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);
2134 /* At ingress, the mac header has already been pulled once.
2135 * At egress, skb_pospull_rcsum has to be done in case that
2136 * the skb is originated from ingress (i.e. a forwarded skb)
2137 * to ensure that rcsum starts at net header.
2139 if (!skb_at_tc_ingress(skb))
2140 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2142 skb_pop_mac_header(skb);
2143 skb_reset_mac_len(skb);
2144 return flags & BPF_F_INGRESS ?
2145 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2148 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2151 /* Verify that a link layer header is carried */
2152 if (unlikely(skb->mac_header >= skb->network_header || skb->len == 0)) {
2157 bpf_push_mac_rcsum(skb);
2158 return flags & BPF_F_INGRESS ?
2159 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2162 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2165 if (dev_is_mac_header_xmit(dev))
2166 return __bpf_redirect_common(skb, dev, flags);
2168 return __bpf_redirect_no_mac(skb, dev, flags);
2171 #if IS_ENABLED(CONFIG_IPV6)
2172 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2173 struct net_device *dev, struct bpf_nh_params *nh)
2175 u32 hh_len = LL_RESERVED_SPACE(dev);
2176 const struct in6_addr *nexthop;
2177 struct dst_entry *dst = NULL;
2178 struct neighbour *neigh;
2180 if (dev_xmit_recursion()) {
2181 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2188 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2189 skb = skb_expand_head(skb, hh_len);
2197 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2198 &ipv6_hdr(skb)->daddr);
2200 nexthop = &nh->ipv6_nh;
2202 neigh = ip_neigh_gw6(dev, nexthop);
2203 if (likely(!IS_ERR(neigh))) {
2206 sock_confirm_neigh(skb, neigh);
2207 dev_xmit_recursion_inc();
2208 ret = neigh_output(neigh, skb, false);
2209 dev_xmit_recursion_dec();
2210 rcu_read_unlock_bh();
2213 rcu_read_unlock_bh();
2215 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2221 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2222 struct bpf_nh_params *nh)
2224 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2225 struct net *net = dev_net(dev);
2226 int err, ret = NET_XMIT_DROP;
2229 struct dst_entry *dst;
2230 struct flowi6 fl6 = {
2231 .flowi6_flags = FLOWI_FLAG_ANYSRC,
2232 .flowi6_mark = skb->mark,
2233 .flowlabel = ip6_flowinfo(ip6h),
2234 .flowi6_oif = dev->ifindex,
2235 .flowi6_proto = ip6h->nexthdr,
2236 .daddr = ip6h->daddr,
2237 .saddr = ip6h->saddr,
2240 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2244 skb_dst_set(skb, dst);
2245 } else if (nh->nh_family != AF_INET6) {
2249 err = bpf_out_neigh_v6(net, skb, dev, nh);
2250 if (unlikely(net_xmit_eval(err)))
2251 dev->stats.tx_errors++;
2253 ret = NET_XMIT_SUCCESS;
2256 dev->stats.tx_errors++;
2262 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2263 struct bpf_nh_params *nh)
2266 return NET_XMIT_DROP;
2268 #endif /* CONFIG_IPV6 */
2270 #if IS_ENABLED(CONFIG_INET)
2271 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2272 struct net_device *dev, struct bpf_nh_params *nh)
2274 u32 hh_len = LL_RESERVED_SPACE(dev);
2275 struct neighbour *neigh;
2276 bool is_v6gw = false;
2278 if (dev_xmit_recursion()) {
2279 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2286 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2287 skb = skb_expand_head(skb, hh_len);
2294 struct dst_entry *dst = skb_dst(skb);
2295 struct rtable *rt = container_of(dst, struct rtable, dst);
2297 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2298 } else if (nh->nh_family == AF_INET6) {
2299 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2301 } else if (nh->nh_family == AF_INET) {
2302 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2304 rcu_read_unlock_bh();
2308 if (likely(!IS_ERR(neigh))) {
2311 sock_confirm_neigh(skb, neigh);
2312 dev_xmit_recursion_inc();
2313 ret = neigh_output(neigh, skb, is_v6gw);
2314 dev_xmit_recursion_dec();
2315 rcu_read_unlock_bh();
2318 rcu_read_unlock_bh();
2324 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2325 struct bpf_nh_params *nh)
2327 const struct iphdr *ip4h = ip_hdr(skb);
2328 struct net *net = dev_net(dev);
2329 int err, ret = NET_XMIT_DROP;
2332 struct flowi4 fl4 = {
2333 .flowi4_flags = FLOWI_FLAG_ANYSRC,
2334 .flowi4_mark = skb->mark,
2335 .flowi4_tos = RT_TOS(ip4h->tos),
2336 .flowi4_oif = dev->ifindex,
2337 .flowi4_proto = ip4h->protocol,
2338 .daddr = ip4h->daddr,
2339 .saddr = ip4h->saddr,
2343 rt = ip_route_output_flow(net, &fl4, NULL);
2346 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2351 skb_dst_set(skb, &rt->dst);
2354 err = bpf_out_neigh_v4(net, skb, dev, nh);
2355 if (unlikely(net_xmit_eval(err)))
2356 dev->stats.tx_errors++;
2358 ret = NET_XMIT_SUCCESS;
2361 dev->stats.tx_errors++;
2367 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2368 struct bpf_nh_params *nh)
2371 return NET_XMIT_DROP;
2373 #endif /* CONFIG_INET */
2375 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2376 struct bpf_nh_params *nh)
2378 struct ethhdr *ethh = eth_hdr(skb);
2380 if (unlikely(skb->mac_header >= skb->network_header))
2382 bpf_push_mac_rcsum(skb);
2383 if (is_multicast_ether_addr(ethh->h_dest))
2386 skb_pull(skb, sizeof(*ethh));
2387 skb_unset_mac_header(skb);
2388 skb_reset_network_header(skb);
2390 if (skb->protocol == htons(ETH_P_IP))
2391 return __bpf_redirect_neigh_v4(skb, dev, nh);
2392 else if (skb->protocol == htons(ETH_P_IPV6))
2393 return __bpf_redirect_neigh_v6(skb, dev, nh);
2399 /* Internal, non-exposed redirect flags. */
2401 BPF_F_NEIGH = (1ULL << 1),
2402 BPF_F_PEER = (1ULL << 2),
2403 BPF_F_NEXTHOP = (1ULL << 3),
2404 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2407 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2409 struct net_device *dev;
2410 struct sk_buff *clone;
2413 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2416 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2420 clone = skb_clone(skb, GFP_ATOMIC);
2421 if (unlikely(!clone))
2424 /* For direct write, we need to keep the invariant that the skbs
2425 * we're dealing with need to be uncloned. Should uncloning fail
2426 * here, we need to free the just generated clone to unclone once
2429 ret = bpf_try_make_head_writable(skb);
2430 if (unlikely(ret)) {
2435 return __bpf_redirect(clone, dev, flags);
2438 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2439 .func = bpf_clone_redirect,
2441 .ret_type = RET_INTEGER,
2442 .arg1_type = ARG_PTR_TO_CTX,
2443 .arg2_type = ARG_ANYTHING,
2444 .arg3_type = ARG_ANYTHING,
2447 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2448 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2450 int skb_do_redirect(struct sk_buff *skb)
2452 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2453 struct net *net = dev_net(skb->dev);
2454 struct net_device *dev;
2455 u32 flags = ri->flags;
2457 dev = dev_get_by_index_rcu(net, ri->tgt_index);
2462 if (flags & BPF_F_PEER) {
2463 const struct net_device_ops *ops = dev->netdev_ops;
2465 if (unlikely(!ops->ndo_get_peer_dev ||
2466 !skb_at_tc_ingress(skb)))
2468 dev = ops->ndo_get_peer_dev(dev);
2469 if (unlikely(!dev ||
2470 !(dev->flags & IFF_UP) ||
2471 net_eq(net, dev_net(dev))))
2476 return flags & BPF_F_NEIGH ?
2477 __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2479 __bpf_redirect(skb, dev, flags);
2485 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2487 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2489 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2493 ri->tgt_index = ifindex;
2495 return TC_ACT_REDIRECT;
2498 static const struct bpf_func_proto bpf_redirect_proto = {
2499 .func = bpf_redirect,
2501 .ret_type = RET_INTEGER,
2502 .arg1_type = ARG_ANYTHING,
2503 .arg2_type = ARG_ANYTHING,
2506 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2508 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2510 if (unlikely(flags))
2513 ri->flags = BPF_F_PEER;
2514 ri->tgt_index = ifindex;
2516 return TC_ACT_REDIRECT;
2519 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2520 .func = bpf_redirect_peer,
2522 .ret_type = RET_INTEGER,
2523 .arg1_type = ARG_ANYTHING,
2524 .arg2_type = ARG_ANYTHING,
2527 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2528 int, plen, u64, flags)
2530 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2532 if (unlikely((plen && plen < sizeof(*params)) || flags))
2535 ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2536 ri->tgt_index = ifindex;
2538 BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2540 memcpy(&ri->nh, params, sizeof(ri->nh));
2542 return TC_ACT_REDIRECT;
2545 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2546 .func = bpf_redirect_neigh,
2548 .ret_type = RET_INTEGER,
2549 .arg1_type = ARG_ANYTHING,
2550 .arg2_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2551 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
2552 .arg4_type = ARG_ANYTHING,
2555 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2557 msg->apply_bytes = bytes;
2561 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2562 .func = bpf_msg_apply_bytes,
2564 .ret_type = RET_INTEGER,
2565 .arg1_type = ARG_PTR_TO_CTX,
2566 .arg2_type = ARG_ANYTHING,
2569 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2571 msg->cork_bytes = bytes;
2575 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2576 .func = bpf_msg_cork_bytes,
2578 .ret_type = RET_INTEGER,
2579 .arg1_type = ARG_PTR_TO_CTX,
2580 .arg2_type = ARG_ANYTHING,
2583 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2584 u32, end, u64, flags)
2586 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2587 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2588 struct scatterlist *sge;
2589 u8 *raw, *to, *from;
2592 if (unlikely(flags || end <= start))
2595 /* First find the starting scatterlist element */
2599 len = sk_msg_elem(msg, i)->length;
2600 if (start < offset + len)
2602 sk_msg_iter_var_next(i);
2603 } while (i != msg->sg.end);
2605 if (unlikely(start >= offset + len))
2609 /* The start may point into the sg element so we need to also
2610 * account for the headroom.
2612 bytes_sg_total = start - offset + bytes;
2613 if (!test_bit(i, &msg->sg.copy) && bytes_sg_total <= len)
2616 /* At this point we need to linearize multiple scatterlist
2617 * elements or a single shared page. Either way we need to
2618 * copy into a linear buffer exclusively owned by BPF. Then
2619 * place the buffer in the scatterlist and fixup the original
2620 * entries by removing the entries now in the linear buffer
2621 * and shifting the remaining entries. For now we do not try
2622 * to copy partial entries to avoid complexity of running out
2623 * of sg_entry slots. The downside is reading a single byte
2624 * will copy the entire sg entry.
2627 copy += sk_msg_elem(msg, i)->length;
2628 sk_msg_iter_var_next(i);
2629 if (bytes_sg_total <= copy)
2631 } while (i != msg->sg.end);
2634 if (unlikely(bytes_sg_total > copy))
2637 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2639 if (unlikely(!page))
2642 raw = page_address(page);
2645 sge = sk_msg_elem(msg, i);
2646 from = sg_virt(sge);
2650 memcpy(to, from, len);
2653 put_page(sg_page(sge));
2655 sk_msg_iter_var_next(i);
2656 } while (i != last_sge);
2658 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2660 /* To repair sg ring we need to shift entries. If we only
2661 * had a single entry though we can just replace it and
2662 * be done. Otherwise walk the ring and shift the entries.
2664 WARN_ON_ONCE(last_sge == first_sge);
2665 shift = last_sge > first_sge ?
2666 last_sge - first_sge - 1 :
2667 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2672 sk_msg_iter_var_next(i);
2676 if (i + shift >= NR_MSG_FRAG_IDS)
2677 move_from = i + shift - NR_MSG_FRAG_IDS;
2679 move_from = i + shift;
2680 if (move_from == msg->sg.end)
2683 msg->sg.data[i] = msg->sg.data[move_from];
2684 msg->sg.data[move_from].length = 0;
2685 msg->sg.data[move_from].page_link = 0;
2686 msg->sg.data[move_from].offset = 0;
2687 sk_msg_iter_var_next(i);
2690 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2691 msg->sg.end - shift + NR_MSG_FRAG_IDS :
2692 msg->sg.end - shift;
2694 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2695 msg->data_end = msg->data + bytes;
2699 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2700 .func = bpf_msg_pull_data,
2702 .ret_type = RET_INTEGER,
2703 .arg1_type = ARG_PTR_TO_CTX,
2704 .arg2_type = ARG_ANYTHING,
2705 .arg3_type = ARG_ANYTHING,
2706 .arg4_type = ARG_ANYTHING,
2709 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2710 u32, len, u64, flags)
2712 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2713 u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2714 u8 *raw, *to, *from;
2717 if (unlikely(flags))
2720 if (unlikely(len == 0))
2723 /* First find the starting scatterlist element */
2727 l = sk_msg_elem(msg, i)->length;
2729 if (start < offset + l)
2731 sk_msg_iter_var_next(i);
2732 } while (i != msg->sg.end);
2734 if (start >= offset + l)
2737 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2739 /* If no space available will fallback to copy, we need at
2740 * least one scatterlist elem available to push data into
2741 * when start aligns to the beginning of an element or two
2742 * when it falls inside an element. We handle the start equals
2743 * offset case because its the common case for inserting a
2746 if (!space || (space == 1 && start != offset))
2747 copy = msg->sg.data[i].length;
2749 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2750 get_order(copy + len));
2751 if (unlikely(!page))
2757 raw = page_address(page);
2759 psge = sk_msg_elem(msg, i);
2760 front = start - offset;
2761 back = psge->length - front;
2762 from = sg_virt(psge);
2765 memcpy(raw, from, front);
2769 to = raw + front + len;
2771 memcpy(to, from, back);
2774 put_page(sg_page(psge));
2775 } else if (start - offset) {
2776 psge = sk_msg_elem(msg, i);
2777 rsge = sk_msg_elem_cpy(msg, i);
2779 psge->length = start - offset;
2780 rsge.length -= psge->length;
2781 rsge.offset += start;
2783 sk_msg_iter_var_next(i);
2784 sg_unmark_end(psge);
2785 sg_unmark_end(&rsge);
2786 sk_msg_iter_next(msg, end);
2789 /* Slot(s) to place newly allocated data */
2792 /* Shift one or two slots as needed */
2794 sge = sk_msg_elem_cpy(msg, i);
2796 sk_msg_iter_var_next(i);
2797 sg_unmark_end(&sge);
2798 sk_msg_iter_next(msg, end);
2800 nsge = sk_msg_elem_cpy(msg, i);
2802 sk_msg_iter_var_next(i);
2803 nnsge = sk_msg_elem_cpy(msg, i);
2806 while (i != msg->sg.end) {
2807 msg->sg.data[i] = sge;
2809 sk_msg_iter_var_next(i);
2812 nnsge = sk_msg_elem_cpy(msg, i);
2814 nsge = sk_msg_elem_cpy(msg, i);
2819 /* Place newly allocated data buffer */
2820 sk_mem_charge(msg->sk, len);
2821 msg->sg.size += len;
2822 __clear_bit(new, &msg->sg.copy);
2823 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2825 get_page(sg_page(&rsge));
2826 sk_msg_iter_var_next(new);
2827 msg->sg.data[new] = rsge;
2830 sk_msg_compute_data_pointers(msg);
2834 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2835 .func = bpf_msg_push_data,
2837 .ret_type = RET_INTEGER,
2838 .arg1_type = ARG_PTR_TO_CTX,
2839 .arg2_type = ARG_ANYTHING,
2840 .arg3_type = ARG_ANYTHING,
2841 .arg4_type = ARG_ANYTHING,
2844 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2850 sk_msg_iter_var_next(i);
2851 msg->sg.data[prev] = msg->sg.data[i];
2852 } while (i != msg->sg.end);
2854 sk_msg_iter_prev(msg, end);
2857 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2859 struct scatterlist tmp, sge;
2861 sk_msg_iter_next(msg, end);
2862 sge = sk_msg_elem_cpy(msg, i);
2863 sk_msg_iter_var_next(i);
2864 tmp = sk_msg_elem_cpy(msg, i);
2866 while (i != msg->sg.end) {
2867 msg->sg.data[i] = sge;
2868 sk_msg_iter_var_next(i);
2870 tmp = sk_msg_elem_cpy(msg, i);
2874 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2875 u32, len, u64, flags)
2877 u32 i = 0, l = 0, space, offset = 0;
2878 u64 last = start + len;
2881 if (unlikely(flags))
2884 /* First find the starting scatterlist element */
2888 l = sk_msg_elem(msg, i)->length;
2890 if (start < offset + l)
2892 sk_msg_iter_var_next(i);
2893 } while (i != msg->sg.end);
2895 /* Bounds checks: start and pop must be inside message */
2896 if (start >= offset + l || last >= msg->sg.size)
2899 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2902 /* --------------| offset
2903 * -| start |-------- len -------|
2905 * |----- a ----|-------- pop -------|----- b ----|
2906 * |______________________________________________| length
2909 * a: region at front of scatter element to save
2910 * b: region at back of scatter element to save when length > A + pop
2911 * pop: region to pop from element, same as input 'pop' here will be
2912 * decremented below per iteration.
2914 * Two top-level cases to handle when start != offset, first B is non
2915 * zero and second B is zero corresponding to when a pop includes more
2918 * Then if B is non-zero AND there is no space allocate space and
2919 * compact A, B regions into page. If there is space shift ring to
2920 * the rigth free'ing the next element in ring to place B, leaving
2921 * A untouched except to reduce length.
2923 if (start != offset) {
2924 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2926 int b = sge->length - pop - a;
2928 sk_msg_iter_var_next(i);
2930 if (pop < sge->length - a) {
2933 sk_msg_shift_right(msg, i);
2934 nsge = sk_msg_elem(msg, i);
2935 get_page(sg_page(sge));
2938 b, sge->offset + pop + a);
2940 struct page *page, *orig;
2943 page = alloc_pages(__GFP_NOWARN |
2944 __GFP_COMP | GFP_ATOMIC,
2946 if (unlikely(!page))
2950 orig = sg_page(sge);
2951 from = sg_virt(sge);
2952 to = page_address(page);
2953 memcpy(to, from, a);
2954 memcpy(to + a, from + a + pop, b);
2955 sg_set_page(sge, page, a + b, 0);
2959 } else if (pop >= sge->length - a) {
2960 pop -= (sge->length - a);
2965 /* From above the current layout _must_ be as follows,
2970 * |---- pop ---|---------------- b ------------|
2971 * |____________________________________________| length
2973 * Offset and start of the current msg elem are equal because in the
2974 * previous case we handled offset != start and either consumed the
2975 * entire element and advanced to the next element OR pop == 0.
2977 * Two cases to handle here are first pop is less than the length
2978 * leaving some remainder b above. Simply adjust the element's layout
2979 * in this case. Or pop >= length of the element so that b = 0. In this
2980 * case advance to next element decrementing pop.
2983 struct scatterlist *sge = sk_msg_elem(msg, i);
2985 if (pop < sge->length) {
2991 sk_msg_shift_left(msg, i);
2993 sk_msg_iter_var_next(i);
2996 sk_mem_uncharge(msg->sk, len - pop);
2997 msg->sg.size -= (len - pop);
2998 sk_msg_compute_data_pointers(msg);
3002 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
3003 .func = bpf_msg_pop_data,
3005 .ret_type = RET_INTEGER,
3006 .arg1_type = ARG_PTR_TO_CTX,
3007 .arg2_type = ARG_ANYTHING,
3008 .arg3_type = ARG_ANYTHING,
3009 .arg4_type = ARG_ANYTHING,
3012 #ifdef CONFIG_CGROUP_NET_CLASSID
3013 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3015 return __task_get_classid(current);
3018 static const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3019 .func = bpf_get_cgroup_classid_curr,
3021 .ret_type = RET_INTEGER,
3024 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3026 struct sock *sk = skb_to_full_sk(skb);
3028 if (!sk || !sk_fullsock(sk))
3031 return sock_cgroup_classid(&sk->sk_cgrp_data);
3034 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3035 .func = bpf_skb_cgroup_classid,
3037 .ret_type = RET_INTEGER,
3038 .arg1_type = ARG_PTR_TO_CTX,
3042 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3044 return task_get_classid(skb);
3047 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3048 .func = bpf_get_cgroup_classid,
3050 .ret_type = RET_INTEGER,
3051 .arg1_type = ARG_PTR_TO_CTX,
3054 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3056 return dst_tclassid(skb);
3059 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3060 .func = bpf_get_route_realm,
3062 .ret_type = RET_INTEGER,
3063 .arg1_type = ARG_PTR_TO_CTX,
3066 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3068 /* If skb_clear_hash() was called due to mangling, we can
3069 * trigger SW recalculation here. Later access to hash
3070 * can then use the inline skb->hash via context directly
3071 * instead of calling this helper again.
3073 return skb_get_hash(skb);
3076 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3077 .func = bpf_get_hash_recalc,
3079 .ret_type = RET_INTEGER,
3080 .arg1_type = ARG_PTR_TO_CTX,
3083 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3085 /* After all direct packet write, this can be used once for
3086 * triggering a lazy recalc on next skb_get_hash() invocation.
3088 skb_clear_hash(skb);
3092 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3093 .func = bpf_set_hash_invalid,
3095 .ret_type = RET_INTEGER,
3096 .arg1_type = ARG_PTR_TO_CTX,
3099 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3101 /* Set user specified hash as L4(+), so that it gets returned
3102 * on skb_get_hash() call unless BPF prog later on triggers a
3105 __skb_set_sw_hash(skb, hash, true);
3109 static const struct bpf_func_proto bpf_set_hash_proto = {
3110 .func = bpf_set_hash,
3112 .ret_type = RET_INTEGER,
3113 .arg1_type = ARG_PTR_TO_CTX,
3114 .arg2_type = ARG_ANYTHING,
3117 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3122 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3123 vlan_proto != htons(ETH_P_8021AD)))
3124 vlan_proto = htons(ETH_P_8021Q);
3126 bpf_push_mac_rcsum(skb);
3127 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3128 bpf_pull_mac_rcsum(skb);
3130 bpf_compute_data_pointers(skb);
3134 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3135 .func = bpf_skb_vlan_push,
3137 .ret_type = RET_INTEGER,
3138 .arg1_type = ARG_PTR_TO_CTX,
3139 .arg2_type = ARG_ANYTHING,
3140 .arg3_type = ARG_ANYTHING,
3143 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3147 bpf_push_mac_rcsum(skb);
3148 ret = skb_vlan_pop(skb);
3149 bpf_pull_mac_rcsum(skb);
3151 bpf_compute_data_pointers(skb);
3155 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3156 .func = bpf_skb_vlan_pop,
3158 .ret_type = RET_INTEGER,
3159 .arg1_type = ARG_PTR_TO_CTX,
3162 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3164 /* Caller already did skb_cow() with len as headroom,
3165 * so no need to do it here.
3168 memmove(skb->data, skb->data + len, off);
3169 memset(skb->data + off, 0, len);
3171 /* No skb_postpush_rcsum(skb, skb->data + off, len)
3172 * needed here as it does not change the skb->csum
3173 * result for checksum complete when summing over
3179 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3181 /* skb_ensure_writable() is not needed here, as we're
3182 * already working on an uncloned skb.
3184 if (unlikely(!pskb_may_pull(skb, off + len)))
3187 skb_postpull_rcsum(skb, skb->data + off, len);
3188 memmove(skb->data + len, skb->data, off);
3189 __skb_pull(skb, len);
3194 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3196 bool trans_same = skb->transport_header == skb->network_header;
3199 /* There's no need for __skb_push()/__skb_pull() pair to
3200 * get to the start of the mac header as we're guaranteed
3201 * to always start from here under eBPF.
3203 ret = bpf_skb_generic_push(skb, off, len);
3205 skb->mac_header -= len;
3206 skb->network_header -= len;
3208 skb->transport_header = skb->network_header;
3214 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3216 bool trans_same = skb->transport_header == skb->network_header;
3219 /* Same here, __skb_push()/__skb_pull() pair not needed. */
3220 ret = bpf_skb_generic_pop(skb, off, len);
3222 skb->mac_header += len;
3223 skb->network_header += len;
3225 skb->transport_header = skb->network_header;
3231 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3233 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3234 u32 off = skb_mac_header_len(skb);
3237 ret = skb_cow(skb, len_diff);
3238 if (unlikely(ret < 0))
3241 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3242 if (unlikely(ret < 0))
3245 if (skb_is_gso(skb)) {
3246 struct skb_shared_info *shinfo = skb_shinfo(skb);
3248 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */
3249 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3250 shinfo->gso_type &= ~SKB_GSO_TCPV4;
3251 shinfo->gso_type |= SKB_GSO_TCPV6;
3255 skb->protocol = htons(ETH_P_IPV6);
3256 skb_clear_hash(skb);
3261 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3263 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3264 u32 off = skb_mac_header_len(skb);
3267 ret = skb_unclone(skb, GFP_ATOMIC);
3268 if (unlikely(ret < 0))
3271 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3272 if (unlikely(ret < 0))
3275 if (skb_is_gso(skb)) {
3276 struct skb_shared_info *shinfo = skb_shinfo(skb);
3278 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */
3279 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3280 shinfo->gso_type &= ~SKB_GSO_TCPV6;
3281 shinfo->gso_type |= SKB_GSO_TCPV4;
3285 skb->protocol = htons(ETH_P_IP);
3286 skb_clear_hash(skb);
3291 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3293 __be16 from_proto = skb->protocol;
3295 if (from_proto == htons(ETH_P_IP) &&
3296 to_proto == htons(ETH_P_IPV6))
3297 return bpf_skb_proto_4_to_6(skb);
3299 if (from_proto == htons(ETH_P_IPV6) &&
3300 to_proto == htons(ETH_P_IP))
3301 return bpf_skb_proto_6_to_4(skb);
3306 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3311 if (unlikely(flags))
3314 /* General idea is that this helper does the basic groundwork
3315 * needed for changing the protocol, and eBPF program fills the
3316 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3317 * and other helpers, rather than passing a raw buffer here.
