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 won't be used at
1246 * this point in time anymore internally after the migration to the eBPF
1247 * instruction representation.
1249 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1250 sizeof(struct bpf_insn));
1252 /* Conversion cannot happen on overlapping memory areas,
1253 * so we need to keep the user BPF around until the 2nd
1254 * pass. At this time, the user BPF is stored in fp->insns.
1256 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1257 GFP_KERNEL | __GFP_NOWARN);
1263 /* 1st pass: calculate the new program length. */
1264 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1269 /* Expand fp for appending the new filter representation. */
1271 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1273 /* The old_fp is still around in case we couldn't
1274 * allocate new memory, so uncharge on that one.
1283 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1284 err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1287 /* 2nd bpf_convert_filter() can fail only if it fails
1288 * to allocate memory, remapping must succeed. Note,
1289 * that at this time old_fp has already been released
1294 fp = bpf_prog_select_runtime(fp, &err);
1304 __bpf_prog_release(fp);
1305 return ERR_PTR(err);
1308 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1309 bpf_aux_classic_check_t trans)
1313 fp->bpf_func = NULL;
1316 err = bpf_check_classic(fp->insns, fp->len);
1318 __bpf_prog_release(fp);
1319 return ERR_PTR(err);
1322 /* There might be additional checks and transformations
1323 * needed on classic filters, f.e. in case of seccomp.
1326 err = trans(fp->insns, fp->len);
1328 __bpf_prog_release(fp);
1329 return ERR_PTR(err);
1333 /* Probe if we can JIT compile the filter and if so, do
1334 * the compilation of the filter.
1336 bpf_jit_compile(fp);
1338 /* JIT compiler couldn't process this filter, so do the eBPF translation
1339 * for the optimized interpreter.
1342 fp = bpf_migrate_filter(fp);
1348 * bpf_prog_create - create an unattached filter
1349 * @pfp: the unattached filter that is created
1350 * @fprog: the filter program
1352 * Create a filter independent of any socket. We first run some
1353 * sanity checks on it to make sure it does not explode on us later.
1354 * If an error occurs or there is insufficient memory for the filter
1355 * a negative errno code is returned. On success the return is zero.
1357 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1359 unsigned int fsize = bpf_classic_proglen(fprog);
1360 struct bpf_prog *fp;
1362 /* Make sure new filter is there and in the right amounts. */
1363 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1366 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1370 memcpy(fp->insns, fprog->filter, fsize);
1372 fp->len = fprog->len;
1373 /* Since unattached filters are not copied back to user
1374 * space through sk_get_filter(), we do not need to hold
1375 * a copy here, and can spare us the work.
1377 fp->orig_prog = NULL;
1379 /* bpf_prepare_filter() already takes care of freeing
1380 * memory in case something goes wrong.
1382 fp = bpf_prepare_filter(fp, NULL);
1389 EXPORT_SYMBOL_GPL(bpf_prog_create);
1392 * bpf_prog_create_from_user - create an unattached filter from user buffer
1393 * @pfp: the unattached filter that is created
1394 * @fprog: the filter program
1395 * @trans: post-classic verifier transformation handler
1396 * @save_orig: save classic BPF program
1398 * This function effectively does the same as bpf_prog_create(), only
1399 * that it builds up its insns buffer from user space provided buffer.
1400 * It also allows for passing a bpf_aux_classic_check_t handler.
1402 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1403 bpf_aux_classic_check_t trans, bool save_orig)
1405 unsigned int fsize = bpf_classic_proglen(fprog);
1406 struct bpf_prog *fp;
1409 /* Make sure new filter is there and in the right amounts. */
1410 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1413 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1417 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1418 __bpf_prog_free(fp);
1422 fp->len = fprog->len;
1423 fp->orig_prog = NULL;
1426 err = bpf_prog_store_orig_filter(fp, fprog);
1428 __bpf_prog_free(fp);
1433 /* bpf_prepare_filter() already takes care of freeing
1434 * memory in case something goes wrong.
1436 fp = bpf_prepare_filter(fp, trans);
1443 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1445 void bpf_prog_destroy(struct bpf_prog *fp)
1447 __bpf_prog_release(fp);
1449 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1451 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1453 struct sk_filter *fp, *old_fp;
1455 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1461 if (!__sk_filter_charge(sk, fp)) {
1465 refcount_set(&fp->refcnt, 1);
1467 old_fp = rcu_dereference_protected(sk->sk_filter,
1468 lockdep_sock_is_held(sk));
1469 rcu_assign_pointer(sk->sk_filter, fp);
1472 sk_filter_uncharge(sk, old_fp);
1478 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1480 unsigned int fsize = bpf_classic_proglen(fprog);
1481 struct bpf_prog *prog;
1484 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1485 return ERR_PTR(-EPERM);
1487 /* Make sure new filter is there and in the right amounts. */
1488 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1489 return ERR_PTR(-EINVAL);
1491 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1493 return ERR_PTR(-ENOMEM);
1495 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1496 __bpf_prog_free(prog);
1497 return ERR_PTR(-EFAULT);
1500 prog->len = fprog->len;
1502 err = bpf_prog_store_orig_filter(prog, fprog);
1504 __bpf_prog_free(prog);
1505 return ERR_PTR(-ENOMEM);
1508 /* bpf_prepare_filter() already takes care of freeing
1509 * memory in case something goes wrong.
1511 return bpf_prepare_filter(prog, NULL);
1515 * sk_attach_filter - attach a socket filter
1516 * @fprog: the filter program
1517 * @sk: the socket to use
1519 * Attach the user's filter code. We first run some sanity checks on
1520 * it to make sure it does not explode on us later. If an error
1521 * occurs or there is insufficient memory for the filter a negative
1522 * errno code is returned. On success the return is zero.
1524 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1526 struct bpf_prog *prog = __get_filter(fprog, sk);
1530 return PTR_ERR(prog);
1532 err = __sk_attach_prog(prog, sk);
1534 __bpf_prog_release(prog);
1540 EXPORT_SYMBOL_GPL(sk_attach_filter);
1542 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1544 struct bpf_prog *prog = __get_filter(fprog, sk);
1548 return PTR_ERR(prog);
1550 if (bpf_prog_size(prog->len) > sysctl_optmem_max)
1553 err = reuseport_attach_prog(sk, prog);
1556 __bpf_prog_release(prog);
1561 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1563 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1564 return ERR_PTR(-EPERM);
1566 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1569 int sk_attach_bpf(u32 ufd, struct sock *sk)
1571 struct bpf_prog *prog = __get_bpf(ufd, sk);
1575 return PTR_ERR(prog);
1577 err = __sk_attach_prog(prog, sk);
1586 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1588 struct bpf_prog *prog;
1591 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1594 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1595 if (PTR_ERR(prog) == -EINVAL)
1596 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1598 return PTR_ERR(prog);
1600 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1601 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1602 * bpf prog (e.g. sockmap). It depends on the
1603 * limitation imposed by bpf_prog_load().
1604 * Hence, sysctl_optmem_max is not checked.
1606 if ((sk->sk_type != SOCK_STREAM &&
1607 sk->sk_type != SOCK_DGRAM) ||
1608 (sk->sk_protocol != IPPROTO_UDP &&
1609 sk->sk_protocol != IPPROTO_TCP) ||
1610 (sk->sk_family != AF_INET &&
1611 sk->sk_family != AF_INET6)) {
1616 /* BPF_PROG_TYPE_SOCKET_FILTER */
1617 if (bpf_prog_size(prog->len) > sysctl_optmem_max) {
1623 err = reuseport_attach_prog(sk, prog);
1631 void sk_reuseport_prog_free(struct bpf_prog *prog)
1636 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1639 bpf_prog_destroy(prog);
1642 struct bpf_scratchpad {
1644 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1645 u8 buff[MAX_BPF_STACK];
1649 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1651 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1652 unsigned int write_len)
1654 return skb_ensure_writable(skb, write_len);
1657 static inline int bpf_try_make_writable(struct sk_buff *skb,
1658 unsigned int write_len)
1660 int err = __bpf_try_make_writable(skb, write_len);
1662 bpf_compute_data_pointers(skb);
1666 static int bpf_try_make_head_writable(struct sk_buff *skb)
1668 return bpf_try_make_writable(skb, skb_headlen(skb));
1671 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1673 if (skb_at_tc_ingress(skb))
1674 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1677 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1679 if (skb_at_tc_ingress(skb))
1680 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1683 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1684 const void *, from, u32, len, u64, flags)
1688 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1690 if (unlikely(offset > 0xffff))
1692 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1695 ptr = skb->data + offset;
1696 if (flags & BPF_F_RECOMPUTE_CSUM)
1697 __skb_postpull_rcsum(skb, ptr, len, offset);
1699 memcpy(ptr, from, len);
1701 if (flags & BPF_F_RECOMPUTE_CSUM)
1702 __skb_postpush_rcsum(skb, ptr, len, offset);
1703 if (flags & BPF_F_INVALIDATE_HASH)
1704 skb_clear_hash(skb);
1709 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1710 .func = bpf_skb_store_bytes,
1712 .ret_type = RET_INTEGER,
1713 .arg1_type = ARG_PTR_TO_CTX,
1714 .arg2_type = ARG_ANYTHING,
1715 .arg3_type = ARG_PTR_TO_MEM,
1716 .arg4_type = ARG_CONST_SIZE,
1717 .arg5_type = ARG_ANYTHING,
1720 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1721 void *, to, u32, len)
1725 if (unlikely(offset > 0xffff))
1728 ptr = skb_header_pointer(skb, offset, len, to);
1732 memcpy(to, ptr, len);
1740 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1741 .func = bpf_skb_load_bytes,
1743 .ret_type = RET_INTEGER,
1744 .arg1_type = ARG_PTR_TO_CTX,
1745 .arg2_type = ARG_ANYTHING,
1746 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1747 .arg4_type = ARG_CONST_SIZE,
1750 BPF_CALL_4(bpf_flow_dissector_load_bytes,
1751 const struct bpf_flow_dissector *, ctx, u32, offset,
1752 void *, to, u32, len)
1756 if (unlikely(offset > 0xffff))
1759 if (unlikely(!ctx->skb))
1762 ptr = skb_header_pointer(ctx->skb, offset, len, to);
1766 memcpy(to, ptr, len);
1774 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1775 .func = bpf_flow_dissector_load_bytes,
1777 .ret_type = RET_INTEGER,
1778 .arg1_type = ARG_PTR_TO_CTX,
1779 .arg2_type = ARG_ANYTHING,
1780 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1781 .arg4_type = ARG_CONST_SIZE,
1784 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1785 u32, offset, void *, to, u32, len, u32, start_header)
1787 u8 *end = skb_tail_pointer(skb);
1790 if (unlikely(offset > 0xffff))
1793 switch (start_header) {
1794 case BPF_HDR_START_MAC:
1795 if (unlikely(!skb_mac_header_was_set(skb)))
1797 start = skb_mac_header(skb);
1799 case BPF_HDR_START_NET:
1800 start = skb_network_header(skb);
1806 ptr = start + offset;
1808 if (likely(ptr + len <= end)) {
1809 memcpy(to, ptr, len);
1818 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1819 .func = bpf_skb_load_bytes_relative,
1821 .ret_type = RET_INTEGER,
1822 .arg1_type = ARG_PTR_TO_CTX,
1823 .arg2_type = ARG_ANYTHING,
1824 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1825 .arg4_type = ARG_CONST_SIZE,
1826 .arg5_type = ARG_ANYTHING,
1829 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1831 /* Idea is the following: should the needed direct read/write
1832 * test fail during runtime, we can pull in more data and redo
1833 * again, since implicitly, we invalidate previous checks here.
1835 * Or, since we know how much we need to make read/writeable,
1836 * this can be done once at the program beginning for direct
1837 * access case. By this we overcome limitations of only current
1838 * headroom being accessible.
1840 return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1843 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1844 .func = bpf_skb_pull_data,
1846 .ret_type = RET_INTEGER,
1847 .arg1_type = ARG_PTR_TO_CTX,
1848 .arg2_type = ARG_ANYTHING,
1851 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1853 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1856 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1857 .func = bpf_sk_fullsock,
1859 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
1860 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
1863 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1864 unsigned int write_len)
1866 return __bpf_try_make_writable(skb, write_len);
1869 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1871 /* Idea is the following: should the needed direct read/write
1872 * test fail during runtime, we can pull in more data and redo
1873 * again, since implicitly, we invalidate previous checks here.
1875 * Or, since we know how much we need to make read/writeable,
1876 * this can be done once at the program beginning for direct
1877 * access case. By this we overcome limitations of only current
1878 * headroom being accessible.
1880 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1883 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1884 .func = sk_skb_pull_data,
1886 .ret_type = RET_INTEGER,
1887 .arg1_type = ARG_PTR_TO_CTX,
1888 .arg2_type = ARG_ANYTHING,
1891 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1892 u64, from, u64, to, u64, flags)
1896 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1898 if (unlikely(offset > 0xffff || offset & 1))
1900 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1903 ptr = (__sum16 *)(skb->data + offset);
1904 switch (flags & BPF_F_HDR_FIELD_MASK) {
1906 if (unlikely(from != 0))
1909 csum_replace_by_diff(ptr, to);
1912 csum_replace2(ptr, from, to);
1915 csum_replace4(ptr, from, to);
1924 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1925 .func = bpf_l3_csum_replace,
1927 .ret_type = RET_INTEGER,
1928 .arg1_type = ARG_PTR_TO_CTX,
1929 .arg2_type = ARG_ANYTHING,
1930 .arg3_type = ARG_ANYTHING,
1931 .arg4_type = ARG_ANYTHING,
1932 .arg5_type = ARG_ANYTHING,
1935 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1936 u64, from, u64, to, u64, flags)
1938 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1939 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1940 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1943 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1944 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1946 if (unlikely(offset > 0xffff || offset & 1))
1948 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1951 ptr = (__sum16 *)(skb->data + offset);
1952 if (is_mmzero && !do_mforce && !*ptr)
1955 switch (flags & BPF_F_HDR_FIELD_MASK) {
1957 if (unlikely(from != 0))
1960 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1963 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1966 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1972 if (is_mmzero && !*ptr)
1973 *ptr = CSUM_MANGLED_0;
1977 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1978 .func = bpf_l4_csum_replace,
1980 .ret_type = RET_INTEGER,
1981 .arg1_type = ARG_PTR_TO_CTX,
1982 .arg2_type = ARG_ANYTHING,
1983 .arg3_type = ARG_ANYTHING,
1984 .arg4_type = ARG_ANYTHING,
1985 .arg5_type = ARG_ANYTHING,
1988 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1989 __be32 *, to, u32, to_size, __wsum, seed)
1991 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1992 u32 diff_size = from_size + to_size;
1995 /* This is quite flexible, some examples:
1997 * from_size == 0, to_size > 0, seed := csum --> pushing data
1998 * from_size > 0, to_size == 0, seed := csum --> pulling data
1999 * from_size > 0, to_size > 0, seed := 0 --> diffing data
2001 * Even for diffing, from_size and to_size don't need to be equal.
2003 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
2004 diff_size > sizeof(sp->diff)))
2007 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
2008 sp->diff[j] = ~from[i];
2009 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
2010 sp->diff[j] = to[i];
2012 return csum_partial(sp->diff, diff_size, seed);
2015 static const struct bpf_func_proto bpf_csum_diff_proto = {
2016 .func = bpf_csum_diff,
2019 .ret_type = RET_INTEGER,
2020 .arg1_type = ARG_PTR_TO_MEM_OR_NULL,
2021 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
2022 .arg3_type = ARG_PTR_TO_MEM_OR_NULL,
2023 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
2024 .arg5_type = ARG_ANYTHING,
2027 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2029 /* The interface is to be used in combination with bpf_csum_diff()
2030 * for direct packet writes. csum rotation for alignment as well
2031 * as emulating csum_sub() can be done from the eBPF program.
2033 if (skb->ip_summed == CHECKSUM_COMPLETE)
2034 return (skb->csum = csum_add(skb->csum, csum));
2039 static const struct bpf_func_proto bpf_csum_update_proto = {
2040 .func = bpf_csum_update,
2042 .ret_type = RET_INTEGER,
2043 .arg1_type = ARG_PTR_TO_CTX,
2044 .arg2_type = ARG_ANYTHING,
2047 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
2049 /* The interface is to be used in combination with bpf_skb_adjust_room()
2050 * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET
2051 * is passed as flags, for example.
2054 case BPF_CSUM_LEVEL_INC:
2055 __skb_incr_checksum_unnecessary(skb);
2057 case BPF_CSUM_LEVEL_DEC:
2058 __skb_decr_checksum_unnecessary(skb);
2060 case BPF_CSUM_LEVEL_RESET:
2061 __skb_reset_checksum_unnecessary(skb);
2063 case BPF_CSUM_LEVEL_QUERY:
2064 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2065 skb->csum_level : -EACCES;
2073 static const struct bpf_func_proto bpf_csum_level_proto = {
2074 .func = bpf_csum_level,
2076 .ret_type = RET_INTEGER,
2077 .arg1_type = ARG_PTR_TO_CTX,
2078 .arg2_type = ARG_ANYTHING,
2081 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2083 return dev_forward_skb_nomtu(dev, skb);
2086 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2087 struct sk_buff *skb)
2089 int ret = ____dev_forward_skb(dev, skb, false);
2093 ret = netif_rx(skb);
2099 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2103 if (dev_xmit_recursion()) {
2104 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2112 dev_xmit_recursion_inc();
2113 ret = dev_queue_xmit(skb);
2114 dev_xmit_recursion_dec();
2119 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2122 unsigned int mlen = skb_network_offset(skb);
2125 __skb_pull(skb, mlen);
2127 /* At ingress, the mac header has already been pulled once.
2128 * At egress, skb_pospull_rcsum has to be done in case that
2129 * the skb is originated from ingress (i.e. a forwarded skb)
2130 * to ensure that rcsum starts at net header.
2132 if (!skb_at_tc_ingress(skb))
2133 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2135 skb_pop_mac_header(skb);
2136 skb_reset_mac_len(skb);
2137 return flags & BPF_F_INGRESS ?
2138 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2141 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2144 /* Verify that a link layer header is carried */
2145 if (unlikely(skb->mac_header >= skb->network_header)) {
2150 bpf_push_mac_rcsum(skb);
2151 return flags & BPF_F_INGRESS ?
2152 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2155 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2158 if (dev_is_mac_header_xmit(dev))
2159 return __bpf_redirect_common(skb, dev, flags);
2161 return __bpf_redirect_no_mac(skb, dev, flags);
2164 #if IS_ENABLED(CONFIG_IPV6)
2165 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2166 struct net_device *dev, struct bpf_nh_params *nh)
2168 u32 hh_len = LL_RESERVED_SPACE(dev);
2169 const struct in6_addr *nexthop;
2170 struct dst_entry *dst = NULL;
2171 struct neighbour *neigh;
2173 if (dev_xmit_recursion()) {
2174 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2181 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2182 skb = skb_expand_head(skb, hh_len);
2190 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2191 &ipv6_hdr(skb)->daddr);
2193 nexthop = &nh->ipv6_nh;
2195 neigh = ip_neigh_gw6(dev, nexthop);
2196 if (likely(!IS_ERR(neigh))) {
2199 sock_confirm_neigh(skb, neigh);
2200 dev_xmit_recursion_inc();
2201 ret = neigh_output(neigh, skb, false);
2202 dev_xmit_recursion_dec();
2203 rcu_read_unlock_bh();
2206 rcu_read_unlock_bh();
2208 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2214 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2215 struct bpf_nh_params *nh)
2217 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2218 struct net *net = dev_net(dev);
2219 int err, ret = NET_XMIT_DROP;
2222 struct dst_entry *dst;
2223 struct flowi6 fl6 = {
2224 .flowi6_flags = FLOWI_FLAG_ANYSRC,
2225 .flowi6_mark = skb->mark,
2226 .flowlabel = ip6_flowinfo(ip6h),
2227 .flowi6_oif = dev->ifindex,
2228 .flowi6_proto = ip6h->nexthdr,
2229 .daddr = ip6h->daddr,
2230 .saddr = ip6h->saddr,
2233 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2237 skb_dst_set(skb, dst);
2238 } else if (nh->nh_family != AF_INET6) {
2242 err = bpf_out_neigh_v6(net, skb, dev, nh);
2243 if (unlikely(net_xmit_eval(err)))
2244 dev->stats.tx_errors++;
2246 ret = NET_XMIT_SUCCESS;
2249 dev->stats.tx_errors++;
2255 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2256 struct bpf_nh_params *nh)
2259 return NET_XMIT_DROP;
2261 #endif /* CONFIG_IPV6 */
2263 #if IS_ENABLED(CONFIG_INET)
2264 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2265 struct net_device *dev, struct bpf_nh_params *nh)
2267 u32 hh_len = LL_RESERVED_SPACE(dev);
2268 struct neighbour *neigh;
2269 bool is_v6gw = false;
2271 if (dev_xmit_recursion()) {
2272 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2279 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2280 skb = skb_expand_head(skb, hh_len);
2287 struct dst_entry *dst = skb_dst(skb);
2288 struct rtable *rt = container_of(dst, struct rtable, dst);
2290 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2291 } else if (nh->nh_family == AF_INET6) {
2292 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2294 } else if (nh->nh_family == AF_INET) {
2295 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2297 rcu_read_unlock_bh();
2301 if (likely(!IS_ERR(neigh))) {
2304 sock_confirm_neigh(skb, neigh);
2305 dev_xmit_recursion_inc();
2306 ret = neigh_output(neigh, skb, is_v6gw);
2307 dev_xmit_recursion_dec();
2308 rcu_read_unlock_bh();
2311 rcu_read_unlock_bh();
2317 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2318 struct bpf_nh_params *nh)
2320 const struct iphdr *ip4h = ip_hdr(skb);
2321 struct net *net = dev_net(dev);
2322 int err, ret = NET_XMIT_DROP;
2325 struct flowi4 fl4 = {
2326 .flowi4_flags = FLOWI_FLAG_ANYSRC,
2327 .flowi4_mark = skb->mark,
2328 .flowi4_tos = RT_TOS(ip4h->tos),
2329 .flowi4_oif = dev->ifindex,
2330 .flowi4_proto = ip4h->protocol,
2331 .daddr = ip4h->daddr,
2332 .saddr = ip4h->saddr,
2336 rt = ip_route_output_flow(net, &fl4, NULL);
2339 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2344 skb_dst_set(skb, &rt->dst);
2347 err = bpf_out_neigh_v4(net, skb, dev, nh);
2348 if (unlikely(net_xmit_eval(err)))
2349 dev->stats.tx_errors++;
2351 ret = NET_XMIT_SUCCESS;
2354 dev->stats.tx_errors++;
2360 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2361 struct bpf_nh_params *nh)
2364 return NET_XMIT_DROP;
2366 #endif /* CONFIG_INET */
2368 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2369 struct bpf_nh_params *nh)
2371 struct ethhdr *ethh = eth_hdr(skb);
2373 if (unlikely(skb->mac_header >= skb->network_header))
2375 bpf_push_mac_rcsum(skb);
2376 if (is_multicast_ether_addr(ethh->h_dest))
2379 skb_pull(skb, sizeof(*ethh));
2380 skb_unset_mac_header(skb);
2381 skb_reset_network_header(skb);
2383 if (skb->protocol == htons(ETH_P_IP))
2384 return __bpf_redirect_neigh_v4(skb, dev, nh);
2385 else if (skb->protocol == htons(ETH_P_IPV6))
2386 return __bpf_redirect_neigh_v6(skb, dev, nh);
2392 /* Internal, non-exposed redirect flags. */
2394 BPF_F_NEIGH = (1ULL << 1),
2395 BPF_F_PEER = (1ULL << 2),
2396 BPF_F_NEXTHOP = (1ULL << 3),
2397 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2400 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2402 struct net_device *dev;
2403 struct sk_buff *clone;
2406 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2409 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2413 clone = skb_clone(skb, GFP_ATOMIC);
2414 if (unlikely(!clone))
2417 /* For direct write, we need to keep the invariant that the skbs
2418 * we're dealing with need to be uncloned. Should uncloning fail
2419 * here, we need to free the just generated clone to unclone once
2422 ret = bpf_try_make_head_writable(skb);
2423 if (unlikely(ret)) {
2428 return __bpf_redirect(clone, dev, flags);
2431 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2432 .func = bpf_clone_redirect,
2434 .ret_type = RET_INTEGER,
2435 .arg1_type = ARG_PTR_TO_CTX,
2436 .arg2_type = ARG_ANYTHING,
2437 .arg3_type = ARG_ANYTHING,
2440 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2441 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2443 int skb_do_redirect(struct sk_buff *skb)
2445 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2446 struct net *net = dev_net(skb->dev);
2447 struct net_device *dev;
2448 u32 flags = ri->flags;
2450 dev = dev_get_by_index_rcu(net, ri->tgt_index);
2455 if (flags & BPF_F_PEER) {
2456 const struct net_device_ops *ops = dev->netdev_ops;
2458 if (unlikely(!ops->ndo_get_peer_dev ||
2459 !skb_at_tc_ingress(skb)))
2461 dev = ops->ndo_get_peer_dev(dev);
2462 if (unlikely(!dev ||
2463 !(dev->flags & IFF_UP) ||
2464 net_eq(net, dev_net(dev))))
2469 return flags & BPF_F_NEIGH ?
2470 __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2472 __bpf_redirect(skb, dev, flags);
2478 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2480 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2482 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2486 ri->tgt_index = ifindex;
2488 return TC_ACT_REDIRECT;
2491 static const struct bpf_func_proto bpf_redirect_proto = {
2492 .func = bpf_redirect,
2494 .ret_type = RET_INTEGER,
2495 .arg1_type = ARG_ANYTHING,
2496 .arg2_type = ARG_ANYTHING,
2499 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2501 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2503 if (unlikely(flags))
2506 ri->flags = BPF_F_PEER;
2507 ri->tgt_index = ifindex;
2509 return TC_ACT_REDIRECT;
2512 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2513 .func = bpf_redirect_peer,
2515 .ret_type = RET_INTEGER,
2516 .arg1_type = ARG_ANYTHING,
2517 .arg2_type = ARG_ANYTHING,
2520 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2521 int, plen, u64, flags)
2523 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2525 if (unlikely((plen && plen < sizeof(*params)) || flags))
2528 ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2529 ri->tgt_index = ifindex;
2531 BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2533 memcpy(&ri->nh, params, sizeof(ri->nh));
2535 return TC_ACT_REDIRECT;
2538 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2539 .func = bpf_redirect_neigh,
2541 .ret_type = RET_INTEGER,
2542 .arg1_type = ARG_ANYTHING,
2543 .arg2_type = ARG_PTR_TO_MEM_OR_NULL,
2544 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
2545 .arg4_type = ARG_ANYTHING,
2548 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2550 msg->apply_bytes = bytes;
2554 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2555 .func = bpf_msg_apply_bytes,
2557 .ret_type = RET_INTEGER,
2558 .arg1_type = ARG_PTR_TO_CTX,
2559 .arg2_type = ARG_ANYTHING,
2562 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2564 msg->cork_bytes = bytes;
2568 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2569 .func = bpf_msg_cork_bytes,
2571 .ret_type = RET_INTEGER,
2572 .arg1_type = ARG_PTR_TO_CTX,
2573 .arg2_type = ARG_ANYTHING,
2576 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2577 u32, end, u64, flags)
2579 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2580 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2581 struct scatterlist *sge;
2582 u8 *raw, *to, *from;
2585 if (unlikely(flags || end <= start))
2588 /* First find the starting scatterlist element */
2592 len = sk_msg_elem(msg, i)->length;
2593 if (start < offset + len)
2595 sk_msg_iter_var_next(i);
2596 } while (i != msg->sg.end);
2598 if (unlikely(start >= offset + len))
2602 /* The start may point into the sg element so we need to also
2603 * account for the headroom.
2605 bytes_sg_total = start - offset + bytes;
2606 if (!test_bit(i, &msg->sg.copy) && bytes_sg_total <= len)
2609 /* At this point we need to linearize multiple scatterlist
2610 * elements or a single shared page. Either way we need to
2611 * copy into a linear buffer exclusively owned by BPF. Then
2612 * place the buffer in the scatterlist and fixup the original
2613 * entries by removing the entries now in the linear buffer
2614 * and shifting the remaining entries. For now we do not try
2615 * to copy partial entries to avoid complexity of running out
2616 * of sg_entry slots. The downside is reading a single byte
2617 * will copy the entire sg entry.
