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/module.h>
21 #include <linux/types.h>
23 #include <linux/fcntl.h>
24 #include <linux/socket.h>
25 #include <linux/sock_diag.h>
27 #include <linux/inet.h>
28 #include <linux/netdevice.h>
29 #include <linux/if_packet.h>
30 #include <linux/if_arp.h>
31 #include <linux/gfp.h>
32 #include <net/inet_common.h>
34 #include <net/protocol.h>
35 #include <net/netlink.h>
36 #include <linux/skbuff.h>
37 #include <linux/skmsg.h>
39 #include <net/flow_dissector.h>
40 #include <linux/errno.h>
41 #include <linux/timer.h>
42 #include <linux/uaccess.h>
43 #include <asm/unaligned.h>
44 #include <asm/cmpxchg.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>
81 static const struct bpf_func_proto *
82 bpf_sk_base_func_proto(enum bpf_func_id func_id);
84 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len)
86 if (in_compat_syscall()) {
87 struct compat_sock_fprog f32;
89 if (len != sizeof(f32))
91 if (copy_from_sockptr(&f32, src, sizeof(f32)))
93 memset(dst, 0, sizeof(*dst));
95 dst->filter = compat_ptr(f32.filter);
97 if (len != sizeof(*dst))
99 if (copy_from_sockptr(dst, src, sizeof(*dst)))
105 EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user);
108 * sk_filter_trim_cap - run a packet through a socket filter
109 * @sk: sock associated with &sk_buff
110 * @skb: buffer to filter
111 * @cap: limit on how short the eBPF program may trim the packet
113 * Run the eBPF program and then cut skb->data to correct size returned by
114 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
115 * than pkt_len we keep whole skb->data. This is the socket level
116 * wrapper to BPF_PROG_RUN. It returns 0 if the packet should
117 * be accepted or -EPERM if the packet should be tossed.
120 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
123 struct sk_filter *filter;
126 * If the skb was allocated from pfmemalloc reserves, only
127 * allow SOCK_MEMALLOC sockets to use it as this socket is
128 * helping free memory
130 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
131 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
134 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
138 err = security_sock_rcv_skb(sk, skb);
143 filter = rcu_dereference(sk->sk_filter);
145 struct sock *save_sk = skb->sk;
146 unsigned int pkt_len;
149 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
151 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
157 EXPORT_SYMBOL(sk_filter_trim_cap);
159 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
161 return skb_get_poff(skb);
164 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
168 if (skb_is_nonlinear(skb))
171 if (skb->len < sizeof(struct nlattr))
174 if (a > skb->len - sizeof(struct nlattr))
177 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
179 return (void *) nla - (void *) skb->data;
184 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
188 if (skb_is_nonlinear(skb))
191 if (skb->len < sizeof(struct nlattr))
194 if (a > skb->len - sizeof(struct nlattr))
197 nla = (struct nlattr *) &skb->data[a];
198 if (nla->nla_len > skb->len - a)
201 nla = nla_find_nested(nla, x);
203 return (void *) nla - (void *) skb->data;
208 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
209 data, int, headlen, int, offset)
212 const int len = sizeof(tmp);
215 if (headlen - offset >= len)
216 return *(u8 *)(data + offset);
217 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
220 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
228 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
231 return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
235 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
236 data, int, headlen, int, offset)
239 const int len = sizeof(tmp);
242 if (headlen - offset >= len)
243 return get_unaligned_be16(data + offset);
244 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
245 return be16_to_cpu(tmp);
247 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
249 return get_unaligned_be16(ptr);
255 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
258 return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
262 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
263 data, int, headlen, int, offset)
266 const int len = sizeof(tmp);
268 if (likely(offset >= 0)) {
269 if (headlen - offset >= len)
270 return get_unaligned_be32(data + offset);
271 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
272 return be32_to_cpu(tmp);
274 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
276 return get_unaligned_be32(ptr);
282 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
285 return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
289 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
290 struct bpf_insn *insn_buf)
292 struct bpf_insn *insn = insn_buf;
296 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4);
298 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
299 offsetof(struct sk_buff, mark));
303 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
304 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
305 #ifdef __BIG_ENDIAN_BITFIELD
306 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
311 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2);
313 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
314 offsetof(struct sk_buff, queue_mapping));
317 case SKF_AD_VLAN_TAG:
318 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2);
320 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
321 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
322 offsetof(struct sk_buff, vlan_tci));
324 case SKF_AD_VLAN_TAG_PRESENT:
325 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_VLAN_PRESENT_OFFSET());
326 if (PKT_VLAN_PRESENT_BIT)
327 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, PKT_VLAN_PRESENT_BIT);
328 if (PKT_VLAN_PRESENT_BIT < 7)
329 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
333 return insn - insn_buf;
336 static bool convert_bpf_extensions(struct sock_filter *fp,
337 struct bpf_insn **insnp)
339 struct bpf_insn *insn = *insnp;
343 case SKF_AD_OFF + SKF_AD_PROTOCOL:
344 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2);
346 /* A = *(u16 *) (CTX + offsetof(protocol)) */
347 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
348 offsetof(struct sk_buff, protocol));
349 /* A = ntohs(A) [emitting a nop or swap16] */
350 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
353 case SKF_AD_OFF + SKF_AD_PKTTYPE:
354 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
358 case SKF_AD_OFF + SKF_AD_IFINDEX:
359 case SKF_AD_OFF + SKF_AD_HATYPE:
360 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4);
361 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2);
363 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
364 BPF_REG_TMP, BPF_REG_CTX,
365 offsetof(struct sk_buff, dev));
366 /* if (tmp != 0) goto pc + 1 */
367 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
368 *insn++ = BPF_EXIT_INSN();
369 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
370 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
371 offsetof(struct net_device, ifindex));
373 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
374 offsetof(struct net_device, type));
377 case SKF_AD_OFF + SKF_AD_MARK:
378 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
382 case SKF_AD_OFF + SKF_AD_RXHASH:
383 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4);
385 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
386 offsetof(struct sk_buff, hash));
389 case SKF_AD_OFF + SKF_AD_QUEUE:
390 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
394 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
395 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
396 BPF_REG_A, BPF_REG_CTX, insn);
400 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
401 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
402 BPF_REG_A, BPF_REG_CTX, insn);
406 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
407 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2);
409 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
410 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
411 offsetof(struct sk_buff, vlan_proto));
412 /* A = ntohs(A) [emitting a nop or swap16] */
413 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
416 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
417 case SKF_AD_OFF + SKF_AD_NLATTR:
418 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
419 case SKF_AD_OFF + SKF_AD_CPU:
420 case SKF_AD_OFF + SKF_AD_RANDOM:
422 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
424 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
426 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
427 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
429 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
430 *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
432 case SKF_AD_OFF + SKF_AD_NLATTR:
433 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
435 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
436 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
438 case SKF_AD_OFF + SKF_AD_CPU:
439 *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
441 case SKF_AD_OFF + SKF_AD_RANDOM:
442 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
443 bpf_user_rnd_init_once();
448 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
450 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
454 /* This is just a dummy call to avoid letting the compiler
455 * evict __bpf_call_base() as an optimization. Placed here
456 * where no-one bothers.
458 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
466 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
468 const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
469 int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
470 bool endian = BPF_SIZE(fp->code) == BPF_H ||
471 BPF_SIZE(fp->code) == BPF_W;
472 bool indirect = BPF_MODE(fp->code) == BPF_IND;
473 const int ip_align = NET_IP_ALIGN;
474 struct bpf_insn *insn = *insnp;
478 ((unaligned_ok && offset >= 0) ||
479 (!unaligned_ok && offset >= 0 &&
480 offset + ip_align >= 0 &&
481 offset + ip_align % size == 0))) {
482 bool ldx_off_ok = offset <= S16_MAX;
484 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
486 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
487 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
488 size, 2 + endian + (!ldx_off_ok * 2));
490 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
493 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
494 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
495 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
499 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
500 *insn++ = BPF_JMP_A(8);
503 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
504 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
505 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
507 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
509 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
511 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
514 switch (BPF_SIZE(fp->code)) {
516 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
519 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
522 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
528 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
529 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
530 *insn = BPF_EXIT_INSN();
537 * bpf_convert_filter - convert filter program
538 * @prog: the user passed filter program
539 * @len: the length of the user passed filter program
540 * @new_prog: allocated 'struct bpf_prog' or NULL
541 * @new_len: pointer to store length of converted program
542 * @seen_ld_abs: bool whether we've seen ld_abs/ind
544 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
545 * style extended BPF (eBPF).
546 * Conversion workflow:
548 * 1) First pass for calculating the new program length:
549 * bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
551 * 2) 2nd pass to remap in two passes: 1st pass finds new
552 * jump offsets, 2nd pass remapping:
553 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
555 static int bpf_convert_filter(struct sock_filter *prog, int len,
556 struct bpf_prog *new_prog, int *new_len,
559 int new_flen = 0, pass = 0, target, i, stack_off;
560 struct bpf_insn *new_insn, *first_insn = NULL;
561 struct sock_filter *fp;
565 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
566 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
568 if (len <= 0 || len > BPF_MAXINSNS)
572 first_insn = new_prog->insnsi;
573 addrs = kcalloc(len, sizeof(*addrs),
574 GFP_KERNEL | __GFP_NOWARN);
580 new_insn = first_insn;
583 /* Classic BPF related prologue emission. */
585 /* Classic BPF expects A and X to be reset first. These need
586 * to be guaranteed to be the first two instructions.
588 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
589 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
591 /* All programs must keep CTX in callee saved BPF_REG_CTX.
592 * In eBPF case it's done by the compiler, here we need to
593 * do this ourself. Initial CTX is present in BPF_REG_ARG1.
595 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
597 /* For packet access in classic BPF, cache skb->data
598 * in callee-saved BPF R8 and skb->len - skb->data_len
599 * (headlen) in BPF R9. Since classic BPF is read-only
600 * on CTX, we only need to cache it once.
602 *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
603 BPF_REG_D, BPF_REG_CTX,
604 offsetof(struct sk_buff, data));
605 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
606 offsetof(struct sk_buff, len));
607 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
608 offsetof(struct sk_buff, data_len));
609 *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
615 for (i = 0; i < len; fp++, i++) {
616 struct bpf_insn tmp_insns[32] = { };
617 struct bpf_insn *insn = tmp_insns;
620 addrs[i] = new_insn - first_insn;
623 /* All arithmetic insns and skb loads map as-is. */
624 case BPF_ALU | BPF_ADD | BPF_X:
625 case BPF_ALU | BPF_ADD | BPF_K:
626 case BPF_ALU | BPF_SUB | BPF_X:
627 case BPF_ALU | BPF_SUB | BPF_K:
628 case BPF_ALU | BPF_AND | BPF_X:
629 case BPF_ALU | BPF_AND | BPF_K:
630 case BPF_ALU | BPF_OR | BPF_X:
631 case BPF_ALU | BPF_OR | BPF_K:
632 case BPF_ALU | BPF_LSH | BPF_X:
633 case BPF_ALU | BPF_LSH | BPF_K:
634 case BPF_ALU | BPF_RSH | BPF_X:
635 case BPF_ALU | BPF_RSH | BPF_K:
636 case BPF_ALU | BPF_XOR | BPF_X:
637 case BPF_ALU | BPF_XOR | BPF_K:
638 case BPF_ALU | BPF_MUL | BPF_X:
639 case BPF_ALU | BPF_MUL | BPF_K:
640 case BPF_ALU | BPF_DIV | BPF_X:
641 case BPF_ALU | BPF_DIV | BPF_K:
642 case BPF_ALU | BPF_MOD | BPF_X:
643 case BPF_ALU | BPF_MOD | BPF_K:
644 case BPF_ALU | BPF_NEG:
645 case BPF_LD | BPF_ABS | BPF_W:
646 case BPF_LD | BPF_ABS | BPF_H:
647 case BPF_LD | BPF_ABS | BPF_B:
648 case BPF_LD | BPF_IND | BPF_W:
649 case BPF_LD | BPF_IND | BPF_H:
650 case BPF_LD | BPF_IND | BPF_B:
651 /* Check for overloaded BPF extension and
652 * directly convert it if found, otherwise
653 * just move on with mapping.