3319 * The rationale is to keep this minimal and without a need to
3320 * deal with raw packet data. F.e. even if we would pass buffers
3321 * here, the program still needs to call the bpf_lX_csum_replace()
3322 * helpers anyway. Plus, this way we keep also separation of
3323 * concerns, since f.e. bpf_skb_store_bytes() should only take
3326 * Currently, additional options and extension header space are
3327 * not supported, but flags register is reserved so we can adapt
3328 * that. For offloads, we mark packet as dodgy, so that headers
3329 * need to be verified first.
3331 ret = bpf_skb_proto_xlat(skb, proto);
3332 bpf_compute_data_pointers(skb);
3336 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3337 .func = bpf_skb_change_proto,
3339 .ret_type = RET_INTEGER,
3340 .arg1_type = ARG_PTR_TO_CTX,
3341 .arg2_type = ARG_ANYTHING,
3342 .arg3_type = ARG_ANYTHING,
3345 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3347 /* We only allow a restricted subset to be changed for now. */
3348 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3349 !skb_pkt_type_ok(pkt_type)))
3352 skb->pkt_type = pkt_type;
3356 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3357 .func = bpf_skb_change_type,
3359 .ret_type = RET_INTEGER,
3360 .arg1_type = ARG_PTR_TO_CTX,
3361 .arg2_type = ARG_ANYTHING,
3364 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3366 switch (skb->protocol) {
3367 case htons(ETH_P_IP):
3368 return sizeof(struct iphdr);
3369 case htons(ETH_P_IPV6):
3370 return sizeof(struct ipv6hdr);
3376 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3377 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3379 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \
3380 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3381 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3382 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3383 BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \
3384 BPF_F_ADJ_ROOM_ENCAP_L2( \
3385 BPF_ADJ_ROOM_ENCAP_L2_MASK))
3387 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3390 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3391 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3392 u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3393 unsigned int gso_type = SKB_GSO_DODGY;
3396 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3397 /* udp gso_size delineates datagrams, only allow if fixed */
3398 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3399 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3403 ret = skb_cow_head(skb, len_diff);
3404 if (unlikely(ret < 0))
3408 if (skb->protocol != htons(ETH_P_IP) &&
3409 skb->protocol != htons(ETH_P_IPV6))
3412 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3413 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3416 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3417 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3420 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH &&
3421 inner_mac_len < ETH_HLEN)
3424 if (skb->encapsulation)
3427 mac_len = skb->network_header - skb->mac_header;
3428 inner_net = skb->network_header;
3429 if (inner_mac_len > len_diff)
3431 inner_trans = skb->transport_header;
3434 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3435 if (unlikely(ret < 0))
3439 skb->inner_mac_header = inner_net - inner_mac_len;
3440 skb->inner_network_header = inner_net;
3441 skb->inner_transport_header = inner_trans;
3443 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH)
3444 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
3446 skb_set_inner_protocol(skb, skb->protocol);
3448 skb->encapsulation = 1;
3449 skb_set_network_header(skb, mac_len);
3451 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3452 gso_type |= SKB_GSO_UDP_TUNNEL;
3453 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3454 gso_type |= SKB_GSO_GRE;
3455 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3456 gso_type |= SKB_GSO_IPXIP6;
3457 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3458 gso_type |= SKB_GSO_IPXIP4;
3460 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3461 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3462 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3463 sizeof(struct ipv6hdr) :
3464 sizeof(struct iphdr);
3466 skb_set_transport_header(skb, mac_len + nh_len);
3469 /* Match skb->protocol to new outer l3 protocol */
3470 if (skb->protocol == htons(ETH_P_IP) &&
3471 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3472 skb->protocol = htons(ETH_P_IPV6);
3473 else if (skb->protocol == htons(ETH_P_IPV6) &&
3474 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3475 skb->protocol = htons(ETH_P_IP);
3478 if (skb_is_gso(skb)) {
3479 struct skb_shared_info *shinfo = skb_shinfo(skb);
3481 /* Due to header grow, MSS needs to be downgraded. */
3482 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3483 skb_decrease_gso_size(shinfo, len_diff);
3485 /* Header must be checked, and gso_segs recomputed. */
3486 shinfo->gso_type |= gso_type;
3487 shinfo->gso_segs = 0;
3493 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3498 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3499 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3502 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3503 /* udp gso_size delineates datagrams, only allow if fixed */
3504 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3505 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3509 ret = skb_unclone(skb, GFP_ATOMIC);
3510 if (unlikely(ret < 0))
3513 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3514 if (unlikely(ret < 0))
3517 if (skb_is_gso(skb)) {
3518 struct skb_shared_info *shinfo = skb_shinfo(skb);
3520 /* Due to header shrink, MSS can be upgraded. */
3521 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3522 skb_increase_gso_size(shinfo, len_diff);
3524 /* Header must be checked, and gso_segs recomputed. */
3525 shinfo->gso_type |= SKB_GSO_DODGY;
3526 shinfo->gso_segs = 0;
3532 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3534 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3535 u32, mode, u64, flags)
3537 u32 len_diff_abs = abs(len_diff);
3538 bool shrink = len_diff < 0;
3541 if (unlikely(flags || mode))
3543 if (unlikely(len_diff_abs > 0xfffU))
3547 ret = skb_cow(skb, len_diff);
3548 if (unlikely(ret < 0))
3550 __skb_push(skb, len_diff_abs);
3551 memset(skb->data, 0, len_diff_abs);
3553 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3555 __skb_pull(skb, len_diff_abs);
3557 if (tls_sw_has_ctx_rx(skb->sk)) {
3558 struct strp_msg *rxm = strp_msg(skb);
3560 rxm->full_len += len_diff;
3565 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3566 .func = sk_skb_adjust_room,
3568 .ret_type = RET_INTEGER,
3569 .arg1_type = ARG_PTR_TO_CTX,
3570 .arg2_type = ARG_ANYTHING,
3571 .arg3_type = ARG_ANYTHING,
3572 .arg4_type = ARG_ANYTHING,
3575 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3576 u32, mode, u64, flags)
3578 u32 len_cur, len_diff_abs = abs(len_diff);
3579 u32 len_min = bpf_skb_net_base_len(skb);
3580 u32 len_max = BPF_SKB_MAX_LEN;
3581 __be16 proto = skb->protocol;
3582 bool shrink = len_diff < 0;
3586 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3587 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3589 if (unlikely(len_diff_abs > 0xfffU))
3591 if (unlikely(proto != htons(ETH_P_IP) &&
3592 proto != htons(ETH_P_IPV6)))
3595 off = skb_mac_header_len(skb);
3597 case BPF_ADJ_ROOM_NET:
3598 off += bpf_skb_net_base_len(skb);
3600 case BPF_ADJ_ROOM_MAC:
3606 len_cur = skb->len - skb_network_offset(skb);
3607 if ((shrink && (len_diff_abs >= len_cur ||
3608 len_cur - len_diff_abs < len_min)) ||
3609 (!shrink && (skb->len + len_diff_abs > len_max &&
3613 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3614 bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3615 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3616 __skb_reset_checksum_unnecessary(skb);
3618 bpf_compute_data_pointers(skb);
3622 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3623 .func = bpf_skb_adjust_room,
3625 .ret_type = RET_INTEGER,
3626 .arg1_type = ARG_PTR_TO_CTX,
3627 .arg2_type = ARG_ANYTHING,
3628 .arg3_type = ARG_ANYTHING,
3629 .arg4_type = ARG_ANYTHING,
3632 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3634 u32 min_len = skb_network_offset(skb);
3636 if (skb_transport_header_was_set(skb))
3637 min_len = skb_transport_offset(skb);
3638 if (skb->ip_summed == CHECKSUM_PARTIAL)
3639 min_len = skb_checksum_start_offset(skb) +
3640 skb->csum_offset + sizeof(__sum16);
3644 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3646 unsigned int old_len = skb->len;
3649 ret = __skb_grow_rcsum(skb, new_len);
3651 memset(skb->data + old_len, 0, new_len - old_len);
3655 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3657 return __skb_trim_rcsum(skb, new_len);
3660 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3663 u32 max_len = BPF_SKB_MAX_LEN;
3664 u32 min_len = __bpf_skb_min_len(skb);
3667 if (unlikely(flags || new_len > max_len || new_len < min_len))
3669 if (skb->encapsulation)
3672 /* The basic idea of this helper is that it's performing the
3673 * needed work to either grow or trim an skb, and eBPF program
3674 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3675 * bpf_lX_csum_replace() and others rather than passing a raw
3676 * buffer here. This one is a slow path helper and intended
3677 * for replies with control messages.
3679 * Like in bpf_skb_change_proto(), we want to keep this rather
3680 * minimal and without protocol specifics so that we are able
3681 * to separate concerns as in bpf_skb_store_bytes() should only
3682 * be the one responsible for writing buffers.
3684 * It's really expected to be a slow path operation here for
3685 * control message replies, so we're implicitly linearizing,
3686 * uncloning and drop offloads from the skb by this.
3688 ret = __bpf_try_make_writable(skb, skb->len);
3690 if (new_len > skb->len)
3691 ret = bpf_skb_grow_rcsum(skb, new_len);
3692 else if (new_len < skb->len)
3693 ret = bpf_skb_trim_rcsum(skb, new_len);
3694 if (!ret && skb_is_gso(skb))
3700 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3703 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3705 bpf_compute_data_pointers(skb);
3709 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3710 .func = bpf_skb_change_tail,
3712 .ret_type = RET_INTEGER,
3713 .arg1_type = ARG_PTR_TO_CTX,
3714 .arg2_type = ARG_ANYTHING,
3715 .arg3_type = ARG_ANYTHING,
3718 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3721 return __bpf_skb_change_tail(skb, new_len, flags);
3724 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3725 .func = sk_skb_change_tail,
3727 .ret_type = RET_INTEGER,
3728 .arg1_type = ARG_PTR_TO_CTX,
3729 .arg2_type = ARG_ANYTHING,
3730 .arg3_type = ARG_ANYTHING,
3733 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3736 u32 max_len = BPF_SKB_MAX_LEN;
3737 u32 new_len = skb->len + head_room;
3740 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3741 new_len < skb->len))
3744 ret = skb_cow(skb, head_room);
3746 /* Idea for this helper is that we currently only
3747 * allow to expand on mac header. This means that
3748 * skb->protocol network header, etc, stay as is.
3749 * Compared to bpf_skb_change_tail(), we're more
3750 * flexible due to not needing to linearize or
3751 * reset GSO. Intention for this helper is to be
3752 * used by an L3 skb that needs to push mac header
3753 * for redirection into L2 device.
3755 __skb_push(skb, head_room);
3756 memset(skb->data, 0, head_room);
3757 skb_reset_mac_header(skb);
3758 skb_reset_mac_len(skb);
3764 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3767 int ret = __bpf_skb_change_head(skb, head_room, flags);
3769 bpf_compute_data_pointers(skb);
3773 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3774 .func = bpf_skb_change_head,
3776 .ret_type = RET_INTEGER,
3777 .arg1_type = ARG_PTR_TO_CTX,
3778 .arg2_type = ARG_ANYTHING,
3779 .arg3_type = ARG_ANYTHING,
3782 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3785 return __bpf_skb_change_head(skb, head_room, flags);
3788 static const struct bpf_func_proto sk_skb_change_head_proto = {
3789 .func = sk_skb_change_head,
3791 .ret_type = RET_INTEGER,
3792 .arg1_type = ARG_PTR_TO_CTX,
3793 .arg2_type = ARG_ANYTHING,
3794 .arg3_type = ARG_ANYTHING,
3796 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3798 return xdp_data_meta_unsupported(xdp) ? 0 :
3799 xdp->data - xdp->data_meta;
3802 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3804 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3805 unsigned long metalen = xdp_get_metalen(xdp);
3806 void *data_start = xdp_frame_end + metalen;
3807 void *data = xdp->data + offset;
3809 if (unlikely(data < data_start ||
3810 data > xdp->data_end - ETH_HLEN))
3814 memmove(xdp->data_meta + offset,
3815 xdp->data_meta, metalen);
3816 xdp->data_meta += offset;
3822 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3823 .func = bpf_xdp_adjust_head,
3825 .ret_type = RET_INTEGER,
3826 .arg1_type = ARG_PTR_TO_CTX,
3827 .arg2_type = ARG_ANYTHING,
3830 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3832 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
3833 void *data_end = xdp->data_end + offset;
3835 /* Notice that xdp_data_hard_end have reserved some tailroom */
3836 if (unlikely(data_end > data_hard_end))
3839 /* ALL drivers MUST init xdp->frame_sz, chicken check below */
3840 if (unlikely(xdp->frame_sz > PAGE_SIZE)) {
3841 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz);
3845 if (unlikely(data_end < xdp->data + ETH_HLEN))
3848 /* Clear memory area on grow, can contain uninit kernel memory */
3850 memset(xdp->data_end, 0, offset);
3852 xdp->data_end = data_end;
3857 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3858 .func = bpf_xdp_adjust_tail,
3860 .ret_type = RET_INTEGER,
3861 .arg1_type = ARG_PTR_TO_CTX,
3862 .arg2_type = ARG_ANYTHING,
3865 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3867 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3868 void *meta = xdp->data_meta + offset;
3869 unsigned long metalen = xdp->data - meta;
3871 if (xdp_data_meta_unsupported(xdp))
3873 if (unlikely(meta < xdp_frame_end ||
3876 if (unlikely(xdp_metalen_invalid(metalen)))
3879 xdp->data_meta = meta;
3884 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3885 .func = bpf_xdp_adjust_meta,
3887 .ret_type = RET_INTEGER,
3888 .arg1_type = ARG_PTR_TO_CTX,
3889 .arg2_type = ARG_ANYTHING,
3892 /* XDP_REDIRECT works by a three-step process, implemented in the functions
3895 * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target
3896 * of the redirect and store it (along with some other metadata) in a per-CPU
3897 * struct bpf_redirect_info.
3899 * 2. When the program returns the XDP_REDIRECT return code, the driver will
3900 * call xdp_do_redirect() which will use the information in struct
3901 * bpf_redirect_info to actually enqueue the frame into a map type-specific
3902 * bulk queue structure.
3904 * 3. Before exiting its NAPI poll loop, the driver will call xdp_do_flush(),
3905 * which will flush all the different bulk queues, thus completing the
3908 * Pointers to the map entries will be kept around for this whole sequence of
3909 * steps, protected by RCU. However, there is no top-level rcu_read_lock() in
3910 * the core code; instead, the RCU protection relies on everything happening
3911 * inside a single NAPI poll sequence, which means it's between a pair of calls
3912 * to local_bh_disable()/local_bh_enable().
3914 * The map entries are marked as __rcu and the map code makes sure to
3915 * dereference those pointers with rcu_dereference_check() in a way that works
3916 * for both sections that to hold an rcu_read_lock() and sections that are
3917 * called from NAPI without a separate rcu_read_lock(). The code below does not
3918 * use RCU annotations, but relies on those in the map code.
3920 void xdp_do_flush(void)
3926 EXPORT_SYMBOL_GPL(xdp_do_flush);
3928 void bpf_clear_redirect_map(struct bpf_map *map)
3930 struct bpf_redirect_info *ri;
3933 for_each_possible_cpu(cpu) {
3934 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3935 /* Avoid polluting remote cacheline due to writes if
3936 * not needed. Once we pass this test, we need the
3937 * cmpxchg() to make sure it hasn't been changed in
3938 * the meantime by remote CPU.
3940 if (unlikely(READ_ONCE(ri->map) == map))
3941 cmpxchg(&ri->map, map, NULL);
3945 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
3946 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key);
3948 u32 xdp_master_redirect(struct xdp_buff *xdp)
3950 struct net_device *master, *slave;
3951 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3953 master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev);
3954 slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp);
3955 if (slave && slave != xdp->rxq->dev) {
3956 /* The target device is different from the receiving device, so
3957 * redirect it to the new device.
3958 * Using XDP_REDIRECT gets the correct behaviour from XDP enabled
3959 * drivers to unmap the packet from their rx ring.
3961 ri->tgt_index = slave->ifindex;
3962 ri->map_id = INT_MAX;
3963 ri->map_type = BPF_MAP_TYPE_UNSPEC;
3964 return XDP_REDIRECT;
3968 EXPORT_SYMBOL_GPL(xdp_master_redirect);
3970 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3971 struct bpf_prog *xdp_prog)
3973 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3974 enum bpf_map_type map_type = ri->map_type;
3975 void *fwd = ri->tgt_value;
3976 u32 map_id = ri->map_id;
3977 struct bpf_map *map;
3980 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
3981 ri->map_type = BPF_MAP_TYPE_UNSPEC;
3984 case BPF_MAP_TYPE_DEVMAP:
3986 case BPF_MAP_TYPE_DEVMAP_HASH:
3987 map = READ_ONCE(ri->map);
3988 if (unlikely(map)) {
3989 WRITE_ONCE(ri->map, NULL);
3990 err = dev_map_enqueue_multi(xdp, dev, map,
3991 ri->flags & BPF_F_EXCLUDE_INGRESS);
3993 err = dev_map_enqueue(fwd, xdp, dev);
3996 case BPF_MAP_TYPE_CPUMAP:
3997 err = cpu_map_enqueue(fwd, xdp, dev);
3999 case BPF_MAP_TYPE_XSKMAP:
4000 err = __xsk_map_redirect(fwd, xdp);
4002 case BPF_MAP_TYPE_UNSPEC:
4003 if (map_id == INT_MAX) {
4004 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4005 if (unlikely(!fwd)) {
4009 err = dev_xdp_enqueue(fwd, xdp, dev);
4020 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4023 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4026 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4028 static int xdp_do_generic_redirect_map(struct net_device *dev,
4029 struct sk_buff *skb,
4030 struct xdp_buff *xdp,
4031 struct bpf_prog *xdp_prog,
4033 enum bpf_map_type map_type, u32 map_id)
4035 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4036 struct bpf_map *map;
4040 case BPF_MAP_TYPE_DEVMAP:
4042 case BPF_MAP_TYPE_DEVMAP_HASH:
4043 map = READ_ONCE(ri->map);
4044 if (unlikely(map)) {
4045 WRITE_ONCE(ri->map, NULL);
4046 err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4047 ri->flags & BPF_F_EXCLUDE_INGRESS);
4049 err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4054 case BPF_MAP_TYPE_XSKMAP:
4055 err = xsk_generic_rcv(fwd, xdp);
4060 case BPF_MAP_TYPE_CPUMAP:
4061 err = cpu_map_generic_redirect(fwd, skb);
4070 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4073 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4077 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4078 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4080 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4081 enum bpf_map_type map_type = ri->map_type;
4082 void *fwd = ri->tgt_value;
4083 u32 map_id = ri->map_id;
4086 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4087 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4089 if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4090 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4091 if (unlikely(!fwd)) {
4096 err = xdp_ok_fwd_dev(fwd, skb->len);
4101 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4102 generic_xdp_tx(skb, xdp_prog);
4106 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id);
4108 _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4112 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4114 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4116 if (unlikely(flags))
4119 /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4120 * by map_idr) is used for ifindex based XDP redirect.
4122 ri->tgt_index = ifindex;
4123 ri->map_id = INT_MAX;
4124 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4126 return XDP_REDIRECT;
4129 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4130 .func = bpf_xdp_redirect,
4132 .ret_type = RET_INTEGER,
4133 .arg1_type = ARG_ANYTHING,
4134 .arg2_type = ARG_ANYTHING,
4137 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
4140 return map->ops->map_redirect(map, ifindex, flags);
4143 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4144 .func = bpf_xdp_redirect_map,
4146 .ret_type = RET_INTEGER,
4147 .arg1_type = ARG_CONST_MAP_PTR,
4148 .arg2_type = ARG_ANYTHING,
4149 .arg3_type = ARG_ANYTHING,
4152 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4153 unsigned long off, unsigned long len)
4155 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4159 if (ptr != dst_buff)
4160 memcpy(dst_buff, ptr, len);
4165 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4166 u64, flags, void *, meta, u64, meta_size)
4168 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4170 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4172 if (unlikely(!skb || skb_size > skb->len))
4175 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4179 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4180 .func = bpf_skb_event_output,
4182 .ret_type = RET_INTEGER,
4183 .arg1_type = ARG_PTR_TO_CTX,
4184 .arg2_type = ARG_CONST_MAP_PTR,
4185 .arg3_type = ARG_ANYTHING,
4186 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4187 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4190 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4192 const struct bpf_func_proto bpf_skb_output_proto = {
4193 .func = bpf_skb_event_output,
4195 .ret_type = RET_INTEGER,
4196 .arg1_type = ARG_PTR_TO_BTF_ID,
4197 .arg1_btf_id = &bpf_skb_output_btf_ids[0],
4198 .arg2_type = ARG_CONST_MAP_PTR,
4199 .arg3_type = ARG_ANYTHING,
4200 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4201 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4204 static unsigned short bpf_tunnel_key_af(u64 flags)
4206 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4209 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4210 u32, size, u64, flags)
4212 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4213 u8 compat[sizeof(struct bpf_tunnel_key)];
4217 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
4221 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4225 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4228 case offsetof(struct bpf_tunnel_key, tunnel_label):
4229 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4231 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4232 /* Fixup deprecated structure layouts here, so we have
4233 * a common path later on.
4235 if (ip_tunnel_info_af(info) != AF_INET)
4238 to = (struct bpf_tunnel_key *)compat;
4245 to->tunnel_id = be64_to_cpu(info->key.tun_id);
4246 to->tunnel_tos = info->key.tos;
4247 to->tunnel_ttl = info->key.ttl;
4250 if (flags & BPF_F_TUNINFO_IPV6) {
4251 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4252 sizeof(to->remote_ipv6));
4253 to->tunnel_label = be32_to_cpu(info->key.label);
4255 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4256 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4257 to->tunnel_label = 0;
4260 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4261 memcpy(to_orig, to, size);
4265 memset(to_orig, 0, size);
4269 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4270 .func = bpf_skb_get_tunnel_key,
4272 .ret_type = RET_INTEGER,
4273 .arg1_type = ARG_PTR_TO_CTX,
4274 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4275 .arg3_type = ARG_CONST_SIZE,
4276 .arg4_type = ARG_ANYTHING,
4279 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4281 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4284 if (unlikely(!info ||
4285 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4289 if (unlikely(size < info->options_len)) {
4294 ip_tunnel_info_opts_get(to, info);
4295 if (size > info->options_len)
4296 memset(to + info->options_len, 0, size - info->options_len);
4298 return info->options_len;
4300 memset(to, 0, size);
4304 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4305 .func = bpf_skb_get_tunnel_opt,
4307 .ret_type = RET_INTEGER,
4308 .arg1_type = ARG_PTR_TO_CTX,
4309 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4310 .arg3_type = ARG_CONST_SIZE,
4313 static struct metadata_dst __percpu *md_dst;
4315 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4316 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4318 struct metadata_dst *md = this_cpu_ptr(md_dst);
4319 u8 compat[sizeof(struct bpf_tunnel_key)];
4320 struct ip_tunnel_info *info;
4322 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4323 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
4325 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4327 case offsetof(struct bpf_tunnel_key, tunnel_label):
4328 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4329 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4330 /* Fixup deprecated structure layouts here, so we have
4331 * a common path later on.
