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);
1216 int optmem_max = READ_ONCE(sysctl_optmem_max);
1218 /* same check as in sock_kmalloc() */
1219 if (filter_size <= optmem_max &&
1220 atomic_read(&sk->sk_omem_alloc) + filter_size < optmem_max) {
1221 atomic_add(filter_size, &sk->sk_omem_alloc);
1227 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1229 if (!refcount_inc_not_zero(&fp->refcnt))
1232 if (!__sk_filter_charge(sk, fp)) {
1233 sk_filter_release(fp);
1239 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1241 struct sock_filter *old_prog;
1242 struct bpf_prog *old_fp;
1243 int err, new_len, old_len = fp->len;
1244 bool seen_ld_abs = false;
1246 /* We are free to overwrite insns et al right here as it
1247 * won't be used at this point in time anymore internally
1248 * after the migration to the internal BPF instruction
1251 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1252 sizeof(struct bpf_insn));
1254 /* Conversion cannot happen on overlapping memory areas,
1255 * so we need to keep the user BPF around until the 2nd
1256 * pass. At this time, the user BPF is stored in fp->insns.
1258 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1259 GFP_KERNEL | __GFP_NOWARN);
1265 /* 1st pass: calculate the new program length. */
1266 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1271 /* Expand fp for appending the new filter representation. */
1273 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1275 /* The old_fp is still around in case we couldn't
1276 * allocate new memory, so uncharge on that one.
1285 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1286 err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1289 /* 2nd bpf_convert_filter() can fail only if it fails
1290 * to allocate memory, remapping must succeed. Note,
1291 * that at this time old_fp has already been released
1296 fp = bpf_prog_select_runtime(fp, &err);
1306 __bpf_prog_release(fp);
1307 return ERR_PTR(err);
1310 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1311 bpf_aux_classic_check_t trans)
1315 fp->bpf_func = NULL;
1318 err = bpf_check_classic(fp->insns, fp->len);
1320 __bpf_prog_release(fp);
1321 return ERR_PTR(err);
1324 /* There might be additional checks and transformations
1325 * needed on classic filters, f.e. in case of seccomp.
1328 err = trans(fp->insns, fp->len);
1330 __bpf_prog_release(fp);
1331 return ERR_PTR(err);
1335 /* Probe if we can JIT compile the filter and if so, do
1336 * the compilation of the filter.
1338 bpf_jit_compile(fp);
1340 /* JIT compiler couldn't process this filter, so do the
1341 * internal BPF translation for the optimized interpreter.
1344 fp = bpf_migrate_filter(fp);
1350 * bpf_prog_create - create an unattached filter
1351 * @pfp: the unattached filter that is created
1352 * @fprog: the filter program
1354 * Create a filter independent of any socket. We first run some
1355 * sanity checks on it to make sure it does not explode on us later.
1356 * If an error occurs or there is insufficient memory for the filter
1357 * a negative errno code is returned. On success the return is zero.
1359 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1361 unsigned int fsize = bpf_classic_proglen(fprog);
1362 struct bpf_prog *fp;
1364 /* Make sure new filter is there and in the right amounts. */
1365 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1368 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1372 memcpy(fp->insns, fprog->filter, fsize);
1374 fp->len = fprog->len;
1375 /* Since unattached filters are not copied back to user
1376 * space through sk_get_filter(), we do not need to hold
1377 * a copy here, and can spare us the work.
1379 fp->orig_prog = NULL;
1381 /* bpf_prepare_filter() already takes care of freeing
1382 * memory in case something goes wrong.
1384 fp = bpf_prepare_filter(fp, NULL);
1391 EXPORT_SYMBOL_GPL(bpf_prog_create);
1394 * bpf_prog_create_from_user - create an unattached filter from user buffer
1395 * @pfp: the unattached filter that is created
1396 * @fprog: the filter program
1397 * @trans: post-classic verifier transformation handler
1398 * @save_orig: save classic BPF program
1400 * This function effectively does the same as bpf_prog_create(), only
1401 * that it builds up its insns buffer from user space provided buffer.
1402 * It also allows for passing a bpf_aux_classic_check_t handler.
1404 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1405 bpf_aux_classic_check_t trans, bool save_orig)
1407 unsigned int fsize = bpf_classic_proglen(fprog);
1408 struct bpf_prog *fp;
1411 /* Make sure new filter is there and in the right amounts. */
1412 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1415 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1419 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1420 __bpf_prog_free(fp);
1424 fp->len = fprog->len;
1425 fp->orig_prog = NULL;
1428 err = bpf_prog_store_orig_filter(fp, fprog);
1430 __bpf_prog_free(fp);
1435 /* bpf_prepare_filter() already takes care of freeing
1436 * memory in case something goes wrong.
1438 fp = bpf_prepare_filter(fp, trans);
1445 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1447 void bpf_prog_destroy(struct bpf_prog *fp)
1449 __bpf_prog_release(fp);
1451 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1453 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1455 struct sk_filter *fp, *old_fp;
1457 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1463 if (!__sk_filter_charge(sk, fp)) {
1467 refcount_set(&fp->refcnt, 1);
1469 old_fp = rcu_dereference_protected(sk->sk_filter,
1470 lockdep_sock_is_held(sk));
1471 rcu_assign_pointer(sk->sk_filter, fp);
1474 sk_filter_uncharge(sk, old_fp);
1480 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1482 unsigned int fsize = bpf_classic_proglen(fprog);
1483 struct bpf_prog *prog;
1486 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1487 return ERR_PTR(-EPERM);
1489 /* Make sure new filter is there and in the right amounts. */
1490 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1491 return ERR_PTR(-EINVAL);
1493 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1495 return ERR_PTR(-ENOMEM);
1497 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1498 __bpf_prog_free(prog);
1499 return ERR_PTR(-EFAULT);
1502 prog->len = fprog->len;
1504 err = bpf_prog_store_orig_filter(prog, fprog);
1506 __bpf_prog_free(prog);
1507 return ERR_PTR(-ENOMEM);
1510 /* bpf_prepare_filter() already takes care of freeing
1511 * memory in case something goes wrong.
1513 return bpf_prepare_filter(prog, NULL);
1517 * sk_attach_filter - attach a socket filter
1518 * @fprog: the filter program
1519 * @sk: the socket to use
1521 * Attach the user's filter code. We first run some sanity checks on
1522 * it to make sure it does not explode on us later. If an error
1523 * occurs or there is insufficient memory for the filter a negative
1524 * errno code is returned. On success the return is zero.
1526 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1528 struct bpf_prog *prog = __get_filter(fprog, sk);
1532 return PTR_ERR(prog);
1534 err = __sk_attach_prog(prog, sk);
1536 __bpf_prog_release(prog);
1542 EXPORT_SYMBOL_GPL(sk_attach_filter);
1544 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1546 struct bpf_prog *prog = __get_filter(fprog, sk);
1550 return PTR_ERR(prog);
1552 if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max))
1555 err = reuseport_attach_prog(sk, prog);
1558 __bpf_prog_release(prog);
1563 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1565 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1566 return ERR_PTR(-EPERM);
1568 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1571 int sk_attach_bpf(u32 ufd, struct sock *sk)
1573 struct bpf_prog *prog = __get_bpf(ufd, sk);
1577 return PTR_ERR(prog);
1579 err = __sk_attach_prog(prog, sk);
1588 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1590 struct bpf_prog *prog;
1593 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1596 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1597 if (PTR_ERR(prog) == -EINVAL)
1598 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1600 return PTR_ERR(prog);
1602 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1603 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1604 * bpf prog (e.g. sockmap). It depends on the
1605 * limitation imposed by bpf_prog_load().
1606 * Hence, sysctl_optmem_max is not checked.
1608 if ((sk->sk_type != SOCK_STREAM &&
1609 sk->sk_type != SOCK_DGRAM) ||
1610 (sk->sk_protocol != IPPROTO_UDP &&
1611 sk->sk_protocol != IPPROTO_TCP) ||
1612 (sk->sk_family != AF_INET &&
1613 sk->sk_family != AF_INET6)) {
1618 /* BPF_PROG_TYPE_SOCKET_FILTER */
1619 if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max)) {
1625 err = reuseport_attach_prog(sk, prog);
1633 void sk_reuseport_prog_free(struct bpf_prog *prog)
1638 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1641 bpf_prog_destroy(prog);
1644 struct bpf_scratchpad {
1646 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1647 u8 buff[MAX_BPF_STACK];
1651 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1653 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1654 unsigned int write_len)
1656 return skb_ensure_writable(skb, write_len);
1659 static inline int bpf_try_make_writable(struct sk_buff *skb,
1660 unsigned int write_len)
1662 int err = __bpf_try_make_writable(skb, write_len);
1664 bpf_compute_data_pointers(skb);
1668 static int bpf_try_make_head_writable(struct sk_buff *skb)
1670 return bpf_try_make_writable(skb, skb_headlen(skb));
1673 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1675 if (skb_at_tc_ingress(skb))
1676 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1679 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1681 if (skb_at_tc_ingress(skb))
1682 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1685 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1686 const void *, from, u32, len, u64, flags)
1690 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1692 if (unlikely(offset > INT_MAX))
1694 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1697 ptr = skb->data + offset;
1698 if (flags & BPF_F_RECOMPUTE_CSUM)
1699 __skb_postpull_rcsum(skb, ptr, len, offset);
1701 memcpy(ptr, from, len);
1703 if (flags & BPF_F_RECOMPUTE_CSUM)
1704 __skb_postpush_rcsum(skb, ptr, len, offset);
1705 if (flags & BPF_F_INVALIDATE_HASH)
1706 skb_clear_hash(skb);
1711 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1712 .func = bpf_skb_store_bytes,
1714 .ret_type = RET_INTEGER,
1715 .arg1_type = ARG_PTR_TO_CTX,
1716 .arg2_type = ARG_ANYTHING,
1717 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1718 .arg4_type = ARG_CONST_SIZE,
1719 .arg5_type = ARG_ANYTHING,
1722 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1723 void *, to, u32, len)
1727 if (unlikely(offset > INT_MAX))
1730 ptr = skb_header_pointer(skb, offset, len, to);
1734 memcpy(to, ptr, len);
1742 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1743 .func = bpf_skb_load_bytes,
1745 .ret_type = RET_INTEGER,
1746 .arg1_type = ARG_PTR_TO_CTX,
1747 .arg2_type = ARG_ANYTHING,
1748 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1749 .arg4_type = ARG_CONST_SIZE,
1752 BPF_CALL_4(bpf_flow_dissector_load_bytes,
1753 const struct bpf_flow_dissector *, ctx, u32, offset,
1754 void *, to, u32, len)
1758 if (unlikely(offset > 0xffff))
1761 if (unlikely(!ctx->skb))
1764 ptr = skb_header_pointer(ctx->skb, offset, len, to);
1768 memcpy(to, ptr, len);
1776 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1777 .func = bpf_flow_dissector_load_bytes,
1779 .ret_type = RET_INTEGER,
1780 .arg1_type = ARG_PTR_TO_CTX,
1781 .arg2_type = ARG_ANYTHING,
1782 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1783 .arg4_type = ARG_CONST_SIZE,
1786 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1787 u32, offset, void *, to, u32, len, u32, start_header)
1789 u8 *end = skb_tail_pointer(skb);
1792 if (unlikely(offset > 0xffff))
1795 switch (start_header) {
1796 case BPF_HDR_START_MAC:
1797 if (unlikely(!skb_mac_header_was_set(skb)))
1799 start = skb_mac_header(skb);
1801 case BPF_HDR_START_NET:
1802 start = skb_network_header(skb);
1808 ptr = start + offset;
1810 if (likely(ptr + len <= end)) {
1811 memcpy(to, ptr, len);
1820 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1821 .func = bpf_skb_load_bytes_relative,
1823 .ret_type = RET_INTEGER,
1824 .arg1_type = ARG_PTR_TO_CTX,
1825 .arg2_type = ARG_ANYTHING,
1826 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1827 .arg4_type = ARG_CONST_SIZE,
1828 .arg5_type = ARG_ANYTHING,
1831 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1833 /* Idea is the following: should the needed direct read/write
1834 * test fail during runtime, we can pull in more data and redo
1835 * again, since implicitly, we invalidate previous checks here.
1837 * Or, since we know how much we need to make read/writeable,
1838 * this can be done once at the program beginning for direct
1839 * access case. By this we overcome limitations of only current
1840 * headroom being accessible.
1842 return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1845 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1846 .func = bpf_skb_pull_data,
1848 .ret_type = RET_INTEGER,
1849 .arg1_type = ARG_PTR_TO_CTX,
1850 .arg2_type = ARG_ANYTHING,
1853 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1855 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1858 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1859 .func = bpf_sk_fullsock,
1861 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
1862 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
1865 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1866 unsigned int write_len)
1868 return __bpf_try_make_writable(skb, write_len);
1871 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1873 /* Idea is the following: should the needed direct read/write
1874 * test fail during runtime, we can pull in more data and redo
1875 * again, since implicitly, we invalidate previous checks here.
1877 * Or, since we know how much we need to make read/writeable,
1878 * this can be done once at the program beginning for direct
1879 * access case. By this we overcome limitations of only current
1880 * headroom being accessible.
1882 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1885 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1886 .func = sk_skb_pull_data,
1888 .ret_type = RET_INTEGER,
1889 .arg1_type = ARG_PTR_TO_CTX,
1890 .arg2_type = ARG_ANYTHING,
1893 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1894 u64, from, u64, to, u64, flags)
1898 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1900 if (unlikely(offset > 0xffff || offset & 1))
1902 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1905 ptr = (__sum16 *)(skb->data + offset);
1906 switch (flags & BPF_F_HDR_FIELD_MASK) {
1908 if (unlikely(from != 0))
1911 csum_replace_by_diff(ptr, to);
1914 csum_replace2(ptr, from, to);
1917 csum_replace4(ptr, from, to);
1926 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1927 .func = bpf_l3_csum_replace,
1929 .ret_type = RET_INTEGER,
1930 .arg1_type = ARG_PTR_TO_CTX,
1931 .arg2_type = ARG_ANYTHING,
1932 .arg3_type = ARG_ANYTHING,
1933 .arg4_type = ARG_ANYTHING,
1934 .arg5_type = ARG_ANYTHING,
1937 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1938 u64, from, u64, to, u64, flags)
1940 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1941 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1942 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1945 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1946 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1948 if (unlikely(offset > 0xffff || offset & 1))
1950 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1953 ptr = (__sum16 *)(skb->data + offset);
1954 if (is_mmzero && !do_mforce && !*ptr)
1957 switch (flags & BPF_F_HDR_FIELD_MASK) {
1959 if (unlikely(from != 0))
1962 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1965 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1968 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1974 if (is_mmzero && !*ptr)
1975 *ptr = CSUM_MANGLED_0;
1979 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1980 .func = bpf_l4_csum_replace,
1982 .ret_type = RET_INTEGER,
1983 .arg1_type = ARG_PTR_TO_CTX,
1984 .arg2_type = ARG_ANYTHING,
1985 .arg3_type = ARG_ANYTHING,
1986 .arg4_type = ARG_ANYTHING,
1987 .arg5_type = ARG_ANYTHING,
1990 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1991 __be32 *, to, u32, to_size, __wsum, seed)
1993 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1994 u32 diff_size = from_size + to_size;
1997 /* This is quite flexible, some examples:
1999 * from_size == 0, to_size > 0, seed := csum --> pushing data
2000 * from_size > 0, to_size == 0, seed := csum --> pulling data
2001 * from_size > 0, to_size > 0, seed := 0 --> diffing data
2003 * Even for diffing, from_size and to_size don't need to be equal.
2005 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
2006 diff_size > sizeof(sp->diff)))
2009 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
2010 sp->diff[j] = ~from[i];
2011 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
2012 sp->diff[j] = to[i];
2014 return csum_partial(sp->diff, diff_size, seed);
2017 static const struct bpf_func_proto bpf_csum_diff_proto = {
2018 .func = bpf_csum_diff,
2021 .ret_type = RET_INTEGER,
2022 .arg1_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2023 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
2024 .arg3_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2025 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
2026 .arg5_type = ARG_ANYTHING,
2029 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2031 /* The interface is to be used in combination with bpf_csum_diff()
2032 * for direct packet writes. csum rotation for alignment as well
2033 * as emulating csum_sub() can be done from the eBPF program.
2035 if (skb->ip_summed == CHECKSUM_COMPLETE)
2036 return (skb->csum = csum_add(skb->csum, csum));
2041 static const struct bpf_func_proto bpf_csum_update_proto = {
2042 .func = bpf_csum_update,
2044 .ret_type = RET_INTEGER,
2045 .arg1_type = ARG_PTR_TO_CTX,
2046 .arg2_type = ARG_ANYTHING,
2049 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
2051 /* The interface is to be used in combination with bpf_skb_adjust_room()
2052 * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET
2053 * is passed as flags, for example.
2056 case BPF_CSUM_LEVEL_INC:
2057 __skb_incr_checksum_unnecessary(skb);
2059 case BPF_CSUM_LEVEL_DEC:
2060 __skb_decr_checksum_unnecessary(skb);
2062 case BPF_CSUM_LEVEL_RESET:
2063 __skb_reset_checksum_unnecessary(skb);
2065 case BPF_CSUM_LEVEL_QUERY:
2066 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2067 skb->csum_level : -EACCES;
2075 static const struct bpf_func_proto bpf_csum_level_proto = {
2076 .func = bpf_csum_level,
2078 .ret_type = RET_INTEGER,
2079 .arg1_type = ARG_PTR_TO_CTX,
2080 .arg2_type = ARG_ANYTHING,
2083 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2085 return dev_forward_skb_nomtu(dev, skb);
2088 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2089 struct sk_buff *skb)
2091 int ret = ____dev_forward_skb(dev, skb, false);
2095 ret = netif_rx(skb);
2101 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2105 if (dev_xmit_recursion()) {
2106 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2114 dev_xmit_recursion_inc();
2115 ret = dev_queue_xmit(skb);
2116 dev_xmit_recursion_dec();
2121 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2124 unsigned int mlen = skb_network_offset(skb);
2126 if (unlikely(skb->len <= mlen)) {
2132 __skb_pull(skb, mlen);
2133 if (unlikely(!skb->len)) {
2138 /* At ingress, the mac header has already been pulled once.
2139 * At egress, skb_pospull_rcsum has to be done in case that
2140 * the skb is originated from ingress (i.e. a forwarded skb)
2141 * to ensure that rcsum starts at net header.
2143 if (!skb_at_tc_ingress(skb))
2144 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2146 skb_pop_mac_header(skb);
2147 skb_reset_mac_len(skb);
2148 return flags & BPF_F_INGRESS ?
2149 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2152 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2155 /* Verify that a link layer header is carried */
2156 if (unlikely(skb->mac_header >= skb->network_header || skb->len == 0)) {
2161 bpf_push_mac_rcsum(skb);
2162 return flags & BPF_F_INGRESS ?
2163 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2166 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2169 if (dev_is_mac_header_xmit(dev))
2170 return __bpf_redirect_common(skb, dev, flags);
2172 return __bpf_redirect_no_mac(skb, dev, flags);
2175 #if IS_ENABLED(CONFIG_IPV6)
2176 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2177 struct net_device *dev, struct bpf_nh_params *nh)
2179 u32 hh_len = LL_RESERVED_SPACE(dev);
2180 const struct in6_addr *nexthop;
2181 struct dst_entry *dst = NULL;
2182 struct neighbour *neigh;
2184 if (dev_xmit_recursion()) {
2185 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2192 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2193 skb = skb_expand_head(skb, hh_len);
2201 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2202 &ipv6_hdr(skb)->daddr);
2204 nexthop = &nh->ipv6_nh;
2206 neigh = ip_neigh_gw6(dev, nexthop);
2207 if (likely(!IS_ERR(neigh))) {
2210 sock_confirm_neigh(skb, neigh);
2211 dev_xmit_recursion_inc();
2212 ret = neigh_output(neigh, skb, false);
2213 dev_xmit_recursion_dec();
2214 rcu_read_unlock_bh();
2217 rcu_read_unlock_bh();
2219 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2225 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2226 struct bpf_nh_params *nh)
2228 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2229 struct net *net = dev_net(dev);
2230 int err, ret = NET_XMIT_DROP;
2233 struct dst_entry *dst;
2234 struct flowi6 fl6 = {
2235 .flowi6_flags = FLOWI_FLAG_ANYSRC,
2236 .flowi6_mark = skb->mark,
2237 .flowlabel = ip6_flowinfo(ip6h),
2238 .flowi6_oif = dev->ifindex,
2239 .flowi6_proto = ip6h->nexthdr,
2240 .daddr = ip6h->daddr,
2241 .saddr = ip6h->saddr,
2244 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2248 skb_dst_set(skb, dst);
2249 } else if (nh->nh_family != AF_INET6) {
2253 err = bpf_out_neigh_v6(net, skb, dev, nh);
2254 if (unlikely(net_xmit_eval(err)))
2255 dev->stats.tx_errors++;
2257 ret = NET_XMIT_SUCCESS;
2260 dev->stats.tx_errors++;
2266 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2267 struct bpf_nh_params *nh)
2270 return NET_XMIT_DROP;
2272 #endif /* CONFIG_IPV6 */
2274 #if IS_ENABLED(CONFIG_INET)
2275 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2276 struct net_device *dev, struct bpf_nh_params *nh)
2278 u32 hh_len = LL_RESERVED_SPACE(dev);
2279 struct neighbour *neigh;
2280 bool is_v6gw = false;
2282 if (dev_xmit_recursion()) {
2283 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2290 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2291 skb = skb_expand_head(skb, hh_len);
2298 struct dst_entry *dst = skb_dst(skb);
2299 struct rtable *rt = container_of(dst, struct rtable, dst);
2301 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2302 } else if (nh->nh_family == AF_INET6) {
2303 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2305 } else if (nh->nh_family == AF_INET) {
2306 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2308 rcu_read_unlock_bh();
2312 if (likely(!IS_ERR(neigh))) {
2315 sock_confirm_neigh(skb, neigh);
2316 dev_xmit_recursion_inc();
2317 ret = neigh_output(neigh, skb, is_v6gw);
2318 dev_xmit_recursion_dec();
2319 rcu_read_unlock_bh();
2322 rcu_read_unlock_bh();
2328 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2329 struct bpf_nh_params *nh)
2331 const struct iphdr *ip4h = ip_hdr(skb);
2332 struct net *net = dev_net(dev);
2333 int err, ret = NET_XMIT_DROP;
2336 struct flowi4 fl4 = {
2337 .flowi4_flags = FLOWI_FLAG_ANYSRC,
2338 .flowi4_mark = skb->mark,
2339 .flowi4_tos = RT_TOS(ip4h->tos),
2340 .flowi4_oif = dev->ifindex,
2341 .flowi4_proto = ip4h->protocol,
2342 .daddr = ip4h->daddr,
2343 .saddr = ip4h->saddr,
2347 rt = ip_route_output_flow(net, &fl4, NULL);
2350 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2355 skb_dst_set(skb, &rt->dst);
2358 err = bpf_out_neigh_v4(net, skb, dev, nh);
2359 if (unlikely(net_xmit_eval(err)))
2360 dev->stats.tx_errors++;
2362 ret = NET_XMIT_SUCCESS;
2365 dev->stats.tx_errors++;
2371 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2372 struct bpf_nh_params *nh)
2375 return NET_XMIT_DROP;
2377 #endif /* CONFIG_INET */
2379 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2380 struct bpf_nh_params *nh)
2382 struct ethhdr *ethh = eth_hdr(skb);
2384 if (unlikely(skb->mac_header >= skb->network_header))
2386 bpf_push_mac_rcsum(skb);
2387 if (is_multicast_ether_addr(ethh->h_dest))
2390 skb_pull(skb, sizeof(*ethh));
2391 skb_unset_mac_header(skb);
2392 skb_reset_network_header(skb);
2394 if (skb->protocol == htons(ETH_P_IP))
2395 return __bpf_redirect_neigh_v4(skb, dev, nh);
2396 else if (skb->protocol == htons(ETH_P_IPV6))
2397 return __bpf_redirect_neigh_v6(skb, dev, nh);
2403 /* Internal, non-exposed redirect flags. */
2405 BPF_F_NEIGH = (1ULL << 1),
2406 BPF_F_PEER = (1ULL << 2),
2407 BPF_F_NEXTHOP = (1ULL << 3),
2408 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2411 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2413 struct net_device *dev;
2414 struct sk_buff *clone;
2417 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2420 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2424 clone = skb_clone(skb, GFP_ATOMIC);
2425 if (unlikely(!clone))
2428 /* For direct write, we need to keep the invariant that the skbs
2429 * we're dealing with need to be uncloned. Should uncloning fail
2430 * here, we need to free the just generated clone to unclone once
2433 ret = bpf_try_make_head_writable(skb);
2434 if (unlikely(ret)) {
2439 return __bpf_redirect(clone, dev, flags);
2442 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2443 .func = bpf_clone_redirect,
2445 .ret_type = RET_INTEGER,
2446 .arg1_type = ARG_PTR_TO_CTX,
2447 .arg2_type = ARG_ANYTHING,
2448 .arg3_type = ARG_ANYTHING,
2451 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2452 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2454 int skb_do_redirect(struct sk_buff *skb)
2456 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2457 struct net *net = dev_net(skb->dev);
2458 struct net_device *dev;
2459 u32 flags = ri->flags;
2461 dev = dev_get_by_index_rcu(net, ri->tgt_index);
2466 if (flags & BPF_F_PEER) {
2467 const struct net_device_ops *ops = dev->netdev_ops;
2469 if (unlikely(!ops->ndo_get_peer_dev ||
2470 !skb_at_tc_ingress(skb)))
2472 dev = ops->ndo_get_peer_dev(dev);
2473 if (unlikely(!dev ||
2474 !(dev->flags & IFF_UP) ||
2475 net_eq(net, dev_net(dev))))
2480 return flags & BPF_F_NEIGH ?
2481 __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2483 __bpf_redirect(skb, dev, flags);
2489 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2491 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2493 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2497 ri->tgt_index = ifindex;
2499 return TC_ACT_REDIRECT;
2502 static const struct bpf_func_proto bpf_redirect_proto = {
2503 .func = bpf_redirect,
2505 .ret_type = RET_INTEGER,
2506 .arg1_type = ARG_ANYTHING,
2507 .arg2_type = ARG_ANYTHING,
2510 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2512 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2514 if (unlikely(flags))
2517 ri->flags = BPF_F_PEER;
2518 ri->tgt_index = ifindex;
2520 return TC_ACT_REDIRECT;
2523 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2524 .func = bpf_redirect_peer,
2526 .ret_type = RET_INTEGER,
2527 .arg1_type = ARG_ANYTHING,
2528 .arg2_type = ARG_ANYTHING,
2531 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2532 int, plen, u64, flags)
2534 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2536 if (unlikely((plen && plen < sizeof(*params)) || flags))
2539 ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2540 ri->tgt_index = ifindex;
2542 BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2544 memcpy(&ri->nh, params, sizeof(ri->nh));
2546 return TC_ACT_REDIRECT;
2549 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2550 .func = bpf_redirect_neigh,
2552 .ret_type = RET_INTEGER,
2553 .arg1_type = ARG_ANYTHING,
2554 .arg2_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2555 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
2556 .arg4_type = ARG_ANYTHING,
2559 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2561 msg->apply_bytes = bytes;
2565 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2566 .func = bpf_msg_apply_bytes,
2568 .ret_type = RET_INTEGER,
2569 .arg1_type = ARG_PTR_TO_CTX,
2570 .arg2_type = ARG_ANYTHING,
2573 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2575 msg->cork_bytes = bytes;
2579 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2580 .func = bpf_msg_cork_bytes,
2582 .ret_type = RET_INTEGER,
2583 .arg1_type = ARG_PTR_TO_CTX,
2584 .arg2_type = ARG_ANYTHING,
2587 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2588 u32, end, u64, flags)
2590 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2591 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2592 struct scatterlist *sge;
2593 u8 *raw, *to, *from;
2596 if (unlikely(flags || end <= start))
2599 /* First find the starting scatterlist element */
2603 len = sk_msg_elem(msg, i)->length;
2604 if (start < offset + len)
2606 sk_msg_iter_var_next(i);
2607 } while (i != msg->sg.end);
2609 if (unlikely(start >= offset + len))
2613 /* The start may point into the sg element so we need to also
2614 * account for the headroom.
2616 bytes_sg_total = start - offset + bytes;
2617 if (!test_bit(i, &msg->sg.copy) && bytes_sg_total <= len)
2620 /* At this point we need to linearize multiple scatterlist
2621 * elements or a single shared page. Either way we need to
2622 * copy into a linear buffer exclusively owned by BPF. Then
2623 * place the buffer in the scatterlist and fixup the original
2624 * entries by removing the entries now in the linear buffer
2625 * and shifting the remaining entries. For now we do not try
2626 * to copy partial entries to avoid complexity of running out
2627 * of sg_entry slots. The downside is reading a single byte
2628 * will copy the entire sg entry.
