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 struct sk_buff *skb2;
2185 skb2 = skb_realloc_headroom(skb, hh_len);
2186 if (unlikely(!skb2)) {
2191 skb_set_owner_w(skb2, skb->sk);
2199 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2200 &ipv6_hdr(skb)->daddr);
2202 nexthop = &nh->ipv6_nh;
2204 neigh = ip_neigh_gw6(dev, nexthop);
2205 if (likely(!IS_ERR(neigh))) {
2208 sock_confirm_neigh(skb, neigh);
2209 dev_xmit_recursion_inc();
2210 ret = neigh_output(neigh, skb, false);
2211 dev_xmit_recursion_dec();
2212 rcu_read_unlock_bh();
2215 rcu_read_unlock_bh();
2217 IP6_INC_STATS(dev_net(dst->dev),
2218 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2224 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2225 struct bpf_nh_params *nh)
2227 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2228 struct net *net = dev_net(dev);
2229 int err, ret = NET_XMIT_DROP;
2232 struct dst_entry *dst;
2233 struct flowi6 fl6 = {
2234 .flowi6_flags = FLOWI_FLAG_ANYSRC,
2235 .flowi6_mark = skb->mark,
2236 .flowlabel = ip6_flowinfo(ip6h),
2237 .flowi6_oif = dev->ifindex,
2238 .flowi6_proto = ip6h->nexthdr,
2239 .daddr = ip6h->daddr,
2240 .saddr = ip6h->saddr,
2243 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2247 skb_dst_set(skb, dst);
2248 } else if (nh->nh_family != AF_INET6) {
2252 err = bpf_out_neigh_v6(net, skb, dev, nh);
2253 if (unlikely(net_xmit_eval(err)))
2254 dev->stats.tx_errors++;
2256 ret = NET_XMIT_SUCCESS;
2259 dev->stats.tx_errors++;
2265 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2266 struct bpf_nh_params *nh)
2269 return NET_XMIT_DROP;
2271 #endif /* CONFIG_IPV6 */
2273 #if IS_ENABLED(CONFIG_INET)
2274 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2275 struct net_device *dev, struct bpf_nh_params *nh)
2277 u32 hh_len = LL_RESERVED_SPACE(dev);
2278 struct neighbour *neigh;
2279 bool is_v6gw = false;
2281 if (dev_xmit_recursion()) {
2282 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2289 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2290 struct sk_buff *skb2;
2292 skb2 = skb_realloc_headroom(skb, hh_len);
2293 if (unlikely(!skb2)) {
2298 skb_set_owner_w(skb2, skb->sk);
2305 struct dst_entry *dst = skb_dst(skb);
2306 struct rtable *rt = container_of(dst, struct rtable, dst);
2308 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2309 } else if (nh->nh_family == AF_INET6) {
2310 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2312 } else if (nh->nh_family == AF_INET) {
2313 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2315 rcu_read_unlock_bh();
2319 if (likely(!IS_ERR(neigh))) {
2322 sock_confirm_neigh(skb, neigh);
2323 dev_xmit_recursion_inc();
2324 ret = neigh_output(neigh, skb, is_v6gw);
2325 dev_xmit_recursion_dec();
2326 rcu_read_unlock_bh();
2329 rcu_read_unlock_bh();
2335 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2336 struct bpf_nh_params *nh)
2338 const struct iphdr *ip4h = ip_hdr(skb);
2339 struct net *net = dev_net(dev);
2340 int err, ret = NET_XMIT_DROP;
2343 struct flowi4 fl4 = {
2344 .flowi4_flags = FLOWI_FLAG_ANYSRC,
2345 .flowi4_mark = skb->mark,
2346 .flowi4_tos = RT_TOS(ip4h->tos),
2347 .flowi4_oif = dev->ifindex,
2348 .flowi4_proto = ip4h->protocol,
2349 .daddr = ip4h->daddr,
2350 .saddr = ip4h->saddr,
2354 rt = ip_route_output_flow(net, &fl4, NULL);
2357 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2362 skb_dst_set(skb, &rt->dst);
2365 err = bpf_out_neigh_v4(net, skb, dev, nh);
2366 if (unlikely(net_xmit_eval(err)))
2367 dev->stats.tx_errors++;
2369 ret = NET_XMIT_SUCCESS;
2372 dev->stats.tx_errors++;
2378 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2379 struct bpf_nh_params *nh)
2382 return NET_XMIT_DROP;
2384 #endif /* CONFIG_INET */
2386 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2387 struct bpf_nh_params *nh)
2389 struct ethhdr *ethh = eth_hdr(skb);
2391 if (unlikely(skb->mac_header >= skb->network_header))
2393 bpf_push_mac_rcsum(skb);
2394 if (is_multicast_ether_addr(ethh->h_dest))
2397 skb_pull(skb, sizeof(*ethh));
2398 skb_unset_mac_header(skb);
2399 skb_reset_network_header(skb);
2401 if (skb->protocol == htons(ETH_P_IP))
2402 return __bpf_redirect_neigh_v4(skb, dev, nh);
2403 else if (skb->protocol == htons(ETH_P_IPV6))
2404 return __bpf_redirect_neigh_v6(skb, dev, nh);
2410 /* Internal, non-exposed redirect flags. */
2412 BPF_F_NEIGH = (1ULL << 1),
2413 BPF_F_PEER = (1ULL << 2),
2414 BPF_F_NEXTHOP = (1ULL << 3),
2415 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2418 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2420 struct net_device *dev;
2421 struct sk_buff *clone;
2424 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2427 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2431 clone = skb_clone(skb, GFP_ATOMIC);
2432 if (unlikely(!clone))
2435 /* For direct write, we need to keep the invariant that the skbs
2436 * we're dealing with need to be uncloned. Should uncloning fail
2437 * here, we need to free the just generated clone to unclone once
2440 ret = bpf_try_make_head_writable(skb);
2441 if (unlikely(ret)) {
2446 return __bpf_redirect(clone, dev, flags);
2449 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2450 .func = bpf_clone_redirect,
2452 .ret_type = RET_INTEGER,
2453 .arg1_type = ARG_PTR_TO_CTX,
2454 .arg2_type = ARG_ANYTHING,
2455 .arg3_type = ARG_ANYTHING,
2458 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2459 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2461 int skb_do_redirect(struct sk_buff *skb)
2463 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2464 struct net *net = dev_net(skb->dev);
2465 struct net_device *dev;
2466 u32 flags = ri->flags;
2468 dev = dev_get_by_index_rcu(net, ri->tgt_index);
2473 if (flags & BPF_F_PEER) {
2474 const struct net_device_ops *ops = dev->netdev_ops;
2476 if (unlikely(!ops->ndo_get_peer_dev ||
2477 !skb_at_tc_ingress(skb)))
2479 dev = ops->ndo_get_peer_dev(dev);
2480 if (unlikely(!dev ||
2481 !(dev->flags & IFF_UP) ||
2482 net_eq(net, dev_net(dev))))
2487 return flags & BPF_F_NEIGH ?
2488 __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2490 __bpf_redirect(skb, dev, flags);
2496 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2498 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2500 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2504 ri->tgt_index = ifindex;
2506 return TC_ACT_REDIRECT;
2509 static const struct bpf_func_proto bpf_redirect_proto = {
2510 .func = bpf_redirect,
2512 .ret_type = RET_INTEGER,
2513 .arg1_type = ARG_ANYTHING,
2514 .arg2_type = ARG_ANYTHING,
2517 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2519 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2521 if (unlikely(flags))
2524 ri->flags = BPF_F_PEER;
2525 ri->tgt_index = ifindex;
2527 return TC_ACT_REDIRECT;
2530 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2531 .func = bpf_redirect_peer,
2533 .ret_type = RET_INTEGER,
2534 .arg1_type = ARG_ANYTHING,
2535 .arg2_type = ARG_ANYTHING,
2538 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2539 int, plen, u64, flags)
2541 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2543 if (unlikely((plen && plen < sizeof(*params)) || flags))
2546 ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2547 ri->tgt_index = ifindex;
2549 BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2551 memcpy(&ri->nh, params, sizeof(ri->nh));
2553 return TC_ACT_REDIRECT;
2556 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2557 .func = bpf_redirect_neigh,
2559 .ret_type = RET_INTEGER,
2560 .arg1_type = ARG_ANYTHING,
2561 .arg2_type = ARG_PTR_TO_MEM_OR_NULL,
2562 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
2563 .arg4_type = ARG_ANYTHING,
2566 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2568 msg->apply_bytes = bytes;
2572 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2573 .func = bpf_msg_apply_bytes,
2575 .ret_type = RET_INTEGER,
2576 .arg1_type = ARG_PTR_TO_CTX,
2577 .arg2_type = ARG_ANYTHING,
2580 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2582 msg->cork_bytes = bytes;
2586 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2587 .func = bpf_msg_cork_bytes,
2589 .ret_type = RET_INTEGER,
2590 .arg1_type = ARG_PTR_TO_CTX,
2591 .arg2_type = ARG_ANYTHING,
2594 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2595 u32, end, u64, flags)
2597 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2598 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2599 struct scatterlist *sge;
2600 u8 *raw, *to, *from;
2603 if (unlikely(flags || end <= start))
2606 /* First find the starting scatterlist element */
2610 len = sk_msg_elem(msg, i)->length;
2611 if (start < offset + len)
2613 sk_msg_iter_var_next(i);
2614 } while (i != msg->sg.end);
2616 if (unlikely(start >= offset + len))
2620 /* The start may point into the sg element so we need to also
2621 * account for the headroom.
2623 bytes_sg_total = start - offset + bytes;
2624 if (!test_bit(i, &msg->sg.copy) && bytes_sg_total <= len)
2627 /* At this point we need to linearize multiple scatterlist
2628 * elements or a single shared page. Either way we need to
2629 * copy into a linear buffer exclusively owned by BPF. Then
2630 * place the buffer in the scatterlist and fixup the original
2631 * entries by removing the entries now in the linear buffer
2632 * and shifting the remaining entries. For now we do not try
2633 * to copy partial entries to avoid complexity of running out
2634 * of sg_entry slots. The downside is reading a single byte
2635 * will copy the entire sg entry.
2638 copy += sk_msg_elem(msg, i)->length;
2639 sk_msg_iter_var_next(i);
2640 if (bytes_sg_total <= copy)
2642 } while (i != msg->sg.end);
2645 if (unlikely(bytes_sg_total > copy))
2648 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2650 if (unlikely(!page))
2653 raw = page_address(page);
2656 sge = sk_msg_elem(msg, i);
2657 from = sg_virt(sge);
2661 memcpy(to, from, len);
2664 put_page(sg_page(sge));
2666 sk_msg_iter_var_next(i);
2667 } while (i != last_sge);
2669 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2671 /* To repair sg ring we need to shift entries. If we only
2672 * had a single entry though we can just replace it and
2673 * be done. Otherwise walk the ring and shift the entries.
2675 WARN_ON_ONCE(last_sge == first_sge);
2676 shift = last_sge > first_sge ?
2677 last_sge - first_sge - 1 :
2678 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2683 sk_msg_iter_var_next(i);
2687 if (i + shift >= NR_MSG_FRAG_IDS)
2688 move_from = i + shift - NR_MSG_FRAG_IDS;
2690 move_from = i + shift;
2691 if (move_from == msg->sg.end)
2694 msg->sg.data[i] = msg->sg.data[move_from];
2695 msg->sg.data[move_from].length = 0;
2696 msg->sg.data[move_from].page_link = 0;
2697 msg->sg.data[move_from].offset = 0;
2698 sk_msg_iter_var_next(i);
2701 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2702 msg->sg.end - shift + NR_MSG_FRAG_IDS :
2703 msg->sg.end - shift;
2705 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2706 msg->data_end = msg->data + bytes;
2710 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2711 .func = bpf_msg_pull_data,
2713 .ret_type = RET_INTEGER,
2714 .arg1_type = ARG_PTR_TO_CTX,
2715 .arg2_type = ARG_ANYTHING,
2716 .arg3_type = ARG_ANYTHING,
2717 .arg4_type = ARG_ANYTHING,
2720 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2721 u32, len, u64, flags)
2723 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2724 u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2725 u8 *raw, *to, *from;
2728 if (unlikely(flags))
2731 /* First find the starting scatterlist element */
2735 l = sk_msg_elem(msg, i)->length;
2737 if (start < offset + l)
2739 sk_msg_iter_var_next(i);
2740 } while (i != msg->sg.end);
2742 if (start >= offset + l)
2745 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2747 /* If no space available will fallback to copy, we need at
2748 * least one scatterlist elem available to push data into
2749 * when start aligns to the beginning of an element or two
2750 * when it falls inside an element. We handle the start equals
2751 * offset case because its the common case for inserting a
2754 if (!space || (space == 1 && start != offset))
2755 copy = msg->sg.data[i].length;
2757 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2758 get_order(copy + len));
2759 if (unlikely(!page))
2765 raw = page_address(page);
2767 psge = sk_msg_elem(msg, i);
2768 front = start - offset;
2769 back = psge->length - front;
2770 from = sg_virt(psge);
2773 memcpy(raw, from, front);
2777 to = raw + front + len;
2779 memcpy(to, from, back);
2782 put_page(sg_page(psge));
2783 } else if (start - offset) {
2784 psge = sk_msg_elem(msg, i);
2785 rsge = sk_msg_elem_cpy(msg, i);
2787 psge->length = start - offset;
2788 rsge.length -= psge->length;
2789 rsge.offset += start;
2791 sk_msg_iter_var_next(i);
2792 sg_unmark_end(psge);
2793 sg_unmark_end(&rsge);
2794 sk_msg_iter_next(msg, end);
2797 /* Slot(s) to place newly allocated data */
2800 /* Shift one or two slots as needed */
2802 sge = sk_msg_elem_cpy(msg, i);
2804 sk_msg_iter_var_next(i);
2805 sg_unmark_end(&sge);
2806 sk_msg_iter_next(msg, end);
2808 nsge = sk_msg_elem_cpy(msg, i);
2810 sk_msg_iter_var_next(i);
2811 nnsge = sk_msg_elem_cpy(msg, i);
2814 while (i != msg->sg.end) {
2815 msg->sg.data[i] = sge;
2817 sk_msg_iter_var_next(i);
2820 nnsge = sk_msg_elem_cpy(msg, i);
2822 nsge = sk_msg_elem_cpy(msg, i);
2827 /* Place newly allocated data buffer */
2828 sk_mem_charge(msg->sk, len);
2829 msg->sg.size += len;
2830 __clear_bit(new, &msg->sg.copy);
2831 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2833 get_page(sg_page(&rsge));
2834 sk_msg_iter_var_next(new);
2835 msg->sg.data[new] = rsge;
2838 sk_msg_compute_data_pointers(msg);
2842 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2843 .func = bpf_msg_push_data,
2845 .ret_type = RET_INTEGER,
2846 .arg1_type = ARG_PTR_TO_CTX,
2847 .arg2_type = ARG_ANYTHING,
2848 .arg3_type = ARG_ANYTHING,
2849 .arg4_type = ARG_ANYTHING,
2852 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2858 sk_msg_iter_var_next(i);
2859 msg->sg.data[prev] = msg->sg.data[i];
2860 } while (i != msg->sg.end);
2862 sk_msg_iter_prev(msg, end);
2865 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2867 struct scatterlist tmp, sge;
2869 sk_msg_iter_next(msg, end);
2870 sge = sk_msg_elem_cpy(msg, i);
2871 sk_msg_iter_var_next(i);
2872 tmp = sk_msg_elem_cpy(msg, i);
2874 while (i != msg->sg.end) {
2875 msg->sg.data[i] = sge;
2876 sk_msg_iter_var_next(i);
2878 tmp = sk_msg_elem_cpy(msg, i);
2882 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2883 u32, len, u64, flags)
2885 u32 i = 0, l = 0, space, offset = 0;
2886 u64 last = start + len;
2889 if (unlikely(flags))
2892 /* First find the starting scatterlist element */
2896 l = sk_msg_elem(msg, i)->length;
2898 if (start < offset + l)
2900 sk_msg_iter_var_next(i);
2901 } while (i != msg->sg.end);
2903 /* Bounds checks: start and pop must be inside message */
2904 if (start >= offset + l || last >= msg->sg.size)
2907 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2910 /* --------------| offset
2911 * -| start |-------- len -------|
2913 * |----- a ----|-------- pop -------|----- b ----|
2914 * |______________________________________________| length
2917 * a: region at front of scatter element to save
2918 * b: region at back of scatter element to save when length > A + pop
2919 * pop: region to pop from element, same as input 'pop' here will be
2920 * decremented below per iteration.
2922 * Two top-level cases to handle when start != offset, first B is non
2923 * zero and second B is zero corresponding to when a pop includes more
2926 * Then if B is non-zero AND there is no space allocate space and
2927 * compact A, B regions into page. If there is space shift ring to
2928 * the rigth free'ing the next element in ring to place B, leaving
2929 * A untouched except to reduce length.
2931 if (start != offset) {
2932 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2934 int b = sge->length - pop - a;
2936 sk_msg_iter_var_next(i);
2938 if (pop < sge->length - a) {
2941 sk_msg_shift_right(msg, i);
2942 nsge = sk_msg_elem(msg, i);
2943 get_page(sg_page(sge));
2946 b, sge->offset + pop + a);
2948 struct page *page, *orig;
2951 page = alloc_pages(__GFP_NOWARN |
2952 __GFP_COMP | GFP_ATOMIC,
2954 if (unlikely(!page))
2958 orig = sg_page(sge);
2959 from = sg_virt(sge);
2960 to = page_address(page);
2961 memcpy(to, from, a);
2962 memcpy(to + a, from + a + pop, b);
2963 sg_set_page(sge, page, a + b, 0);
2967 } else if (pop >= sge->length - a) {
2968 pop -= (sge->length - a);
2973 /* From above the current layout _must_ be as follows,
2978 * |---- pop ---|---------------- b ------------|
2979 * |____________________________________________| length
2981 * Offset and start of the current msg elem are equal because in the
2982 * previous case we handled offset != start and either consumed the
2983 * entire element and advanced to the next element OR pop == 0.
2985 * Two cases to handle here are first pop is less than the length
2986 * leaving some remainder b above. Simply adjust the element's layout
2987 * in this case. Or pop >= length of the element so that b = 0. In this
2988 * case advance to next element decrementing pop.
2991 struct scatterlist *sge = sk_msg_elem(msg, i);
2993 if (pop < sge->length) {
2999 sk_msg_shift_left(msg, i);
3001 sk_msg_iter_var_next(i);
3004 sk_mem_uncharge(msg->sk, len - pop);
3005 msg->sg.size -= (len - pop);
3006 sk_msg_compute_data_pointers(msg);
3010 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
3011 .func = bpf_msg_pop_data,
3013 .ret_type = RET_INTEGER,
3014 .arg1_type = ARG_PTR_TO_CTX,
3015 .arg2_type = ARG_ANYTHING,
3016 .arg3_type = ARG_ANYTHING,
3017 .arg4_type = ARG_ANYTHING,
3020 #ifdef CONFIG_CGROUP_NET_CLASSID
3021 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3023 return __task_get_classid(current);
3026 static const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3027 .func = bpf_get_cgroup_classid_curr,
3029 .ret_type = RET_INTEGER,
3032 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3034 struct sock *sk = skb_to_full_sk(skb);
3036 if (!sk || !sk_fullsock(sk))
3039 return sock_cgroup_classid(&sk->sk_cgrp_data);
3042 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3043 .func = bpf_skb_cgroup_classid,
3045 .ret_type = RET_INTEGER,
3046 .arg1_type = ARG_PTR_TO_CTX,
3050 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3052 return task_get_classid(skb);
3055 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3056 .func = bpf_get_cgroup_classid,
3058 .ret_type = RET_INTEGER,
3059 .arg1_type = ARG_PTR_TO_CTX,
3062 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3064 return dst_tclassid(skb);
3067 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3068 .func = bpf_get_route_realm,
3070 .ret_type = RET_INTEGER,
3071 .arg1_type = ARG_PTR_TO_CTX,
3074 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3076 /* If skb_clear_hash() was called due to mangling, we can
3077 * trigger SW recalculation here. Later access to hash
3078 * can then use the inline skb->hash via context directly
3079 * instead of calling this helper again.
3081 return skb_get_hash(skb);
3084 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3085 .func = bpf_get_hash_recalc,
3087 .ret_type = RET_INTEGER,
3088 .arg1_type = ARG_PTR_TO_CTX,
3091 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3093 /* After all direct packet write, this can be used once for
3094 * triggering a lazy recalc on next skb_get_hash() invocation.
3096 skb_clear_hash(skb);
3100 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3101 .func = bpf_set_hash_invalid,
3103 .ret_type = RET_INTEGER,
3104 .arg1_type = ARG_PTR_TO_CTX,
3107 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3109 /* Set user specified hash as L4(+), so that it gets returned
3110 * on skb_get_hash() call unless BPF prog later on triggers a
3113 __skb_set_sw_hash(skb, hash, true);
3117 static const struct bpf_func_proto bpf_set_hash_proto = {
3118 .func = bpf_set_hash,
3120 .ret_type = RET_INTEGER,
3121 .arg1_type = ARG_PTR_TO_CTX,
3122 .arg2_type = ARG_ANYTHING,
3125 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3130 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3131 vlan_proto != htons(ETH_P_8021AD)))
3132 vlan_proto = htons(ETH_P_8021Q);
3134 bpf_push_mac_rcsum(skb);
3135 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3136 bpf_pull_mac_rcsum(skb);
3138 bpf_compute_data_pointers(skb);
3142 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3143 .func = bpf_skb_vlan_push,
3145 .ret_type = RET_INTEGER,
3146 .arg1_type = ARG_PTR_TO_CTX,
3147 .arg2_type = ARG_ANYTHING,
3148 .arg3_type = ARG_ANYTHING,
3151 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3155 bpf_push_mac_rcsum(skb);
3156 ret = skb_vlan_pop(skb);
3157 bpf_pull_mac_rcsum(skb);
3159 bpf_compute_data_pointers(skb);
3163 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3164 .func = bpf_skb_vlan_pop,
3166 .ret_type = RET_INTEGER,
3167 .arg1_type = ARG_PTR_TO_CTX,
3170 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3172 /* Caller already did skb_cow() with len as headroom,
3173 * so no need to do it here.
3176 memmove(skb->data, skb->data + len, off);
3177 memset(skb->data + off, 0, len);
3179 /* No skb_postpush_rcsum(skb, skb->data + off, len)
3180 * needed here as it does not change the skb->csum
3181 * result for checksum complete when summing over
3187 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3189 /* skb_ensure_writable() is not needed here, as we're
3190 * already working on an uncloned skb.
3192 if (unlikely(!pskb_may_pull(skb, off + len)))
3195 skb_postpull_rcsum(skb, skb->data + off, len);
3196 memmove(skb->data + len, skb->data, off);
3197 __skb_pull(skb, len);
3202 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3204 bool trans_same = skb->transport_header == skb->network_header;
3207 /* There's no need for __skb_push()/__skb_pull() pair to
3208 * get to the start of the mac header as we're guaranteed
3209 * to always start from here under eBPF.
3211 ret = bpf_skb_generic_push(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_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3224 bool trans_same = skb->transport_header == skb->network_header;
3227 /* Same here, __skb_push()/__skb_pull() pair not needed. */
3228 ret = bpf_skb_generic_pop(skb, off, len);
3230 skb->mac_header += len;
3231 skb->network_header += len;
3233 skb->transport_header = skb->network_header;
3239 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3241 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3242 u32 off = skb_mac_header_len(skb);
3245 ret = skb_cow(skb, len_diff);
3246 if (unlikely(ret < 0))
3249 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3250 if (unlikely(ret < 0))
3253 if (skb_is_gso(skb)) {
3254 struct skb_shared_info *shinfo = skb_shinfo(skb);
3256 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */
3257 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3258 shinfo->gso_type &= ~SKB_GSO_TCPV4;
3259 shinfo->gso_type |= SKB_GSO_TCPV6;
3263 skb->protocol = htons(ETH_P_IPV6);
3264 skb_clear_hash(skb);
3269 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3271 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3272 u32 off = skb_mac_header_len(skb);
3275 ret = skb_unclone(skb, GFP_ATOMIC);
3276 if (unlikely(ret < 0))
3279 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3280 if (unlikely(ret < 0))
3283 if (skb_is_gso(skb)) {
3284 struct skb_shared_info *shinfo = skb_shinfo(skb);
3286 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */
3287 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3288 shinfo->gso_type &= ~SKB_GSO_TCPV6;
3289 shinfo->gso_type |= SKB_GSO_TCPV4;
3293 skb->protocol = htons(ETH_P_IP);
3294 skb_clear_hash(skb);
3299 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3301 __be16 from_proto = skb->protocol;
3303 if (from_proto == htons(ETH_P_IP) &&
3304 to_proto == htons(ETH_P_IPV6))
3305 return bpf_skb_proto_4_to_6(skb);
3307 if (from_proto == htons(ETH_P_IPV6) &&
3308 to_proto == htons(ETH_P_IP))
3309 return bpf_skb_proto_6_to_4(skb);
3314 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3319 if (unlikely(flags))
3322 /* General idea is that this helper does the basic groundwork
3323 * needed for changing the protocol, and eBPF program fills the
3324 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3325 * and other helpers, rather than passing a raw buffer here.
3327 * The rationale is to keep this minimal and without a need to
3328 * deal with raw packet data. F.e. even if we would pass buffers
3329 * here, the program still needs to call the bpf_lX_csum_replace()
3330 * helpers anyway. Plus, this way we keep also separation of
3331 * concerns, since f.e. bpf_skb_store_bytes() should only take
3334 * Currently, additional options and extension header space are
3335 * not supported, but flags register is reserved so we can adapt
3336 * that. For offloads, we mark packet as dodgy, so that headers
3337 * need to be verified first.
