1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Linux Socket Filter - Kernel level socket filtering
5 * Based on the design of the Berkeley Packet Filter. The new
6 * internal format has been designed by PLUMgrid:
8 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
12 * Jay Schulist <jschlst@samba.org>
13 * Alexei Starovoitov <ast@plumgrid.com>
14 * Daniel Borkmann <dborkman@redhat.com>
16 * Andi Kleen - Fix a few bad bugs and races.
17 * Kris Katterjohn - Added many additional checks in bpf_check_classic()
20 #include <linux/atomic.h>
21 #include <linux/module.h>
22 #include <linux/types.h>
24 #include <linux/fcntl.h>
25 #include <linux/socket.h>
26 #include <linux/sock_diag.h>
28 #include <linux/inet.h>
29 #include <linux/netdevice.h>
30 #include <linux/if_packet.h>
31 #include <linux/if_arp.h>
32 #include <linux/gfp.h>
33 #include <net/inet_common.h>
35 #include <net/protocol.h>
36 #include <net/netlink.h>
37 #include <linux/skbuff.h>
38 #include <linux/skmsg.h>
40 #include <net/flow_dissector.h>
41 #include <linux/errno.h>
42 #include <linux/timer.h>
43 #include <linux/uaccess.h>
44 #include <asm/unaligned.h>
45 #include <linux/filter.h>
46 #include <linux/ratelimit.h>
47 #include <linux/seccomp.h>
48 #include <linux/if_vlan.h>
49 #include <linux/bpf.h>
50 #include <linux/btf.h>
51 #include <net/sch_generic.h>
52 #include <net/cls_cgroup.h>
53 #include <net/dst_metadata.h>
55 #include <net/sock_reuseport.h>
56 #include <net/busy_poll.h>
60 #include <linux/bpf_trace.h>
61 #include <net/xdp_sock.h>
62 #include <linux/inetdevice.h>
63 #include <net/inet_hashtables.h>
64 #include <net/inet6_hashtables.h>
65 #include <net/ip_fib.h>
66 #include <net/nexthop.h>
70 #include <net/net_namespace.h>
71 #include <linux/seg6_local.h>
73 #include <net/seg6_local.h>
74 #include <net/lwtunnel.h>
75 #include <net/ipv6_stubs.h>
76 #include <net/bpf_sk_storage.h>
77 #include <net/transp_v6.h>
78 #include <linux/btf_ids.h>
82 static const struct bpf_func_proto *
83 bpf_sk_base_func_proto(enum bpf_func_id func_id);
85 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len)
87 if (in_compat_syscall()) {
88 struct compat_sock_fprog f32;
90 if (len != sizeof(f32))
92 if (copy_from_sockptr(&f32, src, sizeof(f32)))
94 memset(dst, 0, sizeof(*dst));
96 dst->filter = compat_ptr(f32.filter);
98 if (len != sizeof(*dst))
100 if (copy_from_sockptr(dst, src, sizeof(*dst)))
106 EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user);
109 * sk_filter_trim_cap - run a packet through a socket filter
110 * @sk: sock associated with &sk_buff
111 * @skb: buffer to filter
112 * @cap: limit on how short the eBPF program may trim the packet
114 * Run the eBPF program and then cut skb->data to correct size returned by
115 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
116 * than pkt_len we keep whole skb->data. This is the socket level
117 * wrapper to bpf_prog_run. It returns 0 if the packet should
118 * be accepted or -EPERM if the packet should be tossed.
121 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
124 struct sk_filter *filter;
127 * If the skb was allocated from pfmemalloc reserves, only
128 * allow SOCK_MEMALLOC sockets to use it as this socket is
129 * helping free memory
131 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
132 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
135 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
139 err = security_sock_rcv_skb(sk, skb);
144 filter = rcu_dereference(sk->sk_filter);
146 struct sock *save_sk = skb->sk;
147 unsigned int pkt_len;
150 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
152 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
158 EXPORT_SYMBOL(sk_filter_trim_cap);
160 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
162 return skb_get_poff(skb);
165 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
169 if (skb_is_nonlinear(skb))
172 if (skb->len < sizeof(struct nlattr))
175 if (a > skb->len - sizeof(struct nlattr))
178 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
180 return (void *) nla - (void *) skb->data;
185 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
189 if (skb_is_nonlinear(skb))
192 if (skb->len < sizeof(struct nlattr))
195 if (a > skb->len - sizeof(struct nlattr))
198 nla = (struct nlattr *) &skb->data[a];
199 if (nla->nla_len > skb->len - a)
202 nla = nla_find_nested(nla, x);
204 return (void *) nla - (void *) skb->data;
209 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
210 data, int, headlen, int, offset)
213 const int len = sizeof(tmp);
216 if (headlen - offset >= len)
217 return *(u8 *)(data + offset);
218 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
221 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
229 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
232 return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
236 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
237 data, int, headlen, int, offset)
240 const int len = sizeof(tmp);
243 if (headlen - offset >= len)
244 return get_unaligned_be16(data + offset);
245 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
246 return be16_to_cpu(tmp);
248 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
250 return get_unaligned_be16(ptr);
256 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
259 return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
263 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
264 data, int, headlen, int, offset)
267 const int len = sizeof(tmp);
269 if (likely(offset >= 0)) {
270 if (headlen - offset >= len)
271 return get_unaligned_be32(data + offset);
272 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
273 return be32_to_cpu(tmp);
275 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
277 return get_unaligned_be32(ptr);
283 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
286 return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
290 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
291 struct bpf_insn *insn_buf)
293 struct bpf_insn *insn = insn_buf;
297 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4);
299 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
300 offsetof(struct sk_buff, mark));
304 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
305 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
306 #ifdef __BIG_ENDIAN_BITFIELD
307 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
312 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2);
314 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
315 offsetof(struct sk_buff, queue_mapping));
318 case SKF_AD_VLAN_TAG:
319 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2);
321 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
322 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
323 offsetof(struct sk_buff, vlan_tci));
325 case SKF_AD_VLAN_TAG_PRESENT:
326 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_VLAN_PRESENT_OFFSET());
327 if (PKT_VLAN_PRESENT_BIT)
328 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, PKT_VLAN_PRESENT_BIT);
329 if (PKT_VLAN_PRESENT_BIT < 7)
330 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
334 return insn - insn_buf;
337 static bool convert_bpf_extensions(struct sock_filter *fp,
338 struct bpf_insn **insnp)
340 struct bpf_insn *insn = *insnp;
344 case SKF_AD_OFF + SKF_AD_PROTOCOL:
345 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2);
347 /* A = *(u16 *) (CTX + offsetof(protocol)) */
348 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
349 offsetof(struct sk_buff, protocol));
350 /* A = ntohs(A) [emitting a nop or swap16] */
351 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
354 case SKF_AD_OFF + SKF_AD_PKTTYPE:
355 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
359 case SKF_AD_OFF + SKF_AD_IFINDEX:
360 case SKF_AD_OFF + SKF_AD_HATYPE:
361 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4);
362 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2);
364 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
365 BPF_REG_TMP, BPF_REG_CTX,
366 offsetof(struct sk_buff, dev));
367 /* if (tmp != 0) goto pc + 1 */
368 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
369 *insn++ = BPF_EXIT_INSN();
370 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
371 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
372 offsetof(struct net_device, ifindex));
374 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
375 offsetof(struct net_device, type));
378 case SKF_AD_OFF + SKF_AD_MARK:
379 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
383 case SKF_AD_OFF + SKF_AD_RXHASH:
384 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4);
386 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
387 offsetof(struct sk_buff, hash));
390 case SKF_AD_OFF + SKF_AD_QUEUE:
391 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
395 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
396 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
397 BPF_REG_A, BPF_REG_CTX, insn);
401 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
402 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
403 BPF_REG_A, BPF_REG_CTX, insn);
407 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
408 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2);
410 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
411 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
412 offsetof(struct sk_buff, vlan_proto));
413 /* A = ntohs(A) [emitting a nop or swap16] */
414 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
417 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
418 case SKF_AD_OFF + SKF_AD_NLATTR:
419 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
420 case SKF_AD_OFF + SKF_AD_CPU:
421 case SKF_AD_OFF + SKF_AD_RANDOM:
423 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
425 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
427 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
428 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
430 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
431 *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
433 case SKF_AD_OFF + SKF_AD_NLATTR:
434 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
436 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
437 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
439 case SKF_AD_OFF + SKF_AD_CPU:
440 *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
442 case SKF_AD_OFF + SKF_AD_RANDOM:
443 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
444 bpf_user_rnd_init_once();
449 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
451 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
455 /* This is just a dummy call to avoid letting the compiler
456 * evict __bpf_call_base() as an optimization. Placed here
457 * where no-one bothers.
459 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
467 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
469 const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
470 int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
471 bool endian = BPF_SIZE(fp->code) == BPF_H ||
472 BPF_SIZE(fp->code) == BPF_W;
473 bool indirect = BPF_MODE(fp->code) == BPF_IND;
474 const int ip_align = NET_IP_ALIGN;
475 struct bpf_insn *insn = *insnp;
479 ((unaligned_ok && offset >= 0) ||
480 (!unaligned_ok && offset >= 0 &&
481 offset + ip_align >= 0 &&
482 offset + ip_align % size == 0))) {
483 bool ldx_off_ok = offset <= S16_MAX;
485 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
487 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
488 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
489 size, 2 + endian + (!ldx_off_ok * 2));
491 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
494 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
495 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
496 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
500 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
501 *insn++ = BPF_JMP_A(8);
504 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
505 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
506 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
508 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
510 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
512 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
515 switch (BPF_SIZE(fp->code)) {
517 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
520 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
523 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
529 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
530 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
531 *insn = BPF_EXIT_INSN();
538 * bpf_convert_filter - convert filter program
539 * @prog: the user passed filter program
540 * @len: the length of the user passed filter program
541 * @new_prog: allocated 'struct bpf_prog' or NULL
542 * @new_len: pointer to store length of converted program
543 * @seen_ld_abs: bool whether we've seen ld_abs/ind
545 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
546 * style extended BPF (eBPF).
547 * Conversion workflow:
549 * 1) First pass for calculating the new program length:
550 * bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
552 * 2) 2nd pass to remap in two passes: 1st pass finds new
553 * jump offsets, 2nd pass remapping:
554 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
556 static int bpf_convert_filter(struct sock_filter *prog, int len,
557 struct bpf_prog *new_prog, int *new_len,
560 int new_flen = 0, pass = 0, target, i, stack_off;
561 struct bpf_insn *new_insn, *first_insn = NULL;
562 struct sock_filter *fp;
566 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
567 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
569 if (len <= 0 || len > BPF_MAXINSNS)
573 first_insn = new_prog->insnsi;
574 addrs = kcalloc(len, sizeof(*addrs),
575 GFP_KERNEL | __GFP_NOWARN);
581 new_insn = first_insn;
584 /* Classic BPF related prologue emission. */
586 /* Classic BPF expects A and X to be reset first. These need
587 * to be guaranteed to be the first two instructions.
589 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
590 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
592 /* All programs must keep CTX in callee saved BPF_REG_CTX.
593 * In eBPF case it's done by the compiler, here we need to
594 * do this ourself. Initial CTX is present in BPF_REG_ARG1.
596 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
598 /* For packet access in classic BPF, cache skb->data
599 * in callee-saved BPF R8 and skb->len - skb->data_len
600 * (headlen) in BPF R9. Since classic BPF is read-only
601 * on CTX, we only need to cache it once.
603 *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
604 BPF_REG_D, BPF_REG_CTX,
605 offsetof(struct sk_buff, data));
606 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
607 offsetof(struct sk_buff, len));
608 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
609 offsetof(struct sk_buff, data_len));
610 *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
616 for (i = 0; i < len; fp++, i++) {
617 struct bpf_insn tmp_insns[32] = { };
618 struct bpf_insn *insn = tmp_insns;
621 addrs[i] = new_insn - first_insn;
624 /* All arithmetic insns and skb loads map as-is. */
625 case BPF_ALU | BPF_ADD | BPF_X:
626 case BPF_ALU | BPF_ADD | BPF_K:
627 case BPF_ALU | BPF_SUB | BPF_X:
628 case BPF_ALU | BPF_SUB | BPF_K:
629 case BPF_ALU | BPF_AND | BPF_X:
630 case BPF_ALU | BPF_AND | BPF_K:
631 case BPF_ALU | BPF_OR | BPF_X:
632 case BPF_ALU | BPF_OR | BPF_K:
633 case BPF_ALU | BPF_LSH | BPF_X:
634 case BPF_ALU | BPF_LSH | BPF_K:
635 case BPF_ALU | BPF_RSH | BPF_X:
636 case BPF_ALU | BPF_RSH | BPF_K:
637 case BPF_ALU | BPF_XOR | BPF_X:
638 case BPF_ALU | BPF_XOR | BPF_K:
639 case BPF_ALU | BPF_MUL | BPF_X:
640 case BPF_ALU | BPF_MUL | BPF_K:
641 case BPF_ALU | BPF_DIV | BPF_X:
642 case BPF_ALU | BPF_DIV | BPF_K:
643 case BPF_ALU | BPF_MOD | BPF_X:
644 case BPF_ALU | BPF_MOD | BPF_K:
645 case BPF_ALU | BPF_NEG:
646 case BPF_LD | BPF_ABS | BPF_W:
647 case BPF_LD | BPF_ABS | BPF_H:
648 case BPF_LD | BPF_ABS | BPF_B:
649 case BPF_LD | BPF_IND | BPF_W:
650 case BPF_LD | BPF_IND | BPF_H:
651 case BPF_LD | BPF_IND | BPF_B:
652 /* Check for overloaded BPF extension and
653 * directly convert it if found, otherwise
654 * just move on with mapping.
656 if (BPF_CLASS(fp->code) == BPF_LD &&
657 BPF_MODE(fp->code) == BPF_ABS &&
658 convert_bpf_extensions(fp, &insn))
660 if (BPF_CLASS(fp->code) == BPF_LD &&
661 convert_bpf_ld_abs(fp, &insn)) {
666 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
667 fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
668 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
669 /* Error with exception code on div/mod by 0.
670 * For cBPF programs, this was always return 0.
672 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
673 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
674 *insn++ = BPF_EXIT_INSN();
677 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
680 /* Jump transformation cannot use BPF block macros
681 * everywhere as offset calculation and target updates
682 * require a bit more work than the rest, i.e. jump
683 * opcodes map as-is, but offsets need adjustment.
686 #define BPF_EMIT_JMP \
688 const s32 off_min = S16_MIN, off_max = S16_MAX; \
691 if (target >= len || target < 0) \
693 off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
694 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
695 off -= insn - tmp_insns; \
696 /* Reject anything not fitting into insn->off. */ \
697 if (off < off_min || off > off_max) \
702 case BPF_JMP | BPF_JA:
703 target = i + fp->k + 1;
704 insn->code = fp->code;
708 case BPF_JMP | BPF_JEQ | BPF_K:
709 case BPF_JMP | BPF_JEQ | BPF_X:
710 case BPF_JMP | BPF_JSET | BPF_K:
711 case BPF_JMP | BPF_JSET | BPF_X:
712 case BPF_JMP | BPF_JGT | BPF_K:
713 case BPF_JMP | BPF_JGT | BPF_X:
714 case BPF_JMP | BPF_JGE | BPF_K:
715 case BPF_JMP | BPF_JGE | BPF_X:
716 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
717 /* BPF immediates are signed, zero extend
718 * immediate into tmp register and use it
721 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
723 insn->dst_reg = BPF_REG_A;
724 insn->src_reg = BPF_REG_TMP;
727 insn->dst_reg = BPF_REG_A;
729 bpf_src = BPF_SRC(fp->code);
730 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
733 /* Common case where 'jump_false' is next insn. */
735 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
736 target = i + fp->jt + 1;
741 /* Convert some jumps when 'jump_true' is next insn. */
743 switch (BPF_OP(fp->code)) {
745 insn->code = BPF_JMP | BPF_JNE | bpf_src;
748 insn->code = BPF_JMP | BPF_JLE | bpf_src;
751 insn->code = BPF_JMP | BPF_JLT | bpf_src;
757 target = i + fp->jf + 1;
762 /* Other jumps are mapped into two insns: Jxx and JA. */
763 target = i + fp->jt + 1;
764 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
768 insn->code = BPF_JMP | BPF_JA;
769 target = i + fp->jf + 1;
773 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
774 case BPF_LDX | BPF_MSH | BPF_B: {
775 struct sock_filter tmp = {
776 .code = BPF_LD | BPF_ABS | BPF_B,
783 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
784 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
785 convert_bpf_ld_abs(&tmp, &insn);
788 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
790 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
792 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
794 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
796 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
799 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
800 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
802 case BPF_RET | BPF_A:
803 case BPF_RET | BPF_K:
804 if (BPF_RVAL(fp->code) == BPF_K)
805 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
807 *insn = BPF_EXIT_INSN();
810 /* Store to stack. */
813 stack_off = fp->k * 4 + 4;
814 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
815 BPF_ST ? BPF_REG_A : BPF_REG_X,
817 /* check_load_and_stores() verifies that classic BPF can
818 * load from stack only after write, so tracking
819 * stack_depth for ST|STX insns is enough
821 if (new_prog && new_prog->aux->stack_depth < stack_off)
822 new_prog->aux->stack_depth = stack_off;
825 /* Load from stack. */
826 case BPF_LD | BPF_MEM:
827 case BPF_LDX | BPF_MEM:
828 stack_off = fp->k * 4 + 4;
829 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
830 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
835 case BPF_LD | BPF_IMM:
836 case BPF_LDX | BPF_IMM:
837 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
838 BPF_REG_A : BPF_REG_X, fp->k);
842 case BPF_MISC | BPF_TAX:
843 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
847 case BPF_MISC | BPF_TXA:
848 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
851 /* A = skb->len or X = skb->len */
852 case BPF_LD | BPF_W | BPF_LEN:
853 case BPF_LDX | BPF_W | BPF_LEN:
854 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
855 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
856 offsetof(struct sk_buff, len));
859 /* Access seccomp_data fields. */
860 case BPF_LDX | BPF_ABS | BPF_W:
861 /* A = *(u32 *) (ctx + K) */
862 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
865 /* Unknown instruction. */
872 memcpy(new_insn, tmp_insns,
873 sizeof(*insn) * (insn - tmp_insns));
874 new_insn += insn - tmp_insns;
878 /* Only calculating new length. */
879 *new_len = new_insn - first_insn;
881 *new_len += 4; /* Prologue bits. */
886 if (new_flen != new_insn - first_insn) {
887 new_flen = new_insn - first_insn;
894 BUG_ON(*new_len != new_flen);
903 * As we dont want to clear mem[] array for each packet going through
904 * __bpf_prog_run(), we check that filter loaded by user never try to read
905 * a cell if not previously written, and we check all branches to be sure
906 * a malicious user doesn't try to abuse us.
908 static int check_load_and_stores(const struct sock_filter *filter, int flen)
910 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
913 BUILD_BUG_ON(BPF_MEMWORDS > 16);
915 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
919 memset(masks, 0xff, flen * sizeof(*masks));
921 for (pc = 0; pc < flen; pc++) {
922 memvalid &= masks[pc];
924 switch (filter[pc].code) {
927 memvalid |= (1 << filter[pc].k);
929 case BPF_LD | BPF_MEM:
930 case BPF_LDX | BPF_MEM:
931 if (!(memvalid & (1 << filter[pc].k))) {
936 case BPF_JMP | BPF_JA:
937 /* A jump must set masks on target */
938 masks[pc + 1 + filter[pc].k] &= memvalid;
941 case BPF_JMP | BPF_JEQ | BPF_K:
942 case BPF_JMP | BPF_JEQ | BPF_X:
943 case BPF_JMP | BPF_JGE | BPF_K:
944 case BPF_JMP | BPF_JGE | BPF_X:
945 case BPF_JMP | BPF_JGT | BPF_K:
946 case BPF_JMP | BPF_JGT | BPF_X:
947 case BPF_JMP | BPF_JSET | BPF_K:
948 case BPF_JMP | BPF_JSET | BPF_X:
949 /* A jump must set masks on targets */
950 masks[pc + 1 + filter[pc].jt] &= memvalid;
951 masks[pc + 1 + filter[pc].jf] &= memvalid;
961 static bool chk_code_allowed(u16 code_to_probe)
963 static const bool codes[] = {
964 /* 32 bit ALU operations */
965 [BPF_ALU | BPF_ADD | BPF_K] = true,
966 [BPF_ALU | BPF_ADD | BPF_X] = true,
967 [BPF_ALU | BPF_SUB | BPF_K] = true,
968 [BPF_ALU | BPF_SUB | BPF_X] = true,
969 [BPF_ALU | BPF_MUL | BPF_K] = true,
970 [BPF_ALU | BPF_MUL | BPF_X] = true,
971 [BPF_ALU | BPF_DIV | BPF_K] = true,
972 [BPF_ALU | BPF_DIV | BPF_X] = true,
973 [BPF_ALU | BPF_MOD | BPF_K] = true,
974 [BPF_ALU | BPF_MOD | BPF_X] = true,
975 [BPF_ALU | BPF_AND | BPF_K] = true,
976 [BPF_ALU | BPF_AND | BPF_X] = true,
977 [BPF_ALU | BPF_OR | BPF_K] = true,
978 [BPF_ALU | BPF_OR | BPF_X] = true,
979 [BPF_ALU | BPF_XOR | BPF_K] = true,
980 [BPF_ALU | BPF_XOR | BPF_X] = true,
981 [BPF_ALU | BPF_LSH | BPF_K] = true,
982 [BPF_ALU | BPF_LSH | BPF_X] = true,
983 [BPF_ALU | BPF_RSH | BPF_K] = true,
984 [BPF_ALU | BPF_RSH | BPF_X] = true,
985 [BPF_ALU | BPF_NEG] = true,
986 /* Load instructions */
987 [BPF_LD | BPF_W | BPF_ABS] = true,
988 [BPF_LD | BPF_H | BPF_ABS] = true,
989 [BPF_LD | BPF_B | BPF_ABS] = true,
990 [BPF_LD | BPF_W | BPF_LEN] = true,
991 [BPF_LD | BPF_W | BPF_IND] = true,
992 [BPF_LD | BPF_H | BPF_IND] = true,
993 [BPF_LD | BPF_B | BPF_IND] = true,
994 [BPF_LD | BPF_IMM] = true,
995 [BPF_LD | BPF_MEM] = true,
996 [BPF_LDX | BPF_W | BPF_LEN] = true,
997 [BPF_LDX | BPF_B | BPF_MSH] = true,
998 [BPF_LDX | BPF_IMM] = true,
999 [BPF_LDX | BPF_MEM] = true,
1000 /* Store instructions */
1003 /* Misc instructions */
1004 [BPF_MISC | BPF_TAX] = true,
1005 [BPF_MISC | BPF_TXA] = true,
1006 /* Return instructions */
1007 [BPF_RET | BPF_K] = true,
1008 [BPF_RET | BPF_A] = true,
1009 /* Jump instructions */
1010 [BPF_JMP | BPF_JA] = true,
1011 [BPF_JMP | BPF_JEQ | BPF_K] = true,
1012 [BPF_JMP | BPF_JEQ | BPF_X] = true,
1013 [BPF_JMP | BPF_JGE | BPF_K] = true,
1014 [BPF_JMP | BPF_JGE | BPF_X] = true,
1015 [BPF_JMP | BPF_JGT | BPF_K] = true,
1016 [BPF_JMP | BPF_JGT | BPF_X] = true,
1017 [BPF_JMP | BPF_JSET | BPF_K] = true,
1018 [BPF_JMP | BPF_JSET | BPF_X] = true,
1021 if (code_to_probe >= ARRAY_SIZE(codes))
1024 return codes[code_to_probe];
1027 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1032 if (flen == 0 || flen > BPF_MAXINSNS)
1039 * bpf_check_classic - verify socket filter code
1040 * @filter: filter to verify
1041 * @flen: length of filter
1043 * Check the user's filter code. If we let some ugly
1044 * filter code slip through kaboom! The filter must contain
1045 * no references or jumps that are out of range, no illegal
1046 * instructions, and must end with a RET instruction.
1048 * All jumps are forward as they are not signed.
1050 * Returns 0 if the rule set is legal or -EINVAL if not.
1052 static int bpf_check_classic(const struct sock_filter *filter,
1058 /* Check the filter code now */
1059 for (pc = 0; pc < flen; pc++) {
1060 const struct sock_filter *ftest = &filter[pc];
1062 /* May we actually operate on this code? */
1063 if (!chk_code_allowed(ftest->code))
1066 /* Some instructions need special checks */
1067 switch (ftest->code) {
1068 case BPF_ALU | BPF_DIV | BPF_K:
1069 case BPF_ALU | BPF_MOD | BPF_K:
1070 /* Check for division by zero */
1074 case BPF_ALU | BPF_LSH | BPF_K:
1075 case BPF_ALU | BPF_RSH | BPF_K:
1079 case BPF_LD | BPF_MEM:
1080 case BPF_LDX | BPF_MEM:
1083 /* Check for invalid memory addresses */
1084 if (ftest->k >= BPF_MEMWORDS)
1087 case BPF_JMP | BPF_JA:
1088 /* Note, the large ftest->k might cause loops.
1089 * Compare this with conditional jumps below,
1090 * where offsets are limited. --ANK (981016)
1092 if (ftest->k >= (unsigned int)(flen - pc - 1))
1095 case BPF_JMP | BPF_JEQ | BPF_K:
1096 case BPF_JMP | BPF_JEQ | BPF_X:
1097 case BPF_JMP | BPF_JGE | BPF_K:
1098 case BPF_JMP | BPF_JGE | BPF_X:
1099 case BPF_JMP | BPF_JGT | BPF_K:
1100 case BPF_JMP | BPF_JGT | BPF_X:
1101 case BPF_JMP | BPF_JSET | BPF_K:
1102 case BPF_JMP | BPF_JSET | BPF_X:
1103 /* Both conditionals must be safe */
1104 if (pc + ftest->jt + 1 >= flen ||
1105 pc + ftest->jf + 1 >= flen)
1108 case BPF_LD | BPF_W | BPF_ABS:
1109 case BPF_LD | BPF_H | BPF_ABS:
1110 case BPF_LD | BPF_B | BPF_ABS:
1112 if (bpf_anc_helper(ftest) & BPF_ANC)
1114 /* Ancillary operation unknown or unsupported */
1115 if (anc_found == false && ftest->k >= SKF_AD_OFF)
1120 /* Last instruction must be a RET code */
1121 switch (filter[flen - 1].code) {
1122 case BPF_RET | BPF_K:
1123 case BPF_RET | BPF_A:
1124 return check_load_and_stores(filter, flen);
1130 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1131 const struct sock_fprog *fprog)
1133 unsigned int fsize = bpf_classic_proglen(fprog);
1134 struct sock_fprog_kern *fkprog;
1136 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1140 fkprog = fp->orig_prog;
1141 fkprog->len = fprog->len;
1143 fkprog->filter = kmemdup(fp->insns, fsize,
1144 GFP_KERNEL | __GFP_NOWARN);
1145 if (!fkprog->filter) {
1146 kfree(fp->orig_prog);
1153 static void bpf_release_orig_filter(struct bpf_prog *fp)
1155 struct sock_fprog_kern *fprog = fp->orig_prog;
1158 kfree(fprog->filter);
1163 static void __bpf_prog_release(struct bpf_prog *prog)
1165 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1168 bpf_release_orig_filter(prog);
1169 bpf_prog_free(prog);
1173 static void __sk_filter_release(struct sk_filter *fp)
1175 __bpf_prog_release(fp->prog);
1180 * sk_filter_release_rcu - Release a socket filter by rcu_head
1181 * @rcu: rcu_head that contains the sk_filter to free
1183 static void sk_filter_release_rcu(struct rcu_head *rcu)
1185 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1187 __sk_filter_release(fp);
1191 * sk_filter_release - release a socket filter
1192 * @fp: filter to remove
1194 * Remove a filter from a socket and release its resources.
1196 static void sk_filter_release(struct sk_filter *fp)
1198 if (refcount_dec_and_test(&fp->refcnt))
1199 call_rcu(&fp->rcu, sk_filter_release_rcu);
1202 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1204 u32 filter_size = bpf_prog_size(fp->prog->len);
1206 atomic_sub(filter_size, &sk->sk_omem_alloc);
1207 sk_filter_release(fp);
1210 /* try to charge the socket memory if there is space available
1211 * return true on success
1213 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1215 u32 filter_size = bpf_prog_size(fp->prog->len);
1217 /* same check as in sock_kmalloc() */
1218 if (filter_size <= sysctl_optmem_max &&
1219 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
1220 atomic_add(filter_size, &sk->sk_omem_alloc);
1226 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1228 if (!refcount_inc_not_zero(&fp->refcnt))
1231 if (!__sk_filter_charge(sk, fp)) {
1232 sk_filter_release(fp);
1238 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1240 struct sock_filter *old_prog;
1241 struct bpf_prog *old_fp;
1242 int err, new_len, old_len = fp->len;
1243 bool seen_ld_abs = false;
1245 /* We are free to overwrite insns et al right here as it
1246 * won't be used at this point in time anymore internally
1247 * after the migration to the internal BPF instruction
1250 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1251 sizeof(struct bpf_insn));
1253 /* Conversion cannot happen on overlapping memory areas,
1254 * so we need to keep the user BPF around until the 2nd
1255 * pass. At this time, the user BPF is stored in fp->insns.
1257 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1258 GFP_KERNEL | __GFP_NOWARN);
1264 /* 1st pass: calculate the new program length. */
1265 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1270 /* Expand fp for appending the new filter representation. */
1272 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1274 /* The old_fp is still around in case we couldn't
1275 * allocate new memory, so uncharge on that one.
1284 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1285 err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1288 /* 2nd bpf_convert_filter() can fail only if it fails
1289 * to allocate memory, remapping must succeed. Note,
1290 * that at this time old_fp has already been released
1295 fp = bpf_prog_select_runtime(fp, &err);
1305 __bpf_prog_release(fp);
1306 return ERR_PTR(err);
1309 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1310 bpf_aux_classic_check_t trans)
1314 fp->bpf_func = NULL;
1317 err = bpf_check_classic(fp->insns, fp->len);
1319 __bpf_prog_release(fp);
1320 return ERR_PTR(err);
1323 /* There might be additional checks and transformations
1324 * needed on classic filters, f.e. in case of seccomp.
1327 err = trans(fp->insns, fp->len);
1329 __bpf_prog_release(fp);
1330 return ERR_PTR(err);
1334 /* Probe if we can JIT compile the filter and if so, do
1335 * the compilation of the filter.
1337 bpf_jit_compile(fp);
1339 /* JIT compiler couldn't process this filter, so do the
1340 * internal BPF translation for the optimized interpreter.
1343 fp = bpf_migrate_filter(fp);
1349 * bpf_prog_create - create an unattached filter
1350 * @pfp: the unattached filter that is created
1351 * @fprog: the filter program
1353 * Create a filter independent of any socket. We first run some
1354 * sanity checks on it to make sure it does not explode on us later.
1355 * If an error occurs or there is insufficient memory for the filter
1356 * a negative errno code is returned. On success the return is zero.
1358 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1360 unsigned int fsize = bpf_classic_proglen(fprog);
1361 struct bpf_prog *fp;
1363 /* Make sure new filter is there and in the right amounts. */
1364 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1367 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1371 memcpy(fp->insns, fprog->filter, fsize);
1373 fp->len = fprog->len;
1374 /* Since unattached filters are not copied back to user
1375 * space through sk_get_filter(), we do not need to hold
1376 * a copy here, and can spare us the work.
1378 fp->orig_prog = NULL;
1380 /* bpf_prepare_filter() already takes care of freeing
1381 * memory in case something goes wrong.
1383 fp = bpf_prepare_filter(fp, NULL);
1390 EXPORT_SYMBOL_GPL(bpf_prog_create);
1393 * bpf_prog_create_from_user - create an unattached filter from user buffer
1394 * @pfp: the unattached filter that is created
1395 * @fprog: the filter program
1396 * @trans: post-classic verifier transformation handler
1397 * @save_orig: save classic BPF program
1399 * This function effectively does the same as bpf_prog_create(), only
1400 * that it builds up its insns buffer from user space provided buffer.
1401 * It also allows for passing a bpf_aux_classic_check_t handler.