2631 copy += sk_msg_elem(msg, i)->length;
2632 sk_msg_iter_var_next(i);
2633 if (bytes_sg_total <= copy)
2635 } while (i != msg->sg.end);
2638 if (unlikely(bytes_sg_total > copy))
2641 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2643 if (unlikely(!page))
2646 raw = page_address(page);
2649 sge = sk_msg_elem(msg, i);
2650 from = sg_virt(sge);
2654 memcpy(to, from, len);
2657 put_page(sg_page(sge));
2659 sk_msg_iter_var_next(i);
2660 } while (i != last_sge);
2662 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2664 /* To repair sg ring we need to shift entries. If we only
2665 * had a single entry though we can just replace it and
2666 * be done. Otherwise walk the ring and shift the entries.
2668 WARN_ON_ONCE(last_sge == first_sge);
2669 shift = last_sge > first_sge ?
2670 last_sge - first_sge - 1 :
2671 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2676 sk_msg_iter_var_next(i);
2680 if (i + shift >= NR_MSG_FRAG_IDS)
2681 move_from = i + shift - NR_MSG_FRAG_IDS;
2683 move_from = i + shift;
2684 if (move_from == msg->sg.end)
2687 msg->sg.data[i] = msg->sg.data[move_from];
2688 msg->sg.data[move_from].length = 0;
2689 msg->sg.data[move_from].page_link = 0;
2690 msg->sg.data[move_from].offset = 0;
2691 sk_msg_iter_var_next(i);
2694 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2695 msg->sg.end - shift + NR_MSG_FRAG_IDS :
2696 msg->sg.end - shift;
2698 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2699 msg->data_end = msg->data + bytes;
2703 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2704 .func = bpf_msg_pull_data,
2706 .ret_type = RET_INTEGER,
2707 .arg1_type = ARG_PTR_TO_CTX,
2708 .arg2_type = ARG_ANYTHING,
2709 .arg3_type = ARG_ANYTHING,
2710 .arg4_type = ARG_ANYTHING,
2713 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2714 u32, len, u64, flags)
2716 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2717 u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2718 u8 *raw, *to, *from;
2721 if (unlikely(flags))
2724 if (unlikely(len == 0))
2727 /* First find the starting scatterlist element */
2731 l = sk_msg_elem(msg, i)->length;
2733 if (start < offset + l)
2735 sk_msg_iter_var_next(i);
2736 } while (i != msg->sg.end);
2738 if (start >= offset + l)
2741 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2743 /* If no space available will fallback to copy, we need at
2744 * least one scatterlist elem available to push data into
2745 * when start aligns to the beginning of an element or two
2746 * when it falls inside an element. We handle the start equals
2747 * offset case because its the common case for inserting a
2750 if (!space || (space == 1 && start != offset))
2751 copy = msg->sg.data[i].length;
2753 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2754 get_order(copy + len));
2755 if (unlikely(!page))
2761 raw = page_address(page);
2763 psge = sk_msg_elem(msg, i);
2764 front = start - offset;
2765 back = psge->length - front;
2766 from = sg_virt(psge);
2769 memcpy(raw, from, front);
2773 to = raw + front + len;
2775 memcpy(to, from, back);
2778 put_page(sg_page(psge));
2779 } else if (start - offset) {
2780 psge = sk_msg_elem(msg, i);
2781 rsge = sk_msg_elem_cpy(msg, i);
2783 psge->length = start - offset;
2784 rsge.length -= psge->length;
2785 rsge.offset += start;
2787 sk_msg_iter_var_next(i);
2788 sg_unmark_end(psge);
2789 sg_unmark_end(&rsge);
2790 sk_msg_iter_next(msg, end);
2793 /* Slot(s) to place newly allocated data */
2796 /* Shift one or two slots as needed */
2798 sge = sk_msg_elem_cpy(msg, i);
2800 sk_msg_iter_var_next(i);
2801 sg_unmark_end(&sge);
2802 sk_msg_iter_next(msg, end);
2804 nsge = sk_msg_elem_cpy(msg, i);
2806 sk_msg_iter_var_next(i);
2807 nnsge = sk_msg_elem_cpy(msg, i);
2810 while (i != msg->sg.end) {
2811 msg->sg.data[i] = sge;
2813 sk_msg_iter_var_next(i);
2816 nnsge = sk_msg_elem_cpy(msg, i);
2818 nsge = sk_msg_elem_cpy(msg, i);
2823 /* Place newly allocated data buffer */
2824 sk_mem_charge(msg->sk, len);
2825 msg->sg.size += len;
2826 __clear_bit(new, &msg->sg.copy);
2827 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2829 get_page(sg_page(&rsge));
2830 sk_msg_iter_var_next(new);
2831 msg->sg.data[new] = rsge;
2834 sk_msg_compute_data_pointers(msg);
2838 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2839 .func = bpf_msg_push_data,
2841 .ret_type = RET_INTEGER,
2842 .arg1_type = ARG_PTR_TO_CTX,
2843 .arg2_type = ARG_ANYTHING,
2844 .arg3_type = ARG_ANYTHING,
2845 .arg4_type = ARG_ANYTHING,
2848 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2854 sk_msg_iter_var_next(i);
2855 msg->sg.data[prev] = msg->sg.data[i];
2856 } while (i != msg->sg.end);
2858 sk_msg_iter_prev(msg, end);
2861 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2863 struct scatterlist tmp, sge;
2865 sk_msg_iter_next(msg, end);
2866 sge = sk_msg_elem_cpy(msg, i);
2867 sk_msg_iter_var_next(i);
2868 tmp = sk_msg_elem_cpy(msg, i);
2870 while (i != msg->sg.end) {
2871 msg->sg.data[i] = sge;
2872 sk_msg_iter_var_next(i);
2874 tmp = sk_msg_elem_cpy(msg, i);
2878 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2879 u32, len, u64, flags)
2881 u32 i = 0, l = 0, space, offset = 0;
2882 u64 last = start + len;
2885 if (unlikely(flags))
2888 /* First find the starting scatterlist element */
2892 l = sk_msg_elem(msg, i)->length;
2894 if (start < offset + l)
2896 sk_msg_iter_var_next(i);
2897 } while (i != msg->sg.end);
2899 /* Bounds checks: start and pop must be inside message */
2900 if (start >= offset + l || last >= msg->sg.size)
2903 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2906 /* --------------| offset
2907 * -| start |-------- len -------|
2909 * |----- a ----|-------- pop -------|----- b ----|
2910 * |______________________________________________| length
2913 * a: region at front of scatter element to save
2914 * b: region at back of scatter element to save when length > A + pop
2915 * pop: region to pop from element, same as input 'pop' here will be
2916 * decremented below per iteration.
2918 * Two top-level cases to handle when start != offset, first B is non
2919 * zero and second B is zero corresponding to when a pop includes more
2922 * Then if B is non-zero AND there is no space allocate space and
2923 * compact A, B regions into page. If there is space shift ring to
2924 * the rigth free'ing the next element in ring to place B, leaving
2925 * A untouched except to reduce length.
2927 if (start != offset) {
2928 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2930 int b = sge->length - pop - a;
2932 sk_msg_iter_var_next(i);
2934 if (pop < sge->length - a) {
2937 sk_msg_shift_right(msg, i);
2938 nsge = sk_msg_elem(msg, i);
2939 get_page(sg_page(sge));
2942 b, sge->offset + pop + a);
2944 struct page *page, *orig;
2947 page = alloc_pages(__GFP_NOWARN |
2948 __GFP_COMP | GFP_ATOMIC,
2950 if (unlikely(!page))
2954 orig = sg_page(sge);
2955 from = sg_virt(sge);
2956 to = page_address(page);
2957 memcpy(to, from, a);
2958 memcpy(to + a, from + a + pop, b);
2959 sg_set_page(sge, page, a + b, 0);
2963 } else if (pop >= sge->length - a) {
2964 pop -= (sge->length - a);
2969 /* From above the current layout _must_ be as follows,
2974 * |---- pop ---|---------------- b ------------|
2975 * |____________________________________________| length
2977 * Offset and start of the current msg elem are equal because in the
2978 * previous case we handled offset != start and either consumed the
2979 * entire element and advanced to the next element OR pop == 0.
2981 * Two cases to handle here are first pop is less than the length
2982 * leaving some remainder b above. Simply adjust the element's layout
2983 * in this case. Or pop >= length of the element so that b = 0. In this
2984 * case advance to next element decrementing pop.
2987 struct scatterlist *sge = sk_msg_elem(msg, i);
2989 if (pop < sge->length) {
2995 sk_msg_shift_left(msg, i);
2997 sk_msg_iter_var_next(i);
3000 sk_mem_uncharge(msg->sk, len - pop);
3001 msg->sg.size -= (len - pop);
3002 sk_msg_compute_data_pointers(msg);
3006 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
3007 .func = bpf_msg_pop_data,
3009 .ret_type = RET_INTEGER,
3010 .arg1_type = ARG_PTR_TO_CTX,
3011 .arg2_type = ARG_ANYTHING,
3012 .arg3_type = ARG_ANYTHING,
3013 .arg4_type = ARG_ANYTHING,
3016 #ifdef CONFIG_CGROUP_NET_CLASSID
3017 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3019 return __task_get_classid(current);
3022 static const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3023 .func = bpf_get_cgroup_classid_curr,
3025 .ret_type = RET_INTEGER,
3028 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3030 struct sock *sk = skb_to_full_sk(skb);
3032 if (!sk || !sk_fullsock(sk))
3035 return sock_cgroup_classid(&sk->sk_cgrp_data);
3038 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3039 .func = bpf_skb_cgroup_classid,
3041 .ret_type = RET_INTEGER,
3042 .arg1_type = ARG_PTR_TO_CTX,
3046 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3048 return task_get_classid(skb);
3051 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3052 .func = bpf_get_cgroup_classid,
3054 .ret_type = RET_INTEGER,
3055 .arg1_type = ARG_PTR_TO_CTX,
3058 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3060 return dst_tclassid(skb);
3063 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3064 .func = bpf_get_route_realm,
3066 .ret_type = RET_INTEGER,
3067 .arg1_type = ARG_PTR_TO_CTX,
3070 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3072 /* If skb_clear_hash() was called due to mangling, we can
3073 * trigger SW recalculation here. Later access to hash
3074 * can then use the inline skb->hash via context directly
3075 * instead of calling this helper again.
3077 return skb_get_hash(skb);
3080 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3081 .func = bpf_get_hash_recalc,
3083 .ret_type = RET_INTEGER,
3084 .arg1_type = ARG_PTR_TO_CTX,
3087 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3089 /* After all direct packet write, this can be used once for
3090 * triggering a lazy recalc on next skb_get_hash() invocation.
3092 skb_clear_hash(skb);
3096 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3097 .func = bpf_set_hash_invalid,
3099 .ret_type = RET_INTEGER,
3100 .arg1_type = ARG_PTR_TO_CTX,
3103 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3105 /* Set user specified hash as L4(+), so that it gets returned
3106 * on skb_get_hash() call unless BPF prog later on triggers a
3109 __skb_set_sw_hash(skb, hash, true);
3113 static const struct bpf_func_proto bpf_set_hash_proto = {
3114 .func = bpf_set_hash,
3116 .ret_type = RET_INTEGER,
3117 .arg1_type = ARG_PTR_TO_CTX,
3118 .arg2_type = ARG_ANYTHING,
3121 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3126 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3127 vlan_proto != htons(ETH_P_8021AD)))
3128 vlan_proto = htons(ETH_P_8021Q);
3130 bpf_push_mac_rcsum(skb);
3131 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3132 bpf_pull_mac_rcsum(skb);
3134 bpf_compute_data_pointers(skb);
3138 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3139 .func = bpf_skb_vlan_push,
3141 .ret_type = RET_INTEGER,
3142 .arg1_type = ARG_PTR_TO_CTX,
3143 .arg2_type = ARG_ANYTHING,
3144 .arg3_type = ARG_ANYTHING,
3147 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3151 bpf_push_mac_rcsum(skb);
3152 ret = skb_vlan_pop(skb);
3153 bpf_pull_mac_rcsum(skb);
3155 bpf_compute_data_pointers(skb);
3159 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3160 .func = bpf_skb_vlan_pop,
3162 .ret_type = RET_INTEGER,
3163 .arg1_type = ARG_PTR_TO_CTX,
3166 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3168 /* Caller already did skb_cow() with len as headroom,
3169 * so no need to do it here.
3172 memmove(skb->data, skb->data + len, off);
3173 memset(skb->data + off, 0, len);
3175 /* No skb_postpush_rcsum(skb, skb->data + off, len)
3176 * needed here as it does not change the skb->csum
3177 * result for checksum complete when summing over
3183 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3187 /* skb_ensure_writable() is not needed here, as we're
3188 * already working on an uncloned skb.
3190 if (unlikely(!pskb_may_pull(skb, off + len)))
3193 old_data = skb->data;
3194 __skb_pull(skb, len);
3195 skb_postpull_rcsum(skb, old_data + off, len);
3196 memmove(skb->data, old_data, off);
3201 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3203 bool trans_same = skb->transport_header == skb->network_header;
3206 /* There's no need for __skb_push()/__skb_pull() pair to
3207 * get to the start of the mac header as we're guaranteed
3208 * to always start from here under eBPF.
3210 ret = bpf_skb_generic_push(skb, off, len);
3212 skb->mac_header -= len;
3213 skb->network_header -= len;
3215 skb->transport_header = skb->network_header;
3221 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3223 bool trans_same = skb->transport_header == skb->network_header;
3226 /* Same here, __skb_push()/__skb_pull() pair not needed. */
3227 ret = bpf_skb_generic_pop(skb, off, len);
3229 skb->mac_header += len;
3230 skb->network_header += len;
3232 skb->transport_header = skb->network_header;
3238 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3240 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3241 u32 off = skb_mac_header_len(skb);
3244 ret = skb_cow(skb, len_diff);
3245 if (unlikely(ret < 0))
3248 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3249 if (unlikely(ret < 0))
3252 if (skb_is_gso(skb)) {
3253 struct skb_shared_info *shinfo = skb_shinfo(skb);
3255 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */
3256 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3257 shinfo->gso_type &= ~SKB_GSO_TCPV4;
3258 shinfo->gso_type |= SKB_GSO_TCPV6;
3262 skb->protocol = htons(ETH_P_IPV6);
3263 skb_clear_hash(skb);
3268 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3270 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3271 u32 off = skb_mac_header_len(skb);
3274 ret = skb_unclone(skb, GFP_ATOMIC);
3275 if (unlikely(ret < 0))
3278 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3279 if (unlikely(ret < 0))
3282 if (skb_is_gso(skb)) {
3283 struct skb_shared_info *shinfo = skb_shinfo(skb);
3285 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */
3286 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3287 shinfo->gso_type &= ~SKB_GSO_TCPV6;
3288 shinfo->gso_type |= SKB_GSO_TCPV4;
3292 skb->protocol = htons(ETH_P_IP);
3293 skb_clear_hash(skb);
3298 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3300 __be16 from_proto = skb->protocol;
3302 if (from_proto == htons(ETH_P_IP) &&
3303 to_proto == htons(ETH_P_IPV6))
3304 return bpf_skb_proto_4_to_6(skb);
3306 if (from_proto == htons(ETH_P_IPV6) &&
3307 to_proto == htons(ETH_P_IP))
3308 return bpf_skb_proto_6_to_4(skb);
3313 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3318 if (unlikely(flags))
3321 /* General idea is that this helper does the basic groundwork
3322 * needed for changing the protocol, and eBPF program fills the
3323 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3324 * and other helpers, rather than passing a raw buffer here.
3326 * The rationale is to keep this minimal and without a need to
3327 * deal with raw packet data. F.e. even if we would pass buffers
3328 * here, the program still needs to call the bpf_lX_csum_replace()
3329 * helpers anyway. Plus, this way we keep also separation of
3330 * concerns, since f.e. bpf_skb_store_bytes() should only take
3333 * Currently, additional options and extension header space are
3334 * not supported, but flags register is reserved so we can adapt
3335 * that. For offloads, we mark packet as dodgy, so that headers
3336 * need to be verified first.
3338 ret = bpf_skb_proto_xlat(skb, proto);
3339 bpf_compute_data_pointers(skb);
3343 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3344 .func = bpf_skb_change_proto,
3346 .ret_type = RET_INTEGER,
3347 .arg1_type = ARG_PTR_TO_CTX,
3348 .arg2_type = ARG_ANYTHING,
3349 .arg3_type = ARG_ANYTHING,
3352 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3354 /* We only allow a restricted subset to be changed for now. */
3355 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3356 !skb_pkt_type_ok(pkt_type)))
3359 skb->pkt_type = pkt_type;
3363 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3364 .func = bpf_skb_change_type,
3366 .ret_type = RET_INTEGER,
3367 .arg1_type = ARG_PTR_TO_CTX,
3368 .arg2_type = ARG_ANYTHING,
3371 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3373 switch (skb->protocol) {
3374 case htons(ETH_P_IP):
3375 return sizeof(struct iphdr);
3376 case htons(ETH_P_IPV6):
3377 return sizeof(struct ipv6hdr);
3383 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3384 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3386 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \
3387 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3388 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3389 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3390 BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \
3391 BPF_F_ADJ_ROOM_ENCAP_L2( \
3392 BPF_ADJ_ROOM_ENCAP_L2_MASK))
3394 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3397 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3398 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3399 u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3400 unsigned int gso_type = SKB_GSO_DODGY;
3403 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3404 /* udp gso_size delineates datagrams, only allow if fixed */
3405 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3406 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3410 ret = skb_cow_head(skb, len_diff);
3411 if (unlikely(ret < 0))
3415 if (skb->protocol != htons(ETH_P_IP) &&
3416 skb->protocol != htons(ETH_P_IPV6))
3419 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3420 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3423 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3424 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3427 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH &&
3428 inner_mac_len < ETH_HLEN)
3431 if (skb->encapsulation)
3434 mac_len = skb->network_header - skb->mac_header;
3435 inner_net = skb->network_header;
3436 if (inner_mac_len > len_diff)
3438 inner_trans = skb->transport_header;
3441 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3442 if (unlikely(ret < 0))
3446 skb->inner_mac_header = inner_net - inner_mac_len;
3447 skb->inner_network_header = inner_net;
3448 skb->inner_transport_header = inner_trans;
3450 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH)
3451 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
3453 skb_set_inner_protocol(skb, skb->protocol);
3455 skb->encapsulation = 1;
3456 skb_set_network_header(skb, mac_len);
3458 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3459 gso_type |= SKB_GSO_UDP_TUNNEL;
3460 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3461 gso_type |= SKB_GSO_GRE;
3462 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3463 gso_type |= SKB_GSO_IPXIP6;
3464 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3465 gso_type |= SKB_GSO_IPXIP4;
3467 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3468 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3469 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3470 sizeof(struct ipv6hdr) :
3471 sizeof(struct iphdr);
3473 skb_set_transport_header(skb, mac_len + nh_len);
3476 /* Match skb->protocol to new outer l3 protocol */
3477 if (skb->protocol == htons(ETH_P_IP) &&
3478 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3479 skb->protocol = htons(ETH_P_IPV6);
3480 else if (skb->protocol == htons(ETH_P_IPV6) &&
3481 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3482 skb->protocol = htons(ETH_P_IP);
3485 if (skb_is_gso(skb)) {
3486 struct skb_shared_info *shinfo = skb_shinfo(skb);
3488 /* Due to header grow, MSS needs to be downgraded. */
3489 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3490 skb_decrease_gso_size(shinfo, len_diff);
3492 /* Header must be checked, and gso_segs recomputed. */
3493 shinfo->gso_type |= gso_type;
3494 shinfo->gso_segs = 0;
3500 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3505 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3506 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3509 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3510 /* udp gso_size delineates datagrams, only allow if fixed */
3511 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3512 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3516 ret = skb_unclone(skb, GFP_ATOMIC);
3517 if (unlikely(ret < 0))
3520 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3521 if (unlikely(ret < 0))
3524 if (skb_is_gso(skb)) {
3525 struct skb_shared_info *shinfo = skb_shinfo(skb);
3527 /* Due to header shrink, MSS can be upgraded. */
3528 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3529 skb_increase_gso_size(shinfo, len_diff);
3531 /* Header must be checked, and gso_segs recomputed. */
3532 shinfo->gso_type |= SKB_GSO_DODGY;
3533 shinfo->gso_segs = 0;
3539 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3541 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3542 u32, mode, u64, flags)
3544 u32 len_diff_abs = abs(len_diff);
3545 bool shrink = len_diff < 0;
3548 if (unlikely(flags || mode))
3550 if (unlikely(len_diff_abs > 0xfffU))
3554 ret = skb_cow(skb, len_diff);
3555 if (unlikely(ret < 0))
3557 __skb_push(skb, len_diff_abs);
3558 memset(skb->data, 0, len_diff_abs);
3560 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3562 __skb_pull(skb, len_diff_abs);
3564 if (tls_sw_has_ctx_rx(skb->sk)) {
3565 struct strp_msg *rxm = strp_msg(skb);
3567 rxm->full_len += len_diff;
3572 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3573 .func = sk_skb_adjust_room,
3575 .ret_type = RET_INTEGER,
3576 .arg1_type = ARG_PTR_TO_CTX,
3577 .arg2_type = ARG_ANYTHING,
3578 .arg3_type = ARG_ANYTHING,
3579 .arg4_type = ARG_ANYTHING,
3582 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3583 u32, mode, u64, flags)
3585 u32 len_cur, len_diff_abs = abs(len_diff);
3586 u32 len_min = bpf_skb_net_base_len(skb);
3587 u32 len_max = BPF_SKB_MAX_LEN;
3588 __be16 proto = skb->protocol;
3589 bool shrink = len_diff < 0;
3593 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3594 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3596 if (unlikely(len_diff_abs > 0xfffU))
3598 if (unlikely(proto != htons(ETH_P_IP) &&
3599 proto != htons(ETH_P_IPV6)))
3602 off = skb_mac_header_len(skb);
3604 case BPF_ADJ_ROOM_NET:
3605 off += bpf_skb_net_base_len(skb);
3607 case BPF_ADJ_ROOM_MAC:
3613 len_cur = skb->len - skb_network_offset(skb);
3614 if ((shrink && (len_diff_abs >= len_cur ||
3615 len_cur - len_diff_abs < len_min)) ||
3616 (!shrink && (skb->len + len_diff_abs > len_max &&
3620 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3621 bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3622 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3623 __skb_reset_checksum_unnecessary(skb);
3625 bpf_compute_data_pointers(skb);
3629 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3630 .func = bpf_skb_adjust_room,
3632 .ret_type = RET_INTEGER,
3633 .arg1_type = ARG_PTR_TO_CTX,
3634 .arg2_type = ARG_ANYTHING,
3635 .arg3_type = ARG_ANYTHING,
3636 .arg4_type = ARG_ANYTHING,
3639 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3641 u32 min_len = skb_network_offset(skb);
3643 if (skb_transport_header_was_set(skb))
3644 min_len = skb_transport_offset(skb);
3645 if (skb->ip_summed == CHECKSUM_PARTIAL)
3646 min_len = skb_checksum_start_offset(skb) +
3647 skb->csum_offset + sizeof(__sum16);
3651 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3653 unsigned int old_len = skb->len;
3656 ret = __skb_grow_rcsum(skb, new_len);
3658 memset(skb->data + old_len, 0, new_len - old_len);
3662 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3664 return __skb_trim_rcsum(skb, new_len);
3667 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3670 u32 max_len = BPF_SKB_MAX_LEN;
3671 u32 min_len = __bpf_skb_min_len(skb);
3674 if (unlikely(flags || new_len > max_len || new_len < min_len))
3676 if (skb->encapsulation)
3679 /* The basic idea of this helper is that it's performing the
3680 * needed work to either grow or trim an skb, and eBPF program
3681 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3682 * bpf_lX_csum_replace() and others rather than passing a raw
3683 * buffer here. This one is a slow path helper and intended
3684 * for replies with control messages.
3686 * Like in bpf_skb_change_proto(), we want to keep this rather
3687 * minimal and without protocol specifics so that we are able
3688 * to separate concerns as in bpf_skb_store_bytes() should only
3689 * be the one responsible for writing buffers.
3691 * It's really expected to be a slow path operation here for
3692 * control message replies, so we're implicitly linearizing,
3693 * uncloning and drop offloads from the skb by this.
3695 ret = __bpf_try_make_writable(skb, skb->len);
3697 if (new_len > skb->len)
3698 ret = bpf_skb_grow_rcsum(skb, new_len);
3699 else if (new_len < skb->len)
3700 ret = bpf_skb_trim_rcsum(skb, new_len);
3701 if (!ret && skb_is_gso(skb))
3707 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3710 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3712 bpf_compute_data_pointers(skb);
3716 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3717 .func = bpf_skb_change_tail,
3719 .ret_type = RET_INTEGER,
3720 .arg1_type = ARG_PTR_TO_CTX,
3721 .arg2_type = ARG_ANYTHING,
3722 .arg3_type = ARG_ANYTHING,
3725 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3728 return __bpf_skb_change_tail(skb, new_len, flags);
3731 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3732 .func = sk_skb_change_tail,
3734 .ret_type = RET_INTEGER,
3735 .arg1_type = ARG_PTR_TO_CTX,
3736 .arg2_type = ARG_ANYTHING,
3737 .arg3_type = ARG_ANYTHING,
3740 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3743 u32 max_len = BPF_SKB_MAX_LEN;
3744 u32 new_len = skb->len + head_room;
3747 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3748 new_len < skb->len))
3751 ret = skb_cow(skb, head_room);
3753 /* Idea for this helper is that we currently only
3754 * allow to expand on mac header. This means that
3755 * skb->protocol network header, etc, stay as is.
3756 * Compared to bpf_skb_change_tail(), we're more
3757 * flexible due to not needing to linearize or
3758 * reset GSO. Intention for this helper is to be
3759 * used by an L3 skb that needs to push mac header
3760 * for redirection into L2 device.
3762 __skb_push(skb, head_room);
3763 memset(skb->data, 0, head_room);
3764 skb_reset_mac_header(skb);
3765 skb_reset_mac_len(skb);
3771 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3774 int ret = __bpf_skb_change_head(skb, head_room, flags);
3776 bpf_compute_data_pointers(skb);
3780 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3781 .func = bpf_skb_change_head,
3783 .ret_type = RET_INTEGER,
3784 .arg1_type = ARG_PTR_TO_CTX,
3785 .arg2_type = ARG_ANYTHING,
3786 .arg3_type = ARG_ANYTHING,
3789 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3792 return __bpf_skb_change_head(skb, head_room, flags);
3795 static const struct bpf_func_proto sk_skb_change_head_proto = {
3796 .func = sk_skb_change_head,
3798 .ret_type = RET_INTEGER,
3799 .arg1_type = ARG_PTR_TO_CTX,
3800 .arg2_type = ARG_ANYTHING,
3801 .arg3_type = ARG_ANYTHING,
3803 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3805 return xdp_data_meta_unsupported(xdp) ? 0 :
3806 xdp->data - xdp->data_meta;
3809 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3811 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3812 unsigned long metalen = xdp_get_metalen(xdp);
3813 void *data_start = xdp_frame_end + metalen;
3814 void *data = xdp->data + offset;
3816 if (unlikely(data < data_start ||
3817 data > xdp->data_end - ETH_HLEN))
3821 memmove(xdp->data_meta + offset,
3822 xdp->data_meta, metalen);
3823 xdp->data_meta += offset;
3829 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3830 .func = bpf_xdp_adjust_head,
3832 .ret_type = RET_INTEGER,
3833 .arg1_type = ARG_PTR_TO_CTX,
3834 .arg2_type = ARG_ANYTHING,
3837 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3839 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
3840 void *data_end = xdp->data_end + offset;
3842 /* Notice that xdp_data_hard_end have reserved some tailroom */
3843 if (unlikely(data_end > data_hard_end))
3846 if (unlikely(data_end < xdp->data + ETH_HLEN))
3849 /* Clear memory area on grow, can contain uninit kernel memory */
3851 memset(xdp->data_end, 0, offset);
3853 xdp->data_end = data_end;
3858 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3859 .func = bpf_xdp_adjust_tail,
3861 .ret_type = RET_INTEGER,
3862 .arg1_type = ARG_PTR_TO_CTX,
3863 .arg2_type = ARG_ANYTHING,
3866 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3868 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3869 void *meta = xdp->data_meta + offset;
3870 unsigned long metalen = xdp->data - meta;
3872 if (xdp_data_meta_unsupported(xdp))
3874 if (unlikely(meta < xdp_frame_end ||
3877 if (unlikely(xdp_metalen_invalid(metalen)))
3880 xdp->data_meta = meta;
3885 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3886 .func = bpf_xdp_adjust_meta,
3888 .ret_type = RET_INTEGER,
3889 .arg1_type = ARG_PTR_TO_CTX,
3890 .arg2_type = ARG_ANYTHING,
3893 /* XDP_REDIRECT works by a three-step process, implemented in the functions
3896 * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target
3897 * of the redirect and store it (along with some other metadata) in a per-CPU
3898 * struct bpf_redirect_info.
3900 * 2. When the program returns the XDP_REDIRECT return code, the driver will
3901 * call xdp_do_redirect() which will use the information in struct
3902 * bpf_redirect_info to actually enqueue the frame into a map type-specific
3903 * bulk queue structure.
