true,
1013 [BPF_JMP | BPF_JGE | BPF_K] = true,
1014 [BPF_JMP | BPF_JGE | BPF_X] = true,
1015 [BPF_JMP | BPF_JGT | BPF_K] = true,
1016 [BPF_JMP | BPF_JGT | BPF_X] = true,
1017 [BPF_JMP | BPF_JSET | BPF_K] = true,
1018 [BPF_JMP | BPF_JSET | BPF_X] = true,
1021 if (code_to_probe >= ARRAY_SIZE(codes))
1024 return codes[code_to_probe];
1027 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1032 if (flen == 0 || flen > BPF_MAXINSNS)
1039 * bpf_check_classic - verify socket filter code
1040 * @filter: filter to verify
1041 * @flen: length of filter
1043 * Check the user's filter code. If we let some ugly
1044 * filter code slip through kaboom! The filter must contain
1045 * no references or jumps that are out of range, no illegal
1046 * instructions, and must end with a RET instruction.
1048 * All jumps are forward as they are not signed.
1050 * Returns 0 if the rule set is legal or -EINVAL if not.
1052 static int bpf_check_classic(const struct sock_filter *filter,
1058 /* Check the filter code now */
1059 for (pc = 0; pc < flen; pc++) {
1060 const struct sock_filter *ftest = &filter[pc];
1062 /* May we actually operate on this code? */
1063 if (!chk_code_allowed(ftest->code))
1066 /* Some instructions need special checks */
1067 switch (ftest->code) {
1068 case BPF_ALU | BPF_DIV | BPF_K:
1069 case BPF_ALU | BPF_MOD | BPF_K:
1070 /* Check for division by zero */
1074 case BPF_ALU | BPF_LSH | BPF_K:
1075 case BPF_ALU | BPF_RSH | BPF_K:
1079 case BPF_LD | BPF_MEM:
1080 case BPF_LDX | BPF_MEM:
1083 /* Check for invalid memory addresses */
1084 if (ftest->k >= BPF_MEMWORDS)
1087 case BPF_JMP | BPF_JA:
1088 /* Note, the large ftest->k might cause loops.
1089 * Compare this with conditional jumps below,
1090 * where offsets are limited. --ANK (981016)
1092 if (ftest->k >= (unsigned int)(flen - pc - 1))
1095 case BPF_JMP | BPF_JEQ | BPF_K:
1096 case BPF_JMP | BPF_JEQ | BPF_X:
1097 case BPF_JMP | BPF_JGE | BPF_K:
1098 case BPF_JMP | BPF_JGE | BPF_X:
1099 case BPF_JMP | BPF_JGT | BPF_K:
1100 case BPF_JMP | BPF_JGT | BPF_X:
1101 case BPF_JMP | BPF_JSET | BPF_K:
1102 case BPF_JMP | BPF_JSET | BPF_X:
1103 /* Both conditionals must be safe */
1104 if (pc + ftest->jt + 1 >= flen ||
1105 pc + ftest->jf + 1 >= flen)
1108 case BPF_LD | BPF_W | BPF_ABS:
1109 case BPF_LD | BPF_H | BPF_ABS:
1110 case BPF_LD | BPF_B | BPF_ABS:
1112 if (bpf_anc_helper(ftest) & BPF_ANC)
1114 /* Ancillary operation unknown or unsupported */
1115 if (anc_found == false && ftest->k >= SKF_AD_OFF)
1120 /* Last instruction must be a RET code */
1121 switch (filter[flen - 1].code) {
1122 case BPF_RET | BPF_K:
1123 case BPF_RET | BPF_A:
1124 return check_load_and_stores(filter, flen);
1130 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1131 const struct sock_fprog *fprog)
1133 unsigned int fsize = bpf_classic_proglen(fprog);
1134 struct sock_fprog_kern *fkprog;
1136 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1140 fkprog = fp->orig_prog;
1141 fkprog->len = fprog->len;
1143 fkprog->filter = kmemdup(fp->insns, fsize,
1144 GFP_KERNEL | __GFP_NOWARN);
1145 if (!fkprog->filter) {
1146 kfree(fp->orig_prog);
1153 static void bpf_release_orig_filter(struct bpf_prog *fp)
1155 struct sock_fprog_kern *fprog = fp->orig_prog;
1158 kfree(fprog->filter);
1163 static void __bpf_prog_release(struct bpf_prog *prog)
1165 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1168 bpf_release_orig_filter(prog);
1169 bpf_prog_free(prog);
1173 static void __sk_filter_release(struct sk_filter *fp)
1175 __bpf_prog_release(fp->prog);
1180 * sk_filter_release_rcu - Release a socket filter by rcu_head
1181 * @rcu: rcu_head that contains the sk_filter to free
1183 static void sk_filter_release_rcu(struct rcu_head *rcu)
1185 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1187 __sk_filter_release(fp);
1191 * sk_filter_release - release a socket filter
1192 * @fp: filter to remove
1194 * Remove a filter from a socket and release its resources.
1196 static void sk_filter_release(struct sk_filter *fp)
1198 if (refcount_dec_and_test(&fp->refcnt))
1199 call_rcu(&fp->rcu, sk_filter_release_rcu);
1202 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1204 u32 filter_size = bpf_prog_size(fp->prog->len);
1206 atomic_sub(filter_size, &sk->sk_omem_alloc);
1207 sk_filter_release(fp);
1210 /* try to charge the socket memory if there is space available
1211 * return true on success
1213 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1215 u32 filter_size = bpf_prog_size(fp->prog->len);
1217 /* same check as in sock_kmalloc() */
1218 if (filter_size <= sysctl_optmem_max &&
1219 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
1220 atomic_add(filter_size, &sk->sk_omem_alloc);
1226 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1228 if (!refcount_inc_not_zero(&fp->refcnt))
1231 if (!__sk_filter_charge(sk, fp)) {
1232 sk_filter_release(fp);
1238 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1240 struct sock_filter *old_prog;
1241 struct bpf_prog *old_fp;
1242 int err, new_len, old_len = fp->len;
1243 bool seen_ld_abs = false;
1245 /* We are free to overwrite insns et al right here as it
1246 * won't be used at this point in time anymore internally
1247 * after the migration to the internal BPF instruction
1250 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1251 sizeof(struct bpf_insn));
1253 /* Conversion cannot happen on overlapping memory areas,
1254 * so we need to keep the user BPF around until the 2nd
1255 * pass. At this time, the user BPF is stored in fp->insns.
1257 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1258 GFP_KERNEL | __GFP_NOWARN);
1264 /* 1st pass: calculate the new program length. */
1265 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1270 /* Expand fp for appending the new filter representation. */
1272 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1274 /* The old_fp is still around in case we couldn't
1275 * allocate new memory, so uncharge on that one.
1284 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1285 err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1288 /* 2nd bpf_convert_filter() can fail only if it fails
1289 * to allocate memory, remapping must succeed. Note,
1290 * that at this time old_fp has already been released
1295 fp = bpf_prog_select_runtime(fp, &err);
1305 __bpf_prog_release(fp);
1306 return ERR_PTR(err);
1309 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1310 bpf_aux_classic_check_t trans)
1314 fp->bpf_func = NULL;
1317 err = bpf_check_classic(fp->insns, fp->len);
1319 __bpf_prog_release(fp);
1320 return ERR_PTR(err);
1323 /* There might be additional checks and transformations
1324 * needed on classic filters, f.e. in case of seccomp.
1327 err = trans(fp->insns, fp->len);
1329 __bpf_prog_release(fp);
1330 return ERR_PTR(err);
1334 /* Probe if we can JIT compile the filter and if so, do
1335 * the compilation of the filter.
1337 bpf_jit_compile(fp);
1339 /* JIT compiler couldn't process this filter, so do the
1340 * internal BPF translation for the optimized interpreter.
1343 fp = bpf_migrate_filter(fp);
1349 * bpf_prog_create - create an unattached filter
1350 * @pfp: the unattached filter that is created
1351 * @fprog: the filter program
1353 * Create a filter independent of any socket. We first run some
1354 * sanity checks on it to make sure it does not explode on us later.
1355 * If an error occurs or there is insufficient memory for the filter
1356 * a negative errno code is returned. On success the return is zero.
1358 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1360 unsigned int fsize = bpf_classic_proglen(fprog);
1361 struct bpf_prog *fp;
1363 /* Make sure new filter is there and in the right amounts. */
1364 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1367 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1371 memcpy(fp->insns, fprog->filter, fsize);
1373 fp->len = fprog->len;
1374 /* Since unattached filters are not copied back to user
1375 * space through sk_get_filter(), we do not need to hold
1376 * a copy here, and can spare us the work.
1378 fp->orig_prog = NULL;
1380 /* bpf_prepare_filter() already takes care of freeing
1381 * memory in case something goes wrong.
1383 fp = bpf_prepare_filter(fp, NULL);
1390 EXPORT_SYMBOL_GPL(bpf_prog_create);
1393 * bpf_prog_create_from_user - create an unattached filter from user buffer
1394 * @pfp: the unattached filter that is created
1395 * @fprog: the filter program
1396 * @trans: post-classic verifier transformation handler
1397 * @save_orig: save classic BPF program
1399 * This function effectively does the same as bpf_prog_create(), only
1400 * that it builds up its insns buffer from user space provided buffer.
1401 * It also allows for passing a bpf_aux_classic_check_t handler.
1403 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1404 bpf_aux_classic_check_t trans, bool save_orig)
1406 unsigned int fsize = bpf_classic_proglen(fprog);
1407 struct bpf_prog *fp;
1410 /* Make sure new filter is there and in the right amounts. */
1411 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1414 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1418 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1419 __bpf_prog_free(fp);
1423 fp->len = fprog->len;
1424 fp->orig_prog = NULL;
1427 err = bpf_prog_store_orig_filter(fp, fprog);
1429 __bpf_prog_free(fp);
1434 /* bpf_prepare_filter() already takes care of freeing
1435 * memory in case something goes wrong.
1437 fp = bpf_prepare_filter(fp, trans);
1444 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1446 void bpf_prog_destroy(struct bpf_prog *fp)
1448 __bpf_prog_release(fp);
1450 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1452 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1454 struct sk_filter *fp, *old_fp;
1456 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1462 if (!__sk_filter_charge(sk, fp)) {
1466 refcount_set(&fp->refcnt, 1);
1468 old_fp = rcu_dereference_protected(sk->sk_filter,
1469 lockdep_sock_is_held(sk));
1470 rcu_assign_pointer(sk->sk_filter, fp);
1473 sk_filter_uncharge(sk, old_fp);
1479 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1481 unsigned int fsize = bpf_classic_proglen(fprog);
1482 struct bpf_prog *prog;
1485 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1486 return ERR_PTR(-EPERM);
1488 /* Make sure new filter is there and in the right amounts. */
1489 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1490 return ERR_PTR(-EINVAL);
1492 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1494 return ERR_PTR(-ENOMEM);
1496 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1497 __bpf_prog_free(prog);
1498 return ERR_PTR(-EFAULT);
1501 prog->len = fprog->len;
1503 err = bpf_prog_store_orig_filter(prog, fprog);
1505 __bpf_prog_free(prog);
1506 return ERR_PTR(-ENOMEM);
1509 /* bpf_prepare_filter() already takes care of freeing
1510 * memory in case something goes wrong.
1512 return bpf_prepare_filter(prog, NULL);
1516 * sk_attach_filter - attach a socket filter
1517 * @fprog: the filter program
1518 * @sk: the socket to use
1520 * Attach the user's filter code. We first run some sanity checks on
1521 * it to make sure it does not explode on us later. If an error
1522 * occurs or there is insufficient memory for the filter a negative
1523 * errno code is returned. On success the return is zero.
1525 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1527 struct bpf_prog *prog = __get_filter(fprog, sk);
1531 return PTR_ERR(prog);
1533 err = __sk_attach_prog(prog, sk);
1535 __bpf_prog_release(prog);
1541 EXPORT_SYMBOL_GPL(sk_attach_filter);
1543 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1545 struct bpf_prog *prog = __get_filter(fprog, sk);
1549 return PTR_ERR(prog);
1551 if (bpf_prog_size(prog->len) > sysctl_optmem_max)
1554 err = reuseport_attach_prog(sk, prog);
1557 __bpf_prog_release(prog);
1562 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1564 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1565 return ERR_PTR(-EPERM);
1567 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1570 int sk_attach_bpf(u32 ufd, struct sock *sk)
1572 struct bpf_prog *prog = __get_bpf(ufd, sk);
1576 return PTR_ERR(prog);
1578 err = __sk_attach_prog(prog, sk);
1587 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1589 struct bpf_prog *prog;
1592 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1595 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1596 if (PTR_ERR(prog) == -EINVAL)
1597 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1599 return PTR_ERR(prog);
1601 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1602 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1603 * bpf prog (e.g. sockmap). It depends on the
1604 * limitation imposed by bpf_prog_load().
1605 * Hence, sysctl_optmem_max is not checked.
1607 if ((sk->sk_type != SOCK_STREAM &&
1608 sk->sk_type != SOCK_DGRAM) ||
1609 (sk->sk_protocol != IPPROTO_UDP &&
1610 sk->sk_protocol != IPPROTO_TCP) ||
1611 (sk->sk_family != AF_INET &&
1612 sk->sk_family != AF_INET6)) {
1617 /* BPF_PROG_TYPE_SOCKET_FILTER */
1618 if (bpf_prog_size(prog->len) > sysctl_optmem_max) {
1624 err = reuseport_attach_prog(sk, prog);
1632 void sk_reuseport_prog_free(struct bpf_prog *prog)
1637 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1640 bpf_prog_destroy(prog);
1643 struct bpf_scratchpad {
1645 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1646 u8 buff[MAX_BPF_STACK];
1650 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1652 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1653 unsigned int write_len)
1655 return skb_ensure_writable(skb, write_len);
1658 static inline int bpf_try_make_writable(struct sk_buff *skb,
1659 unsigned int write_len)
1661 int err = __bpf_try_make_writable(skb, write_len);
1663 bpf_compute_data_pointers(skb);
1667 static int bpf_try_make_head_writable(struct sk_buff *skb)
1669 return bpf_try_make_writable(skb, skb_headlen(skb));
1672 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1674 if (skb_at_tc_ingress(skb))
1675 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1678 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1680 if (skb_at_tc_ingress(skb))
1681 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1684 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1685 const void *, from, u32, len, u64, flags)
1689 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1691 if (unlikely(offset > 0xffff))
1693 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1696 ptr = skb->data + offset;
1697 if (flags & BPF_F_RECOMPUTE_CSUM)
1698 __skb_postpull_rcsum(skb, ptr, len, offset);
1700 memcpy(ptr, from, len);
1702 if (flags & BPF_F_RECOMPUTE_CSUM)
1703 __skb_postpush_rcsum(skb, ptr, len, offset);
1704 if (flags & BPF_F_INVALIDATE_HASH)
1705 skb_clear_hash(skb);
1710 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1711 .func = bpf_skb_store_bytes,
1713 .ret_type = RET_INTEGER,
1714 .arg1_type = ARG_PTR_TO_CTX,
1715 .arg2_type = ARG_ANYTHING,
1716 .arg3_type = ARG_PTR_TO_MEM,
1717 .arg4_type = ARG_CONST_SIZE,
1718 .arg5_type = ARG_ANYTHING,
1721 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1722 void *, to, u32, len)
1726 if (unlikely(offset > 0xffff))
1729 ptr = skb_header_pointer(skb, offset, len, to);
1733 memcpy(to, ptr, len);
1741 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1742 .func = bpf_skb_load_bytes,
1744 .ret_type = RET_INTEGER,
1745 .arg1_type = ARG_PTR_TO_CTX,
1746 .arg2_type = ARG_ANYTHING,
1747 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1748 .arg4_type = ARG_CONST_SIZE,
1751 BPF_CALL_4(bpf_flow_dissector_load_bytes,
1752 const struct bpf_flow_dissector *, ctx, u32, offset,
1753 void *, to, u32, len)
1757 if (unlikely(offset > 0xffff))
1760 if (unlikely(!ctx->skb))
1763 ptr = skb_header_pointer(ctx->skb, offset, len, to);
1767 memcpy(to, ptr, len);
1775 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1776 .func = bpf_flow_dissector_load_bytes,
1778 .ret_type = RET_INTEGER,
1779 .arg1_type = ARG_PTR_TO_CTX,
1780 .arg2_type = ARG_ANYTHING,
1781 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1782 .arg4_type = ARG_CONST_SIZE,
1785 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1786 u32, offset, void *, to, u32, len, u32, start_header)
1788 u8 *end = skb_tail_pointer(skb);
1791 if (unlikely(offset > 0xffff))
1794 switch (start_header) {
1795 case BPF_HDR_START_MAC:
1796 if (unlikely(!skb_mac_header_was_set(skb)))
1798 start = skb_mac_header(skb);
1800 case BPF_HDR_START_NET:
1801 start = skb_network_header(skb);
1807 ptr = start + offset;
1809 if (likely(ptr + len <= end)) {
1810 memcpy(to, ptr, len);
1819 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1820 .func = bpf_skb_load_bytes_relative,
1822 .ret_type = RET_INTEGER,
1823 .arg1_type = ARG_PTR_TO_CTX,
1824 .arg2_type = ARG_ANYTHING,
1825 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1826 .arg4_type = ARG_CONST_SIZE,
1827 .arg5_type = ARG_ANYTHING,
1830 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1832 /* Idea is the following: should the needed direct read/write
1833 * test fail during runtime, we can pull in more data and redo
1834 * again, since implicitly, we invalidate previous checks here.
1836 * Or, since we know how much we need to make read/writeable,
1837 * this can be done once at the program beginning for direct
1838 * access case. By this we overcome limitations of only current
1839 * headroom being accessible.
1841 return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1844 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1845 .func = bpf_skb_pull_data,
1847 .ret_type = RET_INTEGER,
1848 .arg1_type = ARG_PTR_TO_CTX,
1849 .arg2_type = ARG_ANYTHING,
1852 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1854 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1857 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1858 .func = bpf_sk_fullsock,
1860 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
1861 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
1864 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1865 unsigned int write_len)
1867 return __bpf_try_make_writable(skb, write_len);
1870 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1872 /* Idea is the following: should the needed direct read/write
1873 * test fail during runtime, we can pull in more data and redo
1874 * again, since implicitly, we invalidate previous checks here.
1876 * Or, since we know how much we need to make read/writeable,
1877 * this can be done once at the program beginning for direct
1878 * access case. By this we overcome limitations of only current
1879 * headroom being accessible.
1881 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1884 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1885 .func = sk_skb_pull_data,
1887 .ret_type = RET_INTEGER,
1888 .arg1_type = ARG_PTR_TO_CTX,
1889 .arg2_type = ARG_ANYTHING,
1892 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1893 u64, from, u64, to, u64, flags)
1897 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1899 if (unlikely(offset > 0xffff || offset & 1))
1901 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1904 ptr = (__sum16 *)(skb->data + offset);
1905 switch (flags & BPF_F_HDR_FIELD_MASK) {
1907 if (unlikely(from != 0))
1910 csum_replace_by_diff(ptr, to);
1913 csum_replace2(ptr, from, to);
1916 csum_replace4(ptr, from, to);
1925 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1926 .func = bpf_l3_csum_replace,
1928 .ret_type = RET_INTEGER,
1929 .arg1_type = ARG_PTR_TO_CTX,
1930 .arg2_type = ARG_ANYTHING,
1931 .arg3_type = ARG_ANYTHING,
1932 .arg4_type = ARG_ANYTHING,
1933 .arg5_type = ARG_ANYTHING,
1936 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1937 u64, from, u64, to, u64, flags)
1939 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1940 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1941 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1944 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1945 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1947 if (unlikely(offset > 0xffff || offset & 1))
1949 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1952 ptr = (__sum16 *)(skb->data + offset);
1953 if (is_mmzero && !do_mforce && !*ptr)
1956 switch (flags & BPF_F_HDR_FIELD_MASK) {
1958 if (unlikely(from != 0))
1961 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1964 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1967 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1973 if (is_mmzero && !*ptr)
1974 *ptr = CSUM_MANGLED_0;
1978 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1979 .func = bpf_l4_csum_replace,
1981 .ret_type = RET_INTEGER,
1982 .arg1_type = ARG_PTR_TO_CTX,
1983 .arg2_type = ARG_ANYTHING,
1984 .arg3_type = ARG_ANYTHING,
1985 .arg4_type = ARG_ANYTHING,
1986 .arg5_type = ARG_ANYTHING,
1989 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1990 __be32 *, to, u32, to_size, __wsum, seed)
1992 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1993 u32 diff_size = from_size + to_size;
1996 /* This is quite flexible, some examples:
1998 * from_size == 0, to_size > 0, seed := csum --> pushing data
1999 * from_size > 0, to_size == 0, seed := csum --> pulling data
2000 * from_size > 0, to_size > 0, seed := 0 --> diffing data
2002 * Even for diffing, from_size and to_size don't need to be equal.
2004 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
2005 diff_size > sizeof(sp->diff)))
2008 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
2009 sp->diff[j] = ~from[i];
2010 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
2011 sp->diff[j] = to[i];
2013 return csum_partial(sp->diff, diff_size, seed);
2016 static const struct bpf_func_proto bpf_csum_diff_proto = {
2017 .func = bpf_csum_diff,
2020 .ret_type = RET_INTEGER,
2021 .arg1_type = ARG_PTR_TO_MEM_OR_NULL,
2022 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
2023 .arg3_type = ARG_PTR_TO_MEM_OR_NULL,
2024 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
2025 .arg5_type = ARG_ANYTHING,
2028 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2030 /* The interface is to be used in combination with bpf_csum_diff()
2031 * for direct packet writes. csum rotation for alignment as well
2032 * as emulating csum_sub() can be done from the eBPF program.
2034 if (skb->ip_summed == CHECKSUM_COMPLETE)
2035 return (skb->csum = csum_add(skb->csum, csum));
2040 static const struct bpf_func_proto bpf_csum_update_proto = {
2041 .func = bpf_csum_update,
2043 .ret_type = RET_INTEGER,
2044 .arg1_type = ARG_PTR_TO_CTX,
2045 .arg2_type = ARG_ANYTHING,
2048 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
2050 /* The interface is to be used in combination with bpf_skb_adjust_room()
2051 * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET
2052 * is passed as flags, for example.
2055 case BPF_CSUM_LEVEL_INC:
2056 __skb_incr_checksum_unnecessary(skb);
2058 case BPF_CSUM_LEVEL_DEC:
2059 __skb_decr_checksum_unnecessary(skb);
2061 case BPF_CSUM_LEVEL_RESET:
2062 __skb_reset_checksum_unnecessary(skb);
2064 case BPF_CSUM_LEVEL_QUERY:
2065 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2066 skb->csum_level : -EACCES;
2074 static const struct bpf_func_proto bpf_csum_level_proto = {
2075 .func = bpf_csum_level,
2077 .ret_type = RET_INTEGER,
2078 .arg1_type = ARG_PTR_TO_CTX,
2079 .arg2_type = ARG_ANYTHING,
2082 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2084 return dev_forward_skb_nomtu(dev, skb);
2087 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2088 struct sk_buff *skb)
2090 int ret = ____dev_forward_skb(dev, skb, false);
2094 ret = netif_rx(skb);
2100 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2104 if (dev_xmit_recursion()) {
2105 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2113 dev_xmit_recursion_inc();
2114 ret = dev_queue_xmit(skb);
2115 dev_xmit_recursion_dec();
2120 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2123 unsigned int mlen = skb_network_offset(skb);
2126 __skb_pull(skb, mlen);
2128 /* At ingress, the mac header has already been pulled once.
2129 * At egress, skb_pospull_rcsum has to be done in case that
2130 * the skb is originated from ingress (i.e. a forwarded skb)
2131 * to ensure that rcsum starts at net header.
2133 if (!skb_at_tc_ingress(skb))
2134 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2136 skb_pop_mac_header(skb);
2137 skb_reset_mac_len(skb);
2138 return flags & BPF_F_INGRESS ?
2139 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2142 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2145 /* Verify that a link layer header is carried */
2146 if (unlikely(skb->mac_header >= skb->network_header)) {
2151 bpf_push_mac_rcsum(skb);
2152 return flags & BPF_F_INGRESS ?
2153 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2156 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2159 if (dev_is_mac_header_xmit(dev))
2160 return __bpf_redirect_common(skb, dev, flags);
2162 return __bpf_redirect_no_mac(skb, dev, flags);
2165 #if IS_ENABLED(CONFIG_IPV6)
2166 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2167 struct net_device *dev, struct bpf_nh_params *nh)
2169 u32 hh_len = LL_RESERVED_SPACE(dev);
2170 const struct in6_addr *nexthop;
2171 struct dst_entry *dst = NULL;
2172 struct neighbour *neigh;
2174 if (dev_xmit_recursion()) {
2175 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2182 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2183 skb = skb_expand_head(skb, hh_len);
2191 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2192 &ipv6_hdr(skb)->daddr);
2194 nexthop = &nh->ipv6_nh;
2196 neigh = ip_neigh_gw6(dev, nexthop);
2197 if (likely(!IS_ERR(neigh))) {
2200 sock_confirm_neigh(skb, neigh);
2201 dev_xmit_recursion_inc();
2202 ret = neigh_output(neigh, skb, false);
2203 dev_xmit_recursion_dec();
2204 rcu_read_unlock_bh();
2207 rcu_read_unlock_bh();
2209 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2215 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2216 struct bpf_nh_params *nh)
2218 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2219 struct net *net = dev_net(dev);
2220 int err, ret = NET_XMIT_DROP;
2223 struct dst_entry *dst;
2224 struct flowi6 fl6 = {
2225 .flowi6_flags = FLOWI_FLAG_ANYSRC,
2226 .flowi6_mark = skb->mark,
2227 .flowlabel = ip6_flowinfo(ip6h),
2228 .flowi6_oif = dev->ifindex,
2229 .flowi6_proto = ip6h->nexthdr,
2230 .daddr = ip6h->daddr,
2231 .saddr = ip6h->saddr,
2234 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2238 skb_dst_set(skb, dst);
2239 } else if (nh->nh_family != AF_INET6) {
2243 err = bpf_out_neigh_v6(net, skb, dev, nh);
2244 if (unlikely(net_xmit_eval(err)))
2245 dev->stats.tx_errors++;
2247 ret = NET_XMIT_SUCCESS;
2250 dev->stats.tx_errors++;
2256 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2257 struct bpf_nh_params *nh)
2260 return NET_XMIT_DROP;
2262 #endif /* CONFIG_IPV6 */
2264 #if IS_ENABLED(CONFIG_INET)
2265 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2266 struct net_device *dev, struct bpf_nh_params *nh)
2268 u32 hh_len = LL_RESERVED_SPACE(dev);
2269 struct neighbour *neigh;
2270 bool is_v6gw = false;
2272 if (dev_xmit_recursion()) {
2273 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2280 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2281 skb = skb_expand_head(skb, hh_len);
2288 struct dst_entry *dst = skb_dst(skb);
2289 struct rtable *rt = container_of(dst, struct rtable, dst);
2291 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2292 } else if (nh->nh_family == AF_INET6) {
2293 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2295 } else if (nh->nh_family == AF_INET) {
2296 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2298 rcu_read_unlock_bh();
2302 if (likely(!IS_ERR(neigh))) {
2305 sock_confirm_neigh(skb, neigh);
2306 dev_xmit_recursion_inc();
2307 ret = neigh_output(neigh, skb, is_v6gw);
2308 dev_xmit_recursion_dec();
2309 rcu_read_unlock_bh();
2312 rcu_read_unlock_bh();
2318 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2319 struct bpf_nh_params *nh)
2321 const struct iphdr *ip4h = ip_hdr(skb);
2322 struct net *net = dev_net(dev);
2323 int err, ret = NET_XMIT_DROP;
2326 struct flowi4 fl4 = {
2327 .flowi4_flags = FLOWI_FLAG_ANYSRC,
2328 .flowi4_mark = skb->mark,
2329 .flowi4_tos = RT_TOS(ip4h->tos),
2330 .flowi4_oif = dev->ifindex,
2331 .flowi4_proto = ip4h->protocol,
2332 .daddr = ip4h->daddr,
2333 .saddr = ip4h->saddr,
2337 rt = ip_route_output_flow(net, &fl4, NULL);
2340 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2345 skb_dst_set(skb, &rt->dst);
2348 err = bpf_out_neigh_v4(net, skb, dev, nh);
2349 if (unlikely(net_xmit_eval(err)))
2350 dev->stats.tx_errors++;
2352 ret = NET_XMIT_SUCCESS;
2355 dev->stats.tx_errors++;
2361 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2362 struct bpf_nh_params *nh)
2365 return NET_XMIT_DROP;
2367 #endif /* CONFIG_INET */
2369 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2370 struct bpf_nh_params *nh)
2372 struct ethhdr *ethh = eth_hdr(skb);
2374 if (unlikely(skb->mac_header >= skb->network_header))
2376 bpf_push_mac_rcsum(skb);
2377 if (is_multicast_ether_addr(ethh->h_dest))
2380 skb_pull(skb, sizeof(*ethh));
2381 skb_unset_mac_header(skb);
2382 skb_reset_network_header(skb);
2384 if (skb->protocol == htons(ETH_P_IP))
2385 return __bpf_redirect_neigh_v4(skb, dev, nh);
2386 else if (skb->protocol == htons(ETH_P_IPV6))
2387 return __bpf_redirect_neigh_v6(skb, dev, nh);
2393 /* Internal, non-exposed redirect flags. */
2395 BPF_F_NEIGH = (1ULL << 1),
2396 BPF_F_PEER = (1ULL << 2),
2397 BPF_F_NEXTHOP = (1ULL << 3),
2398 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2401 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2403 struct net_device *dev;
2404 struct sk_buff *clone;
2407 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2410 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2414 clone = skb_clone(skb, GFP_ATOMIC);
2415 if (unlikely(!clone))
2418 /* For direct write, we need to keep the invariant that the skbs
2419 * we're dealing with need to be uncloned. Should uncloning fail
2420 * here, we need to free the just generated clone to unclone once
2423 ret = bpf_try_make_head_writable(skb);
2424 if (unlikely(ret)) {
2429 return __bpf_redirect(clone, dev, flags);
2432 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2433 .func = bpf_clone_redirect,
2435 .ret_type = RET_INTEGER,
2436 .arg1_type = ARG_PTR_TO_CTX,
2437 .arg2_type = ARG_ANYTHING,
2438 .arg3_type = ARG_ANYTHING,
2441 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2442 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2444 int skb_do_redirect(struct sk_buff *skb)
2446 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2447 struct net *net = dev_net(skb->dev);
2448 struct net_device *dev;
2449 u32 flags = ri->flags;
2451 dev = dev_get_by_index_rcu(net, ri->tgt_index);
2456 if (flags & BPF_F_PEER) {
2457 const struct net_device_ops *ops = dev->netdev_ops;
2459 if (unlikely(!ops->ndo_get_peer_dev ||
2460 !skb_at_tc_ingress(skb)))
2462 dev = ops->ndo_get_peer_dev(dev);
2463 if (unlikely(!dev ||
2464 !(dev->flags & IFF_UP) ||
2465 net_eq(net, dev_net(dev))))
2470 return flags & BPF_F_NEIGH ?
