1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Linux Socket Filter - Kernel level socket filtering
5 * Based on the design of the Berkeley Packet Filter. The new
6 * internal format has been designed by PLUMgrid:
8 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
12 * Jay Schulist <jschlst@samba.org>
13 * Alexei Starovoitov <ast@plumgrid.com>
14 * Daniel Borkmann <dborkman@redhat.com>
16 * Andi Kleen - Fix a few bad bugs and races.
17 * Kris Katterjohn - Added many additional checks in bpf_check_classic()
20 #include <linux/atomic.h>
21 #include <linux/module.h>
22 #include <linux/types.h>
24 #include <linux/fcntl.h>
25 #include <linux/socket.h>
26 #include <linux/sock_diag.h>
28 #include <linux/inet.h>
29 #include <linux/netdevice.h>
30 #include <linux/if_packet.h>
31 #include <linux/if_arp.h>
32 #include <linux/gfp.h>
33 #include <net/inet_common.h>
35 #include <net/protocol.h>
36 #include <net/netlink.h>
37 #include <linux/skbuff.h>
38 #include <linux/skmsg.h>
40 #include <net/flow_dissector.h>
41 #include <linux/errno.h>
42 #include <linux/timer.h>
43 #include <linux/uaccess.h>
44 #include <asm/unaligned.h>
45 #include <linux/filter.h>
46 #include <linux/ratelimit.h>
47 #include <linux/seccomp.h>
48 #include <linux/if_vlan.h>
49 #include <linux/bpf.h>
50 #include <linux/btf.h>
51 #include <net/sch_generic.h>
52 #include <net/cls_cgroup.h>
53 #include <net/dst_metadata.h>
55 #include <net/sock_reuseport.h>
56 #include <net/busy_poll.h>
60 #include <linux/bpf_trace.h>
61 #include <net/xdp_sock.h>
62 #include <linux/inetdevice.h>
63 #include <net/inet_hashtables.h>
64 #include <net/inet6_hashtables.h>
65 #include <net/ip_fib.h>
66 #include <net/nexthop.h>
70 #include <net/net_namespace.h>
71 #include <linux/seg6_local.h>
73 #include <net/seg6_local.h>
74 #include <net/lwtunnel.h>
75 #include <net/ipv6_stubs.h>
76 #include <net/bpf_sk_storage.h>
77 #include <net/transp_v6.h>
78 #include <linux/btf_ids.h>
82 static const struct bpf_func_proto *
83 bpf_sk_base_func_proto(enum bpf_func_id func_id);
85 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len)
87 if (in_compat_syscall()) {
88 struct compat_sock_fprog f32;
90 if (len != sizeof(f32))
92 if (copy_from_sockptr(&f32, src, sizeof(f32)))
94 memset(dst, 0, sizeof(*dst));
96 dst->filter = compat_ptr(f32.filter);
98 if (len != sizeof(*dst))
100 if (copy_from_sockptr(dst, src, sizeof(*dst)))
106 EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user);
109 * sk_filter_trim_cap - run a packet through a socket filter
110 * @sk: sock associated with &sk_buff
111 * @skb: buffer to filter
112 * @cap: limit on how short the eBPF program may trim the packet
114 * Run the eBPF program and then cut skb->data to correct size returned by
115 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
116 * than pkt_len we keep whole skb->data. This is the socket level
117 * wrapper to bpf_prog_run. It returns 0 if the packet should
118 * be accepted or -EPERM if the packet should be tossed.
121 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
124 struct sk_filter *filter;
127 * If the skb was allocated from pfmemalloc reserves, only
128 * allow SOCK_MEMALLOC sockets to use it as this socket is
129 * helping free memory
131 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
132 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
135 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
139 err = security_sock_rcv_skb(sk, skb);
144 filter = rcu_dereference(sk->sk_filter);
146 struct sock *save_sk = skb->sk;
147 unsigned int pkt_len;
150 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
152 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
158 EXPORT_SYMBOL(sk_filter_trim_cap);
160 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
162 return skb_get_poff(skb);
165 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
169 if (skb_is_nonlinear(skb))
172 if (skb->len < sizeof(struct nlattr))
175 if (a > skb->len - sizeof(struct nlattr))
178 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
180 return (void *) nla - (void *) skb->data;
185 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
189 if (skb_is_nonlinear(skb))
192 if (skb->len < sizeof(struct nlattr))
195 if (a > skb->len - sizeof(struct nlattr))
198 nla = (struct nlattr *) &skb->data[a];
199 if (nla->nla_len > skb->len - a)
202 nla = nla_find_nested(nla, x);
204 return (void *) nla - (void *) skb->data;
209 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
210 data, int, headlen, int, offset)
213 const int len = sizeof(tmp);
216 if (headlen - offset >= len)
217 return *(u8 *)(data + offset);
218 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
221 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
229 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
232 return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
236 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
237 data, int, headlen, int, offset)
240 const int len = sizeof(tmp);
243 if (headlen - offset >= len)
244 return get_unaligned_be16(data + offset);
245 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
246 return be16_to_cpu(tmp);
248 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
250 return get_unaligned_be16(ptr);
256 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
259 return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
263 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
264 data, int, headlen, int, offset)
267 const int len = sizeof(tmp);
269 if (likely(offset >= 0)) {
270 if (headlen - offset >= len)
271 return get_unaligned_be32(data + offset);
272 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
273 return be32_to_cpu(tmp);
275 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
277 return get_unaligned_be32(ptr);
283 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
286 return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
290 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
291 struct bpf_insn *insn_buf)
293 struct bpf_insn *insn = insn_buf;
297 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4);
299 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
300 offsetof(struct sk_buff, mark));
304 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
305 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
306 #ifdef __BIG_ENDIAN_BITFIELD
307 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
312 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2);
314 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
315 offsetof(struct sk_buff, queue_mapping));
318 case SKF_AD_VLAN_TAG:
319 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2);
321 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
322 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
323 offsetof(struct sk_buff, vlan_tci));
325 case SKF_AD_VLAN_TAG_PRESENT:
326 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_VLAN_PRESENT_OFFSET());
327 if (PKT_VLAN_PRESENT_BIT)
328 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, PKT_VLAN_PRESENT_BIT);
329 if (PKT_VLAN_PRESENT_BIT < 7)
330 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
334 return insn - insn_buf;
337 static bool convert_bpf_extensions(struct sock_filter *fp,
338 struct bpf_insn **insnp)
340 struct bpf_insn *insn = *insnp;
344 case SKF_AD_OFF + SKF_AD_PROTOCOL:
345 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2);
347 /* A = *(u16 *) (CTX + offsetof(protocol)) */
348 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
349 offsetof(struct sk_buff, protocol));
350 /* A = ntohs(A) [emitting a nop or swap16] */
351 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
354 case SKF_AD_OFF + SKF_AD_PKTTYPE:
355 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
359 case SKF_AD_OFF + SKF_AD_IFINDEX:
360 case SKF_AD_OFF + SKF_AD_HATYPE:
361 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4);
362 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2);
364 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
365 BPF_REG_TMP, BPF_REG_CTX,
366 offsetof(struct sk_buff, dev));
367 /* if (tmp != 0) goto pc + 1 */
368 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
369 *insn++ = BPF_EXIT_INSN();
370 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
371 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
372 offsetof(struct net_device, ifindex));
374 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
375 offsetof(struct net_device, type));
378 case SKF_AD_OFF + SKF_AD_MARK:
379 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
383 case SKF_AD_OFF + SKF_AD_RXHASH:
384 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4);
386 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
387 offsetof(struct sk_buff, hash));
390 case SKF_AD_OFF + SKF_AD_QUEUE:
391 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
395 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
396 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
397 BPF_REG_A, BPF_REG_CTX, insn);
401 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
402 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
403 BPF_REG_A, BPF_REG_CTX, insn);
407 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
408 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2);
410 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
411 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
412 offsetof(struct sk_buff, vlan_proto));
413 /* A = ntohs(A) [emitting a nop or swap16] */
414 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
417 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
418 case SKF_AD_OFF + SKF_AD_NLATTR:
419 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
420 case SKF_AD_OFF + SKF_AD_CPU:
421 case SKF_AD_OFF + SKF_AD_RANDOM:
423 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
425 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
427 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
428 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
430 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
431 *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
433 case SKF_AD_OFF + SKF_AD_NLATTR:
434 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
436 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
437 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
439 case SKF_AD_OFF + SKF_AD_CPU:
440 *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
442 case SKF_AD_OFF + SKF_AD_RANDOM:
443 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
444 bpf_user_rnd_init_once();
449 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
451 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
455 /* This is just a dummy call to avoid letting the compiler
456 * evict __bpf_call_base() as an optimization. Placed here
457 * where no-one bothers.
459 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
467 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
469 const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
470 int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
471 bool endian = BPF_SIZE(fp->code) == BPF_H ||
472 BPF_SIZE(fp->code) == BPF_W;
473 bool indirect = BPF_MODE(fp->code) == BPF_IND;
474 const int ip_align = NET_IP_ALIGN;
475 struct bpf_insn *insn = *insnp;
479 ((unaligned_ok && offset >= 0) ||
480 (!unaligned_ok && offset >= 0 &&
481 offset + ip_align >= 0 &&
482 offset + ip_align % size == 0))) {
483 bool ldx_off_ok = offset <= S16_MAX;
485 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
487 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
488 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
489 size, 2 + endian + (!ldx_off_ok * 2));
491 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
494 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
495 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
496 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
500 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
501 *insn++ = BPF_JMP_A(8);
504 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
505 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
506 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
508 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
510 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
512 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
515 switch (BPF_SIZE(fp->code)) {
517 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
520 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
523 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
529 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
530 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
531 *insn = BPF_EXIT_INSN();
538 * bpf_convert_filter - convert filter program
539 * @prog: the user passed filter program
540 * @len: the length of the user passed filter program
541 * @new_prog: allocated 'struct bpf_prog' or NULL
542 * @new_len: pointer to store length of converted program
543 * @seen_ld_abs: bool whether we've seen ld_abs/ind
545 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
546 * style extended BPF (eBPF).
547 * Conversion workflow:
549 * 1) First pass for calculating the new program length:
550 * bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
552 * 2) 2nd pass to remap in two passes: 1st pass finds new
553 * jump offsets, 2nd pass remapping:
554 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
556 static int bpf_convert_filter(struct sock_filter *prog, int len,
557 struct bpf_prog *new_prog, int *new_len,
560 int new_flen = 0, pass = 0, target, i, stack_off;
561 struct bpf_insn *new_insn, *first_insn = NULL;
562 struct sock_filter *fp;
566 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
567 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
569 if (len <= 0 || len > BPF_MAXINSNS)
573 first_insn = new_prog->insnsi;
574 addrs = kcalloc(len, sizeof(*addrs),
575 GFP_KERNEL | __GFP_NOWARN);
581 new_insn = first_insn;
584 /* Classic BPF related prologue emission. */
586 /* Classic BPF expects A and X to be reset first. These need
587 * to be guaranteed to be the first two instructions.
589 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
590 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
592 /* All programs must keep CTX in callee saved BPF_REG_CTX.
593 * In eBPF case it's done by the compiler, here we need to
594 * do this ourself. Initial CTX is present in BPF_REG_ARG1.
596 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
598 /* For packet access in classic BPF, cache skb->data
599 * in callee-saved BPF R8 and skb->len - skb->data_len
600 * (headlen) in BPF R9. Since classic BPF is read-only
601 * on CTX, we only need to cache it once.
603 *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
604 BPF_REG_D, BPF_REG_CTX,
605 offsetof(struct sk_buff, data));
606 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
607 offsetof(struct sk_buff, len));
608 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
609 offsetof(struct sk_buff, data_len));
610 *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
616 for (i = 0; i < len; fp++, i++) {
617 struct bpf_insn tmp_insns[32] = { };
618 struct bpf_insn *insn = tmp_insns;
621 addrs[i] = new_insn - first_insn;
624 /* All arithmetic insns and skb loads map as-is. */
625 case BPF_ALU | BPF_ADD | BPF_X:
626 case BPF_ALU | BPF_ADD | BPF_K:
627 case BPF_ALU | BPF_SUB | BPF_X:
628 case BPF_ALU | BPF_SUB | BPF_K:
629 case BPF_ALU | BPF_AND | BPF_X:
630 case BPF_ALU | BPF_AND | BPF_K:
631 case BPF_ALU | BPF_OR | BPF_X:
632 case BPF_ALU | BPF_OR | BPF_K:
633 case BPF_ALU | BPF_LSH | BPF_X:
634 case BPF_ALU | BPF_LSH | BPF_K:
635 case BPF_ALU | BPF_RSH | BPF_X:
636 case BPF_ALU | BPF_RSH | BPF_K:
637 case BPF_ALU | BPF_XOR | BPF_X:
638 case BPF_ALU | BPF_XOR | BPF_K:
639 case BPF_ALU | BPF_MUL | BPF_X:
640 case BPF_ALU | BPF_MUL | BPF_K:
641 case BPF_ALU | BPF_DIV | BPF_X:
642 case BPF_ALU | BPF_DIV | BPF_K:
643 case BPF_ALU | BPF_MOD | BPF_X:
644 case BPF_ALU | BPF_MOD | BPF_K:
645 case BPF_ALU | BPF_NEG:
646 case BPF_LD | BPF_ABS | BPF_W:
647 case BPF_LD | BPF_ABS | BPF_H:
648 case BPF_LD | BPF_ABS | BPF_B:
649 case BPF_LD | BPF_IND | BPF_W:
650 case BPF_LD | BPF_IND | BPF_H:
651 case BPF_LD | BPF_IND | BPF_B:
652 /* Check for overloaded BPF extension and
653 * directly convert it if found, otherwise
654 * just move on with mapping.
656 if (BPF_CLASS(fp->code) == BPF_LD &&
657 BPF_MODE(fp->code) == BPF_ABS &&
658 convert_bpf_extensions(fp, &insn))
660 if (BPF_CLASS(fp->code) == BPF_LD &&
661 convert_bpf_ld_abs(fp, &insn)) {
666 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
667 fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
668 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
669 /* Error with exception code on div/mod by 0.
670 * For cBPF programs, this was always return 0.
672 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
673 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
674 *insn++ = BPF_EXIT_INSN();
677 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
680 /* Jump transformation cannot use BPF block macros
681 * everywhere as offset calculation and target updates
682 * require a bit more work than the rest, i.e. jump
683 * opcodes map as-is, but offsets need adjustment.
686 #define BPF_EMIT_JMP \
688 const s32 off_min = S16_MIN, off_max = S16_MAX; \
691 if (target >= len || target < 0) \
693 off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
694 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
695 off -= insn - tmp_insns; \
696 /* Reject anything not fitting into insn->off. */ \
697 if (off < off_min || off > off_max) \
702 case BPF_JMP | BPF_JA:
703 target = i + fp->k + 1;
704 insn->code = fp->code;
708 case BPF_JMP | BPF_JEQ | BPF_K:
709 case BPF_JMP | BPF_JEQ | BPF_X:
710 case BPF_JMP | BPF_JSET | BPF_K:
711 case BPF_JMP | BPF_JSET | BPF_X:
712 case BPF_JMP | BPF_JGT | BPF_K:
713 case BPF_JMP | BPF_JGT | BPF_X:
714 case BPF_JMP | BPF_JGE | BPF_K:
715 case BPF_JMP | BPF_JGE | BPF_X:
716 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
717 /* BPF immediates are signed, zero extend
718 * immediate into tmp register and use it
721 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
723 insn->dst_reg = BPF_REG_A;
724 insn->src_reg = BPF_REG_TMP;
727 insn->dst_reg = BPF_REG_A;
729 bpf_src = BPF_SRC(fp->code);
730 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
733 /* Common case where 'jump_false' is next insn. */
735 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
736 target = i + fp->jt + 1;
741 /* Convert some jumps when 'jump_true' is next insn. */
743 switch (BPF_OP(fp->code)) {
745 insn->code = BPF_JMP | BPF_JNE | bpf_src;
748 insn->code = BPF_JMP | BPF_JLE | bpf_src;
751 insn->code = BPF_JMP | BPF_JLT | bpf_src;
757 target = i + fp->jf + 1;
762 /* Other jumps are mapped into two insns: Jxx and JA. */
763 target = i + fp->jt + 1;
764 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
768 insn->code = BPF_JMP | BPF_JA;
769 target = i + fp->jf + 1;
773 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
774 case BPF_LDX | BPF_MSH | BPF_B: {
775 struct sock_filter tmp = {
776 .code = BPF_LD | BPF_ABS | BPF_B,
783 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
784 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
785 convert_bpf_ld_abs(&tmp, &insn);
788 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
790 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
792 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
794 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
796 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
799 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
800 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
802 case BPF_RET | BPF_A:
803 case BPF_RET | BPF_K:
804 if (BPF_RVAL(fp->code) == BPF_K)
805 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
807 *insn = BPF_EXIT_INSN();
810 /* Store to stack. */
813 stack_off = fp->k * 4 + 4;
814 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
815 BPF_ST ? BPF_REG_A : BPF_REG_X,
817 /* check_load_and_stores() verifies that classic BPF can
818 * load from stack only after write, so tracking
819 * stack_depth for ST|STX insns is enough
821 if (new_prog && new_prog->aux->stack_depth < stack_off)
822 new_prog->aux->stack_depth = stack_off;
825 /* Load from stack. */
826 case BPF_LD | BPF_MEM:
827 case BPF_LDX | BPF_MEM:
828 stack_off = fp->k * 4 + 4;
829 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
830 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
835 case BPF_LD | BPF_IMM:
836 case BPF_LDX | BPF_IMM:
837 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
838 BPF_REG_A : BPF_REG_X, fp->k);
842 case BPF_MISC | BPF_TAX:
843 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
847 case BPF_MISC | BPF_TXA:
848 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
851 /* A = skb->len or X = skb->len */
852 case BPF_LD | BPF_W | BPF_LEN:
853 case BPF_LDX | BPF_W | BPF_LEN:
854 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
855 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
856 offsetof(struct sk_buff, len));
859 /* Access seccomp_data fields. */
860 case BPF_LDX | BPF_ABS | BPF_W:
861 /* A = *(u32 *) (ctx + K) */
862 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
865 /* Unknown instruction. */
872 memcpy(new_insn, tmp_insns,
873 sizeof(*insn) * (insn - tmp_insns));
874 new_insn += insn - tmp_insns;
878 /* Only calculating new length. */
879 *new_len = new_insn - first_insn;
881 *new_len += 4; /* Prologue bits. */
886 if (new_flen != new_insn - first_insn) {
887 new_flen = new_insn - first_insn;
894 BUG_ON(*new_len != new_flen);
903 * As we dont want to clear mem[] array for each packet going through
904 * __bpf_prog_run(), we check that filter loaded by user never try to read
905 * a cell if not previously written, and we check all branches to be sure
906 * a malicious user doesn't try to abuse us.
908 static int check_load_and_stores(const struct sock_filter *filter, int flen)
910 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
913 BUILD_BUG_ON(BPF_MEMWORDS > 16);
915 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
919 memset(masks, 0xff, flen * sizeof(*masks));
921 for (pc = 0; pc < flen; pc++) {
922 memvalid &= masks[pc];
924 switch (filter[pc].code) {
927 memvalid |= (1 << filter[pc].k);
929 case BPF_LD | BPF_MEM:
930 case BPF_LDX | BPF_MEM:
931 if (!(memvalid & (1 << filter[pc].k))) {
936 case BPF_JMP | BPF_JA:
937 /* A jump must set masks on target */
938 masks[pc + 1 + filter[pc].k] &= memvalid;
941 case BPF_JMP | BPF_JEQ | BPF_K:
942 case BPF_JMP | BPF_JEQ | BPF_X:
943 case BPF_JMP | BPF_JGE | BPF_K:
944 case BPF_JMP | BPF_JGE | BPF_X:
945 case BPF_JMP | BPF_JGT | BPF_K:
946 case BPF_JMP | BPF_JGT | BPF_X:
947 case BPF_JMP | BPF_JSET | BPF_K:
948 case BPF_JMP | BPF_JSET | BPF_X:
949 /* A jump must set masks on targets */
950 masks[pc + 1 + filter[pc].jt] &= memvalid;
951 masks[pc + 1 + filter[pc].jf] &= memvalid;
961 static bool chk_code_allowed(u16 code_to_probe)
963 static const bool codes[] = {
964 /* 32 bit ALU operations */
965 [BPF_ALU | BPF_ADD | BPF_K] = true,
966 [BPF_ALU | BPF_ADD | BPF_X] = true,
967 [BPF_ALU | BPF_SUB | BPF_K] = true,
968 [BPF_ALU | BPF_SUB | BPF_X] = true,
969 [BPF_ALU | BPF_MUL | BPF_K] = true,
970 [BPF_ALU | BPF_MUL | BPF_X] = true,
971 [BPF_ALU | BPF_DIV | BPF_K] = true,
972 [BPF_ALU | BPF_DIV | BPF_X] = true,
973 [BPF_ALU | BPF_MOD | BPF_K] = true,
974 [BPF_ALU | BPF_MOD | BPF_X] = true,
975 [BPF_ALU | BPF_AND | BPF_K] = true,
976 [BPF_ALU | BPF_AND | BPF_X] = true,
977 [BPF_ALU | BPF_OR | BPF_K] = true,
978 [BPF_ALU | BPF_OR | BPF_X] = true,
979 [BPF_ALU | BPF_XOR | BPF_K] = true,
980 [BPF_ALU | BPF_XOR | BPF_X] = true,
981 [BPF_ALU | BPF_LSH | BPF_K] = true,
982 [BPF_ALU | BPF_LSH | BPF_X] = true,
983 [BPF_ALU | BPF_RSH | BPF_K] = true,
984 [BPF_ALU | BPF_RSH | BPF_X] = true,
985 [BPF_ALU | BPF_NEG] = true,
986 /* Load instructions */
987 [BPF_LD | BPF_W | BPF_ABS] = true,
988 [BPF_LD | BPF_H | BPF_ABS] = true,
989 [BPF_LD | BPF_B | BPF_ABS] = true,
990 [BPF_LD | BPF_W | BPF_LEN] = true,
991 [BPF_LD | BPF_W | BPF_IND] = true,
992 [BPF_LD | BPF_H | BPF_IND] = true,
993 [BPF_LD | BPF_B | BPF_IND] = true,
994 [BPF_LD | BPF_IMM] = true,
995 [BPF_LD | BPF_MEM] = true,
996 [BPF_LDX | BPF_W | BPF_LEN] = true,
997 [BPF_LDX | BPF_B | BPF_MSH] = true,
998 [BPF_LDX | BPF_IMM] = true,
999 [BPF_LDX | BPF_MEM] = true,
1000 /* Store instructions */
1003 /* Misc instructions */
1004 [BPF_MISC | BPF_TAX] = true,
1005 [BPF_MISC | BPF_TXA] = true,
1006 /* Return instructions */
1007 [BPF_RET | BPF_K] = true,
1008 [BPF_RET | BPF_A] = true,
1009 /* Jump instructions */
1010 [BPF_JMP | BPF_JA] = true,
1011 [BPF_JMP | BPF_JEQ | BPF_K] = true,
1012 [BPF_JMP | BPF_JEQ | BPF_X] = true,
1013 [BPF_JMP | BPF_JGE | BPF_K] = true,
1014 [BPF_JMP | BPF_JGE | BPF_X] = true,
1015 [BPF_JMP | BPF_JGT | BPF_K] = true,
1016 [BPF_JMP | BPF_JGT | BPF_X] = true,
1017 [BPF_JMP | BPF_JSET | BPF_K] = true,
1018 [BPF_JMP | BPF_JSET | BPF_X] = true,
1021 if (code_to_probe >= ARRAY_SIZE(codes))
1024 return codes[code_to_probe];
1027 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1032 if (flen == 0 || flen > BPF_MAXINSNS)
1039 * bpf_check_classic - verify socket filter code
1040 * @filter: filter to verify
1041 * @flen: length of filter
1043 * Check the user's filter code. If we let some ugly
1044 * filter code slip through kaboom! The filter must contain
1045 * no references or jumps that are out of range, no illegal
1046 * instructions, and must end with a RET instruction.
1048 * All jumps are forward as they are not signed.
1050 * Returns 0 if the rule set is legal or -EINVAL if not.
1052 static int bpf_check_classic(const struct sock_filter *filter,
1058 /* Check the filter code now */
1059 for (pc = 0; pc < flen; pc++) {
1060 const struct sock_filter *ftest = &filter[pc];
1062 /* May we actually operate on this code? */
1063 if (!chk_code_allowed(ftest->code))
1066 /* Some instructions need special checks */
1067 switch (ftest->code) {
1068 case BPF_ALU | BPF_DIV | BPF_K:
1069 case BPF_ALU | BPF_MOD | BPF_K:
1070 /* Check for division by zero */
1074 case BPF_ALU | BPF_LSH | BPF_K:
1075 case BPF_ALU | BPF_RSH | BPF_K:
1079 case BPF_LD | BPF_MEM:
1080 case BPF_LDX | BPF_MEM:
1083 /* Check for invalid memory addresses */
1084 if (ftest->k >= BPF_MEMWORDS)
1087 case BPF_JMP | BPF_JA:
1088 /* Note, the large ftest->k might cause loops.
1089 * Compare this with conditional jumps below,
1090 * where offsets are limited. --ANK (981016)
1092 if (ftest->k >= (unsigned int)(flen - pc - 1))
1095 case BPF_JMP | BPF_JEQ | BPF_K:
1096 case BPF_JMP | BPF_JEQ | BPF_X:
1097 case BPF_JMP | BPF_JGE | BPF_K:
1098 case BPF_JMP | BPF_JGE | BPF_X:
1099 case BPF_JMP | BPF_JGT | BPF_K:
1100 case BPF_JMP | BPF_JGT | BPF_X:
1101 case BPF_JMP | BPF_JSET | BPF_K:
1102 case BPF_JMP | BPF_JSET | BPF_X:
1103 /* Both conditionals must be safe */
1104 if (pc + ftest->jt + 1 >= flen ||
1105 pc + ftest->jf + 1 >= flen)
1108 case BPF_LD | BPF_W | BPF_ABS:
1109 case BPF_LD | BPF_H | BPF_ABS:
1110 case BPF_LD | BPF_B | BPF_ABS:
1112 if (bpf_anc_helper(ftest) & BPF_ANC)
1114 /* Ancillary operation unknown or unsupported */
1115 if (anc_found == false && ftest->k >= SKF_AD_OFF)
1120 /* Last instruction must be a RET code */
1121 switch (filter[flen - 1].code) {
1122 case BPF_RET | BPF_K:
1123 case BPF_RET | BPF_A:
1124 return check_load_and_stores(filter, flen);
1130 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1131 const struct sock_fprog *fprog)
1133 unsigned int fsize = bpf_classic_proglen(fprog);
1134 struct sock_fprog_kern *fkprog;
1136 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1140 fkprog = fp->orig_prog;
1141 fkprog->len = fprog->len;
1143 fkprog->filter = kmemdup(fp->insns, fsize,
1144 GFP_KERNEL | __GFP_NOWARN);
1145 if (!fkprog->filter) {
1146 kfree(fp->orig_prog);
1153 static void bpf_release_orig_filter(struct bpf_prog *fp)
1155 struct sock_fprog_kern *fprog = fp->orig_prog;
1158 kfree(fprog->filter);
1163 static void __bpf_prog_release(struct bpf_prog *prog)
1165 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1168 bpf_release_orig_filter(prog);
1169 bpf_prog_free(prog);
1173 static void __sk_filter_release(struct sk_filter *fp)
1175 __bpf_prog_release(fp->prog);
1180 * sk_filter_release_rcu - Release a socket filter by rcu_head
1181 * @rcu: rcu_head that contains the sk_filter to free
1183 static void sk_filter_release_rcu(struct rcu_head *rcu)
1185 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1187 __sk_filter_release(fp);
1191 * sk_filter_release - release a socket filter
1192 * @fp: filter to remove
1194 * Remove a filter from a socket and release its resources.
1196 static void sk_filter_release(struct sk_filter *fp)
1198 if (refcount_dec_and_test(&fp->refcnt))
1199 call_rcu(&fp->rcu, sk_filter_release_rcu);
1202 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1204 u32 filter_size = bpf_prog_size(fp->prog->len);
1206 atomic_sub(filter_size, &sk->sk_omem_alloc);
1207 sk_filter_release(fp);
1210 /* try to charge the socket memory if there is space available
1211 * return true on success
1213 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1215 u32 filter_size = bpf_prog_size(fp->prog->len);
1216 int optmem_max = READ_ONCE(sysctl_optmem_max);
1218 /* same check as in sock_kmalloc() */
1219 if (filter_size <= optmem_max &&
1220 atomic_read(&sk->sk_omem_alloc) + filter_size < optmem_max) {
1221 atomic_add(filter_size, &sk->sk_omem_alloc);
1227 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1229 if (!refcount_inc_not_zero(&fp->refcnt))
1232 if (!__sk_filter_charge(sk, fp)) {
1233 sk_filter_release(fp);
1239 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1241 struct sock_filter *old_prog;
1242 struct bpf_prog *old_fp;
1243 int err, new_len, old_len = fp->len;
1244 bool seen_ld_abs = false;
1246 /* We are free to overwrite insns et al right here as it
1247 * won't be used at this point in time anymore internally
1248 * after the migration to the internal BPF instruction
1251 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1252 sizeof(struct bpf_insn));
1254 /* Conversion cannot happen on overlapping memory areas,
1255 * so we need to keep the user BPF around until the 2nd
1256 * pass. At this time, the user BPF is stored in fp->insns.
1258 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1259 GFP_KERNEL | __GFP_NOWARN);
1265 /* 1st pass: calculate the new program length. */
1266 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1271 /* Expand fp for appending the new filter representation. */
1273 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1275 /* The old_fp is still around in case we couldn't
1276 * allocate new memory, so uncharge on that one.
1285 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1286 err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1289 /* 2nd bpf_convert_filter() can fail only if it fails
1290 * to allocate memory, remapping must succeed. Note,
1291 * that at this time old_fp has already been released
1296 fp = bpf_prog_select_runtime(fp, &err);
1306 __bpf_prog_release(fp);
1307 return ERR_PTR(err);
1310 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1311 bpf_aux_classic_check_t trans)
1315 fp->bpf_func = NULL;
1318 err = bpf_check_classic(fp->insns, fp->len);
1320 __bpf_prog_release(fp);
1321 return ERR_PTR(err);
1324 /* There might be additional checks and transformations
1325 * needed on classic filters, f.e. in case of seccomp.
1328 err = trans(fp->insns, fp->len);
1330 __bpf_prog_release(fp);
1331 return ERR_PTR(err);
1335 /* Probe if we can JIT compile the filter and if so, do
1336 * the compilation of the filter.
1338 bpf_jit_compile(fp);
1340 /* JIT compiler couldn't process this filter, so do the
1341 * internal BPF translation for the optimized interpreter.
1344 fp = bpf_migrate_filter(fp);
1350 * bpf_prog_create - create an unattached filter
1351 * @pfp: the unattached filter that is created
1352 * @fprog: the filter program
1354 * Create a filter independent of any socket. We first run some
1355 * sanity checks on it to make sure it does not explode on us later.
1356 * If an error occurs or there is insufficient memory for the filter
1357 * a negative errno code is returned. On success the return is zero.
1359 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1361 unsigned int fsize = bpf_classic_proglen(fprog);
1362 struct bpf_prog *fp;
1364 /* Make sure new filter is there and in the right amounts. */
1365 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1368 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1372 memcpy(fp->insns, fprog->filter, fsize);
1374 fp->len = fprog->len;
1375 /* Since unattached filters are not copied back to user
1376 * space through sk_get_filter(), we do not need to hold
1377 * a copy here, and can spare us the work.
1379 fp->orig_prog = NULL;
1381 /* bpf_prepare_filter() already takes care of freeing
1382 * memory in case something goes wrong.
1384 fp = bpf_prepare_filter(fp, NULL);
1391 EXPORT_SYMBOL_GPL(bpf_prog_create);
1394 * bpf_prog_create_from_user - create an unattached filter from user buffer
1395 * @pfp: the unattached filter that is created
1396 * @fprog: the filter program
1397 * @trans: post-classic verifier transformation handler
1398 * @save_orig: save classic BPF program
1400 * This function effectively does the same as bpf_prog_create(), only
1401 * that it builds up its insns buffer from user space provided buffer.