655 if (BPF_CLASS(fp->code) == BPF_LD &&
656 BPF_MODE(fp->code) == BPF_ABS &&
657 convert_bpf_extensions(fp, &insn))
659 if (BPF_CLASS(fp->code) == BPF_LD &&
660 convert_bpf_ld_abs(fp, &insn)) {
665 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
666 fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
667 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
668 /* Error with exception code on div/mod by 0.
669 * For cBPF programs, this was always return 0.
671 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
672 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
673 *insn++ = BPF_EXIT_INSN();
676 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
679 /* Jump transformation cannot use BPF block macros
680 * everywhere as offset calculation and target updates
681 * require a bit more work than the rest, i.e. jump
682 * opcodes map as-is, but offsets need adjustment.
685 #define BPF_EMIT_JMP \
687 const s32 off_min = S16_MIN, off_max = S16_MAX; \
690 if (target >= len || target < 0) \
692 off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
693 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
694 off -= insn - tmp_insns; \
695 /* Reject anything not fitting into insn->off. */ \
696 if (off < off_min || off > off_max) \
701 case BPF_JMP | BPF_JA:
702 target = i + fp->k + 1;
703 insn->code = fp->code;
707 case BPF_JMP | BPF_JEQ | BPF_K:
708 case BPF_JMP | BPF_JEQ | BPF_X:
709 case BPF_JMP | BPF_JSET | BPF_K:
710 case BPF_JMP | BPF_JSET | BPF_X:
711 case BPF_JMP | BPF_JGT | BPF_K:
712 case BPF_JMP | BPF_JGT | BPF_X:
713 case BPF_JMP | BPF_JGE | BPF_K:
714 case BPF_JMP | BPF_JGE | BPF_X:
715 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
716 /* BPF immediates are signed, zero extend
717 * immediate into tmp register and use it
720 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
722 insn->dst_reg = BPF_REG_A;
723 insn->src_reg = BPF_REG_TMP;
726 insn->dst_reg = BPF_REG_A;
728 bpf_src = BPF_SRC(fp->code);
729 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
732 /* Common case where 'jump_false' is next insn. */
734 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
735 target = i + fp->jt + 1;
740 /* Convert some jumps when 'jump_true' is next insn. */
742 switch (BPF_OP(fp->code)) {
744 insn->code = BPF_JMP | BPF_JNE | bpf_src;
747 insn->code = BPF_JMP | BPF_JLE | bpf_src;
750 insn->code = BPF_JMP | BPF_JLT | bpf_src;
756 target = i + fp->jf + 1;
761 /* Other jumps are mapped into two insns: Jxx and JA. */
762 target = i + fp->jt + 1;
763 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
767 insn->code = BPF_JMP | BPF_JA;
768 target = i + fp->jf + 1;
772 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
773 case BPF_LDX | BPF_MSH | BPF_B: {
774 struct sock_filter tmp = {
775 .code = BPF_LD | BPF_ABS | BPF_B,
782 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
783 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
784 convert_bpf_ld_abs(&tmp, &insn);
787 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
789 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
791 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
793 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
795 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
798 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
799 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
801 case BPF_RET | BPF_A:
802 case BPF_RET | BPF_K:
803 if (BPF_RVAL(fp->code) == BPF_K)
804 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
806 *insn = BPF_EXIT_INSN();
809 /* Store to stack. */
812 stack_off = fp->k * 4 + 4;
813 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
814 BPF_ST ? BPF_REG_A : BPF_REG_X,
816 /* check_load_and_stores() verifies that classic BPF can
817 * load from stack only after write, so tracking
818 * stack_depth for ST|STX insns is enough
820 if (new_prog && new_prog->aux->stack_depth < stack_off)
821 new_prog->aux->stack_depth = stack_off;
824 /* Load from stack. */
825 case BPF_LD | BPF_MEM:
826 case BPF_LDX | BPF_MEM:
827 stack_off = fp->k * 4 + 4;
828 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
829 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
834 case BPF_LD | BPF_IMM:
835 case BPF_LDX | BPF_IMM:
836 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
837 BPF_REG_A : BPF_REG_X, fp->k);
841 case BPF_MISC | BPF_TAX:
842 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
846 case BPF_MISC | BPF_TXA:
847 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
850 /* A = skb->len or X = skb->len */
851 case BPF_LD | BPF_W | BPF_LEN:
852 case BPF_LDX | BPF_W | BPF_LEN:
853 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
854 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
855 offsetof(struct sk_buff, len));
858 /* Access seccomp_data fields. */
859 case BPF_LDX | BPF_ABS | BPF_W:
860 /* A = *(u32 *) (ctx + K) */
861 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
864 /* Unknown instruction. */
871 memcpy(new_insn, tmp_insns,
872 sizeof(*insn) * (insn - tmp_insns));
873 new_insn += insn - tmp_insns;
877 /* Only calculating new length. */
878 *new_len = new_insn - first_insn;
880 *new_len += 4; /* Prologue bits. */
885 if (new_flen != new_insn - first_insn) {
886 new_flen = new_insn - first_insn;
893 BUG_ON(*new_len != new_flen);
902 * As we dont want to clear mem[] array for each packet going through
903 * __bpf_prog_run(), we check that filter loaded by user never try to read
904 * a cell if not previously written, and we check all branches to be sure
905 * a malicious user doesn't try to abuse us.
907 static int check_load_and_stores(const struct sock_filter *filter, int flen)
909 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
912 BUILD_BUG_ON(BPF_MEMWORDS > 16);
914 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
918 memset(masks, 0xff, flen * sizeof(*masks));
920 for (pc = 0; pc < flen; pc++) {
921 memvalid &= masks[pc];
923 switch (filter[pc].code) {
926 memvalid |= (1 << filter[pc].k);
928 case BPF_LD | BPF_MEM:
929 case BPF_LDX | BPF_MEM:
930 if (!(memvalid & (1 << filter[pc].k))) {
935 case BPF_JMP | BPF_JA:
936 /* A jump must set masks on target */
937 masks[pc + 1 + filter[pc].k] &= memvalid;
940 case BPF_JMP | BPF_JEQ | BPF_K:
941 case BPF_JMP | BPF_JEQ | BPF_X:
942 case BPF_JMP | BPF_JGE | BPF_K:
943 case BPF_JMP | BPF_JGE | BPF_X:
944 case BPF_JMP | BPF_JGT | BPF_K:
945 case BPF_JMP | BPF_JGT | BPF_X:
946 case BPF_JMP | BPF_JSET | BPF_K:
947 case BPF_JMP | BPF_JSET | BPF_X:
948 /* A jump must set masks on targets */
949 masks[pc + 1 + filter[pc].jt] &= memvalid;
950 masks[pc + 1 + filter[pc].jf] &= memvalid;
960 static bool chk_code_allowed(u16 code_to_probe)
962 static const bool codes[] = {
963 /* 32 bit ALU operations */
964 [BPF_ALU | BPF_ADD | BPF_K] = true,
965 [BPF_ALU | BPF_ADD | BPF_X] = true,
966 [BPF_ALU | BPF_SUB | BPF_K] = true,
967 [BPF_ALU | BPF_SUB | BPF_X] = true,
968 [BPF_ALU | BPF_MUL | BPF_K] = true,
969 [BPF_ALU | BPF_MUL | BPF_X] = true,
970 [BPF_ALU | BPF_DIV | BPF_K] = true,
971 [BPF_ALU | BPF_DIV | BPF_X] = true,
972 [BPF_ALU | BPF_MOD | BPF_K] = true,
973 [BPF_ALU | BPF_MOD | BPF_X] = true,
974 [BPF_ALU | BPF_AND | BPF_K] = true,
975 [BPF_ALU | BPF_AND | BPF_X] = true,
976 [BPF_ALU | BPF_OR | BPF_K] = true,
977 [BPF_ALU | BPF_OR | BPF_X] = true,
978 [BPF_ALU | BPF_XOR | BPF_K] = true,
979 [BPF_ALU | BPF_XOR | BPF_X] = true,
980 [BPF_ALU | BPF_LSH | BPF_K] = true,
981 [BPF_ALU | BPF_LSH | BPF_X] = true,
982 [BPF_ALU | BPF_RSH | BPF_K] = true,
983 [BPF_ALU | BPF_RSH | BPF_X] = true,
984 [BPF_ALU | BPF_NEG] = true,
985 /* Load instructions */
986 [BPF_LD | BPF_W | BPF_ABS] = true,
987 [BPF_LD | BPF_H | BPF_ABS] = true,
988 [BPF_LD | BPF_B | BPF_ABS] = true,
989 [BPF_LD | BPF_W | BPF_LEN] = true,
990 [BPF_LD | BPF_W | BPF_IND] = true,
991 [BPF_LD | BPF_H | BPF_IND] = true,
992 [BPF_LD | BPF_B | BPF_IND] = true,
993 [BPF_LD | BPF_IMM] = true,
994 [BPF_LD | BPF_MEM] = true,
995 [BPF_LDX | BPF_W | BPF_LEN] = true,
996 [BPF_LDX | BPF_B | BPF_MSH] = true,
997 [BPF_LDX | BPF_IMM] = true,
998 [BPF_LDX | BPF_MEM] = true,
999 /* Store instructions */
1002 /* Misc instructions */
1003 [BPF_MISC | BPF_TAX] = true,
1004 [BPF_MISC | BPF_TXA] = true,
1005 /* Return instructions */
1006 [BPF_RET | BPF_K] = true,
1007 [BPF_RET | BPF_A] = true,
1008 /* Jump instructions */
1009 [BPF_JMP | BPF_JA] = true,
1010 [BPF_JMP | BPF_JEQ | BPF_K] = true,
1011 [BPF_JMP | BPF_JEQ | BPF_X] = true,
1012 [BPF_JMP | BPF_JGE | BPF_K] = true,
1013 [BPF_JMP | BPF_JGE | BPF_X] = true,
1014 [BPF_JMP | BPF_JGT | BPF_K] = true,
1015 [BPF_JMP | BPF_JGT | BPF_X] = true,
1016 [BPF_JMP | BPF_JSET | BPF_K] = true,
1017 [BPF_JMP | BPF_JSET | BPF_X] = true,
1020 if (code_to_probe >= ARRAY_SIZE(codes))
1023 return codes[code_to_probe];
1026 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1031 if (flen == 0 || flen > BPF_MAXINSNS)
1038 * bpf_check_classic - verify socket filter code
1039 * @filter: filter to verify
1040 * @flen: length of filter
1042 * Check the user's filter code. If we let some ugly
1043 * filter code slip through kaboom! The filter must contain
1044 * no references or jumps that are out of range, no illegal
1045 * instructions, and must end with a RET instruction.