4333 memcpy(compat, from, size);
4334 memset(compat + size, 0, sizeof(compat) - size);
4335 from = (const struct bpf_tunnel_key *) compat;
4341 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4346 dst_hold((struct dst_entry *) md);
4347 skb_dst_set(skb, (struct dst_entry *) md);
4349 info = &md->u.tun_info;
4350 memset(info, 0, sizeof(*info));
4351 info->mode = IP_TUNNEL_INFO_TX;
4353 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4354 if (flags & BPF_F_DONT_FRAGMENT)
4355 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4356 if (flags & BPF_F_ZERO_CSUM_TX)
4357 info->key.tun_flags &= ~TUNNEL_CSUM;
4358 if (flags & BPF_F_SEQ_NUMBER)
4359 info->key.tun_flags |= TUNNEL_SEQ;
4361 info->key.tun_id = cpu_to_be64(from->tunnel_id);
4362 info->key.tos = from->tunnel_tos;
4363 info->key.ttl = from->tunnel_ttl;
4365 if (flags & BPF_F_TUNINFO_IPV6) {
4366 info->mode |= IP_TUNNEL_INFO_IPV6;
4367 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4368 sizeof(from->remote_ipv6));
4369 info->key.label = cpu_to_be32(from->tunnel_label) &
4370 IPV6_FLOWLABEL_MASK;
4372 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4378 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4379 .func = bpf_skb_set_tunnel_key,
4381 .ret_type = RET_INTEGER,
4382 .arg1_type = ARG_PTR_TO_CTX,
4383 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4384 .arg3_type = ARG_CONST_SIZE,
4385 .arg4_type = ARG_ANYTHING,
4388 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4389 const u8 *, from, u32, size)
4391 struct ip_tunnel_info *info = skb_tunnel_info(skb);
4392 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4394 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4396 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4399 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4404 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4405 .func = bpf_skb_set_tunnel_opt,
4407 .ret_type = RET_INTEGER,
4408 .arg1_type = ARG_PTR_TO_CTX,
4409 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4410 .arg3_type = ARG_CONST_SIZE,
4413 static const struct bpf_func_proto *
4414 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4417 struct metadata_dst __percpu *tmp;
4419 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4424 if (cmpxchg(&md_dst, NULL, tmp))
4425 metadata_dst_free_percpu(tmp);
4429 case BPF_FUNC_skb_set_tunnel_key:
4430 return &bpf_skb_set_tunnel_key_proto;
4431 case BPF_FUNC_skb_set_tunnel_opt:
4432 return &bpf_skb_set_tunnel_opt_proto;
4438 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4441 struct bpf_array *array = container_of(map, struct bpf_array, map);
4442 struct cgroup *cgrp;
4445 sk = skb_to_full_sk(skb);
4446 if (!sk || !sk_fullsock(sk))
4448 if (unlikely(idx >= array->map.max_entries))
4451 cgrp = READ_ONCE(array->ptrs[idx]);
4452 if (unlikely(!cgrp))
4455 return sk_under_cgroup_hierarchy(sk, cgrp);
4458 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4459 .func = bpf_skb_under_cgroup,
4461 .ret_type = RET_INTEGER,
4462 .arg1_type = ARG_PTR_TO_CTX,
4463 .arg2_type = ARG_CONST_MAP_PTR,
4464 .arg3_type = ARG_ANYTHING,
4467 #ifdef CONFIG_SOCK_CGROUP_DATA
4468 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4470 struct cgroup *cgrp;
4472 sk = sk_to_full_sk(sk);
4473 if (!sk || !sk_fullsock(sk))
4476 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4477 return cgroup_id(cgrp);
4480 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4482 return __bpf_sk_cgroup_id(skb->sk);
4485 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4486 .func = bpf_skb_cgroup_id,
4488 .ret_type = RET_INTEGER,
4489 .arg1_type = ARG_PTR_TO_CTX,
4492 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4495 struct cgroup *ancestor;
4496 struct cgroup *cgrp;
4498 sk = sk_to_full_sk(sk);
4499 if (!sk || !sk_fullsock(sk))
4502 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4503 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4507 return cgroup_id(ancestor);
4510 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4513 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4516 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4517 .func = bpf_skb_ancestor_cgroup_id,
4519 .ret_type = RET_INTEGER,
4520 .arg1_type = ARG_PTR_TO_CTX,
4521 .arg2_type = ARG_ANYTHING,
4524 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4526 return __bpf_sk_cgroup_id(sk);
4529 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4530 .func = bpf_sk_cgroup_id,
4532 .ret_type = RET_INTEGER,
4533 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4536 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4538 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4541 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4542 .func = bpf_sk_ancestor_cgroup_id,
4544 .ret_type = RET_INTEGER,
4545 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4546 .arg2_type = ARG_ANYTHING,
4550 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4551 unsigned long off, unsigned long len)
4553 memcpy(dst_buff, src_buff + off, len);
4557 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4558 u64, flags, void *, meta, u64, meta_size)
4560 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4562 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4564 if (unlikely(!xdp ||
4565 xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4568 return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4569 xdp_size, bpf_xdp_copy);
4572 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4573 .func = bpf_xdp_event_output,
4575 .ret_type = RET_INTEGER,
4576 .arg1_type = ARG_PTR_TO_CTX,
4577 .arg2_type = ARG_CONST_MAP_PTR,
4578 .arg3_type = ARG_ANYTHING,
4579 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4580 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4583 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4585 const struct bpf_func_proto bpf_xdp_output_proto = {
4586 .func = bpf_xdp_event_output,
4588 .ret_type = RET_INTEGER,
4589 .arg1_type = ARG_PTR_TO_BTF_ID,
4590 .arg1_btf_id = &bpf_xdp_output_btf_ids[0],
4591 .arg2_type = ARG_CONST_MAP_PTR,
4592 .arg3_type = ARG_ANYTHING,
4593 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4594 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4597 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4599 return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4602 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4603 .func = bpf_get_socket_cookie,
4605 .ret_type = RET_INTEGER,
4606 .arg1_type = ARG_PTR_TO_CTX,
4609 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4611 return __sock_gen_cookie(ctx->sk);
4614 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4615 .func = bpf_get_socket_cookie_sock_addr,
4617 .ret_type = RET_INTEGER,
4618 .arg1_type = ARG_PTR_TO_CTX,
4621 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4623 return __sock_gen_cookie(ctx);
4626 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4627 .func = bpf_get_socket_cookie_sock,
4629 .ret_type = RET_INTEGER,
4630 .arg1_type = ARG_PTR_TO_CTX,
4633 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
4635 return sk ? sock_gen_cookie(sk) : 0;
4638 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
4639 .func = bpf_get_socket_ptr_cookie,
4641 .ret_type = RET_INTEGER,
4642 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4645 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4647 return __sock_gen_cookie(ctx->sk);
4650 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4651 .func = bpf_get_socket_cookie_sock_ops,
4653 .ret_type = RET_INTEGER,
4654 .arg1_type = ARG_PTR_TO_CTX,
4657 static u64 __bpf_get_netns_cookie(struct sock *sk)
4659 const struct net *net = sk ? sock_net(sk) : &init_net;
4661 return net->net_cookie;
4664 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
4666 return __bpf_get_netns_cookie(ctx);
4669 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
4670 .func = bpf_get_netns_cookie_sock,
4672 .ret_type = RET_INTEGER,
4673 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4676 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4678 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4681 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
4682 .func = bpf_get_netns_cookie_sock_addr,
4684 .ret_type = RET_INTEGER,
4685 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4688 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4690 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4693 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = {
4694 .func = bpf_get_netns_cookie_sock_ops,
4696 .ret_type = RET_INTEGER,
4697 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4700 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx)
4702 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4705 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = {
4706 .func = bpf_get_netns_cookie_sk_msg,
4708 .ret_type = RET_INTEGER,
4709 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4712 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4714 struct sock *sk = sk_to_full_sk(skb->sk);
4717 if (!sk || !sk_fullsock(sk))
4719 kuid = sock_net_uid(sock_net(sk), sk);
4720 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4723 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4724 .func = bpf_get_socket_uid,
4726 .ret_type = RET_INTEGER,
4727 .arg1_type = ARG_PTR_TO_CTX,
4730 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
4731 char *optval, int optlen)
4733 char devname[IFNAMSIZ];
4739 if (!sk_fullsock(sk))
4742 sock_owned_by_me(sk);
4744 if (level == SOL_SOCKET) {
4745 if (optlen != sizeof(int) && optname != SO_BINDTODEVICE)
4747 val = *((int *)optval);
4748 valbool = val ? 1 : 0;
4750 /* Only some socketops are supported */
4753 val = min_t(u32, val, READ_ONCE(sysctl_rmem_max));
4754 val = min_t(int, val, INT_MAX / 2);
4755 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4756 WRITE_ONCE(sk->sk_rcvbuf,
4757 max_t(int, val * 2, SOCK_MIN_RCVBUF));
4760 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
4761 val = min_t(int, val, INT_MAX / 2);
4762 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4763 WRITE_ONCE(sk->sk_sndbuf,
4764 max_t(int, val * 2, SOCK_MIN_SNDBUF));
4766 case SO_MAX_PACING_RATE: /* 32bit version */
4768 cmpxchg(&sk->sk_pacing_status,
4771 sk->sk_max_pacing_rate = (val == ~0U) ?
4772 ~0UL : (unsigned int)val;
4773 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4774 sk->sk_max_pacing_rate);
4777 sk->sk_priority = val;
4782 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
4785 if (sk->sk_mark != val) {
4790 case SO_BINDTODEVICE:
4791 optlen = min_t(long, optlen, IFNAMSIZ - 1);
4792 strncpy(devname, optval, optlen);
4793 devname[optlen] = 0;
4796 if (devname[0] != '\0') {
4797 struct net_device *dev;
4802 dev = dev_get_by_name(net, devname);
4805 ifindex = dev->ifindex;
4809 case SO_BINDTOIFINDEX:
4810 if (optname == SO_BINDTOIFINDEX)
4812 ret = sock_bindtoindex(sk, ifindex, false);
4815 if (sk->sk_prot->keepalive)
4816 sk->sk_prot->keepalive(sk, valbool);
4817 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
4820 sk->sk_reuseport = valbool;
4826 } else if (level == SOL_IP) {
4827 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4830 val = *((int *)optval);
4831 /* Only some options are supported */
4834 if (val < -1 || val > 0xff) {
4837 struct inet_sock *inet = inet_sk(sk);
4847 #if IS_ENABLED(CONFIG_IPV6)
4848 } else if (level == SOL_IPV6) {
4849 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4852 val = *((int *)optval);
4853 /* Only some options are supported */
4856 if (val < -1 || val > 0xff) {
4859 struct ipv6_pinfo *np = inet6_sk(sk);
4870 } else if (level == SOL_TCP &&
4871 sk->sk_prot->setsockopt == tcp_setsockopt) {
4872 if (optname == TCP_CONGESTION) {
4873 char name[TCP_CA_NAME_MAX];
4875 strncpy(name, optval, min_t(long, optlen,
4876 TCP_CA_NAME_MAX-1));
4877 name[TCP_CA_NAME_MAX-1] = 0;
4878 ret = tcp_set_congestion_control(sk, name, false, true);
4880 struct inet_connection_sock *icsk = inet_csk(sk);
4881 struct tcp_sock *tp = tcp_sk(sk);
4882 unsigned long timeout;
4884 if (optlen != sizeof(int))
4887 val = *((int *)optval);
4888 /* Only some options are supported */
4891 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4894 tcp_snd_cwnd_set(tp, val);
4896 case TCP_BPF_SNDCWND_CLAMP:
4900 tp->snd_cwnd_clamp = val;
4901 tp->snd_ssthresh = val;
4904 case TCP_BPF_DELACK_MAX:
4905 timeout = usecs_to_jiffies(val);
4906 if (timeout > TCP_DELACK_MAX ||
4907 timeout < TCP_TIMEOUT_MIN)
4909 inet_csk(sk)->icsk_delack_max = timeout;
4911 case TCP_BPF_RTO_MIN:
4912 timeout = usecs_to_jiffies(val);
4913 if (timeout > TCP_RTO_MIN ||
4914 timeout < TCP_TIMEOUT_MIN)
4916 inet_csk(sk)->icsk_rto_min = timeout;
4919 if (val < 0 || val > 1)
4925 ret = tcp_sock_set_keepidle_locked(sk, val);
4928 if (val < 1 || val > MAX_TCP_KEEPINTVL)
4931 tp->keepalive_intvl = val * HZ;
4934 if (val < 1 || val > MAX_TCP_KEEPCNT)
4937 tp->keepalive_probes = val;
4940 if (val < 1 || val > MAX_TCP_SYNCNT)
4943 icsk->icsk_syn_retries = val;
4945 case TCP_USER_TIMEOUT:
4949 icsk->icsk_user_timeout = val;
4951 case TCP_NOTSENT_LOWAT:
4952 tp->notsent_lowat = val;
4953 sk->sk_write_space(sk);
4955 case TCP_WINDOW_CLAMP:
4956 ret = tcp_set_window_clamp(sk, val);
4969 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
4970 char *optval, int optlen)
4972 if (!sk_fullsock(sk))
4975 sock_owned_by_me(sk);
4977 if (level == SOL_SOCKET) {
4978 if (optlen != sizeof(int))
4983 *((int *)optval) = sk->sk_mark;
4986 *((int *)optval) = sk->sk_priority;
4988 case SO_BINDTOIFINDEX:
4989 *((int *)optval) = sk->sk_bound_dev_if;
4992 *((int *)optval) = sk->sk_reuseport;
4998 } else if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4999 struct inet_connection_sock *icsk;
5000 struct tcp_sock *tp;
5003 case TCP_CONGESTION:
5004 icsk = inet_csk(sk);
5006 if (!icsk->icsk_ca_ops || optlen <= 1)
5008 strncpy(optval, icsk->icsk_ca_ops->name, optlen);
5009 optval[optlen - 1] = 0;
5014 if (optlen <= 0 || !tp->saved_syn ||
5015 optlen > tcp_saved_syn_len(tp->saved_syn))
5017 memcpy(optval, tp->saved_syn->data, optlen);
5022 } else if (level == SOL_IP) {
5023 struct inet_sock *inet = inet_sk(sk);
5025 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
5028 /* Only some options are supported */
5031 *((int *)optval) = (int)inet->tos;
5036 #if IS_ENABLED(CONFIG_IPV6)
5037 } else if (level == SOL_IPV6) {
5038 struct ipv6_pinfo *np = inet6_sk(sk);
5040 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
5043 /* Only some options are supported */
5046 *((int *)optval) = (int)np->tclass;
5058 memset(optval, 0, optlen);
5062 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level,
5063 int, optname, char *, optval, int, optlen)
5065 if (level == SOL_TCP && optname == TCP_CONGESTION) {
5066 if (optlen >= sizeof("cdg") - 1 &&
5067 !strncmp("cdg", optval, optlen))
5071 return _bpf_setsockopt(sk, level, optname, optval, optlen);
5074 const struct bpf_func_proto bpf_sk_setsockopt_proto = {
5075 .func = bpf_sk_setsockopt,
5077 .ret_type = RET_INTEGER,
5078 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5079 .arg2_type = ARG_ANYTHING,
5080 .arg3_type = ARG_ANYTHING,
5081 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5082 .arg5_type = ARG_CONST_SIZE,
5085 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level,
5086 int, optname, char *, optval, int, optlen)
5088 return _bpf_getsockopt(sk, level, optname, optval, optlen);
5091 const struct bpf_func_proto bpf_sk_getsockopt_proto = {
5092 .func = bpf_sk_getsockopt,
5094 .ret_type = RET_INTEGER,
5095 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5096 .arg2_type = ARG_ANYTHING,
5097 .arg3_type = ARG_ANYTHING,
5098 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5099 .arg5_type = ARG_CONST_SIZE,
5102 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5103 int, level, int, optname, char *, optval, int, optlen)
5105 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5108 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5109 .func = bpf_sock_addr_setsockopt,
5111 .ret_type = RET_INTEGER,
5112 .arg1_type = ARG_PTR_TO_CTX,
5113 .arg2_type = ARG_ANYTHING,
5114 .arg3_type = ARG_ANYTHING,
5115 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5116 .arg5_type = ARG_CONST_SIZE,
5119 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5120 int, level, int, optname, char *, optval, int, optlen)
5122 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5125 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5126 .func = bpf_sock_addr_getsockopt,
5128 .ret_type = RET_INTEGER,
5129 .arg1_type = ARG_PTR_TO_CTX,
5130 .arg2_type = ARG_ANYTHING,
5131 .arg3_type = ARG_ANYTHING,
5132 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5133 .arg5_type = ARG_CONST_SIZE,
5136 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5137 int, level, int, optname, char *, optval, int, optlen)
5139 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5142 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5143 .func = bpf_sock_ops_setsockopt,
5145 .ret_type = RET_INTEGER,
5146 .arg1_type = ARG_PTR_TO_CTX,
5147 .arg2_type = ARG_ANYTHING,
5148 .arg3_type = ARG_ANYTHING,
5149 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5150 .arg5_type = ARG_CONST_SIZE,
5153 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5154 int optname, const u8 **start)
5156 struct sk_buff *syn_skb = bpf_sock->syn_skb;
5157 const u8 *hdr_start;
5161 /* sk is a request_sock here */
5163 if (optname == TCP_BPF_SYN) {
5164 hdr_start = syn_skb->data;
5165 ret = tcp_hdrlen(syn_skb);
5166 } else if (optname == TCP_BPF_SYN_IP) {
5167 hdr_start = skb_network_header(syn_skb);
5168 ret = skb_network_header_len(syn_skb) +
5169 tcp_hdrlen(syn_skb);
5171 /* optname == TCP_BPF_SYN_MAC */
5172 hdr_start = skb_mac_header(syn_skb);
5173 ret = skb_mac_header_len(syn_skb) +
5174 skb_network_header_len(syn_skb) +
5175 tcp_hdrlen(syn_skb);
5178 struct sock *sk = bpf_sock->sk;
5179 struct saved_syn *saved_syn;
5181 if (sk->sk_state == TCP_NEW_SYN_RECV)
5182 /* synack retransmit. bpf_sock->syn_skb will
5183 * not be available. It has to resort to
5184 * saved_syn (if it is saved).
5186 saved_syn = inet_reqsk(sk)->saved_syn;
5188 saved_syn = tcp_sk(sk)->saved_syn;
5193 if (optname == TCP_BPF_SYN) {
5194 hdr_start = saved_syn->data +
5195 saved_syn->mac_hdrlen +
5196 saved_syn->network_hdrlen;
5197 ret = saved_syn->tcp_hdrlen;
5198 } else if (optname == TCP_BPF_SYN_IP) {
5199 hdr_start = saved_syn->data +
5200 saved_syn->mac_hdrlen;
5201 ret = saved_syn->network_hdrlen +
5202 saved_syn->tcp_hdrlen;
5204 /* optname == TCP_BPF_SYN_MAC */
5206 /* TCP_SAVE_SYN may not have saved the mac hdr */
5207 if (!saved_syn->mac_hdrlen)
5210 hdr_start = saved_syn->data;
5211 ret = saved_syn->mac_hdrlen +
5212 saved_syn->network_hdrlen +
5213 saved_syn->tcp_hdrlen;
5221 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5222 int, level, int, optname, char *, optval, int, optlen)
5224 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5225 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5226 int ret, copy_len = 0;
5229 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5232 if (optlen < copy_len) {
5237 memcpy(optval, start, copy_len);
5240 /* Zero out unused buffer at the end */
5241 memset(optval + copy_len, 0, optlen - copy_len);
5246 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5249 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5250 .func = bpf_sock_ops_getsockopt,
5252 .ret_type = RET_INTEGER,
5253 .arg1_type = ARG_PTR_TO_CTX,
5254 .arg2_type = ARG_ANYTHING,
5255 .arg3_type = ARG_ANYTHING,
5256 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5257 .arg5_type = ARG_CONST_SIZE,
5260 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5263 struct sock *sk = bpf_sock->sk;
5264 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5266 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5269 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5271 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5274 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5275 .func = bpf_sock_ops_cb_flags_set,
5277 .ret_type = RET_INTEGER,
5278 .arg1_type = ARG_PTR_TO_CTX,
5279 .arg2_type = ARG_ANYTHING,
5282 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5283 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5285 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5289 struct sock *sk = ctx->sk;
5290 u32 flags = BIND_FROM_BPF;
5294 if (addr_len < offsetofend(struct sockaddr, sa_family))
5296 if (addr->sa_family == AF_INET) {
5297 if (addr_len < sizeof(struct sockaddr_in))
5299 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5300 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5301 return __inet_bind(sk, addr, addr_len, flags);
5302 #if IS_ENABLED(CONFIG_IPV6)
5303 } else if (addr->sa_family == AF_INET6) {
5304 if (addr_len < SIN6_LEN_RFC2133)
5306 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5307 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5308 /* ipv6_bpf_stub cannot be NULL, since it's called from
5309 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5311 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5312 #endif /* CONFIG_IPV6 */
5314 #endif /* CONFIG_INET */
5316 return -EAFNOSUPPORT;
5319 static const struct bpf_func_proto bpf_bind_proto = {
5322 .ret_type = RET_INTEGER,
5323 .arg1_type = ARG_PTR_TO_CTX,
5324 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5325 .arg3_type = ARG_CONST_SIZE,
5329 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5330 struct bpf_xfrm_state *, to, u32, size, u64, flags)
5332 const struct sec_path *sp = skb_sec_path(skb);
5333 const struct xfrm_state *x;
5335 if (!sp || unlikely(index >= sp->len || flags))
5338 x = sp->xvec[index];
5340 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5343 to->reqid = x->props.reqid;
5344 to->spi = x->id.spi;
5345 to->family = x->props.family;
5348 if (to->family == AF_INET6) {
5349 memcpy(to->remote_ipv6, x->props.saddr.a6,
5350 sizeof(to->remote_ipv6));
5352 to->remote_ipv4 = x->props.saddr.a4;
5353 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5358 memset(to, 0, size);
5362 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5363 .func = bpf_skb_get_xfrm_state,
5365 .ret_type = RET_INTEGER,
5366 .arg1_type = ARG_PTR_TO_CTX,
5367 .arg2_type = ARG_ANYTHING,
5368 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
5369 .arg4_type = ARG_CONST_SIZE,
5370 .arg5_type = ARG_ANYTHING,
5374 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5375 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
5376 const struct neighbour *neigh,
5377 const struct net_device *dev, u32 mtu)
5379 memcpy(params->dmac, neigh->ha, ETH_ALEN);
5380 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5381 params->h_vlan_TCI = 0;
5382 params->h_vlan_proto = 0;
5384 params->mtu_result = mtu; /* union with tot_len */
5390 #if IS_ENABLED(CONFIG_INET)
5391 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5392 u32 flags, bool check_mtu)
5394 struct fib_nh_common *nhc;
5395 struct in_device *in_dev;
5396 struct neighbour *neigh;
5397 struct net_device *dev;
5398 struct fib_result res;
5403 dev = dev_get_by_index_rcu(net, params->ifindex);
5407 /* verify forwarding is enabled on this interface */
5408 in_dev = __in_dev_get_rcu(dev);
5409 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5410 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5412 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5414 fl4.flowi4_oif = params->ifindex;
5416 fl4.flowi4_iif = params->ifindex;
5419 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5420 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5421 fl4.flowi4_flags = 0;
5423 fl4.flowi4_proto = params->l4_protocol;
5424 fl4.daddr = params->ipv4_dst;
5425 fl4.saddr = params->ipv4_src;
5426 fl4.fl4_sport = params->sport;
5427 fl4.fl4_dport = params->dport;
5428 fl4.flowi4_multipath_hash = 0;
5430 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5431 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5432 struct fib_table *tb;
5434 tb = fib_get_table(net, tbid);
5436 return BPF_FIB_LKUP_RET_NOT_FWDED;
5438 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5440 fl4.flowi4_mark = 0;
5441 fl4.flowi4_secid = 0;
5442 fl4.flowi4_tun_key.tun_id = 0;
5443 fl4.flowi4_uid = sock_net_uid(net, NULL);
5445 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5449 /* map fib lookup errors to RTN_ type */
5451 return BPF_FIB_LKUP_RET_BLACKHOLE;
5452 if (err == -EHOSTUNREACH)
5453 return BPF_FIB_LKUP_RET_UNREACHABLE;
5455 return BPF_FIB_LKUP_RET_PROHIBIT;
5457 return BPF_FIB_LKUP_RET_NOT_FWDED;
5460 if (res.type != RTN_UNICAST)
5461 return BPF_FIB_LKUP_RET_NOT_FWDED;
5463 if (fib_info_num_path(res.fi) > 1)
5464 fib_select_path(net, &res, &fl4, NULL);
5467 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5468 if (params->tot_len > mtu) {
5469 params->mtu_result = mtu; /* union with tot_len */
5470 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5476 /* do not handle lwt encaps right now */
5477 if (nhc->nhc_lwtstate)
5478 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5482 params->rt_metric = res.fi->fib_priority;
5483 params->ifindex = dev->ifindex;
5485 /* xdp and cls_bpf programs are run in RCU-bh so
5486 * rcu_read_lock_bh is not needed here
5488 if (likely(nhc->nhc_gw_family != AF_INET6)) {
5489 if (nhc->nhc_gw_family)
5490 params->ipv4_dst = nhc->nhc_gw.ipv4;
5492 neigh = __ipv4_neigh_lookup_noref(dev,
5493 (__force u32)params->ipv4_dst);
5495 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5497 params->family = AF_INET6;
5498 *dst = nhc->nhc_gw.ipv6;
5499 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5503 return BPF_FIB_LKUP_RET_NO_NEIGH;
5505 return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5509 #if IS_ENABLED(CONFIG_IPV6)
5510 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5511 u32 flags, bool check_mtu)
5513 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5514 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5515 struct fib6_result res = {};
5516 struct neighbour *neigh;
5517 struct net_device *dev;
5518 struct inet6_dev *idev;
5524 /* link local addresses are never forwarded */
5525 if (rt6_need_strict(dst) || rt6_need_strict(src))
5526 return BPF_FIB_LKUP_RET_NOT_FWDED;
5528 dev = dev_get_by_index_rcu(net, params->ifindex);
5532 idev = __in6_dev_get_safely(dev);
5533 if (unlikely(!idev || !idev->cnf.forwarding))
5534 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5536 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5538 oif = fl6.flowi6_oif = params->ifindex;
5540 oif = fl6.flowi6_iif = params->ifindex;
5542 strict = RT6_LOOKUP_F_HAS_SADDR;
5544 fl6.flowlabel = params->flowinfo;
5545 fl6.flowi6_scope = 0;
5546 fl6.flowi6_flags = 0;
5549 fl6.flowi6_proto = params->l4_protocol;
5552 fl6.fl6_sport = params->sport;
5553 fl6.fl6_dport = params->dport;
5555 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5556 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5557 struct fib6_table *tb;
5559 tb = ipv6_stub->fib6_get_table(net, tbid);
5561 return BPF_FIB_LKUP_RET_NOT_FWDED;
5563 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5566 fl6.flowi6_mark = 0;
5567 fl6.flowi6_secid = 0;
5568 fl6.flowi6_tun_key.tun_id = 0;
5569 fl6.flowi6_uid = sock_net_uid(net, NULL);
5571 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5574 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5575 res.f6i == net->ipv6.fib6_null_entry))
5576 return BPF_FIB_LKUP_RET_NOT_FWDED;
5578 switch (res.fib6_type) {
5579 /* only unicast is forwarded */
5583 return BPF_FIB_LKUP_RET_BLACKHOLE;
5584 case RTN_UNREACHABLE:
5585 return BPF_FIB_LKUP_RET_UNREACHABLE;
5587 return BPF_FIB_LKUP_RET_PROHIBIT;
5589 return BPF_FIB_LKUP_RET_NOT_FWDED;
5592 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5593 fl6.flowi6_oif != 0, NULL, strict);
5596 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5597 if (params->tot_len > mtu) {
5598 params->mtu_result = mtu; /* union with tot_len */
5599 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5603 if (res.nh->fib_nh_lws)
5604 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5606 if (res.nh->fib_nh_gw_family)
5607 *dst = res.nh->fib_nh_gw6;
5609 dev = res.nh->fib_nh_dev;
5610 params->rt_metric = res.f6i->fib6_metric;
5611 params->ifindex = dev->ifindex;
5613 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
5616 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5618 return BPF_FIB_LKUP_RET_NO_NEIGH;
5620 return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5624 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
5625 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5627 if (plen < sizeof(*params))
5630 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5633 switch (params->family) {
5634 #if IS_ENABLED(CONFIG_INET)
5636 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
5639 #if IS_ENABLED(CONFIG_IPV6)
5641 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
5645 return -EAFNOSUPPORT;
5648 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
5649 .func = bpf_xdp_fib_lookup,
5651 .ret_type = RET_INTEGER,
5652 .arg1_type = ARG_PTR_TO_CTX,
5653 .arg2_type = ARG_PTR_TO_MEM,
5654 .arg3_type = ARG_CONST_SIZE,
5655 .arg4_type = ARG_ANYTHING,
5658 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
5659 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5661 struct net *net = dev_net(skb->dev);
5662 int rc = -EAFNOSUPPORT;
5663 bool check_mtu = false;
5665 if (plen < sizeof(*params))
5668 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5671 if (params->tot_len)
5674 switch (params->family) {
5675 #if IS_ENABLED(CONFIG_INET)
5677 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
5680 #if IS_ENABLED(CONFIG_IPV6)
5682 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
5687 if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
5688 struct net_device *dev;
5690 /* When tot_len isn't provided by user, check skb
5691 * against MTU of FIB lookup resulting net_device
5693 dev = dev_get_by_index_rcu(net, params->ifindex);
5694 if (!is_skb_forwardable(dev, skb))
5695 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
5697 params->mtu_result = dev->mtu; /* union with tot_len */
5703 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
5704 .func = bpf_skb_fib_lookup,
5706 .ret_type = RET_INTEGER,
5707 .arg1_type = ARG_PTR_TO_CTX,
5708 .arg2_type = ARG_PTR_TO_MEM,
5709 .arg3_type = ARG_CONST_SIZE,
5710 .arg4_type = ARG_ANYTHING,
5713 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
5716 struct net *netns = dev_net(dev_curr);
5718 /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
5722 return dev_get_by_index_rcu(netns, ifindex);
5725 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
5726 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5728 int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5729 struct net_device *dev = skb->dev;
5730 int skb_len, dev_len;
5733 if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
5736 if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
5739 dev = __dev_via_ifindex(dev, ifindex);
5743 mtu = READ_ONCE(dev->mtu);
5745 dev_len = mtu + dev->hard_header_len;
5747 /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
5748 skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
5750 skb_len += len_diff; /* minus result pass check */
5751 if (skb_len <= dev_len) {
5752 ret = BPF_MTU_CHK_RET_SUCCESS;
5755 /* At this point, skb->len exceed MTU, but as it include length of all
5756 * segments, it can still be below MTU. The SKB can possibly get
5757 * re-segmented in transmit path (see validate_xmit_skb). Thus, user
5758 * must choose if segs are to be MTU checked.