1402 * It also allows for passing a bpf_aux_classic_check_t handler.
1404 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1405 bpf_aux_classic_check_t trans, bool save_orig)
1407 unsigned int fsize = bpf_classic_proglen(fprog);
1408 struct bpf_prog *fp;
1411 /* Make sure new filter is there and in the right amounts. */
1412 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1415 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1419 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1420 __bpf_prog_free(fp);
1424 fp->len = fprog->len;
1425 fp->orig_prog = NULL;
1428 err = bpf_prog_store_orig_filter(fp, fprog);
1430 __bpf_prog_free(fp);
1435 /* bpf_prepare_filter() already takes care of freeing
1436 * memory in case something goes wrong.
1438 fp = bpf_prepare_filter(fp, trans);
1445 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1447 void bpf_prog_destroy(struct bpf_prog *fp)
1449 __bpf_prog_release(fp);
1451 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1453 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1455 struct sk_filter *fp, *old_fp;
1457 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1463 if (!__sk_filter_charge(sk, fp)) {
1467 refcount_set(&fp->refcnt, 1);
1469 old_fp = rcu_dereference_protected(sk->sk_filter,
1470 lockdep_sock_is_held(sk));
1471 rcu_assign_pointer(sk->sk_filter, fp);
1474 sk_filter_uncharge(sk, old_fp);
1480 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1482 unsigned int fsize = bpf_classic_proglen(fprog);
1483 struct bpf_prog *prog;
1486 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1487 return ERR_PTR(-EPERM);
1489 /* Make sure new filter is there and in the right amounts. */
1490 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1491 return ERR_PTR(-EINVAL);
1493 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1495 return ERR_PTR(-ENOMEM);
1497 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1498 __bpf_prog_free(prog);
1499 return ERR_PTR(-EFAULT);
1502 prog->len = fprog->len;
1504 err = bpf_prog_store_orig_filter(prog, fprog);
1506 __bpf_prog_free(prog);
1507 return ERR_PTR(-ENOMEM);
1510 /* bpf_prepare_filter() already takes care of freeing
1511 * memory in case something goes wrong.
1513 return bpf_prepare_filter(prog, NULL);
1517 * sk_attach_filter - attach a socket filter
1518 * @fprog: the filter program
1519 * @sk: the socket to use
1521 * Attach the user's filter code. We first run some sanity checks on
1522 * it to make sure it does not explode on us later. If an error
1523 * occurs or there is insufficient memory for the filter a negative
1524 * errno code is returned. On success the return is zero.
1526 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1528 struct bpf_prog *prog = __get_filter(fprog, sk);
1532 return PTR_ERR(prog);
1534 err = __sk_attach_prog(prog, sk);
1536 __bpf_prog_release(prog);
1542 EXPORT_SYMBOL_GPL(sk_attach_filter);
1544 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1546 struct bpf_prog *prog = __get_filter(fprog, sk);
1550 return PTR_ERR(prog);
1552 if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max))
1555 err = reuseport_attach_prog(sk, prog);
1558 __bpf_prog_release(prog);
1563 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1565 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1566 return ERR_PTR(-EPERM);
1568 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1571 int sk_attach_bpf(u32 ufd, struct sock *sk)
1573 struct bpf_prog *prog = __get_bpf(ufd, sk);
1577 return PTR_ERR(prog);
1579 err = __sk_attach_prog(prog, sk);
1588 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1590 struct bpf_prog *prog;
1593 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1596 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1597 if (PTR_ERR(prog) == -EINVAL)
1598 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1600 return PTR_ERR(prog);
1602 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1603 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1604 * bpf prog (e.g. sockmap). It depends on the
1605 * limitation imposed by bpf_prog_load().
1606 * Hence, sysctl_optmem_max is not checked.
1608 if ((sk->sk_type != SOCK_STREAM &&
1609 sk->sk_type != SOCK_DGRAM) ||
1610 (sk->sk_protocol != IPPROTO_UDP &&
1611 sk->sk_protocol != IPPROTO_TCP) ||
1612 (sk->sk_family != AF_INET &&
1613 sk->sk_family != AF_INET6)) {
1618 /* BPF_PROG_TYPE_SOCKET_FILTER */
1619 if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max)) {
1625 err = reuseport_attach_prog(sk, prog);
1633 void sk_reuseport_prog_free(struct bpf_prog *prog)
1638 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1641 bpf_prog_destroy(prog);
1644 struct bpf_scratchpad {
1646 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1647 u8 buff[MAX_BPF_STACK];
1651 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1653 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1654 unsigned int write_len)
1656 return skb_ensure_writable(skb, write_len);
1659 static inline int bpf_try_make_writable(struct sk_buff *skb,
1660 unsigned int write_len)
1662 int err = __bpf_try_make_writable(skb, write_len);
1664 bpf_compute_data_pointers(skb);
1668 static int bpf_try_make_head_writable(struct sk_buff *skb)
1670 return bpf_try_make_writable(skb, skb_headlen(skb));
1673 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1675 if (skb_at_tc_ingress(skb))
1676 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1679 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1681 if (skb_at_tc_ingress(skb))
1682 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1685 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1686 const void *, from, u32, len, u64, flags)
1690 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1692 if (unlikely(offset > INT_MAX))
1694 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1697 ptr = skb->data + offset;
1698 if (flags & BPF_F_RECOMPUTE_CSUM)
1699 __skb_postpull_rcsum(skb, ptr, len, offset);
1701 memcpy(ptr, from, len);
1703 if (flags & BPF_F_RECOMPUTE_CSUM)
1704 __skb_postpush_rcsum(skb, ptr, len, offset);
1705 if (flags & BPF_F_INVALIDATE_HASH)
1706 skb_clear_hash(skb);
1711 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1712 .func = bpf_skb_store_bytes,
1714 .ret_type = RET_INTEGER,
1715 .arg1_type = ARG_PTR_TO_CTX,
1716 .arg2_type = ARG_ANYTHING,
1717 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1718 .arg4_type = ARG_CONST_SIZE,
1719 .arg5_type = ARG_ANYTHING,
1722 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1723 void *, to, u32, len)
1727 if (unlikely(offset > INT_MAX))
1730 ptr = skb_header_pointer(skb, offset, len, to);
1734 memcpy(to, ptr, len);
1742 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1743 .func = bpf_skb_load_bytes,
1745 .ret_type = RET_INTEGER,
1746 .arg1_type = ARG_PTR_TO_CTX,
1747 .arg2_type = ARG_ANYTHING,
1748 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1749 .arg4_type = ARG_CONST_SIZE,
1752 BPF_CALL_4(bpf_flow_dissector_load_bytes,
1753 const struct bpf_flow_dissector *, ctx, u32, offset,
1754 void *, to, u32, len)
1758 if (unlikely(offset > 0xffff))
1761 if (unlikely(!ctx->skb))
1764 ptr = skb_header_pointer(ctx->skb, offset, len, to);
1768 memcpy(to, ptr, len);
1776 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1777 .func = bpf_flow_dissector_load_bytes,
1779 .ret_type = RET_INTEGER,
1780 .arg1_type = ARG_PTR_TO_CTX,
1781 .arg2_type = ARG_ANYTHING,
1782 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1783 .arg4_type = ARG_CONST_SIZE,
1786 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1787 u32, offset, void *, to, u32, len, u32, start_header)
1789 u8 *end = skb_tail_pointer(skb);
1792 if (unlikely(offset > 0xffff))
1795 switch (start_header) {
1796 case BPF_HDR_START_MAC:
1797 if (unlikely(!skb_mac_header_was_set(skb)))
1799 start = skb_mac_header(skb);
1801 case BPF_HDR_START_NET:
1802 start = skb_network_header(skb);
1808 ptr = start + offset;
1810 if (likely(ptr + len <= end)) {
1811 memcpy(to, ptr, len);
1820 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1821 .func = bpf_skb_load_bytes_relative,
1823 .ret_type = RET_INTEGER,
1824 .arg1_type = ARG_PTR_TO_CTX,
1825 .arg2_type = ARG_ANYTHING,
1826 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1827 .arg4_type = ARG_CONST_SIZE,
1828 .arg5_type = ARG_ANYTHING,
1831 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1833 /* Idea is the following: should the needed direct read/write
1834 * test fail during runtime, we can pull in more data and redo
1835 * again, since implicitly, we invalidate previous checks here.
1837 * Or, since we know how much we need to make read/writeable,
1838 * this can be done once at the program beginning for direct
1839 * access case. By this we overcome limitations of only current
1840 * headroom being accessible.
1842 return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1845 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1846 .func = bpf_skb_pull_data,
1848 .ret_type = RET_INTEGER,
1849 .arg1_type = ARG_PTR_TO_CTX,
1850 .arg2_type = ARG_ANYTHING,
1853 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1855 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1858 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1859 .func = bpf_sk_fullsock,
1861 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
1862 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
1865 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1866 unsigned int write_len)
1868 return __bpf_try_make_writable(skb, write_len);
1871 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1873 /* Idea is the following: should the needed direct read/write
1874 * test fail during runtime, we can pull in more data and redo
1875 * again, since implicitly, we invalidate previous checks here.
1877 * Or, since we know how much we need to make read/writeable,
1878 * this can be done once at the program beginning for direct
1879 * access case. By this we overcome limitations of only current
1880 * headroom being accessible.
1882 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1885 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1886 .func = sk_skb_pull_data,
1888 .ret_type = RET_INTEGER,
1889 .arg1_type = ARG_PTR_TO_CTX,
1890 .arg2_type = ARG_ANYTHING,
1893 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1894 u64, from, u64, to, u64, flags)
1898 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1900 if (unlikely(offset > 0xffff || offset & 1))
1902 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1905 ptr = (__sum16 *)(skb->data + offset);
1906 switch (flags & BPF_F_HDR_FIELD_MASK) {
1908 if (unlikely(from != 0))
1911 csum_replace_by_diff(ptr, to);
1914 csum_replace2(ptr, from, to);
1917 csum_replace4(ptr, from, to);
1926 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1927 .func = bpf_l3_csum_replace,
1929 .ret_type = RET_INTEGER,
1930 .arg1_type = ARG_PTR_TO_CTX,
1931 .arg2_type = ARG_ANYTHING,
1932 .arg3_type = ARG_ANYTHING,
1933 .arg4_type = ARG_ANYTHING,
1934 .arg5_type = ARG_ANYTHING,
1937 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1938 u64, from, u64, to, u64, flags)
1940 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1941 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1942 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1945 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1946 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1948 if (unlikely(offset > 0xffff || offset & 1))
1950 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1953 ptr = (__sum16 *)(skb->data + offset);
1954 if (is_mmzero && !do_mforce && !*ptr)
1957 switch (flags & BPF_F_HDR_FIELD_MASK) {
1959 if (unlikely(from != 0))
1962 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1965 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1968 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1974 if (is_mmzero && !*ptr)
1975 *ptr = CSUM_MANGLED_0;
1979 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1980 .func = bpf_l4_csum_replace,
1982 .ret_type = RET_INTEGER,
1983 .arg1_type = ARG_PTR_TO_CTX,
1984 .arg2_type = ARG_ANYTHING,
1985 .arg3_type = ARG_ANYTHING,
1986 .arg4_type = ARG_ANYTHING,
1987 .arg5_type = ARG_ANYTHING,
1990 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1991 __be32 *, to, u32, to_size, __wsum, seed)
1993 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1994 u32 diff_size = from_size + to_size;
1997 /* This is quite flexible, some examples:
1999 * from_size == 0, to_size > 0, seed := csum --> pushing data
2000 * from_size > 0, to_size == 0, seed := csum --> pulling data
2001 * from_size > 0, to_size > 0, seed := 0 --> diffing data
2003 * Even for diffing, from_size and to_size don't need to be equal.
2005 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
2006 diff_size > sizeof(sp->diff)))
2009 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
2010 sp->diff[j] = ~from[i];
2011 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
2012 sp->diff[j] = to[i];
2014 return csum_partial(sp->diff, diff_size, seed);
2017 static const struct bpf_func_proto bpf_csum_diff_proto = {
2018 .func = bpf_csum_diff,
2021 .ret_type = RET_INTEGER,
2022 .arg1_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2023 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
2024 .arg3_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2025 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
2026 .arg5_type = ARG_ANYTHING,
2029 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2031 /* The interface is to be used in combination with bpf_csum_diff()
2032 * for direct packet writes. csum rotation for alignment as well
2033 * as emulating csum_sub() can be done from the eBPF program.
2035 if (skb->ip_summed == CHECKSUM_COMPLETE)
2036 return (skb->csum = csum_add(skb->csum, csum));
2041 static const struct bpf_func_proto bpf_csum_update_proto = {
2042 .func = bpf_csum_update,
2044 .ret_type = RET_INTEGER,
2045 .arg1_type = ARG_PTR_TO_CTX,
2046 .arg2_type = ARG_ANYTHING,
2049 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
2051 /* The interface is to be used in combination with bpf_skb_adjust_room()
2052 * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET
2053 * is passed as flags, for example.
2056 case BPF_CSUM_LEVEL_INC:
2057 __skb_incr_checksum_unnecessary(skb);
2059 case BPF_CSUM_LEVEL_DEC:
2060 __skb_decr_checksum_unnecessary(skb);
2062 case BPF_CSUM_LEVEL_RESET:
2063 __skb_reset_checksum_unnecessary(skb);
2065 case BPF_CSUM_LEVEL_QUERY:
2066 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2067 skb->csum_level : -EACCES;
2075 static const struct bpf_func_proto bpf_csum_level_proto = {
2076 .func = bpf_csum_level,
2078 .ret_type = RET_INTEGER,
2079 .arg1_type = ARG_PTR_TO_CTX,
2080 .arg2_type = ARG_ANYTHING,
2083 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2085 return dev_forward_skb_nomtu(dev, skb);
2088 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2089 struct sk_buff *skb)
2091 int ret = ____dev_forward_skb(dev, skb, false);
2095 ret = netif_rx(skb);
2101 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2105 if (dev_xmit_recursion()) {
2106 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2114 dev_xmit_recursion_inc();
2115 ret = dev_queue_xmit(skb);
2116 dev_xmit_recursion_dec();
2121 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2124 unsigned int mlen = skb_network_offset(skb);
2127 __skb_pull(skb, mlen);
2129 /* At ingress, the mac header has already been pulled once.
2130 * At egress, skb_pospull_rcsum has to be done in case that
2131 * the skb is originated from ingress (i.e. a forwarded skb)
2132 * to ensure that rcsum starts at net header.
2134 if (!skb_at_tc_ingress(skb))
2135 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2137 skb_pop_mac_header(skb);
2138 skb_reset_mac_len(skb);
2139 return flags & BPF_F_INGRESS ?
2140 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2143 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2146 /* Verify that a link layer header is carried */
2147 if (unlikely(skb->mac_header >= skb->network_header)) {
2152 bpf_push_mac_rcsum(skb);
2153 return flags & BPF_F_INGRESS ?
2154 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2157 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2160 if (dev_is_mac_header_xmit(dev))
2161 return __bpf_redirect_common(skb, dev, flags);
2163 return __bpf_redirect_no_mac(skb, dev, flags);
2166 #if IS_ENABLED(CONFIG_IPV6)
2167 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2168 struct net_device *dev, struct bpf_nh_params *nh)
2170 u32 hh_len = LL_RESERVED_SPACE(dev);
2171 const struct in6_addr *nexthop;
2172 struct dst_entry *dst = NULL;
2173 struct neighbour *neigh;
2175 if (dev_xmit_recursion()) {
2176 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2183 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2184 skb = skb_expand_head(skb, hh_len);
2192 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2193 &ipv6_hdr(skb)->daddr);
2195 nexthop = &nh->ipv6_nh;
2197 neigh = ip_neigh_gw6(dev, nexthop);
2198 if (likely(!IS_ERR(neigh))) {
2201 sock_confirm_neigh(skb, neigh);
2202 dev_xmit_recursion_inc();
2203 ret = neigh_output(neigh, skb, false);
2204 dev_xmit_recursion_dec();
2205 rcu_read_unlock_bh();
2208 rcu_read_unlock_bh();
2210 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2216 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2217 struct bpf_nh_params *nh)
2219 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2220 struct net *net = dev_net(dev);
2221 int err, ret = NET_XMIT_DROP;
2224 struct dst_entry *dst;
2225 struct flowi6 fl6 = {
2226 .flowi6_flags = FLOWI_FLAG_ANYSRC,
2227 .flowi6_mark = skb->mark,
2228 .flowlabel = ip6_flowinfo(ip6h),
2229 .flowi6_oif = dev->ifindex,
2230 .flowi6_proto = ip6h->nexthdr,
2231 .daddr = ip6h->daddr,
2232 .saddr = ip6h->saddr,
2235 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2239 skb_dst_set(skb, dst);
2240 } else if (nh->nh_family != AF_INET6) {
2244 err = bpf_out_neigh_v6(net, skb, dev, nh);
2245 if (unlikely(net_xmit_eval(err)))
2246 dev->stats.tx_errors++;
2248 ret = NET_XMIT_SUCCESS;
2251 dev->stats.tx_errors++;
2257 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2258 struct bpf_nh_params *nh)
2261 return NET_XMIT_DROP;
2263 #endif /* CONFIG_IPV6 */
2265 #if IS_ENABLED(CONFIG_INET)
2266 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2267 struct net_device *dev, struct bpf_nh_params *nh)
2269 u32 hh_len = LL_RESERVED_SPACE(dev);
2270 struct neighbour *neigh;
2271 bool is_v6gw = false;
2273 if (dev_xmit_recursion()) {
2274 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2281 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2282 skb = skb_expand_head(skb, hh_len);
2289 struct dst_entry *dst = skb_dst(skb);
2290 struct rtable *rt = container_of(dst, struct rtable, dst);
2292 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2293 } else if (nh->nh_family == AF_INET6) {
2294 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2296 } else if (nh->nh_family == AF_INET) {
2297 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2299 rcu_read_unlock_bh();
2303 if (likely(!IS_ERR(neigh))) {
2306 sock_confirm_neigh(skb, neigh);
2307 dev_xmit_recursion_inc();
2308 ret = neigh_output(neigh, skb, is_v6gw);
2309 dev_xmit_recursion_dec();
2310 rcu_read_unlock_bh();
2313 rcu_read_unlock_bh();
2319 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2320 struct bpf_nh_params *nh)
2322 const struct iphdr *ip4h = ip_hdr(skb);
2323 struct net *net = dev_net(dev);
2324 int err, ret = NET_XMIT_DROP;
2327 struct flowi4 fl4 = {
2328 .flowi4_flags = FLOWI_FLAG_ANYSRC,
2329 .flowi4_mark = skb->mark,
2330 .flowi4_tos = RT_TOS(ip4h->tos),
2331 .flowi4_oif = dev->ifindex,
2332 .flowi4_proto = ip4h->protocol,
2333 .daddr = ip4h->daddr,
2334 .saddr = ip4h->saddr,
2338 rt = ip_route_output_flow(net, &fl4, NULL);
2341 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2346 skb_dst_set(skb, &rt->dst);
2349 err = bpf_out_neigh_v4(net, skb, dev, nh);
2350 if (unlikely(net_xmit_eval(err)))
2351 dev->stats.tx_errors++;
2353 ret = NET_XMIT_SUCCESS;
2356 dev->stats.tx_errors++;
2362 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2363 struct bpf_nh_params *nh)
2366 return NET_XMIT_DROP;
2368 #endif /* CONFIG_INET */
2370 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2371 struct bpf_nh_params *nh)
2373 struct ethhdr *ethh = eth_hdr(skb);
2375 if (unlikely(skb->mac_header >= skb->network_header))
2377 bpf_push_mac_rcsum(skb);
2378 if (is_multicast_ether_addr(ethh->h_dest))
2381 skb_pull(skb, sizeof(*ethh));
2382 skb_unset_mac_header(skb);
2383 skb_reset_network_header(skb);
2385 if (skb->protocol == htons(ETH_P_IP))
2386 return __bpf_redirect_neigh_v4(skb, dev, nh);
2387 else if (skb->protocol == htons(ETH_P_IPV6))
2388 return __bpf_redirect_neigh_v6(skb, dev, nh);
2394 /* Internal, non-exposed redirect flags. */
2396 BPF_F_NEIGH = (1ULL << 1),
2397 BPF_F_PEER = (1ULL << 2),
2398 BPF_F_NEXTHOP = (1ULL << 3),
2399 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2402 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2404 struct net_device *dev;
2405 struct sk_buff *clone;
2408 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2411 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2415 clone = skb_clone(skb, GFP_ATOMIC);
2416 if (unlikely(!clone))
2419 /* For direct write, we need to keep the invariant that the skbs
2420 * we're dealing with need to be uncloned. Should uncloning fail
2421 * here, we need to free the just generated clone to unclone once
2424 ret = bpf_try_make_head_writable(skb);
2425 if (unlikely(ret)) {
2430 return __bpf_redirect(clone, dev, flags);
2433 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2434 .func = bpf_clone_redirect,
2436 .ret_type = RET_INTEGER,
2437 .arg1_type = ARG_PTR_TO_CTX,
2438 .arg2_type = ARG_ANYTHING,
2439 .arg3_type = ARG_ANYTHING,
2442 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2443 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2445 int skb_do_redirect(struct sk_buff *skb)
2447 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2448 struct net *net = dev_net(skb->dev);
2449 struct net_device *dev;
2450 u32 flags = ri->flags;
2452 dev = dev_get_by_index_rcu(net, ri->tgt_index);
2457 if (flags & BPF_F_PEER) {
2458 const struct net_device_ops *ops = dev->netdev_ops;
2460 if (unlikely(!ops->ndo_get_peer_dev ||
2461 !skb_at_tc_ingress(skb)))
2463 dev = ops->ndo_get_peer_dev(dev);
2464 if (unlikely(!dev ||
2465 !(dev->flags & IFF_UP) ||
2466 net_eq(net, dev_net(dev))))
2471 return flags & BPF_F_NEIGH ?
2472 __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2474 __bpf_redirect(skb, dev, flags);
2480 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2482 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2484 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2488 ri->tgt_index = ifindex;
2490 return TC_ACT_REDIRECT;
2493 static const struct bpf_func_proto bpf_redirect_proto = {
2494 .func = bpf_redirect,
2496 .ret_type = RET_INTEGER,
2497 .arg1_type = ARG_ANYTHING,
2498 .arg2_type = ARG_ANYTHING,
2501 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2503 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2505 if (unlikely(flags))
2508 ri->flags = BPF_F_PEER;
2509 ri->tgt_index = ifindex;
2511 return TC_ACT_REDIRECT;
2514 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2515 .func = bpf_redirect_peer,
2517 .ret_type = RET_INTEGER,
2518 .arg1_type = ARG_ANYTHING,
2519 .arg2_type = ARG_ANYTHING,
2522 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2523 int, plen, u64, flags)
2525 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2527 if (unlikely((plen && plen < sizeof(*params)) || flags))
2530 ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2531 ri->tgt_index = ifindex;
2533 BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2535 memcpy(&ri->nh, params, sizeof(ri->nh));
2537 return TC_ACT_REDIRECT;
2540 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2541 .func = bpf_redirect_neigh,
2543 .ret_type = RET_INTEGER,
2544 .arg1_type = ARG_ANYTHING,
2545 .arg2_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2546 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
2547 .arg4_type = ARG_ANYTHING,
2550 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2552 msg->apply_bytes = bytes;
2556 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2557 .func = bpf_msg_apply_bytes,
2559 .ret_type = RET_INTEGER,
2560 .arg1_type = ARG_PTR_TO_CTX,
2561 .arg2_type = ARG_ANYTHING,
2564 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2566 msg->cork_bytes = bytes;
2570 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2571 .func = bpf_msg_cork_bytes,
2573 .ret_type = RET_INTEGER,
2574 .arg1_type = ARG_PTR_TO_CTX,
2575 .arg2_type = ARG_ANYTHING,
2578 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2579 u32, end, u64, flags)
2581 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2582 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2583 struct scatterlist *sge;
2584 u8 *raw, *to, *from;
2587 if (unlikely(flags || end <= start))
2590 /* First find the starting scatterlist element */
2594 len = sk_msg_elem(msg, i)->length;
2595 if (start < offset + len)
2597 sk_msg_iter_var_next(i);
2598 } while (i != msg->sg.end);
2600 if (unlikely(start >= offset + len))
2604 /* The start may point into the sg element so we need to also
2605 * account for the headroom.
2607 bytes_sg_total = start - offset + bytes;
2608 if (!test_bit(i, &msg->sg.copy) && bytes_sg_total <= len)
2611 /* At this point we need to linearize multiple scatterlist
2612 * elements or a single shared page. Either way we need to
2613 * copy into a linear buffer exclusively owned by BPF. Then
2614 * place the buffer in the scatterlist and fixup the original
2615 * entries by removing the entries now in the linear buffer
2616 * and shifting the remaining entries. For now we do not try
2617 * to copy partial entries to avoid complexity of running out
2618 * of sg_entry slots. The downside is reading a single byte
2619 * will copy the entire sg entry.
2622 copy += sk_msg_elem(msg, i)->length;
2623 sk_msg_iter_var_next(i);
2624 if (bytes_sg_total <= copy)
2626 } while (i != msg->sg.end);
2629 if (unlikely(bytes_sg_total > copy))
2632 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2634 if (unlikely(!page))
2637 raw = page_address(page);
2640 sge = sk_msg_elem(msg, i);
2641 from = sg_virt(sge);
2645 memcpy(to, from, len);
2648 put_page(sg_page(sge));
2650 sk_msg_iter_var_next(i);
2651 } while (i != last_sge);
2653 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2655 /* To repair sg ring we need to shift entries. If we only
2656 * had a single entry though we can just replace it and
2657 * be done. Otherwise walk the ring and shift the entries.
2659 WARN_ON_ONCE(last_sge == first_sge);
2660 shift = last_sge > first_sge ?
2661 last_sge - first_sge - 1 :
2662 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2667 sk_msg_iter_var_next(i);
2671 if (i + shift >= NR_MSG_FRAG_IDS)
2672 move_from = i + shift - NR_MSG_FRAG_IDS;
2674 move_from = i + shift;
2675 if (move_from == msg->sg.end)
2678 msg->sg.data[i] = msg->sg.data[move_from];
2679 msg->sg.data[move_from].length = 0;
2680 msg->sg.data[move_from].page_link = 0;
2681 msg->sg.data[move_from].offset = 0;
2682 sk_msg_iter_var_next(i);
2685 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2686 msg->sg.end - shift + NR_MSG_FRAG_IDS :
2687 msg->sg.end - shift;
2689 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2690 msg->data_end = msg->data + bytes;
2694 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2695 .func = bpf_msg_pull_data,
2697 .ret_type = RET_INTEGER,
2698 .arg1_type = ARG_PTR_TO_CTX,
2699 .arg2_type = ARG_ANYTHING,
2700 .arg3_type = ARG_ANYTHING,
2701 .arg4_type = ARG_ANYTHING,
2704 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2705 u32, len, u64, flags)
2707 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2708 u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2709 u8 *raw, *to, *from;
2712 if (unlikely(flags))
2715 if (unlikely(len == 0))
2718 /* First find the starting scatterlist element */
2722 l = sk_msg_elem(msg, i)->length;
2724 if (start < offset + l)
2726 sk_msg_iter_var_next(i);
2727 } while (i != msg->sg.end);
2729 if (start >= offset + l)
2732 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2734 /* If no space available will fallback to copy, we need at
2735 * least one scatterlist elem available to push data into
2736 * when start aligns to the beginning of an element or two
2737 * when it falls inside an element. We handle the start equals
2738 * offset case because its the common case for inserting a
2741 if (!space || (space == 1 && start != offset))
2742 copy = msg->sg.data[i].length;
2744 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2745 get_order(copy + len));
2746 if (unlikely(!page))
2752 raw = page_address(page);
2754 psge = sk_msg_elem(msg, i);
2755 front = start - offset;
2756 back = psge->length - front;
2757 from = sg_virt(psge);
2760 memcpy(raw, from, front);
2764 to = raw + front + len;
2766 memcpy(to, from, back);
2769 put_page(sg_page(psge));
2770 } else if (start - offset) {
2771 psge = sk_msg_elem(msg, i);
2772 rsge = sk_msg_elem_cpy(msg, i);
2774 psge->length = start - offset;
2775 rsge.length -= psge->length;
2776 rsge.offset += start;
2778 sk_msg_iter_var_next(i);
2779 sg_unmark_end(psge);
2780 sg_unmark_end(&rsge);
2781 sk_msg_iter_next(msg, end);
2784 /* Slot(s) to place newly allocated data */
2787 /* Shift one or two slots as needed */
2789 sge = sk_msg_elem_cpy(msg, i);
2791 sk_msg_iter_var_next(i);
2792 sg_unmark_end(&sge);
2793 sk_msg_iter_next(msg, end);
2795 nsge = sk_msg_elem_cpy(msg, i);
2797 sk_msg_iter_var_next(i);
2798 nnsge = sk_msg_elem_cpy(msg, i);
2801 while (i != msg->sg.end) {
2802 msg->sg.data[i] = sge;
2804 sk_msg_iter_var_next(i);
2807 nnsge = sk_msg_elem_cpy(msg, i);
2809 nsge = sk_msg_elem_cpy(msg, i);
2814 /* Place newly allocated data buffer */
2815 sk_mem_charge(msg->sk, len);
2816 msg->sg.size += len;
2817 __clear_bit(new, &msg->sg.copy);
2818 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2820 get_page(sg_page(&rsge));
2821 sk_msg_iter_var_next(new);
2822 msg->sg.data[new] = rsge;
2825 sk_msg_compute_data_pointers(msg);
2829 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2830 .func = bpf_msg_push_data,
2832 .ret_type = RET_INTEGER,
2833 .arg1_type = ARG_PTR_TO_CTX,
2834 .arg2_type = ARG_ANYTHING,
2835 .arg3_type = ARG_ANYTHING,
2836 .arg4_type = ARG_ANYTHING,
2839 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2845 sk_msg_iter_var_next(i);
2846 msg->sg.data[prev] = msg->sg.data[i];
2847 } while (i != msg->sg.end);
2849 sk_msg_iter_prev(msg, end);
2852 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2854 struct scatterlist tmp, sge;
2856 sk_msg_iter_next(msg, end);
2857 sge = sk_msg_elem_cpy(msg, i);
2858 sk_msg_iter_var_next(i);
2859 tmp = sk_msg_elem_cpy(msg, i);
2861 while (i != msg->sg.end) {
2862 msg->sg.data[i] = sge;
2863 sk_msg_iter_var_next(i);
2865 tmp = sk_msg_elem_cpy(msg, i);
2869 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2870 u32, len, u64, flags)
2872 u32 i = 0, l = 0, space, offset = 0;
2873 u64 last = start + len;
2876 if (unlikely(flags))
2879 /* First find the starting scatterlist element */
2883 l = sk_msg_elem(msg, i)->length;
2885 if (start < offset + l)
2887 sk_msg_iter_var_next(i);
2888 } while (i != msg->sg.end);
2890 /* Bounds checks: start and pop must be inside message */
2891 if (start >= offset + l || last >= msg->sg.size)
2894 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2897 /* --------------| offset
2898 * -| start |-------- len -------|
2900 * |----- a ----|-------- pop -------|----- b ----|
2901 * |______________________________________________| length
2904 * a: region at front of scatter element to save
2905 * b: region at back of scatter element to save when length > A + pop
2906 * pop: region to pop from element, same as input 'pop' here will be
2907 * decremented below per iteration.
2909 * Two top-level cases to handle when start != offset, first B is non
2910 * zero and second B is zero corresponding to when a pop includes more
2913 * Then if B is non-zero AND there is no space allocate space and
2914 * compact A, B regions into page. If there is space shift ring to
2915 * the rigth free'ing the next element in ring to place B, leaving
2916 * A untouched except to reduce length.