1047 * All jumps are forward as they are not signed.
1049 * Returns 0 if the rule set is legal or -EINVAL if not.
1051 static int bpf_check_classic(const struct sock_filter *filter,
1057 /* Check the filter code now */
1058 for (pc = 0; pc < flen; pc++) {
1059 const struct sock_filter *ftest = &filter[pc];
1061 /* May we actually operate on this code? */
1062 if (!chk_code_allowed(ftest->code))
1065 /* Some instructions need special checks */
1066 switch (ftest->code) {
1067 case BPF_ALU | BPF_DIV | BPF_K:
1068 case BPF_ALU | BPF_MOD | BPF_K:
1069 /* Check for division by zero */
1073 case BPF_ALU | BPF_LSH | BPF_K:
1074 case BPF_ALU | BPF_RSH | BPF_K:
1078 case BPF_LD | BPF_MEM:
1079 case BPF_LDX | BPF_MEM:
1082 /* Check for invalid memory addresses */
1083 if (ftest->k >= BPF_MEMWORDS)
1086 case BPF_JMP | BPF_JA:
1087 /* Note, the large ftest->k might cause loops.
1088 * Compare this with conditional jumps below,
1089 * where offsets are limited. --ANK (981016)
1091 if (ftest->k >= (unsigned int)(flen - pc - 1))
1094 case BPF_JMP | BPF_JEQ | BPF_K:
1095 case BPF_JMP | BPF_JEQ | BPF_X:
1096 case BPF_JMP | BPF_JGE | BPF_K:
1097 case BPF_JMP | BPF_JGE | BPF_X:
1098 case BPF_JMP | BPF_JGT | BPF_K:
1099 case BPF_JMP | BPF_JGT | BPF_X:
1100 case BPF_JMP | BPF_JSET | BPF_K:
1101 case BPF_JMP | BPF_JSET | BPF_X:
1102 /* Both conditionals must be safe */
1103 if (pc + ftest->jt + 1 >= flen ||
1104 pc + ftest->jf + 1 >= flen)
1107 case BPF_LD | BPF_W | BPF_ABS:
1108 case BPF_LD | BPF_H | BPF_ABS:
1109 case BPF_LD | BPF_B | BPF_ABS:
1111 if (bpf_anc_helper(ftest) & BPF_ANC)
1113 /* Ancillary operation unknown or unsupported */
1114 if (anc_found == false && ftest->k >= SKF_AD_OFF)
1119 /* Last instruction must be a RET code */
1120 switch (filter[flen - 1].code) {
1121 case BPF_RET | BPF_K:
1122 case BPF_RET | BPF_A:
1123 return check_load_and_stores(filter, flen);
1129 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1130 const struct sock_fprog *fprog)
1132 unsigned int fsize = bpf_classic_proglen(fprog);
1133 struct sock_fprog_kern *fkprog;
1135 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1139 fkprog = fp->orig_prog;
1140 fkprog->len = fprog->len;
1142 fkprog->filter = kmemdup(fp->insns, fsize,
1143 GFP_KERNEL | __GFP_NOWARN);
1144 if (!fkprog->filter) {
1145 kfree(fp->orig_prog);
1152 static void bpf_release_orig_filter(struct bpf_prog *fp)
1154 struct sock_fprog_kern *fprog = fp->orig_prog;
1157 kfree(fprog->filter);
1162 static void __bpf_prog_release(struct bpf_prog *prog)
1164 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1167 bpf_release_orig_filter(prog);
1168 bpf_prog_free(prog);
1172 static void __sk_filter_release(struct sk_filter *fp)
1174 __bpf_prog_release(fp->prog);
1179 * sk_filter_release_rcu - Release a socket filter by rcu_head
1180 * @rcu: rcu_head that contains the sk_filter to free
1182 static void sk_filter_release_rcu(struct rcu_head *rcu)
1184 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1186 __sk_filter_release(fp);
1190 * sk_filter_release - release a socket filter
1191 * @fp: filter to remove
1193 * Remove a filter from a socket and release its resources.
1195 static void sk_filter_release(struct sk_filter *fp)
1197 if (refcount_dec_and_test(&fp->refcnt))
1198 call_rcu(&fp->rcu, sk_filter_release_rcu);
1201 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1203 u32 filter_size = bpf_prog_size(fp->prog->len);
1205 atomic_sub(filter_size, &sk->sk_omem_alloc);
1206 sk_filter_release(fp);
1209 /* try to charge the socket memory if there is space available
1210 * return true on success
1212 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1214 u32 filter_size = bpf_prog_size(fp->prog->len);
1216 /* same check as in sock_kmalloc() */
1217 if (filter_size <= sysctl_optmem_max &&
1218 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
1219 atomic_add(filter_size, &sk->sk_omem_alloc);
1225 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1227 if (!refcount_inc_not_zero(&fp->refcnt))
1230 if (!__sk_filter_charge(sk, fp)) {
1231 sk_filter_release(fp);
1237 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1239 struct sock_filter *old_prog;
1240 struct bpf_prog *old_fp;
1241 int err, new_len, old_len = fp->len;
1242 bool seen_ld_abs = false;
1244 /* We are free to overwrite insns et al right here as it
1245 * won't be used at this point in time anymore internally
1246 * after the migration to the internal BPF instruction
1249 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1250 sizeof(struct bpf_insn));
1252 /* Conversion cannot happen on overlapping memory areas,
1253 * so we need to keep the user BPF around until the 2nd
1254 * pass. At this time, the user BPF is stored in fp->insns.
1256 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1257 GFP_KERNEL | __GFP_NOWARN);
1263 /* 1st pass: calculate the new program length. */
1264 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1269 /* Expand fp for appending the new filter representation. */
1271 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1273 /* The old_fp is still around in case we couldn't
1274 * allocate new memory, so uncharge on that one.
1283 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1284 err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1287 /* 2nd bpf_convert_filter() can fail only if it fails
1288 * to allocate memory, remapping must succeed. Note,
1289 * that at this time old_fp has already been released
1294 fp = bpf_prog_select_runtime(fp, &err);
1304 __bpf_prog_release(fp);
1305 return ERR_PTR(err);
1308 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1309 bpf_aux_classic_check_t trans)
1313 fp->bpf_func = NULL;
1316 err = bpf_check_classic(fp->insns, fp->len);
1318 __bpf_prog_release(fp);
1319 return ERR_PTR(err);
1322 /* There might be additional checks and transformations
1323 * needed on classic filters, f.e. in case of seccomp.
1326 err = trans(fp->insns, fp->len);
1328 __bpf_prog_release(fp);
1329 return ERR_PTR(err);
1333 /* Probe if we can JIT compile the filter and if so, do
1334 * the compilation of the filter.
1336 bpf_jit_compile(fp);
1338 /* JIT compiler couldn't process this filter, so do the
1339 * internal BPF translation for the optimized interpreter.
1342 fp = bpf_migrate_filter(fp);
1348 * bpf_prog_create - create an unattached filter
1349 * @pfp: the unattached filter that is created
1350 * @fprog: the filter program
1352 * Create a filter independent of any socket. We first run some
1353 * sanity checks on it to make sure it does not explode on us later.
1354 * If an error occurs or there is insufficient memory for the filter
1355 * a negative errno code is returned. On success the return is zero.
1357 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1359 unsigned int fsize = bpf_classic_proglen(fprog);
1360 struct bpf_prog *fp;
1362 /* Make sure new filter is there and in the right amounts. */
1363 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1366 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1370 memcpy(fp->insns, fprog->filter, fsize);
1372 fp->len = fprog->len;
1373 /* Since unattached filters are not copied back to user
1374 * space through sk_get_filter(), we do not need to hold
1375 * a copy here, and can spare us the work.
1377 fp->orig_prog = NULL;
1379 /* bpf_prepare_filter() already takes care of freeing
1380 * memory in case something goes wrong.
1382 fp = bpf_prepare_filter(fp, NULL);
1389 EXPORT_SYMBOL_GPL(bpf_prog_create);
1392 * bpf_prog_create_from_user - create an unattached filter from user buffer
1393 * @pfp: the unattached filter that is created
1394 * @fprog: the filter program
1395 * @trans: post-classic verifier transformation handler
1396 * @save_orig: save classic BPF program
1398 * This function effectively does the same as bpf_prog_create(), only
1399 * that it builds up its insns buffer from user space provided buffer.
1400 * It also allows for passing a bpf_aux_classic_check_t handler.
1402 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1403 bpf_aux_classic_check_t trans, bool save_orig)
1405 unsigned int fsize = bpf_classic_proglen(fprog);
1406 struct bpf_prog *fp;
1409 /* Make sure new filter is there and in the right amounts. */
1410 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1413 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1417 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1418 __bpf_prog_free(fp);
1422 fp->len = fprog->len;
1423 fp->orig_prog = NULL;
1426 err = bpf_prog_store_orig_filter(fp, fprog);
1428 __bpf_prog_free(fp);
1433 /* bpf_prepare_filter() already takes care of freeing
1434 * memory in case something goes wrong.
1436 fp = bpf_prepare_filter(fp, trans);
1443 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1445 void bpf_prog_destroy(struct bpf_prog *fp)
1447 __bpf_prog_release(fp);
1449 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1451 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1453 struct sk_filter *fp, *old_fp;
1455 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1461 if (!__sk_filter_charge(sk, fp)) {
1465 refcount_set(&fp->refcnt, 1);
1467 old_fp = rcu_dereference_protected(sk->sk_filter,
1468 lockdep_sock_is_held(sk));
1469 rcu_assign_pointer(sk->sk_filter, fp);
1472 sk_filter_uncharge(sk, old_fp);
1478 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1480 unsigned int fsize = bpf_classic_proglen(fprog);
1481 struct bpf_prog *prog;
1484 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1485 return ERR_PTR(-EPERM);
1487 /* Make sure new filter is there and in the right amounts. */
1488 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1489 return ERR_PTR(-EINVAL);
1491 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1493 return ERR_PTR(-ENOMEM);
1495 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1496 __bpf_prog_free(prog);
1497 return ERR_PTR(-EFAULT);
1500 prog->len = fprog->len;
1502 err = bpf_prog_store_orig_filter(prog, fprog);
1504 __bpf_prog_free(prog);
1505 return ERR_PTR(-ENOMEM);
1508 /* bpf_prepare_filter() already takes care of freeing
1509 * memory in case something goes wrong.
1511 return bpf_prepare_filter(prog, NULL);
1515 * sk_attach_filter - attach a socket filter
1516 * @fprog: the filter program
1517 * @sk: the socket to use
1519 * Attach the user's filter code. We first run some sanity checks on
1520 * it to make sure it does not explode on us later. If an error
1521 * occurs or there is insufficient memory for the filter a negative
1522 * errno code is returned. On success the return is zero.