2620 copy += sk_msg_elem(msg, i)->length;
2621 sk_msg_iter_var_next(i);
2622 if (bytes_sg_total <= copy)
2624 } while (i != msg->sg.end);
2627 if (unlikely(bytes_sg_total > copy))
2630 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2632 if (unlikely(!page))
2635 raw = page_address(page);
2638 sge = sk_msg_elem(msg, i);
2639 from = sg_virt(sge);
2643 memcpy(to, from, len);
2646 put_page(sg_page(sge));
2648 sk_msg_iter_var_next(i);
2649 } while (i != last_sge);
2651 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2653 /* To repair sg ring we need to shift entries. If we only
2654 * had a single entry though we can just replace it and
2655 * be done. Otherwise walk the ring and shift the entries.
2657 WARN_ON_ONCE(last_sge == first_sge);
2658 shift = last_sge > first_sge ?
2659 last_sge - first_sge - 1 :
2660 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2665 sk_msg_iter_var_next(i);
2669 if (i + shift >= NR_MSG_FRAG_IDS)
2670 move_from = i + shift - NR_MSG_FRAG_IDS;
2672 move_from = i + shift;
2673 if (move_from == msg->sg.end)
2676 msg->sg.data[i] = msg->sg.data[move_from];
2677 msg->sg.data[move_from].length = 0;
2678 msg->sg.data[move_from].page_link = 0;
2679 msg->sg.data[move_from].offset = 0;
2680 sk_msg_iter_var_next(i);
2683 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2684 msg->sg.end - shift + NR_MSG_FRAG_IDS :
2685 msg->sg.end - shift;
2687 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2688 msg->data_end = msg->data + bytes;
2692 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2693 .func = bpf_msg_pull_data,
2695 .ret_type = RET_INTEGER,
2696 .arg1_type = ARG_PTR_TO_CTX,
2697 .arg2_type = ARG_ANYTHING,
2698 .arg3_type = ARG_ANYTHING,
2699 .arg4_type = ARG_ANYTHING,
2702 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2703 u32, len, u64, flags)
2705 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2706 u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2707 u8 *raw, *to, *from;
2710 if (unlikely(flags))
2713 /* First find the starting scatterlist element */
2717 l = sk_msg_elem(msg, i)->length;
2719 if (start < offset + l)
2721 sk_msg_iter_var_next(i);
2722 } while (i != msg->sg.end);
2724 if (start >= offset + l)
2727 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2729 /* If no space available will fallback to copy, we need at
2730 * least one scatterlist elem available to push data into
2731 * when start aligns to the beginning of an element or two
2732 * when it falls inside an element. We handle the start equals
2733 * offset case because its the common case for inserting a
2736 if (!space || (space == 1 && start != offset))
2737 copy = msg->sg.data[i].length;
2739 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2740 get_order(copy + len));
2741 if (unlikely(!page))
2747 raw = page_address(page);
2749 psge = sk_msg_elem(msg, i);
2750 front = start - offset;
2751 back = psge->length - front;
2752 from = sg_virt(psge);
2755 memcpy(raw, from, front);
2759 to = raw + front + len;
2761 memcpy(to, from, back);
2764 put_page(sg_page(psge));
2765 } else if (start - offset) {
2766 psge = sk_msg_elem(msg, i);
2767 rsge = sk_msg_elem_cpy(msg, i);
2769 psge->length = start - offset;
2770 rsge.length -= psge->length;
2771 rsge.offset += start;
2773 sk_msg_iter_var_next(i);
2774 sg_unmark_end(psge);
2775 sg_unmark_end(&rsge);
2776 sk_msg_iter_next(msg, end);
2779 /* Slot(s) to place newly allocated data */
2782 /* Shift one or two slots as needed */
2784 sge = sk_msg_elem_cpy(msg, i);
2786 sk_msg_iter_var_next(i);
2787 sg_unmark_end(&sge);
2788 sk_msg_iter_next(msg, end);
2790 nsge = sk_msg_elem_cpy(msg, i);
2792 sk_msg_iter_var_next(i);
2793 nnsge = sk_msg_elem_cpy(msg, i);
2796 while (i != msg->sg.end) {
2797 msg->sg.data[i] = sge;
2799 sk_msg_iter_var_next(i);
2802 nnsge = sk_msg_elem_cpy(msg, i);
2804 nsge = sk_msg_elem_cpy(msg, i);
2809 /* Place newly allocated data buffer */
2810 sk_mem_charge(msg->sk, len);
2811 msg->sg.size += len;
2812 __clear_bit(new, &msg->sg.copy);
2813 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2815 get_page(sg_page(&rsge));
2816 sk_msg_iter_var_next(new);
2817 msg->sg.data[new] = rsge;
2820 sk_msg_compute_data_pointers(msg);
2824 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2825 .func = bpf_msg_push_data,
2827 .ret_type = RET_INTEGER,
2828 .arg1_type = ARG_PTR_TO_CTX,
2829 .arg2_type = ARG_ANYTHING,
2830 .arg3_type = ARG_ANYTHING,
2831 .arg4_type = ARG_ANYTHING,
2834 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2840 sk_msg_iter_var_next(i);
2841 msg->sg.data[prev] = msg->sg.data[i];
2842 } while (i != msg->sg.end);
2844 sk_msg_iter_prev(msg, end);
2847 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2849 struct scatterlist tmp, sge;
2851 sk_msg_iter_next(msg, end);
2852 sge = sk_msg_elem_cpy(msg, i);
2853 sk_msg_iter_var_next(i);
2854 tmp = sk_msg_elem_cpy(msg, i);
2856 while (i != msg->sg.end) {
2857 msg->sg.data[i] = sge;
2858 sk_msg_iter_var_next(i);
2860 tmp = sk_msg_elem_cpy(msg, i);
2864 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2865 u32, len, u64, flags)
2867 u32 i = 0, l = 0, space, offset = 0;
2868 u64 last = start + len;
2871 if (unlikely(flags))
2874 /* First find the starting scatterlist element */
2878 l = sk_msg_elem(msg, i)->length;
2880 if (start < offset + l)
2882 sk_msg_iter_var_next(i);
2883 } while (i != msg->sg.end);
2885 /* Bounds checks: start and pop must be inside message */
2886 if (start >= offset + l || last >= msg->sg.size)
2889 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2892 /* --------------| offset
2893 * -| start |-------- len -------|
2895 * |----- a ----|-------- pop -------|----- b ----|
2896 * |______________________________________________| length
2899 * a: region at front of scatter element to save
2900 * b: region at back of scatter element to save when length > A + pop
2901 * pop: region to pop from element, same as input 'pop' here will be
2902 * decremented below per iteration.
2904 * Two top-level cases to handle when start != offset, first B is non
2905 * zero and second B is zero corresponding to when a pop includes more
2908 * Then if B is non-zero AND there is no space allocate space and
2909 * compact A, B regions into page. If there is space shift ring to
2910 * the rigth free'ing the next element in ring to place B, leaving
2911 * A untouched except to reduce length.
2913 if (start != offset) {
2914 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2916 int b = sge->length - pop - a;
2918 sk_msg_iter_var_next(i);
2920 if (pop < sge->length - a) {
2923 sk_msg_shift_right(msg, i);
2924 nsge = sk_msg_elem(msg, i);
2925 get_page(sg_page(sge));
2928 b, sge->offset + pop + a);
2930 struct page *page, *orig;
2933 page = alloc_pages(__GFP_NOWARN |
2934 __GFP_COMP | GFP_ATOMIC,
2936 if (unlikely(!page))
2940 orig = sg_page(sge);
2941 from = sg_virt(sge);
2942 to = page_address(page);
2943 memcpy(to, from, a);
2944 memcpy(to + a, from + a + pop, b);
2945 sg_set_page(sge, page, a + b, 0);
2949 } else if (pop >= sge->length - a) {
2950 pop -= (sge->length - a);
2955 /* From above the current layout _must_ be as follows,
2960 * |---- pop ---|---------------- b ------------|
2961 * |____________________________________________| length
2963 * Offset and start of the current msg elem are equal because in the
2964 * previous case we handled offset != start and either consumed the
2965 * entire element and advanced to the next element OR pop == 0.
2967 * Two cases to handle here are first pop is less than the length
2968 * leaving some remainder b above. Simply adjust the element's layout
2969 * in this case. Or pop >= length of the element so that b = 0. In this
2970 * case advance to next element decrementing pop.
2973 struct scatterlist *sge = sk_msg_elem(msg, i);
2975 if (pop < sge->length) {
2981 sk_msg_shift_left(msg, i);
2983 sk_msg_iter_var_next(i);
2986 sk_mem_uncharge(msg->sk, len - pop);
2987 msg->sg.size -= (len - pop);
2988 sk_msg_compute_data_pointers(msg);
2992 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
2993 .func = bpf_msg_pop_data,
2995 .ret_type = RET_INTEGER,
2996 .arg1_type = ARG_PTR_TO_CTX,
2997 .arg2_type = ARG_ANYTHING,
2998 .arg3_type = ARG_ANYTHING,
2999 .arg4_type = ARG_ANYTHING,
3002 #ifdef CONFIG_CGROUP_NET_CLASSID
3003 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3005 return __task_get_classid(current);
3008 static const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3009 .func = bpf_get_cgroup_classid_curr,
3011 .ret_type = RET_INTEGER,
3014 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3016 struct sock *sk = skb_to_full_sk(skb);
3018 if (!sk || !sk_fullsock(sk))
3021 return sock_cgroup_classid(&sk->sk_cgrp_data);
3024 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3025 .func = bpf_skb_cgroup_classid,
3027 .ret_type = RET_INTEGER,
3028 .arg1_type = ARG_PTR_TO_CTX,
3032 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3034 return task_get_classid(skb);
3037 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3038 .func = bpf_get_cgroup_classid,
3040 .ret_type = RET_INTEGER,
3041 .arg1_type = ARG_PTR_TO_CTX,
3044 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3046 return dst_tclassid(skb);
3049 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3050 .func = bpf_get_route_realm,
3052 .ret_type = RET_INTEGER,
3053 .arg1_type = ARG_PTR_TO_CTX,
3056 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3058 /* If skb_clear_hash() was called due to mangling, we can
3059 * trigger SW recalculation here. Later access to hash
3060 * can then use the inline skb->hash via context directly
3061 * instead of calling this helper again.
3063 return skb_get_hash(skb);
3066 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3067 .func = bpf_get_hash_recalc,
3069 .ret_type = RET_INTEGER,
3070 .arg1_type = ARG_PTR_TO_CTX,
3073 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3075 /* After all direct packet write, this can be used once for
3076 * triggering a lazy recalc on next skb_get_hash() invocation.
3078 skb_clear_hash(skb);
3082 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3083 .func = bpf_set_hash_invalid,
3085 .ret_type = RET_INTEGER,
3086 .arg1_type = ARG_PTR_TO_CTX,
3089 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3091 /* Set user specified hash as L4(+), so that it gets returned
3092 * on skb_get_hash() call unless BPF prog later on triggers a
3095 __skb_set_sw_hash(skb, hash, true);
3099 static const struct bpf_func_proto bpf_set_hash_proto = {
3100 .func = bpf_set_hash,
3102 .ret_type = RET_INTEGER,
3103 .arg1_type = ARG_PTR_TO_CTX,
3104 .arg2_type = ARG_ANYTHING,
3107 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3112 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3113 vlan_proto != htons(ETH_P_8021AD)))
3114 vlan_proto = htons(ETH_P_8021Q);
3116 bpf_push_mac_rcsum(skb);
3117 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3118 bpf_pull_mac_rcsum(skb);
3120 bpf_compute_data_pointers(skb);
3124 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3125 .func = bpf_skb_vlan_push,
3127 .ret_type = RET_INTEGER,
3128 .arg1_type = ARG_PTR_TO_CTX,
3129 .arg2_type = ARG_ANYTHING,
3130 .arg3_type = ARG_ANYTHING,
3133 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3137 bpf_push_mac_rcsum(skb);
3138 ret = skb_vlan_pop(skb);
3139 bpf_pull_mac_rcsum(skb);
3141 bpf_compute_data_pointers(skb);
3145 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3146 .func = bpf_skb_vlan_pop,
3148 .ret_type = RET_INTEGER,
3149 .arg1_type = ARG_PTR_TO_CTX,
3152 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3154 /* Caller already did skb_cow() with len as headroom,
3155 * so no need to do it here.
3158 memmove(skb->data, skb->data + len, off);
3159 memset(skb->data + off, 0, len);
3161 /* No skb_postpush_rcsum(skb, skb->data + off, len)
3162 * needed here as it does not change the skb->csum
3163 * result for checksum complete when summing over
3169 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3171 /* skb_ensure_writable() is not needed here, as we're
3172 * already working on an uncloned skb.
3174 if (unlikely(!pskb_may_pull(skb, off + len)))
3177 skb_postpull_rcsum(skb, skb->data + off, len);
3178 memmove(skb->data + len, skb->data, off);
3179 __skb_pull(skb, len);
3184 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3186 bool trans_same = skb->transport_header == skb->network_header;
3189 /* There's no need for __skb_push()/__skb_pull() pair to
3190 * get to the start of the mac header as we're guaranteed
3191 * to always start from here under eBPF.
3193 ret = bpf_skb_generic_push(skb, off, len);
3195 skb->mac_header -= len;
3196 skb->network_header -= len;
3198 skb->transport_header = skb->network_header;
3204 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3206 bool trans_same = skb->transport_header == skb->network_header;
3209 /* Same here, __skb_push()/__skb_pull() pair not needed. */
3210 ret = bpf_skb_generic_pop(skb, off, len);
3212 skb->mac_header += len;
3213 skb->network_header += len;
3215 skb->transport_header = skb->network_header;
3221 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3223 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3224 u32 off = skb_mac_header_len(skb);
3227 ret = skb_cow(skb, len_diff);
3228 if (unlikely(ret < 0))
3231 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3232 if (unlikely(ret < 0))
3235 if (skb_is_gso(skb)) {
3236 struct skb_shared_info *shinfo = skb_shinfo(skb);
3238 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */
3239 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3240 shinfo->gso_type &= ~SKB_GSO_TCPV4;
3241 shinfo->gso_type |= SKB_GSO_TCPV6;
3245 skb->protocol = htons(ETH_P_IPV6);
3246 skb_clear_hash(skb);
3251 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3253 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3254 u32 off = skb_mac_header_len(skb);
3257 ret = skb_unclone(skb, GFP_ATOMIC);
3258 if (unlikely(ret < 0))
3261 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3262 if (unlikely(ret < 0))
3265 if (skb_is_gso(skb)) {
3266 struct skb_shared_info *shinfo = skb_shinfo(skb);
3268 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */
3269 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3270 shinfo->gso_type &= ~SKB_GSO_TCPV6;
3271 shinfo->gso_type |= SKB_GSO_TCPV4;
3275 skb->protocol = htons(ETH_P_IP);
3276 skb_clear_hash(skb);
3281 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3283 __be16 from_proto = skb->protocol;
3285 if (from_proto == htons(ETH_P_IP) &&
3286 to_proto == htons(ETH_P_IPV6))
3287 return bpf_skb_proto_4_to_6(skb);
3289 if (from_proto == htons(ETH_P_IPV6) &&
3290 to_proto == htons(ETH_P_IP))
3291 return bpf_skb_proto_6_to_4(skb);
3296 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3301 if (unlikely(flags))
3304 /* General idea is that this helper does the basic groundwork
3305 * needed for changing the protocol, and eBPF program fills the
3306 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3307 * and other helpers, rather than passing a raw buffer here.
3309 * The rationale is to keep this minimal and without a need to
3310 * deal with raw packet data. F.e. even if we would pass buffers
3311 * here, the program still needs to call the bpf_lX_csum_replace()
3312 * helpers anyway. Plus, this way we keep also separation of
3313 * concerns, since f.e. bpf_skb_store_bytes() should only take
3316 * Currently, additional options and extension header space are
3317 * not supported, but flags register is reserved so we can adapt
3318 * that. For offloads, we mark packet as dodgy, so that headers
3319 * need to be verified first.
3321 ret = bpf_skb_proto_xlat(skb, proto);
3322 bpf_compute_data_pointers(skb);
3326 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3327 .func = bpf_skb_change_proto,
3329 .ret_type = RET_INTEGER,
3330 .arg1_type = ARG_PTR_TO_CTX,
3331 .arg2_type = ARG_ANYTHING,
3332 .arg3_type = ARG_ANYTHING,
3335 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3337 /* We only allow a restricted subset to be changed for now. */
3338 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3339 !skb_pkt_type_ok(pkt_type)))
3342 skb->pkt_type = pkt_type;
3346 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3347 .func = bpf_skb_change_type,
3349 .ret_type = RET_INTEGER,
3350 .arg1_type = ARG_PTR_TO_CTX,
3351 .arg2_type = ARG_ANYTHING,
3354 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3356 switch (skb->protocol) {
3357 case htons(ETH_P_IP):
3358 return sizeof(struct iphdr);
3359 case htons(ETH_P_IPV6):
3360 return sizeof(struct ipv6hdr);
3366 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3367 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3369 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \
3370 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3371 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3372 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3373 BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \
3374 BPF_F_ADJ_ROOM_ENCAP_L2( \
3375 BPF_ADJ_ROOM_ENCAP_L2_MASK))
3377 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3380 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3381 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3382 u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3383 unsigned int gso_type = SKB_GSO_DODGY;
3386 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3387 /* udp gso_size delineates datagrams, only allow if fixed */
3388 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3389 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3393 ret = skb_cow_head(skb, len_diff);
3394 if (unlikely(ret < 0))
3398 if (skb->protocol != htons(ETH_P_IP) &&
3399 skb->protocol != htons(ETH_P_IPV6))
3402 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3403 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3406 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3407 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3410 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH &&
3411 inner_mac_len < ETH_HLEN)
3414 if (skb->encapsulation)
3417 mac_len = skb->network_header - skb->mac_header;
3418 inner_net = skb->network_header;
3419 if (inner_mac_len > len_diff)
3421 inner_trans = skb->transport_header;
3424 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3425 if (unlikely(ret < 0))
3429 skb->inner_mac_header = inner_net - inner_mac_len;
3430 skb->inner_network_header = inner_net;
3431 skb->inner_transport_header = inner_trans;
3433 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH)
3434 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
3436 skb_set_inner_protocol(skb, skb->protocol);
3438 skb->encapsulation = 1;
3439 skb_set_network_header(skb, mac_len);
3441 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3442 gso_type |= SKB_GSO_UDP_TUNNEL;
3443 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3444 gso_type |= SKB_GSO_GRE;
3445 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3446 gso_type |= SKB_GSO_IPXIP6;
3447 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3448 gso_type |= SKB_GSO_IPXIP4;
3450 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3451 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3452 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3453 sizeof(struct ipv6hdr) :
3454 sizeof(struct iphdr);
3456 skb_set_transport_header(skb, mac_len + nh_len);
3459 /* Match skb->protocol to new outer l3 protocol */
3460 if (skb->protocol == htons(ETH_P_IP) &&
3461 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3462 skb->protocol = htons(ETH_P_IPV6);
3463 else if (skb->protocol == htons(ETH_P_IPV6) &&
3464 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3465 skb->protocol = htons(ETH_P_IP);
3468 if (skb_is_gso(skb)) {
3469 struct skb_shared_info *shinfo = skb_shinfo(skb);
3471 /* Due to header grow, MSS needs to be downgraded. */
3472 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3473 skb_decrease_gso_size(shinfo, len_diff);
3475 /* Header must be checked, and gso_segs recomputed. */
3476 shinfo->gso_type |= gso_type;
3477 shinfo->gso_segs = 0;
3483 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3488 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3489 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3492 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3493 /* udp gso_size delineates datagrams, only allow if fixed */
3494 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3495 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3499 ret = skb_unclone(skb, GFP_ATOMIC);
3500 if (unlikely(ret < 0))
3503 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3504 if (unlikely(ret < 0))
3507 if (skb_is_gso(skb)) {
3508 struct skb_shared_info *shinfo = skb_shinfo(skb);
3510 /* Due to header shrink, MSS can be upgraded. */
3511 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3512 skb_increase_gso_size(shinfo, len_diff);
3514 /* Header must be checked, and gso_segs recomputed. */
3515 shinfo->gso_type |= SKB_GSO_DODGY;
3516 shinfo->gso_segs = 0;
3522 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3524 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3525 u32, mode, u64, flags)
3527 u32 len_diff_abs = abs(len_diff);
3528 bool shrink = len_diff < 0;
3531 if (unlikely(flags || mode))
3533 if (unlikely(len_diff_abs > 0xfffU))
3537 ret = skb_cow(skb, len_diff);
3538 if (unlikely(ret < 0))
3540 __skb_push(skb, len_diff_abs);
3541 memset(skb->data, 0, len_diff_abs);
3543 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3545 __skb_pull(skb, len_diff_abs);
3547 if (tls_sw_has_ctx_rx(skb->sk)) {
3548 struct strp_msg *rxm = strp_msg(skb);
3550 rxm->full_len += len_diff;
3555 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3556 .func = sk_skb_adjust_room,
3558 .ret_type = RET_INTEGER,
3559 .arg1_type = ARG_PTR_TO_CTX,
3560 .arg2_type = ARG_ANYTHING,
3561 .arg3_type = ARG_ANYTHING,
3562 .arg4_type = ARG_ANYTHING,
3565 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3566 u32, mode, u64, flags)
3568 u32 len_cur, len_diff_abs = abs(len_diff);
3569 u32 len_min = bpf_skb_net_base_len(skb);
3570 u32 len_max = BPF_SKB_MAX_LEN;
3571 __be16 proto = skb->protocol;
3572 bool shrink = len_diff < 0;
3576 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3577 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3579 if (unlikely(len_diff_abs > 0xfffU))
3581 if (unlikely(proto != htons(ETH_P_IP) &&
3582 proto != htons(ETH_P_IPV6)))
3585 off = skb_mac_header_len(skb);
3587 case BPF_ADJ_ROOM_NET:
3588 off += bpf_skb_net_base_len(skb);
3590 case BPF_ADJ_ROOM_MAC:
3596 len_cur = skb->len - skb_network_offset(skb);
3597 if ((shrink && (len_diff_abs >= len_cur ||
3598 len_cur - len_diff_abs < len_min)) ||
3599 (!shrink && (skb->len + len_diff_abs > len_max &&
3603 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3604 bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3605 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3606 __skb_reset_checksum_unnecessary(skb);
3608 bpf_compute_data_pointers(skb);
3612 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3613 .func = bpf_skb_adjust_room,
3615 .ret_type = RET_INTEGER,
3616 .arg1_type = ARG_PTR_TO_CTX,
3617 .arg2_type = ARG_ANYTHING,
3618 .arg3_type = ARG_ANYTHING,
3619 .arg4_type = ARG_ANYTHING,
3622 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3624 u32 min_len = skb_network_offset(skb);
3626 if (skb_transport_header_was_set(skb))
3627 min_len = skb_transport_offset(skb);
3628 if (skb->ip_summed == CHECKSUM_PARTIAL)
3629 min_len = skb_checksum_start_offset(skb) +
3630 skb->csum_offset + sizeof(__sum16);
3634 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3636 unsigned int old_len = skb->len;
3639 ret = __skb_grow_rcsum(skb, new_len);
3641 memset(skb->data + old_len, 0, new_len - old_len);
3645 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3647 return __skb_trim_rcsum(skb, new_len);
3650 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3653 u32 max_len = BPF_SKB_MAX_LEN;
3654 u32 min_len = __bpf_skb_min_len(skb);
3657 if (unlikely(flags || new_len > max_len || new_len < min_len))
3659 if (skb->encapsulation)
3662 /* The basic idea of this helper is that it's performing the
3663 * needed work to either grow or trim an skb, and eBPF program
3664 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3665 * bpf_lX_csum_replace() and others rather than passing a raw
3666 * buffer here. This one is a slow path helper and intended
3667 * for replies with control messages.
3669 * Like in bpf_skb_change_proto(), we want to keep this rather
3670 * minimal and without protocol specifics so that we are able
3671 * to separate concerns as in bpf_skb_store_bytes() should only
3672 * be the one responsible for writing buffers.
3674 * It's really expected to be a slow path operation here for
3675 * control message replies, so we're implicitly linearizing,
3676 * uncloning and drop offloads from the skb by this.
3678 ret = __bpf_try_make_writable(skb, skb->len);
3680 if (new_len > skb->len)
3681 ret = bpf_skb_grow_rcsum(skb, new_len);
3682 else if (new_len < skb->len)
3683 ret = bpf_skb_trim_rcsum(skb, new_len);
3684 if (!ret && skb_is_gso(skb))
3690 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3693 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3695 bpf_compute_data_pointers(skb);
3699 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3700 .func = bpf_skb_change_tail,
3702 .ret_type = RET_INTEGER,
3703 .arg1_type = ARG_PTR_TO_CTX,
3704 .arg2_type = ARG_ANYTHING,
3705 .arg3_type = ARG_ANYTHING,
3708 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3711 return __bpf_skb_change_tail(skb, new_len, flags);
3714 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3715 .func = sk_skb_change_tail,
3717 .ret_type = RET_INTEGER,
3718 .arg1_type = ARG_PTR_TO_CTX,
3719 .arg2_type = ARG_ANYTHING,
3720 .arg3_type = ARG_ANYTHING,
3723 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3726 u32 max_len = BPF_SKB_MAX_LEN;
3727 u32 new_len = skb->len + head_room;
3730 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3731 new_len < skb->len))
3734 ret = skb_cow(skb, head_room);
3736 /* Idea for this helper is that we currently only
3737 * allow to expand on mac header. This means that
3738 * skb->protocol network header, etc, stay as is.
3739 * Compared to bpf_skb_change_tail(), we're more
3740 * flexible due to not needing to linearize or
3741 * reset GSO. Intention for this helper is to be
3742 * used by an L3 skb that needs to push mac header
3743 * for redirection into L2 device.
3745 __skb_push(skb, head_room);
3746 memset(skb->data, 0, head_room);
3747 skb_reset_mac_header(skb);
3748 skb_reset_mac_len(skb);
3754 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3757 int ret = __bpf_skb_change_head(skb, head_room, flags);
3759 bpf_compute_data_pointers(skb);
3763 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3764 .func = bpf_skb_change_head,
3766 .ret_type = RET_INTEGER,
3767 .arg1_type = ARG_PTR_TO_CTX,
3768 .arg2_type = ARG_ANYTHING,
3769 .arg3_type = ARG_ANYTHING,
3772 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3775 return __bpf_skb_change_head(skb, head_room, flags);
3778 static const struct bpf_func_proto sk_skb_change_head_proto = {
3779 .func = sk_skb_change_head,
3781 .ret_type = RET_INTEGER,
3782 .arg1_type = ARG_PTR_TO_CTX,
3783 .arg2_type = ARG_ANYTHING,
3784 .arg3_type = ARG_ANYTHING,
3786 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3788 return xdp_data_meta_unsupported(xdp) ? 0 :
3789 xdp->data - xdp->data_meta;
3792 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3794 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3795 unsigned long metalen = xdp_get_metalen(xdp);
3796 void *data_start = xdp_frame_end + metalen;
3797 void *data = xdp->data + offset;
3799 if (unlikely(data < data_start ||
3800 data > xdp->data_end - ETH_HLEN))
3804 memmove(xdp->data_meta + offset,
3805 xdp->data_meta, metalen);
3806 xdp->data_meta += offset;
3812 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3813 .func = bpf_xdp_adjust_head,
3815 .ret_type = RET_INTEGER,
3816 .arg1_type = ARG_PTR_TO_CTX,
3817 .arg2_type = ARG_ANYTHING,
3820 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3822 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
3823 void *data_end = xdp->data_end + offset;
3825 /* Notice that xdp_data_hard_end have reserved some tailroom */
3826 if (unlikely(data_end > data_hard_end))
3829 /* ALL drivers MUST init xdp->frame_sz, chicken check below */
3830 if (unlikely(xdp->frame_sz > PAGE_SIZE)) {
3831 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz);
3835 if (unlikely(data_end < xdp->data + ETH_HLEN))
3838 /* Clear memory area on grow, can contain uninit kernel memory */
3840 memset(xdp->data_end, 0, offset);
3842 xdp->data_end = data_end;
3847 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3848 .func = bpf_xdp_adjust_tail,
3850 .ret_type = RET_INTEGER,
3851 .arg1_type = ARG_PTR_TO_CTX,
3852 .arg2_type = ARG_ANYTHING,
3855 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3857 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3858 void *meta = xdp->data_meta + offset;
3859 unsigned long metalen = xdp->data - meta;
3861 if (xdp_data_meta_unsupported(xdp))
3863 if (unlikely(meta < xdp_frame_end ||
3866 if (unlikely(xdp_metalen_invalid(metalen)))
3869 xdp->data_meta = meta;
3874 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3875 .func = bpf_xdp_adjust_meta,
3877 .ret_type = RET_INTEGER,
3878 .arg1_type = ARG_PTR_TO_CTX,
3879 .arg2_type = ARG_ANYTHING,
3882 /* XDP_REDIRECT works by a three-step process, implemented in the functions
3885 * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target
3886 * of the redirect and store it (along with some other metadata) in a per-CPU
3887 * struct bpf_redirect_info.