2918 if (start != offset) {
2919 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2921 int b = sge->length - pop - a;
2923 sk_msg_iter_var_next(i);
2925 if (pop < sge->length - a) {
2928 sk_msg_shift_right(msg, i);
2929 nsge = sk_msg_elem(msg, i);
2930 get_page(sg_page(sge));
2933 b, sge->offset + pop + a);
2935 struct page *page, *orig;
2938 page = alloc_pages(__GFP_NOWARN |
2939 __GFP_COMP | GFP_ATOMIC,
2941 if (unlikely(!page))
2945 orig = sg_page(sge);
2946 from = sg_virt(sge);
2947 to = page_address(page);
2948 memcpy(to, from, a);
2949 memcpy(to + a, from + a + pop, b);
2950 sg_set_page(sge, page, a + b, 0);
2954 } else if (pop >= sge->length - a) {
2955 pop -= (sge->length - a);
2960 /* From above the current layout _must_ be as follows,
2965 * |---- pop ---|---------------- b ------------|
2966 * |____________________________________________| length
2968 * Offset and start of the current msg elem are equal because in the
2969 * previous case we handled offset != start and either consumed the
2970 * entire element and advanced to the next element OR pop == 0.
2972 * Two cases to handle here are first pop is less than the length
2973 * leaving some remainder b above. Simply adjust the element's layout
2974 * in this case. Or pop >= length of the element so that b = 0. In this
2975 * case advance to next element decrementing pop.
2978 struct scatterlist *sge = sk_msg_elem(msg, i);
2980 if (pop < sge->length) {
2986 sk_msg_shift_left(msg, i);
2988 sk_msg_iter_var_next(i);
2991 sk_mem_uncharge(msg->sk, len - pop);
2992 msg->sg.size -= (len - pop);
2993 sk_msg_compute_data_pointers(msg);
2997 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
2998 .func = bpf_msg_pop_data,
3000 .ret_type = RET_INTEGER,
3001 .arg1_type = ARG_PTR_TO_CTX,
3002 .arg2_type = ARG_ANYTHING,
3003 .arg3_type = ARG_ANYTHING,
3004 .arg4_type = ARG_ANYTHING,
3007 #ifdef CONFIG_CGROUP_NET_CLASSID
3008 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3010 return __task_get_classid(current);
3013 static const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3014 .func = bpf_get_cgroup_classid_curr,
3016 .ret_type = RET_INTEGER,
3019 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3021 struct sock *sk = skb_to_full_sk(skb);
3023 if (!sk || !sk_fullsock(sk))
3026 return sock_cgroup_classid(&sk->sk_cgrp_data);
3029 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3030 .func = bpf_skb_cgroup_classid,
3032 .ret_type = RET_INTEGER,
3033 .arg1_type = ARG_PTR_TO_CTX,
3037 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3039 return task_get_classid(skb);
3042 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3043 .func = bpf_get_cgroup_classid,
3045 .ret_type = RET_INTEGER,
3046 .arg1_type = ARG_PTR_TO_CTX,
3049 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3051 return dst_tclassid(skb);
3054 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3055 .func = bpf_get_route_realm,
3057 .ret_type = RET_INTEGER,
3058 .arg1_type = ARG_PTR_TO_CTX,
3061 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3063 /* If skb_clear_hash() was called due to mangling, we can
3064 * trigger SW recalculation here. Later access to hash
3065 * can then use the inline skb->hash via context directly
3066 * instead of calling this helper again.
3068 return skb_get_hash(skb);
3071 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3072 .func = bpf_get_hash_recalc,
3074 .ret_type = RET_INTEGER,
3075 .arg1_type = ARG_PTR_TO_CTX,
3078 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3080 /* After all direct packet write, this can be used once for
3081 * triggering a lazy recalc on next skb_get_hash() invocation.
3083 skb_clear_hash(skb);
3087 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3088 .func = bpf_set_hash_invalid,
3090 .ret_type = RET_INTEGER,
3091 .arg1_type = ARG_PTR_TO_CTX,
3094 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3096 /* Set user specified hash as L4(+), so that it gets returned
3097 * on skb_get_hash() call unless BPF prog later on triggers a
3100 __skb_set_sw_hash(skb, hash, true);
3104 static const struct bpf_func_proto bpf_set_hash_proto = {
3105 .func = bpf_set_hash,
3107 .ret_type = RET_INTEGER,
3108 .arg1_type = ARG_PTR_TO_CTX,
3109 .arg2_type = ARG_ANYTHING,
3112 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3117 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3118 vlan_proto != htons(ETH_P_8021AD)))
3119 vlan_proto = htons(ETH_P_8021Q);
3121 bpf_push_mac_rcsum(skb);
3122 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3123 bpf_pull_mac_rcsum(skb);
3125 bpf_compute_data_pointers(skb);
3129 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3130 .func = bpf_skb_vlan_push,
3132 .ret_type = RET_INTEGER,
3133 .arg1_type = ARG_PTR_TO_CTX,
3134 .arg2_type = ARG_ANYTHING,
3135 .arg3_type = ARG_ANYTHING,
3138 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3142 bpf_push_mac_rcsum(skb);
3143 ret = skb_vlan_pop(skb);
3144 bpf_pull_mac_rcsum(skb);
3146 bpf_compute_data_pointers(skb);
3150 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3151 .func = bpf_skb_vlan_pop,
3153 .ret_type = RET_INTEGER,
3154 .arg1_type = ARG_PTR_TO_CTX,
3157 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3159 /* Caller already did skb_cow() with len as headroom,
3160 * so no need to do it here.
3163 memmove(skb->data, skb->data + len, off);
3164 memset(skb->data + off, 0, len);
3166 /* No skb_postpush_rcsum(skb, skb->data + off, len)
3167 * needed here as it does not change the skb->csum
3168 * result for checksum complete when summing over
3174 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3176 /* skb_ensure_writable() is not needed here, as we're
3177 * already working on an uncloned skb.
3179 if (unlikely(!pskb_may_pull(skb, off + len)))
3182 skb_postpull_rcsum(skb, skb->data + off, len);
3183 memmove(skb->data + len, skb->data, off);
3184 __skb_pull(skb, len);
3189 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3191 bool trans_same = skb->transport_header == skb->network_header;
3194 /* There's no need for __skb_push()/__skb_pull() pair to
3195 * get to the start of the mac header as we're guaranteed
3196 * to always start from here under eBPF.
3198 ret = bpf_skb_generic_push(skb, off, len);
3200 skb->mac_header -= len;
3201 skb->network_header -= len;
3203 skb->transport_header = skb->network_header;
3209 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3211 bool trans_same = skb->transport_header == skb->network_header;
3214 /* Same here, __skb_push()/__skb_pull() pair not needed. */
3215 ret = bpf_skb_generic_pop(skb, off, len);
3217 skb->mac_header += len;
3218 skb->network_header += len;
3220 skb->transport_header = skb->network_header;
3226 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3228 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3229 u32 off = skb_mac_header_len(skb);
3232 ret = skb_cow(skb, len_diff);
3233 if (unlikely(ret < 0))
3236 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3237 if (unlikely(ret < 0))
3240 if (skb_is_gso(skb)) {
3241 struct skb_shared_info *shinfo = skb_shinfo(skb);
3243 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */
3244 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3245 shinfo->gso_type &= ~SKB_GSO_TCPV4;
3246 shinfo->gso_type |= SKB_GSO_TCPV6;
3250 skb->protocol = htons(ETH_P_IPV6);
3251 skb_clear_hash(skb);
3256 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3258 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3259 u32 off = skb_mac_header_len(skb);
3262 ret = skb_unclone(skb, GFP_ATOMIC);
3263 if (unlikely(ret < 0))
3266 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3267 if (unlikely(ret < 0))
3270 if (skb_is_gso(skb)) {
3271 struct skb_shared_info *shinfo = skb_shinfo(skb);
3273 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */
3274 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3275 shinfo->gso_type &= ~SKB_GSO_TCPV6;
3276 shinfo->gso_type |= SKB_GSO_TCPV4;
3280 skb->protocol = htons(ETH_P_IP);
3281 skb_clear_hash(skb);
3286 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3288 __be16 from_proto = skb->protocol;
3290 if (from_proto == htons(ETH_P_IP) &&
3291 to_proto == htons(ETH_P_IPV6))
3292 return bpf_skb_proto_4_to_6(skb);
3294 if (from_proto == htons(ETH_P_IPV6) &&
3295 to_proto == htons(ETH_P_IP))
3296 return bpf_skb_proto_6_to_4(skb);
3301 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3306 if (unlikely(flags))
3309 /* General idea is that this helper does the basic groundwork
3310 * needed for changing the protocol, and eBPF program fills the
3311 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3312 * and other helpers, rather than passing a raw buffer here.
3314 * The rationale is to keep this minimal and without a need to
3315 * deal with raw packet data. F.e. even if we would pass buffers
3316 * here, the program still needs to call the bpf_lX_csum_replace()
3317 * helpers anyway. Plus, this way we keep also separation of
3318 * concerns, since f.e. bpf_skb_store_bytes() should only take
3321 * Currently, additional options and extension header space are
3322 * not supported, but flags register is reserved so we can adapt
3323 * that. For offloads, we mark packet as dodgy, so that headers
3324 * need to be verified first.
3326 ret = bpf_skb_proto_xlat(skb, proto);
3327 bpf_compute_data_pointers(skb);
3331 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3332 .func = bpf_skb_change_proto,
3334 .ret_type = RET_INTEGER,
3335 .arg1_type = ARG_PTR_TO_CTX,
3336 .arg2_type = ARG_ANYTHING,
3337 .arg3_type = ARG_ANYTHING,
3340 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3342 /* We only allow a restricted subset to be changed for now. */
3343 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3344 !skb_pkt_type_ok(pkt_type)))
3347 skb->pkt_type = pkt_type;
3351 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3352 .func = bpf_skb_change_type,
3354 .ret_type = RET_INTEGER,
3355 .arg1_type = ARG_PTR_TO_CTX,
3356 .arg2_type = ARG_ANYTHING,
3359 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3361 switch (skb->protocol) {
3362 case htons(ETH_P_IP):
3363 return sizeof(struct iphdr);
3364 case htons(ETH_P_IPV6):
3365 return sizeof(struct ipv6hdr);
3371 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3372 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3374 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \
3375 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3376 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3377 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3378 BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \
3379 BPF_F_ADJ_ROOM_ENCAP_L2( \
3380 BPF_ADJ_ROOM_ENCAP_L2_MASK))
3382 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3385 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3386 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3387 u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3388 unsigned int gso_type = SKB_GSO_DODGY;
3391 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3392 /* udp gso_size delineates datagrams, only allow if fixed */
3393 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3394 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3398 ret = skb_cow_head(skb, len_diff);
3399 if (unlikely(ret < 0))
3403 if (skb->protocol != htons(ETH_P_IP) &&
3404 skb->protocol != htons(ETH_P_IPV6))
3407 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3408 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3411 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3412 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3415 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH &&
3416 inner_mac_len < ETH_HLEN)
3419 if (skb->encapsulation)
3422 mac_len = skb->network_header - skb->mac_header;
3423 inner_net = skb->network_header;
3424 if (inner_mac_len > len_diff)
3426 inner_trans = skb->transport_header;
3429 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3430 if (unlikely(ret < 0))
3434 skb->inner_mac_header = inner_net - inner_mac_len;
3435 skb->inner_network_header = inner_net;
3436 skb->inner_transport_header = inner_trans;
3438 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH)
3439 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
3441 skb_set_inner_protocol(skb, skb->protocol);
3443 skb->encapsulation = 1;
3444 skb_set_network_header(skb, mac_len);
3446 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3447 gso_type |= SKB_GSO_UDP_TUNNEL;
3448 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3449 gso_type |= SKB_GSO_GRE;
3450 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3451 gso_type |= SKB_GSO_IPXIP6;
3452 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3453 gso_type |= SKB_GSO_IPXIP4;
3455 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3456 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3457 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3458 sizeof(struct ipv6hdr) :
3459 sizeof(struct iphdr);
3461 skb_set_transport_header(skb, mac_len + nh_len);
3464 /* Match skb->protocol to new outer l3 protocol */
3465 if (skb->protocol == htons(ETH_P_IP) &&
3466 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3467 skb->protocol = htons(ETH_P_IPV6);
3468 else if (skb->protocol == htons(ETH_P_IPV6) &&
3469 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3470 skb->protocol = htons(ETH_P_IP);
3473 if (skb_is_gso(skb)) {
3474 struct skb_shared_info *shinfo = skb_shinfo(skb);
3476 /* Due to header grow, MSS needs to be downgraded. */
3477 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3478 skb_decrease_gso_size(shinfo, len_diff);
3480 /* Header must be checked, and gso_segs recomputed. */
3481 shinfo->gso_type |= gso_type;
3482 shinfo->gso_segs = 0;
3488 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3493 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3494 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3497 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3498 /* udp gso_size delineates datagrams, only allow if fixed */
3499 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3500 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3504 ret = skb_unclone(skb, GFP_ATOMIC);
3505 if (unlikely(ret < 0))
3508 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3509 if (unlikely(ret < 0))
3512 if (skb_is_gso(skb)) {
3513 struct skb_shared_info *shinfo = skb_shinfo(skb);
3515 /* Due to header shrink, MSS can be upgraded. */
3516 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3517 skb_increase_gso_size(shinfo, len_diff);
3519 /* Header must be checked, and gso_segs recomputed. */
3520 shinfo->gso_type |= SKB_GSO_DODGY;
3521 shinfo->gso_segs = 0;
3527 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3529 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3530 u32, mode, u64, flags)
3532 u32 len_diff_abs = abs(len_diff);
3533 bool shrink = len_diff < 0;
3536 if (unlikely(flags || mode))
3538 if (unlikely(len_diff_abs > 0xfffU))
3542 ret = skb_cow(skb, len_diff);
3543 if (unlikely(ret < 0))
3545 __skb_push(skb, len_diff_abs);
3546 memset(skb->data, 0, len_diff_abs);
3548 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3550 __skb_pull(skb, len_diff_abs);
3552 if (tls_sw_has_ctx_rx(skb->sk)) {
3553 struct strp_msg *rxm = strp_msg(skb);
3555 rxm->full_len += len_diff;
3560 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3561 .func = sk_skb_adjust_room,
3563 .ret_type = RET_INTEGER,
3564 .arg1_type = ARG_PTR_TO_CTX,
3565 .arg2_type = ARG_ANYTHING,
3566 .arg3_type = ARG_ANYTHING,
3567 .arg4_type = ARG_ANYTHING,
3570 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3571 u32, mode, u64, flags)
3573 u32 len_cur, len_diff_abs = abs(len_diff);
3574 u32 len_min = bpf_skb_net_base_len(skb);
3575 u32 len_max = BPF_SKB_MAX_LEN;
3576 __be16 proto = skb->protocol;
3577 bool shrink = len_diff < 0;
3581 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3582 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3584 if (unlikely(len_diff_abs > 0xfffU))
3586 if (unlikely(proto != htons(ETH_P_IP) &&
3587 proto != htons(ETH_P_IPV6)))
3590 off = skb_mac_header_len(skb);
3592 case BPF_ADJ_ROOM_NET:
3593 off += bpf_skb_net_base_len(skb);
3595 case BPF_ADJ_ROOM_MAC:
3601 len_cur = skb->len - skb_network_offset(skb);
3602 if ((shrink && (len_diff_abs >= len_cur ||
3603 len_cur - len_diff_abs < len_min)) ||
3604 (!shrink && (skb->len + len_diff_abs > len_max &&
3608 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3609 bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3610 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3611 __skb_reset_checksum_unnecessary(skb);
3613 bpf_compute_data_pointers(skb);
3617 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3618 .func = bpf_skb_adjust_room,
3620 .ret_type = RET_INTEGER,
3621 .arg1_type = ARG_PTR_TO_CTX,
3622 .arg2_type = ARG_ANYTHING,
3623 .arg3_type = ARG_ANYTHING,
3624 .arg4_type = ARG_ANYTHING,
3627 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3629 u32 min_len = skb_network_offset(skb);
3631 if (skb_transport_header_was_set(skb))
3632 min_len = skb_transport_offset(skb);
3633 if (skb->ip_summed == CHECKSUM_PARTIAL)
3634 min_len = skb_checksum_start_offset(skb) +
3635 skb->csum_offset + sizeof(__sum16);
3639 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3641 unsigned int old_len = skb->len;
3644 ret = __skb_grow_rcsum(skb, new_len);
3646 memset(skb->data + old_len, 0, new_len - old_len);
3650 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3652 return __skb_trim_rcsum(skb, new_len);
3655 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3658 u32 max_len = BPF_SKB_MAX_LEN;
3659 u32 min_len = __bpf_skb_min_len(skb);
3662 if (unlikely(flags || new_len > max_len || new_len < min_len))
3664 if (skb->encapsulation)
3667 /* The basic idea of this helper is that it's performing the
3668 * needed work to either grow or trim an skb, and eBPF program
3669 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3670 * bpf_lX_csum_replace() and others rather than passing a raw
3671 * buffer here. This one is a slow path helper and intended
3672 * for replies with control messages.
3674 * Like in bpf_skb_change_proto(), we want to keep this rather
3675 * minimal and without protocol specifics so that we are able
3676 * to separate concerns as in bpf_skb_store_bytes() should only
3677 * be the one responsible for writing buffers.
3679 * It's really expected to be a slow path operation here for
3680 * control message replies, so we're implicitly linearizing,
3681 * uncloning and drop offloads from the skb by this.
3683 ret = __bpf_try_make_writable(skb, skb->len);
3685 if (new_len > skb->len)
3686 ret = bpf_skb_grow_rcsum(skb, new_len);
3687 else if (new_len < skb->len)
3688 ret = bpf_skb_trim_rcsum(skb, new_len);
3689 if (!ret && skb_is_gso(skb))
3695 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3698 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3700 bpf_compute_data_pointers(skb);
3704 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3705 .func = bpf_skb_change_tail,
3707 .ret_type = RET_INTEGER,
3708 .arg1_type = ARG_PTR_TO_CTX,
3709 .arg2_type = ARG_ANYTHING,
3710 .arg3_type = ARG_ANYTHING,
3713 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3716 return __bpf_skb_change_tail(skb, new_len, flags);
3719 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3720 .func = sk_skb_change_tail,
3722 .ret_type = RET_INTEGER,
3723 .arg1_type = ARG_PTR_TO_CTX,
3724 .arg2_type = ARG_ANYTHING,
3725 .arg3_type = ARG_ANYTHING,
3728 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3731 u32 max_len = BPF_SKB_MAX_LEN;
3732 u32 new_len = skb->len + head_room;
3735 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3736 new_len < skb->len))
3739 ret = skb_cow(skb, head_room);
3741 /* Idea for this helper is that we currently only
3742 * allow to expand on mac header. This means that
3743 * skb->protocol network header, etc, stay as is.
3744 * Compared to bpf_skb_change_tail(), we're more
3745 * flexible due to not needing to linearize or
3746 * reset GSO. Intention for this helper is to be
3747 * used by an L3 skb that needs to push mac header
3748 * for redirection into L2 device.
3750 __skb_push(skb, head_room);
3751 memset(skb->data, 0, head_room);
3752 skb_reset_mac_header(skb);
3753 skb_reset_mac_len(skb);
3759 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3762 int ret = __bpf_skb_change_head(skb, head_room, flags);
3764 bpf_compute_data_pointers(skb);
3768 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3769 .func = bpf_skb_change_head,
3771 .ret_type = RET_INTEGER,
3772 .arg1_type = ARG_PTR_TO_CTX,
3773 .arg2_type = ARG_ANYTHING,
3774 .arg3_type = ARG_ANYTHING,
3777 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3780 return __bpf_skb_change_head(skb, head_room, flags);
3783 static const struct bpf_func_proto sk_skb_change_head_proto = {
3784 .func = sk_skb_change_head,
3786 .ret_type = RET_INTEGER,
3787 .arg1_type = ARG_PTR_TO_CTX,
3788 .arg2_type = ARG_ANYTHING,
3789 .arg3_type = ARG_ANYTHING,
3791 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3793 return xdp_data_meta_unsupported(xdp) ? 0 :
3794 xdp->data - xdp->data_meta;
3797 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3799 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3800 unsigned long metalen = xdp_get_metalen(xdp);
3801 void *data_start = xdp_frame_end + metalen;
3802 void *data = xdp->data + offset;
3804 if (unlikely(data < data_start ||
3805 data > xdp->data_end - ETH_HLEN))
3809 memmove(xdp->data_meta + offset,
3810 xdp->data_meta, metalen);
3811 xdp->data_meta += offset;
3817 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3818 .func = bpf_xdp_adjust_head,
3820 .ret_type = RET_INTEGER,
3821 .arg1_type = ARG_PTR_TO_CTX,
3822 .arg2_type = ARG_ANYTHING,
3825 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3827 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
3828 void *data_end = xdp->data_end + offset;
3830 /* Notice that xdp_data_hard_end have reserved some tailroom */
3831 if (unlikely(data_end > data_hard_end))
3834 /* ALL drivers MUST init xdp->frame_sz, chicken check below */
3835 if (unlikely(xdp->frame_sz > PAGE_SIZE)) {
3836 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz);
3840 if (unlikely(data_end < xdp->data + ETH_HLEN))
3843 /* Clear memory area on grow, can contain uninit kernel memory */
3845 memset(xdp->data_end, 0, offset);
3847 xdp->data_end = data_end;
3852 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3853 .func = bpf_xdp_adjust_tail,
3855 .ret_type = RET_INTEGER,
3856 .arg1_type = ARG_PTR_TO_CTX,
3857 .arg2_type = ARG_ANYTHING,
3860 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3862 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3863 void *meta = xdp->data_meta + offset;
3864 unsigned long metalen = xdp->data - meta;
3866 if (xdp_data_meta_unsupported(xdp))
3868 if (unlikely(meta < xdp_frame_end ||
3871 if (unlikely(xdp_metalen_invalid(metalen)))
3874 xdp->data_meta = meta;
3879 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3880 .func = bpf_xdp_adjust_meta,
3882 .ret_type = RET_INTEGER,
3883 .arg1_type = ARG_PTR_TO_CTX,
3884 .arg2_type = ARG_ANYTHING,
3887 /* XDP_REDIRECT works by a three-step process, implemented in the functions
3890 * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target
3891 * of the redirect and store it (along with some other metadata) in a per-CPU
3892 * struct bpf_redirect_info.
3894 * 2. When the program returns the XDP_REDIRECT return code, the driver will
3895 * call xdp_do_redirect() which will use the information in struct
3896 * bpf_redirect_info to actually enqueue the frame into a map type-specific
3897 * bulk queue structure.
3899 * 3. Before exiting its NAPI poll loop, the driver will call xdp_do_flush(),
3900 * which will flush all the different bulk queues, thus completing the
3903 * Pointers to the map entries will be kept around for this whole sequence of
3904 * steps, protected by RCU. However, there is no top-level rcu_read_lock() in
3905 * the core code; instead, the RCU protection relies on everything happening
3906 * inside a single NAPI poll sequence, which means it's between a pair of calls
3907 * to local_bh_disable()/local_bh_enable().
3909 * The map entries are marked as __rcu and the map code makes sure to
3910 * dereference those pointers with rcu_dereference_check() in a way that works
3911 * for both sections that to hold an rcu_read_lock() and sections that are
3912 * called from NAPI without a separate rcu_read_lock(). The code below does not
3913 * use RCU annotations, but relies on those in the map code.
3915 void xdp_do_flush(void)
3921 EXPORT_SYMBOL_GPL(xdp_do_flush);
3923 void bpf_clear_redirect_map(struct bpf_map *map)
3925 struct bpf_redirect_info *ri;
3928 for_each_possible_cpu(cpu) {
3929 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3930 /* Avoid polluting remote cacheline due to writes if
3931 * not needed. Once we pass this test, we need the
3932 * cmpxchg() to make sure it hasn't been changed in
3933 * the meantime by remote CPU.
3935 if (unlikely(READ_ONCE(ri->map) == map))
3936 cmpxchg(&ri->map, map, NULL);
3940 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
3941 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key);
3943 u32 xdp_master_redirect(struct xdp_buff *xdp)
3945 struct net_device *master, *slave;
3946 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3948 master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev);
3949 slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp);
3950 if (slave && slave != xdp->rxq->dev) {
3951 /* The target device is different from the receiving device, so
3952 * redirect it to the new device.
3953 * Using XDP_REDIRECT gets the correct behaviour from XDP enabled
3954 * drivers to unmap the packet from their rx ring.
3956 ri->tgt_index = slave->ifindex;
3957 ri->map_id = INT_MAX;
3958 ri->map_type = BPF_MAP_TYPE_UNSPEC;
3959 return XDP_REDIRECT;
3963 EXPORT_SYMBOL_GPL(xdp_master_redirect);
3965 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3966 struct bpf_prog *xdp_prog)
3968 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3969 enum bpf_map_type map_type = ri->map_type;
3970 void *fwd = ri->tgt_value;
3971 u32 map_id = ri->map_id;
3972 struct bpf_map *map;
3975 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
3976 ri->map_type = BPF_MAP_TYPE_UNSPEC;
3979 case BPF_MAP_TYPE_DEVMAP:
3981 case BPF_MAP_TYPE_DEVMAP_HASH:
3982 map = READ_ONCE(ri->map);
3983 if (unlikely(map)) {
3984 WRITE_ONCE(ri->map, NULL);
3985 err = dev_map_enqueue_multi(xdp, dev, map,
3986 ri->flags & BPF_F_EXCLUDE_INGRESS);
3988 err = dev_map_enqueue(fwd, xdp, dev);
3991 case BPF_MAP_TYPE_CPUMAP:
3992 err = cpu_map_enqueue(fwd, xdp, dev);
3994 case BPF_MAP_TYPE_XSKMAP:
3995 err = __xsk_map_redirect(fwd, xdp);
3997 case BPF_MAP_TYPE_UNSPEC:
3998 if (map_id == INT_MAX) {
3999 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4000 if (unlikely(!fwd)) {
4004 err = dev_xdp_enqueue(fwd, xdp, dev);
4015 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4018 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4021 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4023 static int xdp_do_generic_redirect_map(struct net_device *dev,
4024 struct sk_buff *skb,
4025 struct xdp_buff *xdp,
4026 struct bpf_prog *xdp_prog,
4028 enum bpf_map_type map_type, u32 map_id)
4030 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4031 struct bpf_map *map;
4035 case BPF_MAP_TYPE_DEVMAP:
4037 case BPF_MAP_TYPE_DEVMAP_HASH:
4038 map = READ_ONCE(ri->map);
4039 if (unlikely(map)) {
4040 WRITE_ONCE(ri->map, NULL);
4041 err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4042 ri->flags & BPF_F_EXCLUDE_INGRESS);
4044 err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4049 case BPF_MAP_TYPE_XSKMAP:
4050 err = xsk_generic_rcv(fwd, xdp);
4055 case BPF_MAP_TYPE_CPUMAP:
4056 err = cpu_map_generic_redirect(fwd, skb);
4065 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4068 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4072 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4073 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4075 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4076 enum bpf_map_type map_type = ri->map_type;
4077 void *fwd = ri->tgt_value;
4078 u32 map_id = ri->map_id;
4081 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4082 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4084 if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4085 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4086 if (unlikely(!fwd)) {
4091 err = xdp_ok_fwd_dev(fwd, skb->len);
4096 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4097 generic_xdp_tx(skb, xdp_prog);
4101 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id);
4103 _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4107 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4109 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4111 if (unlikely(flags))
4114 /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4115 * by map_idr) is used for ifindex based XDP redirect.
4117 ri->tgt_index = ifindex;
4118 ri->map_id = INT_MAX;
4119 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4121 return XDP_REDIRECT;
4124 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4125 .func = bpf_xdp_redirect,
4127 .ret_type = RET_INTEGER,
4128 .arg1_type = ARG_ANYTHING,
4129 .arg2_type = ARG_ANYTHING,
4132 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
4135 return map->ops->map_redirect(map, ifindex, flags);
4138 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4139 .func = bpf_xdp_redirect_map,
4141 .ret_type = RET_INTEGER,
4142 .arg1_type = ARG_CONST_MAP_PTR,
4143 .arg2_type = ARG_ANYTHING,
4144 .arg3_type = ARG_ANYTHING,
4147 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4148 unsigned long off, unsigned long len)
4150 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4154 if (ptr != dst_buff)
4155 memcpy(dst_buff, ptr, len);
4160 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4161 u64, flags, void *, meta, u64, meta_size)
4163 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4165 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4167 if (unlikely(!skb || skb_size > skb->len))
4170 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4174 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4175 .func = bpf_skb_event_output,
4177 .ret_type = RET_INTEGER,
4178 .arg1_type = ARG_PTR_TO_CTX,
4179 .arg2_type = ARG_CONST_MAP_PTR,
4180 .arg3_type = ARG_ANYTHING,
4181 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4182 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4185 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4187 const struct bpf_func_proto bpf_skb_output_proto = {
4188 .func = bpf_skb_event_output,
4190 .ret_type = RET_INTEGER,
4191 .arg1_type = ARG_PTR_TO_BTF_ID,
4192 .arg1_btf_id = &bpf_skb_output_btf_ids[0],
4193 .arg2_type = ARG_CONST_MAP_PTR,
4194 .arg3_type = ARG_ANYTHING,
4195 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4196 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4199 static unsigned short bpf_tunnel_key_af(u64 flags)
4201 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4204 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4205 u32, size, u64, flags)
4207 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4208 u8 compat[sizeof(struct bpf_tunnel_key)];
4212 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
4216 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4220 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4223 case offsetof(struct bpf_tunnel_key, tunnel_label):
4224 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4226 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4227 /* Fixup deprecated structure layouts here, so we have
4228 * a common path later on.
4230 if (ip_tunnel_info_af(info) != AF_INET)
4233 to = (struct bpf_tunnel_key *)compat;
4240 to->tunnel_id = be64_to_cpu(info->key.tun_id);
4241 to->tunnel_tos = info->key.tos;
4242 to->tunnel_ttl = info->key.ttl;
4245 if (flags & BPF_F_TUNINFO_IPV6) {
4246 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4247 sizeof(to->remote_ipv6));
4248 to->tunnel_label = be32_to_cpu(info->key.label);
4250 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4251 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4252 to->tunnel_label = 0;
4255 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4256 memcpy(to_orig, to, size);
4260 memset(to_orig, 0, size);
4264 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4265 .func = bpf_skb_get_tunnel_key,
4267 .ret_type = RET_INTEGER,
4268 .arg1_type = ARG_PTR_TO_CTX,
4269 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4270 .arg3_type = ARG_CONST_SIZE,
4271 .arg4_type = ARG_ANYTHING,
4274 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4276 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4279 if (unlikely(!info ||
4280 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4284 if (unlikely(size < info->options_len)) {
4289 ip_tunnel_info_opts_get(to, info);
4290 if (size > info->options_len)
4291 memset(to + info->options_len, 0, size - info->options_len);
4293 return info->options_len;
4295 memset(to, 0, size);
4299 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4300 .func = bpf_skb_get_tunnel_opt,
4302 .ret_type = RET_INTEGER,
4303 .arg1_type = ARG_PTR_TO_CTX,
4304 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4305 .arg3_type = ARG_CONST_SIZE,
4308 static struct metadata_dst __percpu *md_dst;
4310 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4311 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4313 struct metadata_dst *md = this_cpu_ptr(md_dst);
4314 u8 compat[sizeof(struct bpf_tunnel_key)];
4315 struct ip_tunnel_info *info;
4317 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4318 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
4320 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4322 case offsetof(struct bpf_tunnel_key, tunnel_label):
4323 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4324 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4325 /* Fixup deprecated structure layouts here, so we have
4326 * a common path later on.