1403 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1404 bpf_aux_classic_check_t trans, bool save_orig)
1406 unsigned int fsize = bpf_classic_proglen(fprog);
1407 struct bpf_prog *fp;
1410 /* Make sure new filter is there and in the right amounts. */
1411 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1414 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1418 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1419 __bpf_prog_free(fp);
1423 fp->len = fprog->len;
1424 fp->orig_prog = NULL;
1427 err = bpf_prog_store_orig_filter(fp, fprog);
1429 __bpf_prog_free(fp);
1434 /* bpf_prepare_filter() already takes care of freeing
1435 * memory in case something goes wrong.
1437 fp = bpf_prepare_filter(fp, trans);
1444 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1446 void bpf_prog_destroy(struct bpf_prog *fp)
1448 __bpf_prog_release(fp);
1450 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1452 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1454 struct sk_filter *fp, *old_fp;
1456 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1462 if (!__sk_filter_charge(sk, fp)) {
1466 refcount_set(&fp->refcnt, 1);
1468 old_fp = rcu_dereference_protected(sk->sk_filter,
1469 lockdep_sock_is_held(sk));
1470 rcu_assign_pointer(sk->sk_filter, fp);
1473 sk_filter_uncharge(sk, old_fp);
1479 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1481 unsigned int fsize = bpf_classic_proglen(fprog);
1482 struct bpf_prog *prog;
1485 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1486 return ERR_PTR(-EPERM);
1488 /* Make sure new filter is there and in the right amounts. */
1489 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1490 return ERR_PTR(-EINVAL);
1492 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1494 return ERR_PTR(-ENOMEM);
1496 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1497 __bpf_prog_free(prog);
1498 return ERR_PTR(-EFAULT);
1501 prog->len = fprog->len;
1503 err = bpf_prog_store_orig_filter(prog, fprog);
1505 __bpf_prog_free(prog);
1506 return ERR_PTR(-ENOMEM);
1509 /* bpf_prepare_filter() already takes care of freeing
1510 * memory in case something goes wrong.
1512 return bpf_prepare_filter(prog, NULL);
1516 * sk_attach_filter - attach a socket filter
1517 * @fprog: the filter program
1518 * @sk: the socket to use
1520 * Attach the user's filter code. We first run some sanity checks on
1521 * it to make sure it does not explode on us later. If an error
1522 * occurs or there is insufficient memory for the filter a negative
1523 * errno code is returned. On success the return is zero.
1525 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1527 struct bpf_prog *prog = __get_filter(fprog, sk);
1531 return PTR_ERR(prog);
1533 err = __sk_attach_prog(prog, sk);
1535 __bpf_prog_release(prog);
1541 EXPORT_SYMBOL_GPL(sk_attach_filter);
1543 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1545 struct bpf_prog *prog = __get_filter(fprog, sk);
1549 return PTR_ERR(prog);
1551 if (bpf_prog_size(prog->len) > sysctl_optmem_max)
1554 err = reuseport_attach_prog(sk, prog);
1557 __bpf_prog_release(prog);
1562 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1564 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1565 return ERR_PTR(-EPERM);
1567 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1570 int sk_attach_bpf(u32 ufd, struct sock *sk)
1572 struct bpf_prog *prog = __get_bpf(ufd, sk);
1576 return PTR_ERR(prog);
1578 err = __sk_attach_prog(prog, sk);
1587 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1589 struct bpf_prog *prog;
1592 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1595 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1596 if (PTR_ERR(prog) == -EINVAL)
1597 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1599 return PTR_ERR(prog);
1601 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1602 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1603 * bpf prog (e.g. sockmap). It depends on the
1604 * limitation imposed by bpf_prog_load().
1605 * Hence, sysctl_optmem_max is not checked.
1607 if ((sk->sk_type != SOCK_STREAM &&
1608 sk->sk_type != SOCK_DGRAM) ||
1609 (sk->sk_protocol != IPPROTO_UDP &&
1610 sk->sk_protocol != IPPROTO_TCP) ||
1611 (sk->sk_family != AF_INET &&
1612 sk->sk_family != AF_INET6)) {
1617 /* BPF_PROG_TYPE_SOCKET_FILTER */
1618 if (bpf_prog_size(prog->len) > sysctl_optmem_max) {
1624 err = reuseport_attach_prog(sk, prog);
1632 void sk_reuseport_prog_free(struct bpf_prog *prog)
1637 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1640 bpf_prog_destroy(prog);
1643 struct bpf_scratchpad {
1645 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1646 u8 buff[MAX_BPF_STACK];
1650 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1652 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1653 unsigned int write_len)
1655 return skb_ensure_writable(skb, write_len);
1658 static inline int bpf_try_make_writable(struct sk_buff *skb,
1659 unsigned int write_len)
1661 int err = __bpf_try_make_writable(skb, write_len);
1663 bpf_compute_data_pointers(skb);
1667 static int bpf_try_make_head_writable(struct sk_buff *skb)
1669 return bpf_try_make_writable(skb, skb_headlen(skb));
1672 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1674 if (skb_at_tc_ingress(skb))
1675 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1678 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1680 if (skb_at_tc_ingress(skb))
1681 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1684 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1685 const void *, from, u32, len, u64, flags)
1689 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1691 if (unlikely(offset > INT_MAX))
1693 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1696 ptr = skb->data + offset;
1697 if (flags & BPF_F_RECOMPUTE_CSUM)
1698 __skb_postpull_rcsum(skb, ptr, len, offset);
1700 memcpy(ptr, from, len);
1702 if (flags & BPF_F_RECOMPUTE_CSUM)
1703 __skb_postpush_rcsum(skb, ptr, len, offset);
1704 if (flags & BPF_F_INVALIDATE_HASH)
1705 skb_clear_hash(skb);
1710 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1711 .func = bpf_skb_store_bytes,
1713 .ret_type = RET_INTEGER,
1714 .arg1_type = ARG_PTR_TO_CTX,
1715 .arg2_type = ARG_ANYTHING,
1716 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1717 .arg4_type = ARG_CONST_SIZE,
1718 .arg5_type = ARG_ANYTHING,
1721 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1722 void *, to, u32, len)
1726 if (unlikely(offset > INT_MAX))
1729 ptr = skb_header_pointer(skb, offset, len, to);
1733 memcpy(to, ptr, len);
1741 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1742 .func = bpf_skb_load_bytes,
1744 .ret_type = RET_INTEGER,
1745 .arg1_type = ARG_PTR_TO_CTX,
1746 .arg2_type = ARG_ANYTHING,
1747 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1748 .arg4_type = ARG_CONST_SIZE,
1751 BPF_CALL_4(bpf_flow_dissector_load_bytes,
1752 const struct bpf_flow_dissector *, ctx, u32, offset,
1753 void *, to, u32, len)
1757 if (unlikely(offset > 0xffff))
1760 if (unlikely(!ctx->skb))
1763 ptr = skb_header_pointer(ctx->skb, offset, len, to);
1767 memcpy(to, ptr, len);
1775 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1776 .func = bpf_flow_dissector_load_bytes,
1778 .ret_type = RET_INTEGER,
1779 .arg1_type = ARG_PTR_TO_CTX,
1780 .arg2_type = ARG_ANYTHING,
1781 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1782 .arg4_type = ARG_CONST_SIZE,
1785 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1786 u32, offset, void *, to, u32, len, u32, start_header)
1788 u8 *end = skb_tail_pointer(skb);
1791 if (unlikely(offset > 0xffff))
1794 switch (start_header) {
1795 case BPF_HDR_START_MAC:
1796 if (unlikely(!skb_mac_header_was_set(skb)))
1798 start = skb_mac_header(skb);
1800 case BPF_HDR_START_NET:
1801 start = skb_network_header(skb);
1807 ptr = start + offset;
1809 if (likely(ptr + len <= end)) {
1810 memcpy(to, ptr, len);
1819 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1820 .func = bpf_skb_load_bytes_relative,
1822 .ret_type = RET_INTEGER,
1823 .arg1_type = ARG_PTR_TO_CTX,
1824 .arg2_type = ARG_ANYTHING,
1825 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1826 .arg4_type = ARG_CONST_SIZE,
1827 .arg5_type = ARG_ANYTHING,
1830 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1832 /* Idea is the following: should the needed direct read/write
1833 * test fail during runtime, we can pull in more data and redo
1834 * again, since implicitly, we invalidate previous checks here.
1836 * Or, since we know how much we need to make read/writeable,
1837 * this can be done once at the program beginning for direct
1838 * access case. By this we overcome limitations of only current
1839 * headroom being accessible.
1841 return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1844 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1845 .func = bpf_skb_pull_data,
1847 .ret_type = RET_INTEGER,
1848 .arg1_type = ARG_PTR_TO_CTX,
1849 .arg2_type = ARG_ANYTHING,
1852 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1854 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1857 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1858 .func = bpf_sk_fullsock,
1860 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
1861 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
1864 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1865 unsigned int write_len)
1867 return __bpf_try_make_writable(skb, write_len);
1870 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1872 /* Idea is the following: should the needed direct read/write
1873 * test fail during runtime, we can pull in more data and redo
1874 * again, since implicitly, we invalidate previous checks here.
1876 * Or, since we know how much we need to make read/writeable,
1877 * this can be done once at the program beginning for direct
1878 * access case. By this we overcome limitations of only current
1879 * headroom being accessible.
1881 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1884 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1885 .func = sk_skb_pull_data,
1887 .ret_type = RET_INTEGER,
1888 .arg1_type = ARG_PTR_TO_CTX,
1889 .arg2_type = ARG_ANYTHING,
1892 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1893 u64, from, u64, to, u64, flags)
1897 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1899 if (unlikely(offset > 0xffff || offset & 1))
1901 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1904 ptr = (__sum16 *)(skb->data + offset);
1905 switch (flags & BPF_F_HDR_FIELD_MASK) {
1907 if (unlikely(from != 0))
1910 csum_replace_by_diff(ptr, to);
1913 csum_replace2(ptr, from, to);
1916 csum_replace4(ptr, from, to);
1925 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1926 .func = bpf_l3_csum_replace,
1928 .ret_type = RET_INTEGER,
1929 .arg1_type = ARG_PTR_TO_CTX,
1930 .arg2_type = ARG_ANYTHING,
1931 .arg3_type = ARG_ANYTHING,
1932 .arg4_type = ARG_ANYTHING,
1933 .arg5_type = ARG_ANYTHING,
1936 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1937 u64, from, u64, to, u64, flags)
1939 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1940 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1941 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1944 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1945 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1947 if (unlikely(offset > 0xffff || offset & 1))
1949 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1952 ptr = (__sum16 *)(skb->data + offset);
1953 if (is_mmzero && !do_mforce && !*ptr)
1956 switch (flags & BPF_F_HDR_FIELD_MASK) {
1958 if (unlikely(from != 0))
1961 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1964 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1967 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1973 if (is_mmzero && !*ptr)
1974 *ptr = CSUM_MANGLED_0;
1978 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1979 .func = bpf_l4_csum_replace,
1981 .ret_type = RET_INTEGER,
1982 .arg1_type = ARG_PTR_TO_CTX,
1983 .arg2_type = ARG_ANYTHING,
1984 .arg3_type = ARG_ANYTHING,
1985 .arg4_type = ARG_ANYTHING,
1986 .arg5_type = ARG_ANYTHING,
1989 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1990 __be32 *, to, u32, to_size, __wsum, seed)
1992 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1993 u32 diff_size = from_size + to_size;
1996 /* This is quite flexible, some examples:
1998 * from_size == 0, to_size > 0, seed := csum --> pushing data
1999 * from_size > 0, to_size == 0, seed := csum --> pulling data
2000 * from_size > 0, to_size > 0, seed := 0 --> diffing data
2002 * Even for diffing, from_size and to_size don't need to be equal.
2004 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
2005 diff_size > sizeof(sp->diff)))
2008 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
2009 sp->diff[j] = ~from[i];
2010 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
2011 sp->diff[j] = to[i];
2013 return csum_partial(sp->diff, diff_size, seed);
2016 static const struct bpf_func_proto bpf_csum_diff_proto = {
2017 .func = bpf_csum_diff,
2020 .ret_type = RET_INTEGER,
2021 .arg1_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2022 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
2023 .arg3_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2024 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
2025 .arg5_type = ARG_ANYTHING,
2028 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2030 /* The interface is to be used in combination with bpf_csum_diff()
2031 * for direct packet writes. csum rotation for alignment as well
2032 * as emulating csum_sub() can be done from the eBPF program.
2034 if (skb->ip_summed == CHECKSUM_COMPLETE)
2035 return (skb->csum = csum_add(skb->csum, csum));
2040 static const struct bpf_func_proto bpf_csum_update_proto = {
2041 .func = bpf_csum_update,
2043 .ret_type = RET_INTEGER,
2044 .arg1_type = ARG_PTR_TO_CTX,
2045 .arg2_type = ARG_ANYTHING,
2048 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
2050 /* The interface is to be used in combination with bpf_skb_adjust_room()
2051 * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET
2052 * is passed as flags, for example.
2055 case BPF_CSUM_LEVEL_INC:
2056 __skb_incr_checksum_unnecessary(skb);
2058 case BPF_CSUM_LEVEL_DEC:
2059 __skb_decr_checksum_unnecessary(skb);
2061 case BPF_CSUM_LEVEL_RESET:
2062 __skb_reset_checksum_unnecessary(skb);
2064 case BPF_CSUM_LEVEL_QUERY:
2065 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2066 skb->csum_level : -EACCES;
2074 static const struct bpf_func_proto bpf_csum_level_proto = {
2075 .func = bpf_csum_level,
2077 .ret_type = RET_INTEGER,
2078 .arg1_type = ARG_PTR_TO_CTX,
2079 .arg2_type = ARG_ANYTHING,
2082 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2084 return dev_forward_skb_nomtu(dev, skb);
2087 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2088 struct sk_buff *skb)
2090 int ret = ____dev_forward_skb(dev, skb, false);
2094 ret = netif_rx(skb);
2100 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2104 if (dev_xmit_recursion()) {
2105 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2113 dev_xmit_recursion_inc();
2114 ret = dev_queue_xmit(skb);
2115 dev_xmit_recursion_dec();
2120 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2123 unsigned int mlen = skb_network_offset(skb);
2126 __skb_pull(skb, mlen);
2128 /* At ingress, the mac header has already been pulled once.
2129 * At egress, skb_pospull_rcsum has to be done in case that
2130 * the skb is originated from ingress (i.e. a forwarded skb)
2131 * to ensure that rcsum starts at net header.
2133 if (!skb_at_tc_ingress(skb))
2134 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2136 skb_pop_mac_header(skb);
2137 skb_reset_mac_len(skb);
2138 return flags & BPF_F_INGRESS ?
2139 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2142 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2145 /* Verify that a link layer header is carried */
2146 if (unlikely(skb->mac_header >= skb->network_header)) {
2151 bpf_push_mac_rcsum(skb);
2152 return flags & BPF_F_INGRESS ?
2153 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2156 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2159 if (dev_is_mac_header_xmit(dev))
2160 return __bpf_redirect_common(skb, dev, flags);
2162 return __bpf_redirect_no_mac(skb, dev, flags);
2165 #if IS_ENABLED(CONFIG_IPV6)
2166 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2167 struct net_device *dev, struct bpf_nh_params *nh)
2169 u32 hh_len = LL_RESERVED_SPACE(dev);
2170 const struct in6_addr *nexthop;
2171 struct dst_entry *dst = NULL;
2172 struct neighbour *neigh;
2174 if (dev_xmit_recursion()) {
2175 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2182 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2183 skb = skb_expand_head(skb, hh_len);
2191 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2192 &ipv6_hdr(skb)->daddr);
2194 nexthop = &nh->ipv6_nh;
2196 neigh = ip_neigh_gw6(dev, nexthop);
2197 if (likely(!IS_ERR(neigh))) {
2200 sock_confirm_neigh(skb, neigh);
2201 dev_xmit_recursion_inc();
2202 ret = neigh_output(neigh, skb, false);
2203 dev_xmit_recursion_dec();
2204 rcu_read_unlock_bh();
2207 rcu_read_unlock_bh();
2209 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2215 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2216 struct bpf_nh_params *nh)
2218 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2219 struct net *net = dev_net(dev);
2220 int err, ret = NET_XMIT_DROP;
2223 struct dst_entry *dst;
2224 struct flowi6 fl6 = {
2225 .flowi6_flags = FLOWI_FLAG_ANYSRC,
2226 .flowi6_mark = skb->mark,
2227 .flowlabel = ip6_flowinfo(ip6h),
2228 .flowi6_oif = dev->ifindex,
2229 .flowi6_proto = ip6h->nexthdr,
2230 .daddr = ip6h->daddr,
2231 .saddr = ip6h->saddr,
2234 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2238 skb_dst_set(skb, dst);
2239 } else if (nh->nh_family != AF_INET6) {
2243 err = bpf_out_neigh_v6(net, skb, dev, nh);
2244 if (unlikely(net_xmit_eval(err)))
2245 dev->stats.tx_errors++;
2247 ret = NET_XMIT_SUCCESS;
2250 dev->stats.tx_errors++;
2256 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2257 struct bpf_nh_params *nh)
2260 return NET_XMIT_DROP;
2262 #endif /* CONFIG_IPV6 */
2264 #if IS_ENABLED(CONFIG_INET)
2265 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2266 struct net_device *dev, struct bpf_nh_params *nh)
2268 u32 hh_len = LL_RESERVED_SPACE(dev);
2269 struct neighbour *neigh;
2270 bool is_v6gw = false;
2272 if (dev_xmit_recursion()) {
2273 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2280 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2281 skb = skb_expand_head(skb, hh_len);
2288 struct dst_entry *dst = skb_dst(skb);
2289 struct rtable *rt = container_of(dst, struct rtable, dst);
2291 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2292 } else if (nh->nh_family == AF_INET6) {
2293 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2295 } else if (nh->nh_family == AF_INET) {
2296 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2298 rcu_read_unlock_bh();
2302 if (likely(!IS_ERR(neigh))) {
2305 sock_confirm_neigh(skb, neigh);
2306 dev_xmit_recursion_inc();
2307 ret = neigh_output(neigh, skb, is_v6gw);
2308 dev_xmit_recursion_dec();
2309 rcu_read_unlock_bh();
2312 rcu_read_unlock_bh();
2318 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2319 struct bpf_nh_params *nh)
2321 const struct iphdr *ip4h = ip_hdr(skb);
2322 struct net *net = dev_net(dev);
2323 int err, ret = NET_XMIT_DROP;
2326 struct flowi4 fl4 = {
2327 .flowi4_flags = FLOWI_FLAG_ANYSRC,
2328 .flowi4_mark = skb->mark,
2329 .flowi4_tos = RT_TOS(ip4h->tos),
2330 .flowi4_oif = dev->ifindex,
2331 .flowi4_proto = ip4h->protocol,
2332 .daddr = ip4h->daddr,
2333 .saddr = ip4h->saddr,
2337 rt = ip_route_output_flow(net, &fl4, NULL);
2340 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2345 skb_dst_set(skb, &rt->dst);
2348 err = bpf_out_neigh_v4(net, skb, dev, nh);
2349 if (unlikely(net_xmit_eval(err)))
2350 dev->stats.tx_errors++;
2352 ret = NET_XMIT_SUCCESS;
2355 dev->stats.tx_errors++;
2361 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2362 struct bpf_nh_params *nh)
2365 return NET_XMIT_DROP;
2367 #endif /* CONFIG_INET */
2369 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2370 struct bpf_nh_params *nh)
2372 struct ethhdr *ethh = eth_hdr(skb);
2374 if (unlikely(skb->mac_header >= skb->network_header))
2376 bpf_push_mac_rcsum(skb);
2377 if (is_multicast_ether_addr(ethh->h_dest))
2380 skb_pull(skb, sizeof(*ethh));
2381 skb_unset_mac_header(skb);
2382 skb_reset_network_header(skb);
2384 if (skb->protocol == htons(ETH_P_IP))
2385 return __bpf_redirect_neigh_v4(skb, dev, nh);
2386 else if (skb->protocol == htons(ETH_P_IPV6))
2387 return __bpf_redirect_neigh_v6(skb, dev, nh);
2393 /* Internal, non-exposed redirect flags. */
2395 BPF_F_NEIGH = (1ULL << 1),
2396 BPF_F_PEER = (1ULL << 2),
2397 BPF_F_NEXTHOP = (1ULL << 3),
2398 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2401 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2403 struct net_device *dev;
2404 struct sk_buff *clone;
2407 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2410 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2414 clone = skb_clone(skb, GFP_ATOMIC);
2415 if (unlikely(!clone))
2418 /* For direct write, we need to keep the invariant that the skbs
2419 * we're dealing with need to be uncloned. Should uncloning fail
2420 * here, we need to free the just generated clone to unclone once
2423 ret = bpf_try_make_head_writable(skb);
2424 if (unlikely(ret)) {
2429 return __bpf_redirect(clone, dev, flags);
2432 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2433 .func = bpf_clone_redirect,
2435 .ret_type = RET_INTEGER,
2436 .arg1_type = ARG_PTR_TO_CTX,
2437 .arg2_type = ARG_ANYTHING,
2438 .arg3_type = ARG_ANYTHING,
2441 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2442 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2444 int skb_do_redirect(struct sk_buff *skb)
2446 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2447 struct net *net = dev_net(skb->dev);
2448 struct net_device *dev;
2449 u32 flags = ri->flags;
2451 dev = dev_get_by_index_rcu(net, ri->tgt_index);
2456 if (flags & BPF_F_PEER) {
2457 const struct net_device_ops *ops = dev->netdev_ops;
2459 if (unlikely(!ops->ndo_get_peer_dev ||
2460 !skb_at_tc_ingress(skb)))
2462 dev = ops->ndo_get_peer_dev(dev);
2463 if (unlikely(!dev ||
2464 !(dev->flags & IFF_UP) ||
2465 net_eq(net, dev_net(dev))))
2470 return flags & BPF_F_NEIGH ?
2471 __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2473 __bpf_redirect(skb, dev, flags);
2479 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2481 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2483 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2487 ri->tgt_index = ifindex;
2489 return TC_ACT_REDIRECT;
2492 static const struct bpf_func_proto bpf_redirect_proto = {
2493 .func = bpf_redirect,
2495 .ret_type = RET_INTEGER,
2496 .arg1_type = ARG_ANYTHING,
2497 .arg2_type = ARG_ANYTHING,
2500 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2502 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2504 if (unlikely(flags))
2507 ri->flags = BPF_F_PEER;
2508 ri->tgt_index = ifindex;
2510 return TC_ACT_REDIRECT;
2513 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2514 .func = bpf_redirect_peer,
2516 .ret_type = RET_INTEGER,
2517 .arg1_type = ARG_ANYTHING,
2518 .arg2_type = ARG_ANYTHING,
2521 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2522 int, plen, u64, flags)
2524 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2526 if (unlikely((plen && plen < sizeof(*params)) || flags))
2529 ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2530 ri->tgt_index = ifindex;
2532 BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2534 memcpy(&ri->nh, params, sizeof(ri->nh));
2536 return TC_ACT_REDIRECT;
2539 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2540 .func = bpf_redirect_neigh,
2542 .ret_type = RET_INTEGER,
2543 .arg1_type = ARG_ANYTHING,
2544 .arg2_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2545 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
2546 .arg4_type = ARG_ANYTHING,
2549 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2551 msg->apply_bytes = bytes;
2555 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2556 .func = bpf_msg_apply_bytes,
2558 .ret_type = RET_INTEGER,
2559 .arg1_type = ARG_PTR_TO_CTX,
2560 .arg2_type = ARG_ANYTHING,
2563 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2565 msg->cork_bytes = bytes;
2569 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2570 .func = bpf_msg_cork_bytes,
2572 .ret_type = RET_INTEGER,
2573 .arg1_type = ARG_PTR_TO_CTX,
2574 .arg2_type = ARG_ANYTHING,
2577 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2578 u32, end, u64, flags)
2580 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2581 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2582 struct scatterlist *sge;
2583 u8 *raw, *to, *from;
2586 if (unlikely(flags || end <= start))
2589 /* First find the starting scatterlist element */
2593 len = sk_msg_elem(msg, i)->length;
2594 if (start < offset + len)
2596 sk_msg_iter_var_next(i);
2597 } while (i != msg->sg.end);
2599 if (unlikely(start >= offset + len))
2603 /* The start may point into the sg element so we need to also
2604 * account for the headroom.
2606 bytes_sg_total = start - offset + bytes;
2607 if (!test_bit(i, &msg->sg.copy) && bytes_sg_total <= len)
2610 /* At this point we need to linearize multiple scatterlist
2611 * elements or a single shared page. Either way we need to
2612 * copy into a linear buffer exclusively owned by BPF. Then
2613 * place the buffer in the scatterlist and fixup the original
2614 * entries by removing the entries now in the linear buffer
2615 * and shifting the remaining entries. For now we do not try
2616 * to copy partial entries to avoid complexity of running out
2617 * of sg_entry slots. The downside is reading a single byte
2618 * will copy the entire sg entry.
2621 copy += sk_msg_elem(msg, i)->length;
2622 sk_msg_iter_var_next(i);
2623 if (bytes_sg_total <= copy)
2625 } while (i != msg->sg.end);
2628 if (unlikely(bytes_sg_total > copy))
2631 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2633 if (unlikely(!page))
2636 raw = page_address(page);
2639 sge = sk_msg_elem(msg, i);
2640 from = sg_virt(sge);
2644 memcpy(to, from, len);
2647 put_page(sg_page(sge));
2649 sk_msg_iter_var_next(i);
2650 } while (i != last_sge);
2652 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2654 /* To repair sg ring we need to shift entries. If we only
2655 * had a single entry though we can just replace it and
2656 * be done. Otherwise walk the ring and shift the entries.
2658 WARN_ON_ONCE(last_sge == first_sge);
2659 shift = last_sge > first_sge ?
2660 last_sge - first_sge - 1 :
2661 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2666 sk_msg_iter_var_next(i);
2670 if (i + shift >= NR_MSG_FRAG_IDS)
2671 move_from = i + shift - NR_MSG_FRAG_IDS;
2673 move_from = i + shift;
2674 if (move_from == msg->sg.end)
2677 msg->sg.data[i] = msg->sg.data[move_from];
2678 msg->sg.data[move_from].length = 0;
2679 msg->sg.data[move_from].page_link = 0;
2680 msg->sg.data[move_from].offset = 0;
2681 sk_msg_iter_var_next(i);
2684 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2685 msg->sg.end - shift + NR_MSG_FRAG_IDS :
2686 msg->sg.end - shift;
2688 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2689 msg->data_end = msg->data + bytes;
2693 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2694 .func = bpf_msg_pull_data,
2696 .ret_type = RET_INTEGER,
2697 .arg1_type = ARG_PTR_TO_CTX,
2698 .arg2_type = ARG_ANYTHING,
2699 .arg3_type = ARG_ANYTHING,
2700 .arg4_type = ARG_ANYTHING,
2703 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2704 u32, len, u64, flags)
2706 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2707 u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2708 u8 *raw, *to, *from;
2711 if (unlikely(flags))
2714 if (unlikely(len == 0))
2717 /* First find the starting scatterlist element */
2721 l = sk_msg_elem(msg, i)->length;
2723 if (start < offset + l)
2725 sk_msg_iter_var_next(i);
2726 } while (i != msg->sg.end);
2728 if (start >= offset + l)
2731 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2733 /* If no space available will fallback to copy, we need at
2734 * least one scatterlist elem available to push data into
2735 * when start aligns to the beginning of an element or two
2736 * when it falls inside an element. We handle the start equals
2737 * offset case because its the common case for inserting a
2740 if (!space || (space == 1 && start != offset))
2741 copy = msg->sg.data[i].length;
2743 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2744 get_order(copy + len));
2745 if (unlikely(!page))
2751 raw = page_address(page);
2753 psge = sk_msg_elem(msg, i);
2754 front = start - offset;
2755 back = psge->length - front;
2756 from = sg_virt(psge);
2759 memcpy(raw, from, front);
2763 to = raw + front + len;
2765 memcpy(to, from, back);
2768 put_page(sg_page(psge));
2769 } else if (start - offset) {
2770 psge = sk_msg_elem(msg, i);
2771 rsge = sk_msg_elem_cpy(msg, i);
2773 psge->length = start - offset;
2774 rsge.length -= psge->length;
2775 rsge.offset += start;
2777 sk_msg_iter_var_next(i);
2778 sg_unmark_end(psge);
2779 sg_unmark_end(&rsge);
2780 sk_msg_iter_next(msg, end);
2783 /* Slot(s) to place newly allocated data */
2786 /* Shift one or two slots as needed */
2788 sge = sk_msg_elem_cpy(msg, i);
2790 sk_msg_iter_var_next(i);
2791 sg_unmark_end(&sge);
2792 sk_msg_iter_next(msg, end);
2794 nsge = sk_msg_elem_cpy(msg, i);
2796 sk_msg_iter_var_next(i);
2797 nnsge = sk_msg_elem_cpy(msg, i);
2800 while (i != msg->sg.end) {
2801 msg->sg.data[i] = sge;
2803 sk_msg_iter_var_next(i);
2806 nnsge = sk_msg_elem_cpy(msg, i);
2808 nsge = sk_msg_elem_cpy(msg, i);
2813 /* Place newly allocated data buffer */
2814 sk_mem_charge(msg->sk, len);
2815 msg->sg.size += len;
2816 __clear_bit(new, &msg->sg.copy);
2817 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2819 get_page(sg_page(&rsge));
2820 sk_msg_iter_var_next(new);
2821 msg->sg.data[new] = rsge;
2824 sk_msg_compute_data_pointers(msg);
2828 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2829 .func = bpf_msg_push_data,
2831 .ret_type = RET_INTEGER,
2832 .arg1_type = ARG_PTR_TO_CTX,
2833 .arg2_type = ARG_ANYTHING,
2834 .arg3_type = ARG_ANYTHING,
2835 .arg4_type = ARG_ANYTHING,
2838 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2844 sk_msg_iter_var_next(i);
2845 msg->sg.data[prev] = msg->sg.data[i];
2846 } while (i != msg->sg.end);
2848 sk_msg_iter_prev(msg, end);
2851 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2853 struct scatterlist tmp, sge;
2855 sk_msg_iter_next(msg, end);
2856 sge = sk_msg_elem_cpy(msg, i);
2857 sk_msg_iter_var_next(i);
2858 tmp = sk_msg_elem_cpy(msg, i);
2860 while (i != msg->sg.end) {
2861 msg->sg.data[i] = sge;
2862 sk_msg_iter_var_next(i);
2864 tmp = sk_msg_elem_cpy(msg, i);
2868 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2869 u32, len, u64, flags)
2871 u32 i = 0, l = 0, space, offset = 0;
2872 u64 last = start + len;
2875 if (unlikely(flags))
2878 /* First find the starting scatterlist element */
2882 l = sk_msg_elem(msg, i)->length;
2884 if (start < offset + l)
2886 sk_msg_iter_var_next(i);
2887 } while (i != msg->sg.end);
2889 /* Bounds checks: start and pop must be inside message */
2890 if (start >= offset + l || last >= msg->sg.size)
2893 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2896 /* --------------| offset
2897 * -| start |-------- len -------|
2899 * |----- a ----|-------- pop -------|----- b ----|
2900 * |______________________________________________| length
2903 * a: region at front of scatter element to save
2904 * b: region at back of scatter element to save when length > A + pop
2905 * pop: region to pop from element, same as input 'pop' here will be
2906 * decremented below per iteration.
2908 * Two top-level cases to handle when start != offset, first B is non
2909 * zero and second B is zero corresponding to when a pop includes more
2912 * Then if B is non-zero AND there is no space allocate space and
2913 * compact A, B regions into page. If there is space shift ring to
2914 * the rigth free'ing the next element in ring to place B, leaving
2915 * A untouched except to reduce length.