3339 ret = bpf_skb_proto_xlat(skb, proto);
3340 bpf_compute_data_pointers(skb);
3344 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3345 .func = bpf_skb_change_proto,
3347 .ret_type = RET_INTEGER,
3348 .arg1_type = ARG_PTR_TO_CTX,
3349 .arg2_type = ARG_ANYTHING,
3350 .arg3_type = ARG_ANYTHING,
3353 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3355 /* We only allow a restricted subset to be changed for now. */
3356 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3357 !skb_pkt_type_ok(pkt_type)))
3360 skb->pkt_type = pkt_type;
3364 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3365 .func = bpf_skb_change_type,
3367 .ret_type = RET_INTEGER,
3368 .arg1_type = ARG_PTR_TO_CTX,
3369 .arg2_type = ARG_ANYTHING,
3372 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3374 switch (skb->protocol) {
3375 case htons(ETH_P_IP):
3376 return sizeof(struct iphdr);
3377 case htons(ETH_P_IPV6):
3378 return sizeof(struct ipv6hdr);
3384 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3385 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3387 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \
3388 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3389 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3390 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3391 BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \
3392 BPF_F_ADJ_ROOM_ENCAP_L2( \
3393 BPF_ADJ_ROOM_ENCAP_L2_MASK))
3395 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3398 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3399 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3400 u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3401 unsigned int gso_type = SKB_GSO_DODGY;
3404 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3405 /* udp gso_size delineates datagrams, only allow if fixed */
3406 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3407 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3411 ret = skb_cow_head(skb, len_diff);
3412 if (unlikely(ret < 0))
3416 if (skb->protocol != htons(ETH_P_IP) &&
3417 skb->protocol != htons(ETH_P_IPV6))
3420 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3421 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3424 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3425 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3428 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH &&
3429 inner_mac_len < ETH_HLEN)
3432 if (skb->encapsulation)
3435 mac_len = skb->network_header - skb->mac_header;
3436 inner_net = skb->network_header;
3437 if (inner_mac_len > len_diff)
3439 inner_trans = skb->transport_header;
3442 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3443 if (unlikely(ret < 0))
3447 skb->inner_mac_header = inner_net - inner_mac_len;
3448 skb->inner_network_header = inner_net;
3449 skb->inner_transport_header = inner_trans;
3451 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH)
3452 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
3454 skb_set_inner_protocol(skb, skb->protocol);
3456 skb->encapsulation = 1;
3457 skb_set_network_header(skb, mac_len);
3459 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3460 gso_type |= SKB_GSO_UDP_TUNNEL;
3461 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3462 gso_type |= SKB_GSO_GRE;
3463 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3464 gso_type |= SKB_GSO_IPXIP6;
3465 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3466 gso_type |= SKB_GSO_IPXIP4;
3468 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3469 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3470 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3471 sizeof(struct ipv6hdr) :
3472 sizeof(struct iphdr);
3474 skb_set_transport_header(skb, mac_len + nh_len);
3477 /* Match skb->protocol to new outer l3 protocol */
3478 if (skb->protocol == htons(ETH_P_IP) &&
3479 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3480 skb->protocol = htons(ETH_P_IPV6);
3481 else if (skb->protocol == htons(ETH_P_IPV6) &&
3482 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3483 skb->protocol = htons(ETH_P_IP);
3486 if (skb_is_gso(skb)) {
3487 struct skb_shared_info *shinfo = skb_shinfo(skb);
3489 /* Due to header grow, MSS needs to be downgraded. */
3490 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3491 skb_decrease_gso_size(shinfo, len_diff);
3493 /* Header must be checked, and gso_segs recomputed. */
3494 shinfo->gso_type |= gso_type;
3495 shinfo->gso_segs = 0;
3501 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3506 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3507 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3510 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3511 /* udp gso_size delineates datagrams, only allow if fixed */
3512 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3513 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3517 ret = skb_unclone(skb, GFP_ATOMIC);
3518 if (unlikely(ret < 0))
3521 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3522 if (unlikely(ret < 0))
3525 if (skb_is_gso(skb)) {
3526 struct skb_shared_info *shinfo = skb_shinfo(skb);
3528 /* Due to header shrink, MSS can be upgraded. */
3529 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3530 skb_increase_gso_size(shinfo, len_diff);
3532 /* Header must be checked, and gso_segs recomputed. */
3533 shinfo->gso_type |= SKB_GSO_DODGY;
3534 shinfo->gso_segs = 0;
3540 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3542 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3543 u32, mode, u64, flags)
3545 u32 len_diff_abs = abs(len_diff);
3546 bool shrink = len_diff < 0;
3549 if (unlikely(flags || mode))
3551 if (unlikely(len_diff_abs > 0xfffU))
3555 ret = skb_cow(skb, len_diff);
3556 if (unlikely(ret < 0))
3558 __skb_push(skb, len_diff_abs);
3559 memset(skb->data, 0, len_diff_abs);
3561 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3563 __skb_pull(skb, len_diff_abs);
3565 if (tls_sw_has_ctx_rx(skb->sk)) {
3566 struct strp_msg *rxm = strp_msg(skb);
3568 rxm->full_len += len_diff;
3573 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3574 .func = sk_skb_adjust_room,
3576 .ret_type = RET_INTEGER,
3577 .arg1_type = ARG_PTR_TO_CTX,
3578 .arg2_type = ARG_ANYTHING,
3579 .arg3_type = ARG_ANYTHING,
3580 .arg4_type = ARG_ANYTHING,
3583 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3584 u32, mode, u64, flags)
3586 u32 len_cur, len_diff_abs = abs(len_diff);
3587 u32 len_min = bpf_skb_net_base_len(skb);
3588 u32 len_max = BPF_SKB_MAX_LEN;
3589 __be16 proto = skb->protocol;
3590 bool shrink = len_diff < 0;
3594 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3595 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3597 if (unlikely(len_diff_abs > 0xfffU))
3599 if (unlikely(proto != htons(ETH_P_IP) &&
3600 proto != htons(ETH_P_IPV6)))
3603 off = skb_mac_header_len(skb);
3605 case BPF_ADJ_ROOM_NET:
3606 off += bpf_skb_net_base_len(skb);
3608 case BPF_ADJ_ROOM_MAC:
3614 len_cur = skb->len - skb_network_offset(skb);
3615 if ((shrink && (len_diff_abs >= len_cur ||
3616 len_cur - len_diff_abs < len_min)) ||
3617 (!shrink && (skb->len + len_diff_abs > len_max &&
3621 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3622 bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3623 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3624 __skb_reset_checksum_unnecessary(skb);
3626 bpf_compute_data_pointers(skb);
3630 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3631 .func = bpf_skb_adjust_room,
3633 .ret_type = RET_INTEGER,
3634 .arg1_type = ARG_PTR_TO_CTX,
3635 .arg2_type = ARG_ANYTHING,
3636 .arg3_type = ARG_ANYTHING,
3637 .arg4_type = ARG_ANYTHING,
3640 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3642 u32 min_len = skb_network_offset(skb);
3644 if (skb_transport_header_was_set(skb))
3645 min_len = skb_transport_offset(skb);
3646 if (skb->ip_summed == CHECKSUM_PARTIAL)
3647 min_len = skb_checksum_start_offset(skb) +
3648 skb->csum_offset + sizeof(__sum16);
3652 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3654 unsigned int old_len = skb->len;
3657 ret = __skb_grow_rcsum(skb, new_len);
3659 memset(skb->data + old_len, 0, new_len - old_len);
3663 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3665 return __skb_trim_rcsum(skb, new_len);
3668 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3671 u32 max_len = BPF_SKB_MAX_LEN;
3672 u32 min_len = __bpf_skb_min_len(skb);
3675 if (unlikely(flags || new_len > max_len || new_len < min_len))
3677 if (skb->encapsulation)
3680 /* The basic idea of this helper is that it's performing the
3681 * needed work to either grow or trim an skb, and eBPF program
3682 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3683 * bpf_lX_csum_replace() and others rather than passing a raw
3684 * buffer here. This one is a slow path helper and intended
3685 * for replies with control messages.
3687 * Like in bpf_skb_change_proto(), we want to keep this rather
3688 * minimal and without protocol specifics so that we are able
3689 * to separate concerns as in bpf_skb_store_bytes() should only
3690 * be the one responsible for writing buffers.
3692 * It's really expected to be a slow path operation here for
3693 * control message replies, so we're implicitly linearizing,
3694 * uncloning and drop offloads from the skb by this.
3696 ret = __bpf_try_make_writable(skb, skb->len);
3698 if (new_len > skb->len)
3699 ret = bpf_skb_grow_rcsum(skb, new_len);
3700 else if (new_len < skb->len)
3701 ret = bpf_skb_trim_rcsum(skb, new_len);
3702 if (!ret && skb_is_gso(skb))
3708 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3711 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3713 bpf_compute_data_pointers(skb);
3717 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3718 .func = bpf_skb_change_tail,
3720 .ret_type = RET_INTEGER,
3721 .arg1_type = ARG_PTR_TO_CTX,
3722 .arg2_type = ARG_ANYTHING,
3723 .arg3_type = ARG_ANYTHING,
3726 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3729 return __bpf_skb_change_tail(skb, new_len, flags);
3732 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3733 .func = sk_skb_change_tail,
3735 .ret_type = RET_INTEGER,
3736 .arg1_type = ARG_PTR_TO_CTX,
3737 .arg2_type = ARG_ANYTHING,
3738 .arg3_type = ARG_ANYTHING,
3741 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3744 u32 max_len = BPF_SKB_MAX_LEN;
3745 u32 new_len = skb->len + head_room;
3748 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3749 new_len < skb->len))
3752 ret = skb_cow(skb, head_room);
3754 /* Idea for this helper is that we currently only
3755 * allow to expand on mac header. This means that
3756 * skb->protocol network header, etc, stay as is.
3757 * Compared to bpf_skb_change_tail(), we're more
3758 * flexible due to not needing to linearize or
3759 * reset GSO. Intention for this helper is to be
3760 * used by an L3 skb that needs to push mac header
3761 * for redirection into L2 device.
3763 __skb_push(skb, head_room);
3764 memset(skb->data, 0, head_room);
3765 skb_reset_mac_header(skb);
3766 skb_reset_mac_len(skb);
3772 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3775 int ret = __bpf_skb_change_head(skb, head_room, flags);
3777 bpf_compute_data_pointers(skb);
3781 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3782 .func = bpf_skb_change_head,
3784 .ret_type = RET_INTEGER,
3785 .arg1_type = ARG_PTR_TO_CTX,
3786 .arg2_type = ARG_ANYTHING,
3787 .arg3_type = ARG_ANYTHING,
3790 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3793 return __bpf_skb_change_head(skb, head_room, flags);
3796 static const struct bpf_func_proto sk_skb_change_head_proto = {
3797 .func = sk_skb_change_head,
3799 .ret_type = RET_INTEGER,
3800 .arg1_type = ARG_PTR_TO_CTX,
3801 .arg2_type = ARG_ANYTHING,
3802 .arg3_type = ARG_ANYTHING,
3804 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3806 return xdp_data_meta_unsupported(xdp) ? 0 :
3807 xdp->data - xdp->data_meta;
3810 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3812 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3813 unsigned long metalen = xdp_get_metalen(xdp);
3814 void *data_start = xdp_frame_end + metalen;
3815 void *data = xdp->data + offset;
3817 if (unlikely(data < data_start ||
3818 data > xdp->data_end - ETH_HLEN))
3822 memmove(xdp->data_meta + offset,
3823 xdp->data_meta, metalen);
3824 xdp->data_meta += offset;
3830 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3831 .func = bpf_xdp_adjust_head,
3833 .ret_type = RET_INTEGER,
3834 .arg1_type = ARG_PTR_TO_CTX,
3835 .arg2_type = ARG_ANYTHING,
3838 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3840 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
3841 void *data_end = xdp->data_end + offset;
3843 /* Notice that xdp_data_hard_end have reserved some tailroom */
3844 if (unlikely(data_end > data_hard_end))
3847 /* ALL drivers MUST init xdp->frame_sz, chicken check below */
3848 if (unlikely(xdp->frame_sz > PAGE_SIZE)) {
3849 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz);
3853 if (unlikely(data_end < xdp->data + ETH_HLEN))
3856 /* Clear memory area on grow, can contain uninit kernel memory */
3858 memset(xdp->data_end, 0, offset);
3860 xdp->data_end = data_end;
3865 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3866 .func = bpf_xdp_adjust_tail,
3868 .ret_type = RET_INTEGER,
3869 .arg1_type = ARG_PTR_TO_CTX,
3870 .arg2_type = ARG_ANYTHING,
3873 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3875 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3876 void *meta = xdp->data_meta + offset;
3877 unsigned long metalen = xdp->data - meta;
3879 if (xdp_data_meta_unsupported(xdp))
3881 if (unlikely(meta < xdp_frame_end ||
3884 if (unlikely(xdp_metalen_invalid(metalen)))
3887 xdp->data_meta = meta;
3892 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3893 .func = bpf_xdp_adjust_meta,
3895 .ret_type = RET_INTEGER,
3896 .arg1_type = ARG_PTR_TO_CTX,
3897 .arg2_type = ARG_ANYTHING,
3900 /* XDP_REDIRECT works by a three-step process, implemented in the functions
3903 * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target
3904 * of the redirect and store it (along with some other metadata) in a per-CPU
3905 * struct bpf_redirect_info.
3907 * 2. When the program returns the XDP_REDIRECT return code, the driver will
3908 * call xdp_do_redirect() which will use the information in struct
3909 * bpf_redirect_info to actually enqueue the frame into a map type-specific
3910 * bulk queue structure.
3912 * 3. Before exiting its NAPI poll loop, the driver will call xdp_do_flush(),
3913 * which will flush all the different bulk queues, thus completing the
3916 * Pointers to the map entries will be kept around for this whole sequence of
3917 * steps, protected by RCU. However, there is no top-level rcu_read_lock() in
3918 * the core code; instead, the RCU protection relies on everything happening
3919 * inside a single NAPI poll sequence, which means it's between a pair of calls
3920 * to local_bh_disable()/local_bh_enable().
3922 * The map entries are marked as __rcu and the map code makes sure to
3923 * dereference those pointers with rcu_dereference_check() in a way that works
3924 * for both sections that to hold an rcu_read_lock() and sections that are
3925 * called from NAPI without a separate rcu_read_lock(). The code below does not
3926 * use RCU annotations, but relies on those in the map code.
3928 void xdp_do_flush(void)
3934 EXPORT_SYMBOL_GPL(xdp_do_flush);
3936 void bpf_clear_redirect_map(struct bpf_map *map)
3938 struct bpf_redirect_info *ri;
3941 for_each_possible_cpu(cpu) {
3942 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3943 /* Avoid polluting remote cacheline due to writes if
3944 * not needed. Once we pass this test, we need the
3945 * cmpxchg() to make sure it hasn't been changed in
3946 * the meantime by remote CPU.
3948 if (unlikely(READ_ONCE(ri->map) == map))
3949 cmpxchg(&ri->map, map, NULL);
3953 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3954 struct bpf_prog *xdp_prog)
3956 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3957 enum bpf_map_type map_type = ri->map_type;
3958 void *fwd = ri->tgt_value;
3959 u32 map_id = ri->map_id;
3960 struct bpf_map *map;
3963 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
3964 ri->map_type = BPF_MAP_TYPE_UNSPEC;
3967 case BPF_MAP_TYPE_DEVMAP:
3969 case BPF_MAP_TYPE_DEVMAP_HASH:
3970 map = READ_ONCE(ri->map);
3971 if (unlikely(map)) {
3972 WRITE_ONCE(ri->map, NULL);
3973 err = dev_map_enqueue_multi(xdp, dev, map,
3974 ri->flags & BPF_F_EXCLUDE_INGRESS);
3976 err = dev_map_enqueue(fwd, xdp, dev);
3979 case BPF_MAP_TYPE_CPUMAP:
3980 err = cpu_map_enqueue(fwd, xdp, dev);
3982 case BPF_MAP_TYPE_XSKMAP:
3983 err = __xsk_map_redirect(fwd, xdp);
3985 case BPF_MAP_TYPE_UNSPEC:
3986 if (map_id == INT_MAX) {
3987 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
3988 if (unlikely(!fwd)) {
3992 err = dev_xdp_enqueue(fwd, xdp, dev);
4003 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4006 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4009 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4011 static int xdp_do_generic_redirect_map(struct net_device *dev,
4012 struct sk_buff *skb,
4013 struct xdp_buff *xdp,
4014 struct bpf_prog *xdp_prog,
4016 enum bpf_map_type map_type, u32 map_id)
4018 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4019 struct bpf_map *map;
4023 case BPF_MAP_TYPE_DEVMAP:
4025 case BPF_MAP_TYPE_DEVMAP_HASH:
4026 map = READ_ONCE(ri->map);
4027 if (unlikely(map)) {
4028 WRITE_ONCE(ri->map, NULL);
4029 err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4030 ri->flags & BPF_F_EXCLUDE_INGRESS);
4032 err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4037 case BPF_MAP_TYPE_XSKMAP:
4038 err = xsk_generic_rcv(fwd, xdp);
4043 case BPF_MAP_TYPE_CPUMAP:
4044 err = cpu_map_generic_redirect(fwd, skb);
4053 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4056 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4060 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4061 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4063 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4064 enum bpf_map_type map_type = ri->map_type;
4065 void *fwd = ri->tgt_value;
4066 u32 map_id = ri->map_id;
4069 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4070 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4072 if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4073 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4074 if (unlikely(!fwd)) {
4079 err = xdp_ok_fwd_dev(fwd, skb->len);
4084 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4085 generic_xdp_tx(skb, xdp_prog);
4089 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id);
4091 _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4095 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4097 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4099 if (unlikely(flags))
4102 /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4103 * by map_idr) is used for ifindex based XDP redirect.
4105 ri->tgt_index = ifindex;
4106 ri->map_id = INT_MAX;
4107 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4109 return XDP_REDIRECT;
4112 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4113 .func = bpf_xdp_redirect,
4115 .ret_type = RET_INTEGER,
4116 .arg1_type = ARG_ANYTHING,
4117 .arg2_type = ARG_ANYTHING,
4120 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
4123 return map->ops->map_redirect(map, ifindex, flags);
4126 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4127 .func = bpf_xdp_redirect_map,
4129 .ret_type = RET_INTEGER,
4130 .arg1_type = ARG_CONST_MAP_PTR,
4131 .arg2_type = ARG_ANYTHING,
4132 .arg3_type = ARG_ANYTHING,
4135 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4136 unsigned long off, unsigned long len)
4138 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4142 if (ptr != dst_buff)
4143 memcpy(dst_buff, ptr, len);
4148 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4149 u64, flags, void *, meta, u64, meta_size)
4151 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4153 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4155 if (unlikely(!skb || skb_size > skb->len))
4158 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4162 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4163 .func = bpf_skb_event_output,
4165 .ret_type = RET_INTEGER,
4166 .arg1_type = ARG_PTR_TO_CTX,
4167 .arg2_type = ARG_CONST_MAP_PTR,
4168 .arg3_type = ARG_ANYTHING,
4169 .arg4_type = ARG_PTR_TO_MEM,
4170 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4173 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4175 const struct bpf_func_proto bpf_skb_output_proto = {
4176 .func = bpf_skb_event_output,
4178 .ret_type = RET_INTEGER,
4179 .arg1_type = ARG_PTR_TO_BTF_ID,
4180 .arg1_btf_id = &bpf_skb_output_btf_ids[0],
4181 .arg2_type = ARG_CONST_MAP_PTR,
4182 .arg3_type = ARG_ANYTHING,
4183 .arg4_type = ARG_PTR_TO_MEM,
4184 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4187 static unsigned short bpf_tunnel_key_af(u64 flags)
4189 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4192 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4193 u32, size, u64, flags)
4195 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4196 u8 compat[sizeof(struct bpf_tunnel_key)];
4200 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
4204 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4208 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4211 case offsetof(struct bpf_tunnel_key, tunnel_label):
4212 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4214 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4215 /* Fixup deprecated structure layouts here, so we have
4216 * a common path later on.
4218 if (ip_tunnel_info_af(info) != AF_INET)
4221 to = (struct bpf_tunnel_key *)compat;
4228 to->tunnel_id = be64_to_cpu(info->key.tun_id);
4229 to->tunnel_tos = info->key.tos;
4230 to->tunnel_ttl = info->key.ttl;
4233 if (flags & BPF_F_TUNINFO_IPV6) {
4234 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4235 sizeof(to->remote_ipv6));
4236 to->tunnel_label = be32_to_cpu(info->key.label);
4238 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4239 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4240 to->tunnel_label = 0;
4243 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4244 memcpy(to_orig, to, size);
4248 memset(to_orig, 0, size);
4252 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4253 .func = bpf_skb_get_tunnel_key,
4255 .ret_type = RET_INTEGER,
4256 .arg1_type = ARG_PTR_TO_CTX,
4257 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4258 .arg3_type = ARG_CONST_SIZE,
4259 .arg4_type = ARG_ANYTHING,
4262 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4264 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4267 if (unlikely(!info ||
4268 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4272 if (unlikely(size < info->options_len)) {
4277 ip_tunnel_info_opts_get(to, info);
4278 if (size > info->options_len)
4279 memset(to + info->options_len, 0, size - info->options_len);
4281 return info->options_len;
4283 memset(to, 0, size);
4287 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4288 .func = bpf_skb_get_tunnel_opt,
4290 .ret_type = RET_INTEGER,
4291 .arg1_type = ARG_PTR_TO_CTX,
4292 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4293 .arg3_type = ARG_CONST_SIZE,
4296 static struct metadata_dst __percpu *md_dst;
4298 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4299 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4301 struct metadata_dst *md = this_cpu_ptr(md_dst);
4302 u8 compat[sizeof(struct bpf_tunnel_key)];
4303 struct ip_tunnel_info *info;
4305 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4306 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
4308 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4310 case offsetof(struct bpf_tunnel_key, tunnel_label):
4311 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4312 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4313 /* Fixup deprecated structure layouts here, so we have
4314 * a common path later on.