4328 memcpy(compat, from, size);
4329 memset(compat + size, 0, sizeof(compat) - size);
4330 from = (const struct bpf_tunnel_key *) compat;
4336 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4341 dst_hold((struct dst_entry *) md);
4342 skb_dst_set(skb, (struct dst_entry *) md);
4344 info = &md->u.tun_info;
4345 memset(info, 0, sizeof(*info));
4346 info->mode = IP_TUNNEL_INFO_TX;
4348 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4349 if (flags & BPF_F_DONT_FRAGMENT)
4350 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4351 if (flags & BPF_F_ZERO_CSUM_TX)
4352 info->key.tun_flags &= ~TUNNEL_CSUM;
4353 if (flags & BPF_F_SEQ_NUMBER)
4354 info->key.tun_flags |= TUNNEL_SEQ;
4356 info->key.tun_id = cpu_to_be64(from->tunnel_id);
4357 info->key.tos = from->tunnel_tos;
4358 info->key.ttl = from->tunnel_ttl;
4360 if (flags & BPF_F_TUNINFO_IPV6) {
4361 info->mode |= IP_TUNNEL_INFO_IPV6;
4362 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4363 sizeof(from->remote_ipv6));
4364 info->key.label = cpu_to_be32(from->tunnel_label) &
4365 IPV6_FLOWLABEL_MASK;
4367 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4373 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4374 .func = bpf_skb_set_tunnel_key,
4376 .ret_type = RET_INTEGER,
4377 .arg1_type = ARG_PTR_TO_CTX,
4378 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4379 .arg3_type = ARG_CONST_SIZE,
4380 .arg4_type = ARG_ANYTHING,
4383 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4384 const u8 *, from, u32, size)
4386 struct ip_tunnel_info *info = skb_tunnel_info(skb);
4387 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4389 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4391 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4394 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4399 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4400 .func = bpf_skb_set_tunnel_opt,
4402 .ret_type = RET_INTEGER,
4403 .arg1_type = ARG_PTR_TO_CTX,
4404 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4405 .arg3_type = ARG_CONST_SIZE,
4408 static const struct bpf_func_proto *
4409 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4412 struct metadata_dst __percpu *tmp;
4414 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4419 if (cmpxchg(&md_dst, NULL, tmp))
4420 metadata_dst_free_percpu(tmp);
4424 case BPF_FUNC_skb_set_tunnel_key:
4425 return &bpf_skb_set_tunnel_key_proto;
4426 case BPF_FUNC_skb_set_tunnel_opt:
4427 return &bpf_skb_set_tunnel_opt_proto;
4433 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4436 struct bpf_array *array = container_of(map, struct bpf_array, map);
4437 struct cgroup *cgrp;
4440 sk = skb_to_full_sk(skb);
4441 if (!sk || !sk_fullsock(sk))
4443 if (unlikely(idx >= array->map.max_entries))
4446 cgrp = READ_ONCE(array->ptrs[idx]);
4447 if (unlikely(!cgrp))
4450 return sk_under_cgroup_hierarchy(sk, cgrp);
4453 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4454 .func = bpf_skb_under_cgroup,
4456 .ret_type = RET_INTEGER,
4457 .arg1_type = ARG_PTR_TO_CTX,
4458 .arg2_type = ARG_CONST_MAP_PTR,
4459 .arg3_type = ARG_ANYTHING,
4462 #ifdef CONFIG_SOCK_CGROUP_DATA
4463 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4465 struct cgroup *cgrp;
4467 sk = sk_to_full_sk(sk);
4468 if (!sk || !sk_fullsock(sk))
4471 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4472 return cgroup_id(cgrp);
4475 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4477 return __bpf_sk_cgroup_id(skb->sk);
4480 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4481 .func = bpf_skb_cgroup_id,
4483 .ret_type = RET_INTEGER,
4484 .arg1_type = ARG_PTR_TO_CTX,
4487 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4490 struct cgroup *ancestor;
4491 struct cgroup *cgrp;
4493 sk = sk_to_full_sk(sk);
4494 if (!sk || !sk_fullsock(sk))
4497 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4498 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4502 return cgroup_id(ancestor);
4505 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4508 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4511 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4512 .func = bpf_skb_ancestor_cgroup_id,
4514 .ret_type = RET_INTEGER,
4515 .arg1_type = ARG_PTR_TO_CTX,
4516 .arg2_type = ARG_ANYTHING,
4519 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4521 return __bpf_sk_cgroup_id(sk);
4524 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4525 .func = bpf_sk_cgroup_id,
4527 .ret_type = RET_INTEGER,
4528 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4531 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4533 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4536 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4537 .func = bpf_sk_ancestor_cgroup_id,
4539 .ret_type = RET_INTEGER,
4540 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4541 .arg2_type = ARG_ANYTHING,
4545 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4546 unsigned long off, unsigned long len)
4548 memcpy(dst_buff, src_buff + off, len);
4552 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4553 u64, flags, void *, meta, u64, meta_size)
4555 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4557 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4559 if (unlikely(!xdp ||
4560 xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4563 return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4564 xdp_size, bpf_xdp_copy);
4567 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4568 .func = bpf_xdp_event_output,
4570 .ret_type = RET_INTEGER,
4571 .arg1_type = ARG_PTR_TO_CTX,
4572 .arg2_type = ARG_CONST_MAP_PTR,
4573 .arg3_type = ARG_ANYTHING,
4574 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4575 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4578 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4580 const struct bpf_func_proto bpf_xdp_output_proto = {
4581 .func = bpf_xdp_event_output,
4583 .ret_type = RET_INTEGER,
4584 .arg1_type = ARG_PTR_TO_BTF_ID,
4585 .arg1_btf_id = &bpf_xdp_output_btf_ids[0],
4586 .arg2_type = ARG_CONST_MAP_PTR,
4587 .arg3_type = ARG_ANYTHING,
4588 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4589 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4592 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4594 return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4597 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4598 .func = bpf_get_socket_cookie,
4600 .ret_type = RET_INTEGER,
4601 .arg1_type = ARG_PTR_TO_CTX,
4604 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4606 return __sock_gen_cookie(ctx->sk);
4609 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4610 .func = bpf_get_socket_cookie_sock_addr,
4612 .ret_type = RET_INTEGER,
4613 .arg1_type = ARG_PTR_TO_CTX,
4616 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4618 return __sock_gen_cookie(ctx);
4621 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4622 .func = bpf_get_socket_cookie_sock,
4624 .ret_type = RET_INTEGER,
4625 .arg1_type = ARG_PTR_TO_CTX,
4628 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
4630 return sk ? sock_gen_cookie(sk) : 0;
4633 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
4634 .func = bpf_get_socket_ptr_cookie,
4636 .ret_type = RET_INTEGER,
4637 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4640 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4642 return __sock_gen_cookie(ctx->sk);
4645 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4646 .func = bpf_get_socket_cookie_sock_ops,
4648 .ret_type = RET_INTEGER,
4649 .arg1_type = ARG_PTR_TO_CTX,
4652 static u64 __bpf_get_netns_cookie(struct sock *sk)
4654 const struct net *net = sk ? sock_net(sk) : &init_net;
4656 return net->net_cookie;
4659 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
4661 return __bpf_get_netns_cookie(ctx);
4664 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
4665 .func = bpf_get_netns_cookie_sock,
4667 .ret_type = RET_INTEGER,
4668 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4671 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4673 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4676 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
4677 .func = bpf_get_netns_cookie_sock_addr,
4679 .ret_type = RET_INTEGER,
4680 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4683 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4685 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4688 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = {
4689 .func = bpf_get_netns_cookie_sock_ops,
4691 .ret_type = RET_INTEGER,
4692 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4695 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx)
4697 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4700 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = {
4701 .func = bpf_get_netns_cookie_sk_msg,
4703 .ret_type = RET_INTEGER,
4704 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4707 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4709 struct sock *sk = sk_to_full_sk(skb->sk);
4712 if (!sk || !sk_fullsock(sk))
4714 kuid = sock_net_uid(sock_net(sk), sk);
4715 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4718 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4719 .func = bpf_get_socket_uid,
4721 .ret_type = RET_INTEGER,
4722 .arg1_type = ARG_PTR_TO_CTX,
4725 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
4726 char *optval, int optlen)
4728 char devname[IFNAMSIZ];
4734 if (!sk_fullsock(sk))
4737 sock_owned_by_me(sk);
4739 if (level == SOL_SOCKET) {
4740 if (optlen != sizeof(int) && optname != SO_BINDTODEVICE)
4742 val = *((int *)optval);
4743 valbool = val ? 1 : 0;
4745 /* Only some socketops are supported */
4748 val = min_t(u32, val, READ_ONCE(sysctl_rmem_max));
4749 val = min_t(int, val, INT_MAX / 2);
4750 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4751 WRITE_ONCE(sk->sk_rcvbuf,
4752 max_t(int, val * 2, SOCK_MIN_RCVBUF));
4755 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
4756 val = min_t(int, val, INT_MAX / 2);
4757 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4758 WRITE_ONCE(sk->sk_sndbuf,
4759 max_t(int, val * 2, SOCK_MIN_SNDBUF));
4761 case SO_MAX_PACING_RATE: /* 32bit version */
4763 cmpxchg(&sk->sk_pacing_status,
4766 sk->sk_max_pacing_rate = (val == ~0U) ?
4767 ~0UL : (unsigned int)val;
4768 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4769 sk->sk_max_pacing_rate);
4772 sk->sk_priority = val;
4777 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
4780 if (sk->sk_mark != val) {
4785 case SO_BINDTODEVICE:
4786 optlen = min_t(long, optlen, IFNAMSIZ - 1);
4787 strncpy(devname, optval, optlen);
4788 devname[optlen] = 0;
4791 if (devname[0] != '\0') {
4792 struct net_device *dev;
4797 dev = dev_get_by_name(net, devname);
4800 ifindex = dev->ifindex;
4804 case SO_BINDTOIFINDEX:
4805 if (optname == SO_BINDTOIFINDEX)
4807 ret = sock_bindtoindex(sk, ifindex, false);
4810 if (sk->sk_prot->keepalive)
4811 sk->sk_prot->keepalive(sk, valbool);
4812 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
4815 sk->sk_reuseport = valbool;
4821 } else if (level == SOL_IP) {
4822 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4825 val = *((int *)optval);
4826 /* Only some options are supported */
4829 if (val < -1 || val > 0xff) {
4832 struct inet_sock *inet = inet_sk(sk);
4842 #if IS_ENABLED(CONFIG_IPV6)
4843 } else if (level == SOL_IPV6) {
4844 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4847 val = *((int *)optval);
4848 /* Only some options are supported */
4851 if (val < -1 || val > 0xff) {
4854 struct ipv6_pinfo *np = inet6_sk(sk);
4865 } else if (level == SOL_TCP &&
4866 sk->sk_prot->setsockopt == tcp_setsockopt) {
4867 if (optname == TCP_CONGESTION) {
4868 char name[TCP_CA_NAME_MAX];
4870 strncpy(name, optval, min_t(long, optlen,
4871 TCP_CA_NAME_MAX-1));
4872 name[TCP_CA_NAME_MAX-1] = 0;
4873 ret = tcp_set_congestion_control(sk, name, false, true);
4875 struct inet_connection_sock *icsk = inet_csk(sk);
4876 struct tcp_sock *tp = tcp_sk(sk);
4877 unsigned long timeout;
4879 if (optlen != sizeof(int))
4882 val = *((int *)optval);
4883 /* Only some options are supported */
4886 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4889 tcp_snd_cwnd_set(tp, val);
4891 case TCP_BPF_SNDCWND_CLAMP:
4895 tp->snd_cwnd_clamp = val;
4896 tp->snd_ssthresh = val;
4899 case TCP_BPF_DELACK_MAX:
4900 timeout = usecs_to_jiffies(val);
4901 if (timeout > TCP_DELACK_MAX ||
4902 timeout < TCP_TIMEOUT_MIN)
4904 inet_csk(sk)->icsk_delack_max = timeout;
4906 case TCP_BPF_RTO_MIN:
4907 timeout = usecs_to_jiffies(val);
4908 if (timeout > TCP_RTO_MIN ||
4909 timeout < TCP_TIMEOUT_MIN)
4911 inet_csk(sk)->icsk_rto_min = timeout;
4914 if (val < 0 || val > 1)
4920 ret = tcp_sock_set_keepidle_locked(sk, val);
4923 if (val < 1 || val > MAX_TCP_KEEPINTVL)
4926 tp->keepalive_intvl = val * HZ;
4929 if (val < 1 || val > MAX_TCP_KEEPCNT)
4932 tp->keepalive_probes = val;
4935 if (val < 1 || val > MAX_TCP_SYNCNT)
4938 icsk->icsk_syn_retries = val;
4940 case TCP_USER_TIMEOUT:
4944 icsk->icsk_user_timeout = val;
4946 case TCP_NOTSENT_LOWAT:
4947 tp->notsent_lowat = val;
4948 sk->sk_write_space(sk);
4950 case TCP_WINDOW_CLAMP:
4951 ret = tcp_set_window_clamp(sk, val);
4964 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
4965 char *optval, int optlen)
4967 if (!sk_fullsock(sk))
4970 sock_owned_by_me(sk);
4972 if (level == SOL_SOCKET) {
4973 if (optlen != sizeof(int))
4978 *((int *)optval) = sk->sk_mark;
4981 *((int *)optval) = sk->sk_priority;
4983 case SO_BINDTOIFINDEX:
4984 *((int *)optval) = sk->sk_bound_dev_if;
4987 *((int *)optval) = sk->sk_reuseport;
4993 } else if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4994 struct inet_connection_sock *icsk;
4995 struct tcp_sock *tp;
4998 case TCP_CONGESTION:
4999 icsk = inet_csk(sk);
5001 if (!icsk->icsk_ca_ops || optlen <= 1)
5003 strncpy(optval, icsk->icsk_ca_ops->name, optlen);
5004 optval[optlen - 1] = 0;
5009 if (optlen <= 0 || !tp->saved_syn ||
5010 optlen > tcp_saved_syn_len(tp->saved_syn))
5012 memcpy(optval, tp->saved_syn->data, optlen);
5017 } else if (level == SOL_IP) {
5018 struct inet_sock *inet = inet_sk(sk);
5020 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
5023 /* Only some options are supported */
5026 *((int *)optval) = (int)inet->tos;
5031 #if IS_ENABLED(CONFIG_IPV6)
5032 } else if (level == SOL_IPV6) {
5033 struct ipv6_pinfo *np = inet6_sk(sk);
5035 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
5038 /* Only some options are supported */
5041 *((int *)optval) = (int)np->tclass;
5053 memset(optval, 0, optlen);
5057 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level,
5058 int, optname, char *, optval, int, optlen)
5060 if (level == SOL_TCP && optname == TCP_CONGESTION) {
5061 if (optlen >= sizeof("cdg") - 1 &&
5062 !strncmp("cdg", optval, optlen))
5066 return _bpf_setsockopt(sk, level, optname, optval, optlen);
5069 const struct bpf_func_proto bpf_sk_setsockopt_proto = {
5070 .func = bpf_sk_setsockopt,
5072 .ret_type = RET_INTEGER,
5073 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5074 .arg2_type = ARG_ANYTHING,
5075 .arg3_type = ARG_ANYTHING,
5076 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5077 .arg5_type = ARG_CONST_SIZE,
5080 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level,
5081 int, optname, char *, optval, int, optlen)
5083 return _bpf_getsockopt(sk, level, optname, optval, optlen);
5086 const struct bpf_func_proto bpf_sk_getsockopt_proto = {
5087 .func = bpf_sk_getsockopt,
5089 .ret_type = RET_INTEGER,
5090 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5091 .arg2_type = ARG_ANYTHING,
5092 .arg3_type = ARG_ANYTHING,
5093 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5094 .arg5_type = ARG_CONST_SIZE,
5097 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5098 int, level, int, optname, char *, optval, int, optlen)
5100 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5103 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5104 .func = bpf_sock_addr_setsockopt,
5106 .ret_type = RET_INTEGER,
5107 .arg1_type = ARG_PTR_TO_CTX,
5108 .arg2_type = ARG_ANYTHING,
5109 .arg3_type = ARG_ANYTHING,
5110 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5111 .arg5_type = ARG_CONST_SIZE,
5114 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5115 int, level, int, optname, char *, optval, int, optlen)
5117 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5120 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5121 .func = bpf_sock_addr_getsockopt,
5123 .ret_type = RET_INTEGER,
5124 .arg1_type = ARG_PTR_TO_CTX,
5125 .arg2_type = ARG_ANYTHING,
5126 .arg3_type = ARG_ANYTHING,
5127 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5128 .arg5_type = ARG_CONST_SIZE,
5131 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5132 int, level, int, optname, char *, optval, int, optlen)
5134 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5137 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5138 .func = bpf_sock_ops_setsockopt,
5140 .ret_type = RET_INTEGER,
5141 .arg1_type = ARG_PTR_TO_CTX,
5142 .arg2_type = ARG_ANYTHING,
5143 .arg3_type = ARG_ANYTHING,
5144 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5145 .arg5_type = ARG_CONST_SIZE,
5148 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5149 int optname, const u8 **start)
5151 struct sk_buff *syn_skb = bpf_sock->syn_skb;
5152 const u8 *hdr_start;
5156 /* sk is a request_sock here */
5158 if (optname == TCP_BPF_SYN) {
5159 hdr_start = syn_skb->data;
5160 ret = tcp_hdrlen(syn_skb);
5161 } else if (optname == TCP_BPF_SYN_IP) {
5162 hdr_start = skb_network_header(syn_skb);
5163 ret = skb_network_header_len(syn_skb) +
5164 tcp_hdrlen(syn_skb);
5166 /* optname == TCP_BPF_SYN_MAC */
5167 hdr_start = skb_mac_header(syn_skb);
5168 ret = skb_mac_header_len(syn_skb) +
5169 skb_network_header_len(syn_skb) +
5170 tcp_hdrlen(syn_skb);
5173 struct sock *sk = bpf_sock->sk;
5174 struct saved_syn *saved_syn;
5176 if (sk->sk_state == TCP_NEW_SYN_RECV)
5177 /* synack retransmit. bpf_sock->syn_skb will
5178 * not be available. It has to resort to
5179 * saved_syn (if it is saved).
5181 saved_syn = inet_reqsk(sk)->saved_syn;
5183 saved_syn = tcp_sk(sk)->saved_syn;
5188 if (optname == TCP_BPF_SYN) {
5189 hdr_start = saved_syn->data +
5190 saved_syn->mac_hdrlen +
5191 saved_syn->network_hdrlen;
5192 ret = saved_syn->tcp_hdrlen;
5193 } else if (optname == TCP_BPF_SYN_IP) {
5194 hdr_start = saved_syn->data +
5195 saved_syn->mac_hdrlen;
5196 ret = saved_syn->network_hdrlen +
5197 saved_syn->tcp_hdrlen;
5199 /* optname == TCP_BPF_SYN_MAC */
5201 /* TCP_SAVE_SYN may not have saved the mac hdr */
5202 if (!saved_syn->mac_hdrlen)
5205 hdr_start = saved_syn->data;
5206 ret = saved_syn->mac_hdrlen +
5207 saved_syn->network_hdrlen +
5208 saved_syn->tcp_hdrlen;
5216 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5217 int, level, int, optname, char *, optval, int, optlen)
5219 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5220 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5221 int ret, copy_len = 0;
5224 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5227 if (optlen < copy_len) {
5232 memcpy(optval, start, copy_len);
5235 /* Zero out unused buffer at the end */
5236 memset(optval + copy_len, 0, optlen - copy_len);
5241 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5244 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5245 .func = bpf_sock_ops_getsockopt,
5247 .ret_type = RET_INTEGER,
5248 .arg1_type = ARG_PTR_TO_CTX,
5249 .arg2_type = ARG_ANYTHING,
5250 .arg3_type = ARG_ANYTHING,
5251 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5252 .arg5_type = ARG_CONST_SIZE,
5255 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5258 struct sock *sk = bpf_sock->sk;
5259 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5261 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5264 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5266 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5269 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5270 .func = bpf_sock_ops_cb_flags_set,
5272 .ret_type = RET_INTEGER,
5273 .arg1_type = ARG_PTR_TO_CTX,
5274 .arg2_type = ARG_ANYTHING,
5277 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5278 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5280 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5284 struct sock *sk = ctx->sk;
5285 u32 flags = BIND_FROM_BPF;
5289 if (addr_len < offsetofend(struct sockaddr, sa_family))
5291 if (addr->sa_family == AF_INET) {
5292 if (addr_len < sizeof(struct sockaddr_in))
5294 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5295 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5296 return __inet_bind(sk, addr, addr_len, flags);
5297 #if IS_ENABLED(CONFIG_IPV6)
5298 } else if (addr->sa_family == AF_INET6) {
5299 if (addr_len < SIN6_LEN_RFC2133)
5301 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5302 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5303 /* ipv6_bpf_stub cannot be NULL, since it's called from
5304 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5306 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5307 #endif /* CONFIG_IPV6 */
5309 #endif /* CONFIG_INET */
5311 return -EAFNOSUPPORT;
5314 static const struct bpf_func_proto bpf_bind_proto = {
5317 .ret_type = RET_INTEGER,
5318 .arg1_type = ARG_PTR_TO_CTX,
5319 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5320 .arg3_type = ARG_CONST_SIZE,
5324 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5325 struct bpf_xfrm_state *, to, u32, size, u64, flags)
5327 const struct sec_path *sp = skb_sec_path(skb);
5328 const struct xfrm_state *x;
5330 if (!sp || unlikely(index >= sp->len || flags))
5333 x = sp->xvec[index];
5335 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5338 to->reqid = x->props.reqid;
5339 to->spi = x->id.spi;
5340 to->family = x->props.family;
5343 if (to->family == AF_INET6) {
5344 memcpy(to->remote_ipv6, x->props.saddr.a6,
5345 sizeof(to->remote_ipv6));
5347 to->remote_ipv4 = x->props.saddr.a4;
5348 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5353 memset(to, 0, size);
5357 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5358 .func = bpf_skb_get_xfrm_state,
5360 .ret_type = RET_INTEGER,
5361 .arg1_type = ARG_PTR_TO_CTX,
5362 .arg2_type = ARG_ANYTHING,
5363 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
5364 .arg4_type = ARG_CONST_SIZE,
5365 .arg5_type = ARG_ANYTHING,
5369 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5370 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
5371 const struct neighbour *neigh,
5372 const struct net_device *dev, u32 mtu)
5374 memcpy(params->dmac, neigh->ha, ETH_ALEN);
5375 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5376 params->h_vlan_TCI = 0;
5377 params->h_vlan_proto = 0;
5379 params->mtu_result = mtu; /* union with tot_len */
5385 #if IS_ENABLED(CONFIG_INET)
5386 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5387 u32 flags, bool check_mtu)
5389 struct fib_nh_common *nhc;
5390 struct in_device *in_dev;
5391 struct neighbour *neigh;
5392 struct net_device *dev;
5393 struct fib_result res;
5398 dev = dev_get_by_index_rcu(net, params->ifindex);
5402 /* verify forwarding is enabled on this interface */
5403 in_dev = __in_dev_get_rcu(dev);
5404 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5405 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5407 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5409 fl4.flowi4_oif = params->ifindex;
5411 fl4.flowi4_iif = params->ifindex;
5414 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5415 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5416 fl4.flowi4_flags = 0;
5418 fl4.flowi4_proto = params->l4_protocol;
5419 fl4.daddr = params->ipv4_dst;
5420 fl4.saddr = params->ipv4_src;
5421 fl4.fl4_sport = params->sport;
5422 fl4.fl4_dport = params->dport;
5423 fl4.flowi4_multipath_hash = 0;
5425 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5426 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5427 struct fib_table *tb;
5429 tb = fib_get_table(net, tbid);
5431 return BPF_FIB_LKUP_RET_NOT_FWDED;
5433 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5435 fl4.flowi4_mark = 0;
5436 fl4.flowi4_secid = 0;
5437 fl4.flowi4_tun_key.tun_id = 0;
5438 fl4.flowi4_uid = sock_net_uid(net, NULL);
5440 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5444 /* map fib lookup errors to RTN_ type */
5446 return BPF_FIB_LKUP_RET_BLACKHOLE;
5447 if (err == -EHOSTUNREACH)
5448 return BPF_FIB_LKUP_RET_UNREACHABLE;
5450 return BPF_FIB_LKUP_RET_PROHIBIT;
5452 return BPF_FIB_LKUP_RET_NOT_FWDED;
5455 if (res.type != RTN_UNICAST)
5456 return BPF_FIB_LKUP_RET_NOT_FWDED;
5458 if (fib_info_num_path(res.fi) > 1)
5459 fib_select_path(net, &res, &fl4, NULL);
5462 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5463 if (params->tot_len > mtu) {
5464 params->mtu_result = mtu; /* union with tot_len */
5465 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5471 /* do not handle lwt encaps right now */
5472 if (nhc->nhc_lwtstate)
5473 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5477 params->rt_metric = res.fi->fib_priority;
5478 params->ifindex = dev->ifindex;
5480 /* xdp and cls_bpf programs are run in RCU-bh so
5481 * rcu_read_lock_bh is not needed here
5483 if (likely(nhc->nhc_gw_family != AF_INET6)) {
5484 if (nhc->nhc_gw_family)
5485 params->ipv4_dst = nhc->nhc_gw.ipv4;
5487 neigh = __ipv4_neigh_lookup_noref(dev,
5488 (__force u32)params->ipv4_dst);
5490 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5492 params->family = AF_INET6;
5493 *dst = nhc->nhc_gw.ipv6;
5494 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5498 return BPF_FIB_LKUP_RET_NO_NEIGH;
5500 return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5504 #if IS_ENABLED(CONFIG_IPV6)
5505 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5506 u32 flags, bool check_mtu)
5508 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5509 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5510 struct fib6_result res = {};
5511 struct neighbour *neigh;
5512 struct net_device *dev;
5513 struct inet6_dev *idev;
5519 /* link local addresses are never forwarded */
5520 if (rt6_need_strict(dst) || rt6_need_strict(src))
5521 return BPF_FIB_LKUP_RET_NOT_FWDED;
5523 dev = dev_get_by_index_rcu(net, params->ifindex);
5527 idev = __in6_dev_get_safely(dev);
5528 if (unlikely(!idev || !idev->cnf.forwarding))
5529 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5531 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5533 oif = fl6.flowi6_oif = params->ifindex;
5535 oif = fl6.flowi6_iif = params->ifindex;
5537 strict = RT6_LOOKUP_F_HAS_SADDR;
5539 fl6.flowlabel = params->flowinfo;
5540 fl6.flowi6_scope = 0;
5541 fl6.flowi6_flags = 0;
5544 fl6.flowi6_proto = params->l4_protocol;
5547 fl6.fl6_sport = params->sport;
5548 fl6.fl6_dport = params->dport;
5550 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5551 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5552 struct fib6_table *tb;
5554 tb = ipv6_stub->fib6_get_table(net, tbid);
5556 return BPF_FIB_LKUP_RET_NOT_FWDED;
5558 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5561 fl6.flowi6_mark = 0;
5562 fl6.flowi6_secid = 0;
5563 fl6.flowi6_tun_key.tun_id = 0;
5564 fl6.flowi6_uid = sock_net_uid(net, NULL);
5566 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5569 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5570 res.f6i == net->ipv6.fib6_null_entry))
5571 return BPF_FIB_LKUP_RET_NOT_FWDED;
5573 switch (res.fib6_type) {
5574 /* only unicast is forwarded */
5578 return BPF_FIB_LKUP_RET_BLACKHOLE;
5579 case RTN_UNREACHABLE:
5580 return BPF_FIB_LKUP_RET_UNREACHABLE;
5582 return BPF_FIB_LKUP_RET_PROHIBIT;
5584 return BPF_FIB_LKUP_RET_NOT_FWDED;
5587 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5588 fl6.flowi6_oif != 0, NULL, strict);
5591 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5592 if (params->tot_len > mtu) {
5593 params->mtu_result = mtu; /* union with tot_len */
5594 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5598 if (res.nh->fib_nh_lws)
5599 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5601 if (res.nh->fib_nh_gw_family)
5602 *dst = res.nh->fib_nh_gw6;
5604 dev = res.nh->fib_nh_dev;
5605 params->rt_metric = res.f6i->fib6_metric;
5606 params->ifindex = dev->ifindex;
5608 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
5611 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5613 return BPF_FIB_LKUP_RET_NO_NEIGH;
5615 return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5619 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
5620 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5622 if (plen < sizeof(*params))
5625 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5628 switch (params->family) {
5629 #if IS_ENABLED(CONFIG_INET)
5631 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
5634 #if IS_ENABLED(CONFIG_IPV6)
5636 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
5640 return -EAFNOSUPPORT;
5643 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
5644 .func = bpf_xdp_fib_lookup,
5646 .ret_type = RET_INTEGER,
5647 .arg1_type = ARG_PTR_TO_CTX,
5648 .arg2_type = ARG_PTR_TO_MEM,
5649 .arg3_type = ARG_CONST_SIZE,
5650 .arg4_type = ARG_ANYTHING,
5653 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
5654 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5656 struct net *net = dev_net(skb->dev);
5657 int rc = -EAFNOSUPPORT;
5658 bool check_mtu = false;
5660 if (plen < sizeof(*params))
5663 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5666 if (params->tot_len)
5669 switch (params->family) {
5670 #if IS_ENABLED(CONFIG_INET)
5672 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
5675 #if IS_ENABLED(CONFIG_IPV6)
5677 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
5682 if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
5683 struct net_device *dev;
5685 /* When tot_len isn't provided by user, check skb
5686 * against MTU of FIB lookup resulting net_device
5688 dev = dev_get_by_index_rcu(net, params->ifindex);
5689 if (!is_skb_forwardable(dev, skb))
5690 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
5692 params->mtu_result = dev->mtu; /* union with tot_len */
5698 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
5699 .func = bpf_skb_fib_lookup,
5701 .ret_type = RET_INTEGER,
5702 .arg1_type = ARG_PTR_TO_CTX,
5703 .arg2_type = ARG_PTR_TO_MEM,
5704 .arg3_type = ARG_CONST_SIZE,
5705 .arg4_type = ARG_ANYTHING,
5708 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
5711 struct net *netns = dev_net(dev_curr);
5713 /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
5717 return dev_get_by_index_rcu(netns, ifindex);
5720 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
5721 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5723 int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5724 struct net_device *dev = skb->dev;
5725 int skb_len, dev_len;
5728 if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
5731 if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
5734 dev = __dev_via_ifindex(dev, ifindex);
5738 mtu = READ_ONCE(dev->mtu);
5740 dev_len = mtu + dev->hard_header_len;
5742 /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
5743 skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
5745 skb_len += len_diff; /* minus result pass check */
5746 if (skb_len <= dev_len) {
5747 ret = BPF_MTU_CHK_RET_SUCCESS;
5750 /* At this point, skb->len exceed MTU, but as it include length of all
5751 * segments, it can still be below MTU. The SKB can possibly get
5752 * re-segmented in transmit path (see validate_xmit_skb). Thus, user
5753 * must choose if segs are to be MTU checked.