3889 * 2. When the program returns the XDP_REDIRECT return code, the driver will
3890 * call xdp_do_redirect() which will use the information in struct
3891 * bpf_redirect_info to actually enqueue the frame into a map type-specific
3892 * bulk queue structure.
3894 * 3. Before exiting its NAPI poll loop, the driver will call xdp_do_flush(),
3895 * which will flush all the different bulk queues, thus completing the
3898 * Pointers to the map entries will be kept around for this whole sequence of
3899 * steps, protected by RCU. However, there is no top-level rcu_read_lock() in
3900 * the core code; instead, the RCU protection relies on everything happening
3901 * inside a single NAPI poll sequence, which means it's between a pair of calls
3902 * to local_bh_disable()/local_bh_enable().
3904 * The map entries are marked as __rcu and the map code makes sure to
3905 * dereference those pointers with rcu_dereference_check() in a way that works
3906 * for both sections that to hold an rcu_read_lock() and sections that are
3907 * called from NAPI without a separate rcu_read_lock(). The code below does not
3908 * use RCU annotations, but relies on those in the map code.
3910 void xdp_do_flush(void)
3916 EXPORT_SYMBOL_GPL(xdp_do_flush);
3918 void bpf_clear_redirect_map(struct bpf_map *map)
3920 struct bpf_redirect_info *ri;
3923 for_each_possible_cpu(cpu) {
3924 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3925 /* Avoid polluting remote cacheline due to writes if
3926 * not needed. Once we pass this test, we need the
3927 * cmpxchg() to make sure it hasn't been changed in
3928 * the meantime by remote CPU.
3930 if (unlikely(READ_ONCE(ri->map) == map))
3931 cmpxchg(&ri->map, map, NULL);
3935 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
3936 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key);
3938 u32 xdp_master_redirect(struct xdp_buff *xdp)
3940 struct net_device *master, *slave;
3941 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3943 master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev);
3944 slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp);
3945 if (slave && slave != xdp->rxq->dev) {
3946 /* The target device is different from the receiving device, so
3947 * redirect it to the new device.
3948 * Using XDP_REDIRECT gets the correct behaviour from XDP enabled
3949 * drivers to unmap the packet from their rx ring.
3951 ri->tgt_index = slave->ifindex;
3952 ri->map_id = INT_MAX;
3953 ri->map_type = BPF_MAP_TYPE_UNSPEC;
3954 return XDP_REDIRECT;
3958 EXPORT_SYMBOL_GPL(xdp_master_redirect);
3960 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3961 struct bpf_prog *xdp_prog)
3963 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3964 enum bpf_map_type map_type = ri->map_type;
3965 void *fwd = ri->tgt_value;
3966 u32 map_id = ri->map_id;
3967 struct bpf_map *map;
3970 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
3971 ri->map_type = BPF_MAP_TYPE_UNSPEC;
3974 case BPF_MAP_TYPE_DEVMAP:
3976 case BPF_MAP_TYPE_DEVMAP_HASH:
3977 map = READ_ONCE(ri->map);
3978 if (unlikely(map)) {
3979 WRITE_ONCE(ri->map, NULL);
3980 err = dev_map_enqueue_multi(xdp, dev, map,
3981 ri->flags & BPF_F_EXCLUDE_INGRESS);
3983 err = dev_map_enqueue(fwd, xdp, dev);
3986 case BPF_MAP_TYPE_CPUMAP:
3987 err = cpu_map_enqueue(fwd, xdp, dev);
3989 case BPF_MAP_TYPE_XSKMAP:
3990 err = __xsk_map_redirect(fwd, xdp);
3992 case BPF_MAP_TYPE_UNSPEC:
3993 if (map_id == INT_MAX) {
3994 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
3995 if (unlikely(!fwd)) {
3999 err = dev_xdp_enqueue(fwd, xdp, dev);
4010 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4013 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4016 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4018 static int xdp_do_generic_redirect_map(struct net_device *dev,
4019 struct sk_buff *skb,
4020 struct xdp_buff *xdp,
4021 struct bpf_prog *xdp_prog,
4023 enum bpf_map_type map_type, u32 map_id)
4025 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4026 struct bpf_map *map;
4030 case BPF_MAP_TYPE_DEVMAP:
4032 case BPF_MAP_TYPE_DEVMAP_HASH:
4033 map = READ_ONCE(ri->map);
4034 if (unlikely(map)) {
4035 WRITE_ONCE(ri->map, NULL);
4036 err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4037 ri->flags & BPF_F_EXCLUDE_INGRESS);
4039 err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4044 case BPF_MAP_TYPE_XSKMAP:
4045 err = xsk_generic_rcv(fwd, xdp);
4050 case BPF_MAP_TYPE_CPUMAP:
4051 err = cpu_map_generic_redirect(fwd, skb);
4060 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4063 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4067 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4068 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4070 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4071 enum bpf_map_type map_type = ri->map_type;
4072 void *fwd = ri->tgt_value;
4073 u32 map_id = ri->map_id;
4076 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4077 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4079 if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4080 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4081 if (unlikely(!fwd)) {
4086 err = xdp_ok_fwd_dev(fwd, skb->len);
4091 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4092 generic_xdp_tx(skb, xdp_prog);
4096 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id);
4098 _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4102 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4104 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4106 if (unlikely(flags))
4109 /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4110 * by map_idr) is used for ifindex based XDP redirect.
4112 ri->tgt_index = ifindex;
4113 ri->map_id = INT_MAX;
4114 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4116 return XDP_REDIRECT;
4119 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4120 .func = bpf_xdp_redirect,
4122 .ret_type = RET_INTEGER,
4123 .arg1_type = ARG_ANYTHING,
4124 .arg2_type = ARG_ANYTHING,
4127 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
4130 return map->ops->map_redirect(map, ifindex, flags);
4133 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4134 .func = bpf_xdp_redirect_map,
4136 .ret_type = RET_INTEGER,
4137 .arg1_type = ARG_CONST_MAP_PTR,
4138 .arg2_type = ARG_ANYTHING,
4139 .arg3_type = ARG_ANYTHING,
4142 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4143 unsigned long off, unsigned long len)
4145 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4149 if (ptr != dst_buff)
4150 memcpy(dst_buff, ptr, len);
4155 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4156 u64, flags, void *, meta, u64, meta_size)
4158 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4160 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4162 if (unlikely(!skb || skb_size > skb->len))
4165 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4169 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4170 .func = bpf_skb_event_output,
4172 .ret_type = RET_INTEGER,
4173 .arg1_type = ARG_PTR_TO_CTX,
4174 .arg2_type = ARG_CONST_MAP_PTR,
4175 .arg3_type = ARG_ANYTHING,
4176 .arg4_type = ARG_PTR_TO_MEM,
4177 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4180 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4182 const struct bpf_func_proto bpf_skb_output_proto = {
4183 .func = bpf_skb_event_output,
4185 .ret_type = RET_INTEGER,
4186 .arg1_type = ARG_PTR_TO_BTF_ID,
4187 .arg1_btf_id = &bpf_skb_output_btf_ids[0],
4188 .arg2_type = ARG_CONST_MAP_PTR,
4189 .arg3_type = ARG_ANYTHING,
4190 .arg4_type = ARG_PTR_TO_MEM,
4191 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4194 static unsigned short bpf_tunnel_key_af(u64 flags)
4196 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4199 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4200 u32, size, u64, flags)
4202 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4203 u8 compat[sizeof(struct bpf_tunnel_key)];
4207 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
4211 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4215 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4218 case offsetof(struct bpf_tunnel_key, tunnel_label):
4219 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4221 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4222 /* Fixup deprecated structure layouts here, so we have
4223 * a common path later on.
4225 if (ip_tunnel_info_af(info) != AF_INET)
4228 to = (struct bpf_tunnel_key *)compat;
4235 to->tunnel_id = be64_to_cpu(info->key.tun_id);
4236 to->tunnel_tos = info->key.tos;
4237 to->tunnel_ttl = info->key.ttl;
4240 if (flags & BPF_F_TUNINFO_IPV6) {
4241 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4242 sizeof(to->remote_ipv6));
4243 to->tunnel_label = be32_to_cpu(info->key.label);
4245 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4246 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4247 to->tunnel_label = 0;
4250 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4251 memcpy(to_orig, to, size);
4255 memset(to_orig, 0, size);
4259 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4260 .func = bpf_skb_get_tunnel_key,
4262 .ret_type = RET_INTEGER,
4263 .arg1_type = ARG_PTR_TO_CTX,
4264 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4265 .arg3_type = ARG_CONST_SIZE,
4266 .arg4_type = ARG_ANYTHING,
4269 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4271 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4274 if (unlikely(!info ||
4275 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4279 if (unlikely(size < info->options_len)) {
4284 ip_tunnel_info_opts_get(to, info);
4285 if (size > info->options_len)
4286 memset(to + info->options_len, 0, size - info->options_len);
4288 return info->options_len;
4290 memset(to, 0, size);
4294 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4295 .func = bpf_skb_get_tunnel_opt,
4297 .ret_type = RET_INTEGER,
4298 .arg1_type = ARG_PTR_TO_CTX,
4299 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4300 .arg3_type = ARG_CONST_SIZE,
4303 static struct metadata_dst __percpu *md_dst;
4305 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4306 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4308 struct metadata_dst *md = this_cpu_ptr(md_dst);
4309 u8 compat[sizeof(struct bpf_tunnel_key)];
4310 struct ip_tunnel_info *info;
4312 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4313 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
4315 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4317 case offsetof(struct bpf_tunnel_key, tunnel_label):
4318 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4319 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4320 /* Fixup deprecated structure layouts here, so we have
4321 * a common path later on.
4323 memcpy(compat, from, size);
4324 memset(compat + size, 0, sizeof(compat) - size);
4325 from = (const struct bpf_tunnel_key *) compat;
4331 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4336 dst_hold((struct dst_entry *) md);
4337 skb_dst_set(skb, (struct dst_entry *) md);
4339 info = &md->u.tun_info;
4340 memset(info, 0, sizeof(*info));
4341 info->mode = IP_TUNNEL_INFO_TX;
4343 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4344 if (flags & BPF_F_DONT_FRAGMENT)
4345 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4346 if (flags & BPF_F_ZERO_CSUM_TX)
4347 info->key.tun_flags &= ~TUNNEL_CSUM;
4348 if (flags & BPF_F_SEQ_NUMBER)
4349 info->key.tun_flags |= TUNNEL_SEQ;
4351 info->key.tun_id = cpu_to_be64(from->tunnel_id);
4352 info->key.tos = from->tunnel_tos;
4353 info->key.ttl = from->tunnel_ttl;
4355 if (flags & BPF_F_TUNINFO_IPV6) {
4356 info->mode |= IP_TUNNEL_INFO_IPV6;
4357 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4358 sizeof(from->remote_ipv6));
4359 info->key.label = cpu_to_be32(from->tunnel_label) &
4360 IPV6_FLOWLABEL_MASK;
4362 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4368 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4369 .func = bpf_skb_set_tunnel_key,
4371 .ret_type = RET_INTEGER,
4372 .arg1_type = ARG_PTR_TO_CTX,
4373 .arg2_type = ARG_PTR_TO_MEM,
4374 .arg3_type = ARG_CONST_SIZE,
4375 .arg4_type = ARG_ANYTHING,
4378 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4379 const u8 *, from, u32, size)
4381 struct ip_tunnel_info *info = skb_tunnel_info(skb);
4382 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4384 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4386 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4389 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4394 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4395 .func = bpf_skb_set_tunnel_opt,
4397 .ret_type = RET_INTEGER,
4398 .arg1_type = ARG_PTR_TO_CTX,
4399 .arg2_type = ARG_PTR_TO_MEM,
4400 .arg3_type = ARG_CONST_SIZE,
4403 static const struct bpf_func_proto *
4404 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4407 struct metadata_dst __percpu *tmp;
4409 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4414 if (cmpxchg(&md_dst, NULL, tmp))
4415 metadata_dst_free_percpu(tmp);
4419 case BPF_FUNC_skb_set_tunnel_key:
4420 return &bpf_skb_set_tunnel_key_proto;
4421 case BPF_FUNC_skb_set_tunnel_opt:
4422 return &bpf_skb_set_tunnel_opt_proto;
4428 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4431 struct bpf_array *array = container_of(map, struct bpf_array, map);
4432 struct cgroup *cgrp;
4435 sk = skb_to_full_sk(skb);
4436 if (!sk || !sk_fullsock(sk))
4438 if (unlikely(idx >= array->map.max_entries))
4441 cgrp = READ_ONCE(array->ptrs[idx]);
4442 if (unlikely(!cgrp))
4445 return sk_under_cgroup_hierarchy(sk, cgrp);
4448 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4449 .func = bpf_skb_under_cgroup,
4451 .ret_type = RET_INTEGER,
4452 .arg1_type = ARG_PTR_TO_CTX,
4453 .arg2_type = ARG_CONST_MAP_PTR,
4454 .arg3_type = ARG_ANYTHING,
4457 #ifdef CONFIG_SOCK_CGROUP_DATA
4458 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4460 struct cgroup *cgrp;
4462 sk = sk_to_full_sk(sk);
4463 if (!sk || !sk_fullsock(sk))
4466 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4467 return cgroup_id(cgrp);
4470 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4472 return __bpf_sk_cgroup_id(skb->sk);
4475 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4476 .func = bpf_skb_cgroup_id,
4478 .ret_type = RET_INTEGER,
4479 .arg1_type = ARG_PTR_TO_CTX,
4482 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4485 struct cgroup *ancestor;
4486 struct cgroup *cgrp;
4488 sk = sk_to_full_sk(sk);
4489 if (!sk || !sk_fullsock(sk))
4492 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4493 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4497 return cgroup_id(ancestor);
4500 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4503 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4506 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4507 .func = bpf_skb_ancestor_cgroup_id,
4509 .ret_type = RET_INTEGER,
4510 .arg1_type = ARG_PTR_TO_CTX,
4511 .arg2_type = ARG_ANYTHING,
4514 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4516 return __bpf_sk_cgroup_id(sk);
4519 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4520 .func = bpf_sk_cgroup_id,
4522 .ret_type = RET_INTEGER,
4523 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4526 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4528 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4531 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4532 .func = bpf_sk_ancestor_cgroup_id,
4534 .ret_type = RET_INTEGER,
4535 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4536 .arg2_type = ARG_ANYTHING,
4540 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4541 unsigned long off, unsigned long len)
4543 memcpy(dst_buff, src_buff + off, len);
4547 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4548 u64, flags, void *, meta, u64, meta_size)
4550 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4552 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4554 if (unlikely(!xdp ||
4555 xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4558 return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4559 xdp_size, bpf_xdp_copy);
4562 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4563 .func = bpf_xdp_event_output,
4565 .ret_type = RET_INTEGER,
4566 .arg1_type = ARG_PTR_TO_CTX,
4567 .arg2_type = ARG_CONST_MAP_PTR,
4568 .arg3_type = ARG_ANYTHING,
4569 .arg4_type = ARG_PTR_TO_MEM,
4570 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4573 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4575 const struct bpf_func_proto bpf_xdp_output_proto = {
4576 .func = bpf_xdp_event_output,
4578 .ret_type = RET_INTEGER,
4579 .arg1_type = ARG_PTR_TO_BTF_ID,
4580 .arg1_btf_id = &bpf_xdp_output_btf_ids[0],
4581 .arg2_type = ARG_CONST_MAP_PTR,
4582 .arg3_type = ARG_ANYTHING,
4583 .arg4_type = ARG_PTR_TO_MEM,
4584 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4587 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4589 return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4592 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4593 .func = bpf_get_socket_cookie,
4595 .ret_type = RET_INTEGER,
4596 .arg1_type = ARG_PTR_TO_CTX,
4599 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4601 return __sock_gen_cookie(ctx->sk);
4604 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4605 .func = bpf_get_socket_cookie_sock_addr,
4607 .ret_type = RET_INTEGER,
4608 .arg1_type = ARG_PTR_TO_CTX,
4611 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4613 return __sock_gen_cookie(ctx);
4616 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4617 .func = bpf_get_socket_cookie_sock,
4619 .ret_type = RET_INTEGER,
4620 .arg1_type = ARG_PTR_TO_CTX,
4623 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
4625 return sk ? sock_gen_cookie(sk) : 0;
4628 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
4629 .func = bpf_get_socket_ptr_cookie,
4631 .ret_type = RET_INTEGER,
4632 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4635 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4637 return __sock_gen_cookie(ctx->sk);
4640 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4641 .func = bpf_get_socket_cookie_sock_ops,
4643 .ret_type = RET_INTEGER,
4644 .arg1_type = ARG_PTR_TO_CTX,
4647 static u64 __bpf_get_netns_cookie(struct sock *sk)
4649 const struct net *net = sk ? sock_net(sk) : &init_net;
4651 return net->net_cookie;
4654 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
4656 return __bpf_get_netns_cookie(ctx);
4659 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
4660 .func = bpf_get_netns_cookie_sock,
4662 .ret_type = RET_INTEGER,
4663 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4666 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4668 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4671 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
4672 .func = bpf_get_netns_cookie_sock_addr,
4674 .ret_type = RET_INTEGER,
4675 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4678 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4680 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4683 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = {
4684 .func = bpf_get_netns_cookie_sock_ops,
4686 .ret_type = RET_INTEGER,
4687 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4690 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx)
4692 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4695 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = {
4696 .func = bpf_get_netns_cookie_sk_msg,
4698 .ret_type = RET_INTEGER,
4699 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4702 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4704 struct sock *sk = sk_to_full_sk(skb->sk);
4707 if (!sk || !sk_fullsock(sk))
4709 kuid = sock_net_uid(sock_net(sk), sk);
4710 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4713 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4714 .func = bpf_get_socket_uid,
4716 .ret_type = RET_INTEGER,
4717 .arg1_type = ARG_PTR_TO_CTX,
4720 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
4721 char *optval, int optlen)
4723 char devname[IFNAMSIZ];
4729 if (!sk_fullsock(sk))
4732 sock_owned_by_me(sk);
4734 if (level == SOL_SOCKET) {
4735 if (optlen != sizeof(int) && optname != SO_BINDTODEVICE)
4737 val = *((int *)optval);
4738 valbool = val ? 1 : 0;
4740 /* Only some socketops are supported */
4743 val = min_t(u32, val, sysctl_rmem_max);
4744 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4745 WRITE_ONCE(sk->sk_rcvbuf,
4746 max_t(int, val * 2, SOCK_MIN_RCVBUF));
4749 val = min_t(u32, val, sysctl_wmem_max);
4750 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4751 WRITE_ONCE(sk->sk_sndbuf,
4752 max_t(int, val * 2, SOCK_MIN_SNDBUF));
4754 case SO_MAX_PACING_RATE: /* 32bit version */
4756 cmpxchg(&sk->sk_pacing_status,
4759 sk->sk_max_pacing_rate = (val == ~0U) ?
4760 ~0UL : (unsigned int)val;
4761 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4762 sk->sk_max_pacing_rate);
4765 sk->sk_priority = val;
4770 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
4773 if (sk->sk_mark != val) {
4778 case SO_BINDTODEVICE:
4779 optlen = min_t(long, optlen, IFNAMSIZ - 1);
4780 strncpy(devname, optval, optlen);
4781 devname[optlen] = 0;
4784 if (devname[0] != '\0') {
4785 struct net_device *dev;
4790 dev = dev_get_by_name(net, devname);
4793 ifindex = dev->ifindex;
4797 case SO_BINDTOIFINDEX:
4798 if (optname == SO_BINDTOIFINDEX)
4800 ret = sock_bindtoindex(sk, ifindex, false);
4803 if (sk->sk_prot->keepalive)
4804 sk->sk_prot->keepalive(sk, valbool);
4805 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
4808 sk->sk_reuseport = valbool;
4814 } else if (level == SOL_IP) {
4815 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4818 val = *((int *)optval);
4819 /* Only some options are supported */
4822 if (val < -1 || val > 0xff) {
4825 struct inet_sock *inet = inet_sk(sk);
4835 #if IS_ENABLED(CONFIG_IPV6)
4836 } else if (level == SOL_IPV6) {
4837 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4840 val = *((int *)optval);
4841 /* Only some options are supported */
4844 if (val < -1 || val > 0xff) {
4847 struct ipv6_pinfo *np = inet6_sk(sk);
4858 } else if (level == SOL_TCP &&
4859 sk->sk_prot->setsockopt == tcp_setsockopt) {
4860 if (optname == TCP_CONGESTION) {
4861 char name[TCP_CA_NAME_MAX];
4863 strncpy(name, optval, min_t(long, optlen,
4864 TCP_CA_NAME_MAX-1));
4865 name[TCP_CA_NAME_MAX-1] = 0;
4866 ret = tcp_set_congestion_control(sk, name, false, true);
4868 struct inet_connection_sock *icsk = inet_csk(sk);
4869 struct tcp_sock *tp = tcp_sk(sk);
4870 unsigned long timeout;
4872 if (optlen != sizeof(int))
4875 val = *((int *)optval);
4876 /* Only some options are supported */
4879 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4884 case TCP_BPF_SNDCWND_CLAMP:
4888 tp->snd_cwnd_clamp = val;
4889 tp->snd_ssthresh = val;
4892 case TCP_BPF_DELACK_MAX:
4893 timeout = usecs_to_jiffies(val);
4894 if (timeout > TCP_DELACK_MAX ||
4895 timeout < TCP_TIMEOUT_MIN)
4897 inet_csk(sk)->icsk_delack_max = timeout;
4899 case TCP_BPF_RTO_MIN:
4900 timeout = usecs_to_jiffies(val);
4901 if (timeout > TCP_RTO_MIN ||
4902 timeout < TCP_TIMEOUT_MIN)
4904 inet_csk(sk)->icsk_rto_min = timeout;
4907 if (val < 0 || val > 1)
4913 ret = tcp_sock_set_keepidle_locked(sk, val);
4916 if (val < 1 || val > MAX_TCP_KEEPINTVL)
4919 tp->keepalive_intvl = val * HZ;
4922 if (val < 1 || val > MAX_TCP_KEEPCNT)
4925 tp->keepalive_probes = val;
4928 if (val < 1 || val > MAX_TCP_SYNCNT)
4931 icsk->icsk_syn_retries = val;
4933 case TCP_USER_TIMEOUT:
4937 icsk->icsk_user_timeout = val;
4939 case TCP_NOTSENT_LOWAT:
4940 tp->notsent_lowat = val;
4941 sk->sk_write_space(sk);
4943 case TCP_WINDOW_CLAMP:
4944 ret = tcp_set_window_clamp(sk, val);
4957 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
4958 char *optval, int optlen)
4960 if (!sk_fullsock(sk))
4963 sock_owned_by_me(sk);
4965 if (level == SOL_SOCKET) {
4966 if (optlen != sizeof(int))
4971 *((int *)optval) = sk->sk_mark;
4974 *((int *)optval) = sk->sk_priority;
4976 case SO_BINDTOIFINDEX:
4977 *((int *)optval) = sk->sk_bound_dev_if;
4980 *((int *)optval) = sk->sk_reuseport;
4986 } else if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4987 struct inet_connection_sock *icsk;
4988 struct tcp_sock *tp;
4991 case TCP_CONGESTION:
4992 icsk = inet_csk(sk);
4994 if (!icsk->icsk_ca_ops || optlen <= 1)
4996 strncpy(optval, icsk->icsk_ca_ops->name, optlen);
4997 optval[optlen - 1] = 0;
5002 if (optlen <= 0 || !tp->saved_syn ||
5003 optlen > tcp_saved_syn_len(tp->saved_syn))
5005 memcpy(optval, tp->saved_syn->data, optlen);
5010 } else if (level == SOL_IP) {
5011 struct inet_sock *inet = inet_sk(sk);
5013 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
5016 /* Only some options are supported */
5019 *((int *)optval) = (int)inet->tos;
5024 #if IS_ENABLED(CONFIG_IPV6)
5025 } else if (level == SOL_IPV6) {
5026 struct ipv6_pinfo *np = inet6_sk(sk);
5028 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
5031 /* Only some options are supported */
5034 *((int *)optval) = (int)np->tclass;
5046 memset(optval, 0, optlen);
5050 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level,
5051 int, optname, char *, optval, int, optlen)
5053 if (level == SOL_TCP && optname == TCP_CONGESTION) {
5054 if (optlen >= sizeof("cdg") - 1 &&
5055 !strncmp("cdg", optval, optlen))
5059 return _bpf_setsockopt(sk, level, optname, optval, optlen);
5062 const struct bpf_func_proto bpf_sk_setsockopt_proto = {
5063 .func = bpf_sk_setsockopt,
5065 .ret_type = RET_INTEGER,
5066 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5067 .arg2_type = ARG_ANYTHING,
5068 .arg3_type = ARG_ANYTHING,
5069 .arg4_type = ARG_PTR_TO_MEM,
5070 .arg5_type = ARG_CONST_SIZE,
5073 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level,
5074 int, optname, char *, optval, int, optlen)
5076 return _bpf_getsockopt(sk, level, optname, optval, optlen);
5079 const struct bpf_func_proto bpf_sk_getsockopt_proto = {
5080 .func = bpf_sk_getsockopt,
5082 .ret_type = RET_INTEGER,
5083 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5084 .arg2_type = ARG_ANYTHING,
5085 .arg3_type = ARG_ANYTHING,
5086 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5087 .arg5_type = ARG_CONST_SIZE,
5090 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5091 int, level, int, optname, char *, optval, int, optlen)
5093 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5096 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5097 .func = bpf_sock_addr_setsockopt,
5099 .ret_type = RET_INTEGER,
5100 .arg1_type = ARG_PTR_TO_CTX,
5101 .arg2_type = ARG_ANYTHING,
5102 .arg3_type = ARG_ANYTHING,
5103 .arg4_type = ARG_PTR_TO_MEM,
5104 .arg5_type = ARG_CONST_SIZE,
5107 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5108 int, level, int, optname, char *, optval, int, optlen)
5110 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5113 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5114 .func = bpf_sock_addr_getsockopt,
5116 .ret_type = RET_INTEGER,
5117 .arg1_type = ARG_PTR_TO_CTX,
5118 .arg2_type = ARG_ANYTHING,
5119 .arg3_type = ARG_ANYTHING,
5120 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5121 .arg5_type = ARG_CONST_SIZE,
5124 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5125 int, level, int, optname, char *, optval, int, optlen)
5127 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5130 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5131 .func = bpf_sock_ops_setsockopt,
5133 .ret_type = RET_INTEGER,
5134 .arg1_type = ARG_PTR_TO_CTX,
5135 .arg2_type = ARG_ANYTHING,
5136 .arg3_type = ARG_ANYTHING,
5137 .arg4_type = ARG_PTR_TO_MEM,
5138 .arg5_type = ARG_CONST_SIZE,
5141 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5142 int optname, const u8 **start)
5144 struct sk_buff *syn_skb = bpf_sock->syn_skb;
5145 const u8 *hdr_start;
5149 /* sk is a request_sock here */
5151 if (optname == TCP_BPF_SYN) {
5152 hdr_start = syn_skb->data;
5153 ret = tcp_hdrlen(syn_skb);
5154 } else if (optname == TCP_BPF_SYN_IP) {
5155 hdr_start = skb_network_header(syn_skb);
5156 ret = skb_network_header_len(syn_skb) +
5157 tcp_hdrlen(syn_skb);
5159 /* optname == TCP_BPF_SYN_MAC */
5160 hdr_start = skb_mac_header(syn_skb);
5161 ret = skb_mac_header_len(syn_skb) +
5162 skb_network_header_len(syn_skb) +
5163 tcp_hdrlen(syn_skb);
5166 struct sock *sk = bpf_sock->sk;
5167 struct saved_syn *saved_syn;
5169 if (sk->sk_state == TCP_NEW_SYN_RECV)
5170 /* synack retransmit. bpf_sock->syn_skb will
5171 * not be available. It has to resort to
5172 * saved_syn (if it is saved).