1524 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1526 struct bpf_prog *prog = __get_filter(fprog, sk);
1530 return PTR_ERR(prog);
1532 err = __sk_attach_prog(prog, sk);
1534 __bpf_prog_release(prog);
1540 EXPORT_SYMBOL_GPL(sk_attach_filter);
1542 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1544 struct bpf_prog *prog = __get_filter(fprog, sk);
1548 return PTR_ERR(prog);
1550 if (bpf_prog_size(prog->len) > sysctl_optmem_max)
1553 err = reuseport_attach_prog(sk, prog);
1556 __bpf_prog_release(prog);
1561 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1563 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1564 return ERR_PTR(-EPERM);
1566 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1569 int sk_attach_bpf(u32 ufd, struct sock *sk)
1571 struct bpf_prog *prog = __get_bpf(ufd, sk);
1575 return PTR_ERR(prog);
1577 err = __sk_attach_prog(prog, sk);
1586 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1588 struct bpf_prog *prog;
1591 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1594 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1595 if (PTR_ERR(prog) == -EINVAL)
1596 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1598 return PTR_ERR(prog);
1600 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1601 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1602 * bpf prog (e.g. sockmap). It depends on the
1603 * limitation imposed by bpf_prog_load().
1604 * Hence, sysctl_optmem_max is not checked.
1606 if ((sk->sk_type != SOCK_STREAM &&
1607 sk->sk_type != SOCK_DGRAM) ||
1608 (sk->sk_protocol != IPPROTO_UDP &&
1609 sk->sk_protocol != IPPROTO_TCP) ||
1610 (sk->sk_family != AF_INET &&
1611 sk->sk_family != AF_INET6)) {
1616 /* BPF_PROG_TYPE_SOCKET_FILTER */
1617 if (bpf_prog_size(prog->len) > sysctl_optmem_max) {
1623 err = reuseport_attach_prog(sk, prog);
1631 void sk_reuseport_prog_free(struct bpf_prog *prog)
1636 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1639 bpf_prog_destroy(prog);
1642 struct bpf_scratchpad {
1644 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1645 u8 buff[MAX_BPF_STACK];
1649 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1651 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1652 unsigned int write_len)
1654 return skb_ensure_writable(skb, write_len);
1657 static inline int bpf_try_make_writable(struct sk_buff *skb,
1658 unsigned int write_len)
1660 int err = __bpf_try_make_writable(skb, write_len);
1662 bpf_compute_data_pointers(skb);
1666 static int bpf_try_make_head_writable(struct sk_buff *skb)
1668 return bpf_try_make_writable(skb, skb_headlen(skb));
1671 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1673 if (skb_at_tc_ingress(skb))
1674 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1677 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1679 if (skb_at_tc_ingress(skb))
1680 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1683 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1684 const void *, from, u32, len, u64, flags)
1688 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1690 if (unlikely(offset > 0xffff))
1692 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1695 ptr = skb->data + offset;
1696 if (flags & BPF_F_RECOMPUTE_CSUM)
1697 __skb_postpull_rcsum(skb, ptr, len, offset);
1699 memcpy(ptr, from, len);
1701 if (flags & BPF_F_RECOMPUTE_CSUM)
1702 __skb_postpush_rcsum(skb, ptr, len, offset);
1703 if (flags & BPF_F_INVALIDATE_HASH)
1704 skb_clear_hash(skb);
1709 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1710 .func = bpf_skb_store_bytes,
1712 .ret_type = RET_INTEGER,
1713 .arg1_type = ARG_PTR_TO_CTX,
1714 .arg2_type = ARG_ANYTHING,
1715 .arg3_type = ARG_PTR_TO_MEM,
1716 .arg4_type = ARG_CONST_SIZE,
1717 .arg5_type = ARG_ANYTHING,
1720 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1721 void *, to, u32, len)
1725 if (unlikely(offset > 0xffff))
1728 ptr = skb_header_pointer(skb, offset, len, to);
1732 memcpy(to, ptr, len);
1740 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1741 .func = bpf_skb_load_bytes,
1743 .ret_type = RET_INTEGER,
1744 .arg1_type = ARG_PTR_TO_CTX,
1745 .arg2_type = ARG_ANYTHING,
1746 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1747 .arg4_type = ARG_CONST_SIZE,
1750 BPF_CALL_4(bpf_flow_dissector_load_bytes,
1751 const struct bpf_flow_dissector *, ctx, u32, offset,
1752 void *, to, u32, len)
1756 if (unlikely(offset > 0xffff))
1759 if (unlikely(!ctx->skb))
1762 ptr = skb_header_pointer(ctx->skb, offset, len, to);
1766 memcpy(to, ptr, len);
1774 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1775 .func = bpf_flow_dissector_load_bytes,
1777 .ret_type = RET_INTEGER,
1778 .arg1_type = ARG_PTR_TO_CTX,
1779 .arg2_type = ARG_ANYTHING,
1780 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1781 .arg4_type = ARG_CONST_SIZE,
1784 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1785 u32, offset, void *, to, u32, len, u32, start_header)
1787 u8 *end = skb_tail_pointer(skb);
1790 if (unlikely(offset > 0xffff))
1793 switch (start_header) {
1794 case BPF_HDR_START_MAC:
1795 if (unlikely(!skb_mac_header_was_set(skb)))
1797 start = skb_mac_header(skb);
1799 case BPF_HDR_START_NET:
1800 start = skb_network_header(skb);
1806 ptr = start + offset;
1808 if (likely(ptr + len <= end)) {
1809 memcpy(to, ptr, len);
1818 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1819 .func = bpf_skb_load_bytes_relative,
1821 .ret_type = RET_INTEGER,
1822 .arg1_type = ARG_PTR_TO_CTX,
1823 .arg2_type = ARG_ANYTHING,
1824 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1825 .arg4_type = ARG_CONST_SIZE,
1826 .arg5_type = ARG_ANYTHING,
1829 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1831 /* Idea is the following: should the needed direct read/write
1832 * test fail during runtime, we can pull in more data and redo
1833 * again, since implicitly, we invalidate previous checks here.
1835 * Or, since we know how much we need to make read/writeable,
1836 * this can be done once at the program beginning for direct
1837 * access case. By this we overcome limitations of only current
1838 * headroom being accessible.
1840 return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1843 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1844 .func = bpf_skb_pull_data,
1846 .ret_type = RET_INTEGER,
1847 .arg1_type = ARG_PTR_TO_CTX,
1848 .arg2_type = ARG_ANYTHING,
1851 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1853 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1856 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1857 .func = bpf_sk_fullsock,
1859 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
1860 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
1863 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1864 unsigned int write_len)
1866 int err = __bpf_try_make_writable(skb, write_len);
1868 bpf_compute_data_end_sk_skb(skb);
1872 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1874 /* Idea is the following: should the needed direct read/write
1875 * test fail during runtime, we can pull in more data and redo
1876 * again, since implicitly, we invalidate previous checks here.
1878 * Or, since we know how much we need to make read/writeable,
1879 * this can be done once at the program beginning for direct
1880 * access case. By this we overcome limitations of only current
1881 * headroom being accessible.
1883 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1886 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1887 .func = sk_skb_pull_data,
1889 .ret_type = RET_INTEGER,
1890 .arg1_type = ARG_PTR_TO_CTX,
1891 .arg2_type = ARG_ANYTHING,
1894 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1895 u64, from, u64, to, u64, flags)
1899 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1901 if (unlikely(offset > 0xffff || offset & 1))
1903 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1906 ptr = (__sum16 *)(skb->data + offset);
1907 switch (flags & BPF_F_HDR_FIELD_MASK) {
1909 if (unlikely(from != 0))
1912 csum_replace_by_diff(ptr, to);
1915 csum_replace2(ptr, from, to);
1918 csum_replace4(ptr, from, to);
1927 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1928 .func = bpf_l3_csum_replace,
1930 .ret_type = RET_INTEGER,
1931 .arg1_type = ARG_PTR_TO_CTX,
1932 .arg2_type = ARG_ANYTHING,
1933 .arg3_type = ARG_ANYTHING,
1934 .arg4_type = ARG_ANYTHING,
1935 .arg5_type = ARG_ANYTHING,
1938 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1939 u64, from, u64, to, u64, flags)
1941 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1942 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1943 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1946 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1947 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1949 if (unlikely(offset > 0xffff || offset & 1))
1951 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1954 ptr = (__sum16 *)(skb->data + offset);
1955 if (is_mmzero && !do_mforce && !*ptr)
1958 switch (flags & BPF_F_HDR_FIELD_MASK) {
1960 if (unlikely(from != 0))
1963 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1966 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1969 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1975 if (is_mmzero && !*ptr)
1976 *ptr = CSUM_MANGLED_0;
1980 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1981 .func = bpf_l4_csum_replace,
1983 .ret_type = RET_INTEGER,
1984 .arg1_type = ARG_PTR_TO_CTX,
1985 .arg2_type = ARG_ANYTHING,
1986 .arg3_type = ARG_ANYTHING,
1987 .arg4_type = ARG_ANYTHING,
1988 .arg5_type = ARG_ANYTHING,
1991 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1992 __be32 *, to, u32, to_size, __wsum, seed)
1994 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1995 u32 diff_size = from_size + to_size;
1998 /* This is quite flexible, some examples:
2000 * from_size == 0, to_size > 0, seed := csum --> pushing data
2001 * from_size > 0, to_size == 0, seed := csum --> pulling data
2002 * from_size > 0, to_size > 0, seed := 0 --> diffing data
2004 * Even for diffing, from_size and to_size don't need to be equal.
2006 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
2007 diff_size > sizeof(sp->diff)))
2010 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
2011 sp->diff[j] = ~from[i];
2012 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
2013 sp->diff[j] = to[i];
2015 return csum_partial(sp->diff, diff_size, seed);
2018 static const struct bpf_func_proto bpf_csum_diff_proto = {
2019 .func = bpf_csum_diff,
2022 .ret_type = RET_INTEGER,
2023 .arg1_type = ARG_PTR_TO_MEM_OR_NULL,
2024 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
2025 .arg3_type = ARG_PTR_TO_MEM_OR_NULL,
2026 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
2027 .arg5_type = ARG_ANYTHING,
2030 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2032 /* The interface is to be used in combination with bpf_csum_diff()
2033 * for direct packet writes. csum rotation for alignment as well
2034 * as emulating csum_sub() can be done from the eBPF program.
2036 if (skb->ip_summed == CHECKSUM_COMPLETE)
2037 return (skb->csum = csum_add(skb->csum, csum));
2042 static const struct bpf_func_proto bpf_csum_update_proto = {
2043 .func = bpf_csum_update,
2045 .ret_type = RET_INTEGER,
2046 .arg1_type = ARG_PTR_TO_CTX,
2047 .arg2_type = ARG_ANYTHING,
2050 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
2052 /* The interface is to be used in combination with bpf_skb_adjust_room()
2053 * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET
2054 * is passed as flags, for example.
2057 case BPF_CSUM_LEVEL_INC:
2058 __skb_incr_checksum_unnecessary(skb);
2060 case BPF_CSUM_LEVEL_DEC:
2061 __skb_decr_checksum_unnecessary(skb);
2063 case BPF_CSUM_LEVEL_RESET:
2064 __skb_reset_checksum_unnecessary(skb);
2066 case BPF_CSUM_LEVEL_QUERY:
2067 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2068 skb->csum_level : -EACCES;
2076 static const struct bpf_func_proto bpf_csum_level_proto = {
2077 .func = bpf_csum_level,
2079 .ret_type = RET_INTEGER,
2080 .arg1_type = ARG_PTR_TO_CTX,
2081 .arg2_type = ARG_ANYTHING,
2084 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2086 return dev_forward_skb(dev, skb);
2089 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2090 struct sk_buff *skb)
2092 int ret = ____dev_forward_skb(dev, skb);
2096 ret = netif_rx(skb);
2102 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2106 if (dev_xmit_recursion()) {
2107 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2115 dev_xmit_recursion_inc();
2116 ret = dev_queue_xmit(skb);
2117 dev_xmit_recursion_dec();
2122 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2125 unsigned int mlen = skb_network_offset(skb);
2128 __skb_pull(skb, mlen);
2130 /* At ingress, the mac header has already been pulled once.