2471 __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2473 __bpf_redirect(skb, dev, flags);
2479 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2481 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2483 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2487 ri->tgt_index = ifindex;
2489 return TC_ACT_REDIRECT;
2492 static const struct bpf_func_proto bpf_redirect_proto = {
2493 .func = bpf_redirect,
2495 .ret_type = RET_INTEGER,
2496 .arg1_type = ARG_ANYTHING,
2497 .arg2_type = ARG_ANYTHING,
2500 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2502 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2504 if (unlikely(flags))
2507 ri->flags = BPF_F_PEER;
2508 ri->tgt_index = ifindex;
2510 return TC_ACT_REDIRECT;
2513 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2514 .func = bpf_redirect_peer,
2516 .ret_type = RET_INTEGER,
2517 .arg1_type = ARG_ANYTHING,
2518 .arg2_type = ARG_ANYTHING,
2521 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2522 int, plen, u64, flags)
2524 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2526 if (unlikely((plen && plen < sizeof(*params)) || flags))
2529 ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2530 ri->tgt_index = ifindex;
2532 BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2534 memcpy(&ri->nh, params, sizeof(ri->nh));
2536 return TC_ACT_REDIRECT;
2539 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2540 .func = bpf_redirect_neigh,
2542 .ret_type = RET_INTEGER,
2543 .arg1_type = ARG_ANYTHING,
2544 .arg2_type = ARG_PTR_TO_MEM_OR_NULL,
2545 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
2546 .arg4_type = ARG_ANYTHING,
2549 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2551 msg->apply_bytes = bytes;
2555 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2556 .func = bpf_msg_apply_bytes,
2558 .ret_type = RET_INTEGER,
2559 .arg1_type = ARG_PTR_TO_CTX,
2560 .arg2_type = ARG_ANYTHING,
2563 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2565 msg->cork_bytes = bytes;
2569 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2570 .func = bpf_msg_cork_bytes,
2572 .ret_type = RET_INTEGER,
2573 .arg1_type = ARG_PTR_TO_CTX,
2574 .arg2_type = ARG_ANYTHING,
2577 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2578 u32, end, u64, flags)
2580 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2581 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2582 struct scatterlist *sge;
2583 u8 *raw, *to, *from;
2586 if (unlikely(flags || end <= start))
2589 /* First find the starting scatterlist element */
2593 len = sk_msg_elem(msg, i)->length;
2594 if (start < offset + len)
2596 sk_msg_iter_var_next(i);
2597 } while (i != msg->sg.end);
2599 if (unlikely(start >= offset + len))
2603 /* The start may point into the sg element so we need to also
2604 * account for the headroom.
2606 bytes_sg_total = start - offset + bytes;
2607 if (!test_bit(i, &msg->sg.copy) && bytes_sg_total <= len)
2610 /* At this point we need to linearize multiple scatterlist
2611 * elements or a single shared page. Either way we need to
2612 * copy into a linear buffer exclusively owned by BPF. Then
2613 * place the buffer in the scatterlist and fixup the original
2614 * entries by removing the entries now in the linear buffer
2615 * and shifting the remaining entries. For now we do not try
2616 * to copy partial entries to avoid complexity of running out
2617 * of sg_entry slots. The downside is reading a single byte
2618 * will copy the entire sg entry.
2621 copy += sk_msg_elem(msg, i)->length;
2622 sk_msg_iter_var_next(i);
2623 if (bytes_sg_total <= copy)
2625 } while (i != msg->sg.end);
2628 if (unlikely(bytes_sg_total > copy))
2631 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2633 if (unlikely(!page))
2636 raw = page_address(page);
2639 sge = sk_msg_elem(msg, i);
2640 from = sg_virt(sge);
2644 memcpy(to, from, len);
2647 put_page(sg_page(sge));
2649 sk_msg_iter_var_next(i);
2650 } while (i != last_sge);
2652 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2654 /* To repair sg ring we need to shift entries. If we only
2655 * had a single entry though we can just replace it and
2656 * be done. Otherwise walk the ring and shift the entries.
2658 WARN_ON_ONCE(last_sge == first_sge);
2659 shift = last_sge > first_sge ?
2660 last_sge - first_sge - 1 :
2661 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2666 sk_msg_iter_var_next(i);
2670 if (i + shift >= NR_MSG_FRAG_IDS)
2671 move_from = i + shift - NR_MSG_FRAG_IDS;
2673 move_from = i + shift;
2674 if (move_from == msg->sg.end)
2677 msg->sg.data[i] = msg->sg.data[move_from];
2678 msg->sg.data[move_from].length = 0;
2679 msg->sg.data[move_from].page_link = 0;
2680 msg->sg.data[move_from].offset = 0;
2681 sk_msg_iter_var_next(i);
2684 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2685 msg->sg.end - shift + NR_MSG_FRAG_IDS :
2686 msg->sg.end - shift;
2688 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2689 msg->data_end = msg->data + bytes;
2693 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2694 .func = bpf_msg_pull_data,
2696 .ret_type = RET_INTEGER,
2697 .arg1_type = ARG_PTR_TO_CTX,
2698 .arg2_type = ARG_ANYTHING,
2699 .arg3_type = ARG_ANYTHING,
2700 .arg4_type = ARG_ANYTHING,
2703 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2704 u32, len, u64, flags)
2706 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2707 u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2708 u8 *raw, *to, *from;
2711 if (unlikely(flags))
2714 /* First find the starting scatterlist element */
2718 l = sk_msg_elem(msg, i)->length;
2720 if (start < offset + l)
2722 sk_msg_iter_var_next(i);
2723 } while (i != msg->sg.end);
2725 if (start >= offset + l)
2728 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2730 /* If no space available will fallback to copy, we need at
2731 * least one scatterlist elem available to push data into
2732 * when start aligns to the beginning of an element or two
2733 * when it falls inside an element. We handle the start equals
2734 * offset case because its the common case for inserting a
2737 if (!space || (space == 1 && start != offset))
2738 copy = msg->sg.data[i].length;
2740 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2741 get_order(copy + len));
2742 if (unlikely(!page))
2748 raw = page_address(page);
2750 psge = sk_msg_elem(msg, i);
2751 front = start - offset;
2752 back = psge->length - front;
2753 from = sg_virt(psge);
2756 memcpy(raw, from, front);
2760 to = raw + front + len;
2762 memcpy(to, from, back);
2765 put_page(sg_page(psge));
2766 } else if (start - offset) {
2767 psge = sk_msg_elem(msg, i);
2768 rsge = sk_msg_elem_cpy(msg, i);
2770 psge->length = start - offset;
2771 rsge.length -= psge->length;
2772 rsge.offset += start;
2774 sk_msg_iter_var_next(i);
2775 sg_unmark_end(psge);
2776 sg_unmark_end(&rsge);
2777 sk_msg_iter_next(msg, end);
2780 /* Slot(s) to place newly allocated data */
2783 /* Shift one or two slots as needed */
2785 sge = sk_msg_elem_cpy(msg, i);
2787 sk_msg_iter_var_next(i);
2788 sg_unmark_end(&sge);
2789 sk_msg_iter_next(msg, end);
2791 nsge = sk_msg_elem_cpy(msg, i);
2793 sk_msg_iter_var_next(i);
2794 nnsge = sk_msg_elem_cpy(msg, i);
2797 while (i != msg->sg.end) {
2798 msg->sg.data[i] = sge;
2800 sk_msg_iter_var_next(i);
2803 nnsge = sk_msg_elem_cpy(msg, i);
2805 nsge = sk_msg_elem_cpy(msg, i);
2810 /* Place newly allocated data buffer */
2811 sk_mem_charge(msg->sk, len);
2812 msg->sg.size += len;
2813 __clear_bit(new, &msg->sg.copy);
2814 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2816 get_page(sg_page(&rsge));
2817 sk_msg_iter_var_next(new);
2818 msg->sg.data[new] = rsge;
2821 sk_msg_compute_data_pointers(msg);
2825 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2826 .func = bpf_msg_push_data,
2828 .ret_type = RET_INTEGER,
2829 .arg1_type = ARG_PTR_TO_CTX,
2830 .arg2_type = ARG_ANYTHING,
2831 .arg3_type = ARG_ANYTHING,
2832 .arg4_type = ARG_ANYTHING,
2835 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2841 sk_msg_iter_var_next(i);
2842 msg->sg.data[prev] = msg->sg.data[i];
2843 } while (i != msg->sg.end);
2845 sk_msg_iter_prev(msg, end);
2848 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2850 struct scatterlist tmp, sge;
2852 sk_msg_iter_next(msg, end);
2853 sge = sk_msg_elem_cpy(msg, i);
2854 sk_msg_iter_var_next(i);
2855 tmp = sk_msg_elem_cpy(msg, i);
2857 while (i != msg->sg.end) {
2858 msg->sg.data[i] = sge;
2859 sk_msg_iter_var_next(i);
2861 tmp = sk_msg_elem_cpy(msg, i);
2865 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2866 u32, len, u64, flags)
2868 u32 i = 0, l = 0, space, offset = 0;
2869 u64 last = start + len;
2872 if (unlikely(flags))
2875 /* First find the starting scatterlist element */
2879 l = sk_msg_elem(msg, i)->length;
2881 if (start < offset + l)
2883 sk_msg_iter_var_next(i);
2884 } while (i != msg->sg.end);
2886 /* Bounds checks: start and pop must be inside message */
2887 if (start >= offset + l || last >= msg->sg.size)
2890 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2893 /* --------------| offset
2894 * -| start |-------- len -------|
2896 * |----- a ----|-------- pop -------|----- b ----|
2897 * |______________________________________________| length
2900 * a: region at front of scatter element to save
2901 * b: region at back of scatter element to save when length > A + pop
2902 * pop: region to pop from element, same as input 'pop' here will be
2903 * decremented below per iteration.
2905 * Two top-level cases to handle when start != offset, first B is non
2906 * zero and second B is zero corresponding to when a pop includes more
2909 * Then if B is non-zero AND there is no space allocate space and
2910 * compact A, B regions into page. If there is space shift ring to
2911 * the rigth free'ing the next element in ring to place B, leaving
2912 * A untouched except to reduce length.
2914 if (start != offset) {
2915 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2917 int b = sge->length - pop - a;
2919 sk_msg_iter_var_next(i);
2921 if (pop < sge->length - a) {
2924 sk_msg_shift_right(msg, i);
2925 nsge = sk_msg_elem(msg, i);
2926 get_page(sg_page(sge));
2929 b, sge->offset + pop + a);
2931 struct page *page, *orig;
2934 page = alloc_pages(__GFP_NOWARN |
2935 __GFP_COMP | GFP_ATOMIC,
2937 if (unlikely(!page))
2941 orig = sg_page(sge);
2942 from = sg_virt(sge);
2943 to = page_address(page);
2944 memcpy(to, from, a);
2945 memcpy(to + a, from + a + pop, b);
2946 sg_set_page(sge, page, a + b, 0);
2950 } else if (pop >= sge->length - a) {
2951 pop -= (sge->length - a);
2956 /* From above the current layout _must_ be as follows,
2961 * |---- pop ---|---------------- b ------------|
2962 * |____________________________________________| length
2964 * Offset and start of the current msg elem are equal because in the
2965 * previous case we handled offset != start and either consumed the
2966 * entire element and advanced to the next element OR pop == 0.
2968 * Two cases to handle here are first pop is less than the length
2969 * leaving some remainder b above. Simply adjust the element's layout
2970 * in this case. Or pop >= length of the element so that b = 0. In this
2971 * case advance to next element decrementing pop.
2974 struct scatterlist *sge = sk_msg_elem(msg, i);
2976 if (pop < sge->length) {
2982 sk_msg_shift_left(msg, i);
2984 sk_msg_iter_var_next(i);
2987 sk_mem_uncharge(msg->sk, len - pop);
2988 msg->sg.size -= (len - pop);
2989 sk_msg_compute_data_pointers(msg);
2993 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
2994 .func = bpf_msg_pop_data,
2996 .ret_type = RET_INTEGER,
2997 .arg1_type = ARG_PTR_TO_CTX,
2998 .arg2_type = ARG_ANYTHING,
2999 .arg3_type = ARG_ANYTHING,
3000 .arg4_type = ARG_ANYTHING,
3003 #ifdef CONFIG_CGROUP_NET_CLASSID
3004 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3006 return __task_get_classid(current);
3009 static const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3010 .func = bpf_get_cgroup_classid_curr,
3012 .ret_type = RET_INTEGER,
3015 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3017 struct sock *sk = skb_to_full_sk(skb);
3019 if (!sk || !sk_fullsock(sk))
3022 return sock_cgroup_classid(&sk->sk_cgrp_data);
3025 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3026 .func = bpf_skb_cgroup_classid,
3028 .ret_type = RET_INTEGER,
3029 .arg1_type = ARG_PTR_TO_CTX,
3033 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3035 return task_get_classid(skb);
3038 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3039 .func = bpf_get_cgroup_classid,
3041 .ret_type = RET_INTEGER,
3042 .arg1_type = ARG_PTR_TO_CTX,
3045 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3047 return dst_tclassid(skb);
3050 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3051 .func = bpf_get_route_realm,
3053 .ret_type = RET_INTEGER,
3054 .arg1_type = ARG_PTR_TO_CTX,
3057 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3059 /* If skb_clear_hash() was called due to mangling, we can
3060 * trigger SW recalculation here. Later access to hash
3061 * can then use the inline skb->hash via context directly
3062 * instead of calling this helper again.
3064 return skb_get_hash(skb);
3067 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3068 .func = bpf_get_hash_recalc,
3070 .ret_type = RET_INTEGER,
3071 .arg1_type = ARG_PTR_TO_CTX,
3074 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3076 /* After all direct packet write, this can be used once for
3077 * triggering a lazy recalc on next skb_get_hash() invocation.
3079 skb_clear_hash(skb);
3083 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3084 .func = bpf_set_hash_invalid,
3086 .ret_type = RET_INTEGER,
3087 .arg1_type = ARG_PTR_TO_CTX,
3090 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3092 /* Set user specified hash as L4(+), so that it gets returned
3093 * on skb_get_hash() call unless BPF prog later on triggers a
3096 __skb_set_sw_hash(skb, hash, true);
3100 static const struct bpf_func_proto bpf_set_hash_proto = {
3101 .func = bpf_set_hash,
3103 .ret_type = RET_INTEGER,
3104 .arg1_type = ARG_PTR_TO_CTX,
3105 .arg2_type = ARG_ANYTHING,
3108 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3113 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3114 vlan_proto != htons(ETH_P_8021AD)))
3115 vlan_proto = htons(ETH_P_8021Q);
3117 bpf_push_mac_rcsum(skb);
3118 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3119 bpf_pull_mac_rcsum(skb);
3121 bpf_compute_data_pointers(skb);
3125 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3126 .func = bpf_skb_vlan_push,
3128 .ret_type = RET_INTEGER,
3129 .arg1_type = ARG_PTR_TO_CTX,
3130 .arg2_type = ARG_ANYTHING,
3131 .arg3_type = ARG_ANYTHING,
3134 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3138 bpf_push_mac_rcsum(skb);
3139 ret = skb_vlan_pop(skb);
3140 bpf_pull_mac_rcsum(skb);
3142 bpf_compute_data_pointers(skb);
3146 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3147 .func = bpf_skb_vlan_pop,
3149 .ret_type = RET_INTEGER,
3150 .arg1_type = ARG_PTR_TO_CTX,
3153 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3155 /* Caller already did skb_cow() with len as headroom,
3156 * so no need to do it here.
3159 memmove(skb->data, skb->data + len, off);
3160 memset(skb->data + off, 0, len);
3162 /* No skb_postpush_rcsum(skb, skb->data + off, len)
3163 * needed here as it does not change the skb->csum
3164 * result for checksum complete when summing over
3170 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3172 /* skb_ensure_writable() is not needed here, as we're
3173 * already working on an uncloned skb.
3175 if (unlikely(!pskb_may_pull(skb, off + len)))
3178 skb_postpull_rcsum(skb, skb->data + off, len);
3179 memmove(skb->data + len, skb->data, off);
3180 __skb_pull(skb, len);
3185 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3187 bool trans_same = skb->transport_header == skb->network_header;
3190 /* There's no need for __skb_push()/__skb_pull() pair to
3191 * get to the start of the mac header as we're guaranteed
3192 * to always start from here under eBPF.
3194 ret = bpf_skb_generic_push(skb, off, len);
3196 skb->mac_header -= len;
3197 skb->network_header -= len;
3199 skb->transport_header = skb->network_header;
3205 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3207 bool trans_same = skb->transport_header == skb->network_header;
3210 /* Same here, __skb_push()/__skb_pull() pair not needed. */
3211 ret = bpf_skb_generic_pop(skb, off, len);
3213 skb->mac_header += len;
3214 skb->network_header += len;
3216 skb->transport_header = skb->network_header;
3222 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3224 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3225 u32 off = skb_mac_header_len(skb);
3228 ret = skb_cow(skb, len_diff);
3229 if (unlikely(ret < 0))
3232 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3233 if (unlikely(ret < 0))
3236 if (skb_is_gso(skb)) {
3237 struct skb_shared_info *shinfo = skb_shinfo(skb);
3239 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */
3240 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3241 shinfo->gso_type &= ~SKB_GSO_TCPV4;
3242 shinfo->gso_type |= SKB_GSO_TCPV6;
3246 skb->protocol = htons(ETH_P_IPV6);
3247 skb_clear_hash(skb);
3252 static int bpf_skb_proto_6_to_4(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_unclone(skb, GFP_ATOMIC);
3259 if (unlikely(ret < 0))
3262 ret = bpf_skb_net_hdr_pop(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_TCPV6 needs to be changed into SKB_GSO_TCPV4. */
3270 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3271 shinfo->gso_type &= ~SKB_GSO_TCPV6;
3272 shinfo->gso_type |= SKB_GSO_TCPV4;
3276 skb->protocol = htons(ETH_P_IP);
3277 skb_clear_hash(skb);
3282 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3284 __be16 from_proto = skb->protocol;
3286 if (from_proto == htons(ETH_P_IP) &&
3287 to_proto == htons(ETH_P_IPV6))
3288 return bpf_skb_proto_4_to_6(skb);
3290 if (from_proto == htons(ETH_P_IPV6) &&
3291 to_proto == htons(ETH_P_IP))
3292 return bpf_skb_proto_6_to_4(skb);
3297 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3302 if (unlikely(flags))
3305 /* General idea is that this helper does the basic groundwork
3306 * needed for changing the protocol, and eBPF program fills the
3307 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3308 * and other helpers, rather than passing a raw buffer here.
3310 * The rationale is to keep this minimal and without a need to
3311 * deal with raw packet data. F.e. even if we would pass buffers
3312 * here, the program still needs to call the bpf_lX_csum_replace()
3313 * helpers anyway. Plus, this way we keep also separation of
3314 * concerns, since f.e. bpf_skb_store_bytes() should only take
3317 * Currently, additional options and extension header space are
3318 * not supported, but flags register is reserved so we can adapt
3319 * that. For offloads, we mark packet as dodgy, so that headers
3320 * need to be verified first.
3322 ret = bpf_skb_proto_xlat(skb, proto);
3323 bpf_compute_data_pointers(skb);
3327 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3328 .func = bpf_skb_change_proto,
3330 .ret_type = RET_INTEGER,
3331 .arg1_type = ARG_PTR_TO_CTX,
3332 .arg2_type = ARG_ANYTHING,
3333 .arg3_type = ARG_ANYTHING,
3336 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3338 /* We only allow a restricted subset to be changed for now. */
3339 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3340 !skb_pkt_type_ok(pkt_type)))
3343 skb->pkt_type = pkt_type;
3347 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3348 .func = bpf_skb_change_type,
3350 .ret_type = RET_INTEGER,
3351 .arg1_type = ARG_PTR_TO_CTX,
3352 .arg2_type = ARG_ANYTHING,
3355 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3357 switch (skb->protocol) {
3358 case htons(ETH_P_IP):
3359 return sizeof(struct iphdr);
3360 case htons(ETH_P_IPV6):
3361 return sizeof(struct ipv6hdr);
3367 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3368 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3370 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \
3371 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3372 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3373 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3374 BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \
3375 BPF_F_ADJ_ROOM_ENCAP_L2( \
3376 BPF_ADJ_ROOM_ENCAP_L2_MASK))
3378 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3381 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3382 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3383 u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3384 unsigned int gso_type = SKB_GSO_DODGY;
3387 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3388 /* udp gso_size delineates datagrams, only allow if fixed */
3389 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3390 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3394 ret = skb_cow_head(skb, len_diff);
3395 if (unlikely(ret < 0))
3399 if (skb->protocol != htons(ETH_P_IP) &&
3400 skb->protocol != htons(ETH_P_IPV6))
3403 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3404 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3407 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3408 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3411 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH &&
3412 inner_mac_len < ETH_HLEN)
3415 if (skb->encapsulation)
3418 mac_len = skb->network_header - skb->mac_header;
3419 inner_net = skb->network_header;
3420 if (inner_mac_len > len_diff)
3422 inner_trans = skb->transport_header;
3425 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3426 if (unlikely(ret < 0))
3430 skb->inner_mac_header = inner_net - inner_mac_len;
3431 skb->inner_network_header = inner_net;
3432 skb->inner_transport_header = inner_trans;
3434 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH)
3435 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
3437 skb_set_inner_protocol(skb, skb->protocol);
3439 skb->encapsulation = 1;
3440 skb_set_network_header(skb, mac_len);
3442 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3443 gso_type |= SKB_GSO_UDP_TUNNEL;
3444 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3445 gso_type |= SKB_GSO_GRE;
3446 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3447 gso_type |= SKB_GSO_IPXIP6;
3448 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3449 gso_type |= SKB_GSO_IPXIP4;
3451 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3452 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3453 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3454 sizeof(struct ipv6hdr) :
3455 sizeof(struct iphdr);
3457 skb_set_transport_header(skb, mac_len + nh_len);
3460 /* Match skb->protocol to new outer l3 protocol */
3461 if (skb->protocol == htons(ETH_P_IP) &&
3462 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3463 skb->protocol = htons(ETH_P_IPV6);
3464 else if (skb->protocol == htons(ETH_P_IPV6) &&
3465 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3466 skb->protocol = htons(ETH_P_IP);
3469 if (skb_is_gso(skb)) {
3470 struct skb_shared_info *shinfo = skb_shinfo(skb);
3472 /* Due to header grow, MSS needs to be downgraded. */
3473 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3474 skb_decrease_gso_size(shinfo, len_diff);
3476 /* Header must be checked, and gso_segs recomputed. */
3477 shinfo->gso_type |= gso_type;
3478 shinfo->gso_segs = 0;
3484 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3489 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3490 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3493 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3494 /* udp gso_size delineates datagrams, only allow if fixed */
3495 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3496 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3500 ret = skb_unclone(skb, GFP_ATOMIC);
3501 if (unlikely(ret < 0))
3504 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3505 if (unlikely(ret < 0))
3508 if (skb_is_gso(skb)) {
3509 struct skb_shared_info *shinfo = skb_shinfo(skb);
3511 /* Due to header shrink, MSS can be upgraded. */
3512 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3513 skb_increase_gso_size(shinfo, len_diff);
3515 /* Header must be checked, and gso_segs recomputed. */
3516 shinfo->gso_type |= SKB_GSO_DODGY;
3517 shinfo->gso_segs = 0;
3523 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3525 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3526 u32, mode, u64, flags)
3528 u32 len_diff_abs = abs(len_diff);
3529 bool shrink = len_diff < 0;
3532 if (unlikely(flags || mode))
3534 if (unlikely(len_diff_abs > 0xfffU))
3538 ret = skb_cow(skb, len_diff);
3539 if (unlikely(ret < 0))
3541 __skb_push(skb, len_diff_abs);
3542 memset(skb->data, 0, len_diff_abs);
3544 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3546 __skb_pull(skb, len_diff_abs);
3548 if (tls_sw_has_ctx_rx(skb->sk)) {
3549 struct strp_msg *rxm = strp_msg(skb);
3551 rxm->full_len += len_diff;
3556 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3557 .func = sk_skb_adjust_room,
3559 .ret_type = RET_INTEGER,
3560 .arg1_type = ARG_PTR_TO_CTX,
3561 .arg2_type = ARG_ANYTHING,
3562 .arg3_type = ARG_ANYTHING,
3563 .arg4_type = ARG_ANYTHING,
3566 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3567 u32, mode, u64, flags)
3569 u32 len_cur, len_diff_abs = abs(len_diff);
3570 u32 len_min = bpf_skb_net_base_len(skb);
3571 u32 len_max = BPF_SKB_MAX_LEN;
3572 __be16 proto = skb->protocol;
3573 bool shrink = len_diff < 0;
3577 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3578 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3580 if (unlikely(len_diff_abs > 0xfffU))
3582 if (unlikely(proto != htons(ETH_P_IP) &&
3583 proto != htons(ETH_P_IPV6)))
3586 off = skb_mac_header_len(skb);
3588 case BPF_ADJ_ROOM_NET:
3589 off += bpf_skb_net_base_len(skb);
3591 case BPF_ADJ_ROOM_MAC:
3597 len_cur = skb->len - skb_network_offset(skb);
3598 if ((shrink && (len_diff_abs >= len_cur ||
3599 len_cur - len_diff_abs < len_min)) ||
3600 (!shrink && (skb->len + len_diff_abs > len_max &&
3604 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3605 bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3606 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3607 __skb_reset_checksum_unnecessary(skb);
3609 bpf_compute_data_pointers(skb);
3613 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3614 .func = bpf_skb_adjust_room,
3616 .ret_type = RET_INTEGER,
3617 .arg1_type = ARG_PTR_TO_CTX,
3618 .arg2_type = ARG_ANYTHING,
3619 .arg3_type = ARG_ANYTHING,
3620 .arg4_type = ARG_ANYTHING,
3623 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3625 u32 min_len = skb_network_offset(skb);
3627 if (skb_transport_header_was_set(skb))
3628 min_len = skb_transport_offset(skb);
3629 if (skb->ip_summed == CHECKSUM_PARTIAL)
3630 min_len = skb_checksum_start_offset(skb) +
3631 skb->csum_offset + sizeof(__sum16);
3635 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3637 unsigned int old_len = skb->len;
3640 ret = __skb_grow_rcsum(skb, new_len);
3642 memset(skb->data + old_len, 0, new_len - old_len);
3646 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3648 return __skb_trim_rcsum(skb, new_len);
3651 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3654 u32 max_len = BPF_SKB_MAX_LEN;
3655 u32 min_len = __bpf_skb_min_len(skb);
3658 if (unlikely(flags || new_len > max_len || new_len < min_len))
3660 if (skb->encapsulation)
3663 /* The basic idea of this helper is that it's performing the
3664 * needed work to either grow or trim an skb, and eBPF program
3665 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3666 * bpf_lX_csum_replace() and others rather than passing a raw
3667 * buffer here. This one is a slow path helper and intended
3668 * for replies with control messages.
3670 * Like in bpf_skb_change_proto(), we want to keep this rather
3671 * minimal and without protocol specifics so that we are able
3672 * to separate concerns as in bpf_skb_store_bytes() should only
3673 * be the one responsible for writing buffers.