5174 saved_syn = inet_reqsk(sk)->saved_syn;
5176 saved_syn = tcp_sk(sk)->saved_syn;
5181 if (optname == TCP_BPF_SYN) {
5182 hdr_start = saved_syn->data +
5183 saved_syn->mac_hdrlen +
5184 saved_syn->network_hdrlen;
5185 ret = saved_syn->tcp_hdrlen;
5186 } else if (optname == TCP_BPF_SYN_IP) {
5187 hdr_start = saved_syn->data +
5188 saved_syn->mac_hdrlen;
5189 ret = saved_syn->network_hdrlen +
5190 saved_syn->tcp_hdrlen;
5192 /* optname == TCP_BPF_SYN_MAC */
5194 /* TCP_SAVE_SYN may not have saved the mac hdr */
5195 if (!saved_syn->mac_hdrlen)
5198 hdr_start = saved_syn->data;
5199 ret = saved_syn->mac_hdrlen +
5200 saved_syn->network_hdrlen +
5201 saved_syn->tcp_hdrlen;
5209 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5210 int, level, int, optname, char *, optval, int, optlen)
5212 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5213 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5214 int ret, copy_len = 0;
5217 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5220 if (optlen < copy_len) {
5225 memcpy(optval, start, copy_len);
5228 /* Zero out unused buffer at the end */
5229 memset(optval + copy_len, 0, optlen - copy_len);
5234 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5237 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5238 .func = bpf_sock_ops_getsockopt,
5240 .ret_type = RET_INTEGER,
5241 .arg1_type = ARG_PTR_TO_CTX,
5242 .arg2_type = ARG_ANYTHING,
5243 .arg3_type = ARG_ANYTHING,
5244 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5245 .arg5_type = ARG_CONST_SIZE,
5248 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5251 struct sock *sk = bpf_sock->sk;
5252 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5254 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5257 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5259 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5262 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5263 .func = bpf_sock_ops_cb_flags_set,
5265 .ret_type = RET_INTEGER,
5266 .arg1_type = ARG_PTR_TO_CTX,
5267 .arg2_type = ARG_ANYTHING,
5270 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5271 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5273 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5277 struct sock *sk = ctx->sk;
5278 u32 flags = BIND_FROM_BPF;
5282 if (addr_len < offsetofend(struct sockaddr, sa_family))
5284 if (addr->sa_family == AF_INET) {
5285 if (addr_len < sizeof(struct sockaddr_in))
5287 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5288 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5289 return __inet_bind(sk, addr, addr_len, flags);
5290 #if IS_ENABLED(CONFIG_IPV6)
5291 } else if (addr->sa_family == AF_INET6) {
5292 if (addr_len < SIN6_LEN_RFC2133)
5294 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5295 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5296 /* ipv6_bpf_stub cannot be NULL, since it's called from
5297 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5299 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5300 #endif /* CONFIG_IPV6 */
5302 #endif /* CONFIG_INET */
5304 return -EAFNOSUPPORT;
5307 static const struct bpf_func_proto bpf_bind_proto = {
5310 .ret_type = RET_INTEGER,
5311 .arg1_type = ARG_PTR_TO_CTX,
5312 .arg2_type = ARG_PTR_TO_MEM,
5313 .arg3_type = ARG_CONST_SIZE,
5317 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5318 struct bpf_xfrm_state *, to, u32, size, u64, flags)
5320 const struct sec_path *sp = skb_sec_path(skb);
5321 const struct xfrm_state *x;
5323 if (!sp || unlikely(index >= sp->len || flags))
5326 x = sp->xvec[index];
5328 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5331 to->reqid = x->props.reqid;
5332 to->spi = x->id.spi;
5333 to->family = x->props.family;
5336 if (to->family == AF_INET6) {
5337 memcpy(to->remote_ipv6, x->props.saddr.a6,
5338 sizeof(to->remote_ipv6));
5340 to->remote_ipv4 = x->props.saddr.a4;
5341 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5346 memset(to, 0, size);
5350 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5351 .func = bpf_skb_get_xfrm_state,
5353 .ret_type = RET_INTEGER,
5354 .arg1_type = ARG_PTR_TO_CTX,
5355 .arg2_type = ARG_ANYTHING,
5356 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
5357 .arg4_type = ARG_CONST_SIZE,
5358 .arg5_type = ARG_ANYTHING,
5362 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5363 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
5364 const struct neighbour *neigh,
5365 const struct net_device *dev, u32 mtu)
5367 memcpy(params->dmac, neigh->ha, ETH_ALEN);
5368 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5369 params->h_vlan_TCI = 0;
5370 params->h_vlan_proto = 0;
5372 params->mtu_result = mtu; /* union with tot_len */
5378 #if IS_ENABLED(CONFIG_INET)
5379 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5380 u32 flags, bool check_mtu)
5382 struct fib_nh_common *nhc;
5383 struct in_device *in_dev;
5384 struct neighbour *neigh;
5385 struct net_device *dev;
5386 struct fib_result res;
5391 dev = dev_get_by_index_rcu(net, params->ifindex);
5395 /* verify forwarding is enabled on this interface */
5396 in_dev = __in_dev_get_rcu(dev);
5397 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5398 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5400 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5402 fl4.flowi4_oif = params->ifindex;
5404 fl4.flowi4_iif = params->ifindex;
5407 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5408 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5409 fl4.flowi4_flags = 0;
5411 fl4.flowi4_proto = params->l4_protocol;
5412 fl4.daddr = params->ipv4_dst;
5413 fl4.saddr = params->ipv4_src;
5414 fl4.fl4_sport = params->sport;
5415 fl4.fl4_dport = params->dport;
5416 fl4.flowi4_multipath_hash = 0;
5418 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5419 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5420 struct fib_table *tb;
5422 tb = fib_get_table(net, tbid);
5424 return BPF_FIB_LKUP_RET_NOT_FWDED;
5426 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5428 fl4.flowi4_mark = 0;
5429 fl4.flowi4_secid = 0;
5430 fl4.flowi4_tun_key.tun_id = 0;
5431 fl4.flowi4_uid = sock_net_uid(net, NULL);
5433 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5437 /* map fib lookup errors to RTN_ type */
5439 return BPF_FIB_LKUP_RET_BLACKHOLE;
5440 if (err == -EHOSTUNREACH)
5441 return BPF_FIB_LKUP_RET_UNREACHABLE;
5443 return BPF_FIB_LKUP_RET_PROHIBIT;
5445 return BPF_FIB_LKUP_RET_NOT_FWDED;
5448 if (res.type != RTN_UNICAST)
5449 return BPF_FIB_LKUP_RET_NOT_FWDED;
5451 if (fib_info_num_path(res.fi) > 1)
5452 fib_select_path(net, &res, &fl4, NULL);
5455 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5456 if (params->tot_len > mtu) {
5457 params->mtu_result = mtu; /* union with tot_len */
5458 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5464 /* do not handle lwt encaps right now */
5465 if (nhc->nhc_lwtstate)
5466 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5470 params->rt_metric = res.fi->fib_priority;
5471 params->ifindex = dev->ifindex;
5473 /* xdp and cls_bpf programs are run in RCU-bh so
5474 * rcu_read_lock_bh is not needed here
5476 if (likely(nhc->nhc_gw_family != AF_INET6)) {
5477 if (nhc->nhc_gw_family)
5478 params->ipv4_dst = nhc->nhc_gw.ipv4;
5480 neigh = __ipv4_neigh_lookup_noref(dev,
5481 (__force u32)params->ipv4_dst);
5483 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5485 params->family = AF_INET6;
5486 *dst = nhc->nhc_gw.ipv6;
5487 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5491 return BPF_FIB_LKUP_RET_NO_NEIGH;
5493 return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5497 #if IS_ENABLED(CONFIG_IPV6)
5498 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5499 u32 flags, bool check_mtu)
5501 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5502 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5503 struct fib6_result res = {};
5504 struct neighbour *neigh;
5505 struct net_device *dev;
5506 struct inet6_dev *idev;
5512 /* link local addresses are never forwarded */
5513 if (rt6_need_strict(dst) || rt6_need_strict(src))
5514 return BPF_FIB_LKUP_RET_NOT_FWDED;
5516 dev = dev_get_by_index_rcu(net, params->ifindex);
5520 idev = __in6_dev_get_safely(dev);
5521 if (unlikely(!idev || !idev->cnf.forwarding))
5522 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5524 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5526 oif = fl6.flowi6_oif = params->ifindex;
5528 oif = fl6.flowi6_iif = params->ifindex;
5530 strict = RT6_LOOKUP_F_HAS_SADDR;
5532 fl6.flowlabel = params->flowinfo;
5533 fl6.flowi6_scope = 0;
5534 fl6.flowi6_flags = 0;
5537 fl6.flowi6_proto = params->l4_protocol;
5540 fl6.fl6_sport = params->sport;
5541 fl6.fl6_dport = params->dport;
5543 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5544 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5545 struct fib6_table *tb;
5547 tb = ipv6_stub->fib6_get_table(net, tbid);
5549 return BPF_FIB_LKUP_RET_NOT_FWDED;
5551 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5554 fl6.flowi6_mark = 0;
5555 fl6.flowi6_secid = 0;
5556 fl6.flowi6_tun_key.tun_id = 0;
5557 fl6.flowi6_uid = sock_net_uid(net, NULL);
5559 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5562 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5563 res.f6i == net->ipv6.fib6_null_entry))
5564 return BPF_FIB_LKUP_RET_NOT_FWDED;
5566 switch (res.fib6_type) {
5567 /* only unicast is forwarded */
5571 return BPF_FIB_LKUP_RET_BLACKHOLE;
5572 case RTN_UNREACHABLE:
5573 return BPF_FIB_LKUP_RET_UNREACHABLE;
5575 return BPF_FIB_LKUP_RET_PROHIBIT;
5577 return BPF_FIB_LKUP_RET_NOT_FWDED;
5580 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5581 fl6.flowi6_oif != 0, NULL, strict);
5584 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5585 if (params->tot_len > mtu) {
5586 params->mtu_result = mtu; /* union with tot_len */
5587 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5591 if (res.nh->fib_nh_lws)
5592 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5594 if (res.nh->fib_nh_gw_family)
5595 *dst = res.nh->fib_nh_gw6;
5597 dev = res.nh->fib_nh_dev;
5598 params->rt_metric = res.f6i->fib6_metric;
5599 params->ifindex = dev->ifindex;
5601 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
5604 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5606 return BPF_FIB_LKUP_RET_NO_NEIGH;
5608 return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5612 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
5613 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5615 if (plen < sizeof(*params))
5618 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5621 switch (params->family) {
5622 #if IS_ENABLED(CONFIG_INET)
5624 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
5627 #if IS_ENABLED(CONFIG_IPV6)
5629 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
5633 return -EAFNOSUPPORT;
5636 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
5637 .func = bpf_xdp_fib_lookup,
5639 .ret_type = RET_INTEGER,
5640 .arg1_type = ARG_PTR_TO_CTX,
5641 .arg2_type = ARG_PTR_TO_MEM,
5642 .arg3_type = ARG_CONST_SIZE,
5643 .arg4_type = ARG_ANYTHING,
5646 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
5647 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5649 struct net *net = dev_net(skb->dev);
5650 int rc = -EAFNOSUPPORT;
5651 bool check_mtu = false;
5653 if (plen < sizeof(*params))
5656 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5659 if (params->tot_len)
5662 switch (params->family) {
5663 #if IS_ENABLED(CONFIG_INET)
5665 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
5668 #if IS_ENABLED(CONFIG_IPV6)
5670 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
5675 if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
5676 struct net_device *dev;
5678 /* When tot_len isn't provided by user, check skb
5679 * against MTU of FIB lookup resulting net_device
5681 dev = dev_get_by_index_rcu(net, params->ifindex);
5682 if (!is_skb_forwardable(dev, skb))
5683 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
5685 params->mtu_result = dev->mtu; /* union with tot_len */
5691 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
5692 .func = bpf_skb_fib_lookup,
5694 .ret_type = RET_INTEGER,
5695 .arg1_type = ARG_PTR_TO_CTX,
5696 .arg2_type = ARG_PTR_TO_MEM,
5697 .arg3_type = ARG_CONST_SIZE,
5698 .arg4_type = ARG_ANYTHING,
5701 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
5704 struct net *netns = dev_net(dev_curr);
5706 /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
5710 return dev_get_by_index_rcu(netns, ifindex);
5713 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
5714 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5716 int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5717 struct net_device *dev = skb->dev;
5718 int skb_len, dev_len;
5721 if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
5724 if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
5727 dev = __dev_via_ifindex(dev, ifindex);
5731 mtu = READ_ONCE(dev->mtu);
5733 dev_len = mtu + dev->hard_header_len;
5735 /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
5736 skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
5738 skb_len += len_diff; /* minus result pass check */
5739 if (skb_len <= dev_len) {
5740 ret = BPF_MTU_CHK_RET_SUCCESS;
5743 /* At this point, skb->len exceed MTU, but as it include length of all
5744 * segments, it can still be below MTU. The SKB can possibly get
5745 * re-segmented in transmit path (see validate_xmit_skb). Thus, user
5746 * must choose if segs are to be MTU checked.
5748 if (skb_is_gso(skb)) {
5749 ret = BPF_MTU_CHK_RET_SUCCESS;
5751 if (flags & BPF_MTU_CHK_SEGS &&
5752 !skb_gso_validate_network_len(skb, mtu))
5753 ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
5756 /* BPF verifier guarantees valid pointer */
5762 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
5763 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5765 struct net_device *dev = xdp->rxq->dev;
5766 int xdp_len = xdp->data_end - xdp->data;
5767 int ret = BPF_MTU_CHK_RET_SUCCESS;
5770 /* XDP variant doesn't support multi-buffer segment check (yet) */
5771 if (unlikely(flags))
5774 dev = __dev_via_ifindex(dev, ifindex);
5778 mtu = READ_ONCE(dev->mtu);
5780 /* Add L2-header as dev MTU is L3 size */
5781 dev_len = mtu + dev->hard_header_len;
5783 /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
5785 xdp_len = *mtu_len + dev->hard_header_len;
5787 xdp_len += len_diff; /* minus result pass check */
5788 if (xdp_len > dev_len)
5789 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5791 /* BPF verifier guarantees valid pointer */
5797 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
5798 .func = bpf_skb_check_mtu,
5800 .ret_type = RET_INTEGER,
5801 .arg1_type = ARG_PTR_TO_CTX,
5802 .arg2_type = ARG_ANYTHING,
5803 .arg3_type = ARG_PTR_TO_INT,
5804 .arg4_type = ARG_ANYTHING,
5805 .arg5_type = ARG_ANYTHING,
5808 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
5809 .func = bpf_xdp_check_mtu,
5811 .ret_type = RET_INTEGER,
5812 .arg1_type = ARG_PTR_TO_CTX,
5813 .arg2_type = ARG_ANYTHING,
5814 .arg3_type = ARG_PTR_TO_INT,
5815 .arg4_type = ARG_ANYTHING,
5816 .arg5_type = ARG_ANYTHING,
5819 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5820 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
5823 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
5825 if (!seg6_validate_srh(srh, len, false))
5829 case BPF_LWT_ENCAP_SEG6_INLINE:
5830 if (skb->protocol != htons(ETH_P_IPV6))
5833 err = seg6_do_srh_inline(skb, srh);
5835 case BPF_LWT_ENCAP_SEG6:
5836 skb_reset_inner_headers(skb);
5837 skb->encapsulation = 1;
5838 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
5844 bpf_compute_data_pointers(skb);
5848 ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5849 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
5851 return seg6_lookup_nexthop(skb, NULL, 0);
5853 #endif /* CONFIG_IPV6_SEG6_BPF */
5855 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5856 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
5859 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
5863 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
5867 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5868 case BPF_LWT_ENCAP_SEG6:
5869 case BPF_LWT_ENCAP_SEG6_INLINE:
5870 return bpf_push_seg6_encap(skb, type, hdr, len);
5872 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5873 case BPF_LWT_ENCAP_IP:
5874 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
5881 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
5882 void *, hdr, u32, len)
5885 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5886 case BPF_LWT_ENCAP_IP:
5887 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
5894 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
5895 .func = bpf_lwt_in_push_encap,
5897 .ret_type = RET_INTEGER,
5898 .arg1_type = ARG_PTR_TO_CTX,
5899 .arg2_type = ARG_ANYTHING,
5900 .arg3_type = ARG_PTR_TO_MEM,
5901 .arg4_type = ARG_CONST_SIZE
5904 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
5905 .func = bpf_lwt_xmit_push_encap,
5907 .ret_type = RET_INTEGER,
5908 .arg1_type = ARG_PTR_TO_CTX,
5909 .arg2_type = ARG_ANYTHING,
5910 .arg3_type = ARG_PTR_TO_MEM,
5911 .arg4_type = ARG_CONST_SIZE
5914 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5915 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
5916 const void *, from, u32, len)
5918 struct seg6_bpf_srh_state *srh_state =
5919 this_cpu_ptr(&seg6_bpf_srh_states);
5920 struct ipv6_sr_hdr *srh = srh_state->srh;
5921 void *srh_tlvs, *srh_end, *ptr;
5927 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
5928 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
5930 ptr = skb->data + offset;
5931 if (ptr >= srh_tlvs && ptr + len <= srh_end)
5932 srh_state->valid = false;
5933 else if (ptr < (void *)&srh->flags ||
5934 ptr + len > (void *)&srh->segments)
5937 if (unlikely(bpf_try_make_writable(skb, offset + len)))
5939 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5941 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5943 memcpy(skb->data + offset, from, len);
5947 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
5948 .func = bpf_lwt_seg6_store_bytes,
5950 .ret_type = RET_INTEGER,
5951 .arg1_type = ARG_PTR_TO_CTX,
5952 .arg2_type = ARG_ANYTHING,
5953 .arg3_type = ARG_PTR_TO_MEM,
5954 .arg4_type = ARG_CONST_SIZE
5957 static void bpf_update_srh_state(struct sk_buff *skb)
5959 struct seg6_bpf_srh_state *srh_state =
5960 this_cpu_ptr(&seg6_bpf_srh_states);
5963 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
5964 srh_state->srh = NULL;
5966 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5967 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
5968 srh_state->valid = true;
5972 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
5973 u32, action, void *, param, u32, param_len)
5975 struct seg6_bpf_srh_state *srh_state =
5976 this_cpu_ptr(&seg6_bpf_srh_states);
5981 case SEG6_LOCAL_ACTION_END_X:
5982 if (!seg6_bpf_has_valid_srh(skb))
5984 if (param_len != sizeof(struct in6_addr))
5986 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
5987 case SEG6_LOCAL_ACTION_END_T:
5988 if (!seg6_bpf_has_valid_srh(skb))
5990 if (param_len != sizeof(int))
5992 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5993 case SEG6_LOCAL_ACTION_END_DT6:
5994 if (!seg6_bpf_has_valid_srh(skb))
5996 if (param_len != sizeof(int))
5999 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
6001 if (!pskb_pull(skb, hdroff))
6004 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
6005 skb_reset_network_header(skb);
6006 skb_reset_transport_header(skb);
6007 skb->encapsulation = 0;
6009 bpf_compute_data_pointers(skb);
6010 bpf_update_srh_state(skb);
6011 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6012 case SEG6_LOCAL_ACTION_END_B6:
6013 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6015 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
6018 bpf_update_srh_state(skb);
6021 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
6022 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6024 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
6027 bpf_update_srh_state(skb);
6035 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
6036 .func = bpf_lwt_seg6_action,
6038 .ret_type = RET_INTEGER,
6039 .arg1_type = ARG_PTR_TO_CTX,
6040 .arg2_type = ARG_ANYTHING,
6041 .arg3_type = ARG_PTR_TO_MEM,
6042 .arg4_type = ARG_CONST_SIZE
6045 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
6048 struct seg6_bpf_srh_state *srh_state =
6049 this_cpu_ptr(&seg6_bpf_srh_states);
6050 struct ipv6_sr_hdr *srh = srh_state->srh;
6051 void *srh_end, *srh_tlvs, *ptr;
6052 struct ipv6hdr *hdr;
6056 if (unlikely(srh == NULL))
6059 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
6060 ((srh->first_segment + 1) << 4));
6061 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
6063 ptr = skb->data + offset;
6065 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
6067 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
6071 ret = skb_cow_head(skb, len);
6072 if (unlikely(ret < 0))
6075 ret = bpf_skb_net_hdr_push(skb, offset, len);
6077 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6080 bpf_compute_data_pointers(skb);
6081 if (unlikely(ret < 0))
6084 hdr = (struct ipv6hdr *)skb->data;
6085 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6087 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6089 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6090 srh_state->hdrlen += len;
6091 srh_state->valid = false;
6095 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6096 .func = bpf_lwt_seg6_adjust_srh,
6098 .ret_type = RET_INTEGER,
6099 .arg1_type = ARG_PTR_TO_CTX,
6100 .arg2_type = ARG_ANYTHING,
6101 .arg3_type = ARG_ANYTHING,
6103 #endif /* CONFIG_IPV6_SEG6_BPF */
6106 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6107 int dif, int sdif, u8 family, u8 proto)
6109 bool refcounted = false;
6110 struct sock *sk = NULL;
6112 if (family == AF_INET) {
6113 __be32 src4 = tuple->ipv4.saddr;
6114 __be32 dst4 = tuple->ipv4.daddr;
6116 if (proto == IPPROTO_TCP)
6117 sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
6118 src4, tuple->ipv4.sport,
6119 dst4, tuple->ipv4.dport,
6120 dif, sdif, &refcounted);
6122 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6123 dst4, tuple->ipv4.dport,
6124 dif, sdif, &udp_table, NULL);
6125 #if IS_ENABLED(CONFIG_IPV6)
6127 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6128 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6130 if (proto == IPPROTO_TCP)
6131 sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
6132 src6, tuple->ipv6.sport,
6133 dst6, ntohs(tuple->ipv6.dport),
6134 dif, sdif, &refcounted);
6135 else if (likely(ipv6_bpf_stub))
6136 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6137 src6, tuple->ipv6.sport,
6138 dst6, tuple->ipv6.dport,
6144 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6145 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6151 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6152 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6153 * Returns the socket as an 'unsigned long' to simplify the casting in the
6154 * callers to satisfy BPF_CALL declarations.