3905 * 3. Before exiting its NAPI poll loop, the driver will call xdp_do_flush(),
3906 * which will flush all the different bulk queues, thus completing the
3909 * Pointers to the map entries will be kept around for this whole sequence of
3910 * steps, protected by RCU. However, there is no top-level rcu_read_lock() in
3911 * the core code; instead, the RCU protection relies on everything happening
3912 * inside a single NAPI poll sequence, which means it's between a pair of calls
3913 * to local_bh_disable()/local_bh_enable().
3915 * The map entries are marked as __rcu and the map code makes sure to
3916 * dereference those pointers with rcu_dereference_check() in a way that works
3917 * for both sections that to hold an rcu_read_lock() and sections that are
3918 * called from NAPI without a separate rcu_read_lock(). The code below does not
3919 * use RCU annotations, but relies on those in the map code.
3921 void xdp_do_flush(void)
3927 EXPORT_SYMBOL_GPL(xdp_do_flush);
3929 void bpf_clear_redirect_map(struct bpf_map *map)
3931 struct bpf_redirect_info *ri;
3934 for_each_possible_cpu(cpu) {
3935 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3936 /* Avoid polluting remote cacheline due to writes if
3937 * not needed. Once we pass this test, we need the
3938 * cmpxchg() to make sure it hasn't been changed in
3939 * the meantime by remote CPU.
3941 if (unlikely(READ_ONCE(ri->map) == map))
3942 cmpxchg(&ri->map, map, NULL);
3946 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
3947 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key);
3949 u32 xdp_master_redirect(struct xdp_buff *xdp)
3951 struct net_device *master, *slave;
3952 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3954 master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev);
3955 slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp);
3956 if (slave && slave != xdp->rxq->dev) {
3957 /* The target device is different from the receiving device, so
3958 * redirect it to the new device.
3959 * Using XDP_REDIRECT gets the correct behaviour from XDP enabled
3960 * drivers to unmap the packet from their rx ring.
3962 ri->tgt_index = slave->ifindex;
3963 ri->map_id = INT_MAX;
3964 ri->map_type = BPF_MAP_TYPE_UNSPEC;
3965 return XDP_REDIRECT;
3969 EXPORT_SYMBOL_GPL(xdp_master_redirect);
3971 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3972 struct bpf_prog *xdp_prog)
3974 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3975 enum bpf_map_type map_type = ri->map_type;
3976 void *fwd = ri->tgt_value;
3977 u32 map_id = ri->map_id;
3978 struct bpf_map *map;
3981 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
3982 ri->map_type = BPF_MAP_TYPE_UNSPEC;
3985 case BPF_MAP_TYPE_DEVMAP:
3987 case BPF_MAP_TYPE_DEVMAP_HASH:
3988 map = READ_ONCE(ri->map);
3989 if (unlikely(map)) {
3990 WRITE_ONCE(ri->map, NULL);
3991 err = dev_map_enqueue_multi(xdp, dev, map,
3992 ri->flags & BPF_F_EXCLUDE_INGRESS);
3994 err = dev_map_enqueue(fwd, xdp, dev);
3997 case BPF_MAP_TYPE_CPUMAP:
3998 err = cpu_map_enqueue(fwd, xdp, dev);
4000 case BPF_MAP_TYPE_XSKMAP:
4001 err = __xsk_map_redirect(fwd, xdp);
4003 case BPF_MAP_TYPE_UNSPEC:
4004 if (map_id == INT_MAX) {
4005 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4006 if (unlikely(!fwd)) {
4010 err = dev_xdp_enqueue(fwd, xdp, dev);
4021 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4024 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4027 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4029 static int xdp_do_generic_redirect_map(struct net_device *dev,
4030 struct sk_buff *skb,
4031 struct xdp_buff *xdp,
4032 struct bpf_prog *xdp_prog,
4034 enum bpf_map_type map_type, u32 map_id)
4036 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4037 struct bpf_map *map;
4041 case BPF_MAP_TYPE_DEVMAP:
4043 case BPF_MAP_TYPE_DEVMAP_HASH:
4044 map = READ_ONCE(ri->map);
4045 if (unlikely(map)) {
4046 WRITE_ONCE(ri->map, NULL);
4047 err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4048 ri->flags & BPF_F_EXCLUDE_INGRESS);
4050 err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4055 case BPF_MAP_TYPE_XSKMAP:
4056 err = xsk_generic_rcv(fwd, xdp);
4061 case BPF_MAP_TYPE_CPUMAP:
4062 err = cpu_map_generic_redirect(fwd, skb);
4071 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4074 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4078 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4079 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4081 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4082 enum bpf_map_type map_type = ri->map_type;
4083 void *fwd = ri->tgt_value;
4084 u32 map_id = ri->map_id;
4087 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4088 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4090 if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4091 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4092 if (unlikely(!fwd)) {
4097 err = xdp_ok_fwd_dev(fwd, skb->len);
4102 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4103 generic_xdp_tx(skb, xdp_prog);
4107 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id);
4109 _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4113 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4115 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4117 if (unlikely(flags))
4120 /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4121 * by map_idr) is used for ifindex based XDP redirect.
4123 ri->tgt_index = ifindex;
4124 ri->map_id = INT_MAX;
4125 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4127 return XDP_REDIRECT;
4130 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4131 .func = bpf_xdp_redirect,
4133 .ret_type = RET_INTEGER,
4134 .arg1_type = ARG_ANYTHING,
4135 .arg2_type = ARG_ANYTHING,
4138 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
4141 return map->ops->map_redirect(map, ifindex, flags);
4144 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4145 .func = bpf_xdp_redirect_map,
4147 .ret_type = RET_INTEGER,
4148 .arg1_type = ARG_CONST_MAP_PTR,
4149 .arg2_type = ARG_ANYTHING,
4150 .arg3_type = ARG_ANYTHING,
4153 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4154 unsigned long off, unsigned long len)
4156 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4160 if (ptr != dst_buff)
4161 memcpy(dst_buff, ptr, len);
4166 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4167 u64, flags, void *, meta, u64, meta_size)
4169 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4171 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4173 if (unlikely(!skb || skb_size > skb->len))
4176 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4180 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4181 .func = bpf_skb_event_output,
4183 .ret_type = RET_INTEGER,
4184 .arg1_type = ARG_PTR_TO_CTX,
4185 .arg2_type = ARG_CONST_MAP_PTR,
4186 .arg3_type = ARG_ANYTHING,
4187 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4188 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4191 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4193 const struct bpf_func_proto bpf_skb_output_proto = {
4194 .func = bpf_skb_event_output,
4196 .ret_type = RET_INTEGER,
4197 .arg1_type = ARG_PTR_TO_BTF_ID,
4198 .arg1_btf_id = &bpf_skb_output_btf_ids[0],
4199 .arg2_type = ARG_CONST_MAP_PTR,
4200 .arg3_type = ARG_ANYTHING,
4201 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4202 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4205 static unsigned short bpf_tunnel_key_af(u64 flags)
4207 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4210 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4211 u32, size, u64, flags)
4213 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4214 u8 compat[sizeof(struct bpf_tunnel_key)];
4218 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
4222 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4226 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4229 case offsetof(struct bpf_tunnel_key, tunnel_label):
4230 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4232 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4233 /* Fixup deprecated structure layouts here, so we have
4234 * a common path later on.
4236 if (ip_tunnel_info_af(info) != AF_INET)
4239 to = (struct bpf_tunnel_key *)compat;
4246 to->tunnel_id = be64_to_cpu(info->key.tun_id);
4247 to->tunnel_tos = info->key.tos;
4248 to->tunnel_ttl = info->key.ttl;
4251 if (flags & BPF_F_TUNINFO_IPV6) {
4252 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4253 sizeof(to->remote_ipv6));
4254 to->tunnel_label = be32_to_cpu(info->key.label);
4256 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4257 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4258 to->tunnel_label = 0;
4261 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4262 memcpy(to_orig, to, size);
4266 memset(to_orig, 0, size);
4270 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4271 .func = bpf_skb_get_tunnel_key,
4273 .ret_type = RET_INTEGER,
4274 .arg1_type = ARG_PTR_TO_CTX,
4275 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4276 .arg3_type = ARG_CONST_SIZE,
4277 .arg4_type = ARG_ANYTHING,
4280 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4282 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4285 if (unlikely(!info ||
4286 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4290 if (unlikely(size < info->options_len)) {
4295 ip_tunnel_info_opts_get(to, info);
4296 if (size > info->options_len)
4297 memset(to + info->options_len, 0, size - info->options_len);
4299 return info->options_len;
4301 memset(to, 0, size);
4305 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4306 .func = bpf_skb_get_tunnel_opt,
4308 .ret_type = RET_INTEGER,
4309 .arg1_type = ARG_PTR_TO_CTX,
4310 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4311 .arg3_type = ARG_CONST_SIZE,
4314 static struct metadata_dst __percpu *md_dst;
4316 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4317 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4319 struct metadata_dst *md = this_cpu_ptr(md_dst);
4320 u8 compat[sizeof(struct bpf_tunnel_key)];
4321 struct ip_tunnel_info *info;
4323 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4324 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
4326 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4328 case offsetof(struct bpf_tunnel_key, tunnel_label):
4329 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4330 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4331 /* Fixup deprecated structure layouts here, so we have
4332 * a common path later on.
4334 memcpy(compat, from, size);
4335 memset(compat + size, 0, sizeof(compat) - size);
4336 from = (const struct bpf_tunnel_key *) compat;
4342 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4347 dst_hold((struct dst_entry *) md);
4348 skb_dst_set(skb, (struct dst_entry *) md);
4350 info = &md->u.tun_info;
4351 memset(info, 0, sizeof(*info));
4352 info->mode = IP_TUNNEL_INFO_TX;
4354 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4355 if (flags & BPF_F_DONT_FRAGMENT)
4356 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4357 if (flags & BPF_F_ZERO_CSUM_TX)
4358 info->key.tun_flags &= ~TUNNEL_CSUM;
4359 if (flags & BPF_F_SEQ_NUMBER)
4360 info->key.tun_flags |= TUNNEL_SEQ;
4362 info->key.tun_id = cpu_to_be64(from->tunnel_id);
4363 info->key.tos = from->tunnel_tos;
4364 info->key.ttl = from->tunnel_ttl;
4366 if (flags & BPF_F_TUNINFO_IPV6) {
4367 info->mode |= IP_TUNNEL_INFO_IPV6;
4368 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4369 sizeof(from->remote_ipv6));
4370 info->key.label = cpu_to_be32(from->tunnel_label) &
4371 IPV6_FLOWLABEL_MASK;
4373 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4379 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4380 .func = bpf_skb_set_tunnel_key,
4382 .ret_type = RET_INTEGER,
4383 .arg1_type = ARG_PTR_TO_CTX,
4384 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4385 .arg3_type = ARG_CONST_SIZE,
4386 .arg4_type = ARG_ANYTHING,
4389 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4390 const u8 *, from, u32, size)
4392 struct ip_tunnel_info *info = skb_tunnel_info(skb);
4393 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4395 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4397 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4400 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4405 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4406 .func = bpf_skb_set_tunnel_opt,
4408 .ret_type = RET_INTEGER,
4409 .arg1_type = ARG_PTR_TO_CTX,
4410 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4411 .arg3_type = ARG_CONST_SIZE,
4414 static const struct bpf_func_proto *
4415 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4418 struct metadata_dst __percpu *tmp;
4420 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4425 if (cmpxchg(&md_dst, NULL, tmp))
4426 metadata_dst_free_percpu(tmp);
4430 case BPF_FUNC_skb_set_tunnel_key:
4431 return &bpf_skb_set_tunnel_key_proto;
4432 case BPF_FUNC_skb_set_tunnel_opt:
4433 return &bpf_skb_set_tunnel_opt_proto;
4439 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4442 struct bpf_array *array = container_of(map, struct bpf_array, map);
4443 struct cgroup *cgrp;
4446 sk = skb_to_full_sk(skb);
4447 if (!sk || !sk_fullsock(sk))
4449 if (unlikely(idx >= array->map.max_entries))
4452 cgrp = READ_ONCE(array->ptrs[idx]);
4453 if (unlikely(!cgrp))
4456 return sk_under_cgroup_hierarchy(sk, cgrp);
4459 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4460 .func = bpf_skb_under_cgroup,
4462 .ret_type = RET_INTEGER,
4463 .arg1_type = ARG_PTR_TO_CTX,
4464 .arg2_type = ARG_CONST_MAP_PTR,
4465 .arg3_type = ARG_ANYTHING,
4468 #ifdef CONFIG_SOCK_CGROUP_DATA
4469 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4471 struct cgroup *cgrp;
4473 sk = sk_to_full_sk(sk);
4474 if (!sk || !sk_fullsock(sk))
4477 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4478 return cgroup_id(cgrp);
4481 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4483 return __bpf_sk_cgroup_id(skb->sk);
4486 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4487 .func = bpf_skb_cgroup_id,
4489 .ret_type = RET_INTEGER,
4490 .arg1_type = ARG_PTR_TO_CTX,
4493 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4496 struct cgroup *ancestor;
4497 struct cgroup *cgrp;
4499 sk = sk_to_full_sk(sk);
4500 if (!sk || !sk_fullsock(sk))
4503 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4504 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4508 return cgroup_id(ancestor);
4511 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4514 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4517 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4518 .func = bpf_skb_ancestor_cgroup_id,
4520 .ret_type = RET_INTEGER,
4521 .arg1_type = ARG_PTR_TO_CTX,
4522 .arg2_type = ARG_ANYTHING,
4525 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4527 return __bpf_sk_cgroup_id(sk);
4530 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4531 .func = bpf_sk_cgroup_id,
4533 .ret_type = RET_INTEGER,
4534 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4537 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4539 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4542 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4543 .func = bpf_sk_ancestor_cgroup_id,
4545 .ret_type = RET_INTEGER,
4546 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4547 .arg2_type = ARG_ANYTHING,
4551 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4552 unsigned long off, unsigned long len)
4554 memcpy(dst_buff, src_buff + off, len);
4558 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4559 u64, flags, void *, meta, u64, meta_size)
4561 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4563 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4565 if (unlikely(!xdp ||
4566 xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4569 return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4570 xdp_size, bpf_xdp_copy);
4573 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4574 .func = bpf_xdp_event_output,
4576 .ret_type = RET_INTEGER,
4577 .arg1_type = ARG_PTR_TO_CTX,
4578 .arg2_type = ARG_CONST_MAP_PTR,
4579 .arg3_type = ARG_ANYTHING,
4580 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4581 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4584 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4586 const struct bpf_func_proto bpf_xdp_output_proto = {
4587 .func = bpf_xdp_event_output,
4589 .ret_type = RET_INTEGER,
4590 .arg1_type = ARG_PTR_TO_BTF_ID,
4591 .arg1_btf_id = &bpf_xdp_output_btf_ids[0],
4592 .arg2_type = ARG_CONST_MAP_PTR,
4593 .arg3_type = ARG_ANYTHING,
4594 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4595 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4598 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4600 return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4603 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4604 .func = bpf_get_socket_cookie,
4606 .ret_type = RET_INTEGER,
4607 .arg1_type = ARG_PTR_TO_CTX,
4610 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4612 return __sock_gen_cookie(ctx->sk);
4615 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4616 .func = bpf_get_socket_cookie_sock_addr,
4618 .ret_type = RET_INTEGER,
4619 .arg1_type = ARG_PTR_TO_CTX,
4622 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4624 return __sock_gen_cookie(ctx);
4627 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4628 .func = bpf_get_socket_cookie_sock,
4630 .ret_type = RET_INTEGER,
4631 .arg1_type = ARG_PTR_TO_CTX,
4634 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
4636 return sk ? sock_gen_cookie(sk) : 0;
4639 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
4640 .func = bpf_get_socket_ptr_cookie,
4642 .ret_type = RET_INTEGER,
4643 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4646 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4648 return __sock_gen_cookie(ctx->sk);
4651 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4652 .func = bpf_get_socket_cookie_sock_ops,
4654 .ret_type = RET_INTEGER,
4655 .arg1_type = ARG_PTR_TO_CTX,
4658 static u64 __bpf_get_netns_cookie(struct sock *sk)
4660 const struct net *net = sk ? sock_net(sk) : &init_net;
4662 return net->net_cookie;
4665 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
4667 return __bpf_get_netns_cookie(ctx);
4670 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
4671 .func = bpf_get_netns_cookie_sock,
4673 .ret_type = RET_INTEGER,
4674 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4677 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4679 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4682 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
4683 .func = bpf_get_netns_cookie_sock_addr,
4685 .ret_type = RET_INTEGER,
4686 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4689 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4691 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4694 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = {
4695 .func = bpf_get_netns_cookie_sock_ops,
4697 .ret_type = RET_INTEGER,
4698 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4701 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx)
4703 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4706 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = {
4707 .func = bpf_get_netns_cookie_sk_msg,
4709 .ret_type = RET_INTEGER,
4710 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4713 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4715 struct sock *sk = sk_to_full_sk(skb->sk);
4718 if (!sk || !sk_fullsock(sk))
4720 kuid = sock_net_uid(sock_net(sk), sk);
4721 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4724 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4725 .func = bpf_get_socket_uid,
4727 .ret_type = RET_INTEGER,
4728 .arg1_type = ARG_PTR_TO_CTX,
4731 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
4732 char *optval, int optlen)
4734 char devname[IFNAMSIZ];
4740 if (!sk_fullsock(sk))
4743 sock_owned_by_me(sk);
4745 if (level == SOL_SOCKET) {
4746 if (optlen != sizeof(int) && optname != SO_BINDTODEVICE)
4748 val = *((int *)optval);
4749 valbool = val ? 1 : 0;
4751 /* Only some socketops are supported */
4754 val = min_t(u32, val, READ_ONCE(sysctl_rmem_max));
4755 val = min_t(int, val, INT_MAX / 2);
4756 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4757 WRITE_ONCE(sk->sk_rcvbuf,
4758 max_t(int, val * 2, SOCK_MIN_RCVBUF));
4761 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
4762 val = min_t(int, val, INT_MAX / 2);
4763 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4764 WRITE_ONCE(sk->sk_sndbuf,
4765 max_t(int, val * 2, SOCK_MIN_SNDBUF));
4767 case SO_MAX_PACING_RATE: /* 32bit version */
4769 cmpxchg(&sk->sk_pacing_status,
4772 sk->sk_max_pacing_rate = (val == ~0U) ?
4773 ~0UL : (unsigned int)val;
4774 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4775 sk->sk_max_pacing_rate);
4778 sk->sk_priority = val;
4783 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
4786 if (sk->sk_mark != val) {
4791 case SO_BINDTODEVICE:
4792 optlen = min_t(long, optlen, IFNAMSIZ - 1);
4793 strncpy(devname, optval, optlen);
4794 devname[optlen] = 0;
4797 if (devname[0] != '\0') {
4798 struct net_device *dev;
4803 dev = dev_get_by_name(net, devname);
4806 ifindex = dev->ifindex;
4810 case SO_BINDTOIFINDEX:
4811 if (optname == SO_BINDTOIFINDEX)
4813 ret = sock_bindtoindex(sk, ifindex, false);
4816 if (sk->sk_prot->keepalive)
4817 sk->sk_prot->keepalive(sk, valbool);
4818 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
4821 sk->sk_reuseport = valbool;
4827 } else if (level == SOL_IP) {
4828 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4831 val = *((int *)optval);
4832 /* Only some options are supported */
4835 if (val < -1 || val > 0xff) {
4838 struct inet_sock *inet = inet_sk(sk);
4848 #if IS_ENABLED(CONFIG_IPV6)
4849 } else if (level == SOL_IPV6) {
4850 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4853 val = *((int *)optval);
4854 /* Only some options are supported */
4857 if (val < -1 || val > 0xff) {
4860 struct ipv6_pinfo *np = inet6_sk(sk);
4871 } else if (level == SOL_TCP &&
4872 sk->sk_prot->setsockopt == tcp_setsockopt) {
4873 if (optname == TCP_CONGESTION) {
4874 char name[TCP_CA_NAME_MAX];
4876 strncpy(name, optval, min_t(long, optlen,
4877 TCP_CA_NAME_MAX-1));
4878 name[TCP_CA_NAME_MAX-1] = 0;
4879 ret = tcp_set_congestion_control(sk, name, false, true);
4881 struct inet_connection_sock *icsk = inet_csk(sk);
4882 struct tcp_sock *tp = tcp_sk(sk);
4883 unsigned long timeout;
4885 if (optlen != sizeof(int))
4888 val = *((int *)optval);
4889 /* Only some options are supported */
4892 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4895 tcp_snd_cwnd_set(tp, val);
4897 case TCP_BPF_SNDCWND_CLAMP:
4901 tp->snd_cwnd_clamp = val;
4902 tp->snd_ssthresh = val;
4905 case TCP_BPF_DELACK_MAX:
4906 timeout = usecs_to_jiffies(val);
4907 if (timeout > TCP_DELACK_MAX ||
4908 timeout < TCP_TIMEOUT_MIN)
4910 inet_csk(sk)->icsk_delack_max = timeout;
4912 case TCP_BPF_RTO_MIN:
4913 timeout = usecs_to_jiffies(val);
4914 if (timeout > TCP_RTO_MIN ||
4915 timeout < TCP_TIMEOUT_MIN)
4917 inet_csk(sk)->icsk_rto_min = timeout;
4920 if (val < 0 || val > 1)
4926 ret = tcp_sock_set_keepidle_locked(sk, val);
4929 if (val < 1 || val > MAX_TCP_KEEPINTVL)
4932 tp->keepalive_intvl = val * HZ;
4935 if (val < 1 || val > MAX_TCP_KEEPCNT)
4938 tp->keepalive_probes = val;
4941 if (val < 1 || val > MAX_TCP_SYNCNT)
4944 icsk->icsk_syn_retries = val;
4946 case TCP_USER_TIMEOUT:
4950 icsk->icsk_user_timeout = val;
4952 case TCP_NOTSENT_LOWAT:
4953 tp->notsent_lowat = val;
4954 sk->sk_write_space(sk);
4956 case TCP_WINDOW_CLAMP:
4957 ret = tcp_set_window_clamp(sk, val);
4970 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
4971 char *optval, int optlen)
4973 if (!sk_fullsock(sk))
4976 sock_owned_by_me(sk);
4978 if (level == SOL_SOCKET) {
4979 if (optlen != sizeof(int))
4984 *((int *)optval) = sk->sk_mark;
4987 *((int *)optval) = sk->sk_priority;
4989 case SO_BINDTOIFINDEX:
4990 *((int *)optval) = sk->sk_bound_dev_if;
4993 *((int *)optval) = sk->sk_reuseport;
4999 } else if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
5000 struct inet_connection_sock *icsk;
5001 struct tcp_sock *tp;
5004 case TCP_CONGESTION:
5005 icsk = inet_csk(sk);
5007 if (!icsk->icsk_ca_ops || optlen <= 1)
5009 strncpy(optval, icsk->icsk_ca_ops->name, optlen);
5010 optval[optlen - 1] = 0;
5015 if (optlen <= 0 || !tp->saved_syn ||
5016 optlen > tcp_saved_syn_len(tp->saved_syn))
5018 memcpy(optval, tp->saved_syn->data, optlen);
5023 } else if (level == SOL_IP) {
5024 struct inet_sock *inet = inet_sk(sk);
5026 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
5029 /* Only some options are supported */
5032 *((int *)optval) = (int)inet->tos;
5037 #if IS_ENABLED(CONFIG_IPV6)
5038 } else if (level == SOL_IPV6) {
5039 struct ipv6_pinfo *np = inet6_sk(sk);
5041 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
5044 /* Only some options are supported */
5047 *((int *)optval) = (int)np->tclass;
5059 memset(optval, 0, optlen);
5063 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level,
5064 int, optname, char *, optval, int, optlen)
5066 if (level == SOL_TCP && optname == TCP_CONGESTION) {
5067 if (optlen >= sizeof("cdg") - 1 &&
5068 !strncmp("cdg", optval, optlen))
5072 return _bpf_setsockopt(sk, level, optname, optval, optlen);
5075 const struct bpf_func_proto bpf_sk_setsockopt_proto = {
5076 .func = bpf_sk_setsockopt,
5078 .ret_type = RET_INTEGER,
5079 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5080 .arg2_type = ARG_ANYTHING,
5081 .arg3_type = ARG_ANYTHING,
5082 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5083 .arg5_type = ARG_CONST_SIZE,
5086 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level,
5087 int, optname, char *, optval, int, optlen)
5089 return _bpf_getsockopt(sk, level, optname, optval, optlen);
5092 const struct bpf_func_proto bpf_sk_getsockopt_proto = {
5093 .func = bpf_sk_getsockopt,
5095 .ret_type = RET_INTEGER,
5096 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5097 .arg2_type = ARG_ANYTHING,
5098 .arg3_type = ARG_ANYTHING,
5099 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5100 .arg5_type = ARG_CONST_SIZE,
5103 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5104 int, level, int, optname, char *, optval, int, optlen)
5106 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5109 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5110 .func = bpf_sock_addr_setsockopt,
5112 .ret_type = RET_INTEGER,
5113 .arg1_type = ARG_PTR_TO_CTX,
5114 .arg2_type = ARG_ANYTHING,
5115 .arg3_type = ARG_ANYTHING,
5116 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5117 .arg5_type = ARG_CONST_SIZE,
5120 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5121 int, level, int, optname, char *, optval, int, optlen)
5123 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5126 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5127 .func = bpf_sock_addr_getsockopt,
5129 .ret_type = RET_INTEGER,
5130 .arg1_type = ARG_PTR_TO_CTX,
5131 .arg2_type = ARG_ANYTHING,
5132 .arg3_type = ARG_ANYTHING,
5133 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5134 .arg5_type = ARG_CONST_SIZE,
5137 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5138 int, level, int, optname, char *, optval, int, optlen)
5140 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5143 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5144 .func = bpf_sock_ops_setsockopt,
5146 .ret_type = RET_INTEGER,
5147 .arg1_type = ARG_PTR_TO_CTX,
5148 .arg2_type = ARG_ANYTHING,
5149 .arg3_type = ARG_ANYTHING,
5150 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5151 .arg5_type = ARG_CONST_SIZE,
5154 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5155 int optname, const u8 **start)
5157 struct sk_buff *syn_skb = bpf_sock->syn_skb;
5158 const u8 *hdr_start;
5162 /* sk is a request_sock here */
5164 if (optname == TCP_BPF_SYN) {
5165 hdr_start = syn_skb->data;
5166 ret = tcp_hdrlen(syn_skb);
5167 } else if (optname == TCP_BPF_SYN_IP) {
5168 hdr_start = skb_network_header(syn_skb);
5169 ret = skb_network_header_len(syn_skb) +
5170 tcp_hdrlen(syn_skb);
5172 /* optname == TCP_BPF_SYN_MAC */
5173 hdr_start = skb_mac_header(syn_skb);
5174 ret = skb_mac_header_len(syn_skb) +
5175 skb_network_header_len(syn_skb) +
5176 tcp_hdrlen(syn_skb);
5179 struct sock *sk = bpf_sock->sk;
5180 struct saved_syn *saved_syn;
5182 if (sk->sk_state == TCP_NEW_SYN_RECV)
5183 /* synack retransmit. bpf_sock->syn_skb will
5184 * not be available. It has to resort to
5185 * saved_syn (if it is saved).