3675 * It's really expected to be a slow path operation here for
3676 * control message replies, so we're implicitly linearizing,
3677 * uncloning and drop offloads from the skb by this.
3679 ret = __bpf_try_make_writable(skb, skb->len);
3681 if (new_len > skb->len)
3682 ret = bpf_skb_grow_rcsum(skb, new_len);
3683 else if (new_len < skb->len)
3684 ret = bpf_skb_trim_rcsum(skb, new_len);
3685 if (!ret && skb_is_gso(skb))
3691 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3694 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3696 bpf_compute_data_pointers(skb);
3700 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3701 .func = bpf_skb_change_tail,
3703 .ret_type = RET_INTEGER,
3704 .arg1_type = ARG_PTR_TO_CTX,
3705 .arg2_type = ARG_ANYTHING,
3706 .arg3_type = ARG_ANYTHING,
3709 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3712 return __bpf_skb_change_tail(skb, new_len, flags);
3715 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3716 .func = sk_skb_change_tail,
3718 .ret_type = RET_INTEGER,
3719 .arg1_type = ARG_PTR_TO_CTX,
3720 .arg2_type = ARG_ANYTHING,
3721 .arg3_type = ARG_ANYTHING,
3724 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3727 u32 max_len = BPF_SKB_MAX_LEN;
3728 u32 new_len = skb->len + head_room;
3731 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3732 new_len < skb->len))
3735 ret = skb_cow(skb, head_room);
3737 /* Idea for this helper is that we currently only
3738 * allow to expand on mac header. This means that
3739 * skb->protocol network header, etc, stay as is.
3740 * Compared to bpf_skb_change_tail(), we're more
3741 * flexible due to not needing to linearize or
3742 * reset GSO. Intention for this helper is to be
3743 * used by an L3 skb that needs to push mac header
3744 * for redirection into L2 device.
3746 __skb_push(skb, head_room);
3747 memset(skb->data, 0, head_room);
3748 skb_reset_mac_header(skb);
3749 skb_reset_mac_len(skb);
3755 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3758 int ret = __bpf_skb_change_head(skb, head_room, flags);
3760 bpf_compute_data_pointers(skb);
3764 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3765 .func = bpf_skb_change_head,
3767 .ret_type = RET_INTEGER,
3768 .arg1_type = ARG_PTR_TO_CTX,
3769 .arg2_type = ARG_ANYTHING,
3770 .arg3_type = ARG_ANYTHING,
3773 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3776 return __bpf_skb_change_head(skb, head_room, flags);
3779 static const struct bpf_func_proto sk_skb_change_head_proto = {
3780 .func = sk_skb_change_head,
3782 .ret_type = RET_INTEGER,
3783 .arg1_type = ARG_PTR_TO_CTX,
3784 .arg2_type = ARG_ANYTHING,
3785 .arg3_type = ARG_ANYTHING,
3787 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3789 return xdp_data_meta_unsupported(xdp) ? 0 :
3790 xdp->data - xdp->data_meta;
3793 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3795 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3796 unsigned long metalen = xdp_get_metalen(xdp);
3797 void *data_start = xdp_frame_end + metalen;
3798 void *data = xdp->data + offset;
3800 if (unlikely(data < data_start ||
3801 data > xdp->data_end - ETH_HLEN))
3805 memmove(xdp->data_meta + offset,
3806 xdp->data_meta, metalen);
3807 xdp->data_meta += offset;
3813 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3814 .func = bpf_xdp_adjust_head,
3816 .ret_type = RET_INTEGER,
3817 .arg1_type = ARG_PTR_TO_CTX,
3818 .arg2_type = ARG_ANYTHING,
3821 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3823 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
3824 void *data_end = xdp->data_end + offset;
3826 /* Notice that xdp_data_hard_end have reserved some tailroom */
3827 if (unlikely(data_end > data_hard_end))
3830 /* ALL drivers MUST init xdp->frame_sz, chicken check below */
3831 if (unlikely(xdp->frame_sz > PAGE_SIZE)) {
3832 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz);
3836 if (unlikely(data_end < xdp->data + ETH_HLEN))
3839 /* Clear memory area on grow, can contain uninit kernel memory */
3841 memset(xdp->data_end, 0, offset);
3843 xdp->data_end = data_end;
3848 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3849 .func = bpf_xdp_adjust_tail,
3851 .ret_type = RET_INTEGER,
3852 .arg1_type = ARG_PTR_TO_CTX,
3853 .arg2_type = ARG_ANYTHING,
3856 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3858 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3859 void *meta = xdp->data_meta + offset;
3860 unsigned long metalen = xdp->data - meta;
3862 if (xdp_data_meta_unsupported(xdp))
3864 if (unlikely(meta < xdp_frame_end ||
3867 if (unlikely(xdp_metalen_invalid(metalen)))
3870 xdp->data_meta = meta;
3875 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3876 .func = bpf_xdp_adjust_meta,
3878 .ret_type = RET_INTEGER,
3879 .arg1_type = ARG_PTR_TO_CTX,
3880 .arg2_type = ARG_ANYTHING,
3883 /* XDP_REDIRECT works by a three-step process, implemented in the functions
3886 * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target
3887 * of the redirect and store it (along with some other metadata) in a per-CPU
3888 * struct bpf_redirect_info.
3890 * 2. When the program returns the XDP_REDIRECT return code, the driver will
3891 * call xdp_do_redirect() which will use the information in struct
3892 * bpf_redirect_info to actually enqueue the frame into a map type-specific
3893 * bulk queue structure.
3895 * 3. Before exiting its NAPI poll loop, the driver will call xdp_do_flush(),
3896 * which will flush all the different bulk queues, thus completing the
3899 * Pointers to the map entries will be kept around for this whole sequence of
3900 * steps, protected by RCU. However, there is no top-level rcu_read_lock() in
3901 * the core code; instead, the RCU protection relies on everything happening
3902 * inside a single NAPI poll sequence, which means it's between a pair of calls
3903 * to local_bh_disable()/local_bh_enable().
3905 * The map entries are marked as __rcu and the map code makes sure to
3906 * dereference those pointers with rcu_dereference_check() in a way that works
3907 * for both sections that to hold an rcu_read_lock() and sections that are
3908 * called from NAPI without a separate rcu_read_lock(). The code below does not
3909 * use RCU annotations, but relies on those in the map code.
3911 void xdp_do_flush(void)
3917 EXPORT_SYMBOL_GPL(xdp_do_flush);
3919 void bpf_clear_redirect_map(struct bpf_map *map)
3921 struct bpf_redirect_info *ri;
3924 for_each_possible_cpu(cpu) {
3925 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3926 /* Avoid polluting remote cacheline due to writes if
3927 * not needed. Once we pass this test, we need the
3928 * cmpxchg() to make sure it hasn't been changed in
3929 * the meantime by remote CPU.
3931 if (unlikely(READ_ONCE(ri->map) == map))
3932 cmpxchg(&ri->map, map, NULL);
3936 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3937 struct bpf_prog *xdp_prog)
3939 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3940 enum bpf_map_type map_type = ri->map_type;
3941 void *fwd = ri->tgt_value;
3942 u32 map_id = ri->map_id;
3943 struct bpf_map *map;
3946 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
3947 ri->map_type = BPF_MAP_TYPE_UNSPEC;
3950 case BPF_MAP_TYPE_DEVMAP:
3952 case BPF_MAP_TYPE_DEVMAP_HASH:
3953 map = READ_ONCE(ri->map);
3954 if (unlikely(map)) {
3955 WRITE_ONCE(ri->map, NULL);
3956 err = dev_map_enqueue_multi(xdp, dev, map,
3957 ri->flags & BPF_F_EXCLUDE_INGRESS);
3959 err = dev_map_enqueue(fwd, xdp, dev);
3962 case BPF_MAP_TYPE_CPUMAP:
3963 err = cpu_map_enqueue(fwd, xdp, dev);
3965 case BPF_MAP_TYPE_XSKMAP:
3966 err = __xsk_map_redirect(fwd, xdp);
3968 case BPF_MAP_TYPE_UNSPEC:
3969 if (map_id == INT_MAX) {
3970 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
3971 if (unlikely(!fwd)) {
3975 err = dev_xdp_enqueue(fwd, xdp, dev);
3986 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
3989 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
3992 EXPORT_SYMBOL_GPL(xdp_do_redirect);
3994 static int xdp_do_generic_redirect_map(struct net_device *dev,
3995 struct sk_buff *skb,
3996 struct xdp_buff *xdp,
3997 struct bpf_prog *xdp_prog,
3999 enum bpf_map_type map_type, u32 map_id)
4001 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4002 struct bpf_map *map;
4006 case BPF_MAP_TYPE_DEVMAP:
4008 case BPF_MAP_TYPE_DEVMAP_HASH:
4009 map = READ_ONCE(ri->map);
4010 if (unlikely(map)) {
4011 WRITE_ONCE(ri->map, NULL);
4012 err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4013 ri->flags & BPF_F_EXCLUDE_INGRESS);
4015 err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4020 case BPF_MAP_TYPE_XSKMAP:
4021 err = xsk_generic_rcv(fwd, xdp);
4026 case BPF_MAP_TYPE_CPUMAP:
4027 err = cpu_map_generic_redirect(fwd, skb);
4036 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4039 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4043 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4044 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4046 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4047 enum bpf_map_type map_type = ri->map_type;
4048 void *fwd = ri->tgt_value;
4049 u32 map_id = ri->map_id;
4052 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4053 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4055 if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4056 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4057 if (unlikely(!fwd)) {
4062 err = xdp_ok_fwd_dev(fwd, skb->len);
4067 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4068 generic_xdp_tx(skb, xdp_prog);
4072 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id);
4074 _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4078 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4080 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4082 if (unlikely(flags))
4085 /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4086 * by map_idr) is used for ifindex based XDP redirect.
4088 ri->tgt_index = ifindex;
4089 ri->map_id = INT_MAX;
4090 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4092 return XDP_REDIRECT;
4095 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4096 .func = bpf_xdp_redirect,
4098 .ret_type = RET_INTEGER,
4099 .arg1_type = ARG_ANYTHING,
4100 .arg2_type = ARG_ANYTHING,
4103 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
4106 return map->ops->map_redirect(map, ifindex, flags);
4109 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4110 .func = bpf_xdp_redirect_map,
4112 .ret_type = RET_INTEGER,
4113 .arg1_type = ARG_CONST_MAP_PTR,
4114 .arg2_type = ARG_ANYTHING,
4115 .arg3_type = ARG_ANYTHING,
4118 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4119 unsigned long off, unsigned long len)
4121 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4125 if (ptr != dst_buff)
4126 memcpy(dst_buff, ptr, len);
4131 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4132 u64, flags, void *, meta, u64, meta_size)
4134 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4136 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4138 if (unlikely(!skb || skb_size > skb->len))
4141 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4145 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4146 .func = bpf_skb_event_output,
4148 .ret_type = RET_INTEGER,
4149 .arg1_type = ARG_PTR_TO_CTX,
4150 .arg2_type = ARG_CONST_MAP_PTR,
4151 .arg3_type = ARG_ANYTHING,
4152 .arg4_type = ARG_PTR_TO_MEM,
4153 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4156 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4158 const struct bpf_func_proto bpf_skb_output_proto = {
4159 .func = bpf_skb_event_output,
4161 .ret_type = RET_INTEGER,
4162 .arg1_type = ARG_PTR_TO_BTF_ID,
4163 .arg1_btf_id = &bpf_skb_output_btf_ids[0],
4164 .arg2_type = ARG_CONST_MAP_PTR,
4165 .arg3_type = ARG_ANYTHING,
4166 .arg4_type = ARG_PTR_TO_MEM,
4167 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4170 static unsigned short bpf_tunnel_key_af(u64 flags)
4172 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4175 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4176 u32, size, u64, flags)
4178 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4179 u8 compat[sizeof(struct bpf_tunnel_key)];
4183 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
4187 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4191 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4194 case offsetof(struct bpf_tunnel_key, tunnel_label):
4195 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4197 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4198 /* Fixup deprecated structure layouts here, so we have
4199 * a common path later on.
4201 if (ip_tunnel_info_af(info) != AF_INET)
4204 to = (struct bpf_tunnel_key *)compat;
4211 to->tunnel_id = be64_to_cpu(info->key.tun_id);
4212 to->tunnel_tos = info->key.tos;
4213 to->tunnel_ttl = info->key.ttl;
4216 if (flags & BPF_F_TUNINFO_IPV6) {
4217 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4218 sizeof(to->remote_ipv6));
4219 to->tunnel_label = be32_to_cpu(info->key.label);
4221 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4222 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4223 to->tunnel_label = 0;
4226 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4227 memcpy(to_orig, to, size);
4231 memset(to_orig, 0, size);
4235 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4236 .func = bpf_skb_get_tunnel_key,
4238 .ret_type = RET_INTEGER,
4239 .arg1_type = ARG_PTR_TO_CTX,
4240 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4241 .arg3_type = ARG_CONST_SIZE,
4242 .arg4_type = ARG_ANYTHING,
4245 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4247 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4250 if (unlikely(!info ||
4251 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4255 if (unlikely(size < info->options_len)) {
4260 ip_tunnel_info_opts_get(to, info);
4261 if (size > info->options_len)
4262 memset(to + info->options_len, 0, size - info->options_len);
4264 return info->options_len;
4266 memset(to, 0, size);
4270 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4271 .func = bpf_skb_get_tunnel_opt,
4273 .ret_type = RET_INTEGER,
4274 .arg1_type = ARG_PTR_TO_CTX,
4275 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4276 .arg3_type = ARG_CONST_SIZE,
4279 static struct metadata_dst __percpu *md_dst;
4281 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4282 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4284 struct metadata_dst *md = this_cpu_ptr(md_dst);
4285 u8 compat[sizeof(struct bpf_tunnel_key)];
4286 struct ip_tunnel_info *info;
4288 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4289 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
4291 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4293 case offsetof(struct bpf_tunnel_key, tunnel_label):
4294 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4295 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4296 /* Fixup deprecated structure layouts here, so we have
4297 * a common path later on.
4299 memcpy(compat, from, size);
4300 memset(compat + size, 0, sizeof(compat) - size);
4301 from = (const struct bpf_tunnel_key *) compat;
4307 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4312 dst_hold((struct dst_entry *) md);
4313 skb_dst_set(skb, (struct dst_entry *) md);
4315 info = &md->u.tun_info;
4316 memset(info, 0, sizeof(*info));
4317 info->mode = IP_TUNNEL_INFO_TX;
4319 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4320 if (flags & BPF_F_DONT_FRAGMENT)
4321 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4322 if (flags & BPF_F_ZERO_CSUM_TX)
4323 info->key.tun_flags &= ~TUNNEL_CSUM;
4324 if (flags & BPF_F_SEQ_NUMBER)
4325 info->key.tun_flags |= TUNNEL_SEQ;
4327 info->key.tun_id = cpu_to_be64(from->tunnel_id);
4328 info->key.tos = from->tunnel_tos;
4329 info->key.ttl = from->tunnel_ttl;
4331 if (flags & BPF_F_TUNINFO_IPV6) {
4332 info->mode |= IP_TUNNEL_INFO_IPV6;
4333 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4334 sizeof(from->remote_ipv6));
4335 info->key.label = cpu_to_be32(from->tunnel_label) &
4336 IPV6_FLOWLABEL_MASK;
4338 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4344 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4345 .func = bpf_skb_set_tunnel_key,
4347 .ret_type = RET_INTEGER,
4348 .arg1_type = ARG_PTR_TO_CTX,
4349 .arg2_type = ARG_PTR_TO_MEM,
4350 .arg3_type = ARG_CONST_SIZE,
4351 .arg4_type = ARG_ANYTHING,
4354 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4355 const u8 *, from, u32, size)
4357 struct ip_tunnel_info *info = skb_tunnel_info(skb);
4358 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4360 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4362 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4365 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4370 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4371 .func = bpf_skb_set_tunnel_opt,
4373 .ret_type = RET_INTEGER,
4374 .arg1_type = ARG_PTR_TO_CTX,
4375 .arg2_type = ARG_PTR_TO_MEM,
4376 .arg3_type = ARG_CONST_SIZE,
4379 static const struct bpf_func_proto *
4380 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4383 struct metadata_dst __percpu *tmp;
4385 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4390 if (cmpxchg(&md_dst, NULL, tmp))
4391 metadata_dst_free_percpu(tmp);
4395 case BPF_FUNC_skb_set_tunnel_key:
4396 return &bpf_skb_set_tunnel_key_proto;
4397 case BPF_FUNC_skb_set_tunnel_opt:
4398 return &bpf_skb_set_tunnel_opt_proto;
4404 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4407 struct bpf_array *array = container_of(map, struct bpf_array, map);
4408 struct cgroup *cgrp;
4411 sk = skb_to_full_sk(skb);
4412 if (!sk || !sk_fullsock(sk))
4414 if (unlikely(idx >= array->map.max_entries))
4417 cgrp = READ_ONCE(array->ptrs[idx]);
4418 if (unlikely(!cgrp))
4421 return sk_under_cgroup_hierarchy(sk, cgrp);
4424 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4425 .func = bpf_skb_under_cgroup,
4427 .ret_type = RET_INTEGER,
4428 .arg1_type = ARG_PTR_TO_CTX,
4429 .arg2_type = ARG_CONST_MAP_PTR,
4430 .arg3_type = ARG_ANYTHING,
4433 #ifdef CONFIG_SOCK_CGROUP_DATA
4434 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4436 struct cgroup *cgrp;
4438 sk = sk_to_full_sk(sk);
4439 if (!sk || !sk_fullsock(sk))
4442 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4443 return cgroup_id(cgrp);
4446 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4448 return __bpf_sk_cgroup_id(skb->sk);
4451 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4452 .func = bpf_skb_cgroup_id,
4454 .ret_type = RET_INTEGER,
4455 .arg1_type = ARG_PTR_TO_CTX,
4458 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4461 struct cgroup *ancestor;
4462 struct cgroup *cgrp;
4464 sk = sk_to_full_sk(sk);
4465 if (!sk || !sk_fullsock(sk))
4468 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4469 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4473 return cgroup_id(ancestor);
4476 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4479 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4482 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4483 .func = bpf_skb_ancestor_cgroup_id,
4485 .ret_type = RET_INTEGER,
4486 .arg1_type = ARG_PTR_TO_CTX,
4487 .arg2_type = ARG_ANYTHING,
4490 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4492 return __bpf_sk_cgroup_id(sk);
4495 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4496 .func = bpf_sk_cgroup_id,
4498 .ret_type = RET_INTEGER,
4499 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4502 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4504 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4507 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4508 .func = bpf_sk_ancestor_cgroup_id,
4510 .ret_type = RET_INTEGER,
4511 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4512 .arg2_type = ARG_ANYTHING,
4516 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4517 unsigned long off, unsigned long len)
4519 memcpy(dst_buff, src_buff + off, len);
4523 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4524 u64, flags, void *, meta, u64, meta_size)
4526 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4528 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4530 if (unlikely(!xdp ||
4531 xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4534 return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4535 xdp_size, bpf_xdp_copy);
4538 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4539 .func = bpf_xdp_event_output,
4541 .ret_type = RET_INTEGER,
4542 .arg1_type = ARG_PTR_TO_CTX,
4543 .arg2_type = ARG_CONST_MAP_PTR,
4544 .arg3_type = ARG_ANYTHING,
4545 .arg4_type = ARG_PTR_TO_MEM,
4546 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4549 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4551 const struct bpf_func_proto bpf_xdp_output_proto = {
4552 .func = bpf_xdp_event_output,
4554 .ret_type = RET_INTEGER,
4555 .arg1_type = ARG_PTR_TO_BTF_ID,
4556 .arg1_btf_id = &bpf_xdp_output_btf_ids[0],
4557 .arg2_type = ARG_CONST_MAP_PTR,
4558 .arg3_type = ARG_ANYTHING,
4559 .arg4_type = ARG_PTR_TO_MEM,
4560 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4563 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4565 return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4568 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4569 .func = bpf_get_socket_cookie,
4571 .ret_type = RET_INTEGER,
4572 .arg1_type = ARG_PTR_TO_CTX,
4575 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4577 return __sock_gen_cookie(ctx->sk);
4580 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4581 .func = bpf_get_socket_cookie_sock_addr,
4583 .ret_type = RET_INTEGER,
4584 .arg1_type = ARG_PTR_TO_CTX,
4587 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4589 return __sock_gen_cookie(ctx);
4592 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4593 .func = bpf_get_socket_cookie_sock,
4595 .ret_type = RET_INTEGER,
4596 .arg1_type = ARG_PTR_TO_CTX,
4599 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
4601 return sk ? sock_gen_cookie(sk) : 0;
4604 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
4605 .func = bpf_get_socket_ptr_cookie,
4607 .ret_type = RET_INTEGER,
4608 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4611 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4613 return __sock_gen_cookie(ctx->sk);
4616 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4617 .func = bpf_get_socket_cookie_sock_ops,
4619 .ret_type = RET_INTEGER,
4620 .arg1_type = ARG_PTR_TO_CTX,
4623 static u64 __bpf_get_netns_cookie(struct sock *sk)
4625 const struct net *net = sk ? sock_net(sk) : &init_net;
4627 return net->net_cookie;
4630 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
4632 return __bpf_get_netns_cookie(ctx);
4635 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
4636 .func = bpf_get_netns_cookie_sock,
4638 .ret_type = RET_INTEGER,
4639 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4642 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4644 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4647 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
4648 .func = bpf_get_netns_cookie_sock_addr,
4650 .ret_type = RET_INTEGER,
4651 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4654 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4656 struct sock *sk = sk_to_full_sk(skb->sk);
4659 if (!sk || !sk_fullsock(sk))
4661 kuid = sock_net_uid(sock_net(sk), sk);
4662 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4665 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4666 .func = bpf_get_socket_uid,
4668 .ret_type = RET_INTEGER,
4669 .arg1_type = ARG_PTR_TO_CTX,
4672 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
4673 char *optval, int optlen)
4675 char devname[IFNAMSIZ];
4681 if (!sk_fullsock(sk))
4684 sock_owned_by_me(sk);
4686 if (level == SOL_SOCKET) {
4687 if (optlen != sizeof(int) && optname != SO_BINDTODEVICE)
4689 val = *((int *)optval);
4690 valbool = val ? 1 : 0;
4692 /* Only some socketops are supported */
4695 val = min_t(u32, val, sysctl_rmem_max);
4696 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4697 WRITE_ONCE(sk->sk_rcvbuf,
4698 max_t(int, val * 2, SOCK_MIN_RCVBUF));
4701 val = min_t(u32, val, sysctl_wmem_max);
4702 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4703 WRITE_ONCE(sk->sk_sndbuf,
4704 max_t(int, val * 2, SOCK_MIN_SNDBUF));
4706 case SO_MAX_PACING_RATE: /* 32bit version */
4708 cmpxchg(&sk->sk_pacing_status,
4711 sk->sk_max_pacing_rate = (val == ~0U) ?
4712 ~0UL : (unsigned int)val;
4713 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4714 sk->sk_max_pacing_rate);
4717 sk->sk_priority = val;
4722 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
4725 if (sk->sk_mark != val) {
4730 case SO_BINDTODEVICE:
4731 optlen = min_t(long, optlen, IFNAMSIZ - 1);
4732 strncpy(devname, optval, optlen);
4733 devname[optlen] = 0;
4736 if (devname[0] != '\0') {
4737 struct net_device *dev;
4742 dev = dev_get_by_name(net, devname);
4745 ifindex = dev->ifindex;
4749 case SO_BINDTOIFINDEX:
4750 if (optname == SO_BINDTOIFINDEX)
4752 ret = sock_bindtoindex(sk, ifindex, false);
4755 if (sk->sk_prot->keepalive)
4756 sk->sk_prot->keepalive(sk, valbool);
4757 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
4760 sk->sk_reuseport = valbool;
4766 } else if (level == SOL_IP) {
4767 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4770 val = *((int *)optval);
4771 /* Only some options are supported */
4774 if (val < -1 || val > 0xff) {
4777 struct inet_sock *inet = inet_sk(sk);
4787 #if IS_ENABLED(CONFIG_IPV6)
4788 } else if (level == SOL_IPV6) {
4789 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4792 val = *((int *)optval);
4793 /* Only some options are supported */
4796 if (val < -1 || val > 0xff) {
4799 struct ipv6_pinfo *np = inet6_sk(sk);
4810 } else if (level == SOL_TCP &&
4811 sk->sk_prot->setsockopt == tcp_setsockopt) {
4812 if (optname == TCP_CONGESTION) {
4813 char name[TCP_CA_NAME_MAX];
4815 strncpy(name, optval, min_t(long, optlen,
4816 TCP_CA_NAME_MAX-1));
4817 name[TCP_CA_NAME_MAX-1] = 0;
4818 ret = tcp_set_congestion_control(sk, name, false, true);
4820 struct inet_connection_sock *icsk = inet_csk(sk);
4821 struct tcp_sock *tp = tcp_sk(sk);
4822 unsigned long timeout;
4824 if (optlen != sizeof(int))
4827 val = *((int *)optval);
4828 /* Only some options are supported */
4831 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4836 case TCP_BPF_SNDCWND_CLAMP:
4840 tp->snd_cwnd_clamp = val;
4841 tp->snd_ssthresh = val;
4844 case TCP_BPF_DELACK_MAX:
4845 timeout = usecs_to_jiffies(val);
4846 if (timeout > TCP_DELACK_MAX ||
4847 timeout < TCP_TIMEOUT_MIN)
4849 inet_csk(sk)->icsk_delack_max = timeout;
4851 case TCP_BPF_RTO_MIN:
4852 timeout = usecs_to_jiffies(val);
4853 if (timeout > TCP_RTO_MIN ||
4854 timeout < TCP_TIMEOUT_MIN)
4856 inet_csk(sk)->icsk_rto_min = timeout;
4859 if (val < 0 || val > 1)
4865 ret = tcp_sock_set_keepidle_locked(sk, val);
4868 if (val < 1 || val > MAX_TCP_KEEPINTVL)
4871 tp->keepalive_intvl = val * HZ;
4874 if (val < 1 || val > MAX_TCP_KEEPCNT)
4877 tp->keepalive_probes = val;
4880 if (val < 1 || val > MAX_TCP_SYNCNT)
4883 icsk->icsk_syn_retries = val;
4885 case TCP_USER_TIMEOUT:
4889 icsk->icsk_user_timeout = val;
4891 case TCP_NOTSENT_LOWAT:
4892 tp->notsent_lowat = val;
4893 sk->sk_write_space(sk);
4895 case TCP_WINDOW_CLAMP:
4896 ret = tcp_set_window_clamp(sk, val);
4909 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
4910 char *optval, int optlen)
4912 if (!sk_fullsock(sk))
4915 sock_owned_by_me(sk);
4917 if (level == SOL_SOCKET) {
4918 if (optlen != sizeof(int))
4923 *((int *)optval) = sk->sk_mark;
4926 *((int *)optval) = sk->sk_priority;
4928 case SO_BINDTOIFINDEX:
4929 *((int *)optval) = sk->sk_bound_dev_if;
4932 *((int *)optval) = sk->sk_reuseport;
4938 } else if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4939 struct inet_connection_sock *icsk;
4940 struct tcp_sock *tp;
4943 case TCP_CONGESTION:
4944 icsk = inet_csk(sk);
4946 if (!icsk->icsk_ca_ops || optlen <= 1)
4948 strncpy(optval, icsk->icsk_ca_ops->name, optlen);
4949 optval[optlen - 1] = 0;
4954 if (optlen <= 0 || !tp->saved_syn ||
4955 optlen > tcp_saved_syn_len(tp->saved_syn))
4957 memcpy(optval, tp->saved_syn->data, optlen);
4962 } else if (level == SOL_IP) {
4963 struct inet_sock *inet = inet_sk(sk);
4965 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4968 /* Only some options are supported */
4971 *((int *)optval) = (int)inet->tos;
4976 #if IS_ENABLED(CONFIG_IPV6)
4977 } else if (level == SOL_IPV6) {
4978 struct ipv6_pinfo *np = inet6_sk(sk);
4980 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4983 /* Only some options are supported */
4986 *((int *)optval) = (int)np->tclass;
4998 memset(optval, 0, optlen);
5002 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level,
5003 int, optname, char *, optval, int, optlen)
5005 return _bpf_setsockopt(sk, level, optname, optval, optlen);
5008 const struct bpf_func_proto bpf_sk_setsockopt_proto = {
5009 .func = bpf_sk_setsockopt,
5011 .ret_type = RET_INTEGER,
5012 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5013 .arg2_type = ARG_ANYTHING,
5014 .arg3_type = ARG_ANYTHING,
5015 .arg4_type = ARG_PTR_TO_MEM,
5016 .arg5_type = ARG_CONST_SIZE,
5019 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level,
5020 int, optname, char *, optval, int, optlen)
5022 return _bpf_getsockopt(sk, level, optname, optval, optlen);
5025 const struct bpf_func_proto bpf_sk_getsockopt_proto = {
5026 .func = bpf_sk_getsockopt,
5028 .ret_type = RET_INTEGER,
5029 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5030 .arg2_type = ARG_ANYTHING,
5031 .arg3_type = ARG_ANYTHING,
5032 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5033 .arg5_type = ARG_CONST_SIZE,
5036 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5037 int, level, int, optname, char *, optval, int, optlen)
5039 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5042 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5043 .func = bpf_sock_addr_setsockopt,
5045 .ret_type = RET_INTEGER,
5046 .arg1_type = ARG_PTR_TO_CTX,
5047 .arg2_type = ARG_ANYTHING,
5048 .arg3_type = ARG_ANYTHING,
5049 .arg4_type = ARG_PTR_TO_MEM,
5050 .arg5_type = ARG_CONST_SIZE,
5053 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5054 int, level, int, optname, char *, optval, int, optlen)
5056 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5059 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5060 .func = bpf_sock_addr_getsockopt,
5062 .ret_type = RET_INTEGER,
5063 .arg1_type = ARG_PTR_TO_CTX,
5064 .arg2_type = ARG_ANYTHING,
5065 .arg3_type = ARG_ANYTHING,
5066 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5067 .arg5_type = ARG_CONST_SIZE,
5070 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5071 int, level, int, optname, char *, optval, int, optlen)
5073 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5076 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5077 .func = bpf_sock_ops_setsockopt,
5079 .ret_type = RET_INTEGER,
5080 .arg1_type = ARG_PTR_TO_CTX,
5081 .arg2_type = ARG_ANYTHING,
5082 .arg3_type = ARG_ANYTHING,
5083 .arg4_type = ARG_PTR_TO_MEM,
5084 .arg5_type = ARG_CONST_SIZE,
5087 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5088 int optname, const u8 **start)
5090 struct sk_buff *syn_skb = bpf_sock->syn_skb;
5091 const u8 *hdr_start;
5095 /* sk is a request_sock here */
5097 if (optname == TCP_BPF_SYN) {
5098 hdr_start = syn_skb->data;
5099 ret = tcp_hdrlen(syn_skb);
5100 } else if (optname == TCP_BPF_SYN_IP) {
5101 hdr_start = skb_network_header(syn_skb);
5102 ret = skb_network_header_len(syn_skb) +
5103 tcp_hdrlen(syn_skb);
5105 /* optname == TCP_BPF_SYN_MAC */
5106 hdr_start = skb_mac_header(syn_skb);
5107 ret = skb_mac_header_len(syn_skb) +
5108 skb_network_header_len(syn_skb) +
5109 tcp_hdrlen(syn_skb);
5112 struct sock *sk = bpf_sock->sk;
5113 struct saved_syn *saved_syn;
5115 if (sk->sk_state == TCP_NEW_SYN_RECV)
5116 /* synack retransmit. bpf_sock->syn_skb will
5117 * not be available. It has to resort to
5118 * saved_syn (if it is saved).