2131 * At egress, skb_pospull_rcsum has to be done in case that
2132 * the skb is originated from ingress (i.e. a forwarded skb)
2133 * to ensure that rcsum starts at net header.
2135 if (!skb_at_tc_ingress(skb))
2136 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2138 skb_pop_mac_header(skb);
2139 skb_reset_mac_len(skb);
2140 return flags & BPF_F_INGRESS ?
2141 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2144 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2147 /* Verify that a link layer header is carried */
2148 if (unlikely(skb->mac_header >= skb->network_header)) {
2153 bpf_push_mac_rcsum(skb);
2154 return flags & BPF_F_INGRESS ?
2155 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2158 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2161 if (dev_is_mac_header_xmit(dev))
2162 return __bpf_redirect_common(skb, dev, flags);
2164 return __bpf_redirect_no_mac(skb, dev, flags);
2167 #if IS_ENABLED(CONFIG_IPV6)
2168 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2169 struct net_device *dev, struct bpf_nh_params *nh)
2171 u32 hh_len = LL_RESERVED_SPACE(dev);
2172 const struct in6_addr *nexthop;
2173 struct dst_entry *dst = NULL;
2174 struct neighbour *neigh;
2176 if (dev_xmit_recursion()) {
2177 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2184 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2185 struct sk_buff *skb2;
2187 skb2 = skb_realloc_headroom(skb, hh_len);
2188 if (unlikely(!skb2)) {
2193 skb_set_owner_w(skb2, skb->sk);
2201 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2202 &ipv6_hdr(skb)->daddr);
2204 nexthop = &nh->ipv6_nh;
2206 neigh = ip_neigh_gw6(dev, nexthop);
2207 if (likely(!IS_ERR(neigh))) {
2210 sock_confirm_neigh(skb, neigh);
2211 dev_xmit_recursion_inc();
2212 ret = neigh_output(neigh, skb, false);
2213 dev_xmit_recursion_dec();
2214 rcu_read_unlock_bh();
2217 rcu_read_unlock_bh();
2219 IP6_INC_STATS(dev_net(dst->dev),
2220 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2226 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2227 struct bpf_nh_params *nh)
2229 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2230 struct net *net = dev_net(dev);
2231 int err, ret = NET_XMIT_DROP;
2234 struct dst_entry *dst;
2235 struct flowi6 fl6 = {
2236 .flowi6_flags = FLOWI_FLAG_ANYSRC,
2237 .flowi6_mark = skb->mark,
2238 .flowlabel = ip6_flowinfo(ip6h),
2239 .flowi6_oif = dev->ifindex,
2240 .flowi6_proto = ip6h->nexthdr,
2241 .daddr = ip6h->daddr,
2242 .saddr = ip6h->saddr,
2245 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2249 skb_dst_set(skb, dst);
2250 } else if (nh->nh_family != AF_INET6) {
2254 err = bpf_out_neigh_v6(net, skb, dev, nh);
2255 if (unlikely(net_xmit_eval(err)))
2256 dev->stats.tx_errors++;
2258 ret = NET_XMIT_SUCCESS;
2261 dev->stats.tx_errors++;
2267 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2268 struct bpf_nh_params *nh)
2271 return NET_XMIT_DROP;
2273 #endif /* CONFIG_IPV6 */
2275 #if IS_ENABLED(CONFIG_INET)
2276 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2277 struct net_device *dev, struct bpf_nh_params *nh)
2279 u32 hh_len = LL_RESERVED_SPACE(dev);
2280 struct neighbour *neigh;
2281 bool is_v6gw = false;
2283 if (dev_xmit_recursion()) {
2284 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2291 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2292 struct sk_buff *skb2;
2294 skb2 = skb_realloc_headroom(skb, hh_len);
2295 if (unlikely(!skb2)) {
2300 skb_set_owner_w(skb2, skb->sk);
2307 struct dst_entry *dst = skb_dst(skb);
2308 struct rtable *rt = container_of(dst, struct rtable, dst);
2310 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2311 } else if (nh->nh_family == AF_INET6) {
2312 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2314 } else if (nh->nh_family == AF_INET) {
2315 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2317 rcu_read_unlock_bh();
2321 if (likely(!IS_ERR(neigh))) {
2324 sock_confirm_neigh(skb, neigh);
2325 dev_xmit_recursion_inc();
2326 ret = neigh_output(neigh, skb, is_v6gw);
2327 dev_xmit_recursion_dec();
2328 rcu_read_unlock_bh();
2331 rcu_read_unlock_bh();
2337 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2338 struct bpf_nh_params *nh)
2340 const struct iphdr *ip4h = ip_hdr(skb);
2341 struct net *net = dev_net(dev);
2342 int err, ret = NET_XMIT_DROP;
2345 struct flowi4 fl4 = {
2346 .flowi4_flags = FLOWI_FLAG_ANYSRC,
2347 .flowi4_mark = skb->mark,
2348 .flowi4_tos = RT_TOS(ip4h->tos),
2349 .flowi4_oif = dev->ifindex,
2350 .flowi4_proto = ip4h->protocol,
2351 .daddr = ip4h->daddr,
2352 .saddr = ip4h->saddr,
2356 rt = ip_route_output_flow(net, &fl4, NULL);
2359 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2364 skb_dst_set(skb, &rt->dst);
2367 err = bpf_out_neigh_v4(net, skb, dev, nh);
2368 if (unlikely(net_xmit_eval(err)))
2369 dev->stats.tx_errors++;
2371 ret = NET_XMIT_SUCCESS;
2374 dev->stats.tx_errors++;
2380 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2381 struct bpf_nh_params *nh)
2384 return NET_XMIT_DROP;
2386 #endif /* CONFIG_INET */
2388 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2389 struct bpf_nh_params *nh)
2391 struct ethhdr *ethh = eth_hdr(skb);
2393 if (unlikely(skb->mac_header >= skb->network_header))
2395 bpf_push_mac_rcsum(skb);
2396 if (is_multicast_ether_addr(ethh->h_dest))
2399 skb_pull(skb, sizeof(*ethh));
2400 skb_unset_mac_header(skb);
2401 skb_reset_network_header(skb);
2403 if (skb->protocol == htons(ETH_P_IP))
2404 return __bpf_redirect_neigh_v4(skb, dev, nh);
2405 else if (skb->protocol == htons(ETH_P_IPV6))
2406 return __bpf_redirect_neigh_v6(skb, dev, nh);
2412 /* Internal, non-exposed redirect flags. */
2414 BPF_F_NEIGH = (1ULL << 1),
2415 BPF_F_PEER = (1ULL << 2),
2416 BPF_F_NEXTHOP = (1ULL << 3),
2417 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2420 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2422 struct net_device *dev;
2423 struct sk_buff *clone;
2426 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2429 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2433 clone = skb_clone(skb, GFP_ATOMIC);
2434 if (unlikely(!clone))
2437 /* For direct write, we need to keep the invariant that the skbs
2438 * we're dealing with need to be uncloned. Should uncloning fail
2439 * here, we need to free the just generated clone to unclone once
2442 ret = bpf_try_make_head_writable(skb);
2443 if (unlikely(ret)) {
2448 return __bpf_redirect(clone, dev, flags);
2451 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2452 .func = bpf_clone_redirect,
2454 .ret_type = RET_INTEGER,
2455 .arg1_type = ARG_PTR_TO_CTX,
2456 .arg2_type = ARG_ANYTHING,
2457 .arg3_type = ARG_ANYTHING,
2460 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2461 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2463 int skb_do_redirect(struct sk_buff *skb)
2465 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2466 struct net *net = dev_net(skb->dev);
2467 struct net_device *dev;
2468 u32 flags = ri->flags;
2470 dev = dev_get_by_index_rcu(net, ri->tgt_index);
2475 if (flags & BPF_F_PEER) {
2476 const struct net_device_ops *ops = dev->netdev_ops;
2478 if (unlikely(!ops->ndo_get_peer_dev ||
2479 !skb_at_tc_ingress(skb)))
2481 dev = ops->ndo_get_peer_dev(dev);
2482 if (unlikely(!dev ||
2483 !is_skb_forwardable(dev, skb) ||
2484 net_eq(net, dev_net(dev))))
2489 return flags & BPF_F_NEIGH ?
2490 __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2492 __bpf_redirect(skb, dev, flags);
2498 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2500 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2502 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2506 ri->tgt_index = ifindex;
2508 return TC_ACT_REDIRECT;
2511 static const struct bpf_func_proto bpf_redirect_proto = {
2512 .func = bpf_redirect,
2514 .ret_type = RET_INTEGER,
2515 .arg1_type = ARG_ANYTHING,
2516 .arg2_type = ARG_ANYTHING,
2519 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2521 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2523 if (unlikely(flags))
2526 ri->flags = BPF_F_PEER;
2527 ri->tgt_index = ifindex;
2529 return TC_ACT_REDIRECT;
2532 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2533 .func = bpf_redirect_peer,
2535 .ret_type = RET_INTEGER,
2536 .arg1_type = ARG_ANYTHING,
2537 .arg2_type = ARG_ANYTHING,
2540 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2541 int, plen, u64, flags)
2543 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2545 if (unlikely((plen && plen < sizeof(*params)) || flags))
2548 ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2549 ri->tgt_index = ifindex;
2551 BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2553 memcpy(&ri->nh, params, sizeof(ri->nh));
2555 return TC_ACT_REDIRECT;
2558 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2559 .func = bpf_redirect_neigh,
2561 .ret_type = RET_INTEGER,
2562 .arg1_type = ARG_ANYTHING,
2563 .arg2_type = ARG_PTR_TO_MEM_OR_NULL,
2564 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
2565 .arg4_type = ARG_ANYTHING,
2568 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2570 msg->apply_bytes = bytes;
2574 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2575 .func = bpf_msg_apply_bytes,
2577 .ret_type = RET_INTEGER,
2578 .arg1_type = ARG_PTR_TO_CTX,
2579 .arg2_type = ARG_ANYTHING,
2582 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2584 msg->cork_bytes = bytes;
2588 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2589 .func = bpf_msg_cork_bytes,
2591 .ret_type = RET_INTEGER,
2592 .arg1_type = ARG_PTR_TO_CTX,
2593 .arg2_type = ARG_ANYTHING,
2596 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2597 u32, end, u64, flags)
2599 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2600 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2601 struct scatterlist *sge;
2602 u8 *raw, *to, *from;
2605 if (unlikely(flags || end <= start))
2608 /* First find the starting scatterlist element */
2612 len = sk_msg_elem(msg, i)->length;
2613 if (start < offset + len)
2615 sk_msg_iter_var_next(i);
2616 } while (i != msg->sg.end);
2618 if (unlikely(start >= offset + len))
2622 /* The start may point into the sg element so we need to also
2623 * account for the headroom.