5760 if (skb_is_gso(skb)) {
5761 ret = BPF_MTU_CHK_RET_SUCCESS;
5763 if (flags & BPF_MTU_CHK_SEGS &&
5764 !skb_gso_validate_network_len(skb, mtu))
5765 ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
5768 /* BPF verifier guarantees valid pointer */
5774 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
5775 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5777 struct net_device *dev = xdp->rxq->dev;
5778 int xdp_len = xdp->data_end - xdp->data;
5779 int ret = BPF_MTU_CHK_RET_SUCCESS;
5782 /* XDP variant doesn't support multi-buffer segment check (yet) */
5783 if (unlikely(flags))
5786 dev = __dev_via_ifindex(dev, ifindex);
5790 mtu = READ_ONCE(dev->mtu);
5792 /* Add L2-header as dev MTU is L3 size */
5793 dev_len = mtu + dev->hard_header_len;
5795 /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
5797 xdp_len = *mtu_len + dev->hard_header_len;
5799 xdp_len += len_diff; /* minus result pass check */
5800 if (xdp_len > dev_len)
5801 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5803 /* BPF verifier guarantees valid pointer */
5809 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
5810 .func = bpf_skb_check_mtu,
5812 .ret_type = RET_INTEGER,
5813 .arg1_type = ARG_PTR_TO_CTX,
5814 .arg2_type = ARG_ANYTHING,
5815 .arg3_type = ARG_PTR_TO_INT,
5816 .arg4_type = ARG_ANYTHING,
5817 .arg5_type = ARG_ANYTHING,
5820 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
5821 .func = bpf_xdp_check_mtu,
5823 .ret_type = RET_INTEGER,
5824 .arg1_type = ARG_PTR_TO_CTX,
5825 .arg2_type = ARG_ANYTHING,
5826 .arg3_type = ARG_PTR_TO_INT,
5827 .arg4_type = ARG_ANYTHING,
5828 .arg5_type = ARG_ANYTHING,
5831 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5832 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
5835 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
5837 if (!seg6_validate_srh(srh, len, false))
5841 case BPF_LWT_ENCAP_SEG6_INLINE:
5842 if (skb->protocol != htons(ETH_P_IPV6))
5845 err = seg6_do_srh_inline(skb, srh);
5847 case BPF_LWT_ENCAP_SEG6:
5848 skb_reset_inner_headers(skb);
5849 skb->encapsulation = 1;
5850 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
5856 bpf_compute_data_pointers(skb);
5860 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
5862 return seg6_lookup_nexthop(skb, NULL, 0);
5864 #endif /* CONFIG_IPV6_SEG6_BPF */
5866 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5867 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
5870 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
5874 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
5878 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5879 case BPF_LWT_ENCAP_SEG6:
5880 case BPF_LWT_ENCAP_SEG6_INLINE:
5881 return bpf_push_seg6_encap(skb, type, hdr, len);
5883 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5884 case BPF_LWT_ENCAP_IP:
5885 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
5892 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
5893 void *, hdr, u32, len)
5896 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5897 case BPF_LWT_ENCAP_IP:
5898 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
5905 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
5906 .func = bpf_lwt_in_push_encap,
5908 .ret_type = RET_INTEGER,
5909 .arg1_type = ARG_PTR_TO_CTX,
5910 .arg2_type = ARG_ANYTHING,
5911 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5912 .arg4_type = ARG_CONST_SIZE
5915 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
5916 .func = bpf_lwt_xmit_push_encap,
5918 .ret_type = RET_INTEGER,
5919 .arg1_type = ARG_PTR_TO_CTX,
5920 .arg2_type = ARG_ANYTHING,
5921 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5922 .arg4_type = ARG_CONST_SIZE
5925 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5926 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
5927 const void *, from, u32, len)
5929 struct seg6_bpf_srh_state *srh_state =
5930 this_cpu_ptr(&seg6_bpf_srh_states);
5931 struct ipv6_sr_hdr *srh = srh_state->srh;
5932 void *srh_tlvs, *srh_end, *ptr;
5938 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
5939 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
5941 ptr = skb->data + offset;
5942 if (ptr >= srh_tlvs && ptr + len <= srh_end)
5943 srh_state->valid = false;
5944 else if (ptr < (void *)&srh->flags ||
5945 ptr + len > (void *)&srh->segments)
5948 if (unlikely(bpf_try_make_writable(skb, offset + len)))
5950 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5952 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5954 memcpy(skb->data + offset, from, len);
5958 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
5959 .func = bpf_lwt_seg6_store_bytes,
5961 .ret_type = RET_INTEGER,
5962 .arg1_type = ARG_PTR_TO_CTX,
5963 .arg2_type = ARG_ANYTHING,
5964 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5965 .arg4_type = ARG_CONST_SIZE
5968 static void bpf_update_srh_state(struct sk_buff *skb)
5970 struct seg6_bpf_srh_state *srh_state =
5971 this_cpu_ptr(&seg6_bpf_srh_states);
5974 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
5975 srh_state->srh = NULL;
5977 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5978 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
5979 srh_state->valid = true;
5983 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
5984 u32, action, void *, param, u32, param_len)
5986 struct seg6_bpf_srh_state *srh_state =
5987 this_cpu_ptr(&seg6_bpf_srh_states);
5992 case SEG6_LOCAL_ACTION_END_X:
5993 if (!seg6_bpf_has_valid_srh(skb))
5995 if (param_len != sizeof(struct in6_addr))
5997 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
5998 case SEG6_LOCAL_ACTION_END_T:
5999 if (!seg6_bpf_has_valid_srh(skb))
6001 if (param_len != sizeof(int))
6003 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6004 case SEG6_LOCAL_ACTION_END_DT6:
6005 if (!seg6_bpf_has_valid_srh(skb))
6007 if (param_len != sizeof(int))
6010 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
6012 if (!pskb_pull(skb, hdroff))
6015 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
6016 skb_reset_network_header(skb);
6017 skb_reset_transport_header(skb);
6018 skb->encapsulation = 0;
6020 bpf_compute_data_pointers(skb);
6021 bpf_update_srh_state(skb);
6022 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6023 case SEG6_LOCAL_ACTION_END_B6:
6024 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6026 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
6029 bpf_update_srh_state(skb);
6032 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
6033 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6035 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
6038 bpf_update_srh_state(skb);
6046 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
6047 .func = bpf_lwt_seg6_action,
6049 .ret_type = RET_INTEGER,
6050 .arg1_type = ARG_PTR_TO_CTX,
6051 .arg2_type = ARG_ANYTHING,
6052 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6053 .arg4_type = ARG_CONST_SIZE
6056 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
6059 struct seg6_bpf_srh_state *srh_state =
6060 this_cpu_ptr(&seg6_bpf_srh_states);
6061 struct ipv6_sr_hdr *srh = srh_state->srh;
6062 void *srh_end, *srh_tlvs, *ptr;
6063 struct ipv6hdr *hdr;
6067 if (unlikely(srh == NULL))
6070 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
6071 ((srh->first_segment + 1) << 4));
6072 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
6074 ptr = skb->data + offset;
6076 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
6078 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
6082 ret = skb_cow_head(skb, len);
6083 if (unlikely(ret < 0))
6086 ret = bpf_skb_net_hdr_push(skb, offset, len);
6088 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6091 bpf_compute_data_pointers(skb);
6092 if (unlikely(ret < 0))
6095 hdr = (struct ipv6hdr *)skb->data;
6096 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6098 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6100 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6101 srh_state->hdrlen += len;
6102 srh_state->valid = false;
6106 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6107 .func = bpf_lwt_seg6_adjust_srh,
6109 .ret_type = RET_INTEGER,
6110 .arg1_type = ARG_PTR_TO_CTX,
6111 .arg2_type = ARG_ANYTHING,
6112 .arg3_type = ARG_ANYTHING,
6114 #endif /* CONFIG_IPV6_SEG6_BPF */
6117 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6118 int dif, int sdif, u8 family, u8 proto)
6120 bool refcounted = false;
6121 struct sock *sk = NULL;
6123 if (family == AF_INET) {
6124 __be32 src4 = tuple->ipv4.saddr;
6125 __be32 dst4 = tuple->ipv4.daddr;
6127 if (proto == IPPROTO_TCP)
6128 sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
6129 src4, tuple->ipv4.sport,
6130 dst4, tuple->ipv4.dport,
6131 dif, sdif, &refcounted);
6133 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6134 dst4, tuple->ipv4.dport,
6135 dif, sdif, &udp_table, NULL);
6136 #if IS_ENABLED(CONFIG_IPV6)
6138 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6139 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6141 if (proto == IPPROTO_TCP)
6142 sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
6143 src6, tuple->ipv6.sport,
6144 dst6, ntohs(tuple->ipv6.dport),
6145 dif, sdif, &refcounted);
6146 else if (likely(ipv6_bpf_stub))
6147 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6148 src6, tuple->ipv6.sport,
6149 dst6, tuple->ipv6.dport,
6155 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6156 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6162 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6163 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6164 * Returns the socket as an 'unsigned long' to simplify the casting in the
6165 * callers to satisfy BPF_CALL declarations.
6167 static struct sock *
6168 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6169 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6172 struct sock *sk = NULL;
6173 u8 family = AF_UNSPEC;
6177 if (len == sizeof(tuple->ipv4))
6179 else if (len == sizeof(tuple->ipv6))
6184 if (unlikely(family == AF_UNSPEC || flags ||
6185 !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6188 if (family == AF_INET)
6189 sdif = inet_sdif(skb);
6191 sdif = inet6_sdif(skb);
6193 if ((s32)netns_id < 0) {
6195 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6197 net = get_net_ns_by_id(caller_net, netns_id);
6200 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6208 static struct sock *
6209 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6210 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6213 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6214 ifindex, proto, netns_id, flags);
6217 struct sock *sk2 = sk_to_full_sk(sk);
6219 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6220 * sock refcnt is decremented to prevent a request_sock leak.
6222 if (!sk_fullsock(sk2))
6226 /* Ensure there is no need to bump sk2 refcnt */
6227 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6228 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6238 static struct sock *
6239 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6240 u8 proto, u64 netns_id, u64 flags)
6242 struct net *caller_net;
6246 caller_net = dev_net(skb->dev);
6247 ifindex = skb->dev->ifindex;
6249 caller_net = sock_net(skb->sk);
6253 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6257 static struct sock *
6258 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6259 u8 proto, u64 netns_id, u64 flags)
6261 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6265 struct sock *sk2 = sk_to_full_sk(sk);
6267 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6268 * sock refcnt is decremented to prevent a request_sock leak.
6270 if (!sk_fullsock(sk2))
6274 /* Ensure there is no need to bump sk2 refcnt */
6275 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6276 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6286 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6287 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6289 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6293 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6294 .func = bpf_skc_lookup_tcp,
6297 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6298 .arg1_type = ARG_PTR_TO_CTX,
6299 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6300 .arg3_type = ARG_CONST_SIZE,
6301 .arg4_type = ARG_ANYTHING,
6302 .arg5_type = ARG_ANYTHING,
6305 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6306 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6308 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6312 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6313 .func = bpf_sk_lookup_tcp,
6316 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6317 .arg1_type = ARG_PTR_TO_CTX,
6318 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6319 .arg3_type = ARG_CONST_SIZE,
6320 .arg4_type = ARG_ANYTHING,
6321 .arg5_type = ARG_ANYTHING,
6324 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6325 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6327 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6331 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6332 .func = bpf_sk_lookup_udp,
6335 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6336 .arg1_type = ARG_PTR_TO_CTX,
6337 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6338 .arg3_type = ARG_CONST_SIZE,
6339 .arg4_type = ARG_ANYTHING,
6340 .arg5_type = ARG_ANYTHING,
6343 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6345 if (sk && sk_is_refcounted(sk))
6350 static const struct bpf_func_proto bpf_sk_release_proto = {
6351 .func = bpf_sk_release,
6353 .ret_type = RET_INTEGER,
6354 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6357 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6358 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6360 struct net *caller_net = dev_net(ctx->rxq->dev);
6361 int ifindex = ctx->rxq->dev->ifindex;
6363 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6364 ifindex, IPPROTO_UDP, netns_id,
6368 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6369 .func = bpf_xdp_sk_lookup_udp,
6372 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6373 .arg1_type = ARG_PTR_TO_CTX,
6374 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6375 .arg3_type = ARG_CONST_SIZE,
6376 .arg4_type = ARG_ANYTHING,
6377 .arg5_type = ARG_ANYTHING,
6380 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6381 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6383 struct net *caller_net = dev_net(ctx->rxq->dev);
6384 int ifindex = ctx->rxq->dev->ifindex;
6386 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6387 ifindex, IPPROTO_TCP, netns_id,
6391 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6392 .func = bpf_xdp_skc_lookup_tcp,
6395 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6396 .arg1_type = ARG_PTR_TO_CTX,
6397 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6398 .arg3_type = ARG_CONST_SIZE,
6399 .arg4_type = ARG_ANYTHING,
6400 .arg5_type = ARG_ANYTHING,
6403 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6404 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6406 struct net *caller_net = dev_net(ctx->rxq->dev);
6407 int ifindex = ctx->rxq->dev->ifindex;
6409 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6410 ifindex, IPPROTO_TCP, netns_id,
6414 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6415 .func = bpf_xdp_sk_lookup_tcp,
6418 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6419 .arg1_type = ARG_PTR_TO_CTX,
6420 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6421 .arg3_type = ARG_CONST_SIZE,
6422 .arg4_type = ARG_ANYTHING,
6423 .arg5_type = ARG_ANYTHING,
6426 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6427 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6429 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6430 sock_net(ctx->sk), 0,
6431 IPPROTO_TCP, netns_id, flags);
6434 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6435 .func = bpf_sock_addr_skc_lookup_tcp,
6437 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6438 .arg1_type = ARG_PTR_TO_CTX,
6439 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6440 .arg3_type = ARG_CONST_SIZE,
6441 .arg4_type = ARG_ANYTHING,
6442 .arg5_type = ARG_ANYTHING,
6445 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6446 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6448 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6449 sock_net(ctx->sk), 0, IPPROTO_TCP,
6453 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6454 .func = bpf_sock_addr_sk_lookup_tcp,
6456 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6457 .arg1_type = ARG_PTR_TO_CTX,
6458 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6459 .arg3_type = ARG_CONST_SIZE,
6460 .arg4_type = ARG_ANYTHING,
6461 .arg5_type = ARG_ANYTHING,
6464 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6465 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6467 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6468 sock_net(ctx->sk), 0, IPPROTO_UDP,
6472 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6473 .func = bpf_sock_addr_sk_lookup_udp,
6475 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6476 .arg1_type = ARG_PTR_TO_CTX,
6477 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6478 .arg3_type = ARG_CONST_SIZE,
6479 .arg4_type = ARG_ANYTHING,
6480 .arg5_type = ARG_ANYTHING,
6483 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6484 struct bpf_insn_access_aux *info)
6486 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6490 if (off % size != 0)
6494 case offsetof(struct bpf_tcp_sock, bytes_received):
6495 case offsetof(struct bpf_tcp_sock, bytes_acked):
6496 return size == sizeof(__u64);
6498 return size == sizeof(__u32);
6502 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6503 const struct bpf_insn *si,
6504 struct bpf_insn *insn_buf,
6505 struct bpf_prog *prog, u32 *target_size)
6507 struct bpf_insn *insn = insn_buf;
6509 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \
6511 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \
6512 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6513 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6514 si->dst_reg, si->src_reg, \
6515 offsetof(struct tcp_sock, FIELD)); \
6518 #define BPF_INET_SOCK_GET_COMMON(FIELD) \
6520 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \
6522 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6523 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6524 struct inet_connection_sock, \
6526 si->dst_reg, si->src_reg, \
6528 struct inet_connection_sock, \
6532 if (insn > insn_buf)
6533 return insn - insn_buf;
6536 case offsetof(struct bpf_tcp_sock, rtt_min):
6537 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
6538 sizeof(struct minmax));
6539 BUILD_BUG_ON(sizeof(struct minmax) <
6540 sizeof(struct minmax_sample));
6542 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6543 offsetof(struct tcp_sock, rtt_min) +
6544 offsetof(struct minmax_sample, v));
6546 case offsetof(struct bpf_tcp_sock, snd_cwnd):
6547 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
6549 case offsetof(struct bpf_tcp_sock, srtt_us):
6550 BPF_TCP_SOCK_GET_COMMON(srtt_us);
6552 case offsetof(struct bpf_tcp_sock, snd_ssthresh):
6553 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
6555 case offsetof(struct bpf_tcp_sock, rcv_nxt):
6556 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
6558 case offsetof(struct bpf_tcp_sock, snd_nxt):
6559 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
6561 case offsetof(struct bpf_tcp_sock, snd_una):
6562 BPF_TCP_SOCK_GET_COMMON(snd_una);
6564 case offsetof(struct bpf_tcp_sock, mss_cache):
6565 BPF_TCP_SOCK_GET_COMMON(mss_cache);
6567 case offsetof(struct bpf_tcp_sock, ecn_flags):
6568 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
6570 case offsetof(struct bpf_tcp_sock, rate_delivered):
6571 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
6573 case offsetof(struct bpf_tcp_sock, rate_interval_us):
6574 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
6576 case offsetof(struct bpf_tcp_sock, packets_out):
6577 BPF_TCP_SOCK_GET_COMMON(packets_out);
6579 case offsetof(struct bpf_tcp_sock, retrans_out):
6580 BPF_TCP_SOCK_GET_COMMON(retrans_out);
6582 case offsetof(struct bpf_tcp_sock, total_retrans):
6583 BPF_TCP_SOCK_GET_COMMON(total_retrans);
6585 case offsetof(struct bpf_tcp_sock, segs_in):
6586 BPF_TCP_SOCK_GET_COMMON(segs_in);
6588 case offsetof(struct bpf_tcp_sock, data_segs_in):
6589 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
6591 case offsetof(struct bpf_tcp_sock, segs_out):
6592 BPF_TCP_SOCK_GET_COMMON(segs_out);
6594 case offsetof(struct bpf_tcp_sock, data_segs_out):
6595 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
6597 case offsetof(struct bpf_tcp_sock, lost_out):
6598 BPF_TCP_SOCK_GET_COMMON(lost_out);
6600 case offsetof(struct bpf_tcp_sock, sacked_out):
6601 BPF_TCP_SOCK_GET_COMMON(sacked_out);
6603 case offsetof(struct bpf_tcp_sock, bytes_received):
6604 BPF_TCP_SOCK_GET_COMMON(bytes_received);
6606 case offsetof(struct bpf_tcp_sock, bytes_acked):
6607 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
6609 case offsetof(struct bpf_tcp_sock, dsack_dups):
6610 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
6612 case offsetof(struct bpf_tcp_sock, delivered):
6613 BPF_TCP_SOCK_GET_COMMON(delivered);
6615 case offsetof(struct bpf_tcp_sock, delivered_ce):
6616 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
6618 case offsetof(struct bpf_tcp_sock, icsk_retransmits):
6619 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
6623 return insn - insn_buf;
6626 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
6628 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
6629 return (unsigned long)sk;
6631 return (unsigned long)NULL;
6634 const struct bpf_func_proto bpf_tcp_sock_proto = {
6635 .func = bpf_tcp_sock,
6637 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
6638 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6641 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
6643 sk = sk_to_full_sk(sk);
6645 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
6646 return (unsigned long)sk;
6648 return (unsigned long)NULL;
6651 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
6652 .func = bpf_get_listener_sock,
6654 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6655 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6658 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
6660 unsigned int iphdr_len;
6662 switch (skb_protocol(skb, true)) {
6663 case cpu_to_be16(ETH_P_IP):
6664 iphdr_len = sizeof(struct iphdr);
6666 case cpu_to_be16(ETH_P_IPV6):
6667 iphdr_len = sizeof(struct ipv6hdr);
6673 if (skb_headlen(skb) < iphdr_len)
6676 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
6679 return INET_ECN_set_ce(skb);
6682 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6683 struct bpf_insn_access_aux *info)
6685 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
6688 if (off % size != 0)
6693 return size == sizeof(__u32);
6697 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
6698 const struct bpf_insn *si,
6699 struct bpf_insn *insn_buf,
6700 struct bpf_prog *prog, u32 *target_size)
6702 struct bpf_insn *insn = insn_buf;
6704 #define BPF_XDP_SOCK_GET(FIELD) \
6706 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \
6707 sizeof_field(struct bpf_xdp_sock, FIELD)); \
6708 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
6709 si->dst_reg, si->src_reg, \
6710 offsetof(struct xdp_sock, FIELD)); \
6714 case offsetof(struct bpf_xdp_sock, queue_id):
6715 BPF_XDP_SOCK_GET(queue_id);
6719 return insn - insn_buf;
6722 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
6723 .func = bpf_skb_ecn_set_ce,
6725 .ret_type = RET_INTEGER,
6726 .arg1_type = ARG_PTR_TO_CTX,
6729 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6730 struct tcphdr *, th, u32, th_len)
6732 #ifdef CONFIG_SYN_COOKIES
6736 if (unlikely(!sk || th_len < sizeof(*th)))
6739 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
6740 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6743 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
6746 if (!th->ack || th->rst || th->syn)
6749 if (unlikely(iph_len < sizeof(struct iphdr)))
6752 if (tcp_synq_no_recent_overflow(sk))
6755 cookie = ntohl(th->ack_seq) - 1;
6757 /* Both struct iphdr and struct ipv6hdr have the version field at the
6758 * same offset so we can cast to the shorter header (struct iphdr).
6760 switch (((struct iphdr *)iph)->version) {
6762 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
6765 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
6768 #if IS_BUILTIN(CONFIG_IPV6)
6770 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6773 if (sk->sk_family != AF_INET6)
6776 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
6778 #endif /* CONFIG_IPV6 */
6781 return -EPROTONOSUPPORT;
6793 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
6794 .func = bpf_tcp_check_syncookie,
6797 .ret_type = RET_INTEGER,
6798 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6799 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6800 .arg3_type = ARG_CONST_SIZE,
6801 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6802 .arg5_type = ARG_CONST_SIZE,
6805 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6806 struct tcphdr *, th, u32, th_len)
6808 #ifdef CONFIG_SYN_COOKIES
6812 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
6815 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6818 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
6821 if (!th->syn || th->ack || th->fin || th->rst)
6824 if (unlikely(iph_len < sizeof(struct iphdr)))
6827 /* Both struct iphdr and struct ipv6hdr have the version field at the
6828 * same offset so we can cast to the shorter header (struct iphdr).
6830 switch (((struct iphdr *)iph)->version) {
6832 if (sk->sk_family == AF_INET6 && sk->sk_ipv6only)
6835 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
6838 #if IS_BUILTIN(CONFIG_IPV6)
6840 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6843 if (sk->sk_family != AF_INET6)
6846 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
6848 #endif /* CONFIG_IPV6 */
6851 return -EPROTONOSUPPORT;
6856 return cookie | ((u64)mss << 32);
6859 #endif /* CONFIG_SYN_COOKIES */
6862 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
6863 .func = bpf_tcp_gen_syncookie,
6864 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */
6866 .ret_type = RET_INTEGER,
6867 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6868 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6869 .arg3_type = ARG_CONST_SIZE,
6870 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6871 .arg5_type = ARG_CONST_SIZE,
6874 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
6876 if (!sk || flags != 0)
6878 if (!skb_at_tc_ingress(skb))
6880 if (unlikely(dev_net(skb->dev) != sock_net(sk)))
6881 return -ENETUNREACH;
6882 if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
6883 return -ESOCKTNOSUPPORT;
6884 if (sk_is_refcounted(sk) &&
6885 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
6890 skb->destructor = sock_pfree;
6895 static const struct bpf_func_proto bpf_sk_assign_proto = {
6896 .func = bpf_sk_assign,
6898 .ret_type = RET_INTEGER,
6899 .arg1_type = ARG_PTR_TO_CTX,
6900 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6901 .arg3_type = ARG_ANYTHING,
6904 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
6905 u8 search_kind, const u8 *magic,
6906 u8 magic_len, bool *eol)
6912 while (op < opend) {
6915 if (kind == TCPOPT_EOL) {
6917 return ERR_PTR(-ENOMSG);
6918 } else if (kind == TCPOPT_NOP) {
6923 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
6924 /* Something is wrong in the received header.
6925 * Follow the TCP stack's tcp_parse_options()
6926 * and just bail here.
6928 return ERR_PTR(-EFAULT);
6931 if (search_kind == kind) {
6935 if (magic_len > kind_len - 2)
6936 return ERR_PTR(-ENOMSG);
6938 if (!memcmp(&op[2], magic, magic_len))
6945 return ERR_PTR(-ENOMSG);
6948 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6949 void *, search_res, u32, len, u64, flags)
6951 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
6952 const u8 *op, *opend, *magic, *search = search_res;
6953 u8 search_kind, search_len, copy_len, magic_len;
6956 /* 2 byte is the minimal option len except TCPOPT_NOP and
6957 * TCPOPT_EOL which are useless for the bpf prog to learn
6958 * and this helper disallow loading them also.