5120 saved_syn = inet_reqsk(sk)->saved_syn;
5122 saved_syn = tcp_sk(sk)->saved_syn;
5127 if (optname == TCP_BPF_SYN) {
5128 hdr_start = saved_syn->data +
5129 saved_syn->mac_hdrlen +
5130 saved_syn->network_hdrlen;
5131 ret = saved_syn->tcp_hdrlen;
5132 } else if (optname == TCP_BPF_SYN_IP) {
5133 hdr_start = saved_syn->data +
5134 saved_syn->mac_hdrlen;
5135 ret = saved_syn->network_hdrlen +
5136 saved_syn->tcp_hdrlen;
5138 /* optname == TCP_BPF_SYN_MAC */
5140 /* TCP_SAVE_SYN may not have saved the mac hdr */
5141 if (!saved_syn->mac_hdrlen)
5144 hdr_start = saved_syn->data;
5145 ret = saved_syn->mac_hdrlen +
5146 saved_syn->network_hdrlen +
5147 saved_syn->tcp_hdrlen;
5155 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5156 int, level, int, optname, char *, optval, int, optlen)
5158 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5159 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5160 int ret, copy_len = 0;
5163 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5166 if (optlen < copy_len) {
5171 memcpy(optval, start, copy_len);
5174 /* Zero out unused buffer at the end */
5175 memset(optval + copy_len, 0, optlen - copy_len);
5180 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5183 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5184 .func = bpf_sock_ops_getsockopt,
5186 .ret_type = RET_INTEGER,
5187 .arg1_type = ARG_PTR_TO_CTX,
5188 .arg2_type = ARG_ANYTHING,
5189 .arg3_type = ARG_ANYTHING,
5190 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5191 .arg5_type = ARG_CONST_SIZE,
5194 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5197 struct sock *sk = bpf_sock->sk;
5198 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5200 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5203 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5205 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5208 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5209 .func = bpf_sock_ops_cb_flags_set,
5211 .ret_type = RET_INTEGER,
5212 .arg1_type = ARG_PTR_TO_CTX,
5213 .arg2_type = ARG_ANYTHING,
5216 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5217 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5219 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5223 struct sock *sk = ctx->sk;
5224 u32 flags = BIND_FROM_BPF;
5228 if (addr_len < offsetofend(struct sockaddr, sa_family))
5230 if (addr->sa_family == AF_INET) {
5231 if (addr_len < sizeof(struct sockaddr_in))
5233 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5234 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5235 return __inet_bind(sk, addr, addr_len, flags);
5236 #if IS_ENABLED(CONFIG_IPV6)
5237 } else if (addr->sa_family == AF_INET6) {
5238 if (addr_len < SIN6_LEN_RFC2133)
5240 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5241 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5242 /* ipv6_bpf_stub cannot be NULL, since it's called from
5243 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5245 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5246 #endif /* CONFIG_IPV6 */
5248 #endif /* CONFIG_INET */
5250 return -EAFNOSUPPORT;
5253 static const struct bpf_func_proto bpf_bind_proto = {
5256 .ret_type = RET_INTEGER,
5257 .arg1_type = ARG_PTR_TO_CTX,
5258 .arg2_type = ARG_PTR_TO_MEM,
5259 .arg3_type = ARG_CONST_SIZE,
5263 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5264 struct bpf_xfrm_state *, to, u32, size, u64, flags)
5266 const struct sec_path *sp = skb_sec_path(skb);
5267 const struct xfrm_state *x;
5269 if (!sp || unlikely(index >= sp->len || flags))
5272 x = sp->xvec[index];
5274 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5277 to->reqid = x->props.reqid;
5278 to->spi = x->id.spi;
5279 to->family = x->props.family;
5282 if (to->family == AF_INET6) {
5283 memcpy(to->remote_ipv6, x->props.saddr.a6,
5284 sizeof(to->remote_ipv6));
5286 to->remote_ipv4 = x->props.saddr.a4;
5287 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5292 memset(to, 0, size);
5296 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5297 .func = bpf_skb_get_xfrm_state,
5299 .ret_type = RET_INTEGER,
5300 .arg1_type = ARG_PTR_TO_CTX,
5301 .arg2_type = ARG_ANYTHING,
5302 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
5303 .arg4_type = ARG_CONST_SIZE,
5304 .arg5_type = ARG_ANYTHING,
5308 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5309 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
5310 const struct neighbour *neigh,
5311 const struct net_device *dev, u32 mtu)
5313 memcpy(params->dmac, neigh->ha, ETH_ALEN);
5314 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5315 params->h_vlan_TCI = 0;
5316 params->h_vlan_proto = 0;
5318 params->mtu_result = mtu; /* union with tot_len */
5324 #if IS_ENABLED(CONFIG_INET)
5325 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5326 u32 flags, bool check_mtu)
5328 struct fib_nh_common *nhc;
5329 struct in_device *in_dev;
5330 struct neighbour *neigh;
5331 struct net_device *dev;
5332 struct fib_result res;
5337 dev = dev_get_by_index_rcu(net, params->ifindex);
5341 /* verify forwarding is enabled on this interface */
5342 in_dev = __in_dev_get_rcu(dev);
5343 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5344 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5346 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5348 fl4.flowi4_oif = params->ifindex;
5350 fl4.flowi4_iif = params->ifindex;
5353 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5354 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5355 fl4.flowi4_flags = 0;
5357 fl4.flowi4_proto = params->l4_protocol;
5358 fl4.daddr = params->ipv4_dst;
5359 fl4.saddr = params->ipv4_src;
5360 fl4.fl4_sport = params->sport;
5361 fl4.fl4_dport = params->dport;
5362 fl4.flowi4_multipath_hash = 0;
5364 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5365 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5366 struct fib_table *tb;
5368 tb = fib_get_table(net, tbid);
5370 return BPF_FIB_LKUP_RET_NOT_FWDED;
5372 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5374 fl4.flowi4_mark = 0;
5375 fl4.flowi4_secid = 0;
5376 fl4.flowi4_tun_key.tun_id = 0;
5377 fl4.flowi4_uid = sock_net_uid(net, NULL);
5379 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5383 /* map fib lookup errors to RTN_ type */
5385 return BPF_FIB_LKUP_RET_BLACKHOLE;
5386 if (err == -EHOSTUNREACH)
5387 return BPF_FIB_LKUP_RET_UNREACHABLE;
5389 return BPF_FIB_LKUP_RET_PROHIBIT;
5391 return BPF_FIB_LKUP_RET_NOT_FWDED;
5394 if (res.type != RTN_UNICAST)
5395 return BPF_FIB_LKUP_RET_NOT_FWDED;
5397 if (fib_info_num_path(res.fi) > 1)
5398 fib_select_path(net, &res, &fl4, NULL);
5401 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5402 if (params->tot_len > mtu) {
5403 params->mtu_result = mtu; /* union with tot_len */
5404 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5410 /* do not handle lwt encaps right now */
5411 if (nhc->nhc_lwtstate)
5412 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5416 params->rt_metric = res.fi->fib_priority;
5417 params->ifindex = dev->ifindex;
5419 /* xdp and cls_bpf programs are run in RCU-bh so
5420 * rcu_read_lock_bh is not needed here
5422 if (likely(nhc->nhc_gw_family != AF_INET6)) {
5423 if (nhc->nhc_gw_family)
5424 params->ipv4_dst = nhc->nhc_gw.ipv4;
5426 neigh = __ipv4_neigh_lookup_noref(dev,
5427 (__force u32)params->ipv4_dst);
5429 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5431 params->family = AF_INET6;
5432 *dst = nhc->nhc_gw.ipv6;
5433 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5437 return BPF_FIB_LKUP_RET_NO_NEIGH;
5439 return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5443 #if IS_ENABLED(CONFIG_IPV6)
5444 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5445 u32 flags, bool check_mtu)
5447 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5448 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5449 struct fib6_result res = {};
5450 struct neighbour *neigh;
5451 struct net_device *dev;
5452 struct inet6_dev *idev;
5458 /* link local addresses are never forwarded */
5459 if (rt6_need_strict(dst) || rt6_need_strict(src))
5460 return BPF_FIB_LKUP_RET_NOT_FWDED;
5462 dev = dev_get_by_index_rcu(net, params->ifindex);
5466 idev = __in6_dev_get_safely(dev);
5467 if (unlikely(!idev || !idev->cnf.forwarding))
5468 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5470 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5472 oif = fl6.flowi6_oif = params->ifindex;
5474 oif = fl6.flowi6_iif = params->ifindex;
5476 strict = RT6_LOOKUP_F_HAS_SADDR;
5478 fl6.flowlabel = params->flowinfo;
5479 fl6.flowi6_scope = 0;
5480 fl6.flowi6_flags = 0;
5483 fl6.flowi6_proto = params->l4_protocol;
5486 fl6.fl6_sport = params->sport;
5487 fl6.fl6_dport = params->dport;
5489 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5490 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5491 struct fib6_table *tb;
5493 tb = ipv6_stub->fib6_get_table(net, tbid);
5495 return BPF_FIB_LKUP_RET_NOT_FWDED;
5497 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5500 fl6.flowi6_mark = 0;
5501 fl6.flowi6_secid = 0;
5502 fl6.flowi6_tun_key.tun_id = 0;
5503 fl6.flowi6_uid = sock_net_uid(net, NULL);
5505 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5508 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5509 res.f6i == net->ipv6.fib6_null_entry))
5510 return BPF_FIB_LKUP_RET_NOT_FWDED;
5512 switch (res.fib6_type) {
5513 /* only unicast is forwarded */
5517 return BPF_FIB_LKUP_RET_BLACKHOLE;
5518 case RTN_UNREACHABLE:
5519 return BPF_FIB_LKUP_RET_UNREACHABLE;
5521 return BPF_FIB_LKUP_RET_PROHIBIT;
5523 return BPF_FIB_LKUP_RET_NOT_FWDED;
5526 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5527 fl6.flowi6_oif != 0, NULL, strict);
5530 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5531 if (params->tot_len > mtu) {
5532 params->mtu_result = mtu; /* union with tot_len */
5533 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5537 if (res.nh->fib_nh_lws)
5538 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5540 if (res.nh->fib_nh_gw_family)
5541 *dst = res.nh->fib_nh_gw6;
5543 dev = res.nh->fib_nh_dev;
5544 params->rt_metric = res.f6i->fib6_metric;
5545 params->ifindex = dev->ifindex;
5547 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
5550 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5552 return BPF_FIB_LKUP_RET_NO_NEIGH;
5554 return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5558 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
5559 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5561 if (plen < sizeof(*params))
5564 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5567 switch (params->family) {
5568 #if IS_ENABLED(CONFIG_INET)
5570 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
5573 #if IS_ENABLED(CONFIG_IPV6)
5575 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
5579 return -EAFNOSUPPORT;
5582 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
5583 .func = bpf_xdp_fib_lookup,
5585 .ret_type = RET_INTEGER,
5586 .arg1_type = ARG_PTR_TO_CTX,
5587 .arg2_type = ARG_PTR_TO_MEM,
5588 .arg3_type = ARG_CONST_SIZE,
5589 .arg4_type = ARG_ANYTHING,
5592 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
5593 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5595 struct net *net = dev_net(skb->dev);
5596 int rc = -EAFNOSUPPORT;
5597 bool check_mtu = false;
5599 if (plen < sizeof(*params))
5602 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5605 if (params->tot_len)
5608 switch (params->family) {
5609 #if IS_ENABLED(CONFIG_INET)
5611 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
5614 #if IS_ENABLED(CONFIG_IPV6)
5616 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
5621 if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
5622 struct net_device *dev;
5624 /* When tot_len isn't provided by user, check skb
5625 * against MTU of FIB lookup resulting net_device
5627 dev = dev_get_by_index_rcu(net, params->ifindex);
5628 if (!is_skb_forwardable(dev, skb))
5629 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
5631 params->mtu_result = dev->mtu; /* union with tot_len */
5637 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
5638 .func = bpf_skb_fib_lookup,
5640 .ret_type = RET_INTEGER,
5641 .arg1_type = ARG_PTR_TO_CTX,
5642 .arg2_type = ARG_PTR_TO_MEM,
5643 .arg3_type = ARG_CONST_SIZE,
5644 .arg4_type = ARG_ANYTHING,
5647 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
5650 struct net *netns = dev_net(dev_curr);
5652 /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
5656 return dev_get_by_index_rcu(netns, ifindex);
5659 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
5660 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5662 int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5663 struct net_device *dev = skb->dev;
5664 int skb_len, dev_len;
5667 if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
5670 if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
5673 dev = __dev_via_ifindex(dev, ifindex);
5677 mtu = READ_ONCE(dev->mtu);
5679 dev_len = mtu + dev->hard_header_len;
5681 /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
5682 skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
5684 skb_len += len_diff; /* minus result pass check */
5685 if (skb_len <= dev_len) {
5686 ret = BPF_MTU_CHK_RET_SUCCESS;
5689 /* At this point, skb->len exceed MTU, but as it include length of all
5690 * segments, it can still be below MTU. The SKB can possibly get
5691 * re-segmented in transmit path (see validate_xmit_skb). Thus, user
5692 * must choose if segs are to be MTU checked.
5694 if (skb_is_gso(skb)) {
5695 ret = BPF_MTU_CHK_RET_SUCCESS;
5697 if (flags & BPF_MTU_CHK_SEGS &&
5698 !skb_gso_validate_network_len(skb, mtu))
5699 ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
5702 /* BPF verifier guarantees valid pointer */
5708 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
5709 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5711 struct net_device *dev = xdp->rxq->dev;
5712 int xdp_len = xdp->data_end - xdp->data;
5713 int ret = BPF_MTU_CHK_RET_SUCCESS;
5716 /* XDP variant doesn't support multi-buffer segment check (yet) */
5717 if (unlikely(flags))
5720 dev = __dev_via_ifindex(dev, ifindex);
5724 mtu = READ_ONCE(dev->mtu);
5726 /* Add L2-header as dev MTU is L3 size */
5727 dev_len = mtu + dev->hard_header_len;
5729 /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
5731 xdp_len = *mtu_len + dev->hard_header_len;
5733 xdp_len += len_diff; /* minus result pass check */
5734 if (xdp_len > dev_len)
5735 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5737 /* BPF verifier guarantees valid pointer */
5743 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
5744 .func = bpf_skb_check_mtu,
5746 .ret_type = RET_INTEGER,
5747 .arg1_type = ARG_PTR_TO_CTX,
5748 .arg2_type = ARG_ANYTHING,
5749 .arg3_type = ARG_PTR_TO_INT,
5750 .arg4_type = ARG_ANYTHING,
5751 .arg5_type = ARG_ANYTHING,
5754 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
5755 .func = bpf_xdp_check_mtu,
5757 .ret_type = RET_INTEGER,
5758 .arg1_type = ARG_PTR_TO_CTX,
5759 .arg2_type = ARG_ANYTHING,
5760 .arg3_type = ARG_PTR_TO_INT,
5761 .arg4_type = ARG_ANYTHING,
5762 .arg5_type = ARG_ANYTHING,
5765 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5766 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
5769 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
5771 if (!seg6_validate_srh(srh, len, false))
5775 case BPF_LWT_ENCAP_SEG6_INLINE:
5776 if (skb->protocol != htons(ETH_P_IPV6))
5779 err = seg6_do_srh_inline(skb, srh);
5781 case BPF_LWT_ENCAP_SEG6:
5782 skb_reset_inner_headers(skb);
5783 skb->encapsulation = 1;
5784 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
5790 bpf_compute_data_pointers(skb);
5794 ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5795 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
5797 return seg6_lookup_nexthop(skb, NULL, 0);
5799 #endif /* CONFIG_IPV6_SEG6_BPF */
5801 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5802 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
5805 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
5809 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
5813 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5814 case BPF_LWT_ENCAP_SEG6:
5815 case BPF_LWT_ENCAP_SEG6_INLINE:
5816 return bpf_push_seg6_encap(skb, type, hdr, len);
5818 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5819 case BPF_LWT_ENCAP_IP:
5820 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
5827 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
5828 void *, hdr, u32, len)
5831 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5832 case BPF_LWT_ENCAP_IP:
5833 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
5840 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
5841 .func = bpf_lwt_in_push_encap,
5843 .ret_type = RET_INTEGER,
5844 .arg1_type = ARG_PTR_TO_CTX,
5845 .arg2_type = ARG_ANYTHING,
5846 .arg3_type = ARG_PTR_TO_MEM,
5847 .arg4_type = ARG_CONST_SIZE
5850 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
5851 .func = bpf_lwt_xmit_push_encap,
5853 .ret_type = RET_INTEGER,
5854 .arg1_type = ARG_PTR_TO_CTX,
5855 .arg2_type = ARG_ANYTHING,
5856 .arg3_type = ARG_PTR_TO_MEM,
5857 .arg4_type = ARG_CONST_SIZE
5860 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5861 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
5862 const void *, from, u32, len)
5864 struct seg6_bpf_srh_state *srh_state =
5865 this_cpu_ptr(&seg6_bpf_srh_states);
5866 struct ipv6_sr_hdr *srh = srh_state->srh;
5867 void *srh_tlvs, *srh_end, *ptr;
5873 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
5874 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
5876 ptr = skb->data + offset;
5877 if (ptr >= srh_tlvs && ptr + len <= srh_end)
5878 srh_state->valid = false;
5879 else if (ptr < (void *)&srh->flags ||
5880 ptr + len > (void *)&srh->segments)
5883 if (unlikely(bpf_try_make_writable(skb, offset + len)))
5885 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5887 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5889 memcpy(skb->data + offset, from, len);
5893 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
5894 .func = bpf_lwt_seg6_store_bytes,
5896 .ret_type = RET_INTEGER,
5897 .arg1_type = ARG_PTR_TO_CTX,
5898 .arg2_type = ARG_ANYTHING,
5899 .arg3_type = ARG_PTR_TO_MEM,
5900 .arg4_type = ARG_CONST_SIZE
5903 static void bpf_update_srh_state(struct sk_buff *skb)
5905 struct seg6_bpf_srh_state *srh_state =
5906 this_cpu_ptr(&seg6_bpf_srh_states);
5909 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
5910 srh_state->srh = NULL;
5912 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5913 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
5914 srh_state->valid = true;
5918 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
5919 u32, action, void *, param, u32, param_len)
5921 struct seg6_bpf_srh_state *srh_state =
5922 this_cpu_ptr(&seg6_bpf_srh_states);
5927 case SEG6_LOCAL_ACTION_END_X:
5928 if (!seg6_bpf_has_valid_srh(skb))
5930 if (param_len != sizeof(struct in6_addr))
5932 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
5933 case SEG6_LOCAL_ACTION_END_T:
5934 if (!seg6_bpf_has_valid_srh(skb))
5936 if (param_len != sizeof(int))
5938 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5939 case SEG6_LOCAL_ACTION_END_DT6:
5940 if (!seg6_bpf_has_valid_srh(skb))
5942 if (param_len != sizeof(int))
5945 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
5947 if (!pskb_pull(skb, hdroff))
5950 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
5951 skb_reset_network_header(skb);
5952 skb_reset_transport_header(skb);
5953 skb->encapsulation = 0;
5955 bpf_compute_data_pointers(skb);
5956 bpf_update_srh_state(skb);
5957 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5958 case SEG6_LOCAL_ACTION_END_B6:
5959 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5961 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
5964 bpf_update_srh_state(skb);
5967 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
5968 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5970 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
5973 bpf_update_srh_state(skb);
5981 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
5982 .func = bpf_lwt_seg6_action,
5984 .ret_type = RET_INTEGER,
5985 .arg1_type = ARG_PTR_TO_CTX,
5986 .arg2_type = ARG_ANYTHING,
5987 .arg3_type = ARG_PTR_TO_MEM,
5988 .arg4_type = ARG_CONST_SIZE
5991 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
5994 struct seg6_bpf_srh_state *srh_state =
5995 this_cpu_ptr(&seg6_bpf_srh_states);
5996 struct ipv6_sr_hdr *srh = srh_state->srh;
5997 void *srh_end, *srh_tlvs, *ptr;
5998 struct ipv6hdr *hdr;
6002 if (unlikely(srh == NULL))
6005 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
6006 ((srh->first_segment + 1) << 4));
6007 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
6009 ptr = skb->data + offset;
6011 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
6013 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
6017 ret = skb_cow_head(skb, len);
6018 if (unlikely(ret < 0))
6021 ret = bpf_skb_net_hdr_push(skb, offset, len);
6023 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6026 bpf_compute_data_pointers(skb);
6027 if (unlikely(ret < 0))
6030 hdr = (struct ipv6hdr *)skb->data;
6031 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6033 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6035 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6036 srh_state->hdrlen += len;
6037 srh_state->valid = false;
6041 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6042 .func = bpf_lwt_seg6_adjust_srh,
6044 .ret_type = RET_INTEGER,
6045 .arg1_type = ARG_PTR_TO_CTX,
6046 .arg2_type = ARG_ANYTHING,
6047 .arg3_type = ARG_ANYTHING,
6049 #endif /* CONFIG_IPV6_SEG6_BPF */
6052 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6053 int dif, int sdif, u8 family, u8 proto)
6055 bool refcounted = false;
6056 struct sock *sk = NULL;
6058 if (family == AF_INET) {
6059 __be32 src4 = tuple->ipv4.saddr;
6060 __be32 dst4 = tuple->ipv4.daddr;
6062 if (proto == IPPROTO_TCP)
6063 sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
6064 src4, tuple->ipv4.sport,
6065 dst4, tuple->ipv4.dport,
6066 dif, sdif, &refcounted);
6068 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6069 dst4, tuple->ipv4.dport,
6070 dif, sdif, &udp_table, NULL);
6071 #if IS_ENABLED(CONFIG_IPV6)
6073 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6074 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6076 if (proto == IPPROTO_TCP)
6077 sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
6078 src6, tuple->ipv6.sport,
6079 dst6, ntohs(tuple->ipv6.dport),
6080 dif, sdif, &refcounted);
6081 else if (likely(ipv6_bpf_stub))
6082 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6083 src6, tuple->ipv6.sport,
6084 dst6, tuple->ipv6.dport,
6090 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6091 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6097 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6098 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6099 * Returns the socket as an 'unsigned long' to simplify the casting in the
6100 * callers to satisfy BPF_CALL declarations.