5187 saved_syn = inet_reqsk(sk)->saved_syn;
5189 saved_syn = tcp_sk(sk)->saved_syn;
5194 if (optname == TCP_BPF_SYN) {
5195 hdr_start = saved_syn->data +
5196 saved_syn->mac_hdrlen +
5197 saved_syn->network_hdrlen;
5198 ret = saved_syn->tcp_hdrlen;
5199 } else if (optname == TCP_BPF_SYN_IP) {
5200 hdr_start = saved_syn->data +
5201 saved_syn->mac_hdrlen;
5202 ret = saved_syn->network_hdrlen +
5203 saved_syn->tcp_hdrlen;
5205 /* optname == TCP_BPF_SYN_MAC */
5207 /* TCP_SAVE_SYN may not have saved the mac hdr */
5208 if (!saved_syn->mac_hdrlen)
5211 hdr_start = saved_syn->data;
5212 ret = saved_syn->mac_hdrlen +
5213 saved_syn->network_hdrlen +
5214 saved_syn->tcp_hdrlen;
5222 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5223 int, level, int, optname, char *, optval, int, optlen)
5225 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5226 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5227 int ret, copy_len = 0;
5230 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5233 if (optlen < copy_len) {
5238 memcpy(optval, start, copy_len);
5241 /* Zero out unused buffer at the end */
5242 memset(optval + copy_len, 0, optlen - copy_len);
5247 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5250 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5251 .func = bpf_sock_ops_getsockopt,
5253 .ret_type = RET_INTEGER,
5254 .arg1_type = ARG_PTR_TO_CTX,
5255 .arg2_type = ARG_ANYTHING,
5256 .arg3_type = ARG_ANYTHING,
5257 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5258 .arg5_type = ARG_CONST_SIZE,
5261 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5264 struct sock *sk = bpf_sock->sk;
5265 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5267 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5270 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5272 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5275 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5276 .func = bpf_sock_ops_cb_flags_set,
5278 .ret_type = RET_INTEGER,
5279 .arg1_type = ARG_PTR_TO_CTX,
5280 .arg2_type = ARG_ANYTHING,
5283 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5284 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5286 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5290 struct sock *sk = ctx->sk;
5291 u32 flags = BIND_FROM_BPF;
5295 if (addr_len < offsetofend(struct sockaddr, sa_family))
5297 if (addr->sa_family == AF_INET) {
5298 if (addr_len < sizeof(struct sockaddr_in))
5300 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5301 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5302 return __inet_bind(sk, addr, addr_len, flags);
5303 #if IS_ENABLED(CONFIG_IPV6)
5304 } else if (addr->sa_family == AF_INET6) {
5305 if (addr_len < SIN6_LEN_RFC2133)
5307 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5308 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5309 /* ipv6_bpf_stub cannot be NULL, since it's called from
5310 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5312 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5313 #endif /* CONFIG_IPV6 */
5315 #endif /* CONFIG_INET */
5317 return -EAFNOSUPPORT;
5320 static const struct bpf_func_proto bpf_bind_proto = {
5323 .ret_type = RET_INTEGER,
5324 .arg1_type = ARG_PTR_TO_CTX,
5325 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5326 .arg3_type = ARG_CONST_SIZE,
5330 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5331 struct bpf_xfrm_state *, to, u32, size, u64, flags)
5333 const struct sec_path *sp = skb_sec_path(skb);
5334 const struct xfrm_state *x;
5336 if (!sp || unlikely(index >= sp->len || flags))
5339 x = sp->xvec[index];
5341 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5344 to->reqid = x->props.reqid;
5345 to->spi = x->id.spi;
5346 to->family = x->props.family;
5349 if (to->family == AF_INET6) {
5350 memcpy(to->remote_ipv6, x->props.saddr.a6,
5351 sizeof(to->remote_ipv6));
5353 to->remote_ipv4 = x->props.saddr.a4;
5354 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5359 memset(to, 0, size);
5363 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5364 .func = bpf_skb_get_xfrm_state,
5366 .ret_type = RET_INTEGER,
5367 .arg1_type = ARG_PTR_TO_CTX,
5368 .arg2_type = ARG_ANYTHING,
5369 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
5370 .arg4_type = ARG_CONST_SIZE,
5371 .arg5_type = ARG_ANYTHING,
5375 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5376 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
5377 const struct neighbour *neigh,
5378 const struct net_device *dev, u32 mtu)
5380 memcpy(params->dmac, neigh->ha, ETH_ALEN);
5381 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5382 params->h_vlan_TCI = 0;
5383 params->h_vlan_proto = 0;
5385 params->mtu_result = mtu; /* union with tot_len */
5391 #if IS_ENABLED(CONFIG_INET)
5392 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5393 u32 flags, bool check_mtu)
5395 struct fib_nh_common *nhc;
5396 struct in_device *in_dev;
5397 struct neighbour *neigh;
5398 struct net_device *dev;
5399 struct fib_result res;
5404 dev = dev_get_by_index_rcu(net, params->ifindex);
5408 /* verify forwarding is enabled on this interface */
5409 in_dev = __in_dev_get_rcu(dev);
5410 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5411 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5413 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5415 fl4.flowi4_oif = params->ifindex;
5417 fl4.flowi4_iif = params->ifindex;
5420 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5421 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5422 fl4.flowi4_flags = 0;
5424 fl4.flowi4_proto = params->l4_protocol;
5425 fl4.daddr = params->ipv4_dst;
5426 fl4.saddr = params->ipv4_src;
5427 fl4.fl4_sport = params->sport;
5428 fl4.fl4_dport = params->dport;
5429 fl4.flowi4_multipath_hash = 0;
5431 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5432 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5433 struct fib_table *tb;
5435 tb = fib_get_table(net, tbid);
5437 return BPF_FIB_LKUP_RET_NOT_FWDED;
5439 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5441 fl4.flowi4_mark = 0;
5442 fl4.flowi4_secid = 0;
5443 fl4.flowi4_tun_key.tun_id = 0;
5444 fl4.flowi4_uid = sock_net_uid(net, NULL);
5446 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5450 /* map fib lookup errors to RTN_ type */
5452 return BPF_FIB_LKUP_RET_BLACKHOLE;
5453 if (err == -EHOSTUNREACH)
5454 return BPF_FIB_LKUP_RET_UNREACHABLE;
5456 return BPF_FIB_LKUP_RET_PROHIBIT;
5458 return BPF_FIB_LKUP_RET_NOT_FWDED;
5461 if (res.type != RTN_UNICAST)
5462 return BPF_FIB_LKUP_RET_NOT_FWDED;
5464 if (fib_info_num_path(res.fi) > 1)
5465 fib_select_path(net, &res, &fl4, NULL);
5468 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5469 if (params->tot_len > mtu) {
5470 params->mtu_result = mtu; /* union with tot_len */
5471 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5477 /* do not handle lwt encaps right now */
5478 if (nhc->nhc_lwtstate)
5479 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5483 params->rt_metric = res.fi->fib_priority;
5484 params->ifindex = dev->ifindex;
5486 /* xdp and cls_bpf programs are run in RCU-bh so
5487 * rcu_read_lock_bh is not needed here
5489 if (likely(nhc->nhc_gw_family != AF_INET6)) {
5490 if (nhc->nhc_gw_family)
5491 params->ipv4_dst = nhc->nhc_gw.ipv4;
5493 neigh = __ipv4_neigh_lookup_noref(dev,
5494 (__force u32)params->ipv4_dst);
5496 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5498 params->family = AF_INET6;
5499 *dst = nhc->nhc_gw.ipv6;
5500 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5503 if (!neigh || !(neigh->nud_state & NUD_VALID))
5504 return BPF_FIB_LKUP_RET_NO_NEIGH;
5506 return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5510 #if IS_ENABLED(CONFIG_IPV6)
5511 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5512 u32 flags, bool check_mtu)
5514 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5515 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5516 struct fib6_result res = {};
5517 struct neighbour *neigh;
5518 struct net_device *dev;
5519 struct inet6_dev *idev;
5525 /* link local addresses are never forwarded */
5526 if (rt6_need_strict(dst) || rt6_need_strict(src))
5527 return BPF_FIB_LKUP_RET_NOT_FWDED;
5529 dev = dev_get_by_index_rcu(net, params->ifindex);
5533 idev = __in6_dev_get_safely(dev);
5534 if (unlikely(!idev || !idev->cnf.forwarding))
5535 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5537 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5539 oif = fl6.flowi6_oif = params->ifindex;
5541 oif = fl6.flowi6_iif = params->ifindex;
5543 strict = RT6_LOOKUP_F_HAS_SADDR;
5545 fl6.flowlabel = params->flowinfo;
5546 fl6.flowi6_scope = 0;
5547 fl6.flowi6_flags = 0;
5550 fl6.flowi6_proto = params->l4_protocol;
5553 fl6.fl6_sport = params->sport;
5554 fl6.fl6_dport = params->dport;
5556 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5557 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5558 struct fib6_table *tb;
5560 tb = ipv6_stub->fib6_get_table(net, tbid);
5562 return BPF_FIB_LKUP_RET_NOT_FWDED;
5564 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5567 fl6.flowi6_mark = 0;
5568 fl6.flowi6_secid = 0;
5569 fl6.flowi6_tun_key.tun_id = 0;
5570 fl6.flowi6_uid = sock_net_uid(net, NULL);
5572 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5575 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5576 res.f6i == net->ipv6.fib6_null_entry))
5577 return BPF_FIB_LKUP_RET_NOT_FWDED;
5579 switch (res.fib6_type) {
5580 /* only unicast is forwarded */
5584 return BPF_FIB_LKUP_RET_BLACKHOLE;
5585 case RTN_UNREACHABLE:
5586 return BPF_FIB_LKUP_RET_UNREACHABLE;
5588 return BPF_FIB_LKUP_RET_PROHIBIT;
5590 return BPF_FIB_LKUP_RET_NOT_FWDED;
5593 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5594 fl6.flowi6_oif != 0, NULL, strict);
5597 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5598 if (params->tot_len > mtu) {
5599 params->mtu_result = mtu; /* union with tot_len */
5600 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5604 if (res.nh->fib_nh_lws)
5605 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5607 if (res.nh->fib_nh_gw_family)
5608 *dst = res.nh->fib_nh_gw6;
5610 dev = res.nh->fib_nh_dev;
5611 params->rt_metric = res.f6i->fib6_metric;
5612 params->ifindex = dev->ifindex;
5614 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
5617 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5618 if (!neigh || !(neigh->nud_state & NUD_VALID))
5619 return BPF_FIB_LKUP_RET_NO_NEIGH;
5621 return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5625 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
5626 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5628 if (plen < sizeof(*params))
5631 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5634 switch (params->family) {
5635 #if IS_ENABLED(CONFIG_INET)
5637 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
5640 #if IS_ENABLED(CONFIG_IPV6)
5642 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
5646 return -EAFNOSUPPORT;
5649 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
5650 .func = bpf_xdp_fib_lookup,
5652 .ret_type = RET_INTEGER,
5653 .arg1_type = ARG_PTR_TO_CTX,
5654 .arg2_type = ARG_PTR_TO_MEM,
5655 .arg3_type = ARG_CONST_SIZE,
5656 .arg4_type = ARG_ANYTHING,
5659 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
5660 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5662 struct net *net = dev_net(skb->dev);
5663 int rc = -EAFNOSUPPORT;
5664 bool check_mtu = false;
5666 if (plen < sizeof(*params))
5669 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5672 if (params->tot_len)
5675 switch (params->family) {
5676 #if IS_ENABLED(CONFIG_INET)
5678 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
5681 #if IS_ENABLED(CONFIG_IPV6)
5683 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
5688 if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
5689 struct net_device *dev;
5691 /* When tot_len isn't provided by user, check skb
5692 * against MTU of FIB lookup resulting net_device
5694 dev = dev_get_by_index_rcu(net, params->ifindex);
5695 if (!is_skb_forwardable(dev, skb))
5696 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
5698 params->mtu_result = dev->mtu; /* union with tot_len */
5704 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
5705 .func = bpf_skb_fib_lookup,
5707 .ret_type = RET_INTEGER,
5708 .arg1_type = ARG_PTR_TO_CTX,
5709 .arg2_type = ARG_PTR_TO_MEM,
5710 .arg3_type = ARG_CONST_SIZE,
5711 .arg4_type = ARG_ANYTHING,
5714 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
5717 struct net *netns = dev_net(dev_curr);
5719 /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
5723 return dev_get_by_index_rcu(netns, ifindex);
5726 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
5727 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5729 int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5730 struct net_device *dev = skb->dev;
5731 int skb_len, dev_len;
5734 if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
5737 if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
5740 dev = __dev_via_ifindex(dev, ifindex);
5744 mtu = READ_ONCE(dev->mtu);
5746 dev_len = mtu + dev->hard_header_len;
5748 /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
5749 skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
5751 skb_len += len_diff; /* minus result pass check */
5752 if (skb_len <= dev_len) {
5753 ret = BPF_MTU_CHK_RET_SUCCESS;
5756 /* At this point, skb->len exceed MTU, but as it include length of all
5757 * segments, it can still be below MTU. The SKB can possibly get
5758 * re-segmented in transmit path (see validate_xmit_skb). Thus, user
5759 * must choose if segs are to be MTU checked.
5761 if (skb_is_gso(skb)) {
5762 ret = BPF_MTU_CHK_RET_SUCCESS;
5764 if (flags & BPF_MTU_CHK_SEGS &&
5765 !skb_gso_validate_network_len(skb, mtu))
5766 ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
5769 /* BPF verifier guarantees valid pointer */
5775 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
5776 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5778 struct net_device *dev = xdp->rxq->dev;
5779 int xdp_len = xdp->data_end - xdp->data;
5780 int ret = BPF_MTU_CHK_RET_SUCCESS;
5783 /* XDP variant doesn't support multi-buffer segment check (yet) */
5784 if (unlikely(flags))
5787 dev = __dev_via_ifindex(dev, ifindex);
5791 mtu = READ_ONCE(dev->mtu);
5793 /* Add L2-header as dev MTU is L3 size */
5794 dev_len = mtu + dev->hard_header_len;
5796 /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
5798 xdp_len = *mtu_len + dev->hard_header_len;
5800 xdp_len += len_diff; /* minus result pass check */
5801 if (xdp_len > dev_len)
5802 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5804 /* BPF verifier guarantees valid pointer */
5810 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
5811 .func = bpf_skb_check_mtu,
5813 .ret_type = RET_INTEGER,
5814 .arg1_type = ARG_PTR_TO_CTX,
5815 .arg2_type = ARG_ANYTHING,
5816 .arg3_type = ARG_PTR_TO_INT,
5817 .arg4_type = ARG_ANYTHING,
5818 .arg5_type = ARG_ANYTHING,
5821 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
5822 .func = bpf_xdp_check_mtu,
5824 .ret_type = RET_INTEGER,
5825 .arg1_type = ARG_PTR_TO_CTX,
5826 .arg2_type = ARG_ANYTHING,
5827 .arg3_type = ARG_PTR_TO_INT,
5828 .arg4_type = ARG_ANYTHING,
5829 .arg5_type = ARG_ANYTHING,
5832 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5833 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
5836 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
5838 if (!seg6_validate_srh(srh, len, false))
5842 case BPF_LWT_ENCAP_SEG6_INLINE:
5843 if (skb->protocol != htons(ETH_P_IPV6))
5846 err = seg6_do_srh_inline(skb, srh);
5848 case BPF_LWT_ENCAP_SEG6:
5849 skb_reset_inner_headers(skb);
5850 skb->encapsulation = 1;
5851 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
5857 bpf_compute_data_pointers(skb);
5861 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
5863 return seg6_lookup_nexthop(skb, NULL, 0);
5865 #endif /* CONFIG_IPV6_SEG6_BPF */
5867 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5868 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
5871 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
5875 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
5879 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5880 case BPF_LWT_ENCAP_SEG6:
5881 case BPF_LWT_ENCAP_SEG6_INLINE:
5882 return bpf_push_seg6_encap(skb, type, hdr, len);
5884 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5885 case BPF_LWT_ENCAP_IP:
5886 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
5893 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
5894 void *, hdr, u32, len)
5897 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5898 case BPF_LWT_ENCAP_IP:
5899 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
5906 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
5907 .func = bpf_lwt_in_push_encap,
5909 .ret_type = RET_INTEGER,
5910 .arg1_type = ARG_PTR_TO_CTX,
5911 .arg2_type = ARG_ANYTHING,
5912 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5913 .arg4_type = ARG_CONST_SIZE
5916 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
5917 .func = bpf_lwt_xmit_push_encap,
5919 .ret_type = RET_INTEGER,
5920 .arg1_type = ARG_PTR_TO_CTX,
5921 .arg2_type = ARG_ANYTHING,
5922 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5923 .arg4_type = ARG_CONST_SIZE
5926 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5927 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
5928 const void *, from, u32, len)
5930 struct seg6_bpf_srh_state *srh_state =
5931 this_cpu_ptr(&seg6_bpf_srh_states);
5932 struct ipv6_sr_hdr *srh = srh_state->srh;
5933 void *srh_tlvs, *srh_end, *ptr;
5939 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
5940 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
5942 ptr = skb->data + offset;
5943 if (ptr >= srh_tlvs && ptr + len <= srh_end)
5944 srh_state->valid = false;
5945 else if (ptr < (void *)&srh->flags ||
5946 ptr + len > (void *)&srh->segments)
5949 if (unlikely(bpf_try_make_writable(skb, offset + len)))
5951 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5953 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5955 memcpy(skb->data + offset, from, len);
5959 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
5960 .func = bpf_lwt_seg6_store_bytes,
5962 .ret_type = RET_INTEGER,
5963 .arg1_type = ARG_PTR_TO_CTX,
5964 .arg2_type = ARG_ANYTHING,
5965 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5966 .arg4_type = ARG_CONST_SIZE
5969 static void bpf_update_srh_state(struct sk_buff *skb)
5971 struct seg6_bpf_srh_state *srh_state =
5972 this_cpu_ptr(&seg6_bpf_srh_states);
5975 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
5976 srh_state->srh = NULL;
5978 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5979 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
5980 srh_state->valid = true;
5984 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
5985 u32, action, void *, param, u32, param_len)
5987 struct seg6_bpf_srh_state *srh_state =
5988 this_cpu_ptr(&seg6_bpf_srh_states);
5993 case SEG6_LOCAL_ACTION_END_X:
5994 if (!seg6_bpf_has_valid_srh(skb))
5996 if (param_len != sizeof(struct in6_addr))
5998 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
5999 case SEG6_LOCAL_ACTION_END_T:
6000 if (!seg6_bpf_has_valid_srh(skb))
6002 if (param_len != sizeof(int))
6004 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6005 case SEG6_LOCAL_ACTION_END_DT6:
6006 if (!seg6_bpf_has_valid_srh(skb))
6008 if (param_len != sizeof(int))
6011 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
6013 if (!pskb_pull(skb, hdroff))
6016 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
6017 skb_reset_network_header(skb);
6018 skb_reset_transport_header(skb);
6019 skb->encapsulation = 0;
6021 bpf_compute_data_pointers(skb);
6022 bpf_update_srh_state(skb);
6023 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6024 case SEG6_LOCAL_ACTION_END_B6:
6025 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6027 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
6030 bpf_update_srh_state(skb);
6033 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
6034 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6036 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
6039 bpf_update_srh_state(skb);
6047 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
6048 .func = bpf_lwt_seg6_action,
6050 .ret_type = RET_INTEGER,
6051 .arg1_type = ARG_PTR_TO_CTX,
6052 .arg2_type = ARG_ANYTHING,
6053 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6054 .arg4_type = ARG_CONST_SIZE
6057 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
6060 struct seg6_bpf_srh_state *srh_state =
6061 this_cpu_ptr(&seg6_bpf_srh_states);
6062 struct ipv6_sr_hdr *srh = srh_state->srh;
6063 void *srh_end, *srh_tlvs, *ptr;
6064 struct ipv6hdr *hdr;
6068 if (unlikely(srh == NULL))
6071 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
6072 ((srh->first_segment + 1) << 4));
6073 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
6075 ptr = skb->data + offset;
6077 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
6079 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
6083 ret = skb_cow_head(skb, len);
6084 if (unlikely(ret < 0))
6087 ret = bpf_skb_net_hdr_push(skb, offset, len);
6089 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6092 bpf_compute_data_pointers(skb);
6093 if (unlikely(ret < 0))
6096 hdr = (struct ipv6hdr *)skb->data;
6097 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6099 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6101 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6102 srh_state->hdrlen += len;
6103 srh_state->valid = false;
6107 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6108 .func = bpf_lwt_seg6_adjust_srh,
6110 .ret_type = RET_INTEGER,
6111 .arg1_type = ARG_PTR_TO_CTX,
6112 .arg2_type = ARG_ANYTHING,
6113 .arg3_type = ARG_ANYTHING,
6115 #endif /* CONFIG_IPV6_SEG6_BPF */
6118 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6119 int dif, int sdif, u8 family, u8 proto)
6121 bool refcounted = false;
6122 struct sock *sk = NULL;
6124 if (family == AF_INET) {
6125 __be32 src4 = tuple->ipv4.saddr;
6126 __be32 dst4 = tuple->ipv4.daddr;
6128 if (proto == IPPROTO_TCP)
6129 sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
6130 src4, tuple->ipv4.sport,
6131 dst4, tuple->ipv4.dport,
6132 dif, sdif, &refcounted);
6134 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6135 dst4, tuple->ipv4.dport,
6136 dif, sdif, &udp_table, NULL);
6137 #if IS_ENABLED(CONFIG_IPV6)
6139 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6140 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6142 if (proto == IPPROTO_TCP)
6143 sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
6144 src6, tuple->ipv6.sport,
6145 dst6, ntohs(tuple->ipv6.dport),
6146 dif, sdif, &refcounted);
6147 else if (likely(ipv6_bpf_stub))
6148 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6149 src6, tuple->ipv6.sport,
6150 dst6, tuple->ipv6.dport,
6156 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6157 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6163 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6164 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6165 * Returns the socket as an 'unsigned long' to simplify the casting in the
6166 * callers to satisfy BPF_CALL declarations.
6168 static struct sock *
6169 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6170 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6171 u64 flags, int sdif)
6173 struct sock *sk = NULL;
6177 if (len == sizeof(tuple->ipv4))
6179 else if (len == sizeof(tuple->ipv6))
6184 if (unlikely(flags || !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6188 if (family == AF_INET)
6189 sdif = inet_sdif(skb);
6191 sdif = inet6_sdif(skb);
6194 if ((s32)netns_id < 0) {
6196 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6198 net = get_net_ns_by_id(caller_net, netns_id);
6201 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6209 static struct sock *
6210 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6211 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6212 u64 flags, int sdif)
6214 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6215 ifindex, proto, netns_id, flags,
6219 struct sock *sk2 = sk_to_full_sk(sk);
6221 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6222 * sock refcnt is decremented to prevent a request_sock leak.
6224 if (!sk_fullsock(sk2))
6228 /* Ensure there is no need to bump sk2 refcnt */
6229 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6230 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6240 static struct sock *
6241 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6242 u8 proto, u64 netns_id, u64 flags)
6244 struct net *caller_net;
6248 caller_net = dev_net(skb->dev);
6249 ifindex = skb->dev->ifindex;
6251 caller_net = sock_net(skb->sk);
6255 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6256 netns_id, flags, -1);
6259 static struct sock *
6260 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6261 u8 proto, u64 netns_id, u64 flags)
6263 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6267 struct sock *sk2 = sk_to_full_sk(sk);
6269 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6270 * sock refcnt is decremented to prevent a request_sock leak.
6272 if (!sk_fullsock(sk2))
6276 /* Ensure there is no need to bump sk2 refcnt */
6277 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6278 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6288 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6289 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6291 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6295 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6296 .func = bpf_skc_lookup_tcp,
6299 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6300 .arg1_type = ARG_PTR_TO_CTX,
6301 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6302 .arg3_type = ARG_CONST_SIZE,
6303 .arg4_type = ARG_ANYTHING,
6304 .arg5_type = ARG_ANYTHING,
6307 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6308 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6310 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6314 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6315 .func = bpf_sk_lookup_tcp,
6318 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6319 .arg1_type = ARG_PTR_TO_CTX,
6320 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6321 .arg3_type = ARG_CONST_SIZE,
6322 .arg4_type = ARG_ANYTHING,
6323 .arg5_type = ARG_ANYTHING,
6326 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6327 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6329 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6333 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6334 .func = bpf_sk_lookup_udp,
6337 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6338 .arg1_type = ARG_PTR_TO_CTX,
6339 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6340 .arg3_type = ARG_CONST_SIZE,
6341 .arg4_type = ARG_ANYTHING,
6342 .arg5_type = ARG_ANYTHING,
6345 BPF_CALL_5(bpf_tc_skc_lookup_tcp, struct sk_buff *, skb,
6346 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6348 struct net_device *dev = skb->dev;
6349 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6350 struct net *caller_net = dev_net(dev);
6352 return (unsigned long)__bpf_skc_lookup(skb, tuple, len, caller_net,
6353 ifindex, IPPROTO_TCP, netns_id,
6357 static const struct bpf_func_proto bpf_tc_skc_lookup_tcp_proto = {
6358 .func = bpf_tc_skc_lookup_tcp,
6361 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6362 .arg1_type = ARG_PTR_TO_CTX,
6363 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6364 .arg3_type = ARG_CONST_SIZE,
6365 .arg4_type = ARG_ANYTHING,
6366 .arg5_type = ARG_ANYTHING,
6369 BPF_CALL_5(bpf_tc_sk_lookup_tcp, struct sk_buff *, skb,
6370 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6372 struct net_device *dev = skb->dev;
6373 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6374 struct net *caller_net = dev_net(dev);
6376 return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6377 ifindex, IPPROTO_TCP, netns_id,
6381 static const struct bpf_func_proto bpf_tc_sk_lookup_tcp_proto = {
6382 .func = bpf_tc_sk_lookup_tcp,
6385 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6386 .arg1_type = ARG_PTR_TO_CTX,
6387 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6388 .arg3_type = ARG_CONST_SIZE,
6389 .arg4_type = ARG_ANYTHING,
6390 .arg5_type = ARG_ANYTHING,
6393 BPF_CALL_5(bpf_tc_sk_lookup_udp, struct sk_buff *, skb,
6394 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6396 struct net_device *dev = skb->dev;
6397 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6398 struct net *caller_net = dev_net(dev);
6400 return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6401 ifindex, IPPROTO_UDP, netns_id,
6405 static const struct bpf_func_proto bpf_tc_sk_lookup_udp_proto = {
6406 .func = bpf_tc_sk_lookup_udp,
6409 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6410 .arg1_type = ARG_PTR_TO_CTX,
6411 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6412 .arg3_type = ARG_CONST_SIZE,
6413 .arg4_type = ARG_ANYTHING,
6414 .arg5_type = ARG_ANYTHING,
6417 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6419 if (sk && sk_is_refcounted(sk))
6424 static const struct bpf_func_proto bpf_sk_release_proto = {
6425 .func = bpf_sk_release,
6427 .ret_type = RET_INTEGER,
6428 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6431 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6432 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6434 struct net_device *dev = ctx->rxq->dev;
6435 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6436 struct net *caller_net = dev_net(dev);
6438 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6439 ifindex, IPPROTO_UDP, netns_id,
6443 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6444 .func = bpf_xdp_sk_lookup_udp,
6447 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6448 .arg1_type = ARG_PTR_TO_CTX,
6449 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6450 .arg3_type = ARG_CONST_SIZE,
6451 .arg4_type = ARG_ANYTHING,
6452 .arg5_type = ARG_ANYTHING,
6455 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6456 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6458 struct net_device *dev = ctx->rxq->dev;
6459 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6460 struct net *caller_net = dev_net(dev);
6462 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6463 ifindex, IPPROTO_TCP, netns_id,
6467 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6468 .func = bpf_xdp_skc_lookup_tcp,
6471 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6472 .arg1_type = ARG_PTR_TO_CTX,
6473 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6474 .arg3_type = ARG_CONST_SIZE,
6475 .arg4_type = ARG_ANYTHING,
6476 .arg5_type = ARG_ANYTHING,
6479 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6480 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6482 struct net_device *dev = ctx->rxq->dev;
6483 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6484 struct net *caller_net = dev_net(dev);
6486 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6487 ifindex, IPPROTO_TCP, netns_id,
6491 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6492 .func = bpf_xdp_sk_lookup_tcp,
6495 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6496 .arg1_type = ARG_PTR_TO_CTX,
6497 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6498 .arg3_type = ARG_CONST_SIZE,
6499 .arg4_type = ARG_ANYTHING,
6500 .arg5_type = ARG_ANYTHING,
6503 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6504 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6506 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6507 sock_net(ctx->sk), 0,
6508 IPPROTO_TCP, netns_id, flags,
6512 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6513 .func = bpf_sock_addr_skc_lookup_tcp,
6515 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6516 .arg1_type = ARG_PTR_TO_CTX,
6517 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6518 .arg3_type = ARG_CONST_SIZE,
6519 .arg4_type = ARG_ANYTHING,
6520 .arg5_type = ARG_ANYTHING,
6523 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6524 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6526 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6527 sock_net(ctx->sk), 0, IPPROTO_TCP,
6528 netns_id, flags, -1);
6531 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6532 .func = bpf_sock_addr_sk_lookup_tcp,
6534 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6535 .arg1_type = ARG_PTR_TO_CTX,
6536 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6537 .arg3_type = ARG_CONST_SIZE,
6538 .arg4_type = ARG_ANYTHING,
6539 .arg5_type = ARG_ANYTHING,
6542 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6543 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6545 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6546 sock_net(ctx->sk), 0, IPPROTO_UDP,
6547 netns_id, flags, -1);
6550 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6551 .func = bpf_sock_addr_sk_lookup_udp,
6553 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6554 .arg1_type = ARG_PTR_TO_CTX,
6555 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6556 .arg3_type = ARG_CONST_SIZE,
6557 .arg4_type = ARG_ANYTHING,
6558 .arg5_type = ARG_ANYTHING,
6561 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6562 struct bpf_insn_access_aux *info)
6564 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6568 if (off % size != 0)
6572 case offsetof(struct bpf_tcp_sock, bytes_received):
6573 case offsetof(struct bpf_tcp_sock, bytes_acked):
6574 return size == sizeof(__u64);
6576 return size == sizeof(__u32);
6580 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6581 const struct bpf_insn *si,
6582 struct bpf_insn *insn_buf,
6583 struct bpf_prog *prog, u32 *target_size)
6585 struct bpf_insn *insn = insn_buf;
6587 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \
6589 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \
6590 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6591 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6592 si->dst_reg, si->src_reg, \
6593 offsetof(struct tcp_sock, FIELD)); \
6596 #define BPF_INET_SOCK_GET_COMMON(FIELD) \
6598 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \
6600 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6601 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6602 struct inet_connection_sock, \
6604 si->dst_reg, si->src_reg, \
6606 struct inet_connection_sock, \
6610 if (insn > insn_buf)
6611 return insn - insn_buf;
6614 case offsetof(struct bpf_tcp_sock, rtt_min):
6615 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
6616 sizeof(struct minmax));
6617 BUILD_BUG_ON(sizeof(struct minmax) <
6618 sizeof(struct minmax_sample));
6620 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6621 offsetof(struct tcp_sock, rtt_min) +
6622 offsetof(struct minmax_sample, v));
6624 case offsetof(struct bpf_tcp_sock, snd_cwnd):
6625 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
6627 case offsetof(struct bpf_tcp_sock, srtt_us):
6628 BPF_TCP_SOCK_GET_COMMON(srtt_us);
6630 case offsetof(struct bpf_tcp_sock, snd_ssthresh):
6631 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
6633 case offsetof(struct bpf_tcp_sock, rcv_nxt):
6634 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
6636 case offsetof(struct bpf_tcp_sock, snd_nxt):
6637 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
6639 case offsetof(struct bpf_tcp_sock, snd_una):
6640 BPF_TCP_SOCK_GET_COMMON(snd_una);
6642 case offsetof(struct bpf_tcp_sock, mss_cache):
6643 BPF_TCP_SOCK_GET_COMMON(mss_cache);
6645 case offsetof(struct bpf_tcp_sock, ecn_flags):
6646 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
6648 case offsetof(struct bpf_tcp_sock, rate_delivered):
6649 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
6651 case offsetof(struct bpf_tcp_sock, rate_interval_us):
6652 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
6654 case offsetof(struct bpf_tcp_sock, packets_out):
6655 BPF_TCP_SOCK_GET_COMMON(packets_out);
6657 case offsetof(struct bpf_tcp_sock, retrans_out):
6658 BPF_TCP_SOCK_GET_COMMON(retrans_out);
6660 case offsetof(struct bpf_tcp_sock, total_retrans):
6661 BPF_TCP_SOCK_GET_COMMON(total_retrans);
6663 case offsetof(struct bpf_tcp_sock, segs_in):
6664 BPF_TCP_SOCK_GET_COMMON(segs_in);
6666 case offsetof(struct bpf_tcp_sock, data_segs_in):
6667 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
6669 case offsetof(struct bpf_tcp_sock, segs_out):
6670 BPF_TCP_SOCK_GET_COMMON(segs_out);
6672 case offsetof(struct bpf_tcp_sock, data_segs_out):
6673 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
6675 case offsetof(struct bpf_tcp_sock, lost_out):
6676 BPF_TCP_SOCK_GET_COMMON(lost_out);
6678 case offsetof(struct bpf_tcp_sock, sacked_out):
6679 BPF_TCP_SOCK_GET_COMMON(sacked_out);
6681 case offsetof(struct bpf_tcp_sock, bytes_received):
6682 BPF_TCP_SOCK_GET_COMMON(bytes_received);
6684 case offsetof(struct bpf_tcp_sock, bytes_acked):
6685 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
6687 case offsetof(struct bpf_tcp_sock, dsack_dups):
6688 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
6690 case offsetof(struct bpf_tcp_sock, delivered):
6691 BPF_TCP_SOCK_GET_COMMON(delivered);
6693 case offsetof(struct bpf_tcp_sock, delivered_ce):
6694 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
6696 case offsetof(struct bpf_tcp_sock, icsk_retransmits):
6697 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
6701 return insn - insn_buf;
6704 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
6706 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
6707 return (unsigned long)sk;
6709 return (unsigned long)NULL;
6712 const struct bpf_func_proto bpf_tcp_sock_proto = {
6713 .func = bpf_tcp_sock,
6715 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
6716 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6719 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
6721 sk = sk_to_full_sk(sk);
6723 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
6724 return (unsigned long)sk;
6726 return (unsigned long)NULL;
6729 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
6730 .func = bpf_get_listener_sock,
6732 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6733 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6736 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
6738 unsigned int iphdr_len;
6740 switch (skb_protocol(skb, true)) {
6741 case cpu_to_be16(ETH_P_IP):
6742 iphdr_len = sizeof(struct iphdr);
6744 case cpu_to_be16(ETH_P_IPV6):
6745 iphdr_len = sizeof(struct ipv6hdr);
6751 if (skb_headlen(skb) < iphdr_len)
6754 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
6757 return INET_ECN_set_ce(skb);
6760 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6761 struct bpf_insn_access_aux *info)
6763 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
6766 if (off % size != 0)
6771 return size == sizeof(__u32);
6775 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
6776 const struct bpf_insn *si,
6777 struct bpf_insn *insn_buf,
6778 struct bpf_prog *prog, u32 *target_size)
6780 struct bpf_insn *insn = insn_buf;
6782 #define BPF_XDP_SOCK_GET(FIELD) \
6784 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \
6785 sizeof_field(struct bpf_xdp_sock, FIELD)); \
6786 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
6787 si->dst_reg, si->src_reg, \
6788 offsetof(struct xdp_sock, FIELD)); \
6792 case offsetof(struct bpf_xdp_sock, queue_id):
6793 BPF_XDP_SOCK_GET(queue_id);
6797 return insn - insn_buf;
6800 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
6801 .func = bpf_skb_ecn_set_ce,
6803 .ret_type = RET_INTEGER,
6804 .arg1_type = ARG_PTR_TO_CTX,
6807 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6808 struct tcphdr *, th, u32, th_len)
6810 #ifdef CONFIG_SYN_COOKIES
6814 if (unlikely(!sk || th_len < sizeof(*th)))
6817 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
6818 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6821 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
6824 if (!th->ack || th->rst || th->syn)
6827 if (unlikely(iph_len < sizeof(struct iphdr)))
6830 if (tcp_synq_no_recent_overflow(sk))
6833 cookie = ntohl(th->ack_seq) - 1;
6835 /* Both struct iphdr and struct ipv6hdr have the version field at the
6836 * same offset so we can cast to the shorter header (struct iphdr).