6960 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
6963 search_kind = search[0];
6964 search_len = search[1];
6966 if (search_len > len || search_kind == TCPOPT_NOP ||
6967 search_kind == TCPOPT_EOL)
6970 if (search_kind == TCPOPT_EXP || search_kind == 253) {
6971 /* 16 or 32 bit magic. +2 for kind and kind length */
6972 if (search_len != 4 && search_len != 6)
6975 magic_len = search_len - 2;
6984 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
6989 op += sizeof(struct tcphdr);
6991 if (!bpf_sock->skb ||
6992 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6993 /* This bpf_sock->op cannot call this helper */
6996 opend = bpf_sock->skb_data_end;
6997 op = bpf_sock->skb->data + sizeof(struct tcphdr);
7000 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
7007 if (copy_len > len) {
7012 memcpy(search_res, op, copy_len);
7016 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
7017 .func = bpf_sock_ops_load_hdr_opt,
7019 .ret_type = RET_INTEGER,
7020 .arg1_type = ARG_PTR_TO_CTX,
7021 .arg2_type = ARG_PTR_TO_MEM,
7022 .arg3_type = ARG_CONST_SIZE,
7023 .arg4_type = ARG_ANYTHING,
7026 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7027 const void *, from, u32, len, u64, flags)
7029 u8 new_kind, new_kind_len, magic_len = 0, *opend;
7030 const u8 *op, *new_op, *magic = NULL;
7031 struct sk_buff *skb;
7034 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
7037 if (len < 2 || flags)
7041 new_kind = new_op[0];
7042 new_kind_len = new_op[1];
7044 if (new_kind_len > len || new_kind == TCPOPT_NOP ||
7045 new_kind == TCPOPT_EOL)
7048 if (new_kind_len > bpf_sock->remaining_opt_len)
7051 /* 253 is another experimental kind */
7052 if (new_kind == TCPOPT_EXP || new_kind == 253) {
7053 if (new_kind_len < 4)
7055 /* Match for the 2 byte magic also.
7056 * RFC 6994: the magic could be 2 or 4 bytes.
7057 * Hence, matching by 2 byte only is on the
7058 * conservative side but it is the right
7059 * thing to do for the 'search-for-duplication'
7066 /* Check for duplication */
7067 skb = bpf_sock->skb;
7068 op = skb->data + sizeof(struct tcphdr);
7069 opend = bpf_sock->skb_data_end;
7071 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
7076 if (PTR_ERR(op) != -ENOMSG)
7080 /* The option has been ended. Treat it as no more
7081 * header option can be written.
7085 /* No duplication found. Store the header option. */
7086 memcpy(opend, from, new_kind_len);
7088 bpf_sock->remaining_opt_len -= new_kind_len;
7089 bpf_sock->skb_data_end += new_kind_len;
7094 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
7095 .func = bpf_sock_ops_store_hdr_opt,
7097 .ret_type = RET_INTEGER,
7098 .arg1_type = ARG_PTR_TO_CTX,
7099 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7100 .arg3_type = ARG_CONST_SIZE,
7101 .arg4_type = ARG_ANYTHING,
7104 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7105 u32, len, u64, flags)
7107 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7110 if (flags || len < 2)
7113 if (len > bpf_sock->remaining_opt_len)
7116 bpf_sock->remaining_opt_len -= len;
7121 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7122 .func = bpf_sock_ops_reserve_hdr_opt,
7124 .ret_type = RET_INTEGER,
7125 .arg1_type = ARG_PTR_TO_CTX,
7126 .arg2_type = ARG_ANYTHING,
7127 .arg3_type = ARG_ANYTHING,
7130 #endif /* CONFIG_INET */
7132 bool bpf_helper_changes_pkt_data(void *func)
7134 if (func == bpf_skb_vlan_push ||
7135 func == bpf_skb_vlan_pop ||
7136 func == bpf_skb_store_bytes ||
7137 func == bpf_skb_change_proto ||
7138 func == bpf_skb_change_head ||
7139 func == sk_skb_change_head ||
7140 func == bpf_skb_change_tail ||
7141 func == sk_skb_change_tail ||
7142 func == bpf_skb_adjust_room ||
7143 func == sk_skb_adjust_room ||
7144 func == bpf_skb_pull_data ||
7145 func == sk_skb_pull_data ||
7146 func == bpf_clone_redirect ||
7147 func == bpf_l3_csum_replace ||
7148 func == bpf_l4_csum_replace ||
7149 func == bpf_xdp_adjust_head ||
7150 func == bpf_xdp_adjust_meta ||
7151 func == bpf_msg_pull_data ||
7152 func == bpf_msg_push_data ||
7153 func == bpf_msg_pop_data ||
7154 func == bpf_xdp_adjust_tail ||
7155 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7156 func == bpf_lwt_seg6_store_bytes ||
7157 func == bpf_lwt_seg6_adjust_srh ||
7158 func == bpf_lwt_seg6_action ||
7161 func == bpf_sock_ops_store_hdr_opt ||
7163 func == bpf_lwt_in_push_encap ||
7164 func == bpf_lwt_xmit_push_encap)
7170 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7171 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7173 static const struct bpf_func_proto *
7174 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7177 /* inet and inet6 sockets are created in a process
7178 * context so there is always a valid uid/gid
7180 case BPF_FUNC_get_current_uid_gid:
7181 return &bpf_get_current_uid_gid_proto;
7182 case BPF_FUNC_get_local_storage:
7183 return &bpf_get_local_storage_proto;
7184 case BPF_FUNC_get_socket_cookie:
7185 return &bpf_get_socket_cookie_sock_proto;
7186 case BPF_FUNC_get_netns_cookie:
7187 return &bpf_get_netns_cookie_sock_proto;
7188 case BPF_FUNC_perf_event_output:
7189 return &bpf_event_output_data_proto;
7190 case BPF_FUNC_get_current_pid_tgid:
7191 return &bpf_get_current_pid_tgid_proto;
7192 case BPF_FUNC_get_current_comm:
7193 return &bpf_get_current_comm_proto;
7194 #ifdef CONFIG_CGROUPS
7195 case BPF_FUNC_get_current_cgroup_id:
7196 return &bpf_get_current_cgroup_id_proto;
7197 case BPF_FUNC_get_current_ancestor_cgroup_id:
7198 return &bpf_get_current_ancestor_cgroup_id_proto;
7200 #ifdef CONFIG_CGROUP_NET_CLASSID
7201 case BPF_FUNC_get_cgroup_classid:
7202 return &bpf_get_cgroup_classid_curr_proto;
7204 case BPF_FUNC_sk_storage_get:
7205 return &bpf_sk_storage_get_cg_sock_proto;
7206 case BPF_FUNC_ktime_get_coarse_ns:
7207 return &bpf_ktime_get_coarse_ns_proto;
7209 return bpf_base_func_proto(func_id);
7213 static const struct bpf_func_proto *
7214 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7217 /* inet and inet6 sockets are created in a process
7218 * context so there is always a valid uid/gid
7220 case BPF_FUNC_get_current_uid_gid:
7221 return &bpf_get_current_uid_gid_proto;
7223 switch (prog->expected_attach_type) {
7224 case BPF_CGROUP_INET4_CONNECT:
7225 case BPF_CGROUP_INET6_CONNECT:
7226 return &bpf_bind_proto;
7230 case BPF_FUNC_get_socket_cookie:
7231 return &bpf_get_socket_cookie_sock_addr_proto;
7232 case BPF_FUNC_get_netns_cookie:
7233 return &bpf_get_netns_cookie_sock_addr_proto;
7234 case BPF_FUNC_get_local_storage:
7235 return &bpf_get_local_storage_proto;
7236 case BPF_FUNC_perf_event_output:
7237 return &bpf_event_output_data_proto;
7238 case BPF_FUNC_get_current_pid_tgid:
7239 return &bpf_get_current_pid_tgid_proto;
7240 case BPF_FUNC_get_current_comm:
7241 return &bpf_get_current_comm_proto;
7242 #ifdef CONFIG_CGROUPS
7243 case BPF_FUNC_get_current_cgroup_id:
7244 return &bpf_get_current_cgroup_id_proto;
7245 case BPF_FUNC_get_current_ancestor_cgroup_id:
7246 return &bpf_get_current_ancestor_cgroup_id_proto;
7248 #ifdef CONFIG_CGROUP_NET_CLASSID
7249 case BPF_FUNC_get_cgroup_classid:
7250 return &bpf_get_cgroup_classid_curr_proto;
7253 case BPF_FUNC_sk_lookup_tcp:
7254 return &bpf_sock_addr_sk_lookup_tcp_proto;
7255 case BPF_FUNC_sk_lookup_udp:
7256 return &bpf_sock_addr_sk_lookup_udp_proto;
7257 case BPF_FUNC_sk_release:
7258 return &bpf_sk_release_proto;
7259 case BPF_FUNC_skc_lookup_tcp:
7260 return &bpf_sock_addr_skc_lookup_tcp_proto;
7261 #endif /* CONFIG_INET */
7262 case BPF_FUNC_sk_storage_get:
7263 return &bpf_sk_storage_get_proto;
7264 case BPF_FUNC_sk_storage_delete:
7265 return &bpf_sk_storage_delete_proto;
7266 case BPF_FUNC_setsockopt:
7267 switch (prog->expected_attach_type) {
7268 case BPF_CGROUP_INET4_BIND:
7269 case BPF_CGROUP_INET6_BIND:
7270 case BPF_CGROUP_INET4_CONNECT:
7271 case BPF_CGROUP_INET6_CONNECT:
7272 case BPF_CGROUP_UDP4_RECVMSG:
7273 case BPF_CGROUP_UDP6_RECVMSG:
7274 case BPF_CGROUP_UDP4_SENDMSG:
7275 case BPF_CGROUP_UDP6_SENDMSG:
7276 case BPF_CGROUP_INET4_GETPEERNAME:
7277 case BPF_CGROUP_INET6_GETPEERNAME:
7278 case BPF_CGROUP_INET4_GETSOCKNAME:
7279 case BPF_CGROUP_INET6_GETSOCKNAME:
7280 return &bpf_sock_addr_setsockopt_proto;
7284 case BPF_FUNC_getsockopt:
7285 switch (prog->expected_attach_type) {
7286 case BPF_CGROUP_INET4_BIND:
7287 case BPF_CGROUP_INET6_BIND:
7288 case BPF_CGROUP_INET4_CONNECT:
7289 case BPF_CGROUP_INET6_CONNECT:
7290 case BPF_CGROUP_UDP4_RECVMSG:
7291 case BPF_CGROUP_UDP6_RECVMSG:
7292 case BPF_CGROUP_UDP4_SENDMSG:
7293 case BPF_CGROUP_UDP6_SENDMSG:
7294 case BPF_CGROUP_INET4_GETPEERNAME:
7295 case BPF_CGROUP_INET6_GETPEERNAME:
7296 case BPF_CGROUP_INET4_GETSOCKNAME:
7297 case BPF_CGROUP_INET6_GETSOCKNAME:
7298 return &bpf_sock_addr_getsockopt_proto;
7303 return bpf_sk_base_func_proto(func_id);
7307 static const struct bpf_func_proto *
7308 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7311 case BPF_FUNC_skb_load_bytes:
7312 return &bpf_skb_load_bytes_proto;
7313 case BPF_FUNC_skb_load_bytes_relative:
7314 return &bpf_skb_load_bytes_relative_proto;
7315 case BPF_FUNC_get_socket_cookie:
7316 return &bpf_get_socket_cookie_proto;
7317 case BPF_FUNC_get_socket_uid:
7318 return &bpf_get_socket_uid_proto;
7319 case BPF_FUNC_perf_event_output:
7320 return &bpf_skb_event_output_proto;
7322 return bpf_sk_base_func_proto(func_id);
7326 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7327 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7329 static const struct bpf_func_proto *
7330 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7333 case BPF_FUNC_get_local_storage:
7334 return &bpf_get_local_storage_proto;
7335 case BPF_FUNC_sk_fullsock:
7336 return &bpf_sk_fullsock_proto;
7337 case BPF_FUNC_sk_storage_get:
7338 return &bpf_sk_storage_get_proto;
7339 case BPF_FUNC_sk_storage_delete:
7340 return &bpf_sk_storage_delete_proto;
7341 case BPF_FUNC_perf_event_output:
7342 return &bpf_skb_event_output_proto;
7343 #ifdef CONFIG_SOCK_CGROUP_DATA
7344 case BPF_FUNC_skb_cgroup_id:
7345 return &bpf_skb_cgroup_id_proto;
7346 case BPF_FUNC_skb_ancestor_cgroup_id:
7347 return &bpf_skb_ancestor_cgroup_id_proto;
7348 case BPF_FUNC_sk_cgroup_id:
7349 return &bpf_sk_cgroup_id_proto;
7350 case BPF_FUNC_sk_ancestor_cgroup_id:
7351 return &bpf_sk_ancestor_cgroup_id_proto;
7354 case BPF_FUNC_sk_lookup_tcp:
7355 return &bpf_sk_lookup_tcp_proto;
7356 case BPF_FUNC_sk_lookup_udp:
7357 return &bpf_sk_lookup_udp_proto;
7358 case BPF_FUNC_sk_release:
7359 return &bpf_sk_release_proto;
7360 case BPF_FUNC_skc_lookup_tcp:
7361 return &bpf_skc_lookup_tcp_proto;
7362 case BPF_FUNC_tcp_sock:
7363 return &bpf_tcp_sock_proto;
7364 case BPF_FUNC_get_listener_sock:
7365 return &bpf_get_listener_sock_proto;
7366 case BPF_FUNC_skb_ecn_set_ce:
7367 return &bpf_skb_ecn_set_ce_proto;
7370 return sk_filter_func_proto(func_id, prog);
7374 static const struct bpf_func_proto *
7375 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7378 case BPF_FUNC_skb_store_bytes:
7379 return &bpf_skb_store_bytes_proto;
7380 case BPF_FUNC_skb_load_bytes:
7381 return &bpf_skb_load_bytes_proto;
7382 case BPF_FUNC_skb_load_bytes_relative:
7383 return &bpf_skb_load_bytes_relative_proto;
7384 case BPF_FUNC_skb_pull_data:
7385 return &bpf_skb_pull_data_proto;
7386 case BPF_FUNC_csum_diff:
7387 return &bpf_csum_diff_proto;
7388 case BPF_FUNC_csum_update:
7389 return &bpf_csum_update_proto;
7390 case BPF_FUNC_csum_level:
7391 return &bpf_csum_level_proto;
7392 case BPF_FUNC_l3_csum_replace:
7393 return &bpf_l3_csum_replace_proto;
7394 case BPF_FUNC_l4_csum_replace:
7395 return &bpf_l4_csum_replace_proto;
7396 case BPF_FUNC_clone_redirect:
7397 return &bpf_clone_redirect_proto;
7398 case BPF_FUNC_get_cgroup_classid:
7399 return &bpf_get_cgroup_classid_proto;
7400 case BPF_FUNC_skb_vlan_push:
7401 return &bpf_skb_vlan_push_proto;
7402 case BPF_FUNC_skb_vlan_pop:
7403 return &bpf_skb_vlan_pop_proto;
7404 case BPF_FUNC_skb_change_proto:
7405 return &bpf_skb_change_proto_proto;
7406 case BPF_FUNC_skb_change_type:
7407 return &bpf_skb_change_type_proto;
7408 case BPF_FUNC_skb_adjust_room:
7409 return &bpf_skb_adjust_room_proto;
7410 case BPF_FUNC_skb_change_tail:
7411 return &bpf_skb_change_tail_proto;
7412 case BPF_FUNC_skb_change_head:
7413 return &bpf_skb_change_head_proto;
7414 case BPF_FUNC_skb_get_tunnel_key:
7415 return &bpf_skb_get_tunnel_key_proto;
7416 case BPF_FUNC_skb_set_tunnel_key:
7417 return bpf_get_skb_set_tunnel_proto(func_id);
7418 case BPF_FUNC_skb_get_tunnel_opt:
7419 return &bpf_skb_get_tunnel_opt_proto;
7420 case BPF_FUNC_skb_set_tunnel_opt:
7421 return bpf_get_skb_set_tunnel_proto(func_id);
7422 case BPF_FUNC_redirect:
7423 return &bpf_redirect_proto;
7424 case BPF_FUNC_redirect_neigh:
7425 return &bpf_redirect_neigh_proto;
7426 case BPF_FUNC_redirect_peer:
7427 return &bpf_redirect_peer_proto;
7428 case BPF_FUNC_get_route_realm:
7429 return &bpf_get_route_realm_proto;
7430 case BPF_FUNC_get_hash_recalc:
7431 return &bpf_get_hash_recalc_proto;
7432 case BPF_FUNC_set_hash_invalid:
7433 return &bpf_set_hash_invalid_proto;
7434 case BPF_FUNC_set_hash:
7435 return &bpf_set_hash_proto;
7436 case BPF_FUNC_perf_event_output:
7437 return &bpf_skb_event_output_proto;
7438 case BPF_FUNC_get_smp_processor_id:
7439 return &bpf_get_smp_processor_id_proto;
7440 case BPF_FUNC_skb_under_cgroup:
7441 return &bpf_skb_under_cgroup_proto;
7442 case BPF_FUNC_get_socket_cookie:
7443 return &bpf_get_socket_cookie_proto;
7444 case BPF_FUNC_get_socket_uid:
7445 return &bpf_get_socket_uid_proto;
7446 case BPF_FUNC_fib_lookup:
7447 return &bpf_skb_fib_lookup_proto;
7448 case BPF_FUNC_check_mtu:
7449 return &bpf_skb_check_mtu_proto;
7450 case BPF_FUNC_sk_fullsock:
7451 return &bpf_sk_fullsock_proto;
7452 case BPF_FUNC_sk_storage_get:
7453 return &bpf_sk_storage_get_proto;
7454 case BPF_FUNC_sk_storage_delete:
7455 return &bpf_sk_storage_delete_proto;
7457 case BPF_FUNC_skb_get_xfrm_state:
7458 return &bpf_skb_get_xfrm_state_proto;
7460 #ifdef CONFIG_CGROUP_NET_CLASSID
7461 case BPF_FUNC_skb_cgroup_classid:
7462 return &bpf_skb_cgroup_classid_proto;
7464 #ifdef CONFIG_SOCK_CGROUP_DATA
7465 case BPF_FUNC_skb_cgroup_id:
7466 return &bpf_skb_cgroup_id_proto;
7467 case BPF_FUNC_skb_ancestor_cgroup_id:
7468 return &bpf_skb_ancestor_cgroup_id_proto;
7471 case BPF_FUNC_sk_lookup_tcp:
7472 return &bpf_sk_lookup_tcp_proto;
7473 case BPF_FUNC_sk_lookup_udp:
7474 return &bpf_sk_lookup_udp_proto;
7475 case BPF_FUNC_sk_release:
7476 return &bpf_sk_release_proto;
7477 case BPF_FUNC_tcp_sock:
7478 return &bpf_tcp_sock_proto;
7479 case BPF_FUNC_get_listener_sock:
7480 return &bpf_get_listener_sock_proto;
7481 case BPF_FUNC_skc_lookup_tcp:
7482 return &bpf_skc_lookup_tcp_proto;
7483 case BPF_FUNC_tcp_check_syncookie:
7484 return &bpf_tcp_check_syncookie_proto;
7485 case BPF_FUNC_skb_ecn_set_ce:
7486 return &bpf_skb_ecn_set_ce_proto;
7487 case BPF_FUNC_tcp_gen_syncookie:
7488 return &bpf_tcp_gen_syncookie_proto;
7489 case BPF_FUNC_sk_assign:
7490 return &bpf_sk_assign_proto;
7493 return bpf_sk_base_func_proto(func_id);
7497 static const struct bpf_func_proto *
7498 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7501 case BPF_FUNC_perf_event_output:
7502 return &bpf_xdp_event_output_proto;
7503 case BPF_FUNC_get_smp_processor_id:
7504 return &bpf_get_smp_processor_id_proto;
7505 case BPF_FUNC_csum_diff:
7506 return &bpf_csum_diff_proto;
7507 case BPF_FUNC_xdp_adjust_head:
7508 return &bpf_xdp_adjust_head_proto;
7509 case BPF_FUNC_xdp_adjust_meta:
7510 return &bpf_xdp_adjust_meta_proto;
7511 case BPF_FUNC_redirect:
7512 return &bpf_xdp_redirect_proto;
7513 case BPF_FUNC_redirect_map:
7514 return &bpf_xdp_redirect_map_proto;
7515 case BPF_FUNC_xdp_adjust_tail:
7516 return &bpf_xdp_adjust_tail_proto;
7517 case BPF_FUNC_fib_lookup:
7518 return &bpf_xdp_fib_lookup_proto;
7519 case BPF_FUNC_check_mtu:
7520 return &bpf_xdp_check_mtu_proto;
7522 case BPF_FUNC_sk_lookup_udp:
7523 return &bpf_xdp_sk_lookup_udp_proto;
7524 case BPF_FUNC_sk_lookup_tcp:
7525 return &bpf_xdp_sk_lookup_tcp_proto;
7526 case BPF_FUNC_sk_release:
7527 return &bpf_sk_release_proto;
7528 case BPF_FUNC_skc_lookup_tcp:
7529 return &bpf_xdp_skc_lookup_tcp_proto;
7530 case BPF_FUNC_tcp_check_syncookie:
7531 return &bpf_tcp_check_syncookie_proto;
7532 case BPF_FUNC_tcp_gen_syncookie:
7533 return &bpf_tcp_gen_syncookie_proto;
7536 return bpf_sk_base_func_proto(func_id);
7540 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
7541 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
7543 static const struct bpf_func_proto *
7544 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7547 case BPF_FUNC_setsockopt:
7548 return &bpf_sock_ops_setsockopt_proto;
7549 case BPF_FUNC_getsockopt:
7550 return &bpf_sock_ops_getsockopt_proto;
7551 case BPF_FUNC_sock_ops_cb_flags_set:
7552 return &bpf_sock_ops_cb_flags_set_proto;
7553 case BPF_FUNC_sock_map_update:
7554 return &bpf_sock_map_update_proto;
7555 case BPF_FUNC_sock_hash_update:
7556 return &bpf_sock_hash_update_proto;
7557 case BPF_FUNC_get_socket_cookie:
7558 return &bpf_get_socket_cookie_sock_ops_proto;
7559 case BPF_FUNC_get_local_storage:
7560 return &bpf_get_local_storage_proto;
7561 case BPF_FUNC_perf_event_output:
7562 return &bpf_event_output_data_proto;
7563 case BPF_FUNC_sk_storage_get:
7564 return &bpf_sk_storage_get_proto;
7565 case BPF_FUNC_sk_storage_delete:
7566 return &bpf_sk_storage_delete_proto;
7567 case BPF_FUNC_get_netns_cookie:
7568 return &bpf_get_netns_cookie_sock_ops_proto;
7570 case BPF_FUNC_load_hdr_opt:
7571 return &bpf_sock_ops_load_hdr_opt_proto;
7572 case BPF_FUNC_store_hdr_opt:
7573 return &bpf_sock_ops_store_hdr_opt_proto;
7574 case BPF_FUNC_reserve_hdr_opt:
7575 return &bpf_sock_ops_reserve_hdr_opt_proto;
7576 case BPF_FUNC_tcp_sock:
7577 return &bpf_tcp_sock_proto;
7578 #endif /* CONFIG_INET */
7580 return bpf_sk_base_func_proto(func_id);
7584 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
7585 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
7587 static const struct bpf_func_proto *
7588 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7591 case BPF_FUNC_msg_redirect_map:
7592 return &bpf_msg_redirect_map_proto;
7593 case BPF_FUNC_msg_redirect_hash:
7594 return &bpf_msg_redirect_hash_proto;
7595 case BPF_FUNC_msg_apply_bytes:
7596 return &bpf_msg_apply_bytes_proto;
7597 case BPF_FUNC_msg_cork_bytes:
7598 return &bpf_msg_cork_bytes_proto;
7599 case BPF_FUNC_msg_pull_data:
7600 return &bpf_msg_pull_data_proto;
7601 case BPF_FUNC_msg_push_data:
7602 return &bpf_msg_push_data_proto;
7603 case BPF_FUNC_msg_pop_data:
7604 return &bpf_msg_pop_data_proto;
7605 case BPF_FUNC_perf_event_output:
7606 return &bpf_event_output_data_proto;
7607 case BPF_FUNC_get_current_uid_gid:
7608 return &bpf_get_current_uid_gid_proto;
7609 case BPF_FUNC_get_current_pid_tgid:
7610 return &bpf_get_current_pid_tgid_proto;
7611 case BPF_FUNC_sk_storage_get:
7612 return &bpf_sk_storage_get_proto;
7613 case BPF_FUNC_sk_storage_delete:
7614 return &bpf_sk_storage_delete_proto;
7615 case BPF_FUNC_get_netns_cookie:
7616 return &bpf_get_netns_cookie_sk_msg_proto;
7617 #ifdef CONFIG_CGROUPS
7618 case BPF_FUNC_get_current_cgroup_id:
7619 return &bpf_get_current_cgroup_id_proto;
7620 case BPF_FUNC_get_current_ancestor_cgroup_id:
7621 return &bpf_get_current_ancestor_cgroup_id_proto;
7623 #ifdef CONFIG_CGROUP_NET_CLASSID
7624 case BPF_FUNC_get_cgroup_classid:
7625 return &bpf_get_cgroup_classid_curr_proto;
7628 return bpf_sk_base_func_proto(func_id);
7632 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
7633 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
7635 static const struct bpf_func_proto *
7636 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7639 case BPF_FUNC_skb_store_bytes:
7640 return &bpf_skb_store_bytes_proto;
7641 case BPF_FUNC_skb_load_bytes:
7642 return &bpf_skb_load_bytes_proto;
7643 case BPF_FUNC_skb_pull_data:
7644 return &sk_skb_pull_data_proto;
7645 case BPF_FUNC_skb_change_tail:
7646 return &sk_skb_change_tail_proto;
7647 case BPF_FUNC_skb_change_head:
7648 return &sk_skb_change_head_proto;
7649 case BPF_FUNC_skb_adjust_room:
7650 return &sk_skb_adjust_room_proto;
7651 case BPF_FUNC_get_socket_cookie:
7652 return &bpf_get_socket_cookie_proto;
7653 case BPF_FUNC_get_socket_uid:
7654 return &bpf_get_socket_uid_proto;
7655 case BPF_FUNC_sk_redirect_map:
7656 return &bpf_sk_redirect_map_proto;
7657 case BPF_FUNC_sk_redirect_hash:
7658 return &bpf_sk_redirect_hash_proto;
7659 case BPF_FUNC_perf_event_output:
7660 return &bpf_skb_event_output_proto;
7662 case BPF_FUNC_sk_lookup_tcp:
7663 return &bpf_sk_lookup_tcp_proto;
7664 case BPF_FUNC_sk_lookup_udp:
7665 return &bpf_sk_lookup_udp_proto;
7666 case BPF_FUNC_sk_release:
7667 return &bpf_sk_release_proto;
7668 case BPF_FUNC_skc_lookup_tcp:
7669 return &bpf_skc_lookup_tcp_proto;
7672 return bpf_sk_base_func_proto(func_id);
7676 static const struct bpf_func_proto *
7677 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7680 case BPF_FUNC_skb_load_bytes:
7681 return &bpf_flow_dissector_load_bytes_proto;
7683 return bpf_sk_base_func_proto(func_id);
7687 static const struct bpf_func_proto *
7688 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7691 case BPF_FUNC_skb_load_bytes:
7692 return &bpf_skb_load_bytes_proto;
7693 case BPF_FUNC_skb_pull_data:
7694 return &bpf_skb_pull_data_proto;
7695 case BPF_FUNC_csum_diff:
7696 return &bpf_csum_diff_proto;
7697 case BPF_FUNC_get_cgroup_classid:
7698 return &bpf_get_cgroup_classid_proto;
7699 case BPF_FUNC_get_route_realm:
7700 return &bpf_get_route_realm_proto;
7701 case BPF_FUNC_get_hash_recalc:
7702 return &bpf_get_hash_recalc_proto;
7703 case BPF_FUNC_perf_event_output:
7704 return &bpf_skb_event_output_proto;
7705 case BPF_FUNC_get_smp_processor_id:
7706 return &bpf_get_smp_processor_id_proto;
7707 case BPF_FUNC_skb_under_cgroup:
7708 return &bpf_skb_under_cgroup_proto;
7710 return bpf_sk_base_func_proto(func_id);
7714 static const struct bpf_func_proto *
7715 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7718 case BPF_FUNC_lwt_push_encap:
7719 return &bpf_lwt_in_push_encap_proto;
7721 return lwt_out_func_proto(func_id, prog);
7725 static const struct bpf_func_proto *
7726 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7729 case BPF_FUNC_skb_get_tunnel_key:
7730 return &bpf_skb_get_tunnel_key_proto;
7731 case BPF_FUNC_skb_set_tunnel_key:
7732 return bpf_get_skb_set_tunnel_proto(func_id);
7733 case BPF_FUNC_skb_get_tunnel_opt:
7734 return &bpf_skb_get_tunnel_opt_proto;
7735 case BPF_FUNC_skb_set_tunnel_opt:
7736 return bpf_get_skb_set_tunnel_proto(func_id);
7737 case BPF_FUNC_redirect:
7738 return &bpf_redirect_proto;
7739 case BPF_FUNC_clone_redirect:
7740 return &bpf_clone_redirect_proto;
7741 case BPF_FUNC_skb_change_tail:
7742 return &bpf_skb_change_tail_proto;
7743 case BPF_FUNC_skb_change_head:
7744 return &bpf_skb_change_head_proto;
7745 case BPF_FUNC_skb_store_bytes:
7746 return &bpf_skb_store_bytes_proto;
7747 case BPF_FUNC_csum_update:
7748 return &bpf_csum_update_proto;
7749 case BPF_FUNC_csum_level:
7750 return &bpf_csum_level_proto;
7751 case BPF_FUNC_l3_csum_replace:
7752 return &bpf_l3_csum_replace_proto;
7753 case BPF_FUNC_l4_csum_replace:
7754 return &bpf_l4_csum_replace_proto;
7755 case BPF_FUNC_set_hash_invalid:
7756 return &bpf_set_hash_invalid_proto;
7757 case BPF_FUNC_lwt_push_encap:
7758 return &bpf_lwt_xmit_push_encap_proto;
7760 return lwt_out_func_proto(func_id, prog);
7764 static const struct bpf_func_proto *
7765 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7768 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7769 case BPF_FUNC_lwt_seg6_store_bytes:
7770 return &bpf_lwt_seg6_store_bytes_proto;
7771 case BPF_FUNC_lwt_seg6_action:
7772 return &bpf_lwt_seg6_action_proto;
7773 case BPF_FUNC_lwt_seg6_adjust_srh:
7774 return &bpf_lwt_seg6_adjust_srh_proto;
7777 return lwt_out_func_proto(func_id, prog);
7781 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
7782 const struct bpf_prog *prog,
7783 struct bpf_insn_access_aux *info)
7785 const int size_default = sizeof(__u32);
7787 if (off < 0 || off >= sizeof(struct __sk_buff))
7790 /* The verifier guarantees that size > 0. */
7791 if (off % size != 0)
7795 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7796 if (off + size > offsetofend(struct __sk_buff, cb[4]))
