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 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
3937 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key);
3939 u32 xdp_master_redirect(struct xdp_buff *xdp)
3941 struct net_device *master, *slave;
3942 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3944 master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev);
3945 slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp);
3946 if (slave && slave != xdp->rxq->dev) {
3947 /* The target device is different from the receiving device, so
3948 * redirect it to the new device.
3949 * Using XDP_REDIRECT gets the correct behaviour from XDP enabled
3950 * drivers to unmap the packet from their rx ring.
3952 ri->tgt_index = slave->ifindex;
3953 ri->map_id = INT_MAX;
3954 ri->map_type = BPF_MAP_TYPE_UNSPEC;
3955 return XDP_REDIRECT;
3959 EXPORT_SYMBOL_GPL(xdp_master_redirect);
3961 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3962 struct bpf_prog *xdp_prog)
3964 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3965 enum bpf_map_type map_type = ri->map_type;
3966 void *fwd = ri->tgt_value;
3967 u32 map_id = ri->map_id;
3968 struct bpf_map *map;
3971 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
3972 ri->map_type = BPF_MAP_TYPE_UNSPEC;
3975 case BPF_MAP_TYPE_DEVMAP:
3977 case BPF_MAP_TYPE_DEVMAP_HASH:
3978 map = READ_ONCE(ri->map);
3979 if (unlikely(map)) {
3980 WRITE_ONCE(ri->map, NULL);
3981 err = dev_map_enqueue_multi(xdp, dev, map,
3982 ri->flags & BPF_F_EXCLUDE_INGRESS);
3984 err = dev_map_enqueue(fwd, xdp, dev);
3987 case BPF_MAP_TYPE_CPUMAP:
3988 err = cpu_map_enqueue(fwd, xdp, dev);
3990 case BPF_MAP_TYPE_XSKMAP:
3991 err = __xsk_map_redirect(fwd, xdp);
3993 case BPF_MAP_TYPE_UNSPEC:
3994 if (map_id == INT_MAX) {
3995 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
3996 if (unlikely(!fwd)) {
4000 err = dev_xdp_enqueue(fwd, xdp, dev);
4011 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4014 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4017 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4019 static int xdp_do_generic_redirect_map(struct net_device *dev,
4020 struct sk_buff *skb,
4021 struct xdp_buff *xdp,
4022 struct bpf_prog *xdp_prog,
4024 enum bpf_map_type map_type, u32 map_id)
4026 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4027 struct bpf_map *map;
4031 case BPF_MAP_TYPE_DEVMAP:
4033 case BPF_MAP_TYPE_DEVMAP_HASH:
4034 map = READ_ONCE(ri->map);
4035 if (unlikely(map)) {
4036 WRITE_ONCE(ri->map, NULL);
4037 err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4038 ri->flags & BPF_F_EXCLUDE_INGRESS);
4040 err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4045 case BPF_MAP_TYPE_XSKMAP:
4046 err = xsk_generic_rcv(fwd, xdp);
4051 case BPF_MAP_TYPE_CPUMAP:
4052 err = cpu_map_generic_redirect(fwd, skb);
4061 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4064 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4068 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4069 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4071 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4072 enum bpf_map_type map_type = ri->map_type;
4073 void *fwd = ri->tgt_value;
4074 u32 map_id = ri->map_id;
4077 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4078 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4080 if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4081 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4082 if (unlikely(!fwd)) {
4087 err = xdp_ok_fwd_dev(fwd, skb->len);
4092 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4093 generic_xdp_tx(skb, xdp_prog);
4097 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id);
4099 _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4103 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4105 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4107 if (unlikely(flags))
4110 /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4111 * by map_idr) is used for ifindex based XDP redirect.
4113 ri->tgt_index = ifindex;
4114 ri->map_id = INT_MAX;
4115 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4117 return XDP_REDIRECT;
4120 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4121 .func = bpf_xdp_redirect,
4123 .ret_type = RET_INTEGER,
4124 .arg1_type = ARG_ANYTHING,
4125 .arg2_type = ARG_ANYTHING,
4128 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
4131 return map->ops->map_redirect(map, ifindex, flags);
4134 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4135 .func = bpf_xdp_redirect_map,
4137 .ret_type = RET_INTEGER,
4138 .arg1_type = ARG_CONST_MAP_PTR,
4139 .arg2_type = ARG_ANYTHING,
4140 .arg3_type = ARG_ANYTHING,
4143 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4144 unsigned long off, unsigned long len)
4146 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4150 if (ptr != dst_buff)
4151 memcpy(dst_buff, ptr, len);
4156 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4157 u64, flags, void *, meta, u64, meta_size)
4159 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4161 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4163 if (unlikely(!skb || skb_size > skb->len))
4166 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4170 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4171 .func = bpf_skb_event_output,
4173 .ret_type = RET_INTEGER,
4174 .arg1_type = ARG_PTR_TO_CTX,
4175 .arg2_type = ARG_CONST_MAP_PTR,
4176 .arg3_type = ARG_ANYTHING,
4177 .arg4_type = ARG_PTR_TO_MEM,
4178 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4181 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4183 const struct bpf_func_proto bpf_skb_output_proto = {
4184 .func = bpf_skb_event_output,
4186 .ret_type = RET_INTEGER,
4187 .arg1_type = ARG_PTR_TO_BTF_ID,
4188 .arg1_btf_id = &bpf_skb_output_btf_ids[0],
4189 .arg2_type = ARG_CONST_MAP_PTR,
4190 .arg3_type = ARG_ANYTHING,
4191 .arg4_type = ARG_PTR_TO_MEM,
4192 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4195 static unsigned short bpf_tunnel_key_af(u64 flags)
4197 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4200 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4201 u32, size, u64, flags)
4203 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4204 u8 compat[sizeof(struct bpf_tunnel_key)];
4208 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
4212 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4216 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4219 case offsetof(struct bpf_tunnel_key, tunnel_label):
4220 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4222 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4223 /* Fixup deprecated structure layouts here, so we have
4224 * a common path later on.
4226 if (ip_tunnel_info_af(info) != AF_INET)
4229 to = (struct bpf_tunnel_key *)compat;
4236 to->tunnel_id = be64_to_cpu(info->key.tun_id);
4237 to->tunnel_tos = info->key.tos;
4238 to->tunnel_ttl = info->key.ttl;
4241 if (flags & BPF_F_TUNINFO_IPV6) {
4242 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4243 sizeof(to->remote_ipv6));
4244 to->tunnel_label = be32_to_cpu(info->key.label);
4246 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4247 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4248 to->tunnel_label = 0;
4251 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4252 memcpy(to_orig, to, size);
4256 memset(to_orig, 0, size);
4260 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4261 .func = bpf_skb_get_tunnel_key,
4263 .ret_type = RET_INTEGER,
4264 .arg1_type = ARG_PTR_TO_CTX,
4265 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4266 .arg3_type = ARG_CONST_SIZE,
4267 .arg4_type = ARG_ANYTHING,
4270 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4272 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4275 if (unlikely(!info ||
4276 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4280 if (unlikely(size < info->options_len)) {
4285 ip_tunnel_info_opts_get(to, info);
4286 if (size > info->options_len)
4287 memset(to + info->options_len, 0, size - info->options_len);
4289 return info->options_len;
4291 memset(to, 0, size);
4295 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4296 .func = bpf_skb_get_tunnel_opt,
4298 .ret_type = RET_INTEGER,
4299 .arg1_type = ARG_PTR_TO_CTX,
4300 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4301 .arg3_type = ARG_CONST_SIZE,
4304 static struct metadata_dst __percpu *md_dst;
4306 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4307 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4309 struct metadata_dst *md = this_cpu_ptr(md_dst);
4310 u8 compat[sizeof(struct bpf_tunnel_key)];
4311 struct ip_tunnel_info *info;
4313 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4314 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
4316 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4318 case offsetof(struct bpf_tunnel_key, tunnel_label):
4319 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4320 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4321 /* Fixup deprecated structure layouts here, so we have
4322 * a common path later on.
4324 memcpy(compat, from, size);
4325 memset(compat + size, 0, sizeof(compat) - size);
4326 from = (const struct bpf_tunnel_key *) compat;
4332 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4337 dst_hold((struct dst_entry *) md);
4338 skb_dst_set(skb, (struct dst_entry *) md);
4340 info = &md->u.tun_info;
4341 memset(info, 0, sizeof(*info));
4342 info->mode = IP_TUNNEL_INFO_TX;
4344 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4345 if (flags & BPF_F_DONT_FRAGMENT)
4346 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4347 if (flags & BPF_F_ZERO_CSUM_TX)
4348 info->key.tun_flags &= ~TUNNEL_CSUM;
4349 if (flags & BPF_F_SEQ_NUMBER)
4350 info->key.tun_flags |= TUNNEL_SEQ;
4352 info->key.tun_id = cpu_to_be64(from->tunnel_id);
4353 info->key.tos = from->tunnel_tos;
4354 info->key.ttl = from->tunnel_ttl;
4356 if (flags & BPF_F_TUNINFO_IPV6) {
4357 info->mode |= IP_TUNNEL_INFO_IPV6;
4358 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4359 sizeof(from->remote_ipv6));
4360 info->key.label = cpu_to_be32(from->tunnel_label) &
4361 IPV6_FLOWLABEL_MASK;
4363 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4369 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4370 .func = bpf_skb_set_tunnel_key,
4372 .ret_type = RET_INTEGER,
4373 .arg1_type = ARG_PTR_TO_CTX,
4374 .arg2_type = ARG_PTR_TO_MEM,
4375 .arg3_type = ARG_CONST_SIZE,
4376 .arg4_type = ARG_ANYTHING,
4379 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4380 const u8 *, from, u32, size)
4382 struct ip_tunnel_info *info = skb_tunnel_info(skb);
4383 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4385 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4387 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4390 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4395 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4396 .func = bpf_skb_set_tunnel_opt,
4398 .ret_type = RET_INTEGER,
4399 .arg1_type = ARG_PTR_TO_CTX,
4400 .arg2_type = ARG_PTR_TO_MEM,
4401 .arg3_type = ARG_CONST_SIZE,
4404 static const struct bpf_func_proto *
4405 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4408 struct metadata_dst __percpu *tmp;
4410 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4415 if (cmpxchg(&md_dst, NULL, tmp))
4416 metadata_dst_free_percpu(tmp);
4420 case BPF_FUNC_skb_set_tunnel_key:
4421 return &bpf_skb_set_tunnel_key_proto;
4422 case BPF_FUNC_skb_set_tunnel_opt:
4423 return &bpf_skb_set_tunnel_opt_proto;
4429 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4432 struct bpf_array *array = container_of(map, struct bpf_array, map);
4433 struct cgroup *cgrp;
4436 sk = skb_to_full_sk(skb);
4437 if (!sk || !sk_fullsock(sk))
4439 if (unlikely(idx >= array->map.max_entries))
4442 cgrp = READ_ONCE(array->ptrs[idx]);
4443 if (unlikely(!cgrp))
4446 return sk_under_cgroup_hierarchy(sk, cgrp);
4449 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4450 .func = bpf_skb_under_cgroup,
4452 .ret_type = RET_INTEGER,
4453 .arg1_type = ARG_PTR_TO_CTX,
4454 .arg2_type = ARG_CONST_MAP_PTR,
4455 .arg3_type = ARG_ANYTHING,
4458 #ifdef CONFIG_SOCK_CGROUP_DATA
4459 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4461 struct cgroup *cgrp;
4463 sk = sk_to_full_sk(sk);
4464 if (!sk || !sk_fullsock(sk))
4467 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4468 return cgroup_id(cgrp);
4471 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4473 return __bpf_sk_cgroup_id(skb->sk);
4476 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4477 .func = bpf_skb_cgroup_id,
4479 .ret_type = RET_INTEGER,
4480 .arg1_type = ARG_PTR_TO_CTX,
4483 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4486 struct cgroup *ancestor;
4487 struct cgroup *cgrp;
4489 sk = sk_to_full_sk(sk);
4490 if (!sk || !sk_fullsock(sk))
4493 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4494 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4498 return cgroup_id(ancestor);
4501 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4504 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4507 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4508 .func = bpf_skb_ancestor_cgroup_id,
4510 .ret_type = RET_INTEGER,
4511 .arg1_type = ARG_PTR_TO_CTX,
4512 .arg2_type = ARG_ANYTHING,
4515 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4517 return __bpf_sk_cgroup_id(sk);
4520 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4521 .func = bpf_sk_cgroup_id,
4523 .ret_type = RET_INTEGER,
4524 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4527 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4529 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4532 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4533 .func = bpf_sk_ancestor_cgroup_id,
4535 .ret_type = RET_INTEGER,
4536 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4537 .arg2_type = ARG_ANYTHING,
4541 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4542 unsigned long off, unsigned long len)
4544 memcpy(dst_buff, src_buff + off, len);
4548 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4549 u64, flags, void *, meta, u64, meta_size)
4551 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4553 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4555 if (unlikely(!xdp ||
4556 xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4559 return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4560 xdp_size, bpf_xdp_copy);
4563 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4564 .func = bpf_xdp_event_output,
4566 .ret_type = RET_INTEGER,
4567 .arg1_type = ARG_PTR_TO_CTX,
4568 .arg2_type = ARG_CONST_MAP_PTR,
4569 .arg3_type = ARG_ANYTHING,
4570 .arg4_type = ARG_PTR_TO_MEM,
4571 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4574 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4576 const struct bpf_func_proto bpf_xdp_output_proto = {
4577 .func = bpf_xdp_event_output,
4579 .ret_type = RET_INTEGER,
4580 .arg1_type = ARG_PTR_TO_BTF_ID,
4581 .arg1_btf_id = &bpf_xdp_output_btf_ids[0],
4582 .arg2_type = ARG_CONST_MAP_PTR,
4583 .arg3_type = ARG_ANYTHING,
4584 .arg4_type = ARG_PTR_TO_MEM,
4585 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4588 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4590 return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4593 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4594 .func = bpf_get_socket_cookie,
4596 .ret_type = RET_INTEGER,
4597 .arg1_type = ARG_PTR_TO_CTX,
4600 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4602 return __sock_gen_cookie(ctx->sk);
4605 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4606 .func = bpf_get_socket_cookie_sock_addr,
4608 .ret_type = RET_INTEGER,
4609 .arg1_type = ARG_PTR_TO_CTX,
4612 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4614 return __sock_gen_cookie(ctx);
4617 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4618 .func = bpf_get_socket_cookie_sock,
4620 .ret_type = RET_INTEGER,
4621 .arg1_type = ARG_PTR_TO_CTX,
4624 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
4626 return sk ? sock_gen_cookie(sk) : 0;
4629 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
4630 .func = bpf_get_socket_ptr_cookie,
4632 .ret_type = RET_INTEGER,
4633 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4636 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4638 return __sock_gen_cookie(ctx->sk);
4641 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4642 .func = bpf_get_socket_cookie_sock_ops,
4644 .ret_type = RET_INTEGER,
4645 .arg1_type = ARG_PTR_TO_CTX,
4648 static u64 __bpf_get_netns_cookie(struct sock *sk)
4650 const struct net *net = sk ? sock_net(sk) : &init_net;
4652 return net->net_cookie;
4655 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
4657 return __bpf_get_netns_cookie(ctx);
4660 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
4661 .func = bpf_get_netns_cookie_sock,
4663 .ret_type = RET_INTEGER,
4664 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4667 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4669 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4672 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
4673 .func = bpf_get_netns_cookie_sock_addr,
4675 .ret_type = RET_INTEGER,
4676 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4679 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4681 struct sock *sk = sk_to_full_sk(skb->sk);
4684 if (!sk || !sk_fullsock(sk))
4686 kuid = sock_net_uid(sock_net(sk), sk);
4687 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4690 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4691 .func = bpf_get_socket_uid,
4693 .ret_type = RET_INTEGER,
4694 .arg1_type = ARG_PTR_TO_CTX,
4697 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
4698 char *optval, int optlen)
4700 char devname[IFNAMSIZ];
4706 if (!sk_fullsock(sk))
4709 sock_owned_by_me(sk);
4711 if (level == SOL_SOCKET) {
4712 if (optlen != sizeof(int) && optname != SO_BINDTODEVICE)
4714 val = *((int *)optval);
4715 valbool = val ? 1 : 0;
4717 /* Only some socketops are supported */
4720 val = min_t(u32, val, sysctl_rmem_max);
4721 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4722 WRITE_ONCE(sk->sk_rcvbuf,
4723 max_t(int, val * 2, SOCK_MIN_RCVBUF));
4726 val = min_t(u32, val, sysctl_wmem_max);
4727 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4728 WRITE_ONCE(sk->sk_sndbuf,
4729 max_t(int, val * 2, SOCK_MIN_SNDBUF));
4731 case SO_MAX_PACING_RATE: /* 32bit version */
4733 cmpxchg(&sk->sk_pacing_status,
4736 sk->sk_max_pacing_rate = (val == ~0U) ?