6156 static struct sock *
6157 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6158 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6161 struct sock *sk = NULL;
6162 u8 family = AF_UNSPEC;
6166 if (len == sizeof(tuple->ipv4))
6168 else if (len == sizeof(tuple->ipv6))
6173 if (unlikely(family == AF_UNSPEC || flags ||
6174 !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6177 if (family == AF_INET)
6178 sdif = inet_sdif(skb);
6180 sdif = inet6_sdif(skb);
6182 if ((s32)netns_id < 0) {
6184 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6186 net = get_net_ns_by_id(caller_net, netns_id);
6189 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6197 static struct sock *
6198 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6199 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6202 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6203 ifindex, proto, netns_id, flags);
6206 sk = sk_to_full_sk(sk);
6207 if (!sk_fullsock(sk)) {
6216 static struct sock *
6217 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6218 u8 proto, u64 netns_id, u64 flags)
6220 struct net *caller_net;
6224 caller_net = dev_net(skb->dev);
6225 ifindex = skb->dev->ifindex;
6227 caller_net = sock_net(skb->sk);
6231 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6235 static struct sock *
6236 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6237 u8 proto, u64 netns_id, u64 flags)
6239 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6243 sk = sk_to_full_sk(sk);
6244 if (!sk_fullsock(sk)) {
6253 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6254 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6256 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6260 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6261 .func = bpf_skc_lookup_tcp,
6264 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6265 .arg1_type = ARG_PTR_TO_CTX,
6266 .arg2_type = ARG_PTR_TO_MEM,
6267 .arg3_type = ARG_CONST_SIZE,
6268 .arg4_type = ARG_ANYTHING,
6269 .arg5_type = ARG_ANYTHING,
6272 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6273 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6275 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6279 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6280 .func = bpf_sk_lookup_tcp,
6283 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6284 .arg1_type = ARG_PTR_TO_CTX,
6285 .arg2_type = ARG_PTR_TO_MEM,
6286 .arg3_type = ARG_CONST_SIZE,
6287 .arg4_type = ARG_ANYTHING,
6288 .arg5_type = ARG_ANYTHING,
6291 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6292 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6294 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6298 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6299 .func = bpf_sk_lookup_udp,
6302 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6303 .arg1_type = ARG_PTR_TO_CTX,
6304 .arg2_type = ARG_PTR_TO_MEM,
6305 .arg3_type = ARG_CONST_SIZE,
6306 .arg4_type = ARG_ANYTHING,
6307 .arg5_type = ARG_ANYTHING,
6310 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6312 if (sk && sk_is_refcounted(sk))
6317 static const struct bpf_func_proto bpf_sk_release_proto = {
6318 .func = bpf_sk_release,
6320 .ret_type = RET_INTEGER,
6321 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6324 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6325 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6327 struct net *caller_net = dev_net(ctx->rxq->dev);
6328 int ifindex = ctx->rxq->dev->ifindex;
6330 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6331 ifindex, IPPROTO_UDP, netns_id,
6335 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6336 .func = bpf_xdp_sk_lookup_udp,
6339 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6340 .arg1_type = ARG_PTR_TO_CTX,
6341 .arg2_type = ARG_PTR_TO_MEM,
6342 .arg3_type = ARG_CONST_SIZE,
6343 .arg4_type = ARG_ANYTHING,
6344 .arg5_type = ARG_ANYTHING,
6347 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6348 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6350 struct net *caller_net = dev_net(ctx->rxq->dev);
6351 int ifindex = ctx->rxq->dev->ifindex;
6353 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6354 ifindex, IPPROTO_TCP, netns_id,
6358 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6359 .func = bpf_xdp_skc_lookup_tcp,
6362 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6363 .arg1_type = ARG_PTR_TO_CTX,
6364 .arg2_type = ARG_PTR_TO_MEM,
6365 .arg3_type = ARG_CONST_SIZE,
6366 .arg4_type = ARG_ANYTHING,
6367 .arg5_type = ARG_ANYTHING,
6370 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6371 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6373 struct net *caller_net = dev_net(ctx->rxq->dev);
6374 int ifindex = ctx->rxq->dev->ifindex;
6376 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6377 ifindex, IPPROTO_TCP, netns_id,
6381 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6382 .func = bpf_xdp_sk_lookup_tcp,
6385 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6386 .arg1_type = ARG_PTR_TO_CTX,
6387 .arg2_type = ARG_PTR_TO_MEM,
6388 .arg3_type = ARG_CONST_SIZE,
6389 .arg4_type = ARG_ANYTHING,
6390 .arg5_type = ARG_ANYTHING,
6393 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6394 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6396 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6397 sock_net(ctx->sk), 0,
6398 IPPROTO_TCP, netns_id, flags);
6401 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6402 .func = bpf_sock_addr_skc_lookup_tcp,
6404 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6405 .arg1_type = ARG_PTR_TO_CTX,
6406 .arg2_type = ARG_PTR_TO_MEM,
6407 .arg3_type = ARG_CONST_SIZE,
6408 .arg4_type = ARG_ANYTHING,
6409 .arg5_type = ARG_ANYTHING,
6412 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6413 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6415 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6416 sock_net(ctx->sk), 0, IPPROTO_TCP,
6420 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6421 .func = bpf_sock_addr_sk_lookup_tcp,
6423 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6424 .arg1_type = ARG_PTR_TO_CTX,
6425 .arg2_type = ARG_PTR_TO_MEM,
6426 .arg3_type = ARG_CONST_SIZE,
6427 .arg4_type = ARG_ANYTHING,
6428 .arg5_type = ARG_ANYTHING,
6431 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6432 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6434 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6435 sock_net(ctx->sk), 0, IPPROTO_UDP,
6439 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6440 .func = bpf_sock_addr_sk_lookup_udp,
6442 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6443 .arg1_type = ARG_PTR_TO_CTX,
6444 .arg2_type = ARG_PTR_TO_MEM,
6445 .arg3_type = ARG_CONST_SIZE,
6446 .arg4_type = ARG_ANYTHING,
6447 .arg5_type = ARG_ANYTHING,
6450 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6451 struct bpf_insn_access_aux *info)
6453 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6457 if (off % size != 0)
6461 case offsetof(struct bpf_tcp_sock, bytes_received):
6462 case offsetof(struct bpf_tcp_sock, bytes_acked):
6463 return size == sizeof(__u64);
6465 return size == sizeof(__u32);
6469 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6470 const struct bpf_insn *si,
6471 struct bpf_insn *insn_buf,
6472 struct bpf_prog *prog, u32 *target_size)
6474 struct bpf_insn *insn = insn_buf;
6476 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \
6478 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \
6479 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6480 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6481 si->dst_reg, si->src_reg, \
6482 offsetof(struct tcp_sock, FIELD)); \
6485 #define BPF_INET_SOCK_GET_COMMON(FIELD) \
6487 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \
6489 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6490 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6491 struct inet_connection_sock, \
6493 si->dst_reg, si->src_reg, \
6495 struct inet_connection_sock, \
6499 if (insn > insn_buf)
6500 return insn - insn_buf;
6503 case offsetof(struct bpf_tcp_sock, rtt_min):
6504 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
6505 sizeof(struct minmax));
6506 BUILD_BUG_ON(sizeof(struct minmax) <
6507 sizeof(struct minmax_sample));
6509 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6510 offsetof(struct tcp_sock, rtt_min) +
6511 offsetof(struct minmax_sample, v));
6513 case offsetof(struct bpf_tcp_sock, snd_cwnd):
6514 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
6516 case offsetof(struct bpf_tcp_sock, srtt_us):
6517 BPF_TCP_SOCK_GET_COMMON(srtt_us);
6519 case offsetof(struct bpf_tcp_sock, snd_ssthresh):
6520 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
6522 case offsetof(struct bpf_tcp_sock, rcv_nxt):
6523 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
6525 case offsetof(struct bpf_tcp_sock, snd_nxt):
6526 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
6528 case offsetof(struct bpf_tcp_sock, snd_una):
6529 BPF_TCP_SOCK_GET_COMMON(snd_una);
6531 case offsetof(struct bpf_tcp_sock, mss_cache):
6532 BPF_TCP_SOCK_GET_COMMON(mss_cache);
6534 case offsetof(struct bpf_tcp_sock, ecn_flags):
6535 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
6537 case offsetof(struct bpf_tcp_sock, rate_delivered):
6538 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
6540 case offsetof(struct bpf_tcp_sock, rate_interval_us):
6541 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
6543 case offsetof(struct bpf_tcp_sock, packets_out):
6544 BPF_TCP_SOCK_GET_COMMON(packets_out);
6546 case offsetof(struct bpf_tcp_sock, retrans_out):
6547 BPF_TCP_SOCK_GET_COMMON(retrans_out);
6549 case offsetof(struct bpf_tcp_sock, total_retrans):
6550 BPF_TCP_SOCK_GET_COMMON(total_retrans);
6552 case offsetof(struct bpf_tcp_sock, segs_in):
6553 BPF_TCP_SOCK_GET_COMMON(segs_in);
6555 case offsetof(struct bpf_tcp_sock, data_segs_in):
6556 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
6558 case offsetof(struct bpf_tcp_sock, segs_out):
6559 BPF_TCP_SOCK_GET_COMMON(segs_out);
6561 case offsetof(struct bpf_tcp_sock, data_segs_out):
6562 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
6564 case offsetof(struct bpf_tcp_sock, lost_out):
6565 BPF_TCP_SOCK_GET_COMMON(lost_out);
6567 case offsetof(struct bpf_tcp_sock, sacked_out):
6568 BPF_TCP_SOCK_GET_COMMON(sacked_out);
6570 case offsetof(struct bpf_tcp_sock, bytes_received):
6571 BPF_TCP_SOCK_GET_COMMON(bytes_received);
6573 case offsetof(struct bpf_tcp_sock, bytes_acked):
6574 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
6576 case offsetof(struct bpf_tcp_sock, dsack_dups):
6577 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
6579 case offsetof(struct bpf_tcp_sock, delivered):
6580 BPF_TCP_SOCK_GET_COMMON(delivered);
6582 case offsetof(struct bpf_tcp_sock, delivered_ce):
6583 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
6585 case offsetof(struct bpf_tcp_sock, icsk_retransmits):
6586 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
6590 return insn - insn_buf;
6593 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
6595 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
6596 return (unsigned long)sk;
6598 return (unsigned long)NULL;
6601 const struct bpf_func_proto bpf_tcp_sock_proto = {
6602 .func = bpf_tcp_sock,
6604 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
6605 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6608 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
6610 sk = sk_to_full_sk(sk);
6612 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
6613 return (unsigned long)sk;
6615 return (unsigned long)NULL;
6618 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
6619 .func = bpf_get_listener_sock,
6621 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6622 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6625 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
6627 unsigned int iphdr_len;
6629 switch (skb_protocol(skb, true)) {
6630 case cpu_to_be16(ETH_P_IP):
6631 iphdr_len = sizeof(struct iphdr);
6633 case cpu_to_be16(ETH_P_IPV6):
6634 iphdr_len = sizeof(struct ipv6hdr);
6640 if (skb_headlen(skb) < iphdr_len)
6643 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
6646 return INET_ECN_set_ce(skb);
6649 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6650 struct bpf_insn_access_aux *info)
6652 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
6655 if (off % size != 0)
6660 return size == sizeof(__u32);
6664 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
6665 const struct bpf_insn *si,
6666 struct bpf_insn *insn_buf,
6667 struct bpf_prog *prog, u32 *target_size)
6669 struct bpf_insn *insn = insn_buf;
6671 #define BPF_XDP_SOCK_GET(FIELD) \
6673 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \
6674 sizeof_field(struct bpf_xdp_sock, FIELD)); \
6675 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
6676 si->dst_reg, si->src_reg, \
6677 offsetof(struct xdp_sock, FIELD)); \
6681 case offsetof(struct bpf_xdp_sock, queue_id):
6682 BPF_XDP_SOCK_GET(queue_id);
6686 return insn - insn_buf;
6689 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
6690 .func = bpf_skb_ecn_set_ce,
6692 .ret_type = RET_INTEGER,
6693 .arg1_type = ARG_PTR_TO_CTX,
6696 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6697 struct tcphdr *, th, u32, th_len)
6699 #ifdef CONFIG_SYN_COOKIES
6703 if (unlikely(!sk || th_len < sizeof(*th)))
6706 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
6707 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6710 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
6713 if (!th->ack || th->rst || th->syn)
6716 if (tcp_synq_no_recent_overflow(sk))
6719 cookie = ntohl(th->ack_seq) - 1;
6721 switch (sk->sk_family) {
6723 if (unlikely(iph_len < sizeof(struct iphdr)))
6726 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
6729 #if IS_BUILTIN(CONFIG_IPV6)
6731 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6734 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
6736 #endif /* CONFIG_IPV6 */
6739 return -EPROTONOSUPPORT;
6751 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
6752 .func = bpf_tcp_check_syncookie,
6755 .ret_type = RET_INTEGER,
6756 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6757 .arg2_type = ARG_PTR_TO_MEM,
6758 .arg3_type = ARG_CONST_SIZE,
6759 .arg4_type = ARG_PTR_TO_MEM,
6760 .arg5_type = ARG_CONST_SIZE,
6763 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6764 struct tcphdr *, th, u32, th_len)
6766 #ifdef CONFIG_SYN_COOKIES
6770 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
6773 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6776 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
6779 if (!th->syn || th->ack || th->fin || th->rst)
6782 if (unlikely(iph_len < sizeof(struct iphdr)))
6785 /* Both struct iphdr and struct ipv6hdr have the version field at the
6786 * same offset so we can cast to the shorter header (struct iphdr).
6788 switch (((struct iphdr *)iph)->version) {
6790 if (sk->sk_family == AF_INET6 && sk->sk_ipv6only)
6793 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
6796 #if IS_BUILTIN(CONFIG_IPV6)
6798 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6801 if (sk->sk_family != AF_INET6)
6804 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
6806 #endif /* CONFIG_IPV6 */
6809 return -EPROTONOSUPPORT;
6814 return cookie | ((u64)mss << 32);
6817 #endif /* CONFIG_SYN_COOKIES */
6820 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
6821 .func = bpf_tcp_gen_syncookie,
6822 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */
6824 .ret_type = RET_INTEGER,
6825 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6826 .arg2_type = ARG_PTR_TO_MEM,
6827 .arg3_type = ARG_CONST_SIZE,
6828 .arg4_type = ARG_PTR_TO_MEM,
6829 .arg5_type = ARG_CONST_SIZE,
6832 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
6834 if (!sk || flags != 0)
6836 if (!skb_at_tc_ingress(skb))
6838 if (unlikely(dev_net(skb->dev) != sock_net(sk)))
6839 return -ENETUNREACH;
6840 if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
6841 return -ESOCKTNOSUPPORT;
6842 if (sk_is_refcounted(sk) &&
6843 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
6848 skb->destructor = sock_pfree;
6853 static const struct bpf_func_proto bpf_sk_assign_proto = {
6854 .func = bpf_sk_assign,
6856 .ret_type = RET_INTEGER,
6857 .arg1_type = ARG_PTR_TO_CTX,
6858 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6859 .arg3_type = ARG_ANYTHING,
6862 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
6863 u8 search_kind, const u8 *magic,
6864 u8 magic_len, bool *eol)
6870 while (op < opend) {
6873 if (kind == TCPOPT_EOL) {
6875 return ERR_PTR(-ENOMSG);
6876 } else if (kind == TCPOPT_NOP) {
6881 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
6882 /* Something is wrong in the received header.
6883 * Follow the TCP stack's tcp_parse_options()
6884 * and just bail here.
6886 return ERR_PTR(-EFAULT);
6889 if (search_kind == kind) {
6893 if (magic_len > kind_len - 2)
6894 return ERR_PTR(-ENOMSG);
6896 if (!memcmp(&op[2], magic, magic_len))
6903 return ERR_PTR(-ENOMSG);
6906 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6907 void *, search_res, u32, len, u64, flags)
6909 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
6910 const u8 *op, *opend, *magic, *search = search_res;
6911 u8 search_kind, search_len, copy_len, magic_len;
6914 /* 2 byte is the minimal option len except TCPOPT_NOP and
6915 * TCPOPT_EOL which are useless for the bpf prog to learn
6916 * and this helper disallow loading them also.
6918 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
6921 search_kind = search[0];
6922 search_len = search[1];
6924 if (search_len > len || search_kind == TCPOPT_NOP ||
6925 search_kind == TCPOPT_EOL)
6928 if (search_kind == TCPOPT_EXP || search_kind == 253) {
6929 /* 16 or 32 bit magic. +2 for kind and kind length */
6930 if (search_len != 4 && search_len != 6)
6933 magic_len = search_len - 2;
6942 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
6947 op += sizeof(struct tcphdr);
6949 if (!bpf_sock->skb ||
6950 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6951 /* This bpf_sock->op cannot call this helper */
6954 opend = bpf_sock->skb_data_end;
6955 op = bpf_sock->skb->data + sizeof(struct tcphdr);
6958 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
6965 if (copy_len > len) {
6970 memcpy(search_res, op, copy_len);
6974 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
6975 .func = bpf_sock_ops_load_hdr_opt,
6977 .ret_type = RET_INTEGER,
6978 .arg1_type = ARG_PTR_TO_CTX,
6979 .arg2_type = ARG_PTR_TO_MEM,
6980 .arg3_type = ARG_CONST_SIZE,
6981 .arg4_type = ARG_ANYTHING,
6984 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6985 const void *, from, u32, len, u64, flags)
6987 u8 new_kind, new_kind_len, magic_len = 0, *opend;
6988 const u8 *op, *new_op, *magic = NULL;
6989 struct sk_buff *skb;
6992 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
6995 if (len < 2 || flags)
6999 new_kind = new_op[0];
7000 new_kind_len = new_op[1];
7002 if (new_kind_len > len || new_kind == TCPOPT_NOP ||
7003 new_kind == TCPOPT_EOL)
7006 if (new_kind_len > bpf_sock->remaining_opt_len)
7009 /* 253 is another experimental kind */
7010 if (new_kind == TCPOPT_EXP || new_kind == 253) {
7011 if (new_kind_len < 4)
7013 /* Match for the 2 byte magic also.
7014 * RFC 6994: the magic could be 2 or 4 bytes.
7015 * Hence, matching by 2 byte only is on the
7016 * conservative side but it is the right
7017 * thing to do for the 'search-for-duplication'
7024 /* Check for duplication */
7025 skb = bpf_sock->skb;
7026 op = skb->data + sizeof(struct tcphdr);
7027 opend = bpf_sock->skb_data_end;
7029 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
7034 if (PTR_ERR(op) != -ENOMSG)
7038 /* The option has been ended. Treat it as no more
7039 * header option can be written.
7043 /* No duplication found. Store the header option. */
7044 memcpy(opend, from, new_kind_len);
7046 bpf_sock->remaining_opt_len -= new_kind_len;
7047 bpf_sock->skb_data_end += new_kind_len;
7052 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
7053 .func = bpf_sock_ops_store_hdr_opt,
7055 .ret_type = RET_INTEGER,
7056 .arg1_type = ARG_PTR_TO_CTX,
7057 .arg2_type = ARG_PTR_TO_MEM,
7058 .arg3_type = ARG_CONST_SIZE,
7059 .arg4_type = ARG_ANYTHING,
7062 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7063 u32, len, u64, flags)
7065 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7068 if (flags || len < 2)
7071 if (len > bpf_sock->remaining_opt_len)
7074 bpf_sock->remaining_opt_len -= len;
7079 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7080 .func = bpf_sock_ops_reserve_hdr_opt,
7082 .ret_type = RET_INTEGER,
7083 .arg1_type = ARG_PTR_TO_CTX,
7084 .arg2_type = ARG_ANYTHING,
7085 .arg3_type = ARG_ANYTHING,
7088 #endif /* CONFIG_INET */
7090 bool bpf_helper_changes_pkt_data(void *func)
7092 if (func == bpf_skb_vlan_push ||
7093 func == bpf_skb_vlan_pop ||
7094 func == bpf_skb_store_bytes ||
7095 func == bpf_skb_change_proto ||
7096 func == bpf_skb_change_head ||
7097 func == sk_skb_change_head ||
7098 func == bpf_skb_change_tail ||
7099 func == sk_skb_change_tail ||
7100 func == bpf_skb_adjust_room ||
7101 func == sk_skb_adjust_room ||
7102 func == bpf_skb_pull_data ||
7103 func == sk_skb_pull_data ||
7104 func == bpf_clone_redirect ||
7105 func == bpf_l3_csum_replace ||
7106 func == bpf_l4_csum_replace ||
7107 func == bpf_xdp_adjust_head ||
7108 func == bpf_xdp_adjust_meta ||
7109 func == bpf_msg_pull_data ||
7110 func == bpf_msg_push_data ||
7111 func == bpf_msg_pop_data ||
7112 func == bpf_xdp_adjust_tail ||
7113 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7114 func == bpf_lwt_seg6_store_bytes ||
7115 func == bpf_lwt_seg6_adjust_srh ||
7116 func == bpf_lwt_seg6_action ||
7119 func == bpf_sock_ops_store_hdr_opt ||
7121 func == bpf_lwt_in_push_encap ||
7122 func == bpf_lwt_xmit_push_encap)
7128 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7129 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7131 static const struct bpf_func_proto *
7132 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7135 /* inet and inet6 sockets are created in a process
7136 * context so there is always a valid uid/gid
7138 case BPF_FUNC_get_current_uid_gid:
7139 return &bpf_get_current_uid_gid_proto;
7140 case BPF_FUNC_get_local_storage:
7141 return &bpf_get_local_storage_proto;
7142 case BPF_FUNC_get_socket_cookie:
7143 return &bpf_get_socket_cookie_sock_proto;
7144 case BPF_FUNC_get_netns_cookie:
7145 return &bpf_get_netns_cookie_sock_proto;
7146 case BPF_FUNC_perf_event_output:
7147 return &bpf_event_output_data_proto;
7148 case BPF_FUNC_get_current_pid_tgid:
7149 return &bpf_get_current_pid_tgid_proto;
7150 case BPF_FUNC_get_current_comm:
7151 return &bpf_get_current_comm_proto;
7152 #ifdef CONFIG_CGROUPS
7153 case BPF_FUNC_get_current_cgroup_id:
7154 return &bpf_get_current_cgroup_id_proto;
7155 case BPF_FUNC_get_current_ancestor_cgroup_id:
7156 return &bpf_get_current_ancestor_cgroup_id_proto;
7158 #ifdef CONFIG_CGROUP_NET_CLASSID
7159 case BPF_FUNC_get_cgroup_classid:
7160 return &bpf_get_cgroup_classid_curr_proto;
7162 case BPF_FUNC_sk_storage_get:
7163 return &bpf_sk_storage_get_cg_sock_proto;
7164 case BPF_FUNC_ktime_get_coarse_ns:
7165 return &bpf_ktime_get_coarse_ns_proto;
7167 return bpf_base_func_proto(func_id);
7171 static const struct bpf_func_proto *
7172 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7175 /* inet and inet6 sockets are created in a process
7176 * context so there is always a valid uid/gid
7178 case BPF_FUNC_get_current_uid_gid:
7179 return &bpf_get_current_uid_gid_proto;
7181 switch (prog->expected_attach_type) {
7182 case BPF_CGROUP_INET4_CONNECT:
7183 case BPF_CGROUP_INET6_CONNECT:
7184 return &bpf_bind_proto;
7188 case BPF_FUNC_get_socket_cookie:
7189 return &bpf_get_socket_cookie_sock_addr_proto;
7190 case BPF_FUNC_get_netns_cookie:
7191 return &bpf_get_netns_cookie_sock_addr_proto;
7192 case BPF_FUNC_get_local_storage:
7193 return &bpf_get_local_storage_proto;
7194 case BPF_FUNC_perf_event_output:
7195 return &bpf_event_output_data_proto;
7196 case BPF_FUNC_get_current_pid_tgid:
7197 return &bpf_get_current_pid_tgid_proto;
7198 case BPF_FUNC_get_current_comm:
7199 return &bpf_get_current_comm_proto;
7200 #ifdef CONFIG_CGROUPS
7201 case BPF_FUNC_get_current_cgroup_id:
7202 return &bpf_get_current_cgroup_id_proto;
7203 case BPF_FUNC_get_current_ancestor_cgroup_id:
7204 return &bpf_get_current_ancestor_cgroup_id_proto;
7206 #ifdef CONFIG_CGROUP_NET_CLASSID
7207 case BPF_FUNC_get_cgroup_classid:
7208 return &bpf_get_cgroup_classid_curr_proto;
7211 case BPF_FUNC_sk_lookup_tcp:
7212 return &bpf_sock_addr_sk_lookup_tcp_proto;
7213 case BPF_FUNC_sk_lookup_udp:
7214 return &bpf_sock_addr_sk_lookup_udp_proto;
7215 case BPF_FUNC_sk_release:
7216 return &bpf_sk_release_proto;
7217 case BPF_FUNC_skc_lookup_tcp:
7218 return &bpf_sock_addr_skc_lookup_tcp_proto;
7219 #endif /* CONFIG_INET */
7220 case BPF_FUNC_sk_storage_get:
7221 return &bpf_sk_storage_get_proto;
7222 case BPF_FUNC_sk_storage_delete:
7223 return &bpf_sk_storage_delete_proto;
7224 case BPF_FUNC_setsockopt:
7225 switch (prog->expected_attach_type) {
7226 case BPF_CGROUP_INET4_BIND:
7227 case BPF_CGROUP_INET6_BIND:
7228 case BPF_CGROUP_INET4_CONNECT:
7229 case BPF_CGROUP_INET6_CONNECT:
7230 case BPF_CGROUP_UDP4_RECVMSG:
7231 case BPF_CGROUP_UDP6_RECVMSG:
7232 case BPF_CGROUP_UDP4_SENDMSG:
7233 case BPF_CGROUP_UDP6_SENDMSG:
7234 case BPF_CGROUP_INET4_GETPEERNAME:
7235 case BPF_CGROUP_INET6_GETPEERNAME:
7236 case BPF_CGROUP_INET4_GETSOCKNAME:
7237 case BPF_CGROUP_INET6_GETSOCKNAME:
7238 return &bpf_sock_addr_setsockopt_proto;
7242 case BPF_FUNC_getsockopt:
7243 switch (prog->expected_attach_type) {
7244 case BPF_CGROUP_INET4_BIND:
7245 case BPF_CGROUP_INET6_BIND:
7246 case BPF_CGROUP_INET4_CONNECT:
7247 case BPF_CGROUP_INET6_CONNECT:
7248 case BPF_CGROUP_UDP4_RECVMSG:
7249 case BPF_CGROUP_UDP6_RECVMSG:
7250 case BPF_CGROUP_UDP4_SENDMSG:
7251 case BPF_CGROUP_UDP6_SENDMSG:
7252 case BPF_CGROUP_INET4_GETPEERNAME:
7253 case BPF_CGROUP_INET6_GETPEERNAME:
7254 case BPF_CGROUP_INET4_GETSOCKNAME:
7255 case BPF_CGROUP_INET6_GETSOCKNAME:
7256 return &bpf_sock_addr_getsockopt_proto;
7261 return bpf_sk_base_func_proto(func_id);
7265 static const struct bpf_func_proto *
7266 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7269 case BPF_FUNC_skb_load_bytes:
7270 return &bpf_skb_load_bytes_proto;
7271 case BPF_FUNC_skb_load_bytes_relative:
7272 return &bpf_skb_load_bytes_relative_proto;
7273 case BPF_FUNC_get_socket_cookie:
7274 return &bpf_get_socket_cookie_proto;
7275 case BPF_FUNC_get_socket_uid:
7276 return &bpf_get_socket_uid_proto;
7277 case BPF_FUNC_perf_event_output:
7278 return &bpf_skb_event_output_proto;
7280 return bpf_sk_base_func_proto(func_id);
7284 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7285 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7287 static const struct bpf_func_proto *
7288 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7291 case BPF_FUNC_get_local_storage:
7292 return &bpf_get_local_storage_proto;
7293 case BPF_FUNC_sk_fullsock:
7294 return &bpf_sk_fullsock_proto;
7295 case BPF_FUNC_sk_storage_get:
7296 return &bpf_sk_storage_get_proto;
7297 case BPF_FUNC_sk_storage_delete:
7298 return &bpf_sk_storage_delete_proto;
7299 case BPF_FUNC_perf_event_output:
7300 return &bpf_skb_event_output_proto;
7301 #ifdef CONFIG_SOCK_CGROUP_DATA
7302 case BPF_FUNC_skb_cgroup_id:
7303 return &bpf_skb_cgroup_id_proto;
7304 case BPF_FUNC_skb_ancestor_cgroup_id:
7305 return &bpf_skb_ancestor_cgroup_id_proto;
7306 case BPF_FUNC_sk_cgroup_id:
7307 return &bpf_sk_cgroup_id_proto;
7308 case BPF_FUNC_sk_ancestor_cgroup_id:
7309 return &bpf_sk_ancestor_cgroup_id_proto;
7312 case BPF_FUNC_sk_lookup_tcp:
7313 return &bpf_sk_lookup_tcp_proto;
7314 case BPF_FUNC_sk_lookup_udp:
7315 return &bpf_sk_lookup_udp_proto;
7316 case BPF_FUNC_sk_release:
7317 return &bpf_sk_release_proto;
7318 case BPF_FUNC_skc_lookup_tcp:
7319 return &bpf_skc_lookup_tcp_proto;
7320 case BPF_FUNC_tcp_sock:
7321 return &bpf_tcp_sock_proto;
7322 case BPF_FUNC_get_listener_sock:
7323 return &bpf_get_listener_sock_proto;
7324 case BPF_FUNC_skb_ecn_set_ce:
7325 return &bpf_skb_ecn_set_ce_proto;
7328 return sk_filter_func_proto(func_id, prog);
7332 static const struct bpf_func_proto *
7333 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7336 case BPF_FUNC_skb_store_bytes:
7337 return &bpf_skb_store_bytes_proto;
7338 case BPF_FUNC_skb_load_bytes:
7339 return &bpf_skb_load_bytes_proto;
7340 case BPF_FUNC_skb_load_bytes_relative:
7341 return &bpf_skb_load_bytes_relative_proto;
7342 case BPF_FUNC_skb_pull_data:
7343 return &bpf_skb_pull_data_proto;
7344 case BPF_FUNC_csum_diff:
7345 return &bpf_csum_diff_proto;
7346 case BPF_FUNC_csum_update:
7347 return &bpf_csum_update_proto;
7348 case BPF_FUNC_csum_level:
7349 return &bpf_csum_level_proto;
7350 case BPF_FUNC_l3_csum_replace:
7351 return &bpf_l3_csum_replace_proto;
7352 case BPF_FUNC_l4_csum_replace:
7353 return &bpf_l4_csum_replace_proto;
7354 case BPF_FUNC_clone_redirect:
7355 return &bpf_clone_redirect_proto;
7356 case BPF_FUNC_get_cgroup_classid:
7357 return &bpf_get_cgroup_classid_proto;
7358 case BPF_FUNC_skb_vlan_push:
7359 return &bpf_skb_vlan_push_proto;
7360 case BPF_FUNC_skb_vlan_pop:
7361 return &bpf_skb_vlan_pop_proto;
7362 case BPF_FUNC_skb_change_proto:
7363 return &bpf_skb_change_proto_proto;
7364 case BPF_FUNC_skb_change_type:
7365 return &bpf_skb_change_type_proto;
7366 case BPF_FUNC_skb_adjust_room:
7367 return &bpf_skb_adjust_room_proto;
7368 case BPF_FUNC_skb_change_tail:
7369 return &bpf_skb_change_tail_proto;
7370 case BPF_FUNC_skb_change_head:
7371 return &bpf_skb_change_head_proto;
7372 case BPF_FUNC_skb_get_tunnel_key:
7373 return &bpf_skb_get_tunnel_key_proto;
7374 case BPF_FUNC_skb_set_tunnel_key:
7375 return bpf_get_skb_set_tunnel_proto(func_id);
7376 case BPF_FUNC_skb_get_tunnel_opt:
7377 return &bpf_skb_get_tunnel_opt_proto;
7378 case BPF_FUNC_skb_set_tunnel_opt:
7379 return bpf_get_skb_set_tunnel_proto(func_id);
7380 case BPF_FUNC_redirect:
7381 return &bpf_redirect_proto;
7382 case BPF_FUNC_redirect_neigh:
7383 return &bpf_redirect_neigh_proto;
7384 case BPF_FUNC_redirect_peer:
7385 return &bpf_redirect_peer_proto;
7386 case BPF_FUNC_get_route_realm:
7387 return &bpf_get_route_realm_proto;
7388 case BPF_FUNC_get_hash_recalc:
7389 return &bpf_get_hash_recalc_proto;
7390 case BPF_FUNC_set_hash_invalid:
7391 return &bpf_set_hash_invalid_proto;
7392 case BPF_FUNC_set_hash:
7393 return &bpf_set_hash_proto;
7394 case BPF_FUNC_perf_event_output:
7395 return &bpf_skb_event_output_proto;
7396 case BPF_FUNC_get_smp_processor_id:
7397 return &bpf_get_smp_processor_id_proto;
7398 case BPF_FUNC_skb_under_cgroup:
7399 return &bpf_skb_under_cgroup_proto;
7400 case BPF_FUNC_get_socket_cookie:
7401 return &bpf_get_socket_cookie_proto;
7402 case BPF_FUNC_get_socket_uid:
7403 return &bpf_get_socket_uid_proto;
7404 case BPF_FUNC_fib_lookup:
7405 return &bpf_skb_fib_lookup_proto;
7406 case BPF_FUNC_check_mtu:
7407 return &bpf_skb_check_mtu_proto;
7408 case BPF_FUNC_sk_fullsock:
7409 return &bpf_sk_fullsock_proto;
7410 case BPF_FUNC_sk_storage_get:
7411 return &bpf_sk_storage_get_proto;
7412 case BPF_FUNC_sk_storage_delete:
7413 return &bpf_sk_storage_delete_proto;
7415 case BPF_FUNC_skb_get_xfrm_state:
7416 return &bpf_skb_get_xfrm_state_proto;