6102 static struct sock *
6103 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6104 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6107 struct sock *sk = NULL;
6108 u8 family = AF_UNSPEC;
6112 if (len == sizeof(tuple->ipv4))
6114 else if (len == sizeof(tuple->ipv6))
6119 if (unlikely(family == AF_UNSPEC || flags ||
6120 !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6123 if (family == AF_INET)
6124 sdif = inet_sdif(skb);
6126 sdif = inet6_sdif(skb);
6128 if ((s32)netns_id < 0) {
6130 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6132 net = get_net_ns_by_id(caller_net, netns_id);
6135 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6143 static struct sock *
6144 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6145 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6148 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6149 ifindex, proto, netns_id, flags);
6152 sk = sk_to_full_sk(sk);
6153 if (!sk_fullsock(sk)) {
6162 static struct sock *
6163 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6164 u8 proto, u64 netns_id, u64 flags)
6166 struct net *caller_net;
6170 caller_net = dev_net(skb->dev);
6171 ifindex = skb->dev->ifindex;
6173 caller_net = sock_net(skb->sk);
6177 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6181 static struct sock *
6182 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6183 u8 proto, u64 netns_id, u64 flags)
6185 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6189 sk = sk_to_full_sk(sk);
6190 if (!sk_fullsock(sk)) {
6199 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6200 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6202 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6206 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6207 .func = bpf_skc_lookup_tcp,
6210 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6211 .arg1_type = ARG_PTR_TO_CTX,
6212 .arg2_type = ARG_PTR_TO_MEM,
6213 .arg3_type = ARG_CONST_SIZE,
6214 .arg4_type = ARG_ANYTHING,
6215 .arg5_type = ARG_ANYTHING,
6218 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6219 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6221 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6225 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6226 .func = bpf_sk_lookup_tcp,
6229 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6230 .arg1_type = ARG_PTR_TO_CTX,
6231 .arg2_type = ARG_PTR_TO_MEM,
6232 .arg3_type = ARG_CONST_SIZE,
6233 .arg4_type = ARG_ANYTHING,
6234 .arg5_type = ARG_ANYTHING,
6237 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6238 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6240 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6244 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6245 .func = bpf_sk_lookup_udp,
6248 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6249 .arg1_type = ARG_PTR_TO_CTX,
6250 .arg2_type = ARG_PTR_TO_MEM,
6251 .arg3_type = ARG_CONST_SIZE,
6252 .arg4_type = ARG_ANYTHING,
6253 .arg5_type = ARG_ANYTHING,
6256 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6258 if (sk && sk_is_refcounted(sk))
6263 static const struct bpf_func_proto bpf_sk_release_proto = {
6264 .func = bpf_sk_release,
6266 .ret_type = RET_INTEGER,
6267 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6270 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6271 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6273 struct net *caller_net = dev_net(ctx->rxq->dev);
6274 int ifindex = ctx->rxq->dev->ifindex;
6276 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6277 ifindex, IPPROTO_UDP, netns_id,
6281 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6282 .func = bpf_xdp_sk_lookup_udp,
6285 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6286 .arg1_type = ARG_PTR_TO_CTX,
6287 .arg2_type = ARG_PTR_TO_MEM,
6288 .arg3_type = ARG_CONST_SIZE,
6289 .arg4_type = ARG_ANYTHING,
6290 .arg5_type = ARG_ANYTHING,
6293 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6294 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6296 struct net *caller_net = dev_net(ctx->rxq->dev);
6297 int ifindex = ctx->rxq->dev->ifindex;
6299 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6300 ifindex, IPPROTO_TCP, netns_id,
6304 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6305 .func = bpf_xdp_skc_lookup_tcp,
6308 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6309 .arg1_type = ARG_PTR_TO_CTX,
6310 .arg2_type = ARG_PTR_TO_MEM,
6311 .arg3_type = ARG_CONST_SIZE,
6312 .arg4_type = ARG_ANYTHING,
6313 .arg5_type = ARG_ANYTHING,
6316 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6317 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6319 struct net *caller_net = dev_net(ctx->rxq->dev);
6320 int ifindex = ctx->rxq->dev->ifindex;
6322 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6323 ifindex, IPPROTO_TCP, netns_id,
6327 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6328 .func = bpf_xdp_sk_lookup_tcp,
6331 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6332 .arg1_type = ARG_PTR_TO_CTX,
6333 .arg2_type = ARG_PTR_TO_MEM,
6334 .arg3_type = ARG_CONST_SIZE,
6335 .arg4_type = ARG_ANYTHING,
6336 .arg5_type = ARG_ANYTHING,
6339 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6340 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6342 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6343 sock_net(ctx->sk), 0,
6344 IPPROTO_TCP, netns_id, flags);
6347 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6348 .func = bpf_sock_addr_skc_lookup_tcp,
6350 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6351 .arg1_type = ARG_PTR_TO_CTX,
6352 .arg2_type = ARG_PTR_TO_MEM,
6353 .arg3_type = ARG_CONST_SIZE,
6354 .arg4_type = ARG_ANYTHING,
6355 .arg5_type = ARG_ANYTHING,
6358 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6359 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6361 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6362 sock_net(ctx->sk), 0, IPPROTO_TCP,
6366 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6367 .func = bpf_sock_addr_sk_lookup_tcp,
6369 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6370 .arg1_type = ARG_PTR_TO_CTX,
6371 .arg2_type = ARG_PTR_TO_MEM,
6372 .arg3_type = ARG_CONST_SIZE,
6373 .arg4_type = ARG_ANYTHING,
6374 .arg5_type = ARG_ANYTHING,
6377 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6378 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6380 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6381 sock_net(ctx->sk), 0, IPPROTO_UDP,
6385 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6386 .func = bpf_sock_addr_sk_lookup_udp,
6388 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6389 .arg1_type = ARG_PTR_TO_CTX,
6390 .arg2_type = ARG_PTR_TO_MEM,
6391 .arg3_type = ARG_CONST_SIZE,
6392 .arg4_type = ARG_ANYTHING,
6393 .arg5_type = ARG_ANYTHING,
6396 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6397 struct bpf_insn_access_aux *info)
6399 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6403 if (off % size != 0)
6407 case offsetof(struct bpf_tcp_sock, bytes_received):
6408 case offsetof(struct bpf_tcp_sock, bytes_acked):
6409 return size == sizeof(__u64);
6411 return size == sizeof(__u32);
6415 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6416 const struct bpf_insn *si,
6417 struct bpf_insn *insn_buf,
6418 struct bpf_prog *prog, u32 *target_size)
6420 struct bpf_insn *insn = insn_buf;
6422 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \
6424 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \
6425 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6426 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6427 si->dst_reg, si->src_reg, \
6428 offsetof(struct tcp_sock, FIELD)); \
6431 #define BPF_INET_SOCK_GET_COMMON(FIELD) \
6433 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \
6435 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6436 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6437 struct inet_connection_sock, \
6439 si->dst_reg, si->src_reg, \
6441 struct inet_connection_sock, \
6445 if (insn > insn_buf)
6446 return insn - insn_buf;
6449 case offsetof(struct bpf_tcp_sock, rtt_min):
6450 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
6451 sizeof(struct minmax));
6452 BUILD_BUG_ON(sizeof(struct minmax) <
6453 sizeof(struct minmax_sample));
6455 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6456 offsetof(struct tcp_sock, rtt_min) +
6457 offsetof(struct minmax_sample, v));
6459 case offsetof(struct bpf_tcp_sock, snd_cwnd):
6460 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
6462 case offsetof(struct bpf_tcp_sock, srtt_us):
6463 BPF_TCP_SOCK_GET_COMMON(srtt_us);
6465 case offsetof(struct bpf_tcp_sock, snd_ssthresh):
6466 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
6468 case offsetof(struct bpf_tcp_sock, rcv_nxt):
6469 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
6471 case offsetof(struct bpf_tcp_sock, snd_nxt):
6472 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
6474 case offsetof(struct bpf_tcp_sock, snd_una):
6475 BPF_TCP_SOCK_GET_COMMON(snd_una);
6477 case offsetof(struct bpf_tcp_sock, mss_cache):
6478 BPF_TCP_SOCK_GET_COMMON(mss_cache);
6480 case offsetof(struct bpf_tcp_sock, ecn_flags):
6481 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
6483 case offsetof(struct bpf_tcp_sock, rate_delivered):
6484 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
6486 case offsetof(struct bpf_tcp_sock, rate_interval_us):
6487 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
6489 case offsetof(struct bpf_tcp_sock, packets_out):
6490 BPF_TCP_SOCK_GET_COMMON(packets_out);
6492 case offsetof(struct bpf_tcp_sock, retrans_out):
6493 BPF_TCP_SOCK_GET_COMMON(retrans_out);
6495 case offsetof(struct bpf_tcp_sock, total_retrans):
6496 BPF_TCP_SOCK_GET_COMMON(total_retrans);
6498 case offsetof(struct bpf_tcp_sock, segs_in):
6499 BPF_TCP_SOCK_GET_COMMON(segs_in);
6501 case offsetof(struct bpf_tcp_sock, data_segs_in):
6502 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
6504 case offsetof(struct bpf_tcp_sock, segs_out):
6505 BPF_TCP_SOCK_GET_COMMON(segs_out);
6507 case offsetof(struct bpf_tcp_sock, data_segs_out):
6508 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
6510 case offsetof(struct bpf_tcp_sock, lost_out):
6511 BPF_TCP_SOCK_GET_COMMON(lost_out);
6513 case offsetof(struct bpf_tcp_sock, sacked_out):
6514 BPF_TCP_SOCK_GET_COMMON(sacked_out);
6516 case offsetof(struct bpf_tcp_sock, bytes_received):
6517 BPF_TCP_SOCK_GET_COMMON(bytes_received);
6519 case offsetof(struct bpf_tcp_sock, bytes_acked):
6520 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
6522 case offsetof(struct bpf_tcp_sock, dsack_dups):
6523 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
6525 case offsetof(struct bpf_tcp_sock, delivered):
6526 BPF_TCP_SOCK_GET_COMMON(delivered);
6528 case offsetof(struct bpf_tcp_sock, delivered_ce):
6529 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
6531 case offsetof(struct bpf_tcp_sock, icsk_retransmits):
6532 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
6536 return insn - insn_buf;
6539 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
6541 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
6542 return (unsigned long)sk;
6544 return (unsigned long)NULL;
6547 const struct bpf_func_proto bpf_tcp_sock_proto = {
6548 .func = bpf_tcp_sock,
6550 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
6551 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6554 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
6556 sk = sk_to_full_sk(sk);
6558 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
6559 return (unsigned long)sk;
6561 return (unsigned long)NULL;
6564 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
6565 .func = bpf_get_listener_sock,
6567 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6568 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6571 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
6573 unsigned int iphdr_len;
6575 switch (skb_protocol(skb, true)) {
6576 case cpu_to_be16(ETH_P_IP):
6577 iphdr_len = sizeof(struct iphdr);
6579 case cpu_to_be16(ETH_P_IPV6):
6580 iphdr_len = sizeof(struct ipv6hdr);
6586 if (skb_headlen(skb) < iphdr_len)
6589 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
6592 return INET_ECN_set_ce(skb);
6595 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6596 struct bpf_insn_access_aux *info)
6598 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
6601 if (off % size != 0)
6606 return size == sizeof(__u32);
6610 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
6611 const struct bpf_insn *si,
6612 struct bpf_insn *insn_buf,
6613 struct bpf_prog *prog, u32 *target_size)
6615 struct bpf_insn *insn = insn_buf;
6617 #define BPF_XDP_SOCK_GET(FIELD) \
6619 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \
6620 sizeof_field(struct bpf_xdp_sock, FIELD)); \
6621 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
6622 si->dst_reg, si->src_reg, \
6623 offsetof(struct xdp_sock, FIELD)); \
6627 case offsetof(struct bpf_xdp_sock, queue_id):
6628 BPF_XDP_SOCK_GET(queue_id);
6632 return insn - insn_buf;
6635 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
6636 .func = bpf_skb_ecn_set_ce,
6638 .ret_type = RET_INTEGER,
6639 .arg1_type = ARG_PTR_TO_CTX,
6642 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6643 struct tcphdr *, th, u32, th_len)
6645 #ifdef CONFIG_SYN_COOKIES
6649 if (unlikely(!sk || th_len < sizeof(*th)))
6652 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
6653 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6656 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
6659 if (!th->ack || th->rst || th->syn)
6662 if (tcp_synq_no_recent_overflow(sk))
6665 cookie = ntohl(th->ack_seq) - 1;
6667 switch (sk->sk_family) {
6669 if (unlikely(iph_len < sizeof(struct iphdr)))
6672 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
6675 #if IS_BUILTIN(CONFIG_IPV6)
6677 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6680 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
6682 #endif /* CONFIG_IPV6 */
6685 return -EPROTONOSUPPORT;
6697 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
6698 .func = bpf_tcp_check_syncookie,
6701 .ret_type = RET_INTEGER,
6702 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6703 .arg2_type = ARG_PTR_TO_MEM,
6704 .arg3_type = ARG_CONST_SIZE,
6705 .arg4_type = ARG_PTR_TO_MEM,
6706 .arg5_type = ARG_CONST_SIZE,
6709 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6710 struct tcphdr *, th, u32, th_len)
6712 #ifdef CONFIG_SYN_COOKIES
6716 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
6719 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6722 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
6725 if (!th->syn || th->ack || th->fin || th->rst)
6728 if (unlikely(iph_len < sizeof(struct iphdr)))
6731 /* Both struct iphdr and struct ipv6hdr have the version field at the
6732 * same offset so we can cast to the shorter header (struct iphdr).
6734 switch (((struct iphdr *)iph)->version) {
6736 if (sk->sk_family == AF_INET6 && sk->sk_ipv6only)
6739 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
6742 #if IS_BUILTIN(CONFIG_IPV6)
6744 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6747 if (sk->sk_family != AF_INET6)
6750 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
6752 #endif /* CONFIG_IPV6 */
6755 return -EPROTONOSUPPORT;
6760 return cookie | ((u64)mss << 32);
6763 #endif /* CONFIG_SYN_COOKIES */
6766 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
6767 .func = bpf_tcp_gen_syncookie,
6768 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */
6770 .ret_type = RET_INTEGER,
6771 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6772 .arg2_type = ARG_PTR_TO_MEM,
6773 .arg3_type = ARG_CONST_SIZE,
6774 .arg4_type = ARG_PTR_TO_MEM,
6775 .arg5_type = ARG_CONST_SIZE,
6778 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
6780 if (!sk || flags != 0)
6782 if (!skb_at_tc_ingress(skb))
6784 if (unlikely(dev_net(skb->dev) != sock_net(sk)))
6785 return -ENETUNREACH;
6786 if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
6787 return -ESOCKTNOSUPPORT;
6788 if (sk_is_refcounted(sk) &&
6789 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
6794 skb->destructor = sock_pfree;
6799 static const struct bpf_func_proto bpf_sk_assign_proto = {
6800 .func = bpf_sk_assign,
6802 .ret_type = RET_INTEGER,
6803 .arg1_type = ARG_PTR_TO_CTX,
6804 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6805 .arg3_type = ARG_ANYTHING,
6808 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
6809 u8 search_kind, const u8 *magic,
6810 u8 magic_len, bool *eol)
6816 while (op < opend) {
6819 if (kind == TCPOPT_EOL) {
6821 return ERR_PTR(-ENOMSG);
6822 } else if (kind == TCPOPT_NOP) {
6827 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
6828 /* Something is wrong in the received header.
6829 * Follow the TCP stack's tcp_parse_options()
6830 * and just bail here.
6832 return ERR_PTR(-EFAULT);
6835 if (search_kind == kind) {
6839 if (magic_len > kind_len - 2)
6840 return ERR_PTR(-ENOMSG);
6842 if (!memcmp(&op[2], magic, magic_len))
6849 return ERR_PTR(-ENOMSG);
6852 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6853 void *, search_res, u32, len, u64, flags)
6855 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
6856 const u8 *op, *opend, *magic, *search = search_res;
6857 u8 search_kind, search_len, copy_len, magic_len;
6860 /* 2 byte is the minimal option len except TCPOPT_NOP and
6861 * TCPOPT_EOL which are useless for the bpf prog to learn
6862 * and this helper disallow loading them also.
6864 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
6867 search_kind = search[0];
6868 search_len = search[1];
6870 if (search_len > len || search_kind == TCPOPT_NOP ||
6871 search_kind == TCPOPT_EOL)
6874 if (search_kind == TCPOPT_EXP || search_kind == 253) {
6875 /* 16 or 32 bit magic. +2 for kind and kind length */
6876 if (search_len != 4 && search_len != 6)
6879 magic_len = search_len - 2;
6888 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
6893 op += sizeof(struct tcphdr);
6895 if (!bpf_sock->skb ||
6896 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6897 /* This bpf_sock->op cannot call this helper */
6900 opend = bpf_sock->skb_data_end;
6901 op = bpf_sock->skb->data + sizeof(struct tcphdr);
6904 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
6911 if (copy_len > len) {
6916 memcpy(search_res, op, copy_len);
6920 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
6921 .func = bpf_sock_ops_load_hdr_opt,
6923 .ret_type = RET_INTEGER,
6924 .arg1_type = ARG_PTR_TO_CTX,
6925 .arg2_type = ARG_PTR_TO_MEM,
6926 .arg3_type = ARG_CONST_SIZE,
6927 .arg4_type = ARG_ANYTHING,
6930 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6931 const void *, from, u32, len, u64, flags)
6933 u8 new_kind, new_kind_len, magic_len = 0, *opend;
6934 const u8 *op, *new_op, *magic = NULL;
6935 struct sk_buff *skb;
6938 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
6941 if (len < 2 || flags)
6945 new_kind = new_op[0];
6946 new_kind_len = new_op[1];
6948 if (new_kind_len > len || new_kind == TCPOPT_NOP ||
6949 new_kind == TCPOPT_EOL)
6952 if (new_kind_len > bpf_sock->remaining_opt_len)
6955 /* 253 is another experimental kind */
6956 if (new_kind == TCPOPT_EXP || new_kind == 253) {
6957 if (new_kind_len < 4)
6959 /* Match for the 2 byte magic also.
6960 * RFC 6994: the magic could be 2 or 4 bytes.
6961 * Hence, matching by 2 byte only is on the
6962 * conservative side but it is the right
6963 * thing to do for the 'search-for-duplication'
6970 /* Check for duplication */
6971 skb = bpf_sock->skb;
6972 op = skb->data + sizeof(struct tcphdr);
6973 opend = bpf_sock->skb_data_end;
6975 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
6980 if (PTR_ERR(op) != -ENOMSG)
6984 /* The option has been ended. Treat it as no more
6985 * header option can be written.
6989 /* No duplication found. Store the header option. */
6990 memcpy(opend, from, new_kind_len);
6992 bpf_sock->remaining_opt_len -= new_kind_len;
6993 bpf_sock->skb_data_end += new_kind_len;
6998 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
6999 .func = bpf_sock_ops_store_hdr_opt,
7001 .ret_type = RET_INTEGER,
7002 .arg1_type = ARG_PTR_TO_CTX,
7003 .arg2_type = ARG_PTR_TO_MEM,
7004 .arg3_type = ARG_CONST_SIZE,
7005 .arg4_type = ARG_ANYTHING,
7008 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7009 u32, len, u64, flags)
7011 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7014 if (flags || len < 2)
7017 if (len > bpf_sock->remaining_opt_len)
7020 bpf_sock->remaining_opt_len -= len;
7025 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7026 .func = bpf_sock_ops_reserve_hdr_opt,
7028 .ret_type = RET_INTEGER,
7029 .arg1_type = ARG_PTR_TO_CTX,
7030 .arg2_type = ARG_ANYTHING,
7031 .arg3_type = ARG_ANYTHING,
7034 #endif /* CONFIG_INET */
7036 bool bpf_helper_changes_pkt_data(void *func)
7038 if (func == bpf_skb_vlan_push ||
7039 func == bpf_skb_vlan_pop ||
7040 func == bpf_skb_store_bytes ||
7041 func == bpf_skb_change_proto ||
7042 func == bpf_skb_change_head ||
7043 func == sk_skb_change_head ||
7044 func == bpf_skb_change_tail ||
7045 func == sk_skb_change_tail ||
7046 func == bpf_skb_adjust_room ||
7047 func == sk_skb_adjust_room ||
7048 func == bpf_skb_pull_data ||
7049 func == sk_skb_pull_data ||
7050 func == bpf_clone_redirect ||
7051 func == bpf_l3_csum_replace ||
7052 func == bpf_l4_csum_replace ||
7053 func == bpf_xdp_adjust_head ||
7054 func == bpf_xdp_adjust_meta ||
7055 func == bpf_msg_pull_data ||
7056 func == bpf_msg_push_data ||
7057 func == bpf_msg_pop_data ||
7058 func == bpf_xdp_adjust_tail ||
7059 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7060 func == bpf_lwt_seg6_store_bytes ||
7061 func == bpf_lwt_seg6_adjust_srh ||
7062 func == bpf_lwt_seg6_action ||
7065 func == bpf_sock_ops_store_hdr_opt ||
7067 func == bpf_lwt_in_push_encap ||
7068 func == bpf_lwt_xmit_push_encap)
7074 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7075 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7077 static const struct bpf_func_proto *
7078 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7081 /* inet and inet6 sockets are created in a process
7082 * context so there is always a valid uid/gid
7084 case BPF_FUNC_get_current_uid_gid:
7085 return &bpf_get_current_uid_gid_proto;
7086 case BPF_FUNC_get_local_storage:
7087 return &bpf_get_local_storage_proto;
7088 case BPF_FUNC_get_socket_cookie:
7089 return &bpf_get_socket_cookie_sock_proto;
7090 case BPF_FUNC_get_netns_cookie:
7091 return &bpf_get_netns_cookie_sock_proto;
7092 case BPF_FUNC_perf_event_output:
7093 return &bpf_event_output_data_proto;
7094 case BPF_FUNC_get_current_pid_tgid:
7095 return &bpf_get_current_pid_tgid_proto;
7096 case BPF_FUNC_get_current_comm:
7097 return &bpf_get_current_comm_proto;
7098 #ifdef CONFIG_CGROUPS
7099 case BPF_FUNC_get_current_cgroup_id:
7100 return &bpf_get_current_cgroup_id_proto;
7101 case BPF_FUNC_get_current_ancestor_cgroup_id:
7102 return &bpf_get_current_ancestor_cgroup_id_proto;
7104 #ifdef CONFIG_CGROUP_NET_CLASSID
7105 case BPF_FUNC_get_cgroup_classid:
7106 return &bpf_get_cgroup_classid_curr_proto;
7108 case BPF_FUNC_sk_storage_get:
7109 return &bpf_sk_storage_get_cg_sock_proto;
7111 return bpf_base_func_proto(func_id);
7115 static const struct bpf_func_proto *
7116 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7119 /* inet and inet6 sockets are created in a process
7120 * context so there is always a valid uid/gid
7122 case BPF_FUNC_get_current_uid_gid:
7123 return &bpf_get_current_uid_gid_proto;
7125 switch (prog->expected_attach_type) {
7126 case BPF_CGROUP_INET4_CONNECT:
7127 case BPF_CGROUP_INET6_CONNECT:
7128 return &bpf_bind_proto;
7132 case BPF_FUNC_get_socket_cookie:
7133 return &bpf_get_socket_cookie_sock_addr_proto;
7134 case BPF_FUNC_get_netns_cookie:
7135 return &bpf_get_netns_cookie_sock_addr_proto;
7136 case BPF_FUNC_get_local_storage:
7137 return &bpf_get_local_storage_proto;
7138 case BPF_FUNC_perf_event_output:
7139 return &bpf_event_output_data_proto;
7140 case BPF_FUNC_get_current_pid_tgid:
7141 return &bpf_get_current_pid_tgid_proto;
7142 case BPF_FUNC_get_current_comm:
7143 return &bpf_get_current_comm_proto;
7144 #ifdef CONFIG_CGROUPS
7145 case BPF_FUNC_get_current_cgroup_id:
7146 return &bpf_get_current_cgroup_id_proto;
7147 case BPF_FUNC_get_current_ancestor_cgroup_id:
7148 return &bpf_get_current_ancestor_cgroup_id_proto;
7150 #ifdef CONFIG_CGROUP_NET_CLASSID
7151 case BPF_FUNC_get_cgroup_classid:
7152 return &bpf_get_cgroup_classid_curr_proto;
7155 case BPF_FUNC_sk_lookup_tcp:
7156 return &bpf_sock_addr_sk_lookup_tcp_proto;
7157 case BPF_FUNC_sk_lookup_udp:
7158 return &bpf_sock_addr_sk_lookup_udp_proto;
7159 case BPF_FUNC_sk_release:
7160 return &bpf_sk_release_proto;
7161 case BPF_FUNC_skc_lookup_tcp:
7162 return &bpf_sock_addr_skc_lookup_tcp_proto;
7163 #endif /* CONFIG_INET */
7164 case BPF_FUNC_sk_storage_get:
7165 return &bpf_sk_storage_get_proto;
7166 case BPF_FUNC_sk_storage_delete:
7167 return &bpf_sk_storage_delete_proto;
7168 case BPF_FUNC_setsockopt:
7169 switch (prog->expected_attach_type) {
7170 case BPF_CGROUP_INET4_BIND:
7171 case BPF_CGROUP_INET6_BIND:
7172 case BPF_CGROUP_INET4_CONNECT:
7173 case BPF_CGROUP_INET6_CONNECT:
7174 case BPF_CGROUP_UDP4_RECVMSG:
7175 case BPF_CGROUP_UDP6_RECVMSG:
7176 case BPF_CGROUP_UDP4_SENDMSG:
7177 case BPF_CGROUP_UDP6_SENDMSG:
7178 case BPF_CGROUP_INET4_GETPEERNAME:
7179 case BPF_CGROUP_INET6_GETPEERNAME:
7180 case BPF_CGROUP_INET4_GETSOCKNAME:
7181 case BPF_CGROUP_INET6_GETSOCKNAME:
7182 return &bpf_sock_addr_setsockopt_proto;
7186 case BPF_FUNC_getsockopt:
7187 switch (prog->expected_attach_type) {
7188 case BPF_CGROUP_INET4_BIND:
7189 case BPF_CGROUP_INET6_BIND:
7190 case BPF_CGROUP_INET4_CONNECT:
7191 case BPF_CGROUP_INET6_CONNECT:
7192 case BPF_CGROUP_UDP4_RECVMSG:
7193 case BPF_CGROUP_UDP6_RECVMSG:
7194 case BPF_CGROUP_UDP4_SENDMSG:
7195 case BPF_CGROUP_UDP6_SENDMSG:
7196 case BPF_CGROUP_INET4_GETPEERNAME:
7197 case BPF_CGROUP_INET6_GETPEERNAME:
7198 case BPF_CGROUP_INET4_GETSOCKNAME:
7199 case BPF_CGROUP_INET6_GETSOCKNAME:
7200 return &bpf_sock_addr_getsockopt_proto;
7205 return bpf_sk_base_func_proto(func_id);
7209 static const struct bpf_func_proto *
7210 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7213 case BPF_FUNC_skb_load_bytes:
7214 return &bpf_skb_load_bytes_proto;
7215 case BPF_FUNC_skb_load_bytes_relative:
7216 return &bpf_skb_load_bytes_relative_proto;
7217 case BPF_FUNC_get_socket_cookie:
7218 return &bpf_get_socket_cookie_proto;
7219 case BPF_FUNC_get_socket_uid:
7220 return &bpf_get_socket_uid_proto;
7221 case BPF_FUNC_perf_event_output:
7222 return &bpf_skb_event_output_proto;
7224 return bpf_sk_base_func_proto(func_id);
7228 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7229 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7231 static const struct bpf_func_proto *
7232 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7235 case BPF_FUNC_get_local_storage:
7236 return &bpf_get_local_storage_proto;
7237 case BPF_FUNC_sk_fullsock:
7238 return &bpf_sk_fullsock_proto;
7239 case BPF_FUNC_sk_storage_get:
7240 return &bpf_sk_storage_get_proto;
7241 case BPF_FUNC_sk_storage_delete:
7242 return &bpf_sk_storage_delete_proto;
7243 case BPF_FUNC_perf_event_output:
7244 return &bpf_skb_event_output_proto;
7245 #ifdef CONFIG_SOCK_CGROUP_DATA
7246 case BPF_FUNC_skb_cgroup_id:
7247 return &bpf_skb_cgroup_id_proto;
7248 case BPF_FUNC_skb_ancestor_cgroup_id:
7249 return &bpf_skb_ancestor_cgroup_id_proto;
7250 case BPF_FUNC_sk_cgroup_id:
7251 return &bpf_sk_cgroup_id_proto;
7252 case BPF_FUNC_sk_ancestor_cgroup_id:
7253 return &bpf_sk_ancestor_cgroup_id_proto;
7256 case BPF_FUNC_sk_lookup_tcp:
7257 return &bpf_sk_lookup_tcp_proto;
7258 case BPF_FUNC_sk_lookup_udp:
7259 return &bpf_sk_lookup_udp_proto;
7260 case BPF_FUNC_sk_release:
7261 return &bpf_sk_release_proto;
7262 case BPF_FUNC_skc_lookup_tcp:
7263 return &bpf_skc_lookup_tcp_proto;
7264 case BPF_FUNC_tcp_sock:
7265 return &bpf_tcp_sock_proto;
7266 case BPF_FUNC_get_listener_sock:
7267 return &bpf_get_listener_sock_proto;
7268 case BPF_FUNC_skb_ecn_set_ce:
7269 return &bpf_skb_ecn_set_ce_proto;
7272 return sk_filter_func_proto(func_id, prog);
7276 static const struct bpf_func_proto *
7277 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7280 case BPF_FUNC_skb_store_bytes:
7281 return &bpf_skb_store_bytes_proto;
7282 case BPF_FUNC_skb_load_bytes:
7283 return &bpf_skb_load_bytes_proto;
7284 case BPF_FUNC_skb_load_bytes_relative:
7285 return &bpf_skb_load_bytes_relative_proto;
7286 case BPF_FUNC_skb_pull_data:
7287 return &bpf_skb_pull_data_proto;
7288 case BPF_FUNC_csum_diff:
7289 return &bpf_csum_diff_proto;
7290 case BPF_FUNC_csum_update:
7291 return &bpf_csum_update_proto;
7292 case BPF_FUNC_csum_level:
7293 return &bpf_csum_level_proto;
7294 case BPF_FUNC_l3_csum_replace:
7295 return &bpf_l3_csum_replace_proto;
7296 case BPF_FUNC_l4_csum_replace:
7297 return &bpf_l4_csum_replace_proto;
7298 case BPF_FUNC_clone_redirect:
7299 return &bpf_clone_redirect_proto;
7300 case BPF_FUNC_get_cgroup_classid:
7301 return &bpf_get_cgroup_classid_proto;
7302 case BPF_FUNC_skb_vlan_push:
7303 return &bpf_skb_vlan_push_proto;
7304 case BPF_FUNC_skb_vlan_pop:
7305 return &bpf_skb_vlan_pop_proto;
7306 case BPF_FUNC_skb_change_proto:
7307 return &bpf_skb_change_proto_proto;
7308 case BPF_FUNC_skb_change_type:
7309 return &bpf_skb_change_type_proto;
7310 case BPF_FUNC_skb_adjust_room:
7311 return &bpf_skb_adjust_room_proto;
7312 case BPF_FUNC_skb_change_tail:
7313 return &bpf_skb_change_tail_proto;
7314 case BPF_FUNC_skb_change_head:
7315 return &bpf_skb_change_head_proto;
7316 case BPF_FUNC_skb_get_tunnel_key:
7317 return &bpf_skb_get_tunnel_key_proto;
7318 case BPF_FUNC_skb_set_tunnel_key:
7319 return bpf_get_skb_set_tunnel_proto(func_id);
7320 case BPF_FUNC_skb_get_tunnel_opt:
7321 return &bpf_skb_get_tunnel_opt_proto;
7322 case BPF_FUNC_skb_set_tunnel_opt:
7323 return bpf_get_skb_set_tunnel_proto(func_id);
7324 case BPF_FUNC_redirect:
7325 return &bpf_redirect_proto;
7326 case BPF_FUNC_redirect_neigh:
7327 return &bpf_redirect_neigh_proto;
7328 case BPF_FUNC_redirect_peer:
7329 return &bpf_redirect_peer_proto;
7330 case BPF_FUNC_get_route_realm:
7331 return &bpf_get_route_realm_proto;
7332 case BPF_FUNC_get_hash_recalc:
7333 return &bpf_get_hash_recalc_proto;
7334 case BPF_FUNC_set_hash_invalid:
7335 return &bpf_set_hash_invalid_proto;
7336 case BPF_FUNC_set_hash:
7337 return &bpf_set_hash_proto;
7338 case BPF_FUNC_perf_event_output:
7339 return &bpf_skb_event_output_proto;
7340 case BPF_FUNC_get_smp_processor_id:
7341 return &bpf_get_smp_processor_id_proto;
7342 case BPF_FUNC_skb_under_cgroup:
7343 return &bpf_skb_under_cgroup_proto;
7344 case BPF_FUNC_get_socket_cookie:
7345 return &bpf_get_socket_cookie_proto;
7346 case BPF_FUNC_get_socket_uid:
7347 return &bpf_get_socket_uid_proto;
7348 case BPF_FUNC_fib_lookup:
7349 return &bpf_skb_fib_lookup_proto;
7350 case BPF_FUNC_check_mtu:
7351 return &bpf_skb_check_mtu_proto;
7352 case BPF_FUNC_sk_fullsock:
7353 return &bpf_sk_fullsock_proto;
7354 case BPF_FUNC_sk_storage_get:
7355 return &bpf_sk_storage_get_proto;
7356 case BPF_FUNC_sk_storage_delete:
7357 return &bpf_sk_storage_delete_proto;
7359 case BPF_FUNC_skb_get_xfrm_state:
7360 return &bpf_skb_get_xfrm_state_proto;
7362 #ifdef CONFIG_CGROUP_NET_CLASSID
7363 case BPF_FUNC_skb_cgroup_classid:
7364 return &bpf_skb_cgroup_classid_proto;
7366 #ifdef CONFIG_SOCK_CGROUP_DATA
7367 case BPF_FUNC_skb_cgroup_id:
7368 return &bpf_skb_cgroup_id_proto;
7369 case BPF_FUNC_skb_ancestor_cgroup_id:
7370 return &bpf_skb_ancestor_cgroup_id_proto;
7373 case BPF_FUNC_sk_lookup_tcp:
7374 return &bpf_sk_lookup_tcp_proto;
7375 case BPF_FUNC_sk_lookup_udp:
7376 return &bpf_sk_lookup_udp_proto;
7377 case BPF_FUNC_sk_release:
7378 return &bpf_sk_release_proto;
7379 case BPF_FUNC_tcp_sock:
7380 return &bpf_tcp_sock_proto;
7381 case BPF_FUNC_get_listener_sock:
7382 return &bpf_get_listener_sock_proto;
7383 case BPF_FUNC_skc_lookup_tcp:
7384 return &bpf_skc_lookup_tcp_proto;
7385 case BPF_FUNC_tcp_check_syncookie:
7386 return &bpf_tcp_check_syncookie_proto;
7387 case BPF_FUNC_skb_ecn_set_ce:
7388 return &bpf_skb_ecn_set_ce_proto;
7389 case BPF_FUNC_tcp_gen_syncookie:
7390 return &bpf_tcp_gen_syncookie_proto;
7391 case BPF_FUNC_sk_assign:
7392 return &bpf_sk_assign_proto;
7395 return bpf_sk_base_func_proto(func_id);
7399 static const struct bpf_func_proto *
7400 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7403 case BPF_FUNC_perf_event_output:
7404 return &bpf_xdp_event_output_proto;
7405 case BPF_FUNC_get_smp_processor_id:
7406 return &bpf_get_smp_processor_id_proto;
7407 case BPF_FUNC_csum_diff:
7408 return &bpf_csum_diff_proto;
7409 case BPF_FUNC_xdp_adjust_head:
7410 return &bpf_xdp_adjust_head_proto;
7411 case BPF_FUNC_xdp_adjust_meta:
7412 return &bpf_xdp_adjust_meta_proto;
7413 case BPF_FUNC_redirect:
7414 return &bpf_xdp_redirect_proto;
7415 case BPF_FUNC_redirect_map:
7416 return &bpf_xdp_redirect_map_proto;
7417 case BPF_FUNC_xdp_adjust_tail:
7418 return &bpf_xdp_adjust_tail_proto;
7419 case BPF_FUNC_fib_lookup:
7420 return &bpf_xdp_fib_lookup_proto;
7421 case BPF_FUNC_check_mtu:
7422 return &bpf_xdp_check_mtu_proto;
7424 case BPF_FUNC_sk_lookup_udp:
7425 return &bpf_xdp_sk_lookup_udp_proto;
7426 case BPF_FUNC_sk_lookup_tcp:
7427 return &bpf_xdp_sk_lookup_tcp_proto;
7428 case BPF_FUNC_sk_release:
7429 return &bpf_sk_release_proto;
7430 case BPF_FUNC_skc_lookup_tcp:
7431 return &bpf_xdp_skc_lookup_tcp_proto;
7432 case BPF_FUNC_tcp_check_syncookie:
7433 return &bpf_tcp_check_syncookie_proto;
7434 case BPF_FUNC_tcp_gen_syncookie:
7435 return &bpf_tcp_gen_syncookie_proto;
7438 return bpf_sk_base_func_proto(func_id);
7442 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
7443 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
7445 static const struct bpf_func_proto *
7446 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7449 case BPF_FUNC_setsockopt:
7450 return &bpf_sock_ops_setsockopt_proto;
7451 case BPF_FUNC_getsockopt:
7452 return &bpf_sock_ops_getsockopt_proto;
7453 case BPF_FUNC_sock_ops_cb_flags_set:
7454 return &bpf_sock_ops_cb_flags_set_proto;
7455 case BPF_FUNC_sock_map_update:
7456 return &bpf_sock_map_update_proto;
7457 case BPF_FUNC_sock_hash_update:
7458 return &bpf_sock_hash_update_proto;
7459 case BPF_FUNC_get_socket_cookie:
7460 return &bpf_get_socket_cookie_sock_ops_proto;
7461 case BPF_FUNC_get_local_storage:
7462 return &bpf_get_local_storage_proto;
7463 case BPF_FUNC_perf_event_output:
7464 return &bpf_event_output_data_proto;
7465 case BPF_FUNC_sk_storage_get:
7466 return &bpf_sk_storage_get_proto;
7467 case BPF_FUNC_sk_storage_delete:
7468 return &bpf_sk_storage_delete_proto;
7470 case BPF_FUNC_load_hdr_opt:
7471 return &bpf_sock_ops_load_hdr_opt_proto;
7472 case BPF_FUNC_store_hdr_opt:
7473 return &bpf_sock_ops_store_hdr_opt_proto;
7474 case BPF_FUNC_reserve_hdr_opt:
7475 return &bpf_sock_ops_reserve_hdr_opt_proto;
7476 case BPF_FUNC_tcp_sock:
7477 return &bpf_tcp_sock_proto;
7478 #endif /* CONFIG_INET */
7480 return bpf_sk_base_func_proto(func_id);
7484 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
7485 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
7487 static const struct bpf_func_proto *
7488 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7491 case BPF_FUNC_msg_redirect_map:
7492 return &bpf_msg_redirect_map_proto;
7493 case BPF_FUNC_msg_redirect_hash:
7494 return &bpf_msg_redirect_hash_proto;
7495 case BPF_FUNC_msg_apply_bytes:
7496 return &bpf_msg_apply_bytes_proto;
7497 case BPF_FUNC_msg_cork_bytes:
7498 return &bpf_msg_cork_bytes_proto;
7499 case BPF_FUNC_msg_pull_data:
7500 return &bpf_msg_pull_data_proto;
7501 case BPF_FUNC_msg_push_data:
7502 return &bpf_msg_push_data_proto;
7503 case BPF_FUNC_msg_pop_data:
7504 return &bpf_msg_pop_data_proto;
7505 case BPF_FUNC_perf_event_output:
7506 return &bpf_event_output_data_proto;
7507 case BPF_FUNC_get_current_uid_gid:
7508 return &bpf_get_current_uid_gid_proto;
7509 case BPF_FUNC_get_current_pid_tgid:
7510 return &bpf_get_current_pid_tgid_proto;
7511 case BPF_FUNC_sk_storage_get:
7512 return &bpf_sk_storage_get_proto;
7513 case BPF_FUNC_sk_storage_delete:
7514 return &bpf_sk_storage_delete_proto;
7515 #ifdef CONFIG_CGROUPS
7516 case BPF_FUNC_get_current_cgroup_id:
7517 return &bpf_get_current_cgroup_id_proto;
7518 case BPF_FUNC_get_current_ancestor_cgroup_id:
7519 return &bpf_get_current_ancestor_cgroup_id_proto;
7521 #ifdef CONFIG_CGROUP_NET_CLASSID
7522 case BPF_FUNC_get_cgroup_classid:
7523 return &bpf_get_cgroup_classid_curr_proto;
7526 return bpf_sk_base_func_proto(func_id);
7530 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
7531 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
7533 static const struct bpf_func_proto *
7534 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7537 case BPF_FUNC_skb_store_bytes:
7538 return &bpf_skb_store_bytes_proto;
7539 case BPF_FUNC_skb_load_bytes:
7540 return &bpf_skb_load_bytes_proto;
7541 case BPF_FUNC_skb_pull_data:
7542 return &sk_skb_pull_data_proto;
7543 case BPF_FUNC_skb_change_tail:
7544 return &sk_skb_change_tail_proto;
7545 case BPF_FUNC_skb_change_head:
7546 return &sk_skb_change_head_proto;
7547 case BPF_FUNC_skb_adjust_room:
7548 return &sk_skb_adjust_room_proto;
7549 case BPF_FUNC_get_socket_cookie:
7550 return &bpf_get_socket_cookie_proto;
7551 case BPF_FUNC_get_socket_uid:
7552 return &bpf_get_socket_uid_proto;
7553 case BPF_FUNC_sk_redirect_map:
7554 return &bpf_sk_redirect_map_proto;
7555 case BPF_FUNC_sk_redirect_hash:
7556 return &bpf_sk_redirect_hash_proto;
7557 case BPF_FUNC_perf_event_output:
7558 return &bpf_skb_event_output_proto;
7560 case BPF_FUNC_sk_lookup_tcp:
7561 return &bpf_sk_lookup_tcp_proto;
7562 case BPF_FUNC_sk_lookup_udp:
7563 return &bpf_sk_lookup_udp_proto;
7564 case BPF_FUNC_sk_release:
7565 return &bpf_sk_release_proto;
7566 case BPF_FUNC_skc_lookup_tcp:
7567 return &bpf_skc_lookup_tcp_proto;
7570 return bpf_sk_base_func_proto(func_id);
7574 static const struct bpf_func_proto *
7575 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7578 case BPF_FUNC_skb_load_bytes:
7579 return &bpf_flow_dissector_load_bytes_proto;
7581 return bpf_sk_base_func_proto(func_id);
7585 static const struct bpf_func_proto *
7586 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7589 case BPF_FUNC_skb_load_bytes:
7590 return &bpf_skb_load_bytes_proto;
7591 case BPF_FUNC_skb_pull_data:
7592 return &bpf_skb_pull_data_proto;
7593 case BPF_FUNC_csum_diff:
7594 return &bpf_csum_diff_proto;
7595 case BPF_FUNC_get_cgroup_classid:
7596 return &bpf_get_cgroup_classid_proto;
7597 case BPF_FUNC_get_route_realm:
7598 return &bpf_get_route_realm_proto;
7599 case BPF_FUNC_get_hash_recalc:
7600 return &bpf_get_hash_recalc_proto;
7601 case BPF_FUNC_perf_event_output:
7602 return &bpf_skb_event_output_proto;
7603 case BPF_FUNC_get_smp_processor_id:
7604 return &bpf_get_smp_processor_id_proto;
7605 case BPF_FUNC_skb_under_cgroup:
7606 return &bpf_skb_under_cgroup_proto;
7608 return bpf_sk_base_func_proto(func_id);
7612 static const struct bpf_func_proto *
7613 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7616 case BPF_FUNC_lwt_push_encap:
7617 return &bpf_lwt_in_push_encap_proto;
7619 return lwt_out_func_proto(func_id, prog);
7623 static const struct bpf_func_proto *
7624 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7627 case BPF_FUNC_skb_get_tunnel_key:
7628 return &bpf_skb_get_tunnel_key_proto;
7629 case BPF_FUNC_skb_set_tunnel_key:
7630 return bpf_get_skb_set_tunnel_proto(func_id);
7631 case BPF_FUNC_skb_get_tunnel_opt:
7632 return &bpf_skb_get_tunnel_opt_proto;
7633 case BPF_FUNC_skb_set_tunnel_opt:
7634 return bpf_get_skb_set_tunnel_proto(func_id);
7635 case BPF_FUNC_redirect:
7636 return &bpf_redirect_proto;
7637 case BPF_FUNC_clone_redirect:
7638 return &bpf_clone_redirect_proto;
7639 case BPF_FUNC_skb_change_tail:
7640 return &bpf_skb_change_tail_proto;
7641 case BPF_FUNC_skb_change_head:
7642 return &bpf_skb_change_head_proto;
7643 case BPF_FUNC_skb_store_bytes:
7644 return &bpf_skb_store_bytes_proto;
7645 case BPF_FUNC_csum_update:
7646 return &bpf_csum_update_proto;
7647 case BPF_FUNC_csum_level:
7648 return &bpf_csum_level_proto;
7649 case BPF_FUNC_l3_csum_replace:
7650 return &bpf_l3_csum_replace_proto;
7651 case BPF_FUNC_l4_csum_replace:
7652 return &bpf_l4_csum_replace_proto;
7653 case BPF_FUNC_set_hash_invalid:
7654 return &bpf_set_hash_invalid_proto;
7655 case BPF_FUNC_lwt_push_encap:
7656 return &bpf_lwt_xmit_push_encap_proto;
7658 return lwt_out_func_proto(func_id, prog);
7662 static const struct bpf_func_proto *
7663 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7666 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7667 case BPF_FUNC_lwt_seg6_store_bytes:
7668 return &bpf_lwt_seg6_store_bytes_proto;
7669 case BPF_FUNC_lwt_seg6_action:
7670 return &bpf_lwt_seg6_action_proto;
7671 case BPF_FUNC_lwt_seg6_adjust_srh:
7672 return &bpf_lwt_seg6_adjust_srh_proto;
7675 return lwt_out_func_proto(func_id, prog);
7679 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
7680 const struct bpf_prog *prog,
7681 struct bpf_insn_access_aux *info)
7683 const int size_default = sizeof(__u32);
7685 if (off < 0 || off >= sizeof(struct __sk_buff))
7688 /* The verifier guarantees that size > 0. */
7689 if (off % size != 0)
7693 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7694 if (off + size > offsetofend(struct __sk_buff, cb[4]))
7697 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
7698 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
7699 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
7700 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
7701 case bpf_ctx_range(struct __sk_buff, data):
7702 case bpf_ctx_range(struct __sk_buff, data_meta):
7703 case bpf_ctx_range(struct __sk_buff, data_end):
7704 if (size != size_default)
7707 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
7709 case bpf_ctx_range(struct __sk_buff, tstamp):
7710 if (size != sizeof(__u64))
7713 case offsetof(struct __sk_buff, sk):
7714 if (type == BPF_WRITE || size != sizeof(__u64))
7716 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
7719 /* Only narrow read access allowed for now. */
7720 if (type == BPF_WRITE) {
7721 if (size != size_default)
7724 bpf_ctx_record_field_size(info, size_default);
7725 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
7733 static bool sk_filter_is_valid_access(int off, int size,
7734 enum bpf_access_type type,
7735 const struct bpf_prog *prog,
7736 struct bpf_insn_access_aux *info)
7739 case bpf_ctx_range(struct __sk_buff, tc_classid):
7740 case bpf_ctx_range(struct __sk_buff, data):
7741 case bpf_ctx_range(struct __sk_buff, data_meta):
7742 case bpf_ctx_range(struct __sk_buff, data_end):
7743 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7744 case bpf_ctx_range(struct __sk_buff, tstamp):
7745 case bpf_ctx_range(struct __sk_buff, wire_len):
7749 if (type == BPF_WRITE) {
7751 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7758 return bpf_skb_is_valid_access(off, size, type, prog, info);
7761 static bool cg_skb_is_valid_access(int off, int size,
7762 enum bpf_access_type type,
7763 const struct bpf_prog *prog,
7764 struct bpf_insn_access_aux *info)
7767 case bpf_ctx_range(struct __sk_buff, tc_classid):
7768 case bpf_ctx_range(struct __sk_buff, data_meta):
7769 case bpf_ctx_range(struct __sk_buff, wire_len):
7771 case bpf_ctx_range(struct __sk_buff, data):
7772 case bpf_ctx_range(struct __sk_buff, data_end):
7778 if (type == BPF_WRITE) {
7780 case bpf_ctx_range(struct __sk_buff, mark):
7781 case bpf_ctx_range(struct __sk_buff, priority):
7782 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7784 case bpf_ctx_range(struct __sk_buff, tstamp):
7794 case bpf_ctx_range(struct __sk_buff, data):
7795 info->reg_type = PTR_TO_PACKET;
7797 case bpf_ctx_range(struct __sk_buff, data_end):
7798 info->reg_type = PTR_TO_PACKET_END;
7802 return bpf_skb_is_valid_access(off, size, type, prog, info);
7805 static bool lwt_is_valid_access(int off, int size,
7806 enum bpf_access_type type,
7807 const struct bpf_prog *prog,
7808 struct bpf_insn_access_aux *info)
7811 case bpf_ctx_range(struct __sk_buff, tc_classid):
7812 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7813 case bpf_ctx_range(struct __sk_buff, data_meta):
7814 case bpf_ctx_range(struct __sk_buff, tstamp):
7815 case bpf_ctx_range(struct __sk_buff, wire_len):
7819 if (type == BPF_WRITE) {
7821 case bpf_ctx_range(struct __sk_buff, mark):
7822 case bpf_ctx_range(struct __sk_buff, priority):
7823 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7831 case bpf_ctx_range(struct __sk_buff, data):
7832 info->reg_type = PTR_TO_PACKET;
7834 case bpf_ctx_range(struct __sk_buff, data_end):
7835 info->reg_type = PTR_TO_PACKET_END;
7839 return bpf_skb_is_valid_access(off, size, type, prog, info);
7842 /* Attach type specific accesses */
7843 static bool __sock_filter_check_attach_type(int off,
7844 enum bpf_access_type access_type,
7845 enum bpf_attach_type attach_type)
7848 case offsetof(struct bpf_sock, bound_dev_if):
7849 case offsetof(struct bpf_sock, mark):
7850 case offsetof(struct bpf_sock, priority):
7851 switch (attach_type) {
7852 case BPF_CGROUP_INET_SOCK_CREATE:
7853 case BPF_CGROUP_INET_SOCK_RELEASE:
7858 case bpf_ctx_range(struct bpf_sock, src_ip4):
7859 switch (attach_type) {
7860 case BPF_CGROUP_INET4_POST_BIND:
7865 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7866 switch (attach_type) {
7867 case BPF_CGROUP_INET6_POST_BIND:
7872 case bpf_ctx_range(struct bpf_sock, src_port):
7873 switch (attach_type) {
7874 case BPF_CGROUP_INET4_POST_BIND:
7875 case BPF_CGROUP_INET6_POST_BIND:
7882 return access_type == BPF_READ;
7887 bool bpf_sock_common_is_valid_access(int off, int size,
7888 enum bpf_access_type type,
7889 struct bpf_insn_access_aux *info)
7892 case bpf_ctx_range_till(struct bpf_sock, type, priority):
7895 return bpf_sock_is_valid_access(off, size, type, info);
7899 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7900 struct bpf_insn_access_aux *info)
7902 const int size_default = sizeof(__u32);
7904 if (off < 0 || off >= sizeof(struct bpf_sock))
7906 if (off % size != 0)
7910 case offsetof(struct bpf_sock, state):
7911 case offsetof(struct bpf_sock, family):
7912 case offsetof(struct bpf_sock, type):
7913 case offsetof(struct bpf_sock, protocol):
7914 case offsetof(struct bpf_sock, dst_port):
7915 case offsetof(struct bpf_sock, src_port):
7916 case offsetof(struct bpf_sock, rx_queue_mapping):
7917 case bpf_ctx_range(struct bpf_sock, src_ip4):
7918 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7919 case bpf_ctx_range(struct bpf_sock, dst_ip4):
7920 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
7921 bpf_ctx_record_field_size(info, size_default);
7922 return bpf_ctx_narrow_access_ok(off, size, size_default);
7925 return size == size_default;
7928 static bool sock_filter_is_valid_access(int off, int size,
7929 enum bpf_access_type type,
7930 const struct bpf_prog *prog,
7931 struct bpf_insn_access_aux *info)
7933 if (!bpf_sock_is_valid_access(off, size, type, info))
7935 return __sock_filter_check_attach_type(off, type,
7936 prog->expected_attach_type);
7939 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
7940 const struct bpf_prog *prog)
7942 /* Neither direct read nor direct write requires any preliminary
7948 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
7949 const struct bpf_prog *prog, int drop_verdict)
7951 struct bpf_insn *insn = insn_buf;
7956 /* if (!skb->cloned)
7959 * (Fast-path, otherwise approximation that we might be
7960 * a clone, do the rest in helper.)
7962 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
7963 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
7964 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
7966 /* ret = bpf_skb_pull_data(skb, 0); */
7967 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
7968 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
7969 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
7970 BPF_FUNC_skb_pull_data);
7973 * return TC_ACT_SHOT;
7975 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
7976 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
7977 *insn++ = BPF_EXIT_INSN();
7980 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
7982 *insn++ = prog->insnsi[0];
7984 return insn - insn_buf;
7987 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
7988 struct bpf_insn *insn_buf)
7990 bool indirect = BPF_MODE(orig->code) == BPF_IND;
7991 struct bpf_insn *insn = insn_buf;
7994 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
7996 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
7998 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
8000 /* We're guaranteed here that CTX is in R6. */
8001 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
8003 switch (BPF_SIZE(orig->code)) {
8005 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
8008 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
8011 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
8015 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
8016 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
8017 *insn++ = BPF_EXIT_INSN();
8019 return insn - insn_buf;
8022 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8023 const struct bpf_prog *prog)
8025 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8028 static bool tc_cls_act_is_valid_access(int off, int size,
8029 enum bpf_access_type type,
8030 const struct bpf_prog *prog,
8031 struct bpf_insn_access_aux *info)
8033 if (type == BPF_WRITE) {
8035 case bpf_ctx_range(struct __sk_buff, mark):
8036 case bpf_ctx_range(struct __sk_buff, tc_index):
8037 case bpf_ctx_range(struct __sk_buff, priority):
8038 case bpf_ctx_range(struct __sk_buff, tc_classid):
8039 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8040 case bpf_ctx_range(struct __sk_buff, tstamp):
8041 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8049 case bpf_ctx_range(struct __sk_buff, data):
8050 info->reg_type = PTR_TO_PACKET;
8052 case bpf_ctx_range(struct __sk_buff, data_meta):
8053 info->reg_type = PTR_TO_PACKET_META;
8055 case bpf_ctx_range(struct __sk_buff, data_end):
8056 info->reg_type = PTR_TO_PACKET_END;
8058 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8062 return bpf_skb_is_valid_access(off, size, type, prog, info);
8065 static bool __is_valid_xdp_access(int off, int size)
8067 if (off < 0 || off >= sizeof(struct xdp_md))
8069 if (off % size != 0)
8071 if (size != sizeof(__u32))
8077 static bool xdp_is_valid_access(int off, int size,
8078 enum bpf_access_type type,
8079 const struct bpf_prog *prog,
8080 struct bpf_insn_access_aux *info)
8082 if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8084 case offsetof(struct xdp_md, egress_ifindex):
8089 if (type == BPF_WRITE) {
8090 if (bpf_prog_is_dev_bound(prog->aux)) {
8092 case offsetof(struct xdp_md, rx_queue_index):
8093 return __is_valid_xdp_access(off, size);
8100 case offsetof(struct xdp_md, data):
8101 info->reg_type = PTR_TO_PACKET;
8103 case offsetof(struct xdp_md, data_meta):
8104 info->reg_type = PTR_TO_PACKET_META;
8106 case offsetof(struct xdp_md, data_end):
8107 info->reg_type = PTR_TO_PACKET_END;
8111 return __is_valid_xdp_access(off, size);
8114 void bpf_warn_invalid_xdp_action(u32 act)
8116 const u32 act_max = XDP_REDIRECT;
8118 WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
8119 act > act_max ? "Illegal" : "Driver unsupported",
8122 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
8124 static bool sock_addr_is_valid_access(int off, int size,
8125 enum bpf_access_type type,
8126 const struct bpf_prog *prog,
8127 struct bpf_insn_access_aux *info)
8129 const int size_default = sizeof(__u32);
8131 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
8133 if (off % size != 0)
8136 /* Disallow access to IPv6 fields from IPv4 contex and vise
8140 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8141 switch (prog->expected_attach_type) {
8142 case BPF_CGROUP_INET4_BIND:
8143 case BPF_CGROUP_INET4_CONNECT:
8144 case BPF_CGROUP_INET4_GETPEERNAME:
8145 case BPF_CGROUP_INET4_GETSOCKNAME:
8146 case BPF_CGROUP_UDP4_SENDMSG:
8147 case BPF_CGROUP_UDP4_RECVMSG:
8153 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8154 switch (prog->expected_attach_type) {
8155 case BPF_CGROUP_INET6_BIND:
8156 case BPF_CGROUP_INET6_CONNECT:
8157 case BPF_CGROUP_INET6_GETPEERNAME:
8158 case BPF_CGROUP_INET6_GETSOCKNAME:
8159 case BPF_CGROUP_UDP6_SENDMSG:
8160 case BPF_CGROUP_UDP6_RECVMSG:
8166 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8167 switch (prog->expected_attach_type) {
8168 case BPF_CGROUP_UDP4_SENDMSG:
8174 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8176 switch (prog->expected_attach_type) {
8177 case BPF_CGROUP_UDP6_SENDMSG:
8186 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8187 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8188 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8189 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8191 case bpf_ctx_range(struct bpf_sock_addr, user_port):
8192 if (type == BPF_READ) {
8193 bpf_ctx_record_field_size(info, size_default);
8195 if (bpf_ctx_wide_access_ok(off, size,
8196 struct bpf_sock_addr,
8200 if (bpf_ctx_wide_access_ok(off, size,
8201 struct bpf_sock_addr,
8205 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8208 if (bpf_ctx_wide_access_ok(off, size,
8209 struct bpf_sock_addr,
8213 if (bpf_ctx_wide_access_ok(off, size,
8214 struct bpf_sock_addr,
8218 if (size != size_default)
8222 case offsetof(struct bpf_sock_addr, sk):
8223 if (type != BPF_READ)
8225 if (size != sizeof(__u64))
8227 info->reg_type = PTR_TO_SOCKET;
8230 if (type == BPF_READ) {
8231 if (size != size_default)
8241 static bool sock_ops_is_valid_access(int off, int size,
8242 enum bpf_access_type type,
8243 const struct bpf_prog *prog,
8244 struct bpf_insn_access_aux *info)
8246 const int size_default = sizeof(__u32);
8248 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
8251 /* The verifier guarantees that size > 0. */
8252 if (off % size != 0)
8255 if (type == BPF_WRITE) {
8257 case offsetof(struct bpf_sock_ops, reply):
8258 case offsetof(struct bpf_sock_ops, sk_txhash):
8259 if (size != size_default)
8267 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
8269 if (size != sizeof(__u64))
8272 case offsetof(struct bpf_sock_ops, sk):
8273 if (size != sizeof(__u64))
8275 info->reg_type = PTR_TO_SOCKET_OR_NULL;
8277 case offsetof(struct bpf_sock_ops, skb_data):
8278 if (size != sizeof(__u64))
8280 info->reg_type = PTR_TO_PACKET;
8282 case offsetof(struct bpf_sock_ops, skb_data_end):
8283 if (size != sizeof(__u64))
8285 info->reg_type = PTR_TO_PACKET_END;
8287 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
8288 bpf_ctx_record_field_size(info, size_default);
8289 return bpf_ctx_narrow_access_ok(off, size,
8292 if (size != size_default)
8301 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
8302 const struct bpf_prog *prog)
8304 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
8307 static bool sk_skb_is_valid_access(int off, int size,
8308 enum bpf_access_type type,
8309 const struct bpf_prog *prog,
8310 struct bpf_insn_access_aux *info)
8313 case bpf_ctx_range(struct __sk_buff, tc_classid):
8314 case bpf_ctx_range(struct __sk_buff, data_meta):
8315 case bpf_ctx_range(struct __sk_buff, tstamp):
8316 case bpf_ctx_range(struct __sk_buff, wire_len):
8320 if (type == BPF_WRITE) {
8322 case bpf_ctx_range(struct __sk_buff, tc_index):
8323 case bpf_ctx_range(struct __sk_buff, priority):
8331 case bpf_ctx_range(struct __sk_buff, mark):
8333 case bpf_ctx_range(struct __sk_buff, data):
8334 info->reg_type = PTR_TO_PACKET;
8336 case bpf_ctx_range(struct __sk_buff, data_end):
8337 info->reg_type = PTR_TO_PACKET_END;
8341 return bpf_skb_is_valid_access(off, size, type, prog, info);
8344 static bool sk_msg_is_valid_access(int off, int size,
8345 enum bpf_access_type type,
8346 const struct bpf_prog *prog,
8347 struct bpf_insn_access_aux *info)
8349 if (type == BPF_WRITE)
8352 if (off % size != 0)
8356 case offsetof(struct sk_msg_md, data):
8357 info->reg_type = PTR_TO_PACKET;
8358 if (size != sizeof(__u64))
8361 case offsetof(struct sk_msg_md, data_end):
8362 info->reg_type = PTR_TO_PACKET_END;
8363 if (size != sizeof(__u64))
8366 case offsetof(struct sk_msg_md, sk):
8367 if (size != sizeof(__u64))
8369 info->reg_type = PTR_TO_SOCKET;
8371 case bpf_ctx_range(struct sk_msg_md, family):
8372 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
8373 case bpf_ctx_range(struct sk_msg_md, local_ip4):
8374 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
8375 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
8376 case bpf_ctx_range(struct sk_msg_md, remote_port):
8377 case bpf_ctx_range(struct sk_msg_md, local_port):
8378 case bpf_ctx_range(struct sk_msg_md, size):
8379 if (size != sizeof(__u32))
8388 static bool flow_dissector_is_valid_access(int off, int size,
8389 enum bpf_access_type type,
8390 const struct bpf_prog *prog,
8391 struct bpf_insn_access_aux *info)
8393 const int size_default = sizeof(__u32);
8395 if (off < 0 || off >= sizeof(struct __sk_buff))
8398 if (type == BPF_WRITE)
8402 case bpf_ctx_range(struct __sk_buff, data):
8403 if (size != size_default)
8405 info->reg_type = PTR_TO_PACKET;
8407 case bpf_ctx_range(struct __sk_buff, data_end):
8408 if (size != size_default)
8410 info->reg_type = PTR_TO_PACKET_END;
8412 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8413 if (size != sizeof(__u64))
8415 info->reg_type = PTR_TO_FLOW_KEYS;
8422 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
8423 const struct bpf_insn *si,
8424 struct bpf_insn *insn_buf,
8425 struct bpf_prog *prog,
8429 struct bpf_insn *insn = insn_buf;
8432 case offsetof(struct __sk_buff, data):
8433 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
8434 si->dst_reg, si->src_reg,
8435 offsetof(struct bpf_flow_dissector, data));
8438 case offsetof(struct __sk_buff, data_end):
8439 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
8440 si->dst_reg, si->src_reg,
8441 offsetof(struct bpf_flow_dissector, data_end));
8444 case offsetof(struct __sk_buff, flow_keys):
8445 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
8446 si->dst_reg, si->src_reg,
8447 offsetof(struct bpf_flow_dissector, flow_keys));
8451 return insn - insn_buf;
8454 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si,
8455 struct bpf_insn *insn)
8457 /* si->dst_reg = skb_shinfo(SKB); */
8458 #ifdef NET_SKBUFF_DATA_USES_OFFSET
8459 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8460 BPF_REG_AX, si->src_reg,
8461 offsetof(struct sk_buff, end));
8462 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
8463 si->dst_reg, si->src_reg,
8464 offsetof(struct sk_buff, head));
8465 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
8467 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8468 si->dst_reg, si->src_reg,
8469 offsetof(struct sk_buff, end));
8475 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
8476 const struct bpf_insn *si,
8477 struct bpf_insn *insn_buf,
8478 struct bpf_prog *prog, u32 *target_size)
8480 struct bpf_insn *insn = insn_buf;
8484 case offsetof(struct __sk_buff, len):
8485 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8486 bpf_target_off(struct sk_buff, len, 4,
8490 case offsetof(struct __sk_buff, protocol):
8491 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8492 bpf_target_off(struct sk_buff, protocol, 2,
8496 case offsetof(struct __sk_buff, vlan_proto):
8497 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8498 bpf_target_off(struct sk_buff, vlan_proto, 2,
8502 case offsetof(struct __sk_buff, priority):
8503 if (type == BPF_WRITE)
8504 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8505 bpf_target_off(struct sk_buff, priority, 4,
8508 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8509 bpf_target_off(struct sk_buff, priority, 4,
8513 case offsetof(struct __sk_buff, ingress_ifindex):
8514 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8515 bpf_target_off(struct sk_buff, skb_iif, 4,
8519 case offsetof(struct __sk_buff, ifindex):
8520 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8521 si->dst_reg, si->src_reg,
8522 offsetof(struct sk_buff, dev));
8523 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
8524 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8525 bpf_target_off(struct net_device, ifindex, 4,
8529 case offsetof(struct __sk_buff, hash):
8530 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8531 bpf_target_off(struct sk_buff, hash, 4,
8535 case offsetof(struct __sk_buff, mark):
8536 if (type == BPF_WRITE)
8537 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8538 bpf_target_off(struct sk_buff, mark, 4,
8541 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8542 bpf_target_off(struct sk_buff, mark, 4,
8546 case offsetof(struct __sk_buff, pkt_type):
8548 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8550 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
8551 #ifdef __BIG_ENDIAN_BITFIELD
8552 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
8556 case offsetof(struct __sk_buff, queue_mapping):
8557 if (type == BPF_WRITE) {
8558 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
8559 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8560 bpf_target_off(struct sk_buff,
8564 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8565 bpf_target_off(struct sk_buff,
8571 case offsetof(struct __sk_buff, vlan_present):
8573 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8574 PKT_VLAN_PRESENT_OFFSET());
8575 if (PKT_VLAN_PRESENT_BIT)
8576 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
8577 if (PKT_VLAN_PRESENT_BIT < 7)
8578 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
8581 case offsetof(struct __sk_buff, vlan_tci):
8582 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8583 bpf_target_off(struct sk_buff, vlan_tci, 2,
8587 case offsetof(struct __sk_buff, cb[0]) ...