4737 ~0UL : (unsigned int)val;
4738 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4739 sk->sk_max_pacing_rate);
4742 sk->sk_priority = val;
4747 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
4750 if (sk->sk_mark != val) {
4755 case SO_BINDTODEVICE:
4756 optlen = min_t(long, optlen, IFNAMSIZ - 1);
4757 strncpy(devname, optval, optlen);
4758 devname[optlen] = 0;
4761 if (devname[0] != '\0') {
4762 struct net_device *dev;
4767 dev = dev_get_by_name(net, devname);
4770 ifindex = dev->ifindex;
4774 case SO_BINDTOIFINDEX:
4775 if (optname == SO_BINDTOIFINDEX)
4777 ret = sock_bindtoindex(sk, ifindex, false);
4780 if (sk->sk_prot->keepalive)
4781 sk->sk_prot->keepalive(sk, valbool);
4782 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
4785 sk->sk_reuseport = valbool;
4791 } else if (level == SOL_IP) {
4792 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4795 val = *((int *)optval);
4796 /* Only some options are supported */
4799 if (val < -1 || val > 0xff) {
4802 struct inet_sock *inet = inet_sk(sk);
4812 #if IS_ENABLED(CONFIG_IPV6)
4813 } else if (level == SOL_IPV6) {
4814 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4817 val = *((int *)optval);
4818 /* Only some options are supported */
4821 if (val < -1 || val > 0xff) {
4824 struct ipv6_pinfo *np = inet6_sk(sk);
4835 } else if (level == SOL_TCP &&
4836 sk->sk_prot->setsockopt == tcp_setsockopt) {
4837 if (optname == TCP_CONGESTION) {
4838 char name[TCP_CA_NAME_MAX];
4840 strncpy(name, optval, min_t(long, optlen,
4841 TCP_CA_NAME_MAX-1));
4842 name[TCP_CA_NAME_MAX-1] = 0;
4843 ret = tcp_set_congestion_control(sk, name, false, true);
4845 struct inet_connection_sock *icsk = inet_csk(sk);
4846 struct tcp_sock *tp = tcp_sk(sk);
4847 unsigned long timeout;
4849 if (optlen != sizeof(int))
4852 val = *((int *)optval);
4853 /* Only some options are supported */
4856 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4861 case TCP_BPF_SNDCWND_CLAMP:
4865 tp->snd_cwnd_clamp = val;
4866 tp->snd_ssthresh = val;
4869 case TCP_BPF_DELACK_MAX:
4870 timeout = usecs_to_jiffies(val);
4871 if (timeout > TCP_DELACK_MAX ||
4872 timeout < TCP_TIMEOUT_MIN)
4874 inet_csk(sk)->icsk_delack_max = timeout;
4876 case TCP_BPF_RTO_MIN:
4877 timeout = usecs_to_jiffies(val);
4878 if (timeout > TCP_RTO_MIN ||
4879 timeout < TCP_TIMEOUT_MIN)
4881 inet_csk(sk)->icsk_rto_min = timeout;
4884 if (val < 0 || val > 1)
4890 ret = tcp_sock_set_keepidle_locked(sk, val);
4893 if (val < 1 || val > MAX_TCP_KEEPINTVL)
4896 tp->keepalive_intvl = val * HZ;
4899 if (val < 1 || val > MAX_TCP_KEEPCNT)
4902 tp->keepalive_probes = val;
4905 if (val < 1 || val > MAX_TCP_SYNCNT)
4908 icsk->icsk_syn_retries = val;
4910 case TCP_USER_TIMEOUT:
4914 icsk->icsk_user_timeout = val;
4916 case TCP_NOTSENT_LOWAT:
4917 tp->notsent_lowat = val;
4918 sk->sk_write_space(sk);
4920 case TCP_WINDOW_CLAMP:
4921 ret = tcp_set_window_clamp(sk, val);
4934 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
4935 char *optval, int optlen)
4937 if (!sk_fullsock(sk))
4940 sock_owned_by_me(sk);
4942 if (level == SOL_SOCKET) {
4943 if (optlen != sizeof(int))
4948 *((int *)optval) = sk->sk_mark;
4951 *((int *)optval) = sk->sk_priority;
4953 case SO_BINDTOIFINDEX:
4954 *((int *)optval) = sk->sk_bound_dev_if;
4957 *((int *)optval) = sk->sk_reuseport;
4963 } else if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4964 struct inet_connection_sock *icsk;
4965 struct tcp_sock *tp;
4968 case TCP_CONGESTION:
4969 icsk = inet_csk(sk);
4971 if (!icsk->icsk_ca_ops || optlen <= 1)
4973 strncpy(optval, icsk->icsk_ca_ops->name, optlen);
4974 optval[optlen - 1] = 0;
4979 if (optlen <= 0 || !tp->saved_syn ||
4980 optlen > tcp_saved_syn_len(tp->saved_syn))
4982 memcpy(optval, tp->saved_syn->data, optlen);
4987 } else if (level == SOL_IP) {
4988 struct inet_sock *inet = inet_sk(sk);
4990 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4993 /* Only some options are supported */
4996 *((int *)optval) = (int)inet->tos;
5001 #if IS_ENABLED(CONFIG_IPV6)
5002 } else if (level == SOL_IPV6) {
5003 struct ipv6_pinfo *np = inet6_sk(sk);
5005 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
5008 /* Only some options are supported */
5011 *((int *)optval) = (int)np->tclass;
5023 memset(optval, 0, optlen);
5027 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level,
5028 int, optname, char *, optval, int, optlen)
5030 return _bpf_setsockopt(sk, level, optname, optval, optlen);
5033 const struct bpf_func_proto bpf_sk_setsockopt_proto = {
5034 .func = bpf_sk_setsockopt,
5036 .ret_type = RET_INTEGER,
5037 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5038 .arg2_type = ARG_ANYTHING,
5039 .arg3_type = ARG_ANYTHING,
5040 .arg4_type = ARG_PTR_TO_MEM,
5041 .arg5_type = ARG_CONST_SIZE,
5044 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level,
5045 int, optname, char *, optval, int, optlen)
5047 return _bpf_getsockopt(sk, level, optname, optval, optlen);
5050 const struct bpf_func_proto bpf_sk_getsockopt_proto = {
5051 .func = bpf_sk_getsockopt,
5053 .ret_type = RET_INTEGER,
5054 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5055 .arg2_type = ARG_ANYTHING,
5056 .arg3_type = ARG_ANYTHING,
5057 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5058 .arg5_type = ARG_CONST_SIZE,
5061 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5062 int, level, int, optname, char *, optval, int, optlen)
5064 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5067 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5068 .func = bpf_sock_addr_setsockopt,
5070 .ret_type = RET_INTEGER,
5071 .arg1_type = ARG_PTR_TO_CTX,
5072 .arg2_type = ARG_ANYTHING,
5073 .arg3_type = ARG_ANYTHING,
5074 .arg4_type = ARG_PTR_TO_MEM,
5075 .arg5_type = ARG_CONST_SIZE,
5078 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5079 int, level, int, optname, char *, optval, int, optlen)
5081 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5084 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5085 .func = bpf_sock_addr_getsockopt,
5087 .ret_type = RET_INTEGER,
5088 .arg1_type = ARG_PTR_TO_CTX,
5089 .arg2_type = ARG_ANYTHING,
5090 .arg3_type = ARG_ANYTHING,
5091 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5092 .arg5_type = ARG_CONST_SIZE,
5095 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5096 int, level, int, optname, char *, optval, int, optlen)
5098 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5101 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5102 .func = bpf_sock_ops_setsockopt,
5104 .ret_type = RET_INTEGER,
5105 .arg1_type = ARG_PTR_TO_CTX,
5106 .arg2_type = ARG_ANYTHING,
5107 .arg3_type = ARG_ANYTHING,
5108 .arg4_type = ARG_PTR_TO_MEM,
5109 .arg5_type = ARG_CONST_SIZE,
5112 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5113 int optname, const u8 **start)
5115 struct sk_buff *syn_skb = bpf_sock->syn_skb;
5116 const u8 *hdr_start;
5120 /* sk is a request_sock here */
5122 if (optname == TCP_BPF_SYN) {
5123 hdr_start = syn_skb->data;
5124 ret = tcp_hdrlen(syn_skb);
5125 } else if (optname == TCP_BPF_SYN_IP) {
5126 hdr_start = skb_network_header(syn_skb);
5127 ret = skb_network_header_len(syn_skb) +
5128 tcp_hdrlen(syn_skb);
5130 /* optname == TCP_BPF_SYN_MAC */
5131 hdr_start = skb_mac_header(syn_skb);
5132 ret = skb_mac_header_len(syn_skb) +
5133 skb_network_header_len(syn_skb) +
5134 tcp_hdrlen(syn_skb);
5137 struct sock *sk = bpf_sock->sk;
5138 struct saved_syn *saved_syn;
5140 if (sk->sk_state == TCP_NEW_SYN_RECV)
5141 /* synack retransmit. bpf_sock->syn_skb will
5142 * not be available. It has to resort to
5143 * saved_syn (if it is saved).
5145 saved_syn = inet_reqsk(sk)->saved_syn;
5147 saved_syn = tcp_sk(sk)->saved_syn;
5152 if (optname == TCP_BPF_SYN) {
5153 hdr_start = saved_syn->data +
5154 saved_syn->mac_hdrlen +
5155 saved_syn->network_hdrlen;
5156 ret = saved_syn->tcp_hdrlen;
5157 } else if (optname == TCP_BPF_SYN_IP) {
5158 hdr_start = saved_syn->data +
5159 saved_syn->mac_hdrlen;
5160 ret = saved_syn->network_hdrlen +
5161 saved_syn->tcp_hdrlen;
5163 /* optname == TCP_BPF_SYN_MAC */
5165 /* TCP_SAVE_SYN may not have saved the mac hdr */
5166 if (!saved_syn->mac_hdrlen)
5169 hdr_start = saved_syn->data;
5170 ret = saved_syn->mac_hdrlen +
5171 saved_syn->network_hdrlen +
5172 saved_syn->tcp_hdrlen;
5180 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5181 int, level, int, optname, char *, optval, int, optlen)
5183 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5184 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5185 int ret, copy_len = 0;
5188 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5191 if (optlen < copy_len) {
5196 memcpy(optval, start, copy_len);
5199 /* Zero out unused buffer at the end */
5200 memset(optval + copy_len, 0, optlen - copy_len);
5205 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5208 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5209 .func = bpf_sock_ops_getsockopt,
5211 .ret_type = RET_INTEGER,
5212 .arg1_type = ARG_PTR_TO_CTX,
5213 .arg2_type = ARG_ANYTHING,
5214 .arg3_type = ARG_ANYTHING,
5215 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5216 .arg5_type = ARG_CONST_SIZE,
5219 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5222 struct sock *sk = bpf_sock->sk;
5223 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5225 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5228 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5230 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5233 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5234 .func = bpf_sock_ops_cb_flags_set,
5236 .ret_type = RET_INTEGER,
5237 .arg1_type = ARG_PTR_TO_CTX,
5238 .arg2_type = ARG_ANYTHING,
5241 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5242 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5244 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5248 struct sock *sk = ctx->sk;
5249 u32 flags = BIND_FROM_BPF;
5253 if (addr_len < offsetofend(struct sockaddr, sa_family))
5255 if (addr->sa_family == AF_INET) {
5256 if (addr_len < sizeof(struct sockaddr_in))
5258 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5259 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5260 return __inet_bind(sk, addr, addr_len, flags);
5261 #if IS_ENABLED(CONFIG_IPV6)
5262 } else if (addr->sa_family == AF_INET6) {
5263 if (addr_len < SIN6_LEN_RFC2133)
5265 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5266 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5267 /* ipv6_bpf_stub cannot be NULL, since it's called from
5268 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5270 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5271 #endif /* CONFIG_IPV6 */
5273 #endif /* CONFIG_INET */
5275 return -EAFNOSUPPORT;
5278 static const struct bpf_func_proto bpf_bind_proto = {
5281 .ret_type = RET_INTEGER,
5282 .arg1_type = ARG_PTR_TO_CTX,
5283 .arg2_type = ARG_PTR_TO_MEM,
5284 .arg3_type = ARG_CONST_SIZE,
5288 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5289 struct bpf_xfrm_state *, to, u32, size, u64, flags)
5291 const struct sec_path *sp = skb_sec_path(skb);
5292 const struct xfrm_state *x;
5294 if (!sp || unlikely(index >= sp->len || flags))
5297 x = sp->xvec[index];
5299 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5302 to->reqid = x->props.reqid;
5303 to->spi = x->id.spi;
5304 to->family = x->props.family;
5307 if (to->family == AF_INET6) {
5308 memcpy(to->remote_ipv6, x->props.saddr.a6,
5309 sizeof(to->remote_ipv6));
5311 to->remote_ipv4 = x->props.saddr.a4;
5312 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5317 memset(to, 0, size);
5321 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5322 .func = bpf_skb_get_xfrm_state,
5324 .ret_type = RET_INTEGER,
5325 .arg1_type = ARG_PTR_TO_CTX,
5326 .arg2_type = ARG_ANYTHING,
5327 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
5328 .arg4_type = ARG_CONST_SIZE,
5329 .arg5_type = ARG_ANYTHING,
5333 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5334 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
5335 const struct neighbour *neigh,
5336 const struct net_device *dev, u32 mtu)
5338 memcpy(params->dmac, neigh->ha, ETH_ALEN);
5339 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5340 params->h_vlan_TCI = 0;
5341 params->h_vlan_proto = 0;
5343 params->mtu_result = mtu; /* union with tot_len */
5349 #if IS_ENABLED(CONFIG_INET)
5350 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5351 u32 flags, bool check_mtu)
5353 struct fib_nh_common *nhc;
5354 struct in_device *in_dev;
5355 struct neighbour *neigh;
5356 struct net_device *dev;
5357 struct fib_result res;
5362 dev = dev_get_by_index_rcu(net, params->ifindex);
5366 /* verify forwarding is enabled on this interface */
5367 in_dev = __in_dev_get_rcu(dev);
5368 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5369 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5371 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5373 fl4.flowi4_oif = params->ifindex;
5375 fl4.flowi4_iif = params->ifindex;
5378 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5379 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5380 fl4.flowi4_flags = 0;
5382 fl4.flowi4_proto = params->l4_protocol;
5383 fl4.daddr = params->ipv4_dst;
5384 fl4.saddr = params->ipv4_src;
5385 fl4.fl4_sport = params->sport;
5386 fl4.fl4_dport = params->dport;
5387 fl4.flowi4_multipath_hash = 0;
5389 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5390 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5391 struct fib_table *tb;
5393 tb = fib_get_table(net, tbid);
5395 return BPF_FIB_LKUP_RET_NOT_FWDED;
5397 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5399 fl4.flowi4_mark = 0;
5400 fl4.flowi4_secid = 0;
5401 fl4.flowi4_tun_key.tun_id = 0;
5402 fl4.flowi4_uid = sock_net_uid(net, NULL);
5404 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5408 /* map fib lookup errors to RTN_ type */
5410 return BPF_FIB_LKUP_RET_BLACKHOLE;
5411 if (err == -EHOSTUNREACH)
5412 return BPF_FIB_LKUP_RET_UNREACHABLE;
5414 return BPF_FIB_LKUP_RET_PROHIBIT;
5416 return BPF_FIB_LKUP_RET_NOT_FWDED;
5419 if (res.type != RTN_UNICAST)
5420 return BPF_FIB_LKUP_RET_NOT_FWDED;
5422 if (fib_info_num_path(res.fi) > 1)
5423 fib_select_path(net, &res, &fl4, NULL);
5426 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5427 if (params->tot_len > mtu) {
5428 params->mtu_result = mtu; /* union with tot_len */
5429 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5435 /* do not handle lwt encaps right now */
5436 if (nhc->nhc_lwtstate)
5437 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5441 params->rt_metric = res.fi->fib_priority;
5442 params->ifindex = dev->ifindex;
5444 /* xdp and cls_bpf programs are run in RCU-bh so
5445 * rcu_read_lock_bh is not needed here
5447 if (likely(nhc->nhc_gw_family != AF_INET6)) {
5448 if (nhc->nhc_gw_family)
5449 params->ipv4_dst = nhc->nhc_gw.ipv4;
5451 neigh = __ipv4_neigh_lookup_noref(dev,
5452 (__force u32)params->ipv4_dst);
5454 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5456 params->family = AF_INET6;
5457 *dst = nhc->nhc_gw.ipv6;
5458 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5462 return BPF_FIB_LKUP_RET_NO_NEIGH;
5464 return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5468 #if IS_ENABLED(CONFIG_IPV6)
5469 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5470 u32 flags, bool check_mtu)
5472 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5473 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5474 struct fib6_result res = {};
5475 struct neighbour *neigh;
5476 struct net_device *dev;
5477 struct inet6_dev *idev;
5483 /* link local addresses are never forwarded */
5484 if (rt6_need_strict(dst) || rt6_need_strict(src))
5485 return BPF_FIB_LKUP_RET_NOT_FWDED;
5487 dev = dev_get_by_index_rcu(net, params->ifindex);
5491 idev = __in6_dev_get_safely(dev);
5492 if (unlikely(!idev || !idev->cnf.forwarding))
5493 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5495 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5497 oif = fl6.flowi6_oif = params->ifindex;
5499 oif = fl6.flowi6_iif = params->ifindex;
5501 strict = RT6_LOOKUP_F_HAS_SADDR;
5503 fl6.flowlabel = params->flowinfo;
5504 fl6.flowi6_scope = 0;
5505 fl6.flowi6_flags = 0;
5508 fl6.flowi6_proto = params->l4_protocol;
5511 fl6.fl6_sport = params->sport;
5512 fl6.fl6_dport = params->dport;
5514 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5515 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5516 struct fib6_table *tb;
5518 tb = ipv6_stub->fib6_get_table(net, tbid);
5520 return BPF_FIB_LKUP_RET_NOT_FWDED;
5522 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5525 fl6.flowi6_mark = 0;
5526 fl6.flowi6_secid = 0;
5527 fl6.flowi6_tun_key.tun_id = 0;
5528 fl6.flowi6_uid = sock_net_uid(net, NULL);
5530 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5533 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5534 res.f6i == net->ipv6.fib6_null_entry))
5535 return BPF_FIB_LKUP_RET_NOT_FWDED;
5537 switch (res.fib6_type) {
5538 /* only unicast is forwarded */
5542 return BPF_FIB_LKUP_RET_BLACKHOLE;
5543 case RTN_UNREACHABLE:
5544 return BPF_FIB_LKUP_RET_UNREACHABLE;
5546 return BPF_FIB_LKUP_RET_PROHIBIT;
5548 return BPF_FIB_LKUP_RET_NOT_FWDED;
5551 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5552 fl6.flowi6_oif != 0, NULL, strict);
5555 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5556 if (params->tot_len > mtu) {
5557 params->mtu_result = mtu; /* union with tot_len */
5558 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5562 if (res.nh->fib_nh_lws)
5563 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5565 if (res.nh->fib_nh_gw_family)
5566 *dst = res.nh->fib_nh_gw6;
5568 dev = res.nh->fib_nh_dev;
5569 params->rt_metric = res.f6i->fib6_metric;
5570 params->ifindex = dev->ifindex;
5572 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
5575 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5577 return BPF_FIB_LKUP_RET_NO_NEIGH;
5579 return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5583 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
5584 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5586 if (plen < sizeof(*params))
5589 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5592 switch (params->family) {
5593 #if IS_ENABLED(CONFIG_INET)
5595 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
5598 #if IS_ENABLED(CONFIG_IPV6)
5600 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
5604 return -EAFNOSUPPORT;
5607 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
5608 .func = bpf_xdp_fib_lookup,
5610 .ret_type = RET_INTEGER,
5611 .arg1_type = ARG_PTR_TO_CTX,
5612 .arg2_type = ARG_PTR_TO_MEM,
5613 .arg3_type = ARG_CONST_SIZE,
5614 .arg4_type = ARG_ANYTHING,
5617 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
5618 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5620 struct net *net = dev_net(skb->dev);
5621 int rc = -EAFNOSUPPORT;
5622 bool check_mtu = false;
5624 if (plen < sizeof(*params))
5627 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5630 if (params->tot_len)
5633 switch (params->family) {
5634 #if IS_ENABLED(CONFIG_INET)
5636 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
5639 #if IS_ENABLED(CONFIG_IPV6)
5641 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
5646 if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
5647 struct net_device *dev;
5649 /* When tot_len isn't provided by user, check skb
5650 * against MTU of FIB lookup resulting net_device
5652 dev = dev_get_by_index_rcu(net, params->ifindex);
5653 if (!is_skb_forwardable(dev, skb))
5654 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
5656 params->mtu_result = dev->mtu; /* union with tot_len */
5662 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
5663 .func = bpf_skb_fib_lookup,
5665 .ret_type = RET_INTEGER,
5666 .arg1_type = ARG_PTR_TO_CTX,
5667 .arg2_type = ARG_PTR_TO_MEM,
5668 .arg3_type = ARG_CONST_SIZE,
5669 .arg4_type = ARG_ANYTHING,
5672 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
5675 struct net *netns = dev_net(dev_curr);
5677 /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
5681 return dev_get_by_index_rcu(netns, ifindex);
5684 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
5685 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5687 int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5688 struct net_device *dev = skb->dev;
5689 int skb_len, dev_len;
5692 if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
5695 if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
5698 dev = __dev_via_ifindex(dev, ifindex);
5702 mtu = READ_ONCE(dev->mtu);
5704 dev_len = mtu + dev->hard_header_len;
5706 /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
5707 skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
5709 skb_len += len_diff; /* minus result pass check */
5710 if (skb_len <= dev_len) {
5711 ret = BPF_MTU_CHK_RET_SUCCESS;
5714 /* At this point, skb->len exceed MTU, but as it include length of all
5715 * segments, it can still be below MTU. The SKB can possibly get
5716 * re-segmented in transmit path (see validate_xmit_skb). Thus, user
5717 * must choose if segs are to be MTU checked.