7799 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
7800 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
7801 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
7802 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
7803 case bpf_ctx_range(struct __sk_buff, data):
7804 case bpf_ctx_range(struct __sk_buff, data_meta):
7805 case bpf_ctx_range(struct __sk_buff, data_end):
7806 if (size != size_default)
7809 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
7811 case bpf_ctx_range(struct __sk_buff, tstamp):
7812 if (size != sizeof(__u64))
7815 case offsetof(struct __sk_buff, sk):
7816 if (type == BPF_WRITE || size != sizeof(__u64))
7818 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
7821 /* Only narrow read access allowed for now. */
7822 if (type == BPF_WRITE) {
7823 if (size != size_default)
7826 bpf_ctx_record_field_size(info, size_default);
7827 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
7835 static bool sk_filter_is_valid_access(int off, int size,
7836 enum bpf_access_type type,
7837 const struct bpf_prog *prog,
7838 struct bpf_insn_access_aux *info)
7841 case bpf_ctx_range(struct __sk_buff, tc_classid):
7842 case bpf_ctx_range(struct __sk_buff, data):
7843 case bpf_ctx_range(struct __sk_buff, data_meta):
7844 case bpf_ctx_range(struct __sk_buff, data_end):
7845 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7846 case bpf_ctx_range(struct __sk_buff, tstamp):
7847 case bpf_ctx_range(struct __sk_buff, wire_len):
7851 if (type == BPF_WRITE) {
7853 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7860 return bpf_skb_is_valid_access(off, size, type, prog, info);
7863 static bool cg_skb_is_valid_access(int off, int size,
7864 enum bpf_access_type type,
7865 const struct bpf_prog *prog,
7866 struct bpf_insn_access_aux *info)
7869 case bpf_ctx_range(struct __sk_buff, tc_classid):
7870 case bpf_ctx_range(struct __sk_buff, data_meta):
7871 case bpf_ctx_range(struct __sk_buff, wire_len):
7873 case bpf_ctx_range(struct __sk_buff, data):
7874 case bpf_ctx_range(struct __sk_buff, data_end):
7880 if (type == BPF_WRITE) {
7882 case bpf_ctx_range(struct __sk_buff, mark):
7883 case bpf_ctx_range(struct __sk_buff, priority):
7884 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7886 case bpf_ctx_range(struct __sk_buff, tstamp):
7896 case bpf_ctx_range(struct __sk_buff, data):
7897 info->reg_type = PTR_TO_PACKET;
7899 case bpf_ctx_range(struct __sk_buff, data_end):
7900 info->reg_type = PTR_TO_PACKET_END;
7904 return bpf_skb_is_valid_access(off, size, type, prog, info);
7907 static bool lwt_is_valid_access(int off, int size,
7908 enum bpf_access_type type,
7909 const struct bpf_prog *prog,
7910 struct bpf_insn_access_aux *info)
7913 case bpf_ctx_range(struct __sk_buff, tc_classid):
7914 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7915 case bpf_ctx_range(struct __sk_buff, data_meta):
7916 case bpf_ctx_range(struct __sk_buff, tstamp):
7917 case bpf_ctx_range(struct __sk_buff, wire_len):
7921 if (type == BPF_WRITE) {
7923 case bpf_ctx_range(struct __sk_buff, mark):
7924 case bpf_ctx_range(struct __sk_buff, priority):
7925 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7933 case bpf_ctx_range(struct __sk_buff, data):
7934 info->reg_type = PTR_TO_PACKET;
7936 case bpf_ctx_range(struct __sk_buff, data_end):
7937 info->reg_type = PTR_TO_PACKET_END;
7941 return bpf_skb_is_valid_access(off, size, type, prog, info);
7944 /* Attach type specific accesses */
7945 static bool __sock_filter_check_attach_type(int off,
7946 enum bpf_access_type access_type,
7947 enum bpf_attach_type attach_type)
7950 case offsetof(struct bpf_sock, bound_dev_if):
7951 case offsetof(struct bpf_sock, mark):
7952 case offsetof(struct bpf_sock, priority):
7953 switch (attach_type) {
7954 case BPF_CGROUP_INET_SOCK_CREATE:
7955 case BPF_CGROUP_INET_SOCK_RELEASE:
7960 case bpf_ctx_range(struct bpf_sock, src_ip4):
7961 switch (attach_type) {
7962 case BPF_CGROUP_INET4_POST_BIND:
7967 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7968 switch (attach_type) {
7969 case BPF_CGROUP_INET6_POST_BIND:
7974 case bpf_ctx_range(struct bpf_sock, src_port):
7975 switch (attach_type) {
7976 case BPF_CGROUP_INET4_POST_BIND:
7977 case BPF_CGROUP_INET6_POST_BIND:
7984 return access_type == BPF_READ;
7989 bool bpf_sock_common_is_valid_access(int off, int size,
7990 enum bpf_access_type type,
7991 struct bpf_insn_access_aux *info)
7994 case bpf_ctx_range_till(struct bpf_sock, type, priority):
7997 return bpf_sock_is_valid_access(off, size, type, info);
8001 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
8002 struct bpf_insn_access_aux *info)
8004 const int size_default = sizeof(__u32);
8007 if (off < 0 || off >= sizeof(struct bpf_sock))
8009 if (off % size != 0)
8013 case offsetof(struct bpf_sock, state):
8014 case offsetof(struct bpf_sock, family):
8015 case offsetof(struct bpf_sock, type):
8016 case offsetof(struct bpf_sock, protocol):
8017 case offsetof(struct bpf_sock, src_port):
8018 case offsetof(struct bpf_sock, rx_queue_mapping):
8019 case bpf_ctx_range(struct bpf_sock, src_ip4):
8020 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8021 case bpf_ctx_range(struct bpf_sock, dst_ip4):
8022 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8023 bpf_ctx_record_field_size(info, size_default);
8024 return bpf_ctx_narrow_access_ok(off, size, size_default);
8025 case bpf_ctx_range(struct bpf_sock, dst_port):
8026 field_size = size == size_default ?
8027 size_default : sizeof_field(struct bpf_sock, dst_port);
8028 bpf_ctx_record_field_size(info, field_size);
8029 return bpf_ctx_narrow_access_ok(off, size, field_size);
8030 case offsetofend(struct bpf_sock, dst_port) ...
8031 offsetof(struct bpf_sock, dst_ip4) - 1:
8035 return size == size_default;
8038 static bool sock_filter_is_valid_access(int off, int size,
8039 enum bpf_access_type type,
8040 const struct bpf_prog *prog,
8041 struct bpf_insn_access_aux *info)
8043 if (!bpf_sock_is_valid_access(off, size, type, info))
8045 return __sock_filter_check_attach_type(off, type,
8046 prog->expected_attach_type);
8049 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
8050 const struct bpf_prog *prog)
8052 /* Neither direct read nor direct write requires any preliminary
8058 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
8059 const struct bpf_prog *prog, int drop_verdict)
8061 struct bpf_insn *insn = insn_buf;
8066 /* if (!skb->cloned)
8069 * (Fast-path, otherwise approximation that we might be
8070 * a clone, do the rest in helper.)
8072 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
8073 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
8074 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
8076 /* ret = bpf_skb_pull_data(skb, 0); */
8077 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
8078 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
8079 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
8080 BPF_FUNC_skb_pull_data);
8083 * return TC_ACT_SHOT;
8085 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
8086 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
8087 *insn++ = BPF_EXIT_INSN();
8090 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
8092 *insn++ = prog->insnsi[0];
8094 return insn - insn_buf;
8097 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
8098 struct bpf_insn *insn_buf)
8100 bool indirect = BPF_MODE(orig->code) == BPF_IND;
8101 struct bpf_insn *insn = insn_buf;
8104 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
8106 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
8108 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
8110 /* We're guaranteed here that CTX is in R6. */
8111 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
8113 switch (BPF_SIZE(orig->code)) {
8115 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
8118 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
8121 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
8125 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
8126 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
8127 *insn++ = BPF_EXIT_INSN();
8129 return insn - insn_buf;
8132 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8133 const struct bpf_prog *prog)
8135 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8138 static bool tc_cls_act_is_valid_access(int off, int size,
8139 enum bpf_access_type type,
8140 const struct bpf_prog *prog,
8141 struct bpf_insn_access_aux *info)
8143 if (type == BPF_WRITE) {
8145 case bpf_ctx_range(struct __sk_buff, mark):
8146 case bpf_ctx_range(struct __sk_buff, tc_index):
8147 case bpf_ctx_range(struct __sk_buff, priority):
8148 case bpf_ctx_range(struct __sk_buff, tc_classid):
8149 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8150 case bpf_ctx_range(struct __sk_buff, tstamp):
8151 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8159 case bpf_ctx_range(struct __sk_buff, data):
8160 info->reg_type = PTR_TO_PACKET;
8162 case bpf_ctx_range(struct __sk_buff, data_meta):
8163 info->reg_type = PTR_TO_PACKET_META;
8165 case bpf_ctx_range(struct __sk_buff, data_end):
8166 info->reg_type = PTR_TO_PACKET_END;
8168 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8172 return bpf_skb_is_valid_access(off, size, type, prog, info);
8175 static bool __is_valid_xdp_access(int off, int size)
8177 if (off < 0 || off >= sizeof(struct xdp_md))
8179 if (off % size != 0)
8181 if (size != sizeof(__u32))
8187 static bool xdp_is_valid_access(int off, int size,
8188 enum bpf_access_type type,
8189 const struct bpf_prog *prog,
8190 struct bpf_insn_access_aux *info)
8192 if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8194 case offsetof(struct xdp_md, egress_ifindex):
8199 if (type == BPF_WRITE) {
8200 if (bpf_prog_is_dev_bound(prog->aux)) {
8202 case offsetof(struct xdp_md, rx_queue_index):
8203 return __is_valid_xdp_access(off, size);
8210 case offsetof(struct xdp_md, data):
8211 info->reg_type = PTR_TO_PACKET;
8213 case offsetof(struct xdp_md, data_meta):
8214 info->reg_type = PTR_TO_PACKET_META;
8216 case offsetof(struct xdp_md, data_end):
8217 info->reg_type = PTR_TO_PACKET_END;
8221 return __is_valid_xdp_access(off, size);
8224 void bpf_warn_invalid_xdp_action(u32 act)
8226 const u32 act_max = XDP_REDIRECT;
8228 pr_warn_once("%s XDP return value %u, expect packet loss!\n",
8229 act > act_max ? "Illegal" : "Driver unsupported",
8232 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
8234 static bool sock_addr_is_valid_access(int off, int size,
8235 enum bpf_access_type type,
8236 const struct bpf_prog *prog,
8237 struct bpf_insn_access_aux *info)
8239 const int size_default = sizeof(__u32);
8241 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
8243 if (off % size != 0)
8246 /* Disallow access to IPv6 fields from IPv4 contex and vise
8250 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8251 switch (prog->expected_attach_type) {
8252 case BPF_CGROUP_INET4_BIND:
8253 case BPF_CGROUP_INET4_CONNECT:
8254 case BPF_CGROUP_INET4_GETPEERNAME:
8255 case BPF_CGROUP_INET4_GETSOCKNAME:
8256 case BPF_CGROUP_UDP4_SENDMSG:
8257 case BPF_CGROUP_UDP4_RECVMSG:
8263 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8264 switch (prog->expected_attach_type) {
8265 case BPF_CGROUP_INET6_BIND:
8266 case BPF_CGROUP_INET6_CONNECT:
8267 case BPF_CGROUP_INET6_GETPEERNAME:
8268 case BPF_CGROUP_INET6_GETSOCKNAME:
8269 case BPF_CGROUP_UDP6_SENDMSG:
8270 case BPF_CGROUP_UDP6_RECVMSG:
8276 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8277 switch (prog->expected_attach_type) {
8278 case BPF_CGROUP_UDP4_SENDMSG:
8284 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8286 switch (prog->expected_attach_type) {
8287 case BPF_CGROUP_UDP6_SENDMSG:
8296 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8297 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8298 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8299 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8301 case bpf_ctx_range(struct bpf_sock_addr, user_port):
8302 if (type == BPF_READ) {
8303 bpf_ctx_record_field_size(info, size_default);
8305 if (bpf_ctx_wide_access_ok(off, size,
8306 struct bpf_sock_addr,
8310 if (bpf_ctx_wide_access_ok(off, size,
8311 struct bpf_sock_addr,
8315 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8318 if (bpf_ctx_wide_access_ok(off, size,
8319 struct bpf_sock_addr,
8323 if (bpf_ctx_wide_access_ok(off, size,
8324 struct bpf_sock_addr,
8328 if (size != size_default)
8332 case offsetof(struct bpf_sock_addr, sk):
8333 if (type != BPF_READ)
8335 if (size != sizeof(__u64))
8337 info->reg_type = PTR_TO_SOCKET;
8340 if (type == BPF_READ) {
8341 if (size != size_default)
8351 static bool sock_ops_is_valid_access(int off, int size,
8352 enum bpf_access_type type,
8353 const struct bpf_prog *prog,
8354 struct bpf_insn_access_aux *info)
8356 const int size_default = sizeof(__u32);
8358 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
8361 /* The verifier guarantees that size > 0. */
8362 if (off % size != 0)
8365 if (type == BPF_WRITE) {
8367 case offsetof(struct bpf_sock_ops, reply):
8368 case offsetof(struct bpf_sock_ops, sk_txhash):
8369 if (size != size_default)
8377 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
8379 if (size != sizeof(__u64))
8382 case offsetof(struct bpf_sock_ops, sk):
8383 if (size != sizeof(__u64))
8385 info->reg_type = PTR_TO_SOCKET_OR_NULL;
8387 case offsetof(struct bpf_sock_ops, skb_data):
8388 if (size != sizeof(__u64))
8390 info->reg_type = PTR_TO_PACKET;
8392 case offsetof(struct bpf_sock_ops, skb_data_end):
8393 if (size != sizeof(__u64))
8395 info->reg_type = PTR_TO_PACKET_END;
8397 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
8398 bpf_ctx_record_field_size(info, size_default);
8399 return bpf_ctx_narrow_access_ok(off, size,
8402 if (size != size_default)
8411 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
8412 const struct bpf_prog *prog)
8414 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
8417 static bool sk_skb_is_valid_access(int off, int size,
8418 enum bpf_access_type type,
8419 const struct bpf_prog *prog,
8420 struct bpf_insn_access_aux *info)
8423 case bpf_ctx_range(struct __sk_buff, tc_classid):
8424 case bpf_ctx_range(struct __sk_buff, data_meta):
8425 case bpf_ctx_range(struct __sk_buff, tstamp):
8426 case bpf_ctx_range(struct __sk_buff, wire_len):
8430 if (type == BPF_WRITE) {
8432 case bpf_ctx_range(struct __sk_buff, tc_index):
8433 case bpf_ctx_range(struct __sk_buff, priority):
8441 case bpf_ctx_range(struct __sk_buff, mark):
8443 case bpf_ctx_range(struct __sk_buff, data):
8444 info->reg_type = PTR_TO_PACKET;
8446 case bpf_ctx_range(struct __sk_buff, data_end):
8447 info->reg_type = PTR_TO_PACKET_END;
8451 return bpf_skb_is_valid_access(off, size, type, prog, info);
8454 static bool sk_msg_is_valid_access(int off, int size,
8455 enum bpf_access_type type,
8456 const struct bpf_prog *prog,
8457 struct bpf_insn_access_aux *info)
8459 if (type == BPF_WRITE)
8462 if (off % size != 0)
8466 case offsetof(struct sk_msg_md, data):
8467 info->reg_type = PTR_TO_PACKET;
8468 if (size != sizeof(__u64))
8471 case offsetof(struct sk_msg_md, data_end):
8472 info->reg_type = PTR_TO_PACKET_END;
8473 if (size != sizeof(__u64))
8476 case offsetof(struct sk_msg_md, sk):
8477 if (size != sizeof(__u64))
8479 info->reg_type = PTR_TO_SOCKET;
8481 case bpf_ctx_range(struct sk_msg_md, family):
8482 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
8483 case bpf_ctx_range(struct sk_msg_md, local_ip4):
8484 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
8485 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
8486 case bpf_ctx_range(struct sk_msg_md, remote_port):
8487 case bpf_ctx_range(struct sk_msg_md, local_port):
8488 case bpf_ctx_range(struct sk_msg_md, size):
8489 if (size != sizeof(__u32))
8498 static bool flow_dissector_is_valid_access(int off, int size,
8499 enum bpf_access_type type,
8500 const struct bpf_prog *prog,
8501 struct bpf_insn_access_aux *info)
8503 const int size_default = sizeof(__u32);
8505 if (off < 0 || off >= sizeof(struct __sk_buff))
8508 if (type == BPF_WRITE)
8512 case bpf_ctx_range(struct __sk_buff, data):
8513 if (size != size_default)
8515 info->reg_type = PTR_TO_PACKET;
8517 case bpf_ctx_range(struct __sk_buff, data_end):
8518 if (size != size_default)
8520 info->reg_type = PTR_TO_PACKET_END;
8522 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8523 if (size != sizeof(__u64))
8525 info->reg_type = PTR_TO_FLOW_KEYS;
8532 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
8533 const struct bpf_insn *si,
8534 struct bpf_insn *insn_buf,
8535 struct bpf_prog *prog,
8539 struct bpf_insn *insn = insn_buf;
8542 case offsetof(struct __sk_buff, data):
8543 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
8544 si->dst_reg, si->src_reg,
8545 offsetof(struct bpf_flow_dissector, data));
8548 case offsetof(struct __sk_buff, data_end):
8549 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
8550 si->dst_reg, si->src_reg,
8551 offsetof(struct bpf_flow_dissector, data_end));
8554 case offsetof(struct __sk_buff, flow_keys):
8555 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
8556 si->dst_reg, si->src_reg,
8557 offsetof(struct bpf_flow_dissector, flow_keys));
8561 return insn - insn_buf;
8564 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si,
8565 struct bpf_insn *insn)
8567 /* si->dst_reg = skb_shinfo(SKB); */
8568 #ifdef NET_SKBUFF_DATA_USES_OFFSET
8569 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8570 BPF_REG_AX, si->src_reg,
8571 offsetof(struct sk_buff, end));
8572 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
8573 si->dst_reg, si->src_reg,
8574 offsetof(struct sk_buff, head));
8575 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
8577 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8578 si->dst_reg, si->src_reg,
8579 offsetof(struct sk_buff, end));
8585 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
8586 const struct bpf_insn *si,
8587 struct bpf_insn *insn_buf,
8588 struct bpf_prog *prog, u32 *target_size)
8590 struct bpf_insn *insn = insn_buf;
8594 case offsetof(struct __sk_buff, len):
8595 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8596 bpf_target_off(struct sk_buff, len, 4,
8600 case offsetof(struct __sk_buff, protocol):
8601 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8602 bpf_target_off(struct sk_buff, protocol, 2,
8606 case offsetof(struct __sk_buff, vlan_proto):
8607 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8608 bpf_target_off(struct sk_buff, vlan_proto, 2,
8612 case offsetof(struct __sk_buff, priority):
8613 if (type == BPF_WRITE)
8614 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8615 bpf_target_off(struct sk_buff, priority, 4,
8618 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8619 bpf_target_off(struct sk_buff, priority, 4,
8623 case offsetof(struct __sk_buff, ingress_ifindex):
8624 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8625 bpf_target_off(struct sk_buff, skb_iif, 4,
8629 case offsetof(struct __sk_buff, ifindex):
8630 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8631 si->dst_reg, si->src_reg,
8632 offsetof(struct sk_buff, dev));
8633 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
8634 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8635 bpf_target_off(struct net_device, ifindex, 4,
8639 case offsetof(struct __sk_buff, hash):
8640 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8641 bpf_target_off(struct sk_buff, hash, 4,
8645 case offsetof(struct __sk_buff, mark):
8646 if (type == BPF_WRITE)
8647 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8648 bpf_target_off(struct sk_buff, mark, 4,
8651 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8652 bpf_target_off(struct sk_buff, mark, 4,
8656 case offsetof(struct __sk_buff, pkt_type):
8658 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8660 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
8661 #ifdef __BIG_ENDIAN_BITFIELD
8662 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
8666 case offsetof(struct __sk_buff, queue_mapping):
8667 if (type == BPF_WRITE) {
8668 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
8669 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8670 bpf_target_off(struct sk_buff,
8674 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8675 bpf_target_off(struct sk_buff,
8681 case offsetof(struct __sk_buff, vlan_present):
8683 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8684 PKT_VLAN_PRESENT_OFFSET());
8685 if (PKT_VLAN_PRESENT_BIT)
8686 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
8687 if (PKT_VLAN_PRESENT_BIT < 7)
8688 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
8691 case offsetof(struct __sk_buff, vlan_tci):
8692 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8693 bpf_target_off(struct sk_buff, vlan_tci, 2,
8697 case offsetof(struct __sk_buff, cb[0]) ...
8698 offsetofend(struct __sk_buff, cb[4]) - 1:
8699 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
8700 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
8701 offsetof(struct qdisc_skb_cb, data)) %
8704 prog->cb_access = 1;
8706 off -= offsetof(struct __sk_buff, cb[0]);
8707 off += offsetof(struct sk_buff, cb);
8708 off += offsetof(struct qdisc_skb_cb, data);
8709 if (type == BPF_WRITE)
8710 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
8713 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
8717 case offsetof(struct __sk_buff, tc_classid):
8718 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
8721 off -= offsetof(struct __sk_buff, tc_classid);
8722 off += offsetof(struct sk_buff, cb);
8723 off += offsetof(struct qdisc_skb_cb, tc_classid);
8725 if (type == BPF_WRITE)
8726 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
8729 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
8733 case offsetof(struct __sk_buff, data):
8734 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
8735 si->dst_reg, si->src_reg,
8736 offsetof(struct sk_buff, data));
8739 case offsetof(struct __sk_buff, data_meta):
8741 off -= offsetof(struct __sk_buff, data_meta);
8742 off += offsetof(struct sk_buff, cb);
8743 off += offsetof(struct bpf_skb_data_end, data_meta);
8744 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8748 case offsetof(struct __sk_buff, data_end):
8750 off -= offsetof(struct __sk_buff, data_end);
8751 off += offsetof(struct sk_buff, cb);
8752 off += offsetof(struct bpf_skb_data_end, data_end);
8753 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8757 case offsetof(struct __sk_buff, tc_index):
8758 #ifdef CONFIG_NET_SCHED
8759 if (type == BPF_WRITE)
8760 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8761 bpf_target_off(struct sk_buff, tc_index, 2,
8764 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8765 bpf_target_off(struct sk_buff, tc_index, 2,
8769 if (type == BPF_WRITE)
8770 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
8772 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8776 case offsetof(struct __sk_buff, napi_id):
8777 #if defined(CONFIG_NET_RX_BUSY_POLL)
8778 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8779 bpf_target_off(struct sk_buff, napi_id, 4,
8781 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
8782 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8785 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8788 case offsetof(struct __sk_buff, family):
8789 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
8791 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8792 si->dst_reg, si->src_reg,
8793 offsetof(struct sk_buff, sk));
8794 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8795 bpf_target_off(struct sock_common,
8799 case offsetof(struct __sk_buff, remote_ip4):
8800 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
8802 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8803 si->dst_reg, si->src_reg,
8804 offsetof(struct sk_buff, sk));
8805 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8806 bpf_target_off(struct sock_common,
8810 case offsetof(struct __sk_buff, local_ip4):
8811 BUILD_BUG_ON(sizeof_field(struct sock_common,
8812 skc_rcv_saddr) != 4);
8814 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8815 si->dst_reg, si->src_reg,
8816 offsetof(struct sk_buff, sk));
8817 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8818 bpf_target_off(struct sock_common,
8822 case offsetof(struct __sk_buff, remote_ip6[0]) ...