5755 if (skb_is_gso(skb)) {
5756 ret = BPF_MTU_CHK_RET_SUCCESS;
5758 if (flags & BPF_MTU_CHK_SEGS &&
5759 !skb_gso_validate_network_len(skb, mtu))
5760 ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
5763 /* BPF verifier guarantees valid pointer */
5769 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
5770 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5772 struct net_device *dev = xdp->rxq->dev;
5773 int xdp_len = xdp->data_end - xdp->data;
5774 int ret = BPF_MTU_CHK_RET_SUCCESS;
5777 /* XDP variant doesn't support multi-buffer segment check (yet) */
5778 if (unlikely(flags))
5781 dev = __dev_via_ifindex(dev, ifindex);
5785 mtu = READ_ONCE(dev->mtu);
5787 /* Add L2-header as dev MTU is L3 size */
5788 dev_len = mtu + dev->hard_header_len;
5790 /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
5792 xdp_len = *mtu_len + dev->hard_header_len;
5794 xdp_len += len_diff; /* minus result pass check */
5795 if (xdp_len > dev_len)
5796 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5798 /* BPF verifier guarantees valid pointer */
5804 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
5805 .func = bpf_skb_check_mtu,
5807 .ret_type = RET_INTEGER,
5808 .arg1_type = ARG_PTR_TO_CTX,
5809 .arg2_type = ARG_ANYTHING,
5810 .arg3_type = ARG_PTR_TO_INT,
5811 .arg4_type = ARG_ANYTHING,
5812 .arg5_type = ARG_ANYTHING,
5815 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
5816 .func = bpf_xdp_check_mtu,
5818 .ret_type = RET_INTEGER,
5819 .arg1_type = ARG_PTR_TO_CTX,
5820 .arg2_type = ARG_ANYTHING,
5821 .arg3_type = ARG_PTR_TO_INT,
5822 .arg4_type = ARG_ANYTHING,
5823 .arg5_type = ARG_ANYTHING,
5826 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5827 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
5830 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
5832 if (!seg6_validate_srh(srh, len, false))
5836 case BPF_LWT_ENCAP_SEG6_INLINE:
5837 if (skb->protocol != htons(ETH_P_IPV6))
5840 err = seg6_do_srh_inline(skb, srh);
5842 case BPF_LWT_ENCAP_SEG6:
5843 skb_reset_inner_headers(skb);
5844 skb->encapsulation = 1;
5845 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
5851 bpf_compute_data_pointers(skb);
5855 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
5857 return seg6_lookup_nexthop(skb, NULL, 0);
5859 #endif /* CONFIG_IPV6_SEG6_BPF */
5861 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5862 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
5865 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
5869 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
5873 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5874 case BPF_LWT_ENCAP_SEG6:
5875 case BPF_LWT_ENCAP_SEG6_INLINE:
5876 return bpf_push_seg6_encap(skb, type, hdr, len);
5878 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5879 case BPF_LWT_ENCAP_IP:
5880 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
5887 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
5888 void *, hdr, u32, len)
5891 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5892 case BPF_LWT_ENCAP_IP:
5893 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
5900 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
5901 .func = bpf_lwt_in_push_encap,
5903 .ret_type = RET_INTEGER,
5904 .arg1_type = ARG_PTR_TO_CTX,
5905 .arg2_type = ARG_ANYTHING,
5906 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5907 .arg4_type = ARG_CONST_SIZE
5910 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
5911 .func = bpf_lwt_xmit_push_encap,
5913 .ret_type = RET_INTEGER,
5914 .arg1_type = ARG_PTR_TO_CTX,
5915 .arg2_type = ARG_ANYTHING,
5916 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5917 .arg4_type = ARG_CONST_SIZE
5920 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5921 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
5922 const void *, from, u32, len)
5924 struct seg6_bpf_srh_state *srh_state =
5925 this_cpu_ptr(&seg6_bpf_srh_states);
5926 struct ipv6_sr_hdr *srh = srh_state->srh;
5927 void *srh_tlvs, *srh_end, *ptr;
5933 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
5934 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
5936 ptr = skb->data + offset;
5937 if (ptr >= srh_tlvs && ptr + len <= srh_end)
5938 srh_state->valid = false;
5939 else if (ptr < (void *)&srh->flags ||
5940 ptr + len > (void *)&srh->segments)
5943 if (unlikely(bpf_try_make_writable(skb, offset + len)))
5945 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5947 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5949 memcpy(skb->data + offset, from, len);
5953 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
5954 .func = bpf_lwt_seg6_store_bytes,
5956 .ret_type = RET_INTEGER,
5957 .arg1_type = ARG_PTR_TO_CTX,
5958 .arg2_type = ARG_ANYTHING,
5959 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5960 .arg4_type = ARG_CONST_SIZE
5963 static void bpf_update_srh_state(struct sk_buff *skb)
5965 struct seg6_bpf_srh_state *srh_state =
5966 this_cpu_ptr(&seg6_bpf_srh_states);
5969 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
5970 srh_state->srh = NULL;
5972 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5973 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
5974 srh_state->valid = true;
5978 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
5979 u32, action, void *, param, u32, param_len)
5981 struct seg6_bpf_srh_state *srh_state =
5982 this_cpu_ptr(&seg6_bpf_srh_states);
5987 case SEG6_LOCAL_ACTION_END_X:
5988 if (!seg6_bpf_has_valid_srh(skb))
5990 if (param_len != sizeof(struct in6_addr))
5992 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
5993 case SEG6_LOCAL_ACTION_END_T:
5994 if (!seg6_bpf_has_valid_srh(skb))
5996 if (param_len != sizeof(int))
5998 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5999 case SEG6_LOCAL_ACTION_END_DT6:
6000 if (!seg6_bpf_has_valid_srh(skb))
6002 if (param_len != sizeof(int))
6005 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
6007 if (!pskb_pull(skb, hdroff))
6010 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
6011 skb_reset_network_header(skb);
6012 skb_reset_transport_header(skb);
6013 skb->encapsulation = 0;
6015 bpf_compute_data_pointers(skb);
6016 bpf_update_srh_state(skb);
6017 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6018 case SEG6_LOCAL_ACTION_END_B6:
6019 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6021 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
6024 bpf_update_srh_state(skb);
6027 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
6028 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6030 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
6033 bpf_update_srh_state(skb);
6041 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
6042 .func = bpf_lwt_seg6_action,
6044 .ret_type = RET_INTEGER,
6045 .arg1_type = ARG_PTR_TO_CTX,
6046 .arg2_type = ARG_ANYTHING,
6047 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6048 .arg4_type = ARG_CONST_SIZE
6051 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
6054 struct seg6_bpf_srh_state *srh_state =
6055 this_cpu_ptr(&seg6_bpf_srh_states);
6056 struct ipv6_sr_hdr *srh = srh_state->srh;
6057 void *srh_end, *srh_tlvs, *ptr;
6058 struct ipv6hdr *hdr;
6062 if (unlikely(srh == NULL))
6065 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
6066 ((srh->first_segment + 1) << 4));
6067 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
6069 ptr = skb->data + offset;
6071 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
6073 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
6077 ret = skb_cow_head(skb, len);
6078 if (unlikely(ret < 0))
6081 ret = bpf_skb_net_hdr_push(skb, offset, len);
6083 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6086 bpf_compute_data_pointers(skb);
6087 if (unlikely(ret < 0))
6090 hdr = (struct ipv6hdr *)skb->data;
6091 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6093 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6095 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6096 srh_state->hdrlen += len;
6097 srh_state->valid = false;
6101 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6102 .func = bpf_lwt_seg6_adjust_srh,
6104 .ret_type = RET_INTEGER,
6105 .arg1_type = ARG_PTR_TO_CTX,
6106 .arg2_type = ARG_ANYTHING,
6107 .arg3_type = ARG_ANYTHING,
6109 #endif /* CONFIG_IPV6_SEG6_BPF */
6112 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6113 int dif, int sdif, u8 family, u8 proto)
6115 bool refcounted = false;
6116 struct sock *sk = NULL;
6118 if (family == AF_INET) {
6119 __be32 src4 = tuple->ipv4.saddr;
6120 __be32 dst4 = tuple->ipv4.daddr;
6122 if (proto == IPPROTO_TCP)
6123 sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
6124 src4, tuple->ipv4.sport,
6125 dst4, tuple->ipv4.dport,
6126 dif, sdif, &refcounted);
6128 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6129 dst4, tuple->ipv4.dport,
6130 dif, sdif, &udp_table, NULL);
6131 #if IS_ENABLED(CONFIG_IPV6)
6133 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6134 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6136 if (proto == IPPROTO_TCP)
6137 sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
6138 src6, tuple->ipv6.sport,
6139 dst6, ntohs(tuple->ipv6.dport),
6140 dif, sdif, &refcounted);
6141 else if (likely(ipv6_bpf_stub))
6142 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6143 src6, tuple->ipv6.sport,
6144 dst6, tuple->ipv6.dport,
6150 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6151 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6157 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6158 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6159 * Returns the socket as an 'unsigned long' to simplify the casting in the
6160 * callers to satisfy BPF_CALL declarations.
6162 static struct sock *
6163 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6164 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6167 struct sock *sk = NULL;
6168 u8 family = AF_UNSPEC;
6172 if (len == sizeof(tuple->ipv4))
6174 else if (len == sizeof(tuple->ipv6))
6179 if (unlikely(family == AF_UNSPEC || flags ||
6180 !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6183 if (family == AF_INET)
6184 sdif = inet_sdif(skb);
6186 sdif = inet6_sdif(skb);
6188 if ((s32)netns_id < 0) {
6190 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6192 net = get_net_ns_by_id(caller_net, netns_id);
6195 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6203 static struct sock *
6204 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6205 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6208 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6209 ifindex, proto, netns_id, flags);
6212 struct sock *sk2 = sk_to_full_sk(sk);
6214 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6215 * sock refcnt is decremented to prevent a request_sock leak.
6217 if (!sk_fullsock(sk2))
6221 /* Ensure there is no need to bump sk2 refcnt */
6222 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6223 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6233 static struct sock *
6234 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6235 u8 proto, u64 netns_id, u64 flags)
6237 struct net *caller_net;
6241 caller_net = dev_net(skb->dev);
6242 ifindex = skb->dev->ifindex;
6244 caller_net = sock_net(skb->sk);
6248 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6252 static struct sock *
6253 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6254 u8 proto, u64 netns_id, u64 flags)
6256 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6260 struct sock *sk2 = sk_to_full_sk(sk);
6262 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6263 * sock refcnt is decremented to prevent a request_sock leak.
6265 if (!sk_fullsock(sk2))
6269 /* Ensure there is no need to bump sk2 refcnt */
6270 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6271 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6281 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6282 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6284 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6288 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6289 .func = bpf_skc_lookup_tcp,
6292 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6293 .arg1_type = ARG_PTR_TO_CTX,
6294 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6295 .arg3_type = ARG_CONST_SIZE,
6296 .arg4_type = ARG_ANYTHING,
6297 .arg5_type = ARG_ANYTHING,
6300 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6301 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6303 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6307 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6308 .func = bpf_sk_lookup_tcp,
6311 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6312 .arg1_type = ARG_PTR_TO_CTX,
6313 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6314 .arg3_type = ARG_CONST_SIZE,
6315 .arg4_type = ARG_ANYTHING,
6316 .arg5_type = ARG_ANYTHING,
6319 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6320 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6322 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6326 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6327 .func = bpf_sk_lookup_udp,
6330 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6331 .arg1_type = ARG_PTR_TO_CTX,
6332 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6333 .arg3_type = ARG_CONST_SIZE,
6334 .arg4_type = ARG_ANYTHING,
6335 .arg5_type = ARG_ANYTHING,
6338 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6340 if (sk && sk_is_refcounted(sk))
6345 static const struct bpf_func_proto bpf_sk_release_proto = {
6346 .func = bpf_sk_release,
6348 .ret_type = RET_INTEGER,
6349 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6352 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6353 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6355 struct net *caller_net = dev_net(ctx->rxq->dev);
6356 int ifindex = ctx->rxq->dev->ifindex;
6358 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6359 ifindex, IPPROTO_UDP, netns_id,
6363 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6364 .func = bpf_xdp_sk_lookup_udp,
6367 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6368 .arg1_type = ARG_PTR_TO_CTX,
6369 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6370 .arg3_type = ARG_CONST_SIZE,
6371 .arg4_type = ARG_ANYTHING,
6372 .arg5_type = ARG_ANYTHING,
6375 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6376 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6378 struct net *caller_net = dev_net(ctx->rxq->dev);
6379 int ifindex = ctx->rxq->dev->ifindex;
6381 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6382 ifindex, IPPROTO_TCP, netns_id,
6386 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6387 .func = bpf_xdp_skc_lookup_tcp,
6390 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6391 .arg1_type = ARG_PTR_TO_CTX,
6392 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6393 .arg3_type = ARG_CONST_SIZE,
6394 .arg4_type = ARG_ANYTHING,
6395 .arg5_type = ARG_ANYTHING,
6398 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6399 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6401 struct net *caller_net = dev_net(ctx->rxq->dev);
6402 int ifindex = ctx->rxq->dev->ifindex;
6404 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6405 ifindex, IPPROTO_TCP, netns_id,
6409 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6410 .func = bpf_xdp_sk_lookup_tcp,
6413 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6414 .arg1_type = ARG_PTR_TO_CTX,
6415 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6416 .arg3_type = ARG_CONST_SIZE,
6417 .arg4_type = ARG_ANYTHING,
6418 .arg5_type = ARG_ANYTHING,
6421 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6422 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6424 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6425 sock_net(ctx->sk), 0,
6426 IPPROTO_TCP, netns_id, flags);
6429 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6430 .func = bpf_sock_addr_skc_lookup_tcp,
6432 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6433 .arg1_type = ARG_PTR_TO_CTX,
6434 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6435 .arg3_type = ARG_CONST_SIZE,
6436 .arg4_type = ARG_ANYTHING,
6437 .arg5_type = ARG_ANYTHING,
6440 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6441 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6443 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6444 sock_net(ctx->sk), 0, IPPROTO_TCP,
6448 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6449 .func = bpf_sock_addr_sk_lookup_tcp,
6451 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6452 .arg1_type = ARG_PTR_TO_CTX,
6453 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6454 .arg3_type = ARG_CONST_SIZE,
6455 .arg4_type = ARG_ANYTHING,
6456 .arg5_type = ARG_ANYTHING,
6459 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6460 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6462 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6463 sock_net(ctx->sk), 0, IPPROTO_UDP,
6467 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6468 .func = bpf_sock_addr_sk_lookup_udp,
6470 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6471 .arg1_type = ARG_PTR_TO_CTX,
6472 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6473 .arg3_type = ARG_CONST_SIZE,
6474 .arg4_type = ARG_ANYTHING,
6475 .arg5_type = ARG_ANYTHING,
6478 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6479 struct bpf_insn_access_aux *info)
6481 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6485 if (off % size != 0)
6489 case offsetof(struct bpf_tcp_sock, bytes_received):
6490 case offsetof(struct bpf_tcp_sock, bytes_acked):
6491 return size == sizeof(__u64);
6493 return size == sizeof(__u32);
6497 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6498 const struct bpf_insn *si,
6499 struct bpf_insn *insn_buf,
6500 struct bpf_prog *prog, u32 *target_size)
6502 struct bpf_insn *insn = insn_buf;
6504 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \
6506 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \
6507 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6508 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6509 si->dst_reg, si->src_reg, \
6510 offsetof(struct tcp_sock, FIELD)); \
6513 #define BPF_INET_SOCK_GET_COMMON(FIELD) \
6515 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \
6517 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6518 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6519 struct inet_connection_sock, \
6521 si->dst_reg, si->src_reg, \
6523 struct inet_connection_sock, \
6527 if (insn > insn_buf)
6528 return insn - insn_buf;
6531 case offsetof(struct bpf_tcp_sock, rtt_min):
6532 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
6533 sizeof(struct minmax));
6534 BUILD_BUG_ON(sizeof(struct minmax) <
6535 sizeof(struct minmax_sample));
6537 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6538 offsetof(struct tcp_sock, rtt_min) +
6539 offsetof(struct minmax_sample, v));
6541 case offsetof(struct bpf_tcp_sock, snd_cwnd):
6542 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
6544 case offsetof(struct bpf_tcp_sock, srtt_us):
6545 BPF_TCP_SOCK_GET_COMMON(srtt_us);
6547 case offsetof(struct bpf_tcp_sock, snd_ssthresh):
6548 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
6550 case offsetof(struct bpf_tcp_sock, rcv_nxt):
6551 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
6553 case offsetof(struct bpf_tcp_sock, snd_nxt):
6554 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
6556 case offsetof(struct bpf_tcp_sock, snd_una):
6557 BPF_TCP_SOCK_GET_COMMON(snd_una);
6559 case offsetof(struct bpf_tcp_sock, mss_cache):
6560 BPF_TCP_SOCK_GET_COMMON(mss_cache);
6562 case offsetof(struct bpf_tcp_sock, ecn_flags):
6563 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
6565 case offsetof(struct bpf_tcp_sock, rate_delivered):
6566 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
6568 case offsetof(struct bpf_tcp_sock, rate_interval_us):
6569 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
6571 case offsetof(struct bpf_tcp_sock, packets_out):
6572 BPF_TCP_SOCK_GET_COMMON(packets_out);
6574 case offsetof(struct bpf_tcp_sock, retrans_out):
6575 BPF_TCP_SOCK_GET_COMMON(retrans_out);
6577 case offsetof(struct bpf_tcp_sock, total_retrans):
6578 BPF_TCP_SOCK_GET_COMMON(total_retrans);
6580 case offsetof(struct bpf_tcp_sock, segs_in):
6581 BPF_TCP_SOCK_GET_COMMON(segs_in);
6583 case offsetof(struct bpf_tcp_sock, data_segs_in):
6584 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
6586 case offsetof(struct bpf_tcp_sock, segs_out):
6587 BPF_TCP_SOCK_GET_COMMON(segs_out);
6589 case offsetof(struct bpf_tcp_sock, data_segs_out):
6590 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
6592 case offsetof(struct bpf_tcp_sock, lost_out):
6593 BPF_TCP_SOCK_GET_COMMON(lost_out);
6595 case offsetof(struct bpf_tcp_sock, sacked_out):
6596 BPF_TCP_SOCK_GET_COMMON(sacked_out);
6598 case offsetof(struct bpf_tcp_sock, bytes_received):
6599 BPF_TCP_SOCK_GET_COMMON(bytes_received);
6601 case offsetof(struct bpf_tcp_sock, bytes_acked):
6602 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
6604 case offsetof(struct bpf_tcp_sock, dsack_dups):
6605 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
6607 case offsetof(struct bpf_tcp_sock, delivered):
6608 BPF_TCP_SOCK_GET_COMMON(delivered);
6610 case offsetof(struct bpf_tcp_sock, delivered_ce):
6611 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
6613 case offsetof(struct bpf_tcp_sock, icsk_retransmits):
6614 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
6618 return insn - insn_buf;
6621 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
6623 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
6624 return (unsigned long)sk;
6626 return (unsigned long)NULL;
6629 const struct bpf_func_proto bpf_tcp_sock_proto = {
6630 .func = bpf_tcp_sock,
6632 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
6633 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6636 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
6638 sk = sk_to_full_sk(sk);
6640 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
6641 return (unsigned long)sk;
6643 return (unsigned long)NULL;
6646 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
6647 .func = bpf_get_listener_sock,
6649 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6650 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6653 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
6655 unsigned int iphdr_len;
6657 switch (skb_protocol(skb, true)) {
6658 case cpu_to_be16(ETH_P_IP):
6659 iphdr_len = sizeof(struct iphdr);
6661 case cpu_to_be16(ETH_P_IPV6):
6662 iphdr_len = sizeof(struct ipv6hdr);
6668 if (skb_headlen(skb) < iphdr_len)
6671 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
6674 return INET_ECN_set_ce(skb);
6677 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6678 struct bpf_insn_access_aux *info)
6680 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
6683 if (off % size != 0)
6688 return size == sizeof(__u32);
6692 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
6693 const struct bpf_insn *si,
6694 struct bpf_insn *insn_buf,
6695 struct bpf_prog *prog, u32 *target_size)
6697 struct bpf_insn *insn = insn_buf;
6699 #define BPF_XDP_SOCK_GET(FIELD) \
6701 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \
6702 sizeof_field(struct bpf_xdp_sock, FIELD)); \
6703 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
6704 si->dst_reg, si->src_reg, \
6705 offsetof(struct xdp_sock, FIELD)); \
6709 case offsetof(struct bpf_xdp_sock, queue_id):
6710 BPF_XDP_SOCK_GET(queue_id);
6714 return insn - insn_buf;
6717 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
6718 .func = bpf_skb_ecn_set_ce,
6720 .ret_type = RET_INTEGER,
6721 .arg1_type = ARG_PTR_TO_CTX,
6724 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6725 struct tcphdr *, th, u32, th_len)
6727 #ifdef CONFIG_SYN_COOKIES
6731 if (unlikely(!sk || th_len < sizeof(*th)))
6734 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
6735 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6738 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
6741 if (!th->ack || th->rst || th->syn)
6744 if (unlikely(iph_len < sizeof(struct iphdr)))
6747 if (tcp_synq_no_recent_overflow(sk))
6750 cookie = ntohl(th->ack_seq) - 1;
6752 /* Both struct iphdr and struct ipv6hdr have the version field at the
6753 * same offset so we can cast to the shorter header (struct iphdr).
6755 switch (((struct iphdr *)iph)->version) {
6757 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
6760 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
6763 #if IS_BUILTIN(CONFIG_IPV6)
6765 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6768 if (sk->sk_family != AF_INET6)
6771 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
6773 #endif /* CONFIG_IPV6 */
6776 return -EPROTONOSUPPORT;
6788 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
6789 .func = bpf_tcp_check_syncookie,
6792 .ret_type = RET_INTEGER,
6793 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6794 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6795 .arg3_type = ARG_CONST_SIZE,
6796 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6797 .arg5_type = ARG_CONST_SIZE,
6800 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6801 struct tcphdr *, th, u32, th_len)
6803 #ifdef CONFIG_SYN_COOKIES
6807 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
6810 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6813 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
6816 if (!th->syn || th->ack || th->fin || th->rst)
6819 if (unlikely(iph_len < sizeof(struct iphdr)))
6822 /* Both struct iphdr and struct ipv6hdr have the version field at the
6823 * same offset so we can cast to the shorter header (struct iphdr).
6825 switch (((struct iphdr *)iph)->version) {
6827 if (sk->sk_family == AF_INET6 && sk->sk_ipv6only)
6830 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
6833 #if IS_BUILTIN(CONFIG_IPV6)
6835 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6838 if (sk->sk_family != AF_INET6)
6841 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
6843 #endif /* CONFIG_IPV6 */
6846 return -EPROTONOSUPPORT;
6851 return cookie | ((u64)mss << 32);
6854 #endif /* CONFIG_SYN_COOKIES */
6857 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
6858 .func = bpf_tcp_gen_syncookie,
6859 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */
6861 .ret_type = RET_INTEGER,
6862 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6863 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6864 .arg3_type = ARG_CONST_SIZE,
6865 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6866 .arg5_type = ARG_CONST_SIZE,
6869 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
6871 if (!sk || flags != 0)
6873 if (!skb_at_tc_ingress(skb))
6875 if (unlikely(dev_net(skb->dev) != sock_net(sk)))
6876 return -ENETUNREACH;
6877 if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
6878 return -ESOCKTNOSUPPORT;
6879 if (sk_is_refcounted(sk) &&
6880 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
6885 skb->destructor = sock_pfree;
6890 static const struct bpf_func_proto bpf_sk_assign_proto = {
6891 .func = bpf_sk_assign,
6893 .ret_type = RET_INTEGER,
6894 .arg1_type = ARG_PTR_TO_CTX,
6895 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6896 .arg3_type = ARG_ANYTHING,
6899 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
6900 u8 search_kind, const u8 *magic,
6901 u8 magic_len, bool *eol)
6907 while (op < opend) {
6910 if (kind == TCPOPT_EOL) {
6912 return ERR_PTR(-ENOMSG);
6913 } else if (kind == TCPOPT_NOP) {
6918 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
6919 /* Something is wrong in the received header.
6920 * Follow the TCP stack's tcp_parse_options()
6921 * and just bail here.
6923 return ERR_PTR(-EFAULT);
6926 if (search_kind == kind) {
6930 if (magic_len > kind_len - 2)
6931 return ERR_PTR(-ENOMSG);
6933 if (!memcmp(&op[2], magic, magic_len))
6940 return ERR_PTR(-ENOMSG);
6943 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6944 void *, search_res, u32, len, u64, flags)
6946 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
6947 const u8 *op, *opend, *magic, *search = search_res;
6948 u8 search_kind, search_len, copy_len, magic_len;
6951 /* 2 byte is the minimal option len except TCPOPT_NOP and
6952 * TCPOPT_EOL which are useless for the bpf prog to learn
6953 * and this helper disallow loading them also.
6955 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
6958 search_kind = search[0];
6959 search_len = search[1];
6961 if (search_len > len || search_kind == TCPOPT_NOP ||
6962 search_kind == TCPOPT_EOL)
6965 if (search_kind == TCPOPT_EXP || search_kind == 253) {
6966 /* 16 or 32 bit magic. +2 for kind and kind length */
6967 if (search_len != 4 && search_len != 6)
6970 magic_len = search_len - 2;
6979 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
6984 op += sizeof(struct tcphdr);
6986 if (!bpf_sock->skb ||
6987 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6988 /* This bpf_sock->op cannot call this helper */
6991 opend = bpf_sock->skb_data_end;
6992 op = bpf_sock->skb->data + sizeof(struct tcphdr);
6995 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
7002 if (copy_len > len) {
7007 memcpy(search_res, op, copy_len);
7011 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
7012 .func = bpf_sock_ops_load_hdr_opt,
7014 .ret_type = RET_INTEGER,
7015 .arg1_type = ARG_PTR_TO_CTX,
7016 .arg2_type = ARG_PTR_TO_MEM,
7017 .arg3_type = ARG_CONST_SIZE,
7018 .arg4_type = ARG_ANYTHING,
7021 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7022 const void *, from, u32, len, u64, flags)
7024 u8 new_kind, new_kind_len, magic_len = 0, *opend;
7025 const u8 *op, *new_op, *magic = NULL;
7026 struct sk_buff *skb;
7029 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
7032 if (len < 2 || flags)
7036 new_kind = new_op[0];
7037 new_kind_len = new_op[1];
7039 if (new_kind_len > len || new_kind == TCPOPT_NOP ||
7040 new_kind == TCPOPT_EOL)
7043 if (new_kind_len > bpf_sock->remaining_opt_len)
7046 /* 253 is another experimental kind */
7047 if (new_kind == TCPOPT_EXP || new_kind == 253) {
7048 if (new_kind_len < 4)
7050 /* Match for the 2 byte magic also.
7051 * RFC 6994: the magic could be 2 or 4 bytes.
7052 * Hence, matching by 2 byte only is on the
7053 * conservative side but it is the right
7054 * thing to do for the 'search-for-duplication'
7061 /* Check for duplication */
7062 skb = bpf_sock->skb;
7063 op = skb->data + sizeof(struct tcphdr);
7064 opend = bpf_sock->skb_data_end;
7066 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
7071 if (PTR_ERR(op) != -ENOMSG)
7075 /* The option has been ended. Treat it as no more
7076 * header option can be written.