5719 if (skb_is_gso(skb)) {
5720 ret = BPF_MTU_CHK_RET_SUCCESS;
5722 if (flags & BPF_MTU_CHK_SEGS &&
5723 !skb_gso_validate_network_len(skb, mtu))
5724 ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
5727 /* BPF verifier guarantees valid pointer */
5733 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
5734 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5736 struct net_device *dev = xdp->rxq->dev;
5737 int xdp_len = xdp->data_end - xdp->data;
5738 int ret = BPF_MTU_CHK_RET_SUCCESS;
5741 /* XDP variant doesn't support multi-buffer segment check (yet) */
5742 if (unlikely(flags))
5745 dev = __dev_via_ifindex(dev, ifindex);
5749 mtu = READ_ONCE(dev->mtu);
5751 /* Add L2-header as dev MTU is L3 size */
5752 dev_len = mtu + dev->hard_header_len;
5754 /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
5756 xdp_len = *mtu_len + dev->hard_header_len;
5758 xdp_len += len_diff; /* minus result pass check */
5759 if (xdp_len > dev_len)
5760 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5762 /* BPF verifier guarantees valid pointer */
5768 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
5769 .func = bpf_skb_check_mtu,
5771 .ret_type = RET_INTEGER,
5772 .arg1_type = ARG_PTR_TO_CTX,
5773 .arg2_type = ARG_ANYTHING,
5774 .arg3_type = ARG_PTR_TO_INT,
5775 .arg4_type = ARG_ANYTHING,
5776 .arg5_type = ARG_ANYTHING,
5779 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
5780 .func = bpf_xdp_check_mtu,
5782 .ret_type = RET_INTEGER,
5783 .arg1_type = ARG_PTR_TO_CTX,
5784 .arg2_type = ARG_ANYTHING,
5785 .arg3_type = ARG_PTR_TO_INT,
5786 .arg4_type = ARG_ANYTHING,
5787 .arg5_type = ARG_ANYTHING,
5790 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5791 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
5794 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
5796 if (!seg6_validate_srh(srh, len, false))
5800 case BPF_LWT_ENCAP_SEG6_INLINE:
5801 if (skb->protocol != htons(ETH_P_IPV6))
5804 err = seg6_do_srh_inline(skb, srh);
5806 case BPF_LWT_ENCAP_SEG6:
5807 skb_reset_inner_headers(skb);
5808 skb->encapsulation = 1;
5809 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
5815 bpf_compute_data_pointers(skb);
5819 ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5820 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
5822 return seg6_lookup_nexthop(skb, NULL, 0);
5824 #endif /* CONFIG_IPV6_SEG6_BPF */
5826 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5827 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
5830 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
5834 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
5838 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5839 case BPF_LWT_ENCAP_SEG6:
5840 case BPF_LWT_ENCAP_SEG6_INLINE:
5841 return bpf_push_seg6_encap(skb, type, hdr, len);
5843 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5844 case BPF_LWT_ENCAP_IP:
5845 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
5852 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
5853 void *, hdr, u32, len)
5856 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5857 case BPF_LWT_ENCAP_IP:
5858 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
5865 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
5866 .func = bpf_lwt_in_push_encap,
5868 .ret_type = RET_INTEGER,
5869 .arg1_type = ARG_PTR_TO_CTX,
5870 .arg2_type = ARG_ANYTHING,
5871 .arg3_type = ARG_PTR_TO_MEM,
5872 .arg4_type = ARG_CONST_SIZE
5875 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
5876 .func = bpf_lwt_xmit_push_encap,
5878 .ret_type = RET_INTEGER,
5879 .arg1_type = ARG_PTR_TO_CTX,
5880 .arg2_type = ARG_ANYTHING,
5881 .arg3_type = ARG_PTR_TO_MEM,
5882 .arg4_type = ARG_CONST_SIZE
5885 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5886 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
5887 const void *, from, u32, len)
5889 struct seg6_bpf_srh_state *srh_state =
5890 this_cpu_ptr(&seg6_bpf_srh_states);
5891 struct ipv6_sr_hdr *srh = srh_state->srh;
5892 void *srh_tlvs, *srh_end, *ptr;
5898 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
5899 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
5901 ptr = skb->data + offset;
5902 if (ptr >= srh_tlvs && ptr + len <= srh_end)
5903 srh_state->valid = false;
5904 else if (ptr < (void *)&srh->flags ||
5905 ptr + len > (void *)&srh->segments)
5908 if (unlikely(bpf_try_make_writable(skb, offset + len)))
5910 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5912 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5914 memcpy(skb->data + offset, from, len);
5918 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
5919 .func = bpf_lwt_seg6_store_bytes,
5921 .ret_type = RET_INTEGER,
5922 .arg1_type = ARG_PTR_TO_CTX,
5923 .arg2_type = ARG_ANYTHING,
5924 .arg3_type = ARG_PTR_TO_MEM,
5925 .arg4_type = ARG_CONST_SIZE
5928 static void bpf_update_srh_state(struct sk_buff *skb)
5930 struct seg6_bpf_srh_state *srh_state =
5931 this_cpu_ptr(&seg6_bpf_srh_states);
5934 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
5935 srh_state->srh = NULL;
5937 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5938 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
5939 srh_state->valid = true;
5943 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
5944 u32, action, void *, param, u32, param_len)
5946 struct seg6_bpf_srh_state *srh_state =
5947 this_cpu_ptr(&seg6_bpf_srh_states);
5952 case SEG6_LOCAL_ACTION_END_X:
5953 if (!seg6_bpf_has_valid_srh(skb))
5955 if (param_len != sizeof(struct in6_addr))
5957 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
5958 case SEG6_LOCAL_ACTION_END_T:
5959 if (!seg6_bpf_has_valid_srh(skb))
5961 if (param_len != sizeof(int))
5963 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5964 case SEG6_LOCAL_ACTION_END_DT6:
5965 if (!seg6_bpf_has_valid_srh(skb))
5967 if (param_len != sizeof(int))
5970 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
5972 if (!pskb_pull(skb, hdroff))
5975 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
5976 skb_reset_network_header(skb);
5977 skb_reset_transport_header(skb);
5978 skb->encapsulation = 0;
5980 bpf_compute_data_pointers(skb);
5981 bpf_update_srh_state(skb);
5982 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5983 case SEG6_LOCAL_ACTION_END_B6:
5984 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5986 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
5989 bpf_update_srh_state(skb);
5992 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
5993 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5995 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
5998 bpf_update_srh_state(skb);
6006 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
6007 .func = bpf_lwt_seg6_action,
6009 .ret_type = RET_INTEGER,
6010 .arg1_type = ARG_PTR_TO_CTX,
6011 .arg2_type = ARG_ANYTHING,
6012 .arg3_type = ARG_PTR_TO_MEM,
6013 .arg4_type = ARG_CONST_SIZE
6016 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
6019 struct seg6_bpf_srh_state *srh_state =
6020 this_cpu_ptr(&seg6_bpf_srh_states);
6021 struct ipv6_sr_hdr *srh = srh_state->srh;
6022 void *srh_end, *srh_tlvs, *ptr;
6023 struct ipv6hdr *hdr;
6027 if (unlikely(srh == NULL))
6030 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
6031 ((srh->first_segment + 1) << 4));
6032 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
6034 ptr = skb->data + offset;
6036 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
6038 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
6042 ret = skb_cow_head(skb, len);
6043 if (unlikely(ret < 0))
6046 ret = bpf_skb_net_hdr_push(skb, offset, len);
6048 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6051 bpf_compute_data_pointers(skb);
6052 if (unlikely(ret < 0))
6055 hdr = (struct ipv6hdr *)skb->data;
6056 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6058 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6060 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6061 srh_state->hdrlen += len;
6062 srh_state->valid = false;
6066 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6067 .func = bpf_lwt_seg6_adjust_srh,
6069 .ret_type = RET_INTEGER,
6070 .arg1_type = ARG_PTR_TO_CTX,
6071 .arg2_type = ARG_ANYTHING,
6072 .arg3_type = ARG_ANYTHING,
6074 #endif /* CONFIG_IPV6_SEG6_BPF */
6077 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6078 int dif, int sdif, u8 family, u8 proto)
6080 bool refcounted = false;
6081 struct sock *sk = NULL;
6083 if (family == AF_INET) {
6084 __be32 src4 = tuple->ipv4.saddr;
6085 __be32 dst4 = tuple->ipv4.daddr;
6087 if (proto == IPPROTO_TCP)
6088 sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
6089 src4, tuple->ipv4.sport,
6090 dst4, tuple->ipv4.dport,
6091 dif, sdif, &refcounted);
6093 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6094 dst4, tuple->ipv4.dport,
6095 dif, sdif, &udp_table, NULL);
6096 #if IS_ENABLED(CONFIG_IPV6)
6098 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6099 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6101 if (proto == IPPROTO_TCP)
6102 sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
6103 src6, tuple->ipv6.sport,
6104 dst6, ntohs(tuple->ipv6.dport),
6105 dif, sdif, &refcounted);
6106 else if (likely(ipv6_bpf_stub))
6107 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6108 src6, tuple->ipv6.sport,
6109 dst6, tuple->ipv6.dport,
6115 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6116 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6122 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6123 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6124 * Returns the socket as an 'unsigned long' to simplify the casting in the
6125 * callers to satisfy BPF_CALL declarations.
6127 static struct sock *
6128 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6129 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6132 struct sock *sk = NULL;
6133 u8 family = AF_UNSPEC;
6137 if (len == sizeof(tuple->ipv4))
6139 else if (len == sizeof(tuple->ipv6))
6144 if (unlikely(family == AF_UNSPEC || flags ||
6145 !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6148 if (family == AF_INET)
6149 sdif = inet_sdif(skb);
6151 sdif = inet6_sdif(skb);
6153 if ((s32)netns_id < 0) {
6155 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6157 net = get_net_ns_by_id(caller_net, netns_id);
6160 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6168 static struct sock *
6169 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6170 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6173 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6174 ifindex, proto, netns_id, flags);
6177 sk = sk_to_full_sk(sk);
6178 if (!sk_fullsock(sk)) {
6187 static struct sock *
6188 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6189 u8 proto, u64 netns_id, u64 flags)
6191 struct net *caller_net;
6195 caller_net = dev_net(skb->dev);
6196 ifindex = skb->dev->ifindex;
6198 caller_net = sock_net(skb->sk);
6202 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6206 static struct sock *
6207 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6208 u8 proto, u64 netns_id, u64 flags)
6210 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6214 sk = sk_to_full_sk(sk);
6215 if (!sk_fullsock(sk)) {
6224 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6225 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6227 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6231 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6232 .func = bpf_skc_lookup_tcp,
6235 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6236 .arg1_type = ARG_PTR_TO_CTX,
6237 .arg2_type = ARG_PTR_TO_MEM,
6238 .arg3_type = ARG_CONST_SIZE,
6239 .arg4_type = ARG_ANYTHING,
6240 .arg5_type = ARG_ANYTHING,
6243 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6244 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6246 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6250 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6251 .func = bpf_sk_lookup_tcp,
6254 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6255 .arg1_type = ARG_PTR_TO_CTX,
6256 .arg2_type = ARG_PTR_TO_MEM,
6257 .arg3_type = ARG_CONST_SIZE,
6258 .arg4_type = ARG_ANYTHING,
6259 .arg5_type = ARG_ANYTHING,
6262 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6263 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6265 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6269 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6270 .func = bpf_sk_lookup_udp,
6273 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6274 .arg1_type = ARG_PTR_TO_CTX,
6275 .arg2_type = ARG_PTR_TO_MEM,
6276 .arg3_type = ARG_CONST_SIZE,
6277 .arg4_type = ARG_ANYTHING,
6278 .arg5_type = ARG_ANYTHING,
6281 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6283 if (sk && sk_is_refcounted(sk))
6288 static const struct bpf_func_proto bpf_sk_release_proto = {
6289 .func = bpf_sk_release,
6291 .ret_type = RET_INTEGER,
6292 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6295 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6296 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6298 struct net *caller_net = dev_net(ctx->rxq->dev);
6299 int ifindex = ctx->rxq->dev->ifindex;
6301 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6302 ifindex, IPPROTO_UDP, netns_id,
6306 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6307 .func = bpf_xdp_sk_lookup_udp,
6310 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6311 .arg1_type = ARG_PTR_TO_CTX,
6312 .arg2_type = ARG_PTR_TO_MEM,
6313 .arg3_type = ARG_CONST_SIZE,
6314 .arg4_type = ARG_ANYTHING,
6315 .arg5_type = ARG_ANYTHING,
6318 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6319 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6321 struct net *caller_net = dev_net(ctx->rxq->dev);
6322 int ifindex = ctx->rxq->dev->ifindex;
6324 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6325 ifindex, IPPROTO_TCP, netns_id,
6329 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6330 .func = bpf_xdp_skc_lookup_tcp,
6333 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6334 .arg1_type = ARG_PTR_TO_CTX,
6335 .arg2_type = ARG_PTR_TO_MEM,
6336 .arg3_type = ARG_CONST_SIZE,
6337 .arg4_type = ARG_ANYTHING,
6338 .arg5_type = ARG_ANYTHING,
6341 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6342 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6344 struct net *caller_net = dev_net(ctx->rxq->dev);
6345 int ifindex = ctx->rxq->dev->ifindex;
6347 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6348 ifindex, IPPROTO_TCP, netns_id,
6352 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6353 .func = bpf_xdp_sk_lookup_tcp,
6356 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6357 .arg1_type = ARG_PTR_TO_CTX,
6358 .arg2_type = ARG_PTR_TO_MEM,
6359 .arg3_type = ARG_CONST_SIZE,
6360 .arg4_type = ARG_ANYTHING,
6361 .arg5_type = ARG_ANYTHING,
6364 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6365 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6367 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6368 sock_net(ctx->sk), 0,
6369 IPPROTO_TCP, netns_id, flags);
6372 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6373 .func = bpf_sock_addr_skc_lookup_tcp,
6375 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6376 .arg1_type = ARG_PTR_TO_CTX,
6377 .arg2_type = ARG_PTR_TO_MEM,
6378 .arg3_type = ARG_CONST_SIZE,
6379 .arg4_type = ARG_ANYTHING,
6380 .arg5_type = ARG_ANYTHING,
6383 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6384 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6386 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6387 sock_net(ctx->sk), 0, IPPROTO_TCP,
6391 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6392 .func = bpf_sock_addr_sk_lookup_tcp,
6394 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6395 .arg1_type = ARG_PTR_TO_CTX,
6396 .arg2_type = ARG_PTR_TO_MEM,
6397 .arg3_type = ARG_CONST_SIZE,
6398 .arg4_type = ARG_ANYTHING,
6399 .arg5_type = ARG_ANYTHING,
6402 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6403 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6405 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6406 sock_net(ctx->sk), 0, IPPROTO_UDP,
6410 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6411 .func = bpf_sock_addr_sk_lookup_udp,
6413 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6414 .arg1_type = ARG_PTR_TO_CTX,
6415 .arg2_type = ARG_PTR_TO_MEM,
6416 .arg3_type = ARG_CONST_SIZE,
6417 .arg4_type = ARG_ANYTHING,
6418 .arg5_type = ARG_ANYTHING,
6421 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6422 struct bpf_insn_access_aux *info)
6424 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6428 if (off % size != 0)
6432 case offsetof(struct bpf_tcp_sock, bytes_received):
6433 case offsetof(struct bpf_tcp_sock, bytes_acked):
6434 return size == sizeof(__u64);
6436 return size == sizeof(__u32);
6440 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6441 const struct bpf_insn *si,
6442 struct bpf_insn *insn_buf,
6443 struct bpf_prog *prog, u32 *target_size)
6445 struct bpf_insn *insn = insn_buf;
6447 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \
6449 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \
6450 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6451 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6452 si->dst_reg, si->src_reg, \
6453 offsetof(struct tcp_sock, FIELD)); \
6456 #define BPF_INET_SOCK_GET_COMMON(FIELD) \
6458 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \
6460 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6461 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6462 struct inet_connection_sock, \
6464 si->dst_reg, si->src_reg, \
6466 struct inet_connection_sock, \
6470 if (insn > insn_buf)
6471 return insn - insn_buf;
6474 case offsetof(struct bpf_tcp_sock, rtt_min):
6475 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
6476 sizeof(struct minmax));
6477 BUILD_BUG_ON(sizeof(struct minmax) <
6478 sizeof(struct minmax_sample));
6480 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6481 offsetof(struct tcp_sock, rtt_min) +
6482 offsetof(struct minmax_sample, v));
6484 case offsetof(struct bpf_tcp_sock, snd_cwnd):
6485 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
6487 case offsetof(struct bpf_tcp_sock, srtt_us):
6488 BPF_TCP_SOCK_GET_COMMON(srtt_us);
6490 case offsetof(struct bpf_tcp_sock, snd_ssthresh):
6491 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
6493 case offsetof(struct bpf_tcp_sock, rcv_nxt):
6494 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
6496 case offsetof(struct bpf_tcp_sock, snd_nxt):
6497 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
6499 case offsetof(struct bpf_tcp_sock, snd_una):
6500 BPF_TCP_SOCK_GET_COMMON(snd_una);
6502 case offsetof(struct bpf_tcp_sock, mss_cache):
6503 BPF_TCP_SOCK_GET_COMMON(mss_cache);
6505 case offsetof(struct bpf_tcp_sock, ecn_flags):
6506 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
6508 case offsetof(struct bpf_tcp_sock, rate_delivered):
6509 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
6511 case offsetof(struct bpf_tcp_sock, rate_interval_us):
6512 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
6514 case offsetof(struct bpf_tcp_sock, packets_out):
6515 BPF_TCP_SOCK_GET_COMMON(packets_out);
6517 case offsetof(struct bpf_tcp_sock, retrans_out):
6518 BPF_TCP_SOCK_GET_COMMON(retrans_out);
6520 case offsetof(struct bpf_tcp_sock, total_retrans):
6521 BPF_TCP_SOCK_GET_COMMON(total_retrans);
6523 case offsetof(struct bpf_tcp_sock, segs_in):
6524 BPF_TCP_SOCK_GET_COMMON(segs_in);
6526 case offsetof(struct bpf_tcp_sock, data_segs_in):
6527 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
6529 case offsetof(struct bpf_tcp_sock, segs_out):
6530 BPF_TCP_SOCK_GET_COMMON(segs_out);
6532 case offsetof(struct bpf_tcp_sock, data_segs_out):
6533 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
6535 case offsetof(struct bpf_tcp_sock, lost_out):
6536 BPF_TCP_SOCK_GET_COMMON(lost_out);
6538 case offsetof(struct bpf_tcp_sock, sacked_out):
6539 BPF_TCP_SOCK_GET_COMMON(sacked_out);
6541 case offsetof(struct bpf_tcp_sock, bytes_received):
6542 BPF_TCP_SOCK_GET_COMMON(bytes_received);
6544 case offsetof(struct bpf_tcp_sock, bytes_acked):
6545 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
6547 case offsetof(struct bpf_tcp_sock, dsack_dups):
6548 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
6550 case offsetof(struct bpf_tcp_sock, delivered):
6551 BPF_TCP_SOCK_GET_COMMON(delivered);
6553 case offsetof(struct bpf_tcp_sock, delivered_ce):
6554 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
6556 case offsetof(struct bpf_tcp_sock, icsk_retransmits):
6557 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
6561 return insn - insn_buf;
6564 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
6566 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
6567 return (unsigned long)sk;
6569 return (unsigned long)NULL;
6572 const struct bpf_func_proto bpf_tcp_sock_proto = {
6573 .func = bpf_tcp_sock,
6575 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
6576 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6579 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
6581 sk = sk_to_full_sk(sk);
6583 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
6584 return (unsigned long)sk;
6586 return (unsigned long)NULL;
6589 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
6590 .func = bpf_get_listener_sock,
6592 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6593 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6596 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
6598 unsigned int iphdr_len;
6600 switch (skb_protocol(skb, true)) {
6601 case cpu_to_be16(ETH_P_IP):
6602 iphdr_len = sizeof(struct iphdr);
6604 case cpu_to_be16(ETH_P_IPV6):
6605 iphdr_len = sizeof(struct ipv6hdr);
6611 if (skb_headlen(skb) < iphdr_len)
6614 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
6617 return INET_ECN_set_ce(skb);
6620 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6621 struct bpf_insn_access_aux *info)
6623 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
6626 if (off % size != 0)
6631 return size == sizeof(__u32);
6635 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
6636 const struct bpf_insn *si,
6637 struct bpf_insn *insn_buf,
6638 struct bpf_prog *prog, u32 *target_size)
6640 struct bpf_insn *insn = insn_buf;
6642 #define BPF_XDP_SOCK_GET(FIELD) \
6644 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \
6645 sizeof_field(struct bpf_xdp_sock, FIELD)); \
6646 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
6647 si->dst_reg, si->src_reg, \
6648 offsetof(struct xdp_sock, FIELD)); \
6652 case offsetof(struct bpf_xdp_sock, queue_id):
6653 BPF_XDP_SOCK_GET(queue_id);
6657 return insn - insn_buf;
6660 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
6661 .func = bpf_skb_ecn_set_ce,
6663 .ret_type = RET_INTEGER,
6664 .arg1_type = ARG_PTR_TO_CTX,
6667 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6668 struct tcphdr *, th, u32, th_len)
6670 #ifdef CONFIG_SYN_COOKIES
6674 if (unlikely(!sk || th_len < sizeof(*th)))
6677 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
6678 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6681 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
6684 if (!th->ack || th->rst || th->syn)
6687 if (tcp_synq_no_recent_overflow(sk))
6690 cookie = ntohl(th->ack_seq) - 1;
6692 switch (sk->sk_family) {
6694 if (unlikely(iph_len < sizeof(struct iphdr)))
6697 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
6700 #if IS_BUILTIN(CONFIG_IPV6)
6702 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6705 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
6707 #endif /* CONFIG_IPV6 */
6710 return -EPROTONOSUPPORT;
6722 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
6723 .func = bpf_tcp_check_syncookie,
6726 .ret_type = RET_INTEGER,
6727 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6728 .arg2_type = ARG_PTR_TO_MEM,
6729 .arg3_type = ARG_CONST_SIZE,
6730 .arg4_type = ARG_PTR_TO_MEM,
6731 .arg5_type = ARG_CONST_SIZE,
6734 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6735 struct tcphdr *, th, u32, th_len)
6737 #ifdef CONFIG_SYN_COOKIES
6741 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
6744 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6747 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
6750 if (!th->syn || th->ack || th->fin || th->rst)
6753 if (unlikely(iph_len < sizeof(struct iphdr)))
6756 /* Both struct iphdr and struct ipv6hdr have the version field at the
6757 * same offset so we can cast to the shorter header (struct iphdr).
6759 switch (((struct iphdr *)iph)->version) {
6761 if (sk->sk_family == AF_INET6 && sk->sk_ipv6only)
6764 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
6767 #if IS_BUILTIN(CONFIG_IPV6)
6769 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6772 if (sk->sk_family != AF_INET6)
6775 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
6777 #endif /* CONFIG_IPV6 */
6780 return -EPROTONOSUPPORT;
6785 return cookie | ((u64)mss << 32);
6788 #endif /* CONFIG_SYN_COOKIES */
6791 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
6792 .func = bpf_tcp_gen_syncookie,
6793 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */
6795 .ret_type = RET_INTEGER,
6796 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6797 .arg2_type = ARG_PTR_TO_MEM,
6798 .arg3_type = ARG_CONST_SIZE,
6799 .arg4_type = ARG_PTR_TO_MEM,
6800 .arg5_type = ARG_CONST_SIZE,
6803 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
6805 if (!sk || flags != 0)
6807 if (!skb_at_tc_ingress(skb))
6809 if (unlikely(dev_net(skb->dev) != sock_net(sk)))
6810 return -ENETUNREACH;
6811 if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
6812 return -ESOCKTNOSUPPORT;
6813 if (sk_is_refcounted(sk) &&
6814 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
6819 skb->destructor = sock_pfree;
6824 static const struct bpf_func_proto bpf_sk_assign_proto = {
6825 .func = bpf_sk_assign,
6827 .ret_type = RET_INTEGER,
6828 .arg1_type = ARG_PTR_TO_CTX,
6829 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6830 .arg3_type = ARG_ANYTHING,
6833 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
6834 u8 search_kind, const u8 *magic,
6835 u8 magic_len, bool *eol)
6841 while (op < opend) {
6844 if (kind == TCPOPT_EOL) {
6846 return ERR_PTR(-ENOMSG);
6847 } else if (kind == TCPOPT_NOP) {
6852 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
6853 /* Something is wrong in the received header.
6854 * Follow the TCP stack's tcp_parse_options()
6855 * and just bail here.