6838 switch (((struct iphdr *)iph)->version) {
6840 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
6843 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
6846 #if IS_BUILTIN(CONFIG_IPV6)
6848 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6851 if (sk->sk_family != AF_INET6)
6854 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
6856 #endif /* CONFIG_IPV6 */
6859 return -EPROTONOSUPPORT;
6871 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
6872 .func = bpf_tcp_check_syncookie,
6875 .ret_type = RET_INTEGER,
6876 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6877 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6878 .arg3_type = ARG_CONST_SIZE,
6879 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6880 .arg5_type = ARG_CONST_SIZE,
6883 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6884 struct tcphdr *, th, u32, th_len)
6886 #ifdef CONFIG_SYN_COOKIES
6890 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
6893 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6896 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
6899 if (!th->syn || th->ack || th->fin || th->rst)
6902 if (unlikely(iph_len < sizeof(struct iphdr)))
6905 /* Both struct iphdr and struct ipv6hdr have the version field at the
6906 * same offset so we can cast to the shorter header (struct iphdr).
6908 switch (((struct iphdr *)iph)->version) {
6910 if (sk->sk_family == AF_INET6 && sk->sk_ipv6only)
6913 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
6916 #if IS_BUILTIN(CONFIG_IPV6)
6918 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6921 if (sk->sk_family != AF_INET6)
6924 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
6926 #endif /* CONFIG_IPV6 */
6929 return -EPROTONOSUPPORT;
6934 return cookie | ((u64)mss << 32);
6937 #endif /* CONFIG_SYN_COOKIES */
6940 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
6941 .func = bpf_tcp_gen_syncookie,
6942 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */
6944 .ret_type = RET_INTEGER,
6945 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6946 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6947 .arg3_type = ARG_CONST_SIZE,
6948 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6949 .arg5_type = ARG_CONST_SIZE,
6952 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
6954 if (!sk || flags != 0)
6956 if (!skb_at_tc_ingress(skb))
6958 if (unlikely(dev_net(skb->dev) != sock_net(sk)))
6959 return -ENETUNREACH;
6960 if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
6961 return -ESOCKTNOSUPPORT;
6962 if (sk_unhashed(sk))
6964 if (sk_is_refcounted(sk) &&
6965 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
6970 skb->destructor = sock_pfree;
6975 static const struct bpf_func_proto bpf_sk_assign_proto = {
6976 .func = bpf_sk_assign,
6978 .ret_type = RET_INTEGER,
6979 .arg1_type = ARG_PTR_TO_CTX,
6980 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6981 .arg3_type = ARG_ANYTHING,
6984 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
6985 u8 search_kind, const u8 *magic,
6986 u8 magic_len, bool *eol)
6992 while (op < opend) {
6995 if (kind == TCPOPT_EOL) {
6997 return ERR_PTR(-ENOMSG);
6998 } else if (kind == TCPOPT_NOP) {
7003 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
7004 /* Something is wrong in the received header.
7005 * Follow the TCP stack's tcp_parse_options()
7006 * and just bail here.
7008 return ERR_PTR(-EFAULT);
7011 if (search_kind == kind) {
7015 if (magic_len > kind_len - 2)
7016 return ERR_PTR(-ENOMSG);
7018 if (!memcmp(&op[2], magic, magic_len))
7025 return ERR_PTR(-ENOMSG);
7028 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7029 void *, search_res, u32, len, u64, flags)
7031 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
7032 const u8 *op, *opend, *magic, *search = search_res;
7033 u8 search_kind, search_len, copy_len, magic_len;
7036 /* 2 byte is the minimal option len except TCPOPT_NOP and
7037 * TCPOPT_EOL which are useless for the bpf prog to learn
7038 * and this helper disallow loading them also.
7040 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
7043 search_kind = search[0];
7044 search_len = search[1];
7046 if (search_len > len || search_kind == TCPOPT_NOP ||
7047 search_kind == TCPOPT_EOL)
7050 if (search_kind == TCPOPT_EXP || search_kind == 253) {
7051 /* 16 or 32 bit magic. +2 for kind and kind length */
7052 if (search_len != 4 && search_len != 6)
7055 magic_len = search_len - 2;
7064 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
7069 op += sizeof(struct tcphdr);
7071 if (!bpf_sock->skb ||
7072 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7073 /* This bpf_sock->op cannot call this helper */
7076 opend = bpf_sock->skb_data_end;
7077 op = bpf_sock->skb->data + sizeof(struct tcphdr);
7080 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
7087 if (copy_len > len) {
7092 memcpy(search_res, op, copy_len);
7096 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
7097 .func = bpf_sock_ops_load_hdr_opt,
7099 .ret_type = RET_INTEGER,
7100 .arg1_type = ARG_PTR_TO_CTX,
7101 .arg2_type = ARG_PTR_TO_MEM,
7102 .arg3_type = ARG_CONST_SIZE,
7103 .arg4_type = ARG_ANYTHING,
7106 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7107 const void *, from, u32, len, u64, flags)
7109 u8 new_kind, new_kind_len, magic_len = 0, *opend;
7110 const u8 *op, *new_op, *magic = NULL;
7111 struct sk_buff *skb;
7114 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
7117 if (len < 2 || flags)
7121 new_kind = new_op[0];
7122 new_kind_len = new_op[1];
7124 if (new_kind_len > len || new_kind == TCPOPT_NOP ||
7125 new_kind == TCPOPT_EOL)
7128 if (new_kind_len > bpf_sock->remaining_opt_len)
7131 /* 253 is another experimental kind */
7132 if (new_kind == TCPOPT_EXP || new_kind == 253) {
7133 if (new_kind_len < 4)
7135 /* Match for the 2 byte magic also.
7136 * RFC 6994: the magic could be 2 or 4 bytes.
7137 * Hence, matching by 2 byte only is on the
7138 * conservative side but it is the right
7139 * thing to do for the 'search-for-duplication'
7146 /* Check for duplication */
7147 skb = bpf_sock->skb;
7148 op = skb->data + sizeof(struct tcphdr);
7149 opend = bpf_sock->skb_data_end;
7151 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
7156 if (PTR_ERR(op) != -ENOMSG)
7160 /* The option has been ended. Treat it as no more
7161 * header option can be written.
7165 /* No duplication found. Store the header option. */
7166 memcpy(opend, from, new_kind_len);
7168 bpf_sock->remaining_opt_len -= new_kind_len;
7169 bpf_sock->skb_data_end += new_kind_len;
7174 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
7175 .func = bpf_sock_ops_store_hdr_opt,
7177 .ret_type = RET_INTEGER,
7178 .arg1_type = ARG_PTR_TO_CTX,
7179 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7180 .arg3_type = ARG_CONST_SIZE,
7181 .arg4_type = ARG_ANYTHING,
7184 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7185 u32, len, u64, flags)
7187 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7190 if (flags || len < 2)
7193 if (len > bpf_sock->remaining_opt_len)
7196 bpf_sock->remaining_opt_len -= len;
7201 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7202 .func = bpf_sock_ops_reserve_hdr_opt,
7204 .ret_type = RET_INTEGER,
7205 .arg1_type = ARG_PTR_TO_CTX,
7206 .arg2_type = ARG_ANYTHING,
7207 .arg3_type = ARG_ANYTHING,
7210 #endif /* CONFIG_INET */
7212 bool bpf_helper_changes_pkt_data(void *func)
7214 if (func == bpf_skb_vlan_push ||
7215 func == bpf_skb_vlan_pop ||
7216 func == bpf_skb_store_bytes ||
7217 func == bpf_skb_change_proto ||
7218 func == bpf_skb_change_head ||
7219 func == sk_skb_change_head ||
7220 func == bpf_skb_change_tail ||
7221 func == sk_skb_change_tail ||
7222 func == bpf_skb_adjust_room ||
7223 func == sk_skb_adjust_room ||
7224 func == bpf_skb_pull_data ||
7225 func == sk_skb_pull_data ||
7226 func == bpf_clone_redirect ||
7227 func == bpf_l3_csum_replace ||
7228 func == bpf_l4_csum_replace ||
7229 func == bpf_xdp_adjust_head ||
7230 func == bpf_xdp_adjust_meta ||
7231 func == bpf_msg_pull_data ||
7232 func == bpf_msg_push_data ||
7233 func == bpf_msg_pop_data ||
7234 func == bpf_xdp_adjust_tail ||
7235 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7236 func == bpf_lwt_seg6_store_bytes ||
7237 func == bpf_lwt_seg6_adjust_srh ||
7238 func == bpf_lwt_seg6_action ||
7241 func == bpf_sock_ops_store_hdr_opt ||
7243 func == bpf_lwt_in_push_encap ||
7244 func == bpf_lwt_xmit_push_encap)
7250 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7251 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7253 static const struct bpf_func_proto *
7254 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7257 /* inet and inet6 sockets are created in a process
7258 * context so there is always a valid uid/gid
7260 case BPF_FUNC_get_current_uid_gid:
7261 return &bpf_get_current_uid_gid_proto;
7262 case BPF_FUNC_get_local_storage:
7263 return &bpf_get_local_storage_proto;
7264 case BPF_FUNC_get_socket_cookie:
7265 return &bpf_get_socket_cookie_sock_proto;
7266 case BPF_FUNC_get_netns_cookie:
7267 return &bpf_get_netns_cookie_sock_proto;
7268 case BPF_FUNC_perf_event_output:
7269 return &bpf_event_output_data_proto;
7270 case BPF_FUNC_get_current_pid_tgid:
7271 return &bpf_get_current_pid_tgid_proto;
7272 case BPF_FUNC_get_current_comm:
7273 return &bpf_get_current_comm_proto;
7274 #ifdef CONFIG_CGROUPS
7275 case BPF_FUNC_get_current_cgroup_id:
7276 return &bpf_get_current_cgroup_id_proto;
7277 case BPF_FUNC_get_current_ancestor_cgroup_id:
7278 return &bpf_get_current_ancestor_cgroup_id_proto;
7280 #ifdef CONFIG_CGROUP_NET_CLASSID
7281 case BPF_FUNC_get_cgroup_classid:
7282 return &bpf_get_cgroup_classid_curr_proto;
7284 case BPF_FUNC_sk_storage_get:
7285 return &bpf_sk_storage_get_cg_sock_proto;
7286 case BPF_FUNC_ktime_get_coarse_ns:
7287 return &bpf_ktime_get_coarse_ns_proto;
7289 return bpf_base_func_proto(func_id);
7293 static const struct bpf_func_proto *
7294 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7297 /* inet and inet6 sockets are created in a process
7298 * context so there is always a valid uid/gid
7300 case BPF_FUNC_get_current_uid_gid:
7301 return &bpf_get_current_uid_gid_proto;
7303 switch (prog->expected_attach_type) {
7304 case BPF_CGROUP_INET4_CONNECT:
7305 case BPF_CGROUP_INET6_CONNECT:
7306 return &bpf_bind_proto;
7310 case BPF_FUNC_get_socket_cookie:
7311 return &bpf_get_socket_cookie_sock_addr_proto;
7312 case BPF_FUNC_get_netns_cookie:
7313 return &bpf_get_netns_cookie_sock_addr_proto;
7314 case BPF_FUNC_get_local_storage:
7315 return &bpf_get_local_storage_proto;
7316 case BPF_FUNC_perf_event_output:
7317 return &bpf_event_output_data_proto;
7318 case BPF_FUNC_get_current_pid_tgid:
7319 return &bpf_get_current_pid_tgid_proto;
7320 case BPF_FUNC_get_current_comm:
7321 return &bpf_get_current_comm_proto;
7322 #ifdef CONFIG_CGROUPS
7323 case BPF_FUNC_get_current_cgroup_id:
7324 return &bpf_get_current_cgroup_id_proto;
7325 case BPF_FUNC_get_current_ancestor_cgroup_id:
7326 return &bpf_get_current_ancestor_cgroup_id_proto;
7328 #ifdef CONFIG_CGROUP_NET_CLASSID
7329 case BPF_FUNC_get_cgroup_classid:
7330 return &bpf_get_cgroup_classid_curr_proto;
7333 case BPF_FUNC_sk_lookup_tcp:
7334 return &bpf_sock_addr_sk_lookup_tcp_proto;
7335 case BPF_FUNC_sk_lookup_udp:
7336 return &bpf_sock_addr_sk_lookup_udp_proto;
7337 case BPF_FUNC_sk_release:
7338 return &bpf_sk_release_proto;
7339 case BPF_FUNC_skc_lookup_tcp:
7340 return &bpf_sock_addr_skc_lookup_tcp_proto;
7341 #endif /* CONFIG_INET */
7342 case BPF_FUNC_sk_storage_get:
7343 return &bpf_sk_storage_get_proto;
7344 case BPF_FUNC_sk_storage_delete:
7345 return &bpf_sk_storage_delete_proto;
7346 case BPF_FUNC_setsockopt:
7347 switch (prog->expected_attach_type) {
7348 case BPF_CGROUP_INET4_BIND:
7349 case BPF_CGROUP_INET6_BIND:
7350 case BPF_CGROUP_INET4_CONNECT:
7351 case BPF_CGROUP_INET6_CONNECT:
7352 case BPF_CGROUP_UDP4_RECVMSG:
7353 case BPF_CGROUP_UDP6_RECVMSG:
7354 case BPF_CGROUP_UDP4_SENDMSG:
7355 case BPF_CGROUP_UDP6_SENDMSG:
7356 case BPF_CGROUP_INET4_GETPEERNAME:
7357 case BPF_CGROUP_INET6_GETPEERNAME:
7358 case BPF_CGROUP_INET4_GETSOCKNAME:
7359 case BPF_CGROUP_INET6_GETSOCKNAME:
7360 return &bpf_sock_addr_setsockopt_proto;
7364 case BPF_FUNC_getsockopt:
7365 switch (prog->expected_attach_type) {
7366 case BPF_CGROUP_INET4_BIND:
7367 case BPF_CGROUP_INET6_BIND:
7368 case BPF_CGROUP_INET4_CONNECT:
7369 case BPF_CGROUP_INET6_CONNECT:
7370 case BPF_CGROUP_UDP4_RECVMSG:
7371 case BPF_CGROUP_UDP6_RECVMSG:
7372 case BPF_CGROUP_UDP4_SENDMSG:
7373 case BPF_CGROUP_UDP6_SENDMSG:
7374 case BPF_CGROUP_INET4_GETPEERNAME:
7375 case BPF_CGROUP_INET6_GETPEERNAME:
7376 case BPF_CGROUP_INET4_GETSOCKNAME:
7377 case BPF_CGROUP_INET6_GETSOCKNAME:
7378 return &bpf_sock_addr_getsockopt_proto;
7383 return bpf_sk_base_func_proto(func_id);
7387 static const struct bpf_func_proto *
7388 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7391 case BPF_FUNC_skb_load_bytes:
7392 return &bpf_skb_load_bytes_proto;
7393 case BPF_FUNC_skb_load_bytes_relative:
7394 return &bpf_skb_load_bytes_relative_proto;
7395 case BPF_FUNC_get_socket_cookie:
7396 return &bpf_get_socket_cookie_proto;
7397 case BPF_FUNC_get_socket_uid:
7398 return &bpf_get_socket_uid_proto;
7399 case BPF_FUNC_perf_event_output:
7400 return &bpf_skb_event_output_proto;
7402 return bpf_sk_base_func_proto(func_id);
7406 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7407 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7409 static const struct bpf_func_proto *
7410 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7413 case BPF_FUNC_get_local_storage:
7414 return &bpf_get_local_storage_proto;
7415 case BPF_FUNC_sk_fullsock:
7416 return &bpf_sk_fullsock_proto;
7417 case BPF_FUNC_sk_storage_get:
7418 return &bpf_sk_storage_get_proto;
7419 case BPF_FUNC_sk_storage_delete:
7420 return &bpf_sk_storage_delete_proto;
7421 case BPF_FUNC_perf_event_output:
7422 return &bpf_skb_event_output_proto;
7423 #ifdef CONFIG_SOCK_CGROUP_DATA
7424 case BPF_FUNC_skb_cgroup_id:
7425 return &bpf_skb_cgroup_id_proto;
7426 case BPF_FUNC_skb_ancestor_cgroup_id:
7427 return &bpf_skb_ancestor_cgroup_id_proto;
7428 case BPF_FUNC_sk_cgroup_id:
7429 return &bpf_sk_cgroup_id_proto;
7430 case BPF_FUNC_sk_ancestor_cgroup_id:
7431 return &bpf_sk_ancestor_cgroup_id_proto;
7434 case BPF_FUNC_sk_lookup_tcp:
7435 return &bpf_sk_lookup_tcp_proto;
7436 case BPF_FUNC_sk_lookup_udp:
7437 return &bpf_sk_lookup_udp_proto;
7438 case BPF_FUNC_sk_release:
7439 return &bpf_sk_release_proto;
7440 case BPF_FUNC_skc_lookup_tcp:
7441 return &bpf_skc_lookup_tcp_proto;
7442 case BPF_FUNC_tcp_sock:
7443 return &bpf_tcp_sock_proto;
7444 case BPF_FUNC_get_listener_sock:
7445 return &bpf_get_listener_sock_proto;
7446 case BPF_FUNC_skb_ecn_set_ce:
7447 return &bpf_skb_ecn_set_ce_proto;
7450 return sk_filter_func_proto(func_id, prog);
7454 static const struct bpf_func_proto *
7455 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7458 case BPF_FUNC_skb_store_bytes:
7459 return &bpf_skb_store_bytes_proto;
7460 case BPF_FUNC_skb_load_bytes:
7461 return &bpf_skb_load_bytes_proto;
7462 case BPF_FUNC_skb_load_bytes_relative:
7463 return &bpf_skb_load_bytes_relative_proto;
7464 case BPF_FUNC_skb_pull_data:
7465 return &bpf_skb_pull_data_proto;
7466 case BPF_FUNC_csum_diff:
7467 return &bpf_csum_diff_proto;
7468 case BPF_FUNC_csum_update:
7469 return &bpf_csum_update_proto;
7470 case BPF_FUNC_csum_level:
7471 return &bpf_csum_level_proto;
7472 case BPF_FUNC_l3_csum_replace:
7473 return &bpf_l3_csum_replace_proto;
7474 case BPF_FUNC_l4_csum_replace:
7475 return &bpf_l4_csum_replace_proto;
7476 case BPF_FUNC_clone_redirect:
7477 return &bpf_clone_redirect_proto;
7478 case BPF_FUNC_get_cgroup_classid:
7479 return &bpf_get_cgroup_classid_proto;
7480 case BPF_FUNC_skb_vlan_push:
7481 return &bpf_skb_vlan_push_proto;
7482 case BPF_FUNC_skb_vlan_pop:
7483 return &bpf_skb_vlan_pop_proto;
7484 case BPF_FUNC_skb_change_proto:
7485 return &bpf_skb_change_proto_proto;
7486 case BPF_FUNC_skb_change_type:
7487 return &bpf_skb_change_type_proto;
7488 case BPF_FUNC_skb_adjust_room:
7489 return &bpf_skb_adjust_room_proto;
7490 case BPF_FUNC_skb_change_tail:
7491 return &bpf_skb_change_tail_proto;
7492 case BPF_FUNC_skb_change_head:
7493 return &bpf_skb_change_head_proto;
7494 case BPF_FUNC_skb_get_tunnel_key:
7495 return &bpf_skb_get_tunnel_key_proto;
7496 case BPF_FUNC_skb_set_tunnel_key:
7497 return bpf_get_skb_set_tunnel_proto(func_id);
7498 case BPF_FUNC_skb_get_tunnel_opt:
7499 return &bpf_skb_get_tunnel_opt_proto;
7500 case BPF_FUNC_skb_set_tunnel_opt:
7501 return bpf_get_skb_set_tunnel_proto(func_id);
7502 case BPF_FUNC_redirect:
7503 return &bpf_redirect_proto;
7504 case BPF_FUNC_redirect_neigh:
7505 return &bpf_redirect_neigh_proto;
7506 case BPF_FUNC_redirect_peer:
7507 return &bpf_redirect_peer_proto;
7508 case BPF_FUNC_get_route_realm:
7509 return &bpf_get_route_realm_proto;
7510 case BPF_FUNC_get_hash_recalc:
7511 return &bpf_get_hash_recalc_proto;
7512 case BPF_FUNC_set_hash_invalid:
7513 return &bpf_set_hash_invalid_proto;
7514 case BPF_FUNC_set_hash:
7515 return &bpf_set_hash_proto;
7516 case BPF_FUNC_perf_event_output:
7517 return &bpf_skb_event_output_proto;
7518 case BPF_FUNC_get_smp_processor_id:
7519 return &bpf_get_smp_processor_id_proto;
7520 case BPF_FUNC_skb_under_cgroup:
7521 return &bpf_skb_under_cgroup_proto;
7522 case BPF_FUNC_get_socket_cookie:
7523 return &bpf_get_socket_cookie_proto;
7524 case BPF_FUNC_get_socket_uid:
7525 return &bpf_get_socket_uid_proto;
7526 case BPF_FUNC_fib_lookup:
7527 return &bpf_skb_fib_lookup_proto;
7528 case BPF_FUNC_check_mtu:
7529 return &bpf_skb_check_mtu_proto;
7530 case BPF_FUNC_sk_fullsock:
7531 return &bpf_sk_fullsock_proto;
7532 case BPF_FUNC_sk_storage_get:
7533 return &bpf_sk_storage_get_proto;
7534 case BPF_FUNC_sk_storage_delete:
7535 return &bpf_sk_storage_delete_proto;
7537 case BPF_FUNC_skb_get_xfrm_state:
7538 return &bpf_skb_get_xfrm_state_proto;
7540 #ifdef CONFIG_CGROUP_NET_CLASSID
7541 case BPF_FUNC_skb_cgroup_classid:
7542 return &bpf_skb_cgroup_classid_proto;
7544 #ifdef CONFIG_SOCK_CGROUP_DATA
7545 case BPF_FUNC_skb_cgroup_id:
7546 return &bpf_skb_cgroup_id_proto;
7547 case BPF_FUNC_skb_ancestor_cgroup_id:
7548 return &bpf_skb_ancestor_cgroup_id_proto;
7551 case BPF_FUNC_sk_lookup_tcp:
7552 return &bpf_tc_sk_lookup_tcp_proto;
7553 case BPF_FUNC_sk_lookup_udp:
7554 return &bpf_tc_sk_lookup_udp_proto;
7555 case BPF_FUNC_sk_release:
7556 return &bpf_sk_release_proto;
7557 case BPF_FUNC_tcp_sock:
7558 return &bpf_tcp_sock_proto;
7559 case BPF_FUNC_get_listener_sock:
7560 return &bpf_get_listener_sock_proto;
7561 case BPF_FUNC_skc_lookup_tcp:
7562 return &bpf_tc_skc_lookup_tcp_proto;
7563 case BPF_FUNC_tcp_check_syncookie:
7564 return &bpf_tcp_check_syncookie_proto;
7565 case BPF_FUNC_skb_ecn_set_ce:
7566 return &bpf_skb_ecn_set_ce_proto;
7567 case BPF_FUNC_tcp_gen_syncookie:
7568 return &bpf_tcp_gen_syncookie_proto;
7569 case BPF_FUNC_sk_assign:
7570 return &bpf_sk_assign_proto;
7573 return bpf_sk_base_func_proto(func_id);
7577 static const struct bpf_func_proto *
7578 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7581 case BPF_FUNC_perf_event_output:
7582 return &bpf_xdp_event_output_proto;
7583 case BPF_FUNC_get_smp_processor_id:
7584 return &bpf_get_smp_processor_id_proto;
7585 case BPF_FUNC_csum_diff:
7586 return &bpf_csum_diff_proto;
7587 case BPF_FUNC_xdp_adjust_head:
7588 return &bpf_xdp_adjust_head_proto;
7589 case BPF_FUNC_xdp_adjust_meta:
7590 return &bpf_xdp_adjust_meta_proto;
7591 case BPF_FUNC_redirect:
7592 return &bpf_xdp_redirect_proto;
7593 case BPF_FUNC_redirect_map:
7594 return &bpf_xdp_redirect_map_proto;
7595 case BPF_FUNC_xdp_adjust_tail:
7596 return &bpf_xdp_adjust_tail_proto;
7597 case BPF_FUNC_fib_lookup:
7598 return &bpf_xdp_fib_lookup_proto;
7599 case BPF_FUNC_check_mtu:
7600 return &bpf_xdp_check_mtu_proto;
7602 case BPF_FUNC_sk_lookup_udp:
7603 return &bpf_xdp_sk_lookup_udp_proto;
7604 case BPF_FUNC_sk_lookup_tcp:
7605 return &bpf_xdp_sk_lookup_tcp_proto;
7606 case BPF_FUNC_sk_release:
7607 return &bpf_sk_release_proto;
7608 case BPF_FUNC_skc_lookup_tcp:
7609 return &bpf_xdp_skc_lookup_tcp_proto;
7610 case BPF_FUNC_tcp_check_syncookie:
7611 return &bpf_tcp_check_syncookie_proto;
7612 case BPF_FUNC_tcp_gen_syncookie:
7613 return &bpf_tcp_gen_syncookie_proto;
7616 return bpf_sk_base_func_proto(func_id);
7620 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
7621 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
7623 static const struct bpf_func_proto *
7624 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7627 case BPF_FUNC_setsockopt:
7628 return &bpf_sock_ops_setsockopt_proto;
7629 case BPF_FUNC_getsockopt:
7630 return &bpf_sock_ops_getsockopt_proto;
7631 case BPF_FUNC_sock_ops_cb_flags_set:
7632 return &bpf_sock_ops_cb_flags_set_proto;
7633 case BPF_FUNC_sock_map_update:
7634 return &bpf_sock_map_update_proto;
7635 case BPF_FUNC_sock_hash_update:
7636 return &bpf_sock_hash_update_proto;
7637 case BPF_FUNC_get_socket_cookie:
7638 return &bpf_get_socket_cookie_sock_ops_proto;
7639 case BPF_FUNC_get_local_storage:
7640 return &bpf_get_local_storage_proto;
7641 case BPF_FUNC_perf_event_output:
7642 return &bpf_event_output_data_proto;
7643 case BPF_FUNC_sk_storage_get:
7644 return &bpf_sk_storage_get_proto;
7645 case BPF_FUNC_sk_storage_delete:
7646 return &bpf_sk_storage_delete_proto;
7647 case BPF_FUNC_get_netns_cookie:
7648 return &bpf_get_netns_cookie_sock_ops_proto;
7650 case BPF_FUNC_load_hdr_opt:
7651 return &bpf_sock_ops_load_hdr_opt_proto;
7652 case BPF_FUNC_store_hdr_opt:
7653 return &bpf_sock_ops_store_hdr_opt_proto;
7654 case BPF_FUNC_reserve_hdr_opt:
7655 return &bpf_sock_ops_reserve_hdr_opt_proto;
7656 case BPF_FUNC_tcp_sock:
7657 return &bpf_tcp_sock_proto;
7658 #endif /* CONFIG_INET */
7660 return bpf_sk_base_func_proto(func_id);
7664 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
7665 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
7667 static const struct bpf_func_proto *
7668 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7671 case BPF_FUNC_msg_redirect_map:
7672 return &bpf_msg_redirect_map_proto;
7673 case BPF_FUNC_msg_redirect_hash:
7674 return &bpf_msg_redirect_hash_proto;
7675 case BPF_FUNC_msg_apply_bytes:
7676 return &bpf_msg_apply_bytes_proto;
7677 case BPF_FUNC_msg_cork_bytes:
7678 return &bpf_msg_cork_bytes_proto;
7679 case BPF_FUNC_msg_pull_data:
7680 return &bpf_msg_pull_data_proto;
7681 case BPF_FUNC_msg_push_data:
7682 return &bpf_msg_push_data_proto;
7683 case BPF_FUNC_msg_pop_data:
7684 return &bpf_msg_pop_data_proto;
7685 case BPF_FUNC_perf_event_output:
7686 return &bpf_event_output_data_proto;
7687 case BPF_FUNC_get_current_uid_gid:
7688 return &bpf_get_current_uid_gid_proto;
7689 case BPF_FUNC_get_current_pid_tgid:
7690 return &bpf_get_current_pid_tgid_proto;
7691 case BPF_FUNC_sk_storage_get:
7692 return &bpf_sk_storage_get_proto;
7693 case BPF_FUNC_sk_storage_delete:
7694 return &bpf_sk_storage_delete_proto;
7695 case BPF_FUNC_get_netns_cookie:
7696 return &bpf_get_netns_cookie_sk_msg_proto;
7697 #ifdef CONFIG_CGROUPS
7698 case BPF_FUNC_get_current_cgroup_id:
7699 return &bpf_get_current_cgroup_id_proto;
7700 case BPF_FUNC_get_current_ancestor_cgroup_id:
7701 return &bpf_get_current_ancestor_cgroup_id_proto;
7703 #ifdef CONFIG_CGROUP_NET_CLASSID
7704 case BPF_FUNC_get_cgroup_classid:
7705 return &bpf_get_cgroup_classid_curr_proto;
7708 return bpf_sk_base_func_proto(func_id);
7712 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
7713 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
7715 static const struct bpf_func_proto *
7716 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7719 case BPF_FUNC_skb_store_bytes:
7720 return &bpf_skb_store_bytes_proto;
7721 case BPF_FUNC_skb_load_bytes:
7722 return &bpf_skb_load_bytes_proto;
7723 case BPF_FUNC_skb_pull_data:
7724 return &sk_skb_pull_data_proto;
7725 case BPF_FUNC_skb_change_tail:
7726 return &sk_skb_change_tail_proto;
7727 case BPF_FUNC_skb_change_head:
7728 return &sk_skb_change_head_proto;
7729 case BPF_FUNC_skb_adjust_room:
7730 return &sk_skb_adjust_room_proto;
7731 case BPF_FUNC_get_socket_cookie:
7732 return &bpf_get_socket_cookie_proto;
7733 case BPF_FUNC_get_socket_uid:
7734 return &bpf_get_socket_uid_proto;
7735 case BPF_FUNC_sk_redirect_map:
7736 return &bpf_sk_redirect_map_proto;
7737 case BPF_FUNC_sk_redirect_hash:
7738 return &bpf_sk_redirect_hash_proto;
7739 case BPF_FUNC_perf_event_output:
7740 return &bpf_skb_event_output_proto;
7742 case BPF_FUNC_sk_lookup_tcp:
7743 return &bpf_sk_lookup_tcp_proto;
7744 case BPF_FUNC_sk_lookup_udp:
7745 return &bpf_sk_lookup_udp_proto;
7746 case BPF_FUNC_sk_release:
7747 return &bpf_sk_release_proto;
7748 case BPF_FUNC_skc_lookup_tcp:
7749 return &bpf_skc_lookup_tcp_proto;
7752 return bpf_sk_base_func_proto(func_id);
7756 static const struct bpf_func_proto *
7757 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7760 case BPF_FUNC_skb_load_bytes:
7761 return &bpf_flow_dissector_load_bytes_proto;
7763 return bpf_sk_base_func_proto(func_id);
7767 static const struct bpf_func_proto *
7768 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7771 case BPF_FUNC_skb_load_bytes:
7772 return &bpf_skb_load_bytes_proto;
7773 case BPF_FUNC_skb_pull_data:
7774 return &bpf_skb_pull_data_proto;
7775 case BPF_FUNC_csum_diff:
7776 return &bpf_csum_diff_proto;
7777 case BPF_FUNC_get_cgroup_classid:
7778 return &bpf_get_cgroup_classid_proto;
7779 case BPF_FUNC_get_route_realm:
7780 return &bpf_get_route_realm_proto;
7781 case BPF_FUNC_get_hash_recalc:
7782 return &bpf_get_hash_recalc_proto;
7783 case BPF_FUNC_perf_event_output:
7784 return &bpf_skb_event_output_proto;
7785 case BPF_FUNC_get_smp_processor_id:
7786 return &bpf_get_smp_processor_id_proto;
7787 case BPF_FUNC_skb_under_cgroup:
7788 return &bpf_skb_under_cgroup_proto;
7790 return bpf_sk_base_func_proto(func_id);
7794 static const struct bpf_func_proto *
7795 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7798 case BPF_FUNC_lwt_push_encap:
7799 return &bpf_lwt_in_push_encap_proto;
7801 return lwt_out_func_proto(func_id, prog);
7805 static const struct bpf_func_proto *
7806 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7809 case BPF_FUNC_skb_get_tunnel_key:
7810 return &bpf_skb_get_tunnel_key_proto;
7811 case BPF_FUNC_skb_set_tunnel_key:
7812 return bpf_get_skb_set_tunnel_proto(func_id);
7813 case BPF_FUNC_skb_get_tunnel_opt:
7814 return &bpf_skb_get_tunnel_opt_proto;
7815 case BPF_FUNC_skb_set_tunnel_opt:
7816 return bpf_get_skb_set_tunnel_proto(func_id);
7817 case BPF_FUNC_redirect:
7818 return &bpf_redirect_proto;
7819 case BPF_FUNC_clone_redirect:
7820 return &bpf_clone_redirect_proto;
7821 case BPF_FUNC_skb_change_tail:
7822 return &bpf_skb_change_tail_proto;
7823 case BPF_FUNC_skb_change_head:
7824 return &bpf_skb_change_head_proto;
7825 case BPF_FUNC_skb_store_bytes:
7826 return &bpf_skb_store_bytes_proto;
7827 case BPF_FUNC_csum_update:
7828 return &bpf_csum_update_proto;
7829 case BPF_FUNC_csum_level:
7830 return &bpf_csum_level_proto;
7831 case BPF_FUNC_l3_csum_replace:
7832 return &bpf_l3_csum_replace_proto;
7833 case BPF_FUNC_l4_csum_replace:
7834 return &bpf_l4_csum_replace_proto;
7835 case BPF_FUNC_set_hash_invalid:
7836 return &bpf_set_hash_invalid_proto;
7837 case BPF_FUNC_lwt_push_encap:
7838 return &bpf_lwt_xmit_push_encap_proto;
7840 return lwt_out_func_proto(func_id, prog);
7844 static const struct bpf_func_proto *
7845 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7848 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7849 case BPF_FUNC_lwt_seg6_store_bytes:
7850 return &bpf_lwt_seg6_store_bytes_proto;
7851 case BPF_FUNC_lwt_seg6_action:
7852 return &bpf_lwt_seg6_action_proto;
7853 case BPF_FUNC_lwt_seg6_adjust_srh:
7854 return &bpf_lwt_seg6_adjust_srh_proto;
7857 return lwt_out_func_proto(func_id, prog);
7861 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
7862 const struct bpf_prog *prog,
7863 struct bpf_insn_access_aux *info)
7865 const int size_default = sizeof(__u32);
7867 if (off < 0 || off >= sizeof(struct __sk_buff))
7870 /* The verifier guarantees that size > 0. */
7871 if (off % size != 0)
7875 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7876 if (off + size > offsetofend(struct __sk_buff, cb[4]))
7879 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
7880 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
7881 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
7882 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
7883 case bpf_ctx_range(struct __sk_buff, data):
7884 case bpf_ctx_range(struct __sk_buff, data_meta):
7885 case bpf_ctx_range(struct __sk_buff, data_end):
7886 if (size != size_default)
7889 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
7891 case bpf_ctx_range(struct __sk_buff, tstamp):
7892 if (size != sizeof(__u64))
7895 case offsetof(struct __sk_buff, sk):
7896 if (type == BPF_WRITE || size != sizeof(__u64))
7898 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
7901 /* Only narrow read access allowed for now. */
7902 if (type == BPF_WRITE) {
7903 if (size != size_default)
7906 bpf_ctx_record_field_size(info, size_default);
7907 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
7915 static bool sk_filter_is_valid_access(int off, int size,
7916 enum bpf_access_type type,
7917 const struct bpf_prog *prog,
7918 struct bpf_insn_access_aux *info)
7921 case bpf_ctx_range(struct __sk_buff, tc_classid):
7922 case bpf_ctx_range(struct __sk_buff, data):
7923 case bpf_ctx_range(struct __sk_buff, data_meta):
7924 case bpf_ctx_range(struct __sk_buff, data_end):
7925 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7926 case bpf_ctx_range(struct __sk_buff, tstamp):
7927 case bpf_ctx_range(struct __sk_buff, wire_len):
7931 if (type == BPF_WRITE) {
7933 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7940 return bpf_skb_is_valid_access(off, size, type, prog, info);
7943 static bool cg_skb_is_valid_access(int off, int size,
7944 enum bpf_access_type type,
7945 const struct bpf_prog *prog,
7946 struct bpf_insn_access_aux *info)
7949 case bpf_ctx_range(struct __sk_buff, tc_classid):
7950 case bpf_ctx_range(struct __sk_buff, data_meta):
7951 case bpf_ctx_range(struct __sk_buff, wire_len):
7953 case bpf_ctx_range(struct __sk_buff, data):
7954 case bpf_ctx_range(struct __sk_buff, data_end):
7960 if (type == BPF_WRITE) {
7962 case bpf_ctx_range(struct __sk_buff, mark):
7963 case bpf_ctx_range(struct __sk_buff, priority):
7964 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7966 case bpf_ctx_range(struct __sk_buff, tstamp):
7976 case bpf_ctx_range(struct __sk_buff, data):
7977 info->reg_type = PTR_TO_PACKET;
7979 case bpf_ctx_range(struct __sk_buff, data_end):
7980 info->reg_type = PTR_TO_PACKET_END;
7984 return bpf_skb_is_valid_access(off, size, type, prog, info);
7987 static bool lwt_is_valid_access(int off, int size,
7988 enum bpf_access_type type,
7989 const struct bpf_prog *prog,
7990 struct bpf_insn_access_aux *info)
7993 case bpf_ctx_range(struct __sk_buff, tc_classid):
7994 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7995 case bpf_ctx_range(struct __sk_buff, data_meta):
7996 case bpf_ctx_range(struct __sk_buff, tstamp):
7997 case bpf_ctx_range(struct __sk_buff, wire_len):
8001 if (type == BPF_WRITE) {
8003 case bpf_ctx_range(struct __sk_buff, mark):
8004 case bpf_ctx_range(struct __sk_buff, priority):
8005 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8013 case bpf_ctx_range(struct __sk_buff, data):
8014 info->reg_type = PTR_TO_PACKET;
8016 case bpf_ctx_range(struct __sk_buff, data_end):
8017 info->reg_type = PTR_TO_PACKET_END;
8021 return bpf_skb_is_valid_access(off, size, type, prog, info);
8024 /* Attach type specific accesses */
8025 static bool __sock_filter_check_attach_type(int off,
8026 enum bpf_access_type access_type,
8027 enum bpf_attach_type attach_type)
8030 case offsetof(struct bpf_sock, bound_dev_if):
8031 case offsetof(struct bpf_sock, mark):
8032 case offsetof(struct bpf_sock, priority):
8033 switch (attach_type) {
8034 case BPF_CGROUP_INET_SOCK_CREATE:
8035 case BPF_CGROUP_INET_SOCK_RELEASE:
8040 case bpf_ctx_range(struct bpf_sock, src_ip4):
8041 switch (attach_type) {
8042 case BPF_CGROUP_INET4_POST_BIND:
8047 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8048 switch (attach_type) {
8049 case BPF_CGROUP_INET6_POST_BIND:
8054 case bpf_ctx_range(struct bpf_sock, src_port):
8055 switch (attach_type) {
8056 case BPF_CGROUP_INET4_POST_BIND:
8057 case BPF_CGROUP_INET6_POST_BIND:
8064 return access_type == BPF_READ;
8069 bool bpf_sock_common_is_valid_access(int off, int size,
8070 enum bpf_access_type type,
8071 struct bpf_insn_access_aux *info)
8074 case bpf_ctx_range_till(struct bpf_sock, type, priority):
8077 return bpf_sock_is_valid_access(off, size, type, info);
8081 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
8082 struct bpf_insn_access_aux *info)
8084 const int size_default = sizeof(__u32);
8087 if (off < 0 || off >= sizeof(struct bpf_sock))
8089 if (off % size != 0)
8093 case offsetof(struct bpf_sock, state):
8094 case offsetof(struct bpf_sock, family):
8095 case offsetof(struct bpf_sock, type):
8096 case offsetof(struct bpf_sock, protocol):
8097 case offsetof(struct bpf_sock, src_port):
8098 case offsetof(struct bpf_sock, rx_queue_mapping):
8099 case bpf_ctx_range(struct bpf_sock, src_ip4):
8100 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8101 case bpf_ctx_range(struct bpf_sock, dst_ip4):
8102 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8103 bpf_ctx_record_field_size(info, size_default);
8104 return bpf_ctx_narrow_access_ok(off, size, size_default);
8105 case bpf_ctx_range(struct bpf_sock, dst_port):
8106 field_size = size == size_default ?
8107 size_default : sizeof_field(struct bpf_sock, dst_port);
8108 bpf_ctx_record_field_size(info, field_size);
8109 return bpf_ctx_narrow_access_ok(off, size, field_size);
8110 case offsetofend(struct bpf_sock, dst_port) ...
8111 offsetof(struct bpf_sock, dst_ip4) - 1:
8115 return size == size_default;
8118 static bool sock_filter_is_valid_access(int off, int size,
8119 enum bpf_access_type type,
8120 const struct bpf_prog *prog,
8121 struct bpf_insn_access_aux *info)
8123 if (!bpf_sock_is_valid_access(off, size, type, info))
8125 return __sock_filter_check_attach_type(off, type,
8126 prog->expected_attach_type);
8129 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
8130 const struct bpf_prog *prog)
8132 /* Neither direct read nor direct write requires any preliminary
8138 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
8139 const struct bpf_prog *prog, int drop_verdict)
8141 struct bpf_insn *insn = insn_buf;
8146 /* if (!skb->cloned)
8149 * (Fast-path, otherwise approximation that we might be
8150 * a clone, do the rest in helper.)
8152 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
8153 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
8154 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
8156 /* ret = bpf_skb_pull_data(skb, 0); */
8157 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
8158 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
8159 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
8160 BPF_FUNC_skb_pull_data);
8163 * return TC_ACT_SHOT;
8165 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
8166 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
8167 *insn++ = BPF_EXIT_INSN();
8170 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
8172 *insn++ = prog->insnsi[0];
8174 return insn - insn_buf;
8177 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
8178 struct bpf_insn *insn_buf)
8180 bool indirect = BPF_MODE(orig->code) == BPF_IND;
8181 struct bpf_insn *insn = insn_buf;
8184 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
8186 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
8188 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
8190 /* We're guaranteed here that CTX is in R6. */
8191 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
8193 switch (BPF_SIZE(orig->code)) {
8195 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
8198 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
8201 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
8205 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
8206 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
8207 *insn++ = BPF_EXIT_INSN();
8209 return insn - insn_buf;
8212 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8213 const struct bpf_prog *prog)
8215 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8218 static bool tc_cls_act_is_valid_access(int off, int size,
8219 enum bpf_access_type type,
8220 const struct bpf_prog *prog,
8221 struct bpf_insn_access_aux *info)
8223 if (type == BPF_WRITE) {
8225 case bpf_ctx_range(struct __sk_buff, mark):
8226 case bpf_ctx_range(struct __sk_buff, tc_index):
8227 case bpf_ctx_range(struct __sk_buff, priority):
8228 case bpf_ctx_range(struct __sk_buff, tc_classid):
8229 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8230 case bpf_ctx_range(struct __sk_buff, tstamp):
8231 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8239 case bpf_ctx_range(struct __sk_buff, data):
8240 info->reg_type = PTR_TO_PACKET;
8242 case bpf_ctx_range(struct __sk_buff, data_meta):
8243 info->reg_type = PTR_TO_PACKET_META;
8245 case bpf_ctx_range(struct __sk_buff, data_end):
8246 info->reg_type = PTR_TO_PACKET_END;
8248 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8252 return bpf_skb_is_valid_access(off, size, type, prog, info);
8255 static bool __is_valid_xdp_access(int off, int size)
8257 if (off < 0 || off >= sizeof(struct xdp_md))
8259 if (off % size != 0)
8261 if (size != sizeof(__u32))
8267 static bool xdp_is_valid_access(int off, int size,
8268 enum bpf_access_type type,
8269 const struct bpf_prog *prog,
8270 struct bpf_insn_access_aux *info)
8272 if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8274 case offsetof(struct xdp_md, egress_ifindex):
8279 if (type == BPF_WRITE) {
8280 if (bpf_prog_is_dev_bound(prog->aux)) {
8282 case offsetof(struct xdp_md, rx_queue_index):
8283 return __is_valid_xdp_access(off, size);
8290 case offsetof(struct xdp_md, data):
8291 info->reg_type = PTR_TO_PACKET;
8293 case offsetof(struct xdp_md, data_meta):
8294 info->reg_type = PTR_TO_PACKET_META;
8296 case offsetof(struct xdp_md, data_end):
8297 info->reg_type = PTR_TO_PACKET_END;
8301 return __is_valid_xdp_access(off, size);
8304 void bpf_warn_invalid_xdp_action(u32 act)
8306 const u32 act_max = XDP_REDIRECT;
8308 pr_warn_once("%s XDP return value %u, expect packet loss!\n",
8309 act > act_max ? "Illegal" : "Driver unsupported",
8312 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
8314 static bool sock_addr_is_valid_access(int off, int size,
8315 enum bpf_access_type type,
8316 const struct bpf_prog *prog,
8317 struct bpf_insn_access_aux *info)
8319 const int size_default = sizeof(__u32);
8321 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
8323 if (off % size != 0)
8326 /* Disallow access to IPv6 fields from IPv4 contex and vise
8330 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8331 switch (prog->expected_attach_type) {
8332 case BPF_CGROUP_INET4_BIND:
8333 case BPF_CGROUP_INET4_CONNECT:
8334 case BPF_CGROUP_INET4_GETPEERNAME:
8335 case BPF_CGROUP_INET4_GETSOCKNAME:
8336 case BPF_CGROUP_UDP4_SENDMSG:
8337 case BPF_CGROUP_UDP4_RECVMSG:
8343 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8344 switch (prog->expected_attach_type) {
8345 case BPF_CGROUP_INET6_BIND:
8346 case BPF_CGROUP_INET6_CONNECT:
8347 case BPF_CGROUP_INET6_GETPEERNAME:
8348 case BPF_CGROUP_INET6_GETSOCKNAME:
8349 case BPF_CGROUP_UDP6_SENDMSG:
8350 case BPF_CGROUP_UDP6_RECVMSG:
8356 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8357 switch (prog->expected_attach_type) {
8358 case BPF_CGROUP_UDP4_SENDMSG:
8364 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8366 switch (prog->expected_attach_type) {
8367 case BPF_CGROUP_UDP6_SENDMSG:
8376 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8377 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8378 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8379 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8381 case bpf_ctx_range(struct bpf_sock_addr, user_port):
8382 if (type == BPF_READ) {
8383 bpf_ctx_record_field_size(info, size_default);
8385 if (bpf_ctx_wide_access_ok(off, size,
8386 struct bpf_sock_addr,
8390 if (bpf_ctx_wide_access_ok(off, size,
8391 struct bpf_sock_addr,
8395 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8398 if (bpf_ctx_wide_access_ok(off, size,
8399 struct bpf_sock_addr,
8403 if (bpf_ctx_wide_access_ok(off, size,
8404 struct bpf_sock_addr,
8408 if (size != size_default)
8412 case offsetof(struct bpf_sock_addr, sk):
8413 if (type != BPF_READ)
8415 if (size != sizeof(__u64))
8417 info->reg_type = PTR_TO_SOCKET;
8420 if (type == BPF_READ) {
8421 if (size != size_default)
8431 static bool sock_ops_is_valid_access(int off, int size,
8432 enum bpf_access_type type,
8433 const struct bpf_prog *prog,
8434 struct bpf_insn_access_aux *info)
8436 const int size_default = sizeof(__u32);
8438 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
8441 /* The verifier guarantees that size > 0. */
8442 if (off % size != 0)
8445 if (type == BPF_WRITE) {
8447 case offsetof(struct bpf_sock_ops, reply):
8448 case offsetof(struct bpf_sock_ops, sk_txhash):
8449 if (size != size_default)
8457 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
8459 if (size != sizeof(__u64))
8462 case offsetof(struct bpf_sock_ops, sk):
8463 if (size != sizeof(__u64))
8465 info->reg_type = PTR_TO_SOCKET_OR_NULL;
8467 case offsetof(struct bpf_sock_ops, skb_data):
8468 if (size != sizeof(__u64))
8470 info->reg_type = PTR_TO_PACKET;
8472 case offsetof(struct bpf_sock_ops, skb_data_end):
8473 if (size != sizeof(__u64))
8475 info->reg_type = PTR_TO_PACKET_END;
8477 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
8478 bpf_ctx_record_field_size(info, size_default);
8479 return bpf_ctx_narrow_access_ok(off, size,
8482 if (size != size_default)
8491 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
8492 const struct bpf_prog *prog)
8494 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
8497 static bool sk_skb_is_valid_access(int off, int size,
8498 enum bpf_access_type type,
8499 const struct bpf_prog *prog,
8500 struct bpf_insn_access_aux *info)
8503 case bpf_ctx_range(struct __sk_buff, tc_classid):
8504 case bpf_ctx_range(struct __sk_buff, data_meta):
8505 case bpf_ctx_range(struct __sk_buff, tstamp):
8506 case bpf_ctx_range(struct __sk_buff, wire_len):
8510 if (type == BPF_WRITE) {
8512 case bpf_ctx_range(struct __sk_buff, tc_index):
8513 case bpf_ctx_range(struct __sk_buff, priority):
8521 case bpf_ctx_range(struct __sk_buff, mark):
8523 case bpf_ctx_range(struct __sk_buff, data):
8524 info->reg_type = PTR_TO_PACKET;
8526 case bpf_ctx_range(struct __sk_buff, data_end):
8527 info->reg_type = PTR_TO_PACKET_END;
8531 return bpf_skb_is_valid_access(off, size, type, prog, info);
8534 static bool sk_msg_is_valid_access(int off, int size,
8535 enum bpf_access_type type,
8536 const struct bpf_prog *prog,
8537 struct bpf_insn_access_aux *info)
8539 if (type == BPF_WRITE)
8542 if (off % size != 0)
8546 case offsetof(struct sk_msg_md, data):
8547 info->reg_type = PTR_TO_PACKET;
8548 if (size != sizeof(__u64))
8551 case offsetof(struct sk_msg_md, data_end):
8552 info->reg_type = PTR_TO_PACKET_END;
8553 if (size != sizeof(__u64))
8556 case offsetof(struct sk_msg_md, sk):
8557 if (size != sizeof(__u64))
8559 info->reg_type = PTR_TO_SOCKET;
8561 case bpf_ctx_range(struct sk_msg_md, family):
8562 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
8563 case bpf_ctx_range(struct sk_msg_md, local_ip4):
8564 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
8565 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
8566 case bpf_ctx_range(struct sk_msg_md, remote_port):
8567 case bpf_ctx_range(struct sk_msg_md, local_port):
8568 case bpf_ctx_range(struct sk_msg_md, size):
8569 if (size != sizeof(__u32))
8578 static bool flow_dissector_is_valid_access(int off, int size,
8579 enum bpf_access_type type,
8580 const struct bpf_prog *prog,
8581 struct bpf_insn_access_aux *info)
8583 const int size_default = sizeof(__u32);
8585 if (off < 0 || off >= sizeof(struct __sk_buff))
8588 if (type == BPF_WRITE)
8592 case bpf_ctx_range(struct __sk_buff, data):
8593 if (size != size_default)
8595 info->reg_type = PTR_TO_PACKET;
8597 case bpf_ctx_range(struct __sk_buff, data_end):
8598 if (size != size_default)
8600 info->reg_type = PTR_TO_PACKET_END;
8602 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8603 if (size != sizeof(__u64))
8605 info->reg_type = PTR_TO_FLOW_KEYS;
8612 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
8613 const struct bpf_insn *si,
8614 struct bpf_insn *insn_buf,
8615 struct bpf_prog *prog,
8619 struct bpf_insn *insn = insn_buf;
8622 case offsetof(struct __sk_buff, data):
8623 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
8624 si->dst_reg, si->src_reg,
8625 offsetof(struct bpf_flow_dissector, data));
8628 case offsetof(struct __sk_buff, data_end):
8629 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
8630 si->dst_reg, si->src_reg,
8631 offsetof(struct bpf_flow_dissector, data_end));
8634 case offsetof(struct __sk_buff, flow_keys):
8635 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
8636 si->dst_reg, si->src_reg,
8637 offsetof(struct bpf_flow_dissector, flow_keys));
8641 return insn - insn_buf;
8644 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si,
8645 struct bpf_insn *insn)
8647 /* si->dst_reg = skb_shinfo(SKB); */
8648 #ifdef NET_SKBUFF_DATA_USES_OFFSET
8649 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8650 BPF_REG_AX, si->src_reg,
8651 offsetof(struct sk_buff, end));
8652 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
8653 si->dst_reg, si->src_reg,
8654 offsetof(struct sk_buff, head));
8655 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
8657 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8658 si->dst_reg, si->src_reg,
8659 offsetof(struct sk_buff, end));
8665 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
8666 const struct bpf_insn *si,
8667 struct bpf_insn *insn_buf,
8668 struct bpf_prog *prog, u32 *target_size)
8670 struct bpf_insn *insn = insn_buf;
8674 case offsetof(struct __sk_buff, len):
8675 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8676 bpf_target_off(struct sk_buff, len, 4,
8680 case offsetof(struct __sk_buff, protocol):
8681 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8682 bpf_target_off(struct sk_buff, protocol, 2,
8686 case offsetof(struct __sk_buff, vlan_proto):
8687 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8688 bpf_target_off(struct sk_buff, vlan_proto, 2,
8692 case offsetof(struct __sk_buff, priority):
8693 if (type == BPF_WRITE)
8694 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8695 bpf_target_off(struct sk_buff, priority, 4,
8698 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8699 bpf_target_off(struct sk_buff, priority, 4,
8703 case offsetof(struct __sk_buff, ingress_ifindex):
8704 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8705 bpf_target_off(struct sk_buff, skb_iif, 4,
8709 case offsetof(struct __sk_buff, ifindex):
8710 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8711 si->dst_reg, si->src_reg,
8712 offsetof(struct sk_buff, dev));
8713 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
8714 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8715 bpf_target_off(struct net_device, ifindex, 4,
8719 case offsetof(struct __sk_buff, hash):
8720 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8721 bpf_target_off(struct sk_buff, hash, 4,
8725 case offsetof(struct __sk_buff, mark):
8726 if (type == BPF_WRITE)
8727 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8728 bpf_target_off(struct sk_buff, mark, 4,
8731 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8732 bpf_target_off(struct sk_buff, mark, 4,
8736 case offsetof(struct __sk_buff, pkt_type):
8738 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8740 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
8741 #ifdef __BIG_ENDIAN_BITFIELD
8742 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
8746 case offsetof(struct __sk_buff, queue_mapping):
8747 if (type == BPF_WRITE) {
8748 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
8749 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8750 bpf_target_off(struct sk_buff,
8754 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8755 bpf_target_off(struct sk_buff,
8761 case offsetof(struct __sk_buff, vlan_present):
8763 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8764 PKT_VLAN_PRESENT_OFFSET());
8765 if (PKT_VLAN_PRESENT_BIT)
8766 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
8767 if (PKT_VLAN_PRESENT_BIT < 7)
8768 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
8771 case offsetof(struct __sk_buff, vlan_tci):
8772 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8773 bpf_target_off(struct sk_buff, vlan_tci, 2,
8777 case offsetof(struct __sk_buff, cb[0]) ...