7080 /* No duplication found. Store the header option. */
7081 memcpy(opend, from, new_kind_len);
7083 bpf_sock->remaining_opt_len -= new_kind_len;
7084 bpf_sock->skb_data_end += new_kind_len;
7089 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
7090 .func = bpf_sock_ops_store_hdr_opt,
7092 .ret_type = RET_INTEGER,
7093 .arg1_type = ARG_PTR_TO_CTX,
7094 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7095 .arg3_type = ARG_CONST_SIZE,
7096 .arg4_type = ARG_ANYTHING,
7099 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7100 u32, len, u64, flags)
7102 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7105 if (flags || len < 2)
7108 if (len > bpf_sock->remaining_opt_len)
7111 bpf_sock->remaining_opt_len -= len;
7116 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7117 .func = bpf_sock_ops_reserve_hdr_opt,
7119 .ret_type = RET_INTEGER,
7120 .arg1_type = ARG_PTR_TO_CTX,
7121 .arg2_type = ARG_ANYTHING,
7122 .arg3_type = ARG_ANYTHING,
7125 #endif /* CONFIG_INET */
7127 bool bpf_helper_changes_pkt_data(void *func)
7129 if (func == bpf_skb_vlan_push ||
7130 func == bpf_skb_vlan_pop ||
7131 func == bpf_skb_store_bytes ||
7132 func == bpf_skb_change_proto ||
7133 func == bpf_skb_change_head ||
7134 func == sk_skb_change_head ||
7135 func == bpf_skb_change_tail ||
7136 func == sk_skb_change_tail ||
7137 func == bpf_skb_adjust_room ||
7138 func == sk_skb_adjust_room ||
7139 func == bpf_skb_pull_data ||
7140 func == sk_skb_pull_data ||
7141 func == bpf_clone_redirect ||
7142 func == bpf_l3_csum_replace ||
7143 func == bpf_l4_csum_replace ||
7144 func == bpf_xdp_adjust_head ||
7145 func == bpf_xdp_adjust_meta ||
7146 func == bpf_msg_pull_data ||
7147 func == bpf_msg_push_data ||
7148 func == bpf_msg_pop_data ||
7149 func == bpf_xdp_adjust_tail ||
7150 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7151 func == bpf_lwt_seg6_store_bytes ||
7152 func == bpf_lwt_seg6_adjust_srh ||
7153 func == bpf_lwt_seg6_action ||
7156 func == bpf_sock_ops_store_hdr_opt ||
7158 func == bpf_lwt_in_push_encap ||
7159 func == bpf_lwt_xmit_push_encap)
7165 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7166 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7168 static const struct bpf_func_proto *
7169 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7172 /* inet and inet6 sockets are created in a process
7173 * context so there is always a valid uid/gid
7175 case BPF_FUNC_get_current_uid_gid:
7176 return &bpf_get_current_uid_gid_proto;
7177 case BPF_FUNC_get_local_storage:
7178 return &bpf_get_local_storage_proto;
7179 case BPF_FUNC_get_socket_cookie:
7180 return &bpf_get_socket_cookie_sock_proto;
7181 case BPF_FUNC_get_netns_cookie:
7182 return &bpf_get_netns_cookie_sock_proto;
7183 case BPF_FUNC_perf_event_output:
7184 return &bpf_event_output_data_proto;
7185 case BPF_FUNC_get_current_pid_tgid:
7186 return &bpf_get_current_pid_tgid_proto;
7187 case BPF_FUNC_get_current_comm:
7188 return &bpf_get_current_comm_proto;
7189 #ifdef CONFIG_CGROUPS
7190 case BPF_FUNC_get_current_cgroup_id:
7191 return &bpf_get_current_cgroup_id_proto;
7192 case BPF_FUNC_get_current_ancestor_cgroup_id:
7193 return &bpf_get_current_ancestor_cgroup_id_proto;
7195 #ifdef CONFIG_CGROUP_NET_CLASSID
7196 case BPF_FUNC_get_cgroup_classid:
7197 return &bpf_get_cgroup_classid_curr_proto;
7199 case BPF_FUNC_sk_storage_get:
7200 return &bpf_sk_storage_get_cg_sock_proto;
7201 case BPF_FUNC_ktime_get_coarse_ns:
7202 return &bpf_ktime_get_coarse_ns_proto;
7204 return bpf_base_func_proto(func_id);
7208 static const struct bpf_func_proto *
7209 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7212 /* inet and inet6 sockets are created in a process
7213 * context so there is always a valid uid/gid
7215 case BPF_FUNC_get_current_uid_gid:
7216 return &bpf_get_current_uid_gid_proto;
7218 switch (prog->expected_attach_type) {
7219 case BPF_CGROUP_INET4_CONNECT:
7220 case BPF_CGROUP_INET6_CONNECT:
7221 return &bpf_bind_proto;
7225 case BPF_FUNC_get_socket_cookie:
7226 return &bpf_get_socket_cookie_sock_addr_proto;
7227 case BPF_FUNC_get_netns_cookie:
7228 return &bpf_get_netns_cookie_sock_addr_proto;
7229 case BPF_FUNC_get_local_storage:
7230 return &bpf_get_local_storage_proto;
7231 case BPF_FUNC_perf_event_output:
7232 return &bpf_event_output_data_proto;
7233 case BPF_FUNC_get_current_pid_tgid:
7234 return &bpf_get_current_pid_tgid_proto;
7235 case BPF_FUNC_get_current_comm:
7236 return &bpf_get_current_comm_proto;
7237 #ifdef CONFIG_CGROUPS
7238 case BPF_FUNC_get_current_cgroup_id:
7239 return &bpf_get_current_cgroup_id_proto;
7240 case BPF_FUNC_get_current_ancestor_cgroup_id:
7241 return &bpf_get_current_ancestor_cgroup_id_proto;
7243 #ifdef CONFIG_CGROUP_NET_CLASSID
7244 case BPF_FUNC_get_cgroup_classid:
7245 return &bpf_get_cgroup_classid_curr_proto;
7248 case BPF_FUNC_sk_lookup_tcp:
7249 return &bpf_sock_addr_sk_lookup_tcp_proto;
7250 case BPF_FUNC_sk_lookup_udp:
7251 return &bpf_sock_addr_sk_lookup_udp_proto;
7252 case BPF_FUNC_sk_release:
7253 return &bpf_sk_release_proto;
7254 case BPF_FUNC_skc_lookup_tcp:
7255 return &bpf_sock_addr_skc_lookup_tcp_proto;
7256 #endif /* CONFIG_INET */
7257 case BPF_FUNC_sk_storage_get:
7258 return &bpf_sk_storage_get_proto;
7259 case BPF_FUNC_sk_storage_delete:
7260 return &bpf_sk_storage_delete_proto;
7261 case BPF_FUNC_setsockopt:
7262 switch (prog->expected_attach_type) {
7263 case BPF_CGROUP_INET4_BIND:
7264 case BPF_CGROUP_INET6_BIND:
7265 case BPF_CGROUP_INET4_CONNECT:
7266 case BPF_CGROUP_INET6_CONNECT:
7267 case BPF_CGROUP_UDP4_RECVMSG:
7268 case BPF_CGROUP_UDP6_RECVMSG:
7269 case BPF_CGROUP_UDP4_SENDMSG:
7270 case BPF_CGROUP_UDP6_SENDMSG:
7271 case BPF_CGROUP_INET4_GETPEERNAME:
7272 case BPF_CGROUP_INET6_GETPEERNAME:
7273 case BPF_CGROUP_INET4_GETSOCKNAME:
7274 case BPF_CGROUP_INET6_GETSOCKNAME:
7275 return &bpf_sock_addr_setsockopt_proto;
7279 case BPF_FUNC_getsockopt:
7280 switch (prog->expected_attach_type) {
7281 case BPF_CGROUP_INET4_BIND:
7282 case BPF_CGROUP_INET6_BIND:
7283 case BPF_CGROUP_INET4_CONNECT:
7284 case BPF_CGROUP_INET6_CONNECT:
7285 case BPF_CGROUP_UDP4_RECVMSG:
7286 case BPF_CGROUP_UDP6_RECVMSG:
7287 case BPF_CGROUP_UDP4_SENDMSG:
7288 case BPF_CGROUP_UDP6_SENDMSG:
7289 case BPF_CGROUP_INET4_GETPEERNAME:
7290 case BPF_CGROUP_INET6_GETPEERNAME:
7291 case BPF_CGROUP_INET4_GETSOCKNAME:
7292 case BPF_CGROUP_INET6_GETSOCKNAME:
7293 return &bpf_sock_addr_getsockopt_proto;
7298 return bpf_sk_base_func_proto(func_id);
7302 static const struct bpf_func_proto *
7303 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7306 case BPF_FUNC_skb_load_bytes:
7307 return &bpf_skb_load_bytes_proto;
7308 case BPF_FUNC_skb_load_bytes_relative:
7309 return &bpf_skb_load_bytes_relative_proto;
7310 case BPF_FUNC_get_socket_cookie:
7311 return &bpf_get_socket_cookie_proto;
7312 case BPF_FUNC_get_socket_uid:
7313 return &bpf_get_socket_uid_proto;
7314 case BPF_FUNC_perf_event_output:
7315 return &bpf_skb_event_output_proto;
7317 return bpf_sk_base_func_proto(func_id);
7321 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7322 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7324 static const struct bpf_func_proto *
7325 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7328 case BPF_FUNC_get_local_storage:
7329 return &bpf_get_local_storage_proto;
7330 case BPF_FUNC_sk_fullsock:
7331 return &bpf_sk_fullsock_proto;
7332 case BPF_FUNC_sk_storage_get:
7333 return &bpf_sk_storage_get_proto;
7334 case BPF_FUNC_sk_storage_delete:
7335 return &bpf_sk_storage_delete_proto;
7336 case BPF_FUNC_perf_event_output:
7337 return &bpf_skb_event_output_proto;
7338 #ifdef CONFIG_SOCK_CGROUP_DATA
7339 case BPF_FUNC_skb_cgroup_id:
7340 return &bpf_skb_cgroup_id_proto;
7341 case BPF_FUNC_skb_ancestor_cgroup_id:
7342 return &bpf_skb_ancestor_cgroup_id_proto;
7343 case BPF_FUNC_sk_cgroup_id:
7344 return &bpf_sk_cgroup_id_proto;
7345 case BPF_FUNC_sk_ancestor_cgroup_id:
7346 return &bpf_sk_ancestor_cgroup_id_proto;
7349 case BPF_FUNC_sk_lookup_tcp:
7350 return &bpf_sk_lookup_tcp_proto;
7351 case BPF_FUNC_sk_lookup_udp:
7352 return &bpf_sk_lookup_udp_proto;
7353 case BPF_FUNC_sk_release:
7354 return &bpf_sk_release_proto;
7355 case BPF_FUNC_skc_lookup_tcp:
7356 return &bpf_skc_lookup_tcp_proto;
7357 case BPF_FUNC_tcp_sock:
7358 return &bpf_tcp_sock_proto;
7359 case BPF_FUNC_get_listener_sock:
7360 return &bpf_get_listener_sock_proto;
7361 case BPF_FUNC_skb_ecn_set_ce:
7362 return &bpf_skb_ecn_set_ce_proto;
7365 return sk_filter_func_proto(func_id, prog);
7369 static const struct bpf_func_proto *
7370 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7373 case BPF_FUNC_skb_store_bytes:
7374 return &bpf_skb_store_bytes_proto;
7375 case BPF_FUNC_skb_load_bytes:
7376 return &bpf_skb_load_bytes_proto;
7377 case BPF_FUNC_skb_load_bytes_relative:
7378 return &bpf_skb_load_bytes_relative_proto;
7379 case BPF_FUNC_skb_pull_data:
7380 return &bpf_skb_pull_data_proto;
7381 case BPF_FUNC_csum_diff:
7382 return &bpf_csum_diff_proto;
7383 case BPF_FUNC_csum_update:
7384 return &bpf_csum_update_proto;
7385 case BPF_FUNC_csum_level:
7386 return &bpf_csum_level_proto;
7387 case BPF_FUNC_l3_csum_replace:
7388 return &bpf_l3_csum_replace_proto;
7389 case BPF_FUNC_l4_csum_replace:
7390 return &bpf_l4_csum_replace_proto;
7391 case BPF_FUNC_clone_redirect:
7392 return &bpf_clone_redirect_proto;
7393 case BPF_FUNC_get_cgroup_classid:
7394 return &bpf_get_cgroup_classid_proto;
7395 case BPF_FUNC_skb_vlan_push:
7396 return &bpf_skb_vlan_push_proto;
7397 case BPF_FUNC_skb_vlan_pop:
7398 return &bpf_skb_vlan_pop_proto;
7399 case BPF_FUNC_skb_change_proto:
7400 return &bpf_skb_change_proto_proto;
7401 case BPF_FUNC_skb_change_type:
7402 return &bpf_skb_change_type_proto;
7403 case BPF_FUNC_skb_adjust_room:
7404 return &bpf_skb_adjust_room_proto;
7405 case BPF_FUNC_skb_change_tail:
7406 return &bpf_skb_change_tail_proto;
7407 case BPF_FUNC_skb_change_head:
7408 return &bpf_skb_change_head_proto;
7409 case BPF_FUNC_skb_get_tunnel_key:
7410 return &bpf_skb_get_tunnel_key_proto;
7411 case BPF_FUNC_skb_set_tunnel_key:
7412 return bpf_get_skb_set_tunnel_proto(func_id);
7413 case BPF_FUNC_skb_get_tunnel_opt:
7414 return &bpf_skb_get_tunnel_opt_proto;
7415 case BPF_FUNC_skb_set_tunnel_opt:
7416 return bpf_get_skb_set_tunnel_proto(func_id);
7417 case BPF_FUNC_redirect:
7418 return &bpf_redirect_proto;
7419 case BPF_FUNC_redirect_neigh:
7420 return &bpf_redirect_neigh_proto;
7421 case BPF_FUNC_redirect_peer:
7422 return &bpf_redirect_peer_proto;
7423 case BPF_FUNC_get_route_realm:
7424 return &bpf_get_route_realm_proto;
7425 case BPF_FUNC_get_hash_recalc:
7426 return &bpf_get_hash_recalc_proto;
7427 case BPF_FUNC_set_hash_invalid:
7428 return &bpf_set_hash_invalid_proto;
7429 case BPF_FUNC_set_hash:
7430 return &bpf_set_hash_proto;
7431 case BPF_FUNC_perf_event_output:
7432 return &bpf_skb_event_output_proto;
7433 case BPF_FUNC_get_smp_processor_id:
7434 return &bpf_get_smp_processor_id_proto;
7435 case BPF_FUNC_skb_under_cgroup:
7436 return &bpf_skb_under_cgroup_proto;
7437 case BPF_FUNC_get_socket_cookie:
7438 return &bpf_get_socket_cookie_proto;
7439 case BPF_FUNC_get_socket_uid:
7440 return &bpf_get_socket_uid_proto;
7441 case BPF_FUNC_fib_lookup:
7442 return &bpf_skb_fib_lookup_proto;
7443 case BPF_FUNC_check_mtu:
7444 return &bpf_skb_check_mtu_proto;
7445 case BPF_FUNC_sk_fullsock:
7446 return &bpf_sk_fullsock_proto;
7447 case BPF_FUNC_sk_storage_get:
7448 return &bpf_sk_storage_get_proto;
7449 case BPF_FUNC_sk_storage_delete:
7450 return &bpf_sk_storage_delete_proto;
7452 case BPF_FUNC_skb_get_xfrm_state:
7453 return &bpf_skb_get_xfrm_state_proto;
7455 #ifdef CONFIG_CGROUP_NET_CLASSID
7456 case BPF_FUNC_skb_cgroup_classid:
7457 return &bpf_skb_cgroup_classid_proto;
7459 #ifdef CONFIG_SOCK_CGROUP_DATA
7460 case BPF_FUNC_skb_cgroup_id:
7461 return &bpf_skb_cgroup_id_proto;
7462 case BPF_FUNC_skb_ancestor_cgroup_id:
7463 return &bpf_skb_ancestor_cgroup_id_proto;
7466 case BPF_FUNC_sk_lookup_tcp:
7467 return &bpf_sk_lookup_tcp_proto;
7468 case BPF_FUNC_sk_lookup_udp:
7469 return &bpf_sk_lookup_udp_proto;
7470 case BPF_FUNC_sk_release:
7471 return &bpf_sk_release_proto;
7472 case BPF_FUNC_tcp_sock:
7473 return &bpf_tcp_sock_proto;
7474 case BPF_FUNC_get_listener_sock:
7475 return &bpf_get_listener_sock_proto;
7476 case BPF_FUNC_skc_lookup_tcp:
7477 return &bpf_skc_lookup_tcp_proto;
7478 case BPF_FUNC_tcp_check_syncookie:
7479 return &bpf_tcp_check_syncookie_proto;
7480 case BPF_FUNC_skb_ecn_set_ce:
7481 return &bpf_skb_ecn_set_ce_proto;
7482 case BPF_FUNC_tcp_gen_syncookie:
7483 return &bpf_tcp_gen_syncookie_proto;
7484 case BPF_FUNC_sk_assign:
7485 return &bpf_sk_assign_proto;
7488 return bpf_sk_base_func_proto(func_id);
7492 static const struct bpf_func_proto *
7493 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7496 case BPF_FUNC_perf_event_output:
7497 return &bpf_xdp_event_output_proto;
7498 case BPF_FUNC_get_smp_processor_id:
7499 return &bpf_get_smp_processor_id_proto;
7500 case BPF_FUNC_csum_diff:
7501 return &bpf_csum_diff_proto;
7502 case BPF_FUNC_xdp_adjust_head:
7503 return &bpf_xdp_adjust_head_proto;
7504 case BPF_FUNC_xdp_adjust_meta:
7505 return &bpf_xdp_adjust_meta_proto;
7506 case BPF_FUNC_redirect:
7507 return &bpf_xdp_redirect_proto;
7508 case BPF_FUNC_redirect_map:
7509 return &bpf_xdp_redirect_map_proto;
7510 case BPF_FUNC_xdp_adjust_tail:
7511 return &bpf_xdp_adjust_tail_proto;
7512 case BPF_FUNC_fib_lookup:
7513 return &bpf_xdp_fib_lookup_proto;
7514 case BPF_FUNC_check_mtu:
7515 return &bpf_xdp_check_mtu_proto;
7517 case BPF_FUNC_sk_lookup_udp:
7518 return &bpf_xdp_sk_lookup_udp_proto;
7519 case BPF_FUNC_sk_lookup_tcp:
7520 return &bpf_xdp_sk_lookup_tcp_proto;
7521 case BPF_FUNC_sk_release:
7522 return &bpf_sk_release_proto;
7523 case BPF_FUNC_skc_lookup_tcp:
7524 return &bpf_xdp_skc_lookup_tcp_proto;
7525 case BPF_FUNC_tcp_check_syncookie:
7526 return &bpf_tcp_check_syncookie_proto;
7527 case BPF_FUNC_tcp_gen_syncookie:
7528 return &bpf_tcp_gen_syncookie_proto;
7531 return bpf_sk_base_func_proto(func_id);
7535 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
7536 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
7538 static const struct bpf_func_proto *
7539 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7542 case BPF_FUNC_setsockopt:
7543 return &bpf_sock_ops_setsockopt_proto;
7544 case BPF_FUNC_getsockopt:
7545 return &bpf_sock_ops_getsockopt_proto;
7546 case BPF_FUNC_sock_ops_cb_flags_set:
7547 return &bpf_sock_ops_cb_flags_set_proto;
7548 case BPF_FUNC_sock_map_update:
7549 return &bpf_sock_map_update_proto;
7550 case BPF_FUNC_sock_hash_update:
7551 return &bpf_sock_hash_update_proto;
7552 case BPF_FUNC_get_socket_cookie:
7553 return &bpf_get_socket_cookie_sock_ops_proto;
7554 case BPF_FUNC_get_local_storage:
7555 return &bpf_get_local_storage_proto;
7556 case BPF_FUNC_perf_event_output:
7557 return &bpf_event_output_data_proto;
7558 case BPF_FUNC_sk_storage_get:
7559 return &bpf_sk_storage_get_proto;
7560 case BPF_FUNC_sk_storage_delete:
7561 return &bpf_sk_storage_delete_proto;
7562 case BPF_FUNC_get_netns_cookie:
7563 return &bpf_get_netns_cookie_sock_ops_proto;
7565 case BPF_FUNC_load_hdr_opt:
7566 return &bpf_sock_ops_load_hdr_opt_proto;
7567 case BPF_FUNC_store_hdr_opt:
7568 return &bpf_sock_ops_store_hdr_opt_proto;
7569 case BPF_FUNC_reserve_hdr_opt:
7570 return &bpf_sock_ops_reserve_hdr_opt_proto;
7571 case BPF_FUNC_tcp_sock:
7572 return &bpf_tcp_sock_proto;
7573 #endif /* CONFIG_INET */
7575 return bpf_sk_base_func_proto(func_id);
7579 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
7580 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
7582 static const struct bpf_func_proto *
7583 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7586 case BPF_FUNC_msg_redirect_map:
7587 return &bpf_msg_redirect_map_proto;
7588 case BPF_FUNC_msg_redirect_hash:
7589 return &bpf_msg_redirect_hash_proto;
7590 case BPF_FUNC_msg_apply_bytes:
7591 return &bpf_msg_apply_bytes_proto;
7592 case BPF_FUNC_msg_cork_bytes:
7593 return &bpf_msg_cork_bytes_proto;
7594 case BPF_FUNC_msg_pull_data:
7595 return &bpf_msg_pull_data_proto;
7596 case BPF_FUNC_msg_push_data:
7597 return &bpf_msg_push_data_proto;
7598 case BPF_FUNC_msg_pop_data:
7599 return &bpf_msg_pop_data_proto;
7600 case BPF_FUNC_perf_event_output:
7601 return &bpf_event_output_data_proto;
7602 case BPF_FUNC_get_current_uid_gid:
7603 return &bpf_get_current_uid_gid_proto;
7604 case BPF_FUNC_get_current_pid_tgid:
7605 return &bpf_get_current_pid_tgid_proto;
7606 case BPF_FUNC_sk_storage_get:
7607 return &bpf_sk_storage_get_proto;
7608 case BPF_FUNC_sk_storage_delete:
7609 return &bpf_sk_storage_delete_proto;
7610 case BPF_FUNC_get_netns_cookie:
7611 return &bpf_get_netns_cookie_sk_msg_proto;
7612 #ifdef CONFIG_CGROUPS
7613 case BPF_FUNC_get_current_cgroup_id:
7614 return &bpf_get_current_cgroup_id_proto;
7615 case BPF_FUNC_get_current_ancestor_cgroup_id:
7616 return &bpf_get_current_ancestor_cgroup_id_proto;
7618 #ifdef CONFIG_CGROUP_NET_CLASSID
7619 case BPF_FUNC_get_cgroup_classid:
7620 return &bpf_get_cgroup_classid_curr_proto;
7623 return bpf_sk_base_func_proto(func_id);
7627 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
7628 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
7630 static const struct bpf_func_proto *
7631 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7634 case BPF_FUNC_skb_store_bytes:
7635 return &bpf_skb_store_bytes_proto;
7636 case BPF_FUNC_skb_load_bytes:
7637 return &bpf_skb_load_bytes_proto;
7638 case BPF_FUNC_skb_pull_data:
7639 return &sk_skb_pull_data_proto;
7640 case BPF_FUNC_skb_change_tail:
7641 return &sk_skb_change_tail_proto;
7642 case BPF_FUNC_skb_change_head:
7643 return &sk_skb_change_head_proto;
7644 case BPF_FUNC_skb_adjust_room:
7645 return &sk_skb_adjust_room_proto;
7646 case BPF_FUNC_get_socket_cookie:
7647 return &bpf_get_socket_cookie_proto;
7648 case BPF_FUNC_get_socket_uid:
7649 return &bpf_get_socket_uid_proto;
7650 case BPF_FUNC_sk_redirect_map:
7651 return &bpf_sk_redirect_map_proto;
7652 case BPF_FUNC_sk_redirect_hash:
7653 return &bpf_sk_redirect_hash_proto;
7654 case BPF_FUNC_perf_event_output:
7655 return &bpf_skb_event_output_proto;
7657 case BPF_FUNC_sk_lookup_tcp:
7658 return &bpf_sk_lookup_tcp_proto;
7659 case BPF_FUNC_sk_lookup_udp:
7660 return &bpf_sk_lookup_udp_proto;
7661 case BPF_FUNC_sk_release:
7662 return &bpf_sk_release_proto;
7663 case BPF_FUNC_skc_lookup_tcp:
7664 return &bpf_skc_lookup_tcp_proto;
7667 return bpf_sk_base_func_proto(func_id);
7671 static const struct bpf_func_proto *
7672 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7675 case BPF_FUNC_skb_load_bytes:
7676 return &bpf_flow_dissector_load_bytes_proto;
7678 return bpf_sk_base_func_proto(func_id);
7682 static const struct bpf_func_proto *
7683 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7686 case BPF_FUNC_skb_load_bytes:
7687 return &bpf_skb_load_bytes_proto;
7688 case BPF_FUNC_skb_pull_data:
7689 return &bpf_skb_pull_data_proto;
7690 case BPF_FUNC_csum_diff:
7691 return &bpf_csum_diff_proto;
7692 case BPF_FUNC_get_cgroup_classid:
7693 return &bpf_get_cgroup_classid_proto;
7694 case BPF_FUNC_get_route_realm:
7695 return &bpf_get_route_realm_proto;
7696 case BPF_FUNC_get_hash_recalc:
7697 return &bpf_get_hash_recalc_proto;
7698 case BPF_FUNC_perf_event_output:
7699 return &bpf_skb_event_output_proto;
7700 case BPF_FUNC_get_smp_processor_id:
7701 return &bpf_get_smp_processor_id_proto;
7702 case BPF_FUNC_skb_under_cgroup:
7703 return &bpf_skb_under_cgroup_proto;
7705 return bpf_sk_base_func_proto(func_id);
7709 static const struct bpf_func_proto *
7710 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7713 case BPF_FUNC_lwt_push_encap:
7714 return &bpf_lwt_in_push_encap_proto;
7716 return lwt_out_func_proto(func_id, prog);
7720 static const struct bpf_func_proto *
7721 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7724 case BPF_FUNC_skb_get_tunnel_key:
7725 return &bpf_skb_get_tunnel_key_proto;
7726 case BPF_FUNC_skb_set_tunnel_key:
7727 return bpf_get_skb_set_tunnel_proto(func_id);
7728 case BPF_FUNC_skb_get_tunnel_opt:
7729 return &bpf_skb_get_tunnel_opt_proto;
7730 case BPF_FUNC_skb_set_tunnel_opt:
7731 return bpf_get_skb_set_tunnel_proto(func_id);
7732 case BPF_FUNC_redirect:
7733 return &bpf_redirect_proto;
7734 case BPF_FUNC_clone_redirect:
7735 return &bpf_clone_redirect_proto;
7736 case BPF_FUNC_skb_change_tail:
7737 return &bpf_skb_change_tail_proto;
7738 case BPF_FUNC_skb_change_head:
7739 return &bpf_skb_change_head_proto;
7740 case BPF_FUNC_skb_store_bytes:
7741 return &bpf_skb_store_bytes_proto;
7742 case BPF_FUNC_csum_update:
7743 return &bpf_csum_update_proto;
7744 case BPF_FUNC_csum_level:
7745 return &bpf_csum_level_proto;
7746 case BPF_FUNC_l3_csum_replace:
7747 return &bpf_l3_csum_replace_proto;
7748 case BPF_FUNC_l4_csum_replace:
7749 return &bpf_l4_csum_replace_proto;
7750 case BPF_FUNC_set_hash_invalid:
7751 return &bpf_set_hash_invalid_proto;
7752 case BPF_FUNC_lwt_push_encap:
7753 return &bpf_lwt_xmit_push_encap_proto;
7755 return lwt_out_func_proto(func_id, prog);
7759 static const struct bpf_func_proto *
7760 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7763 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7764 case BPF_FUNC_lwt_seg6_store_bytes:
7765 return &bpf_lwt_seg6_store_bytes_proto;
7766 case BPF_FUNC_lwt_seg6_action:
7767 return &bpf_lwt_seg6_action_proto;
7768 case BPF_FUNC_lwt_seg6_adjust_srh:
7769 return &bpf_lwt_seg6_adjust_srh_proto;
7772 return lwt_out_func_proto(func_id, prog);
7776 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
7777 const struct bpf_prog *prog,
7778 struct bpf_insn_access_aux *info)
7780 const int size_default = sizeof(__u32);
7782 if (off < 0 || off >= sizeof(struct __sk_buff))
7785 /* The verifier guarantees that size > 0. */
7786 if (off % size != 0)
7790 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7791 if (off + size > offsetofend(struct __sk_buff, cb[4]))
7794 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
7795 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
7796 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
7797 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
7798 case bpf_ctx_range(struct __sk_buff, data):
7799 case bpf_ctx_range(struct __sk_buff, data_meta):
7800 case bpf_ctx_range(struct __sk_buff, data_end):
7801 if (size != size_default)
7804 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
7806 case bpf_ctx_range(struct __sk_buff, tstamp):
7807 if (size != sizeof(__u64))
7810 case offsetof(struct __sk_buff, sk):
7811 if (type == BPF_WRITE || size != sizeof(__u64))
7813 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
7816 /* Only narrow read access allowed for now. */
7817 if (type == BPF_WRITE) {
7818 if (size != size_default)
7821 bpf_ctx_record_field_size(info, size_default);
7822 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
7830 static bool sk_filter_is_valid_access(int off, int size,
7831 enum bpf_access_type type,
7832 const struct bpf_prog *prog,
7833 struct bpf_insn_access_aux *info)
7836 case bpf_ctx_range(struct __sk_buff, tc_classid):
7837 case bpf_ctx_range(struct __sk_buff, data):
7838 case bpf_ctx_range(struct __sk_buff, data_meta):
7839 case bpf_ctx_range(struct __sk_buff, data_end):
7840 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7841 case bpf_ctx_range(struct __sk_buff, tstamp):
7842 case bpf_ctx_range(struct __sk_buff, wire_len):
7846 if (type == BPF_WRITE) {
7848 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7855 return bpf_skb_is_valid_access(off, size, type, prog, info);
7858 static bool cg_skb_is_valid_access(int off, int size,
7859 enum bpf_access_type type,
7860 const struct bpf_prog *prog,
7861 struct bpf_insn_access_aux *info)
7864 case bpf_ctx_range(struct __sk_buff, tc_classid):
7865 case bpf_ctx_range(struct __sk_buff, data_meta):
7866 case bpf_ctx_range(struct __sk_buff, wire_len):
7868 case bpf_ctx_range(struct __sk_buff, data):
7869 case bpf_ctx_range(struct __sk_buff, data_end):
7875 if (type == BPF_WRITE) {
7877 case bpf_ctx_range(struct __sk_buff, mark):
7878 case bpf_ctx_range(struct __sk_buff, priority):
7879 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7881 case bpf_ctx_range(struct __sk_buff, tstamp):
7891 case bpf_ctx_range(struct __sk_buff, data):
7892 info->reg_type = PTR_TO_PACKET;
7894 case bpf_ctx_range(struct __sk_buff, data_end):
7895 info->reg_type = PTR_TO_PACKET_END;
7899 return bpf_skb_is_valid_access(off, size, type, prog, info);
7902 static bool lwt_is_valid_access(int off, int size,
7903 enum bpf_access_type type,
7904 const struct bpf_prog *prog,
7905 struct bpf_insn_access_aux *info)
7908 case bpf_ctx_range(struct __sk_buff, tc_classid):
7909 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7910 case bpf_ctx_range(struct __sk_buff, data_meta):
7911 case bpf_ctx_range(struct __sk_buff, tstamp):
7912 case bpf_ctx_range(struct __sk_buff, wire_len):
7916 if (type == BPF_WRITE) {
7918 case bpf_ctx_range(struct __sk_buff, mark):
7919 case bpf_ctx_range(struct __sk_buff, priority):
7920 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7928 case bpf_ctx_range(struct __sk_buff, data):
7929 info->reg_type = PTR_TO_PACKET;
7931 case bpf_ctx_range(struct __sk_buff, data_end):
7932 info->reg_type = PTR_TO_PACKET_END;
7936 return bpf_skb_is_valid_access(off, size, type, prog, info);
7939 /* Attach type specific accesses */
7940 static bool __sock_filter_check_attach_type(int off,
7941 enum bpf_access_type access_type,
7942 enum bpf_attach_type attach_type)
7945 case offsetof(struct bpf_sock, bound_dev_if):
7946 case offsetof(struct bpf_sock, mark):
7947 case offsetof(struct bpf_sock, priority):
7948 switch (attach_type) {
7949 case BPF_CGROUP_INET_SOCK_CREATE:
7950 case BPF_CGROUP_INET_SOCK_RELEASE:
7955 case bpf_ctx_range(struct bpf_sock, src_ip4):
7956 switch (attach_type) {
7957 case BPF_CGROUP_INET4_POST_BIND:
7962 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7963 switch (attach_type) {
7964 case BPF_CGROUP_INET6_POST_BIND:
7969 case bpf_ctx_range(struct bpf_sock, src_port):
7970 switch (attach_type) {
7971 case BPF_CGROUP_INET4_POST_BIND:
7972 case BPF_CGROUP_INET6_POST_BIND:
7979 return access_type == BPF_READ;
7984 bool bpf_sock_common_is_valid_access(int off, int size,
7985 enum bpf_access_type type,
7986 struct bpf_insn_access_aux *info)
7989 case bpf_ctx_range_till(struct bpf_sock, type, priority):
7992 return bpf_sock_is_valid_access(off, size, type, info);
7996 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7997 struct bpf_insn_access_aux *info)
7999 const int size_default = sizeof(__u32);
8002 if (off < 0 || off >= sizeof(struct bpf_sock))
8004 if (off % size != 0)
8008 case offsetof(struct bpf_sock, state):
8009 case offsetof(struct bpf_sock, family):
8010 case offsetof(struct bpf_sock, type):
8011 case offsetof(struct bpf_sock, protocol):
8012 case offsetof(struct bpf_sock, src_port):
8013 case offsetof(struct bpf_sock, rx_queue_mapping):
8014 case bpf_ctx_range(struct bpf_sock, src_ip4):
8015 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8016 case bpf_ctx_range(struct bpf_sock, dst_ip4):
8017 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8018 bpf_ctx_record_field_size(info, size_default);
8019 return bpf_ctx_narrow_access_ok(off, size, size_default);
8020 case bpf_ctx_range(struct bpf_sock, dst_port):
8021 field_size = size == size_default ?
8022 size_default : sizeof_field(struct bpf_sock, dst_port);
8023 bpf_ctx_record_field_size(info, field_size);
8024 return bpf_ctx_narrow_access_ok(off, size, field_size);
8025 case offsetofend(struct bpf_sock, dst_port) ...
8026 offsetof(struct bpf_sock, dst_ip4) - 1:
8030 return size == size_default;
8033 static bool sock_filter_is_valid_access(int off, int size,
8034 enum bpf_access_type type,
8035 const struct bpf_prog *prog,
8036 struct bpf_insn_access_aux *info)
8038 if (!bpf_sock_is_valid_access(off, size, type, info))
8040 return __sock_filter_check_attach_type(off, type,
8041 prog->expected_attach_type);
8044 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
8045 const struct bpf_prog *prog)
8047 /* Neither direct read nor direct write requires any preliminary
8053 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
8054 const struct bpf_prog *prog, int drop_verdict)
8056 struct bpf_insn *insn = insn_buf;
8061 /* if (!skb->cloned)
8064 * (Fast-path, otherwise approximation that we might be
8065 * a clone, do the rest in helper.)