6857 return ERR_PTR(-EFAULT);
6860 if (search_kind == kind) {
6864 if (magic_len > kind_len - 2)
6865 return ERR_PTR(-ENOMSG);
6867 if (!memcmp(&op[2], magic, magic_len))
6874 return ERR_PTR(-ENOMSG);
6877 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6878 void *, search_res, u32, len, u64, flags)
6880 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
6881 const u8 *op, *opend, *magic, *search = search_res;
6882 u8 search_kind, search_len, copy_len, magic_len;
6885 /* 2 byte is the minimal option len except TCPOPT_NOP and
6886 * TCPOPT_EOL which are useless for the bpf prog to learn
6887 * and this helper disallow loading them also.
6889 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
6892 search_kind = search[0];
6893 search_len = search[1];
6895 if (search_len > len || search_kind == TCPOPT_NOP ||
6896 search_kind == TCPOPT_EOL)
6899 if (search_kind == TCPOPT_EXP || search_kind == 253) {
6900 /* 16 or 32 bit magic. +2 for kind and kind length */
6901 if (search_len != 4 && search_len != 6)
6904 magic_len = search_len - 2;
6913 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
6918 op += sizeof(struct tcphdr);
6920 if (!bpf_sock->skb ||
6921 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6922 /* This bpf_sock->op cannot call this helper */
6925 opend = bpf_sock->skb_data_end;
6926 op = bpf_sock->skb->data + sizeof(struct tcphdr);
6929 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
6936 if (copy_len > len) {
6941 memcpy(search_res, op, copy_len);
6945 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
6946 .func = bpf_sock_ops_load_hdr_opt,
6948 .ret_type = RET_INTEGER,
6949 .arg1_type = ARG_PTR_TO_CTX,
6950 .arg2_type = ARG_PTR_TO_MEM,
6951 .arg3_type = ARG_CONST_SIZE,
6952 .arg4_type = ARG_ANYTHING,
6955 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6956 const void *, from, u32, len, u64, flags)
6958 u8 new_kind, new_kind_len, magic_len = 0, *opend;
6959 const u8 *op, *new_op, *magic = NULL;
6960 struct sk_buff *skb;
6963 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
6966 if (len < 2 || flags)
6970 new_kind = new_op[0];
6971 new_kind_len = new_op[1];
6973 if (new_kind_len > len || new_kind == TCPOPT_NOP ||
6974 new_kind == TCPOPT_EOL)
6977 if (new_kind_len > bpf_sock->remaining_opt_len)
6980 /* 253 is another experimental kind */
6981 if (new_kind == TCPOPT_EXP || new_kind == 253) {
6982 if (new_kind_len < 4)
6984 /* Match for the 2 byte magic also.
6985 * RFC 6994: the magic could be 2 or 4 bytes.
6986 * Hence, matching by 2 byte only is on the
6987 * conservative side but it is the right
6988 * thing to do for the 'search-for-duplication'
6995 /* Check for duplication */
6996 skb = bpf_sock->skb;
6997 op = skb->data + sizeof(struct tcphdr);
6998 opend = bpf_sock->skb_data_end;
7000 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
7005 if (PTR_ERR(op) != -ENOMSG)
7009 /* The option has been ended. Treat it as no more
7010 * header option can be written.
7014 /* No duplication found. Store the header option. */
7015 memcpy(opend, from, new_kind_len);
7017 bpf_sock->remaining_opt_len -= new_kind_len;
7018 bpf_sock->skb_data_end += new_kind_len;
7023 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
7024 .func = bpf_sock_ops_store_hdr_opt,
7026 .ret_type = RET_INTEGER,
7027 .arg1_type = ARG_PTR_TO_CTX,
7028 .arg2_type = ARG_PTR_TO_MEM,
7029 .arg3_type = ARG_CONST_SIZE,
7030 .arg4_type = ARG_ANYTHING,
7033 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7034 u32, len, u64, flags)
7036 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7039 if (flags || len < 2)
7042 if (len > bpf_sock->remaining_opt_len)
7045 bpf_sock->remaining_opt_len -= len;
7050 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7051 .func = bpf_sock_ops_reserve_hdr_opt,
7053 .ret_type = RET_INTEGER,
7054 .arg1_type = ARG_PTR_TO_CTX,
7055 .arg2_type = ARG_ANYTHING,
7056 .arg3_type = ARG_ANYTHING,
7059 #endif /* CONFIG_INET */
7061 bool bpf_helper_changes_pkt_data(void *func)
7063 if (func == bpf_skb_vlan_push ||
7064 func == bpf_skb_vlan_pop ||
7065 func == bpf_skb_store_bytes ||
7066 func == bpf_skb_change_proto ||
7067 func == bpf_skb_change_head ||
7068 func == sk_skb_change_head ||
7069 func == bpf_skb_change_tail ||
7070 func == sk_skb_change_tail ||
7071 func == bpf_skb_adjust_room ||
7072 func == sk_skb_adjust_room ||
7073 func == bpf_skb_pull_data ||
7074 func == sk_skb_pull_data ||
7075 func == bpf_clone_redirect ||
7076 func == bpf_l3_csum_replace ||
7077 func == bpf_l4_csum_replace ||
7078 func == bpf_xdp_adjust_head ||
7079 func == bpf_xdp_adjust_meta ||
7080 func == bpf_msg_pull_data ||
7081 func == bpf_msg_push_data ||
7082 func == bpf_msg_pop_data ||
7083 func == bpf_xdp_adjust_tail ||
7084 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7085 func == bpf_lwt_seg6_store_bytes ||
7086 func == bpf_lwt_seg6_adjust_srh ||
7087 func == bpf_lwt_seg6_action ||
7090 func == bpf_sock_ops_store_hdr_opt ||
7092 func == bpf_lwt_in_push_encap ||
7093 func == bpf_lwt_xmit_push_encap)
7099 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7100 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7102 static const struct bpf_func_proto *
7103 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7106 /* inet and inet6 sockets are created in a process
7107 * context so there is always a valid uid/gid
7109 case BPF_FUNC_get_current_uid_gid:
7110 return &bpf_get_current_uid_gid_proto;
7111 case BPF_FUNC_get_local_storage:
7112 return &bpf_get_local_storage_proto;
7113 case BPF_FUNC_get_socket_cookie:
7114 return &bpf_get_socket_cookie_sock_proto;
7115 case BPF_FUNC_get_netns_cookie:
7116 return &bpf_get_netns_cookie_sock_proto;
7117 case BPF_FUNC_perf_event_output:
7118 return &bpf_event_output_data_proto;
7119 case BPF_FUNC_get_current_pid_tgid:
7120 return &bpf_get_current_pid_tgid_proto;
7121 case BPF_FUNC_get_current_comm:
7122 return &bpf_get_current_comm_proto;
7123 #ifdef CONFIG_CGROUPS
7124 case BPF_FUNC_get_current_cgroup_id:
7125 return &bpf_get_current_cgroup_id_proto;
7126 case BPF_FUNC_get_current_ancestor_cgroup_id:
7127 return &bpf_get_current_ancestor_cgroup_id_proto;
7129 #ifdef CONFIG_CGROUP_NET_CLASSID
7130 case BPF_FUNC_get_cgroup_classid:
7131 return &bpf_get_cgroup_classid_curr_proto;
7133 case BPF_FUNC_sk_storage_get:
7134 return &bpf_sk_storage_get_cg_sock_proto;
7136 return bpf_base_func_proto(func_id);
7140 static const struct bpf_func_proto *
7141 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7144 /* inet and inet6 sockets are created in a process
7145 * context so there is always a valid uid/gid
7147 case BPF_FUNC_get_current_uid_gid:
7148 return &bpf_get_current_uid_gid_proto;
7150 switch (prog->expected_attach_type) {
7151 case BPF_CGROUP_INET4_CONNECT:
7152 case BPF_CGROUP_INET6_CONNECT:
7153 return &bpf_bind_proto;
7157 case BPF_FUNC_get_socket_cookie:
7158 return &bpf_get_socket_cookie_sock_addr_proto;
7159 case BPF_FUNC_get_netns_cookie:
7160 return &bpf_get_netns_cookie_sock_addr_proto;
7161 case BPF_FUNC_get_local_storage:
7162 return &bpf_get_local_storage_proto;
7163 case BPF_FUNC_perf_event_output:
7164 return &bpf_event_output_data_proto;
7165 case BPF_FUNC_get_current_pid_tgid:
7166 return &bpf_get_current_pid_tgid_proto;
7167 case BPF_FUNC_get_current_comm:
7168 return &bpf_get_current_comm_proto;
7169 #ifdef CONFIG_CGROUPS
7170 case BPF_FUNC_get_current_cgroup_id:
7171 return &bpf_get_current_cgroup_id_proto;
7172 case BPF_FUNC_get_current_ancestor_cgroup_id:
7173 return &bpf_get_current_ancestor_cgroup_id_proto;
7175 #ifdef CONFIG_CGROUP_NET_CLASSID
7176 case BPF_FUNC_get_cgroup_classid:
7177 return &bpf_get_cgroup_classid_curr_proto;
7180 case BPF_FUNC_sk_lookup_tcp:
7181 return &bpf_sock_addr_sk_lookup_tcp_proto;
7182 case BPF_FUNC_sk_lookup_udp:
7183 return &bpf_sock_addr_sk_lookup_udp_proto;
7184 case BPF_FUNC_sk_release:
7185 return &bpf_sk_release_proto;
7186 case BPF_FUNC_skc_lookup_tcp:
7187 return &bpf_sock_addr_skc_lookup_tcp_proto;
7188 #endif /* CONFIG_INET */
7189 case BPF_FUNC_sk_storage_get:
7190 return &bpf_sk_storage_get_proto;
7191 case BPF_FUNC_sk_storage_delete:
7192 return &bpf_sk_storage_delete_proto;
7193 case BPF_FUNC_setsockopt:
7194 switch (prog->expected_attach_type) {
7195 case BPF_CGROUP_INET4_BIND:
7196 case BPF_CGROUP_INET6_BIND:
7197 case BPF_CGROUP_INET4_CONNECT:
7198 case BPF_CGROUP_INET6_CONNECT:
7199 case BPF_CGROUP_UDP4_RECVMSG:
7200 case BPF_CGROUP_UDP6_RECVMSG:
7201 case BPF_CGROUP_UDP4_SENDMSG:
7202 case BPF_CGROUP_UDP6_SENDMSG:
7203 case BPF_CGROUP_INET4_GETPEERNAME:
7204 case BPF_CGROUP_INET6_GETPEERNAME:
7205 case BPF_CGROUP_INET4_GETSOCKNAME:
7206 case BPF_CGROUP_INET6_GETSOCKNAME:
7207 return &bpf_sock_addr_setsockopt_proto;
7211 case BPF_FUNC_getsockopt:
7212 switch (prog->expected_attach_type) {
7213 case BPF_CGROUP_INET4_BIND:
7214 case BPF_CGROUP_INET6_BIND:
7215 case BPF_CGROUP_INET4_CONNECT:
7216 case BPF_CGROUP_INET6_CONNECT:
7217 case BPF_CGROUP_UDP4_RECVMSG:
7218 case BPF_CGROUP_UDP6_RECVMSG:
7219 case BPF_CGROUP_UDP4_SENDMSG:
7220 case BPF_CGROUP_UDP6_SENDMSG:
7221 case BPF_CGROUP_INET4_GETPEERNAME:
7222 case BPF_CGROUP_INET6_GETPEERNAME:
7223 case BPF_CGROUP_INET4_GETSOCKNAME:
7224 case BPF_CGROUP_INET6_GETSOCKNAME:
7225 return &bpf_sock_addr_getsockopt_proto;
7230 return bpf_sk_base_func_proto(func_id);
7234 static const struct bpf_func_proto *
7235 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7238 case BPF_FUNC_skb_load_bytes:
7239 return &bpf_skb_load_bytes_proto;
7240 case BPF_FUNC_skb_load_bytes_relative:
7241 return &bpf_skb_load_bytes_relative_proto;
7242 case BPF_FUNC_get_socket_cookie:
7243 return &bpf_get_socket_cookie_proto;
7244 case BPF_FUNC_get_socket_uid:
7245 return &bpf_get_socket_uid_proto;
7246 case BPF_FUNC_perf_event_output:
7247 return &bpf_skb_event_output_proto;
7249 return bpf_sk_base_func_proto(func_id);
7253 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7254 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7256 static const struct bpf_func_proto *
7257 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7260 case BPF_FUNC_get_local_storage:
7261 return &bpf_get_local_storage_proto;
7262 case BPF_FUNC_sk_fullsock:
7263 return &bpf_sk_fullsock_proto;
7264 case BPF_FUNC_sk_storage_get:
7265 return &bpf_sk_storage_get_proto;
7266 case BPF_FUNC_sk_storage_delete:
7267 return &bpf_sk_storage_delete_proto;
7268 case BPF_FUNC_perf_event_output:
7269 return &bpf_skb_event_output_proto;
7270 #ifdef CONFIG_SOCK_CGROUP_DATA
7271 case BPF_FUNC_skb_cgroup_id:
7272 return &bpf_skb_cgroup_id_proto;
7273 case BPF_FUNC_skb_ancestor_cgroup_id:
7274 return &bpf_skb_ancestor_cgroup_id_proto;
7275 case BPF_FUNC_sk_cgroup_id:
7276 return &bpf_sk_cgroup_id_proto;
7277 case BPF_FUNC_sk_ancestor_cgroup_id:
7278 return &bpf_sk_ancestor_cgroup_id_proto;
7281 case BPF_FUNC_sk_lookup_tcp:
7282 return &bpf_sk_lookup_tcp_proto;
7283 case BPF_FUNC_sk_lookup_udp:
7284 return &bpf_sk_lookup_udp_proto;
7285 case BPF_FUNC_sk_release:
7286 return &bpf_sk_release_proto;
7287 case BPF_FUNC_skc_lookup_tcp:
7288 return &bpf_skc_lookup_tcp_proto;
7289 case BPF_FUNC_tcp_sock:
7290 return &bpf_tcp_sock_proto;
7291 case BPF_FUNC_get_listener_sock:
7292 return &bpf_get_listener_sock_proto;
7293 case BPF_FUNC_skb_ecn_set_ce:
7294 return &bpf_skb_ecn_set_ce_proto;
7297 return sk_filter_func_proto(func_id, prog);
7301 static const struct bpf_func_proto *
7302 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7305 case BPF_FUNC_skb_store_bytes:
7306 return &bpf_skb_store_bytes_proto;
7307 case BPF_FUNC_skb_load_bytes:
7308 return &bpf_skb_load_bytes_proto;
7309 case BPF_FUNC_skb_load_bytes_relative:
7310 return &bpf_skb_load_bytes_relative_proto;
7311 case BPF_FUNC_skb_pull_data:
7312 return &bpf_skb_pull_data_proto;
7313 case BPF_FUNC_csum_diff:
7314 return &bpf_csum_diff_proto;
7315 case BPF_FUNC_csum_update:
7316 return &bpf_csum_update_proto;
7317 case BPF_FUNC_csum_level:
7318 return &bpf_csum_level_proto;
7319 case BPF_FUNC_l3_csum_replace:
7320 return &bpf_l3_csum_replace_proto;
7321 case BPF_FUNC_l4_csum_replace:
7322 return &bpf_l4_csum_replace_proto;
7323 case BPF_FUNC_clone_redirect:
7324 return &bpf_clone_redirect_proto;
7325 case BPF_FUNC_get_cgroup_classid:
7326 return &bpf_get_cgroup_classid_proto;
7327 case BPF_FUNC_skb_vlan_push:
7328 return &bpf_skb_vlan_push_proto;
7329 case BPF_FUNC_skb_vlan_pop:
7330 return &bpf_skb_vlan_pop_proto;
7331 case BPF_FUNC_skb_change_proto:
7332 return &bpf_skb_change_proto_proto;
7333 case BPF_FUNC_skb_change_type:
7334 return &bpf_skb_change_type_proto;
7335 case BPF_FUNC_skb_adjust_room:
7336 return &bpf_skb_adjust_room_proto;
7337 case BPF_FUNC_skb_change_tail:
7338 return &bpf_skb_change_tail_proto;
7339 case BPF_FUNC_skb_change_head:
7340 return &bpf_skb_change_head_proto;
7341 case BPF_FUNC_skb_get_tunnel_key:
7342 return &bpf_skb_get_tunnel_key_proto;
7343 case BPF_FUNC_skb_set_tunnel_key:
7344 return bpf_get_skb_set_tunnel_proto(func_id);
7345 case BPF_FUNC_skb_get_tunnel_opt:
7346 return &bpf_skb_get_tunnel_opt_proto;
7347 case BPF_FUNC_skb_set_tunnel_opt:
7348 return bpf_get_skb_set_tunnel_proto(func_id);
7349 case BPF_FUNC_redirect:
7350 return &bpf_redirect_proto;
7351 case BPF_FUNC_redirect_neigh:
7352 return &bpf_redirect_neigh_proto;
7353 case BPF_FUNC_redirect_peer:
7354 return &bpf_redirect_peer_proto;
7355 case BPF_FUNC_get_route_realm:
7356 return &bpf_get_route_realm_proto;
7357 case BPF_FUNC_get_hash_recalc:
7358 return &bpf_get_hash_recalc_proto;
7359 case BPF_FUNC_set_hash_invalid:
7360 return &bpf_set_hash_invalid_proto;
7361 case BPF_FUNC_set_hash:
7362 return &bpf_set_hash_proto;
7363 case BPF_FUNC_perf_event_output:
7364 return &bpf_skb_event_output_proto;
7365 case BPF_FUNC_get_smp_processor_id:
7366 return &bpf_get_smp_processor_id_proto;
7367 case BPF_FUNC_skb_under_cgroup:
7368 return &bpf_skb_under_cgroup_proto;
7369 case BPF_FUNC_get_socket_cookie:
7370 return &bpf_get_socket_cookie_proto;
7371 case BPF_FUNC_get_socket_uid:
7372 return &bpf_get_socket_uid_proto;
7373 case BPF_FUNC_fib_lookup:
7374 return &bpf_skb_fib_lookup_proto;
7375 case BPF_FUNC_check_mtu:
7376 return &bpf_skb_check_mtu_proto;
7377 case BPF_FUNC_sk_fullsock:
7378 return &bpf_sk_fullsock_proto;
7379 case BPF_FUNC_sk_storage_get:
7380 return &bpf_sk_storage_get_proto;
7381 case BPF_FUNC_sk_storage_delete:
7382 return &bpf_sk_storage_delete_proto;
7384 case BPF_FUNC_skb_get_xfrm_state:
7385 return &bpf_skb_get_xfrm_state_proto;
7387 #ifdef CONFIG_CGROUP_NET_CLASSID
7388 case BPF_FUNC_skb_cgroup_classid:
7389 return &bpf_skb_cgroup_classid_proto;
7391 #ifdef CONFIG_SOCK_CGROUP_DATA
7392 case BPF_FUNC_skb_cgroup_id:
7393 return &bpf_skb_cgroup_id_proto;
7394 case BPF_FUNC_skb_ancestor_cgroup_id:
7395 return &bpf_skb_ancestor_cgroup_id_proto;
7398 case BPF_FUNC_sk_lookup_tcp:
7399 return &bpf_sk_lookup_tcp_proto;
7400 case BPF_FUNC_sk_lookup_udp:
7401 return &bpf_sk_lookup_udp_proto;
7402 case BPF_FUNC_sk_release:
7403 return &bpf_sk_release_proto;
7404 case BPF_FUNC_tcp_sock:
7405 return &bpf_tcp_sock_proto;
7406 case BPF_FUNC_get_listener_sock:
7407 return &bpf_get_listener_sock_proto;
7408 case BPF_FUNC_skc_lookup_tcp:
7409 return &bpf_skc_lookup_tcp_proto;
7410 case BPF_FUNC_tcp_check_syncookie:
7411 return &bpf_tcp_check_syncookie_proto;
7412 case BPF_FUNC_skb_ecn_set_ce:
7413 return &bpf_skb_ecn_set_ce_proto;
7414 case BPF_FUNC_tcp_gen_syncookie:
7415 return &bpf_tcp_gen_syncookie_proto;
7416 case BPF_FUNC_sk_assign:
7417 return &bpf_sk_assign_proto;
7420 return bpf_sk_base_func_proto(func_id);
7424 static const struct bpf_func_proto *
7425 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7428 case BPF_FUNC_perf_event_output:
7429 return &bpf_xdp_event_output_proto;
7430 case BPF_FUNC_get_smp_processor_id:
7431 return &bpf_get_smp_processor_id_proto;
7432 case BPF_FUNC_csum_diff:
7433 return &bpf_csum_diff_proto;
7434 case BPF_FUNC_xdp_adjust_head:
7435 return &bpf_xdp_adjust_head_proto;
7436 case BPF_FUNC_xdp_adjust_meta:
7437 return &bpf_xdp_adjust_meta_proto;
7438 case BPF_FUNC_redirect:
7439 return &bpf_xdp_redirect_proto;
7440 case BPF_FUNC_redirect_map:
7441 return &bpf_xdp_redirect_map_proto;
7442 case BPF_FUNC_xdp_adjust_tail:
7443 return &bpf_xdp_adjust_tail_proto;
7444 case BPF_FUNC_fib_lookup:
7445 return &bpf_xdp_fib_lookup_proto;
7446 case BPF_FUNC_check_mtu:
7447 return &bpf_xdp_check_mtu_proto;
7449 case BPF_FUNC_sk_lookup_udp:
7450 return &bpf_xdp_sk_lookup_udp_proto;
7451 case BPF_FUNC_sk_lookup_tcp:
7452 return &bpf_xdp_sk_lookup_tcp_proto;
7453 case BPF_FUNC_sk_release:
7454 return &bpf_sk_release_proto;
7455 case BPF_FUNC_skc_lookup_tcp:
7456 return &bpf_xdp_skc_lookup_tcp_proto;
7457 case BPF_FUNC_tcp_check_syncookie:
7458 return &bpf_tcp_check_syncookie_proto;
7459 case BPF_FUNC_tcp_gen_syncookie:
7460 return &bpf_tcp_gen_syncookie_proto;
7463 return bpf_sk_base_func_proto(func_id);
7467 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
7468 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
7470 static const struct bpf_func_proto *
7471 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7474 case BPF_FUNC_setsockopt:
7475 return &bpf_sock_ops_setsockopt_proto;
7476 case BPF_FUNC_getsockopt:
7477 return &bpf_sock_ops_getsockopt_proto;
7478 case BPF_FUNC_sock_ops_cb_flags_set:
7479 return &bpf_sock_ops_cb_flags_set_proto;
7480 case BPF_FUNC_sock_map_update:
7481 return &bpf_sock_map_update_proto;
7482 case BPF_FUNC_sock_hash_update:
7483 return &bpf_sock_hash_update_proto;
7484 case BPF_FUNC_get_socket_cookie:
7485 return &bpf_get_socket_cookie_sock_ops_proto;
7486 case BPF_FUNC_get_local_storage:
7487 return &bpf_get_local_storage_proto;
7488 case BPF_FUNC_perf_event_output:
7489 return &bpf_event_output_data_proto;
7490 case BPF_FUNC_sk_storage_get:
7491 return &bpf_sk_storage_get_proto;
7492 case BPF_FUNC_sk_storage_delete:
7493 return &bpf_sk_storage_delete_proto;
7495 case BPF_FUNC_load_hdr_opt:
7496 return &bpf_sock_ops_load_hdr_opt_proto;
7497 case BPF_FUNC_store_hdr_opt:
7498 return &bpf_sock_ops_store_hdr_opt_proto;
7499 case BPF_FUNC_reserve_hdr_opt:
7500 return &bpf_sock_ops_reserve_hdr_opt_proto;
7501 case BPF_FUNC_tcp_sock:
7502 return &bpf_tcp_sock_proto;
7503 #endif /* CONFIG_INET */
7505 return bpf_sk_base_func_proto(func_id);
7509 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
7510 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
7512 static const struct bpf_func_proto *
7513 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7516 case BPF_FUNC_msg_redirect_map:
7517 return &bpf_msg_redirect_map_proto;
7518 case BPF_FUNC_msg_redirect_hash:
7519 return &bpf_msg_redirect_hash_proto;
7520 case BPF_FUNC_msg_apply_bytes:
7521 return &bpf_msg_apply_bytes_proto;
7522 case BPF_FUNC_msg_cork_bytes:
7523 return &bpf_msg_cork_bytes_proto;
7524 case BPF_FUNC_msg_pull_data:
7525 return &bpf_msg_pull_data_proto;
7526 case BPF_FUNC_msg_push_data:
7527 return &bpf_msg_push_data_proto;
7528 case BPF_FUNC_msg_pop_data:
7529 return &bpf_msg_pop_data_proto;
7530 case BPF_FUNC_perf_event_output:
7531 return &bpf_event_output_data_proto;
7532 case BPF_FUNC_get_current_uid_gid:
7533 return &bpf_get_current_uid_gid_proto;
7534 case BPF_FUNC_get_current_pid_tgid:
7535 return &bpf_get_current_pid_tgid_proto;
7536 case BPF_FUNC_sk_storage_get:
7537 return &bpf_sk_storage_get_proto;
7538 case BPF_FUNC_sk_storage_delete:
7539 return &bpf_sk_storage_delete_proto;
7540 #ifdef CONFIG_CGROUPS
7541 case BPF_FUNC_get_current_cgroup_id:
7542 return &bpf_get_current_cgroup_id_proto;
7543 case BPF_FUNC_get_current_ancestor_cgroup_id:
7544 return &bpf_get_current_ancestor_cgroup_id_proto;
7546 #ifdef CONFIG_CGROUP_NET_CLASSID
7547 case BPF_FUNC_get_cgroup_classid:
7548 return &bpf_get_cgroup_classid_curr_proto;
7551 return bpf_sk_base_func_proto(func_id);
7555 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
7556 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
7558 static const struct bpf_func_proto *
7559 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7562 case BPF_FUNC_skb_store_bytes:
7563 return &bpf_skb_store_bytes_proto;
7564 case BPF_FUNC_skb_load_bytes:
7565 return &bpf_skb_load_bytes_proto;
7566 case BPF_FUNC_skb_pull_data:
7567 return &sk_skb_pull_data_proto;
7568 case BPF_FUNC_skb_change_tail:
7569 return &sk_skb_change_tail_proto;
7570 case BPF_FUNC_skb_change_head:
7571 return &sk_skb_change_head_proto;
7572 case BPF_FUNC_skb_adjust_room:
7573 return &sk_skb_adjust_room_proto;
7574 case BPF_FUNC_get_socket_cookie:
7575 return &bpf_get_socket_cookie_proto;
7576 case BPF_FUNC_get_socket_uid:
7577 return &bpf_get_socket_uid_proto;
7578 case BPF_FUNC_sk_redirect_map:
7579 return &bpf_sk_redirect_map_proto;
7580 case BPF_FUNC_sk_redirect_hash:
7581 return &bpf_sk_redirect_hash_proto;
7582 case BPF_FUNC_perf_event_output:
7583 return &bpf_skb_event_output_proto;
7585 case BPF_FUNC_sk_lookup_tcp:
7586 return &bpf_sk_lookup_tcp_proto;
7587 case BPF_FUNC_sk_lookup_udp:
7588 return &bpf_sk_lookup_udp_proto;
7589 case BPF_FUNC_sk_release:
7590 return &bpf_sk_release_proto;
7591 case BPF_FUNC_skc_lookup_tcp:
7592 return &bpf_skc_lookup_tcp_proto;
7595 return bpf_sk_base_func_proto(func_id);
7599 static const struct bpf_func_proto *
7600 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7603 case BPF_FUNC_skb_load_bytes:
7604 return &bpf_flow_dissector_load_bytes_proto;
7606 return bpf_sk_base_func_proto(func_id);
7610 static const struct bpf_func_proto *
7611 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7614 case BPF_FUNC_skb_load_bytes:
7615 return &bpf_skb_load_bytes_proto;
7616 case BPF_FUNC_skb_pull_data:
7617 return &bpf_skb_pull_data_proto;
7618 case BPF_FUNC_csum_diff:
7619 return &bpf_csum_diff_proto;
7620 case BPF_FUNC_get_cgroup_classid:
7621 return &bpf_get_cgroup_classid_proto;
7622 case BPF_FUNC_get_route_realm:
7623 return &bpf_get_route_realm_proto;
7624 case BPF_FUNC_get_hash_recalc:
7625 return &bpf_get_hash_recalc_proto;
7626 case BPF_FUNC_perf_event_output:
7627 return &bpf_skb_event_output_proto;
7628 case BPF_FUNC_get_smp_processor_id:
7629 return &bpf_get_smp_processor_id_proto;
7630 case BPF_FUNC_skb_under_cgroup:
7631 return &bpf_skb_under_cgroup_proto;
7633 return bpf_sk_base_func_proto(func_id);
7637 static const struct bpf_func_proto *
7638 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7641 case BPF_FUNC_lwt_push_encap:
7642 return &bpf_lwt_in_push_encap_proto;
7644 return lwt_out_func_proto(func_id, prog);
7648 static const struct bpf_func_proto *
7649 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7652 case BPF_FUNC_skb_get_tunnel_key:
7653 return &bpf_skb_get_tunnel_key_proto;
7654 case BPF_FUNC_skb_set_tunnel_key:
7655 return bpf_get_skb_set_tunnel_proto(func_id);
7656 case BPF_FUNC_skb_get_tunnel_opt:
7657 return &bpf_skb_get_tunnel_opt_proto;
7658 case BPF_FUNC_skb_set_tunnel_opt:
7659 return bpf_get_skb_set_tunnel_proto(func_id);
7660 case BPF_FUNC_redirect:
7661 return &bpf_redirect_proto;
7662 case BPF_FUNC_clone_redirect:
7663 return &bpf_clone_redirect_proto;
7664 case BPF_FUNC_skb_change_tail:
7665 return &bpf_skb_change_tail_proto;
7666 case BPF_FUNC_skb_change_head:
7667 return &bpf_skb_change_head_proto;
7668 case BPF_FUNC_skb_store_bytes:
7669 return &bpf_skb_store_bytes_proto;
7670 case BPF_FUNC_csum_update:
7671 return &bpf_csum_update_proto;
7672 case BPF_FUNC_csum_level:
7673 return &bpf_csum_level_proto;
7674 case BPF_FUNC_l3_csum_replace:
7675 return &bpf_l3_csum_replace_proto;
7676 case BPF_FUNC_l4_csum_replace:
7677 return &bpf_l4_csum_replace_proto;
7678 case BPF_FUNC_set_hash_invalid:
7679 return &bpf_set_hash_invalid_proto;
7680 case BPF_FUNC_lwt_push_encap:
7681 return &bpf_lwt_xmit_push_encap_proto;
7683 return lwt_out_func_proto(func_id, prog);
7687 static const struct bpf_func_proto *
7688 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7691 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7692 case BPF_FUNC_lwt_seg6_store_bytes:
7693 return &bpf_lwt_seg6_store_bytes_proto;
7694 case BPF_FUNC_lwt_seg6_action:
7695 return &bpf_lwt_seg6_action_proto;
7696 case BPF_FUNC_lwt_seg6_adjust_srh:
7697 return &bpf_lwt_seg6_adjust_srh_proto;
7700 return lwt_out_func_proto(func_id, prog);
7704 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
7705 const struct bpf_prog *prog,
7706 struct bpf_insn_access_aux *info)
7708 const int size_default = sizeof(__u32);
7710 if (off < 0 || off >= sizeof(struct __sk_buff))
7713 /* The verifier guarantees that size > 0. */
7714 if (off % size != 0)
7718 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7719 if (off + size > offsetofend(struct __sk_buff, cb[4]))
7722 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
7723 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
7724 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
7725 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
7726 case bpf_ctx_range(struct __sk_buff, data):
7727 case bpf_ctx_range(struct __sk_buff, data_meta):
7728 case bpf_ctx_range(struct __sk_buff, data_end):
7729 if (size != size_default)
7732 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
7734 case bpf_ctx_range(struct __sk_buff, tstamp):
7735 if (size != sizeof(__u64))
7738 case offsetof(struct __sk_buff, sk):
7739 if (type == BPF_WRITE || size != sizeof(__u64))
7741 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
7744 /* Only narrow read access allowed for now. */
7745 if (type == BPF_WRITE) {
7746 if (size != size_default)
7749 bpf_ctx_record_field_size(info, size_default);
7750 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
7758 static bool sk_filter_is_valid_access(int off, int size,
7759 enum bpf_access_type type,
7760 const struct bpf_prog *prog,
7761 struct bpf_insn_access_aux *info)
7764 case bpf_ctx_range(struct __sk_buff, tc_classid):
7765 case bpf_ctx_range(struct __sk_buff, data):
7766 case bpf_ctx_range(struct __sk_buff, data_meta):
7767 case bpf_ctx_range(struct __sk_buff, data_end):
7768 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7769 case bpf_ctx_range(struct __sk_buff, tstamp):
7770 case bpf_ctx_range(struct __sk_buff, wire_len):
7774 if (type == BPF_WRITE) {
7776 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7783 return bpf_skb_is_valid_access(off, size, type, prog, info);
7786 static bool cg_skb_is_valid_access(int off, int size,
7787 enum bpf_access_type type,
7788 const struct bpf_prog *prog,
7789 struct bpf_insn_access_aux *info)
7792 case bpf_ctx_range(struct __sk_buff, tc_classid):
7793 case bpf_ctx_range(struct __sk_buff, data_meta):
7794 case bpf_ctx_range(struct __sk_buff, wire_len):
7796 case bpf_ctx_range(struct __sk_buff, data):
7797 case bpf_ctx_range(struct __sk_buff, data_end):
7803 if (type == BPF_WRITE) {
7805 case bpf_ctx_range(struct __sk_buff, mark):
7806 case bpf_ctx_range(struct __sk_buff, priority):
7807 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7809 case bpf_ctx_range(struct __sk_buff, tstamp):
7819 case bpf_ctx_range(struct __sk_buff, data):
7820 info->reg_type = PTR_TO_PACKET;
7822 case bpf_ctx_range(struct __sk_buff, data_end):
7823 info->reg_type = PTR_TO_PACKET_END;
7827 return bpf_skb_is_valid_access(off, size, type, prog, info);
7830 static bool lwt_is_valid_access(int off, int size,
7831 enum bpf_access_type type,
7832 const struct bpf_prog *prog,
7833 struct bpf_insn_access_aux *info)
7836 case bpf_ctx_range(struct __sk_buff, tc_classid):
7837 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7838 case bpf_ctx_range(struct __sk_buff, data_meta):
7839 case bpf_ctx_range(struct __sk_buff, tstamp):
7840 case bpf_ctx_range(struct __sk_buff, wire_len):
7844 if (type == BPF_WRITE) {
7846 case bpf_ctx_range(struct __sk_buff, mark):
7847 case bpf_ctx_range(struct __sk_buff, priority):
7848 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7856 case bpf_ctx_range(struct __sk_buff, data):
7857 info->reg_type = PTR_TO_PACKET;
7859 case bpf_ctx_range(struct __sk_buff, data_end):
7860 info->reg_type = PTR_TO_PACKET_END;
7864 return bpf_skb_is_valid_access(off, size, type, prog, info);
7867 /* Attach type specific accesses */
7868 static bool __sock_filter_check_attach_type(int off,
7869 enum bpf_access_type access_type,
7870 enum bpf_attach_type attach_type)
7873 case offsetof(struct bpf_sock, bound_dev_if):
7874 case offsetof(struct bpf_sock, mark):
7875 case offsetof(struct bpf_sock, priority):
7876 switch (attach_type) {
7877 case BPF_CGROUP_INET_SOCK_CREATE:
7878 case BPF_CGROUP_INET_SOCK_RELEASE:
7883 case bpf_ctx_range(struct bpf_sock, src_ip4):
7884 switch (attach_type) {
7885 case BPF_CGROUP_INET4_POST_BIND:
7890 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7891 switch (attach_type) {
7892 case BPF_CGROUP_INET6_POST_BIND:
7897 case bpf_ctx_range(struct bpf_sock, src_port):
7898 switch (attach_type) {
7899 case BPF_CGROUP_INET4_POST_BIND:
7900 case BPF_CGROUP_INET6_POST_BIND:
7907 return access_type == BPF_READ;
7912 bool bpf_sock_common_is_valid_access(int off, int size,
7913 enum bpf_access_type type,
7914 struct bpf_insn_access_aux *info)
7917 case bpf_ctx_range_till(struct bpf_sock, type, priority):
7920 return bpf_sock_is_valid_access(off, size, type, info);
7924 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7925 struct bpf_insn_access_aux *info)
7927 const int size_default = sizeof(__u32);
7929 if (off < 0 || off >= sizeof(struct bpf_sock))
7931 if (off % size != 0)
7935 case offsetof(struct bpf_sock, state):
7936 case offsetof(struct bpf_sock, family):
7937 case offsetof(struct bpf_sock, type):
7938 case offsetof(struct bpf_sock, protocol):
7939 case offsetof(struct bpf_sock, dst_port):
7940 case offsetof(struct bpf_sock, src_port):
7941 case offsetof(struct bpf_sock, rx_queue_mapping):
7942 case bpf_ctx_range(struct bpf_sock, src_ip4):
7943 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7944 case bpf_ctx_range(struct bpf_sock, dst_ip4):
7945 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
7946 bpf_ctx_record_field_size(info, size_default);
7947 return bpf_ctx_narrow_access_ok(off, size, size_default);
7950 return size == size_default;
7953 static bool sock_filter_is_valid_access(int off, int size,
7954 enum bpf_access_type type,
7955 const struct bpf_prog *prog,
7956 struct bpf_insn_access_aux *info)
7958 if (!bpf_sock_is_valid_access(off, size, type, info))
7960 return __sock_filter_check_attach_type(off, type,
7961 prog->expected_attach_type);
7964 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
7965 const struct bpf_prog *prog)
7967 /* Neither direct read nor direct write requires any preliminary
7973 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
7974 const struct bpf_prog *prog, int drop_verdict)
7976 struct bpf_insn *insn = insn_buf;
7981 /* if (!skb->cloned)
7984 * (Fast-path, otherwise approximation that we might be
7985 * a clone, do the rest in helper.)