8778 offsetofend(struct __sk_buff, cb[4]) - 1:
8779 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
8780 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
8781 offsetof(struct qdisc_skb_cb, data)) %
8784 prog->cb_access = 1;
8786 off -= offsetof(struct __sk_buff, cb[0]);
8787 off += offsetof(struct sk_buff, cb);
8788 off += offsetof(struct qdisc_skb_cb, data);
8789 if (type == BPF_WRITE)
8790 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
8793 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
8797 case offsetof(struct __sk_buff, tc_classid):
8798 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
8801 off -= offsetof(struct __sk_buff, tc_classid);
8802 off += offsetof(struct sk_buff, cb);
8803 off += offsetof(struct qdisc_skb_cb, tc_classid);
8805 if (type == BPF_WRITE)
8806 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
8809 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
8813 case offsetof(struct __sk_buff, data):
8814 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
8815 si->dst_reg, si->src_reg,
8816 offsetof(struct sk_buff, data));
8819 case offsetof(struct __sk_buff, data_meta):
8821 off -= offsetof(struct __sk_buff, data_meta);
8822 off += offsetof(struct sk_buff, cb);
8823 off += offsetof(struct bpf_skb_data_end, data_meta);
8824 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8828 case offsetof(struct __sk_buff, data_end):
8830 off -= offsetof(struct __sk_buff, data_end);
8831 off += offsetof(struct sk_buff, cb);
8832 off += offsetof(struct bpf_skb_data_end, data_end);
8833 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8837 case offsetof(struct __sk_buff, tc_index):
8838 #ifdef CONFIG_NET_SCHED
8839 if (type == BPF_WRITE)
8840 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8841 bpf_target_off(struct sk_buff, tc_index, 2,
8844 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8845 bpf_target_off(struct sk_buff, tc_index, 2,
8849 if (type == BPF_WRITE)
8850 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
8852 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8856 case offsetof(struct __sk_buff, napi_id):
8857 #if defined(CONFIG_NET_RX_BUSY_POLL)
8858 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8859 bpf_target_off(struct sk_buff, napi_id, 4,
8861 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
8862 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8865 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8868 case offsetof(struct __sk_buff, family):
8869 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
8871 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8872 si->dst_reg, si->src_reg,
8873 offsetof(struct sk_buff, sk));
8874 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8875 bpf_target_off(struct sock_common,
8879 case offsetof(struct __sk_buff, remote_ip4):
8880 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
8882 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8883 si->dst_reg, si->src_reg,
8884 offsetof(struct sk_buff, sk));
8885 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8886 bpf_target_off(struct sock_common,
8890 case offsetof(struct __sk_buff, local_ip4):
8891 BUILD_BUG_ON(sizeof_field(struct sock_common,
8892 skc_rcv_saddr) != 4);
8894 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8895 si->dst_reg, si->src_reg,
8896 offsetof(struct sk_buff, sk));
8897 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8898 bpf_target_off(struct sock_common,
8902 case offsetof(struct __sk_buff, remote_ip6[0]) ...
8903 offsetof(struct __sk_buff, remote_ip6[3]):
8904 #if IS_ENABLED(CONFIG_IPV6)
8905 BUILD_BUG_ON(sizeof_field(struct sock_common,
8906 skc_v6_daddr.s6_addr32[0]) != 4);
8909 off -= offsetof(struct __sk_buff, remote_ip6[0]);
8911 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8912 si->dst_reg, si->src_reg,
8913 offsetof(struct sk_buff, sk));
8914 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8915 offsetof(struct sock_common,
8916 skc_v6_daddr.s6_addr32[0]) +
8919 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8922 case offsetof(struct __sk_buff, local_ip6[0]) ...
8923 offsetof(struct __sk_buff, local_ip6[3]):
8924 #if IS_ENABLED(CONFIG_IPV6)
8925 BUILD_BUG_ON(sizeof_field(struct sock_common,
8926 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8929 off -= offsetof(struct __sk_buff, local_ip6[0]);
8931 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8932 si->dst_reg, si->src_reg,
8933 offsetof(struct sk_buff, sk));
8934 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8935 offsetof(struct sock_common,
8936 skc_v6_rcv_saddr.s6_addr32[0]) +
8939 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8943 case offsetof(struct __sk_buff, remote_port):
8944 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
8946 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8947 si->dst_reg, si->src_reg,
8948 offsetof(struct sk_buff, sk));
8949 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8950 bpf_target_off(struct sock_common,
8953 #ifndef __BIG_ENDIAN_BITFIELD
8954 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8958 case offsetof(struct __sk_buff, local_port):
8959 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
8961 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8962 si->dst_reg, si->src_reg,
8963 offsetof(struct sk_buff, sk));
8964 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8965 bpf_target_off(struct sock_common,
8966 skc_num, 2, target_size));
8969 case offsetof(struct __sk_buff, tstamp):
8970 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
8972 if (type == BPF_WRITE)
8973 *insn++ = BPF_STX_MEM(BPF_DW,
8974 si->dst_reg, si->src_reg,
8975 bpf_target_off(struct sk_buff,
8979 *insn++ = BPF_LDX_MEM(BPF_DW,
8980 si->dst_reg, si->src_reg,
8981 bpf_target_off(struct sk_buff,
8986 case offsetof(struct __sk_buff, gso_segs):
8987 insn = bpf_convert_shinfo_access(si, insn);
8988 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
8989 si->dst_reg, si->dst_reg,
8990 bpf_target_off(struct skb_shared_info,
8994 case offsetof(struct __sk_buff, gso_size):
8995 insn = bpf_convert_shinfo_access(si, insn);
8996 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
8997 si->dst_reg, si->dst_reg,
8998 bpf_target_off(struct skb_shared_info,
9002 case offsetof(struct __sk_buff, wire_len):
9003 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
9006 off -= offsetof(struct __sk_buff, wire_len);
9007 off += offsetof(struct sk_buff, cb);
9008 off += offsetof(struct qdisc_skb_cb, pkt_len);
9010 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
9013 case offsetof(struct __sk_buff, sk):
9014 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9015 si->dst_reg, si->src_reg,
9016 offsetof(struct sk_buff, sk));
9020 return insn - insn_buf;
9023 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
9024 const struct bpf_insn *si,
9025 struct bpf_insn *insn_buf,
9026 struct bpf_prog *prog, u32 *target_size)
9028 struct bpf_insn *insn = insn_buf;
9032 case offsetof(struct bpf_sock, bound_dev_if):
9033 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
9035 if (type == BPF_WRITE)
9036 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9037 offsetof(struct sock, sk_bound_dev_if));
9039 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9040 offsetof(struct sock, sk_bound_dev_if));
9043 case offsetof(struct bpf_sock, mark):
9044 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
9046 if (type == BPF_WRITE)
9047 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9048 offsetof(struct sock, sk_mark));
9050 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9051 offsetof(struct sock, sk_mark));
9054 case offsetof(struct bpf_sock, priority):
9055 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
9057 if (type == BPF_WRITE)
9058 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9059 offsetof(struct sock, sk_priority));
9061 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9062 offsetof(struct sock, sk_priority));
9065 case offsetof(struct bpf_sock, family):
9066 *insn++ = BPF_LDX_MEM(
9067 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
9068 si->dst_reg, si->src_reg,
9069 bpf_target_off(struct sock_common,
9071 sizeof_field(struct sock_common,
9076 case offsetof(struct bpf_sock, type):
9077 *insn++ = BPF_LDX_MEM(
9078 BPF_FIELD_SIZEOF(struct sock, sk_type),
9079 si->dst_reg, si->src_reg,
9080 bpf_target_off(struct sock, sk_type,
9081 sizeof_field(struct sock, sk_type),
9085 case offsetof(struct bpf_sock, protocol):
9086 *insn++ = BPF_LDX_MEM(
9087 BPF_FIELD_SIZEOF(struct sock, sk_protocol),
9088 si->dst_reg, si->src_reg,
9089 bpf_target_off(struct sock, sk_protocol,
9090 sizeof_field(struct sock, sk_protocol),
9094 case offsetof(struct bpf_sock, src_ip4):
9095 *insn++ = BPF_LDX_MEM(
9096 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9097 bpf_target_off(struct sock_common, skc_rcv_saddr,
9098 sizeof_field(struct sock_common,
9103 case offsetof(struct bpf_sock, dst_ip4):
9104 *insn++ = BPF_LDX_MEM(
9105 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9106 bpf_target_off(struct sock_common, skc_daddr,
9107 sizeof_field(struct sock_common,
9112 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
9113 #if IS_ENABLED(CONFIG_IPV6)
9115 off -= offsetof(struct bpf_sock, src_ip6[0]);
9116 *insn++ = BPF_LDX_MEM(
9117 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9120 skc_v6_rcv_saddr.s6_addr32[0],
9121 sizeof_field(struct sock_common,
9122 skc_v6_rcv_saddr.s6_addr32[0]),
9123 target_size) + off);
9126 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9130 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
9131 #if IS_ENABLED(CONFIG_IPV6)
9133 off -= offsetof(struct bpf_sock, dst_ip6[0]);
9134 *insn++ = BPF_LDX_MEM(
9135 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9136 bpf_target_off(struct sock_common,
9137 skc_v6_daddr.s6_addr32[0],
9138 sizeof_field(struct sock_common,
9139 skc_v6_daddr.s6_addr32[0]),
9140 target_size) + off);
9142 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9147 case offsetof(struct bpf_sock, src_port):
9148 *insn++ = BPF_LDX_MEM(
9149 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
9150 si->dst_reg, si->src_reg,
9151 bpf_target_off(struct sock_common, skc_num,
9152 sizeof_field(struct sock_common,
9157 case offsetof(struct bpf_sock, dst_port):
9158 *insn++ = BPF_LDX_MEM(
9159 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
9160 si->dst_reg, si->src_reg,
9161 bpf_target_off(struct sock_common, skc_dport,
9162 sizeof_field(struct sock_common,
9167 case offsetof(struct bpf_sock, state):
9168 *insn++ = BPF_LDX_MEM(
9169 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
9170 si->dst_reg, si->src_reg,
9171 bpf_target_off(struct sock_common, skc_state,
9172 sizeof_field(struct sock_common,
9176 case offsetof(struct bpf_sock, rx_queue_mapping):
9177 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
9178 *insn++ = BPF_LDX_MEM(
9179 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
9180 si->dst_reg, si->src_reg,
9181 bpf_target_off(struct sock, sk_rx_queue_mapping,
9182 sizeof_field(struct sock,
9183 sk_rx_queue_mapping),
9185 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
9187 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9189 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9195 return insn - insn_buf;
9198 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
9199 const struct bpf_insn *si,
9200 struct bpf_insn *insn_buf,
9201 struct bpf_prog *prog, u32 *target_size)
9203 struct bpf_insn *insn = insn_buf;
9206 case offsetof(struct __sk_buff, ifindex):
9207 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9208 si->dst_reg, si->src_reg,
9209 offsetof(struct sk_buff, dev));
9210 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9211 bpf_target_off(struct net_device, ifindex, 4,
9215 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9219 return insn - insn_buf;
9222 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
9223 const struct bpf_insn *si,
9224 struct bpf_insn *insn_buf,
9225 struct bpf_prog *prog, u32 *target_size)
9227 struct bpf_insn *insn = insn_buf;
9230 case offsetof(struct xdp_md, data):
9231 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
9232 si->dst_reg, si->src_reg,
9233 offsetof(struct xdp_buff, data));
9235 case offsetof(struct xdp_md, data_meta):
9236 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
9237 si->dst_reg, si->src_reg,
9238 offsetof(struct xdp_buff, data_meta));
9240 case offsetof(struct xdp_md, data_end):
9241 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
9242 si->dst_reg, si->src_reg,
9243 offsetof(struct xdp_buff, data_end));
9245 case offsetof(struct xdp_md, ingress_ifindex):
9246 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9247 si->dst_reg, si->src_reg,
9248 offsetof(struct xdp_buff, rxq));
9249 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
9250 si->dst_reg, si->dst_reg,
9251 offsetof(struct xdp_rxq_info, dev));
9252 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9253 offsetof(struct net_device, ifindex));
9255 case offsetof(struct xdp_md, rx_queue_index):
9256 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9257 si->dst_reg, si->src_reg,
9258 offsetof(struct xdp_buff, rxq));
9259 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9260 offsetof(struct xdp_rxq_info,
9263 case offsetof(struct xdp_md, egress_ifindex):
9264 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
9265 si->dst_reg, si->src_reg,
9266 offsetof(struct xdp_buff, txq));
9267 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
9268 si->dst_reg, si->dst_reg,
9269 offsetof(struct xdp_txq_info, dev));
9270 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9271 offsetof(struct net_device, ifindex));
9275 return insn - insn_buf;
9278 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
9279 * context Structure, F is Field in context structure that contains a pointer
9280 * to Nested Structure of type NS that has the field NF.
9282 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
9283 * sure that SIZE is not greater than actual size of S.F.NF.
9285 * If offset OFF is provided, the load happens from that offset relative to
9288 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
9290 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
9291 si->src_reg, offsetof(S, F)); \
9292 *insn++ = BPF_LDX_MEM( \
9293 SIZE, si->dst_reg, si->dst_reg, \
9294 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
9299 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
9300 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
9301 BPF_FIELD_SIZEOF(NS, NF), 0)
9303 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
9304 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
9306 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
9307 * "register" since two registers available in convert_ctx_access are not
9308 * enough: we can't override neither SRC, since it contains value to store, nor
9309 * DST since it contains pointer to context that may be used by later
9310 * instructions. But we need a temporary place to save pointer to nested
9311 * structure whose field we want to store to.
9313 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
9315 int tmp_reg = BPF_REG_9; \
9316 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
9318 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
9320 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
9322 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
9323 si->dst_reg, offsetof(S, F)); \
9324 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \
9325 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
9328 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
9332 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
9335 if (type == BPF_WRITE) { \
9336 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
9339 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
9340 S, NS, F, NF, SIZE, OFF); \
9344 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
9345 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
9346 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
9348 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
9349 const struct bpf_insn *si,
9350 struct bpf_insn *insn_buf,
9351 struct bpf_prog *prog, u32 *target_size)
9353 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
9354 struct bpf_insn *insn = insn_buf;
9357 case offsetof(struct bpf_sock_addr, user_family):
9358 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9359 struct sockaddr, uaddr, sa_family);
9362 case offsetof(struct bpf_sock_addr, user_ip4):
9363 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9364 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
9365 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
9368 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9370 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
9371 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9372 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9373 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
9377 case offsetof(struct bpf_sock_addr, user_port):
9378 /* To get port we need to know sa_family first and then treat
9379 * sockaddr as either sockaddr_in or sockaddr_in6.
9380 * Though we can simplify since port field has same offset and
9381 * size in both structures.
9382 * Here we check this invariant and use just one of the
9383 * structures if it's true.
9385 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
9386 offsetof(struct sockaddr_in6, sin6_port));
9387 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
9388 sizeof_field(struct sockaddr_in6, sin6_port));
9389 /* Account for sin6_port being smaller than user_port. */
9390 port_size = min(port_size, BPF_LDST_BYTES(si));
9391 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9392 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9393 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
9396 case offsetof(struct bpf_sock_addr, family):
9397 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9398 struct sock, sk, sk_family);
9401 case offsetof(struct bpf_sock_addr, type):
9402 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9403 struct sock, sk, sk_type);
9406 case offsetof(struct bpf_sock_addr, protocol):
9407 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9408 struct sock, sk, sk_protocol);
9411 case offsetof(struct bpf_sock_addr, msg_src_ip4):
9412 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
9413 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9414 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
9415 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
9418 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9421 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
9422 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
9423 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9424 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
9425 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
9427 case offsetof(struct bpf_sock_addr, sk):
9428 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
9429 si->dst_reg, si->src_reg,
9430 offsetof(struct bpf_sock_addr_kern, sk));
9434 return insn - insn_buf;
9437 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
9438 const struct bpf_insn *si,
9439 struct bpf_insn *insn_buf,
9440 struct bpf_prog *prog,
9443 struct bpf_insn *insn = insn_buf;
9446 /* Helper macro for adding read access to tcp_sock or sock fields. */
9447 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
9449 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \
9450 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
9451 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
9452 if (si->dst_reg == reg || si->src_reg == reg) \
9454 if (si->dst_reg == reg || si->src_reg == reg) \
9456 if (si->dst_reg == si->src_reg) { \
9457 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
9458 offsetof(struct bpf_sock_ops_kern, \
9460 fullsock_reg = reg; \
9463 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9464 struct bpf_sock_ops_kern, \
9466 fullsock_reg, si->src_reg, \
9467 offsetof(struct bpf_sock_ops_kern, \
9469 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
9470 if (si->dst_reg == si->src_reg) \
9471 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9472 offsetof(struct bpf_sock_ops_kern, \
9474 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9475 struct bpf_sock_ops_kern, sk),\
9476 si->dst_reg, si->src_reg, \
9477 offsetof(struct bpf_sock_ops_kern, sk));\
9478 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
9480 si->dst_reg, si->dst_reg, \
9481 offsetof(OBJ, OBJ_FIELD)); \
9482 if (si->dst_reg == si->src_reg) { \
9483 *insn++ = BPF_JMP_A(1); \
9484 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9485 offsetof(struct bpf_sock_ops_kern, \
9490 #define SOCK_OPS_GET_SK() \
9492 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \
9493 if (si->dst_reg == reg || si->src_reg == reg) \
9495 if (si->dst_reg == reg || si->src_reg == reg) \
9497 if (si->dst_reg == si->src_reg) { \
9498 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
9499 offsetof(struct bpf_sock_ops_kern, \
9501 fullsock_reg = reg; \
9504 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9505 struct bpf_sock_ops_kern, \
9507 fullsock_reg, si->src_reg, \
9508 offsetof(struct bpf_sock_ops_kern, \
9510 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
9511 if (si->dst_reg == si->src_reg) \
9512 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9513 offsetof(struct bpf_sock_ops_kern, \
9515 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9516 struct bpf_sock_ops_kern, sk),\
9517 si->dst_reg, si->src_reg, \
9518 offsetof(struct bpf_sock_ops_kern, sk));\
9519 if (si->dst_reg == si->src_reg) { \
9520 *insn++ = BPF_JMP_A(1); \
9521 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9522 offsetof(struct bpf_sock_ops_kern, \
9527 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
9528 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
9530 /* Helper macro for adding write access to tcp_sock or sock fields.
9531 * The macro is called with two registers, dst_reg which contains a pointer
9532 * to ctx (context) and src_reg which contains the value that should be
9533 * stored. However, we need an additional register since we cannot overwrite
9534 * dst_reg because it may be used later in the program.
9535 * Instead we "borrow" one of the other register. We first save its value
9536 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
9537 * it at the end of the macro.
9539 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
9541 int reg = BPF_REG_9; \
9542 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
9543 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
9544 if (si->dst_reg == reg || si->src_reg == reg) \
9546 if (si->dst_reg == reg || si->src_reg == reg) \
9548 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
9549 offsetof(struct bpf_sock_ops_kern, \
9551 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9552 struct bpf_sock_ops_kern, \
9555 offsetof(struct bpf_sock_ops_kern, \
9557 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
9558 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9559 struct bpf_sock_ops_kern, sk),\
9561 offsetof(struct bpf_sock_ops_kern, sk));\
9562 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \
9564 offsetof(OBJ, OBJ_FIELD)); \
9565 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
9566 offsetof(struct bpf_sock_ops_kern, \
9570 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
9572 if (TYPE == BPF_WRITE) \
9573 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
9575 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
9578 if (insn > insn_buf)
9579 return insn - insn_buf;
9582 case offsetof(struct bpf_sock_ops, op):
9583 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9585 si->dst_reg, si->src_reg,
9586 offsetof(struct bpf_sock_ops_kern, op));
9589 case offsetof(struct bpf_sock_ops, replylong[0]) ...
9590 offsetof(struct bpf_sock_ops, replylong[3]):
9591 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
9592 sizeof_field(struct bpf_sock_ops_kern, reply));
9593 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
9594 sizeof_field(struct bpf_sock_ops_kern, replylong));
9596 off -= offsetof(struct bpf_sock_ops, replylong[0]);
9597 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
9598 if (type == BPF_WRITE)
9599 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9602 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9606 case offsetof(struct bpf_sock_ops, family):
9607 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9609 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9610 struct bpf_sock_ops_kern, sk),
9611 si->dst_reg, si->src_reg,
9612 offsetof(struct bpf_sock_ops_kern, sk));
9613 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9614 offsetof(struct sock_common, skc_family));
9617 case offsetof(struct bpf_sock_ops, remote_ip4):
9618 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9620 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9621 struct bpf_sock_ops_kern, sk),
9622 si->dst_reg, si->src_reg,
9623 offsetof(struct bpf_sock_ops_kern, sk));
9624 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9625 offsetof(struct sock_common, skc_daddr));
9628 case offsetof(struct bpf_sock_ops, local_ip4):
9629 BUILD_BUG_ON(sizeof_field(struct sock_common,
9630 skc_rcv_saddr) != 4);
9632 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9633 struct bpf_sock_ops_kern, sk),
9634 si->dst_reg, si->src_reg,
9635 offsetof(struct bpf_sock_ops_kern, sk));
9636 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9637 offsetof(struct sock_common,
9641 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
9642 offsetof(struct bpf_sock_ops, remote_ip6[3]):
9643 #if IS_ENABLED(CONFIG_IPV6)
9644 BUILD_BUG_ON(sizeof_field(struct sock_common,
9645 skc_v6_daddr.s6_addr32[0]) != 4);
9648 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
9649 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9650 struct bpf_sock_ops_kern, sk),
9651 si->dst_reg, si->src_reg,
9652 offsetof(struct bpf_sock_ops_kern, sk));
9653 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9654 offsetof(struct sock_common,
9655 skc_v6_daddr.s6_addr32[0]) +
9658 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9662 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
9663 offsetof(struct bpf_sock_ops, local_ip6[3]):
9664 #if IS_ENABLED(CONFIG_IPV6)
9665 BUILD_BUG_ON(sizeof_field(struct sock_common,
9666 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9669 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
9670 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9671 struct bpf_sock_ops_kern, sk),
9672 si->dst_reg, si->src_reg,
9673 offsetof(struct bpf_sock_ops_kern, sk));
9674 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9675 offsetof(struct sock_common,
9676 skc_v6_rcv_saddr.s6_addr32[0]) +
9679 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9683 case offsetof(struct bpf_sock_ops, remote_port):
9684 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9686 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9687 struct bpf_sock_ops_kern, sk),
9688 si->dst_reg, si->src_reg,
9689 offsetof(struct bpf_sock_ops_kern, sk));
9690 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9691 offsetof(struct sock_common, skc_dport));
9692 #ifndef __BIG_ENDIAN_BITFIELD
9693 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9697 case offsetof(struct bpf_sock_ops, local_port):
9698 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9700 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9701 struct bpf_sock_ops_kern, sk),
9702 si->dst_reg, si->src_reg,
9703 offsetof(struct bpf_sock_ops_kern, sk));
9704 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9705 offsetof(struct sock_common, skc_num));
9708 case offsetof(struct bpf_sock_ops, is_fullsock):
9709 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9710 struct bpf_sock_ops_kern,
9712 si->dst_reg, si->src_reg,
9713 offsetof(struct bpf_sock_ops_kern,
9717 case offsetof(struct bpf_sock_ops, state):
9718 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
9720 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9721 struct bpf_sock_ops_kern, sk),
9722 si->dst_reg, si->src_reg,
9723 offsetof(struct bpf_sock_ops_kern, sk));
9724 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
9725 offsetof(struct sock_common, skc_state));
9728 case offsetof(struct bpf_sock_ops, rtt_min):
9729 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
9730 sizeof(struct minmax));
9731 BUILD_BUG_ON(sizeof(struct minmax) <
9732 sizeof(struct minmax_sample));
9734 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9735 struct bpf_sock_ops_kern, sk),
9736 si->dst_reg, si->src_reg,
9737 offsetof(struct bpf_sock_ops_kern, sk));
9738 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9739 offsetof(struct tcp_sock, rtt_min) +
9740 sizeof_field(struct minmax_sample, t));
9743 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
9744 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
9748 case offsetof(struct bpf_sock_ops, sk_txhash):
9749 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
9752 case offsetof(struct bpf_sock_ops, snd_cwnd):
9753 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
9755 case offsetof(struct bpf_sock_ops, srtt_us):
9756 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
9758 case offsetof(struct bpf_sock_ops, snd_ssthresh):
9759 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
9761 case offsetof(struct bpf_sock_ops, rcv_nxt):
9762 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
9764 case offsetof(struct bpf_sock_ops, snd_nxt):
9765 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
9767 case offsetof(struct bpf_sock_ops, snd_una):
9768 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
9770 case offsetof(struct bpf_sock_ops, mss_cache):
9771 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
9773 case offsetof(struct bpf_sock_ops, ecn_flags):
9774 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
9776 case offsetof(struct bpf_sock_ops, rate_delivered):
9777 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
9779 case offsetof(struct bpf_sock_ops, rate_interval_us):
9780 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
9782 case offsetof(struct bpf_sock_ops, packets_out):
9783 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
9785 case offsetof(struct bpf_sock_ops, retrans_out):
9786 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
9788 case offsetof(struct bpf_sock_ops, total_retrans):
9789 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
9791 case offsetof(struct bpf_sock_ops, segs_in):
9792 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
9794 case offsetof(struct bpf_sock_ops, data_segs_in):
9795 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
9797 case offsetof(struct bpf_sock_ops, segs_out):
9798 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
9800 case offsetof(struct bpf_sock_ops, data_segs_out):
9801 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
9803 case offsetof(struct bpf_sock_ops, lost_out):
9804 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
9806 case offsetof(struct bpf_sock_ops, sacked_out):
9807 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
9809 case offsetof(struct bpf_sock_ops, bytes_received):
9810 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
9812 case offsetof(struct bpf_sock_ops, bytes_acked):
9813 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
9815 case offsetof(struct bpf_sock_ops, sk):
9818 case offsetof(struct bpf_sock_ops, skb_data_end):
9819 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9821 si->dst_reg, si->src_reg,
9822 offsetof(struct bpf_sock_ops_kern,
9825 case offsetof(struct bpf_sock_ops, skb_data):
9826 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9828 si->dst_reg, si->src_reg,
9829 offsetof(struct bpf_sock_ops_kern,
9831 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9832 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9833 si->dst_reg, si->dst_reg,
9834 offsetof(struct sk_buff, data));
9836 case offsetof(struct bpf_sock_ops, skb_len):
9837 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9839 si->dst_reg, si->src_reg,
9840 offsetof(struct bpf_sock_ops_kern,
9842 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9843 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9844 si->dst_reg, si->dst_reg,
9845 offsetof(struct sk_buff, len));
9847 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9848 off = offsetof(struct sk_buff, cb);
9849 off += offsetof(struct tcp_skb_cb, tcp_flags);
9850 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
9851 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9853 si->dst_reg, si->src_reg,
9854 offsetof(struct bpf_sock_ops_kern,
9856 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9857 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
9859 si->dst_reg, si->dst_reg, off);
9862 return insn - insn_buf;
9865 /* data_end = skb->data + skb_headlen() */
9866 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
9867 struct bpf_insn *insn)
9870 int temp_reg_off = offsetof(struct sk_buff, cb) +
9871 offsetof(struct sk_skb_cb, temp_reg);
9873 if (si->src_reg == si->dst_reg) {
9874 /* We need an extra register, choose and save a register. */
9876 if (si->src_reg == reg || si->dst_reg == reg)
9878 if (si->src_reg == reg || si->dst_reg == reg)
9880 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off);
9885 /* reg = skb->data */
9886 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9888 offsetof(struct sk_buff, data));
9890 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9891 BPF_REG_AX, si->src_reg,
9892 offsetof(struct sk_buff, len));
9893 /* reg = skb->data + skb->len */
9894 *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX);
9895 /* AX = skb->data_len */
9896 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
9897 BPF_REG_AX, si->src_reg,
9898 offsetof(struct sk_buff, data_len));
9900 /* reg = skb->data + skb->len - skb->data_len */
9901 *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX);
9903 if (si->src_reg == si->dst_reg) {
9904 /* Restore the saved register */
9905 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg);
9906 *insn++ = BPF_MOV64_REG(si->dst_reg, reg);
9907 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off);
9913 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
9914 const struct bpf_insn *si,
9915 struct bpf_insn *insn_buf,
9916 struct bpf_prog *prog, u32 *target_size)
9918 struct bpf_insn *insn = insn_buf;
9922 case offsetof(struct __sk_buff, data_end):
9923 insn = bpf_convert_data_end_access(si, insn);
9925 case offsetof(struct __sk_buff, cb[0]) ...