2917 if (start != offset) {
2918 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2920 int b = sge->length - pop - a;
2922 sk_msg_iter_var_next(i);
2924 if (pop < sge->length - a) {
2927 sk_msg_shift_right(msg, i);
2928 nsge = sk_msg_elem(msg, i);
2929 get_page(sg_page(sge));
2932 b, sge->offset + pop + a);
2934 struct page *page, *orig;
2937 page = alloc_pages(__GFP_NOWARN |
2938 __GFP_COMP | GFP_ATOMIC,
2940 if (unlikely(!page))
2944 orig = sg_page(sge);
2945 from = sg_virt(sge);
2946 to = page_address(page);
2947 memcpy(to, from, a);
2948 memcpy(to + a, from + a + pop, b);
2949 sg_set_page(sge, page, a + b, 0);
2953 } else if (pop >= sge->length - a) {
2954 pop -= (sge->length - a);
2959 /* From above the current layout _must_ be as follows,
2964 * |---- pop ---|---------------- b ------------|
2965 * |____________________________________________| length
2967 * Offset and start of the current msg elem are equal because in the
2968 * previous case we handled offset != start and either consumed the
2969 * entire element and advanced to the next element OR pop == 0.
2971 * Two cases to handle here are first pop is less than the length
2972 * leaving some remainder b above. Simply adjust the element's layout
2973 * in this case. Or pop >= length of the element so that b = 0. In this
2974 * case advance to next element decrementing pop.
2977 struct scatterlist *sge = sk_msg_elem(msg, i);
2979 if (pop < sge->length) {
2985 sk_msg_shift_left(msg, i);
2987 sk_msg_iter_var_next(i);
2990 sk_mem_uncharge(msg->sk, len - pop);
2991 msg->sg.size -= (len - pop);
2992 sk_msg_compute_data_pointers(msg);
2996 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
2997 .func = bpf_msg_pop_data,
2999 .ret_type = RET_INTEGER,
3000 .arg1_type = ARG_PTR_TO_CTX,
3001 .arg2_type = ARG_ANYTHING,
3002 .arg3_type = ARG_ANYTHING,
3003 .arg4_type = ARG_ANYTHING,
3006 #ifdef CONFIG_CGROUP_NET_CLASSID
3007 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3009 return __task_get_classid(current);
3012 static const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3013 .func = bpf_get_cgroup_classid_curr,
3015 .ret_type = RET_INTEGER,
3018 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3020 struct sock *sk = skb_to_full_sk(skb);
3022 if (!sk || !sk_fullsock(sk))
3025 return sock_cgroup_classid(&sk->sk_cgrp_data);
3028 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3029 .func = bpf_skb_cgroup_classid,
3031 .ret_type = RET_INTEGER,
3032 .arg1_type = ARG_PTR_TO_CTX,
3036 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3038 return task_get_classid(skb);
3041 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3042 .func = bpf_get_cgroup_classid,
3044 .ret_type = RET_INTEGER,
3045 .arg1_type = ARG_PTR_TO_CTX,
3048 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3050 return dst_tclassid(skb);
3053 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3054 .func = bpf_get_route_realm,
3056 .ret_type = RET_INTEGER,
3057 .arg1_type = ARG_PTR_TO_CTX,
3060 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3062 /* If skb_clear_hash() was called due to mangling, we can
3063 * trigger SW recalculation here. Later access to hash
3064 * can then use the inline skb->hash via context directly
3065 * instead of calling this helper again.
3067 return skb_get_hash(skb);
3070 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3071 .func = bpf_get_hash_recalc,
3073 .ret_type = RET_INTEGER,
3074 .arg1_type = ARG_PTR_TO_CTX,
3077 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3079 /* After all direct packet write, this can be used once for
3080 * triggering a lazy recalc on next skb_get_hash() invocation.
3082 skb_clear_hash(skb);
3086 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3087 .func = bpf_set_hash_invalid,
3089 .ret_type = RET_INTEGER,
3090 .arg1_type = ARG_PTR_TO_CTX,
3093 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3095 /* Set user specified hash as L4(+), so that it gets returned
3096 * on skb_get_hash() call unless BPF prog later on triggers a
3099 __skb_set_sw_hash(skb, hash, true);
3103 static const struct bpf_func_proto bpf_set_hash_proto = {
3104 .func = bpf_set_hash,
3106 .ret_type = RET_INTEGER,
3107 .arg1_type = ARG_PTR_TO_CTX,
3108 .arg2_type = ARG_ANYTHING,
3111 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3116 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3117 vlan_proto != htons(ETH_P_8021AD)))
3118 vlan_proto = htons(ETH_P_8021Q);
3120 bpf_push_mac_rcsum(skb);
3121 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3122 bpf_pull_mac_rcsum(skb);
3124 bpf_compute_data_pointers(skb);
3128 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3129 .func = bpf_skb_vlan_push,
3131 .ret_type = RET_INTEGER,
3132 .arg1_type = ARG_PTR_TO_CTX,
3133 .arg2_type = ARG_ANYTHING,
3134 .arg3_type = ARG_ANYTHING,
3137 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3141 bpf_push_mac_rcsum(skb);
3142 ret = skb_vlan_pop(skb);
3143 bpf_pull_mac_rcsum(skb);
3145 bpf_compute_data_pointers(skb);
3149 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3150 .func = bpf_skb_vlan_pop,
3152 .ret_type = RET_INTEGER,
3153 .arg1_type = ARG_PTR_TO_CTX,
3156 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3158 /* Caller already did skb_cow() with len as headroom,
3159 * so no need to do it here.
3162 memmove(skb->data, skb->data + len, off);
3163 memset(skb->data + off, 0, len);
3165 /* No skb_postpush_rcsum(skb, skb->data + off, len)
3166 * needed here as it does not change the skb->csum
3167 * result for checksum complete when summing over
3173 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3175 /* skb_ensure_writable() is not needed here, as we're
3176 * already working on an uncloned skb.
3178 if (unlikely(!pskb_may_pull(skb, off + len)))
3181 skb_postpull_rcsum(skb, skb->data + off, len);
3182 memmove(skb->data + len, skb->data, off);
3183 __skb_pull(skb, len);
3188 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3190 bool trans_same = skb->transport_header == skb->network_header;
3193 /* There's no need for __skb_push()/__skb_pull() pair to
3194 * get to the start of the mac header as we're guaranteed
3195 * to always start from here under eBPF.
3197 ret = bpf_skb_generic_push(skb, off, len);
3199 skb->mac_header -= len;
3200 skb->network_header -= len;
3202 skb->transport_header = skb->network_header;
3208 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3210 bool trans_same = skb->transport_header == skb->network_header;
3213 /* Same here, __skb_push()/__skb_pull() pair not needed. */
3214 ret = bpf_skb_generic_pop(skb, off, len);
3216 skb->mac_header += len;
3217 skb->network_header += len;
3219 skb->transport_header = skb->network_header;
3225 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3227 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3228 u32 off = skb_mac_header_len(skb);
3231 ret = skb_cow(skb, len_diff);
3232 if (unlikely(ret < 0))
3235 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3236 if (unlikely(ret < 0))
3239 if (skb_is_gso(skb)) {
3240 struct skb_shared_info *shinfo = skb_shinfo(skb);
3242 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */
3243 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3244 shinfo->gso_type &= ~SKB_GSO_TCPV4;
3245 shinfo->gso_type |= SKB_GSO_TCPV6;
3249 skb->protocol = htons(ETH_P_IPV6);
3250 skb_clear_hash(skb);
3255 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3257 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3258 u32 off = skb_mac_header_len(skb);
3261 ret = skb_unclone(skb, GFP_ATOMIC);
3262 if (unlikely(ret < 0))
3265 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3266 if (unlikely(ret < 0))
3269 if (skb_is_gso(skb)) {
3270 struct skb_shared_info *shinfo = skb_shinfo(skb);
3272 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */
3273 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3274 shinfo->gso_type &= ~SKB_GSO_TCPV6;
3275 shinfo->gso_type |= SKB_GSO_TCPV4;
3279 skb->protocol = htons(ETH_P_IP);
3280 skb_clear_hash(skb);
3285 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3287 __be16 from_proto = skb->protocol;
3289 if (from_proto == htons(ETH_P_IP) &&
3290 to_proto == htons(ETH_P_IPV6))
3291 return bpf_skb_proto_4_to_6(skb);
3293 if (from_proto == htons(ETH_P_IPV6) &&
3294 to_proto == htons(ETH_P_IP))
3295 return bpf_skb_proto_6_to_4(skb);
3300 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3305 if (unlikely(flags))
3308 /* General idea is that this helper does the basic groundwork
3309 * needed for changing the protocol, and eBPF program fills the
3310 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3311 * and other helpers, rather than passing a raw buffer here.
3313 * The rationale is to keep this minimal and without a need to
3314 * deal with raw packet data. F.e. even if we would pass buffers
3315 * here, the program still needs to call the bpf_lX_csum_replace()
3316 * helpers anyway. Plus, this way we keep also separation of
3317 * concerns, since f.e. bpf_skb_store_bytes() should only take
3320 * Currently, additional options and extension header space are
3321 * not supported, but flags register is reserved so we can adapt
3322 * that. For offloads, we mark packet as dodgy, so that headers
3323 * need to be verified first.
3325 ret = bpf_skb_proto_xlat(skb, proto);
3326 bpf_compute_data_pointers(skb);
3330 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3331 .func = bpf_skb_change_proto,
3333 .ret_type = RET_INTEGER,
3334 .arg1_type = ARG_PTR_TO_CTX,
3335 .arg2_type = ARG_ANYTHING,
3336 .arg3_type = ARG_ANYTHING,
3339 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3341 /* We only allow a restricted subset to be changed for now. */
3342 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3343 !skb_pkt_type_ok(pkt_type)))
3346 skb->pkt_type = pkt_type;
3350 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3351 .func = bpf_skb_change_type,
3353 .ret_type = RET_INTEGER,
3354 .arg1_type = ARG_PTR_TO_CTX,
3355 .arg2_type = ARG_ANYTHING,
3358 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3360 switch (skb->protocol) {
3361 case htons(ETH_P_IP):
3362 return sizeof(struct iphdr);
3363 case htons(ETH_P_IPV6):
3364 return sizeof(struct ipv6hdr);
3370 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3371 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3373 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \
3374 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3375 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3376 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3377 BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \
3378 BPF_F_ADJ_ROOM_ENCAP_L2( \
3379 BPF_ADJ_ROOM_ENCAP_L2_MASK))
3381 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3384 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3385 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3386 u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3387 unsigned int gso_type = SKB_GSO_DODGY;
3390 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3391 /* udp gso_size delineates datagrams, only allow if fixed */
3392 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3393 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3397 ret = skb_cow_head(skb, len_diff);
3398 if (unlikely(ret < 0))
3402 if (skb->protocol != htons(ETH_P_IP) &&
3403 skb->protocol != htons(ETH_P_IPV6))
3406 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3407 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3410 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3411 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3414 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH &&
3415 inner_mac_len < ETH_HLEN)
3418 if (skb->encapsulation)
3421 mac_len = skb->network_header - skb->mac_header;
3422 inner_net = skb->network_header;
3423 if (inner_mac_len > len_diff)
3425 inner_trans = skb->transport_header;
3428 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3429 if (unlikely(ret < 0))
3433 skb->inner_mac_header = inner_net - inner_mac_len;
3434 skb->inner_network_header = inner_net;
3435 skb->inner_transport_header = inner_trans;
3437 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH)
3438 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
3440 skb_set_inner_protocol(skb, skb->protocol);
3442 skb->encapsulation = 1;
3443 skb_set_network_header(skb, mac_len);
3445 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3446 gso_type |= SKB_GSO_UDP_TUNNEL;
3447 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3448 gso_type |= SKB_GSO_GRE;
3449 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3450 gso_type |= SKB_GSO_IPXIP6;
3451 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3452 gso_type |= SKB_GSO_IPXIP4;
3454 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3455 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3456 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3457 sizeof(struct ipv6hdr) :
3458 sizeof(struct iphdr);
3460 skb_set_transport_header(skb, mac_len + nh_len);
3463 /* Match skb->protocol to new outer l3 protocol */
3464 if (skb->protocol == htons(ETH_P_IP) &&
3465 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3466 skb->protocol = htons(ETH_P_IPV6);
3467 else if (skb->protocol == htons(ETH_P_IPV6) &&
3468 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3469 skb->protocol = htons(ETH_P_IP);
3472 if (skb_is_gso(skb)) {
3473 struct skb_shared_info *shinfo = skb_shinfo(skb);
3475 /* Due to header grow, MSS needs to be downgraded. */
3476 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3477 skb_decrease_gso_size(shinfo, len_diff);
3479 /* Header must be checked, and gso_segs recomputed. */
3480 shinfo->gso_type |= gso_type;
3481 shinfo->gso_segs = 0;
3487 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3492 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3493 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3496 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3497 /* udp gso_size delineates datagrams, only allow if fixed */
3498 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3499 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3503 ret = skb_unclone(skb, GFP_ATOMIC);
3504 if (unlikely(ret < 0))
3507 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3508 if (unlikely(ret < 0))
3511 if (skb_is_gso(skb)) {
3512 struct skb_shared_info *shinfo = skb_shinfo(skb);
3514 /* Due to header shrink, MSS can be upgraded. */
3515 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3516 skb_increase_gso_size(shinfo, len_diff);
3518 /* Header must be checked, and gso_segs recomputed. */
3519 shinfo->gso_type |= SKB_GSO_DODGY;
3520 shinfo->gso_segs = 0;
3526 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3528 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3529 u32, mode, u64, flags)
3531 u32 len_diff_abs = abs(len_diff);
3532 bool shrink = len_diff < 0;
3535 if (unlikely(flags || mode))
3537 if (unlikely(len_diff_abs > 0xfffU))
3541 ret = skb_cow(skb, len_diff);
3542 if (unlikely(ret < 0))
3544 __skb_push(skb, len_diff_abs);
3545 memset(skb->data, 0, len_diff_abs);
3547 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3549 __skb_pull(skb, len_diff_abs);
3551 if (tls_sw_has_ctx_rx(skb->sk)) {
3552 struct strp_msg *rxm = strp_msg(skb);
3554 rxm->full_len += len_diff;
3559 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3560 .func = sk_skb_adjust_room,
3562 .ret_type = RET_INTEGER,
3563 .arg1_type = ARG_PTR_TO_CTX,
3564 .arg2_type = ARG_ANYTHING,
3565 .arg3_type = ARG_ANYTHING,
3566 .arg4_type = ARG_ANYTHING,
3569 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3570 u32, mode, u64, flags)
3572 u32 len_cur, len_diff_abs = abs(len_diff);
3573 u32 len_min = bpf_skb_net_base_len(skb);
3574 u32 len_max = BPF_SKB_MAX_LEN;
3575 __be16 proto = skb->protocol;
3576 bool shrink = len_diff < 0;
3580 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3581 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3583 if (unlikely(len_diff_abs > 0xfffU))
3585 if (unlikely(proto != htons(ETH_P_IP) &&
3586 proto != htons(ETH_P_IPV6)))
3589 off = skb_mac_header_len(skb);
3591 case BPF_ADJ_ROOM_NET:
3592 off += bpf_skb_net_base_len(skb);
3594 case BPF_ADJ_ROOM_MAC:
3600 len_cur = skb->len - skb_network_offset(skb);
3601 if ((shrink && (len_diff_abs >= len_cur ||
3602 len_cur - len_diff_abs < len_min)) ||
3603 (!shrink && (skb->len + len_diff_abs > len_max &&
3607 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3608 bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3609 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3610 __skb_reset_checksum_unnecessary(skb);
3612 bpf_compute_data_pointers(skb);
3616 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3617 .func = bpf_skb_adjust_room,
3619 .ret_type = RET_INTEGER,
3620 .arg1_type = ARG_PTR_TO_CTX,
3621 .arg2_type = ARG_ANYTHING,
3622 .arg3_type = ARG_ANYTHING,
3623 .arg4_type = ARG_ANYTHING,
3626 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3628 u32 min_len = skb_network_offset(skb);
3630 if (skb_transport_header_was_set(skb))
3631 min_len = skb_transport_offset(skb);
3632 if (skb->ip_summed == CHECKSUM_PARTIAL)
3633 min_len = skb_checksum_start_offset(skb) +
3634 skb->csum_offset + sizeof(__sum16);
3638 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3640 unsigned int old_len = skb->len;
3643 ret = __skb_grow_rcsum(skb, new_len);
3645 memset(skb->data + old_len, 0, new_len - old_len);
3649 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3651 return __skb_trim_rcsum(skb, new_len);
3654 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3657 u32 max_len = BPF_SKB_MAX_LEN;
3658 u32 min_len = __bpf_skb_min_len(skb);
3661 if (unlikely(flags || new_len > max_len || new_len < min_len))
3663 if (skb->encapsulation)
3666 /* The basic idea of this helper is that it's performing the
3667 * needed work to either grow or trim an skb, and eBPF program
3668 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3669 * bpf_lX_csum_replace() and others rather than passing a raw
3670 * buffer here. This one is a slow path helper and intended
3671 * for replies with control messages.
3673 * Like in bpf_skb_change_proto(), we want to keep this rather
3674 * minimal and without protocol specifics so that we are able
3675 * to separate concerns as in bpf_skb_store_bytes() should only
3676 * be the one responsible for writing buffers.
3678 * It's really expected to be a slow path operation here for
3679 * control message replies, so we're implicitly linearizing,
3680 * uncloning and drop offloads from the skb by this.
3682 ret = __bpf_try_make_writable(skb, skb->len);
3684 if (new_len > skb->len)
3685 ret = bpf_skb_grow_rcsum(skb, new_len);
3686 else if (new_len < skb->len)
3687 ret = bpf_skb_trim_rcsum(skb, new_len);
3688 if (!ret && skb_is_gso(skb))
3694 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3697 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3699 bpf_compute_data_pointers(skb);
3703 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3704 .func = bpf_skb_change_tail,
3706 .ret_type = RET_INTEGER,
3707 .arg1_type = ARG_PTR_TO_CTX,
3708 .arg2_type = ARG_ANYTHING,
3709 .arg3_type = ARG_ANYTHING,
3712 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3715 return __bpf_skb_change_tail(skb, new_len, flags);
3718 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3719 .func = sk_skb_change_tail,
3721 .ret_type = RET_INTEGER,
3722 .arg1_type = ARG_PTR_TO_CTX,
3723 .arg2_type = ARG_ANYTHING,
3724 .arg3_type = ARG_ANYTHING,
3727 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3730 u32 max_len = BPF_SKB_MAX_LEN;
3731 u32 new_len = skb->len + head_room;
3734 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3735 new_len < skb->len))
3738 ret = skb_cow(skb, head_room);
3740 /* Idea for this helper is that we currently only
3741 * allow to expand on mac header. This means that
3742 * skb->protocol network header, etc, stay as is.
3743 * Compared to bpf_skb_change_tail(), we're more
3744 * flexible due to not needing to linearize or
3745 * reset GSO. Intention for this helper is to be
3746 * used by an L3 skb that needs to push mac header
3747 * for redirection into L2 device.
3749 __skb_push(skb, head_room);
3750 memset(skb->data, 0, head_room);
3751 skb_reset_mac_header(skb);
3752 skb_reset_mac_len(skb);
3758 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3761 int ret = __bpf_skb_change_head(skb, head_room, flags);
3763 bpf_compute_data_pointers(skb);
3767 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3768 .func = bpf_skb_change_head,
3770 .ret_type = RET_INTEGER,
3771 .arg1_type = ARG_PTR_TO_CTX,
3772 .arg2_type = ARG_ANYTHING,
3773 .arg3_type = ARG_ANYTHING,
3776 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3779 return __bpf_skb_change_head(skb, head_room, flags);
3782 static const struct bpf_func_proto sk_skb_change_head_proto = {
3783 .func = sk_skb_change_head,
3785 .ret_type = RET_INTEGER,
3786 .arg1_type = ARG_PTR_TO_CTX,
3787 .arg2_type = ARG_ANYTHING,
3788 .arg3_type = ARG_ANYTHING,
3790 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3792 return xdp_data_meta_unsupported(xdp) ? 0 :
3793 xdp->data - xdp->data_meta;
3796 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3798 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3799 unsigned long metalen = xdp_get_metalen(xdp);
3800 void *data_start = xdp_frame_end + metalen;
3801 void *data = xdp->data + offset;
3803 if (unlikely(data < data_start ||
3804 data > xdp->data_end - ETH_HLEN))
3808 memmove(xdp->data_meta + offset,
3809 xdp->data_meta, metalen);
3810 xdp->data_meta += offset;
3816 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3817 .func = bpf_xdp_adjust_head,
3819 .ret_type = RET_INTEGER,
3820 .arg1_type = ARG_PTR_TO_CTX,
3821 .arg2_type = ARG_ANYTHING,
3824 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3826 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
3827 void *data_end = xdp->data_end + offset;
3829 /* Notice that xdp_data_hard_end have reserved some tailroom */
3830 if (unlikely(data_end > data_hard_end))
3833 /* ALL drivers MUST init xdp->frame_sz, chicken check below */
3834 if (unlikely(xdp->frame_sz > PAGE_SIZE)) {
3835 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz);
3839 if (unlikely(data_end < xdp->data + ETH_HLEN))
3842 /* Clear memory area on grow, can contain uninit kernel memory */
3844 memset(xdp->data_end, 0, offset);
3846 xdp->data_end = data_end;
3851 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3852 .func = bpf_xdp_adjust_tail,
3854 .ret_type = RET_INTEGER,
3855 .arg1_type = ARG_PTR_TO_CTX,
3856 .arg2_type = ARG_ANYTHING,
3859 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3861 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3862 void *meta = xdp->data_meta + offset;
3863 unsigned long metalen = xdp->data - meta;
3865 if (xdp_data_meta_unsupported(xdp))
3867 if (unlikely(meta < xdp_frame_end ||
3870 if (unlikely(xdp_metalen_invalid(metalen)))
3873 xdp->data_meta = meta;
3878 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3879 .func = bpf_xdp_adjust_meta,
3881 .ret_type = RET_INTEGER,
3882 .arg1_type = ARG_PTR_TO_CTX,
3883 .arg2_type = ARG_ANYTHING,
3886 /* XDP_REDIRECT works by a three-step process, implemented in the functions
3889 * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target
3890 * of the redirect and store it (along with some other metadata) in a per-CPU
3891 * struct bpf_redirect_info.
3893 * 2. When the program returns the XDP_REDIRECT return code, the driver will
3894 * call xdp_do_redirect() which will use the information in struct
3895 * bpf_redirect_info to actually enqueue the frame into a map type-specific
3896 * bulk queue structure.
3898 * 3. Before exiting its NAPI poll loop, the driver will call xdp_do_flush(),
3899 * which will flush all the different bulk queues, thus completing the
3902 * Pointers to the map entries will be kept around for this whole sequence of
3903 * steps, protected by RCU. However, there is no top-level rcu_read_lock() in
3904 * the core code; instead, the RCU protection relies on everything happening
3905 * inside a single NAPI poll sequence, which means it's between a pair of calls
3906 * to local_bh_disable()/local_bh_enable().
3908 * The map entries are marked as __rcu and the map code makes sure to
3909 * dereference those pointers with rcu_dereference_check() in a way that works
3910 * for both sections that to hold an rcu_read_lock() and sections that are
3911 * called from NAPI without a separate rcu_read_lock(). The code below does not
3912 * use RCU annotations, but relies on those in the map code.
3914 void xdp_do_flush(void)
3920 EXPORT_SYMBOL_GPL(xdp_do_flush);
3922 void bpf_clear_redirect_map(struct bpf_map *map)
3924 struct bpf_redirect_info *ri;
3927 for_each_possible_cpu(cpu) {
3928 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3929 /* Avoid polluting remote cacheline due to writes if
3930 * not needed. Once we pass this test, we need the
3931 * cmpxchg() to make sure it hasn't been changed in
3932 * the meantime by remote CPU.
3934 if (unlikely(READ_ONCE(ri->map) == map))
3935 cmpxchg(&ri->map, map, NULL);
3939 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
3940 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key);
3942 u32 xdp_master_redirect(struct xdp_buff *xdp)
3944 struct net_device *master, *slave;
3945 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3947 master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev);
3948 slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp);
3949 if (slave && slave != xdp->rxq->dev) {
3950 /* The target device is different from the receiving device, so
3951 * redirect it to the new device.
3952 * Using XDP_REDIRECT gets the correct behaviour from XDP enabled
3953 * drivers to unmap the packet from their rx ring.
3955 ri->tgt_index = slave->ifindex;
3956 ri->map_id = INT_MAX;
3957 ri->map_type = BPF_MAP_TYPE_UNSPEC;
3958 return XDP_REDIRECT;
3962 EXPORT_SYMBOL_GPL(xdp_master_redirect);
3964 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3965 struct bpf_prog *xdp_prog)
3967 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3968 enum bpf_map_type map_type = ri->map_type;
3969 void *fwd = ri->tgt_value;
3970 u32 map_id = ri->map_id;
3971 struct bpf_map *map;
3974 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
3975 ri->map_type = BPF_MAP_TYPE_UNSPEC;
3978 case BPF_MAP_TYPE_DEVMAP:
3980 case BPF_MAP_TYPE_DEVMAP_HASH:
3981 map = READ_ONCE(ri->map);
3982 if (unlikely(map)) {
3983 WRITE_ONCE(ri->map, NULL);
3984 err = dev_map_enqueue_multi(xdp, dev, map,
3985 ri->flags & BPF_F_EXCLUDE_INGRESS);
3987 err = dev_map_enqueue(fwd, xdp, dev);
3990 case BPF_MAP_TYPE_CPUMAP:
3991 err = cpu_map_enqueue(fwd, xdp, dev);
3993 case BPF_MAP_TYPE_XSKMAP:
3994 err = __xsk_map_redirect(fwd, xdp);
3996 case BPF_MAP_TYPE_UNSPEC:
3997 if (map_id == INT_MAX) {
3998 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
3999 if (unlikely(!fwd)) {
4003 err = dev_xdp_enqueue(fwd, xdp, dev);
4014 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4017 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4020 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4022 static int xdp_do_generic_redirect_map(struct net_device *dev,
4023 struct sk_buff *skb,
4024 struct xdp_buff *xdp,
4025 struct bpf_prog *xdp_prog,
4027 enum bpf_map_type map_type, u32 map_id)
4029 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4030 struct bpf_map *map;
4034 case BPF_MAP_TYPE_DEVMAP:
4036 case BPF_MAP_TYPE_DEVMAP_HASH:
4037 map = READ_ONCE(ri->map);
4038 if (unlikely(map)) {
4039 WRITE_ONCE(ri->map, NULL);
4040 err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4041 ri->flags & BPF_F_EXCLUDE_INGRESS);
4043 err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4048 case BPF_MAP_TYPE_XSKMAP:
4049 err = xsk_generic_rcv(fwd, xdp);
4054 case BPF_MAP_TYPE_CPUMAP:
4055 err = cpu_map_generic_redirect(fwd, skb);
4064 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4067 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4071 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4072 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4074 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4075 enum bpf_map_type map_type = ri->map_type;
4076 void *fwd = ri->tgt_value;
4077 u32 map_id = ri->map_id;
4080 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4081 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4083 if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4084 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4085 if (unlikely(!fwd)) {
4090 err = xdp_ok_fwd_dev(fwd, skb->len);
4095 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4096 generic_xdp_tx(skb, xdp_prog);
4100 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id);
4102 _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4106 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4108 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4110 if (unlikely(flags))
4113 /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4114 * by map_idr) is used for ifindex based XDP redirect.
4116 ri->tgt_index = ifindex;
4117 ri->map_id = INT_MAX;
4118 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4120 return XDP_REDIRECT;
4123 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4124 .func = bpf_xdp_redirect,
4126 .ret_type = RET_INTEGER,
4127 .arg1_type = ARG_ANYTHING,
4128 .arg2_type = ARG_ANYTHING,
4131 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
4134 return map->ops->map_redirect(map, ifindex, flags);
4137 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4138 .func = bpf_xdp_redirect_map,
4140 .ret_type = RET_INTEGER,
4141 .arg1_type = ARG_CONST_MAP_PTR,
4142 .arg2_type = ARG_ANYTHING,
4143 .arg3_type = ARG_ANYTHING,
4146 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4147 unsigned long off, unsigned long len)
4149 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4153 if (ptr != dst_buff)
4154 memcpy(dst_buff, ptr, len);
4159 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4160 u64, flags, void *, meta, u64, meta_size)
4162 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4164 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4166 if (unlikely(!skb || skb_size > skb->len))
4169 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4173 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4174 .func = bpf_skb_event_output,
4176 .ret_type = RET_INTEGER,
4177 .arg1_type = ARG_PTR_TO_CTX,
4178 .arg2_type = ARG_CONST_MAP_PTR,
4179 .arg3_type = ARG_ANYTHING,
4180 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4181 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4184 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4186 const struct bpf_func_proto bpf_skb_output_proto = {
4187 .func = bpf_skb_event_output,
4189 .ret_type = RET_INTEGER,
4190 .arg1_type = ARG_PTR_TO_BTF_ID,
4191 .arg1_btf_id = &bpf_skb_output_btf_ids[0],
4192 .arg2_type = ARG_CONST_MAP_PTR,
4193 .arg3_type = ARG_ANYTHING,
4194 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4195 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4198 static unsigned short bpf_tunnel_key_af(u64 flags)
4200 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4203 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4204 u32, size, u64, flags)
4206 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4207 u8 compat[sizeof(struct bpf_tunnel_key)];
4211 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
4215 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4219 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4222 case offsetof(struct bpf_tunnel_key, tunnel_label):
4223 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4225 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4226 /* Fixup deprecated structure layouts here, so we have
4227 * a common path later on.
4229 if (ip_tunnel_info_af(info) != AF_INET)
4232 to = (struct bpf_tunnel_key *)compat;
4239 to->tunnel_id = be64_to_cpu(info->key.tun_id);
4240 to->tunnel_tos = info->key.tos;
4241 to->tunnel_ttl = info->key.ttl;
4244 if (flags & BPF_F_TUNINFO_IPV6) {
4245 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4246 sizeof(to->remote_ipv6));
4247 to->tunnel_label = be32_to_cpu(info->key.label);
4249 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4250 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4251 to->tunnel_label = 0;
4254 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4255 memcpy(to_orig, to, size);
4259 memset(to_orig, 0, size);
4263 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4264 .func = bpf_skb_get_tunnel_key,
4266 .ret_type = RET_INTEGER,
4267 .arg1_type = ARG_PTR_TO_CTX,
4268 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4269 .arg3_type = ARG_CONST_SIZE,
4270 .arg4_type = ARG_ANYTHING,
4273 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4275 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4278 if (unlikely(!info ||
4279 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4283 if (unlikely(size < info->options_len)) {
4288 ip_tunnel_info_opts_get(to, info);
4289 if (size > info->options_len)
4290 memset(to + info->options_len, 0, size - info->options_len);
4292 return info->options_len;
4294 memset(to, 0, size);
4298 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4299 .func = bpf_skb_get_tunnel_opt,
4301 .ret_type = RET_INTEGER,
4302 .arg1_type = ARG_PTR_TO_CTX,
4303 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4304 .arg3_type = ARG_CONST_SIZE,
4307 static struct metadata_dst __percpu *md_dst;
4309 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4310 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4312 struct metadata_dst *md = this_cpu_ptr(md_dst);
4313 u8 compat[sizeof(struct bpf_tunnel_key)];
4314 struct ip_tunnel_info *info;
4316 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4317 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
4319 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4321 case offsetof(struct bpf_tunnel_key, tunnel_label):
4322 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4323 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4324 /* Fixup deprecated structure layouts here, so we have
4325 * a common path later on.