7987 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
7988 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
7989 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
7991 /* ret = bpf_skb_pull_data(skb, 0); */
7992 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
7993 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
7994 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
7995 BPF_FUNC_skb_pull_data);
7998 * return TC_ACT_SHOT;
8000 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
8001 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
8002 *insn++ = BPF_EXIT_INSN();
8005 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
8007 *insn++ = prog->insnsi[0];
8009 return insn - insn_buf;
8012 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
8013 struct bpf_insn *insn_buf)
8015 bool indirect = BPF_MODE(orig->code) == BPF_IND;
8016 struct bpf_insn *insn = insn_buf;
8019 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
8021 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
8023 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
8025 /* We're guaranteed here that CTX is in R6. */
8026 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
8028 switch (BPF_SIZE(orig->code)) {
8030 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
8033 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
8036 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
8040 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
8041 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
8042 *insn++ = BPF_EXIT_INSN();
8044 return insn - insn_buf;
8047 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8048 const struct bpf_prog *prog)
8050 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8053 static bool tc_cls_act_is_valid_access(int off, int size,
8054 enum bpf_access_type type,
8055 const struct bpf_prog *prog,
8056 struct bpf_insn_access_aux *info)
8058 if (type == BPF_WRITE) {
8060 case bpf_ctx_range(struct __sk_buff, mark):
8061 case bpf_ctx_range(struct __sk_buff, tc_index):
8062 case bpf_ctx_range(struct __sk_buff, priority):
8063 case bpf_ctx_range(struct __sk_buff, tc_classid):
8064 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8065 case bpf_ctx_range(struct __sk_buff, tstamp):
8066 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8074 case bpf_ctx_range(struct __sk_buff, data):
8075 info->reg_type = PTR_TO_PACKET;
8077 case bpf_ctx_range(struct __sk_buff, data_meta):
8078 info->reg_type = PTR_TO_PACKET_META;
8080 case bpf_ctx_range(struct __sk_buff, data_end):
8081 info->reg_type = PTR_TO_PACKET_END;
8083 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8087 return bpf_skb_is_valid_access(off, size, type, prog, info);
8090 static bool __is_valid_xdp_access(int off, int size)
8092 if (off < 0 || off >= sizeof(struct xdp_md))
8094 if (off % size != 0)
8096 if (size != sizeof(__u32))
8102 static bool xdp_is_valid_access(int off, int size,
8103 enum bpf_access_type type,
8104 const struct bpf_prog *prog,
8105 struct bpf_insn_access_aux *info)
8107 if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8109 case offsetof(struct xdp_md, egress_ifindex):
8114 if (type == BPF_WRITE) {
8115 if (bpf_prog_is_dev_bound(prog->aux)) {
8117 case offsetof(struct xdp_md, rx_queue_index):
8118 return __is_valid_xdp_access(off, size);
8125 case offsetof(struct xdp_md, data):
8126 info->reg_type = PTR_TO_PACKET;
8128 case offsetof(struct xdp_md, data_meta):
8129 info->reg_type = PTR_TO_PACKET_META;
8131 case offsetof(struct xdp_md, data_end):
8132 info->reg_type = PTR_TO_PACKET_END;
8136 return __is_valid_xdp_access(off, size);
8139 void bpf_warn_invalid_xdp_action(u32 act)
8141 const u32 act_max = XDP_REDIRECT;
8143 WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
8144 act > act_max ? "Illegal" : "Driver unsupported",
8147 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
8149 static bool sock_addr_is_valid_access(int off, int size,
8150 enum bpf_access_type type,
8151 const struct bpf_prog *prog,
8152 struct bpf_insn_access_aux *info)
8154 const int size_default = sizeof(__u32);
8156 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
8158 if (off % size != 0)
8161 /* Disallow access to IPv6 fields from IPv4 contex and vise
8165 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8166 switch (prog->expected_attach_type) {
8167 case BPF_CGROUP_INET4_BIND:
8168 case BPF_CGROUP_INET4_CONNECT:
8169 case BPF_CGROUP_INET4_GETPEERNAME:
8170 case BPF_CGROUP_INET4_GETSOCKNAME:
8171 case BPF_CGROUP_UDP4_SENDMSG:
8172 case BPF_CGROUP_UDP4_RECVMSG:
8178 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8179 switch (prog->expected_attach_type) {
8180 case BPF_CGROUP_INET6_BIND:
8181 case BPF_CGROUP_INET6_CONNECT:
8182 case BPF_CGROUP_INET6_GETPEERNAME:
8183 case BPF_CGROUP_INET6_GETSOCKNAME:
8184 case BPF_CGROUP_UDP6_SENDMSG:
8185 case BPF_CGROUP_UDP6_RECVMSG:
8191 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8192 switch (prog->expected_attach_type) {
8193 case BPF_CGROUP_UDP4_SENDMSG:
8199 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8201 switch (prog->expected_attach_type) {
8202 case BPF_CGROUP_UDP6_SENDMSG:
8211 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8212 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8213 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8214 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8216 case bpf_ctx_range(struct bpf_sock_addr, user_port):
8217 if (type == BPF_READ) {
8218 bpf_ctx_record_field_size(info, size_default);
8220 if (bpf_ctx_wide_access_ok(off, size,
8221 struct bpf_sock_addr,
8225 if (bpf_ctx_wide_access_ok(off, size,
8226 struct bpf_sock_addr,
8230 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8233 if (bpf_ctx_wide_access_ok(off, size,
8234 struct bpf_sock_addr,
8238 if (bpf_ctx_wide_access_ok(off, size,
8239 struct bpf_sock_addr,
8243 if (size != size_default)
8247 case offsetof(struct bpf_sock_addr, sk):
8248 if (type != BPF_READ)
8250 if (size != sizeof(__u64))
8252 info->reg_type = PTR_TO_SOCKET;
8255 if (type == BPF_READ) {
8256 if (size != size_default)
8266 static bool sock_ops_is_valid_access(int off, int size,
8267 enum bpf_access_type type,
8268 const struct bpf_prog *prog,
8269 struct bpf_insn_access_aux *info)
8271 const int size_default = sizeof(__u32);
8273 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
8276 /* The verifier guarantees that size > 0. */
8277 if (off % size != 0)
8280 if (type == BPF_WRITE) {
8282 case offsetof(struct bpf_sock_ops, reply):
8283 case offsetof(struct bpf_sock_ops, sk_txhash):
8284 if (size != size_default)
8292 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
8294 if (size != sizeof(__u64))
8297 case offsetof(struct bpf_sock_ops, sk):
8298 if (size != sizeof(__u64))
8300 info->reg_type = PTR_TO_SOCKET_OR_NULL;
8302 case offsetof(struct bpf_sock_ops, skb_data):
8303 if (size != sizeof(__u64))
8305 info->reg_type = PTR_TO_PACKET;
8307 case offsetof(struct bpf_sock_ops, skb_data_end):
8308 if (size != sizeof(__u64))
8310 info->reg_type = PTR_TO_PACKET_END;
8312 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
8313 bpf_ctx_record_field_size(info, size_default);
8314 return bpf_ctx_narrow_access_ok(off, size,
8317 if (size != size_default)
8326 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
8327 const struct bpf_prog *prog)
8329 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
8332 static bool sk_skb_is_valid_access(int off, int size,
8333 enum bpf_access_type type,
8334 const struct bpf_prog *prog,
8335 struct bpf_insn_access_aux *info)
8338 case bpf_ctx_range(struct __sk_buff, tc_classid):
8339 case bpf_ctx_range(struct __sk_buff, data_meta):
8340 case bpf_ctx_range(struct __sk_buff, tstamp):
8341 case bpf_ctx_range(struct __sk_buff, wire_len):
8345 if (type == BPF_WRITE) {
8347 case bpf_ctx_range(struct __sk_buff, tc_index):
8348 case bpf_ctx_range(struct __sk_buff, priority):
8356 case bpf_ctx_range(struct __sk_buff, mark):
8358 case bpf_ctx_range(struct __sk_buff, data):
8359 info->reg_type = PTR_TO_PACKET;
8361 case bpf_ctx_range(struct __sk_buff, data_end):
8362 info->reg_type = PTR_TO_PACKET_END;
8366 return bpf_skb_is_valid_access(off, size, type, prog, info);
8369 static bool sk_msg_is_valid_access(int off, int size,
8370 enum bpf_access_type type,
8371 const struct bpf_prog *prog,
8372 struct bpf_insn_access_aux *info)
8374 if (type == BPF_WRITE)
8377 if (off % size != 0)
8381 case offsetof(struct sk_msg_md, data):
8382 info->reg_type = PTR_TO_PACKET;
8383 if (size != sizeof(__u64))
8386 case offsetof(struct sk_msg_md, data_end):
8387 info->reg_type = PTR_TO_PACKET_END;
8388 if (size != sizeof(__u64))
8391 case offsetof(struct sk_msg_md, sk):
8392 if (size != sizeof(__u64))
8394 info->reg_type = PTR_TO_SOCKET;
8396 case bpf_ctx_range(struct sk_msg_md, family):
8397 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
8398 case bpf_ctx_range(struct sk_msg_md, local_ip4):
8399 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
8400 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
8401 case bpf_ctx_range(struct sk_msg_md, remote_port):
8402 case bpf_ctx_range(struct sk_msg_md, local_port):
8403 case bpf_ctx_range(struct sk_msg_md, size):
8404 if (size != sizeof(__u32))
8413 static bool flow_dissector_is_valid_access(int off, int size,
8414 enum bpf_access_type type,
8415 const struct bpf_prog *prog,
8416 struct bpf_insn_access_aux *info)
8418 const int size_default = sizeof(__u32);
8420 if (off < 0 || off >= sizeof(struct __sk_buff))
8423 if (type == BPF_WRITE)
8427 case bpf_ctx_range(struct __sk_buff, data):
8428 if (size != size_default)
8430 info->reg_type = PTR_TO_PACKET;
8432 case bpf_ctx_range(struct __sk_buff, data_end):
8433 if (size != size_default)
8435 info->reg_type = PTR_TO_PACKET_END;
8437 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8438 if (size != sizeof(__u64))
8440 info->reg_type = PTR_TO_FLOW_KEYS;
8447 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
8448 const struct bpf_insn *si,
8449 struct bpf_insn *insn_buf,
8450 struct bpf_prog *prog,
8454 struct bpf_insn *insn = insn_buf;
8457 case offsetof(struct __sk_buff, data):
8458 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
8459 si->dst_reg, si->src_reg,
8460 offsetof(struct bpf_flow_dissector, data));
8463 case offsetof(struct __sk_buff, data_end):
8464 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
8465 si->dst_reg, si->src_reg,
8466 offsetof(struct bpf_flow_dissector, data_end));
8469 case offsetof(struct __sk_buff, flow_keys):
8470 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
8471 si->dst_reg, si->src_reg,
8472 offsetof(struct bpf_flow_dissector, flow_keys));
8476 return insn - insn_buf;
8479 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si,
8480 struct bpf_insn *insn)
8482 /* si->dst_reg = skb_shinfo(SKB); */
8483 #ifdef NET_SKBUFF_DATA_USES_OFFSET
8484 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8485 BPF_REG_AX, si->src_reg,
8486 offsetof(struct sk_buff, end));
8487 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
8488 si->dst_reg, si->src_reg,
8489 offsetof(struct sk_buff, head));
8490 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
8492 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8493 si->dst_reg, si->src_reg,
8494 offsetof(struct sk_buff, end));
8500 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
8501 const struct bpf_insn *si,
8502 struct bpf_insn *insn_buf,
8503 struct bpf_prog *prog, u32 *target_size)
8505 struct bpf_insn *insn = insn_buf;
8509 case offsetof(struct __sk_buff, len):
8510 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8511 bpf_target_off(struct sk_buff, len, 4,
8515 case offsetof(struct __sk_buff, protocol):
8516 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8517 bpf_target_off(struct sk_buff, protocol, 2,
8521 case offsetof(struct __sk_buff, vlan_proto):
8522 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8523 bpf_target_off(struct sk_buff, vlan_proto, 2,
8527 case offsetof(struct __sk_buff, priority):
8528 if (type == BPF_WRITE)
8529 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8530 bpf_target_off(struct sk_buff, priority, 4,
8533 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8534 bpf_target_off(struct sk_buff, priority, 4,
8538 case offsetof(struct __sk_buff, ingress_ifindex):
8539 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8540 bpf_target_off(struct sk_buff, skb_iif, 4,
8544 case offsetof(struct __sk_buff, ifindex):
8545 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8546 si->dst_reg, si->src_reg,
8547 offsetof(struct sk_buff, dev));
8548 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
8549 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8550 bpf_target_off(struct net_device, ifindex, 4,
8554 case offsetof(struct __sk_buff, hash):
8555 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8556 bpf_target_off(struct sk_buff, hash, 4,
8560 case offsetof(struct __sk_buff, mark):
8561 if (type == BPF_WRITE)
8562 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8563 bpf_target_off(struct sk_buff, mark, 4,
8566 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8567 bpf_target_off(struct sk_buff, mark, 4,
8571 case offsetof(struct __sk_buff, pkt_type):
8573 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8575 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
8576 #ifdef __BIG_ENDIAN_BITFIELD
8577 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
8581 case offsetof(struct __sk_buff, queue_mapping):
8582 if (type == BPF_WRITE) {
8583 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
8584 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8585 bpf_target_off(struct sk_buff,
8589 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8590 bpf_target_off(struct sk_buff,
8596 case offsetof(struct __sk_buff, vlan_present):
8598 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8599 PKT_VLAN_PRESENT_OFFSET());
8600 if (PKT_VLAN_PRESENT_BIT)
8601 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
8602 if (PKT_VLAN_PRESENT_BIT < 7)
8603 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
8606 case offsetof(struct __sk_buff, vlan_tci):
8607 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8608 bpf_target_off(struct sk_buff, vlan_tci, 2,
8612 case offsetof(struct __sk_buff, cb[0]) ...
8613 offsetofend(struct __sk_buff, cb[4]) - 1:
8614 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
8615 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
8616 offsetof(struct qdisc_skb_cb, data)) %
8619 prog->cb_access = 1;
8621 off -= offsetof(struct __sk_buff, cb[0]);
8622 off += offsetof(struct sk_buff, cb);
8623 off += offsetof(struct qdisc_skb_cb, data);
8624 if (type == BPF_WRITE)
8625 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
8628 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
8632 case offsetof(struct __sk_buff, tc_classid):
8633 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
8636 off -= offsetof(struct __sk_buff, tc_classid);
8637 off += offsetof(struct sk_buff, cb);
8638 off += offsetof(struct qdisc_skb_cb, tc_classid);
8640 if (type == BPF_WRITE)
8641 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
8644 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
8648 case offsetof(struct __sk_buff, data):
8649 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
8650 si->dst_reg, si->src_reg,
8651 offsetof(struct sk_buff, data));
8654 case offsetof(struct __sk_buff, data_meta):
8656 off -= offsetof(struct __sk_buff, data_meta);
8657 off += offsetof(struct sk_buff, cb);
8658 off += offsetof(struct bpf_skb_data_end, data_meta);
8659 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8663 case offsetof(struct __sk_buff, data_end):
8665 off -= offsetof(struct __sk_buff, data_end);
8666 off += offsetof(struct sk_buff, cb);
8667 off += offsetof(struct bpf_skb_data_end, data_end);
8668 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8672 case offsetof(struct __sk_buff, tc_index):
8673 #ifdef CONFIG_NET_SCHED
8674 if (type == BPF_WRITE)
8675 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8676 bpf_target_off(struct sk_buff, tc_index, 2,
8679 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8680 bpf_target_off(struct sk_buff, tc_index, 2,
8684 if (type == BPF_WRITE)
8685 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
8687 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8691 case offsetof(struct __sk_buff, napi_id):
8692 #if defined(CONFIG_NET_RX_BUSY_POLL)
8693 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8694 bpf_target_off(struct sk_buff, napi_id, 4,
8696 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
8697 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8700 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8703 case offsetof(struct __sk_buff, family):
8704 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
8706 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8707 si->dst_reg, si->src_reg,
8708 offsetof(struct sk_buff, sk));
8709 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8710 bpf_target_off(struct sock_common,
8714 case offsetof(struct __sk_buff, remote_ip4):
8715 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
8717 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8718 si->dst_reg, si->src_reg,
8719 offsetof(struct sk_buff, sk));
8720 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8721 bpf_target_off(struct sock_common,
8725 case offsetof(struct __sk_buff, local_ip4):
8726 BUILD_BUG_ON(sizeof_field(struct sock_common,
8727 skc_rcv_saddr) != 4);
8729 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8730 si->dst_reg, si->src_reg,
8731 offsetof(struct sk_buff, sk));
8732 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8733 bpf_target_off(struct sock_common,
8737 case offsetof(struct __sk_buff, remote_ip6[0]) ...
8738 offsetof(struct __sk_buff, remote_ip6[3]):
8739 #if IS_ENABLED(CONFIG_IPV6)
8740 BUILD_BUG_ON(sizeof_field(struct sock_common,
8741 skc_v6_daddr.s6_addr32[0]) != 4);
8744 off -= offsetof(struct __sk_buff, remote_ip6[0]);
8746 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8747 si->dst_reg, si->src_reg,
8748 offsetof(struct sk_buff, sk));
8749 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8750 offsetof(struct sock_common,
8751 skc_v6_daddr.s6_addr32[0]) +
8754 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8757 case offsetof(struct __sk_buff, local_ip6[0]) ...
8758 offsetof(struct __sk_buff, local_ip6[3]):
8759 #if IS_ENABLED(CONFIG_IPV6)
8760 BUILD_BUG_ON(sizeof_field(struct sock_common,
8761 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8764 off -= offsetof(struct __sk_buff, local_ip6[0]);
8766 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8767 si->dst_reg, si->src_reg,
8768 offsetof(struct sk_buff, sk));
8769 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8770 offsetof(struct sock_common,
8771 skc_v6_rcv_saddr.s6_addr32[0]) +
8774 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8778 case offsetof(struct __sk_buff, remote_port):
8779 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
8781 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8782 si->dst_reg, si->src_reg,
8783 offsetof(struct sk_buff, sk));
8784 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8785 bpf_target_off(struct sock_common,
8788 #ifndef __BIG_ENDIAN_BITFIELD
8789 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8793 case offsetof(struct __sk_buff, local_port):
8794 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
8796 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8797 si->dst_reg, si->src_reg,
8798 offsetof(struct sk_buff, sk));
8799 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8800 bpf_target_off(struct sock_common,
8801 skc_num, 2, target_size));
8804 case offsetof(struct __sk_buff, tstamp):
8805 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
8807 if (type == BPF_WRITE)
8808 *insn++ = BPF_STX_MEM(BPF_DW,
8809 si->dst_reg, si->src_reg,
8810 bpf_target_off(struct sk_buff,
8814 *insn++ = BPF_LDX_MEM(BPF_DW,
8815 si->dst_reg, si->src_reg,
8816 bpf_target_off(struct sk_buff,
8821 case offsetof(struct __sk_buff, gso_segs):
8822 insn = bpf_convert_shinfo_access(si, insn);
8823 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
8824 si->dst_reg, si->dst_reg,
8825 bpf_target_off(struct skb_shared_info,
8829 case offsetof(struct __sk_buff, gso_size):
8830 insn = bpf_convert_shinfo_access(si, insn);
8831 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
8832 si->dst_reg, si->dst_reg,
8833 bpf_target_off(struct skb_shared_info,
8837 case offsetof(struct __sk_buff, wire_len):
8838 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
8841 off -= offsetof(struct __sk_buff, wire_len);
8842 off += offsetof(struct sk_buff, cb);
8843 off += offsetof(struct qdisc_skb_cb, pkt_len);
8845 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
8848 case offsetof(struct __sk_buff, sk):
8849 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8850 si->dst_reg, si->src_reg,
8851 offsetof(struct sk_buff, sk));
8855 return insn - insn_buf;
8858 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
8859 const struct bpf_insn *si,
8860 struct bpf_insn *insn_buf,
8861 struct bpf_prog *prog, u32 *target_size)
8863 struct bpf_insn *insn = insn_buf;
8867 case offsetof(struct bpf_sock, bound_dev_if):
8868 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
8870 if (type == BPF_WRITE)
8871 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8872 offsetof(struct sock, sk_bound_dev_if));
8874 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8875 offsetof(struct sock, sk_bound_dev_if));
8878 case offsetof(struct bpf_sock, mark):
8879 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
8881 if (type == BPF_WRITE)
8882 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8883 offsetof(struct sock, sk_mark));
8885 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8886 offsetof(struct sock, sk_mark));
8889 case offsetof(struct bpf_sock, priority):
8890 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
8892 if (type == BPF_WRITE)
8893 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8894 offsetof(struct sock, sk_priority));
8896 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8897 offsetof(struct sock, sk_priority));
8900 case offsetof(struct bpf_sock, family):
8901 *insn++ = BPF_LDX_MEM(
8902 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
8903 si->dst_reg, si->src_reg,
8904 bpf_target_off(struct sock_common,
8906 sizeof_field(struct sock_common,
8911 case offsetof(struct bpf_sock, type):
8912 *insn++ = BPF_LDX_MEM(
8913 BPF_FIELD_SIZEOF(struct sock, sk_type),
8914 si->dst_reg, si->src_reg,
8915 bpf_target_off(struct sock, sk_type,
8916 sizeof_field(struct sock, sk_type),
8920 case offsetof(struct bpf_sock, protocol):
8921 *insn++ = BPF_LDX_MEM(
8922 BPF_FIELD_SIZEOF(struct sock, sk_protocol),
8923 si->dst_reg, si->src_reg,
8924 bpf_target_off(struct sock, sk_protocol,
8925 sizeof_field(struct sock, sk_protocol),
8929 case offsetof(struct bpf_sock, src_ip4):
8930 *insn++ = BPF_LDX_MEM(
8931 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8932 bpf_target_off(struct sock_common, skc_rcv_saddr,
8933 sizeof_field(struct sock_common,
8938 case offsetof(struct bpf_sock, dst_ip4):
8939 *insn++ = BPF_LDX_MEM(
8940 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8941 bpf_target_off(struct sock_common, skc_daddr,
8942 sizeof_field(struct sock_common,
8947 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8948 #if IS_ENABLED(CONFIG_IPV6)
8950 off -= offsetof(struct bpf_sock, src_ip6[0]);
8951 *insn++ = BPF_LDX_MEM(
8952 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8955 skc_v6_rcv_saddr.s6_addr32[0],
8956 sizeof_field(struct sock_common,
8957 skc_v6_rcv_saddr.s6_addr32[0]),
8958 target_size) + off);
8961 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8965 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8966 #if IS_ENABLED(CONFIG_IPV6)
8968 off -= offsetof(struct bpf_sock, dst_ip6[0]);
8969 *insn++ = BPF_LDX_MEM(
8970 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8971 bpf_target_off(struct sock_common,
8972 skc_v6_daddr.s6_addr32[0],
8973 sizeof_field(struct sock_common,
8974 skc_v6_daddr.s6_addr32[0]),
8975 target_size) + off);
8977 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8982 case offsetof(struct bpf_sock, src_port):
8983 *insn++ = BPF_LDX_MEM(
8984 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
8985 si->dst_reg, si->src_reg,
8986 bpf_target_off(struct sock_common, skc_num,
8987 sizeof_field(struct sock_common,
8992 case offsetof(struct bpf_sock, dst_port):
8993 *insn++ = BPF_LDX_MEM(
8994 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
8995 si->dst_reg, si->src_reg,
8996 bpf_target_off(struct sock_common, skc_dport,
8997 sizeof_field(struct sock_common,
9002 case offsetof(struct bpf_sock, state):
9003 *insn++ = BPF_LDX_MEM(
9004 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
9005 si->dst_reg, si->src_reg,
9006 bpf_target_off(struct sock_common, skc_state,
9007 sizeof_field(struct sock_common,
9011 case offsetof(struct bpf_sock, rx_queue_mapping):
9012 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
9013 *insn++ = BPF_LDX_MEM(
9014 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
9015 si->dst_reg, si->src_reg,
9016 bpf_target_off(struct sock, sk_rx_queue_mapping,
9017 sizeof_field(struct sock,
9018 sk_rx_queue_mapping),
9020 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
9022 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9024 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9030 return insn - insn_buf;
9033 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
9034 const struct bpf_insn *si,
9035 struct bpf_insn *insn_buf,
9036 struct bpf_prog *prog, u32 *target_size)
9038 struct bpf_insn *insn = insn_buf;
9041 case offsetof(struct __sk_buff, ifindex):
9042 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9043 si->dst_reg, si->src_reg,
9044 offsetof(struct sk_buff, dev));
9045 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9046 bpf_target_off(struct net_device, ifindex, 4,
9050 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9054 return insn - insn_buf;
9057 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
9058 const struct bpf_insn *si,
9059 struct bpf_insn *insn_buf,
9060 struct bpf_prog *prog, u32 *target_size)
9062 struct bpf_insn *insn = insn_buf;
9065 case offsetof(struct xdp_md, data):
9066 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
9067 si->dst_reg, si->src_reg,
9068 offsetof(struct xdp_buff, data));
9070 case offsetof(struct xdp_md, data_meta):
9071 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
9072 si->dst_reg, si->src_reg,
9073 offsetof(struct xdp_buff, data_meta));
9075 case offsetof(struct xdp_md, data_end):
9076 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
9077 si->dst_reg, si->src_reg,
9078 offsetof(struct xdp_buff, data_end));
9080 case offsetof(struct xdp_md, ingress_ifindex):
9081 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9082 si->dst_reg, si->src_reg,
9083 offsetof(struct xdp_buff, rxq));
9084 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
9085 si->dst_reg, si->dst_reg,
9086 offsetof(struct xdp_rxq_info, dev));
9087 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9088 offsetof(struct net_device, ifindex));
9090 case offsetof(struct xdp_md, rx_queue_index):
9091 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9092 si->dst_reg, si->src_reg,
9093 offsetof(struct xdp_buff, rxq));
9094 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9095 offsetof(struct xdp_rxq_info,
9098 case offsetof(struct xdp_md, egress_ifindex):
9099 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
9100 si->dst_reg, si->src_reg,
9101 offsetof(struct xdp_buff, txq));
9102 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
9103 si->dst_reg, si->dst_reg,
9104 offsetof(struct xdp_txq_info, dev));
9105 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9106 offsetof(struct net_device, ifindex));
9110 return insn - insn_buf;
9113 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
9114 * context Structure, F is Field in context structure that contains a pointer
9115 * to Nested Structure of type NS that has the field NF.