4316 memcpy(compat, from, size);
4317 memset(compat + size, 0, sizeof(compat) - size);
4318 from = (const struct bpf_tunnel_key *) compat;
4324 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4329 dst_hold((struct dst_entry *) md);
4330 skb_dst_set(skb, (struct dst_entry *) md);
4332 info = &md->u.tun_info;
4333 memset(info, 0, sizeof(*info));
4334 info->mode = IP_TUNNEL_INFO_TX;
4336 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4337 if (flags & BPF_F_DONT_FRAGMENT)
4338 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4339 if (flags & BPF_F_ZERO_CSUM_TX)
4340 info->key.tun_flags &= ~TUNNEL_CSUM;
4341 if (flags & BPF_F_SEQ_NUMBER)
4342 info->key.tun_flags |= TUNNEL_SEQ;
4344 info->key.tun_id = cpu_to_be64(from->tunnel_id);
4345 info->key.tos = from->tunnel_tos;
4346 info->key.ttl = from->tunnel_ttl;
4348 if (flags & BPF_F_TUNINFO_IPV6) {
4349 info->mode |= IP_TUNNEL_INFO_IPV6;
4350 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4351 sizeof(from->remote_ipv6));
4352 info->key.label = cpu_to_be32(from->tunnel_label) &
4353 IPV6_FLOWLABEL_MASK;
4355 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4361 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4362 .func = bpf_skb_set_tunnel_key,
4364 .ret_type = RET_INTEGER,
4365 .arg1_type = ARG_PTR_TO_CTX,
4366 .arg2_type = ARG_PTR_TO_MEM,
4367 .arg3_type = ARG_CONST_SIZE,
4368 .arg4_type = ARG_ANYTHING,
4371 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4372 const u8 *, from, u32, size)
4374 struct ip_tunnel_info *info = skb_tunnel_info(skb);
4375 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4377 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4379 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4382 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4387 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4388 .func = bpf_skb_set_tunnel_opt,
4390 .ret_type = RET_INTEGER,
4391 .arg1_type = ARG_PTR_TO_CTX,
4392 .arg2_type = ARG_PTR_TO_MEM,
4393 .arg3_type = ARG_CONST_SIZE,
4396 static const struct bpf_func_proto *
4397 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4400 struct metadata_dst __percpu *tmp;
4402 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4407 if (cmpxchg(&md_dst, NULL, tmp))
4408 metadata_dst_free_percpu(tmp);
4412 case BPF_FUNC_skb_set_tunnel_key:
4413 return &bpf_skb_set_tunnel_key_proto;
4414 case BPF_FUNC_skb_set_tunnel_opt:
4415 return &bpf_skb_set_tunnel_opt_proto;
4421 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4424 struct bpf_array *array = container_of(map, struct bpf_array, map);
4425 struct cgroup *cgrp;
4428 sk = skb_to_full_sk(skb);
4429 if (!sk || !sk_fullsock(sk))
4431 if (unlikely(idx >= array->map.max_entries))
4434 cgrp = READ_ONCE(array->ptrs[idx]);
4435 if (unlikely(!cgrp))
4438 return sk_under_cgroup_hierarchy(sk, cgrp);
4441 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4442 .func = bpf_skb_under_cgroup,
4444 .ret_type = RET_INTEGER,
4445 .arg1_type = ARG_PTR_TO_CTX,
4446 .arg2_type = ARG_CONST_MAP_PTR,
4447 .arg3_type = ARG_ANYTHING,
4450 #ifdef CONFIG_SOCK_CGROUP_DATA
4451 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4453 struct cgroup *cgrp;
4455 sk = sk_to_full_sk(sk);
4456 if (!sk || !sk_fullsock(sk))
4459 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4460 return cgroup_id(cgrp);
4463 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4465 return __bpf_sk_cgroup_id(skb->sk);
4468 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4469 .func = bpf_skb_cgroup_id,
4471 .ret_type = RET_INTEGER,
4472 .arg1_type = ARG_PTR_TO_CTX,
4475 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4478 struct cgroup *ancestor;
4479 struct cgroup *cgrp;
4481 sk = sk_to_full_sk(sk);
4482 if (!sk || !sk_fullsock(sk))
4485 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4486 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4490 return cgroup_id(ancestor);
4493 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4496 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4499 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4500 .func = bpf_skb_ancestor_cgroup_id,
4502 .ret_type = RET_INTEGER,
4503 .arg1_type = ARG_PTR_TO_CTX,
4504 .arg2_type = ARG_ANYTHING,
4507 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4509 return __bpf_sk_cgroup_id(sk);
4512 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4513 .func = bpf_sk_cgroup_id,
4515 .ret_type = RET_INTEGER,
4516 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4519 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4521 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4524 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4525 .func = bpf_sk_ancestor_cgroup_id,
4527 .ret_type = RET_INTEGER,
4528 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4529 .arg2_type = ARG_ANYTHING,
4533 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4534 unsigned long off, unsigned long len)
4536 memcpy(dst_buff, src_buff + off, len);
4540 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4541 u64, flags, void *, meta, u64, meta_size)
4543 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4545 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4547 if (unlikely(!xdp ||
4548 xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4551 return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4552 xdp_size, bpf_xdp_copy);
4555 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4556 .func = bpf_xdp_event_output,
4558 .ret_type = RET_INTEGER,
4559 .arg1_type = ARG_PTR_TO_CTX,
4560 .arg2_type = ARG_CONST_MAP_PTR,
4561 .arg3_type = ARG_ANYTHING,
4562 .arg4_type = ARG_PTR_TO_MEM,
4563 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4566 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4568 const struct bpf_func_proto bpf_xdp_output_proto = {
4569 .func = bpf_xdp_event_output,
4571 .ret_type = RET_INTEGER,
4572 .arg1_type = ARG_PTR_TO_BTF_ID,
4573 .arg1_btf_id = &bpf_xdp_output_btf_ids[0],
4574 .arg2_type = ARG_CONST_MAP_PTR,
4575 .arg3_type = ARG_ANYTHING,
4576 .arg4_type = ARG_PTR_TO_MEM,
4577 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4580 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4582 return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4585 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4586 .func = bpf_get_socket_cookie,
4588 .ret_type = RET_INTEGER,
4589 .arg1_type = ARG_PTR_TO_CTX,
4592 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4594 return __sock_gen_cookie(ctx->sk);
4597 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4598 .func = bpf_get_socket_cookie_sock_addr,
4600 .ret_type = RET_INTEGER,
4601 .arg1_type = ARG_PTR_TO_CTX,
4604 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4606 return __sock_gen_cookie(ctx);
4609 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4610 .func = bpf_get_socket_cookie_sock,
4612 .ret_type = RET_INTEGER,
4613 .arg1_type = ARG_PTR_TO_CTX,
4616 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
4618 return sk ? sock_gen_cookie(sk) : 0;
4621 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
4622 .func = bpf_get_socket_ptr_cookie,
4624 .ret_type = RET_INTEGER,
4625 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4628 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4630 return __sock_gen_cookie(ctx->sk);
4633 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4634 .func = bpf_get_socket_cookie_sock_ops,
4636 .ret_type = RET_INTEGER,
4637 .arg1_type = ARG_PTR_TO_CTX,
4640 static u64 __bpf_get_netns_cookie(struct sock *sk)
4642 const struct net *net = sk ? sock_net(sk) : &init_net;
4644 return net->net_cookie;
4647 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
4649 return __bpf_get_netns_cookie(ctx);
4652 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
4653 .func = bpf_get_netns_cookie_sock,
4655 .ret_type = RET_INTEGER,
4656 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4659 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4661 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4664 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
4665 .func = bpf_get_netns_cookie_sock_addr,
4667 .ret_type = RET_INTEGER,
4668 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4671 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4673 struct sock *sk = sk_to_full_sk(skb->sk);
4676 if (!sk || !sk_fullsock(sk))
4678 kuid = sock_net_uid(sock_net(sk), sk);
4679 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4682 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4683 .func = bpf_get_socket_uid,
4685 .ret_type = RET_INTEGER,
4686 .arg1_type = ARG_PTR_TO_CTX,
4689 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
4690 char *optval, int optlen)
4692 char devname[IFNAMSIZ];
4698 if (!sk_fullsock(sk))
4701 sock_owned_by_me(sk);
4703 if (level == SOL_SOCKET) {
4704 if (optlen != sizeof(int) && optname != SO_BINDTODEVICE)
4706 val = *((int *)optval);
4707 valbool = val ? 1 : 0;
4709 /* Only some socketops are supported */
4712 val = min_t(u32, val, sysctl_rmem_max);
4713 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4714 WRITE_ONCE(sk->sk_rcvbuf,
4715 max_t(int, val * 2, SOCK_MIN_RCVBUF));
4718 val = min_t(u32, val, sysctl_wmem_max);
4719 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4720 WRITE_ONCE(sk->sk_sndbuf,
4721 max_t(int, val * 2, SOCK_MIN_SNDBUF));
4723 case SO_MAX_PACING_RATE: /* 32bit version */
4725 cmpxchg(&sk->sk_pacing_status,
4728 sk->sk_max_pacing_rate = (val == ~0U) ?
4729 ~0UL : (unsigned int)val;
4730 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4731 sk->sk_max_pacing_rate);
4734 sk->sk_priority = val;
4739 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
4742 if (sk->sk_mark != val) {
4747 case SO_BINDTODEVICE:
4748 optlen = min_t(long, optlen, IFNAMSIZ - 1);
4749 strncpy(devname, optval, optlen);
4750 devname[optlen] = 0;
4753 if (devname[0] != '\0') {
4754 struct net_device *dev;
4759 dev = dev_get_by_name(net, devname);
4762 ifindex = dev->ifindex;
4766 case SO_BINDTOIFINDEX:
4767 if (optname == SO_BINDTOIFINDEX)
4769 ret = sock_bindtoindex(sk, ifindex, false);
4772 if (sk->sk_prot->keepalive)
4773 sk->sk_prot->keepalive(sk, valbool);
4774 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
4777 sk->sk_reuseport = valbool;
4783 } else if (level == SOL_IP) {
4784 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4787 val = *((int *)optval);
4788 /* Only some options are supported */
4791 if (val < -1 || val > 0xff) {
4794 struct inet_sock *inet = inet_sk(sk);
4804 #if IS_ENABLED(CONFIG_IPV6)
4805 } else if (level == SOL_IPV6) {
4806 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4809 val = *((int *)optval);
4810 /* Only some options are supported */
4813 if (val < -1 || val > 0xff) {
4816 struct ipv6_pinfo *np = inet6_sk(sk);
4827 } else if (level == SOL_TCP &&
4828 sk->sk_prot->setsockopt == tcp_setsockopt) {
4829 if (optname == TCP_CONGESTION) {
4830 char name[TCP_CA_NAME_MAX];
4832 strncpy(name, optval, min_t(long, optlen,
4833 TCP_CA_NAME_MAX-1));
4834 name[TCP_CA_NAME_MAX-1] = 0;
4835 ret = tcp_set_congestion_control(sk, name, false, true);
4837 struct inet_connection_sock *icsk = inet_csk(sk);
4838 struct tcp_sock *tp = tcp_sk(sk);
4839 unsigned long timeout;
4841 if (optlen != sizeof(int))
4844 val = *((int *)optval);
4845 /* Only some options are supported */
4848 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4853 case TCP_BPF_SNDCWND_CLAMP:
4857 tp->snd_cwnd_clamp = val;
4858 tp->snd_ssthresh = val;
4861 case TCP_BPF_DELACK_MAX:
4862 timeout = usecs_to_jiffies(val);
4863 if (timeout > TCP_DELACK_MAX ||
4864 timeout < TCP_TIMEOUT_MIN)
4866 inet_csk(sk)->icsk_delack_max = timeout;
4868 case TCP_BPF_RTO_MIN:
4869 timeout = usecs_to_jiffies(val);
4870 if (timeout > TCP_RTO_MIN ||
4871 timeout < TCP_TIMEOUT_MIN)
4873 inet_csk(sk)->icsk_rto_min = timeout;
4876 if (val < 0 || val > 1)
4882 ret = tcp_sock_set_keepidle_locked(sk, val);
4885 if (val < 1 || val > MAX_TCP_KEEPINTVL)
4888 tp->keepalive_intvl = val * HZ;
4891 if (val < 1 || val > MAX_TCP_KEEPCNT)
4894 tp->keepalive_probes = val;
4897 if (val < 1 || val > MAX_TCP_SYNCNT)
4900 icsk->icsk_syn_retries = val;
4902 case TCP_USER_TIMEOUT:
4906 icsk->icsk_user_timeout = val;
4908 case TCP_NOTSENT_LOWAT:
4909 tp->notsent_lowat = val;
4910 sk->sk_write_space(sk);
4912 case TCP_WINDOW_CLAMP:
4913 ret = tcp_set_window_clamp(sk, val);
4926 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
4927 char *optval, int optlen)
4929 if (!sk_fullsock(sk))
4932 sock_owned_by_me(sk);
4934 if (level == SOL_SOCKET) {
4935 if (optlen != sizeof(int))
4940 *((int *)optval) = sk->sk_mark;
4943 *((int *)optval) = sk->sk_priority;
4945 case SO_BINDTOIFINDEX:
4946 *((int *)optval) = sk->sk_bound_dev_if;
4949 *((int *)optval) = sk->sk_reuseport;
4955 } else if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4956 struct inet_connection_sock *icsk;
4957 struct tcp_sock *tp;
4960 case TCP_CONGESTION:
4961 icsk = inet_csk(sk);
4963 if (!icsk->icsk_ca_ops || optlen <= 1)
4965 strncpy(optval, icsk->icsk_ca_ops->name, optlen);
4966 optval[optlen - 1] = 0;
4971 if (optlen <= 0 || !tp->saved_syn ||
4972 optlen > tcp_saved_syn_len(tp->saved_syn))
4974 memcpy(optval, tp->saved_syn->data, optlen);
4979 } else if (level == SOL_IP) {
4980 struct inet_sock *inet = inet_sk(sk);
4982 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4985 /* Only some options are supported */
4988 *((int *)optval) = (int)inet->tos;
4993 #if IS_ENABLED(CONFIG_IPV6)
4994 } else if (level == SOL_IPV6) {
4995 struct ipv6_pinfo *np = inet6_sk(sk);
4997 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
5000 /* Only some options are supported */
5003 *((int *)optval) = (int)np->tclass;
5015 memset(optval, 0, optlen);
5019 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5020 int, level, int, optname, char *, optval, int, optlen)
5022 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5025 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5026 .func = bpf_sock_addr_setsockopt,
5028 .ret_type = RET_INTEGER,
5029 .arg1_type = ARG_PTR_TO_CTX,
5030 .arg2_type = ARG_ANYTHING,
5031 .arg3_type = ARG_ANYTHING,
5032 .arg4_type = ARG_PTR_TO_MEM,
5033 .arg5_type = ARG_CONST_SIZE,
5036 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5037 int, level, int, optname, char *, optval, int, optlen)
5039 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5042 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5043 .func = bpf_sock_addr_getsockopt,
5045 .ret_type = RET_INTEGER,
5046 .arg1_type = ARG_PTR_TO_CTX,
5047 .arg2_type = ARG_ANYTHING,
5048 .arg3_type = ARG_ANYTHING,
5049 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5050 .arg5_type = ARG_CONST_SIZE,
5053 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5054 int, level, int, optname, char *, optval, int, optlen)
5056 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5059 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5060 .func = bpf_sock_ops_setsockopt,
5062 .ret_type = RET_INTEGER,
5063 .arg1_type = ARG_PTR_TO_CTX,
5064 .arg2_type = ARG_ANYTHING,
5065 .arg3_type = ARG_ANYTHING,
5066 .arg4_type = ARG_PTR_TO_MEM,
5067 .arg5_type = ARG_CONST_SIZE,
5070 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5071 int optname, const u8 **start)
5073 struct sk_buff *syn_skb = bpf_sock->syn_skb;
5074 const u8 *hdr_start;
5078 /* sk is a request_sock here */
5080 if (optname == TCP_BPF_SYN) {
5081 hdr_start = syn_skb->data;
5082 ret = tcp_hdrlen(syn_skb);
5083 } else if (optname == TCP_BPF_SYN_IP) {
5084 hdr_start = skb_network_header(syn_skb);
5085 ret = skb_network_header_len(syn_skb) +
5086 tcp_hdrlen(syn_skb);
5088 /* optname == TCP_BPF_SYN_MAC */
5089 hdr_start = skb_mac_header(syn_skb);
5090 ret = skb_mac_header_len(syn_skb) +
5091 skb_network_header_len(syn_skb) +
5092 tcp_hdrlen(syn_skb);
5095 struct sock *sk = bpf_sock->sk;
5096 struct saved_syn *saved_syn;
5098 if (sk->sk_state == TCP_NEW_SYN_RECV)
5099 /* synack retransmit. bpf_sock->syn_skb will
5100 * not be available. It has to resort to
5101 * saved_syn (if it is saved).
5103 saved_syn = inet_reqsk(sk)->saved_syn;
5105 saved_syn = tcp_sk(sk)->saved_syn;
5110 if (optname == TCP_BPF_SYN) {
5111 hdr_start = saved_syn->data +
5112 saved_syn->mac_hdrlen +
5113 saved_syn->network_hdrlen;
5114 ret = saved_syn->tcp_hdrlen;
5115 } else if (optname == TCP_BPF_SYN_IP) {
5116 hdr_start = saved_syn->data +
5117 saved_syn->mac_hdrlen;
5118 ret = saved_syn->network_hdrlen +
5119 saved_syn->tcp_hdrlen;
5121 /* optname == TCP_BPF_SYN_MAC */
5123 /* TCP_SAVE_SYN may not have saved the mac hdr */
5124 if (!saved_syn->mac_hdrlen)
5127 hdr_start = saved_syn->data;
5128 ret = saved_syn->mac_hdrlen +
5129 saved_syn->network_hdrlen +
5130 saved_syn->tcp_hdrlen;
5138 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5139 int, level, int, optname, char *, optval, int, optlen)
5141 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5142 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5143 int ret, copy_len = 0;
5146 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5149 if (optlen < copy_len) {
5154 memcpy(optval, start, copy_len);
5157 /* Zero out unused buffer at the end */
5158 memset(optval + copy_len, 0, optlen - copy_len);
5163 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5166 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5167 .func = bpf_sock_ops_getsockopt,
5169 .ret_type = RET_INTEGER,
5170 .arg1_type = ARG_PTR_TO_CTX,
5171 .arg2_type = ARG_ANYTHING,
5172 .arg3_type = ARG_ANYTHING,
5173 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5174 .arg5_type = ARG_CONST_SIZE,
5177 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5180 struct sock *sk = bpf_sock->sk;
5181 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5183 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5186 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5188 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5191 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5192 .func = bpf_sock_ops_cb_flags_set,
5194 .ret_type = RET_INTEGER,
5195 .arg1_type = ARG_PTR_TO_CTX,
5196 .arg2_type = ARG_ANYTHING,
5199 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5200 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5202 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5206 struct sock *sk = ctx->sk;
5207 u32 flags = BIND_FROM_BPF;
5211 if (addr_len < offsetofend(struct sockaddr, sa_family))
5213 if (addr->sa_family == AF_INET) {
5214 if (addr_len < sizeof(struct sockaddr_in))
5216 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5217 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5218 return __inet_bind(sk, addr, addr_len, flags);
5219 #if IS_ENABLED(CONFIG_IPV6)
5220 } else if (addr->sa_family == AF_INET6) {
5221 if (addr_len < SIN6_LEN_RFC2133)
5223 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5224 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5225 /* ipv6_bpf_stub cannot be NULL, since it's called from
5226 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5228 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5229 #endif /* CONFIG_IPV6 */
5231 #endif /* CONFIG_INET */
5233 return -EAFNOSUPPORT;
5236 static const struct bpf_func_proto bpf_bind_proto = {
5239 .ret_type = RET_INTEGER,
5240 .arg1_type = ARG_PTR_TO_CTX,
5241 .arg2_type = ARG_PTR_TO_MEM,
5242 .arg3_type = ARG_CONST_SIZE,
5246 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5247 struct bpf_xfrm_state *, to, u32, size, u64, flags)
5249 const struct sec_path *sp = skb_sec_path(skb);
5250 const struct xfrm_state *x;
5252 if (!sp || unlikely(index >= sp->len || flags))
5255 x = sp->xvec[index];
5257 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5260 to->reqid = x->props.reqid;
5261 to->spi = x->id.spi;
5262 to->family = x->props.family;
5265 if (to->family == AF_INET6) {
5266 memcpy(to->remote_ipv6, x->props.saddr.a6,
5267 sizeof(to->remote_ipv6));
5269 to->remote_ipv4 = x->props.saddr.a4;
5270 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5275 memset(to, 0, size);
5279 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5280 .func = bpf_skb_get_xfrm_state,
5282 .ret_type = RET_INTEGER,
5283 .arg1_type = ARG_PTR_TO_CTX,
5284 .arg2_type = ARG_ANYTHING,
5285 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
5286 .arg4_type = ARG_CONST_SIZE,
5287 .arg5_type = ARG_ANYTHING,
5291 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5292 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
5293 const struct neighbour *neigh,
5294 const struct net_device *dev, u32 mtu)
5296 memcpy(params->dmac, neigh->ha, ETH_ALEN);
5297 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5298 params->h_vlan_TCI = 0;
5299 params->h_vlan_proto = 0;
5301 params->mtu_result = mtu; /* union with tot_len */
5307 #if IS_ENABLED(CONFIG_INET)
5308 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5309 u32 flags, bool check_mtu)
5311 struct fib_nh_common *nhc;
5312 struct in_device *in_dev;
5313 struct neighbour *neigh;
5314 struct net_device *dev;
5315 struct fib_result res;
5320 dev = dev_get_by_index_rcu(net, params->ifindex);
5324 /* verify forwarding is enabled on this interface */
5325 in_dev = __in_dev_get_rcu(dev);
5326 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5327 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5329 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5331 fl4.flowi4_oif = params->ifindex;
5333 fl4.flowi4_iif = params->ifindex;
5336 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5337 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5338 fl4.flowi4_flags = 0;
5340 fl4.flowi4_proto = params->l4_protocol;
5341 fl4.daddr = params->ipv4_dst;
5342 fl4.saddr = params->ipv4_src;
5343 fl4.fl4_sport = params->sport;
5344 fl4.fl4_dport = params->dport;
5345 fl4.flowi4_multipath_hash = 0;
5347 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5348 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5349 struct fib_table *tb;
5351 tb = fib_get_table(net, tbid);
5353 return BPF_FIB_LKUP_RET_NOT_FWDED;
5355 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5357 fl4.flowi4_mark = 0;
5358 fl4.flowi4_secid = 0;
5359 fl4.flowi4_tun_key.tun_id = 0;
5360 fl4.flowi4_uid = sock_net_uid(net, NULL);
5362 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5366 /* map fib lookup errors to RTN_ type */
5368 return BPF_FIB_LKUP_RET_BLACKHOLE;
5369 if (err == -EHOSTUNREACH)
5370 return BPF_FIB_LKUP_RET_UNREACHABLE;
5372 return BPF_FIB_LKUP_RET_PROHIBIT;
5374 return BPF_FIB_LKUP_RET_NOT_FWDED;
5377 if (res.type != RTN_UNICAST)
5378 return BPF_FIB_LKUP_RET_NOT_FWDED;
5380 if (fib_info_num_path(res.fi) > 1)
5381 fib_select_path(net, &res, &fl4, NULL);
5384 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5385 if (params->tot_len > mtu) {
5386 params->mtu_result = mtu; /* union with tot_len */
5387 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5393 /* do not handle lwt encaps right now */
5394 if (nhc->nhc_lwtstate)
5395 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5399 params->rt_metric = res.fi->fib_priority;
5400 params->ifindex = dev->ifindex;
5402 /* xdp and cls_bpf programs are run in RCU-bh so
5403 * rcu_read_lock_bh is not needed here
5405 if (likely(nhc->nhc_gw_family != AF_INET6)) {
5406 if (nhc->nhc_gw_family)
5407 params->ipv4_dst = nhc->nhc_gw.ipv4;
5409 neigh = __ipv4_neigh_lookup_noref(dev,
5410 (__force u32)params->ipv4_dst);
5412 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5414 params->family = AF_INET6;
5415 *dst = nhc->nhc_gw.ipv6;
5416 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5420 return BPF_FIB_LKUP_RET_NO_NEIGH;
5422 return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5426 #if IS_ENABLED(CONFIG_IPV6)
5427 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5428 u32 flags, bool check_mtu)
5430 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5431 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5432 struct fib6_result res = {};
5433 struct neighbour *neigh;
5434 struct net_device *dev;
5435 struct inet6_dev *idev;
5441 /* link local addresses are never forwarded */
5442 if (rt6_need_strict(dst) || rt6_need_strict(src))
5443 return BPF_FIB_LKUP_RET_NOT_FWDED;
5445 dev = dev_get_by_index_rcu(net, params->ifindex);
5449 idev = __in6_dev_get_safely(dev);
5450 if (unlikely(!idev || !idev->cnf.forwarding))
5451 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5453 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5455 oif = fl6.flowi6_oif = params->ifindex;
5457 oif = fl6.flowi6_iif = params->ifindex;
5459 strict = RT6_LOOKUP_F_HAS_SADDR;
5461 fl6.flowlabel = params->flowinfo;
5462 fl6.flowi6_scope = 0;
5463 fl6.flowi6_flags = 0;
5466 fl6.flowi6_proto = params->l4_protocol;
5469 fl6.fl6_sport = params->sport;
5470 fl6.fl6_dport = params->dport;
5472 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5473 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5474 struct fib6_table *tb;
5476 tb = ipv6_stub->fib6_get_table(net, tbid);
5478 return BPF_FIB_LKUP_RET_NOT_FWDED;
5480 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5483 fl6.flowi6_mark = 0;
5484 fl6.flowi6_secid = 0;
5485 fl6.flowi6_tun_key.tun_id = 0;
5486 fl6.flowi6_uid = sock_net_uid(net, NULL);
5488 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5491 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5492 res.f6i == net->ipv6.fib6_null_entry))
5493 return BPF_FIB_LKUP_RET_NOT_FWDED;
5495 switch (res.fib6_type) {
5496 /* only unicast is forwarded */
5500 return BPF_FIB_LKUP_RET_BLACKHOLE;
5501 case RTN_UNREACHABLE:
5502 return BPF_FIB_LKUP_RET_UNREACHABLE;
5504 return BPF_FIB_LKUP_RET_PROHIBIT;
5506 return BPF_FIB_LKUP_RET_NOT_FWDED;
5509 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5510 fl6.flowi6_oif != 0, NULL, strict);
5513 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5514 if (params->tot_len > mtu) {
5515 params->mtu_result = mtu; /* union with tot_len */
5516 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5520 if (res.nh->fib_nh_lws)
5521 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5523 if (res.nh->fib_nh_gw_family)
5524 *dst = res.nh->fib_nh_gw6;
5526 dev = res.nh->fib_nh_dev;
5527 params->rt_metric = res.f6i->fib6_metric;
5528 params->ifindex = dev->ifindex;
5530 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
5533 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5535 return BPF_FIB_LKUP_RET_NO_NEIGH;
5537 return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5541 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
5542 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5544 if (plen < sizeof(*params))
5547 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5550 switch (params->family) {
5551 #if IS_ENABLED(CONFIG_INET)
5553 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
5556 #if IS_ENABLED(CONFIG_IPV6)
5558 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
5562 return -EAFNOSUPPORT;
5565 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
5566 .func = bpf_xdp_fib_lookup,
5568 .ret_type = RET_INTEGER,
5569 .arg1_type = ARG_PTR_TO_CTX,
5570 .arg2_type = ARG_PTR_TO_MEM,
5571 .arg3_type = ARG_CONST_SIZE,
5572 .arg4_type = ARG_ANYTHING,
5575 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
5576 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5578 struct net *net = dev_net(skb->dev);
5579 int rc = -EAFNOSUPPORT;
5580 bool check_mtu = false;
5582 if (plen < sizeof(*params))
5585 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5588 if (params->tot_len)
5591 switch (params->family) {
5592 #if IS_ENABLED(CONFIG_INET)
5594 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
5597 #if IS_ENABLED(CONFIG_IPV6)
5599 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
5604 if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
5605 struct net_device *dev;
5607 /* When tot_len isn't provided by user, check skb
5608 * against MTU of FIB lookup resulting net_device
5610 dev = dev_get_by_index_rcu(net, params->ifindex);
5611 if (!is_skb_forwardable(dev, skb))
5612 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
5614 params->mtu_result = dev->mtu; /* union with tot_len */
5620 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
5621 .func = bpf_skb_fib_lookup,
5623 .ret_type = RET_INTEGER,
5624 .arg1_type = ARG_PTR_TO_CTX,
5625 .arg2_type = ARG_PTR_TO_MEM,
5626 .arg3_type = ARG_CONST_SIZE,
5627 .arg4_type = ARG_ANYTHING,
5630 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
5633 struct net *netns = dev_net(dev_curr);
5635 /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
5639 return dev_get_by_index_rcu(netns, ifindex);
5642 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
5643 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5645 int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5646 struct net_device *dev = skb->dev;
5647 int skb_len, dev_len;
5650 if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
5653 if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
5656 dev = __dev_via_ifindex(dev, ifindex);
5660 mtu = READ_ONCE(dev->mtu);
5662 dev_len = mtu + dev->hard_header_len;
5664 /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
5665 skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
5667 skb_len += len_diff; /* minus result pass check */
5668 if (skb_len <= dev_len) {
5669 ret = BPF_MTU_CHK_RET_SUCCESS;
5672 /* At this point, skb->len exceed MTU, but as it include length of all
5673 * segments, it can still be below MTU. The SKB can possibly get
5674 * re-segmented in transmit path (see validate_xmit_skb). Thus, user
5675 * must choose if segs are to be MTU checked.