8067 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
8068 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
8069 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
8071 /* ret = bpf_skb_pull_data(skb, 0); */
8072 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
8073 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
8074 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
8075 BPF_FUNC_skb_pull_data);
8078 * return TC_ACT_SHOT;
8080 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
8081 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
8082 *insn++ = BPF_EXIT_INSN();
8085 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
8087 *insn++ = prog->insnsi[0];
8089 return insn - insn_buf;
8092 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
8093 struct bpf_insn *insn_buf)
8095 bool indirect = BPF_MODE(orig->code) == BPF_IND;
8096 struct bpf_insn *insn = insn_buf;
8099 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
8101 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
8103 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
8105 /* We're guaranteed here that CTX is in R6. */
8106 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
8108 switch (BPF_SIZE(orig->code)) {
8110 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
8113 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
8116 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
8120 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
8121 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
8122 *insn++ = BPF_EXIT_INSN();
8124 return insn - insn_buf;
8127 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8128 const struct bpf_prog *prog)
8130 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8133 static bool tc_cls_act_is_valid_access(int off, int size,
8134 enum bpf_access_type type,
8135 const struct bpf_prog *prog,
8136 struct bpf_insn_access_aux *info)
8138 if (type == BPF_WRITE) {
8140 case bpf_ctx_range(struct __sk_buff, mark):
8141 case bpf_ctx_range(struct __sk_buff, tc_index):
8142 case bpf_ctx_range(struct __sk_buff, priority):
8143 case bpf_ctx_range(struct __sk_buff, tc_classid):
8144 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8145 case bpf_ctx_range(struct __sk_buff, tstamp):
8146 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8154 case bpf_ctx_range(struct __sk_buff, data):
8155 info->reg_type = PTR_TO_PACKET;
8157 case bpf_ctx_range(struct __sk_buff, data_meta):
8158 info->reg_type = PTR_TO_PACKET_META;
8160 case bpf_ctx_range(struct __sk_buff, data_end):
8161 info->reg_type = PTR_TO_PACKET_END;
8163 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8167 return bpf_skb_is_valid_access(off, size, type, prog, info);
8170 static bool __is_valid_xdp_access(int off, int size)
8172 if (off < 0 || off >= sizeof(struct xdp_md))
8174 if (off % size != 0)
8176 if (size != sizeof(__u32))
8182 static bool xdp_is_valid_access(int off, int size,
8183 enum bpf_access_type type,
8184 const struct bpf_prog *prog,
8185 struct bpf_insn_access_aux *info)
8187 if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8189 case offsetof(struct xdp_md, egress_ifindex):
8194 if (type == BPF_WRITE) {
8195 if (bpf_prog_is_dev_bound(prog->aux)) {
8197 case offsetof(struct xdp_md, rx_queue_index):
8198 return __is_valid_xdp_access(off, size);
8205 case offsetof(struct xdp_md, data):
8206 info->reg_type = PTR_TO_PACKET;
8208 case offsetof(struct xdp_md, data_meta):
8209 info->reg_type = PTR_TO_PACKET_META;
8211 case offsetof(struct xdp_md, data_end):
8212 info->reg_type = PTR_TO_PACKET_END;
8216 return __is_valid_xdp_access(off, size);
8219 void bpf_warn_invalid_xdp_action(u32 act)
8221 const u32 act_max = XDP_REDIRECT;
8223 pr_warn_once("%s XDP return value %u, expect packet loss!\n",
8224 act > act_max ? "Illegal" : "Driver unsupported",
8227 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
8229 static bool sock_addr_is_valid_access(int off, int size,
8230 enum bpf_access_type type,
8231 const struct bpf_prog *prog,
8232 struct bpf_insn_access_aux *info)
8234 const int size_default = sizeof(__u32);
8236 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
8238 if (off % size != 0)
8241 /* Disallow access to IPv6 fields from IPv4 contex and vise
8245 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8246 switch (prog->expected_attach_type) {
8247 case BPF_CGROUP_INET4_BIND:
8248 case BPF_CGROUP_INET4_CONNECT:
8249 case BPF_CGROUP_INET4_GETPEERNAME:
8250 case BPF_CGROUP_INET4_GETSOCKNAME:
8251 case BPF_CGROUP_UDP4_SENDMSG:
8252 case BPF_CGROUP_UDP4_RECVMSG:
8258 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8259 switch (prog->expected_attach_type) {
8260 case BPF_CGROUP_INET6_BIND:
8261 case BPF_CGROUP_INET6_CONNECT:
8262 case BPF_CGROUP_INET6_GETPEERNAME:
8263 case BPF_CGROUP_INET6_GETSOCKNAME:
8264 case BPF_CGROUP_UDP6_SENDMSG:
8265 case BPF_CGROUP_UDP6_RECVMSG:
8271 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8272 switch (prog->expected_attach_type) {
8273 case BPF_CGROUP_UDP4_SENDMSG:
8279 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8281 switch (prog->expected_attach_type) {
8282 case BPF_CGROUP_UDP6_SENDMSG:
8291 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8292 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8293 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8294 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8296 case bpf_ctx_range(struct bpf_sock_addr, user_port):
8297 if (type == BPF_READ) {
8298 bpf_ctx_record_field_size(info, size_default);
8300 if (bpf_ctx_wide_access_ok(off, size,
8301 struct bpf_sock_addr,
8305 if (bpf_ctx_wide_access_ok(off, size,
8306 struct bpf_sock_addr,
8310 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8313 if (bpf_ctx_wide_access_ok(off, size,
8314 struct bpf_sock_addr,
8318 if (bpf_ctx_wide_access_ok(off, size,
8319 struct bpf_sock_addr,
8323 if (size != size_default)
8327 case offsetof(struct bpf_sock_addr, sk):
8328 if (type != BPF_READ)
8330 if (size != sizeof(__u64))
8332 info->reg_type = PTR_TO_SOCKET;
8335 if (type == BPF_READ) {
8336 if (size != size_default)
8346 static bool sock_ops_is_valid_access(int off, int size,
8347 enum bpf_access_type type,
8348 const struct bpf_prog *prog,
8349 struct bpf_insn_access_aux *info)
8351 const int size_default = sizeof(__u32);
8353 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
8356 /* The verifier guarantees that size > 0. */
8357 if (off % size != 0)
8360 if (type == BPF_WRITE) {
8362 case offsetof(struct bpf_sock_ops, reply):
8363 case offsetof(struct bpf_sock_ops, sk_txhash):
8364 if (size != size_default)
8372 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
8374 if (size != sizeof(__u64))
8377 case offsetof(struct bpf_sock_ops, sk):
8378 if (size != sizeof(__u64))
8380 info->reg_type = PTR_TO_SOCKET_OR_NULL;
8382 case offsetof(struct bpf_sock_ops, skb_data):
8383 if (size != sizeof(__u64))
8385 info->reg_type = PTR_TO_PACKET;
8387 case offsetof(struct bpf_sock_ops, skb_data_end):
8388 if (size != sizeof(__u64))
8390 info->reg_type = PTR_TO_PACKET_END;
8392 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
8393 bpf_ctx_record_field_size(info, size_default);
8394 return bpf_ctx_narrow_access_ok(off, size,
8397 if (size != size_default)
8406 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
8407 const struct bpf_prog *prog)
8409 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
8412 static bool sk_skb_is_valid_access(int off, int size,
8413 enum bpf_access_type type,
8414 const struct bpf_prog *prog,
8415 struct bpf_insn_access_aux *info)
8418 case bpf_ctx_range(struct __sk_buff, tc_classid):
8419 case bpf_ctx_range(struct __sk_buff, data_meta):
8420 case bpf_ctx_range(struct __sk_buff, tstamp):
8421 case bpf_ctx_range(struct __sk_buff, wire_len):
8425 if (type == BPF_WRITE) {
8427 case bpf_ctx_range(struct __sk_buff, tc_index):
8428 case bpf_ctx_range(struct __sk_buff, priority):
8436 case bpf_ctx_range(struct __sk_buff, mark):
8438 case bpf_ctx_range(struct __sk_buff, data):
8439 info->reg_type = PTR_TO_PACKET;
8441 case bpf_ctx_range(struct __sk_buff, data_end):
8442 info->reg_type = PTR_TO_PACKET_END;
8446 return bpf_skb_is_valid_access(off, size, type, prog, info);
8449 static bool sk_msg_is_valid_access(int off, int size,
8450 enum bpf_access_type type,
8451 const struct bpf_prog *prog,
8452 struct bpf_insn_access_aux *info)
8454 if (type == BPF_WRITE)
8457 if (off % size != 0)
8461 case offsetof(struct sk_msg_md, data):
8462 info->reg_type = PTR_TO_PACKET;
8463 if (size != sizeof(__u64))
8466 case offsetof(struct sk_msg_md, data_end):
8467 info->reg_type = PTR_TO_PACKET_END;
8468 if (size != sizeof(__u64))
8471 case offsetof(struct sk_msg_md, sk):
8472 if (size != sizeof(__u64))
8474 info->reg_type = PTR_TO_SOCKET;
8476 case bpf_ctx_range(struct sk_msg_md, family):
8477 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
8478 case bpf_ctx_range(struct sk_msg_md, local_ip4):
8479 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
8480 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
8481 case bpf_ctx_range(struct sk_msg_md, remote_port):
8482 case bpf_ctx_range(struct sk_msg_md, local_port):
8483 case bpf_ctx_range(struct sk_msg_md, size):
8484 if (size != sizeof(__u32))
8493 static bool flow_dissector_is_valid_access(int off, int size,
8494 enum bpf_access_type type,
8495 const struct bpf_prog *prog,
8496 struct bpf_insn_access_aux *info)
8498 const int size_default = sizeof(__u32);
8500 if (off < 0 || off >= sizeof(struct __sk_buff))
8503 if (type == BPF_WRITE)
8507 case bpf_ctx_range(struct __sk_buff, data):
8508 if (size != size_default)
8510 info->reg_type = PTR_TO_PACKET;
8512 case bpf_ctx_range(struct __sk_buff, data_end):
8513 if (size != size_default)
8515 info->reg_type = PTR_TO_PACKET_END;
8517 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8518 if (size != sizeof(__u64))
8520 info->reg_type = PTR_TO_FLOW_KEYS;
8527 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
8528 const struct bpf_insn *si,
8529 struct bpf_insn *insn_buf,
8530 struct bpf_prog *prog,
8534 struct bpf_insn *insn = insn_buf;
8537 case offsetof(struct __sk_buff, data):
8538 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
8539 si->dst_reg, si->src_reg,
8540 offsetof(struct bpf_flow_dissector, data));
8543 case offsetof(struct __sk_buff, data_end):
8544 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
8545 si->dst_reg, si->src_reg,
8546 offsetof(struct bpf_flow_dissector, data_end));
8549 case offsetof(struct __sk_buff, flow_keys):
8550 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
8551 si->dst_reg, si->src_reg,
8552 offsetof(struct bpf_flow_dissector, flow_keys));
8556 return insn - insn_buf;
8559 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si,
8560 struct bpf_insn *insn)
8562 /* si->dst_reg = skb_shinfo(SKB); */
8563 #ifdef NET_SKBUFF_DATA_USES_OFFSET
8564 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8565 BPF_REG_AX, si->src_reg,
8566 offsetof(struct sk_buff, end));
8567 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
8568 si->dst_reg, si->src_reg,
8569 offsetof(struct sk_buff, head));
8570 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
8572 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8573 si->dst_reg, si->src_reg,
8574 offsetof(struct sk_buff, end));
8580 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
8581 const struct bpf_insn *si,
8582 struct bpf_insn *insn_buf,
8583 struct bpf_prog *prog, u32 *target_size)
8585 struct bpf_insn *insn = insn_buf;
8589 case offsetof(struct __sk_buff, len):
8590 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8591 bpf_target_off(struct sk_buff, len, 4,
8595 case offsetof(struct __sk_buff, protocol):
8596 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8597 bpf_target_off(struct sk_buff, protocol, 2,
8601 case offsetof(struct __sk_buff, vlan_proto):
8602 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8603 bpf_target_off(struct sk_buff, vlan_proto, 2,
8607 case offsetof(struct __sk_buff, priority):
8608 if (type == BPF_WRITE)
8609 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8610 bpf_target_off(struct sk_buff, priority, 4,
8613 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8614 bpf_target_off(struct sk_buff, priority, 4,
8618 case offsetof(struct __sk_buff, ingress_ifindex):
8619 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8620 bpf_target_off(struct sk_buff, skb_iif, 4,
8624 case offsetof(struct __sk_buff, ifindex):
8625 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8626 si->dst_reg, si->src_reg,
8627 offsetof(struct sk_buff, dev));
8628 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
8629 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8630 bpf_target_off(struct net_device, ifindex, 4,
8634 case offsetof(struct __sk_buff, hash):
8635 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8636 bpf_target_off(struct sk_buff, hash, 4,
8640 case offsetof(struct __sk_buff, mark):
8641 if (type == BPF_WRITE)
8642 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8643 bpf_target_off(struct sk_buff, mark, 4,
8646 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8647 bpf_target_off(struct sk_buff, mark, 4,
8651 case offsetof(struct __sk_buff, pkt_type):
8653 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8655 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
8656 #ifdef __BIG_ENDIAN_BITFIELD
8657 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
8661 case offsetof(struct __sk_buff, queue_mapping):
8662 if (type == BPF_WRITE) {
8663 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
8664 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8665 bpf_target_off(struct sk_buff,
8669 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8670 bpf_target_off(struct sk_buff,
8676 case offsetof(struct __sk_buff, vlan_present):
8678 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8679 PKT_VLAN_PRESENT_OFFSET());
8680 if (PKT_VLAN_PRESENT_BIT)
8681 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
8682 if (PKT_VLAN_PRESENT_BIT < 7)
8683 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
8686 case offsetof(struct __sk_buff, vlan_tci):
8687 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8688 bpf_target_off(struct sk_buff, vlan_tci, 2,
8692 case offsetof(struct __sk_buff, cb[0]) ...
8693 offsetofend(struct __sk_buff, cb[4]) - 1:
8694 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
8695 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
8696 offsetof(struct qdisc_skb_cb, data)) %
8699 prog->cb_access = 1;
8701 off -= offsetof(struct __sk_buff, cb[0]);
8702 off += offsetof(struct sk_buff, cb);
8703 off += offsetof(struct qdisc_skb_cb, data);
8704 if (type == BPF_WRITE)
8705 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
8708 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
8712 case offsetof(struct __sk_buff, tc_classid):
8713 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
8716 off -= offsetof(struct __sk_buff, tc_classid);
8717 off += offsetof(struct sk_buff, cb);
8718 off += offsetof(struct qdisc_skb_cb, tc_classid);
8720 if (type == BPF_WRITE)
8721 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
8724 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
8728 case offsetof(struct __sk_buff, data):
8729 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
8730 si->dst_reg, si->src_reg,
8731 offsetof(struct sk_buff, data));
8734 case offsetof(struct __sk_buff, data_meta):
8736 off -= offsetof(struct __sk_buff, data_meta);
8737 off += offsetof(struct sk_buff, cb);
8738 off += offsetof(struct bpf_skb_data_end, data_meta);
8739 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8743 case offsetof(struct __sk_buff, data_end):
8745 off -= offsetof(struct __sk_buff, data_end);
8746 off += offsetof(struct sk_buff, cb);
8747 off += offsetof(struct bpf_skb_data_end, data_end);
8748 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8752 case offsetof(struct __sk_buff, tc_index):
8753 #ifdef CONFIG_NET_SCHED
8754 if (type == BPF_WRITE)
8755 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8756 bpf_target_off(struct sk_buff, tc_index, 2,
8759 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8760 bpf_target_off(struct sk_buff, tc_index, 2,
8764 if (type == BPF_WRITE)
8765 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
8767 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8771 case offsetof(struct __sk_buff, napi_id):
8772 #if defined(CONFIG_NET_RX_BUSY_POLL)
8773 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8774 bpf_target_off(struct sk_buff, napi_id, 4,
8776 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
8777 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8780 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8783 case offsetof(struct __sk_buff, family):
8784 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
8786 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8787 si->dst_reg, si->src_reg,
8788 offsetof(struct sk_buff, sk));
8789 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8790 bpf_target_off(struct sock_common,
8794 case offsetof(struct __sk_buff, remote_ip4):
8795 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
8797 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8798 si->dst_reg, si->src_reg,
8799 offsetof(struct sk_buff, sk));
8800 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8801 bpf_target_off(struct sock_common,
8805 case offsetof(struct __sk_buff, local_ip4):
8806 BUILD_BUG_ON(sizeof_field(struct sock_common,
8807 skc_rcv_saddr) != 4);
8809 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8810 si->dst_reg, si->src_reg,
8811 offsetof(struct sk_buff, sk));
8812 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8813 bpf_target_off(struct sock_common,
8817 case offsetof(struct __sk_buff, remote_ip6[0]) ...
8818 offsetof(struct __sk_buff, remote_ip6[3]):
8819 #if IS_ENABLED(CONFIG_IPV6)
8820 BUILD_BUG_ON(sizeof_field(struct sock_common,
8821 skc_v6_daddr.s6_addr32[0]) != 4);
8824 off -= offsetof(struct __sk_buff, remote_ip6[0]);
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_W, si->dst_reg, si->dst_reg,
8830 offsetof(struct sock_common,
8831 skc_v6_daddr.s6_addr32[0]) +
8834 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8837 case offsetof(struct __sk_buff, local_ip6[0]) ...
8838 offsetof(struct __sk_buff, local_ip6[3]):
8839 #if IS_ENABLED(CONFIG_IPV6)
8840 BUILD_BUG_ON(sizeof_field(struct sock_common,
8841 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8844 off -= offsetof(struct __sk_buff, local_ip6[0]);
8846 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8847 si->dst_reg, si->src_reg,
8848 offsetof(struct sk_buff, sk));
8849 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8850 offsetof(struct sock_common,
8851 skc_v6_rcv_saddr.s6_addr32[0]) +
8854 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8858 case offsetof(struct __sk_buff, remote_port):
8859 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
8861 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8862 si->dst_reg, si->src_reg,
8863 offsetof(struct sk_buff, sk));
8864 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8865 bpf_target_off(struct sock_common,
8868 #ifndef __BIG_ENDIAN_BITFIELD
8869 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8873 case offsetof(struct __sk_buff, local_port):
8874 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
8876 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8877 si->dst_reg, si->src_reg,
8878 offsetof(struct sk_buff, sk));
8879 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8880 bpf_target_off(struct sock_common,
8881 skc_num, 2, target_size));
8884 case offsetof(struct __sk_buff, tstamp):
8885 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
8887 if (type == BPF_WRITE)
8888 *insn++ = BPF_STX_MEM(BPF_DW,
8889 si->dst_reg, si->src_reg,
8890 bpf_target_off(struct sk_buff,
8894 *insn++ = BPF_LDX_MEM(BPF_DW,
8895 si->dst_reg, si->src_reg,
8896 bpf_target_off(struct sk_buff,
8901 case offsetof(struct __sk_buff, gso_segs):
8902 insn = bpf_convert_shinfo_access(si, insn);
8903 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
8904 si->dst_reg, si->dst_reg,
8905 bpf_target_off(struct skb_shared_info,
8909 case offsetof(struct __sk_buff, gso_size):
8910 insn = bpf_convert_shinfo_access(si, insn);
8911 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
8912 si->dst_reg, si->dst_reg,
8913 bpf_target_off(struct skb_shared_info,
8917 case offsetof(struct __sk_buff, wire_len):
8918 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
8921 off -= offsetof(struct __sk_buff, wire_len);
8922 off += offsetof(struct sk_buff, cb);
8923 off += offsetof(struct qdisc_skb_cb, pkt_len);
8925 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
8928 case offsetof(struct __sk_buff, sk):
8929 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8930 si->dst_reg, si->src_reg,
8931 offsetof(struct sk_buff, sk));
8935 return insn - insn_buf;
8938 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
8939 const struct bpf_insn *si,
8940 struct bpf_insn *insn_buf,
8941 struct bpf_prog *prog, u32 *target_size)
8943 struct bpf_insn *insn = insn_buf;
8947 case offsetof(struct bpf_sock, bound_dev_if):
8948 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
8950 if (type == BPF_WRITE)
8951 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8952 offsetof(struct sock, sk_bound_dev_if));
8954 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8955 offsetof(struct sock, sk_bound_dev_if));
8958 case offsetof(struct bpf_sock, mark):
8959 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
8961 if (type == BPF_WRITE)
8962 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8963 offsetof(struct sock, sk_mark));
8965 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8966 offsetof(struct sock, sk_mark));
8969 case offsetof(struct bpf_sock, priority):
8970 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
8972 if (type == BPF_WRITE)
8973 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8974 offsetof(struct sock, sk_priority));
8976 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8977 offsetof(struct sock, sk_priority));
8980 case offsetof(struct bpf_sock, family):
8981 *insn++ = BPF_LDX_MEM(
8982 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
8983 si->dst_reg, si->src_reg,
8984 bpf_target_off(struct sock_common,
8986 sizeof_field(struct sock_common,
8991 case offsetof(struct bpf_sock, type):
8992 *insn++ = BPF_LDX_MEM(
8993 BPF_FIELD_SIZEOF(struct sock, sk_type),
8994 si->dst_reg, si->src_reg,
8995 bpf_target_off(struct sock, sk_type,
8996 sizeof_field(struct sock, sk_type),
9000 case offsetof(struct bpf_sock, protocol):
9001 *insn++ = BPF_LDX_MEM(
9002 BPF_FIELD_SIZEOF(struct sock, sk_protocol),
9003 si->dst_reg, si->src_reg,
9004 bpf_target_off(struct sock, sk_protocol,
9005 sizeof_field(struct sock, sk_protocol),
9009 case offsetof(struct bpf_sock, src_ip4):
9010 *insn++ = BPF_LDX_MEM(
9011 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9012 bpf_target_off(struct sock_common, skc_rcv_saddr,
9013 sizeof_field(struct sock_common,
9018 case offsetof(struct bpf_sock, dst_ip4):
9019 *insn++ = BPF_LDX_MEM(
9020 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9021 bpf_target_off(struct sock_common, skc_daddr,
9022 sizeof_field(struct sock_common,
9027 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
9028 #if IS_ENABLED(CONFIG_IPV6)
9030 off -= offsetof(struct bpf_sock, src_ip6[0]);
9031 *insn++ = BPF_LDX_MEM(
9032 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9035 skc_v6_rcv_saddr.s6_addr32[0],
9036 sizeof_field(struct sock_common,
9037 skc_v6_rcv_saddr.s6_addr32[0]),
9038 target_size) + off);
9041 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9045 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
9046 #if IS_ENABLED(CONFIG_IPV6)
9048 off -= offsetof(struct bpf_sock, dst_ip6[0]);
9049 *insn++ = BPF_LDX_MEM(
9050 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9051 bpf_target_off(struct sock_common,
9052 skc_v6_daddr.s6_addr32[0],
9053 sizeof_field(struct sock_common,
9054 skc_v6_daddr.s6_addr32[0]),
9055 target_size) + off);
9057 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9062 case offsetof(struct bpf_sock, src_port):
9063 *insn++ = BPF_LDX_MEM(
9064 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
9065 si->dst_reg, si->src_reg,
9066 bpf_target_off(struct sock_common, skc_num,
9067 sizeof_field(struct sock_common,
9072 case offsetof(struct bpf_sock, dst_port):
9073 *insn++ = BPF_LDX_MEM(
9074 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
9075 si->dst_reg, si->src_reg,
9076 bpf_target_off(struct sock_common, skc_dport,
9077 sizeof_field(struct sock_common,
9082 case offsetof(struct bpf_sock, state):
9083 *insn++ = BPF_LDX_MEM(
9084 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
9085 si->dst_reg, si->src_reg,
9086 bpf_target_off(struct sock_common, skc_state,
9087 sizeof_field(struct sock_common,
9091 case offsetof(struct bpf_sock, rx_queue_mapping):
9092 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
9093 *insn++ = BPF_LDX_MEM(
9094 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
9095 si->dst_reg, si->src_reg,
9096 bpf_target_off(struct sock, sk_rx_queue_mapping,
9097 sizeof_field(struct sock,
9098 sk_rx_queue_mapping),
9100 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
9102 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9104 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9110 return insn - insn_buf;
9113 static u32 tc_cls_act_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 __sk_buff, ifindex):
9122 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9123 si->dst_reg, si->src_reg,
9124 offsetof(struct sk_buff, dev));
9125 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9126 bpf_target_off(struct net_device, ifindex, 4,
9130 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9134 return insn - insn_buf;
9137 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
9138 const struct bpf_insn *si,
9139 struct bpf_insn *insn_buf,
9140 struct bpf_prog *prog, u32 *target_size)
9142 struct bpf_insn *insn = insn_buf;
9145 case offsetof(struct xdp_md, data):
9146 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
9147 si->dst_reg, si->src_reg,
9148 offsetof(struct xdp_buff, data));
9150 case offsetof(struct xdp_md, data_meta):
9151 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
9152 si->dst_reg, si->src_reg,
9153 offsetof(struct xdp_buff, data_meta));
9155 case offsetof(struct xdp_md, data_end):
9156 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
9157 si->dst_reg, si->src_reg,
9158 offsetof(struct xdp_buff, data_end));
9160 case offsetof(struct xdp_md, ingress_ifindex):
9161 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9162 si->dst_reg, si->src_reg,
9163 offsetof(struct xdp_buff, rxq));
9164 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
9165 si->dst_reg, si->dst_reg,
9166 offsetof(struct xdp_rxq_info, dev));
9167 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9168 offsetof(struct net_device, ifindex));
9170 case offsetof(struct xdp_md, rx_queue_index):
9171 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9172 si->dst_reg, si->src_reg,
9173 offsetof(struct xdp_buff, rxq));
9174 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9175 offsetof(struct xdp_rxq_info,
9178 case offsetof(struct xdp_md, egress_ifindex):
9179 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
9180 si->dst_reg, si->src_reg,
9181 offsetof(struct xdp_buff, txq));
9182 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
9183 si->dst_reg, si->dst_reg,
9184 offsetof(struct xdp_txq_info, dev));
9185 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9186 offsetof(struct net_device, ifindex));
9190 return insn - insn_buf;
9193 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
9194 * context Structure, F is Field in context structure that contains a pointer
9195 * to Nested Structure of type NS that has the field NF.
9197 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
9198 * sure that SIZE is not greater than actual size of S.F.NF.
9200 * If offset OFF is provided, the load happens from that offset relative to
9203 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
9205 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
9206 si->src_reg, offsetof(S, F)); \
9207 *insn++ = BPF_LDX_MEM( \
9208 SIZE, si->dst_reg, si->dst_reg, \
9209 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
9214 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
9215 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
9216 BPF_FIELD_SIZEOF(NS, NF), 0)
9218 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
9219 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
9221 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
9222 * "register" since two registers available in convert_ctx_access are not
9223 * enough: we can't override neither SRC, since it contains value to store, nor
9224 * DST since it contains pointer to context that may be used by later
9225 * instructions. But we need a temporary place to save pointer to nested
9226 * structure whose field we want to store to.
9228 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
9230 int tmp_reg = BPF_REG_9; \
9231 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
9233 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
9235 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
9237 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
9238 si->dst_reg, offsetof(S, F)); \
9239 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \
9240 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
9243 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
9247 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
9250 if (type == BPF_WRITE) { \
9251 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
9254 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
9255 S, NS, F, NF, SIZE, OFF); \
9259 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
9260 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
9261 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
9263 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
9264 const struct bpf_insn *si,
9265 struct bpf_insn *insn_buf,
9266 struct bpf_prog *prog, u32 *target_size)
9268 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
9269 struct bpf_insn *insn = insn_buf;
9272 case offsetof(struct bpf_sock_addr, user_family):
9273 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9274 struct sockaddr, uaddr, sa_family);
9277 case offsetof(struct bpf_sock_addr, user_ip4):
9278 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9279 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
9280 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
9283 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9285 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
9286 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9287 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9288 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
9292 case offsetof(struct bpf_sock_addr, user_port):
9293 /* To get port we need to know sa_family first and then treat
9294 * sockaddr as either sockaddr_in or sockaddr_in6.
9295 * Though we can simplify since port field has same offset and
9296 * size in both structures.
9297 * Here we check this invariant and use just one of the
9298 * structures if it's true.
9300 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
9301 offsetof(struct sockaddr_in6, sin6_port));
9302 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
9303 sizeof_field(struct sockaddr_in6, sin6_port));
9304 /* Account for sin6_port being smaller than user_port. */
9305 port_size = min(port_size, BPF_LDST_BYTES(si));
9306 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9307 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9308 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
9311 case offsetof(struct bpf_sock_addr, family):
9312 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9313 struct sock, sk, sk_family);
9316 case offsetof(struct bpf_sock_addr, type):
9317 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9318 struct sock, sk, sk_type);
9321 case offsetof(struct bpf_sock_addr, protocol):
9322 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9323 struct sock, sk, sk_protocol);
9326 case offsetof(struct bpf_sock_addr, msg_src_ip4):
9327 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
9328 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9329 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
9330 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
9333 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9336 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
9337 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
9338 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9339 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
9340 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
9342 case offsetof(struct bpf_sock_addr, sk):
9343 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
9344 si->dst_reg, si->src_reg,
9345 offsetof(struct bpf_sock_addr_kern, sk));
9349 return insn - insn_buf;
9352 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
9353 const struct bpf_insn *si,
9354 struct bpf_insn *insn_buf,
9355 struct bpf_prog *prog,
9358 struct bpf_insn *insn = insn_buf;
9361 /* Helper macro for adding read access to tcp_sock or sock fields. */
9362 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
9364 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \
9365 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
9366 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
9367 if (si->dst_reg == reg || si->src_reg == reg) \
9369 if (si->dst_reg == reg || si->src_reg == reg) \
9371 if (si->dst_reg == si->src_reg) { \
9372 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
9373 offsetof(struct bpf_sock_ops_kern, \
9375 fullsock_reg = reg; \
9378 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9379 struct bpf_sock_ops_kern, \
9381 fullsock_reg, si->src_reg, \
9382 offsetof(struct bpf_sock_ops_kern, \
9384 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
9385 if (si->dst_reg == si->src_reg) \
9386 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9387 offsetof(struct bpf_sock_ops_kern, \
9389 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9390 struct bpf_sock_ops_kern, sk),\
9391 si->dst_reg, si->src_reg, \
9392 offsetof(struct bpf_sock_ops_kern, sk));\
9393 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
9395 si->dst_reg, si->dst_reg, \
9396 offsetof(OBJ, OBJ_FIELD)); \
9397 if (si->dst_reg == si->src_reg) { \
9398 *insn++ = BPF_JMP_A(1); \
9399 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9400 offsetof(struct bpf_sock_ops_kern, \
9405 #define SOCK_OPS_GET_SK() \
9407 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \
9408 if (si->dst_reg == reg || si->src_reg == reg) \
9410 if (si->dst_reg == reg || si->src_reg == reg) \
9412 if (si->dst_reg == si->src_reg) { \
9413 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
9414 offsetof(struct bpf_sock_ops_kern, \
9416 fullsock_reg = reg; \
9419 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9420 struct bpf_sock_ops_kern, \
9422 fullsock_reg, si->src_reg, \
9423 offsetof(struct bpf_sock_ops_kern, \
9425 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
9426 if (si->dst_reg == si->src_reg) \
9427 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9428 offsetof(struct bpf_sock_ops_kern, \
9430 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9431 struct bpf_sock_ops_kern, sk),\
9432 si->dst_reg, si->src_reg, \
9433 offsetof(struct bpf_sock_ops_kern, sk));\
9434 if (si->dst_reg == si->src_reg) { \
9435 *insn++ = BPF_JMP_A(1); \
9436 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9437 offsetof(struct bpf_sock_ops_kern, \
9442 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
9443 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
9445 /* Helper macro for adding write access to tcp_sock or sock fields.
9446 * The macro is called with two registers, dst_reg which contains a pointer
9447 * to ctx (context) and src_reg which contains the value that should be
9448 * stored. However, we need an additional register since we cannot overwrite
9449 * dst_reg because it may be used later in the program.
9450 * Instead we "borrow" one of the other register. We first save its value
9451 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
9452 * it at the end of the macro.
9454 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
9456 int reg = BPF_REG_9; \
9457 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
9458 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
9459 if (si->dst_reg == reg || si->src_reg == reg) \
9461 if (si->dst_reg == reg || si->src_reg == reg) \
9463 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
9464 offsetof(struct bpf_sock_ops_kern, \
9466 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9467 struct bpf_sock_ops_kern, \
9470 offsetof(struct bpf_sock_ops_kern, \
9472 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
9473 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9474 struct bpf_sock_ops_kern, sk),\
9476 offsetof(struct bpf_sock_ops_kern, sk));\
9477 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \
9479 offsetof(OBJ, OBJ_FIELD)); \
9480 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
9481 offsetof(struct bpf_sock_ops_kern, \
9485 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
9487 if (TYPE == BPF_WRITE) \
9488 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
9490 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
9493 if (insn > insn_buf)
9494 return insn - insn_buf;
9497 case offsetof(struct bpf_sock_ops, op):
9498 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9500 si->dst_reg, si->src_reg,
9501 offsetof(struct bpf_sock_ops_kern, op));
9504 case offsetof(struct bpf_sock_ops, replylong[0]) ...