9117 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
9118 * sure that SIZE is not greater than actual size of S.F.NF.
9120 * If offset OFF is provided, the load happens from that offset relative to
9123 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
9125 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
9126 si->src_reg, offsetof(S, F)); \
9127 *insn++ = BPF_LDX_MEM( \
9128 SIZE, si->dst_reg, si->dst_reg, \
9129 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
9134 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
9135 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
9136 BPF_FIELD_SIZEOF(NS, NF), 0)
9138 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
9139 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
9141 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
9142 * "register" since two registers available in convert_ctx_access are not
9143 * enough: we can't override neither SRC, since it contains value to store, nor
9144 * DST since it contains pointer to context that may be used by later
9145 * instructions. But we need a temporary place to save pointer to nested
9146 * structure whose field we want to store to.
9148 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
9150 int tmp_reg = BPF_REG_9; \
9151 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
9153 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
9155 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
9157 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
9158 si->dst_reg, offsetof(S, F)); \
9159 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \
9160 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
9163 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
9167 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
9170 if (type == BPF_WRITE) { \
9171 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
9174 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
9175 S, NS, F, NF, SIZE, OFF); \
9179 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
9180 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
9181 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
9183 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
9184 const struct bpf_insn *si,
9185 struct bpf_insn *insn_buf,
9186 struct bpf_prog *prog, u32 *target_size)
9188 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
9189 struct bpf_insn *insn = insn_buf;
9192 case offsetof(struct bpf_sock_addr, user_family):
9193 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9194 struct sockaddr, uaddr, sa_family);
9197 case offsetof(struct bpf_sock_addr, user_ip4):
9198 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9199 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
9200 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
9203 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9205 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
9206 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9207 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9208 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
9212 case offsetof(struct bpf_sock_addr, user_port):
9213 /* To get port we need to know sa_family first and then treat
9214 * sockaddr as either sockaddr_in or sockaddr_in6.
9215 * Though we can simplify since port field has same offset and
9216 * size in both structures.
9217 * Here we check this invariant and use just one of the
9218 * structures if it's true.
9220 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
9221 offsetof(struct sockaddr_in6, sin6_port));
9222 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
9223 sizeof_field(struct sockaddr_in6, sin6_port));
9224 /* Account for sin6_port being smaller than user_port. */
9225 port_size = min(port_size, BPF_LDST_BYTES(si));
9226 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9227 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9228 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
9231 case offsetof(struct bpf_sock_addr, family):
9232 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9233 struct sock, sk, sk_family);
9236 case offsetof(struct bpf_sock_addr, type):
9237 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9238 struct sock, sk, sk_type);
9241 case offsetof(struct bpf_sock_addr, protocol):
9242 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9243 struct sock, sk, sk_protocol);
9246 case offsetof(struct bpf_sock_addr, msg_src_ip4):
9247 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
9248 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9249 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
9250 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
9253 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9256 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
9257 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
9258 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9259 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
9260 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
9262 case offsetof(struct bpf_sock_addr, sk):
9263 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
9264 si->dst_reg, si->src_reg,
9265 offsetof(struct bpf_sock_addr_kern, sk));
9269 return insn - insn_buf;
9272 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
9273 const struct bpf_insn *si,
9274 struct bpf_insn *insn_buf,
9275 struct bpf_prog *prog,
9278 struct bpf_insn *insn = insn_buf;
9281 /* Helper macro for adding read access to tcp_sock or sock fields. */
9282 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
9284 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \
9285 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
9286 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
9287 if (si->dst_reg == reg || si->src_reg == reg) \
9289 if (si->dst_reg == reg || si->src_reg == reg) \
9291 if (si->dst_reg == si->src_reg) { \
9292 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
9293 offsetof(struct bpf_sock_ops_kern, \
9295 fullsock_reg = reg; \
9298 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9299 struct bpf_sock_ops_kern, \
9301 fullsock_reg, si->src_reg, \
9302 offsetof(struct bpf_sock_ops_kern, \
9304 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
9305 if (si->dst_reg == si->src_reg) \
9306 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9307 offsetof(struct bpf_sock_ops_kern, \
9309 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9310 struct bpf_sock_ops_kern, sk),\
9311 si->dst_reg, si->src_reg, \
9312 offsetof(struct bpf_sock_ops_kern, sk));\
9313 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
9315 si->dst_reg, si->dst_reg, \
9316 offsetof(OBJ, OBJ_FIELD)); \
9317 if (si->dst_reg == si->src_reg) { \
9318 *insn++ = BPF_JMP_A(1); \
9319 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9320 offsetof(struct bpf_sock_ops_kern, \
9325 #define SOCK_OPS_GET_SK() \
9327 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \
9328 if (si->dst_reg == reg || si->src_reg == reg) \
9330 if (si->dst_reg == reg || si->src_reg == reg) \
9332 if (si->dst_reg == si->src_reg) { \
9333 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
9334 offsetof(struct bpf_sock_ops_kern, \
9336 fullsock_reg = reg; \
9339 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9340 struct bpf_sock_ops_kern, \
9342 fullsock_reg, si->src_reg, \
9343 offsetof(struct bpf_sock_ops_kern, \
9345 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
9346 if (si->dst_reg == si->src_reg) \
9347 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9348 offsetof(struct bpf_sock_ops_kern, \
9350 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9351 struct bpf_sock_ops_kern, sk),\
9352 si->dst_reg, si->src_reg, \
9353 offsetof(struct bpf_sock_ops_kern, sk));\
9354 if (si->dst_reg == si->src_reg) { \
9355 *insn++ = BPF_JMP_A(1); \
9356 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9357 offsetof(struct bpf_sock_ops_kern, \
9362 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
9363 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
9365 /* Helper macro for adding write access to tcp_sock or sock fields.
9366 * The macro is called with two registers, dst_reg which contains a pointer
9367 * to ctx (context) and src_reg which contains the value that should be
9368 * stored. However, we need an additional register since we cannot overwrite
9369 * dst_reg because it may be used later in the program.
9370 * Instead we "borrow" one of the other register. We first save its value
9371 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
9372 * it at the end of the macro.
9374 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
9376 int reg = BPF_REG_9; \
9377 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
9378 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
9379 if (si->dst_reg == reg || si->src_reg == reg) \
9381 if (si->dst_reg == reg || si->src_reg == reg) \
9383 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
9384 offsetof(struct bpf_sock_ops_kern, \
9386 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9387 struct bpf_sock_ops_kern, \
9390 offsetof(struct bpf_sock_ops_kern, \
9392 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
9393 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9394 struct bpf_sock_ops_kern, sk),\
9396 offsetof(struct bpf_sock_ops_kern, sk));\
9397 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \
9399 offsetof(OBJ, OBJ_FIELD)); \
9400 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
9401 offsetof(struct bpf_sock_ops_kern, \
9405 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
9407 if (TYPE == BPF_WRITE) \
9408 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
9410 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
9413 if (insn > insn_buf)
9414 return insn - insn_buf;
9417 case offsetof(struct bpf_sock_ops, op):
9418 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9420 si->dst_reg, si->src_reg,
9421 offsetof(struct bpf_sock_ops_kern, op));
9424 case offsetof(struct bpf_sock_ops, replylong[0]) ...
9425 offsetof(struct bpf_sock_ops, replylong[3]):
9426 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
9427 sizeof_field(struct bpf_sock_ops_kern, reply));
9428 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
9429 sizeof_field(struct bpf_sock_ops_kern, replylong));
9431 off -= offsetof(struct bpf_sock_ops, replylong[0]);
9432 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
9433 if (type == BPF_WRITE)
9434 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9437 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9441 case offsetof(struct bpf_sock_ops, family):
9442 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9444 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9445 struct bpf_sock_ops_kern, sk),
9446 si->dst_reg, si->src_reg,
9447 offsetof(struct bpf_sock_ops_kern, sk));
9448 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9449 offsetof(struct sock_common, skc_family));
9452 case offsetof(struct bpf_sock_ops, remote_ip4):
9453 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9455 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9456 struct bpf_sock_ops_kern, sk),
9457 si->dst_reg, si->src_reg,
9458 offsetof(struct bpf_sock_ops_kern, sk));
9459 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9460 offsetof(struct sock_common, skc_daddr));
9463 case offsetof(struct bpf_sock_ops, local_ip4):
9464 BUILD_BUG_ON(sizeof_field(struct sock_common,
9465 skc_rcv_saddr) != 4);
9467 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9468 struct bpf_sock_ops_kern, sk),
9469 si->dst_reg, si->src_reg,
9470 offsetof(struct bpf_sock_ops_kern, sk));
9471 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9472 offsetof(struct sock_common,
9476 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
9477 offsetof(struct bpf_sock_ops, remote_ip6[3]):
9478 #if IS_ENABLED(CONFIG_IPV6)
9479 BUILD_BUG_ON(sizeof_field(struct sock_common,
9480 skc_v6_daddr.s6_addr32[0]) != 4);
9483 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
9484 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9485 struct bpf_sock_ops_kern, sk),
9486 si->dst_reg, si->src_reg,
9487 offsetof(struct bpf_sock_ops_kern, sk));
9488 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9489 offsetof(struct sock_common,
9490 skc_v6_daddr.s6_addr32[0]) +
9493 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9497 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
9498 offsetof(struct bpf_sock_ops, local_ip6[3]):
9499 #if IS_ENABLED(CONFIG_IPV6)
9500 BUILD_BUG_ON(sizeof_field(struct sock_common,
9501 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9504 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
9505 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9506 struct bpf_sock_ops_kern, sk),
9507 si->dst_reg, si->src_reg,
9508 offsetof(struct bpf_sock_ops_kern, sk));
9509 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9510 offsetof(struct sock_common,
9511 skc_v6_rcv_saddr.s6_addr32[0]) +
9514 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9518 case offsetof(struct bpf_sock_ops, remote_port):
9519 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9521 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9522 struct bpf_sock_ops_kern, sk),
9523 si->dst_reg, si->src_reg,
9524 offsetof(struct bpf_sock_ops_kern, sk));
9525 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9526 offsetof(struct sock_common, skc_dport));
9527 #ifndef __BIG_ENDIAN_BITFIELD
9528 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9532 case offsetof(struct bpf_sock_ops, local_port):
9533 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9535 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9536 struct bpf_sock_ops_kern, sk),
9537 si->dst_reg, si->src_reg,
9538 offsetof(struct bpf_sock_ops_kern, sk));
9539 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9540 offsetof(struct sock_common, skc_num));
9543 case offsetof(struct bpf_sock_ops, is_fullsock):
9544 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9545 struct bpf_sock_ops_kern,
9547 si->dst_reg, si->src_reg,
9548 offsetof(struct bpf_sock_ops_kern,
9552 case offsetof(struct bpf_sock_ops, state):
9553 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
9555 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9556 struct bpf_sock_ops_kern, sk),
9557 si->dst_reg, si->src_reg,
9558 offsetof(struct bpf_sock_ops_kern, sk));
9559 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
9560 offsetof(struct sock_common, skc_state));
9563 case offsetof(struct bpf_sock_ops, rtt_min):
9564 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
9565 sizeof(struct minmax));
9566 BUILD_BUG_ON(sizeof(struct minmax) <
9567 sizeof(struct minmax_sample));
9569 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9570 struct bpf_sock_ops_kern, sk),
9571 si->dst_reg, si->src_reg,
9572 offsetof(struct bpf_sock_ops_kern, sk));
9573 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9574 offsetof(struct tcp_sock, rtt_min) +
9575 sizeof_field(struct minmax_sample, t));
9578 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
9579 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
9583 case offsetof(struct bpf_sock_ops, sk_txhash):
9584 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
9587 case offsetof(struct bpf_sock_ops, snd_cwnd):
9588 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
9590 case offsetof(struct bpf_sock_ops, srtt_us):
9591 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
9593 case offsetof(struct bpf_sock_ops, snd_ssthresh):
9594 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
9596 case offsetof(struct bpf_sock_ops, rcv_nxt):
9597 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
9599 case offsetof(struct bpf_sock_ops, snd_nxt):
9600 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
9602 case offsetof(struct bpf_sock_ops, snd_una):
9603 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
9605 case offsetof(struct bpf_sock_ops, mss_cache):
9606 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
9608 case offsetof(struct bpf_sock_ops, ecn_flags):
9609 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
9611 case offsetof(struct bpf_sock_ops, rate_delivered):
9612 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
9614 case offsetof(struct bpf_sock_ops, rate_interval_us):
9615 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
9617 case offsetof(struct bpf_sock_ops, packets_out):
9618 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
9620 case offsetof(struct bpf_sock_ops, retrans_out):
9621 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
9623 case offsetof(struct bpf_sock_ops, total_retrans):
9624 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
9626 case offsetof(struct bpf_sock_ops, segs_in):
9627 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
9629 case offsetof(struct bpf_sock_ops, data_segs_in):
9630 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
9632 case offsetof(struct bpf_sock_ops, segs_out):
9633 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
9635 case offsetof(struct bpf_sock_ops, data_segs_out):
9636 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
9638 case offsetof(struct bpf_sock_ops, lost_out):
9639 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
9641 case offsetof(struct bpf_sock_ops, sacked_out):
9642 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
9644 case offsetof(struct bpf_sock_ops, bytes_received):
9645 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
9647 case offsetof(struct bpf_sock_ops, bytes_acked):
9648 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
9650 case offsetof(struct bpf_sock_ops, sk):
9653 case offsetof(struct bpf_sock_ops, skb_data_end):
9654 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9656 si->dst_reg, si->src_reg,
9657 offsetof(struct bpf_sock_ops_kern,
9660 case offsetof(struct bpf_sock_ops, skb_data):
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 sk_buff, data),
9668 si->dst_reg, si->dst_reg,
9669 offsetof(struct sk_buff, data));
9671 case offsetof(struct bpf_sock_ops, skb_len):
9672 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9674 si->dst_reg, si->src_reg,
9675 offsetof(struct bpf_sock_ops_kern,
9677 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9678 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9679 si->dst_reg, si->dst_reg,
9680 offsetof(struct sk_buff, len));
9682 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9683 off = offsetof(struct sk_buff, cb);
9684 off += offsetof(struct tcp_skb_cb, tcp_flags);
9685 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
9686 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9688 si->dst_reg, si->src_reg,
9689 offsetof(struct bpf_sock_ops_kern,
9691 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9692 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
9694 si->dst_reg, si->dst_reg, off);
9697 return insn - insn_buf;
9700 /* data_end = skb->data + skb_headlen() */
9701 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
9702 struct bpf_insn *insn)
9704 /* si->dst_reg = skb->data */
9705 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9706 si->dst_reg, si->src_reg,
9707 offsetof(struct sk_buff, data));
9709 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9710 BPF_REG_AX, si->src_reg,
9711 offsetof(struct sk_buff, len));
9712 /* si->dst_reg = skb->data + skb->len */
9713 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
9714 /* AX = skb->data_len */
9715 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
9716 BPF_REG_AX, si->src_reg,
9717 offsetof(struct sk_buff, data_len));
9718 /* si->dst_reg = skb->data + skb->len - skb->data_len */
9719 *insn++ = BPF_ALU64_REG(BPF_SUB, si->dst_reg, BPF_REG_AX);
9724 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
9725 const struct bpf_insn *si,
9726 struct bpf_insn *insn_buf,
9727 struct bpf_prog *prog, u32 *target_size)
9729 struct bpf_insn *insn = insn_buf;
9732 case offsetof(struct __sk_buff, data_end):
9733 insn = bpf_convert_data_end_access(si, insn);
9736 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9740 return insn - insn_buf;
9743 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
9744 const struct bpf_insn *si,
9745 struct bpf_insn *insn_buf,
9746 struct bpf_prog *prog, u32 *target_size)
9748 struct bpf_insn *insn = insn_buf;
9749 #if IS_ENABLED(CONFIG_IPV6)
9753 /* convert ctx uses the fact sg element is first in struct */
9754 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
9757 case offsetof(struct sk_msg_md, data):
9758 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
9759 si->dst_reg, si->src_reg,
9760 offsetof(struct sk_msg, data));
9762 case offsetof(struct sk_msg_md, data_end):
9763 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
9764 si->dst_reg, si->src_reg,
9765 offsetof(struct sk_msg, data_end));
9767 case offsetof(struct sk_msg_md, family):
9768 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9770 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9772 si->dst_reg, si->src_reg,
9773 offsetof(struct sk_msg, sk));
9774 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9775 offsetof(struct sock_common, skc_family));
9778 case offsetof(struct sk_msg_md, remote_ip4):
9779 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9781 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9783 si->dst_reg, si->src_reg,
9784 offsetof(struct sk_msg, sk));
9785 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9786 offsetof(struct sock_common, skc_daddr));
9789 case offsetof(struct sk_msg_md, local_ip4):
9790 BUILD_BUG_ON(sizeof_field(struct sock_common,
9791 skc_rcv_saddr) != 4);
9793 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9795 si->dst_reg, si->src_reg,
9796 offsetof(struct sk_msg, sk));
9797 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9798 offsetof(struct sock_common,
9802 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
9803 offsetof(struct sk_msg_md, remote_ip6[3]):
9804 #if IS_ENABLED(CONFIG_IPV6)
9805 BUILD_BUG_ON(sizeof_field(struct sock_common,
9806 skc_v6_daddr.s6_addr32[0]) != 4);
9809 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
9810 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9812 si->dst_reg, si->src_reg,
9813 offsetof(struct sk_msg, sk));
9814 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9815 offsetof(struct sock_common,
9816 skc_v6_daddr.s6_addr32[0]) +
9819 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9823 case offsetof(struct sk_msg_md, local_ip6[0]) ...