7418 #ifdef CONFIG_CGROUP_NET_CLASSID
7419 case BPF_FUNC_skb_cgroup_classid:
7420 return &bpf_skb_cgroup_classid_proto;
7422 #ifdef CONFIG_SOCK_CGROUP_DATA
7423 case BPF_FUNC_skb_cgroup_id:
7424 return &bpf_skb_cgroup_id_proto;
7425 case BPF_FUNC_skb_ancestor_cgroup_id:
7426 return &bpf_skb_ancestor_cgroup_id_proto;
7429 case BPF_FUNC_sk_lookup_tcp:
7430 return &bpf_sk_lookup_tcp_proto;
7431 case BPF_FUNC_sk_lookup_udp:
7432 return &bpf_sk_lookup_udp_proto;
7433 case BPF_FUNC_sk_release:
7434 return &bpf_sk_release_proto;
7435 case BPF_FUNC_tcp_sock:
7436 return &bpf_tcp_sock_proto;
7437 case BPF_FUNC_get_listener_sock:
7438 return &bpf_get_listener_sock_proto;
7439 case BPF_FUNC_skc_lookup_tcp:
7440 return &bpf_skc_lookup_tcp_proto;
7441 case BPF_FUNC_tcp_check_syncookie:
7442 return &bpf_tcp_check_syncookie_proto;
7443 case BPF_FUNC_skb_ecn_set_ce:
7444 return &bpf_skb_ecn_set_ce_proto;
7445 case BPF_FUNC_tcp_gen_syncookie:
7446 return &bpf_tcp_gen_syncookie_proto;
7447 case BPF_FUNC_sk_assign:
7448 return &bpf_sk_assign_proto;
7451 return bpf_sk_base_func_proto(func_id);
7455 static const struct bpf_func_proto *
7456 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7459 case BPF_FUNC_perf_event_output:
7460 return &bpf_xdp_event_output_proto;
7461 case BPF_FUNC_get_smp_processor_id:
7462 return &bpf_get_smp_processor_id_proto;
7463 case BPF_FUNC_csum_diff:
7464 return &bpf_csum_diff_proto;
7465 case BPF_FUNC_xdp_adjust_head:
7466 return &bpf_xdp_adjust_head_proto;
7467 case BPF_FUNC_xdp_adjust_meta:
7468 return &bpf_xdp_adjust_meta_proto;
7469 case BPF_FUNC_redirect:
7470 return &bpf_xdp_redirect_proto;
7471 case BPF_FUNC_redirect_map:
7472 return &bpf_xdp_redirect_map_proto;
7473 case BPF_FUNC_xdp_adjust_tail:
7474 return &bpf_xdp_adjust_tail_proto;
7475 case BPF_FUNC_fib_lookup:
7476 return &bpf_xdp_fib_lookup_proto;
7477 case BPF_FUNC_check_mtu:
7478 return &bpf_xdp_check_mtu_proto;
7480 case BPF_FUNC_sk_lookup_udp:
7481 return &bpf_xdp_sk_lookup_udp_proto;
7482 case BPF_FUNC_sk_lookup_tcp:
7483 return &bpf_xdp_sk_lookup_tcp_proto;
7484 case BPF_FUNC_sk_release:
7485 return &bpf_sk_release_proto;
7486 case BPF_FUNC_skc_lookup_tcp:
7487 return &bpf_xdp_skc_lookup_tcp_proto;
7488 case BPF_FUNC_tcp_check_syncookie:
7489 return &bpf_tcp_check_syncookie_proto;
7490 case BPF_FUNC_tcp_gen_syncookie:
7491 return &bpf_tcp_gen_syncookie_proto;
7494 return bpf_sk_base_func_proto(func_id);
7498 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
7499 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
7501 static const struct bpf_func_proto *
7502 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7505 case BPF_FUNC_setsockopt:
7506 return &bpf_sock_ops_setsockopt_proto;
7507 case BPF_FUNC_getsockopt:
7508 return &bpf_sock_ops_getsockopt_proto;
7509 case BPF_FUNC_sock_ops_cb_flags_set:
7510 return &bpf_sock_ops_cb_flags_set_proto;
7511 case BPF_FUNC_sock_map_update:
7512 return &bpf_sock_map_update_proto;
7513 case BPF_FUNC_sock_hash_update:
7514 return &bpf_sock_hash_update_proto;
7515 case BPF_FUNC_get_socket_cookie:
7516 return &bpf_get_socket_cookie_sock_ops_proto;
7517 case BPF_FUNC_get_local_storage:
7518 return &bpf_get_local_storage_proto;
7519 case BPF_FUNC_perf_event_output:
7520 return &bpf_event_output_data_proto;
7521 case BPF_FUNC_sk_storage_get:
7522 return &bpf_sk_storage_get_proto;
7523 case BPF_FUNC_sk_storage_delete:
7524 return &bpf_sk_storage_delete_proto;
7525 case BPF_FUNC_get_netns_cookie:
7526 return &bpf_get_netns_cookie_sock_ops_proto;
7528 case BPF_FUNC_load_hdr_opt:
7529 return &bpf_sock_ops_load_hdr_opt_proto;
7530 case BPF_FUNC_store_hdr_opt:
7531 return &bpf_sock_ops_store_hdr_opt_proto;
7532 case BPF_FUNC_reserve_hdr_opt:
7533 return &bpf_sock_ops_reserve_hdr_opt_proto;
7534 case BPF_FUNC_tcp_sock:
7535 return &bpf_tcp_sock_proto;
7536 #endif /* CONFIG_INET */
7538 return bpf_sk_base_func_proto(func_id);
7542 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
7543 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
7545 static const struct bpf_func_proto *
7546 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7549 case BPF_FUNC_msg_redirect_map:
7550 return &bpf_msg_redirect_map_proto;
7551 case BPF_FUNC_msg_redirect_hash:
7552 return &bpf_msg_redirect_hash_proto;
7553 case BPF_FUNC_msg_apply_bytes:
7554 return &bpf_msg_apply_bytes_proto;
7555 case BPF_FUNC_msg_cork_bytes:
7556 return &bpf_msg_cork_bytes_proto;
7557 case BPF_FUNC_msg_pull_data:
7558 return &bpf_msg_pull_data_proto;
7559 case BPF_FUNC_msg_push_data:
7560 return &bpf_msg_push_data_proto;
7561 case BPF_FUNC_msg_pop_data:
7562 return &bpf_msg_pop_data_proto;
7563 case BPF_FUNC_perf_event_output:
7564 return &bpf_event_output_data_proto;
7565 case BPF_FUNC_get_current_uid_gid:
7566 return &bpf_get_current_uid_gid_proto;
7567 case BPF_FUNC_get_current_pid_tgid:
7568 return &bpf_get_current_pid_tgid_proto;
7569 case BPF_FUNC_sk_storage_get:
7570 return &bpf_sk_storage_get_proto;
7571 case BPF_FUNC_sk_storage_delete:
7572 return &bpf_sk_storage_delete_proto;
7573 case BPF_FUNC_get_netns_cookie:
7574 return &bpf_get_netns_cookie_sk_msg_proto;
7575 #ifdef CONFIG_CGROUPS
7576 case BPF_FUNC_get_current_cgroup_id:
7577 return &bpf_get_current_cgroup_id_proto;
7578 case BPF_FUNC_get_current_ancestor_cgroup_id:
7579 return &bpf_get_current_ancestor_cgroup_id_proto;
7581 #ifdef CONFIG_CGROUP_NET_CLASSID
7582 case BPF_FUNC_get_cgroup_classid:
7583 return &bpf_get_cgroup_classid_curr_proto;
7586 return bpf_sk_base_func_proto(func_id);
7590 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
7591 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
7593 static const struct bpf_func_proto *
7594 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7597 case BPF_FUNC_skb_store_bytes:
7598 return &bpf_skb_store_bytes_proto;
7599 case BPF_FUNC_skb_load_bytes:
7600 return &bpf_skb_load_bytes_proto;
7601 case BPF_FUNC_skb_pull_data:
7602 return &sk_skb_pull_data_proto;
7603 case BPF_FUNC_skb_change_tail:
7604 return &sk_skb_change_tail_proto;
7605 case BPF_FUNC_skb_change_head:
7606 return &sk_skb_change_head_proto;
7607 case BPF_FUNC_skb_adjust_room:
7608 return &sk_skb_adjust_room_proto;
7609 case BPF_FUNC_get_socket_cookie:
7610 return &bpf_get_socket_cookie_proto;
7611 case BPF_FUNC_get_socket_uid:
7612 return &bpf_get_socket_uid_proto;
7613 case BPF_FUNC_sk_redirect_map:
7614 return &bpf_sk_redirect_map_proto;
7615 case BPF_FUNC_sk_redirect_hash:
7616 return &bpf_sk_redirect_hash_proto;
7617 case BPF_FUNC_perf_event_output:
7618 return &bpf_skb_event_output_proto;
7620 case BPF_FUNC_sk_lookup_tcp:
7621 return &bpf_sk_lookup_tcp_proto;
7622 case BPF_FUNC_sk_lookup_udp:
7623 return &bpf_sk_lookup_udp_proto;
7624 case BPF_FUNC_sk_release:
7625 return &bpf_sk_release_proto;
7626 case BPF_FUNC_skc_lookup_tcp:
7627 return &bpf_skc_lookup_tcp_proto;
7630 return bpf_sk_base_func_proto(func_id);
7634 static const struct bpf_func_proto *
7635 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7638 case BPF_FUNC_skb_load_bytes:
7639 return &bpf_flow_dissector_load_bytes_proto;
7641 return bpf_sk_base_func_proto(func_id);
7645 static const struct bpf_func_proto *
7646 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7649 case BPF_FUNC_skb_load_bytes:
7650 return &bpf_skb_load_bytes_proto;
7651 case BPF_FUNC_skb_pull_data:
7652 return &bpf_skb_pull_data_proto;
7653 case BPF_FUNC_csum_diff:
7654 return &bpf_csum_diff_proto;
7655 case BPF_FUNC_get_cgroup_classid:
7656 return &bpf_get_cgroup_classid_proto;
7657 case BPF_FUNC_get_route_realm:
7658 return &bpf_get_route_realm_proto;
7659 case BPF_FUNC_get_hash_recalc:
7660 return &bpf_get_hash_recalc_proto;
7661 case BPF_FUNC_perf_event_output:
7662 return &bpf_skb_event_output_proto;
7663 case BPF_FUNC_get_smp_processor_id:
7664 return &bpf_get_smp_processor_id_proto;
7665 case BPF_FUNC_skb_under_cgroup:
7666 return &bpf_skb_under_cgroup_proto;
7668 return bpf_sk_base_func_proto(func_id);
7672 static const struct bpf_func_proto *
7673 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7676 case BPF_FUNC_lwt_push_encap:
7677 return &bpf_lwt_in_push_encap_proto;
7679 return lwt_out_func_proto(func_id, prog);
7683 static const struct bpf_func_proto *
7684 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7687 case BPF_FUNC_skb_get_tunnel_key:
7688 return &bpf_skb_get_tunnel_key_proto;
7689 case BPF_FUNC_skb_set_tunnel_key:
7690 return bpf_get_skb_set_tunnel_proto(func_id);
7691 case BPF_FUNC_skb_get_tunnel_opt:
7692 return &bpf_skb_get_tunnel_opt_proto;
7693 case BPF_FUNC_skb_set_tunnel_opt:
7694 return bpf_get_skb_set_tunnel_proto(func_id);
7695 case BPF_FUNC_redirect:
7696 return &bpf_redirect_proto;
7697 case BPF_FUNC_clone_redirect:
7698 return &bpf_clone_redirect_proto;
7699 case BPF_FUNC_skb_change_tail:
7700 return &bpf_skb_change_tail_proto;
7701 case BPF_FUNC_skb_change_head:
7702 return &bpf_skb_change_head_proto;
7703 case BPF_FUNC_skb_store_bytes:
7704 return &bpf_skb_store_bytes_proto;
7705 case BPF_FUNC_csum_update:
7706 return &bpf_csum_update_proto;
7707 case BPF_FUNC_csum_level:
7708 return &bpf_csum_level_proto;
7709 case BPF_FUNC_l3_csum_replace:
7710 return &bpf_l3_csum_replace_proto;
7711 case BPF_FUNC_l4_csum_replace:
7712 return &bpf_l4_csum_replace_proto;
7713 case BPF_FUNC_set_hash_invalid:
7714 return &bpf_set_hash_invalid_proto;
7715 case BPF_FUNC_lwt_push_encap:
7716 return &bpf_lwt_xmit_push_encap_proto;
7718 return lwt_out_func_proto(func_id, prog);
7722 static const struct bpf_func_proto *
7723 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7726 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7727 case BPF_FUNC_lwt_seg6_store_bytes:
7728 return &bpf_lwt_seg6_store_bytes_proto;
7729 case BPF_FUNC_lwt_seg6_action:
7730 return &bpf_lwt_seg6_action_proto;
7731 case BPF_FUNC_lwt_seg6_adjust_srh:
7732 return &bpf_lwt_seg6_adjust_srh_proto;
7735 return lwt_out_func_proto(func_id, prog);
7739 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
7740 const struct bpf_prog *prog,
7741 struct bpf_insn_access_aux *info)
7743 const int size_default = sizeof(__u32);
7745 if (off < 0 || off >= sizeof(struct __sk_buff))
7748 /* The verifier guarantees that size > 0. */
7749 if (off % size != 0)
7753 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7754 if (off + size > offsetofend(struct __sk_buff, cb[4]))
7757 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
7758 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
7759 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
7760 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
7761 case bpf_ctx_range(struct __sk_buff, data):
7762 case bpf_ctx_range(struct __sk_buff, data_meta):
7763 case bpf_ctx_range(struct __sk_buff, data_end):
7764 if (size != size_default)
7767 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
7769 case bpf_ctx_range(struct __sk_buff, hwtstamp):
7770 if (type == BPF_WRITE || size != sizeof(__u64))
7773 case bpf_ctx_range(struct __sk_buff, tstamp):
7774 if (size != sizeof(__u64))
7777 case offsetof(struct __sk_buff, sk):
7778 if (type == BPF_WRITE || size != sizeof(__u64))
7780 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
7782 case offsetofend(struct __sk_buff, gso_size) ... offsetof(struct __sk_buff, hwtstamp) - 1:
7783 /* Explicitly prohibit access to padding in __sk_buff. */
7786 /* Only narrow read access allowed for now. */
7787 if (type == BPF_WRITE) {
7788 if (size != size_default)
7791 bpf_ctx_record_field_size(info, size_default);
7792 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
7800 static bool sk_filter_is_valid_access(int off, int size,
7801 enum bpf_access_type type,
7802 const struct bpf_prog *prog,
7803 struct bpf_insn_access_aux *info)
7806 case bpf_ctx_range(struct __sk_buff, tc_classid):
7807 case bpf_ctx_range(struct __sk_buff, data):
7808 case bpf_ctx_range(struct __sk_buff, data_meta):
7809 case bpf_ctx_range(struct __sk_buff, data_end):
7810 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7811 case bpf_ctx_range(struct __sk_buff, tstamp):
7812 case bpf_ctx_range(struct __sk_buff, wire_len):
7813 case bpf_ctx_range(struct __sk_buff, hwtstamp):
7817 if (type == BPF_WRITE) {
7819 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7826 return bpf_skb_is_valid_access(off, size, type, prog, info);
7829 static bool cg_skb_is_valid_access(int off, int size,
7830 enum bpf_access_type type,
7831 const struct bpf_prog *prog,
7832 struct bpf_insn_access_aux *info)
7835 case bpf_ctx_range(struct __sk_buff, tc_classid):
7836 case bpf_ctx_range(struct __sk_buff, data_meta):
7837 case bpf_ctx_range(struct __sk_buff, wire_len):
7839 case bpf_ctx_range(struct __sk_buff, data):
7840 case bpf_ctx_range(struct __sk_buff, data_end):
7846 if (type == BPF_WRITE) {
7848 case bpf_ctx_range(struct __sk_buff, mark):
7849 case bpf_ctx_range(struct __sk_buff, priority):
7850 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7852 case bpf_ctx_range(struct __sk_buff, tstamp):
7862 case bpf_ctx_range(struct __sk_buff, data):
7863 info->reg_type = PTR_TO_PACKET;
7865 case bpf_ctx_range(struct __sk_buff, data_end):
7866 info->reg_type = PTR_TO_PACKET_END;
7870 return bpf_skb_is_valid_access(off, size, type, prog, info);
7873 static bool lwt_is_valid_access(int off, int size,
7874 enum bpf_access_type type,
7875 const struct bpf_prog *prog,
7876 struct bpf_insn_access_aux *info)
7879 case bpf_ctx_range(struct __sk_buff, tc_classid):
7880 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7881 case bpf_ctx_range(struct __sk_buff, data_meta):
7882 case bpf_ctx_range(struct __sk_buff, tstamp):
7883 case bpf_ctx_range(struct __sk_buff, wire_len):
7884 case bpf_ctx_range(struct __sk_buff, hwtstamp):
7888 if (type == BPF_WRITE) {
7890 case bpf_ctx_range(struct __sk_buff, mark):
7891 case bpf_ctx_range(struct __sk_buff, priority):
7892 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7900 case bpf_ctx_range(struct __sk_buff, data):
7901 info->reg_type = PTR_TO_PACKET;
7903 case bpf_ctx_range(struct __sk_buff, data_end):
7904 info->reg_type = PTR_TO_PACKET_END;
7908 return bpf_skb_is_valid_access(off, size, type, prog, info);
7911 /* Attach type specific accesses */
7912 static bool __sock_filter_check_attach_type(int off,
7913 enum bpf_access_type access_type,
7914 enum bpf_attach_type attach_type)
7917 case offsetof(struct bpf_sock, bound_dev_if):
7918 case offsetof(struct bpf_sock, mark):
7919 case offsetof(struct bpf_sock, priority):
7920 switch (attach_type) {
7921 case BPF_CGROUP_INET_SOCK_CREATE:
7922 case BPF_CGROUP_INET_SOCK_RELEASE:
7927 case bpf_ctx_range(struct bpf_sock, src_ip4):
7928 switch (attach_type) {
7929 case BPF_CGROUP_INET4_POST_BIND:
7934 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7935 switch (attach_type) {
7936 case BPF_CGROUP_INET6_POST_BIND:
7941 case bpf_ctx_range(struct bpf_sock, src_port):
7942 switch (attach_type) {
7943 case BPF_CGROUP_INET4_POST_BIND:
7944 case BPF_CGROUP_INET6_POST_BIND:
7951 return access_type == BPF_READ;
7956 bool bpf_sock_common_is_valid_access(int off, int size,
7957 enum bpf_access_type type,
7958 struct bpf_insn_access_aux *info)
7961 case bpf_ctx_range_till(struct bpf_sock, type, priority):
7964 return bpf_sock_is_valid_access(off, size, type, info);
7968 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7969 struct bpf_insn_access_aux *info)
7971 const int size_default = sizeof(__u32);
7973 if (off < 0 || off >= sizeof(struct bpf_sock))
7975 if (off % size != 0)
7979 case offsetof(struct bpf_sock, state):
7980 case offsetof(struct bpf_sock, family):
7981 case offsetof(struct bpf_sock, type):
7982 case offsetof(struct bpf_sock, protocol):
7983 case offsetof(struct bpf_sock, dst_port):
7984 case offsetof(struct bpf_sock, src_port):
7985 case offsetof(struct bpf_sock, rx_queue_mapping):
7986 case bpf_ctx_range(struct bpf_sock, src_ip4):
7987 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7988 case bpf_ctx_range(struct bpf_sock, dst_ip4):
7989 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
7990 bpf_ctx_record_field_size(info, size_default);
7991 return bpf_ctx_narrow_access_ok(off, size, size_default);
7994 return size == size_default;
7997 static bool sock_filter_is_valid_access(int off, int size,
7998 enum bpf_access_type type,
7999 const struct bpf_prog *prog,
8000 struct bpf_insn_access_aux *info)
8002 if (!bpf_sock_is_valid_access(off, size, type, info))
8004 return __sock_filter_check_attach_type(off, type,
8005 prog->expected_attach_type);
8008 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
8009 const struct bpf_prog *prog)
8011 /* Neither direct read nor direct write requires any preliminary
8017 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
8018 const struct bpf_prog *prog, int drop_verdict)
8020 struct bpf_insn *insn = insn_buf;
8025 /* if (!skb->cloned)
8028 * (Fast-path, otherwise approximation that we might be
8029 * a clone, do the rest in helper.)
8031 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET);
8032 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
8033 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
8035 /* ret = bpf_skb_pull_data(skb, 0); */
8036 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
8037 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
8038 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
8039 BPF_FUNC_skb_pull_data);
8042 * return TC_ACT_SHOT;
8044 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
8045 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
8046 *insn++ = BPF_EXIT_INSN();
8049 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
8051 *insn++ = prog->insnsi[0];
8053 return insn - insn_buf;
8056 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
8057 struct bpf_insn *insn_buf)
8059 bool indirect = BPF_MODE(orig->code) == BPF_IND;
8060 struct bpf_insn *insn = insn_buf;
8063 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
8065 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
8067 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
8069 /* We're guaranteed here that CTX is in R6. */
8070 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
8072 switch (BPF_SIZE(orig->code)) {
8074 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
8077 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
8080 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
8084 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
8085 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
8086 *insn++ = BPF_EXIT_INSN();
8088 return insn - insn_buf;
8091 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8092 const struct bpf_prog *prog)
8094 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8097 static bool tc_cls_act_is_valid_access(int off, int size,
8098 enum bpf_access_type type,
8099 const struct bpf_prog *prog,
8100 struct bpf_insn_access_aux *info)
8102 if (type == BPF_WRITE) {
8104 case bpf_ctx_range(struct __sk_buff, mark):
8105 case bpf_ctx_range(struct __sk_buff, tc_index):
8106 case bpf_ctx_range(struct __sk_buff, priority):
8107 case bpf_ctx_range(struct __sk_buff, tc_classid):
8108 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8109 case bpf_ctx_range(struct __sk_buff, tstamp):
8110 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8118 case bpf_ctx_range(struct __sk_buff, data):
8119 info->reg_type = PTR_TO_PACKET;
8121 case bpf_ctx_range(struct __sk_buff, data_meta):
8122 info->reg_type = PTR_TO_PACKET_META;
8124 case bpf_ctx_range(struct __sk_buff, data_end):
8125 info->reg_type = PTR_TO_PACKET_END;
8127 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8131 return bpf_skb_is_valid_access(off, size, type, prog, info);
8134 static bool __is_valid_xdp_access(int off, int size)
8136 if (off < 0 || off >= sizeof(struct xdp_md))
8138 if (off % size != 0)
8140 if (size != sizeof(__u32))
8146 static bool xdp_is_valid_access(int off, int size,
8147 enum bpf_access_type type,
8148 const struct bpf_prog *prog,
8149 struct bpf_insn_access_aux *info)
8151 if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8153 case offsetof(struct xdp_md, egress_ifindex):
8158 if (type == BPF_WRITE) {
8159 if (bpf_prog_is_dev_bound(prog->aux)) {
8161 case offsetof(struct xdp_md, rx_queue_index):
8162 return __is_valid_xdp_access(off, size);
8169 case offsetof(struct xdp_md, data):
8170 info->reg_type = PTR_TO_PACKET;
8172 case offsetof(struct xdp_md, data_meta):
8173 info->reg_type = PTR_TO_PACKET_META;
8175 case offsetof(struct xdp_md, data_end):
8176 info->reg_type = PTR_TO_PACKET_END;
8180 return __is_valid_xdp_access(off, size);
8183 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act)
8185 const u32 act_max = XDP_REDIRECT;
8187 pr_warn_once("%s XDP return value %u on prog %s (id %d) dev %s, expect packet loss!\n",
8188 act > act_max ? "Illegal" : "Driver unsupported",
8189 act, prog->aux->name, prog->aux->id, dev ? dev->name : "N/A");
8191 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
8193 static bool sock_addr_is_valid_access(int off, int size,
8194 enum bpf_access_type type,
8195 const struct bpf_prog *prog,
8196 struct bpf_insn_access_aux *info)
8198 const int size_default = sizeof(__u32);
8200 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
8202 if (off % size != 0)
8205 /* Disallow access to IPv6 fields from IPv4 contex and vise
8209 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8210 switch (prog->expected_attach_type) {
8211 case BPF_CGROUP_INET4_BIND:
8212 case BPF_CGROUP_INET4_CONNECT:
8213 case BPF_CGROUP_INET4_GETPEERNAME:
8214 case BPF_CGROUP_INET4_GETSOCKNAME:
8215 case BPF_CGROUP_UDP4_SENDMSG:
8216 case BPF_CGROUP_UDP4_RECVMSG:
8222 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8223 switch (prog->expected_attach_type) {
8224 case BPF_CGROUP_INET6_BIND:
8225 case BPF_CGROUP_INET6_CONNECT:
8226 case BPF_CGROUP_INET6_GETPEERNAME:
8227 case BPF_CGROUP_INET6_GETSOCKNAME:
8228 case BPF_CGROUP_UDP6_SENDMSG:
8229 case BPF_CGROUP_UDP6_RECVMSG:
8235 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8236 switch (prog->expected_attach_type) {
8237 case BPF_CGROUP_UDP4_SENDMSG:
8243 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8245 switch (prog->expected_attach_type) {
8246 case BPF_CGROUP_UDP6_SENDMSG:
8255 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8256 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8257 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8258 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8260 case bpf_ctx_range(struct bpf_sock_addr, user_port):
8261 if (type == BPF_READ) {
8262 bpf_ctx_record_field_size(info, size_default);
8264 if (bpf_ctx_wide_access_ok(off, size,
8265 struct bpf_sock_addr,
8269 if (bpf_ctx_wide_access_ok(off, size,
8270 struct bpf_sock_addr,
8274 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8277 if (bpf_ctx_wide_access_ok(off, size,
8278 struct bpf_sock_addr,
8282 if (bpf_ctx_wide_access_ok(off, size,
8283 struct bpf_sock_addr,
8287 if (size != size_default)
8291 case offsetof(struct bpf_sock_addr, sk):
8292 if (type != BPF_READ)
8294 if (size != sizeof(__u64))
8296 info->reg_type = PTR_TO_SOCKET;
8299 if (type == BPF_READ) {
8300 if (size != size_default)
8310 static bool sock_ops_is_valid_access(int off, int size,
8311 enum bpf_access_type type,
8312 const struct bpf_prog *prog,
8313 struct bpf_insn_access_aux *info)
8315 const int size_default = sizeof(__u32);
8317 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
8320 /* The verifier guarantees that size > 0. */
8321 if (off % size != 0)
8324 if (type == BPF_WRITE) {
8326 case offsetof(struct bpf_sock_ops, reply):
8327 case offsetof(struct bpf_sock_ops, sk_txhash):
8328 if (size != size_default)
8336 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
8338 if (size != sizeof(__u64))
8341 case offsetof(struct bpf_sock_ops, sk):
8342 if (size != sizeof(__u64))
8344 info->reg_type = PTR_TO_SOCKET_OR_NULL;
8346 case offsetof(struct bpf_sock_ops, skb_data):
8347 if (size != sizeof(__u64))
8349 info->reg_type = PTR_TO_PACKET;
8351 case offsetof(struct bpf_sock_ops, skb_data_end):
8352 if (size != sizeof(__u64))
8354 info->reg_type = PTR_TO_PACKET_END;
8356 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
8357 bpf_ctx_record_field_size(info, size_default);
8358 return bpf_ctx_narrow_access_ok(off, size,
8361 if (size != size_default)
8370 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
8371 const struct bpf_prog *prog)
8373 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
8376 static bool sk_skb_is_valid_access(int off, int size,
8377 enum bpf_access_type type,
8378 const struct bpf_prog *prog,
8379 struct bpf_insn_access_aux *info)
8382 case bpf_ctx_range(struct __sk_buff, tc_classid):
8383 case bpf_ctx_range(struct __sk_buff, data_meta):
8384 case bpf_ctx_range(struct __sk_buff, tstamp):
8385 case bpf_ctx_range(struct __sk_buff, wire_len):
8386 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8390 if (type == BPF_WRITE) {
8392 case bpf_ctx_range(struct __sk_buff, tc_index):
8393 case bpf_ctx_range(struct __sk_buff, priority):
8401 case bpf_ctx_range(struct __sk_buff, mark):
8403 case bpf_ctx_range(struct __sk_buff, data):
8404 info->reg_type = PTR_TO_PACKET;
8406 case bpf_ctx_range(struct __sk_buff, data_end):
8407 info->reg_type = PTR_TO_PACKET_END;
8411 return bpf_skb_is_valid_access(off, size, type, prog, info);
8414 static bool sk_msg_is_valid_access(int off, int size,
8415 enum bpf_access_type type,
8416 const struct bpf_prog *prog,
8417 struct bpf_insn_access_aux *info)
8419 if (type == BPF_WRITE)
8422 if (off % size != 0)
8426 case offsetof(struct sk_msg_md, data):
8427 info->reg_type = PTR_TO_PACKET;
8428 if (size != sizeof(__u64))
8431 case offsetof(struct sk_msg_md, data_end):
8432 info->reg_type = PTR_TO_PACKET_END;
8433 if (size != sizeof(__u64))
8436 case offsetof(struct sk_msg_md, sk):
8437 if (size != sizeof(__u64))
8439 info->reg_type = PTR_TO_SOCKET;
8441 case bpf_ctx_range(struct sk_msg_md, family):
8442 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
8443 case bpf_ctx_range(struct sk_msg_md, local_ip4):
8444 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
8445 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
8446 case bpf_ctx_range(struct sk_msg_md, remote_port):
8447 case bpf_ctx_range(struct sk_msg_md, local_port):
8448 case bpf_ctx_range(struct sk_msg_md, size):
8449 if (size != sizeof(__u32))
8458 static bool flow_dissector_is_valid_access(int off, int size,
8459 enum bpf_access_type type,
8460 const struct bpf_prog *prog,
8461 struct bpf_insn_access_aux *info)
8463 const int size_default = sizeof(__u32);
8465 if (off < 0 || off >= sizeof(struct __sk_buff))
8468 if (type == BPF_WRITE)
8472 case bpf_ctx_range(struct __sk_buff, data):
8473 if (size != size_default)
8475 info->reg_type = PTR_TO_PACKET;
8477 case bpf_ctx_range(struct __sk_buff, data_end):
8478 if (size != size_default)
8480 info->reg_type = PTR_TO_PACKET_END;
8482 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8483 if (size != sizeof(__u64))
8485 info->reg_type = PTR_TO_FLOW_KEYS;
8492 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
8493 const struct bpf_insn *si,
8494 struct bpf_insn *insn_buf,
8495 struct bpf_prog *prog,
8499 struct bpf_insn *insn = insn_buf;
8502 case offsetof(struct __sk_buff, data):
8503 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
8504 si->dst_reg, si->src_reg,
8505 offsetof(struct bpf_flow_dissector, data));
8508 case offsetof(struct __sk_buff, data_end):
8509 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
8510 si->dst_reg, si->src_reg,
8511 offsetof(struct bpf_flow_dissector, data_end));
8514 case offsetof(struct __sk_buff, flow_keys):
8515 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
8516 si->dst_reg, si->src_reg,
8517 offsetof(struct bpf_flow_dissector, flow_keys));
8521 return insn - insn_buf;
8524 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si,
8525 struct bpf_insn *insn)
8527 /* si->dst_reg = skb_shinfo(SKB); */
8528 #ifdef NET_SKBUFF_DATA_USES_OFFSET
8529 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8530 BPF_REG_AX, si->src_reg,
8531 offsetof(struct sk_buff, end));
8532 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
8533 si->dst_reg, si->src_reg,
8534 offsetof(struct sk_buff, head));
8535 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
8537 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8538 si->dst_reg, si->src_reg,
8539 offsetof(struct sk_buff, end));
8545 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
8546 const struct bpf_insn *si,
8547 struct bpf_insn *insn_buf,
8548 struct bpf_prog *prog, u32 *target_size)
8550 struct bpf_insn *insn = insn_buf;
8554 case offsetof(struct __sk_buff, len):
8555 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8556 bpf_target_off(struct sk_buff, len, 4,
8560 case offsetof(struct __sk_buff, protocol):
8561 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8562 bpf_target_off(struct sk_buff, protocol, 2,
8566 case offsetof(struct __sk_buff, vlan_proto):
8567 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8568 bpf_target_off(struct sk_buff, vlan_proto, 2,
8572 case offsetof(struct __sk_buff, priority):
8573 if (type == BPF_WRITE)
8574 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8575 bpf_target_off(struct sk_buff, priority, 4,
8578 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8579 bpf_target_off(struct sk_buff, priority, 4,
8583 case offsetof(struct __sk_buff, ingress_ifindex):
8584 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8585 bpf_target_off(struct sk_buff, skb_iif, 4,
8589 case offsetof(struct __sk_buff, ifindex):
8590 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8591 si->dst_reg, si->src_reg,
8592 offsetof(struct sk_buff, dev));
8593 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
8594 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8595 bpf_target_off(struct net_device, ifindex, 4,
8599 case offsetof(struct __sk_buff, hash):
8600 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8601 bpf_target_off(struct sk_buff, hash, 4,
8605 case offsetof(struct __sk_buff, mark):
8606 if (type == BPF_WRITE)
8607 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8608 bpf_target_off(struct sk_buff, mark, 4,
8611 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8612 bpf_target_off(struct sk_buff, mark, 4,
8616 case offsetof(struct __sk_buff, pkt_type):
8618 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8620 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
8621 #ifdef __BIG_ENDIAN_BITFIELD
8622 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
8626 case offsetof(struct __sk_buff, queue_mapping):
8627 if (type == BPF_WRITE) {
8628 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
8629 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8630 bpf_target_off(struct sk_buff,
8634 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8635 bpf_target_off(struct sk_buff,
8641 case offsetof(struct __sk_buff, vlan_present):
8643 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8644 PKT_VLAN_PRESENT_OFFSET);
8645 if (PKT_VLAN_PRESENT_BIT)
8646 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
8647 if (PKT_VLAN_PRESENT_BIT < 7)
8648 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
8651 case offsetof(struct __sk_buff, vlan_tci):
8652 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8653 bpf_target_off(struct sk_buff, vlan_tci, 2,
8657 case offsetof(struct __sk_buff, cb[0]) ...