4327 memcpy(compat, from, size);
4328 memset(compat + size, 0, sizeof(compat) - size);
4329 from = (const struct bpf_tunnel_key *) compat;
4335 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4340 dst_hold((struct dst_entry *) md);
4341 skb_dst_set(skb, (struct dst_entry *) md);
4343 info = &md->u.tun_info;
4344 memset(info, 0, sizeof(*info));
4345 info->mode = IP_TUNNEL_INFO_TX;
4347 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4348 if (flags & BPF_F_DONT_FRAGMENT)
4349 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4350 if (flags & BPF_F_ZERO_CSUM_TX)
4351 info->key.tun_flags &= ~TUNNEL_CSUM;
4352 if (flags & BPF_F_SEQ_NUMBER)
4353 info->key.tun_flags |= TUNNEL_SEQ;
4355 info->key.tun_id = cpu_to_be64(from->tunnel_id);
4356 info->key.tos = from->tunnel_tos;
4357 info->key.ttl = from->tunnel_ttl;
4359 if (flags & BPF_F_TUNINFO_IPV6) {
4360 info->mode |= IP_TUNNEL_INFO_IPV6;
4361 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4362 sizeof(from->remote_ipv6));
4363 info->key.label = cpu_to_be32(from->tunnel_label) &
4364 IPV6_FLOWLABEL_MASK;
4366 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4372 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4373 .func = bpf_skb_set_tunnel_key,
4375 .ret_type = RET_INTEGER,
4376 .arg1_type = ARG_PTR_TO_CTX,
4377 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4378 .arg3_type = ARG_CONST_SIZE,
4379 .arg4_type = ARG_ANYTHING,
4382 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4383 const u8 *, from, u32, size)
4385 struct ip_tunnel_info *info = skb_tunnel_info(skb);
4386 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4388 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4390 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4393 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4398 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4399 .func = bpf_skb_set_tunnel_opt,
4401 .ret_type = RET_INTEGER,
4402 .arg1_type = ARG_PTR_TO_CTX,
4403 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4404 .arg3_type = ARG_CONST_SIZE,
4407 static const struct bpf_func_proto *
4408 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4411 struct metadata_dst __percpu *tmp;
4413 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4418 if (cmpxchg(&md_dst, NULL, tmp))
4419 metadata_dst_free_percpu(tmp);
4423 case BPF_FUNC_skb_set_tunnel_key:
4424 return &bpf_skb_set_tunnel_key_proto;
4425 case BPF_FUNC_skb_set_tunnel_opt:
4426 return &bpf_skb_set_tunnel_opt_proto;
4432 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4435 struct bpf_array *array = container_of(map, struct bpf_array, map);
4436 struct cgroup *cgrp;
4439 sk = skb_to_full_sk(skb);
4440 if (!sk || !sk_fullsock(sk))
4442 if (unlikely(idx >= array->map.max_entries))
4445 cgrp = READ_ONCE(array->ptrs[idx]);
4446 if (unlikely(!cgrp))
4449 return sk_under_cgroup_hierarchy(sk, cgrp);
4452 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4453 .func = bpf_skb_under_cgroup,
4455 .ret_type = RET_INTEGER,
4456 .arg1_type = ARG_PTR_TO_CTX,
4457 .arg2_type = ARG_CONST_MAP_PTR,
4458 .arg3_type = ARG_ANYTHING,
4461 #ifdef CONFIG_SOCK_CGROUP_DATA
4462 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4464 struct cgroup *cgrp;
4466 sk = sk_to_full_sk(sk);
4467 if (!sk || !sk_fullsock(sk))
4470 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4471 return cgroup_id(cgrp);
4474 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4476 return __bpf_sk_cgroup_id(skb->sk);
4479 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4480 .func = bpf_skb_cgroup_id,
4482 .ret_type = RET_INTEGER,
4483 .arg1_type = ARG_PTR_TO_CTX,
4486 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4489 struct cgroup *ancestor;
4490 struct cgroup *cgrp;
4492 sk = sk_to_full_sk(sk);
4493 if (!sk || !sk_fullsock(sk))
4496 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4497 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4501 return cgroup_id(ancestor);
4504 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4507 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4510 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4511 .func = bpf_skb_ancestor_cgroup_id,
4513 .ret_type = RET_INTEGER,
4514 .arg1_type = ARG_PTR_TO_CTX,
4515 .arg2_type = ARG_ANYTHING,
4518 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4520 return __bpf_sk_cgroup_id(sk);
4523 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4524 .func = bpf_sk_cgroup_id,
4526 .ret_type = RET_INTEGER,
4527 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4530 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4532 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4535 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4536 .func = bpf_sk_ancestor_cgroup_id,
4538 .ret_type = RET_INTEGER,
4539 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4540 .arg2_type = ARG_ANYTHING,
4544 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4545 unsigned long off, unsigned long len)
4547 memcpy(dst_buff, src_buff + off, len);
4551 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4552 u64, flags, void *, meta, u64, meta_size)
4554 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4556 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4558 if (unlikely(!xdp ||
4559 xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4562 return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4563 xdp_size, bpf_xdp_copy);
4566 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4567 .func = bpf_xdp_event_output,
4569 .ret_type = RET_INTEGER,
4570 .arg1_type = ARG_PTR_TO_CTX,
4571 .arg2_type = ARG_CONST_MAP_PTR,
4572 .arg3_type = ARG_ANYTHING,
4573 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4574 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4577 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4579 const struct bpf_func_proto bpf_xdp_output_proto = {
4580 .func = bpf_xdp_event_output,
4582 .ret_type = RET_INTEGER,
4583 .arg1_type = ARG_PTR_TO_BTF_ID,
4584 .arg1_btf_id = &bpf_xdp_output_btf_ids[0],
4585 .arg2_type = ARG_CONST_MAP_PTR,
4586 .arg3_type = ARG_ANYTHING,
4587 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4588 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4591 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4593 return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4596 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4597 .func = bpf_get_socket_cookie,
4599 .ret_type = RET_INTEGER,
4600 .arg1_type = ARG_PTR_TO_CTX,
4603 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4605 return __sock_gen_cookie(ctx->sk);
4608 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4609 .func = bpf_get_socket_cookie_sock_addr,
4611 .ret_type = RET_INTEGER,
4612 .arg1_type = ARG_PTR_TO_CTX,
4615 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4617 return __sock_gen_cookie(ctx);
4620 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4621 .func = bpf_get_socket_cookie_sock,
4623 .ret_type = RET_INTEGER,
4624 .arg1_type = ARG_PTR_TO_CTX,
4627 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
4629 return sk ? sock_gen_cookie(sk) : 0;
4632 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
4633 .func = bpf_get_socket_ptr_cookie,
4635 .ret_type = RET_INTEGER,
4636 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4639 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4641 return __sock_gen_cookie(ctx->sk);
4644 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4645 .func = bpf_get_socket_cookie_sock_ops,
4647 .ret_type = RET_INTEGER,
4648 .arg1_type = ARG_PTR_TO_CTX,
4651 static u64 __bpf_get_netns_cookie(struct sock *sk)
4653 const struct net *net = sk ? sock_net(sk) : &init_net;
4655 return net->net_cookie;
4658 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
4660 return __bpf_get_netns_cookie(ctx);
4663 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
4664 .func = bpf_get_netns_cookie_sock,
4666 .ret_type = RET_INTEGER,
4667 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4670 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4672 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4675 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
4676 .func = bpf_get_netns_cookie_sock_addr,
4678 .ret_type = RET_INTEGER,
4679 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4682 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4684 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4687 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = {
4688 .func = bpf_get_netns_cookie_sock_ops,
4690 .ret_type = RET_INTEGER,
4691 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4694 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx)
4696 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4699 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = {
4700 .func = bpf_get_netns_cookie_sk_msg,
4702 .ret_type = RET_INTEGER,
4703 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4706 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4708 struct sock *sk = sk_to_full_sk(skb->sk);
4711 if (!sk || !sk_fullsock(sk))
4713 kuid = sock_net_uid(sock_net(sk), sk);
4714 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4717 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4718 .func = bpf_get_socket_uid,
4720 .ret_type = RET_INTEGER,
4721 .arg1_type = ARG_PTR_TO_CTX,
4724 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
4725 char *optval, int optlen)
4727 char devname[IFNAMSIZ];
4733 if (!sk_fullsock(sk))
4736 sock_owned_by_me(sk);
4738 if (level == SOL_SOCKET) {
4739 if (optlen != sizeof(int) && optname != SO_BINDTODEVICE)
4741 val = *((int *)optval);
4742 valbool = val ? 1 : 0;
4744 /* Only some socketops are supported */
4747 val = min_t(u32, val, sysctl_rmem_max);
4748 val = min_t(int, val, INT_MAX / 2);
4749 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4750 WRITE_ONCE(sk->sk_rcvbuf,
4751 max_t(int, val * 2, SOCK_MIN_RCVBUF));
4754 val = min_t(u32, val, sysctl_wmem_max);
4755 val = min_t(int, val, INT_MAX / 2);
4756 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4757 WRITE_ONCE(sk->sk_sndbuf,
4758 max_t(int, val * 2, SOCK_MIN_SNDBUF));
4760 case SO_MAX_PACING_RATE: /* 32bit version */
4762 cmpxchg(&sk->sk_pacing_status,
4765 sk->sk_max_pacing_rate = (val == ~0U) ?
4766 ~0UL : (unsigned int)val;
4767 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4768 sk->sk_max_pacing_rate);
4771 sk->sk_priority = val;
4776 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
4779 if (sk->sk_mark != val) {
4784 case SO_BINDTODEVICE:
4785 optlen = min_t(long, optlen, IFNAMSIZ - 1);
4786 strncpy(devname, optval, optlen);
4787 devname[optlen] = 0;
4790 if (devname[0] != '\0') {
4791 struct net_device *dev;
4796 dev = dev_get_by_name(net, devname);
4799 ifindex = dev->ifindex;
4803 case SO_BINDTOIFINDEX:
4804 if (optname == SO_BINDTOIFINDEX)
4806 ret = sock_bindtoindex(sk, ifindex, false);
4809 if (sk->sk_prot->keepalive)
4810 sk->sk_prot->keepalive(sk, valbool);
4811 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
4814 sk->sk_reuseport = valbool;
4820 } else if (level == SOL_IP) {
4821 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4824 val = *((int *)optval);
4825 /* Only some options are supported */
4828 if (val < -1 || val > 0xff) {
4831 struct inet_sock *inet = inet_sk(sk);
4841 #if IS_ENABLED(CONFIG_IPV6)
4842 } else if (level == SOL_IPV6) {
4843 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4846 val = *((int *)optval);
4847 /* Only some options are supported */
4850 if (val < -1 || val > 0xff) {
4853 struct ipv6_pinfo *np = inet6_sk(sk);
4864 } else if (level == SOL_TCP &&
4865 sk->sk_prot->setsockopt == tcp_setsockopt) {
4866 if (optname == TCP_CONGESTION) {
4867 char name[TCP_CA_NAME_MAX];
4869 strncpy(name, optval, min_t(long, optlen,
4870 TCP_CA_NAME_MAX-1));
4871 name[TCP_CA_NAME_MAX-1] = 0;
4872 ret = tcp_set_congestion_control(sk, name, false, true);
4874 struct inet_connection_sock *icsk = inet_csk(sk);
4875 struct tcp_sock *tp = tcp_sk(sk);
4876 unsigned long timeout;
4878 if (optlen != sizeof(int))
4881 val = *((int *)optval);
4882 /* Only some options are supported */
4885 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4888 tcp_snd_cwnd_set(tp, val);
4890 case TCP_BPF_SNDCWND_CLAMP:
4894 tp->snd_cwnd_clamp = val;
4895 tp->snd_ssthresh = val;
4898 case TCP_BPF_DELACK_MAX:
4899 timeout = usecs_to_jiffies(val);
4900 if (timeout > TCP_DELACK_MAX ||
4901 timeout < TCP_TIMEOUT_MIN)
4903 inet_csk(sk)->icsk_delack_max = timeout;
4905 case TCP_BPF_RTO_MIN:
4906 timeout = usecs_to_jiffies(val);
4907 if (timeout > TCP_RTO_MIN ||
4908 timeout < TCP_TIMEOUT_MIN)
4910 inet_csk(sk)->icsk_rto_min = timeout;
4913 if (val < 0 || val > 1)
4919 ret = tcp_sock_set_keepidle_locked(sk, val);
4922 if (val < 1 || val > MAX_TCP_KEEPINTVL)
4925 tp->keepalive_intvl = val * HZ;
4928 if (val < 1 || val > MAX_TCP_KEEPCNT)
4931 tp->keepalive_probes = val;
4934 if (val < 1 || val > MAX_TCP_SYNCNT)
4937 icsk->icsk_syn_retries = val;
4939 case TCP_USER_TIMEOUT:
4943 icsk->icsk_user_timeout = val;
4945 case TCP_NOTSENT_LOWAT:
4946 tp->notsent_lowat = val;
4947 sk->sk_write_space(sk);
4949 case TCP_WINDOW_CLAMP:
4950 ret = tcp_set_window_clamp(sk, val);
4963 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
4964 char *optval, int optlen)
4966 if (!sk_fullsock(sk))
4969 sock_owned_by_me(sk);
4971 if (level == SOL_SOCKET) {
4972 if (optlen != sizeof(int))
4977 *((int *)optval) = sk->sk_mark;
4980 *((int *)optval) = sk->sk_priority;
4982 case SO_BINDTOIFINDEX:
4983 *((int *)optval) = sk->sk_bound_dev_if;
4986 *((int *)optval) = sk->sk_reuseport;
4992 } else if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4993 struct inet_connection_sock *icsk;
4994 struct tcp_sock *tp;
4997 case TCP_CONGESTION:
4998 icsk = inet_csk(sk);
5000 if (!icsk->icsk_ca_ops || optlen <= 1)
5002 strncpy(optval, icsk->icsk_ca_ops->name, optlen);
5003 optval[optlen - 1] = 0;
5008 if (optlen <= 0 || !tp->saved_syn ||
5009 optlen > tcp_saved_syn_len(tp->saved_syn))
5011 memcpy(optval, tp->saved_syn->data, optlen);
5016 } else if (level == SOL_IP) {
5017 struct inet_sock *inet = inet_sk(sk);
5019 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
5022 /* Only some options are supported */
5025 *((int *)optval) = (int)inet->tos;
5030 #if IS_ENABLED(CONFIG_IPV6)
5031 } else if (level == SOL_IPV6) {
5032 struct ipv6_pinfo *np = inet6_sk(sk);
5034 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
5037 /* Only some options are supported */
5040 *((int *)optval) = (int)np->tclass;
5052 memset(optval, 0, optlen);
5056 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level,
5057 int, optname, char *, optval, int, optlen)
5059 if (level == SOL_TCP && optname == TCP_CONGESTION) {
5060 if (optlen >= sizeof("cdg") - 1 &&
5061 !strncmp("cdg", optval, optlen))
5065 return _bpf_setsockopt(sk, level, optname, optval, optlen);
5068 const struct bpf_func_proto bpf_sk_setsockopt_proto = {
5069 .func = bpf_sk_setsockopt,
5071 .ret_type = RET_INTEGER,
5072 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5073 .arg2_type = ARG_ANYTHING,
5074 .arg3_type = ARG_ANYTHING,
5075 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5076 .arg5_type = ARG_CONST_SIZE,
5079 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level,
5080 int, optname, char *, optval, int, optlen)
5082 return _bpf_getsockopt(sk, level, optname, optval, optlen);
5085 const struct bpf_func_proto bpf_sk_getsockopt_proto = {
5086 .func = bpf_sk_getsockopt,
5088 .ret_type = RET_INTEGER,
5089 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5090 .arg2_type = ARG_ANYTHING,
5091 .arg3_type = ARG_ANYTHING,
5092 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5093 .arg5_type = ARG_CONST_SIZE,
5096 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5097 int, level, int, optname, char *, optval, int, optlen)
5099 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5102 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5103 .func = bpf_sock_addr_setsockopt,
5105 .ret_type = RET_INTEGER,
5106 .arg1_type = ARG_PTR_TO_CTX,
5107 .arg2_type = ARG_ANYTHING,
5108 .arg3_type = ARG_ANYTHING,
5109 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5110 .arg5_type = ARG_CONST_SIZE,
5113 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5114 int, level, int, optname, char *, optval, int, optlen)
5116 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5119 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5120 .func = bpf_sock_addr_getsockopt,
5122 .ret_type = RET_INTEGER,
5123 .arg1_type = ARG_PTR_TO_CTX,
5124 .arg2_type = ARG_ANYTHING,
5125 .arg3_type = ARG_ANYTHING,
5126 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5127 .arg5_type = ARG_CONST_SIZE,
5130 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5131 int, level, int, optname, char *, optval, int, optlen)
5133 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5136 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5137 .func = bpf_sock_ops_setsockopt,
5139 .ret_type = RET_INTEGER,
5140 .arg1_type = ARG_PTR_TO_CTX,
5141 .arg2_type = ARG_ANYTHING,
5142 .arg3_type = ARG_ANYTHING,
5143 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5144 .arg5_type = ARG_CONST_SIZE,
5147 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5148 int optname, const u8 **start)
5150 struct sk_buff *syn_skb = bpf_sock->syn_skb;
5151 const u8 *hdr_start;
5155 /* sk is a request_sock here */
5157 if (optname == TCP_BPF_SYN) {
5158 hdr_start = syn_skb->data;
5159 ret = tcp_hdrlen(syn_skb);
5160 } else if (optname == TCP_BPF_SYN_IP) {
5161 hdr_start = skb_network_header(syn_skb);
5162 ret = skb_network_header_len(syn_skb) +
5163 tcp_hdrlen(syn_skb);
5165 /* optname == TCP_BPF_SYN_MAC */
5166 hdr_start = skb_mac_header(syn_skb);
5167 ret = skb_mac_header_len(syn_skb) +
5168 skb_network_header_len(syn_skb) +
5169 tcp_hdrlen(syn_skb);
5172 struct sock *sk = bpf_sock->sk;
5173 struct saved_syn *saved_syn;
5175 if (sk->sk_state == TCP_NEW_SYN_RECV)
5176 /* synack retransmit. bpf_sock->syn_skb will
5177 * not be available. It has to resort to
5178 * saved_syn (if it is saved).
5180 saved_syn = inet_reqsk(sk)->saved_syn;
5182 saved_syn = tcp_sk(sk)->saved_syn;
5187 if (optname == TCP_BPF_SYN) {
5188 hdr_start = saved_syn->data +
5189 saved_syn->mac_hdrlen +
5190 saved_syn->network_hdrlen;
5191 ret = saved_syn->tcp_hdrlen;
5192 } else if (optname == TCP_BPF_SYN_IP) {
5193 hdr_start = saved_syn->data +
5194 saved_syn->mac_hdrlen;
5195 ret = saved_syn->network_hdrlen +
5196 saved_syn->tcp_hdrlen;
5198 /* optname == TCP_BPF_SYN_MAC */
5200 /* TCP_SAVE_SYN may not have saved the mac hdr */
5201 if (!saved_syn->mac_hdrlen)
5204 hdr_start = saved_syn->data;
5205 ret = saved_syn->mac_hdrlen +
5206 saved_syn->network_hdrlen +
5207 saved_syn->tcp_hdrlen;
5215 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5216 int, level, int, optname, char *, optval, int, optlen)
5218 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5219 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5220 int ret, copy_len = 0;
5223 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5226 if (optlen < copy_len) {
5231 memcpy(optval, start, copy_len);
5234 /* Zero out unused buffer at the end */
5235 memset(optval + copy_len, 0, optlen - copy_len);
5240 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5243 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5244 .func = bpf_sock_ops_getsockopt,
5246 .ret_type = RET_INTEGER,
5247 .arg1_type = ARG_PTR_TO_CTX,
5248 .arg2_type = ARG_ANYTHING,
5249 .arg3_type = ARG_ANYTHING,
5250 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5251 .arg5_type = ARG_CONST_SIZE,
5254 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5257 struct sock *sk = bpf_sock->sk;
5258 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5260 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5263 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5265 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5268 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5269 .func = bpf_sock_ops_cb_flags_set,
5271 .ret_type = RET_INTEGER,
5272 .arg1_type = ARG_PTR_TO_CTX,
5273 .arg2_type = ARG_ANYTHING,
5276 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5277 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5279 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5283 struct sock *sk = ctx->sk;
5284 u32 flags = BIND_FROM_BPF;
5288 if (addr_len < offsetofend(struct sockaddr, sa_family))
5290 if (addr->sa_family == AF_INET) {
5291 if (addr_len < sizeof(struct sockaddr_in))
5293 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5294 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5295 return __inet_bind(sk, addr, addr_len, flags);
5296 #if IS_ENABLED(CONFIG_IPV6)
5297 } else if (addr->sa_family == AF_INET6) {
5298 if (addr_len < SIN6_LEN_RFC2133)
5300 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5301 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5302 /* ipv6_bpf_stub cannot be NULL, since it's called from
5303 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5305 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5306 #endif /* CONFIG_IPV6 */
5308 #endif /* CONFIG_INET */
5310 return -EAFNOSUPPORT;
5313 static const struct bpf_func_proto bpf_bind_proto = {
5316 .ret_type = RET_INTEGER,
5317 .arg1_type = ARG_PTR_TO_CTX,
5318 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5319 .arg3_type = ARG_CONST_SIZE,
5323 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5324 struct bpf_xfrm_state *, to, u32, size, u64, flags)
5326 const struct sec_path *sp = skb_sec_path(skb);
5327 const struct xfrm_state *x;
5329 if (!sp || unlikely(index >= sp->len || flags))
5332 x = sp->xvec[index];
5334 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5337 to->reqid = x->props.reqid;
5338 to->spi = x->id.spi;
5339 to->family = x->props.family;
5342 if (to->family == AF_INET6) {
5343 memcpy(to->remote_ipv6, x->props.saddr.a6,
5344 sizeof(to->remote_ipv6));
5346 to->remote_ipv4 = x->props.saddr.a4;
5347 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5352 memset(to, 0, size);
5356 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5357 .func = bpf_skb_get_xfrm_state,
5359 .ret_type = RET_INTEGER,
5360 .arg1_type = ARG_PTR_TO_CTX,
5361 .arg2_type = ARG_ANYTHING,
5362 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
5363 .arg4_type = ARG_CONST_SIZE,
5364 .arg5_type = ARG_ANYTHING,
5368 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5369 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
5370 const struct neighbour *neigh,
5371 const struct net_device *dev, u32 mtu)
5373 memcpy(params->dmac, neigh->ha, ETH_ALEN);
5374 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5375 params->h_vlan_TCI = 0;
5376 params->h_vlan_proto = 0;
5378 params->mtu_result = mtu; /* union with tot_len */
5384 #if IS_ENABLED(CONFIG_INET)
5385 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5386 u32 flags, bool check_mtu)
5388 struct fib_nh_common *nhc;
5389 struct in_device *in_dev;
5390 struct neighbour *neigh;
5391 struct net_device *dev;
5392 struct fib_result res;
5397 dev = dev_get_by_index_rcu(net, params->ifindex);
5401 /* verify forwarding is enabled on this interface */
5402 in_dev = __in_dev_get_rcu(dev);
5403 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5404 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5406 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5408 fl4.flowi4_oif = params->ifindex;
5410 fl4.flowi4_iif = params->ifindex;
5413 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5414 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5415 fl4.flowi4_flags = 0;
5417 fl4.flowi4_proto = params->l4_protocol;
5418 fl4.daddr = params->ipv4_dst;
5419 fl4.saddr = params->ipv4_src;
5420 fl4.fl4_sport = params->sport;
5421 fl4.fl4_dport = params->dport;
5422 fl4.flowi4_multipath_hash = 0;
5424 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5425 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5426 struct fib_table *tb;
5428 tb = fib_get_table(net, tbid);
5430 return BPF_FIB_LKUP_RET_NOT_FWDED;
5432 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5434 fl4.flowi4_mark = 0;
5435 fl4.flowi4_secid = 0;
5436 fl4.flowi4_tun_key.tun_id = 0;
5437 fl4.flowi4_uid = sock_net_uid(net, NULL);
5439 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5443 /* map fib lookup errors to RTN_ type */
5445 return BPF_FIB_LKUP_RET_BLACKHOLE;
5446 if (err == -EHOSTUNREACH)
5447 return BPF_FIB_LKUP_RET_UNREACHABLE;
5449 return BPF_FIB_LKUP_RET_PROHIBIT;
5451 return BPF_FIB_LKUP_RET_NOT_FWDED;
5454 if (res.type != RTN_UNICAST)
5455 return BPF_FIB_LKUP_RET_NOT_FWDED;
5457 if (fib_info_num_path(res.fi) > 1)
5458 fib_select_path(net, &res, &fl4, NULL);
5461 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5462 if (params->tot_len > mtu) {
5463 params->mtu_result = mtu; /* union with tot_len */
5464 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5470 /* do not handle lwt encaps right now */
5471 if (nhc->nhc_lwtstate)
5472 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5476 params->rt_metric = res.fi->fib_priority;
5477 params->ifindex = dev->ifindex;
5479 /* xdp and cls_bpf programs are run in RCU-bh so
5480 * rcu_read_lock_bh is not needed here
5482 if (likely(nhc->nhc_gw_family != AF_INET6)) {
5483 if (nhc->nhc_gw_family)
5484 params->ipv4_dst = nhc->nhc_gw.ipv4;
5486 neigh = __ipv4_neigh_lookup_noref(dev,
5487 (__force u32)params->ipv4_dst);
5489 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5491 params->family = AF_INET6;
5492 *dst = nhc->nhc_gw.ipv6;
5493 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5497 return BPF_FIB_LKUP_RET_NO_NEIGH;
5499 return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5503 #if IS_ENABLED(CONFIG_IPV6)
5504 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5505 u32 flags, bool check_mtu)
5507 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5508 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5509 struct fib6_result res = {};
5510 struct neighbour *neigh;
5511 struct net_device *dev;
5512 struct inet6_dev *idev;
5518 /* link local addresses are never forwarded */
5519 if (rt6_need_strict(dst) || rt6_need_strict(src))
5520 return BPF_FIB_LKUP_RET_NOT_FWDED;
5522 dev = dev_get_by_index_rcu(net, params->ifindex);
5526 idev = __in6_dev_get_safely(dev);
5527 if (unlikely(!idev || !idev->cnf.forwarding))
5528 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5530 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5532 oif = fl6.flowi6_oif = params->ifindex;
5534 oif = fl6.flowi6_iif = params->ifindex;
5536 strict = RT6_LOOKUP_F_HAS_SADDR;
5538 fl6.flowlabel = params->flowinfo;
5539 fl6.flowi6_scope = 0;
5540 fl6.flowi6_flags = 0;
5543 fl6.flowi6_proto = params->l4_protocol;
5546 fl6.fl6_sport = params->sport;
5547 fl6.fl6_dport = params->dport;
5549 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5550 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5551 struct fib6_table *tb;
5553 tb = ipv6_stub->fib6_get_table(net, tbid);
5555 return BPF_FIB_LKUP_RET_NOT_FWDED;
5557 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5560 fl6.flowi6_mark = 0;
5561 fl6.flowi6_secid = 0;
5562 fl6.flowi6_tun_key.tun_id = 0;
5563 fl6.flowi6_uid = sock_net_uid(net, NULL);
5565 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5568 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5569 res.f6i == net->ipv6.fib6_null_entry))
5570 return BPF_FIB_LKUP_RET_NOT_FWDED;
5572 switch (res.fib6_type) {
5573 /* only unicast is forwarded */
5577 return BPF_FIB_LKUP_RET_BLACKHOLE;
5578 case RTN_UNREACHABLE:
5579 return BPF_FIB_LKUP_RET_UNREACHABLE;
5581 return BPF_FIB_LKUP_RET_PROHIBIT;
5583 return BPF_FIB_LKUP_RET_NOT_FWDED;
5586 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5587 fl6.flowi6_oif != 0, NULL, strict);
5590 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5591 if (params->tot_len > mtu) {
5592 params->mtu_result = mtu; /* union with tot_len */
5593 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5597 if (res.nh->fib_nh_lws)
5598 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5600 if (res.nh->fib_nh_gw_family)
5601 *dst = res.nh->fib_nh_gw6;
5603 dev = res.nh->fib_nh_dev;
5604 params->rt_metric = res.f6i->fib6_metric;
5605 params->ifindex = dev->ifindex;
5607 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
5610 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5612 return BPF_FIB_LKUP_RET_NO_NEIGH;
5614 return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5618 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
5619 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5621 if (plen < sizeof(*params))
5624 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5627 switch (params->family) {
5628 #if IS_ENABLED(CONFIG_INET)
5630 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
5633 #if IS_ENABLED(CONFIG_IPV6)
5635 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
5639 return -EAFNOSUPPORT;
5642 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
5643 .func = bpf_xdp_fib_lookup,
5645 .ret_type = RET_INTEGER,
5646 .arg1_type = ARG_PTR_TO_CTX,
5647 .arg2_type = ARG_PTR_TO_MEM,
5648 .arg3_type = ARG_CONST_SIZE,
5649 .arg4_type = ARG_ANYTHING,
5652 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
5653 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5655 struct net *net = dev_net(skb->dev);
5656 int rc = -EAFNOSUPPORT;
5657 bool check_mtu = false;
5659 if (plen < sizeof(*params))
5662 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5665 if (params->tot_len)
5668 switch (params->family) {
5669 #if IS_ENABLED(CONFIG_INET)
5671 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
5674 #if IS_ENABLED(CONFIG_IPV6)
5676 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
5681 if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
5682 struct net_device *dev;
5684 /* When tot_len isn't provided by user, check skb
5685 * against MTU of FIB lookup resulting net_device
5687 dev = dev_get_by_index_rcu(net, params->ifindex);
5688 if (!is_skb_forwardable(dev, skb))
5689 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
5691 params->mtu_result = dev->mtu; /* union with tot_len */
5697 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
5698 .func = bpf_skb_fib_lookup,
5700 .ret_type = RET_INTEGER,
5701 .arg1_type = ARG_PTR_TO_CTX,
5702 .arg2_type = ARG_PTR_TO_MEM,
5703 .arg3_type = ARG_CONST_SIZE,
5704 .arg4_type = ARG_ANYTHING,
5707 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
5710 struct net *netns = dev_net(dev_curr);
5712 /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
5716 return dev_get_by_index_rcu(netns, ifindex);
5719 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
5720 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5722 int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5723 struct net_device *dev = skb->dev;
5724 int skb_len, dev_len;
5727 if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
5730 if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
5733 dev = __dev_via_ifindex(dev, ifindex);
5737 mtu = READ_ONCE(dev->mtu);
5739 dev_len = mtu + dev->hard_header_len;
5741 /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
5742 skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
5744 skb_len += len_diff; /* minus result pass check */
5745 if (skb_len <= dev_len) {
5746 ret = BPF_MTU_CHK_RET_SUCCESS;
5749 /* At this point, skb->len exceed MTU, but as it include length of all
5750 * segments, it can still be below MTU. The SKB can possibly get
5751 * re-segmented in transmit path (see validate_xmit_skb). Thus, user
5752 * must choose if segs are to be MTU checked.