9824 offsetof(struct sk_msg_md, local_ip6[3]):
9825 #if IS_ENABLED(CONFIG_IPV6)
9826 BUILD_BUG_ON(sizeof_field(struct sock_common,
9827 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9830 off -= offsetof(struct sk_msg_md, local_ip6[0]);
9831 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9833 si->dst_reg, si->src_reg,
9834 offsetof(struct sk_msg, sk));
9835 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9836 offsetof(struct sock_common,
9837 skc_v6_rcv_saddr.s6_addr32[0]) +
9840 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9844 case offsetof(struct sk_msg_md, remote_port):
9845 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9847 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9849 si->dst_reg, si->src_reg,
9850 offsetof(struct sk_msg, sk));
9851 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9852 offsetof(struct sock_common, skc_dport));
9853 #ifndef __BIG_ENDIAN_BITFIELD
9854 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9858 case offsetof(struct sk_msg_md, local_port):
9859 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9861 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9863 si->dst_reg, si->src_reg,
9864 offsetof(struct sk_msg, sk));
9865 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9866 offsetof(struct sock_common, skc_num));
9869 case offsetof(struct sk_msg_md, size):
9870 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
9871 si->dst_reg, si->src_reg,
9872 offsetof(struct sk_msg_sg, size));
9875 case offsetof(struct sk_msg_md, sk):
9876 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
9877 si->dst_reg, si->src_reg,
9878 offsetof(struct sk_msg, sk));
9882 return insn - insn_buf;
9885 const struct bpf_verifier_ops sk_filter_verifier_ops = {
9886 .get_func_proto = sk_filter_func_proto,
9887 .is_valid_access = sk_filter_is_valid_access,
9888 .convert_ctx_access = bpf_convert_ctx_access,
9889 .gen_ld_abs = bpf_gen_ld_abs,
9892 const struct bpf_prog_ops sk_filter_prog_ops = {
9893 .test_run = bpf_prog_test_run_skb,
9896 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
9897 .get_func_proto = tc_cls_act_func_proto,
9898 .is_valid_access = tc_cls_act_is_valid_access,
9899 .convert_ctx_access = tc_cls_act_convert_ctx_access,
9900 .gen_prologue = tc_cls_act_prologue,
9901 .gen_ld_abs = bpf_gen_ld_abs,
9902 .check_kfunc_call = bpf_prog_test_check_kfunc_call,
9905 const struct bpf_prog_ops tc_cls_act_prog_ops = {
9906 .test_run = bpf_prog_test_run_skb,
9909 const struct bpf_verifier_ops xdp_verifier_ops = {
9910 .get_func_proto = xdp_func_proto,
9911 .is_valid_access = xdp_is_valid_access,
9912 .convert_ctx_access = xdp_convert_ctx_access,
9913 .gen_prologue = bpf_noop_prologue,
9916 const struct bpf_prog_ops xdp_prog_ops = {
9917 .test_run = bpf_prog_test_run_xdp,
9920 const struct bpf_verifier_ops cg_skb_verifier_ops = {
9921 .get_func_proto = cg_skb_func_proto,
9922 .is_valid_access = cg_skb_is_valid_access,
9923 .convert_ctx_access = bpf_convert_ctx_access,
9926 const struct bpf_prog_ops cg_skb_prog_ops = {
9927 .test_run = bpf_prog_test_run_skb,
9930 const struct bpf_verifier_ops lwt_in_verifier_ops = {
9931 .get_func_proto = lwt_in_func_proto,
9932 .is_valid_access = lwt_is_valid_access,
9933 .convert_ctx_access = bpf_convert_ctx_access,
9936 const struct bpf_prog_ops lwt_in_prog_ops = {
9937 .test_run = bpf_prog_test_run_skb,
9940 const struct bpf_verifier_ops lwt_out_verifier_ops = {
9941 .get_func_proto = lwt_out_func_proto,
9942 .is_valid_access = lwt_is_valid_access,
9943 .convert_ctx_access = bpf_convert_ctx_access,
9946 const struct bpf_prog_ops lwt_out_prog_ops = {
9947 .test_run = bpf_prog_test_run_skb,
9950 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
9951 .get_func_proto = lwt_xmit_func_proto,
9952 .is_valid_access = lwt_is_valid_access,
9953 .convert_ctx_access = bpf_convert_ctx_access,
9954 .gen_prologue = tc_cls_act_prologue,
9957 const struct bpf_prog_ops lwt_xmit_prog_ops = {
9958 .test_run = bpf_prog_test_run_skb,
9961 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
9962 .get_func_proto = lwt_seg6local_func_proto,
9963 .is_valid_access = lwt_is_valid_access,
9964 .convert_ctx_access = bpf_convert_ctx_access,
9967 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
9968 .test_run = bpf_prog_test_run_skb,
9971 const struct bpf_verifier_ops cg_sock_verifier_ops = {
9972 .get_func_proto = sock_filter_func_proto,
9973 .is_valid_access = sock_filter_is_valid_access,
9974 .convert_ctx_access = bpf_sock_convert_ctx_access,
9977 const struct bpf_prog_ops cg_sock_prog_ops = {
9980 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
9981 .get_func_proto = sock_addr_func_proto,
9982 .is_valid_access = sock_addr_is_valid_access,
9983 .convert_ctx_access = sock_addr_convert_ctx_access,
9986 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
9989 const struct bpf_verifier_ops sock_ops_verifier_ops = {
9990 .get_func_proto = sock_ops_func_proto,
9991 .is_valid_access = sock_ops_is_valid_access,
9992 .convert_ctx_access = sock_ops_convert_ctx_access,
9995 const struct bpf_prog_ops sock_ops_prog_ops = {
9998 const struct bpf_verifier_ops sk_skb_verifier_ops = {
9999 .get_func_proto = sk_skb_func_proto,
10000 .is_valid_access = sk_skb_is_valid_access,
10001 .convert_ctx_access = sk_skb_convert_ctx_access,
10002 .gen_prologue = sk_skb_prologue,
10005 const struct bpf_prog_ops sk_skb_prog_ops = {
10008 const struct bpf_verifier_ops sk_msg_verifier_ops = {
10009 .get_func_proto = sk_msg_func_proto,
10010 .is_valid_access = sk_msg_is_valid_access,
10011 .convert_ctx_access = sk_msg_convert_ctx_access,
10012 .gen_prologue = bpf_noop_prologue,
10015 const struct bpf_prog_ops sk_msg_prog_ops = {
10018 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
10019 .get_func_proto = flow_dissector_func_proto,
10020 .is_valid_access = flow_dissector_is_valid_access,
10021 .convert_ctx_access = flow_dissector_convert_ctx_access,
10024 const struct bpf_prog_ops flow_dissector_prog_ops = {
10025 .test_run = bpf_prog_test_run_flow_dissector,
10028 int sk_detach_filter(struct sock *sk)
10031 struct sk_filter *filter;
10033 if (sock_flag(sk, SOCK_FILTER_LOCKED))
10036 filter = rcu_dereference_protected(sk->sk_filter,
10037 lockdep_sock_is_held(sk));
10039 RCU_INIT_POINTER(sk->sk_filter, NULL);
10040 sk_filter_uncharge(sk, filter);
10046 EXPORT_SYMBOL_GPL(sk_detach_filter);
10048 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
10051 struct sock_fprog_kern *fprog;
10052 struct sk_filter *filter;
10056 filter = rcu_dereference_protected(sk->sk_filter,
10057 lockdep_sock_is_held(sk));
10061 /* We're copying the filter that has been originally attached,
10062 * so no conversion/decode needed anymore. eBPF programs that
10063 * have no original program cannot be dumped through this.
10066 fprog = filter->prog->orig_prog;
10072 /* User space only enquires number of filter blocks. */
10076 if (len < fprog->len)
10080 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
10083 /* Instead of bytes, the API requests to return the number
10084 * of filter blocks.
10093 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
10094 struct sock_reuseport *reuse,
10095 struct sock *sk, struct sk_buff *skb,
10096 struct sock *migrating_sk,
10099 reuse_kern->skb = skb;
10100 reuse_kern->sk = sk;
10101 reuse_kern->selected_sk = NULL;
10102 reuse_kern->migrating_sk = migrating_sk;
10103 reuse_kern->data_end = skb->data + skb_headlen(skb);
10104 reuse_kern->hash = hash;
10105 reuse_kern->reuseport_id = reuse->reuseport_id;
10106 reuse_kern->bind_inany = reuse->bind_inany;
10109 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
10110 struct bpf_prog *prog, struct sk_buff *skb,
10111 struct sock *migrating_sk,
10114 struct sk_reuseport_kern reuse_kern;
10115 enum sk_action action;
10117 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
10118 action = BPF_PROG_RUN(prog, &reuse_kern);
10120 if (action == SK_PASS)
10121 return reuse_kern.selected_sk;
10123 return ERR_PTR(-ECONNREFUSED);
10126 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
10127 struct bpf_map *, map, void *, key, u32, flags)
10129 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
10130 struct sock_reuseport *reuse;
10131 struct sock *selected_sk;
10133 selected_sk = map->ops->map_lookup_elem(map, key);
10137 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
10139 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
10140 if (sk_is_refcounted(selected_sk))
10141 sock_put(selected_sk);
10143 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
10144 * The only (!reuse) case here is - the sk has already been
10145 * unhashed (e.g. by close()), so treat it as -ENOENT.
10147 * Other maps (e.g. sock_map) do not provide this guarantee and
10148 * the sk may never be in the reuseport group to begin with.
10150 return is_sockarray ? -ENOENT : -EINVAL;
10153 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
10154 struct sock *sk = reuse_kern->sk;
10156 if (sk->sk_protocol != selected_sk->sk_protocol)
10157 return -EPROTOTYPE;
10158 else if (sk->sk_family != selected_sk->sk_family)
10159 return -EAFNOSUPPORT;
10161 /* Catch all. Likely bound to a different sockaddr. */
10165 reuse_kern->selected_sk = selected_sk;
10170 static const struct bpf_func_proto sk_select_reuseport_proto = {
10171 .func = sk_select_reuseport,
10173 .ret_type = RET_INTEGER,
10174 .arg1_type = ARG_PTR_TO_CTX,
10175 .arg2_type = ARG_CONST_MAP_PTR,
10176 .arg3_type = ARG_PTR_TO_MAP_KEY,
10177 .arg4_type = ARG_ANYTHING,
10180 BPF_CALL_4(sk_reuseport_load_bytes,
10181 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10182 void *, to, u32, len)
10184 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
10187 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
10188 .func = sk_reuseport_load_bytes,
10190 .ret_type = RET_INTEGER,
10191 .arg1_type = ARG_PTR_TO_CTX,
10192 .arg2_type = ARG_ANYTHING,
10193 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
10194 .arg4_type = ARG_CONST_SIZE,
10197 BPF_CALL_5(sk_reuseport_load_bytes_relative,
10198 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10199 void *, to, u32, len, u32, start_header)
10201 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
10202 len, start_header);
10205 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
10206 .func = sk_reuseport_load_bytes_relative,
10208 .ret_type = RET_INTEGER,
10209 .arg1_type = ARG_PTR_TO_CTX,
10210 .arg2_type = ARG_ANYTHING,
10211 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
10212 .arg4_type = ARG_CONST_SIZE,
10213 .arg5_type = ARG_ANYTHING,
10216 static const struct bpf_func_proto *
10217 sk_reuseport_func_proto(enum bpf_func_id func_id,
10218 const struct bpf_prog *prog)
10221 case BPF_FUNC_sk_select_reuseport:
10222 return &sk_select_reuseport_proto;
10223 case BPF_FUNC_skb_load_bytes:
10224 return &sk_reuseport_load_bytes_proto;
10225 case BPF_FUNC_skb_load_bytes_relative:
10226 return &sk_reuseport_load_bytes_relative_proto;
10227 case BPF_FUNC_get_socket_cookie:
10228 return &bpf_get_socket_ptr_cookie_proto;
10230 return bpf_base_func_proto(func_id);
10235 sk_reuseport_is_valid_access(int off, int size,
10236 enum bpf_access_type type,
10237 const struct bpf_prog *prog,
10238 struct bpf_insn_access_aux *info)
10240 const u32 size_default = sizeof(__u32);
10242 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
10243 off % size || type != BPF_READ)
10247 case offsetof(struct sk_reuseport_md, data):
10248 info->reg_type = PTR_TO_PACKET;
10249 return size == sizeof(__u64);
10251 case offsetof(struct sk_reuseport_md, data_end):
10252 info->reg_type = PTR_TO_PACKET_END;
10253 return size == sizeof(__u64);
10255 case offsetof(struct sk_reuseport_md, hash):
10256 return size == size_default;
10258 case offsetof(struct sk_reuseport_md, sk):
10259 info->reg_type = PTR_TO_SOCKET;
10260 return size == sizeof(__u64);
10262 case offsetof(struct sk_reuseport_md, migrating_sk):
10263 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
10264 return size == sizeof(__u64);
10266 /* Fields that allow narrowing */
10267 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
10268 if (size < sizeof_field(struct sk_buff, protocol))
10271 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
10272 case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
10273 case bpf_ctx_range(struct sk_reuseport_md, len):
10274 bpf_ctx_record_field_size(info, size_default);
10275 return bpf_ctx_narrow_access_ok(off, size, size_default);
10282 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
10283 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
10284 si->dst_reg, si->src_reg, \
10285 bpf_target_off(struct sk_reuseport_kern, F, \
10286 sizeof_field(struct sk_reuseport_kern, F), \
10290 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
10291 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
10296 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \
10297 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
10302 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
10303 const struct bpf_insn *si,
10304 struct bpf_insn *insn_buf,
10305 struct bpf_prog *prog,
10308 struct bpf_insn *insn = insn_buf;
10311 case offsetof(struct sk_reuseport_md, data):
10312 SK_REUSEPORT_LOAD_SKB_FIELD(data);
10315 case offsetof(struct sk_reuseport_md, len):
10316 SK_REUSEPORT_LOAD_SKB_FIELD(len);
10319 case offsetof(struct sk_reuseport_md, eth_protocol):
10320 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
10323 case offsetof(struct sk_reuseport_md, ip_protocol):
10324 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
10327 case offsetof(struct sk_reuseport_md, data_end):
10328 SK_REUSEPORT_LOAD_FIELD(data_end);
10331 case offsetof(struct sk_reuseport_md, hash):
10332 SK_REUSEPORT_LOAD_FIELD(hash);
10335 case offsetof(struct sk_reuseport_md, bind_inany):
10336 SK_REUSEPORT_LOAD_FIELD(bind_inany);
10339 case offsetof(struct sk_reuseport_md, sk):
10340 SK_REUSEPORT_LOAD_FIELD(sk);
10343 case offsetof(struct sk_reuseport_md, migrating_sk):
10344 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
10348 return insn - insn_buf;
10351 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
10352 .get_func_proto = sk_reuseport_func_proto,
10353 .is_valid_access = sk_reuseport_is_valid_access,
10354 .convert_ctx_access = sk_reuseport_convert_ctx_access,
10357 const struct bpf_prog_ops sk_reuseport_prog_ops = {
10360 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
10361 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
10363 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
10364 struct sock *, sk, u64, flags)
10366 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
10367 BPF_SK_LOOKUP_F_NO_REUSEPORT)))
10369 if (unlikely(sk && sk_is_refcounted(sk)))
10370 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
10371 if (unlikely(sk && sk->sk_state == TCP_ESTABLISHED))
10372 return -ESOCKTNOSUPPORT; /* reject connected sockets */
10374 /* Check if socket is suitable for packet L3/L4 protocol */
10375 if (sk && sk->sk_protocol != ctx->protocol)
10376 return -EPROTOTYPE;
10377 if (sk && sk->sk_family != ctx->family &&
10378 (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
10379 return -EAFNOSUPPORT;
10381 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
10384 /* Select socket as lookup result */
10385 ctx->selected_sk = sk;
10386 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
10390 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
10391 .func = bpf_sk_lookup_assign,
10393 .ret_type = RET_INTEGER,
10394 .arg1_type = ARG_PTR_TO_CTX,
10395 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL,
10396 .arg3_type = ARG_ANYTHING,
10399 static const struct bpf_func_proto *
10400 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
10403 case BPF_FUNC_perf_event_output:
10404 return &bpf_event_output_data_proto;
10405 case BPF_FUNC_sk_assign:
10406 return &bpf_sk_lookup_assign_proto;
10407 case BPF_FUNC_sk_release:
10408 return &bpf_sk_release_proto;
10410 return bpf_sk_base_func_proto(func_id);
10414 static bool sk_lookup_is_valid_access(int off, int size,
10415 enum bpf_access_type type,
10416 const struct bpf_prog *prog,
10417 struct bpf_insn_access_aux *info)
10419 if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
10421 if (off % size != 0)
10423 if (type != BPF_READ)
10427 case offsetof(struct bpf_sk_lookup, sk):
10428 info->reg_type = PTR_TO_SOCKET_OR_NULL;
10429 return size == sizeof(__u64);
10431 case bpf_ctx_range(struct bpf_sk_lookup, family):
10432 case bpf_ctx_range(struct bpf_sk_lookup, protocol):
10433 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
10434 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
10435 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
10436 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
10437 case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
10438 case bpf_ctx_range(struct bpf_sk_lookup, local_port):
10439 bpf_ctx_record_field_size(info, sizeof(__u32));
10440 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
10447 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
10448 const struct bpf_insn *si,
10449 struct bpf_insn *insn_buf,
10450 struct bpf_prog *prog,
10453 struct bpf_insn *insn = insn_buf;
10456 case offsetof(struct bpf_sk_lookup, sk):
10457 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10458 offsetof(struct bpf_sk_lookup_kern, selected_sk));
10461 case offsetof(struct bpf_sk_lookup, family):
10462 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10463 bpf_target_off(struct bpf_sk_lookup_kern,
10464 family, 2, target_size));
10467 case offsetof(struct bpf_sk_lookup, protocol):
10468 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10469 bpf_target_off(struct bpf_sk_lookup_kern,
10470 protocol, 2, target_size));
10473 case offsetof(struct bpf_sk_lookup, remote_ip4):
10474 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10475 bpf_target_off(struct bpf_sk_lookup_kern,
10476 v4.saddr, 4, target_size));
10479 case offsetof(struct bpf_sk_lookup, local_ip4):
10480 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10481 bpf_target_off(struct bpf_sk_lookup_kern,
10482 v4.daddr, 4, target_size));
10485 case bpf_ctx_range_till(struct bpf_sk_lookup,
10486 remote_ip6[0], remote_ip6[3]): {
10487 #if IS_ENABLED(CONFIG_IPV6)
10490 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
10491 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10492 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10493 offsetof(struct bpf_sk_lookup_kern, v6.saddr));
10494 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10495 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10497 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10501 case bpf_ctx_range_till(struct bpf_sk_lookup,
10502 local_ip6[0], local_ip6[3]): {
10503 #if IS_ENABLED(CONFIG_IPV6)
10506 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
10507 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10508 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10509 offsetof(struct bpf_sk_lookup_kern, v6.daddr));
10510 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10511 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10513 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10517 case offsetof(struct bpf_sk_lookup, remote_port):
10518 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10519 bpf_target_off(struct bpf_sk_lookup_kern,
10520 sport, 2, target_size));
10523 case offsetof(struct bpf_sk_lookup, local_port):
10524 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10525 bpf_target_off(struct bpf_sk_lookup_kern,
10526 dport, 2, target_size));
10530 return insn - insn_buf;
10533 const struct bpf_prog_ops sk_lookup_prog_ops = {
10534 .test_run = bpf_prog_test_run_sk_lookup,
10537 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
10538 .get_func_proto = sk_lookup_func_proto,
10539 .is_valid_access = sk_lookup_is_valid_access,
10540 .convert_ctx_access = sk_lookup_convert_ctx_access,
10543 #endif /* CONFIG_INET */
10545 DEFINE_BPF_DISPATCHER(xdp)
10547 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
10549 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
10552 #ifdef CONFIG_DEBUG_INFO_BTF
10553 BTF_ID_LIST_GLOBAL(btf_sock_ids)
10554 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
10556 #undef BTF_SOCK_TYPE
10558 u32 btf_sock_ids[MAX_BTF_SOCK_TYPE];
10561 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
10563 /* tcp6_sock type is not generated in dwarf and hence btf,
10564 * trigger an explicit type generation here.
10566 BTF_TYPE_EMIT(struct tcp6_sock);
10567 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
10568 sk->sk_family == AF_INET6)
10569 return (unsigned long)sk;
10571 return (unsigned long)NULL;
10574 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
10575 .func = bpf_skc_to_tcp6_sock,
10577 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10578 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10579 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
10582 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
10584 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
10585 return (unsigned long)sk;
10587 return (unsigned long)NULL;
10590 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
10591 .func = bpf_skc_to_tcp_sock,
10593 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10594 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10595 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
10598 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
10600 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
10601 * generated if CONFIG_INET=n. Trigger an explicit generation here.
10603 BTF_TYPE_EMIT(struct inet_timewait_sock);
10604 BTF_TYPE_EMIT(struct tcp_timewait_sock);
10607 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
10608 return (unsigned long)sk;
10611 #if IS_BUILTIN(CONFIG_IPV6)
10612 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
10613 return (unsigned long)sk;
10616 return (unsigned long)NULL;
10619 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
10620 .func = bpf_skc_to_tcp_timewait_sock,
10622 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10623 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10624 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
10627 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
10630 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10631 return (unsigned long)sk;
10634 #if IS_BUILTIN(CONFIG_IPV6)
10635 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10636 return (unsigned long)sk;
10639 return (unsigned long)NULL;
10642 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
10643 .func = bpf_skc_to_tcp_request_sock,
10645 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10646 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10647 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
10650 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
10652 /* udp6_sock type is not generated in dwarf and hence btf,
10653 * trigger an explicit type generation here.
10655 BTF_TYPE_EMIT(struct udp6_sock);
10656 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
10657 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
10658 return (unsigned long)sk;
10660 return (unsigned long)NULL;
10663 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
10664 .func = bpf_skc_to_udp6_sock,
10666 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10667 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10668 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
10671 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
10673 return (unsigned long)sock_from_file(file);
10676 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
10677 BTF_ID(struct, socket)
10678 BTF_ID(struct, file)
10680 const struct bpf_func_proto bpf_sock_from_file_proto = {
10681 .func = bpf_sock_from_file,
10683 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10684 .ret_btf_id = &bpf_sock_from_file_btf_ids[0],
10685 .arg1_type = ARG_PTR_TO_BTF_ID,
10686 .arg1_btf_id = &bpf_sock_from_file_btf_ids[1],
10689 static const struct bpf_func_proto *
10690 bpf_sk_base_func_proto(enum bpf_func_id func_id)
10692 const struct bpf_func_proto *func;
10695 case BPF_FUNC_skc_to_tcp6_sock:
10696 func = &bpf_skc_to_tcp6_sock_proto;
10698 case BPF_FUNC_skc_to_tcp_sock:
10699 func = &bpf_skc_to_tcp_sock_proto;
10701 case BPF_FUNC_skc_to_tcp_timewait_sock:
10702 func = &bpf_skc_to_tcp_timewait_sock_proto;
10704 case BPF_FUNC_skc_to_tcp_request_sock:
10705 func = &bpf_skc_to_tcp_request_sock_proto;
10707 case BPF_FUNC_skc_to_udp6_sock:
10708 func = &bpf_skc_to_udp6_sock_proto;
10711 return bpf_base_func_proto(func_id);
10714 if (!perfmon_capable())