8588 offsetofend(struct __sk_buff, cb[4]) - 1:
8589 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
8590 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
8591 offsetof(struct qdisc_skb_cb, data)) %
8594 prog->cb_access = 1;
8596 off -= offsetof(struct __sk_buff, cb[0]);
8597 off += offsetof(struct sk_buff, cb);
8598 off += offsetof(struct qdisc_skb_cb, data);
8599 if (type == BPF_WRITE)
8600 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
8603 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
8607 case offsetof(struct __sk_buff, tc_classid):
8608 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
8611 off -= offsetof(struct __sk_buff, tc_classid);
8612 off += offsetof(struct sk_buff, cb);
8613 off += offsetof(struct qdisc_skb_cb, tc_classid);
8615 if (type == BPF_WRITE)
8616 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
8619 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
8623 case offsetof(struct __sk_buff, data):
8624 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
8625 si->dst_reg, si->src_reg,
8626 offsetof(struct sk_buff, data));
8629 case offsetof(struct __sk_buff, data_meta):
8631 off -= offsetof(struct __sk_buff, data_meta);
8632 off += offsetof(struct sk_buff, cb);
8633 off += offsetof(struct bpf_skb_data_end, data_meta);
8634 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8638 case offsetof(struct __sk_buff, data_end):
8640 off -= offsetof(struct __sk_buff, data_end);
8641 off += offsetof(struct sk_buff, cb);
8642 off += offsetof(struct bpf_skb_data_end, data_end);
8643 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8647 case offsetof(struct __sk_buff, tc_index):
8648 #ifdef CONFIG_NET_SCHED
8649 if (type == BPF_WRITE)
8650 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8651 bpf_target_off(struct sk_buff, tc_index, 2,
8654 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8655 bpf_target_off(struct sk_buff, tc_index, 2,
8659 if (type == BPF_WRITE)
8660 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
8662 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8666 case offsetof(struct __sk_buff, napi_id):
8667 #if defined(CONFIG_NET_RX_BUSY_POLL)
8668 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8669 bpf_target_off(struct sk_buff, napi_id, 4,
8671 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
8672 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8675 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8678 case offsetof(struct __sk_buff, family):
8679 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
8681 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8682 si->dst_reg, si->src_reg,
8683 offsetof(struct sk_buff, sk));
8684 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8685 bpf_target_off(struct sock_common,
8689 case offsetof(struct __sk_buff, remote_ip4):
8690 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
8692 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8693 si->dst_reg, si->src_reg,
8694 offsetof(struct sk_buff, sk));
8695 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8696 bpf_target_off(struct sock_common,
8700 case offsetof(struct __sk_buff, local_ip4):
8701 BUILD_BUG_ON(sizeof_field(struct sock_common,
8702 skc_rcv_saddr) != 4);
8704 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8705 si->dst_reg, si->src_reg,
8706 offsetof(struct sk_buff, sk));
8707 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8708 bpf_target_off(struct sock_common,
8712 case offsetof(struct __sk_buff, remote_ip6[0]) ...
8713 offsetof(struct __sk_buff, remote_ip6[3]):
8714 #if IS_ENABLED(CONFIG_IPV6)
8715 BUILD_BUG_ON(sizeof_field(struct sock_common,
8716 skc_v6_daddr.s6_addr32[0]) != 4);
8719 off -= offsetof(struct __sk_buff, remote_ip6[0]);
8721 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8722 si->dst_reg, si->src_reg,
8723 offsetof(struct sk_buff, sk));
8724 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8725 offsetof(struct sock_common,
8726 skc_v6_daddr.s6_addr32[0]) +
8729 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8732 case offsetof(struct __sk_buff, local_ip6[0]) ...
8733 offsetof(struct __sk_buff, local_ip6[3]):
8734 #if IS_ENABLED(CONFIG_IPV6)
8735 BUILD_BUG_ON(sizeof_field(struct sock_common,
8736 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8739 off -= offsetof(struct __sk_buff, local_ip6[0]);
8741 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8742 si->dst_reg, si->src_reg,
8743 offsetof(struct sk_buff, sk));
8744 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8745 offsetof(struct sock_common,
8746 skc_v6_rcv_saddr.s6_addr32[0]) +
8749 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8753 case offsetof(struct __sk_buff, remote_port):
8754 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
8756 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8757 si->dst_reg, si->src_reg,
8758 offsetof(struct sk_buff, sk));
8759 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8760 bpf_target_off(struct sock_common,
8763 #ifndef __BIG_ENDIAN_BITFIELD
8764 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8768 case offsetof(struct __sk_buff, local_port):
8769 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
8771 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8772 si->dst_reg, si->src_reg,
8773 offsetof(struct sk_buff, sk));
8774 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8775 bpf_target_off(struct sock_common,
8776 skc_num, 2, target_size));
8779 case offsetof(struct __sk_buff, tstamp):
8780 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
8782 if (type == BPF_WRITE)
8783 *insn++ = BPF_STX_MEM(BPF_DW,
8784 si->dst_reg, si->src_reg,
8785 bpf_target_off(struct sk_buff,
8789 *insn++ = BPF_LDX_MEM(BPF_DW,
8790 si->dst_reg, si->src_reg,
8791 bpf_target_off(struct sk_buff,
8796 case offsetof(struct __sk_buff, gso_segs):
8797 insn = bpf_convert_shinfo_access(si, insn);
8798 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
8799 si->dst_reg, si->dst_reg,
8800 bpf_target_off(struct skb_shared_info,
8804 case offsetof(struct __sk_buff, gso_size):
8805 insn = bpf_convert_shinfo_access(si, insn);
8806 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
8807 si->dst_reg, si->dst_reg,
8808 bpf_target_off(struct skb_shared_info,
8812 case offsetof(struct __sk_buff, wire_len):
8813 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
8816 off -= offsetof(struct __sk_buff, wire_len);
8817 off += offsetof(struct sk_buff, cb);
8818 off += offsetof(struct qdisc_skb_cb, pkt_len);
8820 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
8823 case offsetof(struct __sk_buff, sk):
8824 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8825 si->dst_reg, si->src_reg,
8826 offsetof(struct sk_buff, sk));
8830 return insn - insn_buf;
8833 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
8834 const struct bpf_insn *si,
8835 struct bpf_insn *insn_buf,
8836 struct bpf_prog *prog, u32 *target_size)
8838 struct bpf_insn *insn = insn_buf;
8842 case offsetof(struct bpf_sock, bound_dev_if):
8843 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
8845 if (type == BPF_WRITE)
8846 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8847 offsetof(struct sock, sk_bound_dev_if));
8849 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8850 offsetof(struct sock, sk_bound_dev_if));
8853 case offsetof(struct bpf_sock, mark):
8854 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
8856 if (type == BPF_WRITE)
8857 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8858 offsetof(struct sock, sk_mark));
8860 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8861 offsetof(struct sock, sk_mark));
8864 case offsetof(struct bpf_sock, priority):
8865 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
8867 if (type == BPF_WRITE)
8868 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8869 offsetof(struct sock, sk_priority));
8871 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8872 offsetof(struct sock, sk_priority));
8875 case offsetof(struct bpf_sock, family):
8876 *insn++ = BPF_LDX_MEM(
8877 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
8878 si->dst_reg, si->src_reg,
8879 bpf_target_off(struct sock_common,
8881 sizeof_field(struct sock_common,
8886 case offsetof(struct bpf_sock, type):
8887 *insn++ = BPF_LDX_MEM(
8888 BPF_FIELD_SIZEOF(struct sock, sk_type),
8889 si->dst_reg, si->src_reg,
8890 bpf_target_off(struct sock, sk_type,
8891 sizeof_field(struct sock, sk_type),
8895 case offsetof(struct bpf_sock, protocol):
8896 *insn++ = BPF_LDX_MEM(
8897 BPF_FIELD_SIZEOF(struct sock, sk_protocol),
8898 si->dst_reg, si->src_reg,
8899 bpf_target_off(struct sock, sk_protocol,
8900 sizeof_field(struct sock, sk_protocol),
8904 case offsetof(struct bpf_sock, src_ip4):
8905 *insn++ = BPF_LDX_MEM(
8906 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8907 bpf_target_off(struct sock_common, skc_rcv_saddr,
8908 sizeof_field(struct sock_common,
8913 case offsetof(struct bpf_sock, dst_ip4):
8914 *insn++ = BPF_LDX_MEM(
8915 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8916 bpf_target_off(struct sock_common, skc_daddr,
8917 sizeof_field(struct sock_common,
8922 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8923 #if IS_ENABLED(CONFIG_IPV6)
8925 off -= offsetof(struct bpf_sock, src_ip6[0]);
8926 *insn++ = BPF_LDX_MEM(
8927 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8930 skc_v6_rcv_saddr.s6_addr32[0],
8931 sizeof_field(struct sock_common,
8932 skc_v6_rcv_saddr.s6_addr32[0]),
8933 target_size) + off);
8936 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8940 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8941 #if IS_ENABLED(CONFIG_IPV6)
8943 off -= offsetof(struct bpf_sock, dst_ip6[0]);
8944 *insn++ = BPF_LDX_MEM(
8945 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8946 bpf_target_off(struct sock_common,
8947 skc_v6_daddr.s6_addr32[0],
8948 sizeof_field(struct sock_common,
8949 skc_v6_daddr.s6_addr32[0]),
8950 target_size) + off);
8952 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8957 case offsetof(struct bpf_sock, src_port):
8958 *insn++ = BPF_LDX_MEM(
8959 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
8960 si->dst_reg, si->src_reg,
8961 bpf_target_off(struct sock_common, skc_num,
8962 sizeof_field(struct sock_common,
8967 case offsetof(struct bpf_sock, dst_port):
8968 *insn++ = BPF_LDX_MEM(
8969 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
8970 si->dst_reg, si->src_reg,
8971 bpf_target_off(struct sock_common, skc_dport,
8972 sizeof_field(struct sock_common,
8977 case offsetof(struct bpf_sock, state):
8978 *insn++ = BPF_LDX_MEM(
8979 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
8980 si->dst_reg, si->src_reg,
8981 bpf_target_off(struct sock_common, skc_state,
8982 sizeof_field(struct sock_common,
8986 case offsetof(struct bpf_sock, rx_queue_mapping):
8987 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
8988 *insn++ = BPF_LDX_MEM(
8989 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
8990 si->dst_reg, si->src_reg,
8991 bpf_target_off(struct sock, sk_rx_queue_mapping,
8992 sizeof_field(struct sock,
8993 sk_rx_queue_mapping),
8995 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
8997 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
8999 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9005 return insn - insn_buf;
9008 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
9009 const struct bpf_insn *si,
9010 struct bpf_insn *insn_buf,
9011 struct bpf_prog *prog, u32 *target_size)
9013 struct bpf_insn *insn = insn_buf;
9016 case offsetof(struct __sk_buff, ifindex):
9017 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9018 si->dst_reg, si->src_reg,
9019 offsetof(struct sk_buff, dev));
9020 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9021 bpf_target_off(struct net_device, ifindex, 4,
9025 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9029 return insn - insn_buf;
9032 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
9033 const struct bpf_insn *si,
9034 struct bpf_insn *insn_buf,
9035 struct bpf_prog *prog, u32 *target_size)
9037 struct bpf_insn *insn = insn_buf;
9040 case offsetof(struct xdp_md, data):
9041 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
9042 si->dst_reg, si->src_reg,
9043 offsetof(struct xdp_buff, data));
9045 case offsetof(struct xdp_md, data_meta):
9046 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
9047 si->dst_reg, si->src_reg,
9048 offsetof(struct xdp_buff, data_meta));
9050 case offsetof(struct xdp_md, data_end):
9051 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
9052 si->dst_reg, si->src_reg,
9053 offsetof(struct xdp_buff, data_end));
9055 case offsetof(struct xdp_md, ingress_ifindex):
9056 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9057 si->dst_reg, si->src_reg,
9058 offsetof(struct xdp_buff, rxq));
9059 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
9060 si->dst_reg, si->dst_reg,
9061 offsetof(struct xdp_rxq_info, dev));
9062 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9063 offsetof(struct net_device, ifindex));
9065 case offsetof(struct xdp_md, rx_queue_index):
9066 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9067 si->dst_reg, si->src_reg,
9068 offsetof(struct xdp_buff, rxq));
9069 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9070 offsetof(struct xdp_rxq_info,
9073 case offsetof(struct xdp_md, egress_ifindex):
9074 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
9075 si->dst_reg, si->src_reg,
9076 offsetof(struct xdp_buff, txq));
9077 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
9078 si->dst_reg, si->dst_reg,
9079 offsetof(struct xdp_txq_info, dev));
9080 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9081 offsetof(struct net_device, ifindex));
9085 return insn - insn_buf;
9088 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
9089 * context Structure, F is Field in context structure that contains a pointer
9090 * to Nested Structure of type NS that has the field NF.
9092 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
9093 * sure that SIZE is not greater than actual size of S.F.NF.
9095 * If offset OFF is provided, the load happens from that offset relative to
9098 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
9100 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
9101 si->src_reg, offsetof(S, F)); \
9102 *insn++ = BPF_LDX_MEM( \
9103 SIZE, si->dst_reg, si->dst_reg, \
9104 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
9109 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
9110 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
9111 BPF_FIELD_SIZEOF(NS, NF), 0)
9113 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
9114 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
9116 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
9117 * "register" since two registers available in convert_ctx_access are not
9118 * enough: we can't override neither SRC, since it contains value to store, nor
9119 * DST since it contains pointer to context that may be used by later
9120 * instructions. But we need a temporary place to save pointer to nested
9121 * structure whose field we want to store to.
9123 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
9125 int tmp_reg = BPF_REG_9; \
9126 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
9128 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
9130 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
9132 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
9133 si->dst_reg, offsetof(S, F)); \
9134 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \
9135 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
9138 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
9142 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
9145 if (type == BPF_WRITE) { \
9146 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
9149 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
9150 S, NS, F, NF, SIZE, OFF); \
9154 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
9155 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
9156 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
9158 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
9159 const struct bpf_insn *si,
9160 struct bpf_insn *insn_buf,
9161 struct bpf_prog *prog, u32 *target_size)
9163 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
9164 struct bpf_insn *insn = insn_buf;
9167 case offsetof(struct bpf_sock_addr, user_family):
9168 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9169 struct sockaddr, uaddr, sa_family);
9172 case offsetof(struct bpf_sock_addr, user_ip4):
9173 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9174 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
9175 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
9178 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9180 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
9181 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9182 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9183 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
9187 case offsetof(struct bpf_sock_addr, user_port):
9188 /* To get port we need to know sa_family first and then treat
9189 * sockaddr as either sockaddr_in or sockaddr_in6.
9190 * Though we can simplify since port field has same offset and
9191 * size in both structures.
9192 * Here we check this invariant and use just one of the
9193 * structures if it's true.
9195 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
9196 offsetof(struct sockaddr_in6, sin6_port));
9197 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
9198 sizeof_field(struct sockaddr_in6, sin6_port));
9199 /* Account for sin6_port being smaller than user_port. */
9200 port_size = min(port_size, BPF_LDST_BYTES(si));
9201 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9202 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9203 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
9206 case offsetof(struct bpf_sock_addr, family):
9207 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9208 struct sock, sk, sk_family);
9211 case offsetof(struct bpf_sock_addr, type):
9212 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9213 struct sock, sk, sk_type);
9216 case offsetof(struct bpf_sock_addr, protocol):
9217 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9218 struct sock, sk, sk_protocol);
9221 case offsetof(struct bpf_sock_addr, msg_src_ip4):
9222 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
9223 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9224 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
9225 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
9228 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9231 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
9232 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
9233 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9234 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
9235 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
9237 case offsetof(struct bpf_sock_addr, sk):
9238 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
9239 si->dst_reg, si->src_reg,
9240 offsetof(struct bpf_sock_addr_kern, sk));
9244 return insn - insn_buf;
9247 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
9248 const struct bpf_insn *si,
9249 struct bpf_insn *insn_buf,
9250 struct bpf_prog *prog,
9253 struct bpf_insn *insn = insn_buf;
9256 /* Helper macro for adding read access to tcp_sock or sock fields. */
9257 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
9259 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \
9260 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
9261 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
9262 if (si->dst_reg == reg || si->src_reg == reg) \
9264 if (si->dst_reg == reg || si->src_reg == reg) \
9266 if (si->dst_reg == si->src_reg) { \
9267 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
9268 offsetof(struct bpf_sock_ops_kern, \
9270 fullsock_reg = reg; \
9273 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9274 struct bpf_sock_ops_kern, \
9276 fullsock_reg, si->src_reg, \
9277 offsetof(struct bpf_sock_ops_kern, \
9279 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
9280 if (si->dst_reg == si->src_reg) \
9281 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9282 offsetof(struct bpf_sock_ops_kern, \
9284 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9285 struct bpf_sock_ops_kern, sk),\
9286 si->dst_reg, si->src_reg, \
9287 offsetof(struct bpf_sock_ops_kern, sk));\
9288 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
9290 si->dst_reg, si->dst_reg, \
9291 offsetof(OBJ, OBJ_FIELD)); \
9292 if (si->dst_reg == si->src_reg) { \
9293 *insn++ = BPF_JMP_A(1); \
9294 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9295 offsetof(struct bpf_sock_ops_kern, \
9300 #define SOCK_OPS_GET_SK() \
9302 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \
9303 if (si->dst_reg == reg || si->src_reg == reg) \
9305 if (si->dst_reg == reg || si->src_reg == reg) \
9307 if (si->dst_reg == si->src_reg) { \
9308 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
9309 offsetof(struct bpf_sock_ops_kern, \
9311 fullsock_reg = reg; \
9314 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9315 struct bpf_sock_ops_kern, \
9317 fullsock_reg, si->src_reg, \
9318 offsetof(struct bpf_sock_ops_kern, \
9320 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
9321 if (si->dst_reg == si->src_reg) \
9322 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9323 offsetof(struct bpf_sock_ops_kern, \
9325 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9326 struct bpf_sock_ops_kern, sk),\
9327 si->dst_reg, si->src_reg, \
9328 offsetof(struct bpf_sock_ops_kern, sk));\
9329 if (si->dst_reg == si->src_reg) { \
9330 *insn++ = BPF_JMP_A(1); \
9331 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9332 offsetof(struct bpf_sock_ops_kern, \
9337 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
9338 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
9340 /* Helper macro for adding write access to tcp_sock or sock fields.
9341 * The macro is called with two registers, dst_reg which contains a pointer
9342 * to ctx (context) and src_reg which contains the value that should be
9343 * stored. However, we need an additional register since we cannot overwrite
9344 * dst_reg because it may be used later in the program.
9345 * Instead we "borrow" one of the other register. We first save its value
9346 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
9347 * it at the end of the macro.
9349 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
9351 int reg = BPF_REG_9; \
9352 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
9353 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
9354 if (si->dst_reg == reg || si->src_reg == reg) \
9356 if (si->dst_reg == reg || si->src_reg == reg) \
9358 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
9359 offsetof(struct bpf_sock_ops_kern, \
9361 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9362 struct bpf_sock_ops_kern, \
9365 offsetof(struct bpf_sock_ops_kern, \
9367 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
9368 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9369 struct bpf_sock_ops_kern, sk),\
9371 offsetof(struct bpf_sock_ops_kern, sk));\
9372 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \
9374 offsetof(OBJ, OBJ_FIELD)); \
9375 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
9376 offsetof(struct bpf_sock_ops_kern, \
9380 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
9382 if (TYPE == BPF_WRITE) \
9383 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
9385 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
9388 if (insn > insn_buf)
9389 return insn - insn_buf;
9392 case offsetof(struct bpf_sock_ops, op):
9393 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9395 si->dst_reg, si->src_reg,
9396 offsetof(struct bpf_sock_ops_kern, op));
9399 case offsetof(struct bpf_sock_ops, replylong[0]) ...