5677 if (skb_is_gso(skb)) {
5678 ret = BPF_MTU_CHK_RET_SUCCESS;
5680 if (flags & BPF_MTU_CHK_SEGS &&
5681 !skb_gso_validate_network_len(skb, mtu))
5682 ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
5685 /* BPF verifier guarantees valid pointer */
5691 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
5692 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5694 struct net_device *dev = xdp->rxq->dev;
5695 int xdp_len = xdp->data_end - xdp->data;
5696 int ret = BPF_MTU_CHK_RET_SUCCESS;
5699 /* XDP variant doesn't support multi-buffer segment check (yet) */
5700 if (unlikely(flags))
5703 dev = __dev_via_ifindex(dev, ifindex);
5707 mtu = READ_ONCE(dev->mtu);
5709 /* Add L2-header as dev MTU is L3 size */
5710 dev_len = mtu + dev->hard_header_len;
5712 /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
5714 xdp_len = *mtu_len + dev->hard_header_len;
5716 xdp_len += len_diff; /* minus result pass check */
5717 if (xdp_len > dev_len)
5718 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5720 /* BPF verifier guarantees valid pointer */
5726 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
5727 .func = bpf_skb_check_mtu,
5729 .ret_type = RET_INTEGER,
5730 .arg1_type = ARG_PTR_TO_CTX,
5731 .arg2_type = ARG_ANYTHING,
5732 .arg3_type = ARG_PTR_TO_INT,
5733 .arg4_type = ARG_ANYTHING,
5734 .arg5_type = ARG_ANYTHING,
5737 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
5738 .func = bpf_xdp_check_mtu,
5740 .ret_type = RET_INTEGER,
5741 .arg1_type = ARG_PTR_TO_CTX,
5742 .arg2_type = ARG_ANYTHING,
5743 .arg3_type = ARG_PTR_TO_INT,
5744 .arg4_type = ARG_ANYTHING,
5745 .arg5_type = ARG_ANYTHING,
5748 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5749 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
5752 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
5754 if (!seg6_validate_srh(srh, len, false))
5758 case BPF_LWT_ENCAP_SEG6_INLINE:
5759 if (skb->protocol != htons(ETH_P_IPV6))
5762 err = seg6_do_srh_inline(skb, srh);
5764 case BPF_LWT_ENCAP_SEG6:
5765 skb_reset_inner_headers(skb);
5766 skb->encapsulation = 1;
5767 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
5773 bpf_compute_data_pointers(skb);
5777 ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5778 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
5780 return seg6_lookup_nexthop(skb, NULL, 0);
5782 #endif /* CONFIG_IPV6_SEG6_BPF */
5784 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5785 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
5788 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
5792 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
5796 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5797 case BPF_LWT_ENCAP_SEG6:
5798 case BPF_LWT_ENCAP_SEG6_INLINE:
5799 return bpf_push_seg6_encap(skb, type, hdr, len);
5801 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5802 case BPF_LWT_ENCAP_IP:
5803 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
5810 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
5811 void *, hdr, u32, len)
5814 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5815 case BPF_LWT_ENCAP_IP:
5816 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
5823 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
5824 .func = bpf_lwt_in_push_encap,
5826 .ret_type = RET_INTEGER,
5827 .arg1_type = ARG_PTR_TO_CTX,
5828 .arg2_type = ARG_ANYTHING,
5829 .arg3_type = ARG_PTR_TO_MEM,
5830 .arg4_type = ARG_CONST_SIZE
5833 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
5834 .func = bpf_lwt_xmit_push_encap,
5836 .ret_type = RET_INTEGER,
5837 .arg1_type = ARG_PTR_TO_CTX,
5838 .arg2_type = ARG_ANYTHING,
5839 .arg3_type = ARG_PTR_TO_MEM,
5840 .arg4_type = ARG_CONST_SIZE
5843 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5844 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
5845 const void *, from, u32, len)
5847 struct seg6_bpf_srh_state *srh_state =
5848 this_cpu_ptr(&seg6_bpf_srh_states);
5849 struct ipv6_sr_hdr *srh = srh_state->srh;
5850 void *srh_tlvs, *srh_end, *ptr;
5856 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
5857 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
5859 ptr = skb->data + offset;
5860 if (ptr >= srh_tlvs && ptr + len <= srh_end)
5861 srh_state->valid = false;
5862 else if (ptr < (void *)&srh->flags ||
5863 ptr + len > (void *)&srh->segments)
5866 if (unlikely(bpf_try_make_writable(skb, offset + len)))
5868 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5870 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5872 memcpy(skb->data + offset, from, len);
5876 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
5877 .func = bpf_lwt_seg6_store_bytes,
5879 .ret_type = RET_INTEGER,
5880 .arg1_type = ARG_PTR_TO_CTX,
5881 .arg2_type = ARG_ANYTHING,
5882 .arg3_type = ARG_PTR_TO_MEM,
5883 .arg4_type = ARG_CONST_SIZE
5886 static void bpf_update_srh_state(struct sk_buff *skb)
5888 struct seg6_bpf_srh_state *srh_state =
5889 this_cpu_ptr(&seg6_bpf_srh_states);
5892 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
5893 srh_state->srh = NULL;
5895 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5896 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
5897 srh_state->valid = true;
5901 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
5902 u32, action, void *, param, u32, param_len)
5904 struct seg6_bpf_srh_state *srh_state =
5905 this_cpu_ptr(&seg6_bpf_srh_states);
5910 case SEG6_LOCAL_ACTION_END_X:
5911 if (!seg6_bpf_has_valid_srh(skb))
5913 if (param_len != sizeof(struct in6_addr))
5915 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
5916 case SEG6_LOCAL_ACTION_END_T:
5917 if (!seg6_bpf_has_valid_srh(skb))
5919 if (param_len != sizeof(int))
5921 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5922 case SEG6_LOCAL_ACTION_END_DT6:
5923 if (!seg6_bpf_has_valid_srh(skb))
5925 if (param_len != sizeof(int))
5928 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
5930 if (!pskb_pull(skb, hdroff))
5933 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
5934 skb_reset_network_header(skb);
5935 skb_reset_transport_header(skb);
5936 skb->encapsulation = 0;
5938 bpf_compute_data_pointers(skb);
5939 bpf_update_srh_state(skb);
5940 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5941 case SEG6_LOCAL_ACTION_END_B6:
5942 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5944 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
5947 bpf_update_srh_state(skb);
5950 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
5951 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5953 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
5956 bpf_update_srh_state(skb);
5964 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
5965 .func = bpf_lwt_seg6_action,
5967 .ret_type = RET_INTEGER,
5968 .arg1_type = ARG_PTR_TO_CTX,
5969 .arg2_type = ARG_ANYTHING,
5970 .arg3_type = ARG_PTR_TO_MEM,
5971 .arg4_type = ARG_CONST_SIZE
5974 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
5977 struct seg6_bpf_srh_state *srh_state =
5978 this_cpu_ptr(&seg6_bpf_srh_states);
5979 struct ipv6_sr_hdr *srh = srh_state->srh;
5980 void *srh_end, *srh_tlvs, *ptr;
5981 struct ipv6hdr *hdr;
5985 if (unlikely(srh == NULL))
5988 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
5989 ((srh->first_segment + 1) << 4));
5990 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
5992 ptr = skb->data + offset;
5994 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
5996 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
6000 ret = skb_cow_head(skb, len);
6001 if (unlikely(ret < 0))
6004 ret = bpf_skb_net_hdr_push(skb, offset, len);
6006 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6009 bpf_compute_data_pointers(skb);
6010 if (unlikely(ret < 0))
6013 hdr = (struct ipv6hdr *)skb->data;
6014 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6016 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6018 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6019 srh_state->hdrlen += len;
6020 srh_state->valid = false;
6024 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6025 .func = bpf_lwt_seg6_adjust_srh,
6027 .ret_type = RET_INTEGER,
6028 .arg1_type = ARG_PTR_TO_CTX,
6029 .arg2_type = ARG_ANYTHING,
6030 .arg3_type = ARG_ANYTHING,
6032 #endif /* CONFIG_IPV6_SEG6_BPF */
6035 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6036 int dif, int sdif, u8 family, u8 proto)
6038 bool refcounted = false;
6039 struct sock *sk = NULL;
6041 if (family == AF_INET) {
6042 __be32 src4 = tuple->ipv4.saddr;
6043 __be32 dst4 = tuple->ipv4.daddr;
6045 if (proto == IPPROTO_TCP)
6046 sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
6047 src4, tuple->ipv4.sport,
6048 dst4, tuple->ipv4.dport,
6049 dif, sdif, &refcounted);
6051 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6052 dst4, tuple->ipv4.dport,
6053 dif, sdif, &udp_table, NULL);
6054 #if IS_ENABLED(CONFIG_IPV6)
6056 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6057 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6059 if (proto == IPPROTO_TCP)
6060 sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
6061 src6, tuple->ipv6.sport,
6062 dst6, ntohs(tuple->ipv6.dport),
6063 dif, sdif, &refcounted);
6064 else if (likely(ipv6_bpf_stub))
6065 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6066 src6, tuple->ipv6.sport,
6067 dst6, tuple->ipv6.dport,
6073 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6074 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6080 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6081 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6082 * Returns the socket as an 'unsigned long' to simplify the casting in the
6083 * callers to satisfy BPF_CALL declarations.
6085 static struct sock *
6086 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6087 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6090 struct sock *sk = NULL;
6091 u8 family = AF_UNSPEC;
6095 if (len == sizeof(tuple->ipv4))
6097 else if (len == sizeof(tuple->ipv6))
6102 if (unlikely(family == AF_UNSPEC || flags ||
6103 !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6106 if (family == AF_INET)
6107 sdif = inet_sdif(skb);
6109 sdif = inet6_sdif(skb);
6111 if ((s32)netns_id < 0) {
6113 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6115 net = get_net_ns_by_id(caller_net, netns_id);
6118 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6126 static struct sock *
6127 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6128 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6131 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6132 ifindex, proto, netns_id, flags);
6135 sk = sk_to_full_sk(sk);
6136 if (!sk_fullsock(sk)) {
6145 static struct sock *
6146 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6147 u8 proto, u64 netns_id, u64 flags)
6149 struct net *caller_net;
6153 caller_net = dev_net(skb->dev);
6154 ifindex = skb->dev->ifindex;
6156 caller_net = sock_net(skb->sk);
6160 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6164 static struct sock *
6165 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6166 u8 proto, u64 netns_id, u64 flags)
6168 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6172 sk = sk_to_full_sk(sk);
6173 if (!sk_fullsock(sk)) {
6182 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6183 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6185 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6189 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6190 .func = bpf_skc_lookup_tcp,
6193 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6194 .arg1_type = ARG_PTR_TO_CTX,
6195 .arg2_type = ARG_PTR_TO_MEM,
6196 .arg3_type = ARG_CONST_SIZE,
6197 .arg4_type = ARG_ANYTHING,
6198 .arg5_type = ARG_ANYTHING,
6201 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6202 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6204 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6208 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6209 .func = bpf_sk_lookup_tcp,
6212 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6213 .arg1_type = ARG_PTR_TO_CTX,
6214 .arg2_type = ARG_PTR_TO_MEM,
6215 .arg3_type = ARG_CONST_SIZE,
6216 .arg4_type = ARG_ANYTHING,
6217 .arg5_type = ARG_ANYTHING,
6220 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6221 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6223 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6227 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6228 .func = bpf_sk_lookup_udp,
6231 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6232 .arg1_type = ARG_PTR_TO_CTX,
6233 .arg2_type = ARG_PTR_TO_MEM,
6234 .arg3_type = ARG_CONST_SIZE,
6235 .arg4_type = ARG_ANYTHING,
6236 .arg5_type = ARG_ANYTHING,
6239 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6241 if (sk && sk_is_refcounted(sk))
6246 static const struct bpf_func_proto bpf_sk_release_proto = {
6247 .func = bpf_sk_release,
6249 .ret_type = RET_INTEGER,
6250 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6253 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6254 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6256 struct net *caller_net = dev_net(ctx->rxq->dev);
6257 int ifindex = ctx->rxq->dev->ifindex;
6259 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6260 ifindex, IPPROTO_UDP, netns_id,
6264 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6265 .func = bpf_xdp_sk_lookup_udp,
6268 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6269 .arg1_type = ARG_PTR_TO_CTX,
6270 .arg2_type = ARG_PTR_TO_MEM,
6271 .arg3_type = ARG_CONST_SIZE,
6272 .arg4_type = ARG_ANYTHING,
6273 .arg5_type = ARG_ANYTHING,
6276 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6277 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6279 struct net *caller_net = dev_net(ctx->rxq->dev);
6280 int ifindex = ctx->rxq->dev->ifindex;
6282 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6283 ifindex, IPPROTO_TCP, netns_id,
6287 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6288 .func = bpf_xdp_skc_lookup_tcp,
6291 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6292 .arg1_type = ARG_PTR_TO_CTX,
6293 .arg2_type = ARG_PTR_TO_MEM,
6294 .arg3_type = ARG_CONST_SIZE,
6295 .arg4_type = ARG_ANYTHING,
6296 .arg5_type = ARG_ANYTHING,
6299 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6300 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6302 struct net *caller_net = dev_net(ctx->rxq->dev);
6303 int ifindex = ctx->rxq->dev->ifindex;
6305 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6306 ifindex, IPPROTO_TCP, netns_id,
6310 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6311 .func = bpf_xdp_sk_lookup_tcp,
6314 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6315 .arg1_type = ARG_PTR_TO_CTX,
6316 .arg2_type = ARG_PTR_TO_MEM,
6317 .arg3_type = ARG_CONST_SIZE,
6318 .arg4_type = ARG_ANYTHING,
6319 .arg5_type = ARG_ANYTHING,
6322 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6323 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6325 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6326 sock_net(ctx->sk), 0,
6327 IPPROTO_TCP, netns_id, flags);
6330 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6331 .func = bpf_sock_addr_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_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6342 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6344 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6345 sock_net(ctx->sk), 0, IPPROTO_TCP,
6349 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6350 .func = bpf_sock_addr_sk_lookup_tcp,
6352 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6353 .arg1_type = ARG_PTR_TO_CTX,
6354 .arg2_type = ARG_PTR_TO_MEM,
6355 .arg3_type = ARG_CONST_SIZE,
6356 .arg4_type = ARG_ANYTHING,
6357 .arg5_type = ARG_ANYTHING,
6360 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6361 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6363 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6364 sock_net(ctx->sk), 0, IPPROTO_UDP,
6368 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6369 .func = bpf_sock_addr_sk_lookup_udp,
6371 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6372 .arg1_type = ARG_PTR_TO_CTX,
6373 .arg2_type = ARG_PTR_TO_MEM,
6374 .arg3_type = ARG_CONST_SIZE,
6375 .arg4_type = ARG_ANYTHING,
6376 .arg5_type = ARG_ANYTHING,
6379 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6380 struct bpf_insn_access_aux *info)
6382 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6386 if (off % size != 0)
6390 case offsetof(struct bpf_tcp_sock, bytes_received):
6391 case offsetof(struct bpf_tcp_sock, bytes_acked):
6392 return size == sizeof(__u64);
6394 return size == sizeof(__u32);
6398 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6399 const struct bpf_insn *si,
6400 struct bpf_insn *insn_buf,
6401 struct bpf_prog *prog, u32 *target_size)
6403 struct bpf_insn *insn = insn_buf;
6405 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \
6407 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \
6408 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6409 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6410 si->dst_reg, si->src_reg, \
6411 offsetof(struct tcp_sock, FIELD)); \
6414 #define BPF_INET_SOCK_GET_COMMON(FIELD) \
6416 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \
6418 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6419 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6420 struct inet_connection_sock, \
6422 si->dst_reg, si->src_reg, \
6424 struct inet_connection_sock, \
6428 if (insn > insn_buf)
6429 return insn - insn_buf;
6432 case offsetof(struct bpf_tcp_sock, rtt_min):
6433 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
6434 sizeof(struct minmax));
6435 BUILD_BUG_ON(sizeof(struct minmax) <
6436 sizeof(struct minmax_sample));
6438 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6439 offsetof(struct tcp_sock, rtt_min) +
6440 offsetof(struct minmax_sample, v));
6442 case offsetof(struct bpf_tcp_sock, snd_cwnd):
6443 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
6445 case offsetof(struct bpf_tcp_sock, srtt_us):
6446 BPF_TCP_SOCK_GET_COMMON(srtt_us);
6448 case offsetof(struct bpf_tcp_sock, snd_ssthresh):
6449 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
6451 case offsetof(struct bpf_tcp_sock, rcv_nxt):
6452 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
6454 case offsetof(struct bpf_tcp_sock, snd_nxt):
6455 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
6457 case offsetof(struct bpf_tcp_sock, snd_una):
6458 BPF_TCP_SOCK_GET_COMMON(snd_una);
6460 case offsetof(struct bpf_tcp_sock, mss_cache):
6461 BPF_TCP_SOCK_GET_COMMON(mss_cache);
6463 case offsetof(struct bpf_tcp_sock, ecn_flags):
6464 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
6466 case offsetof(struct bpf_tcp_sock, rate_delivered):
6467 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
6469 case offsetof(struct bpf_tcp_sock, rate_interval_us):
6470 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
6472 case offsetof(struct bpf_tcp_sock, packets_out):
6473 BPF_TCP_SOCK_GET_COMMON(packets_out);
6475 case offsetof(struct bpf_tcp_sock, retrans_out):
6476 BPF_TCP_SOCK_GET_COMMON(retrans_out);
6478 case offsetof(struct bpf_tcp_sock, total_retrans):
6479 BPF_TCP_SOCK_GET_COMMON(total_retrans);
6481 case offsetof(struct bpf_tcp_sock, segs_in):
6482 BPF_TCP_SOCK_GET_COMMON(segs_in);
6484 case offsetof(struct bpf_tcp_sock, data_segs_in):
6485 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
6487 case offsetof(struct bpf_tcp_sock, segs_out):
6488 BPF_TCP_SOCK_GET_COMMON(segs_out);
6490 case offsetof(struct bpf_tcp_sock, data_segs_out):
6491 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
6493 case offsetof(struct bpf_tcp_sock, lost_out):
6494 BPF_TCP_SOCK_GET_COMMON(lost_out);
6496 case offsetof(struct bpf_tcp_sock, sacked_out):
6497 BPF_TCP_SOCK_GET_COMMON(sacked_out);
6499 case offsetof(struct bpf_tcp_sock, bytes_received):
6500 BPF_TCP_SOCK_GET_COMMON(bytes_received);
6502 case offsetof(struct bpf_tcp_sock, bytes_acked):
6503 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
6505 case offsetof(struct bpf_tcp_sock, dsack_dups):
6506 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
6508 case offsetof(struct bpf_tcp_sock, delivered):
6509 BPF_TCP_SOCK_GET_COMMON(delivered);
6511 case offsetof(struct bpf_tcp_sock, delivered_ce):
6512 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
6514 case offsetof(struct bpf_tcp_sock, icsk_retransmits):
6515 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
6519 return insn - insn_buf;
6522 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
6524 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
6525 return (unsigned long)sk;
6527 return (unsigned long)NULL;
6530 const struct bpf_func_proto bpf_tcp_sock_proto = {
6531 .func = bpf_tcp_sock,
6533 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
6534 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6537 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
6539 sk = sk_to_full_sk(sk);
6541 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
6542 return (unsigned long)sk;
6544 return (unsigned long)NULL;
6547 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
6548 .func = bpf_get_listener_sock,
6550 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6551 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6554 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
6556 unsigned int iphdr_len;
6558 switch (skb_protocol(skb, true)) {
6559 case cpu_to_be16(ETH_P_IP):
6560 iphdr_len = sizeof(struct iphdr);
6562 case cpu_to_be16(ETH_P_IPV6):
6563 iphdr_len = sizeof(struct ipv6hdr);
6569 if (skb_headlen(skb) < iphdr_len)
6572 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
6575 return INET_ECN_set_ce(skb);
6578 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6579 struct bpf_insn_access_aux *info)
6581 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
6584 if (off % size != 0)
6589 return size == sizeof(__u32);
6593 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
6594 const struct bpf_insn *si,
6595 struct bpf_insn *insn_buf,
6596 struct bpf_prog *prog, u32 *target_size)
6598 struct bpf_insn *insn = insn_buf;
6600 #define BPF_XDP_SOCK_GET(FIELD) \
6602 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \
6603 sizeof_field(struct bpf_xdp_sock, FIELD)); \
6604 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
6605 si->dst_reg, si->src_reg, \
6606 offsetof(struct xdp_sock, FIELD)); \
6610 case offsetof(struct bpf_xdp_sock, queue_id):
6611 BPF_XDP_SOCK_GET(queue_id);
6615 return insn - insn_buf;
6618 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
6619 .func = bpf_skb_ecn_set_ce,
6621 .ret_type = RET_INTEGER,
6622 .arg1_type = ARG_PTR_TO_CTX,
6625 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6626 struct tcphdr *, th, u32, th_len)
6628 #ifdef CONFIG_SYN_COOKIES
6632 if (unlikely(!sk || th_len < sizeof(*th)))
6635 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
6636 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6639 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
6642 if (!th->ack || th->rst || th->syn)
6645 if (tcp_synq_no_recent_overflow(sk))
6648 cookie = ntohl(th->ack_seq) - 1;
6650 switch (sk->sk_family) {
6652 if (unlikely(iph_len < sizeof(struct iphdr)))
6655 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
6658 #if IS_BUILTIN(CONFIG_IPV6)
6660 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6663 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
6665 #endif /* CONFIG_IPV6 */
6668 return -EPROTONOSUPPORT;
6680 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
6681 .func = bpf_tcp_check_syncookie,
6684 .ret_type = RET_INTEGER,
6685 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6686 .arg2_type = ARG_PTR_TO_MEM,
6687 .arg3_type = ARG_CONST_SIZE,
6688 .arg4_type = ARG_PTR_TO_MEM,
6689 .arg5_type = ARG_CONST_SIZE,
6692 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6693 struct tcphdr *, th, u32, th_len)
6695 #ifdef CONFIG_SYN_COOKIES
6699 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
6702 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6705 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
6708 if (!th->syn || th->ack || th->fin || th->rst)
6711 if (unlikely(iph_len < sizeof(struct iphdr)))
6714 /* Both struct iphdr and struct ipv6hdr have the version field at the
6715 * same offset so we can cast to the shorter header (struct iphdr).
6717 switch (((struct iphdr *)iph)->version) {
6719 if (sk->sk_family == AF_INET6 && sk->sk_ipv6only)
6722 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
6725 #if IS_BUILTIN(CONFIG_IPV6)
6727 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6730 if (sk->sk_family != AF_INET6)
6733 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
6735 #endif /* CONFIG_IPV6 */
6738 return -EPROTONOSUPPORT;
6743 return cookie | ((u64)mss << 32);
6746 #endif /* CONFIG_SYN_COOKIES */
6749 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
6750 .func = bpf_tcp_gen_syncookie,
6751 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */
6753 .ret_type = RET_INTEGER,
6754 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6755 .arg2_type = ARG_PTR_TO_MEM,
6756 .arg3_type = ARG_CONST_SIZE,
6757 .arg4_type = ARG_PTR_TO_MEM,
6758 .arg5_type = ARG_CONST_SIZE,
6761 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
6763 if (!sk || flags != 0)
6765 if (!skb_at_tc_ingress(skb))
6767 if (unlikely(dev_net(skb->dev) != sock_net(sk)))
6768 return -ENETUNREACH;
6769 if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
6770 return -ESOCKTNOSUPPORT;
6771 if (sk_is_refcounted(sk) &&
6772 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
6777 skb->destructor = sock_pfree;
6782 static const struct bpf_func_proto bpf_sk_assign_proto = {
6783 .func = bpf_sk_assign,
6785 .ret_type = RET_INTEGER,
6786 .arg1_type = ARG_PTR_TO_CTX,
6787 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6788 .arg3_type = ARG_ANYTHING,
6791 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
6792 u8 search_kind, const u8 *magic,
6793 u8 magic_len, bool *eol)
6799 while (op < opend) {
6802 if (kind == TCPOPT_EOL) {
6804 return ERR_PTR(-ENOMSG);
6805 } else if (kind == TCPOPT_NOP) {
6810 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
6811 /* Something is wrong in the received header.
6812 * Follow the TCP stack's tcp_parse_options()
6813 * and just bail here.
6815 return ERR_PTR(-EFAULT);
6818 if (search_kind == kind) {
6822 if (magic_len > kind_len - 2)
6823 return ERR_PTR(-ENOMSG);
6825 if (!memcmp(&op[2], magic, magic_len))
6832 return ERR_PTR(-ENOMSG);
6835 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6836 void *, search_res, u32, len, u64, flags)
6838 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
6839 const u8 *op, *opend, *magic, *search = search_res;
6840 u8 search_kind, search_len, copy_len, magic_len;
6843 /* 2 byte is the minimal option len except TCPOPT_NOP and
6844 * TCPOPT_EOL which are useless for the bpf prog to learn
6845 * and this helper disallow loading them also.
6847 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
6850 search_kind = search[0];
6851 search_len = search[1];
6853 if (search_len > len || search_kind == TCPOPT_NOP ||
6854 search_kind == TCPOPT_EOL)
6857 if (search_kind == TCPOPT_EXP || search_kind == 253) {
6858 /* 16 or 32 bit magic. +2 for kind and kind length */
6859 if (search_len != 4 && search_len != 6)
6862 magic_len = search_len - 2;
6871 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
6876 op += sizeof(struct tcphdr);
6878 if (!bpf_sock->skb ||
6879 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6880 /* This bpf_sock->op cannot call this helper */
6883 opend = bpf_sock->skb_data_end;
6884 op = bpf_sock->skb->data + sizeof(struct tcphdr);
6887 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
6894 if (copy_len > len) {
6899 memcpy(search_res, op, copy_len);
6903 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
6904 .func = bpf_sock_ops_load_hdr_opt,
6906 .ret_type = RET_INTEGER,
6907 .arg1_type = ARG_PTR_TO_CTX,
6908 .arg2_type = ARG_PTR_TO_MEM,
6909 .arg3_type = ARG_CONST_SIZE,
6910 .arg4_type = ARG_ANYTHING,
6913 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6914 const void *, from, u32, len, u64, flags)
6916 u8 new_kind, new_kind_len, magic_len = 0, *opend;
6917 const u8 *op, *new_op, *magic = NULL;
6918 struct sk_buff *skb;
6921 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
6924 if (len < 2 || flags)
6928 new_kind = new_op[0];
6929 new_kind_len = new_op[1];
6931 if (new_kind_len > len || new_kind == TCPOPT_NOP ||
6932 new_kind == TCPOPT_EOL)
6935 if (new_kind_len > bpf_sock->remaining_opt_len)
6938 /* 253 is another experimental kind */
6939 if (new_kind == TCPOPT_EXP || new_kind == 253) {
6940 if (new_kind_len < 4)
6942 /* Match for the 2 byte magic also.