8658 offsetofend(struct __sk_buff, cb[4]) - 1:
8659 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
8660 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
8661 offsetof(struct qdisc_skb_cb, data)) %
8664 prog->cb_access = 1;
8666 off -= offsetof(struct __sk_buff, cb[0]);
8667 off += offsetof(struct sk_buff, cb);
8668 off += offsetof(struct qdisc_skb_cb, data);
8669 if (type == BPF_WRITE)
8670 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
8673 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
8677 case offsetof(struct __sk_buff, tc_classid):
8678 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
8681 off -= offsetof(struct __sk_buff, tc_classid);
8682 off += offsetof(struct sk_buff, cb);
8683 off += offsetof(struct qdisc_skb_cb, tc_classid);
8685 if (type == BPF_WRITE)
8686 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
8689 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
8693 case offsetof(struct __sk_buff, data):
8694 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
8695 si->dst_reg, si->src_reg,
8696 offsetof(struct sk_buff, data));
8699 case offsetof(struct __sk_buff, data_meta):
8701 off -= offsetof(struct __sk_buff, data_meta);
8702 off += offsetof(struct sk_buff, cb);
8703 off += offsetof(struct bpf_skb_data_end, data_meta);
8704 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8708 case offsetof(struct __sk_buff, data_end):
8710 off -= offsetof(struct __sk_buff, data_end);
8711 off += offsetof(struct sk_buff, cb);
8712 off += offsetof(struct bpf_skb_data_end, data_end);
8713 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8717 case offsetof(struct __sk_buff, tc_index):
8718 #ifdef CONFIG_NET_SCHED
8719 if (type == BPF_WRITE)
8720 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8721 bpf_target_off(struct sk_buff, tc_index, 2,
8724 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8725 bpf_target_off(struct sk_buff, tc_index, 2,
8729 if (type == BPF_WRITE)
8730 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
8732 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8736 case offsetof(struct __sk_buff, napi_id):
8737 #if defined(CONFIG_NET_RX_BUSY_POLL)
8738 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8739 bpf_target_off(struct sk_buff, napi_id, 4,
8741 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
8742 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8745 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8748 case offsetof(struct __sk_buff, family):
8749 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
8751 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8752 si->dst_reg, si->src_reg,
8753 offsetof(struct sk_buff, sk));
8754 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8755 bpf_target_off(struct sock_common,
8759 case offsetof(struct __sk_buff, remote_ip4):
8760 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
8762 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8763 si->dst_reg, si->src_reg,
8764 offsetof(struct sk_buff, sk));
8765 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8766 bpf_target_off(struct sock_common,
8770 case offsetof(struct __sk_buff, local_ip4):
8771 BUILD_BUG_ON(sizeof_field(struct sock_common,
8772 skc_rcv_saddr) != 4);
8774 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8775 si->dst_reg, si->src_reg,
8776 offsetof(struct sk_buff, sk));
8777 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8778 bpf_target_off(struct sock_common,
8782 case offsetof(struct __sk_buff, remote_ip6[0]) ...
8783 offsetof(struct __sk_buff, remote_ip6[3]):
8784 #if IS_ENABLED(CONFIG_IPV6)
8785 BUILD_BUG_ON(sizeof_field(struct sock_common,
8786 skc_v6_daddr.s6_addr32[0]) != 4);
8789 off -= offsetof(struct __sk_buff, remote_ip6[0]);
8791 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8792 si->dst_reg, si->src_reg,
8793 offsetof(struct sk_buff, sk));
8794 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8795 offsetof(struct sock_common,
8796 skc_v6_daddr.s6_addr32[0]) +
8799 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8802 case offsetof(struct __sk_buff, local_ip6[0]) ...
8803 offsetof(struct __sk_buff, local_ip6[3]):
8804 #if IS_ENABLED(CONFIG_IPV6)
8805 BUILD_BUG_ON(sizeof_field(struct sock_common,
8806 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8809 off -= offsetof(struct __sk_buff, local_ip6[0]);
8811 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8812 si->dst_reg, si->src_reg,
8813 offsetof(struct sk_buff, sk));
8814 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8815 offsetof(struct sock_common,
8816 skc_v6_rcv_saddr.s6_addr32[0]) +
8819 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8823 case offsetof(struct __sk_buff, remote_port):
8824 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
8826 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8827 si->dst_reg, si->src_reg,
8828 offsetof(struct sk_buff, sk));
8829 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8830 bpf_target_off(struct sock_common,
8833 #ifndef __BIG_ENDIAN_BITFIELD
8834 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8838 case offsetof(struct __sk_buff, local_port):
8839 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
8841 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8842 si->dst_reg, si->src_reg,
8843 offsetof(struct sk_buff, sk));
8844 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8845 bpf_target_off(struct sock_common,
8846 skc_num, 2, target_size));
8849 case offsetof(struct __sk_buff, tstamp):
8850 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
8852 if (type == BPF_WRITE)
8853 *insn++ = BPF_STX_MEM(BPF_DW,
8854 si->dst_reg, si->src_reg,
8855 bpf_target_off(struct sk_buff,
8859 *insn++ = BPF_LDX_MEM(BPF_DW,
8860 si->dst_reg, si->src_reg,
8861 bpf_target_off(struct sk_buff,
8866 case offsetof(struct __sk_buff, gso_segs):
8867 insn = bpf_convert_shinfo_access(si, insn);
8868 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
8869 si->dst_reg, si->dst_reg,
8870 bpf_target_off(struct skb_shared_info,
8874 case offsetof(struct __sk_buff, gso_size):
8875 insn = bpf_convert_shinfo_access(si, insn);
8876 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
8877 si->dst_reg, si->dst_reg,
8878 bpf_target_off(struct skb_shared_info,
8882 case offsetof(struct __sk_buff, wire_len):
8883 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
8886 off -= offsetof(struct __sk_buff, wire_len);
8887 off += offsetof(struct sk_buff, cb);
8888 off += offsetof(struct qdisc_skb_cb, pkt_len);
8890 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
8893 case offsetof(struct __sk_buff, sk):
8894 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8895 si->dst_reg, si->src_reg,
8896 offsetof(struct sk_buff, sk));
8898 case offsetof(struct __sk_buff, hwtstamp):
8899 BUILD_BUG_ON(sizeof_field(struct skb_shared_hwtstamps, hwtstamp) != 8);
8900 BUILD_BUG_ON(offsetof(struct skb_shared_hwtstamps, hwtstamp) != 0);
8902 insn = bpf_convert_shinfo_access(si, insn);
8903 *insn++ = BPF_LDX_MEM(BPF_DW,
8904 si->dst_reg, si->dst_reg,
8905 bpf_target_off(struct skb_shared_info,
8911 return insn - insn_buf;
8914 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
8915 const struct bpf_insn *si,
8916 struct bpf_insn *insn_buf,
8917 struct bpf_prog *prog, u32 *target_size)
8919 struct bpf_insn *insn = insn_buf;
8923 case offsetof(struct bpf_sock, bound_dev_if):
8924 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
8926 if (type == BPF_WRITE)
8927 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8928 offsetof(struct sock, sk_bound_dev_if));
8930 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8931 offsetof(struct sock, sk_bound_dev_if));
8934 case offsetof(struct bpf_sock, mark):
8935 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
8937 if (type == BPF_WRITE)
8938 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8939 offsetof(struct sock, sk_mark));
8941 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8942 offsetof(struct sock, sk_mark));
8945 case offsetof(struct bpf_sock, priority):
8946 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
8948 if (type == BPF_WRITE)
8949 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8950 offsetof(struct sock, sk_priority));
8952 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8953 offsetof(struct sock, sk_priority));
8956 case offsetof(struct bpf_sock, family):
8957 *insn++ = BPF_LDX_MEM(
8958 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
8959 si->dst_reg, si->src_reg,
8960 bpf_target_off(struct sock_common,
8962 sizeof_field(struct sock_common,
8967 case offsetof(struct bpf_sock, type):
8968 *insn++ = BPF_LDX_MEM(
8969 BPF_FIELD_SIZEOF(struct sock, sk_type),
8970 si->dst_reg, si->src_reg,
8971 bpf_target_off(struct sock, sk_type,
8972 sizeof_field(struct sock, sk_type),
8976 case offsetof(struct bpf_sock, protocol):
8977 *insn++ = BPF_LDX_MEM(
8978 BPF_FIELD_SIZEOF(struct sock, sk_protocol),
8979 si->dst_reg, si->src_reg,
8980 bpf_target_off(struct sock, sk_protocol,
8981 sizeof_field(struct sock, sk_protocol),
8985 case offsetof(struct bpf_sock, src_ip4):
8986 *insn++ = BPF_LDX_MEM(
8987 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8988 bpf_target_off(struct sock_common, skc_rcv_saddr,
8989 sizeof_field(struct sock_common,
8994 case offsetof(struct bpf_sock, dst_ip4):
8995 *insn++ = BPF_LDX_MEM(
8996 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8997 bpf_target_off(struct sock_common, skc_daddr,
8998 sizeof_field(struct sock_common,
9003 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
9004 #if IS_ENABLED(CONFIG_IPV6)
9006 off -= offsetof(struct bpf_sock, src_ip6[0]);
9007 *insn++ = BPF_LDX_MEM(
9008 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9011 skc_v6_rcv_saddr.s6_addr32[0],
9012 sizeof_field(struct sock_common,
9013 skc_v6_rcv_saddr.s6_addr32[0]),
9014 target_size) + off);
9017 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9021 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
9022 #if IS_ENABLED(CONFIG_IPV6)
9024 off -= offsetof(struct bpf_sock, dst_ip6[0]);
9025 *insn++ = BPF_LDX_MEM(
9026 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9027 bpf_target_off(struct sock_common,
9028 skc_v6_daddr.s6_addr32[0],
9029 sizeof_field(struct sock_common,
9030 skc_v6_daddr.s6_addr32[0]),
9031 target_size) + off);
9033 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9038 case offsetof(struct bpf_sock, src_port):
9039 *insn++ = BPF_LDX_MEM(
9040 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
9041 si->dst_reg, si->src_reg,
9042 bpf_target_off(struct sock_common, skc_num,
9043 sizeof_field(struct sock_common,
9048 case offsetof(struct bpf_sock, dst_port):
9049 *insn++ = BPF_LDX_MEM(
9050 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
9051 si->dst_reg, si->src_reg,
9052 bpf_target_off(struct sock_common, skc_dport,
9053 sizeof_field(struct sock_common,
9058 case offsetof(struct bpf_sock, state):
9059 *insn++ = BPF_LDX_MEM(
9060 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
9061 si->dst_reg, si->src_reg,
9062 bpf_target_off(struct sock_common, skc_state,
9063 sizeof_field(struct sock_common,
9067 case offsetof(struct bpf_sock, rx_queue_mapping):
9068 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
9069 *insn++ = BPF_LDX_MEM(
9070 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
9071 si->dst_reg, si->src_reg,
9072 bpf_target_off(struct sock, sk_rx_queue_mapping,
9073 sizeof_field(struct sock,
9074 sk_rx_queue_mapping),
9076 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
9078 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9080 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9086 return insn - insn_buf;
9089 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
9090 const struct bpf_insn *si,
9091 struct bpf_insn *insn_buf,
9092 struct bpf_prog *prog, u32 *target_size)
9094 struct bpf_insn *insn = insn_buf;
9097 case offsetof(struct __sk_buff, ifindex):
9098 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9099 si->dst_reg, si->src_reg,
9100 offsetof(struct sk_buff, dev));
9101 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9102 bpf_target_off(struct net_device, ifindex, 4,
9106 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9110 return insn - insn_buf;
9113 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
9114 const struct bpf_insn *si,
9115 struct bpf_insn *insn_buf,
9116 struct bpf_prog *prog, u32 *target_size)
9118 struct bpf_insn *insn = insn_buf;
9121 case offsetof(struct xdp_md, data):
9122 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
9123 si->dst_reg, si->src_reg,
9124 offsetof(struct xdp_buff, data));
9126 case offsetof(struct xdp_md, data_meta):
9127 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
9128 si->dst_reg, si->src_reg,
9129 offsetof(struct xdp_buff, data_meta));
9131 case offsetof(struct xdp_md, data_end):
9132 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
9133 si->dst_reg, si->src_reg,
9134 offsetof(struct xdp_buff, data_end));
9136 case offsetof(struct xdp_md, ingress_ifindex):
9137 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9138 si->dst_reg, si->src_reg,
9139 offsetof(struct xdp_buff, rxq));
9140 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
9141 si->dst_reg, si->dst_reg,
9142 offsetof(struct xdp_rxq_info, dev));
9143 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9144 offsetof(struct net_device, ifindex));
9146 case offsetof(struct xdp_md, rx_queue_index):
9147 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9148 si->dst_reg, si->src_reg,
9149 offsetof(struct xdp_buff, rxq));
9150 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9151 offsetof(struct xdp_rxq_info,
9154 case offsetof(struct xdp_md, egress_ifindex):
9155 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
9156 si->dst_reg, si->src_reg,
9157 offsetof(struct xdp_buff, txq));
9158 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
9159 si->dst_reg, si->dst_reg,
9160 offsetof(struct xdp_txq_info, dev));
9161 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9162 offsetof(struct net_device, ifindex));
9166 return insn - insn_buf;
9169 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
9170 * context Structure, F is Field in context structure that contains a pointer
9171 * to Nested Structure of type NS that has the field NF.
9173 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
9174 * sure that SIZE is not greater than actual size of S.F.NF.
9176 * If offset OFF is provided, the load happens from that offset relative to
9179 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
9181 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
9182 si->src_reg, offsetof(S, F)); \
9183 *insn++ = BPF_LDX_MEM( \
9184 SIZE, si->dst_reg, si->dst_reg, \
9185 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
9190 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
9191 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
9192 BPF_FIELD_SIZEOF(NS, NF), 0)
9194 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
9195 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
9197 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
9198 * "register" since two registers available in convert_ctx_access are not
9199 * enough: we can't override neither SRC, since it contains value to store, nor
9200 * DST since it contains pointer to context that may be used by later
9201 * instructions. But we need a temporary place to save pointer to nested
9202 * structure whose field we want to store to.
9204 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
9206 int tmp_reg = BPF_REG_9; \
9207 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
9209 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
9211 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
9213 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
9214 si->dst_reg, offsetof(S, F)); \
9215 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \
9216 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
9219 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
9223 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
9226 if (type == BPF_WRITE) { \
9227 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
9230 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
9231 S, NS, F, NF, SIZE, OFF); \
9235 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
9236 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
9237 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
9239 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
9240 const struct bpf_insn *si,
9241 struct bpf_insn *insn_buf,
9242 struct bpf_prog *prog, u32 *target_size)
9244 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
9245 struct bpf_insn *insn = insn_buf;
9248 case offsetof(struct bpf_sock_addr, user_family):
9249 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9250 struct sockaddr, uaddr, sa_family);
9253 case offsetof(struct bpf_sock_addr, user_ip4):
9254 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9255 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
9256 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
9259 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9261 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
9262 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9263 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9264 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
9268 case offsetof(struct bpf_sock_addr, user_port):
9269 /* To get port we need to know sa_family first and then treat
9270 * sockaddr as either sockaddr_in or sockaddr_in6.
9271 * Though we can simplify since port field has same offset and
9272 * size in both structures.
9273 * Here we check this invariant and use just one of the
9274 * structures if it's true.
9276 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
9277 offsetof(struct sockaddr_in6, sin6_port));
9278 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
9279 sizeof_field(struct sockaddr_in6, sin6_port));
9280 /* Account for sin6_port being smaller than user_port. */
9281 port_size = min(port_size, BPF_LDST_BYTES(si));
9282 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9283 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9284 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
9287 case offsetof(struct bpf_sock_addr, family):
9288 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9289 struct sock, sk, sk_family);
9292 case offsetof(struct bpf_sock_addr, type):
9293 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9294 struct sock, sk, sk_type);
9297 case offsetof(struct bpf_sock_addr, protocol):
9298 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9299 struct sock, sk, sk_protocol);
9302 case offsetof(struct bpf_sock_addr, msg_src_ip4):
9303 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
9304 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9305 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
9306 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
9309 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9312 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
9313 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
9314 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9315 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
9316 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
9318 case offsetof(struct bpf_sock_addr, sk):
9319 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
9320 si->dst_reg, si->src_reg,
9321 offsetof(struct bpf_sock_addr_kern, sk));
9325 return insn - insn_buf;
9328 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
9329 const struct bpf_insn *si,
9330 struct bpf_insn *insn_buf,
9331 struct bpf_prog *prog,
9334 struct bpf_insn *insn = insn_buf;
9337 /* Helper macro for adding read access to tcp_sock or sock fields. */
9338 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
9340 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \
9341 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
9342 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
9343 if (si->dst_reg == reg || si->src_reg == reg) \
9345 if (si->dst_reg == reg || si->src_reg == reg) \
9347 if (si->dst_reg == si->src_reg) { \
9348 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
9349 offsetof(struct bpf_sock_ops_kern, \
9351 fullsock_reg = reg; \
9354 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9355 struct bpf_sock_ops_kern, \
9357 fullsock_reg, si->src_reg, \
9358 offsetof(struct bpf_sock_ops_kern, \
9360 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
9361 if (si->dst_reg == si->src_reg) \
9362 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9363 offsetof(struct bpf_sock_ops_kern, \
9365 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9366 struct bpf_sock_ops_kern, sk),\
9367 si->dst_reg, si->src_reg, \
9368 offsetof(struct bpf_sock_ops_kern, sk));\
9369 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
9371 si->dst_reg, si->dst_reg, \
9372 offsetof(OBJ, OBJ_FIELD)); \
9373 if (si->dst_reg == si->src_reg) { \
9374 *insn++ = BPF_JMP_A(1); \
9375 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9376 offsetof(struct bpf_sock_ops_kern, \
9381 #define SOCK_OPS_GET_SK() \
9383 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \
9384 if (si->dst_reg == reg || si->src_reg == reg) \
9386 if (si->dst_reg == reg || si->src_reg == reg) \
9388 if (si->dst_reg == si->src_reg) { \
9389 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
9390 offsetof(struct bpf_sock_ops_kern, \
9392 fullsock_reg = reg; \
9395 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9396 struct bpf_sock_ops_kern, \
9398 fullsock_reg, si->src_reg, \
9399 offsetof(struct bpf_sock_ops_kern, \
9401 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
9402 if (si->dst_reg == si->src_reg) \
9403 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9404 offsetof(struct bpf_sock_ops_kern, \
9406 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9407 struct bpf_sock_ops_kern, sk),\
9408 si->dst_reg, si->src_reg, \
9409 offsetof(struct bpf_sock_ops_kern, sk));\
9410 if (si->dst_reg == si->src_reg) { \
9411 *insn++ = BPF_JMP_A(1); \
9412 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9413 offsetof(struct bpf_sock_ops_kern, \
9418 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
9419 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
9421 /* Helper macro for adding write access to tcp_sock or sock fields.
9422 * The macro is called with two registers, dst_reg which contains a pointer
9423 * to ctx (context) and src_reg which contains the value that should be
9424 * stored. However, we need an additional register since we cannot overwrite
9425 * dst_reg because it may be used later in the program.
9426 * Instead we "borrow" one of the other register. We first save its value
9427 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
9428 * it at the end of the macro.
9430 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
9432 int reg = BPF_REG_9; \
9433 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
9434 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
9435 if (si->dst_reg == reg || si->src_reg == reg) \
9437 if (si->dst_reg == reg || si->src_reg == reg) \
9439 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
9440 offsetof(struct bpf_sock_ops_kern, \
9442 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9443 struct bpf_sock_ops_kern, \
9446 offsetof(struct bpf_sock_ops_kern, \
9448 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
9449 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9450 struct bpf_sock_ops_kern, sk),\
9452 offsetof(struct bpf_sock_ops_kern, sk));\
9453 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \
9455 offsetof(OBJ, OBJ_FIELD)); \
9456 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
9457 offsetof(struct bpf_sock_ops_kern, \
9461 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
9463 if (TYPE == BPF_WRITE) \
9464 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
9466 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
9469 if (insn > insn_buf)
9470 return insn - insn_buf;
9473 case offsetof(struct bpf_sock_ops, op):
9474 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9476 si->dst_reg, si->src_reg,
9477 offsetof(struct bpf_sock_ops_kern, op));
9480 case offsetof(struct bpf_sock_ops, replylong[0]) ...