5754 if (skb_is_gso(skb)) {
5755 ret = BPF_MTU_CHK_RET_SUCCESS;
5757 if (flags & BPF_MTU_CHK_SEGS &&
5758 !skb_gso_validate_network_len(skb, mtu))
5759 ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
5762 /* BPF verifier guarantees valid pointer */
5768 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
5769 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5771 struct net_device *dev = xdp->rxq->dev;
5772 int xdp_len = xdp->data_end - xdp->data;
5773 int ret = BPF_MTU_CHK_RET_SUCCESS;
5776 /* XDP variant doesn't support multi-buffer segment check (yet) */
5777 if (unlikely(flags))
5780 dev = __dev_via_ifindex(dev, ifindex);
5784 mtu = READ_ONCE(dev->mtu);
5786 /* Add L2-header as dev MTU is L3 size */
5787 dev_len = mtu + dev->hard_header_len;
5789 /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
5791 xdp_len = *mtu_len + dev->hard_header_len;
5793 xdp_len += len_diff; /* minus result pass check */
5794 if (xdp_len > dev_len)
5795 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5797 /* BPF verifier guarantees valid pointer */
5803 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
5804 .func = bpf_skb_check_mtu,
5806 .ret_type = RET_INTEGER,
5807 .arg1_type = ARG_PTR_TO_CTX,
5808 .arg2_type = ARG_ANYTHING,
5809 .arg3_type = ARG_PTR_TO_INT,
5810 .arg4_type = ARG_ANYTHING,
5811 .arg5_type = ARG_ANYTHING,
5814 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
5815 .func = bpf_xdp_check_mtu,
5817 .ret_type = RET_INTEGER,
5818 .arg1_type = ARG_PTR_TO_CTX,
5819 .arg2_type = ARG_ANYTHING,
5820 .arg3_type = ARG_PTR_TO_INT,
5821 .arg4_type = ARG_ANYTHING,
5822 .arg5_type = ARG_ANYTHING,
5825 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5826 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
5829 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
5831 if (!seg6_validate_srh(srh, len, false))
5835 case BPF_LWT_ENCAP_SEG6_INLINE:
5836 if (skb->protocol != htons(ETH_P_IPV6))
5839 err = seg6_do_srh_inline(skb, srh);
5841 case BPF_LWT_ENCAP_SEG6:
5842 skb_reset_inner_headers(skb);
5843 skb->encapsulation = 1;
5844 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
5850 bpf_compute_data_pointers(skb);
5854 ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5855 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
5857 return seg6_lookup_nexthop(skb, NULL, 0);
5859 #endif /* CONFIG_IPV6_SEG6_BPF */
5861 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5862 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
5865 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
5869 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
5873 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5874 case BPF_LWT_ENCAP_SEG6:
5875 case BPF_LWT_ENCAP_SEG6_INLINE:
5876 return bpf_push_seg6_encap(skb, type, hdr, len);
5878 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5879 case BPF_LWT_ENCAP_IP:
5880 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
5887 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
5888 void *, hdr, u32, len)
5891 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5892 case BPF_LWT_ENCAP_IP:
5893 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
5900 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
5901 .func = bpf_lwt_in_push_encap,
5903 .ret_type = RET_INTEGER,
5904 .arg1_type = ARG_PTR_TO_CTX,
5905 .arg2_type = ARG_ANYTHING,
5906 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5907 .arg4_type = ARG_CONST_SIZE
5910 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
5911 .func = bpf_lwt_xmit_push_encap,
5913 .ret_type = RET_INTEGER,
5914 .arg1_type = ARG_PTR_TO_CTX,
5915 .arg2_type = ARG_ANYTHING,
5916 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5917 .arg4_type = ARG_CONST_SIZE
5920 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5921 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
5922 const void *, from, u32, len)
5924 struct seg6_bpf_srh_state *srh_state =
5925 this_cpu_ptr(&seg6_bpf_srh_states);
5926 struct ipv6_sr_hdr *srh = srh_state->srh;
5927 void *srh_tlvs, *srh_end, *ptr;
5933 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
5934 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
5936 ptr = skb->data + offset;
5937 if (ptr >= srh_tlvs && ptr + len <= srh_end)
5938 srh_state->valid = false;
5939 else if (ptr < (void *)&srh->flags ||
5940 ptr + len > (void *)&srh->segments)
5943 if (unlikely(bpf_try_make_writable(skb, offset + len)))
5945 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5947 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5949 memcpy(skb->data + offset, from, len);
5953 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
5954 .func = bpf_lwt_seg6_store_bytes,
5956 .ret_type = RET_INTEGER,
5957 .arg1_type = ARG_PTR_TO_CTX,
5958 .arg2_type = ARG_ANYTHING,
5959 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5960 .arg4_type = ARG_CONST_SIZE
5963 static void bpf_update_srh_state(struct sk_buff *skb)
5965 struct seg6_bpf_srh_state *srh_state =
5966 this_cpu_ptr(&seg6_bpf_srh_states);
5969 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
5970 srh_state->srh = NULL;
5972 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5973 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
5974 srh_state->valid = true;
5978 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
5979 u32, action, void *, param, u32, param_len)
5981 struct seg6_bpf_srh_state *srh_state =
5982 this_cpu_ptr(&seg6_bpf_srh_states);
5987 case SEG6_LOCAL_ACTION_END_X:
5988 if (!seg6_bpf_has_valid_srh(skb))
5990 if (param_len != sizeof(struct in6_addr))
5992 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
5993 case SEG6_LOCAL_ACTION_END_T:
5994 if (!seg6_bpf_has_valid_srh(skb))
5996 if (param_len != sizeof(int))
5998 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5999 case SEG6_LOCAL_ACTION_END_DT6:
6000 if (!seg6_bpf_has_valid_srh(skb))
6002 if (param_len != sizeof(int))
6005 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
6007 if (!pskb_pull(skb, hdroff))
6010 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
6011 skb_reset_network_header(skb);
6012 skb_reset_transport_header(skb);
6013 skb->encapsulation = 0;
6015 bpf_compute_data_pointers(skb);
6016 bpf_update_srh_state(skb);
6017 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6018 case SEG6_LOCAL_ACTION_END_B6:
6019 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6021 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
6024 bpf_update_srh_state(skb);
6027 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
6028 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6030 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
6033 bpf_update_srh_state(skb);
6041 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
6042 .func = bpf_lwt_seg6_action,
6044 .ret_type = RET_INTEGER,
6045 .arg1_type = ARG_PTR_TO_CTX,
6046 .arg2_type = ARG_ANYTHING,
6047 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6048 .arg4_type = ARG_CONST_SIZE
6051 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
6054 struct seg6_bpf_srh_state *srh_state =
6055 this_cpu_ptr(&seg6_bpf_srh_states);
6056 struct ipv6_sr_hdr *srh = srh_state->srh;
6057 void *srh_end, *srh_tlvs, *ptr;
6058 struct ipv6hdr *hdr;
6062 if (unlikely(srh == NULL))
6065 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
6066 ((srh->first_segment + 1) << 4));
6067 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
6069 ptr = skb->data + offset;
6071 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
6073 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
6077 ret = skb_cow_head(skb, len);
6078 if (unlikely(ret < 0))
6081 ret = bpf_skb_net_hdr_push(skb, offset, len);
6083 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6086 bpf_compute_data_pointers(skb);
6087 if (unlikely(ret < 0))
6090 hdr = (struct ipv6hdr *)skb->data;
6091 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6093 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6095 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6096 srh_state->hdrlen += len;
6097 srh_state->valid = false;
6101 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6102 .func = bpf_lwt_seg6_adjust_srh,
6104 .ret_type = RET_INTEGER,
6105 .arg1_type = ARG_PTR_TO_CTX,
6106 .arg2_type = ARG_ANYTHING,
6107 .arg3_type = ARG_ANYTHING,
6109 #endif /* CONFIG_IPV6_SEG6_BPF */
6112 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6113 int dif, int sdif, u8 family, u8 proto)
6115 bool refcounted = false;
6116 struct sock *sk = NULL;
6118 if (family == AF_INET) {
6119 __be32 src4 = tuple->ipv4.saddr;
6120 __be32 dst4 = tuple->ipv4.daddr;
6122 if (proto == IPPROTO_TCP)
6123 sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
6124 src4, tuple->ipv4.sport,
6125 dst4, tuple->ipv4.dport,
6126 dif, sdif, &refcounted);
6128 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6129 dst4, tuple->ipv4.dport,
6130 dif, sdif, &udp_table, NULL);
6131 #if IS_ENABLED(CONFIG_IPV6)
6133 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6134 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6136 if (proto == IPPROTO_TCP)
6137 sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
6138 src6, tuple->ipv6.sport,
6139 dst6, ntohs(tuple->ipv6.dport),
6140 dif, sdif, &refcounted);
6141 else if (likely(ipv6_bpf_stub))
6142 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6143 src6, tuple->ipv6.sport,
6144 dst6, tuple->ipv6.dport,
6150 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6151 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6157 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6158 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6159 * Returns the socket as an 'unsigned long' to simplify the casting in the
6160 * callers to satisfy BPF_CALL declarations.
6162 static struct sock *
6163 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6164 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6167 struct sock *sk = NULL;
6168 u8 family = AF_UNSPEC;
6172 if (len == sizeof(tuple->ipv4))
6174 else if (len == sizeof(tuple->ipv6))
6179 if (unlikely(family == AF_UNSPEC || flags ||
6180 !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6183 if (family == AF_INET)
6184 sdif = inet_sdif(skb);
6186 sdif = inet6_sdif(skb);
6188 if ((s32)netns_id < 0) {
6190 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6192 net = get_net_ns_by_id(caller_net, netns_id);
6195 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6203 static struct sock *
6204 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6205 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6208 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6209 ifindex, proto, netns_id, flags);
6212 sk = sk_to_full_sk(sk);
6213 if (!sk_fullsock(sk)) {
6222 static struct sock *
6223 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6224 u8 proto, u64 netns_id, u64 flags)
6226 struct net *caller_net;
6230 caller_net = dev_net(skb->dev);
6231 ifindex = skb->dev->ifindex;
6233 caller_net = sock_net(skb->sk);
6237 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6241 static struct sock *
6242 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6243 u8 proto, u64 netns_id, u64 flags)
6245 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6249 sk = sk_to_full_sk(sk);
6250 if (!sk_fullsock(sk)) {
6259 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6260 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6262 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6266 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6267 .func = bpf_skc_lookup_tcp,
6270 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6271 .arg1_type = ARG_PTR_TO_CTX,
6272 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6273 .arg3_type = ARG_CONST_SIZE,
6274 .arg4_type = ARG_ANYTHING,
6275 .arg5_type = ARG_ANYTHING,
6278 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6279 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6281 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6285 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6286 .func = bpf_sk_lookup_tcp,
6289 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6290 .arg1_type = ARG_PTR_TO_CTX,
6291 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6292 .arg3_type = ARG_CONST_SIZE,
6293 .arg4_type = ARG_ANYTHING,
6294 .arg5_type = ARG_ANYTHING,
6297 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6298 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6300 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6304 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6305 .func = bpf_sk_lookup_udp,
6308 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6309 .arg1_type = ARG_PTR_TO_CTX,
6310 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6311 .arg3_type = ARG_CONST_SIZE,
6312 .arg4_type = ARG_ANYTHING,
6313 .arg5_type = ARG_ANYTHING,
6316 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6318 if (sk && sk_is_refcounted(sk))
6323 static const struct bpf_func_proto bpf_sk_release_proto = {
6324 .func = bpf_sk_release,
6326 .ret_type = RET_INTEGER,
6327 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6330 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6331 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6333 struct net *caller_net = dev_net(ctx->rxq->dev);
6334 int ifindex = ctx->rxq->dev->ifindex;
6336 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6337 ifindex, IPPROTO_UDP, netns_id,
6341 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6342 .func = bpf_xdp_sk_lookup_udp,
6345 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6346 .arg1_type = ARG_PTR_TO_CTX,
6347 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6348 .arg3_type = ARG_CONST_SIZE,
6349 .arg4_type = ARG_ANYTHING,
6350 .arg5_type = ARG_ANYTHING,
6353 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6354 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6356 struct net *caller_net = dev_net(ctx->rxq->dev);
6357 int ifindex = ctx->rxq->dev->ifindex;
6359 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6360 ifindex, IPPROTO_TCP, netns_id,
6364 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6365 .func = bpf_xdp_skc_lookup_tcp,
6368 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6369 .arg1_type = ARG_PTR_TO_CTX,
6370 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6371 .arg3_type = ARG_CONST_SIZE,
6372 .arg4_type = ARG_ANYTHING,
6373 .arg5_type = ARG_ANYTHING,
6376 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6377 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6379 struct net *caller_net = dev_net(ctx->rxq->dev);
6380 int ifindex = ctx->rxq->dev->ifindex;
6382 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6383 ifindex, IPPROTO_TCP, netns_id,
6387 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6388 .func = bpf_xdp_sk_lookup_tcp,
6391 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6392 .arg1_type = ARG_PTR_TO_CTX,
6393 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6394 .arg3_type = ARG_CONST_SIZE,
6395 .arg4_type = ARG_ANYTHING,
6396 .arg5_type = ARG_ANYTHING,
6399 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6400 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6402 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6403 sock_net(ctx->sk), 0,
6404 IPPROTO_TCP, netns_id, flags);
6407 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6408 .func = bpf_sock_addr_skc_lookup_tcp,
6410 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6411 .arg1_type = ARG_PTR_TO_CTX,
6412 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6413 .arg3_type = ARG_CONST_SIZE,
6414 .arg4_type = ARG_ANYTHING,
6415 .arg5_type = ARG_ANYTHING,
6418 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6419 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6421 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6422 sock_net(ctx->sk), 0, IPPROTO_TCP,
6426 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6427 .func = bpf_sock_addr_sk_lookup_tcp,
6429 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6430 .arg1_type = ARG_PTR_TO_CTX,
6431 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6432 .arg3_type = ARG_CONST_SIZE,
6433 .arg4_type = ARG_ANYTHING,
6434 .arg5_type = ARG_ANYTHING,
6437 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6438 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6440 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6441 sock_net(ctx->sk), 0, IPPROTO_UDP,
6445 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6446 .func = bpf_sock_addr_sk_lookup_udp,
6448 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6449 .arg1_type = ARG_PTR_TO_CTX,
6450 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6451 .arg3_type = ARG_CONST_SIZE,
6452 .arg4_type = ARG_ANYTHING,
6453 .arg5_type = ARG_ANYTHING,
6456 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6457 struct bpf_insn_access_aux *info)
6459 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6463 if (off % size != 0)
6467 case offsetof(struct bpf_tcp_sock, bytes_received):
6468 case offsetof(struct bpf_tcp_sock, bytes_acked):
6469 return size == sizeof(__u64);
6471 return size == sizeof(__u32);
6475 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6476 const struct bpf_insn *si,
6477 struct bpf_insn *insn_buf,
6478 struct bpf_prog *prog, u32 *target_size)
6480 struct bpf_insn *insn = insn_buf;
6482 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \
6484 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \
6485 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6486 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6487 si->dst_reg, si->src_reg, \
6488 offsetof(struct tcp_sock, FIELD)); \
6491 #define BPF_INET_SOCK_GET_COMMON(FIELD) \
6493 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \
6495 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6496 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6497 struct inet_connection_sock, \
6499 si->dst_reg, si->src_reg, \
6501 struct inet_connection_sock, \
6505 if (insn > insn_buf)
6506 return insn - insn_buf;
6509 case offsetof(struct bpf_tcp_sock, rtt_min):
6510 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
6511 sizeof(struct minmax));
6512 BUILD_BUG_ON(sizeof(struct minmax) <
6513 sizeof(struct minmax_sample));
6515 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6516 offsetof(struct tcp_sock, rtt_min) +
6517 offsetof(struct minmax_sample, v));
6519 case offsetof(struct bpf_tcp_sock, snd_cwnd):
6520 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
6522 case offsetof(struct bpf_tcp_sock, srtt_us):
6523 BPF_TCP_SOCK_GET_COMMON(srtt_us);
6525 case offsetof(struct bpf_tcp_sock, snd_ssthresh):
6526 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
6528 case offsetof(struct bpf_tcp_sock, rcv_nxt):
6529 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
6531 case offsetof(struct bpf_tcp_sock, snd_nxt):
6532 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
6534 case offsetof(struct bpf_tcp_sock, snd_una):
6535 BPF_TCP_SOCK_GET_COMMON(snd_una);
6537 case offsetof(struct bpf_tcp_sock, mss_cache):
6538 BPF_TCP_SOCK_GET_COMMON(mss_cache);
6540 case offsetof(struct bpf_tcp_sock, ecn_flags):
6541 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
6543 case offsetof(struct bpf_tcp_sock, rate_delivered):
6544 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
6546 case offsetof(struct bpf_tcp_sock, rate_interval_us):
6547 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
6549 case offsetof(struct bpf_tcp_sock, packets_out):
6550 BPF_TCP_SOCK_GET_COMMON(packets_out);
6552 case offsetof(struct bpf_tcp_sock, retrans_out):
6553 BPF_TCP_SOCK_GET_COMMON(retrans_out);
6555 case offsetof(struct bpf_tcp_sock, total_retrans):
6556 BPF_TCP_SOCK_GET_COMMON(total_retrans);
6558 case offsetof(struct bpf_tcp_sock, segs_in):
6559 BPF_TCP_SOCK_GET_COMMON(segs_in);
6561 case offsetof(struct bpf_tcp_sock, data_segs_in):
6562 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
6564 case offsetof(struct bpf_tcp_sock, segs_out):
6565 BPF_TCP_SOCK_GET_COMMON(segs_out);
6567 case offsetof(struct bpf_tcp_sock, data_segs_out):
6568 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
6570 case offsetof(struct bpf_tcp_sock, lost_out):
6571 BPF_TCP_SOCK_GET_COMMON(lost_out);
6573 case offsetof(struct bpf_tcp_sock, sacked_out):
6574 BPF_TCP_SOCK_GET_COMMON(sacked_out);
6576 case offsetof(struct bpf_tcp_sock, bytes_received):
6577 BPF_TCP_SOCK_GET_COMMON(bytes_received);
6579 case offsetof(struct bpf_tcp_sock, bytes_acked):
6580 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
6582 case offsetof(struct bpf_tcp_sock, dsack_dups):
6583 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
6585 case offsetof(struct bpf_tcp_sock, delivered):
6586 BPF_TCP_SOCK_GET_COMMON(delivered);
6588 case offsetof(struct bpf_tcp_sock, delivered_ce):
6589 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
6591 case offsetof(struct bpf_tcp_sock, icsk_retransmits):
6592 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
6596 return insn - insn_buf;
6599 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
6601 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
6602 return (unsigned long)sk;
6604 return (unsigned long)NULL;
6607 const struct bpf_func_proto bpf_tcp_sock_proto = {
6608 .func = bpf_tcp_sock,
6610 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
6611 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6614 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
6616 sk = sk_to_full_sk(sk);
6618 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
6619 return (unsigned long)sk;
6621 return (unsigned long)NULL;
6624 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
6625 .func = bpf_get_listener_sock,
6627 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6628 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6631 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
6633 unsigned int iphdr_len;
6635 switch (skb_protocol(skb, true)) {
6636 case cpu_to_be16(ETH_P_IP):
6637 iphdr_len = sizeof(struct iphdr);
6639 case cpu_to_be16(ETH_P_IPV6):
6640 iphdr_len = sizeof(struct ipv6hdr);
6646 if (skb_headlen(skb) < iphdr_len)
6649 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
6652 return INET_ECN_set_ce(skb);
6655 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6656 struct bpf_insn_access_aux *info)
6658 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
6661 if (off % size != 0)
6666 return size == sizeof(__u32);
6670 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
6671 const struct bpf_insn *si,
6672 struct bpf_insn *insn_buf,
6673 struct bpf_prog *prog, u32 *target_size)
6675 struct bpf_insn *insn = insn_buf;
6677 #define BPF_XDP_SOCK_GET(FIELD) \
6679 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \
6680 sizeof_field(struct bpf_xdp_sock, FIELD)); \
6681 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
6682 si->dst_reg, si->src_reg, \
6683 offsetof(struct xdp_sock, FIELD)); \
6687 case offsetof(struct bpf_xdp_sock, queue_id):
6688 BPF_XDP_SOCK_GET(queue_id);
6692 return insn - insn_buf;
6695 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
6696 .func = bpf_skb_ecn_set_ce,
6698 .ret_type = RET_INTEGER,
6699 .arg1_type = ARG_PTR_TO_CTX,
6702 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6703 struct tcphdr *, th, u32, th_len)
6705 #ifdef CONFIG_SYN_COOKIES
6709 if (unlikely(!sk || th_len < sizeof(*th)))
6712 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
6713 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6716 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
6719 if (!th->ack || th->rst || th->syn)
6722 if (unlikely(iph_len < sizeof(struct iphdr)))
6725 if (tcp_synq_no_recent_overflow(sk))
6728 cookie = ntohl(th->ack_seq) - 1;
6730 /* Both struct iphdr and struct ipv6hdr have the version field at the
6731 * same offset so we can cast to the shorter header (struct iphdr).
6733 switch (((struct iphdr *)iph)->version) {
6735 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
6738 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
6741 #if IS_BUILTIN(CONFIG_IPV6)
6743 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6746 if (sk->sk_family != AF_INET6)
6749 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
6751 #endif /* CONFIG_IPV6 */
6754 return -EPROTONOSUPPORT;
6766 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
6767 .func = bpf_tcp_check_syncookie,
6770 .ret_type = RET_INTEGER,
6771 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6772 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6773 .arg3_type = ARG_CONST_SIZE,
6774 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6775 .arg5_type = ARG_CONST_SIZE,
6778 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6779 struct tcphdr *, th, u32, th_len)
6781 #ifdef CONFIG_SYN_COOKIES
6785 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
6788 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6791 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
6794 if (!th->syn || th->ack || th->fin || th->rst)
6797 if (unlikely(iph_len < sizeof(struct iphdr)))
6800 /* Both struct iphdr and struct ipv6hdr have the version field at the
6801 * same offset so we can cast to the shorter header (struct iphdr).
6803 switch (((struct iphdr *)iph)->version) {
6805 if (sk->sk_family == AF_INET6 && sk->sk_ipv6only)
6808 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
6811 #if IS_BUILTIN(CONFIG_IPV6)
6813 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6816 if (sk->sk_family != AF_INET6)
6819 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
6821 #endif /* CONFIG_IPV6 */
6824 return -EPROTONOSUPPORT;
6829 return cookie | ((u64)mss << 32);
6832 #endif /* CONFIG_SYN_COOKIES */
6835 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
6836 .func = bpf_tcp_gen_syncookie,
6837 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */
6839 .ret_type = RET_INTEGER,
6840 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6841 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6842 .arg3_type = ARG_CONST_SIZE,
6843 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6844 .arg5_type = ARG_CONST_SIZE,
6847 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
6849 if (!sk || flags != 0)
6851 if (!skb_at_tc_ingress(skb))
6853 if (unlikely(dev_net(skb->dev) != sock_net(sk)))
6854 return -ENETUNREACH;
6855 if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
6856 return -ESOCKTNOSUPPORT;
6857 if (sk_is_refcounted(sk) &&
6858 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
6863 skb->destructor = sock_pfree;
6868 static const struct bpf_func_proto bpf_sk_assign_proto = {
6869 .func = bpf_sk_assign,
6871 .ret_type = RET_INTEGER,
6872 .arg1_type = ARG_PTR_TO_CTX,
6873 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6874 .arg3_type = ARG_ANYTHING,
6877 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
6878 u8 search_kind, const u8 *magic,
6879 u8 magic_len, bool *eol)
6885 while (op < opend) {
6888 if (kind == TCPOPT_EOL) {
6890 return ERR_PTR(-ENOMSG);
6891 } else if (kind == TCPOPT_NOP) {
6896 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
6897 /* Something is wrong in the received header.
6898 * Follow the TCP stack's tcp_parse_options()
6899 * and just bail here.
6901 return ERR_PTR(-EFAULT);
6904 if (search_kind == kind) {
6908 if (magic_len > kind_len - 2)
6909 return ERR_PTR(-ENOMSG);
6911 if (!memcmp(&op[2], magic, magic_len))
6918 return ERR_PTR(-ENOMSG);
6921 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6922 void *, search_res, u32, len, u64, flags)
6924 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
6925 const u8 *op, *opend, *magic, *search = search_res;
6926 u8 search_kind, search_len, copy_len, magic_len;
6929 /* 2 byte is the minimal option len except TCPOPT_NOP and
6930 * TCPOPT_EOL which are useless for the bpf prog to learn
6931 * and this helper disallow loading them also.
6933 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
6936 search_kind = search[0];
6937 search_len = search[1];
6939 if (search_len > len || search_kind == TCPOPT_NOP ||
6940 search_kind == TCPOPT_EOL)
6943 if (search_kind == TCPOPT_EXP || search_kind == 253) {
6944 /* 16 or 32 bit magic. +2 for kind and kind length */
6945 if (search_len != 4 && search_len != 6)
6948 magic_len = search_len - 2;
6957 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
6962 op += sizeof(struct tcphdr);
6964 if (!bpf_sock->skb ||
6965 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6966 /* This bpf_sock->op cannot call this helper */
6969 opend = bpf_sock->skb_data_end;
6970 op = bpf_sock->skb->data + sizeof(struct tcphdr);
6973 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
6980 if (copy_len > len) {
6985 memcpy(search_res, op, copy_len);
6989 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
6990 .func = bpf_sock_ops_load_hdr_opt,
6992 .ret_type = RET_INTEGER,
6993 .arg1_type = ARG_PTR_TO_CTX,
6994 .arg2_type = ARG_PTR_TO_MEM,
6995 .arg3_type = ARG_CONST_SIZE,
6996 .arg4_type = ARG_ANYTHING,
6999 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7000 const void *, from, u32, len, u64, flags)
7002 u8 new_kind, new_kind_len, magic_len = 0, *opend;
7003 const u8 *op, *new_op, *magic = NULL;
7004 struct sk_buff *skb;
7007 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
7010 if (len < 2 || flags)
7014 new_kind = new_op[0];
7015 new_kind_len = new_op[1];
7017 if (new_kind_len > len || new_kind == TCPOPT_NOP ||
7018 new_kind == TCPOPT_EOL)
7021 if (new_kind_len > bpf_sock->remaining_opt_len)
7024 /* 253 is another experimental kind */
7025 if (new_kind == TCPOPT_EXP || new_kind == 253) {
7026 if (new_kind_len < 4)
7028 /* Match for the 2 byte magic also.
7029 * RFC 6994: the magic could be 2 or 4 bytes.
7030 * Hence, matching by 2 byte only is on the
7031 * conservative side but it is the right
7032 * thing to do for the 'search-for-duplication'
7039 /* Check for duplication */
7040 skb = bpf_sock->skb;
7041 op = skb->data + sizeof(struct tcphdr);
7042 opend = bpf_sock->skb_data_end;
7044 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
7049 if (PTR_ERR(op) != -ENOMSG)
7053 /* The option has been ended. Treat it as no more
7054 * header option can be written.