8823 offsetof(struct __sk_buff, remote_ip6[3]):
8824 #if IS_ENABLED(CONFIG_IPV6)
8825 BUILD_BUG_ON(sizeof_field(struct sock_common,
8826 skc_v6_daddr.s6_addr32[0]) != 4);
8829 off -= offsetof(struct __sk_buff, remote_ip6[0]);
8831 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8832 si->dst_reg, si->src_reg,
8833 offsetof(struct sk_buff, sk));
8834 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8835 offsetof(struct sock_common,
8836 skc_v6_daddr.s6_addr32[0]) +
8839 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8842 case offsetof(struct __sk_buff, local_ip6[0]) ...
8843 offsetof(struct __sk_buff, local_ip6[3]):
8844 #if IS_ENABLED(CONFIG_IPV6)
8845 BUILD_BUG_ON(sizeof_field(struct sock_common,
8846 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8849 off -= offsetof(struct __sk_buff, local_ip6[0]);
8851 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8852 si->dst_reg, si->src_reg,
8853 offsetof(struct sk_buff, sk));
8854 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8855 offsetof(struct sock_common,
8856 skc_v6_rcv_saddr.s6_addr32[0]) +
8859 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8863 case offsetof(struct __sk_buff, remote_port):
8864 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
8866 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8867 si->dst_reg, si->src_reg,
8868 offsetof(struct sk_buff, sk));
8869 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8870 bpf_target_off(struct sock_common,
8873 #ifndef __BIG_ENDIAN_BITFIELD
8874 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8878 case offsetof(struct __sk_buff, local_port):
8879 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
8881 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8882 si->dst_reg, si->src_reg,
8883 offsetof(struct sk_buff, sk));
8884 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8885 bpf_target_off(struct sock_common,
8886 skc_num, 2, target_size));
8889 case offsetof(struct __sk_buff, tstamp):
8890 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
8892 if (type == BPF_WRITE)
8893 *insn++ = BPF_STX_MEM(BPF_DW,
8894 si->dst_reg, si->src_reg,
8895 bpf_target_off(struct sk_buff,
8899 *insn++ = BPF_LDX_MEM(BPF_DW,
8900 si->dst_reg, si->src_reg,
8901 bpf_target_off(struct sk_buff,
8906 case offsetof(struct __sk_buff, gso_segs):
8907 insn = bpf_convert_shinfo_access(si, insn);
8908 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
8909 si->dst_reg, si->dst_reg,
8910 bpf_target_off(struct skb_shared_info,
8914 case offsetof(struct __sk_buff, gso_size):
8915 insn = bpf_convert_shinfo_access(si, insn);
8916 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
8917 si->dst_reg, si->dst_reg,
8918 bpf_target_off(struct skb_shared_info,
8922 case offsetof(struct __sk_buff, wire_len):
8923 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
8926 off -= offsetof(struct __sk_buff, wire_len);
8927 off += offsetof(struct sk_buff, cb);
8928 off += offsetof(struct qdisc_skb_cb, pkt_len);
8930 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
8933 case offsetof(struct __sk_buff, sk):
8934 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8935 si->dst_reg, si->src_reg,
8936 offsetof(struct sk_buff, sk));
8940 return insn - insn_buf;
8943 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
8944 const struct bpf_insn *si,
8945 struct bpf_insn *insn_buf,
8946 struct bpf_prog *prog, u32 *target_size)
8948 struct bpf_insn *insn = insn_buf;
8952 case offsetof(struct bpf_sock, bound_dev_if):
8953 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
8955 if (type == BPF_WRITE)
8956 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8957 offsetof(struct sock, sk_bound_dev_if));
8959 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8960 offsetof(struct sock, sk_bound_dev_if));
8963 case offsetof(struct bpf_sock, mark):
8964 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
8966 if (type == BPF_WRITE)
8967 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8968 offsetof(struct sock, sk_mark));
8970 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8971 offsetof(struct sock, sk_mark));
8974 case offsetof(struct bpf_sock, priority):
8975 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
8977 if (type == BPF_WRITE)
8978 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8979 offsetof(struct sock, sk_priority));
8981 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8982 offsetof(struct sock, sk_priority));
8985 case offsetof(struct bpf_sock, family):
8986 *insn++ = BPF_LDX_MEM(
8987 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
8988 si->dst_reg, si->src_reg,
8989 bpf_target_off(struct sock_common,
8991 sizeof_field(struct sock_common,
8996 case offsetof(struct bpf_sock, type):
8997 *insn++ = BPF_LDX_MEM(
8998 BPF_FIELD_SIZEOF(struct sock, sk_type),
8999 si->dst_reg, si->src_reg,
9000 bpf_target_off(struct sock, sk_type,
9001 sizeof_field(struct sock, sk_type),
9005 case offsetof(struct bpf_sock, protocol):
9006 *insn++ = BPF_LDX_MEM(
9007 BPF_FIELD_SIZEOF(struct sock, sk_protocol),
9008 si->dst_reg, si->src_reg,
9009 bpf_target_off(struct sock, sk_protocol,
9010 sizeof_field(struct sock, sk_protocol),
9014 case offsetof(struct bpf_sock, src_ip4):
9015 *insn++ = BPF_LDX_MEM(
9016 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9017 bpf_target_off(struct sock_common, skc_rcv_saddr,
9018 sizeof_field(struct sock_common,
9023 case offsetof(struct bpf_sock, dst_ip4):
9024 *insn++ = BPF_LDX_MEM(
9025 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9026 bpf_target_off(struct sock_common, skc_daddr,
9027 sizeof_field(struct sock_common,
9032 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
9033 #if IS_ENABLED(CONFIG_IPV6)
9035 off -= offsetof(struct bpf_sock, src_ip6[0]);
9036 *insn++ = BPF_LDX_MEM(
9037 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9040 skc_v6_rcv_saddr.s6_addr32[0],
9041 sizeof_field(struct sock_common,
9042 skc_v6_rcv_saddr.s6_addr32[0]),
9043 target_size) + off);
9046 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9050 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
9051 #if IS_ENABLED(CONFIG_IPV6)
9053 off -= offsetof(struct bpf_sock, dst_ip6[0]);
9054 *insn++ = BPF_LDX_MEM(
9055 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9056 bpf_target_off(struct sock_common,
9057 skc_v6_daddr.s6_addr32[0],
9058 sizeof_field(struct sock_common,
9059 skc_v6_daddr.s6_addr32[0]),
9060 target_size) + off);
9062 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9067 case offsetof(struct bpf_sock, src_port):
9068 *insn++ = BPF_LDX_MEM(
9069 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
9070 si->dst_reg, si->src_reg,
9071 bpf_target_off(struct sock_common, skc_num,
9072 sizeof_field(struct sock_common,
9077 case offsetof(struct bpf_sock, dst_port):
9078 *insn++ = BPF_LDX_MEM(
9079 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
9080 si->dst_reg, si->src_reg,
9081 bpf_target_off(struct sock_common, skc_dport,
9082 sizeof_field(struct sock_common,
9087 case offsetof(struct bpf_sock, state):
9088 *insn++ = BPF_LDX_MEM(
9089 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
9090 si->dst_reg, si->src_reg,
9091 bpf_target_off(struct sock_common, skc_state,
9092 sizeof_field(struct sock_common,
9096 case offsetof(struct bpf_sock, rx_queue_mapping):
9097 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
9098 *insn++ = BPF_LDX_MEM(
9099 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
9100 si->dst_reg, si->src_reg,
9101 bpf_target_off(struct sock, sk_rx_queue_mapping,
9102 sizeof_field(struct sock,
9103 sk_rx_queue_mapping),
9105 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
9107 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9109 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9115 return insn - insn_buf;
9118 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
9119 const struct bpf_insn *si,
9120 struct bpf_insn *insn_buf,
9121 struct bpf_prog *prog, u32 *target_size)
9123 struct bpf_insn *insn = insn_buf;
9126 case offsetof(struct __sk_buff, ifindex):
9127 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9128 si->dst_reg, si->src_reg,
9129 offsetof(struct sk_buff, dev));
9130 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9131 bpf_target_off(struct net_device, ifindex, 4,
9135 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9139 return insn - insn_buf;
9142 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
9143 const struct bpf_insn *si,
9144 struct bpf_insn *insn_buf,
9145 struct bpf_prog *prog, u32 *target_size)
9147 struct bpf_insn *insn = insn_buf;
9150 case offsetof(struct xdp_md, data):
9151 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
9152 si->dst_reg, si->src_reg,
9153 offsetof(struct xdp_buff, data));
9155 case offsetof(struct xdp_md, data_meta):
9156 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
9157 si->dst_reg, si->src_reg,
9158 offsetof(struct xdp_buff, data_meta));
9160 case offsetof(struct xdp_md, data_end):
9161 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
9162 si->dst_reg, si->src_reg,
9163 offsetof(struct xdp_buff, data_end));
9165 case offsetof(struct xdp_md, ingress_ifindex):
9166 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9167 si->dst_reg, si->src_reg,
9168 offsetof(struct xdp_buff, rxq));
9169 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
9170 si->dst_reg, si->dst_reg,
9171 offsetof(struct xdp_rxq_info, dev));
9172 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9173 offsetof(struct net_device, ifindex));
9175 case offsetof(struct xdp_md, rx_queue_index):
9176 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9177 si->dst_reg, si->src_reg,
9178 offsetof(struct xdp_buff, rxq));
9179 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9180 offsetof(struct xdp_rxq_info,
9183 case offsetof(struct xdp_md, egress_ifindex):
9184 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
9185 si->dst_reg, si->src_reg,
9186 offsetof(struct xdp_buff, txq));
9187 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
9188 si->dst_reg, si->dst_reg,
9189 offsetof(struct xdp_txq_info, dev));
9190 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9191 offsetof(struct net_device, ifindex));
9195 return insn - insn_buf;
9198 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
9199 * context Structure, F is Field in context structure that contains a pointer
9200 * to Nested Structure of type NS that has the field NF.
9202 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
9203 * sure that SIZE is not greater than actual size of S.F.NF.
9205 * If offset OFF is provided, the load happens from that offset relative to
9208 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
9210 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
9211 si->src_reg, offsetof(S, F)); \
9212 *insn++ = BPF_LDX_MEM( \
9213 SIZE, si->dst_reg, si->dst_reg, \
9214 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
9219 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
9220 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
9221 BPF_FIELD_SIZEOF(NS, NF), 0)
9223 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
9224 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
9226 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
9227 * "register" since two registers available in convert_ctx_access are not
9228 * enough: we can't override neither SRC, since it contains value to store, nor
9229 * DST since it contains pointer to context that may be used by later
9230 * instructions. But we need a temporary place to save pointer to nested
9231 * structure whose field we want to store to.
9233 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
9235 int tmp_reg = BPF_REG_9; \
9236 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
9238 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
9240 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
9242 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
9243 si->dst_reg, offsetof(S, F)); \
9244 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \
9245 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
9248 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
9252 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
9255 if (type == BPF_WRITE) { \
9256 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
9259 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
9260 S, NS, F, NF, SIZE, OFF); \
9264 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
9265 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
9266 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
9268 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
9269 const struct bpf_insn *si,
9270 struct bpf_insn *insn_buf,
9271 struct bpf_prog *prog, u32 *target_size)
9273 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
9274 struct bpf_insn *insn = insn_buf;
9277 case offsetof(struct bpf_sock_addr, user_family):
9278 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9279 struct sockaddr, uaddr, sa_family);
9282 case offsetof(struct bpf_sock_addr, user_ip4):
9283 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9284 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
9285 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
9288 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9290 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
9291 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9292 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9293 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
9297 case offsetof(struct bpf_sock_addr, user_port):
9298 /* To get port we need to know sa_family first and then treat
9299 * sockaddr as either sockaddr_in or sockaddr_in6.
9300 * Though we can simplify since port field has same offset and
9301 * size in both structures.
9302 * Here we check this invariant and use just one of the
9303 * structures if it's true.
9305 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
9306 offsetof(struct sockaddr_in6, sin6_port));
9307 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
9308 sizeof_field(struct sockaddr_in6, sin6_port));
9309 /* Account for sin6_port being smaller than user_port. */
9310 port_size = min(port_size, BPF_LDST_BYTES(si));
9311 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9312 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9313 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
9316 case offsetof(struct bpf_sock_addr, family):
9317 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9318 struct sock, sk, sk_family);
9321 case offsetof(struct bpf_sock_addr, type):
9322 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9323 struct sock, sk, sk_type);
9326 case offsetof(struct bpf_sock_addr, protocol):
9327 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9328 struct sock, sk, sk_protocol);
9331 case offsetof(struct bpf_sock_addr, msg_src_ip4):
9332 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
9333 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9334 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
9335 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
9338 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9341 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
9342 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
9343 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9344 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
9345 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
9347 case offsetof(struct bpf_sock_addr, sk):
9348 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
9349 si->dst_reg, si->src_reg,
9350 offsetof(struct bpf_sock_addr_kern, sk));
9354 return insn - insn_buf;
9357 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
9358 const struct bpf_insn *si,
9359 struct bpf_insn *insn_buf,
9360 struct bpf_prog *prog,
9363 struct bpf_insn *insn = insn_buf;
9366 /* Helper macro for adding read access to tcp_sock or sock fields. */
9367 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
9369 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \
9370 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
9371 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
9372 if (si->dst_reg == reg || si->src_reg == reg) \
9374 if (si->dst_reg == reg || si->src_reg == reg) \
9376 if (si->dst_reg == si->src_reg) { \
9377 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
9378 offsetof(struct bpf_sock_ops_kern, \
9380 fullsock_reg = reg; \
9383 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9384 struct bpf_sock_ops_kern, \
9386 fullsock_reg, si->src_reg, \
9387 offsetof(struct bpf_sock_ops_kern, \
9389 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
9390 if (si->dst_reg == si->src_reg) \
9391 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9392 offsetof(struct bpf_sock_ops_kern, \
9394 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9395 struct bpf_sock_ops_kern, sk),\
9396 si->dst_reg, si->src_reg, \
9397 offsetof(struct bpf_sock_ops_kern, sk));\
9398 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
9400 si->dst_reg, si->dst_reg, \
9401 offsetof(OBJ, OBJ_FIELD)); \
9402 if (si->dst_reg == si->src_reg) { \
9403 *insn++ = BPF_JMP_A(1); \
9404 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9405 offsetof(struct bpf_sock_ops_kern, \
9410 #define SOCK_OPS_GET_SK() \
9412 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \
9413 if (si->dst_reg == reg || si->src_reg == reg) \
9415 if (si->dst_reg == reg || si->src_reg == reg) \
9417 if (si->dst_reg == si->src_reg) { \
9418 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
9419 offsetof(struct bpf_sock_ops_kern, \
9421 fullsock_reg = reg; \
9424 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9425 struct bpf_sock_ops_kern, \
9427 fullsock_reg, si->src_reg, \
9428 offsetof(struct bpf_sock_ops_kern, \
9430 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
9431 if (si->dst_reg == si->src_reg) \
9432 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9433 offsetof(struct bpf_sock_ops_kern, \
9435 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9436 struct bpf_sock_ops_kern, sk),\
9437 si->dst_reg, si->src_reg, \
9438 offsetof(struct bpf_sock_ops_kern, sk));\
9439 if (si->dst_reg == si->src_reg) { \
9440 *insn++ = BPF_JMP_A(1); \
9441 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9442 offsetof(struct bpf_sock_ops_kern, \
9447 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
9448 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
9450 /* Helper macro for adding write access to tcp_sock or sock fields.
9451 * The macro is called with two registers, dst_reg which contains a pointer
9452 * to ctx (context) and src_reg which contains the value that should be
9453 * stored. However, we need an additional register since we cannot overwrite
9454 * dst_reg because it may be used later in the program.
9455 * Instead we "borrow" one of the other register. We first save its value
9456 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
9457 * it at the end of the macro.
9459 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
9461 int reg = BPF_REG_9; \
9462 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
9463 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
9464 if (si->dst_reg == reg || si->src_reg == reg) \
9466 if (si->dst_reg == reg || si->src_reg == reg) \
9468 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
9469 offsetof(struct bpf_sock_ops_kern, \
9471 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9472 struct bpf_sock_ops_kern, \
9475 offsetof(struct bpf_sock_ops_kern, \
9477 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
9478 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9479 struct bpf_sock_ops_kern, sk),\
9481 offsetof(struct bpf_sock_ops_kern, sk));\
9482 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \
9484 offsetof(OBJ, OBJ_FIELD)); \
9485 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
9486 offsetof(struct bpf_sock_ops_kern, \
9490 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
9492 if (TYPE == BPF_WRITE) \
9493 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
9495 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
9498 if (insn > insn_buf)
9499 return insn - insn_buf;
9502 case offsetof(struct bpf_sock_ops, op):
9503 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9505 si->dst_reg, si->src_reg,
9506 offsetof(struct bpf_sock_ops_kern, op));
9509 case offsetof(struct bpf_sock_ops, replylong[0]) ...
9510 offsetof(struct bpf_sock_ops, replylong[3]):
9511 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
9512 sizeof_field(struct bpf_sock_ops_kern, reply));
9513 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
9514 sizeof_field(struct bpf_sock_ops_kern, replylong));
9516 off -= offsetof(struct bpf_sock_ops, replylong[0]);
9517 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
9518 if (type == BPF_WRITE)
9519 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9522 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9526 case offsetof(struct bpf_sock_ops, family):
9527 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9529 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9530 struct bpf_sock_ops_kern, sk),
9531 si->dst_reg, si->src_reg,
9532 offsetof(struct bpf_sock_ops_kern, sk));
9533 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9534 offsetof(struct sock_common, skc_family));
9537 case offsetof(struct bpf_sock_ops, remote_ip4):
9538 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9540 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9541 struct bpf_sock_ops_kern, sk),
9542 si->dst_reg, si->src_reg,
9543 offsetof(struct bpf_sock_ops_kern, sk));
9544 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9545 offsetof(struct sock_common, skc_daddr));
9548 case offsetof(struct bpf_sock_ops, local_ip4):
9549 BUILD_BUG_ON(sizeof_field(struct sock_common,
9550 skc_rcv_saddr) != 4);
9552 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9553 struct bpf_sock_ops_kern, sk),
9554 si->dst_reg, si->src_reg,
9555 offsetof(struct bpf_sock_ops_kern, sk));
9556 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9557 offsetof(struct sock_common,
9561 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
9562 offsetof(struct bpf_sock_ops, remote_ip6[3]):
9563 #if IS_ENABLED(CONFIG_IPV6)
9564 BUILD_BUG_ON(sizeof_field(struct sock_common,
9565 skc_v6_daddr.s6_addr32[0]) != 4);
9568 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
9569 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9570 struct bpf_sock_ops_kern, sk),
9571 si->dst_reg, si->src_reg,
9572 offsetof(struct bpf_sock_ops_kern, sk));
9573 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9574 offsetof(struct sock_common,
9575 skc_v6_daddr.s6_addr32[0]) +
9578 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9582 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
9583 offsetof(struct bpf_sock_ops, local_ip6[3]):
9584 #if IS_ENABLED(CONFIG_IPV6)
9585 BUILD_BUG_ON(sizeof_field(struct sock_common,
9586 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9589 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
9590 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9591 struct bpf_sock_ops_kern, sk),
9592 si->dst_reg, si->src_reg,
9593 offsetof(struct bpf_sock_ops_kern, sk));
9594 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9595 offsetof(struct sock_common,
9596 skc_v6_rcv_saddr.s6_addr32[0]) +
9599 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9603 case offsetof(struct bpf_sock_ops, remote_port):
9604 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9606 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9607 struct bpf_sock_ops_kern, sk),
9608 si->dst_reg, si->src_reg,
9609 offsetof(struct bpf_sock_ops_kern, sk));
9610 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9611 offsetof(struct sock_common, skc_dport));
9612 #ifndef __BIG_ENDIAN_BITFIELD
9613 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9617 case offsetof(struct bpf_sock_ops, local_port):
9618 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
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_H, si->dst_reg, si->dst_reg,
9625 offsetof(struct sock_common, skc_num));
9628 case offsetof(struct bpf_sock_ops, is_fullsock):
9629 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9630 struct bpf_sock_ops_kern,
9632 si->dst_reg, si->src_reg,
9633 offsetof(struct bpf_sock_ops_kern,
9637 case offsetof(struct bpf_sock_ops, state):
9638 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
9640 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9641 struct bpf_sock_ops_kern, sk),
9642 si->dst_reg, si->src_reg,
9643 offsetof(struct bpf_sock_ops_kern, sk));
9644 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
9645 offsetof(struct sock_common, skc_state));
9648 case offsetof(struct bpf_sock_ops, rtt_min):
9649 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
9650 sizeof(struct minmax));
9651 BUILD_BUG_ON(sizeof(struct minmax) <
9652 sizeof(struct minmax_sample));
9654 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9655 struct bpf_sock_ops_kern, sk),
9656 si->dst_reg, si->src_reg,
9657 offsetof(struct bpf_sock_ops_kern, sk));
9658 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9659 offsetof(struct tcp_sock, rtt_min) +
9660 sizeof_field(struct minmax_sample, t));
9663 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
9664 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
9668 case offsetof(struct bpf_sock_ops, sk_txhash):
9669 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
9672 case offsetof(struct bpf_sock_ops, snd_cwnd):
9673 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
9675 case offsetof(struct bpf_sock_ops, srtt_us):
9676 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
9678 case offsetof(struct bpf_sock_ops, snd_ssthresh):
9679 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
9681 case offsetof(struct bpf_sock_ops, rcv_nxt):
9682 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
9684 case offsetof(struct bpf_sock_ops, snd_nxt):
9685 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
9687 case offsetof(struct bpf_sock_ops, snd_una):
9688 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
9690 case offsetof(struct bpf_sock_ops, mss_cache):
9691 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
9693 case offsetof(struct bpf_sock_ops, ecn_flags):
9694 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
9696 case offsetof(struct bpf_sock_ops, rate_delivered):
9697 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
9699 case offsetof(struct bpf_sock_ops, rate_interval_us):
9700 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
9702 case offsetof(struct bpf_sock_ops, packets_out):
9703 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
9705 case offsetof(struct bpf_sock_ops, retrans_out):
9706 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
9708 case offsetof(struct bpf_sock_ops, total_retrans):
9709 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
9711 case offsetof(struct bpf_sock_ops, segs_in):
9712 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
9714 case offsetof(struct bpf_sock_ops, data_segs_in):
9715 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
9717 case offsetof(struct bpf_sock_ops, segs_out):
9718 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
9720 case offsetof(struct bpf_sock_ops, data_segs_out):
9721 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
9723 case offsetof(struct bpf_sock_ops, lost_out):
9724 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
9726 case offsetof(struct bpf_sock_ops, sacked_out):
9727 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
9729 case offsetof(struct bpf_sock_ops, bytes_received):
9730 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
9732 case offsetof(struct bpf_sock_ops, bytes_acked):
9733 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
9735 case offsetof(struct bpf_sock_ops, sk):
9738 case offsetof(struct bpf_sock_ops, skb_data_end):
9739 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9741 si->dst_reg, si->src_reg,
9742 offsetof(struct bpf_sock_ops_kern,
9745 case offsetof(struct bpf_sock_ops, skb_data):
9746 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9748 si->dst_reg, si->src_reg,
9749 offsetof(struct bpf_sock_ops_kern,
9751 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9752 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9753 si->dst_reg, si->dst_reg,
9754 offsetof(struct sk_buff, data));
9756 case offsetof(struct bpf_sock_ops, skb_len):
9757 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9759 si->dst_reg, si->src_reg,
9760 offsetof(struct bpf_sock_ops_kern,
9762 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9763 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9764 si->dst_reg, si->dst_reg,
9765 offsetof(struct sk_buff, len));
9767 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9768 off = offsetof(struct sk_buff, cb);
9769 off += offsetof(struct tcp_skb_cb, tcp_flags);
9770 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
9771 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9773 si->dst_reg, si->src_reg,
9774 offsetof(struct bpf_sock_ops_kern,
9776 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9777 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
9779 si->dst_reg, si->dst_reg, off);
9782 return insn - insn_buf;
9785 /* data_end = skb->data + skb_headlen() */
9786 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
9787 struct bpf_insn *insn)
9790 int temp_reg_off = offsetof(struct sk_buff, cb) +
9791 offsetof(struct sk_skb_cb, temp_reg);
9793 if (si->src_reg == si->dst_reg) {
9794 /* We need an extra register, choose and save a register. */
9796 if (si->src_reg == reg || si->dst_reg == reg)
9798 if (si->src_reg == reg || si->dst_reg == reg)
9800 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off);
9805 /* reg = skb->data */
9806 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9808 offsetof(struct sk_buff, data));
9810 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9811 BPF_REG_AX, si->src_reg,
9812 offsetof(struct sk_buff, len));
9813 /* reg = skb->data + skb->len */
9814 *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX);
9815 /* AX = skb->data_len */
9816 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
9817 BPF_REG_AX, si->src_reg,
9818 offsetof(struct sk_buff, data_len));
9820 /* reg = skb->data + skb->len - skb->data_len */
9821 *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX);
9823 if (si->src_reg == si->dst_reg) {
9824 /* Restore the saved register */
9825 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg);
9826 *insn++ = BPF_MOV64_REG(si->dst_reg, reg);
9827 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off);
9833 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
9834 const struct bpf_insn *si,
9835 struct bpf_insn *insn_buf,
9836 struct bpf_prog *prog, u32 *target_size)
9838 struct bpf_insn *insn = insn_buf;
9842 case offsetof(struct __sk_buff, data_end):
9843 insn = bpf_convert_data_end_access(si, insn);
9845 case offsetof(struct __sk_buff, cb[0]) ...