6943 * RFC 6994: the magic could be 2 or 4 bytes.
6944 * Hence, matching by 2 byte only is on the
6945 * conservative side but it is the right
6946 * thing to do for the 'search-for-duplication'
6953 /* Check for duplication */
6954 skb = bpf_sock->skb;
6955 op = skb->data + sizeof(struct tcphdr);
6956 opend = bpf_sock->skb_data_end;
6958 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
6963 if (PTR_ERR(op) != -ENOMSG)
6967 /* The option has been ended. Treat it as no more
6968 * header option can be written.
6972 /* No duplication found. Store the header option. */
6973 memcpy(opend, from, new_kind_len);
6975 bpf_sock->remaining_opt_len -= new_kind_len;
6976 bpf_sock->skb_data_end += new_kind_len;
6981 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
6982 .func = bpf_sock_ops_store_hdr_opt,
6984 .ret_type = RET_INTEGER,
6985 .arg1_type = ARG_PTR_TO_CTX,
6986 .arg2_type = ARG_PTR_TO_MEM,
6987 .arg3_type = ARG_CONST_SIZE,
6988 .arg4_type = ARG_ANYTHING,
6991 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6992 u32, len, u64, flags)
6994 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6997 if (flags || len < 2)
7000 if (len > bpf_sock->remaining_opt_len)
7003 bpf_sock->remaining_opt_len -= len;
7008 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7009 .func = bpf_sock_ops_reserve_hdr_opt,
7011 .ret_type = RET_INTEGER,
7012 .arg1_type = ARG_PTR_TO_CTX,
7013 .arg2_type = ARG_ANYTHING,
7014 .arg3_type = ARG_ANYTHING,
7017 #endif /* CONFIG_INET */
7019 bool bpf_helper_changes_pkt_data(void *func)
7021 if (func == bpf_skb_vlan_push ||
7022 func == bpf_skb_vlan_pop ||
7023 func == bpf_skb_store_bytes ||
7024 func == bpf_skb_change_proto ||
7025 func == bpf_skb_change_head ||
7026 func == sk_skb_change_head ||
7027 func == bpf_skb_change_tail ||
7028 func == sk_skb_change_tail ||
7029 func == bpf_skb_adjust_room ||
7030 func == sk_skb_adjust_room ||
7031 func == bpf_skb_pull_data ||
7032 func == sk_skb_pull_data ||
7033 func == bpf_clone_redirect ||
7034 func == bpf_l3_csum_replace ||
7035 func == bpf_l4_csum_replace ||
7036 func == bpf_xdp_adjust_head ||
7037 func == bpf_xdp_adjust_meta ||
7038 func == bpf_msg_pull_data ||
7039 func == bpf_msg_push_data ||
7040 func == bpf_msg_pop_data ||
7041 func == bpf_xdp_adjust_tail ||
7042 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7043 func == bpf_lwt_seg6_store_bytes ||
7044 func == bpf_lwt_seg6_adjust_srh ||
7045 func == bpf_lwt_seg6_action ||
7048 func == bpf_sock_ops_store_hdr_opt ||
7050 func == bpf_lwt_in_push_encap ||
7051 func == bpf_lwt_xmit_push_encap)
7057 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7058 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7060 static const struct bpf_func_proto *
7061 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7064 /* inet and inet6 sockets are created in a process
7065 * context so there is always a valid uid/gid
7067 case BPF_FUNC_get_current_uid_gid:
7068 return &bpf_get_current_uid_gid_proto;
7069 case BPF_FUNC_get_local_storage:
7070 return &bpf_get_local_storage_proto;
7071 case BPF_FUNC_get_socket_cookie:
7072 return &bpf_get_socket_cookie_sock_proto;
7073 case BPF_FUNC_get_netns_cookie:
7074 return &bpf_get_netns_cookie_sock_proto;
7075 case BPF_FUNC_perf_event_output:
7076 return &bpf_event_output_data_proto;
7077 case BPF_FUNC_get_current_pid_tgid:
7078 return &bpf_get_current_pid_tgid_proto;
7079 case BPF_FUNC_get_current_comm:
7080 return &bpf_get_current_comm_proto;
7081 #ifdef CONFIG_CGROUPS
7082 case BPF_FUNC_get_current_cgroup_id:
7083 return &bpf_get_current_cgroup_id_proto;
7084 case BPF_FUNC_get_current_ancestor_cgroup_id:
7085 return &bpf_get_current_ancestor_cgroup_id_proto;
7087 #ifdef CONFIG_CGROUP_NET_CLASSID
7088 case BPF_FUNC_get_cgroup_classid:
7089 return &bpf_get_cgroup_classid_curr_proto;
7091 case BPF_FUNC_sk_storage_get:
7092 return &bpf_sk_storage_get_cg_sock_proto;
7094 return bpf_base_func_proto(func_id);
7098 static const struct bpf_func_proto *
7099 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7102 /* inet and inet6 sockets are created in a process
7103 * context so there is always a valid uid/gid
7105 case BPF_FUNC_get_current_uid_gid:
7106 return &bpf_get_current_uid_gid_proto;
7108 switch (prog->expected_attach_type) {
7109 case BPF_CGROUP_INET4_CONNECT:
7110 case BPF_CGROUP_INET6_CONNECT:
7111 return &bpf_bind_proto;
7115 case BPF_FUNC_get_socket_cookie:
7116 return &bpf_get_socket_cookie_sock_addr_proto;
7117 case BPF_FUNC_get_netns_cookie:
7118 return &bpf_get_netns_cookie_sock_addr_proto;
7119 case BPF_FUNC_get_local_storage:
7120 return &bpf_get_local_storage_proto;
7121 case BPF_FUNC_perf_event_output:
7122 return &bpf_event_output_data_proto;
7123 case BPF_FUNC_get_current_pid_tgid:
7124 return &bpf_get_current_pid_tgid_proto;
7125 case BPF_FUNC_get_current_comm:
7126 return &bpf_get_current_comm_proto;
7127 #ifdef CONFIG_CGROUPS
7128 case BPF_FUNC_get_current_cgroup_id:
7129 return &bpf_get_current_cgroup_id_proto;
7130 case BPF_FUNC_get_current_ancestor_cgroup_id:
7131 return &bpf_get_current_ancestor_cgroup_id_proto;
7133 #ifdef CONFIG_CGROUP_NET_CLASSID
7134 case BPF_FUNC_get_cgroup_classid:
7135 return &bpf_get_cgroup_classid_curr_proto;
7138 case BPF_FUNC_sk_lookup_tcp:
7139 return &bpf_sock_addr_sk_lookup_tcp_proto;
7140 case BPF_FUNC_sk_lookup_udp:
7141 return &bpf_sock_addr_sk_lookup_udp_proto;
7142 case BPF_FUNC_sk_release:
7143 return &bpf_sk_release_proto;
7144 case BPF_FUNC_skc_lookup_tcp:
7145 return &bpf_sock_addr_skc_lookup_tcp_proto;
7146 #endif /* CONFIG_INET */
7147 case BPF_FUNC_sk_storage_get:
7148 return &bpf_sk_storage_get_proto;
7149 case BPF_FUNC_sk_storage_delete:
7150 return &bpf_sk_storage_delete_proto;
7151 case BPF_FUNC_setsockopt:
7152 switch (prog->expected_attach_type) {
7153 case BPF_CGROUP_INET4_BIND:
7154 case BPF_CGROUP_INET6_BIND:
7155 case BPF_CGROUP_INET4_CONNECT:
7156 case BPF_CGROUP_INET6_CONNECT:
7157 case BPF_CGROUP_UDP4_RECVMSG:
7158 case BPF_CGROUP_UDP6_RECVMSG:
7159 case BPF_CGROUP_UDP4_SENDMSG:
7160 case BPF_CGROUP_UDP6_SENDMSG:
7161 case BPF_CGROUP_INET4_GETPEERNAME:
7162 case BPF_CGROUP_INET6_GETPEERNAME:
7163 case BPF_CGROUP_INET4_GETSOCKNAME:
7164 case BPF_CGROUP_INET6_GETSOCKNAME:
7165 return &bpf_sock_addr_setsockopt_proto;
7169 case BPF_FUNC_getsockopt:
7170 switch (prog->expected_attach_type) {
7171 case BPF_CGROUP_INET4_BIND:
7172 case BPF_CGROUP_INET6_BIND:
7173 case BPF_CGROUP_INET4_CONNECT:
7174 case BPF_CGROUP_INET6_CONNECT:
7175 case BPF_CGROUP_UDP4_RECVMSG:
7176 case BPF_CGROUP_UDP6_RECVMSG:
7177 case BPF_CGROUP_UDP4_SENDMSG:
7178 case BPF_CGROUP_UDP6_SENDMSG:
7179 case BPF_CGROUP_INET4_GETPEERNAME:
7180 case BPF_CGROUP_INET6_GETPEERNAME:
7181 case BPF_CGROUP_INET4_GETSOCKNAME:
7182 case BPF_CGROUP_INET6_GETSOCKNAME:
7183 return &bpf_sock_addr_getsockopt_proto;
7188 return bpf_sk_base_func_proto(func_id);
7192 static const struct bpf_func_proto *
7193 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7196 case BPF_FUNC_skb_load_bytes:
7197 return &bpf_skb_load_bytes_proto;
7198 case BPF_FUNC_skb_load_bytes_relative:
7199 return &bpf_skb_load_bytes_relative_proto;
7200 case BPF_FUNC_get_socket_cookie:
7201 return &bpf_get_socket_cookie_proto;
7202 case BPF_FUNC_get_socket_uid:
7203 return &bpf_get_socket_uid_proto;
7204 case BPF_FUNC_perf_event_output:
7205 return &bpf_skb_event_output_proto;
7207 return bpf_sk_base_func_proto(func_id);
7211 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7212 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7214 static const struct bpf_func_proto *
7215 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7218 case BPF_FUNC_get_local_storage:
7219 return &bpf_get_local_storage_proto;
7220 case BPF_FUNC_sk_fullsock:
7221 return &bpf_sk_fullsock_proto;
7222 case BPF_FUNC_sk_storage_get:
7223 return &bpf_sk_storage_get_proto;
7224 case BPF_FUNC_sk_storage_delete:
7225 return &bpf_sk_storage_delete_proto;
7226 case BPF_FUNC_perf_event_output:
7227 return &bpf_skb_event_output_proto;
7228 #ifdef CONFIG_SOCK_CGROUP_DATA
7229 case BPF_FUNC_skb_cgroup_id:
7230 return &bpf_skb_cgroup_id_proto;
7231 case BPF_FUNC_skb_ancestor_cgroup_id:
7232 return &bpf_skb_ancestor_cgroup_id_proto;
7233 case BPF_FUNC_sk_cgroup_id:
7234 return &bpf_sk_cgroup_id_proto;
7235 case BPF_FUNC_sk_ancestor_cgroup_id:
7236 return &bpf_sk_ancestor_cgroup_id_proto;
7239 case BPF_FUNC_sk_lookup_tcp:
7240 return &bpf_sk_lookup_tcp_proto;
7241 case BPF_FUNC_sk_lookup_udp:
7242 return &bpf_sk_lookup_udp_proto;
7243 case BPF_FUNC_sk_release:
7244 return &bpf_sk_release_proto;
7245 case BPF_FUNC_skc_lookup_tcp:
7246 return &bpf_skc_lookup_tcp_proto;
7247 case BPF_FUNC_tcp_sock:
7248 return &bpf_tcp_sock_proto;
7249 case BPF_FUNC_get_listener_sock:
7250 return &bpf_get_listener_sock_proto;
7251 case BPF_FUNC_skb_ecn_set_ce:
7252 return &bpf_skb_ecn_set_ce_proto;
7255 return sk_filter_func_proto(func_id, prog);
7259 static const struct bpf_func_proto *
7260 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7263 case BPF_FUNC_skb_store_bytes:
7264 return &bpf_skb_store_bytes_proto;
7265 case BPF_FUNC_skb_load_bytes:
7266 return &bpf_skb_load_bytes_proto;
7267 case BPF_FUNC_skb_load_bytes_relative:
7268 return &bpf_skb_load_bytes_relative_proto;
7269 case BPF_FUNC_skb_pull_data:
7270 return &bpf_skb_pull_data_proto;
7271 case BPF_FUNC_csum_diff:
7272 return &bpf_csum_diff_proto;
7273 case BPF_FUNC_csum_update:
7274 return &bpf_csum_update_proto;
7275 case BPF_FUNC_csum_level:
7276 return &bpf_csum_level_proto;
7277 case BPF_FUNC_l3_csum_replace:
7278 return &bpf_l3_csum_replace_proto;
7279 case BPF_FUNC_l4_csum_replace:
7280 return &bpf_l4_csum_replace_proto;
7281 case BPF_FUNC_clone_redirect:
7282 return &bpf_clone_redirect_proto;
7283 case BPF_FUNC_get_cgroup_classid:
7284 return &bpf_get_cgroup_classid_proto;
7285 case BPF_FUNC_skb_vlan_push:
7286 return &bpf_skb_vlan_push_proto;
7287 case BPF_FUNC_skb_vlan_pop:
7288 return &bpf_skb_vlan_pop_proto;
7289 case BPF_FUNC_skb_change_proto:
7290 return &bpf_skb_change_proto_proto;
7291 case BPF_FUNC_skb_change_type:
7292 return &bpf_skb_change_type_proto;
7293 case BPF_FUNC_skb_adjust_room:
7294 return &bpf_skb_adjust_room_proto;
7295 case BPF_FUNC_skb_change_tail:
7296 return &bpf_skb_change_tail_proto;
7297 case BPF_FUNC_skb_change_head:
7298 return &bpf_skb_change_head_proto;
7299 case BPF_FUNC_skb_get_tunnel_key:
7300 return &bpf_skb_get_tunnel_key_proto;
7301 case BPF_FUNC_skb_set_tunnel_key:
7302 return bpf_get_skb_set_tunnel_proto(func_id);
7303 case BPF_FUNC_skb_get_tunnel_opt:
7304 return &bpf_skb_get_tunnel_opt_proto;
7305 case BPF_FUNC_skb_set_tunnel_opt:
7306 return bpf_get_skb_set_tunnel_proto(func_id);
7307 case BPF_FUNC_redirect:
7308 return &bpf_redirect_proto;
7309 case BPF_FUNC_redirect_neigh:
7310 return &bpf_redirect_neigh_proto;
7311 case BPF_FUNC_redirect_peer:
7312 return &bpf_redirect_peer_proto;
7313 case BPF_FUNC_get_route_realm:
7314 return &bpf_get_route_realm_proto;
7315 case BPF_FUNC_get_hash_recalc:
7316 return &bpf_get_hash_recalc_proto;
7317 case BPF_FUNC_set_hash_invalid:
7318 return &bpf_set_hash_invalid_proto;
7319 case BPF_FUNC_set_hash:
7320 return &bpf_set_hash_proto;
7321 case BPF_FUNC_perf_event_output:
7322 return &bpf_skb_event_output_proto;
7323 case BPF_FUNC_get_smp_processor_id:
7324 return &bpf_get_smp_processor_id_proto;
7325 case BPF_FUNC_skb_under_cgroup:
7326 return &bpf_skb_under_cgroup_proto;
7327 case BPF_FUNC_get_socket_cookie:
7328 return &bpf_get_socket_cookie_proto;
7329 case BPF_FUNC_get_socket_uid:
7330 return &bpf_get_socket_uid_proto;
7331 case BPF_FUNC_fib_lookup:
7332 return &bpf_skb_fib_lookup_proto;
7333 case BPF_FUNC_check_mtu:
7334 return &bpf_skb_check_mtu_proto;
7335 case BPF_FUNC_sk_fullsock:
7336 return &bpf_sk_fullsock_proto;
7337 case BPF_FUNC_sk_storage_get:
7338 return &bpf_sk_storage_get_proto;
7339 case BPF_FUNC_sk_storage_delete:
7340 return &bpf_sk_storage_delete_proto;
7342 case BPF_FUNC_skb_get_xfrm_state:
7343 return &bpf_skb_get_xfrm_state_proto;
7345 #ifdef CONFIG_CGROUP_NET_CLASSID
7346 case BPF_FUNC_skb_cgroup_classid:
7347 return &bpf_skb_cgroup_classid_proto;
7349 #ifdef CONFIG_SOCK_CGROUP_DATA
7350 case BPF_FUNC_skb_cgroup_id:
7351 return &bpf_skb_cgroup_id_proto;
7352 case BPF_FUNC_skb_ancestor_cgroup_id:
7353 return &bpf_skb_ancestor_cgroup_id_proto;
7356 case BPF_FUNC_sk_lookup_tcp:
7357 return &bpf_sk_lookup_tcp_proto;
7358 case BPF_FUNC_sk_lookup_udp:
7359 return &bpf_sk_lookup_udp_proto;
7360 case BPF_FUNC_sk_release:
7361 return &bpf_sk_release_proto;
7362 case BPF_FUNC_tcp_sock:
7363 return &bpf_tcp_sock_proto;
7364 case BPF_FUNC_get_listener_sock:
7365 return &bpf_get_listener_sock_proto;
7366 case BPF_FUNC_skc_lookup_tcp:
7367 return &bpf_skc_lookup_tcp_proto;
7368 case BPF_FUNC_tcp_check_syncookie:
7369 return &bpf_tcp_check_syncookie_proto;
7370 case BPF_FUNC_skb_ecn_set_ce:
7371 return &bpf_skb_ecn_set_ce_proto;
7372 case BPF_FUNC_tcp_gen_syncookie:
7373 return &bpf_tcp_gen_syncookie_proto;
7374 case BPF_FUNC_sk_assign:
7375 return &bpf_sk_assign_proto;
7378 return bpf_sk_base_func_proto(func_id);
7382 static const struct bpf_func_proto *
7383 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7386 case BPF_FUNC_perf_event_output:
7387 return &bpf_xdp_event_output_proto;
7388 case BPF_FUNC_get_smp_processor_id:
7389 return &bpf_get_smp_processor_id_proto;
7390 case BPF_FUNC_csum_diff:
7391 return &bpf_csum_diff_proto;
7392 case BPF_FUNC_xdp_adjust_head:
7393 return &bpf_xdp_adjust_head_proto;
7394 case BPF_FUNC_xdp_adjust_meta:
7395 return &bpf_xdp_adjust_meta_proto;
7396 case BPF_FUNC_redirect:
7397 return &bpf_xdp_redirect_proto;
7398 case BPF_FUNC_redirect_map:
7399 return &bpf_xdp_redirect_map_proto;
7400 case BPF_FUNC_xdp_adjust_tail:
7401 return &bpf_xdp_adjust_tail_proto;
7402 case BPF_FUNC_fib_lookup:
7403 return &bpf_xdp_fib_lookup_proto;
7404 case BPF_FUNC_check_mtu:
7405 return &bpf_xdp_check_mtu_proto;
7407 case BPF_FUNC_sk_lookup_udp:
7408 return &bpf_xdp_sk_lookup_udp_proto;
7409 case BPF_FUNC_sk_lookup_tcp:
7410 return &bpf_xdp_sk_lookup_tcp_proto;
7411 case BPF_FUNC_sk_release:
7412 return &bpf_sk_release_proto;
7413 case BPF_FUNC_skc_lookup_tcp:
7414 return &bpf_xdp_skc_lookup_tcp_proto;
7415 case BPF_FUNC_tcp_check_syncookie:
7416 return &bpf_tcp_check_syncookie_proto;
7417 case BPF_FUNC_tcp_gen_syncookie:
7418 return &bpf_tcp_gen_syncookie_proto;
7421 return bpf_sk_base_func_proto(func_id);
7425 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
7426 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
7428 static const struct bpf_func_proto *
7429 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7432 case BPF_FUNC_setsockopt:
7433 return &bpf_sock_ops_setsockopt_proto;
7434 case BPF_FUNC_getsockopt:
7435 return &bpf_sock_ops_getsockopt_proto;
7436 case BPF_FUNC_sock_ops_cb_flags_set:
7437 return &bpf_sock_ops_cb_flags_set_proto;
7438 case BPF_FUNC_sock_map_update:
7439 return &bpf_sock_map_update_proto;
7440 case BPF_FUNC_sock_hash_update:
7441 return &bpf_sock_hash_update_proto;
7442 case BPF_FUNC_get_socket_cookie:
7443 return &bpf_get_socket_cookie_sock_ops_proto;
7444 case BPF_FUNC_get_local_storage:
7445 return &bpf_get_local_storage_proto;
7446 case BPF_FUNC_perf_event_output:
7447 return &bpf_event_output_data_proto;
7448 case BPF_FUNC_sk_storage_get:
7449 return &bpf_sk_storage_get_proto;
7450 case BPF_FUNC_sk_storage_delete:
7451 return &bpf_sk_storage_delete_proto;
7453 case BPF_FUNC_load_hdr_opt:
7454 return &bpf_sock_ops_load_hdr_opt_proto;
7455 case BPF_FUNC_store_hdr_opt:
7456 return &bpf_sock_ops_store_hdr_opt_proto;
7457 case BPF_FUNC_reserve_hdr_opt:
7458 return &bpf_sock_ops_reserve_hdr_opt_proto;
7459 case BPF_FUNC_tcp_sock:
7460 return &bpf_tcp_sock_proto;
7461 #endif /* CONFIG_INET */
7463 return bpf_sk_base_func_proto(func_id);
7467 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
7468 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
7470 static const struct bpf_func_proto *
7471 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7474 case BPF_FUNC_msg_redirect_map:
7475 return &bpf_msg_redirect_map_proto;
7476 case BPF_FUNC_msg_redirect_hash:
7477 return &bpf_msg_redirect_hash_proto;
7478 case BPF_FUNC_msg_apply_bytes:
7479 return &bpf_msg_apply_bytes_proto;
7480 case BPF_FUNC_msg_cork_bytes:
7481 return &bpf_msg_cork_bytes_proto;
7482 case BPF_FUNC_msg_pull_data:
7483 return &bpf_msg_pull_data_proto;
7484 case BPF_FUNC_msg_push_data:
7485 return &bpf_msg_push_data_proto;
7486 case BPF_FUNC_msg_pop_data:
7487 return &bpf_msg_pop_data_proto;
7488 case BPF_FUNC_perf_event_output:
7489 return &bpf_event_output_data_proto;
7490 case BPF_FUNC_get_current_uid_gid:
7491 return &bpf_get_current_uid_gid_proto;
7492 case BPF_FUNC_get_current_pid_tgid:
7493 return &bpf_get_current_pid_tgid_proto;
7494 case BPF_FUNC_sk_storage_get:
7495 return &bpf_sk_storage_get_proto;
7496 case BPF_FUNC_sk_storage_delete:
7497 return &bpf_sk_storage_delete_proto;
7498 #ifdef CONFIG_CGROUPS
7499 case BPF_FUNC_get_current_cgroup_id:
7500 return &bpf_get_current_cgroup_id_proto;
7501 case BPF_FUNC_get_current_ancestor_cgroup_id:
7502 return &bpf_get_current_ancestor_cgroup_id_proto;
7504 #ifdef CONFIG_CGROUP_NET_CLASSID
7505 case BPF_FUNC_get_cgroup_classid:
7506 return &bpf_get_cgroup_classid_curr_proto;
7509 return bpf_sk_base_func_proto(func_id);
7513 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
7514 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
7516 static const struct bpf_func_proto *
7517 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7520 case BPF_FUNC_skb_store_bytes:
7521 return &bpf_skb_store_bytes_proto;
7522 case BPF_FUNC_skb_load_bytes:
7523 return &bpf_skb_load_bytes_proto;
7524 case BPF_FUNC_skb_pull_data:
7525 return &sk_skb_pull_data_proto;
7526 case BPF_FUNC_skb_change_tail:
7527 return &sk_skb_change_tail_proto;
7528 case BPF_FUNC_skb_change_head:
7529 return &sk_skb_change_head_proto;
7530 case BPF_FUNC_skb_adjust_room:
7531 return &sk_skb_adjust_room_proto;
7532 case BPF_FUNC_get_socket_cookie:
7533 return &bpf_get_socket_cookie_proto;
7534 case BPF_FUNC_get_socket_uid:
7535 return &bpf_get_socket_uid_proto;
7536 case BPF_FUNC_sk_redirect_map:
7537 return &bpf_sk_redirect_map_proto;
7538 case BPF_FUNC_sk_redirect_hash:
7539 return &bpf_sk_redirect_hash_proto;
7540 case BPF_FUNC_perf_event_output:
7541 return &bpf_skb_event_output_proto;
7543 case BPF_FUNC_sk_lookup_tcp:
7544 return &bpf_sk_lookup_tcp_proto;
7545 case BPF_FUNC_sk_lookup_udp:
7546 return &bpf_sk_lookup_udp_proto;
7547 case BPF_FUNC_sk_release:
7548 return &bpf_sk_release_proto;
7549 case BPF_FUNC_skc_lookup_tcp:
7550 return &bpf_skc_lookup_tcp_proto;
7553 return bpf_sk_base_func_proto(func_id);
7557 static const struct bpf_func_proto *
7558 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7561 case BPF_FUNC_skb_load_bytes:
7562 return &bpf_flow_dissector_load_bytes_proto;
7564 return bpf_sk_base_func_proto(func_id);
7568 static const struct bpf_func_proto *
7569 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7572 case BPF_FUNC_skb_load_bytes:
7573 return &bpf_skb_load_bytes_proto;
7574 case BPF_FUNC_skb_pull_data:
7575 return &bpf_skb_pull_data_proto;
7576 case BPF_FUNC_csum_diff:
7577 return &bpf_csum_diff_proto;
7578 case BPF_FUNC_get_cgroup_classid:
7579 return &bpf_get_cgroup_classid_proto;
7580 case BPF_FUNC_get_route_realm:
7581 return &bpf_get_route_realm_proto;
7582 case BPF_FUNC_get_hash_recalc:
7583 return &bpf_get_hash_recalc_proto;
7584 case BPF_FUNC_perf_event_output:
7585 return &bpf_skb_event_output_proto;
7586 case BPF_FUNC_get_smp_processor_id:
7587 return &bpf_get_smp_processor_id_proto;
7588 case BPF_FUNC_skb_under_cgroup:
7589 return &bpf_skb_under_cgroup_proto;
7591 return bpf_sk_base_func_proto(func_id);
7595 static const struct bpf_func_proto *
7596 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7599 case BPF_FUNC_lwt_push_encap:
7600 return &bpf_lwt_in_push_encap_proto;
7602 return lwt_out_func_proto(func_id, prog);
7606 static const struct bpf_func_proto *
7607 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7610 case BPF_FUNC_skb_get_tunnel_key:
7611 return &bpf_skb_get_tunnel_key_proto;
7612 case BPF_FUNC_skb_set_tunnel_key:
7613 return bpf_get_skb_set_tunnel_proto(func_id);
7614 case BPF_FUNC_skb_get_tunnel_opt:
7615 return &bpf_skb_get_tunnel_opt_proto;
7616 case BPF_FUNC_skb_set_tunnel_opt:
7617 return bpf_get_skb_set_tunnel_proto(func_id);
7618 case BPF_FUNC_redirect:
7619 return &bpf_redirect_proto;
7620 case BPF_FUNC_clone_redirect:
7621 return &bpf_clone_redirect_proto;
7622 case BPF_FUNC_skb_change_tail:
7623 return &bpf_skb_change_tail_proto;
7624 case BPF_FUNC_skb_change_head:
7625 return &bpf_skb_change_head_proto;
7626 case BPF_FUNC_skb_store_bytes:
7627 return &bpf_skb_store_bytes_proto;
7628 case BPF_FUNC_csum_update:
7629 return &bpf_csum_update_proto;
7630 case BPF_FUNC_csum_level:
7631 return &bpf_csum_level_proto;
7632 case BPF_FUNC_l3_csum_replace:
7633 return &bpf_l3_csum_replace_proto;
7634 case BPF_FUNC_l4_csum_replace:
7635 return &bpf_l4_csum_replace_proto;
7636 case BPF_FUNC_set_hash_invalid:
7637 return &bpf_set_hash_invalid_proto;
7638 case BPF_FUNC_lwt_push_encap:
7639 return &bpf_lwt_xmit_push_encap_proto;
7641 return lwt_out_func_proto(func_id, prog);
7645 static const struct bpf_func_proto *
7646 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7649 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7650 case BPF_FUNC_lwt_seg6_store_bytes:
7651 return &bpf_lwt_seg6_store_bytes_proto;
7652 case BPF_FUNC_lwt_seg6_action:
7653 return &bpf_lwt_seg6_action_proto;
7654 case BPF_FUNC_lwt_seg6_adjust_srh:
7655 return &bpf_lwt_seg6_adjust_srh_proto;
7658 return lwt_out_func_proto(func_id, prog);
7662 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
7663 const struct bpf_prog *prog,
7664 struct bpf_insn_access_aux *info)
7666 const int size_default = sizeof(__u32);
7668 if (off < 0 || off >= sizeof(struct __sk_buff))
7671 /* The verifier guarantees that size > 0. */
7672 if (off % size != 0)
7676 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7677 if (off + size > offsetofend(struct __sk_buff, cb[4]))
7680 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
7681 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
7682 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
7683 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
7684 case bpf_ctx_range(struct __sk_buff, data):
7685 case bpf_ctx_range(struct __sk_buff, data_meta):
7686 case bpf_ctx_range(struct __sk_buff, data_end):
7687 if (size != size_default)
7690 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
7692 case bpf_ctx_range(struct __sk_buff, tstamp):
7693 if (size != sizeof(__u64))
7696 case offsetof(struct __sk_buff, sk):
7697 if (type == BPF_WRITE || size != sizeof(__u64))
7699 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
7702 /* Only narrow read access allowed for now. */
7703 if (type == BPF_WRITE) {
7704 if (size != size_default)
7707 bpf_ctx_record_field_size(info, size_default);