7058 /* No duplication found. Store the header option. */
7059 memcpy(opend, from, new_kind_len);
7061 bpf_sock->remaining_opt_len -= new_kind_len;
7062 bpf_sock->skb_data_end += new_kind_len;
7067 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
7068 .func = bpf_sock_ops_store_hdr_opt,
7070 .ret_type = RET_INTEGER,
7071 .arg1_type = ARG_PTR_TO_CTX,
7072 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7073 .arg3_type = ARG_CONST_SIZE,
7074 .arg4_type = ARG_ANYTHING,
7077 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7078 u32, len, u64, flags)
7080 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7083 if (flags || len < 2)
7086 if (len > bpf_sock->remaining_opt_len)
7089 bpf_sock->remaining_opt_len -= len;
7094 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7095 .func = bpf_sock_ops_reserve_hdr_opt,
7097 .ret_type = RET_INTEGER,
7098 .arg1_type = ARG_PTR_TO_CTX,
7099 .arg2_type = ARG_ANYTHING,
7100 .arg3_type = ARG_ANYTHING,
7103 #endif /* CONFIG_INET */
7105 bool bpf_helper_changes_pkt_data(void *func)
7107 if (func == bpf_skb_vlan_push ||
7108 func == bpf_skb_vlan_pop ||
7109 func == bpf_skb_store_bytes ||
7110 func == bpf_skb_change_proto ||
7111 func == bpf_skb_change_head ||
7112 func == sk_skb_change_head ||
7113 func == bpf_skb_change_tail ||
7114 func == sk_skb_change_tail ||
7115 func == bpf_skb_adjust_room ||
7116 func == sk_skb_adjust_room ||
7117 func == bpf_skb_pull_data ||
7118 func == sk_skb_pull_data ||
7119 func == bpf_clone_redirect ||
7120 func == bpf_l3_csum_replace ||
7121 func == bpf_l4_csum_replace ||
7122 func == bpf_xdp_adjust_head ||
7123 func == bpf_xdp_adjust_meta ||
7124 func == bpf_msg_pull_data ||
7125 func == bpf_msg_push_data ||
7126 func == bpf_msg_pop_data ||
7127 func == bpf_xdp_adjust_tail ||
7128 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7129 func == bpf_lwt_seg6_store_bytes ||
7130 func == bpf_lwt_seg6_adjust_srh ||
7131 func == bpf_lwt_seg6_action ||
7134 func == bpf_sock_ops_store_hdr_opt ||
7136 func == bpf_lwt_in_push_encap ||
7137 func == bpf_lwt_xmit_push_encap)
7143 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7144 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7146 static const struct bpf_func_proto *
7147 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7150 /* inet and inet6 sockets are created in a process
7151 * context so there is always a valid uid/gid
7153 case BPF_FUNC_get_current_uid_gid:
7154 return &bpf_get_current_uid_gid_proto;
7155 case BPF_FUNC_get_local_storage:
7156 return &bpf_get_local_storage_proto;
7157 case BPF_FUNC_get_socket_cookie:
7158 return &bpf_get_socket_cookie_sock_proto;
7159 case BPF_FUNC_get_netns_cookie:
7160 return &bpf_get_netns_cookie_sock_proto;
7161 case BPF_FUNC_perf_event_output:
7162 return &bpf_event_output_data_proto;
7163 case BPF_FUNC_get_current_pid_tgid:
7164 return &bpf_get_current_pid_tgid_proto;
7165 case BPF_FUNC_get_current_comm:
7166 return &bpf_get_current_comm_proto;
7167 #ifdef CONFIG_CGROUPS
7168 case BPF_FUNC_get_current_cgroup_id:
7169 return &bpf_get_current_cgroup_id_proto;
7170 case BPF_FUNC_get_current_ancestor_cgroup_id:
7171 return &bpf_get_current_ancestor_cgroup_id_proto;
7173 #ifdef CONFIG_CGROUP_NET_CLASSID
7174 case BPF_FUNC_get_cgroup_classid:
7175 return &bpf_get_cgroup_classid_curr_proto;
7177 case BPF_FUNC_sk_storage_get:
7178 return &bpf_sk_storage_get_cg_sock_proto;
7179 case BPF_FUNC_ktime_get_coarse_ns:
7180 return &bpf_ktime_get_coarse_ns_proto;
7182 return bpf_base_func_proto(func_id);
7186 static const struct bpf_func_proto *
7187 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7190 /* inet and inet6 sockets are created in a process
7191 * context so there is always a valid uid/gid
7193 case BPF_FUNC_get_current_uid_gid:
7194 return &bpf_get_current_uid_gid_proto;
7196 switch (prog->expected_attach_type) {
7197 case BPF_CGROUP_INET4_CONNECT:
7198 case BPF_CGROUP_INET6_CONNECT:
7199 return &bpf_bind_proto;
7203 case BPF_FUNC_get_socket_cookie:
7204 return &bpf_get_socket_cookie_sock_addr_proto;
7205 case BPF_FUNC_get_netns_cookie:
7206 return &bpf_get_netns_cookie_sock_addr_proto;
7207 case BPF_FUNC_get_local_storage:
7208 return &bpf_get_local_storage_proto;
7209 case BPF_FUNC_perf_event_output:
7210 return &bpf_event_output_data_proto;
7211 case BPF_FUNC_get_current_pid_tgid:
7212 return &bpf_get_current_pid_tgid_proto;
7213 case BPF_FUNC_get_current_comm:
7214 return &bpf_get_current_comm_proto;
7215 #ifdef CONFIG_CGROUPS
7216 case BPF_FUNC_get_current_cgroup_id:
7217 return &bpf_get_current_cgroup_id_proto;
7218 case BPF_FUNC_get_current_ancestor_cgroup_id:
7219 return &bpf_get_current_ancestor_cgroup_id_proto;
7221 #ifdef CONFIG_CGROUP_NET_CLASSID
7222 case BPF_FUNC_get_cgroup_classid:
7223 return &bpf_get_cgroup_classid_curr_proto;
7226 case BPF_FUNC_sk_lookup_tcp:
7227 return &bpf_sock_addr_sk_lookup_tcp_proto;
7228 case BPF_FUNC_sk_lookup_udp:
7229 return &bpf_sock_addr_sk_lookup_udp_proto;
7230 case BPF_FUNC_sk_release:
7231 return &bpf_sk_release_proto;
7232 case BPF_FUNC_skc_lookup_tcp:
7233 return &bpf_sock_addr_skc_lookup_tcp_proto;
7234 #endif /* CONFIG_INET */
7235 case BPF_FUNC_sk_storage_get:
7236 return &bpf_sk_storage_get_proto;
7237 case BPF_FUNC_sk_storage_delete:
7238 return &bpf_sk_storage_delete_proto;
7239 case BPF_FUNC_setsockopt:
7240 switch (prog->expected_attach_type) {
7241 case BPF_CGROUP_INET4_BIND:
7242 case BPF_CGROUP_INET6_BIND:
7243 case BPF_CGROUP_INET4_CONNECT:
7244 case BPF_CGROUP_INET6_CONNECT:
7245 case BPF_CGROUP_UDP4_RECVMSG:
7246 case BPF_CGROUP_UDP6_RECVMSG:
7247 case BPF_CGROUP_UDP4_SENDMSG:
7248 case BPF_CGROUP_UDP6_SENDMSG:
7249 case BPF_CGROUP_INET4_GETPEERNAME:
7250 case BPF_CGROUP_INET6_GETPEERNAME:
7251 case BPF_CGROUP_INET4_GETSOCKNAME:
7252 case BPF_CGROUP_INET6_GETSOCKNAME:
7253 return &bpf_sock_addr_setsockopt_proto;
7257 case BPF_FUNC_getsockopt:
7258 switch (prog->expected_attach_type) {
7259 case BPF_CGROUP_INET4_BIND:
7260 case BPF_CGROUP_INET6_BIND:
7261 case BPF_CGROUP_INET4_CONNECT:
7262 case BPF_CGROUP_INET6_CONNECT:
7263 case BPF_CGROUP_UDP4_RECVMSG:
7264 case BPF_CGROUP_UDP6_RECVMSG:
7265 case BPF_CGROUP_UDP4_SENDMSG:
7266 case BPF_CGROUP_UDP6_SENDMSG:
7267 case BPF_CGROUP_INET4_GETPEERNAME:
7268 case BPF_CGROUP_INET6_GETPEERNAME:
7269 case BPF_CGROUP_INET4_GETSOCKNAME:
7270 case BPF_CGROUP_INET6_GETSOCKNAME:
7271 return &bpf_sock_addr_getsockopt_proto;
7276 return bpf_sk_base_func_proto(func_id);
7280 static const struct bpf_func_proto *
7281 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7284 case BPF_FUNC_skb_load_bytes:
7285 return &bpf_skb_load_bytes_proto;
7286 case BPF_FUNC_skb_load_bytes_relative:
7287 return &bpf_skb_load_bytes_relative_proto;
7288 case BPF_FUNC_get_socket_cookie:
7289 return &bpf_get_socket_cookie_proto;
7290 case BPF_FUNC_get_socket_uid:
7291 return &bpf_get_socket_uid_proto;
7292 case BPF_FUNC_perf_event_output:
7293 return &bpf_skb_event_output_proto;
7295 return bpf_sk_base_func_proto(func_id);
7299 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7300 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7302 static const struct bpf_func_proto *
7303 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7306 case BPF_FUNC_get_local_storage:
7307 return &bpf_get_local_storage_proto;
7308 case BPF_FUNC_sk_fullsock:
7309 return &bpf_sk_fullsock_proto;
7310 case BPF_FUNC_sk_storage_get:
7311 return &bpf_sk_storage_get_proto;
7312 case BPF_FUNC_sk_storage_delete:
7313 return &bpf_sk_storage_delete_proto;
7314 case BPF_FUNC_perf_event_output:
7315 return &bpf_skb_event_output_proto;
7316 #ifdef CONFIG_SOCK_CGROUP_DATA
7317 case BPF_FUNC_skb_cgroup_id:
7318 return &bpf_skb_cgroup_id_proto;
7319 case BPF_FUNC_skb_ancestor_cgroup_id:
7320 return &bpf_skb_ancestor_cgroup_id_proto;
7321 case BPF_FUNC_sk_cgroup_id:
7322 return &bpf_sk_cgroup_id_proto;
7323 case BPF_FUNC_sk_ancestor_cgroup_id:
7324 return &bpf_sk_ancestor_cgroup_id_proto;
7327 case BPF_FUNC_sk_lookup_tcp:
7328 return &bpf_sk_lookup_tcp_proto;
7329 case BPF_FUNC_sk_lookup_udp:
7330 return &bpf_sk_lookup_udp_proto;
7331 case BPF_FUNC_sk_release:
7332 return &bpf_sk_release_proto;
7333 case BPF_FUNC_skc_lookup_tcp:
7334 return &bpf_skc_lookup_tcp_proto;
7335 case BPF_FUNC_tcp_sock:
7336 return &bpf_tcp_sock_proto;
7337 case BPF_FUNC_get_listener_sock:
7338 return &bpf_get_listener_sock_proto;
7339 case BPF_FUNC_skb_ecn_set_ce:
7340 return &bpf_skb_ecn_set_ce_proto;
7343 return sk_filter_func_proto(func_id, prog);
7347 static const struct bpf_func_proto *
7348 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7351 case BPF_FUNC_skb_store_bytes:
7352 return &bpf_skb_store_bytes_proto;
7353 case BPF_FUNC_skb_load_bytes:
7354 return &bpf_skb_load_bytes_proto;
7355 case BPF_FUNC_skb_load_bytes_relative:
7356 return &bpf_skb_load_bytes_relative_proto;
7357 case BPF_FUNC_skb_pull_data:
7358 return &bpf_skb_pull_data_proto;
7359 case BPF_FUNC_csum_diff:
7360 return &bpf_csum_diff_proto;
7361 case BPF_FUNC_csum_update:
7362 return &bpf_csum_update_proto;
7363 case BPF_FUNC_csum_level:
7364 return &bpf_csum_level_proto;
7365 case BPF_FUNC_l3_csum_replace:
7366 return &bpf_l3_csum_replace_proto;
7367 case BPF_FUNC_l4_csum_replace:
7368 return &bpf_l4_csum_replace_proto;
7369 case BPF_FUNC_clone_redirect:
7370 return &bpf_clone_redirect_proto;
7371 case BPF_FUNC_get_cgroup_classid:
7372 return &bpf_get_cgroup_classid_proto;
7373 case BPF_FUNC_skb_vlan_push:
7374 return &bpf_skb_vlan_push_proto;
7375 case BPF_FUNC_skb_vlan_pop:
7376 return &bpf_skb_vlan_pop_proto;
7377 case BPF_FUNC_skb_change_proto:
7378 return &bpf_skb_change_proto_proto;
7379 case BPF_FUNC_skb_change_type:
7380 return &bpf_skb_change_type_proto;
7381 case BPF_FUNC_skb_adjust_room:
7382 return &bpf_skb_adjust_room_proto;
7383 case BPF_FUNC_skb_change_tail:
7384 return &bpf_skb_change_tail_proto;
7385 case BPF_FUNC_skb_change_head:
7386 return &bpf_skb_change_head_proto;
7387 case BPF_FUNC_skb_get_tunnel_key:
7388 return &bpf_skb_get_tunnel_key_proto;
7389 case BPF_FUNC_skb_set_tunnel_key:
7390 return bpf_get_skb_set_tunnel_proto(func_id);
7391 case BPF_FUNC_skb_get_tunnel_opt:
7392 return &bpf_skb_get_tunnel_opt_proto;
7393 case BPF_FUNC_skb_set_tunnel_opt:
7394 return bpf_get_skb_set_tunnel_proto(func_id);
7395 case BPF_FUNC_redirect:
7396 return &bpf_redirect_proto;
7397 case BPF_FUNC_redirect_neigh:
7398 return &bpf_redirect_neigh_proto;
7399 case BPF_FUNC_redirect_peer:
7400 return &bpf_redirect_peer_proto;
7401 case BPF_FUNC_get_route_realm:
7402 return &bpf_get_route_realm_proto;
7403 case BPF_FUNC_get_hash_recalc:
7404 return &bpf_get_hash_recalc_proto;
7405 case BPF_FUNC_set_hash_invalid:
7406 return &bpf_set_hash_invalid_proto;
7407 case BPF_FUNC_set_hash:
7408 return &bpf_set_hash_proto;
7409 case BPF_FUNC_perf_event_output:
7410 return &bpf_skb_event_output_proto;
7411 case BPF_FUNC_get_smp_processor_id:
7412 return &bpf_get_smp_processor_id_proto;
7413 case BPF_FUNC_skb_under_cgroup:
7414 return &bpf_skb_under_cgroup_proto;
7415 case BPF_FUNC_get_socket_cookie:
7416 return &bpf_get_socket_cookie_proto;
7417 case BPF_FUNC_get_socket_uid:
7418 return &bpf_get_socket_uid_proto;
7419 case BPF_FUNC_fib_lookup:
7420 return &bpf_skb_fib_lookup_proto;
7421 case BPF_FUNC_check_mtu:
7422 return &bpf_skb_check_mtu_proto;
7423 case BPF_FUNC_sk_fullsock:
7424 return &bpf_sk_fullsock_proto;
7425 case BPF_FUNC_sk_storage_get:
7426 return &bpf_sk_storage_get_proto;
7427 case BPF_FUNC_sk_storage_delete:
7428 return &bpf_sk_storage_delete_proto;
7430 case BPF_FUNC_skb_get_xfrm_state:
7431 return &bpf_skb_get_xfrm_state_proto;
7433 #ifdef CONFIG_CGROUP_NET_CLASSID
7434 case BPF_FUNC_skb_cgroup_classid:
7435 return &bpf_skb_cgroup_classid_proto;
7437 #ifdef CONFIG_SOCK_CGROUP_DATA
7438 case BPF_FUNC_skb_cgroup_id:
7439 return &bpf_skb_cgroup_id_proto;
7440 case BPF_FUNC_skb_ancestor_cgroup_id:
7441 return &bpf_skb_ancestor_cgroup_id_proto;
7444 case BPF_FUNC_sk_lookup_tcp:
7445 return &bpf_sk_lookup_tcp_proto;
7446 case BPF_FUNC_sk_lookup_udp:
7447 return &bpf_sk_lookup_udp_proto;
7448 case BPF_FUNC_sk_release:
7449 return &bpf_sk_release_proto;
7450 case BPF_FUNC_tcp_sock:
7451 return &bpf_tcp_sock_proto;
7452 case BPF_FUNC_get_listener_sock:
7453 return &bpf_get_listener_sock_proto;
7454 case BPF_FUNC_skc_lookup_tcp:
7455 return &bpf_skc_lookup_tcp_proto;
7456 case BPF_FUNC_tcp_check_syncookie:
7457 return &bpf_tcp_check_syncookie_proto;
7458 case BPF_FUNC_skb_ecn_set_ce:
7459 return &bpf_skb_ecn_set_ce_proto;
7460 case BPF_FUNC_tcp_gen_syncookie:
7461 return &bpf_tcp_gen_syncookie_proto;
7462 case BPF_FUNC_sk_assign:
7463 return &bpf_sk_assign_proto;
7466 return bpf_sk_base_func_proto(func_id);
7470 static const struct bpf_func_proto *
7471 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7474 case BPF_FUNC_perf_event_output:
7475 return &bpf_xdp_event_output_proto;
7476 case BPF_FUNC_get_smp_processor_id:
7477 return &bpf_get_smp_processor_id_proto;
7478 case BPF_FUNC_csum_diff:
7479 return &bpf_csum_diff_proto;
7480 case BPF_FUNC_xdp_adjust_head:
7481 return &bpf_xdp_adjust_head_proto;
7482 case BPF_FUNC_xdp_adjust_meta:
7483 return &bpf_xdp_adjust_meta_proto;
7484 case BPF_FUNC_redirect:
7485 return &bpf_xdp_redirect_proto;
7486 case BPF_FUNC_redirect_map:
7487 return &bpf_xdp_redirect_map_proto;
7488 case BPF_FUNC_xdp_adjust_tail:
7489 return &bpf_xdp_adjust_tail_proto;
7490 case BPF_FUNC_fib_lookup:
7491 return &bpf_xdp_fib_lookup_proto;
7492 case BPF_FUNC_check_mtu:
7493 return &bpf_xdp_check_mtu_proto;
7495 case BPF_FUNC_sk_lookup_udp:
7496 return &bpf_xdp_sk_lookup_udp_proto;
7497 case BPF_FUNC_sk_lookup_tcp:
7498 return &bpf_xdp_sk_lookup_tcp_proto;
7499 case BPF_FUNC_sk_release:
7500 return &bpf_sk_release_proto;
7501 case BPF_FUNC_skc_lookup_tcp:
7502 return &bpf_xdp_skc_lookup_tcp_proto;
7503 case BPF_FUNC_tcp_check_syncookie:
7504 return &bpf_tcp_check_syncookie_proto;
7505 case BPF_FUNC_tcp_gen_syncookie:
7506 return &bpf_tcp_gen_syncookie_proto;
7509 return bpf_sk_base_func_proto(func_id);
7513 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
7514 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
7516 static const struct bpf_func_proto *
7517 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7520 case BPF_FUNC_setsockopt:
7521 return &bpf_sock_ops_setsockopt_proto;
7522 case BPF_FUNC_getsockopt:
7523 return &bpf_sock_ops_getsockopt_proto;
7524 case BPF_FUNC_sock_ops_cb_flags_set:
7525 return &bpf_sock_ops_cb_flags_set_proto;
7526 case BPF_FUNC_sock_map_update:
7527 return &bpf_sock_map_update_proto;
7528 case BPF_FUNC_sock_hash_update:
7529 return &bpf_sock_hash_update_proto;
7530 case BPF_FUNC_get_socket_cookie:
7531 return &bpf_get_socket_cookie_sock_ops_proto;
7532 case BPF_FUNC_get_local_storage:
7533 return &bpf_get_local_storage_proto;
7534 case BPF_FUNC_perf_event_output:
7535 return &bpf_event_output_data_proto;
7536 case BPF_FUNC_sk_storage_get:
7537 return &bpf_sk_storage_get_proto;
7538 case BPF_FUNC_sk_storage_delete:
7539 return &bpf_sk_storage_delete_proto;
7540 case BPF_FUNC_get_netns_cookie:
7541 return &bpf_get_netns_cookie_sock_ops_proto;
7543 case BPF_FUNC_load_hdr_opt:
7544 return &bpf_sock_ops_load_hdr_opt_proto;
7545 case BPF_FUNC_store_hdr_opt:
7546 return &bpf_sock_ops_store_hdr_opt_proto;
7547 case BPF_FUNC_reserve_hdr_opt:
7548 return &bpf_sock_ops_reserve_hdr_opt_proto;
7549 case BPF_FUNC_tcp_sock:
7550 return &bpf_tcp_sock_proto;
7551 #endif /* CONFIG_INET */
7553 return bpf_sk_base_func_proto(func_id);
7557 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
7558 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
7560 static const struct bpf_func_proto *
7561 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7564 case BPF_FUNC_msg_redirect_map:
7565 return &bpf_msg_redirect_map_proto;
7566 case BPF_FUNC_msg_redirect_hash:
7567 return &bpf_msg_redirect_hash_proto;
7568 case BPF_FUNC_msg_apply_bytes:
7569 return &bpf_msg_apply_bytes_proto;
7570 case BPF_FUNC_msg_cork_bytes:
7571 return &bpf_msg_cork_bytes_proto;
7572 case BPF_FUNC_msg_pull_data:
7573 return &bpf_msg_pull_data_proto;
7574 case BPF_FUNC_msg_push_data:
7575 return &bpf_msg_push_data_proto;
7576 case BPF_FUNC_msg_pop_data:
7577 return &bpf_msg_pop_data_proto;
7578 case BPF_FUNC_perf_event_output:
7579 return &bpf_event_output_data_proto;
7580 case BPF_FUNC_get_current_uid_gid:
7581 return &bpf_get_current_uid_gid_proto;
7582 case BPF_FUNC_get_current_pid_tgid:
7583 return &bpf_get_current_pid_tgid_proto;
7584 case BPF_FUNC_sk_storage_get:
7585 return &bpf_sk_storage_get_proto;
7586 case BPF_FUNC_sk_storage_delete:
7587 return &bpf_sk_storage_delete_proto;
7588 case BPF_FUNC_get_netns_cookie:
7589 return &bpf_get_netns_cookie_sk_msg_proto;
7590 #ifdef CONFIG_CGROUPS
7591 case BPF_FUNC_get_current_cgroup_id:
7592 return &bpf_get_current_cgroup_id_proto;
7593 case BPF_FUNC_get_current_ancestor_cgroup_id:
7594 return &bpf_get_current_ancestor_cgroup_id_proto;
7596 #ifdef CONFIG_CGROUP_NET_CLASSID
7597 case BPF_FUNC_get_cgroup_classid:
7598 return &bpf_get_cgroup_classid_curr_proto;
7601 return bpf_sk_base_func_proto(func_id);
7605 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
7606 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
7608 static const struct bpf_func_proto *
7609 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7612 case BPF_FUNC_skb_store_bytes:
7613 return &bpf_skb_store_bytes_proto;
7614 case BPF_FUNC_skb_load_bytes:
7615 return &bpf_skb_load_bytes_proto;
7616 case BPF_FUNC_skb_pull_data:
7617 return &sk_skb_pull_data_proto;
7618 case BPF_FUNC_skb_change_tail:
7619 return &sk_skb_change_tail_proto;
7620 case BPF_FUNC_skb_change_head:
7621 return &sk_skb_change_head_proto;
7622 case BPF_FUNC_skb_adjust_room:
7623 return &sk_skb_adjust_room_proto;
7624 case BPF_FUNC_get_socket_cookie:
7625 return &bpf_get_socket_cookie_proto;
7626 case BPF_FUNC_get_socket_uid:
7627 return &bpf_get_socket_uid_proto;
7628 case BPF_FUNC_sk_redirect_map:
7629 return &bpf_sk_redirect_map_proto;
7630 case BPF_FUNC_sk_redirect_hash:
7631 return &bpf_sk_redirect_hash_proto;
7632 case BPF_FUNC_perf_event_output:
7633 return &bpf_skb_event_output_proto;
7635 case BPF_FUNC_sk_lookup_tcp:
7636 return &bpf_sk_lookup_tcp_proto;
7637 case BPF_FUNC_sk_lookup_udp:
7638 return &bpf_sk_lookup_udp_proto;
7639 case BPF_FUNC_sk_release:
7640 return &bpf_sk_release_proto;
7641 case BPF_FUNC_skc_lookup_tcp:
7642 return &bpf_skc_lookup_tcp_proto;
7645 return bpf_sk_base_func_proto(func_id);
7649 static const struct bpf_func_proto *
7650 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7653 case BPF_FUNC_skb_load_bytes:
7654 return &bpf_flow_dissector_load_bytes_proto;
7656 return bpf_sk_base_func_proto(func_id);
7660 static const struct bpf_func_proto *
7661 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7664 case BPF_FUNC_skb_load_bytes:
7665 return &bpf_skb_load_bytes_proto;
7666 case BPF_FUNC_skb_pull_data:
7667 return &bpf_skb_pull_data_proto;
7668 case BPF_FUNC_csum_diff:
7669 return &bpf_csum_diff_proto;
7670 case BPF_FUNC_get_cgroup_classid:
7671 return &bpf_get_cgroup_classid_proto;
7672 case BPF_FUNC_get_route_realm:
7673 return &bpf_get_route_realm_proto;
7674 case BPF_FUNC_get_hash_recalc:
7675 return &bpf_get_hash_recalc_proto;
7676 case BPF_FUNC_perf_event_output:
7677 return &bpf_skb_event_output_proto;
7678 case BPF_FUNC_get_smp_processor_id:
7679 return &bpf_get_smp_processor_id_proto;
7680 case BPF_FUNC_skb_under_cgroup:
7681 return &bpf_skb_under_cgroup_proto;
7683 return bpf_sk_base_func_proto(func_id);
7687 static const struct bpf_func_proto *
7688 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7691 case BPF_FUNC_lwt_push_encap:
7692 return &bpf_lwt_in_push_encap_proto;
7694 return lwt_out_func_proto(func_id, prog);
7698 static const struct bpf_func_proto *
7699 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7702 case BPF_FUNC_skb_get_tunnel_key:
7703 return &bpf_skb_get_tunnel_key_proto;
7704 case BPF_FUNC_skb_set_tunnel_key:
7705 return bpf_get_skb_set_tunnel_proto(func_id);
7706 case BPF_FUNC_skb_get_tunnel_opt:
7707 return &bpf_skb_get_tunnel_opt_proto;
7708 case BPF_FUNC_skb_set_tunnel_opt:
7709 return bpf_get_skb_set_tunnel_proto(func_id);
7710 case BPF_FUNC_redirect:
7711 return &bpf_redirect_proto;
7712 case BPF_FUNC_clone_redirect:
7713 return &bpf_clone_redirect_proto;
7714 case BPF_FUNC_skb_change_tail:
7715 return &bpf_skb_change_tail_proto;
7716 case BPF_FUNC_skb_change_head:
7717 return &bpf_skb_change_head_proto;
7718 case BPF_FUNC_skb_store_bytes:
7719 return &bpf_skb_store_bytes_proto;
7720 case BPF_FUNC_csum_update:
7721 return &bpf_csum_update_proto;
7722 case BPF_FUNC_csum_level:
7723 return &bpf_csum_level_proto;
7724 case BPF_FUNC_l3_csum_replace:
7725 return &bpf_l3_csum_replace_proto;
7726 case BPF_FUNC_l4_csum_replace:
7727 return &bpf_l4_csum_replace_proto;
7728 case BPF_FUNC_set_hash_invalid:
7729 return &bpf_set_hash_invalid_proto;
7730 case BPF_FUNC_lwt_push_encap:
7731 return &bpf_lwt_xmit_push_encap_proto;
7733 return lwt_out_func_proto(func_id, prog);
7737 static const struct bpf_func_proto *
7738 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7741 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7742 case BPF_FUNC_lwt_seg6_store_bytes:
7743 return &bpf_lwt_seg6_store_bytes_proto;
7744 case BPF_FUNC_lwt_seg6_action:
7745 return &bpf_lwt_seg6_action_proto;
7746 case BPF_FUNC_lwt_seg6_adjust_srh:
7747 return &bpf_lwt_seg6_adjust_srh_proto;
7750 return lwt_out_func_proto(func_id, prog);
7754 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
7755 const struct bpf_prog *prog,
7756 struct bpf_insn_access_aux *info)
7758 const int size_default = sizeof(__u32);
7760 if (off < 0 || off >= sizeof(struct __sk_buff))
7763 /* The verifier guarantees that size > 0. */
7764 if (off % size != 0)
7768 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7769 if (off + size > offsetofend(struct __sk_buff, cb[4]))
7772 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
7773 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
7774 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
7775 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
7776 case bpf_ctx_range(struct __sk_buff, data):
7777 case bpf_ctx_range(struct __sk_buff, data_meta):
7778 case bpf_ctx_range(struct __sk_buff, data_end):
7779 if (size != size_default)
7782 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
7784 case bpf_ctx_range(struct __sk_buff, tstamp):
7785 if (size != sizeof(__u64))
7788 case offsetof(struct __sk_buff, sk):
7789 if (type == BPF_WRITE || size != sizeof(__u64))
7791 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
7794 /* Only narrow read access allowed for now. */
7795 if (type == BPF_WRITE) {
7796 if (size != size_default)
7799 bpf_ctx_record_field_size(info, size_default);
7800 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
7808 static bool sk_filter_is_valid_access(int off, int size,
7809 enum bpf_access_type type,
7810 const struct bpf_prog *prog,
7811 struct bpf_insn_access_aux *info)
7814 case bpf_ctx_range(struct __sk_buff, tc_classid):
7815 case bpf_ctx_range(struct __sk_buff, data):
7816 case bpf_ctx_range(struct __sk_buff, data_meta):
7817 case bpf_ctx_range(struct __sk_buff, data_end):
7818 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7819 case bpf_ctx_range(struct __sk_buff, tstamp):
7820 case bpf_ctx_range(struct __sk_buff, wire_len):
7824 if (type == BPF_WRITE) {
7826 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7833 return bpf_skb_is_valid_access(off, size, type, prog, info);
7836 static bool cg_skb_is_valid_access(int off, int size,
7837 enum bpf_access_type type,
7838 const struct bpf_prog *prog,
7839 struct bpf_insn_access_aux *info)
7842 case bpf_ctx_range(struct __sk_buff, tc_classid):
7843 case bpf_ctx_range(struct __sk_buff, data_meta):
7844 case bpf_ctx_range(struct __sk_buff, wire_len):
7846 case bpf_ctx_range(struct __sk_buff, data):
7847 case bpf_ctx_range(struct __sk_buff, data_end):
7853 if (type == BPF_WRITE) {
7855 case bpf_ctx_range(struct __sk_buff, mark):
7856 case bpf_ctx_range(struct __sk_buff, priority):
7857 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7859 case bpf_ctx_range(struct __sk_buff, tstamp):
7869 case bpf_ctx_range(struct __sk_buff, data):
7870 info->reg_type = PTR_TO_PACKET;
7872 case bpf_ctx_range(struct __sk_buff, data_end):
7873 info->reg_type = PTR_TO_PACKET_END;
7877 return bpf_skb_is_valid_access(off, size, type, prog, info);
7880 static bool lwt_is_valid_access(int off, int size,
7881 enum bpf_access_type type,
7882 const struct bpf_prog *prog,
7883 struct bpf_insn_access_aux *info)
7886 case bpf_ctx_range(struct __sk_buff, tc_classid):
7887 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7888 case bpf_ctx_range(struct __sk_buff, data_meta):
7889 case bpf_ctx_range(struct __sk_buff, tstamp):
7890 case bpf_ctx_range(struct __sk_buff, wire_len):
7894 if (type == BPF_WRITE) {
7896 case bpf_ctx_range(struct __sk_buff, mark):
7897 case bpf_ctx_range(struct __sk_buff, priority):
7898 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7906 case bpf_ctx_range(struct __sk_buff, data):
7907 info->reg_type = PTR_TO_PACKET;
7909 case bpf_ctx_range(struct __sk_buff, data_end):
7910 info->reg_type = PTR_TO_PACKET_END;
7914 return bpf_skb_is_valid_access(off, size, type, prog, info);
7917 /* Attach type specific accesses */
7918 static bool __sock_filter_check_attach_type(int off,
7919 enum bpf_access_type access_type,
7920 enum bpf_attach_type attach_type)
7923 case offsetof(struct bpf_sock, bound_dev_if):
7924 case offsetof(struct bpf_sock, mark):
7925 case offsetof(struct bpf_sock, priority):
7926 switch (attach_type) {
7927 case BPF_CGROUP_INET_SOCK_CREATE:
7928 case BPF_CGROUP_INET_SOCK_RELEASE:
7933 case bpf_ctx_range(struct bpf_sock, src_ip4):
7934 switch (attach_type) {
7935 case BPF_CGROUP_INET4_POST_BIND:
7940 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7941 switch (attach_type) {
7942 case BPF_CGROUP_INET6_POST_BIND:
7947 case bpf_ctx_range(struct bpf_sock, src_port):
7948 switch (attach_type) {
7949 case BPF_CGROUP_INET4_POST_BIND:
7950 case BPF_CGROUP_INET6_POST_BIND:
7957 return access_type == BPF_READ;
7962 bool bpf_sock_common_is_valid_access(int off, int size,
7963 enum bpf_access_type type,
7964 struct bpf_insn_access_aux *info)
7967 case bpf_ctx_range_till(struct bpf_sock, type, priority):
7970 return bpf_sock_is_valid_access(off, size, type, info);
7974 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7975 struct bpf_insn_access_aux *info)
7977 const int size_default = sizeof(__u32);
7980 if (off < 0 || off >= sizeof(struct bpf_sock))
7982 if (off % size != 0)
7986 case offsetof(struct bpf_sock, state):
7987 case offsetof(struct bpf_sock, family):
7988 case offsetof(struct bpf_sock, type):
7989 case offsetof(struct bpf_sock, protocol):
7990 case offsetof(struct bpf_sock, src_port):
7991 case offsetof(struct bpf_sock, rx_queue_mapping):
7992 case bpf_ctx_range(struct bpf_sock, src_ip4):
7993 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7994 case bpf_ctx_range(struct bpf_sock, dst_ip4):
7995 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
7996 bpf_ctx_record_field_size(info, size_default);
7997 return bpf_ctx_narrow_access_ok(off, size, size_default);
7998 case bpf_ctx_range(struct bpf_sock, dst_port):
7999 field_size = size == size_default ?
8000 size_default : sizeof_field(struct bpf_sock, dst_port);
8001 bpf_ctx_record_field_size(info, field_size);
8002 return bpf_ctx_narrow_access_ok(off, size, field_size);
8003 case offsetofend(struct bpf_sock, dst_port) ...
8004 offsetof(struct bpf_sock, dst_ip4) - 1:
8008 return size == size_default;
8011 static bool sock_filter_is_valid_access(int off, int size,
8012 enum bpf_access_type type,
8013 const struct bpf_prog *prog,
8014 struct bpf_insn_access_aux *info)
8016 if (!bpf_sock_is_valid_access(off, size, type, info))
8018 return __sock_filter_check_attach_type(off, type,
8019 prog->expected_attach_type);
8022 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
8023 const struct bpf_prog *prog)
8025 /* Neither direct read nor direct write requires any preliminary
8031 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
8032 const struct bpf_prog *prog, int drop_verdict)
8034 struct bpf_insn *insn = insn_buf;
8039 /* if (!skb->cloned)
8042 * (Fast-path, otherwise approximation that we might be
8043 * a clone, do the rest in helper.)