2625 bytes_sg_total = start - offset + bytes;
2626 if (!test_bit(i, &msg->sg.copy) && bytes_sg_total <= len)
2629 /* At this point we need to linearize multiple scatterlist
2630 * elements or a single shared page. Either way we need to
2631 * copy into a linear buffer exclusively owned by BPF. Then
2632 * place the buffer in the scatterlist and fixup the original
2633 * entries by removing the entries now in the linear buffer
2634 * and shifting the remaining entries. For now we do not try
2635 * to copy partial entries to avoid complexity of running out
2636 * of sg_entry slots. The downside is reading a single byte
2637 * will copy the entire sg entry.
2640 copy += sk_msg_elem(msg, i)->length;
2641 sk_msg_iter_var_next(i);
2642 if (bytes_sg_total <= copy)
2644 } while (i != msg->sg.end);
2647 if (unlikely(bytes_sg_total > copy))
2650 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2652 if (unlikely(!page))
2655 raw = page_address(page);
2658 sge = sk_msg_elem(msg, i);
2659 from = sg_virt(sge);
2663 memcpy(to, from, len);
2666 put_page(sg_page(sge));
2668 sk_msg_iter_var_next(i);
2669 } while (i != last_sge);
2671 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2673 /* To repair sg ring we need to shift entries. If we only
2674 * had a single entry though we can just replace it and
2675 * be done. Otherwise walk the ring and shift the entries.
2677 WARN_ON_ONCE(last_sge == first_sge);
2678 shift = last_sge > first_sge ?
2679 last_sge - first_sge - 1 :
2680 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2685 sk_msg_iter_var_next(i);
2689 if (i + shift >= NR_MSG_FRAG_IDS)
2690 move_from = i + shift - NR_MSG_FRAG_IDS;
2692 move_from = i + shift;
2693 if (move_from == msg->sg.end)
2696 msg->sg.data[i] = msg->sg.data[move_from];
2697 msg->sg.data[move_from].length = 0;
2698 msg->sg.data[move_from].page_link = 0;
2699 msg->sg.data[move_from].offset = 0;
2700 sk_msg_iter_var_next(i);
2703 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2704 msg->sg.end - shift + NR_MSG_FRAG_IDS :
2705 msg->sg.end - shift;
2707 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2708 msg->data_end = msg->data + bytes;
2712 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2713 .func = bpf_msg_pull_data,
2715 .ret_type = RET_INTEGER,
2716 .arg1_type = ARG_PTR_TO_CTX,
2717 .arg2_type = ARG_ANYTHING,
2718 .arg3_type = ARG_ANYTHING,
2719 .arg4_type = ARG_ANYTHING,
2722 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2723 u32, len, u64, flags)
2725 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2726 u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2727 u8 *raw, *to, *from;
2730 if (unlikely(flags))
2733 /* First find the starting scatterlist element */
2737 l = sk_msg_elem(msg, i)->length;
2739 if (start < offset + l)
2741 sk_msg_iter_var_next(i);
2742 } while (i != msg->sg.end);
2744 if (start >= offset + l)
2747 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2749 /* If no space available will fallback to copy, we need at
2750 * least one scatterlist elem available to push data into
2751 * when start aligns to the beginning of an element or two
2752 * when it falls inside an element. We handle the start equals
2753 * offset case because its the common case for inserting a
2756 if (!space || (space == 1 && start != offset))
2757 copy = msg->sg.data[i].length;
2759 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2760 get_order(copy + len));
2761 if (unlikely(!page))
2767 raw = page_address(page);
2769 psge = sk_msg_elem(msg, i);
2770 front = start - offset;
2771 back = psge->length - front;
2772 from = sg_virt(psge);
2775 memcpy(raw, from, front);
2779 to = raw + front + len;
2781 memcpy(to, from, back);
2784 put_page(sg_page(psge));
2785 } else if (start - offset) {
2786 psge = sk_msg_elem(msg, i);
2787 rsge = sk_msg_elem_cpy(msg, i);
2789 psge->length = start - offset;
2790 rsge.length -= psge->length;
2791 rsge.offset += start;
2793 sk_msg_iter_var_next(i);
2794 sg_unmark_end(psge);
2795 sg_unmark_end(&rsge);
2796 sk_msg_iter_next(msg, end);
2799 /* Slot(s) to place newly allocated data */
2802 /* Shift one or two slots as needed */
2804 sge = sk_msg_elem_cpy(msg, i);
2806 sk_msg_iter_var_next(i);
2807 sg_unmark_end(&sge);
2808 sk_msg_iter_next(msg, end);
2810 nsge = sk_msg_elem_cpy(msg, i);
2812 sk_msg_iter_var_next(i);
2813 nnsge = sk_msg_elem_cpy(msg, i);
2816 while (i != msg->sg.end) {
2817 msg->sg.data[i] = sge;
2819 sk_msg_iter_var_next(i);
2822 nnsge = sk_msg_elem_cpy(msg, i);
2824 nsge = sk_msg_elem_cpy(msg, i);
2829 /* Place newly allocated data buffer */
2830 sk_mem_charge(msg->sk, len);
2831 msg->sg.size += len;
2832 __clear_bit(new, &msg->sg.copy);
2833 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2835 get_page(sg_page(&rsge));
2836 sk_msg_iter_var_next(new);
2837 msg->sg.data[new] = rsge;
2840 sk_msg_compute_data_pointers(msg);
2844 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2845 .func = bpf_msg_push_data,
2847 .ret_type = RET_INTEGER,
2848 .arg1_type = ARG_PTR_TO_CTX,
2849 .arg2_type = ARG_ANYTHING,
2850 .arg3_type = ARG_ANYTHING,
2851 .arg4_type = ARG_ANYTHING,
2854 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2860 sk_msg_iter_var_next(i);
2861 msg->sg.data[prev] = msg->sg.data[i];
2862 } while (i != msg->sg.end);
2864 sk_msg_iter_prev(msg, end);
2867 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2869 struct scatterlist tmp, sge;
2871 sk_msg_iter_next(msg, end);
2872 sge = sk_msg_elem_cpy(msg, i);
2873 sk_msg_iter_var_next(i);
2874 tmp = sk_msg_elem_cpy(msg, i);
2876 while (i != msg->sg.end) {
2877 msg->sg.data[i] = sge;
2878 sk_msg_iter_var_next(i);
2880 tmp = sk_msg_elem_cpy(msg, i);
2884 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2885 u32, len, u64, flags)
2887 u32 i = 0, l = 0, space, offset = 0;
2888 u64 last = start + len;
2891 if (unlikely(flags))
2894 /* First find the starting scatterlist element */
2898 l = sk_msg_elem(msg, i)->length;
2900 if (start < offset + l)
2902 sk_msg_iter_var_next(i);
2903 } while (i != msg->sg.end);
2905 /* Bounds checks: start and pop must be inside message */
2906 if (start >= offset + l || last >= msg->sg.size)
2909 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2912 /* --------------| offset
2913 * -| start |-------- len -------|
2915 * |----- a ----|-------- pop -------|----- b ----|
2916 * |______________________________________________| length
2919 * a: region at front of scatter element to save
2920 * b: region at back of scatter element to save when length > A + pop
2921 * pop: region to pop from element, same as input 'pop' here will be
2922 * decremented below per iteration.
2924 * Two top-level cases to handle when start != offset, first B is non
2925 * zero and second B is zero corresponding to when a pop includes more
2928 * Then if B is non-zero AND there is no space allocate space and
2929 * compact A, B regions into page. If there is space shift ring to
2930 * the rigth free'ing the next element in ring to place B, leaving
2931 * A untouched except to reduce length.
2933 if (start != offset) {
2934 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2936 int b = sge->length - pop - a;
2938 sk_msg_iter_var_next(i);
2940 if (pop < sge->length - a) {
2943 sk_msg_shift_right(msg, i);
2944 nsge = sk_msg_elem(msg, i);
2945 get_page(sg_page(sge));
2948 b, sge->offset + pop + a);
2950 struct page *page, *orig;
2953 page = alloc_pages(__GFP_NOWARN |
2954 __GFP_COMP | GFP_ATOMIC,
2956 if (unlikely(!page))
2960 orig = sg_page(sge);
2961 from = sg_virt(sge);
2962 to = page_address(page);
2963 memcpy(to, from, a);
2964 memcpy(to + a, from + a + pop, b);
2965 sg_set_page(sge, page, a + b, 0);
2969 } else if (pop >= sge->length - a) {
2970 pop -= (sge->length - a);
2975 /* From above the current layout _must_ be as follows,
2980 * |---- pop ---|---------------- b ------------|
2981 * |____________________________________________| length
2983 * Offset and start of the current msg elem are equal because in the
2984 * previous case we handled offset != start and either consumed the
2985 * entire element and advanced to the next element OR pop == 0.
2987 * Two cases to handle here are first pop is less than the length
2988 * leaving some remainder b above. Simply adjust the element's layout
2989 * in this case. Or pop >= length of the element so that b = 0. In this
2990 * case advance to next element decrementing pop.
2993 struct scatterlist *sge = sk_msg_elem(msg, i);
2995 if (pop < sge->length) {
3001 sk_msg_shift_left(msg, i);
3003 sk_msg_iter_var_next(i);
3006 sk_mem_uncharge(msg->sk, len - pop);
3007 msg->sg.size -= (len - pop);
3008 sk_msg_compute_data_pointers(msg);
3012 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
3013 .func = bpf_msg_pop_data,
3015 .ret_type = RET_INTEGER,
3016 .arg1_type = ARG_PTR_TO_CTX,
3017 .arg2_type = ARG_ANYTHING,
3018 .arg3_type = ARG_ANYTHING,
3019 .arg4_type = ARG_ANYTHING,
3022 #ifdef CONFIG_CGROUP_NET_CLASSID
3023 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3025 return __task_get_classid(current);
3028 static const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3029 .func = bpf_get_cgroup_classid_curr,
3031 .ret_type = RET_INTEGER,
3034 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3036 struct sock *sk = skb_to_full_sk(skb);
3038 if (!sk || !sk_fullsock(sk))
3041 return sock_cgroup_classid(&sk->sk_cgrp_data);
3044 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3045 .func = bpf_skb_cgroup_classid,
3047 .ret_type = RET_INTEGER,
3048 .arg1_type = ARG_PTR_TO_CTX,
3052 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3054 return task_get_classid(skb);
3057 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3058 .func = bpf_get_cgroup_classid,
3060 .ret_type = RET_INTEGER,
3061 .arg1_type = ARG_PTR_TO_CTX,
3064 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3066 return dst_tclassid(skb);
3069 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3070 .func = bpf_get_route_realm,
3072 .ret_type = RET_INTEGER,
3073 .arg1_type = ARG_PTR_TO_CTX,
3076 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3078 /* If skb_clear_hash() was called due to mangling, we can
3079 * trigger SW recalculation here. Later access to hash
3080 * can then use the inline skb->hash via context directly
3081 * instead of calling this helper again.
3083 return skb_get_hash(skb);
3086 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3087 .func = bpf_get_hash_recalc,
3089 .ret_type = RET_INTEGER,
3090 .arg1_type = ARG_PTR_TO_CTX,
3093 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3095 /* After all direct packet write, this can be used once for
3096 * triggering a lazy recalc on next skb_get_hash() invocation.