7708 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
7716 static bool sk_filter_is_valid_access(int off, int size,
7717 enum bpf_access_type type,
7718 const struct bpf_prog *prog,
7719 struct bpf_insn_access_aux *info)
7722 case bpf_ctx_range(struct __sk_buff, tc_classid):
7723 case bpf_ctx_range(struct __sk_buff, data):
7724 case bpf_ctx_range(struct __sk_buff, data_meta):
7725 case bpf_ctx_range(struct __sk_buff, data_end):
7726 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7727 case bpf_ctx_range(struct __sk_buff, tstamp):
7728 case bpf_ctx_range(struct __sk_buff, wire_len):
7732 if (type == BPF_WRITE) {
7734 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7741 return bpf_skb_is_valid_access(off, size, type, prog, info);
7744 static bool cg_skb_is_valid_access(int off, int size,
7745 enum bpf_access_type type,
7746 const struct bpf_prog *prog,
7747 struct bpf_insn_access_aux *info)
7750 case bpf_ctx_range(struct __sk_buff, tc_classid):
7751 case bpf_ctx_range(struct __sk_buff, data_meta):
7752 case bpf_ctx_range(struct __sk_buff, wire_len):
7754 case bpf_ctx_range(struct __sk_buff, data):
7755 case bpf_ctx_range(struct __sk_buff, data_end):
7761 if (type == BPF_WRITE) {
7763 case bpf_ctx_range(struct __sk_buff, mark):
7764 case bpf_ctx_range(struct __sk_buff, priority):
7765 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7767 case bpf_ctx_range(struct __sk_buff, tstamp):
7777 case bpf_ctx_range(struct __sk_buff, data):
7778 info->reg_type = PTR_TO_PACKET;
7780 case bpf_ctx_range(struct __sk_buff, data_end):
7781 info->reg_type = PTR_TO_PACKET_END;
7785 return bpf_skb_is_valid_access(off, size, type, prog, info);
7788 static bool lwt_is_valid_access(int off, int size,
7789 enum bpf_access_type type,
7790 const struct bpf_prog *prog,
7791 struct bpf_insn_access_aux *info)
7794 case bpf_ctx_range(struct __sk_buff, tc_classid):
7795 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7796 case bpf_ctx_range(struct __sk_buff, data_meta):
7797 case bpf_ctx_range(struct __sk_buff, tstamp):
7798 case bpf_ctx_range(struct __sk_buff, wire_len):
7802 if (type == BPF_WRITE) {
7804 case bpf_ctx_range(struct __sk_buff, mark):
7805 case bpf_ctx_range(struct __sk_buff, priority):
7806 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7814 case bpf_ctx_range(struct __sk_buff, data):
7815 info->reg_type = PTR_TO_PACKET;
7817 case bpf_ctx_range(struct __sk_buff, data_end):
7818 info->reg_type = PTR_TO_PACKET_END;
7822 return bpf_skb_is_valid_access(off, size, type, prog, info);
7825 /* Attach type specific accesses */
7826 static bool __sock_filter_check_attach_type(int off,
7827 enum bpf_access_type access_type,
7828 enum bpf_attach_type attach_type)
7831 case offsetof(struct bpf_sock, bound_dev_if):
7832 case offsetof(struct bpf_sock, mark):
7833 case offsetof(struct bpf_sock, priority):
7834 switch (attach_type) {
7835 case BPF_CGROUP_INET_SOCK_CREATE:
7836 case BPF_CGROUP_INET_SOCK_RELEASE:
7841 case bpf_ctx_range(struct bpf_sock, src_ip4):
7842 switch (attach_type) {
7843 case BPF_CGROUP_INET4_POST_BIND:
7848 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7849 switch (attach_type) {
7850 case BPF_CGROUP_INET6_POST_BIND:
7855 case bpf_ctx_range(struct bpf_sock, src_port):
7856 switch (attach_type) {
7857 case BPF_CGROUP_INET4_POST_BIND:
7858 case BPF_CGROUP_INET6_POST_BIND:
7865 return access_type == BPF_READ;
7870 bool bpf_sock_common_is_valid_access(int off, int size,
7871 enum bpf_access_type type,
7872 struct bpf_insn_access_aux *info)
7875 case bpf_ctx_range_till(struct bpf_sock, type, priority):
7878 return bpf_sock_is_valid_access(off, size, type, info);
7882 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7883 struct bpf_insn_access_aux *info)
7885 const int size_default = sizeof(__u32);
7887 if (off < 0 || off >= sizeof(struct bpf_sock))
7889 if (off % size != 0)
7893 case offsetof(struct bpf_sock, state):
7894 case offsetof(struct bpf_sock, family):
7895 case offsetof(struct bpf_sock, type):
7896 case offsetof(struct bpf_sock, protocol):
7897 case offsetof(struct bpf_sock, dst_port):
7898 case offsetof(struct bpf_sock, src_port):
7899 case offsetof(struct bpf_sock, rx_queue_mapping):
7900 case bpf_ctx_range(struct bpf_sock, src_ip4):
7901 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7902 case bpf_ctx_range(struct bpf_sock, dst_ip4):
7903 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
7904 bpf_ctx_record_field_size(info, size_default);
7905 return bpf_ctx_narrow_access_ok(off, size, size_default);
7908 return size == size_default;
7911 static bool sock_filter_is_valid_access(int off, int size,
7912 enum bpf_access_type type,
7913 const struct bpf_prog *prog,
7914 struct bpf_insn_access_aux *info)
7916 if (!bpf_sock_is_valid_access(off, size, type, info))
7918 return __sock_filter_check_attach_type(off, type,
7919 prog->expected_attach_type);
7922 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
7923 const struct bpf_prog *prog)
7925 /* Neither direct read nor direct write requires any preliminary
7931 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
7932 const struct bpf_prog *prog, int drop_verdict)
7934 struct bpf_insn *insn = insn_buf;
7939 /* if (!skb->cloned)
7942 * (Fast-path, otherwise approximation that we might be
7943 * a clone, do the rest in helper.)
7945 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
7946 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
7947 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
7949 /* ret = bpf_skb_pull_data(skb, 0); */
7950 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
7951 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
7952 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
7953 BPF_FUNC_skb_pull_data);
7956 * return TC_ACT_SHOT;
7958 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
7959 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
7960 *insn++ = BPF_EXIT_INSN();
7963 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
7965 *insn++ = prog->insnsi[0];
7967 return insn - insn_buf;
7970 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
7971 struct bpf_insn *insn_buf)
7973 bool indirect = BPF_MODE(orig->code) == BPF_IND;
7974 struct bpf_insn *insn = insn_buf;
7977 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
7979 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
7981 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
7983 /* We're guaranteed here that CTX is in R6. */
7984 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
7986 switch (BPF_SIZE(orig->code)) {
7988 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
7991 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
7994 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
7998 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
7999 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
8000 *insn++ = BPF_EXIT_INSN();
8002 return insn - insn_buf;
8005 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8006 const struct bpf_prog *prog)
8008 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8011 static bool tc_cls_act_is_valid_access(int off, int size,
8012 enum bpf_access_type type,
8013 const struct bpf_prog *prog,
8014 struct bpf_insn_access_aux *info)
8016 if (type == BPF_WRITE) {
8018 case bpf_ctx_range(struct __sk_buff, mark):
8019 case bpf_ctx_range(struct __sk_buff, tc_index):
8020 case bpf_ctx_range(struct __sk_buff, priority):
8021 case bpf_ctx_range(struct __sk_buff, tc_classid):
8022 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8023 case bpf_ctx_range(struct __sk_buff, tstamp):
8024 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8032 case bpf_ctx_range(struct __sk_buff, data):
8033 info->reg_type = PTR_TO_PACKET;
8035 case bpf_ctx_range(struct __sk_buff, data_meta):
8036 info->reg_type = PTR_TO_PACKET_META;
8038 case bpf_ctx_range(struct __sk_buff, data_end):
8039 info->reg_type = PTR_TO_PACKET_END;
8041 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8045 return bpf_skb_is_valid_access(off, size, type, prog, info);
8048 static bool __is_valid_xdp_access(int off, int size)
8050 if (off < 0 || off >= sizeof(struct xdp_md))
8052 if (off % size != 0)
8054 if (size != sizeof(__u32))
8060 static bool xdp_is_valid_access(int off, int size,
8061 enum bpf_access_type type,
8062 const struct bpf_prog *prog,
8063 struct bpf_insn_access_aux *info)
8065 if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8067 case offsetof(struct xdp_md, egress_ifindex):
8072 if (type == BPF_WRITE) {
8073 if (bpf_prog_is_dev_bound(prog->aux)) {
8075 case offsetof(struct xdp_md, rx_queue_index):
8076 return __is_valid_xdp_access(off, size);
8083 case offsetof(struct xdp_md, data):
8084 info->reg_type = PTR_TO_PACKET;
8086 case offsetof(struct xdp_md, data_meta):
8087 info->reg_type = PTR_TO_PACKET_META;
8089 case offsetof(struct xdp_md, data_end):
8090 info->reg_type = PTR_TO_PACKET_END;
8094 return __is_valid_xdp_access(off, size);
8097 void bpf_warn_invalid_xdp_action(u32 act)
8099 const u32 act_max = XDP_REDIRECT;
8101 WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
8102 act > act_max ? "Illegal" : "Driver unsupported",
8105 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
8107 static bool sock_addr_is_valid_access(int off, int size,
8108 enum bpf_access_type type,
8109 const struct bpf_prog *prog,
8110 struct bpf_insn_access_aux *info)
8112 const int size_default = sizeof(__u32);
8114 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
8116 if (off % size != 0)
8119 /* Disallow access to IPv6 fields from IPv4 contex and vise
8123 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8124 switch (prog->expected_attach_type) {
8125 case BPF_CGROUP_INET4_BIND:
8126 case BPF_CGROUP_INET4_CONNECT:
8127 case BPF_CGROUP_INET4_GETPEERNAME:
8128 case BPF_CGROUP_INET4_GETSOCKNAME:
8129 case BPF_CGROUP_UDP4_SENDMSG:
8130 case BPF_CGROUP_UDP4_RECVMSG:
8136 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8137 switch (prog->expected_attach_type) {
8138 case BPF_CGROUP_INET6_BIND:
8139 case BPF_CGROUP_INET6_CONNECT:
8140 case BPF_CGROUP_INET6_GETPEERNAME:
8141 case BPF_CGROUP_INET6_GETSOCKNAME:
8142 case BPF_CGROUP_UDP6_SENDMSG:
8143 case BPF_CGROUP_UDP6_RECVMSG:
8149 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8150 switch (prog->expected_attach_type) {
8151 case BPF_CGROUP_UDP4_SENDMSG:
8157 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8159 switch (prog->expected_attach_type) {
8160 case BPF_CGROUP_UDP6_SENDMSG:
8169 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8170 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8171 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8172 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8174 case bpf_ctx_range(struct bpf_sock_addr, user_port):
8175 if (type == BPF_READ) {
8176 bpf_ctx_record_field_size(info, size_default);
8178 if (bpf_ctx_wide_access_ok(off, size,
8179 struct bpf_sock_addr,
8183 if (bpf_ctx_wide_access_ok(off, size,
8184 struct bpf_sock_addr,
8188 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8191 if (bpf_ctx_wide_access_ok(off, size,
8192 struct bpf_sock_addr,
8196 if (bpf_ctx_wide_access_ok(off, size,
8197 struct bpf_sock_addr,
8201 if (size != size_default)
8205 case offsetof(struct bpf_sock_addr, sk):
8206 if (type != BPF_READ)
8208 if (size != sizeof(__u64))
8210 info->reg_type = PTR_TO_SOCKET;
8213 if (type == BPF_READ) {
8214 if (size != size_default)
8224 static bool sock_ops_is_valid_access(int off, int size,
8225 enum bpf_access_type type,
8226 const struct bpf_prog *prog,
8227 struct bpf_insn_access_aux *info)
8229 const int size_default = sizeof(__u32);
8231 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
8234 /* The verifier guarantees that size > 0. */
8235 if (off % size != 0)
8238 if (type == BPF_WRITE) {
8240 case offsetof(struct bpf_sock_ops, reply):
8241 case offsetof(struct bpf_sock_ops, sk_txhash):
8242 if (size != size_default)
8250 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
8252 if (size != sizeof(__u64))
8255 case offsetof(struct bpf_sock_ops, sk):
8256 if (size != sizeof(__u64))
8258 info->reg_type = PTR_TO_SOCKET_OR_NULL;
8260 case offsetof(struct bpf_sock_ops, skb_data):
8261 if (size != sizeof(__u64))
8263 info->reg_type = PTR_TO_PACKET;
8265 case offsetof(struct bpf_sock_ops, skb_data_end):
8266 if (size != sizeof(__u64))
8268 info->reg_type = PTR_TO_PACKET_END;
8270 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
8271 bpf_ctx_record_field_size(info, size_default);
8272 return bpf_ctx_narrow_access_ok(off, size,
8275 if (size != size_default)
8284 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
8285 const struct bpf_prog *prog)
8287 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
8290 static bool sk_skb_is_valid_access(int off, int size,
8291 enum bpf_access_type type,
8292 const struct bpf_prog *prog,
8293 struct bpf_insn_access_aux *info)
8296 case bpf_ctx_range(struct __sk_buff, tc_classid):
8297 case bpf_ctx_range(struct __sk_buff, data_meta):
8298 case bpf_ctx_range(struct __sk_buff, tstamp):
8299 case bpf_ctx_range(struct __sk_buff, wire_len):
8303 if (type == BPF_WRITE) {
8305 case bpf_ctx_range(struct __sk_buff, tc_index):
8306 case bpf_ctx_range(struct __sk_buff, priority):
8314 case bpf_ctx_range(struct __sk_buff, mark):
8316 case bpf_ctx_range(struct __sk_buff, data):
8317 info->reg_type = PTR_TO_PACKET;
8319 case bpf_ctx_range(struct __sk_buff, data_end):
8320 info->reg_type = PTR_TO_PACKET_END;
8324 return bpf_skb_is_valid_access(off, size, type, prog, info);
8327 static bool sk_msg_is_valid_access(int off, int size,
8328 enum bpf_access_type type,
8329 const struct bpf_prog *prog,
8330 struct bpf_insn_access_aux *info)
8332 if (type == BPF_WRITE)
8335 if (off % size != 0)
8339 case offsetof(struct sk_msg_md, data):
8340 info->reg_type = PTR_TO_PACKET;
8341 if (size != sizeof(__u64))
8344 case offsetof(struct sk_msg_md, data_end):
8345 info->reg_type = PTR_TO_PACKET_END;
8346 if (size != sizeof(__u64))
8349 case offsetof(struct sk_msg_md, sk):
8350 if (size != sizeof(__u64))
8352 info->reg_type = PTR_TO_SOCKET;
8354 case bpf_ctx_range(struct sk_msg_md, family):
8355 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
8356 case bpf_ctx_range(struct sk_msg_md, local_ip4):
8357 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
8358 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
8359 case bpf_ctx_range(struct sk_msg_md, remote_port):
8360 case bpf_ctx_range(struct sk_msg_md, local_port):
8361 case bpf_ctx_range(struct sk_msg_md, size):
8362 if (size != sizeof(__u32))
8371 static bool flow_dissector_is_valid_access(int off, int size,
8372 enum bpf_access_type type,
8373 const struct bpf_prog *prog,
8374 struct bpf_insn_access_aux *info)
8376 const int size_default = sizeof(__u32);
8378 if (off < 0 || off >= sizeof(struct __sk_buff))
8381 if (type == BPF_WRITE)
8385 case bpf_ctx_range(struct __sk_buff, data):
8386 if (size != size_default)
8388 info->reg_type = PTR_TO_PACKET;
8390 case bpf_ctx_range(struct __sk_buff, data_end):
8391 if (size != size_default)
8393 info->reg_type = PTR_TO_PACKET_END;
8395 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8396 if (size != sizeof(__u64))
8398 info->reg_type = PTR_TO_FLOW_KEYS;
8405 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
8406 const struct bpf_insn *si,
8407 struct bpf_insn *insn_buf,
8408 struct bpf_prog *prog,
8412 struct bpf_insn *insn = insn_buf;
8415 case offsetof(struct __sk_buff, data):
8416 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
8417 si->dst_reg, si->src_reg,
8418 offsetof(struct bpf_flow_dissector, data));
8421 case offsetof(struct __sk_buff, data_end):
8422 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
8423 si->dst_reg, si->src_reg,
8424 offsetof(struct bpf_flow_dissector, data_end));
8427 case offsetof(struct __sk_buff, flow_keys):
8428 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
8429 si->dst_reg, si->src_reg,
8430 offsetof(struct bpf_flow_dissector, flow_keys));
8434 return insn - insn_buf;
8437 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si,
8438 struct bpf_insn *insn)
8440 /* si->dst_reg = skb_shinfo(SKB); */
8441 #ifdef NET_SKBUFF_DATA_USES_OFFSET
8442 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8443 BPF_REG_AX, si->src_reg,
8444 offsetof(struct sk_buff, end));
8445 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
8446 si->dst_reg, si->src_reg,
8447 offsetof(struct sk_buff, head));
8448 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
8450 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8451 si->dst_reg, si->src_reg,
8452 offsetof(struct sk_buff, end));
8458 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
8459 const struct bpf_insn *si,
8460 struct bpf_insn *insn_buf,
8461 struct bpf_prog *prog, u32 *target_size)
8463 struct bpf_insn *insn = insn_buf;
8467 case offsetof(struct __sk_buff, len):
8468 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8469 bpf_target_off(struct sk_buff, len, 4,
8473 case offsetof(struct __sk_buff, protocol):
8474 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8475 bpf_target_off(struct sk_buff, protocol, 2,
8479 case offsetof(struct __sk_buff, vlan_proto):
8480 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8481 bpf_target_off(struct sk_buff, vlan_proto, 2,
8485 case offsetof(struct __sk_buff, priority):
8486 if (type == BPF_WRITE)
8487 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8488 bpf_target_off(struct sk_buff, priority, 4,
8491 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8492 bpf_target_off(struct sk_buff, priority, 4,
8496 case offsetof(struct __sk_buff, ingress_ifindex):
8497 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8498 bpf_target_off(struct sk_buff, skb_iif, 4,
8502 case offsetof(struct __sk_buff, ifindex):
8503 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8504 si->dst_reg, si->src_reg,
8505 offsetof(struct sk_buff, dev));
8506 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
8507 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8508 bpf_target_off(struct net_device, ifindex, 4,
8512 case offsetof(struct __sk_buff, hash):
8513 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8514 bpf_target_off(struct sk_buff, hash, 4,
8518 case offsetof(struct __sk_buff, mark):
8519 if (type == BPF_WRITE)
8520 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8521 bpf_target_off(struct sk_buff, mark, 4,
8524 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8525 bpf_target_off(struct sk_buff, mark, 4,
8529 case offsetof(struct __sk_buff, pkt_type):
8531 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8533 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
8534 #ifdef __BIG_ENDIAN_BITFIELD
8535 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
8539 case offsetof(struct __sk_buff, queue_mapping):
8540 if (type == BPF_WRITE) {
8541 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
8542 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8543 bpf_target_off(struct sk_buff,
8547 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8548 bpf_target_off(struct sk_buff,
8554 case offsetof(struct __sk_buff, vlan_present):
8556 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8557 PKT_VLAN_PRESENT_OFFSET());
8558 if (PKT_VLAN_PRESENT_BIT)
8559 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
8560 if (PKT_VLAN_PRESENT_BIT < 7)
8561 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
8564 case offsetof(struct __sk_buff, vlan_tci):
8565 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8566 bpf_target_off(struct sk_buff, vlan_tci, 2,
8570 case offsetof(struct __sk_buff, cb[0]) ...
8571 offsetofend(struct __sk_buff, cb[4]) - 1:
8572 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
8573 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
8574 offsetof(struct qdisc_skb_cb, data)) %
8577 prog->cb_access = 1;
8579 off -= offsetof(struct __sk_buff, cb[0]);
8580 off += offsetof(struct sk_buff, cb);
8581 off += offsetof(struct qdisc_skb_cb, data);
8582 if (type == BPF_WRITE)
8583 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
8586 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
8590 case offsetof(struct __sk_buff, tc_classid):
8591 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
8594 off -= offsetof(struct __sk_buff, tc_classid);
8595 off += offsetof(struct sk_buff, cb);
8596 off += offsetof(struct qdisc_skb_cb, tc_classid);
8598 if (type == BPF_WRITE)
8599 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
8602 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
8606 case offsetof(struct __sk_buff, data):
8607 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
8608 si->dst_reg, si->src_reg,
8609 offsetof(struct sk_buff, data));
8612 case offsetof(struct __sk_buff, data_meta):
8614 off -= offsetof(struct __sk_buff, data_meta);
8615 off += offsetof(struct sk_buff, cb);
8616 off += offsetof(struct bpf_skb_data_end, data_meta);
8617 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8621 case offsetof(struct __sk_buff, data_end):
8623 off -= offsetof(struct __sk_buff, data_end);
8624 off += offsetof(struct sk_buff, cb);
8625 off += offsetof(struct bpf_skb_data_end, data_end);
8626 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8630 case offsetof(struct __sk_buff, tc_index):
8631 #ifdef CONFIG_NET_SCHED
8632 if (type == BPF_WRITE)
8633 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8634 bpf_target_off(struct sk_buff, tc_index, 2,
8637 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8638 bpf_target_off(struct sk_buff, tc_index, 2,
8642 if (type == BPF_WRITE)
8643 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
8645 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8649 case offsetof(struct __sk_buff, napi_id):
8650 #if defined(CONFIG_NET_RX_BUSY_POLL)
8651 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8652 bpf_target_off(struct sk_buff, napi_id, 4,
8654 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
8655 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8658 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8661 case offsetof(struct __sk_buff, family):
8662 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
8664 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8665 si->dst_reg, si->src_reg,
8666 offsetof(struct sk_buff, sk));
8667 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8668 bpf_target_off(struct sock_common,
8672 case offsetof(struct __sk_buff, remote_ip4):
8673 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
8675 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8676 si->dst_reg, si->src_reg,
8677 offsetof(struct sk_buff, sk));
8678 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8679 bpf_target_off(struct sock_common,
8683 case offsetof(struct __sk_buff, local_ip4):
8684 BUILD_BUG_ON(sizeof_field(struct sock_common,
8685 skc_rcv_saddr) != 4);
8687 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8688 si->dst_reg, si->src_reg,
8689 offsetof(struct sk_buff, sk));
8690 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8691 bpf_target_off(struct sock_common,
8695 case offsetof(struct __sk_buff, remote_ip6[0]) ...