9481 offsetof(struct bpf_sock_ops, replylong[3]):
9482 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
9483 sizeof_field(struct bpf_sock_ops_kern, reply));
9484 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
9485 sizeof_field(struct bpf_sock_ops_kern, replylong));
9487 off -= offsetof(struct bpf_sock_ops, replylong[0]);
9488 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
9489 if (type == BPF_WRITE)
9490 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9493 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9497 case offsetof(struct bpf_sock_ops, family):
9498 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9500 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9501 struct bpf_sock_ops_kern, sk),
9502 si->dst_reg, si->src_reg,
9503 offsetof(struct bpf_sock_ops_kern, sk));
9504 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9505 offsetof(struct sock_common, skc_family));
9508 case offsetof(struct bpf_sock_ops, remote_ip4):
9509 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9511 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9512 struct bpf_sock_ops_kern, sk),
9513 si->dst_reg, si->src_reg,
9514 offsetof(struct bpf_sock_ops_kern, sk));
9515 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9516 offsetof(struct sock_common, skc_daddr));
9519 case offsetof(struct bpf_sock_ops, local_ip4):
9520 BUILD_BUG_ON(sizeof_field(struct sock_common,
9521 skc_rcv_saddr) != 4);
9523 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9524 struct bpf_sock_ops_kern, sk),
9525 si->dst_reg, si->src_reg,
9526 offsetof(struct bpf_sock_ops_kern, sk));
9527 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9528 offsetof(struct sock_common,
9532 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
9533 offsetof(struct bpf_sock_ops, remote_ip6[3]):
9534 #if IS_ENABLED(CONFIG_IPV6)
9535 BUILD_BUG_ON(sizeof_field(struct sock_common,
9536 skc_v6_daddr.s6_addr32[0]) != 4);
9539 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
9540 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9541 struct bpf_sock_ops_kern, sk),
9542 si->dst_reg, si->src_reg,
9543 offsetof(struct bpf_sock_ops_kern, sk));
9544 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9545 offsetof(struct sock_common,
9546 skc_v6_daddr.s6_addr32[0]) +
9549 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9553 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
9554 offsetof(struct bpf_sock_ops, local_ip6[3]):
9555 #if IS_ENABLED(CONFIG_IPV6)
9556 BUILD_BUG_ON(sizeof_field(struct sock_common,
9557 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9560 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
9561 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9562 struct bpf_sock_ops_kern, sk),
9563 si->dst_reg, si->src_reg,
9564 offsetof(struct bpf_sock_ops_kern, sk));
9565 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9566 offsetof(struct sock_common,
9567 skc_v6_rcv_saddr.s6_addr32[0]) +
9570 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9574 case offsetof(struct bpf_sock_ops, remote_port):
9575 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9577 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9578 struct bpf_sock_ops_kern, sk),
9579 si->dst_reg, si->src_reg,
9580 offsetof(struct bpf_sock_ops_kern, sk));
9581 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9582 offsetof(struct sock_common, skc_dport));
9583 #ifndef __BIG_ENDIAN_BITFIELD
9584 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9588 case offsetof(struct bpf_sock_ops, local_port):
9589 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9591 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9592 struct bpf_sock_ops_kern, sk),
9593 si->dst_reg, si->src_reg,
9594 offsetof(struct bpf_sock_ops_kern, sk));
9595 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9596 offsetof(struct sock_common, skc_num));
9599 case offsetof(struct bpf_sock_ops, is_fullsock):
9600 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9601 struct bpf_sock_ops_kern,
9603 si->dst_reg, si->src_reg,
9604 offsetof(struct bpf_sock_ops_kern,
9608 case offsetof(struct bpf_sock_ops, state):
9609 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
9611 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9612 struct bpf_sock_ops_kern, sk),
9613 si->dst_reg, si->src_reg,
9614 offsetof(struct bpf_sock_ops_kern, sk));
9615 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
9616 offsetof(struct sock_common, skc_state));
9619 case offsetof(struct bpf_sock_ops, rtt_min):
9620 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
9621 sizeof(struct minmax));
9622 BUILD_BUG_ON(sizeof(struct minmax) <
9623 sizeof(struct minmax_sample));
9625 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9626 struct bpf_sock_ops_kern, sk),
9627 si->dst_reg, si->src_reg,
9628 offsetof(struct bpf_sock_ops_kern, sk));
9629 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9630 offsetof(struct tcp_sock, rtt_min) +
9631 sizeof_field(struct minmax_sample, t));
9634 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
9635 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
9639 case offsetof(struct bpf_sock_ops, sk_txhash):
9640 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
9643 case offsetof(struct bpf_sock_ops, snd_cwnd):
9644 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
9646 case offsetof(struct bpf_sock_ops, srtt_us):
9647 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
9649 case offsetof(struct bpf_sock_ops, snd_ssthresh):
9650 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
9652 case offsetof(struct bpf_sock_ops, rcv_nxt):
9653 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
9655 case offsetof(struct bpf_sock_ops, snd_nxt):
9656 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
9658 case offsetof(struct bpf_sock_ops, snd_una):
9659 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
9661 case offsetof(struct bpf_sock_ops, mss_cache):
9662 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
9664 case offsetof(struct bpf_sock_ops, ecn_flags):
9665 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
9667 case offsetof(struct bpf_sock_ops, rate_delivered):
9668 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
9670 case offsetof(struct bpf_sock_ops, rate_interval_us):
9671 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
9673 case offsetof(struct bpf_sock_ops, packets_out):
9674 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
9676 case offsetof(struct bpf_sock_ops, retrans_out):
9677 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
9679 case offsetof(struct bpf_sock_ops, total_retrans):
9680 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
9682 case offsetof(struct bpf_sock_ops, segs_in):
9683 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
9685 case offsetof(struct bpf_sock_ops, data_segs_in):
9686 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
9688 case offsetof(struct bpf_sock_ops, segs_out):
9689 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
9691 case offsetof(struct bpf_sock_ops, data_segs_out):
9692 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
9694 case offsetof(struct bpf_sock_ops, lost_out):
9695 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
9697 case offsetof(struct bpf_sock_ops, sacked_out):
9698 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
9700 case offsetof(struct bpf_sock_ops, bytes_received):
9701 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
9703 case offsetof(struct bpf_sock_ops, bytes_acked):
9704 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
9706 case offsetof(struct bpf_sock_ops, sk):
9709 case offsetof(struct bpf_sock_ops, skb_data_end):
9710 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9712 si->dst_reg, si->src_reg,
9713 offsetof(struct bpf_sock_ops_kern,
9716 case offsetof(struct bpf_sock_ops, skb_data):
9717 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9719 si->dst_reg, si->src_reg,
9720 offsetof(struct bpf_sock_ops_kern,
9722 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9723 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9724 si->dst_reg, si->dst_reg,
9725 offsetof(struct sk_buff, data));
9727 case offsetof(struct bpf_sock_ops, skb_len):
9728 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9730 si->dst_reg, si->src_reg,
9731 offsetof(struct bpf_sock_ops_kern,
9733 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9734 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9735 si->dst_reg, si->dst_reg,
9736 offsetof(struct sk_buff, len));
9738 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9739 off = offsetof(struct sk_buff, cb);
9740 off += offsetof(struct tcp_skb_cb, tcp_flags);
9741 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
9742 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9744 si->dst_reg, si->src_reg,
9745 offsetof(struct bpf_sock_ops_kern,
9747 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9748 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
9750 si->dst_reg, si->dst_reg, off);
9753 return insn - insn_buf;
9756 /* data_end = skb->data + skb_headlen() */
9757 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
9758 struct bpf_insn *insn)
9761 int temp_reg_off = offsetof(struct sk_buff, cb) +
9762 offsetof(struct sk_skb_cb, temp_reg);
9764 if (si->src_reg == si->dst_reg) {
9765 /* We need an extra register, choose and save a register. */
9767 if (si->src_reg == reg || si->dst_reg == reg)
9769 if (si->src_reg == reg || si->dst_reg == reg)
9771 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off);
9776 /* reg = skb->data */
9777 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9779 offsetof(struct sk_buff, data));
9781 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9782 BPF_REG_AX, si->src_reg,
9783 offsetof(struct sk_buff, len));
9784 /* reg = skb->data + skb->len */
9785 *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX);
9786 /* AX = skb->data_len */
9787 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
9788 BPF_REG_AX, si->src_reg,
9789 offsetof(struct sk_buff, data_len));
9791 /* reg = skb->data + skb->len - skb->data_len */
9792 *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX);
9794 if (si->src_reg == si->dst_reg) {
9795 /* Restore the saved register */
9796 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg);
9797 *insn++ = BPF_MOV64_REG(si->dst_reg, reg);
9798 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off);
9804 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
9805 const struct bpf_insn *si,
9806 struct bpf_insn *insn_buf,
9807 struct bpf_prog *prog, u32 *target_size)
9809 struct bpf_insn *insn = insn_buf;
9813 case offsetof(struct __sk_buff, data_end):
9814 insn = bpf_convert_data_end_access(si, insn);
9816 case offsetof(struct __sk_buff, cb[0]) ...
9817 offsetofend(struct __sk_buff, cb[4]) - 1:
9818 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20);
9819 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
9820 offsetof(struct sk_skb_cb, data)) %
9823 prog->cb_access = 1;
9825 off -= offsetof(struct __sk_buff, cb[0]);
9826 off += offsetof(struct sk_buff, cb);
9827 off += offsetof(struct sk_skb_cb, data);
9828 if (type == BPF_WRITE)
9829 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
9832 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
9838 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9842 return insn - insn_buf;
9845 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
9846 const struct bpf_insn *si,
9847 struct bpf_insn *insn_buf,
9848 struct bpf_prog *prog, u32 *target_size)
9850 struct bpf_insn *insn = insn_buf;
9851 #if IS_ENABLED(CONFIG_IPV6)
9855 /* convert ctx uses the fact sg element is first in struct */
9856 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
9859 case offsetof(struct sk_msg_md, data):
9860 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
9861 si->dst_reg, si->src_reg,
9862 offsetof(struct sk_msg, data));
9864 case offsetof(struct sk_msg_md, data_end):
9865 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
9866 si->dst_reg, si->src_reg,
9867 offsetof(struct sk_msg, data_end));
9869 case offsetof(struct sk_msg_md, family):
9870 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9872 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9874 si->dst_reg, si->src_reg,
9875 offsetof(struct sk_msg, sk));
9876 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9877 offsetof(struct sock_common, skc_family));
9880 case offsetof(struct sk_msg_md, remote_ip4):
9881 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9883 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9885 si->dst_reg, si->src_reg,
9886 offsetof(struct sk_msg, sk));
9887 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9888 offsetof(struct sock_common, skc_daddr));
9891 case offsetof(struct sk_msg_md, local_ip4):
9892 BUILD_BUG_ON(sizeof_field(struct sock_common,
9893 skc_rcv_saddr) != 4);
9895 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9897 si->dst_reg, si->src_reg,
9898 offsetof(struct sk_msg, sk));
9899 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9900 offsetof(struct sock_common,
9904 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
9905 offsetof(struct sk_msg_md, remote_ip6[3]):
9906 #if IS_ENABLED(CONFIG_IPV6)
9907 BUILD_BUG_ON(sizeof_field(struct sock_common,
9908 skc_v6_daddr.s6_addr32[0]) != 4);
9911 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
9912 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9914 si->dst_reg, si->src_reg,
9915 offsetof(struct sk_msg, sk));
9916 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9917 offsetof(struct sock_common,
9918 skc_v6_daddr.s6_addr32[0]) +
9921 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9925 case offsetof(struct sk_msg_md, local_ip6[0]) ...
9926 offsetof(struct sk_msg_md, local_ip6[3]):
9927 #if IS_ENABLED(CONFIG_IPV6)
9928 BUILD_BUG_ON(sizeof_field(struct sock_common,
9929 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9932 off -= offsetof(struct sk_msg_md, local_ip6[0]);
9933 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9935 si->dst_reg, si->src_reg,
9936 offsetof(struct sk_msg, sk));
9937 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9938 offsetof(struct sock_common,
9939 skc_v6_rcv_saddr.s6_addr32[0]) +
9942 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9946 case offsetof(struct sk_msg_md, remote_port):
9947 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9949 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9951 si->dst_reg, si->src_reg,
9952 offsetof(struct sk_msg, sk));
9953 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9954 offsetof(struct sock_common, skc_dport));
9955 #ifndef __BIG_ENDIAN_BITFIELD
9956 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9960 case offsetof(struct sk_msg_md, local_port):
9961 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9963 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9965 si->dst_reg, si->src_reg,
9966 offsetof(struct sk_msg, sk));
9967 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9968 offsetof(struct sock_common, skc_num));
9971 case offsetof(struct sk_msg_md, size):
9972 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
9973 si->dst_reg, si->src_reg,
9974 offsetof(struct sk_msg_sg, size));
9977 case offsetof(struct sk_msg_md, sk):
9978 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
9979 si->dst_reg, si->src_reg,
9980 offsetof(struct sk_msg, sk));
9984 return insn - insn_buf;
9987 const struct bpf_verifier_ops sk_filter_verifier_ops = {
9988 .get_func_proto = sk_filter_func_proto,
9989 .is_valid_access = sk_filter_is_valid_access,
9990 .convert_ctx_access = bpf_convert_ctx_access,
9991 .gen_ld_abs = bpf_gen_ld_abs,
9994 const struct bpf_prog_ops sk_filter_prog_ops = {
9995 .test_run = bpf_prog_test_run_skb,
9998 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
9999 .get_func_proto = tc_cls_act_func_proto,
10000 .is_valid_access = tc_cls_act_is_valid_access,
10001 .convert_ctx_access = tc_cls_act_convert_ctx_access,
10002 .gen_prologue = tc_cls_act_prologue,
10003 .gen_ld_abs = bpf_gen_ld_abs,
10004 .check_kfunc_call = bpf_prog_test_check_kfunc_call,
10007 const struct bpf_prog_ops tc_cls_act_prog_ops = {
10008 .test_run = bpf_prog_test_run_skb,
10011 const struct bpf_verifier_ops xdp_verifier_ops = {
10012 .get_func_proto = xdp_func_proto,
10013 .is_valid_access = xdp_is_valid_access,
10014 .convert_ctx_access = xdp_convert_ctx_access,
10015 .gen_prologue = bpf_noop_prologue,
10018 const struct bpf_prog_ops xdp_prog_ops = {
10019 .test_run = bpf_prog_test_run_xdp,
10022 const struct bpf_verifier_ops cg_skb_verifier_ops = {
10023 .get_func_proto = cg_skb_func_proto,
10024 .is_valid_access = cg_skb_is_valid_access,
10025 .convert_ctx_access = bpf_convert_ctx_access,
10028 const struct bpf_prog_ops cg_skb_prog_ops = {
10029 .test_run = bpf_prog_test_run_skb,
10032 const struct bpf_verifier_ops lwt_in_verifier_ops = {
10033 .get_func_proto = lwt_in_func_proto,
10034 .is_valid_access = lwt_is_valid_access,
10035 .convert_ctx_access = bpf_convert_ctx_access,
10038 const struct bpf_prog_ops lwt_in_prog_ops = {
10039 .test_run = bpf_prog_test_run_skb,
10042 const struct bpf_verifier_ops lwt_out_verifier_ops = {
10043 .get_func_proto = lwt_out_func_proto,
10044 .is_valid_access = lwt_is_valid_access,
10045 .convert_ctx_access = bpf_convert_ctx_access,
10048 const struct bpf_prog_ops lwt_out_prog_ops = {
10049 .test_run = bpf_prog_test_run_skb,
10052 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
10053 .get_func_proto = lwt_xmit_func_proto,
10054 .is_valid_access = lwt_is_valid_access,
10055 .convert_ctx_access = bpf_convert_ctx_access,
10056 .gen_prologue = tc_cls_act_prologue,
10059 const struct bpf_prog_ops lwt_xmit_prog_ops = {
10060 .test_run = bpf_prog_test_run_skb,
10063 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
10064 .get_func_proto = lwt_seg6local_func_proto,
10065 .is_valid_access = lwt_is_valid_access,
10066 .convert_ctx_access = bpf_convert_ctx_access,
10069 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
10070 .test_run = bpf_prog_test_run_skb,
10073 const struct bpf_verifier_ops cg_sock_verifier_ops = {
10074 .get_func_proto = sock_filter_func_proto,
10075 .is_valid_access = sock_filter_is_valid_access,
10076 .convert_ctx_access = bpf_sock_convert_ctx_access,
10079 const struct bpf_prog_ops cg_sock_prog_ops = {
10082 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
10083 .get_func_proto = sock_addr_func_proto,
10084 .is_valid_access = sock_addr_is_valid_access,
10085 .convert_ctx_access = sock_addr_convert_ctx_access,
10088 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
10091 const struct bpf_verifier_ops sock_ops_verifier_ops = {
10092 .get_func_proto = sock_ops_func_proto,
10093 .is_valid_access = sock_ops_is_valid_access,
10094 .convert_ctx_access = sock_ops_convert_ctx_access,
10097 const struct bpf_prog_ops sock_ops_prog_ops = {
10100 const struct bpf_verifier_ops sk_skb_verifier_ops = {
10101 .get_func_proto = sk_skb_func_proto,
10102 .is_valid_access = sk_skb_is_valid_access,
10103 .convert_ctx_access = sk_skb_convert_ctx_access,
10104 .gen_prologue = sk_skb_prologue,
10107 const struct bpf_prog_ops sk_skb_prog_ops = {
10110 const struct bpf_verifier_ops sk_msg_verifier_ops = {
10111 .get_func_proto = sk_msg_func_proto,
10112 .is_valid_access = sk_msg_is_valid_access,
10113 .convert_ctx_access = sk_msg_convert_ctx_access,
10114 .gen_prologue = bpf_noop_prologue,
10117 const struct bpf_prog_ops sk_msg_prog_ops = {
10120 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
10121 .get_func_proto = flow_dissector_func_proto,
10122 .is_valid_access = flow_dissector_is_valid_access,
10123 .convert_ctx_access = flow_dissector_convert_ctx_access,
10126 const struct bpf_prog_ops flow_dissector_prog_ops = {
10127 .test_run = bpf_prog_test_run_flow_dissector,
10130 int sk_detach_filter(struct sock *sk)
10133 struct sk_filter *filter;
10135 if (sock_flag(sk, SOCK_FILTER_LOCKED))
10138 filter = rcu_dereference_protected(sk->sk_filter,
10139 lockdep_sock_is_held(sk));
10141 RCU_INIT_POINTER(sk->sk_filter, NULL);
10142 sk_filter_uncharge(sk, filter);
10148 EXPORT_SYMBOL_GPL(sk_detach_filter);
10150 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
10153 struct sock_fprog_kern *fprog;
10154 struct sk_filter *filter;
10158 filter = rcu_dereference_protected(sk->sk_filter,
10159 lockdep_sock_is_held(sk));
10163 /* We're copying the filter that has been originally attached,
10164 * so no conversion/decode needed anymore. eBPF programs that
10165 * have no original program cannot be dumped through this.
10168 fprog = filter->prog->orig_prog;
10174 /* User space only enquires number of filter blocks. */
10178 if (len < fprog->len)
10182 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
10185 /* Instead of bytes, the API requests to return the number
10186 * of filter blocks.
10195 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
10196 struct sock_reuseport *reuse,
10197 struct sock *sk, struct sk_buff *skb,
10198 struct sock *migrating_sk,
10201 reuse_kern->skb = skb;
10202 reuse_kern->sk = sk;
10203 reuse_kern->selected_sk = NULL;
10204 reuse_kern->migrating_sk = migrating_sk;
10205 reuse_kern->data_end = skb->data + skb_headlen(skb);
10206 reuse_kern->hash = hash;
10207 reuse_kern->reuseport_id = reuse->reuseport_id;
10208 reuse_kern->bind_inany = reuse->bind_inany;
10211 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
10212 struct bpf_prog *prog, struct sk_buff *skb,
10213 struct sock *migrating_sk,
10216 struct sk_reuseport_kern reuse_kern;
10217 enum sk_action action;
10219 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
10220 action = bpf_prog_run(prog, &reuse_kern);
10222 if (action == SK_PASS)
10223 return reuse_kern.selected_sk;
10225 return ERR_PTR(-ECONNREFUSED);
10228 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
10229 struct bpf_map *, map, void *, key, u32, flags)
10231 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
10232 struct sock_reuseport *reuse;
10233 struct sock *selected_sk;
10235 selected_sk = map->ops->map_lookup_elem(map, key);
10239 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
10241 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
10242 if (sk_is_refcounted(selected_sk))
10243 sock_put(selected_sk);
10245 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
10246 * The only (!reuse) case here is - the sk has already been
10247 * unhashed (e.g. by close()), so treat it as -ENOENT.
10249 * Other maps (e.g. sock_map) do not provide this guarantee and
10250 * the sk may never be in the reuseport group to begin with.
10252 return is_sockarray ? -ENOENT : -EINVAL;
10255 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
10256 struct sock *sk = reuse_kern->sk;
10258 if (sk->sk_protocol != selected_sk->sk_protocol)
10259 return -EPROTOTYPE;
10260 else if (sk->sk_family != selected_sk->sk_family)
10261 return -EAFNOSUPPORT;
10263 /* Catch all. Likely bound to a different sockaddr. */
10267 reuse_kern->selected_sk = selected_sk;
10272 static const struct bpf_func_proto sk_select_reuseport_proto = {
10273 .func = sk_select_reuseport,
10275 .ret_type = RET_INTEGER,
10276 .arg1_type = ARG_PTR_TO_CTX,
10277 .arg2_type = ARG_CONST_MAP_PTR,
10278 .arg3_type = ARG_PTR_TO_MAP_KEY,
10279 .arg4_type = ARG_ANYTHING,
10282 BPF_CALL_4(sk_reuseport_load_bytes,
10283 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10284 void *, to, u32, len)
10286 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
10289 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
10290 .func = sk_reuseport_load_bytes,
10292 .ret_type = RET_INTEGER,
10293 .arg1_type = ARG_PTR_TO_CTX,
10294 .arg2_type = ARG_ANYTHING,
10295 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
10296 .arg4_type = ARG_CONST_SIZE,
10299 BPF_CALL_5(sk_reuseport_load_bytes_relative,
10300 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10301 void *, to, u32, len, u32, start_header)
10303 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
10304 len, start_header);
10307 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
10308 .func = sk_reuseport_load_bytes_relative,
10310 .ret_type = RET_INTEGER,
10311 .arg1_type = ARG_PTR_TO_CTX,
10312 .arg2_type = ARG_ANYTHING,
10313 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
10314 .arg4_type = ARG_CONST_SIZE,
10315 .arg5_type = ARG_ANYTHING,
10318 static const struct bpf_func_proto *
10319 sk_reuseport_func_proto(enum bpf_func_id func_id,
10320 const struct bpf_prog *prog)
10323 case BPF_FUNC_sk_select_reuseport:
10324 return &sk_select_reuseport_proto;
10325 case BPF_FUNC_skb_load_bytes:
10326 return &sk_reuseport_load_bytes_proto;
10327 case BPF_FUNC_skb_load_bytes_relative:
10328 return &sk_reuseport_load_bytes_relative_proto;
10329 case BPF_FUNC_get_socket_cookie:
10330 return &bpf_get_socket_ptr_cookie_proto;
10331 case BPF_FUNC_ktime_get_coarse_ns:
10332 return &bpf_ktime_get_coarse_ns_proto;
10334 return bpf_base_func_proto(func_id);
10339 sk_reuseport_is_valid_access(int off, int size,
10340 enum bpf_access_type type,
10341 const struct bpf_prog *prog,
10342 struct bpf_insn_access_aux *info)
10344 const u32 size_default = sizeof(__u32);
10346 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
10347 off % size || type != BPF_READ)
10351 case offsetof(struct sk_reuseport_md, data):
10352 info->reg_type = PTR_TO_PACKET;
10353 return size == sizeof(__u64);
10355 case offsetof(struct sk_reuseport_md, data_end):
10356 info->reg_type = PTR_TO_PACKET_END;
10357 return size == sizeof(__u64);
10359 case offsetof(struct sk_reuseport_md, hash):
10360 return size == size_default;
10362 case offsetof(struct sk_reuseport_md, sk):
10363 info->reg_type = PTR_TO_SOCKET;
10364 return size == sizeof(__u64);
10366 case offsetof(struct sk_reuseport_md, migrating_sk):
10367 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
10368 return size == sizeof(__u64);
10370 /* Fields that allow narrowing */
10371 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
10372 if (size < sizeof_field(struct sk_buff, protocol))
10375 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
10376 case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
10377 case bpf_ctx_range(struct sk_reuseport_md, len):
10378 bpf_ctx_record_field_size(info, size_default);
10379 return bpf_ctx_narrow_access_ok(off, size, size_default);
10386 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
10387 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
10388 si->dst_reg, si->src_reg, \
10389 bpf_target_off(struct sk_reuseport_kern, F, \
10390 sizeof_field(struct sk_reuseport_kern, F), \
10394 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
10395 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
10400 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \
10401 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
10406 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
10407 const struct bpf_insn *si,
10408 struct bpf_insn *insn_buf,
10409 struct bpf_prog *prog,
10412 struct bpf_insn *insn = insn_buf;
10415 case offsetof(struct sk_reuseport_md, data):
10416 SK_REUSEPORT_LOAD_SKB_FIELD(data);
10419 case offsetof(struct sk_reuseport_md, len):
10420 SK_REUSEPORT_LOAD_SKB_FIELD(len);
10423 case offsetof(struct sk_reuseport_md, eth_protocol):
10424 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
10427 case offsetof(struct sk_reuseport_md, ip_protocol):
10428 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
10431 case offsetof(struct sk_reuseport_md, data_end):
10432 SK_REUSEPORT_LOAD_FIELD(data_end);
10435 case offsetof(struct sk_reuseport_md, hash):
10436 SK_REUSEPORT_LOAD_FIELD(hash);
10439 case offsetof(struct sk_reuseport_md, bind_inany):
10440 SK_REUSEPORT_LOAD_FIELD(bind_inany);
10443 case offsetof(struct sk_reuseport_md, sk):
10444 SK_REUSEPORT_LOAD_FIELD(sk);
10447 case offsetof(struct sk_reuseport_md, migrating_sk):
10448 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
10452 return insn - insn_buf;
10455 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
10456 .get_func_proto = sk_reuseport_func_proto,
10457 .is_valid_access = sk_reuseport_is_valid_access,
10458 .convert_ctx_access = sk_reuseport_convert_ctx_access,
10461 const struct bpf_prog_ops sk_reuseport_prog_ops = {
10464 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
10465 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
10467 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
10468 struct sock *, sk, u64, flags)
10470 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
10471 BPF_SK_LOOKUP_F_NO_REUSEPORT)))
10473 if (unlikely(sk && sk_is_refcounted(sk)))
10474 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
10475 if (unlikely(sk && sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN))
10476 return -ESOCKTNOSUPPORT; /* only accept TCP socket in LISTEN */
10477 if (unlikely(sk && sk_is_udp(sk) && sk->sk_state != TCP_CLOSE))
10478 return -ESOCKTNOSUPPORT; /* only accept UDP socket in CLOSE */
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 bpf_ctx_range(struct bpf_sk_lookup, remote_port):
10544 case bpf_ctx_range(struct bpf_sk_lookup, local_port):
10545 case bpf_ctx_range(struct bpf_sk_lookup, ingress_ifindex):
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));
10636 case offsetof(struct bpf_sk_lookup, ingress_ifindex):
10637 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10638 bpf_target_off(struct bpf_sk_lookup_kern,
10639 ingress_ifindex, 4, target_size));
10643 return insn - insn_buf;
10646 const struct bpf_prog_ops sk_lookup_prog_ops = {
10647 .test_run = bpf_prog_test_run_sk_lookup,
10650 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
10651 .get_func_proto = sk_lookup_func_proto,
10652 .is_valid_access = sk_lookup_is_valid_access,
10653 .convert_ctx_access = sk_lookup_convert_ctx_access,
10656 #endif /* CONFIG_INET */
10658 DEFINE_BPF_DISPATCHER(xdp)
10660 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
10662 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
10665 BTF_ID_LIST_GLOBAL(btf_sock_ids, MAX_BTF_SOCK_TYPE)
10666 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
10668 #undef BTF_SOCK_TYPE
10670 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
10672 /* tcp6_sock type is not generated in dwarf and hence btf,
10673 * trigger an explicit type generation here.
10675 BTF_TYPE_EMIT(struct tcp6_sock);
10676 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
10677 sk->sk_family == AF_INET6)
10678 return (unsigned long)sk;
10680 return (unsigned long)NULL;
10683 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
10684 .func = bpf_skc_to_tcp6_sock,
10686 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10687 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10688 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
10691 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
10693 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
10694 return (unsigned long)sk;
10696 return (unsigned long)NULL;
10699 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
10700 .func = bpf_skc_to_tcp_sock,
10702 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10703 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10704 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
10707 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
10709 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
10710 * generated if CONFIG_INET=n. Trigger an explicit generation here.
10712 BTF_TYPE_EMIT(struct inet_timewait_sock);
10713 BTF_TYPE_EMIT(struct tcp_timewait_sock);
10716 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
10717 return (unsigned long)sk;
10720 #if IS_BUILTIN(CONFIG_IPV6)
10721 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
10722 return (unsigned long)sk;
10725 return (unsigned long)NULL;
10728 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
10729 .func = bpf_skc_to_tcp_timewait_sock,
10731 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10732 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10733 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
10736 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
10739 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10740 return (unsigned long)sk;
10743 #if IS_BUILTIN(CONFIG_IPV6)
10744 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10745 return (unsigned long)sk;
10748 return (unsigned long)NULL;
10751 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
10752 .func = bpf_skc_to_tcp_request_sock,
10754 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10755 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10756 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
10759 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
10761 /* udp6_sock type is not generated in dwarf and hence btf,
10762 * trigger an explicit type generation here.
10764 BTF_TYPE_EMIT(struct udp6_sock);
10765 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
10766 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
10767 return (unsigned long)sk;
10769 return (unsigned long)NULL;
10772 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
10773 .func = bpf_skc_to_udp6_sock,
10775 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10776 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10777 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
10780 BPF_CALL_1(bpf_skc_to_unix_sock, struct sock *, sk)
10782 /* unix_sock type is not generated in dwarf and hence btf,
10783 * trigger an explicit type generation here.
10785 BTF_TYPE_EMIT(struct unix_sock);
10786 if (sk && sk_fullsock(sk) && sk->sk_family == AF_UNIX)
10787 return (unsigned long)sk;
10789 return (unsigned long)NULL;
10792 const struct bpf_func_proto bpf_skc_to_unix_sock_proto = {
10793 .func = bpf_skc_to_unix_sock,
10795 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10796 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10797 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UNIX],
10800 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
10802 return (unsigned long)sock_from_file(file);
10805 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
10806 BTF_ID(struct, socket)
10807 BTF_ID(struct, file)
10809 const struct bpf_func_proto bpf_sock_from_file_proto = {
10810 .func = bpf_sock_from_file,
10812 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10813 .ret_btf_id = &bpf_sock_from_file_btf_ids[0],
10814 .arg1_type = ARG_PTR_TO_BTF_ID,
10815 .arg1_btf_id = &bpf_sock_from_file_btf_ids[1],
10818 static const struct bpf_func_proto *
10819 bpf_sk_base_func_proto(enum bpf_func_id func_id)
10821 const struct bpf_func_proto *func;
10824 case BPF_FUNC_skc_to_tcp6_sock:
10825 func = &bpf_skc_to_tcp6_sock_proto;
10827 case BPF_FUNC_skc_to_tcp_sock:
10828 func = &bpf_skc_to_tcp_sock_proto;
10830 case BPF_FUNC_skc_to_tcp_timewait_sock:
10831 func = &bpf_skc_to_tcp_timewait_sock_proto;
10833 case BPF_FUNC_skc_to_tcp_request_sock:
10834 func = &bpf_skc_to_tcp_request_sock_proto;
10836 case BPF_FUNC_skc_to_udp6_sock:
10837 func = &bpf_skc_to_udp6_sock_proto;
10839 case BPF_FUNC_skc_to_unix_sock:
10840 func = &bpf_skc_to_unix_sock_proto;
10842 case BPF_FUNC_ktime_get_coarse_ns:
10843 return &bpf_ktime_get_coarse_ns_proto;
10845 return bpf_base_func_proto(func_id);
10848 if (!perfmon_capable())