8045 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
8046 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
8047 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
8049 /* ret = bpf_skb_pull_data(skb, 0); */
8050 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
8051 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
8052 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
8053 BPF_FUNC_skb_pull_data);
8056 * return TC_ACT_SHOT;
8058 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
8059 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
8060 *insn++ = BPF_EXIT_INSN();
8063 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
8065 *insn++ = prog->insnsi[0];
8067 return insn - insn_buf;
8070 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
8071 struct bpf_insn *insn_buf)
8073 bool indirect = BPF_MODE(orig->code) == BPF_IND;
8074 struct bpf_insn *insn = insn_buf;
8077 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
8079 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
8081 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
8083 /* We're guaranteed here that CTX is in R6. */
8084 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
8086 switch (BPF_SIZE(orig->code)) {
8088 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
8091 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
8094 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
8098 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
8099 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
8100 *insn++ = BPF_EXIT_INSN();
8102 return insn - insn_buf;
8105 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8106 const struct bpf_prog *prog)
8108 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8111 static bool tc_cls_act_is_valid_access(int off, int size,
8112 enum bpf_access_type type,
8113 const struct bpf_prog *prog,
8114 struct bpf_insn_access_aux *info)
8116 if (type == BPF_WRITE) {
8118 case bpf_ctx_range(struct __sk_buff, mark):
8119 case bpf_ctx_range(struct __sk_buff, tc_index):
8120 case bpf_ctx_range(struct __sk_buff, priority):
8121 case bpf_ctx_range(struct __sk_buff, tc_classid):
8122 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8123 case bpf_ctx_range(struct __sk_buff, tstamp):
8124 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8132 case bpf_ctx_range(struct __sk_buff, data):
8133 info->reg_type = PTR_TO_PACKET;
8135 case bpf_ctx_range(struct __sk_buff, data_meta):
8136 info->reg_type = PTR_TO_PACKET_META;
8138 case bpf_ctx_range(struct __sk_buff, data_end):
8139 info->reg_type = PTR_TO_PACKET_END;
8141 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8145 return bpf_skb_is_valid_access(off, size, type, prog, info);
8148 static bool __is_valid_xdp_access(int off, int size)
8150 if (off < 0 || off >= sizeof(struct xdp_md))
8152 if (off % size != 0)
8154 if (size != sizeof(__u32))
8160 static bool xdp_is_valid_access(int off, int size,
8161 enum bpf_access_type type,
8162 const struct bpf_prog *prog,
8163 struct bpf_insn_access_aux *info)
8165 if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8167 case offsetof(struct xdp_md, egress_ifindex):
8172 if (type == BPF_WRITE) {
8173 if (bpf_prog_is_dev_bound(prog->aux)) {
8175 case offsetof(struct xdp_md, rx_queue_index):
8176 return __is_valid_xdp_access(off, size);
8183 case offsetof(struct xdp_md, data):
8184 info->reg_type = PTR_TO_PACKET;
8186 case offsetof(struct xdp_md, data_meta):
8187 info->reg_type = PTR_TO_PACKET_META;
8189 case offsetof(struct xdp_md, data_end):
8190 info->reg_type = PTR_TO_PACKET_END;
8194 return __is_valid_xdp_access(off, size);
8197 void bpf_warn_invalid_xdp_action(u32 act)
8199 const u32 act_max = XDP_REDIRECT;
8201 pr_warn_once("%s XDP return value %u, expect packet loss!\n",
8202 act > act_max ? "Illegal" : "Driver unsupported",
8205 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
8207 static bool sock_addr_is_valid_access(int off, int size,
8208 enum bpf_access_type type,
8209 const struct bpf_prog *prog,
8210 struct bpf_insn_access_aux *info)
8212 const int size_default = sizeof(__u32);
8214 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
8216 if (off % size != 0)
8219 /* Disallow access to IPv6 fields from IPv4 contex and vise
8223 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8224 switch (prog->expected_attach_type) {
8225 case BPF_CGROUP_INET4_BIND:
8226 case BPF_CGROUP_INET4_CONNECT:
8227 case BPF_CGROUP_INET4_GETPEERNAME:
8228 case BPF_CGROUP_INET4_GETSOCKNAME:
8229 case BPF_CGROUP_UDP4_SENDMSG:
8230 case BPF_CGROUP_UDP4_RECVMSG:
8236 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8237 switch (prog->expected_attach_type) {
8238 case BPF_CGROUP_INET6_BIND:
8239 case BPF_CGROUP_INET6_CONNECT:
8240 case BPF_CGROUP_INET6_GETPEERNAME:
8241 case BPF_CGROUP_INET6_GETSOCKNAME:
8242 case BPF_CGROUP_UDP6_SENDMSG:
8243 case BPF_CGROUP_UDP6_RECVMSG:
8249 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8250 switch (prog->expected_attach_type) {
8251 case BPF_CGROUP_UDP4_SENDMSG:
8257 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8259 switch (prog->expected_attach_type) {
8260 case BPF_CGROUP_UDP6_SENDMSG:
8269 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8270 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8271 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8272 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8274 case bpf_ctx_range(struct bpf_sock_addr, user_port):
8275 if (type == BPF_READ) {
8276 bpf_ctx_record_field_size(info, size_default);
8278 if (bpf_ctx_wide_access_ok(off, size,
8279 struct bpf_sock_addr,
8283 if (bpf_ctx_wide_access_ok(off, size,
8284 struct bpf_sock_addr,
8288 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8291 if (bpf_ctx_wide_access_ok(off, size,
8292 struct bpf_sock_addr,
8296 if (bpf_ctx_wide_access_ok(off, size,
8297 struct bpf_sock_addr,
8301 if (size != size_default)
8305 case offsetof(struct bpf_sock_addr, sk):
8306 if (type != BPF_READ)
8308 if (size != sizeof(__u64))
8310 info->reg_type = PTR_TO_SOCKET;
8313 if (type == BPF_READ) {
8314 if (size != size_default)
8324 static bool sock_ops_is_valid_access(int off, int size,
8325 enum bpf_access_type type,
8326 const struct bpf_prog *prog,
8327 struct bpf_insn_access_aux *info)
8329 const int size_default = sizeof(__u32);
8331 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
8334 /* The verifier guarantees that size > 0. */
8335 if (off % size != 0)
8338 if (type == BPF_WRITE) {
8340 case offsetof(struct bpf_sock_ops, reply):
8341 case offsetof(struct bpf_sock_ops, sk_txhash):
8342 if (size != size_default)
8350 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
8352 if (size != sizeof(__u64))
8355 case offsetof(struct bpf_sock_ops, sk):
8356 if (size != sizeof(__u64))
8358 info->reg_type = PTR_TO_SOCKET_OR_NULL;
8360 case offsetof(struct bpf_sock_ops, skb_data):
8361 if (size != sizeof(__u64))
8363 info->reg_type = PTR_TO_PACKET;
8365 case offsetof(struct bpf_sock_ops, skb_data_end):
8366 if (size != sizeof(__u64))
8368 info->reg_type = PTR_TO_PACKET_END;
8370 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
8371 bpf_ctx_record_field_size(info, size_default);
8372 return bpf_ctx_narrow_access_ok(off, size,
8375 if (size != size_default)
8384 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
8385 const struct bpf_prog *prog)
8387 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
8390 static bool sk_skb_is_valid_access(int off, int size,
8391 enum bpf_access_type type,
8392 const struct bpf_prog *prog,
8393 struct bpf_insn_access_aux *info)
8396 case bpf_ctx_range(struct __sk_buff, tc_classid):
8397 case bpf_ctx_range(struct __sk_buff, data_meta):
8398 case bpf_ctx_range(struct __sk_buff, tstamp):
8399 case bpf_ctx_range(struct __sk_buff, wire_len):
8403 if (type == BPF_WRITE) {
8405 case bpf_ctx_range(struct __sk_buff, tc_index):
8406 case bpf_ctx_range(struct __sk_buff, priority):
8414 case bpf_ctx_range(struct __sk_buff, mark):
8416 case bpf_ctx_range(struct __sk_buff, data):
8417 info->reg_type = PTR_TO_PACKET;
8419 case bpf_ctx_range(struct __sk_buff, data_end):
8420 info->reg_type = PTR_TO_PACKET_END;
8424 return bpf_skb_is_valid_access(off, size, type, prog, info);
8427 static bool sk_msg_is_valid_access(int off, int size,
8428 enum bpf_access_type type,
8429 const struct bpf_prog *prog,
8430 struct bpf_insn_access_aux *info)
8432 if (type == BPF_WRITE)
8435 if (off % size != 0)
8439 case offsetof(struct sk_msg_md, data):
8440 info->reg_type = PTR_TO_PACKET;
8441 if (size != sizeof(__u64))
8444 case offsetof(struct sk_msg_md, data_end):
8445 info->reg_type = PTR_TO_PACKET_END;
8446 if (size != sizeof(__u64))
8449 case offsetof(struct sk_msg_md, sk):
8450 if (size != sizeof(__u64))
8452 info->reg_type = PTR_TO_SOCKET;
8454 case bpf_ctx_range(struct sk_msg_md, family):
8455 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
8456 case bpf_ctx_range(struct sk_msg_md, local_ip4):
8457 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
8458 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
8459 case bpf_ctx_range(struct sk_msg_md, remote_port):
8460 case bpf_ctx_range(struct sk_msg_md, local_port):
8461 case bpf_ctx_range(struct sk_msg_md, size):
8462 if (size != sizeof(__u32))
8471 static bool flow_dissector_is_valid_access(int off, int size,
8472 enum bpf_access_type type,
8473 const struct bpf_prog *prog,
8474 struct bpf_insn_access_aux *info)
8476 const int size_default = sizeof(__u32);
8478 if (off < 0 || off >= sizeof(struct __sk_buff))
8481 if (type == BPF_WRITE)
8485 case bpf_ctx_range(struct __sk_buff, data):
8486 if (size != size_default)
8488 info->reg_type = PTR_TO_PACKET;
8490 case bpf_ctx_range(struct __sk_buff, data_end):
8491 if (size != size_default)
8493 info->reg_type = PTR_TO_PACKET_END;
8495 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8496 if (size != sizeof(__u64))
8498 info->reg_type = PTR_TO_FLOW_KEYS;
8505 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
8506 const struct bpf_insn *si,
8507 struct bpf_insn *insn_buf,
8508 struct bpf_prog *prog,
8512 struct bpf_insn *insn = insn_buf;
8515 case offsetof(struct __sk_buff, data):
8516 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
8517 si->dst_reg, si->src_reg,
8518 offsetof(struct bpf_flow_dissector, data));
8521 case offsetof(struct __sk_buff, data_end):
8522 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
8523 si->dst_reg, si->src_reg,
8524 offsetof(struct bpf_flow_dissector, data_end));
8527 case offsetof(struct __sk_buff, flow_keys):
8528 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
8529 si->dst_reg, si->src_reg,
8530 offsetof(struct bpf_flow_dissector, flow_keys));
8534 return insn - insn_buf;
8537 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si,
8538 struct bpf_insn *insn)
8540 /* si->dst_reg = skb_shinfo(SKB); */
8541 #ifdef NET_SKBUFF_DATA_USES_OFFSET
8542 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8543 BPF_REG_AX, si->src_reg,
8544 offsetof(struct sk_buff, end));
8545 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
8546 si->dst_reg, si->src_reg,
8547 offsetof(struct sk_buff, head));
8548 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
8550 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8551 si->dst_reg, si->src_reg,
8552 offsetof(struct sk_buff, end));
8558 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
8559 const struct bpf_insn *si,
8560 struct bpf_insn *insn_buf,
8561 struct bpf_prog *prog, u32 *target_size)
8563 struct bpf_insn *insn = insn_buf;
8567 case offsetof(struct __sk_buff, len):
8568 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8569 bpf_target_off(struct sk_buff, len, 4,
8573 case offsetof(struct __sk_buff, protocol):
8574 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8575 bpf_target_off(struct sk_buff, protocol, 2,
8579 case offsetof(struct __sk_buff, vlan_proto):
8580 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8581 bpf_target_off(struct sk_buff, vlan_proto, 2,
8585 case offsetof(struct __sk_buff, priority):
8586 if (type == BPF_WRITE)
8587 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8588 bpf_target_off(struct sk_buff, priority, 4,
8591 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8592 bpf_target_off(struct sk_buff, priority, 4,
8596 case offsetof(struct __sk_buff, ingress_ifindex):
8597 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8598 bpf_target_off(struct sk_buff, skb_iif, 4,
8602 case offsetof(struct __sk_buff, ifindex):
8603 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8604 si->dst_reg, si->src_reg,
8605 offsetof(struct sk_buff, dev));
8606 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
8607 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8608 bpf_target_off(struct net_device, ifindex, 4,
8612 case offsetof(struct __sk_buff, hash):
8613 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8614 bpf_target_off(struct sk_buff, hash, 4,
8618 case offsetof(struct __sk_buff, mark):
8619 if (type == BPF_WRITE)
8620 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8621 bpf_target_off(struct sk_buff, mark, 4,
8624 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8625 bpf_target_off(struct sk_buff, mark, 4,
8629 case offsetof(struct __sk_buff, pkt_type):
8631 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8633 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
8634 #ifdef __BIG_ENDIAN_BITFIELD
8635 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
8639 case offsetof(struct __sk_buff, queue_mapping):
8640 if (type == BPF_WRITE) {
8641 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
8642 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8643 bpf_target_off(struct sk_buff,
8647 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8648 bpf_target_off(struct sk_buff,
8654 case offsetof(struct __sk_buff, vlan_present):
8656 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8657 PKT_VLAN_PRESENT_OFFSET());
8658 if (PKT_VLAN_PRESENT_BIT)
8659 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
8660 if (PKT_VLAN_PRESENT_BIT < 7)
8661 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
8664 case offsetof(struct __sk_buff, vlan_tci):
8665 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8666 bpf_target_off(struct sk_buff, vlan_tci, 2,
8670 case offsetof(struct __sk_buff, cb[0]) ...
8671 offsetofend(struct __sk_buff, cb[4]) - 1:
8672 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
8673 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
8674 offsetof(struct qdisc_skb_cb, data)) %
8677 prog->cb_access = 1;
8679 off -= offsetof(struct __sk_buff, cb[0]);
8680 off += offsetof(struct sk_buff, cb);
8681 off += offsetof(struct qdisc_skb_cb, data);
8682 if (type == BPF_WRITE)
8683 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
8686 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
8690 case offsetof(struct __sk_buff, tc_classid):
8691 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
8694 off -= offsetof(struct __sk_buff, tc_classid);
8695 off += offsetof(struct sk_buff, cb);
8696 off += offsetof(struct qdisc_skb_cb, tc_classid);
8698 if (type == BPF_WRITE)
8699 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
8702 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
8706 case offsetof(struct __sk_buff, data):
8707 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
8708 si->dst_reg, si->src_reg,
8709 offsetof(struct sk_buff, data));
8712 case offsetof(struct __sk_buff, data_meta):
8714 off -= offsetof(struct __sk_buff, data_meta);
8715 off += offsetof(struct sk_buff, cb);
8716 off += offsetof(struct bpf_skb_data_end, data_meta);
8717 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8721 case offsetof(struct __sk_buff, data_end):
8723 off -= offsetof(struct __sk_buff, data_end);
8724 off += offsetof(struct sk_buff, cb);
8725 off += offsetof(struct bpf_skb_data_end, data_end);
8726 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8730 case offsetof(struct __sk_buff, tc_index):
8731 #ifdef CONFIG_NET_SCHED
8732 if (type == BPF_WRITE)
8733 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8734 bpf_target_off(struct sk_buff, tc_index, 2,
8737 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8738 bpf_target_off(struct sk_buff, tc_index, 2,
8742 if (type == BPF_WRITE)
8743 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
8745 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8749 case offsetof(struct __sk_buff, napi_id):
8750 #if defined(CONFIG_NET_RX_BUSY_POLL)
8751 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8752 bpf_target_off(struct sk_buff, napi_id, 4,
8754 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
8755 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8758 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8761 case offsetof(struct __sk_buff, family):
8762 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
8764 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8765 si->dst_reg, si->src_reg,
8766 offsetof(struct sk_buff, sk));
8767 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8768 bpf_target_off(struct sock_common,
8772 case offsetof(struct __sk_buff, remote_ip4):
8773 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
8775 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8776 si->dst_reg, si->src_reg,
8777 offsetof(struct sk_buff, sk));
8778 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8779 bpf_target_off(struct sock_common,
8783 case offsetof(struct __sk_buff, local_ip4):
8784 BUILD_BUG_ON(sizeof_field(struct sock_common,
8785 skc_rcv_saddr) != 4);
8787 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8788 si->dst_reg, si->src_reg,
8789 offsetof(struct sk_buff, sk));
8790 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8791 bpf_target_off(struct sock_common,
8795 case offsetof(struct __sk_buff, remote_ip6[0]) ...
8796 offsetof(struct __sk_buff, remote_ip6[3]):
8797 #if IS_ENABLED(CONFIG_IPV6)
8798 BUILD_BUG_ON(sizeof_field(struct sock_common,
8799 skc_v6_daddr.s6_addr32[0]) != 4);
8802 off -= offsetof(struct __sk_buff, remote_ip6[0]);
8804 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8805 si->dst_reg, si->src_reg,
8806 offsetof(struct sk_buff, sk));
8807 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8808 offsetof(struct sock_common,
8809 skc_v6_daddr.s6_addr32[0]) +
8812 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8815 case offsetof(struct __sk_buff, local_ip6[0]) ...
8816 offsetof(struct __sk_buff, local_ip6[3]):
8817 #if IS_ENABLED(CONFIG_IPV6)
8818 BUILD_BUG_ON(sizeof_field(struct sock_common,
8819 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8822 off -= offsetof(struct __sk_buff, local_ip6[0]);
8824 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8825 si->dst_reg, si->src_reg,
8826 offsetof(struct sk_buff, sk));
8827 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8828 offsetof(struct sock_common,
8829 skc_v6_rcv_saddr.s6_addr32[0]) +
8832 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8836 case offsetof(struct __sk_buff, remote_port):
8837 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
8839 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8840 si->dst_reg, si->src_reg,
8841 offsetof(struct sk_buff, sk));
8842 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8843 bpf_target_off(struct sock_common,
8846 #ifndef __BIG_ENDIAN_BITFIELD
8847 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8851 case offsetof(struct __sk_buff, local_port):
8852 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
8854 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8855 si->dst_reg, si->src_reg,
8856 offsetof(struct sk_buff, sk));
8857 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8858 bpf_target_off(struct sock_common,
8859 skc_num, 2, target_size));
8862 case offsetof(struct __sk_buff, tstamp):
8863 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
8865 if (type == BPF_WRITE)
8866 *insn++ = BPF_STX_MEM(BPF_DW,
8867 si->dst_reg, si->src_reg,
8868 bpf_target_off(struct sk_buff,
8872 *insn++ = BPF_LDX_MEM(BPF_DW,
8873 si->dst_reg, si->src_reg,
8874 bpf_target_off(struct sk_buff,
8879 case offsetof(struct __sk_buff, gso_segs):
8880 insn = bpf_convert_shinfo_access(si, insn);
8881 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
8882 si->dst_reg, si->dst_reg,
8883 bpf_target_off(struct skb_shared_info,
8887 case offsetof(struct __sk_buff, gso_size):
8888 insn = bpf_convert_shinfo_access(si, insn);
8889 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
8890 si->dst_reg, si->dst_reg,
8891 bpf_target_off(struct skb_shared_info,
8895 case offsetof(struct __sk_buff, wire_len):
8896 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
8899 off -= offsetof(struct __sk_buff, wire_len);
8900 off += offsetof(struct sk_buff, cb);
8901 off += offsetof(struct qdisc_skb_cb, pkt_len);
8903 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
8906 case offsetof(struct __sk_buff, sk):
8907 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8908 si->dst_reg, si->src_reg,
8909 offsetof(struct sk_buff, sk));
8913 return insn - insn_buf;
8916 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
8917 const struct bpf_insn *si,
8918 struct bpf_insn *insn_buf,
8919 struct bpf_prog *prog, u32 *target_size)
8921 struct bpf_insn *insn = insn_buf;
8925 case offsetof(struct bpf_sock, bound_dev_if):
8926 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
8928 if (type == BPF_WRITE)
8929 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8930 offsetof(struct sock, sk_bound_dev_if));
8932 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8933 offsetof(struct sock, sk_bound_dev_if));
8936 case offsetof(struct bpf_sock, mark):
8937 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
8939 if (type == BPF_WRITE)
8940 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8941 offsetof(struct sock, sk_mark));
8943 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8944 offsetof(struct sock, sk_mark));
8947 case offsetof(struct bpf_sock, priority):
8948 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
8950 if (type == BPF_WRITE)
8951 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8952 offsetof(struct sock, sk_priority));
8954 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8955 offsetof(struct sock, sk_priority));
8958 case offsetof(struct bpf_sock, family):
8959 *insn++ = BPF_LDX_MEM(
8960 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
8961 si->dst_reg, si->src_reg,
8962 bpf_target_off(struct sock_common,
8964 sizeof_field(struct sock_common,
8969 case offsetof(struct bpf_sock, type):
8970 *insn++ = BPF_LDX_MEM(
8971 BPF_FIELD_SIZEOF(struct sock, sk_type),
8972 si->dst_reg, si->src_reg,
8973 bpf_target_off(struct sock, sk_type,
8974 sizeof_field(struct sock, sk_type),
8978 case offsetof(struct bpf_sock, protocol):
8979 *insn++ = BPF_LDX_MEM(
8980 BPF_FIELD_SIZEOF(struct sock, sk_protocol),
8981 si->dst_reg, si->src_reg,
8982 bpf_target_off(struct sock, sk_protocol,
8983 sizeof_field(struct sock, sk_protocol),
8987 case offsetof(struct bpf_sock, src_ip4):
8988 *insn++ = BPF_LDX_MEM(
8989 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8990 bpf_target_off(struct sock_common, skc_rcv_saddr,
8991 sizeof_field(struct sock_common,
8996 case offsetof(struct bpf_sock, dst_ip4):
8997 *insn++ = BPF_LDX_MEM(
8998 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8999 bpf_target_off(struct sock_common, skc_daddr,
9000 sizeof_field(struct sock_common,
9005 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
9006 #if IS_ENABLED(CONFIG_IPV6)
9008 off -= offsetof(struct bpf_sock, src_ip6[0]);
9009 *insn++ = BPF_LDX_MEM(
9010 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9013 skc_v6_rcv_saddr.s6_addr32[0],
9014 sizeof_field(struct sock_common,
9015 skc_v6_rcv_saddr.s6_addr32[0]),
9016 target_size) + off);
9019 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9023 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
9024 #if IS_ENABLED(CONFIG_IPV6)
9026 off -= offsetof(struct bpf_sock, dst_ip6[0]);
9027 *insn++ = BPF_LDX_MEM(
9028 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9029 bpf_target_off(struct sock_common,
9030 skc_v6_daddr.s6_addr32[0],
9031 sizeof_field(struct sock_common,
9032 skc_v6_daddr.s6_addr32[0]),
9033 target_size) + off);
9035 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9040 case offsetof(struct bpf_sock, src_port):
9041 *insn++ = BPF_LDX_MEM(
9042 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
9043 si->dst_reg, si->src_reg,
9044 bpf_target_off(struct sock_common, skc_num,
9045 sizeof_field(struct sock_common,
9050 case offsetof(struct bpf_sock, dst_port):
9051 *insn++ = BPF_LDX_MEM(
9052 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
9053 si->dst_reg, si->src_reg,
9054 bpf_target_off(struct sock_common, skc_dport,
9055 sizeof_field(struct sock_common,
9060 case offsetof(struct bpf_sock, state):
9061 *insn++ = BPF_LDX_MEM(
9062 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
9063 si->dst_reg, si->src_reg,
9064 bpf_target_off(struct sock_common, skc_state,
9065 sizeof_field(struct sock_common,
9069 case offsetof(struct bpf_sock, rx_queue_mapping):
9070 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
9071 *insn++ = BPF_LDX_MEM(
9072 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
9073 si->dst_reg, si->src_reg,
9074 bpf_target_off(struct sock, sk_rx_queue_mapping,
9075 sizeof_field(struct sock,
9076 sk_rx_queue_mapping),
9078 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
9080 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9082 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9088 return insn - insn_buf;
9091 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
9092 const struct bpf_insn *si,
9093 struct bpf_insn *insn_buf,
9094 struct bpf_prog *prog, u32 *target_size)
9096 struct bpf_insn *insn = insn_buf;
9099 case offsetof(struct __sk_buff, ifindex):
9100 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9101 si->dst_reg, si->src_reg,
9102 offsetof(struct sk_buff, dev));
9103 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9104 bpf_target_off(struct net_device, ifindex, 4,
9108 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9112 return insn - insn_buf;
9115 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
9116 const struct bpf_insn *si,
9117 struct bpf_insn *insn_buf,
9118 struct bpf_prog *prog, u32 *target_size)
9120 struct bpf_insn *insn = insn_buf;
9123 case offsetof(struct xdp_md, data):
9124 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
9125 si->dst_reg, si->src_reg,
9126 offsetof(struct xdp_buff, data));
9128 case offsetof(struct xdp_md, data_meta):
9129 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
9130 si->dst_reg, si->src_reg,
9131 offsetof(struct xdp_buff, data_meta));
9133 case offsetof(struct xdp_md, data_end):
9134 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
9135 si->dst_reg, si->src_reg,
9136 offsetof(struct xdp_buff, data_end));
9138 case offsetof(struct xdp_md, ingress_ifindex):
9139 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9140 si->dst_reg, si->src_reg,
9141 offsetof(struct xdp_buff, rxq));
9142 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
9143 si->dst_reg, si->dst_reg,
9144 offsetof(struct xdp_rxq_info, dev));
9145 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9146 offsetof(struct net_device, ifindex));
9148 case offsetof(struct xdp_md, rx_queue_index):
9149 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9150 si->dst_reg, si->src_reg,
9151 offsetof(struct xdp_buff, rxq));
9152 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9153 offsetof(struct xdp_rxq_info,
9156 case offsetof(struct xdp_md, egress_ifindex):
9157 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
9158 si->dst_reg, si->src_reg,
9159 offsetof(struct xdp_buff, txq));
9160 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
9161 si->dst_reg, si->dst_reg,
9162 offsetof(struct xdp_txq_info, dev));
9163 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9164 offsetof(struct net_device, ifindex));
9168 return insn - insn_buf;
9171 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
9172 * context Structure, F is Field in context structure that contains a pointer
9173 * to Nested Structure of type NS that has the field NF.
9175 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
9176 * sure that SIZE is not greater than actual size of S.F.NF.
9178 * If offset OFF is provided, the load happens from that offset relative to
9181 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
9183 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
9184 si->src_reg, offsetof(S, F)); \
9185 *insn++ = BPF_LDX_MEM( \
9186 SIZE, si->dst_reg, si->dst_reg, \
9187 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
9192 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
9193 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
9194 BPF_FIELD_SIZEOF(NS, NF), 0)
9196 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
9197 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
9199 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
9200 * "register" since two registers available in convert_ctx_access are not
9201 * enough: we can't override neither SRC, since it contains value to store, nor
9202 * DST since it contains pointer to context that may be used by later
9203 * instructions. But we need a temporary place to save pointer to nested
9204 * structure whose field we want to store to.
9206 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
9208 int tmp_reg = BPF_REG_9; \
9209 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
9211 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
9213 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
9215 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
9216 si->dst_reg, offsetof(S, F)); \
9217 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \
9218 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
9221 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
9225 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
9228 if (type == BPF_WRITE) { \
9229 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
9232 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
9233 S, NS, F, NF, SIZE, OFF); \
9237 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
9238 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
9239 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
9241 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
9242 const struct bpf_insn *si,
9243 struct bpf_insn *insn_buf,
9244 struct bpf_prog *prog, u32 *target_size)
9246 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
9247 struct bpf_insn *insn = insn_buf;
9250 case offsetof(struct bpf_sock_addr, user_family):
9251 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9252 struct sockaddr, uaddr, sa_family);
9255 case offsetof(struct bpf_sock_addr, user_ip4):
9256 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9257 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
9258 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
9261 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9263 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
9264 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9265 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9266 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
9270 case offsetof(struct bpf_sock_addr, user_port):
9271 /* To get port we need to know sa_family first and then treat
9272 * sockaddr as either sockaddr_in or sockaddr_in6.
9273 * Though we can simplify since port field has same offset and
9274 * size in both structures.
9275 * Here we check this invariant and use just one of the
9276 * structures if it's true.
9278 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
9279 offsetof(struct sockaddr_in6, sin6_port));
9280 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
9281 sizeof_field(struct sockaddr_in6, sin6_port));
9282 /* Account for sin6_port being smaller than user_port. */
9283 port_size = min(port_size, BPF_LDST_BYTES(si));
9284 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9285 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9286 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
9289 case offsetof(struct bpf_sock_addr, family):
9290 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9291 struct sock, sk, sk_family);
9294 case offsetof(struct bpf_sock_addr, type):
9295 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9296 struct sock, sk, sk_type);
9299 case offsetof(struct bpf_sock_addr, protocol):
9300 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9301 struct sock, sk, sk_protocol);
9304 case offsetof(struct bpf_sock_addr, msg_src_ip4):
9305 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
9306 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9307 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
9308 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
9311 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9314 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
9315 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
9316 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9317 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
9318 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
9320 case offsetof(struct bpf_sock_addr, sk):
9321 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
9322 si->dst_reg, si->src_reg,
9323 offsetof(struct bpf_sock_addr_kern, sk));
9327 return insn - insn_buf;
9330 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
9331 const struct bpf_insn *si,
9332 struct bpf_insn *insn_buf,
9333 struct bpf_prog *prog,
9336 struct bpf_insn *insn = insn_buf;
9339 /* Helper macro for adding read access to tcp_sock or sock fields. */
9340 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
9342 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \
9343 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
9344 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
9345 if (si->dst_reg == reg || si->src_reg == reg) \
9347 if (si->dst_reg == reg || si->src_reg == reg) \
9349 if (si->dst_reg == si->src_reg) { \
9350 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
9351 offsetof(struct bpf_sock_ops_kern, \
9353 fullsock_reg = reg; \
9356 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9357 struct bpf_sock_ops_kern, \
9359 fullsock_reg, si->src_reg, \
9360 offsetof(struct bpf_sock_ops_kern, \
9362 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
9363 if (si->dst_reg == si->src_reg) \
9364 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9365 offsetof(struct bpf_sock_ops_kern, \
9367 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9368 struct bpf_sock_ops_kern, sk),\
9369 si->dst_reg, si->src_reg, \
9370 offsetof(struct bpf_sock_ops_kern, sk));\
9371 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
9373 si->dst_reg, si->dst_reg, \
9374 offsetof(OBJ, OBJ_FIELD)); \
9375 if (si->dst_reg == si->src_reg) { \
9376 *insn++ = BPF_JMP_A(1); \
9377 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9378 offsetof(struct bpf_sock_ops_kern, \
9383 #define SOCK_OPS_GET_SK() \
9385 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \
9386 if (si->dst_reg == reg || si->src_reg == reg) \
9388 if (si->dst_reg == reg || si->src_reg == reg) \
9390 if (si->dst_reg == si->src_reg) { \
9391 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
9392 offsetof(struct bpf_sock_ops_kern, \
9394 fullsock_reg = reg; \
9397 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9398 struct bpf_sock_ops_kern, \
9400 fullsock_reg, si->src_reg, \
9401 offsetof(struct bpf_sock_ops_kern, \
9403 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
9404 if (si->dst_reg == si->src_reg) \
9405 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9406 offsetof(struct bpf_sock_ops_kern, \
9408 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9409 struct bpf_sock_ops_kern, sk),\
9410 si->dst_reg, si->src_reg, \
9411 offsetof(struct bpf_sock_ops_kern, sk));\
9412 if (si->dst_reg == si->src_reg) { \
9413 *insn++ = BPF_JMP_A(1); \
9414 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9415 offsetof(struct bpf_sock_ops_kern, \
9420 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
9421 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
9423 /* Helper macro for adding write access to tcp_sock or sock fields.
9424 * The macro is called with two registers, dst_reg which contains a pointer
9425 * to ctx (context) and src_reg which contains the value that should be
9426 * stored. However, we need an additional register since we cannot overwrite
9427 * dst_reg because it may be used later in the program.
9428 * Instead we "borrow" one of the other register. We first save its value
9429 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
9430 * it at the end of the macro.
9432 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
9434 int reg = BPF_REG_9; \
9435 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
9436 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
9437 if (si->dst_reg == reg || si->src_reg == reg) \
9439 if (si->dst_reg == reg || si->src_reg == reg) \
9441 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
9442 offsetof(struct bpf_sock_ops_kern, \
9444 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9445 struct bpf_sock_ops_kern, \
9448 offsetof(struct bpf_sock_ops_kern, \
9450 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
9451 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9452 struct bpf_sock_ops_kern, sk),\
9454 offsetof(struct bpf_sock_ops_kern, sk));\
9455 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \
9457 offsetof(OBJ, OBJ_FIELD)); \
9458 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
9459 offsetof(struct bpf_sock_ops_kern, \
9463 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
9465 if (TYPE == BPF_WRITE) \
9466 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
9468 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
9471 if (insn > insn_buf)
9472 return insn - insn_buf;
9475 case offsetof(struct bpf_sock_ops, op):
9476 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9478 si->dst_reg, si->src_reg,
9479 offsetof(struct bpf_sock_ops_kern, op));
9482 case offsetof(struct bpf_sock_ops, replylong[0]) ...