8696 offsetof(struct __sk_buff, remote_ip6[3]):
8697 #if IS_ENABLED(CONFIG_IPV6)
8698 BUILD_BUG_ON(sizeof_field(struct sock_common,
8699 skc_v6_daddr.s6_addr32[0]) != 4);
8702 off -= offsetof(struct __sk_buff, remote_ip6[0]);
8704 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8705 si->dst_reg, si->src_reg,
8706 offsetof(struct sk_buff, sk));
8707 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8708 offsetof(struct sock_common,
8709 skc_v6_daddr.s6_addr32[0]) +
8712 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8715 case offsetof(struct __sk_buff, local_ip6[0]) ...
8716 offsetof(struct __sk_buff, local_ip6[3]):
8717 #if IS_ENABLED(CONFIG_IPV6)
8718 BUILD_BUG_ON(sizeof_field(struct sock_common,
8719 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8722 off -= offsetof(struct __sk_buff, local_ip6[0]);
8724 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8725 si->dst_reg, si->src_reg,
8726 offsetof(struct sk_buff, sk));
8727 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8728 offsetof(struct sock_common,
8729 skc_v6_rcv_saddr.s6_addr32[0]) +
8732 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8736 case offsetof(struct __sk_buff, remote_port):
8737 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
8739 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8740 si->dst_reg, si->src_reg,
8741 offsetof(struct sk_buff, sk));
8742 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8743 bpf_target_off(struct sock_common,
8746 #ifndef __BIG_ENDIAN_BITFIELD
8747 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8751 case offsetof(struct __sk_buff, local_port):
8752 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
8754 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8755 si->dst_reg, si->src_reg,
8756 offsetof(struct sk_buff, sk));
8757 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8758 bpf_target_off(struct sock_common,
8759 skc_num, 2, target_size));
8762 case offsetof(struct __sk_buff, tstamp):
8763 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
8765 if (type == BPF_WRITE)
8766 *insn++ = BPF_STX_MEM(BPF_DW,
8767 si->dst_reg, si->src_reg,
8768 bpf_target_off(struct sk_buff,
8772 *insn++ = BPF_LDX_MEM(BPF_DW,
8773 si->dst_reg, si->src_reg,
8774 bpf_target_off(struct sk_buff,
8779 case offsetof(struct __sk_buff, gso_segs):
8780 insn = bpf_convert_shinfo_access(si, insn);
8781 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
8782 si->dst_reg, si->dst_reg,
8783 bpf_target_off(struct skb_shared_info,
8787 case offsetof(struct __sk_buff, gso_size):
8788 insn = bpf_convert_shinfo_access(si, insn);
8789 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
8790 si->dst_reg, si->dst_reg,
8791 bpf_target_off(struct skb_shared_info,
8795 case offsetof(struct __sk_buff, wire_len):
8796 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
8799 off -= offsetof(struct __sk_buff, wire_len);
8800 off += offsetof(struct sk_buff, cb);
8801 off += offsetof(struct qdisc_skb_cb, pkt_len);
8803 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
8806 case offsetof(struct __sk_buff, sk):
8807 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8808 si->dst_reg, si->src_reg,
8809 offsetof(struct sk_buff, sk));
8813 return insn - insn_buf;
8816 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
8817 const struct bpf_insn *si,
8818 struct bpf_insn *insn_buf,
8819 struct bpf_prog *prog, u32 *target_size)
8821 struct bpf_insn *insn = insn_buf;
8825 case offsetof(struct bpf_sock, bound_dev_if):
8826 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
8828 if (type == BPF_WRITE)
8829 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8830 offsetof(struct sock, sk_bound_dev_if));
8832 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8833 offsetof(struct sock, sk_bound_dev_if));
8836 case offsetof(struct bpf_sock, mark):
8837 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
8839 if (type == BPF_WRITE)
8840 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8841 offsetof(struct sock, sk_mark));
8843 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8844 offsetof(struct sock, sk_mark));
8847 case offsetof(struct bpf_sock, priority):
8848 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
8850 if (type == BPF_WRITE)
8851 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8852 offsetof(struct sock, sk_priority));
8854 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8855 offsetof(struct sock, sk_priority));
8858 case offsetof(struct bpf_sock, family):
8859 *insn++ = BPF_LDX_MEM(
8860 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
8861 si->dst_reg, si->src_reg,
8862 bpf_target_off(struct sock_common,
8864 sizeof_field(struct sock_common,
8869 case offsetof(struct bpf_sock, type):
8870 *insn++ = BPF_LDX_MEM(
8871 BPF_FIELD_SIZEOF(struct sock, sk_type),
8872 si->dst_reg, si->src_reg,
8873 bpf_target_off(struct sock, sk_type,
8874 sizeof_field(struct sock, sk_type),
8878 case offsetof(struct bpf_sock, protocol):
8879 *insn++ = BPF_LDX_MEM(
8880 BPF_FIELD_SIZEOF(struct sock, sk_protocol),
8881 si->dst_reg, si->src_reg,
8882 bpf_target_off(struct sock, sk_protocol,
8883 sizeof_field(struct sock, sk_protocol),
8887 case offsetof(struct bpf_sock, src_ip4):
8888 *insn++ = BPF_LDX_MEM(
8889 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8890 bpf_target_off(struct sock_common, skc_rcv_saddr,
8891 sizeof_field(struct sock_common,
8896 case offsetof(struct bpf_sock, dst_ip4):
8897 *insn++ = BPF_LDX_MEM(
8898 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8899 bpf_target_off(struct sock_common, skc_daddr,
8900 sizeof_field(struct sock_common,
8905 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8906 #if IS_ENABLED(CONFIG_IPV6)
8908 off -= offsetof(struct bpf_sock, src_ip6[0]);
8909 *insn++ = BPF_LDX_MEM(
8910 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8913 skc_v6_rcv_saddr.s6_addr32[0],
8914 sizeof_field(struct sock_common,
8915 skc_v6_rcv_saddr.s6_addr32[0]),
8916 target_size) + off);
8919 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8923 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8924 #if IS_ENABLED(CONFIG_IPV6)
8926 off -= offsetof(struct bpf_sock, dst_ip6[0]);
8927 *insn++ = BPF_LDX_MEM(
8928 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8929 bpf_target_off(struct sock_common,
8930 skc_v6_daddr.s6_addr32[0],
8931 sizeof_field(struct sock_common,
8932 skc_v6_daddr.s6_addr32[0]),
8933 target_size) + off);
8935 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8940 case offsetof(struct bpf_sock, src_port):
8941 *insn++ = BPF_LDX_MEM(
8942 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
8943 si->dst_reg, si->src_reg,
8944 bpf_target_off(struct sock_common, skc_num,
8945 sizeof_field(struct sock_common,
8950 case offsetof(struct bpf_sock, dst_port):
8951 *insn++ = BPF_LDX_MEM(
8952 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
8953 si->dst_reg, si->src_reg,
8954 bpf_target_off(struct sock_common, skc_dport,
8955 sizeof_field(struct sock_common,
8960 case offsetof(struct bpf_sock, state):
8961 *insn++ = BPF_LDX_MEM(
8962 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
8963 si->dst_reg, si->src_reg,
8964 bpf_target_off(struct sock_common, skc_state,
8965 sizeof_field(struct sock_common,
8969 case offsetof(struct bpf_sock, rx_queue_mapping):
8970 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
8971 *insn++ = BPF_LDX_MEM(
8972 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
8973 si->dst_reg, si->src_reg,
8974 bpf_target_off(struct sock, sk_rx_queue_mapping,
8975 sizeof_field(struct sock,
8976 sk_rx_queue_mapping),
8978 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
8980 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
8982 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
8988 return insn - insn_buf;
8991 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
8992 const struct bpf_insn *si,
8993 struct bpf_insn *insn_buf,
8994 struct bpf_prog *prog, u32 *target_size)
8996 struct bpf_insn *insn = insn_buf;
8999 case offsetof(struct __sk_buff, ifindex):
9000 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9001 si->dst_reg, si->src_reg,
9002 offsetof(struct sk_buff, dev));
9003 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9004 bpf_target_off(struct net_device, ifindex, 4,
9008 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9012 return insn - insn_buf;
9015 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
9016 const struct bpf_insn *si,
9017 struct bpf_insn *insn_buf,
9018 struct bpf_prog *prog, u32 *target_size)
9020 struct bpf_insn *insn = insn_buf;
9023 case offsetof(struct xdp_md, data):
9024 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
9025 si->dst_reg, si->src_reg,
9026 offsetof(struct xdp_buff, data));
9028 case offsetof(struct xdp_md, data_meta):
9029 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
9030 si->dst_reg, si->src_reg,
9031 offsetof(struct xdp_buff, data_meta));
9033 case offsetof(struct xdp_md, data_end):
9034 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
9035 si->dst_reg, si->src_reg,
9036 offsetof(struct xdp_buff, data_end));
9038 case offsetof(struct xdp_md, ingress_ifindex):
9039 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9040 si->dst_reg, si->src_reg,
9041 offsetof(struct xdp_buff, rxq));
9042 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
9043 si->dst_reg, si->dst_reg,
9044 offsetof(struct xdp_rxq_info, dev));
9045 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9046 offsetof(struct net_device, ifindex));
9048 case offsetof(struct xdp_md, rx_queue_index):
9049 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9050 si->dst_reg, si->src_reg,
9051 offsetof(struct xdp_buff, rxq));
9052 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9053 offsetof(struct xdp_rxq_info,
9056 case offsetof(struct xdp_md, egress_ifindex):
9057 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
9058 si->dst_reg, si->src_reg,
9059 offsetof(struct xdp_buff, txq));
9060 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
9061 si->dst_reg, si->dst_reg,
9062 offsetof(struct xdp_txq_info, dev));
9063 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9064 offsetof(struct net_device, ifindex));
9068 return insn - insn_buf;
9071 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
9072 * context Structure, F is Field in context structure that contains a pointer
9073 * to Nested Structure of type NS that has the field NF.
9075 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
9076 * sure that SIZE is not greater than actual size of S.F.NF.
9078 * If offset OFF is provided, the load happens from that offset relative to
9081 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
9083 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
9084 si->src_reg, offsetof(S, F)); \
9085 *insn++ = BPF_LDX_MEM( \
9086 SIZE, si->dst_reg, si->dst_reg, \
9087 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
9092 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
9093 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
9094 BPF_FIELD_SIZEOF(NS, NF), 0)
9096 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
9097 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
9099 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
9100 * "register" since two registers available in convert_ctx_access are not
9101 * enough: we can't override neither SRC, since it contains value to store, nor
9102 * DST since it contains pointer to context that may be used by later
9103 * instructions. But we need a temporary place to save pointer to nested
9104 * structure whose field we want to store to.
9106 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
9108 int tmp_reg = BPF_REG_9; \
9109 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
9111 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
9113 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
9115 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
9116 si->dst_reg, offsetof(S, F)); \
9117 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \
9118 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
9121 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
9125 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
9128 if (type == BPF_WRITE) { \
9129 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
9132 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
9133 S, NS, F, NF, SIZE, OFF); \
9137 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
9138 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
9139 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
9141 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
9142 const struct bpf_insn *si,
9143 struct bpf_insn *insn_buf,
9144 struct bpf_prog *prog, u32 *target_size)
9146 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
9147 struct bpf_insn *insn = insn_buf;
9150 case offsetof(struct bpf_sock_addr, user_family):
9151 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9152 struct sockaddr, uaddr, sa_family);
9155 case offsetof(struct bpf_sock_addr, user_ip4):
9156 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9157 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
9158 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
9161 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9163 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
9164 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9165 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9166 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
9170 case offsetof(struct bpf_sock_addr, user_port):
9171 /* To get port we need to know sa_family first and then treat
9172 * sockaddr as either sockaddr_in or sockaddr_in6.
9173 * Though we can simplify since port field has same offset and
9174 * size in both structures.
9175 * Here we check this invariant and use just one of the
9176 * structures if it's true.
9178 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
9179 offsetof(struct sockaddr_in6, sin6_port));
9180 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
9181 sizeof_field(struct sockaddr_in6, sin6_port));
9182 /* Account for sin6_port being smaller than user_port. */
9183 port_size = min(port_size, BPF_LDST_BYTES(si));
9184 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9185 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9186 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
9189 case offsetof(struct bpf_sock_addr, family):
9190 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9191 struct sock, sk, sk_family);
9194 case offsetof(struct bpf_sock_addr, type):
9195 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9196 struct sock, sk, sk_type);
9199 case offsetof(struct bpf_sock_addr, protocol):
9200 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9201 struct sock, sk, sk_protocol);
9204 case offsetof(struct bpf_sock_addr, msg_src_ip4):
9205 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
9206 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9207 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
9208 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
9211 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9214 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
9215 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
9216 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9217 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
9218 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
9220 case offsetof(struct bpf_sock_addr, sk):
9221 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
9222 si->dst_reg, si->src_reg,
9223 offsetof(struct bpf_sock_addr_kern, sk));
9227 return insn - insn_buf;
9230 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
9231 const struct bpf_insn *si,
9232 struct bpf_insn *insn_buf,
9233 struct bpf_prog *prog,
9236 struct bpf_insn *insn = insn_buf;
9239 /* Helper macro for adding read access to tcp_sock or sock fields. */
9240 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
9242 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \
9243 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
9244 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
9245 if (si->dst_reg == reg || si->src_reg == reg) \
9247 if (si->dst_reg == reg || si->src_reg == reg) \
9249 if (si->dst_reg == si->src_reg) { \
9250 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
9251 offsetof(struct bpf_sock_ops_kern, \
9253 fullsock_reg = reg; \
9256 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9257 struct bpf_sock_ops_kern, \
9259 fullsock_reg, si->src_reg, \
9260 offsetof(struct bpf_sock_ops_kern, \
9262 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
9263 if (si->dst_reg == si->src_reg) \
9264 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9265 offsetof(struct bpf_sock_ops_kern, \
9267 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9268 struct bpf_sock_ops_kern, sk),\
9269 si->dst_reg, si->src_reg, \
9270 offsetof(struct bpf_sock_ops_kern, sk));\
9271 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
9273 si->dst_reg, si->dst_reg, \
9274 offsetof(OBJ, OBJ_FIELD)); \
9275 if (si->dst_reg == si->src_reg) { \
9276 *insn++ = BPF_JMP_A(1); \
9277 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9278 offsetof(struct bpf_sock_ops_kern, \
9283 #define SOCK_OPS_GET_SK() \
9285 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \
9286 if (si->dst_reg == reg || si->src_reg == reg) \
9288 if (si->dst_reg == reg || si->src_reg == reg) \
9290 if (si->dst_reg == si->src_reg) { \
9291 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
9292 offsetof(struct bpf_sock_ops_kern, \
9294 fullsock_reg = reg; \
9297 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9298 struct bpf_sock_ops_kern, \
9300 fullsock_reg, si->src_reg, \
9301 offsetof(struct bpf_sock_ops_kern, \
9303 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
9304 if (si->dst_reg == si->src_reg) \
9305 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9306 offsetof(struct bpf_sock_ops_kern, \
9308 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9309 struct bpf_sock_ops_kern, sk),\
9310 si->dst_reg, si->src_reg, \
9311 offsetof(struct bpf_sock_ops_kern, sk));\
9312 if (si->dst_reg == si->src_reg) { \
9313 *insn++ = BPF_JMP_A(1); \
9314 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9315 offsetof(struct bpf_sock_ops_kern, \
9320 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
9321 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
9323 /* Helper macro for adding write access to tcp_sock or sock fields.
9324 * The macro is called with two registers, dst_reg which contains a pointer
9325 * to ctx (context) and src_reg which contains the value that should be
9326 * stored. However, we need an additional register since we cannot overwrite
9327 * dst_reg because it may be used later in the program.
9328 * Instead we "borrow" one of the other register. We first save its value
9329 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
9330 * it at the end of the macro.
9332 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
9334 int reg = BPF_REG_9; \
9335 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
9336 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
9337 if (si->dst_reg == reg || si->src_reg == reg) \
9339 if (si->dst_reg == reg || si->src_reg == reg) \
9341 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
9342 offsetof(struct bpf_sock_ops_kern, \
9344 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9345 struct bpf_sock_ops_kern, \
9348 offsetof(struct bpf_sock_ops_kern, \
9350 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
9351 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9352 struct bpf_sock_ops_kern, sk),\
9354 offsetof(struct bpf_sock_ops_kern, sk));\
9355 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \
9357 offsetof(OBJ, OBJ_FIELD)); \
9358 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
9359 offsetof(struct bpf_sock_ops_kern, \
9363 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
9365 if (TYPE == BPF_WRITE) \
9366 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
9368 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
9371 if (insn > insn_buf)
9372 return insn - insn_buf;
9375 case offsetof(struct bpf_sock_ops, op):
9376 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9378 si->dst_reg, si->src_reg,
9379 offsetof(struct bpf_sock_ops_kern, op));
9382 case offsetof(struct bpf_sock_ops, replylong[0]) ...
9383 offsetof(struct bpf_sock_ops, replylong[3]):
9384 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
9385 sizeof_field(struct bpf_sock_ops_kern, reply));
9386 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
9387 sizeof_field(struct bpf_sock_ops_kern, replylong));
9389 off -= offsetof(struct bpf_sock_ops, replylong[0]);
9390 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
9391 if (type == BPF_WRITE)
9392 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9395 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9399 case offsetof(struct bpf_sock_ops, family):
9400 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9402 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9403 struct bpf_sock_ops_kern, sk),
9404 si->dst_reg, si->src_reg,
9405 offsetof(struct bpf_sock_ops_kern, sk));
9406 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9407 offsetof(struct sock_common, skc_family));
9410 case offsetof(struct bpf_sock_ops, remote_ip4):
9411 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9413 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9414 struct bpf_sock_ops_kern, sk),
9415 si->dst_reg, si->src_reg,
9416 offsetof(struct bpf_sock_ops_kern, sk));
9417 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9418 offsetof(struct sock_common, skc_daddr));
9421 case offsetof(struct bpf_sock_ops, local_ip4):
9422 BUILD_BUG_ON(sizeof_field(struct sock_common,
9423 skc_rcv_saddr) != 4);
9425 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9426 struct bpf_sock_ops_kern, sk),
9427 si->dst_reg, si->src_reg,
9428 offsetof(struct bpf_sock_ops_kern, sk));
9429 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9430 offsetof(struct sock_common,
9434 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
9435 offsetof(struct bpf_sock_ops, remote_ip6[3]):
9436 #if IS_ENABLED(CONFIG_IPV6)
9437 BUILD_BUG_ON(sizeof_field(struct sock_common,
9438 skc_v6_daddr.s6_addr32[0]) != 4);
9441 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
9442 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9443 struct bpf_sock_ops_kern, sk),
9444 si->dst_reg, si->src_reg,
9445 offsetof(struct bpf_sock_ops_kern, sk));
9446 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9447 offsetof(struct sock_common,
9448 skc_v6_daddr.s6_addr32[0]) +
9451 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9455 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
9456 offsetof(struct bpf_sock_ops, local_ip6[3]):
9457 #if IS_ENABLED(CONFIG_IPV6)
9458 BUILD_BUG_ON(sizeof_field(struct sock_common,
9459 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9462 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
9463 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9464 struct bpf_sock_ops_kern, sk),
9465 si->dst_reg, si->src_reg,
9466 offsetof(struct bpf_sock_ops_kern, sk));
9467 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9468 offsetof(struct sock_common,
9469 skc_v6_rcv_saddr.s6_addr32[0]) +
9472 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9476 case offsetof(struct bpf_sock_ops, remote_port):
9477 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9479 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9480 struct bpf_sock_ops_kern, sk),
9481 si->dst_reg, si->src_reg,
9482 offsetof(struct bpf_sock_ops_kern, sk));
9483 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9484 offsetof(struct sock_common, skc_dport));
9485 #ifndef __BIG_ENDIAN_BITFIELD
9486 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9490 case offsetof(struct bpf_sock_ops, local_port):
9491 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9493 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9494 struct bpf_sock_ops_kern, sk),
9495 si->dst_reg, si->src_reg,
9496 offsetof(struct bpf_sock_ops_kern, sk));
9497 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9498 offsetof(struct sock_common, skc_num));
9501 case offsetof(struct bpf_sock_ops, is_fullsock):
9502 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9503 struct bpf_sock_ops_kern,
9505 si->dst_reg, si->src_reg,
9506 offsetof(struct bpf_sock_ops_kern,
9510 case offsetof(struct bpf_sock_ops, state):
9511 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
9513 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9514 struct bpf_sock_ops_kern, sk),
9515 si->dst_reg, si->src_reg,
9516 offsetof(struct bpf_sock_ops_kern, sk));
9517 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
9518 offsetof(struct sock_common, skc_state));
9521 case offsetof(struct bpf_sock_ops, rtt_min):
9522 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
9523 sizeof(struct minmax));
9524 BUILD_BUG_ON(sizeof(struct minmax) <
9525 sizeof(struct minmax_sample));
9527 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9528 struct bpf_sock_ops_kern, sk),
9529 si->dst_reg, si->src_reg,
9530 offsetof(struct bpf_sock_ops_kern, sk));
9531 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9532 offsetof(struct tcp_sock, rtt_min) +
9533 sizeof_field(struct minmax_sample, t));
9536 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
9537 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
9541 case offsetof(struct bpf_sock_ops, sk_txhash):
9542 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
9545 case offsetof(struct bpf_sock_ops, snd_cwnd):
9546 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
9548 case offsetof(struct bpf_sock_ops, srtt_us):
9549 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
9551 case offsetof(struct bpf_sock_ops, snd_ssthresh):
9552 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
9554 case offsetof(struct bpf_sock_ops, rcv_nxt):
9555 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
9557 case offsetof(struct bpf_sock_ops, snd_nxt):
9558 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
9560 case offsetof(struct bpf_sock_ops, snd_una):
9561 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
9563 case offsetof(struct bpf_sock_ops, mss_cache):
9564 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
9566 case offsetof(struct bpf_sock_ops, ecn_flags):
9567 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
9569 case offsetof(struct bpf_sock_ops, rate_delivered):
9570 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
9572 case offsetof(struct bpf_sock_ops, rate_interval_us):
9573 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
9575 case offsetof(struct bpf_sock_ops, packets_out):
9576 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
9578 case offsetof(struct bpf_sock_ops, retrans_out):
9579 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
9581 case offsetof(struct bpf_sock_ops, total_retrans):
9582 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
9584 case offsetof(struct bpf_sock_ops, segs_in):
9585 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
9587 case offsetof(struct bpf_sock_ops, data_segs_in):
9588 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
9590 case offsetof(struct bpf_sock_ops, segs_out):
9591 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
9593 case offsetof(struct bpf_sock_ops, data_segs_out):
9594 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
9596 case offsetof(struct bpf_sock_ops, lost_out):
9597 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
9599 case offsetof(struct bpf_sock_ops, sacked_out):
9600 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
9602 case offsetof(struct bpf_sock_ops, bytes_received):
9603 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
9605 case offsetof(struct bpf_sock_ops, bytes_acked):
9606 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
9608 case offsetof(struct bpf_sock_ops, sk):
9611 case offsetof(struct bpf_sock_ops, skb_data_end):
9612 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9614 si->dst_reg, si->src_reg,
9615 offsetof(struct bpf_sock_ops_kern,
9618 case offsetof(struct bpf_sock_ops, skb_data):
9619 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9621 si->dst_reg, si->src_reg,
9622 offsetof(struct bpf_sock_ops_kern,
9624 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9625 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9626 si->dst_reg, si->dst_reg,
9627 offsetof(struct sk_buff, data));
9629 case offsetof(struct bpf_sock_ops, skb_len):
9630 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9632 si->dst_reg, si->src_reg,
9633 offsetof(struct bpf_sock_ops_kern,
9635 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9636 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9637 si->dst_reg, si->dst_reg,
9638 offsetof(struct sk_buff, len));
9640 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9641 off = offsetof(struct sk_buff, cb);
9642 off += offsetof(struct tcp_skb_cb, tcp_flags);
9643 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
9644 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9646 si->dst_reg, si->src_reg,
9647 offsetof(struct bpf_sock_ops_kern,
9649 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9650 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
9652 si->dst_reg, si->dst_reg, off);
9655 return insn - insn_buf;
9658 /* data_end = skb->data + skb_headlen() */
9659 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
9660 struct bpf_insn *insn)
9662 /* si->dst_reg = skb->data */
9663 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9664 si->dst_reg, si->src_reg,
9665 offsetof(struct sk_buff, data));
9667 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9668 BPF_REG_AX, si->src_reg,
9669 offsetof(struct sk_buff, len));
9670 /* si->dst_reg = skb->data + skb->len */
9671 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
9672 /* AX = skb->data_len */
9673 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
9674 BPF_REG_AX, si->src_reg,
9675 offsetof(struct sk_buff, data_len));
9676 /* si->dst_reg = skb->data + skb->len - skb->data_len */
9677 *insn++ = BPF_ALU64_REG(BPF_SUB, si->dst_reg, BPF_REG_AX);
9682 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
9683 const struct bpf_insn *si,
9684 struct bpf_insn *insn_buf,
9685 struct bpf_prog *prog, u32 *target_size)
9687 struct bpf_insn *insn = insn_buf;
9690 case offsetof(struct __sk_buff, data_end):
9691 insn = bpf_convert_data_end_access(si, insn);
9694 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9698 return insn - insn_buf;
9701 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
9702 const struct bpf_insn *si,
9703 struct bpf_insn *insn_buf,
9704 struct bpf_prog *prog, u32 *target_size)
9706 struct bpf_insn *insn = insn_buf;
9707 #if IS_ENABLED(CONFIG_IPV6)
9711 /* convert ctx uses the fact sg element is first in struct */
9712 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
9715 case offsetof(struct sk_msg_md, data):
9716 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
9717 si->dst_reg, si->src_reg,
9718 offsetof(struct sk_msg, data));
9720 case offsetof(struct sk_msg_md, data_end):
9721 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
9722 si->dst_reg, si->src_reg,
9723 offsetof(struct sk_msg, data_end));
9725 case offsetof(struct sk_msg_md, family):
9726 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9728 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9730 si->dst_reg, si->src_reg,
9731 offsetof(struct sk_msg, sk));
9732 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9733 offsetof(struct sock_common, skc_family));
9736 case offsetof(struct sk_msg_md, remote_ip4):
9737 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9739 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9741 si->dst_reg, si->src_reg,
9742 offsetof(struct sk_msg, sk));
9743 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9744 offsetof(struct sock_common, skc_daddr));
9747 case offsetof(struct sk_msg_md, local_ip4):
9748 BUILD_BUG_ON(sizeof_field(struct sock_common,
9749 skc_rcv_saddr) != 4);
9751 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9753 si->dst_reg, si->src_reg,
9754 offsetof(struct sk_msg, sk));
9755 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9756 offsetof(struct sock_common,
9760 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
9761 offsetof(struct sk_msg_md, remote_ip6[3]):
9762 #if IS_ENABLED(CONFIG_IPV6)
9763 BUILD_BUG_ON(sizeof_field(struct sock_common,
9764 skc_v6_daddr.s6_addr32[0]) != 4);
9767 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
9768 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9770 si->dst_reg, si->src_reg,
9771 offsetof(struct sk_msg, sk));
9772 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9773 offsetof(struct sock_common,
9774 skc_v6_daddr.s6_addr32[0]) +
9777 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9781 case offsetof(struct sk_msg_md, local_ip6[0]) ...