9846 offsetofend(struct __sk_buff, cb[4]) - 1:
9847 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20);
9848 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
9849 offsetof(struct sk_skb_cb, data)) %
9852 prog->cb_access = 1;
9854 off -= offsetof(struct __sk_buff, cb[0]);
9855 off += offsetof(struct sk_buff, cb);
9856 off += offsetof(struct sk_skb_cb, data);
9857 if (type == BPF_WRITE)
9858 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
9861 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
9867 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9871 return insn - insn_buf;
9874 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
9875 const struct bpf_insn *si,
9876 struct bpf_insn *insn_buf,
9877 struct bpf_prog *prog, u32 *target_size)
9879 struct bpf_insn *insn = insn_buf;
9880 #if IS_ENABLED(CONFIG_IPV6)
9884 /* convert ctx uses the fact sg element is first in struct */
9885 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
9888 case offsetof(struct sk_msg_md, data):
9889 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
9890 si->dst_reg, si->src_reg,
9891 offsetof(struct sk_msg, data));
9893 case offsetof(struct sk_msg_md, data_end):
9894 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
9895 si->dst_reg, si->src_reg,
9896 offsetof(struct sk_msg, data_end));
9898 case offsetof(struct sk_msg_md, family):
9899 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9901 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9903 si->dst_reg, si->src_reg,
9904 offsetof(struct sk_msg, sk));
9905 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9906 offsetof(struct sock_common, skc_family));
9909 case offsetof(struct sk_msg_md, remote_ip4):
9910 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9912 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9914 si->dst_reg, si->src_reg,
9915 offsetof(struct sk_msg, sk));
9916 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9917 offsetof(struct sock_common, skc_daddr));
9920 case offsetof(struct sk_msg_md, local_ip4):
9921 BUILD_BUG_ON(sizeof_field(struct sock_common,
9922 skc_rcv_saddr) != 4);
9924 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9926 si->dst_reg, si->src_reg,
9927 offsetof(struct sk_msg, sk));
9928 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9929 offsetof(struct sock_common,
9933 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
9934 offsetof(struct sk_msg_md, remote_ip6[3]):
9935 #if IS_ENABLED(CONFIG_IPV6)
9936 BUILD_BUG_ON(sizeof_field(struct sock_common,
9937 skc_v6_daddr.s6_addr32[0]) != 4);
9940 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
9941 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9943 si->dst_reg, si->src_reg,
9944 offsetof(struct sk_msg, sk));
9945 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9946 offsetof(struct sock_common,
9947 skc_v6_daddr.s6_addr32[0]) +
9950 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9954 case offsetof(struct sk_msg_md, local_ip6[0]) ...
9955 offsetof(struct sk_msg_md, local_ip6[3]):
9956 #if IS_ENABLED(CONFIG_IPV6)
9957 BUILD_BUG_ON(sizeof_field(struct sock_common,
9958 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9961 off -= offsetof(struct sk_msg_md, local_ip6[0]);
9962 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9964 si->dst_reg, si->src_reg,
9965 offsetof(struct sk_msg, sk));
9966 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9967 offsetof(struct sock_common,
9968 skc_v6_rcv_saddr.s6_addr32[0]) +
9971 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9975 case offsetof(struct sk_msg_md, remote_port):
9976 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9978 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9980 si->dst_reg, si->src_reg,
9981 offsetof(struct sk_msg, sk));
9982 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9983 offsetof(struct sock_common, skc_dport));
9984 #ifndef __BIG_ENDIAN_BITFIELD
9985 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9989 case offsetof(struct sk_msg_md, local_port):
9990 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
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_H, si->dst_reg, si->dst_reg,
9997 offsetof(struct sock_common, skc_num));
10000 case offsetof(struct sk_msg_md, size):
10001 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
10002 si->dst_reg, si->src_reg,
10003 offsetof(struct sk_msg_sg, size));
10006 case offsetof(struct sk_msg_md, sk):
10007 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
10008 si->dst_reg, si->src_reg,
10009 offsetof(struct sk_msg, sk));
10013 return insn - insn_buf;
10016 const struct bpf_verifier_ops sk_filter_verifier_ops = {
10017 .get_func_proto = sk_filter_func_proto,
10018 .is_valid_access = sk_filter_is_valid_access,
10019 .convert_ctx_access = bpf_convert_ctx_access,
10020 .gen_ld_abs = bpf_gen_ld_abs,
10023 const struct bpf_prog_ops sk_filter_prog_ops = {
10024 .test_run = bpf_prog_test_run_skb,
10027 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
10028 .get_func_proto = tc_cls_act_func_proto,
10029 .is_valid_access = tc_cls_act_is_valid_access,
10030 .convert_ctx_access = tc_cls_act_convert_ctx_access,
10031 .gen_prologue = tc_cls_act_prologue,
10032 .gen_ld_abs = bpf_gen_ld_abs,
10033 .check_kfunc_call = bpf_prog_test_check_kfunc_call,
10036 const struct bpf_prog_ops tc_cls_act_prog_ops = {
10037 .test_run = bpf_prog_test_run_skb,
10040 const struct bpf_verifier_ops xdp_verifier_ops = {
10041 .get_func_proto = xdp_func_proto,
10042 .is_valid_access = xdp_is_valid_access,
10043 .convert_ctx_access = xdp_convert_ctx_access,
10044 .gen_prologue = bpf_noop_prologue,
10047 const struct bpf_prog_ops xdp_prog_ops = {
10048 .test_run = bpf_prog_test_run_xdp,
10051 const struct bpf_verifier_ops cg_skb_verifier_ops = {
10052 .get_func_proto = cg_skb_func_proto,
10053 .is_valid_access = cg_skb_is_valid_access,
10054 .convert_ctx_access = bpf_convert_ctx_access,
10057 const struct bpf_prog_ops cg_skb_prog_ops = {
10058 .test_run = bpf_prog_test_run_skb,
10061 const struct bpf_verifier_ops lwt_in_verifier_ops = {
10062 .get_func_proto = lwt_in_func_proto,
10063 .is_valid_access = lwt_is_valid_access,
10064 .convert_ctx_access = bpf_convert_ctx_access,
10067 const struct bpf_prog_ops lwt_in_prog_ops = {
10068 .test_run = bpf_prog_test_run_skb,
10071 const struct bpf_verifier_ops lwt_out_verifier_ops = {
10072 .get_func_proto = lwt_out_func_proto,
10073 .is_valid_access = lwt_is_valid_access,
10074 .convert_ctx_access = bpf_convert_ctx_access,
10077 const struct bpf_prog_ops lwt_out_prog_ops = {
10078 .test_run = bpf_prog_test_run_skb,
10081 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
10082 .get_func_proto = lwt_xmit_func_proto,
10083 .is_valid_access = lwt_is_valid_access,
10084 .convert_ctx_access = bpf_convert_ctx_access,
10085 .gen_prologue = tc_cls_act_prologue,
10088 const struct bpf_prog_ops lwt_xmit_prog_ops = {
10089 .test_run = bpf_prog_test_run_skb,
10092 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
10093 .get_func_proto = lwt_seg6local_func_proto,
10094 .is_valid_access = lwt_is_valid_access,
10095 .convert_ctx_access = bpf_convert_ctx_access,
10098 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
10099 .test_run = bpf_prog_test_run_skb,
10102 const struct bpf_verifier_ops cg_sock_verifier_ops = {
10103 .get_func_proto = sock_filter_func_proto,
10104 .is_valid_access = sock_filter_is_valid_access,
10105 .convert_ctx_access = bpf_sock_convert_ctx_access,
10108 const struct bpf_prog_ops cg_sock_prog_ops = {
10111 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
10112 .get_func_proto = sock_addr_func_proto,
10113 .is_valid_access = sock_addr_is_valid_access,
10114 .convert_ctx_access = sock_addr_convert_ctx_access,
10117 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
10120 const struct bpf_verifier_ops sock_ops_verifier_ops = {
10121 .get_func_proto = sock_ops_func_proto,
10122 .is_valid_access = sock_ops_is_valid_access,
10123 .convert_ctx_access = sock_ops_convert_ctx_access,
10126 const struct bpf_prog_ops sock_ops_prog_ops = {
10129 const struct bpf_verifier_ops sk_skb_verifier_ops = {
10130 .get_func_proto = sk_skb_func_proto,
10131 .is_valid_access = sk_skb_is_valid_access,
10132 .convert_ctx_access = sk_skb_convert_ctx_access,
10133 .gen_prologue = sk_skb_prologue,
10136 const struct bpf_prog_ops sk_skb_prog_ops = {
10139 const struct bpf_verifier_ops sk_msg_verifier_ops = {
10140 .get_func_proto = sk_msg_func_proto,
10141 .is_valid_access = sk_msg_is_valid_access,
10142 .convert_ctx_access = sk_msg_convert_ctx_access,
10143 .gen_prologue = bpf_noop_prologue,
10146 const struct bpf_prog_ops sk_msg_prog_ops = {
10149 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
10150 .get_func_proto = flow_dissector_func_proto,
10151 .is_valid_access = flow_dissector_is_valid_access,
10152 .convert_ctx_access = flow_dissector_convert_ctx_access,
10155 const struct bpf_prog_ops flow_dissector_prog_ops = {
10156 .test_run = bpf_prog_test_run_flow_dissector,
10159 int sk_detach_filter(struct sock *sk)
10162 struct sk_filter *filter;
10164 if (sock_flag(sk, SOCK_FILTER_LOCKED))
10167 filter = rcu_dereference_protected(sk->sk_filter,
10168 lockdep_sock_is_held(sk));
10170 RCU_INIT_POINTER(sk->sk_filter, NULL);
10171 sk_filter_uncharge(sk, filter);
10177 EXPORT_SYMBOL_GPL(sk_detach_filter);
10179 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
10182 struct sock_fprog_kern *fprog;
10183 struct sk_filter *filter;
10187 filter = rcu_dereference_protected(sk->sk_filter,
10188 lockdep_sock_is_held(sk));
10192 /* We're copying the filter that has been originally attached,
10193 * so no conversion/decode needed anymore. eBPF programs that
10194 * have no original program cannot be dumped through this.
10197 fprog = filter->prog->orig_prog;
10203 /* User space only enquires number of filter blocks. */
10207 if (len < fprog->len)
10211 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
10214 /* Instead of bytes, the API requests to return the number
10215 * of filter blocks.
10224 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
10225 struct sock_reuseport *reuse,
10226 struct sock *sk, struct sk_buff *skb,
10227 struct sock *migrating_sk,
10230 reuse_kern->skb = skb;
10231 reuse_kern->sk = sk;
10232 reuse_kern->selected_sk = NULL;
10233 reuse_kern->migrating_sk = migrating_sk;
10234 reuse_kern->data_end = skb->data + skb_headlen(skb);
10235 reuse_kern->hash = hash;
10236 reuse_kern->reuseport_id = reuse->reuseport_id;
10237 reuse_kern->bind_inany = reuse->bind_inany;
10240 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
10241 struct bpf_prog *prog, struct sk_buff *skb,
10242 struct sock *migrating_sk,
10245 struct sk_reuseport_kern reuse_kern;
10246 enum sk_action action;
10248 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
10249 action = bpf_prog_run(prog, &reuse_kern);
10251 if (action == SK_PASS)
10252 return reuse_kern.selected_sk;
10254 return ERR_PTR(-ECONNREFUSED);
10257 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
10258 struct bpf_map *, map, void *, key, u32, flags)
10260 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
10261 struct sock_reuseport *reuse;
10262 struct sock *selected_sk;
10264 selected_sk = map->ops->map_lookup_elem(map, key);
10268 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
10270 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
10271 if (sk_is_refcounted(selected_sk))
10272 sock_put(selected_sk);
10274 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
10275 * The only (!reuse) case here is - the sk has already been
10276 * unhashed (e.g. by close()), so treat it as -ENOENT.
10278 * Other maps (e.g. sock_map) do not provide this guarantee and
10279 * the sk may never be in the reuseport group to begin with.
10281 return is_sockarray ? -ENOENT : -EINVAL;
10284 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
10285 struct sock *sk = reuse_kern->sk;
10287 if (sk->sk_protocol != selected_sk->sk_protocol)
10288 return -EPROTOTYPE;
10289 else if (sk->sk_family != selected_sk->sk_family)
10290 return -EAFNOSUPPORT;
10292 /* Catch all. Likely bound to a different sockaddr. */
10296 reuse_kern->selected_sk = selected_sk;
10301 static const struct bpf_func_proto sk_select_reuseport_proto = {
10302 .func = sk_select_reuseport,
10304 .ret_type = RET_INTEGER,
10305 .arg1_type = ARG_PTR_TO_CTX,
10306 .arg2_type = ARG_CONST_MAP_PTR,
10307 .arg3_type = ARG_PTR_TO_MAP_KEY,
10308 .arg4_type = ARG_ANYTHING,
10311 BPF_CALL_4(sk_reuseport_load_bytes,
10312 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10313 void *, to, u32, len)
10315 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
10318 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
10319 .func = sk_reuseport_load_bytes,
10321 .ret_type = RET_INTEGER,
10322 .arg1_type = ARG_PTR_TO_CTX,
10323 .arg2_type = ARG_ANYTHING,
10324 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
10325 .arg4_type = ARG_CONST_SIZE,
10328 BPF_CALL_5(sk_reuseport_load_bytes_relative,
10329 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10330 void *, to, u32, len, u32, start_header)
10332 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
10333 len, start_header);
10336 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
10337 .func = sk_reuseport_load_bytes_relative,
10339 .ret_type = RET_INTEGER,
10340 .arg1_type = ARG_PTR_TO_CTX,
10341 .arg2_type = ARG_ANYTHING,
10342 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
10343 .arg4_type = ARG_CONST_SIZE,
10344 .arg5_type = ARG_ANYTHING,
10347 static const struct bpf_func_proto *
10348 sk_reuseport_func_proto(enum bpf_func_id func_id,
10349 const struct bpf_prog *prog)
10352 case BPF_FUNC_sk_select_reuseport:
10353 return &sk_select_reuseport_proto;
10354 case BPF_FUNC_skb_load_bytes:
10355 return &sk_reuseport_load_bytes_proto;
10356 case BPF_FUNC_skb_load_bytes_relative:
10357 return &sk_reuseport_load_bytes_relative_proto;
10358 case BPF_FUNC_get_socket_cookie:
10359 return &bpf_get_socket_ptr_cookie_proto;
10360 case BPF_FUNC_ktime_get_coarse_ns:
10361 return &bpf_ktime_get_coarse_ns_proto;
10363 return bpf_base_func_proto(func_id);
10368 sk_reuseport_is_valid_access(int off, int size,
10369 enum bpf_access_type type,
10370 const struct bpf_prog *prog,
10371 struct bpf_insn_access_aux *info)
10373 const u32 size_default = sizeof(__u32);
10375 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
10376 off % size || type != BPF_READ)
10380 case offsetof(struct sk_reuseport_md, data):
10381 info->reg_type = PTR_TO_PACKET;
10382 return size == sizeof(__u64);
10384 case offsetof(struct sk_reuseport_md, data_end):
10385 info->reg_type = PTR_TO_PACKET_END;
10386 return size == sizeof(__u64);
10388 case offsetof(struct sk_reuseport_md, hash):
10389 return size == size_default;
10391 case offsetof(struct sk_reuseport_md, sk):
10392 info->reg_type = PTR_TO_SOCKET;
10393 return size == sizeof(__u64);
10395 case offsetof(struct sk_reuseport_md, migrating_sk):
10396 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
10397 return size == sizeof(__u64);
10399 /* Fields that allow narrowing */
10400 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
10401 if (size < sizeof_field(struct sk_buff, protocol))
10404 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
10405 case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
10406 case bpf_ctx_range(struct sk_reuseport_md, len):
10407 bpf_ctx_record_field_size(info, size_default);
10408 return bpf_ctx_narrow_access_ok(off, size, size_default);
10415 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
10416 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
10417 si->dst_reg, si->src_reg, \
10418 bpf_target_off(struct sk_reuseport_kern, F, \
10419 sizeof_field(struct sk_reuseport_kern, F), \
10423 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
10424 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
10429 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \
10430 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
10435 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
10436 const struct bpf_insn *si,
10437 struct bpf_insn *insn_buf,
10438 struct bpf_prog *prog,
10441 struct bpf_insn *insn = insn_buf;
10444 case offsetof(struct sk_reuseport_md, data):
10445 SK_REUSEPORT_LOAD_SKB_FIELD(data);
10448 case offsetof(struct sk_reuseport_md, len):
10449 SK_REUSEPORT_LOAD_SKB_FIELD(len);
10452 case offsetof(struct sk_reuseport_md, eth_protocol):
10453 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
10456 case offsetof(struct sk_reuseport_md, ip_protocol):
10457 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
10460 case offsetof(struct sk_reuseport_md, data_end):
10461 SK_REUSEPORT_LOAD_FIELD(data_end);
10464 case offsetof(struct sk_reuseport_md, hash):
10465 SK_REUSEPORT_LOAD_FIELD(hash);
10468 case offsetof(struct sk_reuseport_md, bind_inany):
10469 SK_REUSEPORT_LOAD_FIELD(bind_inany);
10472 case offsetof(struct sk_reuseport_md, sk):
10473 SK_REUSEPORT_LOAD_FIELD(sk);
10476 case offsetof(struct sk_reuseport_md, migrating_sk):
10477 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
10481 return insn - insn_buf;
10484 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
10485 .get_func_proto = sk_reuseport_func_proto,
10486 .is_valid_access = sk_reuseport_is_valid_access,
10487 .convert_ctx_access = sk_reuseport_convert_ctx_access,
10490 const struct bpf_prog_ops sk_reuseport_prog_ops = {
10493 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
10494 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
10496 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
10497 struct sock *, sk, u64, flags)
10499 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
10500 BPF_SK_LOOKUP_F_NO_REUSEPORT)))
10502 if (unlikely(sk && sk_is_refcounted(sk)))
10503 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
10504 if (unlikely(sk && sk->sk_state == TCP_ESTABLISHED))
10505 return -ESOCKTNOSUPPORT; /* reject connected sockets */
10507 /* Check if socket is suitable for packet L3/L4 protocol */
10508 if (sk && sk->sk_protocol != ctx->protocol)
10509 return -EPROTOTYPE;
10510 if (sk && sk->sk_family != ctx->family &&
10511 (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
10512 return -EAFNOSUPPORT;
10514 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
10517 /* Select socket as lookup result */
10518 ctx->selected_sk = sk;
10519 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
10523 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
10524 .func = bpf_sk_lookup_assign,
10526 .ret_type = RET_INTEGER,
10527 .arg1_type = ARG_PTR_TO_CTX,
10528 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL,
10529 .arg3_type = ARG_ANYTHING,
10532 static const struct bpf_func_proto *
10533 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
10536 case BPF_FUNC_perf_event_output:
10537 return &bpf_event_output_data_proto;
10538 case BPF_FUNC_sk_assign:
10539 return &bpf_sk_lookup_assign_proto;
10540 case BPF_FUNC_sk_release:
10541 return &bpf_sk_release_proto;
10543 return bpf_sk_base_func_proto(func_id);
10547 static bool sk_lookup_is_valid_access(int off, int size,
10548 enum bpf_access_type type,
10549 const struct bpf_prog *prog,
10550 struct bpf_insn_access_aux *info)
10552 if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
10554 if (off % size != 0)
10556 if (type != BPF_READ)
10560 case offsetof(struct bpf_sk_lookup, sk):
10561 info->reg_type = PTR_TO_SOCKET_OR_NULL;
10562 return size == sizeof(__u64);
10564 case bpf_ctx_range(struct bpf_sk_lookup, family):
10565 case bpf_ctx_range(struct bpf_sk_lookup, protocol):
10566 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
10567 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
10568 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
10569 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
10570 case offsetof(struct bpf_sk_lookup, remote_port) ...
10571 offsetof(struct bpf_sk_lookup, local_ip4) - 1:
10572 case bpf_ctx_range(struct bpf_sk_lookup, local_port):
10573 bpf_ctx_record_field_size(info, sizeof(__u32));
10574 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
10581 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
10582 const struct bpf_insn *si,
10583 struct bpf_insn *insn_buf,
10584 struct bpf_prog *prog,
10587 struct bpf_insn *insn = insn_buf;
10590 case offsetof(struct bpf_sk_lookup, sk):
10591 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10592 offsetof(struct bpf_sk_lookup_kern, selected_sk));
10595 case offsetof(struct bpf_sk_lookup, family):
10596 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10597 bpf_target_off(struct bpf_sk_lookup_kern,
10598 family, 2, target_size));
10601 case offsetof(struct bpf_sk_lookup, protocol):
10602 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10603 bpf_target_off(struct bpf_sk_lookup_kern,
10604 protocol, 2, target_size));
10607 case offsetof(struct bpf_sk_lookup, remote_ip4):
10608 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10609 bpf_target_off(struct bpf_sk_lookup_kern,
10610 v4.saddr, 4, target_size));
10613 case offsetof(struct bpf_sk_lookup, local_ip4):
10614 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10615 bpf_target_off(struct bpf_sk_lookup_kern,
10616 v4.daddr, 4, target_size));
10619 case bpf_ctx_range_till(struct bpf_sk_lookup,
10620 remote_ip6[0], remote_ip6[3]): {
10621 #if IS_ENABLED(CONFIG_IPV6)
10624 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
10625 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10626 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10627 offsetof(struct bpf_sk_lookup_kern, v6.saddr));
10628 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10629 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10631 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10635 case bpf_ctx_range_till(struct bpf_sk_lookup,
10636 local_ip6[0], local_ip6[3]): {
10637 #if IS_ENABLED(CONFIG_IPV6)
10640 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
10641 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10642 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10643 offsetof(struct bpf_sk_lookup_kern, v6.daddr));
10644 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10645 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10647 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10651 case offsetof(struct bpf_sk_lookup, remote_port):
10652 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10653 bpf_target_off(struct bpf_sk_lookup_kern,
10654 sport, 2, target_size));
10657 case offsetof(struct bpf_sk_lookup, local_port):
10658 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10659 bpf_target_off(struct bpf_sk_lookup_kern,
10660 dport, 2, target_size));
10664 return insn - insn_buf;
10667 const struct bpf_prog_ops sk_lookup_prog_ops = {
10668 .test_run = bpf_prog_test_run_sk_lookup,
10671 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
10672 .get_func_proto = sk_lookup_func_proto,
10673 .is_valid_access = sk_lookup_is_valid_access,
10674 .convert_ctx_access = sk_lookup_convert_ctx_access,
10677 #endif /* CONFIG_INET */
10679 DEFINE_BPF_DISPATCHER(xdp)
10681 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
10683 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
10686 #ifdef CONFIG_DEBUG_INFO_BTF
10687 BTF_ID_LIST_GLOBAL(btf_sock_ids)
10688 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
10690 #undef BTF_SOCK_TYPE
10692 u32 btf_sock_ids[MAX_BTF_SOCK_TYPE];
10695 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
10697 /* tcp6_sock type is not generated in dwarf and hence btf,
10698 * trigger an explicit type generation here.
10700 BTF_TYPE_EMIT(struct tcp6_sock);
10701 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
10702 sk->sk_family == AF_INET6)
10703 return (unsigned long)sk;
10705 return (unsigned long)NULL;
10708 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
10709 .func = bpf_skc_to_tcp6_sock,
10711 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10712 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10713 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
10716 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
10718 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
10719 return (unsigned long)sk;
10721 return (unsigned long)NULL;
10724 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
10725 .func = bpf_skc_to_tcp_sock,
10727 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10728 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10729 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
10732 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
10734 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
10735 * generated if CONFIG_INET=n. Trigger an explicit generation here.
10737 BTF_TYPE_EMIT(struct inet_timewait_sock);
10738 BTF_TYPE_EMIT(struct tcp_timewait_sock);
10741 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
10742 return (unsigned long)sk;
10745 #if IS_BUILTIN(CONFIG_IPV6)
10746 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
10747 return (unsigned long)sk;
10750 return (unsigned long)NULL;
10753 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
10754 .func = bpf_skc_to_tcp_timewait_sock,
10756 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10757 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10758 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
10761 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
10764 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10765 return (unsigned long)sk;
10768 #if IS_BUILTIN(CONFIG_IPV6)
10769 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10770 return (unsigned long)sk;
10773 return (unsigned long)NULL;
10776 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
10777 .func = bpf_skc_to_tcp_request_sock,
10779 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10780 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10781 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
10784 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
10786 /* udp6_sock type is not generated in dwarf and hence btf,
10787 * trigger an explicit type generation here.
10789 BTF_TYPE_EMIT(struct udp6_sock);
10790 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
10791 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
10792 return (unsigned long)sk;
10794 return (unsigned long)NULL;
10797 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
10798 .func = bpf_skc_to_udp6_sock,
10800 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10801 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10802 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
10805 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
10807 return (unsigned long)sock_from_file(file);
10810 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
10811 BTF_ID(struct, socket)
10812 BTF_ID(struct, file)
10814 const struct bpf_func_proto bpf_sock_from_file_proto = {
10815 .func = bpf_sock_from_file,
10817 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10818 .ret_btf_id = &bpf_sock_from_file_btf_ids[0],
10819 .arg1_type = ARG_PTR_TO_BTF_ID,
10820 .arg1_btf_id = &bpf_sock_from_file_btf_ids[1],
10823 static const struct bpf_func_proto *
10824 bpf_sk_base_func_proto(enum bpf_func_id func_id)
10826 const struct bpf_func_proto *func;
10829 case BPF_FUNC_skc_to_tcp6_sock:
10830 func = &bpf_skc_to_tcp6_sock_proto;
10832 case BPF_FUNC_skc_to_tcp_sock:
10833 func = &bpf_skc_to_tcp_sock_proto;
10835 case BPF_FUNC_skc_to_tcp_timewait_sock:
10836 func = &bpf_skc_to_tcp_timewait_sock_proto;
10838 case BPF_FUNC_skc_to_tcp_request_sock:
10839 func = &bpf_skc_to_tcp_request_sock_proto;
10841 case BPF_FUNC_skc_to_udp6_sock:
10842 func = &bpf_skc_to_udp6_sock_proto;
10844 case BPF_FUNC_ktime_get_coarse_ns:
10845 return &bpf_ktime_get_coarse_ns_proto;
10847 return bpf_base_func_proto(func_id);
10850 if (!perfmon_capable())