9483 offsetof(struct bpf_sock_ops, replylong[3]):
9484 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
9485 sizeof_field(struct bpf_sock_ops_kern, reply));
9486 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
9487 sizeof_field(struct bpf_sock_ops_kern, replylong));
9489 off -= offsetof(struct bpf_sock_ops, replylong[0]);
9490 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
9491 if (type == BPF_WRITE)
9492 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9495 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9499 case offsetof(struct bpf_sock_ops, family):
9500 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9502 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9503 struct bpf_sock_ops_kern, sk),
9504 si->dst_reg, si->src_reg,
9505 offsetof(struct bpf_sock_ops_kern, sk));
9506 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9507 offsetof(struct sock_common, skc_family));
9510 case offsetof(struct bpf_sock_ops, remote_ip4):
9511 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9513 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9514 struct bpf_sock_ops_kern, sk),
9515 si->dst_reg, si->src_reg,
9516 offsetof(struct bpf_sock_ops_kern, sk));
9517 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9518 offsetof(struct sock_common, skc_daddr));
9521 case offsetof(struct bpf_sock_ops, local_ip4):
9522 BUILD_BUG_ON(sizeof_field(struct sock_common,
9523 skc_rcv_saddr) != 4);
9525 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9526 struct bpf_sock_ops_kern, sk),
9527 si->dst_reg, si->src_reg,
9528 offsetof(struct bpf_sock_ops_kern, sk));
9529 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9530 offsetof(struct sock_common,
9534 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
9535 offsetof(struct bpf_sock_ops, remote_ip6[3]):
9536 #if IS_ENABLED(CONFIG_IPV6)
9537 BUILD_BUG_ON(sizeof_field(struct sock_common,
9538 skc_v6_daddr.s6_addr32[0]) != 4);
9541 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
9542 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9543 struct bpf_sock_ops_kern, sk),
9544 si->dst_reg, si->src_reg,
9545 offsetof(struct bpf_sock_ops_kern, sk));
9546 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9547 offsetof(struct sock_common,
9548 skc_v6_daddr.s6_addr32[0]) +
9551 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9555 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
9556 offsetof(struct bpf_sock_ops, local_ip6[3]):
9557 #if IS_ENABLED(CONFIG_IPV6)
9558 BUILD_BUG_ON(sizeof_field(struct sock_common,
9559 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9562 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
9563 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9564 struct bpf_sock_ops_kern, sk),
9565 si->dst_reg, si->src_reg,
9566 offsetof(struct bpf_sock_ops_kern, sk));
9567 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9568 offsetof(struct sock_common,
9569 skc_v6_rcv_saddr.s6_addr32[0]) +
9572 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9576 case offsetof(struct bpf_sock_ops, remote_port):
9577 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9579 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9580 struct bpf_sock_ops_kern, sk),
9581 si->dst_reg, si->src_reg,
9582 offsetof(struct bpf_sock_ops_kern, sk));
9583 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9584 offsetof(struct sock_common, skc_dport));
9585 #ifndef __BIG_ENDIAN_BITFIELD
9586 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9590 case offsetof(struct bpf_sock_ops, local_port):
9591 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9593 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9594 struct bpf_sock_ops_kern, sk),
9595 si->dst_reg, si->src_reg,
9596 offsetof(struct bpf_sock_ops_kern, sk));
9597 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9598 offsetof(struct sock_common, skc_num));
9601 case offsetof(struct bpf_sock_ops, is_fullsock):
9602 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9603 struct bpf_sock_ops_kern,
9605 si->dst_reg, si->src_reg,
9606 offsetof(struct bpf_sock_ops_kern,
9610 case offsetof(struct bpf_sock_ops, state):
9611 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
9613 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9614 struct bpf_sock_ops_kern, sk),
9615 si->dst_reg, si->src_reg,
9616 offsetof(struct bpf_sock_ops_kern, sk));
9617 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
9618 offsetof(struct sock_common, skc_state));
9621 case offsetof(struct bpf_sock_ops, rtt_min):
9622 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
9623 sizeof(struct minmax));
9624 BUILD_BUG_ON(sizeof(struct minmax) <
9625 sizeof(struct minmax_sample));
9627 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9628 struct bpf_sock_ops_kern, sk),
9629 si->dst_reg, si->src_reg,
9630 offsetof(struct bpf_sock_ops_kern, sk));
9631 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9632 offsetof(struct tcp_sock, rtt_min) +
9633 sizeof_field(struct minmax_sample, t));
9636 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
9637 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
9641 case offsetof(struct bpf_sock_ops, sk_txhash):
9642 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
9645 case offsetof(struct bpf_sock_ops, snd_cwnd):
9646 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
9648 case offsetof(struct bpf_sock_ops, srtt_us):
9649 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
9651 case offsetof(struct bpf_sock_ops, snd_ssthresh):
9652 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
9654 case offsetof(struct bpf_sock_ops, rcv_nxt):
9655 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
9657 case offsetof(struct bpf_sock_ops, snd_nxt):
9658 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
9660 case offsetof(struct bpf_sock_ops, snd_una):
9661 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
9663 case offsetof(struct bpf_sock_ops, mss_cache):
9664 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
9666 case offsetof(struct bpf_sock_ops, ecn_flags):
9667 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
9669 case offsetof(struct bpf_sock_ops, rate_delivered):
9670 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
9672 case offsetof(struct bpf_sock_ops, rate_interval_us):
9673 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
9675 case offsetof(struct bpf_sock_ops, packets_out):
9676 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
9678 case offsetof(struct bpf_sock_ops, retrans_out):
9679 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
9681 case offsetof(struct bpf_sock_ops, total_retrans):
9682 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
9684 case offsetof(struct bpf_sock_ops, segs_in):
9685 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
9687 case offsetof(struct bpf_sock_ops, data_segs_in):
9688 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
9690 case offsetof(struct bpf_sock_ops, segs_out):
9691 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
9693 case offsetof(struct bpf_sock_ops, data_segs_out):
9694 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
9696 case offsetof(struct bpf_sock_ops, lost_out):
9697 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
9699 case offsetof(struct bpf_sock_ops, sacked_out):
9700 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
9702 case offsetof(struct bpf_sock_ops, bytes_received):
9703 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
9705 case offsetof(struct bpf_sock_ops, bytes_acked):
9706 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
9708 case offsetof(struct bpf_sock_ops, sk):
9711 case offsetof(struct bpf_sock_ops, skb_data_end):
9712 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9714 si->dst_reg, si->src_reg,
9715 offsetof(struct bpf_sock_ops_kern,
9718 case offsetof(struct bpf_sock_ops, skb_data):
9719 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9721 si->dst_reg, si->src_reg,
9722 offsetof(struct bpf_sock_ops_kern,
9724 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9725 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9726 si->dst_reg, si->dst_reg,
9727 offsetof(struct sk_buff, data));
9729 case offsetof(struct bpf_sock_ops, skb_len):
9730 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9732 si->dst_reg, si->src_reg,
9733 offsetof(struct bpf_sock_ops_kern,
9735 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9736 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9737 si->dst_reg, si->dst_reg,
9738 offsetof(struct sk_buff, len));
9740 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9741 off = offsetof(struct sk_buff, cb);
9742 off += offsetof(struct tcp_skb_cb, tcp_flags);
9743 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
9744 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9746 si->dst_reg, si->src_reg,
9747 offsetof(struct bpf_sock_ops_kern,
9749 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9750 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
9752 si->dst_reg, si->dst_reg, off);
9755 return insn - insn_buf;
9758 /* data_end = skb->data + skb_headlen() */
9759 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
9760 struct bpf_insn *insn)
9763 int temp_reg_off = offsetof(struct sk_buff, cb) +
9764 offsetof(struct sk_skb_cb, temp_reg);
9766 if (si->src_reg == si->dst_reg) {
9767 /* We need an extra register, choose and save a register. */
9769 if (si->src_reg == reg || si->dst_reg == reg)
9771 if (si->src_reg == reg || si->dst_reg == reg)
9773 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off);
9778 /* reg = skb->data */
9779 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9781 offsetof(struct sk_buff, data));
9783 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9784 BPF_REG_AX, si->src_reg,
9785 offsetof(struct sk_buff, len));
9786 /* reg = skb->data + skb->len */
9787 *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX);
9788 /* AX = skb->data_len */
9789 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
9790 BPF_REG_AX, si->src_reg,
9791 offsetof(struct sk_buff, data_len));
9793 /* reg = skb->data + skb->len - skb->data_len */
9794 *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX);
9796 if (si->src_reg == si->dst_reg) {
9797 /* Restore the saved register */
9798 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg);
9799 *insn++ = BPF_MOV64_REG(si->dst_reg, reg);
9800 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off);
9806 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
9807 const struct bpf_insn *si,
9808 struct bpf_insn *insn_buf,
9809 struct bpf_prog *prog, u32 *target_size)
9811 struct bpf_insn *insn = insn_buf;
9815 case offsetof(struct __sk_buff, data_end):
9816 insn = bpf_convert_data_end_access(si, insn);
9818 case offsetof(struct __sk_buff, cb[0]) ...
9819 offsetofend(struct __sk_buff, cb[4]) - 1:
9820 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20);
9821 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
9822 offsetof(struct sk_skb_cb, data)) %
9825 prog->cb_access = 1;
9827 off -= offsetof(struct __sk_buff, cb[0]);
9828 off += offsetof(struct sk_buff, cb);
9829 off += offsetof(struct sk_skb_cb, data);
9830 if (type == BPF_WRITE)
9831 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
9834 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
9840 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9844 return insn - insn_buf;
9847 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
9848 const struct bpf_insn *si,
9849 struct bpf_insn *insn_buf,
9850 struct bpf_prog *prog, u32 *target_size)
9852 struct bpf_insn *insn = insn_buf;
9853 #if IS_ENABLED(CONFIG_IPV6)
9857 /* convert ctx uses the fact sg element is first in struct */
9858 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
9861 case offsetof(struct sk_msg_md, data):
9862 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
9863 si->dst_reg, si->src_reg,
9864 offsetof(struct sk_msg, data));
9866 case offsetof(struct sk_msg_md, data_end):
9867 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
9868 si->dst_reg, si->src_reg,
9869 offsetof(struct sk_msg, data_end));
9871 case offsetof(struct sk_msg_md, family):
9872 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9874 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9876 si->dst_reg, si->src_reg,
9877 offsetof(struct sk_msg, sk));
9878 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9879 offsetof(struct sock_common, skc_family));
9882 case offsetof(struct sk_msg_md, remote_ip4):
9883 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9885 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9887 si->dst_reg, si->src_reg,
9888 offsetof(struct sk_msg, sk));
9889 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9890 offsetof(struct sock_common, skc_daddr));
9893 case offsetof(struct sk_msg_md, local_ip4):
9894 BUILD_BUG_ON(sizeof_field(struct sock_common,
9895 skc_rcv_saddr) != 4);
9897 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9899 si->dst_reg, si->src_reg,
9900 offsetof(struct sk_msg, sk));
9901 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9902 offsetof(struct sock_common,
9906 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
9907 offsetof(struct sk_msg_md, remote_ip6[3]):
9908 #if IS_ENABLED(CONFIG_IPV6)
9909 BUILD_BUG_ON(sizeof_field(struct sock_common,
9910 skc_v6_daddr.s6_addr32[0]) != 4);
9913 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
9914 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9916 si->dst_reg, si->src_reg,
9917 offsetof(struct sk_msg, sk));
9918 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9919 offsetof(struct sock_common,
9920 skc_v6_daddr.s6_addr32[0]) +
9923 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9927 case offsetof(struct sk_msg_md, local_ip6[0]) ...
9928 offsetof(struct sk_msg_md, local_ip6[3]):
9929 #if IS_ENABLED(CONFIG_IPV6)
9930 BUILD_BUG_ON(sizeof_field(struct sock_common,
9931 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9934 off -= offsetof(struct sk_msg_md, local_ip6[0]);
9935 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9937 si->dst_reg, si->src_reg,
9938 offsetof(struct sk_msg, sk));
9939 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9940 offsetof(struct sock_common,
9941 skc_v6_rcv_saddr.s6_addr32[0]) +
9944 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9948 case offsetof(struct sk_msg_md, remote_port):
9949 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9951 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9953 si->dst_reg, si->src_reg,
9954 offsetof(struct sk_msg, sk));
9955 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9956 offsetof(struct sock_common, skc_dport));
9957 #ifndef __BIG_ENDIAN_BITFIELD
9958 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9962 case offsetof(struct sk_msg_md, local_port):
9963 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9965 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9967 si->dst_reg, si->src_reg,
9968 offsetof(struct sk_msg, sk));
9969 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9970 offsetof(struct sock_common, skc_num));
9973 case offsetof(struct sk_msg_md, size):
9974 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
9975 si->dst_reg, si->src_reg,
9976 offsetof(struct sk_msg_sg, size));
9979 case offsetof(struct sk_msg_md, sk):
9980 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
9981 si->dst_reg, si->src_reg,
9982 offsetof(struct sk_msg, sk));
9986 return insn - insn_buf;
9989 const struct bpf_verifier_ops sk_filter_verifier_ops = {
9990 .get_func_proto = sk_filter_func_proto,
9991 .is_valid_access = sk_filter_is_valid_access,
9992 .convert_ctx_access = bpf_convert_ctx_access,
9993 .gen_ld_abs = bpf_gen_ld_abs,
9996 const struct bpf_prog_ops sk_filter_prog_ops = {
9997 .test_run = bpf_prog_test_run_skb,
10000 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
10001 .get_func_proto = tc_cls_act_func_proto,
10002 .is_valid_access = tc_cls_act_is_valid_access,
10003 .convert_ctx_access = tc_cls_act_convert_ctx_access,
10004 .gen_prologue = tc_cls_act_prologue,
10005 .gen_ld_abs = bpf_gen_ld_abs,
10006 .check_kfunc_call = bpf_prog_test_check_kfunc_call,
10009 const struct bpf_prog_ops tc_cls_act_prog_ops = {
10010 .test_run = bpf_prog_test_run_skb,
10013 const struct bpf_verifier_ops xdp_verifier_ops = {
10014 .get_func_proto = xdp_func_proto,
10015 .is_valid_access = xdp_is_valid_access,
10016 .convert_ctx_access = xdp_convert_ctx_access,
10017 .gen_prologue = bpf_noop_prologue,
10020 const struct bpf_prog_ops xdp_prog_ops = {
10021 .test_run = bpf_prog_test_run_xdp,
10024 const struct bpf_verifier_ops cg_skb_verifier_ops = {
10025 .get_func_proto = cg_skb_func_proto,
10026 .is_valid_access = cg_skb_is_valid_access,
10027 .convert_ctx_access = bpf_convert_ctx_access,
10030 const struct bpf_prog_ops cg_skb_prog_ops = {
10031 .test_run = bpf_prog_test_run_skb,
10034 const struct bpf_verifier_ops lwt_in_verifier_ops = {
10035 .get_func_proto = lwt_in_func_proto,
10036 .is_valid_access = lwt_is_valid_access,
10037 .convert_ctx_access = bpf_convert_ctx_access,
10040 const struct bpf_prog_ops lwt_in_prog_ops = {
10041 .test_run = bpf_prog_test_run_skb,
10044 const struct bpf_verifier_ops lwt_out_verifier_ops = {
10045 .get_func_proto = lwt_out_func_proto,
10046 .is_valid_access = lwt_is_valid_access,
10047 .convert_ctx_access = bpf_convert_ctx_access,
10050 const struct bpf_prog_ops lwt_out_prog_ops = {
10051 .test_run = bpf_prog_test_run_skb,
10054 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
10055 .get_func_proto = lwt_xmit_func_proto,
10056 .is_valid_access = lwt_is_valid_access,
10057 .convert_ctx_access = bpf_convert_ctx_access,
10058 .gen_prologue = tc_cls_act_prologue,
10061 const struct bpf_prog_ops lwt_xmit_prog_ops = {
10062 .test_run = bpf_prog_test_run_skb,
10065 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
10066 .get_func_proto = lwt_seg6local_func_proto,
10067 .is_valid_access = lwt_is_valid_access,
10068 .convert_ctx_access = bpf_convert_ctx_access,
10071 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
10072 .test_run = bpf_prog_test_run_skb,
10075 const struct bpf_verifier_ops cg_sock_verifier_ops = {
10076 .get_func_proto = sock_filter_func_proto,
10077 .is_valid_access = sock_filter_is_valid_access,
10078 .convert_ctx_access = bpf_sock_convert_ctx_access,
10081 const struct bpf_prog_ops cg_sock_prog_ops = {
10084 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
10085 .get_func_proto = sock_addr_func_proto,
10086 .is_valid_access = sock_addr_is_valid_access,
10087 .convert_ctx_access = sock_addr_convert_ctx_access,
10090 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
10093 const struct bpf_verifier_ops sock_ops_verifier_ops = {
10094 .get_func_proto = sock_ops_func_proto,
10095 .is_valid_access = sock_ops_is_valid_access,
10096 .convert_ctx_access = sock_ops_convert_ctx_access,
10099 const struct bpf_prog_ops sock_ops_prog_ops = {
10102 const struct bpf_verifier_ops sk_skb_verifier_ops = {
10103 .get_func_proto = sk_skb_func_proto,
10104 .is_valid_access = sk_skb_is_valid_access,
10105 .convert_ctx_access = sk_skb_convert_ctx_access,
10106 .gen_prologue = sk_skb_prologue,
10109 const struct bpf_prog_ops sk_skb_prog_ops = {
10112 const struct bpf_verifier_ops sk_msg_verifier_ops = {
10113 .get_func_proto = sk_msg_func_proto,
10114 .is_valid_access = sk_msg_is_valid_access,
10115 .convert_ctx_access = sk_msg_convert_ctx_access,
10116 .gen_prologue = bpf_noop_prologue,
10119 const struct bpf_prog_ops sk_msg_prog_ops = {
10122 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
10123 .get_func_proto = flow_dissector_func_proto,
10124 .is_valid_access = flow_dissector_is_valid_access,
10125 .convert_ctx_access = flow_dissector_convert_ctx_access,
10128 const struct bpf_prog_ops flow_dissector_prog_ops = {
10129 .test_run = bpf_prog_test_run_flow_dissector,
10132 int sk_detach_filter(struct sock *sk)
10135 struct sk_filter *filter;
10137 if (sock_flag(sk, SOCK_FILTER_LOCKED))
10140 filter = rcu_dereference_protected(sk->sk_filter,
10141 lockdep_sock_is_held(sk));
10143 RCU_INIT_POINTER(sk->sk_filter, NULL);
10144 sk_filter_uncharge(sk, filter);
10150 EXPORT_SYMBOL_GPL(sk_detach_filter);
10152 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
10155 struct sock_fprog_kern *fprog;
10156 struct sk_filter *filter;
10160 filter = rcu_dereference_protected(sk->sk_filter,
10161 lockdep_sock_is_held(sk));
10165 /* We're copying the filter that has been originally attached,
10166 * so no conversion/decode needed anymore. eBPF programs that
10167 * have no original program cannot be dumped through this.
10170 fprog = filter->prog->orig_prog;
10176 /* User space only enquires number of filter blocks. */
10180 if (len < fprog->len)
10184 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
10187 /* Instead of bytes, the API requests to return the number
10188 * of filter blocks.
10197 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
10198 struct sock_reuseport *reuse,
10199 struct sock *sk, struct sk_buff *skb,
10200 struct sock *migrating_sk,
10203 reuse_kern->skb = skb;
10204 reuse_kern->sk = sk;
10205 reuse_kern->selected_sk = NULL;
10206 reuse_kern->migrating_sk = migrating_sk;
10207 reuse_kern->data_end = skb->data + skb_headlen(skb);
10208 reuse_kern->hash = hash;
10209 reuse_kern->reuseport_id = reuse->reuseport_id;
10210 reuse_kern->bind_inany = reuse->bind_inany;
10213 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
10214 struct bpf_prog *prog, struct sk_buff *skb,
10215 struct sock *migrating_sk,
10218 struct sk_reuseport_kern reuse_kern;
10219 enum sk_action action;
10221 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
10222 action = bpf_prog_run(prog, &reuse_kern);
10224 if (action == SK_PASS)
10225 return reuse_kern.selected_sk;
10227 return ERR_PTR(-ECONNREFUSED);
10230 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
10231 struct bpf_map *, map, void *, key, u32, flags)
10233 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
10234 struct sock_reuseport *reuse;
10235 struct sock *selected_sk;
10237 selected_sk = map->ops->map_lookup_elem(map, key);
10241 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
10243 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
10244 if (sk_is_refcounted(selected_sk))
10245 sock_put(selected_sk);
10247 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
10248 * The only (!reuse) case here is - the sk has already been
10249 * unhashed (e.g. by close()), so treat it as -ENOENT.
10251 * Other maps (e.g. sock_map) do not provide this guarantee and
10252 * the sk may never be in the reuseport group to begin with.
10254 return is_sockarray ? -ENOENT : -EINVAL;
10257 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
10258 struct sock *sk = reuse_kern->sk;
10260 if (sk->sk_protocol != selected_sk->sk_protocol)
10261 return -EPROTOTYPE;
10262 else if (sk->sk_family != selected_sk->sk_family)
10263 return -EAFNOSUPPORT;
10265 /* Catch all. Likely bound to a different sockaddr. */
10269 reuse_kern->selected_sk = selected_sk;
10274 static const struct bpf_func_proto sk_select_reuseport_proto = {
10275 .func = sk_select_reuseport,
10277 .ret_type = RET_INTEGER,
10278 .arg1_type = ARG_PTR_TO_CTX,
10279 .arg2_type = ARG_CONST_MAP_PTR,
10280 .arg3_type = ARG_PTR_TO_MAP_KEY,
10281 .arg4_type = ARG_ANYTHING,
10284 BPF_CALL_4(sk_reuseport_load_bytes,
10285 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10286 void *, to, u32, len)
10288 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
10291 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
10292 .func = sk_reuseport_load_bytes,
10294 .ret_type = RET_INTEGER,
10295 .arg1_type = ARG_PTR_TO_CTX,
10296 .arg2_type = ARG_ANYTHING,
10297 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
10298 .arg4_type = ARG_CONST_SIZE,
10301 BPF_CALL_5(sk_reuseport_load_bytes_relative,
10302 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10303 void *, to, u32, len, u32, start_header)
10305 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
10306 len, start_header);
10309 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
10310 .func = sk_reuseport_load_bytes_relative,
10312 .ret_type = RET_INTEGER,
10313 .arg1_type = ARG_PTR_TO_CTX,
10314 .arg2_type = ARG_ANYTHING,
10315 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
10316 .arg4_type = ARG_CONST_SIZE,
10317 .arg5_type = ARG_ANYTHING,
10320 static const struct bpf_func_proto *
10321 sk_reuseport_func_proto(enum bpf_func_id func_id,
10322 const struct bpf_prog *prog)
10325 case BPF_FUNC_sk_select_reuseport:
10326 return &sk_select_reuseport_proto;
10327 case BPF_FUNC_skb_load_bytes:
10328 return &sk_reuseport_load_bytes_proto;
10329 case BPF_FUNC_skb_load_bytes_relative:
10330 return &sk_reuseport_load_bytes_relative_proto;
10331 case BPF_FUNC_get_socket_cookie:
10332 return &bpf_get_socket_ptr_cookie_proto;
10333 case BPF_FUNC_ktime_get_coarse_ns:
10334 return &bpf_ktime_get_coarse_ns_proto;
10336 return bpf_base_func_proto(func_id);
10341 sk_reuseport_is_valid_access(int off, int size,
10342 enum bpf_access_type type,
10343 const struct bpf_prog *prog,
10344 struct bpf_insn_access_aux *info)
10346 const u32 size_default = sizeof(__u32);
10348 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
10349 off % size || type != BPF_READ)
10353 case offsetof(struct sk_reuseport_md, data):
10354 info->reg_type = PTR_TO_PACKET;
10355 return size == sizeof(__u64);
10357 case offsetof(struct sk_reuseport_md, data_end):
10358 info->reg_type = PTR_TO_PACKET_END;
10359 return size == sizeof(__u64);
10361 case offsetof(struct sk_reuseport_md, hash):
10362 return size == size_default;
10364 case offsetof(struct sk_reuseport_md, sk):
10365 info->reg_type = PTR_TO_SOCKET;
10366 return size == sizeof(__u64);
10368 case offsetof(struct sk_reuseport_md, migrating_sk):
10369 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
10370 return size == sizeof(__u64);
10372 /* Fields that allow narrowing */
10373 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
10374 if (size < sizeof_field(struct sk_buff, protocol))
10377 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
10378 case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
10379 case bpf_ctx_range(struct sk_reuseport_md, len):
10380 bpf_ctx_record_field_size(info, size_default);
10381 return bpf_ctx_narrow_access_ok(off, size, size_default);
10388 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
10389 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
10390 si->dst_reg, si->src_reg, \
10391 bpf_target_off(struct sk_reuseport_kern, F, \
10392 sizeof_field(struct sk_reuseport_kern, F), \
10396 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
10397 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
10402 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \
10403 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
10408 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
10409 const struct bpf_insn *si,
10410 struct bpf_insn *insn_buf,
10411 struct bpf_prog *prog,
10414 struct bpf_insn *insn = insn_buf;
10417 case offsetof(struct sk_reuseport_md, data):
10418 SK_REUSEPORT_LOAD_SKB_FIELD(data);
10421 case offsetof(struct sk_reuseport_md, len):
10422 SK_REUSEPORT_LOAD_SKB_FIELD(len);
10425 case offsetof(struct sk_reuseport_md, eth_protocol):
10426 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
10429 case offsetof(struct sk_reuseport_md, ip_protocol):
10430 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
10433 case offsetof(struct sk_reuseport_md, data_end):
10434 SK_REUSEPORT_LOAD_FIELD(data_end);
10437 case offsetof(struct sk_reuseport_md, hash):
10438 SK_REUSEPORT_LOAD_FIELD(hash);
10441 case offsetof(struct sk_reuseport_md, bind_inany):
10442 SK_REUSEPORT_LOAD_FIELD(bind_inany);
10445 case offsetof(struct sk_reuseport_md, sk):
10446 SK_REUSEPORT_LOAD_FIELD(sk);
10449 case offsetof(struct sk_reuseport_md, migrating_sk):
10450 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
10454 return insn - insn_buf;
10457 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
10458 .get_func_proto = sk_reuseport_func_proto,
10459 .is_valid_access = sk_reuseport_is_valid_access,
10460 .convert_ctx_access = sk_reuseport_convert_ctx_access,
10463 const struct bpf_prog_ops sk_reuseport_prog_ops = {
10466 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
10467 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
10469 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
10470 struct sock *, sk, u64, flags)
10472 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
10473 BPF_SK_LOOKUP_F_NO_REUSEPORT)))
10475 if (unlikely(sk && sk_is_refcounted(sk)))
10476 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
10477 if (unlikely(sk && sk->sk_state == TCP_ESTABLISHED))
10478 return -ESOCKTNOSUPPORT; /* reject connected sockets */
10480 /* Check if socket is suitable for packet L3/L4 protocol */
10481 if (sk && sk->sk_protocol != ctx->protocol)
10482 return -EPROTOTYPE;
10483 if (sk && sk->sk_family != ctx->family &&
10484 (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
10485 return -EAFNOSUPPORT;
10487 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
10490 /* Select socket as lookup result */
10491 ctx->selected_sk = sk;
10492 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
10496 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
10497 .func = bpf_sk_lookup_assign,
10499 .ret_type = RET_INTEGER,
10500 .arg1_type = ARG_PTR_TO_CTX,
10501 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL,
10502 .arg3_type = ARG_ANYTHING,
10505 static const struct bpf_func_proto *
10506 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
10509 case BPF_FUNC_perf_event_output:
10510 return &bpf_event_output_data_proto;
10511 case BPF_FUNC_sk_assign:
10512 return &bpf_sk_lookup_assign_proto;
10513 case BPF_FUNC_sk_release:
10514 return &bpf_sk_release_proto;
10516 return bpf_sk_base_func_proto(func_id);
10520 static bool sk_lookup_is_valid_access(int off, int size,
10521 enum bpf_access_type type,
10522 const struct bpf_prog *prog,
10523 struct bpf_insn_access_aux *info)
10525 if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
10527 if (off % size != 0)
10529 if (type != BPF_READ)
10533 case offsetof(struct bpf_sk_lookup, sk):
10534 info->reg_type = PTR_TO_SOCKET_OR_NULL;
10535 return size == sizeof(__u64);
10537 case bpf_ctx_range(struct bpf_sk_lookup, family):
10538 case bpf_ctx_range(struct bpf_sk_lookup, protocol):
10539 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
10540 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
10541 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
10542 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
10543 case offsetof(struct bpf_sk_lookup, remote_port) ...
10544 offsetof(struct bpf_sk_lookup, local_ip4) - 1:
10545 case bpf_ctx_range(struct bpf_sk_lookup, local_port):
10546 bpf_ctx_record_field_size(info, sizeof(__u32));
10547 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
10554 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
10555 const struct bpf_insn *si,
10556 struct bpf_insn *insn_buf,
10557 struct bpf_prog *prog,
10560 struct bpf_insn *insn = insn_buf;
10563 case offsetof(struct bpf_sk_lookup, sk):
10564 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10565 offsetof(struct bpf_sk_lookup_kern, selected_sk));
10568 case offsetof(struct bpf_sk_lookup, family):
10569 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10570 bpf_target_off(struct bpf_sk_lookup_kern,
10571 family, 2, target_size));
10574 case offsetof(struct bpf_sk_lookup, protocol):
10575 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10576 bpf_target_off(struct bpf_sk_lookup_kern,
10577 protocol, 2, target_size));
10580 case offsetof(struct bpf_sk_lookup, remote_ip4):
10581 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10582 bpf_target_off(struct bpf_sk_lookup_kern,
10583 v4.saddr, 4, target_size));
10586 case offsetof(struct bpf_sk_lookup, local_ip4):
10587 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10588 bpf_target_off(struct bpf_sk_lookup_kern,
10589 v4.daddr, 4, target_size));
10592 case bpf_ctx_range_till(struct bpf_sk_lookup,
10593 remote_ip6[0], remote_ip6[3]): {
10594 #if IS_ENABLED(CONFIG_IPV6)
10597 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
10598 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10599 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10600 offsetof(struct bpf_sk_lookup_kern, v6.saddr));
10601 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10602 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10604 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10608 case bpf_ctx_range_till(struct bpf_sk_lookup,
10609 local_ip6[0], local_ip6[3]): {
10610 #if IS_ENABLED(CONFIG_IPV6)
10613 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
10614 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10615 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10616 offsetof(struct bpf_sk_lookup_kern, v6.daddr));
10617 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10618 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10620 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10624 case offsetof(struct bpf_sk_lookup, remote_port):
10625 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10626 bpf_target_off(struct bpf_sk_lookup_kern,
10627 sport, 2, target_size));
10630 case offsetof(struct bpf_sk_lookup, local_port):
10631 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10632 bpf_target_off(struct bpf_sk_lookup_kern,
10633 dport, 2, target_size));
10637 return insn - insn_buf;
10640 const struct bpf_prog_ops sk_lookup_prog_ops = {
10641 .test_run = bpf_prog_test_run_sk_lookup,
10644 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
10645 .get_func_proto = sk_lookup_func_proto,
10646 .is_valid_access = sk_lookup_is_valid_access,
10647 .convert_ctx_access = sk_lookup_convert_ctx_access,
10650 #endif /* CONFIG_INET */
10652 DEFINE_BPF_DISPATCHER(xdp)
10654 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
10656 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
10659 #ifdef CONFIG_DEBUG_INFO_BTF
10660 BTF_ID_LIST_GLOBAL(btf_sock_ids)
10661 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
10663 #undef BTF_SOCK_TYPE
10665 u32 btf_sock_ids[MAX_BTF_SOCK_TYPE];
10668 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
10670 /* tcp6_sock type is not generated in dwarf and hence btf,
10671 * trigger an explicit type generation here.
10673 BTF_TYPE_EMIT(struct tcp6_sock);
10674 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
10675 sk->sk_family == AF_INET6)
10676 return (unsigned long)sk;
10678 return (unsigned long)NULL;
10681 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
10682 .func = bpf_skc_to_tcp6_sock,
10684 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10685 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10686 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
10689 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
10691 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
10692 return (unsigned long)sk;
10694 return (unsigned long)NULL;
10697 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
10698 .func = bpf_skc_to_tcp_sock,
10700 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10701 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10702 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
10705 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
10707 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
10708 * generated if CONFIG_INET=n. Trigger an explicit generation here.
10710 BTF_TYPE_EMIT(struct inet_timewait_sock);
10711 BTF_TYPE_EMIT(struct tcp_timewait_sock);
10714 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
10715 return (unsigned long)sk;
10718 #if IS_BUILTIN(CONFIG_IPV6)
10719 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
10720 return (unsigned long)sk;
10723 return (unsigned long)NULL;
10726 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
10727 .func = bpf_skc_to_tcp_timewait_sock,
10729 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10730 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10731 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
10734 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
10737 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10738 return (unsigned long)sk;
10741 #if IS_BUILTIN(CONFIG_IPV6)
10742 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10743 return (unsigned long)sk;
10746 return (unsigned long)NULL;
10749 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
10750 .func = bpf_skc_to_tcp_request_sock,
10752 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10753 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10754 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
10757 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
10759 /* udp6_sock type is not generated in dwarf and hence btf,
10760 * trigger an explicit type generation here.
10762 BTF_TYPE_EMIT(struct udp6_sock);
10763 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
10764 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
10765 return (unsigned long)sk;
10767 return (unsigned long)NULL;
10770 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
10771 .func = bpf_skc_to_udp6_sock,
10773 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10774 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10775 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
10778 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
10780 return (unsigned long)sock_from_file(file);
10783 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
10784 BTF_ID(struct, socket)
10785 BTF_ID(struct, file)
10787 const struct bpf_func_proto bpf_sock_from_file_proto = {
10788 .func = bpf_sock_from_file,
10790 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10791 .ret_btf_id = &bpf_sock_from_file_btf_ids[0],
10792 .arg1_type = ARG_PTR_TO_BTF_ID,
10793 .arg1_btf_id = &bpf_sock_from_file_btf_ids[1],
10796 static const struct bpf_func_proto *
10797 bpf_sk_base_func_proto(enum bpf_func_id func_id)
10799 const struct bpf_func_proto *func;
10802 case BPF_FUNC_skc_to_tcp6_sock:
10803 func = &bpf_skc_to_tcp6_sock_proto;
10805 case BPF_FUNC_skc_to_tcp_sock:
10806 func = &bpf_skc_to_tcp_sock_proto;
10808 case BPF_FUNC_skc_to_tcp_timewait_sock:
10809 func = &bpf_skc_to_tcp_timewait_sock_proto;
10811 case BPF_FUNC_skc_to_tcp_request_sock:
10812 func = &bpf_skc_to_tcp_request_sock_proto;
10814 case BPF_FUNC_skc_to_udp6_sock:
10815 func = &bpf_skc_to_udp6_sock_proto;
10817 case BPF_FUNC_ktime_get_coarse_ns:
10818 return &bpf_ktime_get_coarse_ns_proto;
10820 return bpf_base_func_proto(func_id);
10823 if (!perfmon_capable())