9400 offsetof(struct bpf_sock_ops, replylong[3]):
9401 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
9402 sizeof_field(struct bpf_sock_ops_kern, reply));
9403 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
9404 sizeof_field(struct bpf_sock_ops_kern, replylong));
9406 off -= offsetof(struct bpf_sock_ops, replylong[0]);
9407 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
9408 if (type == BPF_WRITE)
9409 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9412 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9416 case offsetof(struct bpf_sock_ops, family):
9417 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9419 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9420 struct bpf_sock_ops_kern, sk),
9421 si->dst_reg, si->src_reg,
9422 offsetof(struct bpf_sock_ops_kern, sk));
9423 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9424 offsetof(struct sock_common, skc_family));
9427 case offsetof(struct bpf_sock_ops, remote_ip4):
9428 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9430 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9431 struct bpf_sock_ops_kern, sk),
9432 si->dst_reg, si->src_reg,
9433 offsetof(struct bpf_sock_ops_kern, sk));
9434 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9435 offsetof(struct sock_common, skc_daddr));
9438 case offsetof(struct bpf_sock_ops, local_ip4):
9439 BUILD_BUG_ON(sizeof_field(struct sock_common,
9440 skc_rcv_saddr) != 4);
9442 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9443 struct bpf_sock_ops_kern, sk),
9444 si->dst_reg, si->src_reg,
9445 offsetof(struct bpf_sock_ops_kern, sk));
9446 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9447 offsetof(struct sock_common,
9451 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
9452 offsetof(struct bpf_sock_ops, remote_ip6[3]):
9453 #if IS_ENABLED(CONFIG_IPV6)
9454 BUILD_BUG_ON(sizeof_field(struct sock_common,
9455 skc_v6_daddr.s6_addr32[0]) != 4);
9458 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
9459 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9460 struct bpf_sock_ops_kern, sk),
9461 si->dst_reg, si->src_reg,
9462 offsetof(struct bpf_sock_ops_kern, sk));
9463 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9464 offsetof(struct sock_common,
9465 skc_v6_daddr.s6_addr32[0]) +
9468 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9472 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
9473 offsetof(struct bpf_sock_ops, local_ip6[3]):
9474 #if IS_ENABLED(CONFIG_IPV6)
9475 BUILD_BUG_ON(sizeof_field(struct sock_common,
9476 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9479 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
9480 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9481 struct bpf_sock_ops_kern, sk),
9482 si->dst_reg, si->src_reg,
9483 offsetof(struct bpf_sock_ops_kern, sk));
9484 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9485 offsetof(struct sock_common,
9486 skc_v6_rcv_saddr.s6_addr32[0]) +
9489 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9493 case offsetof(struct bpf_sock_ops, remote_port):
9494 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9496 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9497 struct bpf_sock_ops_kern, sk),
9498 si->dst_reg, si->src_reg,
9499 offsetof(struct bpf_sock_ops_kern, sk));
9500 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9501 offsetof(struct sock_common, skc_dport));
9502 #ifndef __BIG_ENDIAN_BITFIELD
9503 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9507 case offsetof(struct bpf_sock_ops, local_port):
9508 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9510 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9511 struct bpf_sock_ops_kern, sk),
9512 si->dst_reg, si->src_reg,
9513 offsetof(struct bpf_sock_ops_kern, sk));
9514 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9515 offsetof(struct sock_common, skc_num));
9518 case offsetof(struct bpf_sock_ops, is_fullsock):
9519 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9520 struct bpf_sock_ops_kern,
9522 si->dst_reg, si->src_reg,
9523 offsetof(struct bpf_sock_ops_kern,
9527 case offsetof(struct bpf_sock_ops, state):
9528 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
9530 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9531 struct bpf_sock_ops_kern, sk),
9532 si->dst_reg, si->src_reg,
9533 offsetof(struct bpf_sock_ops_kern, sk));
9534 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
9535 offsetof(struct sock_common, skc_state));
9538 case offsetof(struct bpf_sock_ops, rtt_min):
9539 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
9540 sizeof(struct minmax));
9541 BUILD_BUG_ON(sizeof(struct minmax) <
9542 sizeof(struct minmax_sample));
9544 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9545 struct bpf_sock_ops_kern, sk),
9546 si->dst_reg, si->src_reg,
9547 offsetof(struct bpf_sock_ops_kern, sk));
9548 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9549 offsetof(struct tcp_sock, rtt_min) +
9550 sizeof_field(struct minmax_sample, t));
9553 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
9554 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
9558 case offsetof(struct bpf_sock_ops, sk_txhash):
9559 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
9562 case offsetof(struct bpf_sock_ops, snd_cwnd):
9563 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
9565 case offsetof(struct bpf_sock_ops, srtt_us):
9566 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
9568 case offsetof(struct bpf_sock_ops, snd_ssthresh):
9569 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
9571 case offsetof(struct bpf_sock_ops, rcv_nxt):
9572 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
9574 case offsetof(struct bpf_sock_ops, snd_nxt):
9575 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
9577 case offsetof(struct bpf_sock_ops, snd_una):
9578 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
9580 case offsetof(struct bpf_sock_ops, mss_cache):
9581 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
9583 case offsetof(struct bpf_sock_ops, ecn_flags):
9584 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
9586 case offsetof(struct bpf_sock_ops, rate_delivered):
9587 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
9589 case offsetof(struct bpf_sock_ops, rate_interval_us):
9590 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
9592 case offsetof(struct bpf_sock_ops, packets_out):
9593 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
9595 case offsetof(struct bpf_sock_ops, retrans_out):
9596 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
9598 case offsetof(struct bpf_sock_ops, total_retrans):
9599 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
9601 case offsetof(struct bpf_sock_ops, segs_in):
9602 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
9604 case offsetof(struct bpf_sock_ops, data_segs_in):
9605 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
9607 case offsetof(struct bpf_sock_ops, segs_out):
9608 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
9610 case offsetof(struct bpf_sock_ops, data_segs_out):
9611 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
9613 case offsetof(struct bpf_sock_ops, lost_out):
9614 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
9616 case offsetof(struct bpf_sock_ops, sacked_out):
9617 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
9619 case offsetof(struct bpf_sock_ops, bytes_received):
9620 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
9622 case offsetof(struct bpf_sock_ops, bytes_acked):
9623 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
9625 case offsetof(struct bpf_sock_ops, sk):
9628 case offsetof(struct bpf_sock_ops, skb_data_end):
9629 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9631 si->dst_reg, si->src_reg,
9632 offsetof(struct bpf_sock_ops_kern,
9635 case offsetof(struct bpf_sock_ops, skb_data):
9636 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9638 si->dst_reg, si->src_reg,
9639 offsetof(struct bpf_sock_ops_kern,
9641 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9642 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9643 si->dst_reg, si->dst_reg,
9644 offsetof(struct sk_buff, data));
9646 case offsetof(struct bpf_sock_ops, skb_len):
9647 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9649 si->dst_reg, si->src_reg,
9650 offsetof(struct bpf_sock_ops_kern,
9652 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9653 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9654 si->dst_reg, si->dst_reg,
9655 offsetof(struct sk_buff, len));
9657 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9658 off = offsetof(struct sk_buff, cb);
9659 off += offsetof(struct tcp_skb_cb, tcp_flags);
9660 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
9661 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9663 si->dst_reg, si->src_reg,
9664 offsetof(struct bpf_sock_ops_kern,
9666 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9667 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
9669 si->dst_reg, si->dst_reg, off);
9672 return insn - insn_buf;
9675 /* data_end = skb->data + skb_headlen() */
9676 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
9677 struct bpf_insn *insn)
9679 /* si->dst_reg = skb->data */
9680 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9681 si->dst_reg, si->src_reg,
9682 offsetof(struct sk_buff, data));
9684 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9685 BPF_REG_AX, si->src_reg,
9686 offsetof(struct sk_buff, len));
9687 /* si->dst_reg = skb->data + skb->len */
9688 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
9689 /* AX = skb->data_len */
9690 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
9691 BPF_REG_AX, si->src_reg,
9692 offsetof(struct sk_buff, data_len));
9693 /* si->dst_reg = skb->data + skb->len - skb->data_len */
9694 *insn++ = BPF_ALU64_REG(BPF_SUB, si->dst_reg, BPF_REG_AX);
9699 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
9700 const struct bpf_insn *si,
9701 struct bpf_insn *insn_buf,
9702 struct bpf_prog *prog, u32 *target_size)
9704 struct bpf_insn *insn = insn_buf;
9707 case offsetof(struct __sk_buff, data_end):
9708 insn = bpf_convert_data_end_access(si, insn);
9711 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9715 return insn - insn_buf;
9718 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
9719 const struct bpf_insn *si,
9720 struct bpf_insn *insn_buf,
9721 struct bpf_prog *prog, u32 *target_size)
9723 struct bpf_insn *insn = insn_buf;
9724 #if IS_ENABLED(CONFIG_IPV6)
9728 /* convert ctx uses the fact sg element is first in struct */
9729 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
9732 case offsetof(struct sk_msg_md, data):
9733 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
9734 si->dst_reg, si->src_reg,
9735 offsetof(struct sk_msg, data));
9737 case offsetof(struct sk_msg_md, data_end):
9738 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
9739 si->dst_reg, si->src_reg,
9740 offsetof(struct sk_msg, data_end));
9742 case offsetof(struct sk_msg_md, family):
9743 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9745 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9747 si->dst_reg, si->src_reg,
9748 offsetof(struct sk_msg, sk));
9749 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9750 offsetof(struct sock_common, skc_family));
9753 case offsetof(struct sk_msg_md, remote_ip4):
9754 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9756 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9758 si->dst_reg, si->src_reg,
9759 offsetof(struct sk_msg, sk));
9760 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9761 offsetof(struct sock_common, skc_daddr));
9764 case offsetof(struct sk_msg_md, local_ip4):
9765 BUILD_BUG_ON(sizeof_field(struct sock_common,
9766 skc_rcv_saddr) != 4);
9768 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9770 si->dst_reg, si->src_reg,
9771 offsetof(struct sk_msg, sk));
9772 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9773 offsetof(struct sock_common,
9777 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
9778 offsetof(struct sk_msg_md, remote_ip6[3]):
9779 #if IS_ENABLED(CONFIG_IPV6)
9780 BUILD_BUG_ON(sizeof_field(struct sock_common,
9781 skc_v6_daddr.s6_addr32[0]) != 4);
9784 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
9785 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9787 si->dst_reg, si->src_reg,
9788 offsetof(struct sk_msg, sk));
9789 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9790 offsetof(struct sock_common,
9791 skc_v6_daddr.s6_addr32[0]) +
9794 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9798 case offsetof(struct sk_msg_md, local_ip6[0]) ...
9799 offsetof(struct sk_msg_md, local_ip6[3]):
9800 #if IS_ENABLED(CONFIG_IPV6)
9801 BUILD_BUG_ON(sizeof_field(struct sock_common,
9802 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9805 off -= offsetof(struct sk_msg_md, local_ip6[0]);
9806 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9808 si->dst_reg, si->src_reg,
9809 offsetof(struct sk_msg, sk));
9810 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9811 offsetof(struct sock_common,
9812 skc_v6_rcv_saddr.s6_addr32[0]) +
9815 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9819 case offsetof(struct sk_msg_md, remote_port):
9820 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9822 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9824 si->dst_reg, si->src_reg,
9825 offsetof(struct sk_msg, sk));
9826 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9827 offsetof(struct sock_common, skc_dport));
9828 #ifndef __BIG_ENDIAN_BITFIELD
9829 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9833 case offsetof(struct sk_msg_md, local_port):
9834 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9836 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9838 si->dst_reg, si->src_reg,
9839 offsetof(struct sk_msg, sk));
9840 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9841 offsetof(struct sock_common, skc_num));
9844 case offsetof(struct sk_msg_md, size):
9845 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
9846 si->dst_reg, si->src_reg,
9847 offsetof(struct sk_msg_sg, size));
9850 case offsetof(struct sk_msg_md, sk):
9851 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
9852 si->dst_reg, si->src_reg,
9853 offsetof(struct sk_msg, sk));
9857 return insn - insn_buf;
9860 const struct bpf_verifier_ops sk_filter_verifier_ops = {
9861 .get_func_proto = sk_filter_func_proto,
9862 .is_valid_access = sk_filter_is_valid_access,
9863 .convert_ctx_access = bpf_convert_ctx_access,
9864 .gen_ld_abs = bpf_gen_ld_abs,
9867 const struct bpf_prog_ops sk_filter_prog_ops = {
9868 .test_run = bpf_prog_test_run_skb,
9871 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
9872 .get_func_proto = tc_cls_act_func_proto,
9873 .is_valid_access = tc_cls_act_is_valid_access,
9874 .convert_ctx_access = tc_cls_act_convert_ctx_access,
9875 .gen_prologue = tc_cls_act_prologue,
9876 .gen_ld_abs = bpf_gen_ld_abs,
9877 .check_kfunc_call = bpf_prog_test_check_kfunc_call,
9880 const struct bpf_prog_ops tc_cls_act_prog_ops = {
9881 .test_run = bpf_prog_test_run_skb,
9884 const struct bpf_verifier_ops xdp_verifier_ops = {
9885 .get_func_proto = xdp_func_proto,
9886 .is_valid_access = xdp_is_valid_access,
9887 .convert_ctx_access = xdp_convert_ctx_access,
9888 .gen_prologue = bpf_noop_prologue,
9891 const struct bpf_prog_ops xdp_prog_ops = {
9892 .test_run = bpf_prog_test_run_xdp,
9895 const struct bpf_verifier_ops cg_skb_verifier_ops = {
9896 .get_func_proto = cg_skb_func_proto,
9897 .is_valid_access = cg_skb_is_valid_access,
9898 .convert_ctx_access = bpf_convert_ctx_access,
9901 const struct bpf_prog_ops cg_skb_prog_ops = {
9902 .test_run = bpf_prog_test_run_skb,
9905 const struct bpf_verifier_ops lwt_in_verifier_ops = {
9906 .get_func_proto = lwt_in_func_proto,
9907 .is_valid_access = lwt_is_valid_access,
9908 .convert_ctx_access = bpf_convert_ctx_access,
9911 const struct bpf_prog_ops lwt_in_prog_ops = {
9912 .test_run = bpf_prog_test_run_skb,
9915 const struct bpf_verifier_ops lwt_out_verifier_ops = {
9916 .get_func_proto = lwt_out_func_proto,
9917 .is_valid_access = lwt_is_valid_access,
9918 .convert_ctx_access = bpf_convert_ctx_access,
9921 const struct bpf_prog_ops lwt_out_prog_ops = {
9922 .test_run = bpf_prog_test_run_skb,
9925 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
9926 .get_func_proto = lwt_xmit_func_proto,
9927 .is_valid_access = lwt_is_valid_access,
9928 .convert_ctx_access = bpf_convert_ctx_access,
9929 .gen_prologue = tc_cls_act_prologue,
9932 const struct bpf_prog_ops lwt_xmit_prog_ops = {
9933 .test_run = bpf_prog_test_run_skb,
9936 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
9937 .get_func_proto = lwt_seg6local_func_proto,
9938 .is_valid_access = lwt_is_valid_access,
9939 .convert_ctx_access = bpf_convert_ctx_access,
9942 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
9943 .test_run = bpf_prog_test_run_skb,
9946 const struct bpf_verifier_ops cg_sock_verifier_ops = {
9947 .get_func_proto = sock_filter_func_proto,
9948 .is_valid_access = sock_filter_is_valid_access,
9949 .convert_ctx_access = bpf_sock_convert_ctx_access,
9952 const struct bpf_prog_ops cg_sock_prog_ops = {
9955 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
9956 .get_func_proto = sock_addr_func_proto,
9957 .is_valid_access = sock_addr_is_valid_access,
9958 .convert_ctx_access = sock_addr_convert_ctx_access,
9961 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
9964 const struct bpf_verifier_ops sock_ops_verifier_ops = {
9965 .get_func_proto = sock_ops_func_proto,
9966 .is_valid_access = sock_ops_is_valid_access,
9967 .convert_ctx_access = sock_ops_convert_ctx_access,
9970 const struct bpf_prog_ops sock_ops_prog_ops = {
9973 const struct bpf_verifier_ops sk_skb_verifier_ops = {
9974 .get_func_proto = sk_skb_func_proto,
9975 .is_valid_access = sk_skb_is_valid_access,
9976 .convert_ctx_access = sk_skb_convert_ctx_access,
9977 .gen_prologue = sk_skb_prologue,
9980 const struct bpf_prog_ops sk_skb_prog_ops = {
9983 const struct bpf_verifier_ops sk_msg_verifier_ops = {
9984 .get_func_proto = sk_msg_func_proto,
9985 .is_valid_access = sk_msg_is_valid_access,
9986 .convert_ctx_access = sk_msg_convert_ctx_access,
9987 .gen_prologue = bpf_noop_prologue,
9990 const struct bpf_prog_ops sk_msg_prog_ops = {
9993 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
9994 .get_func_proto = flow_dissector_func_proto,
9995 .is_valid_access = flow_dissector_is_valid_access,
9996 .convert_ctx_access = flow_dissector_convert_ctx_access,
9999 const struct bpf_prog_ops flow_dissector_prog_ops = {
10000 .test_run = bpf_prog_test_run_flow_dissector,
10003 int sk_detach_filter(struct sock *sk)
10006 struct sk_filter *filter;
10008 if (sock_flag(sk, SOCK_FILTER_LOCKED))
10011 filter = rcu_dereference_protected(sk->sk_filter,
10012 lockdep_sock_is_held(sk));
10014 RCU_INIT_POINTER(sk->sk_filter, NULL);
10015 sk_filter_uncharge(sk, filter);
10021 EXPORT_SYMBOL_GPL(sk_detach_filter);
10023 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
10026 struct sock_fprog_kern *fprog;
10027 struct sk_filter *filter;
10031 filter = rcu_dereference_protected(sk->sk_filter,
10032 lockdep_sock_is_held(sk));
10036 /* We're copying the filter that has been originally attached,
10037 * so no conversion/decode needed anymore. eBPF programs that
10038 * have no original program cannot be dumped through this.
10041 fprog = filter->prog->orig_prog;
10047 /* User space only enquires number of filter blocks. */
10051 if (len < fprog->len)
10055 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
10058 /* Instead of bytes, the API requests to return the number
10059 * of filter blocks.
10068 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
10069 struct sock_reuseport *reuse,
10070 struct sock *sk, struct sk_buff *skb,
10071 struct sock *migrating_sk,
10074 reuse_kern->skb = skb;
10075 reuse_kern->sk = sk;
10076 reuse_kern->selected_sk = NULL;
10077 reuse_kern->migrating_sk = migrating_sk;
10078 reuse_kern->data_end = skb->data + skb_headlen(skb);
10079 reuse_kern->hash = hash;
10080 reuse_kern->reuseport_id = reuse->reuseport_id;
10081 reuse_kern->bind_inany = reuse->bind_inany;
10084 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
10085 struct bpf_prog *prog, struct sk_buff *skb,
10086 struct sock *migrating_sk,
10089 struct sk_reuseport_kern reuse_kern;
10090 enum sk_action action;
10092 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
10093 action = BPF_PROG_RUN(prog, &reuse_kern);
10095 if (action == SK_PASS)
10096 return reuse_kern.selected_sk;
10098 return ERR_PTR(-ECONNREFUSED);
10101 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
10102 struct bpf_map *, map, void *, key, u32, flags)
10104 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
10105 struct sock_reuseport *reuse;
10106 struct sock *selected_sk;
10108 selected_sk = map->ops->map_lookup_elem(map, key);
10112 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
10114 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
10115 if (sk_is_refcounted(selected_sk))
10116 sock_put(selected_sk);
10118 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
10119 * The only (!reuse) case here is - the sk has already been
10120 * unhashed (e.g. by close()), so treat it as -ENOENT.
10122 * Other maps (e.g. sock_map) do not provide this guarantee and
10123 * the sk may never be in the reuseport group to begin with.
10125 return is_sockarray ? -ENOENT : -EINVAL;
10128 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
10129 struct sock *sk = reuse_kern->sk;
10131 if (sk->sk_protocol != selected_sk->sk_protocol)
10132 return -EPROTOTYPE;
10133 else if (sk->sk_family != selected_sk->sk_family)
10134 return -EAFNOSUPPORT;
10136 /* Catch all. Likely bound to a different sockaddr. */
10140 reuse_kern->selected_sk = selected_sk;
10145 static const struct bpf_func_proto sk_select_reuseport_proto = {
10146 .func = sk_select_reuseport,
10148 .ret_type = RET_INTEGER,
10149 .arg1_type = ARG_PTR_TO_CTX,
10150 .arg2_type = ARG_CONST_MAP_PTR,
10151 .arg3_type = ARG_PTR_TO_MAP_KEY,
10152 .arg4_type = ARG_ANYTHING,
10155 BPF_CALL_4(sk_reuseport_load_bytes,
10156 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10157 void *, to, u32, len)
10159 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
10162 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
10163 .func = sk_reuseport_load_bytes,
10165 .ret_type = RET_INTEGER,
10166 .arg1_type = ARG_PTR_TO_CTX,
10167 .arg2_type = ARG_ANYTHING,
10168 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
10169 .arg4_type = ARG_CONST_SIZE,
10172 BPF_CALL_5(sk_reuseport_load_bytes_relative,
10173 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10174 void *, to, u32, len, u32, start_header)
10176 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
10177 len, start_header);
10180 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
10181 .func = sk_reuseport_load_bytes_relative,
10183 .ret_type = RET_INTEGER,
10184 .arg1_type = ARG_PTR_TO_CTX,
10185 .arg2_type = ARG_ANYTHING,
10186 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
10187 .arg4_type = ARG_CONST_SIZE,
10188 .arg5_type = ARG_ANYTHING,
10191 static const struct bpf_func_proto *
10192 sk_reuseport_func_proto(enum bpf_func_id func_id,
10193 const struct bpf_prog *prog)
10196 case BPF_FUNC_sk_select_reuseport:
10197 return &sk_select_reuseport_proto;
10198 case BPF_FUNC_skb_load_bytes:
10199 return &sk_reuseport_load_bytes_proto;
10200 case BPF_FUNC_skb_load_bytes_relative:
10201 return &sk_reuseport_load_bytes_relative_proto;
10202 case BPF_FUNC_get_socket_cookie:
10203 return &bpf_get_socket_ptr_cookie_proto;
10205 return bpf_base_func_proto(func_id);
10210 sk_reuseport_is_valid_access(int off, int size,
10211 enum bpf_access_type type,
10212 const struct bpf_prog *prog,
10213 struct bpf_insn_access_aux *info)
10215 const u32 size_default = sizeof(__u32);
10217 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
10218 off % size || type != BPF_READ)
10222 case offsetof(struct sk_reuseport_md, data):
10223 info->reg_type = PTR_TO_PACKET;
10224 return size == sizeof(__u64);
10226 case offsetof(struct sk_reuseport_md, data_end):
10227 info->reg_type = PTR_TO_PACKET_END;
10228 return size == sizeof(__u64);
10230 case offsetof(struct sk_reuseport_md, hash):
10231 return size == size_default;
10233 case offsetof(struct sk_reuseport_md, sk):
10234 info->reg_type = PTR_TO_SOCKET;
10235 return size == sizeof(__u64);
10237 case offsetof(struct sk_reuseport_md, migrating_sk):
10238 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
10239 return size == sizeof(__u64);
10241 /* Fields that allow narrowing */
10242 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
10243 if (size < sizeof_field(struct sk_buff, protocol))
10246 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
10247 case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
10248 case bpf_ctx_range(struct sk_reuseport_md, len):
10249 bpf_ctx_record_field_size(info, size_default);
10250 return bpf_ctx_narrow_access_ok(off, size, size_default);
10257 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
10258 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
10259 si->dst_reg, si->src_reg, \
10260 bpf_target_off(struct sk_reuseport_kern, F, \
10261 sizeof_field(struct sk_reuseport_kern, F), \
10265 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
10266 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
10271 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \
10272 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
10277 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
10278 const struct bpf_insn *si,
10279 struct bpf_insn *insn_buf,
10280 struct bpf_prog *prog,
10283 struct bpf_insn *insn = insn_buf;
10286 case offsetof(struct sk_reuseport_md, data):
10287 SK_REUSEPORT_LOAD_SKB_FIELD(data);
10290 case offsetof(struct sk_reuseport_md, len):
10291 SK_REUSEPORT_LOAD_SKB_FIELD(len);
10294 case offsetof(struct sk_reuseport_md, eth_protocol):
10295 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
10298 case offsetof(struct sk_reuseport_md, ip_protocol):
10299 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
10302 case offsetof(struct sk_reuseport_md, data_end):
10303 SK_REUSEPORT_LOAD_FIELD(data_end);
10306 case offsetof(struct sk_reuseport_md, hash):
10307 SK_REUSEPORT_LOAD_FIELD(hash);
10310 case offsetof(struct sk_reuseport_md, bind_inany):
10311 SK_REUSEPORT_LOAD_FIELD(bind_inany);
10314 case offsetof(struct sk_reuseport_md, sk):
10315 SK_REUSEPORT_LOAD_FIELD(sk);
10318 case offsetof(struct sk_reuseport_md, migrating_sk):
10319 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
10323 return insn - insn_buf;
10326 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
10327 .get_func_proto = sk_reuseport_func_proto,
10328 .is_valid_access = sk_reuseport_is_valid_access,
10329 .convert_ctx_access = sk_reuseport_convert_ctx_access,
10332 const struct bpf_prog_ops sk_reuseport_prog_ops = {
10335 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
10336 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
10338 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
10339 struct sock *, sk, u64, flags)
10341 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
10342 BPF_SK_LOOKUP_F_NO_REUSEPORT)))
10344 if (unlikely(sk && sk_is_refcounted(sk)))
10345 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
10346 if (unlikely(sk && sk->sk_state == TCP_ESTABLISHED))
10347 return -ESOCKTNOSUPPORT; /* reject connected sockets */
10349 /* Check if socket is suitable for packet L3/L4 protocol */
10350 if (sk && sk->sk_protocol != ctx->protocol)
10351 return -EPROTOTYPE;
10352 if (sk && sk->sk_family != ctx->family &&
10353 (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
10354 return -EAFNOSUPPORT;
10356 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
10359 /* Select socket as lookup result */
10360 ctx->selected_sk = sk;
10361 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
10365 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
10366 .func = bpf_sk_lookup_assign,
10368 .ret_type = RET_INTEGER,
10369 .arg1_type = ARG_PTR_TO_CTX,
10370 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL,
10371 .arg3_type = ARG_ANYTHING,
10374 static const struct bpf_func_proto *
10375 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
10378 case BPF_FUNC_perf_event_output:
10379 return &bpf_event_output_data_proto;
10380 case BPF_FUNC_sk_assign:
10381 return &bpf_sk_lookup_assign_proto;
10382 case BPF_FUNC_sk_release:
10383 return &bpf_sk_release_proto;
10385 return bpf_sk_base_func_proto(func_id);
10389 static bool sk_lookup_is_valid_access(int off, int size,
10390 enum bpf_access_type type,
10391 const struct bpf_prog *prog,
10392 struct bpf_insn_access_aux *info)
10394 if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
10396 if (off % size != 0)
10398 if (type != BPF_READ)
10402 case offsetof(struct bpf_sk_lookup, sk):
10403 info->reg_type = PTR_TO_SOCKET_OR_NULL;
10404 return size == sizeof(__u64);
10406 case bpf_ctx_range(struct bpf_sk_lookup, family):
10407 case bpf_ctx_range(struct bpf_sk_lookup, protocol):
10408 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
10409 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
10410 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
10411 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
10412 case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
10413 case bpf_ctx_range(struct bpf_sk_lookup, local_port):
10414 bpf_ctx_record_field_size(info, sizeof(__u32));
10415 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
10422 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
10423 const struct bpf_insn *si,
10424 struct bpf_insn *insn_buf,
10425 struct bpf_prog *prog,
10428 struct bpf_insn *insn = insn_buf;
10431 case offsetof(struct bpf_sk_lookup, sk):
10432 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10433 offsetof(struct bpf_sk_lookup_kern, selected_sk));
10436 case offsetof(struct bpf_sk_lookup, family):
10437 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10438 bpf_target_off(struct bpf_sk_lookup_kern,
10439 family, 2, target_size));
10442 case offsetof(struct bpf_sk_lookup, protocol):
10443 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10444 bpf_target_off(struct bpf_sk_lookup_kern,
10445 protocol, 2, target_size));
10448 case offsetof(struct bpf_sk_lookup, remote_ip4):
10449 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10450 bpf_target_off(struct bpf_sk_lookup_kern,
10451 v4.saddr, 4, target_size));
10454 case offsetof(struct bpf_sk_lookup, local_ip4):
10455 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10456 bpf_target_off(struct bpf_sk_lookup_kern,
10457 v4.daddr, 4, target_size));
10460 case bpf_ctx_range_till(struct bpf_sk_lookup,
10461 remote_ip6[0], remote_ip6[3]): {
10462 #if IS_ENABLED(CONFIG_IPV6)
10465 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
10466 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10467 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10468 offsetof(struct bpf_sk_lookup_kern, v6.saddr));
10469 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10470 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10472 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10476 case bpf_ctx_range_till(struct bpf_sk_lookup,
10477 local_ip6[0], local_ip6[3]): {
10478 #if IS_ENABLED(CONFIG_IPV6)
10481 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
10482 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10483 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10484 offsetof(struct bpf_sk_lookup_kern, v6.daddr));
10485 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10486 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10488 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10492 case offsetof(struct bpf_sk_lookup, remote_port):
10493 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10494 bpf_target_off(struct bpf_sk_lookup_kern,
10495 sport, 2, target_size));
10498 case offsetof(struct bpf_sk_lookup, local_port):
10499 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10500 bpf_target_off(struct bpf_sk_lookup_kern,
10501 dport, 2, target_size));
10505 return insn - insn_buf;
10508 const struct bpf_prog_ops sk_lookup_prog_ops = {
10509 .test_run = bpf_prog_test_run_sk_lookup,
10512 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
10513 .get_func_proto = sk_lookup_func_proto,
10514 .is_valid_access = sk_lookup_is_valid_access,
10515 .convert_ctx_access = sk_lookup_convert_ctx_access,
10518 #endif /* CONFIG_INET */
10520 DEFINE_BPF_DISPATCHER(xdp)
10522 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
10524 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
10527 #ifdef CONFIG_DEBUG_INFO_BTF
10528 BTF_ID_LIST_GLOBAL(btf_sock_ids)
10529 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
10531 #undef BTF_SOCK_TYPE
10533 u32 btf_sock_ids[MAX_BTF_SOCK_TYPE];
10536 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
10538 /* tcp6_sock type is not generated in dwarf and hence btf,
10539 * trigger an explicit type generation here.
10541 BTF_TYPE_EMIT(struct tcp6_sock);
10542 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
10543 sk->sk_family == AF_INET6)
10544 return (unsigned long)sk;
10546 return (unsigned long)NULL;
10549 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
10550 .func = bpf_skc_to_tcp6_sock,
10552 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10553 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10554 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
10557 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
10559 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
10560 return (unsigned long)sk;
10562 return (unsigned long)NULL;
10565 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
10566 .func = bpf_skc_to_tcp_sock,
10568 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10569 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10570 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
10573 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
10575 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
10576 * generated if CONFIG_INET=n. Trigger an explicit generation here.
10578 BTF_TYPE_EMIT(struct inet_timewait_sock);
10579 BTF_TYPE_EMIT(struct tcp_timewait_sock);
10582 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
10583 return (unsigned long)sk;
10586 #if IS_BUILTIN(CONFIG_IPV6)
10587 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
10588 return (unsigned long)sk;
10591 return (unsigned long)NULL;
10594 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
10595 .func = bpf_skc_to_tcp_timewait_sock,
10597 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10598 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10599 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
10602 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
10605 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10606 return (unsigned long)sk;
10609 #if IS_BUILTIN(CONFIG_IPV6)
10610 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10611 return (unsigned long)sk;
10614 return (unsigned long)NULL;
10617 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
10618 .func = bpf_skc_to_tcp_request_sock,
10620 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10621 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10622 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
10625 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
10627 /* udp6_sock type is not generated in dwarf and hence btf,
10628 * trigger an explicit type generation here.
10630 BTF_TYPE_EMIT(struct udp6_sock);
10631 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
10632 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
10633 return (unsigned long)sk;
10635 return (unsigned long)NULL;
10638 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
10639 .func = bpf_skc_to_udp6_sock,
10641 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10642 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10643 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
10646 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
10648 return (unsigned long)sock_from_file(file);
10651 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
10652 BTF_ID(struct, socket)
10653 BTF_ID(struct, file)
10655 const struct bpf_func_proto bpf_sock_from_file_proto = {
10656 .func = bpf_sock_from_file,
10658 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10659 .ret_btf_id = &bpf_sock_from_file_btf_ids[0],
10660 .arg1_type = ARG_PTR_TO_BTF_ID,
10661 .arg1_btf_id = &bpf_sock_from_file_btf_ids[1],
10664 static const struct bpf_func_proto *
10665 bpf_sk_base_func_proto(enum bpf_func_id func_id)
10667 const struct bpf_func_proto *func;
10670 case BPF_FUNC_skc_to_tcp6_sock:
10671 func = &bpf_skc_to_tcp6_sock_proto;
10673 case BPF_FUNC_skc_to_tcp_sock:
10674 func = &bpf_skc_to_tcp_sock_proto;
10676 case BPF_FUNC_skc_to_tcp_timewait_sock:
10677 func = &bpf_skc_to_tcp_timewait_sock_proto;
10679 case BPF_FUNC_skc_to_tcp_request_sock:
10680 func = &bpf_skc_to_tcp_request_sock_proto;
10682 case BPF_FUNC_skc_to_udp6_sock:
10683 func = &bpf_skc_to_udp6_sock_proto;
10686 return bpf_base_func_proto(func_id);
10689 if (!perfmon_capable())