3098 skb_clear_hash(skb);
3102 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3103 .func = bpf_set_hash_invalid,
3105 .ret_type = RET_INTEGER,
3106 .arg1_type = ARG_PTR_TO_CTX,
3109 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3111 /* Set user specified hash as L4(+), so that it gets returned
3112 * on skb_get_hash() call unless BPF prog later on triggers a
3115 __skb_set_sw_hash(skb, hash, true);
3119 static const struct bpf_func_proto bpf_set_hash_proto = {
3120 .func = bpf_set_hash,
3122 .ret_type = RET_INTEGER,
3123 .arg1_type = ARG_PTR_TO_CTX,
3124 .arg2_type = ARG_ANYTHING,
3127 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3132 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3133 vlan_proto != htons(ETH_P_8021AD)))
3134 vlan_proto = htons(ETH_P_8021Q);
3136 bpf_push_mac_rcsum(skb);
3137 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3138 bpf_pull_mac_rcsum(skb);
3140 bpf_compute_data_pointers(skb);
3144 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3145 .func = bpf_skb_vlan_push,
3147 .ret_type = RET_INTEGER,
3148 .arg1_type = ARG_PTR_TO_CTX,
3149 .arg2_type = ARG_ANYTHING,
3150 .arg3_type = ARG_ANYTHING,
3153 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3157 bpf_push_mac_rcsum(skb);
3158 ret = skb_vlan_pop(skb);
3159 bpf_pull_mac_rcsum(skb);
3161 bpf_compute_data_pointers(skb);
3165 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3166 .func = bpf_skb_vlan_pop,
3168 .ret_type = RET_INTEGER,
3169 .arg1_type = ARG_PTR_TO_CTX,
3172 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3174 /* Caller already did skb_cow() with len as headroom,
3175 * so no need to do it here.
3178 memmove(skb->data, skb->data + len, off);
3179 memset(skb->data + off, 0, len);
3181 /* No skb_postpush_rcsum(skb, skb->data + off, len)
3182 * needed here as it does not change the skb->csum
3183 * result for checksum complete when summing over
3189 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3191 /* skb_ensure_writable() is not needed here, as we're
3192 * already working on an uncloned skb.
3194 if (unlikely(!pskb_may_pull(skb, off + len)))
3197 skb_postpull_rcsum(skb, skb->data + off, len);
3198 memmove(skb->data + len, skb->data, off);
3199 __skb_pull(skb, len);
3204 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3206 bool trans_same = skb->transport_header == skb->network_header;
3209 /* There's no need for __skb_push()/__skb_pull() pair to
3210 * get to the start of the mac header as we're guaranteed
3211 * to always start from here under eBPF.
3213 ret = bpf_skb_generic_push(skb, off, len);
3215 skb->mac_header -= len;
3216 skb->network_header -= len;
3218 skb->transport_header = skb->network_header;
3224 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3226 bool trans_same = skb->transport_header == skb->network_header;
3229 /* Same here, __skb_push()/__skb_pull() pair not needed. */
3230 ret = bpf_skb_generic_pop(skb, off, len);
3232 skb->mac_header += len;
3233 skb->network_header += len;
3235 skb->transport_header = skb->network_header;
3241 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3243 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3244 u32 off = skb_mac_header_len(skb);
3247 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
3250 ret = skb_cow(skb, len_diff);
3251 if (unlikely(ret < 0))
3254 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3255 if (unlikely(ret < 0))
3258 if (skb_is_gso(skb)) {
3259 struct skb_shared_info *shinfo = skb_shinfo(skb);
3261 /* SKB_GSO_TCPV4 needs to be changed into
3264 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3265 shinfo->gso_type &= ~SKB_GSO_TCPV4;
3266 shinfo->gso_type |= SKB_GSO_TCPV6;
3269 /* Due to IPv6 header, MSS needs to be downgraded. */
3270 skb_decrease_gso_size(shinfo, len_diff);
3271 /* Header must be checked, and gso_segs recomputed. */
3272 shinfo->gso_type |= SKB_GSO_DODGY;
3273 shinfo->gso_segs = 0;
3276 skb->protocol = htons(ETH_P_IPV6);
3277 skb_clear_hash(skb);
3282 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3284 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3285 u32 off = skb_mac_header_len(skb);
3288 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
3291 ret = skb_unclone(skb, GFP_ATOMIC);
3292 if (unlikely(ret < 0))
3295 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3296 if (unlikely(ret < 0))
3299 if (skb_is_gso(skb)) {
3300 struct skb_shared_info *shinfo = skb_shinfo(skb);
3302 /* SKB_GSO_TCPV6 needs to be changed into
3305 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3306 shinfo->gso_type &= ~SKB_GSO_TCPV6;
3307 shinfo->gso_type |= SKB_GSO_TCPV4;
3310 /* Due to IPv4 header, MSS can be upgraded. */
3311 skb_increase_gso_size(shinfo, len_diff);
3312 /* Header must be checked, and gso_segs recomputed. */
3313 shinfo->gso_type |= SKB_GSO_DODGY;
3314 shinfo->gso_segs = 0;
3317 skb->protocol = htons(ETH_P_IP);
3318 skb_clear_hash(skb);
3323 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3325 __be16 from_proto = skb->protocol;
3327 if (from_proto == htons(ETH_P_IP) &&
3328 to_proto == htons(ETH_P_IPV6))
3329 return bpf_skb_proto_4_to_6(skb);
3331 if (from_proto == htons(ETH_P_IPV6) &&
3332 to_proto == htons(ETH_P_IP))
3333 return bpf_skb_proto_6_to_4(skb);
3338 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3343 if (unlikely(flags))
3346 /* General idea is that this helper does the basic groundwork
3347 * needed for changing the protocol, and eBPF program fills the
3348 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3349 * and other helpers, rather than passing a raw buffer here.
3351 * The rationale is to keep this minimal and without a need to
3352 * deal with raw packet data. F.e. even if we would pass buffers
3353 * here, the program still needs to call the bpf_lX_csum_replace()
3354 * helpers anyway. Plus, this way we keep also separation of
3355 * concerns, since f.e. bpf_skb_store_bytes() should only take
3358 * Currently, additional options and extension header space are
3359 * not supported, but flags register is reserved so we can adapt
3360 * that. For offloads, we mark packet as dodgy, so that headers
3361 * need to be verified first.
3363 ret = bpf_skb_proto_xlat(skb, proto);
3364 bpf_compute_data_pointers(skb);
3368 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3369 .func = bpf_skb_change_proto,
3371 .ret_type = RET_INTEGER,
3372 .arg1_type = ARG_PTR_TO_CTX,
3373 .arg2_type = ARG_ANYTHING,
3374 .arg3_type = ARG_ANYTHING,
3377 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3379 /* We only allow a restricted subset to be changed for now. */
3380 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3381 !skb_pkt_type_ok(pkt_type)))
3384 skb->pkt_type = pkt_type;
3388 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3389 .func = bpf_skb_change_type,
3391 .ret_type = RET_INTEGER,
3392 .arg1_type = ARG_PTR_TO_CTX,
3393 .arg2_type = ARG_ANYTHING,
3396 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3398 switch (skb->protocol) {
3399 case htons(ETH_P_IP):
3400 return sizeof(struct iphdr);
3401 case htons(ETH_P_IPV6):
3402 return sizeof(struct ipv6hdr);
3408 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3409 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3411 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \
3412 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3413 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3414 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3415 BPF_F_ADJ_ROOM_ENCAP_L2( \
3416 BPF_ADJ_ROOM_ENCAP_L2_MASK))
3418 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3421 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3422 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3423 u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3424 unsigned int gso_type = SKB_GSO_DODGY;
3427 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3428 /* udp gso_size delineates datagrams, only allow if fixed */
3429 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3430 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3434 ret = skb_cow_head(skb, len_diff);
3435 if (unlikely(ret < 0))
3439 if (skb->protocol != htons(ETH_P_IP) &&
3440 skb->protocol != htons(ETH_P_IPV6))
3443 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3444 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3447 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3448 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3451 if (skb->encapsulation)
3454 mac_len = skb->network_header - skb->mac_header;
3455 inner_net = skb->network_header;
3456 if (inner_mac_len > len_diff)
3458 inner_trans = skb->transport_header;
3461 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3462 if (unlikely(ret < 0))
3466 skb->inner_mac_header = inner_net - inner_mac_len;
3467 skb->inner_network_header = inner_net;
3468 skb->inner_transport_header = inner_trans;
3469 skb_set_inner_protocol(skb, skb->protocol);
3471 skb->encapsulation = 1;
3472 skb_set_network_header(skb, mac_len);
3474 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3475 gso_type |= SKB_GSO_UDP_TUNNEL;
3476 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3477 gso_type |= SKB_GSO_GRE;
3478 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3479 gso_type |= SKB_GSO_IPXIP6;
3480 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3481 gso_type |= SKB_GSO_IPXIP4;
3483 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3484 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3485 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3486 sizeof(struct ipv6hdr) :
3487 sizeof(struct iphdr);
3489 skb_set_transport_header(skb, mac_len + nh_len);
3492 /* Match skb->protocol to new outer l3 protocol */
3493 if (skb->protocol == htons(ETH_P_IP) &&
3494 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3495 skb->protocol = htons(ETH_P_IPV6);
3496 else if (skb->protocol == htons(ETH_P_IPV6) &&
3497 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3498 skb->protocol = htons(ETH_P_IP);
3501 if (skb_is_gso(skb)) {
3502 struct skb_shared_info *shinfo = skb_shinfo(skb);
3504 /* Due to header grow, MSS needs to be downgraded. */
3505 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3506 skb_decrease_gso_size(shinfo, len_diff);
3508 /* Header must be checked, and gso_segs recomputed. */
3509 shinfo->gso_type |= gso_type;
3510 shinfo->gso_segs = 0;
3516 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3521 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3522 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3525 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3526 /* udp gso_size delineates datagrams, only allow if fixed */
3527 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3528 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3532 ret = skb_unclone(skb, GFP_ATOMIC);
3533 if (unlikely(ret < 0))
3536 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3537 if (unlikely(ret < 0))
3540 if (skb_is_gso(skb)) {
3541 struct skb_shared_info *shinfo = skb_shinfo(skb);
3543 /* Due to header shrink, MSS can be upgraded. */
3544 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3545 skb_increase_gso_size(shinfo, len_diff);
3547 /* Header must be checked, and gso_segs recomputed. */
3548 shinfo->gso_type |= SKB_GSO_DODGY;
3549 shinfo->gso_segs = 0;
3555 static u32 __bpf_skb_max_len(const struct sk_buff *skb)
3557 return skb->dev ? skb->dev->mtu + skb->dev->hard_header_len :
3561 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3562 u32, mode, u64, flags)
3564 u32 len_diff_abs = abs(len_diff);
3565 bool shrink = len_diff < 0;
3568 if (unlikely(flags || mode))
3570 if (unlikely(len_diff_abs > 0xfffU))
3574 ret = skb_cow(skb, len_diff);
3575 if (unlikely(ret < 0))
3577 __skb_push(skb, len_diff_abs);
3578 memset(skb->data, 0, len_diff_abs);
3580 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3582 __skb_pull(skb, len_diff_abs);
3584 bpf_compute_data_end_sk_skb(skb);
3585 if (tls_sw_has_ctx_rx(skb->sk)) {
3586 struct strp_msg *rxm = strp_msg(skb);
3588 rxm->full_len += len_diff;
3593 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3594 .func = sk_skb_adjust_room,
3596 .ret_type = RET_INTEGER,
3597 .arg1_type = ARG_PTR_TO_CTX,
3598 .arg2_type = ARG_ANYTHING,
3599 .arg3_type = ARG_ANYTHING,
3600 .arg4_type = ARG_ANYTHING,
3603 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3604 u32, mode, u64, flags)
3606 u32 len_cur, len_diff_abs = abs(len_diff);
3607 u32 len_min = bpf_skb_net_base_len(skb);
3608 u32 len_max = __bpf_skb_max_len(skb);
3609 __be16 proto = skb->protocol;
3610 bool shrink = len_diff < 0;
3614 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3615 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3617 if (unlikely(len_diff_abs > 0xfffU))
3619 if (unlikely(proto != htons(ETH_P_IP) &&
3620 proto != htons(ETH_P_IPV6)))
3623 off = skb_mac_header_len(skb);
3625 case BPF_ADJ_ROOM_NET:
3626 off += bpf_skb_net_base_len(skb);
3628 case BPF_ADJ_ROOM_MAC:
3634 len_cur = skb->len - skb_network_offset(skb);
3635 if ((shrink && (len_diff_abs >= len_cur ||
3636 len_cur - len_diff_abs < len_min)) ||
3637 (!shrink && (skb->len + len_diff_abs > len_max &&
3641 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3642 bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3643 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3644 __skb_reset_checksum_unnecessary(skb);
3646 bpf_compute_data_pointers(skb);
3650 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3651 .func = bpf_skb_adjust_room,
3653 .ret_type = RET_INTEGER,
3654 .arg1_type = ARG_PTR_TO_CTX,
3655 .arg2_type = ARG_ANYTHING,
3656 .arg3_type = ARG_ANYTHING,
3657 .arg4_type = ARG_ANYTHING,
3660 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3662 u32 min_len = skb_network_offset(skb);
3664 if (skb_transport_header_was_set(skb))
3665 min_len = skb_transport_offset(skb);
3666 if (skb->ip_summed == CHECKSUM_PARTIAL)
3667 min_len = skb_checksum_start_offset(skb) +
3668 skb->csum_offset + sizeof(__sum16);
3672 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3674 unsigned int old_len = skb->len;
3677 ret = __skb_grow_rcsum(skb, new_len);
3679 memset(skb->data + old_len, 0, new_len - old_len);
3683 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3685 return __skb_trim_rcsum(skb, new_len);
3688 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3691 u32 max_len = __bpf_skb_max_len(skb);
3692 u32 min_len = __bpf_skb_min_len(skb);
3695 if (unlikely(flags || new_len > max_len || new_len < min_len))
3697 if (skb->encapsulation)
3700 /* The basic idea of this helper is that it's performing the
3701 * needed work to either grow or trim an skb, and eBPF program
3702 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3703 * bpf_lX_csum_replace() and others rather than passing a raw
3704 * buffer here. This one is a slow path helper and intended
3705 * for replies with control messages.