9926 offsetofend(struct __sk_buff, cb[4]) - 1:
9927 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20);
9928 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
9929 offsetof(struct sk_skb_cb, data)) %
9932 prog->cb_access = 1;
9934 off -= offsetof(struct __sk_buff, cb[0]);
9935 off += offsetof(struct sk_buff, cb);
9936 off += offsetof(struct sk_skb_cb, data);
9937 if (type == BPF_WRITE)
9938 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
9941 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
9947 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9951 return insn - insn_buf;
9954 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
9955 const struct bpf_insn *si,
9956 struct bpf_insn *insn_buf,
9957 struct bpf_prog *prog, u32 *target_size)
9959 struct bpf_insn *insn = insn_buf;
9960 #if IS_ENABLED(CONFIG_IPV6)
9964 /* convert ctx uses the fact sg element is first in struct */
9965 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
9968 case offsetof(struct sk_msg_md, data):
9969 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
9970 si->dst_reg, si->src_reg,
9971 offsetof(struct sk_msg, data));
9973 case offsetof(struct sk_msg_md, data_end):
9974 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
9975 si->dst_reg, si->src_reg,
9976 offsetof(struct sk_msg, data_end));
9978 case offsetof(struct sk_msg_md, family):
9979 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9981 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9983 si->dst_reg, si->src_reg,
9984 offsetof(struct sk_msg, sk));
9985 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9986 offsetof(struct sock_common, skc_family));
9989 case offsetof(struct sk_msg_md, remote_ip4):
9990 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9992 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9994 si->dst_reg, si->src_reg,
9995 offsetof(struct sk_msg, sk));
9996 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9997 offsetof(struct sock_common, skc_daddr));
10000 case offsetof(struct sk_msg_md, local_ip4):
10001 BUILD_BUG_ON(sizeof_field(struct sock_common,
10002 skc_rcv_saddr) != 4);
10004 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10005 struct sk_msg, sk),
10006 si->dst_reg, si->src_reg,
10007 offsetof(struct sk_msg, sk));
10008 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10009 offsetof(struct sock_common,
10013 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
10014 offsetof(struct sk_msg_md, remote_ip6[3]):
10015 #if IS_ENABLED(CONFIG_IPV6)
10016 BUILD_BUG_ON(sizeof_field(struct sock_common,
10017 skc_v6_daddr.s6_addr32[0]) != 4);
10020 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
10021 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10022 struct sk_msg, sk),
10023 si->dst_reg, si->src_reg,
10024 offsetof(struct sk_msg, sk));
10025 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10026 offsetof(struct sock_common,
10027 skc_v6_daddr.s6_addr32[0]) +
10030 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10034 case offsetof(struct sk_msg_md, local_ip6[0]) ...
10035 offsetof(struct sk_msg_md, local_ip6[3]):
10036 #if IS_ENABLED(CONFIG_IPV6)
10037 BUILD_BUG_ON(sizeof_field(struct sock_common,
10038 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10041 off -= offsetof(struct sk_msg_md, local_ip6[0]);
10042 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10043 struct sk_msg, sk),
10044 si->dst_reg, si->src_reg,
10045 offsetof(struct sk_msg, sk));
10046 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10047 offsetof(struct sock_common,
10048 skc_v6_rcv_saddr.s6_addr32[0]) +
10051 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10055 case offsetof(struct sk_msg_md, remote_port):
10056 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10058 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10059 struct sk_msg, sk),
10060 si->dst_reg, si->src_reg,
10061 offsetof(struct sk_msg, sk));
10062 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10063 offsetof(struct sock_common, skc_dport));
10064 #ifndef __BIG_ENDIAN_BITFIELD
10065 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10069 case offsetof(struct sk_msg_md, local_port):
10070 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10072 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10073 struct sk_msg, sk),
10074 si->dst_reg, si->src_reg,
10075 offsetof(struct sk_msg, sk));
10076 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10077 offsetof(struct sock_common, skc_num));
10080 case offsetof(struct sk_msg_md, size):
10081 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
10082 si->dst_reg, si->src_reg,
10083 offsetof(struct sk_msg_sg, size));
10086 case offsetof(struct sk_msg_md, sk):
10087 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
10088 si->dst_reg, si->src_reg,
10089 offsetof(struct sk_msg, sk));
10093 return insn - insn_buf;
10096 const struct bpf_verifier_ops sk_filter_verifier_ops = {
10097 .get_func_proto = sk_filter_func_proto,
10098 .is_valid_access = sk_filter_is_valid_access,
10099 .convert_ctx_access = bpf_convert_ctx_access,
10100 .gen_ld_abs = bpf_gen_ld_abs,
10103 const struct bpf_prog_ops sk_filter_prog_ops = {
10104 .test_run = bpf_prog_test_run_skb,
10107 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
10108 .get_func_proto = tc_cls_act_func_proto,
10109 .is_valid_access = tc_cls_act_is_valid_access,
10110 .convert_ctx_access = tc_cls_act_convert_ctx_access,
10111 .gen_prologue = tc_cls_act_prologue,
10112 .gen_ld_abs = bpf_gen_ld_abs,
10113 .check_kfunc_call = bpf_prog_test_check_kfunc_call,
10116 const struct bpf_prog_ops tc_cls_act_prog_ops = {
10117 .test_run = bpf_prog_test_run_skb,
10120 const struct bpf_verifier_ops xdp_verifier_ops = {
10121 .get_func_proto = xdp_func_proto,
10122 .is_valid_access = xdp_is_valid_access,
10123 .convert_ctx_access = xdp_convert_ctx_access,
10124 .gen_prologue = bpf_noop_prologue,
10127 const struct bpf_prog_ops xdp_prog_ops = {
10128 .test_run = bpf_prog_test_run_xdp,
10131 const struct bpf_verifier_ops cg_skb_verifier_ops = {
10132 .get_func_proto = cg_skb_func_proto,
10133 .is_valid_access = cg_skb_is_valid_access,
10134 .convert_ctx_access = bpf_convert_ctx_access,
10137 const struct bpf_prog_ops cg_skb_prog_ops = {
10138 .test_run = bpf_prog_test_run_skb,
10141 const struct bpf_verifier_ops lwt_in_verifier_ops = {
10142 .get_func_proto = lwt_in_func_proto,
10143 .is_valid_access = lwt_is_valid_access,
10144 .convert_ctx_access = bpf_convert_ctx_access,
10147 const struct bpf_prog_ops lwt_in_prog_ops = {
10148 .test_run = bpf_prog_test_run_skb,
10151 const struct bpf_verifier_ops lwt_out_verifier_ops = {
10152 .get_func_proto = lwt_out_func_proto,
10153 .is_valid_access = lwt_is_valid_access,
10154 .convert_ctx_access = bpf_convert_ctx_access,
10157 const struct bpf_prog_ops lwt_out_prog_ops = {
10158 .test_run = bpf_prog_test_run_skb,
10161 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
10162 .get_func_proto = lwt_xmit_func_proto,
10163 .is_valid_access = lwt_is_valid_access,
10164 .convert_ctx_access = bpf_convert_ctx_access,
10165 .gen_prologue = tc_cls_act_prologue,
10168 const struct bpf_prog_ops lwt_xmit_prog_ops = {
10169 .test_run = bpf_prog_test_run_skb,
10172 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
10173 .get_func_proto = lwt_seg6local_func_proto,
10174 .is_valid_access = lwt_is_valid_access,
10175 .convert_ctx_access = bpf_convert_ctx_access,
10178 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
10179 .test_run = bpf_prog_test_run_skb,
10182 const struct bpf_verifier_ops cg_sock_verifier_ops = {
10183 .get_func_proto = sock_filter_func_proto,
10184 .is_valid_access = sock_filter_is_valid_access,
10185 .convert_ctx_access = bpf_sock_convert_ctx_access,
10188 const struct bpf_prog_ops cg_sock_prog_ops = {
10191 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
10192 .get_func_proto = sock_addr_func_proto,
10193 .is_valid_access = sock_addr_is_valid_access,
10194 .convert_ctx_access = sock_addr_convert_ctx_access,
10197 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
10200 const struct bpf_verifier_ops sock_ops_verifier_ops = {
10201 .get_func_proto = sock_ops_func_proto,
10202 .is_valid_access = sock_ops_is_valid_access,
10203 .convert_ctx_access = sock_ops_convert_ctx_access,
10206 const struct bpf_prog_ops sock_ops_prog_ops = {
10209 const struct bpf_verifier_ops sk_skb_verifier_ops = {
10210 .get_func_proto = sk_skb_func_proto,
10211 .is_valid_access = sk_skb_is_valid_access,
10212 .convert_ctx_access = sk_skb_convert_ctx_access,
10213 .gen_prologue = sk_skb_prologue,
10216 const struct bpf_prog_ops sk_skb_prog_ops = {
10219 const struct bpf_verifier_ops sk_msg_verifier_ops = {
10220 .get_func_proto = sk_msg_func_proto,
10221 .is_valid_access = sk_msg_is_valid_access,
10222 .convert_ctx_access = sk_msg_convert_ctx_access,
10223 .gen_prologue = bpf_noop_prologue,
10226 const struct bpf_prog_ops sk_msg_prog_ops = {
10229 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
10230 .get_func_proto = flow_dissector_func_proto,
10231 .is_valid_access = flow_dissector_is_valid_access,
10232 .convert_ctx_access = flow_dissector_convert_ctx_access,
10235 const struct bpf_prog_ops flow_dissector_prog_ops = {
10236 .test_run = bpf_prog_test_run_flow_dissector,
10239 int sk_detach_filter(struct sock *sk)
10242 struct sk_filter *filter;
10244 if (sock_flag(sk, SOCK_FILTER_LOCKED))
10247 filter = rcu_dereference_protected(sk->sk_filter,
10248 lockdep_sock_is_held(sk));
10250 RCU_INIT_POINTER(sk->sk_filter, NULL);
10251 sk_filter_uncharge(sk, filter);
10257 EXPORT_SYMBOL_GPL(sk_detach_filter);
10259 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
10262 struct sock_fprog_kern *fprog;
10263 struct sk_filter *filter;
10267 filter = rcu_dereference_protected(sk->sk_filter,
10268 lockdep_sock_is_held(sk));
10272 /* We're copying the filter that has been originally attached,
10273 * so no conversion/decode needed anymore. eBPF programs that
10274 * have no original program cannot be dumped through this.
10277 fprog = filter->prog->orig_prog;
10283 /* User space only enquires number of filter blocks. */
10287 if (len < fprog->len)
10291 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
10294 /* Instead of bytes, the API requests to return the number
10295 * of filter blocks.
10304 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
10305 struct sock_reuseport *reuse,
10306 struct sock *sk, struct sk_buff *skb,
10307 struct sock *migrating_sk,
10310 reuse_kern->skb = skb;
10311 reuse_kern->sk = sk;
10312 reuse_kern->selected_sk = NULL;
10313 reuse_kern->migrating_sk = migrating_sk;
10314 reuse_kern->data_end = skb->data + skb_headlen(skb);
10315 reuse_kern->hash = hash;
10316 reuse_kern->reuseport_id = reuse->reuseport_id;
10317 reuse_kern->bind_inany = reuse->bind_inany;
10320 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
10321 struct bpf_prog *prog, struct sk_buff *skb,
10322 struct sock *migrating_sk,
10325 struct sk_reuseport_kern reuse_kern;
10326 enum sk_action action;
10328 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
10329 action = bpf_prog_run(prog, &reuse_kern);
10331 if (action == SK_PASS)
10332 return reuse_kern.selected_sk;
10334 return ERR_PTR(-ECONNREFUSED);
10337 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
10338 struct bpf_map *, map, void *, key, u32, flags)
10340 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
10341 struct sock_reuseport *reuse;
10342 struct sock *selected_sk;
10344 selected_sk = map->ops->map_lookup_elem(map, key);
10348 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
10350 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
10351 if (sk_is_refcounted(selected_sk))
10352 sock_put(selected_sk);
10354 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
10355 * The only (!reuse) case here is - the sk has already been
10356 * unhashed (e.g. by close()), so treat it as -ENOENT.
10358 * Other maps (e.g. sock_map) do not provide this guarantee and
10359 * the sk may never be in the reuseport group to begin with.
10361 return is_sockarray ? -ENOENT : -EINVAL;
10364 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
10365 struct sock *sk = reuse_kern->sk;
10367 if (sk->sk_protocol != selected_sk->sk_protocol)
10368 return -EPROTOTYPE;
10369 else if (sk->sk_family != selected_sk->sk_family)
10370 return -EAFNOSUPPORT;
10372 /* Catch all. Likely bound to a different sockaddr. */
10376 reuse_kern->selected_sk = selected_sk;
10381 static const struct bpf_func_proto sk_select_reuseport_proto = {
10382 .func = sk_select_reuseport,
10384 .ret_type = RET_INTEGER,
10385 .arg1_type = ARG_PTR_TO_CTX,
10386 .arg2_type = ARG_CONST_MAP_PTR,
10387 .arg3_type = ARG_PTR_TO_MAP_KEY,
10388 .arg4_type = ARG_ANYTHING,
10391 BPF_CALL_4(sk_reuseport_load_bytes,
10392 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10393 void *, to, u32, len)
10395 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
10398 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
10399 .func = sk_reuseport_load_bytes,
10401 .ret_type = RET_INTEGER,
10402 .arg1_type = ARG_PTR_TO_CTX,
10403 .arg2_type = ARG_ANYTHING,
10404 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
10405 .arg4_type = ARG_CONST_SIZE,
10408 BPF_CALL_5(sk_reuseport_load_bytes_relative,
10409 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10410 void *, to, u32, len, u32, start_header)
10412 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
10413 len, start_header);
10416 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
10417 .func = sk_reuseport_load_bytes_relative,
10419 .ret_type = RET_INTEGER,
10420 .arg1_type = ARG_PTR_TO_CTX,
10421 .arg2_type = ARG_ANYTHING,
10422 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
10423 .arg4_type = ARG_CONST_SIZE,
10424 .arg5_type = ARG_ANYTHING,
10427 static const struct bpf_func_proto *
10428 sk_reuseport_func_proto(enum bpf_func_id func_id,
10429 const struct bpf_prog *prog)
10432 case BPF_FUNC_sk_select_reuseport:
10433 return &sk_select_reuseport_proto;
10434 case BPF_FUNC_skb_load_bytes:
10435 return &sk_reuseport_load_bytes_proto;
10436 case BPF_FUNC_skb_load_bytes_relative:
10437 return &sk_reuseport_load_bytes_relative_proto;
10438 case BPF_FUNC_get_socket_cookie:
10439 return &bpf_get_socket_ptr_cookie_proto;
10440 case BPF_FUNC_ktime_get_coarse_ns:
10441 return &bpf_ktime_get_coarse_ns_proto;
10443 return bpf_base_func_proto(func_id);
10448 sk_reuseport_is_valid_access(int off, int size,
10449 enum bpf_access_type type,
10450 const struct bpf_prog *prog,
10451 struct bpf_insn_access_aux *info)
10453 const u32 size_default = sizeof(__u32);
10455 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
10456 off % size || type != BPF_READ)
10460 case offsetof(struct sk_reuseport_md, data):
10461 info->reg_type = PTR_TO_PACKET;
10462 return size == sizeof(__u64);
10464 case offsetof(struct sk_reuseport_md, data_end):
10465 info->reg_type = PTR_TO_PACKET_END;
10466 return size == sizeof(__u64);
10468 case offsetof(struct sk_reuseport_md, hash):
10469 return size == size_default;
10471 case offsetof(struct sk_reuseport_md, sk):
10472 info->reg_type = PTR_TO_SOCKET;
10473 return size == sizeof(__u64);
10475 case offsetof(struct sk_reuseport_md, migrating_sk):
10476 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
10477 return size == sizeof(__u64);
10479 /* Fields that allow narrowing */
10480 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
10481 if (size < sizeof_field(struct sk_buff, protocol))
10484 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
10485 case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
10486 case bpf_ctx_range(struct sk_reuseport_md, len):
10487 bpf_ctx_record_field_size(info, size_default);
10488 return bpf_ctx_narrow_access_ok(off, size, size_default);
10495 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
10496 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
10497 si->dst_reg, si->src_reg, \
10498 bpf_target_off(struct sk_reuseport_kern, F, \
10499 sizeof_field(struct sk_reuseport_kern, F), \
10503 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
10504 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
10509 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \
10510 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
10515 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
10516 const struct bpf_insn *si,
10517 struct bpf_insn *insn_buf,
10518 struct bpf_prog *prog,
10521 struct bpf_insn *insn = insn_buf;
10524 case offsetof(struct sk_reuseport_md, data):
10525 SK_REUSEPORT_LOAD_SKB_FIELD(data);
10528 case offsetof(struct sk_reuseport_md, len):
10529 SK_REUSEPORT_LOAD_SKB_FIELD(len);
10532 case offsetof(struct sk_reuseport_md, eth_protocol):
10533 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
10536 case offsetof(struct sk_reuseport_md, ip_protocol):
10537 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
10540 case offsetof(struct sk_reuseport_md, data_end):
10541 SK_REUSEPORT_LOAD_FIELD(data_end);
10544 case offsetof(struct sk_reuseport_md, hash):
10545 SK_REUSEPORT_LOAD_FIELD(hash);
10548 case offsetof(struct sk_reuseport_md, bind_inany):
10549 SK_REUSEPORT_LOAD_FIELD(bind_inany);
10552 case offsetof(struct sk_reuseport_md, sk):
10553 SK_REUSEPORT_LOAD_FIELD(sk);
10556 case offsetof(struct sk_reuseport_md, migrating_sk):
10557 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
10561 return insn - insn_buf;
10564 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
10565 .get_func_proto = sk_reuseport_func_proto,
10566 .is_valid_access = sk_reuseport_is_valid_access,
10567 .convert_ctx_access = sk_reuseport_convert_ctx_access,
10570 const struct bpf_prog_ops sk_reuseport_prog_ops = {
10573 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
10574 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
10576 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
10577 struct sock *, sk, u64, flags)
10579 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
10580 BPF_SK_LOOKUP_F_NO_REUSEPORT)))
10582 if (unlikely(sk && sk_is_refcounted(sk)))
10583 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
10584 if (unlikely(sk && sk->sk_state == TCP_ESTABLISHED))
10585 return -ESOCKTNOSUPPORT; /* reject connected sockets */
10587 /* Check if socket is suitable for packet L3/L4 protocol */
10588 if (sk && sk->sk_protocol != ctx->protocol)
10589 return -EPROTOTYPE;
10590 if (sk && sk->sk_family != ctx->family &&
10591 (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
10592 return -EAFNOSUPPORT;
10594 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
10597 /* Select socket as lookup result */
10598 ctx->selected_sk = sk;
10599 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
10603 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
10604 .func = bpf_sk_lookup_assign,
10606 .ret_type = RET_INTEGER,
10607 .arg1_type = ARG_PTR_TO_CTX,
10608 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL,
10609 .arg3_type = ARG_ANYTHING,
10612 static const struct bpf_func_proto *
10613 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
10616 case BPF_FUNC_perf_event_output:
10617 return &bpf_event_output_data_proto;
10618 case BPF_FUNC_sk_assign:
10619 return &bpf_sk_lookup_assign_proto;
10620 case BPF_FUNC_sk_release:
10621 return &bpf_sk_release_proto;
10623 return bpf_sk_base_func_proto(func_id);
10627 static bool sk_lookup_is_valid_access(int off, int size,
10628 enum bpf_access_type type,
10629 const struct bpf_prog *prog,
10630 struct bpf_insn_access_aux *info)
10632 if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
10634 if (off % size != 0)
10636 if (type != BPF_READ)
10640 case offsetof(struct bpf_sk_lookup, sk):
10641 info->reg_type = PTR_TO_SOCKET_OR_NULL;
10642 return size == sizeof(__u64);
10644 case bpf_ctx_range(struct bpf_sk_lookup, family):
10645 case bpf_ctx_range(struct bpf_sk_lookup, protocol):
10646 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
10647 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
10648 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
10649 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
10650 case offsetof(struct bpf_sk_lookup, remote_port) ...
10651 offsetof(struct bpf_sk_lookup, local_ip4) - 1:
10652 case bpf_ctx_range(struct bpf_sk_lookup, local_port):
10653 bpf_ctx_record_field_size(info, sizeof(__u32));
10654 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
10661 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
10662 const struct bpf_insn *si,
10663 struct bpf_insn *insn_buf,
10664 struct bpf_prog *prog,
10667 struct bpf_insn *insn = insn_buf;
10670 case offsetof(struct bpf_sk_lookup, sk):
10671 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10672 offsetof(struct bpf_sk_lookup_kern, selected_sk));
10675 case offsetof(struct bpf_sk_lookup, family):
10676 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10677 bpf_target_off(struct bpf_sk_lookup_kern,
10678 family, 2, target_size));
10681 case offsetof(struct bpf_sk_lookup, protocol):
10682 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10683 bpf_target_off(struct bpf_sk_lookup_kern,
10684 protocol, 2, target_size));
10687 case offsetof(struct bpf_sk_lookup, remote_ip4):
10688 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10689 bpf_target_off(struct bpf_sk_lookup_kern,
10690 v4.saddr, 4, target_size));
10693 case offsetof(struct bpf_sk_lookup, local_ip4):
10694 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10695 bpf_target_off(struct bpf_sk_lookup_kern,
10696 v4.daddr, 4, target_size));
10699 case bpf_ctx_range_till(struct bpf_sk_lookup,
10700 remote_ip6[0], remote_ip6[3]): {
10701 #if IS_ENABLED(CONFIG_IPV6)
10704 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
10705 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10706 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10707 offsetof(struct bpf_sk_lookup_kern, v6.saddr));
10708 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10709 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10711 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10715 case bpf_ctx_range_till(struct bpf_sk_lookup,
10716 local_ip6[0], local_ip6[3]): {
10717 #if IS_ENABLED(CONFIG_IPV6)
10720 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
10721 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10722 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10723 offsetof(struct bpf_sk_lookup_kern, v6.daddr));
10724 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10725 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10727 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10731 case offsetof(struct bpf_sk_lookup, remote_port):
10732 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10733 bpf_target_off(struct bpf_sk_lookup_kern,
10734 sport, 2, target_size));
10737 case offsetof(struct bpf_sk_lookup, local_port):
10738 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10739 bpf_target_off(struct bpf_sk_lookup_kern,
10740 dport, 2, target_size));
10744 return insn - insn_buf;
10747 const struct bpf_prog_ops sk_lookup_prog_ops = {
10748 .test_run = bpf_prog_test_run_sk_lookup,
10751 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
10752 .get_func_proto = sk_lookup_func_proto,
10753 .is_valid_access = sk_lookup_is_valid_access,
10754 .convert_ctx_access = sk_lookup_convert_ctx_access,
10757 #endif /* CONFIG_INET */
10759 DEFINE_BPF_DISPATCHER(xdp)
10761 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
10763 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
10766 #ifdef CONFIG_DEBUG_INFO_BTF
10767 BTF_ID_LIST_GLOBAL(btf_sock_ids)
10768 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
10770 #undef BTF_SOCK_TYPE
10772 u32 btf_sock_ids[MAX_BTF_SOCK_TYPE];
10775 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
10777 /* tcp6_sock type is not generated in dwarf and hence btf,
10778 * trigger an explicit type generation here.
10780 BTF_TYPE_EMIT(struct tcp6_sock);
10781 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
10782 sk->sk_family == AF_INET6)
10783 return (unsigned long)sk;
10785 return (unsigned long)NULL;
10788 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
10789 .func = bpf_skc_to_tcp6_sock,
10791 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10792 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10793 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
10796 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
10798 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
10799 return (unsigned long)sk;
10801 return (unsigned long)NULL;
10804 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
10805 .func = bpf_skc_to_tcp_sock,
10807 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10808 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10809 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
10812 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
10814 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
10815 * generated if CONFIG_INET=n. Trigger an explicit generation here.
10817 BTF_TYPE_EMIT(struct inet_timewait_sock);
10818 BTF_TYPE_EMIT(struct tcp_timewait_sock);
10821 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
10822 return (unsigned long)sk;
10825 #if IS_BUILTIN(CONFIG_IPV6)
10826 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
10827 return (unsigned long)sk;
10830 return (unsigned long)NULL;
10833 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
10834 .func = bpf_skc_to_tcp_timewait_sock,
10836 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10837 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10838 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
10841 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
10844 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10845 return (unsigned long)sk;
10848 #if IS_BUILTIN(CONFIG_IPV6)
10849 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10850 return (unsigned long)sk;
10853 return (unsigned long)NULL;
10856 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
10857 .func = bpf_skc_to_tcp_request_sock,
10859 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10860 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10861 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
10864 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
10866 /* udp6_sock type is not generated in dwarf and hence btf,
10867 * trigger an explicit type generation here.
10869 BTF_TYPE_EMIT(struct udp6_sock);
10870 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
10871 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
10872 return (unsigned long)sk;
10874 return (unsigned long)NULL;
10877 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
10878 .func = bpf_skc_to_udp6_sock,
10880 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10881 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10882 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
10885 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
10887 return (unsigned long)sock_from_file(file);
10890 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
10891 BTF_ID(struct, socket)
10892 BTF_ID(struct, file)
10894 const struct bpf_func_proto bpf_sock_from_file_proto = {
10895 .func = bpf_sock_from_file,
10897 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10898 .ret_btf_id = &bpf_sock_from_file_btf_ids[0],
10899 .arg1_type = ARG_PTR_TO_BTF_ID,
10900 .arg1_btf_id = &bpf_sock_from_file_btf_ids[1],
10903 static const struct bpf_func_proto *
10904 bpf_sk_base_func_proto(enum bpf_func_id func_id)
10906 const struct bpf_func_proto *func;
10909 case BPF_FUNC_skc_to_tcp6_sock:
10910 func = &bpf_skc_to_tcp6_sock_proto;
10912 case BPF_FUNC_skc_to_tcp_sock:
10913 func = &bpf_skc_to_tcp_sock_proto;
10915 case BPF_FUNC_skc_to_tcp_timewait_sock:
10916 func = &bpf_skc_to_tcp_timewait_sock_proto;
10918 case BPF_FUNC_skc_to_tcp_request_sock:
10919 func = &bpf_skc_to_tcp_request_sock_proto;
10921 case BPF_FUNC_skc_to_udp6_sock:
10922 func = &bpf_skc_to_udp6_sock_proto;
10924 case BPF_FUNC_ktime_get_coarse_ns:
10925 return &bpf_ktime_get_coarse_ns_proto;
10927 return bpf_base_func_proto(func_id);
10930 if (!perfmon_capable())