9505 offsetof(struct bpf_sock_ops, replylong[3]):
9506 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
9507 sizeof_field(struct bpf_sock_ops_kern, reply));
9508 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
9509 sizeof_field(struct bpf_sock_ops_kern, replylong));
9511 off -= offsetof(struct bpf_sock_ops, replylong[0]);
9512 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
9513 if (type == BPF_WRITE)
9514 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9517 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9521 case offsetof(struct bpf_sock_ops, family):
9522 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9524 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9525 struct bpf_sock_ops_kern, sk),
9526 si->dst_reg, si->src_reg,
9527 offsetof(struct bpf_sock_ops_kern, sk));
9528 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9529 offsetof(struct sock_common, skc_family));
9532 case offsetof(struct bpf_sock_ops, remote_ip4):
9533 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9535 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9536 struct bpf_sock_ops_kern, sk),
9537 si->dst_reg, si->src_reg,
9538 offsetof(struct bpf_sock_ops_kern, sk));
9539 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9540 offsetof(struct sock_common, skc_daddr));
9543 case offsetof(struct bpf_sock_ops, local_ip4):
9544 BUILD_BUG_ON(sizeof_field(struct sock_common,
9545 skc_rcv_saddr) != 4);
9547 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9548 struct bpf_sock_ops_kern, sk),
9549 si->dst_reg, si->src_reg,
9550 offsetof(struct bpf_sock_ops_kern, sk));
9551 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9552 offsetof(struct sock_common,
9556 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
9557 offsetof(struct bpf_sock_ops, remote_ip6[3]):
9558 #if IS_ENABLED(CONFIG_IPV6)
9559 BUILD_BUG_ON(sizeof_field(struct sock_common,
9560 skc_v6_daddr.s6_addr32[0]) != 4);
9563 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
9564 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9565 struct bpf_sock_ops_kern, sk),
9566 si->dst_reg, si->src_reg,
9567 offsetof(struct bpf_sock_ops_kern, sk));
9568 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9569 offsetof(struct sock_common,
9570 skc_v6_daddr.s6_addr32[0]) +
9573 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9577 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
9578 offsetof(struct bpf_sock_ops, local_ip6[3]):
9579 #if IS_ENABLED(CONFIG_IPV6)
9580 BUILD_BUG_ON(sizeof_field(struct sock_common,
9581 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9584 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
9585 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9586 struct bpf_sock_ops_kern, sk),
9587 si->dst_reg, si->src_reg,
9588 offsetof(struct bpf_sock_ops_kern, sk));
9589 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9590 offsetof(struct sock_common,
9591 skc_v6_rcv_saddr.s6_addr32[0]) +
9594 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9598 case offsetof(struct bpf_sock_ops, remote_port):
9599 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9601 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9602 struct bpf_sock_ops_kern, sk),
9603 si->dst_reg, si->src_reg,
9604 offsetof(struct bpf_sock_ops_kern, sk));
9605 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9606 offsetof(struct sock_common, skc_dport));
9607 #ifndef __BIG_ENDIAN_BITFIELD
9608 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9612 case offsetof(struct bpf_sock_ops, local_port):
9613 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9615 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9616 struct bpf_sock_ops_kern, sk),
9617 si->dst_reg, si->src_reg,
9618 offsetof(struct bpf_sock_ops_kern, sk));
9619 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9620 offsetof(struct sock_common, skc_num));
9623 case offsetof(struct bpf_sock_ops, is_fullsock):
9624 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9625 struct bpf_sock_ops_kern,
9627 si->dst_reg, si->src_reg,
9628 offsetof(struct bpf_sock_ops_kern,
9632 case offsetof(struct bpf_sock_ops, state):
9633 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
9635 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9636 struct bpf_sock_ops_kern, sk),
9637 si->dst_reg, si->src_reg,
9638 offsetof(struct bpf_sock_ops_kern, sk));
9639 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
9640 offsetof(struct sock_common, skc_state));
9643 case offsetof(struct bpf_sock_ops, rtt_min):
9644 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
9645 sizeof(struct minmax));
9646 BUILD_BUG_ON(sizeof(struct minmax) <
9647 sizeof(struct minmax_sample));
9649 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9650 struct bpf_sock_ops_kern, sk),
9651 si->dst_reg, si->src_reg,
9652 offsetof(struct bpf_sock_ops_kern, sk));
9653 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9654 offsetof(struct tcp_sock, rtt_min) +
9655 sizeof_field(struct minmax_sample, t));
9658 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
9659 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
9663 case offsetof(struct bpf_sock_ops, sk_txhash):
9664 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
9667 case offsetof(struct bpf_sock_ops, snd_cwnd):
9668 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
9670 case offsetof(struct bpf_sock_ops, srtt_us):
9671 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
9673 case offsetof(struct bpf_sock_ops, snd_ssthresh):
9674 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
9676 case offsetof(struct bpf_sock_ops, rcv_nxt):
9677 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
9679 case offsetof(struct bpf_sock_ops, snd_nxt):
9680 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
9682 case offsetof(struct bpf_sock_ops, snd_una):
9683 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
9685 case offsetof(struct bpf_sock_ops, mss_cache):
9686 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
9688 case offsetof(struct bpf_sock_ops, ecn_flags):
9689 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
9691 case offsetof(struct bpf_sock_ops, rate_delivered):
9692 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
9694 case offsetof(struct bpf_sock_ops, rate_interval_us):
9695 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
9697 case offsetof(struct bpf_sock_ops, packets_out):
9698 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
9700 case offsetof(struct bpf_sock_ops, retrans_out):
9701 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
9703 case offsetof(struct bpf_sock_ops, total_retrans):
9704 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
9706 case offsetof(struct bpf_sock_ops, segs_in):
9707 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
9709 case offsetof(struct bpf_sock_ops, data_segs_in):
9710 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
9712 case offsetof(struct bpf_sock_ops, segs_out):
9713 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
9715 case offsetof(struct bpf_sock_ops, data_segs_out):
9716 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
9718 case offsetof(struct bpf_sock_ops, lost_out):
9719 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
9721 case offsetof(struct bpf_sock_ops, sacked_out):
9722 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
9724 case offsetof(struct bpf_sock_ops, bytes_received):
9725 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
9727 case offsetof(struct bpf_sock_ops, bytes_acked):
9728 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
9730 case offsetof(struct bpf_sock_ops, sk):
9733 case offsetof(struct bpf_sock_ops, skb_data_end):
9734 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9736 si->dst_reg, si->src_reg,
9737 offsetof(struct bpf_sock_ops_kern,
9740 case offsetof(struct bpf_sock_ops, skb_data):
9741 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9743 si->dst_reg, si->src_reg,
9744 offsetof(struct bpf_sock_ops_kern,
9746 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9747 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9748 si->dst_reg, si->dst_reg,
9749 offsetof(struct sk_buff, data));
9751 case offsetof(struct bpf_sock_ops, skb_len):
9752 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9754 si->dst_reg, si->src_reg,
9755 offsetof(struct bpf_sock_ops_kern,
9757 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9758 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9759 si->dst_reg, si->dst_reg,
9760 offsetof(struct sk_buff, len));
9762 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9763 off = offsetof(struct sk_buff, cb);
9764 off += offsetof(struct tcp_skb_cb, tcp_flags);
9765 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
9766 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9768 si->dst_reg, si->src_reg,
9769 offsetof(struct bpf_sock_ops_kern,
9771 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9772 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
9774 si->dst_reg, si->dst_reg, off);
9777 return insn - insn_buf;
9780 /* data_end = skb->data + skb_headlen() */
9781 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
9782 struct bpf_insn *insn)
9785 int temp_reg_off = offsetof(struct sk_buff, cb) +
9786 offsetof(struct sk_skb_cb, temp_reg);
9788 if (si->src_reg == si->dst_reg) {
9789 /* We need an extra register, choose and save a register. */
9791 if (si->src_reg == reg || si->dst_reg == reg)
9793 if (si->src_reg == reg || si->dst_reg == reg)
9795 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off);
9800 /* reg = skb->data */
9801 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9803 offsetof(struct sk_buff, data));
9805 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9806 BPF_REG_AX, si->src_reg,
9807 offsetof(struct sk_buff, len));
9808 /* reg = skb->data + skb->len */
9809 *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX);
9810 /* AX = skb->data_len */
9811 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
9812 BPF_REG_AX, si->src_reg,
9813 offsetof(struct sk_buff, data_len));
9815 /* reg = skb->data + skb->len - skb->data_len */
9816 *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX);
9818 if (si->src_reg == si->dst_reg) {
9819 /* Restore the saved register */
9820 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg);
9821 *insn++ = BPF_MOV64_REG(si->dst_reg, reg);
9822 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off);
9828 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
9829 const struct bpf_insn *si,
9830 struct bpf_insn *insn_buf,
9831 struct bpf_prog *prog, u32 *target_size)
9833 struct bpf_insn *insn = insn_buf;
9837 case offsetof(struct __sk_buff, data_end):
9838 insn = bpf_convert_data_end_access(si, insn);
9840 case offsetof(struct __sk_buff, cb[0]) ...
9841 offsetofend(struct __sk_buff, cb[4]) - 1:
9842 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20);
9843 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
9844 offsetof(struct sk_skb_cb, data)) %
9847 prog->cb_access = 1;
9849 off -= offsetof(struct __sk_buff, cb[0]);
9850 off += offsetof(struct sk_buff, cb);
9851 off += offsetof(struct sk_skb_cb, data);
9852 if (type == BPF_WRITE)
9853 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
9856 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
9862 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9866 return insn - insn_buf;
9869 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
9870 const struct bpf_insn *si,
9871 struct bpf_insn *insn_buf,
9872 struct bpf_prog *prog, u32 *target_size)
9874 struct bpf_insn *insn = insn_buf;
9875 #if IS_ENABLED(CONFIG_IPV6)
9879 /* convert ctx uses the fact sg element is first in struct */
9880 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
9883 case offsetof(struct sk_msg_md, data):
9884 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
9885 si->dst_reg, si->src_reg,
9886 offsetof(struct sk_msg, data));
9888 case offsetof(struct sk_msg_md, data_end):
9889 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
9890 si->dst_reg, si->src_reg,
9891 offsetof(struct sk_msg, data_end));
9893 case offsetof(struct sk_msg_md, family):
9894 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9896 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9898 si->dst_reg, si->src_reg,
9899 offsetof(struct sk_msg, sk));
9900 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9901 offsetof(struct sock_common, skc_family));
9904 case offsetof(struct sk_msg_md, remote_ip4):
9905 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9907 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9909 si->dst_reg, si->src_reg,
9910 offsetof(struct sk_msg, sk));
9911 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9912 offsetof(struct sock_common, skc_daddr));
9915 case offsetof(struct sk_msg_md, local_ip4):
9916 BUILD_BUG_ON(sizeof_field(struct sock_common,
9917 skc_rcv_saddr) != 4);
9919 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9921 si->dst_reg, si->src_reg,
9922 offsetof(struct sk_msg, sk));
9923 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9924 offsetof(struct sock_common,
9928 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
9929 offsetof(struct sk_msg_md, remote_ip6[3]):
9930 #if IS_ENABLED(CONFIG_IPV6)
9931 BUILD_BUG_ON(sizeof_field(struct sock_common,
9932 skc_v6_daddr.s6_addr32[0]) != 4);
9935 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
9936 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9938 si->dst_reg, si->src_reg,
9939 offsetof(struct sk_msg, sk));
9940 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9941 offsetof(struct sock_common,
9942 skc_v6_daddr.s6_addr32[0]) +
9945 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9949 case offsetof(struct sk_msg_md, local_ip6[0]) ...
9950 offsetof(struct sk_msg_md, local_ip6[3]):
9951 #if IS_ENABLED(CONFIG_IPV6)
9952 BUILD_BUG_ON(sizeof_field(struct sock_common,
9953 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9956 off -= offsetof(struct sk_msg_md, local_ip6[0]);
9957 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9959 si->dst_reg, si->src_reg,
9960 offsetof(struct sk_msg, sk));
9961 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9962 offsetof(struct sock_common,
9963 skc_v6_rcv_saddr.s6_addr32[0]) +
9966 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9970 case offsetof(struct sk_msg_md, remote_port):
9971 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9973 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9975 si->dst_reg, si->src_reg,
9976 offsetof(struct sk_msg, sk));
9977 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9978 offsetof(struct sock_common, skc_dport));
9979 #ifndef __BIG_ENDIAN_BITFIELD
9980 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9984 case offsetof(struct sk_msg_md, local_port):
9985 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9987 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9989 si->dst_reg, si->src_reg,
9990 offsetof(struct sk_msg, sk));
9991 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9992 offsetof(struct sock_common, skc_num));
9995 case offsetof(struct sk_msg_md, size):
9996 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
9997 si->dst_reg, si->src_reg,
9998 offsetof(struct sk_msg_sg, size));
10001 case offsetof(struct sk_msg_md, sk):
10002 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
10003 si->dst_reg, si->src_reg,
10004 offsetof(struct sk_msg, sk));
10008 return insn - insn_buf;
10011 const struct bpf_verifier_ops sk_filter_verifier_ops = {
10012 .get_func_proto = sk_filter_func_proto,
10013 .is_valid_access = sk_filter_is_valid_access,
10014 .convert_ctx_access = bpf_convert_ctx_access,
10015 .gen_ld_abs = bpf_gen_ld_abs,
10018 const struct bpf_prog_ops sk_filter_prog_ops = {
10019 .test_run = bpf_prog_test_run_skb,
10022 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
10023 .get_func_proto = tc_cls_act_func_proto,
10024 .is_valid_access = tc_cls_act_is_valid_access,
10025 .convert_ctx_access = tc_cls_act_convert_ctx_access,
10026 .gen_prologue = tc_cls_act_prologue,
10027 .gen_ld_abs = bpf_gen_ld_abs,
10028 .check_kfunc_call = bpf_prog_test_check_kfunc_call,
10031 const struct bpf_prog_ops tc_cls_act_prog_ops = {
10032 .test_run = bpf_prog_test_run_skb,
10035 const struct bpf_verifier_ops xdp_verifier_ops = {
10036 .get_func_proto = xdp_func_proto,
10037 .is_valid_access = xdp_is_valid_access,
10038 .convert_ctx_access = xdp_convert_ctx_access,
10039 .gen_prologue = bpf_noop_prologue,
10042 const struct bpf_prog_ops xdp_prog_ops = {
10043 .test_run = bpf_prog_test_run_xdp,
10046 const struct bpf_verifier_ops cg_skb_verifier_ops = {
10047 .get_func_proto = cg_skb_func_proto,
10048 .is_valid_access = cg_skb_is_valid_access,
10049 .convert_ctx_access = bpf_convert_ctx_access,
10052 const struct bpf_prog_ops cg_skb_prog_ops = {
10053 .test_run = bpf_prog_test_run_skb,
10056 const struct bpf_verifier_ops lwt_in_verifier_ops = {
10057 .get_func_proto = lwt_in_func_proto,
10058 .is_valid_access = lwt_is_valid_access,
10059 .convert_ctx_access = bpf_convert_ctx_access,
10062 const struct bpf_prog_ops lwt_in_prog_ops = {
10063 .test_run = bpf_prog_test_run_skb,
10066 const struct bpf_verifier_ops lwt_out_verifier_ops = {
10067 .get_func_proto = lwt_out_func_proto,
10068 .is_valid_access = lwt_is_valid_access,
10069 .convert_ctx_access = bpf_convert_ctx_access,
10072 const struct bpf_prog_ops lwt_out_prog_ops = {
10073 .test_run = bpf_prog_test_run_skb,
10076 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
10077 .get_func_proto = lwt_xmit_func_proto,
10078 .is_valid_access = lwt_is_valid_access,
10079 .convert_ctx_access = bpf_convert_ctx_access,
10080 .gen_prologue = tc_cls_act_prologue,
10083 const struct bpf_prog_ops lwt_xmit_prog_ops = {
10084 .test_run = bpf_prog_test_run_skb,
10087 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
10088 .get_func_proto = lwt_seg6local_func_proto,
10089 .is_valid_access = lwt_is_valid_access,
10090 .convert_ctx_access = bpf_convert_ctx_access,
10093 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
10094 .test_run = bpf_prog_test_run_skb,
10097 const struct bpf_verifier_ops cg_sock_verifier_ops = {
10098 .get_func_proto = sock_filter_func_proto,
10099 .is_valid_access = sock_filter_is_valid_access,
10100 .convert_ctx_access = bpf_sock_convert_ctx_access,
10103 const struct bpf_prog_ops cg_sock_prog_ops = {
10106 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
10107 .get_func_proto = sock_addr_func_proto,
10108 .is_valid_access = sock_addr_is_valid_access,
10109 .convert_ctx_access = sock_addr_convert_ctx_access,
10112 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
10115 const struct bpf_verifier_ops sock_ops_verifier_ops = {
10116 .get_func_proto = sock_ops_func_proto,
10117 .is_valid_access = sock_ops_is_valid_access,
10118 .convert_ctx_access = sock_ops_convert_ctx_access,
10121 const struct bpf_prog_ops sock_ops_prog_ops = {
10124 const struct bpf_verifier_ops sk_skb_verifier_ops = {
10125 .get_func_proto = sk_skb_func_proto,
10126 .is_valid_access = sk_skb_is_valid_access,
10127 .convert_ctx_access = sk_skb_convert_ctx_access,
10128 .gen_prologue = sk_skb_prologue,
10131 const struct bpf_prog_ops sk_skb_prog_ops = {
10134 const struct bpf_verifier_ops sk_msg_verifier_ops = {
10135 .get_func_proto = sk_msg_func_proto,
10136 .is_valid_access = sk_msg_is_valid_access,
10137 .convert_ctx_access = sk_msg_convert_ctx_access,
10138 .gen_prologue = bpf_noop_prologue,
10141 const struct bpf_prog_ops sk_msg_prog_ops = {
10144 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
10145 .get_func_proto = flow_dissector_func_proto,
10146 .is_valid_access = flow_dissector_is_valid_access,
10147 .convert_ctx_access = flow_dissector_convert_ctx_access,
10150 const struct bpf_prog_ops flow_dissector_prog_ops = {
10151 .test_run = bpf_prog_test_run_flow_dissector,
10154 int sk_detach_filter(struct sock *sk)
10157 struct sk_filter *filter;
10159 if (sock_flag(sk, SOCK_FILTER_LOCKED))
10162 filter = rcu_dereference_protected(sk->sk_filter,
10163 lockdep_sock_is_held(sk));
10165 RCU_INIT_POINTER(sk->sk_filter, NULL);
10166 sk_filter_uncharge(sk, filter);
10172 EXPORT_SYMBOL_GPL(sk_detach_filter);
10174 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
10177 struct sock_fprog_kern *fprog;
10178 struct sk_filter *filter;
10182 filter = rcu_dereference_protected(sk->sk_filter,
10183 lockdep_sock_is_held(sk));
10187 /* We're copying the filter that has been originally attached,
10188 * so no conversion/decode needed anymore. eBPF programs that
10189 * have no original program cannot be dumped through this.
10192 fprog = filter->prog->orig_prog;
10198 /* User space only enquires number of filter blocks. */
10202 if (len < fprog->len)
10206 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
10209 /* Instead of bytes, the API requests to return the number
10210 * of filter blocks.
10219 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
10220 struct sock_reuseport *reuse,
10221 struct sock *sk, struct sk_buff *skb,
10222 struct sock *migrating_sk,
10225 reuse_kern->skb = skb;
10226 reuse_kern->sk = sk;
10227 reuse_kern->selected_sk = NULL;
10228 reuse_kern->migrating_sk = migrating_sk;
10229 reuse_kern->data_end = skb->data + skb_headlen(skb);
10230 reuse_kern->hash = hash;
10231 reuse_kern->reuseport_id = reuse->reuseport_id;
10232 reuse_kern->bind_inany = reuse->bind_inany;
10235 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
10236 struct bpf_prog *prog, struct sk_buff *skb,
10237 struct sock *migrating_sk,
10240 struct sk_reuseport_kern reuse_kern;
10241 enum sk_action action;
10243 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
10244 action = bpf_prog_run(prog, &reuse_kern);
10246 if (action == SK_PASS)
10247 return reuse_kern.selected_sk;
10249 return ERR_PTR(-ECONNREFUSED);
10252 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
10253 struct bpf_map *, map, void *, key, u32, flags)
10255 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
10256 struct sock_reuseport *reuse;
10257 struct sock *selected_sk;
10259 selected_sk = map->ops->map_lookup_elem(map, key);
10263 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
10265 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
10266 if (sk_is_refcounted(selected_sk))
10267 sock_put(selected_sk);
10269 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
10270 * The only (!reuse) case here is - the sk has already been
10271 * unhashed (e.g. by close()), so treat it as -ENOENT.
10273 * Other maps (e.g. sock_map) do not provide this guarantee and
10274 * the sk may never be in the reuseport group to begin with.
10276 return is_sockarray ? -ENOENT : -EINVAL;
10279 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
10280 struct sock *sk = reuse_kern->sk;
10282 if (sk->sk_protocol != selected_sk->sk_protocol)
10283 return -EPROTOTYPE;
10284 else if (sk->sk_family != selected_sk->sk_family)
10285 return -EAFNOSUPPORT;
10287 /* Catch all. Likely bound to a different sockaddr. */
10291 reuse_kern->selected_sk = selected_sk;
10296 static const struct bpf_func_proto sk_select_reuseport_proto = {
10297 .func = sk_select_reuseport,
10299 .ret_type = RET_INTEGER,
10300 .arg1_type = ARG_PTR_TO_CTX,
10301 .arg2_type = ARG_CONST_MAP_PTR,
10302 .arg3_type = ARG_PTR_TO_MAP_KEY,
10303 .arg4_type = ARG_ANYTHING,
10306 BPF_CALL_4(sk_reuseport_load_bytes,
10307 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10308 void *, to, u32, len)
10310 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
10313 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
10314 .func = sk_reuseport_load_bytes,
10316 .ret_type = RET_INTEGER,
10317 .arg1_type = ARG_PTR_TO_CTX,
10318 .arg2_type = ARG_ANYTHING,
10319 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
10320 .arg4_type = ARG_CONST_SIZE,
10323 BPF_CALL_5(sk_reuseport_load_bytes_relative,
10324 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10325 void *, to, u32, len, u32, start_header)
10327 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
10328 len, start_header);
10331 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
10332 .func = sk_reuseport_load_bytes_relative,
10334 .ret_type = RET_INTEGER,
10335 .arg1_type = ARG_PTR_TO_CTX,
10336 .arg2_type = ARG_ANYTHING,
10337 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
10338 .arg4_type = ARG_CONST_SIZE,
10339 .arg5_type = ARG_ANYTHING,
10342 static const struct bpf_func_proto *
10343 sk_reuseport_func_proto(enum bpf_func_id func_id,
10344 const struct bpf_prog *prog)
10347 case BPF_FUNC_sk_select_reuseport:
10348 return &sk_select_reuseport_proto;
10349 case BPF_FUNC_skb_load_bytes:
10350 return &sk_reuseport_load_bytes_proto;
10351 case BPF_FUNC_skb_load_bytes_relative:
10352 return &sk_reuseport_load_bytes_relative_proto;
10353 case BPF_FUNC_get_socket_cookie:
10354 return &bpf_get_socket_ptr_cookie_proto;
10355 case BPF_FUNC_ktime_get_coarse_ns:
10356 return &bpf_ktime_get_coarse_ns_proto;
10358 return bpf_base_func_proto(func_id);
10363 sk_reuseport_is_valid_access(int off, int size,
10364 enum bpf_access_type type,
10365 const struct bpf_prog *prog,
10366 struct bpf_insn_access_aux *info)
10368 const u32 size_default = sizeof(__u32);
10370 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
10371 off % size || type != BPF_READ)
10375 case offsetof(struct sk_reuseport_md, data):
10376 info->reg_type = PTR_TO_PACKET;
10377 return size == sizeof(__u64);
10379 case offsetof(struct sk_reuseport_md, data_end):
10380 info->reg_type = PTR_TO_PACKET_END;
10381 return size == sizeof(__u64);
10383 case offsetof(struct sk_reuseport_md, hash):
10384 return size == size_default;
10386 case offsetof(struct sk_reuseport_md, sk):
10387 info->reg_type = PTR_TO_SOCKET;
10388 return size == sizeof(__u64);
10390 case offsetof(struct sk_reuseport_md, migrating_sk):
10391 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
10392 return size == sizeof(__u64);
10394 /* Fields that allow narrowing */
10395 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
10396 if (size < sizeof_field(struct sk_buff, protocol))
10399 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
10400 case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
10401 case bpf_ctx_range(struct sk_reuseport_md, len):
10402 bpf_ctx_record_field_size(info, size_default);
10403 return bpf_ctx_narrow_access_ok(off, size, size_default);
10410 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
10411 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
10412 si->dst_reg, si->src_reg, \
10413 bpf_target_off(struct sk_reuseport_kern, F, \
10414 sizeof_field(struct sk_reuseport_kern, F), \
10418 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
10419 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
10424 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \
10425 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
10430 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
10431 const struct bpf_insn *si,
10432 struct bpf_insn *insn_buf,
10433 struct bpf_prog *prog,
10436 struct bpf_insn *insn = insn_buf;
10439 case offsetof(struct sk_reuseport_md, data):
10440 SK_REUSEPORT_LOAD_SKB_FIELD(data);
10443 case offsetof(struct sk_reuseport_md, len):
10444 SK_REUSEPORT_LOAD_SKB_FIELD(len);
10447 case offsetof(struct sk_reuseport_md, eth_protocol):
10448 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
10451 case offsetof(struct sk_reuseport_md, ip_protocol):
10452 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
10455 case offsetof(struct sk_reuseport_md, data_end):
10456 SK_REUSEPORT_LOAD_FIELD(data_end);
10459 case offsetof(struct sk_reuseport_md, hash):
10460 SK_REUSEPORT_LOAD_FIELD(hash);
10463 case offsetof(struct sk_reuseport_md, bind_inany):
10464 SK_REUSEPORT_LOAD_FIELD(bind_inany);
10467 case offsetof(struct sk_reuseport_md, sk):
10468 SK_REUSEPORT_LOAD_FIELD(sk);
10471 case offsetof(struct sk_reuseport_md, migrating_sk):
10472 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
10476 return insn - insn_buf;
10479 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
10480 .get_func_proto = sk_reuseport_func_proto,
10481 .is_valid_access = sk_reuseport_is_valid_access,
10482 .convert_ctx_access = sk_reuseport_convert_ctx_access,
10485 const struct bpf_prog_ops sk_reuseport_prog_ops = {
10488 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
10489 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
10491 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
10492 struct sock *, sk, u64, flags)
10494 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
10495 BPF_SK_LOOKUP_F_NO_REUSEPORT)))
10497 if (unlikely(sk && sk_is_refcounted(sk)))
10498 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
10499 if (unlikely(sk && sk->sk_state == TCP_ESTABLISHED))
10500 return -ESOCKTNOSUPPORT; /* reject connected sockets */
10502 /* Check if socket is suitable for packet L3/L4 protocol */
10503 if (sk && sk->sk_protocol != ctx->protocol)
10504 return -EPROTOTYPE;
10505 if (sk && sk->sk_family != ctx->family &&
10506 (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
10507 return -EAFNOSUPPORT;
10509 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
10512 /* Select socket as lookup result */
10513 ctx->selected_sk = sk;
10514 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
10518 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
10519 .func = bpf_sk_lookup_assign,
10521 .ret_type = RET_INTEGER,
10522 .arg1_type = ARG_PTR_TO_CTX,
10523 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL,
10524 .arg3_type = ARG_ANYTHING,
10527 static const struct bpf_func_proto *
10528 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
10531 case BPF_FUNC_perf_event_output:
10532 return &bpf_event_output_data_proto;
10533 case BPF_FUNC_sk_assign:
10534 return &bpf_sk_lookup_assign_proto;
10535 case BPF_FUNC_sk_release:
10536 return &bpf_sk_release_proto;
10538 return bpf_sk_base_func_proto(func_id);
10542 static bool sk_lookup_is_valid_access(int off, int size,
10543 enum bpf_access_type type,
10544 const struct bpf_prog *prog,
10545 struct bpf_insn_access_aux *info)
10547 if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
10549 if (off % size != 0)
10551 if (type != BPF_READ)
10555 case offsetof(struct bpf_sk_lookup, sk):
10556 info->reg_type = PTR_TO_SOCKET_OR_NULL;
10557 return size == sizeof(__u64);
10559 case bpf_ctx_range(struct bpf_sk_lookup, family):
10560 case bpf_ctx_range(struct bpf_sk_lookup, protocol):
10561 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
10562 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
10563 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
10564 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
10565 case offsetof(struct bpf_sk_lookup, remote_port) ...
10566 offsetof(struct bpf_sk_lookup, local_ip4) - 1:
10567 case bpf_ctx_range(struct bpf_sk_lookup, local_port):
10568 bpf_ctx_record_field_size(info, sizeof(__u32));
10569 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
10576 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
10577 const struct bpf_insn *si,
10578 struct bpf_insn *insn_buf,
10579 struct bpf_prog *prog,
10582 struct bpf_insn *insn = insn_buf;
10585 case offsetof(struct bpf_sk_lookup, sk):
10586 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10587 offsetof(struct bpf_sk_lookup_kern, selected_sk));
10590 case offsetof(struct bpf_sk_lookup, family):
10591 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10592 bpf_target_off(struct bpf_sk_lookup_kern,
10593 family, 2, target_size));
10596 case offsetof(struct bpf_sk_lookup, protocol):
10597 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10598 bpf_target_off(struct bpf_sk_lookup_kern,
10599 protocol, 2, target_size));
10602 case offsetof(struct bpf_sk_lookup, remote_ip4):
10603 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10604 bpf_target_off(struct bpf_sk_lookup_kern,
10605 v4.saddr, 4, target_size));
10608 case offsetof(struct bpf_sk_lookup, local_ip4):
10609 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10610 bpf_target_off(struct bpf_sk_lookup_kern,
10611 v4.daddr, 4, target_size));
10614 case bpf_ctx_range_till(struct bpf_sk_lookup,
10615 remote_ip6[0], remote_ip6[3]): {
10616 #if IS_ENABLED(CONFIG_IPV6)
10619 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
10620 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10621 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10622 offsetof(struct bpf_sk_lookup_kern, v6.saddr));
10623 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10624 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10626 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10630 case bpf_ctx_range_till(struct bpf_sk_lookup,
10631 local_ip6[0], local_ip6[3]): {
10632 #if IS_ENABLED(CONFIG_IPV6)
10635 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
10636 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10637 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10638 offsetof(struct bpf_sk_lookup_kern, v6.daddr));
10639 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10640 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10642 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10646 case offsetof(struct bpf_sk_lookup, remote_port):
10647 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10648 bpf_target_off(struct bpf_sk_lookup_kern,
10649 sport, 2, target_size));
10652 case offsetof(struct bpf_sk_lookup, local_port):
10653 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10654 bpf_target_off(struct bpf_sk_lookup_kern,
10655 dport, 2, target_size));
10659 return insn - insn_buf;
10662 const struct bpf_prog_ops sk_lookup_prog_ops = {
10663 .test_run = bpf_prog_test_run_sk_lookup,
10666 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
10667 .get_func_proto = sk_lookup_func_proto,
10668 .is_valid_access = sk_lookup_is_valid_access,
10669 .convert_ctx_access = sk_lookup_convert_ctx_access,
10672 #endif /* CONFIG_INET */
10674 DEFINE_BPF_DISPATCHER(xdp)
10676 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
10678 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
10681 #ifdef CONFIG_DEBUG_INFO_BTF
10682 BTF_ID_LIST_GLOBAL(btf_sock_ids)
10683 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
10685 #undef BTF_SOCK_TYPE
10687 u32 btf_sock_ids[MAX_BTF_SOCK_TYPE];
10690 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
10692 /* tcp6_sock type is not generated in dwarf and hence btf,
10693 * trigger an explicit type generation here.
10695 BTF_TYPE_EMIT(struct tcp6_sock);
10696 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
10697 sk->sk_family == AF_INET6)
10698 return (unsigned long)sk;
10700 return (unsigned long)NULL;
10703 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
10704 .func = bpf_skc_to_tcp6_sock,
10706 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10707 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10708 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
10711 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
10713 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
10714 return (unsigned long)sk;
10716 return (unsigned long)NULL;
10719 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
10720 .func = bpf_skc_to_tcp_sock,
10722 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10723 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10724 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
10727 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
10729 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
10730 * generated if CONFIG_INET=n. Trigger an explicit generation here.
10732 BTF_TYPE_EMIT(struct inet_timewait_sock);
10733 BTF_TYPE_EMIT(struct tcp_timewait_sock);
10736 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
10737 return (unsigned long)sk;
10740 #if IS_BUILTIN(CONFIG_IPV6)
10741 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
10742 return (unsigned long)sk;
10745 return (unsigned long)NULL;
10748 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
10749 .func = bpf_skc_to_tcp_timewait_sock,
10751 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10752 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10753 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
10756 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
10759 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10760 return (unsigned long)sk;
10763 #if IS_BUILTIN(CONFIG_IPV6)
10764 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10765 return (unsigned long)sk;
10768 return (unsigned long)NULL;
10771 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
10772 .func = bpf_skc_to_tcp_request_sock,
10774 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10775 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10776 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
10779 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
10781 /* udp6_sock type is not generated in dwarf and hence btf,
10782 * trigger an explicit type generation here.
10784 BTF_TYPE_EMIT(struct udp6_sock);
10785 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
10786 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
10787 return (unsigned long)sk;
10789 return (unsigned long)NULL;
10792 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
10793 .func = bpf_skc_to_udp6_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_UDP6],
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_ktime_get_coarse_ns:
10840 return &bpf_ktime_get_coarse_ns_proto;
10842 return bpf_base_func_proto(func_id);
10845 if (!perfmon_capable())