9782 offsetof(struct sk_msg_md, local_ip6[3]):
9783 #if IS_ENABLED(CONFIG_IPV6)
9784 BUILD_BUG_ON(sizeof_field(struct sock_common,
9785 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9788 off -= offsetof(struct sk_msg_md, local_ip6[0]);
9789 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9791 si->dst_reg, si->src_reg,
9792 offsetof(struct sk_msg, sk));
9793 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9794 offsetof(struct sock_common,
9795 skc_v6_rcv_saddr.s6_addr32[0]) +
9798 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9802 case offsetof(struct sk_msg_md, remote_port):
9803 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9805 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9807 si->dst_reg, si->src_reg,
9808 offsetof(struct sk_msg, sk));
9809 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9810 offsetof(struct sock_common, skc_dport));
9811 #ifndef __BIG_ENDIAN_BITFIELD
9812 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9816 case offsetof(struct sk_msg_md, local_port):
9817 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9819 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9821 si->dst_reg, si->src_reg,
9822 offsetof(struct sk_msg, sk));
9823 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9824 offsetof(struct sock_common, skc_num));
9827 case offsetof(struct sk_msg_md, size):
9828 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
9829 si->dst_reg, si->src_reg,
9830 offsetof(struct sk_msg_sg, size));
9833 case offsetof(struct sk_msg_md, sk):
9834 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
9835 si->dst_reg, si->src_reg,
9836 offsetof(struct sk_msg, sk));
9840 return insn - insn_buf;
9843 const struct bpf_verifier_ops sk_filter_verifier_ops = {
9844 .get_func_proto = sk_filter_func_proto,
9845 .is_valid_access = sk_filter_is_valid_access,
9846 .convert_ctx_access = bpf_convert_ctx_access,
9847 .gen_ld_abs = bpf_gen_ld_abs,
9850 const struct bpf_prog_ops sk_filter_prog_ops = {
9851 .test_run = bpf_prog_test_run_skb,
9854 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
9855 .get_func_proto = tc_cls_act_func_proto,
9856 .is_valid_access = tc_cls_act_is_valid_access,
9857 .convert_ctx_access = tc_cls_act_convert_ctx_access,
9858 .gen_prologue = tc_cls_act_prologue,
9859 .gen_ld_abs = bpf_gen_ld_abs,
9860 .check_kfunc_call = bpf_prog_test_check_kfunc_call,
9863 const struct bpf_prog_ops tc_cls_act_prog_ops = {
9864 .test_run = bpf_prog_test_run_skb,
9867 const struct bpf_verifier_ops xdp_verifier_ops = {
9868 .get_func_proto = xdp_func_proto,
9869 .is_valid_access = xdp_is_valid_access,
9870 .convert_ctx_access = xdp_convert_ctx_access,
9871 .gen_prologue = bpf_noop_prologue,
9874 const struct bpf_prog_ops xdp_prog_ops = {
9875 .test_run = bpf_prog_test_run_xdp,
9878 const struct bpf_verifier_ops cg_skb_verifier_ops = {
9879 .get_func_proto = cg_skb_func_proto,
9880 .is_valid_access = cg_skb_is_valid_access,
9881 .convert_ctx_access = bpf_convert_ctx_access,
9884 const struct bpf_prog_ops cg_skb_prog_ops = {
9885 .test_run = bpf_prog_test_run_skb,
9888 const struct bpf_verifier_ops lwt_in_verifier_ops = {
9889 .get_func_proto = lwt_in_func_proto,
9890 .is_valid_access = lwt_is_valid_access,
9891 .convert_ctx_access = bpf_convert_ctx_access,
9894 const struct bpf_prog_ops lwt_in_prog_ops = {
9895 .test_run = bpf_prog_test_run_skb,
9898 const struct bpf_verifier_ops lwt_out_verifier_ops = {
9899 .get_func_proto = lwt_out_func_proto,
9900 .is_valid_access = lwt_is_valid_access,
9901 .convert_ctx_access = bpf_convert_ctx_access,
9904 const struct bpf_prog_ops lwt_out_prog_ops = {
9905 .test_run = bpf_prog_test_run_skb,
9908 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
9909 .get_func_proto = lwt_xmit_func_proto,
9910 .is_valid_access = lwt_is_valid_access,
9911 .convert_ctx_access = bpf_convert_ctx_access,
9912 .gen_prologue = tc_cls_act_prologue,
9915 const struct bpf_prog_ops lwt_xmit_prog_ops = {
9916 .test_run = bpf_prog_test_run_skb,
9919 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
9920 .get_func_proto = lwt_seg6local_func_proto,
9921 .is_valid_access = lwt_is_valid_access,
9922 .convert_ctx_access = bpf_convert_ctx_access,
9925 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
9926 .test_run = bpf_prog_test_run_skb,
9929 const struct bpf_verifier_ops cg_sock_verifier_ops = {
9930 .get_func_proto = sock_filter_func_proto,
9931 .is_valid_access = sock_filter_is_valid_access,
9932 .convert_ctx_access = bpf_sock_convert_ctx_access,
9935 const struct bpf_prog_ops cg_sock_prog_ops = {
9938 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
9939 .get_func_proto = sock_addr_func_proto,
9940 .is_valid_access = sock_addr_is_valid_access,
9941 .convert_ctx_access = sock_addr_convert_ctx_access,
9944 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
9947 const struct bpf_verifier_ops sock_ops_verifier_ops = {
9948 .get_func_proto = sock_ops_func_proto,
9949 .is_valid_access = sock_ops_is_valid_access,
9950 .convert_ctx_access = sock_ops_convert_ctx_access,
9953 const struct bpf_prog_ops sock_ops_prog_ops = {
9956 const struct bpf_verifier_ops sk_skb_verifier_ops = {
9957 .get_func_proto = sk_skb_func_proto,
9958 .is_valid_access = sk_skb_is_valid_access,
9959 .convert_ctx_access = sk_skb_convert_ctx_access,
9960 .gen_prologue = sk_skb_prologue,
9963 const struct bpf_prog_ops sk_skb_prog_ops = {
9966 const struct bpf_verifier_ops sk_msg_verifier_ops = {
9967 .get_func_proto = sk_msg_func_proto,
9968 .is_valid_access = sk_msg_is_valid_access,
9969 .convert_ctx_access = sk_msg_convert_ctx_access,
9970 .gen_prologue = bpf_noop_prologue,
9973 const struct bpf_prog_ops sk_msg_prog_ops = {
9976 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
9977 .get_func_proto = flow_dissector_func_proto,
9978 .is_valid_access = flow_dissector_is_valid_access,
9979 .convert_ctx_access = flow_dissector_convert_ctx_access,
9982 const struct bpf_prog_ops flow_dissector_prog_ops = {
9983 .test_run = bpf_prog_test_run_flow_dissector,
9986 int sk_detach_filter(struct sock *sk)
9989 struct sk_filter *filter;
9991 if (sock_flag(sk, SOCK_FILTER_LOCKED))
9994 filter = rcu_dereference_protected(sk->sk_filter,
9995 lockdep_sock_is_held(sk));
9997 RCU_INIT_POINTER(sk->sk_filter, NULL);
9998 sk_filter_uncharge(sk, filter);
10004 EXPORT_SYMBOL_GPL(sk_detach_filter);
10006 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
10009 struct sock_fprog_kern *fprog;
10010 struct sk_filter *filter;
10014 filter = rcu_dereference_protected(sk->sk_filter,
10015 lockdep_sock_is_held(sk));
10019 /* We're copying the filter that has been originally attached,
10020 * so no conversion/decode needed anymore. eBPF programs that
10021 * have no original program cannot be dumped through this.
10024 fprog = filter->prog->orig_prog;
10030 /* User space only enquires number of filter blocks. */
10034 if (len < fprog->len)
10038 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
10041 /* Instead of bytes, the API requests to return the number
10042 * of filter blocks.
10051 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
10052 struct sock_reuseport *reuse,
10053 struct sock *sk, struct sk_buff *skb,
10054 struct sock *migrating_sk,
10057 reuse_kern->skb = skb;
10058 reuse_kern->sk = sk;
10059 reuse_kern->selected_sk = NULL;
10060 reuse_kern->migrating_sk = migrating_sk;
10061 reuse_kern->data_end = skb->data + skb_headlen(skb);
10062 reuse_kern->hash = hash;
10063 reuse_kern->reuseport_id = reuse->reuseport_id;
10064 reuse_kern->bind_inany = reuse->bind_inany;
10067 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
10068 struct bpf_prog *prog, struct sk_buff *skb,
10069 struct sock *migrating_sk,
10072 struct sk_reuseport_kern reuse_kern;
10073 enum sk_action action;
10075 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
10076 action = BPF_PROG_RUN(prog, &reuse_kern);
10078 if (action == SK_PASS)
10079 return reuse_kern.selected_sk;
10081 return ERR_PTR(-ECONNREFUSED);
10084 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
10085 struct bpf_map *, map, void *, key, u32, flags)
10087 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
10088 struct sock_reuseport *reuse;
10089 struct sock *selected_sk;
10091 selected_sk = map->ops->map_lookup_elem(map, key);
10095 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
10097 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
10098 if (sk_is_refcounted(selected_sk))
10099 sock_put(selected_sk);
10101 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
10102 * The only (!reuse) case here is - the sk has already been
10103 * unhashed (e.g. by close()), so treat it as -ENOENT.
10105 * Other maps (e.g. sock_map) do not provide this guarantee and
10106 * the sk may never be in the reuseport group to begin with.
10108 return is_sockarray ? -ENOENT : -EINVAL;
10111 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
10112 struct sock *sk = reuse_kern->sk;
10114 if (sk->sk_protocol != selected_sk->sk_protocol)
10115 return -EPROTOTYPE;
10116 else if (sk->sk_family != selected_sk->sk_family)
10117 return -EAFNOSUPPORT;
10119 /* Catch all. Likely bound to a different sockaddr. */
10123 reuse_kern->selected_sk = selected_sk;
10128 static const struct bpf_func_proto sk_select_reuseport_proto = {
10129 .func = sk_select_reuseport,
10131 .ret_type = RET_INTEGER,
10132 .arg1_type = ARG_PTR_TO_CTX,
10133 .arg2_type = ARG_CONST_MAP_PTR,
10134 .arg3_type = ARG_PTR_TO_MAP_KEY,
10135 .arg4_type = ARG_ANYTHING,
10138 BPF_CALL_4(sk_reuseport_load_bytes,
10139 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10140 void *, to, u32, len)
10142 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
10145 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
10146 .func = sk_reuseport_load_bytes,
10148 .ret_type = RET_INTEGER,
10149 .arg1_type = ARG_PTR_TO_CTX,
10150 .arg2_type = ARG_ANYTHING,
10151 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
10152 .arg4_type = ARG_CONST_SIZE,
10155 BPF_CALL_5(sk_reuseport_load_bytes_relative,
10156 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10157 void *, to, u32, len, u32, start_header)
10159 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
10160 len, start_header);
10163 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
10164 .func = sk_reuseport_load_bytes_relative,
10166 .ret_type = RET_INTEGER,
10167 .arg1_type = ARG_PTR_TO_CTX,
10168 .arg2_type = ARG_ANYTHING,
10169 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
10170 .arg4_type = ARG_CONST_SIZE,
10171 .arg5_type = ARG_ANYTHING,
10174 static const struct bpf_func_proto *
10175 sk_reuseport_func_proto(enum bpf_func_id func_id,
10176 const struct bpf_prog *prog)
10179 case BPF_FUNC_sk_select_reuseport:
10180 return &sk_select_reuseport_proto;
10181 case BPF_FUNC_skb_load_bytes:
10182 return &sk_reuseport_load_bytes_proto;
10183 case BPF_FUNC_skb_load_bytes_relative:
10184 return &sk_reuseport_load_bytes_relative_proto;
10185 case BPF_FUNC_get_socket_cookie:
10186 return &bpf_get_socket_ptr_cookie_proto;
10188 return bpf_base_func_proto(func_id);
10193 sk_reuseport_is_valid_access(int off, int size,
10194 enum bpf_access_type type,
10195 const struct bpf_prog *prog,
10196 struct bpf_insn_access_aux *info)
10198 const u32 size_default = sizeof(__u32);
10200 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
10201 off % size || type != BPF_READ)
10205 case offsetof(struct sk_reuseport_md, data):
10206 info->reg_type = PTR_TO_PACKET;
10207 return size == sizeof(__u64);
10209 case offsetof(struct sk_reuseport_md, data_end):
10210 info->reg_type = PTR_TO_PACKET_END;
10211 return size == sizeof(__u64);
10213 case offsetof(struct sk_reuseport_md, hash):
10214 return size == size_default;
10216 case offsetof(struct sk_reuseport_md, sk):
10217 info->reg_type = PTR_TO_SOCKET;
10218 return size == sizeof(__u64);
10220 case offsetof(struct sk_reuseport_md, migrating_sk):
10221 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
10222 return size == sizeof(__u64);
10224 /* Fields that allow narrowing */
10225 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
10226 if (size < sizeof_field(struct sk_buff, protocol))
10229 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
10230 case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
10231 case bpf_ctx_range(struct sk_reuseport_md, len):
10232 bpf_ctx_record_field_size(info, size_default);
10233 return bpf_ctx_narrow_access_ok(off, size, size_default);
10240 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
10241 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
10242 si->dst_reg, si->src_reg, \
10243 bpf_target_off(struct sk_reuseport_kern, F, \
10244 sizeof_field(struct sk_reuseport_kern, F), \
10248 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
10249 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
10254 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \
10255 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
10260 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
10261 const struct bpf_insn *si,
10262 struct bpf_insn *insn_buf,
10263 struct bpf_prog *prog,
10266 struct bpf_insn *insn = insn_buf;
10269 case offsetof(struct sk_reuseport_md, data):
10270 SK_REUSEPORT_LOAD_SKB_FIELD(data);
10273 case offsetof(struct sk_reuseport_md, len):
10274 SK_REUSEPORT_LOAD_SKB_FIELD(len);
10277 case offsetof(struct sk_reuseport_md, eth_protocol):
10278 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
10281 case offsetof(struct sk_reuseport_md, ip_protocol):
10282 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
10285 case offsetof(struct sk_reuseport_md, data_end):
10286 SK_REUSEPORT_LOAD_FIELD(data_end);
10289 case offsetof(struct sk_reuseport_md, hash):
10290 SK_REUSEPORT_LOAD_FIELD(hash);
10293 case offsetof(struct sk_reuseport_md, bind_inany):
10294 SK_REUSEPORT_LOAD_FIELD(bind_inany);
10297 case offsetof(struct sk_reuseport_md, sk):
10298 SK_REUSEPORT_LOAD_FIELD(sk);
10301 case offsetof(struct sk_reuseport_md, migrating_sk):
10302 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
10306 return insn - insn_buf;
10309 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
10310 .get_func_proto = sk_reuseport_func_proto,
10311 .is_valid_access = sk_reuseport_is_valid_access,
10312 .convert_ctx_access = sk_reuseport_convert_ctx_access,
10315 const struct bpf_prog_ops sk_reuseport_prog_ops = {
10318 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
10319 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
10321 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
10322 struct sock *, sk, u64, flags)
10324 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
10325 BPF_SK_LOOKUP_F_NO_REUSEPORT)))
10327 if (unlikely(sk && sk_is_refcounted(sk)))
10328 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
10329 if (unlikely(sk && sk->sk_state == TCP_ESTABLISHED))
10330 return -ESOCKTNOSUPPORT; /* reject connected sockets */
10332 /* Check if socket is suitable for packet L3/L4 protocol */
10333 if (sk && sk->sk_protocol != ctx->protocol)
10334 return -EPROTOTYPE;
10335 if (sk && sk->sk_family != ctx->family &&
10336 (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
10337 return -EAFNOSUPPORT;
10339 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
10342 /* Select socket as lookup result */
10343 ctx->selected_sk = sk;
10344 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
10348 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
10349 .func = bpf_sk_lookup_assign,
10351 .ret_type = RET_INTEGER,
10352 .arg1_type = ARG_PTR_TO_CTX,
10353 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL,
10354 .arg3_type = ARG_ANYTHING,
10357 static const struct bpf_func_proto *
10358 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
10361 case BPF_FUNC_perf_event_output:
10362 return &bpf_event_output_data_proto;
10363 case BPF_FUNC_sk_assign:
10364 return &bpf_sk_lookup_assign_proto;
10365 case BPF_FUNC_sk_release:
10366 return &bpf_sk_release_proto;
10368 return bpf_sk_base_func_proto(func_id);
10372 static bool sk_lookup_is_valid_access(int off, int size,
10373 enum bpf_access_type type,
10374 const struct bpf_prog *prog,
10375 struct bpf_insn_access_aux *info)
10377 if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
10379 if (off % size != 0)
10381 if (type != BPF_READ)
10385 case offsetof(struct bpf_sk_lookup, sk):
10386 info->reg_type = PTR_TO_SOCKET_OR_NULL;
10387 return size == sizeof(__u64);
10389 case bpf_ctx_range(struct bpf_sk_lookup, family):
10390 case bpf_ctx_range(struct bpf_sk_lookup, protocol):
10391 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
10392 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
10393 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
10394 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
10395 case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
10396 case bpf_ctx_range(struct bpf_sk_lookup, local_port):
10397 bpf_ctx_record_field_size(info, sizeof(__u32));
10398 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
10405 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
10406 const struct bpf_insn *si,
10407 struct bpf_insn *insn_buf,
10408 struct bpf_prog *prog,
10411 struct bpf_insn *insn = insn_buf;
10414 case offsetof(struct bpf_sk_lookup, sk):
10415 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10416 offsetof(struct bpf_sk_lookup_kern, selected_sk));
10419 case offsetof(struct bpf_sk_lookup, family):
10420 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10421 bpf_target_off(struct bpf_sk_lookup_kern,
10422 family, 2, target_size));
10425 case offsetof(struct bpf_sk_lookup, protocol):
10426 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10427 bpf_target_off(struct bpf_sk_lookup_kern,
10428 protocol, 2, target_size));
10431 case offsetof(struct bpf_sk_lookup, remote_ip4):
10432 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10433 bpf_target_off(struct bpf_sk_lookup_kern,
10434 v4.saddr, 4, target_size));
10437 case offsetof(struct bpf_sk_lookup, local_ip4):
10438 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10439 bpf_target_off(struct bpf_sk_lookup_kern,
10440 v4.daddr, 4, target_size));
10443 case bpf_ctx_range_till(struct bpf_sk_lookup,
10444 remote_ip6[0], remote_ip6[3]): {
10445 #if IS_ENABLED(CONFIG_IPV6)
10448 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
10449 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10450 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10451 offsetof(struct bpf_sk_lookup_kern, v6.saddr));
10452 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10453 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10455 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10459 case bpf_ctx_range_till(struct bpf_sk_lookup,
10460 local_ip6[0], local_ip6[3]): {
10461 #if IS_ENABLED(CONFIG_IPV6)
10464 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
10465 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10466 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10467 offsetof(struct bpf_sk_lookup_kern, v6.daddr));
10468 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10469 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10471 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10475 case offsetof(struct bpf_sk_lookup, remote_port):
10476 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10477 bpf_target_off(struct bpf_sk_lookup_kern,
10478 sport, 2, target_size));
10481 case offsetof(struct bpf_sk_lookup, local_port):
10482 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10483 bpf_target_off(struct bpf_sk_lookup_kern,
10484 dport, 2, target_size));
10488 return insn - insn_buf;
10491 const struct bpf_prog_ops sk_lookup_prog_ops = {
10492 .test_run = bpf_prog_test_run_sk_lookup,
10495 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
10496 .get_func_proto = sk_lookup_func_proto,
10497 .is_valid_access = sk_lookup_is_valid_access,
10498 .convert_ctx_access = sk_lookup_convert_ctx_access,
10501 #endif /* CONFIG_INET */
10503 DEFINE_BPF_DISPATCHER(xdp)
10505 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
10507 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
10510 #ifdef CONFIG_DEBUG_INFO_BTF
10511 BTF_ID_LIST_GLOBAL(btf_sock_ids)
10512 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
10514 #undef BTF_SOCK_TYPE
10516 u32 btf_sock_ids[MAX_BTF_SOCK_TYPE];
10519 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
10521 /* tcp6_sock type is not generated in dwarf and hence btf,
10522 * trigger an explicit type generation here.
10524 BTF_TYPE_EMIT(struct tcp6_sock);
10525 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
10526 sk->sk_family == AF_INET6)
10527 return (unsigned long)sk;
10529 return (unsigned long)NULL;
10532 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
10533 .func = bpf_skc_to_tcp6_sock,
10535 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10536 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10537 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
10540 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
10542 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
10543 return (unsigned long)sk;
10545 return (unsigned long)NULL;
10548 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
10549 .func = bpf_skc_to_tcp_sock,
10551 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10552 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10553 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
10556 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
10558 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
10559 * generated if CONFIG_INET=n. Trigger an explicit generation here.
10561 BTF_TYPE_EMIT(struct inet_timewait_sock);
10562 BTF_TYPE_EMIT(struct tcp_timewait_sock);
10565 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
10566 return (unsigned long)sk;
10569 #if IS_BUILTIN(CONFIG_IPV6)
10570 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
10571 return (unsigned long)sk;
10574 return (unsigned long)NULL;
10577 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
10578 .func = bpf_skc_to_tcp_timewait_sock,
10580 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10581 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10582 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
10585 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
10588 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10589 return (unsigned long)sk;
10592 #if IS_BUILTIN(CONFIG_IPV6)
10593 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10594 return (unsigned long)sk;
10597 return (unsigned long)NULL;
10600 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
10601 .func = bpf_skc_to_tcp_request_sock,
10603 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10604 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10605 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
10608 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
10610 /* udp6_sock type is not generated in dwarf and hence btf,
10611 * trigger an explicit type generation here.
10613 BTF_TYPE_EMIT(struct udp6_sock);
10614 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
10615 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
10616 return (unsigned long)sk;
10618 return (unsigned long)NULL;
10621 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
10622 .func = bpf_skc_to_udp6_sock,
10624 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10625 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10626 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
10629 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
10631 return (unsigned long)sock_from_file(file);
10634 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
10635 BTF_ID(struct, socket)
10636 BTF_ID(struct, file)
10638 const struct bpf_func_proto bpf_sock_from_file_proto = {
10639 .func = bpf_sock_from_file,
10641 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10642 .ret_btf_id = &bpf_sock_from_file_btf_ids[0],
10643 .arg1_type = ARG_PTR_TO_BTF_ID,
10644 .arg1_btf_id = &bpf_sock_from_file_btf_ids[1],
10647 static const struct bpf_func_proto *
10648 bpf_sk_base_func_proto(enum bpf_func_id func_id)
10650 const struct bpf_func_proto *func;
10653 case BPF_FUNC_skc_to_tcp6_sock:
10654 func = &bpf_skc_to_tcp6_sock_proto;
10656 case BPF_FUNC_skc_to_tcp_sock:
10657 func = &bpf_skc_to_tcp_sock_proto;
10659 case BPF_FUNC_skc_to_tcp_timewait_sock:
10660 func = &bpf_skc_to_tcp_timewait_sock_proto;
10662 case BPF_FUNC_skc_to_tcp_request_sock:
10663 func = &bpf_skc_to_tcp_request_sock_proto;
10665 case BPF_FUNC_skc_to_udp6_sock:
10666 func = &bpf_skc_to_udp6_sock_proto;
10669 return bpf_base_func_proto(func_id);
10672 if (!perfmon_capable())