3707 * Like in bpf_skb_change_proto(), we want to keep this rather
3708 * minimal and without protocol specifics so that we are able
3709 * to separate concerns as in bpf_skb_store_bytes() should only
3710 * be the one responsible for writing buffers.
3712 * It's really expected to be a slow path operation here for
3713 * control message replies, so we're implicitly linearizing,
3714 * uncloning and drop offloads from the skb by this.
3716 ret = __bpf_try_make_writable(skb, skb->len);
3718 if (new_len > skb->len)
3719 ret = bpf_skb_grow_rcsum(skb, new_len);
3720 else if (new_len < skb->len)
3721 ret = bpf_skb_trim_rcsum(skb, new_len);
3722 if (!ret && skb_is_gso(skb))
3728 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3731 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3733 bpf_compute_data_pointers(skb);
3737 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3738 .func = bpf_skb_change_tail,
3740 .ret_type = RET_INTEGER,
3741 .arg1_type = ARG_PTR_TO_CTX,
3742 .arg2_type = ARG_ANYTHING,
3743 .arg3_type = ARG_ANYTHING,
3746 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3749 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3751 bpf_compute_data_end_sk_skb(skb);
3755 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3756 .func = sk_skb_change_tail,
3758 .ret_type = RET_INTEGER,
3759 .arg1_type = ARG_PTR_TO_CTX,
3760 .arg2_type = ARG_ANYTHING,
3761 .arg3_type = ARG_ANYTHING,
3764 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3767 u32 max_len = __bpf_skb_max_len(skb);
3768 u32 new_len = skb->len + head_room;
3771 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3772 new_len < skb->len))
3775 ret = skb_cow(skb, head_room);
3777 /* Idea for this helper is that we currently only
3778 * allow to expand on mac header. This means that
3779 * skb->protocol network header, etc, stay as is.
3780 * Compared to bpf_skb_change_tail(), we're more
3781 * flexible due to not needing to linearize or
3782 * reset GSO. Intention for this helper is to be
3783 * used by an L3 skb that needs to push mac header
3784 * for redirection into L2 device.
3786 __skb_push(skb, head_room);
3787 memset(skb->data, 0, head_room);
3788 skb_reset_mac_header(skb);
3794 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3797 int ret = __bpf_skb_change_head(skb, head_room, flags);
3799 bpf_compute_data_pointers(skb);
3803 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3804 .func = bpf_skb_change_head,
3806 .ret_type = RET_INTEGER,
3807 .arg1_type = ARG_PTR_TO_CTX,
3808 .arg2_type = ARG_ANYTHING,
3809 .arg3_type = ARG_ANYTHING,
3812 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3815 int ret = __bpf_skb_change_head(skb, head_room, flags);
3817 bpf_compute_data_end_sk_skb(skb);
3821 static const struct bpf_func_proto sk_skb_change_head_proto = {
3822 .func = sk_skb_change_head,
3824 .ret_type = RET_INTEGER,
3825 .arg1_type = ARG_PTR_TO_CTX,
3826 .arg2_type = ARG_ANYTHING,
3827 .arg3_type = ARG_ANYTHING,
3829 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3831 return xdp_data_meta_unsupported(xdp) ? 0 :
3832 xdp->data - xdp->data_meta;
3835 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3837 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3838 unsigned long metalen = xdp_get_metalen(xdp);
3839 void *data_start = xdp_frame_end + metalen;
3840 void *data = xdp->data + offset;
3842 if (unlikely(data < data_start ||
3843 data > xdp->data_end - ETH_HLEN))
3847 memmove(xdp->data_meta + offset,
3848 xdp->data_meta, metalen);
3849 xdp->data_meta += offset;
3855 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3856 .func = bpf_xdp_adjust_head,
3858 .ret_type = RET_INTEGER,
3859 .arg1_type = ARG_PTR_TO_CTX,
3860 .arg2_type = ARG_ANYTHING,
3863 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3865 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
3866 void *data_end = xdp->data_end + offset;
3868 /* Notice that xdp_data_hard_end have reserved some tailroom */
3869 if (unlikely(data_end > data_hard_end))
3872 /* ALL drivers MUST init xdp->frame_sz, chicken check below */
3873 if (unlikely(xdp->frame_sz > PAGE_SIZE)) {
3874 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz);
3878 if (unlikely(data_end < xdp->data + ETH_HLEN))
3881 /* Clear memory area on grow, can contain uninit kernel memory */
3883 memset(xdp->data_end, 0, offset);
3885 xdp->data_end = data_end;
3890 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3891 .func = bpf_xdp_adjust_tail,
3893 .ret_type = RET_INTEGER,
3894 .arg1_type = ARG_PTR_TO_CTX,
3895 .arg2_type = ARG_ANYTHING,
3898 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3900 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3901 void *meta = xdp->data_meta + offset;
3902 unsigned long metalen = xdp->data - meta;
3904 if (xdp_data_meta_unsupported(xdp))
3906 if (unlikely(meta < xdp_frame_end ||
3909 if (unlikely((metalen & (sizeof(__u32) - 1)) ||
3913 xdp->data_meta = meta;
3918 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3919 .func = bpf_xdp_adjust_meta,
3921 .ret_type = RET_INTEGER,
3922 .arg1_type = ARG_PTR_TO_CTX,
3923 .arg2_type = ARG_ANYTHING,
3926 static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd,
3927 struct bpf_map *map, struct xdp_buff *xdp)
3929 switch (map->map_type) {
3930 case BPF_MAP_TYPE_DEVMAP:
3931 case BPF_MAP_TYPE_DEVMAP_HASH:
3932 return dev_map_enqueue(fwd, xdp, dev_rx);
3933 case BPF_MAP_TYPE_CPUMAP:
3934 return cpu_map_enqueue(fwd, xdp, dev_rx);
3935 case BPF_MAP_TYPE_XSKMAP:
3936 return __xsk_map_redirect(fwd, xdp);
3943 void xdp_do_flush(void)
3949 EXPORT_SYMBOL_GPL(xdp_do_flush);
3951 static inline void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index)
3953 switch (map->map_type) {
3954 case BPF_MAP_TYPE_DEVMAP:
3955 return __dev_map_lookup_elem(map, index);
3956 case BPF_MAP_TYPE_DEVMAP_HASH:
3957 return __dev_map_hash_lookup_elem(map, index);
3958 case BPF_MAP_TYPE_CPUMAP:
3959 return __cpu_map_lookup_elem(map, index);
3960 case BPF_MAP_TYPE_XSKMAP:
3961 return __xsk_map_lookup_elem(map, index);
3967 void bpf_clear_redirect_map(struct bpf_map *map)
3969 struct bpf_redirect_info *ri;
3972 for_each_possible_cpu(cpu) {
3973 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3974 /* Avoid polluting remote cacheline due to writes if
3975 * not needed. Once we pass this test, we need the
3976 * cmpxchg() to make sure it hasn't been changed in
3977 * the meantime by remote CPU.
3979 if (unlikely(READ_ONCE(ri->map) == map))
3980 cmpxchg(&ri->map, map, NULL);
3984 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3985 struct bpf_prog *xdp_prog)
3987 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3988 struct bpf_map *map = READ_ONCE(ri->map);
3989 u32 index = ri->tgt_index;
3990 void *fwd = ri->tgt_value;
3994 ri->tgt_value = NULL;
3995 WRITE_ONCE(ri->map, NULL);
3997 if (unlikely(!map)) {
3998 fwd = dev_get_by_index_rcu(dev_net(dev), index);
3999 if (unlikely(!fwd)) {
4004 err = dev_xdp_enqueue(fwd, xdp, dev);
4006 err = __bpf_tx_xdp_map(dev, fwd, map, xdp);
4012 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
4015 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
4018 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4020 static int xdp_do_generic_redirect_map(struct net_device *dev,
4021 struct sk_buff *skb,
4022 struct xdp_buff *xdp,
4023 struct bpf_prog *xdp_prog,
4024 struct bpf_map *map)
4026 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4027 u32 index = ri->tgt_index;
4028 void *fwd = ri->tgt_value;
4032 ri->tgt_value = NULL;
4033 WRITE_ONCE(ri->map, NULL);
4035 if (map->map_type == BPF_MAP_TYPE_DEVMAP ||
4036 map->map_type == BPF_MAP_TYPE_DEVMAP_HASH) {
4037 struct bpf_dtab_netdev *dst = fwd;
4039 err = dev_map_generic_redirect(dst, skb, xdp_prog);
4042 } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) {
4043 struct xdp_sock *xs = fwd;
4045 err = xsk_generic_rcv(xs, xdp);
4050 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
4055 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
4058 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
4062 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4063 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4065 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4066 struct bpf_map *map = READ_ONCE(ri->map);
4067 u32 index = ri->tgt_index;
4068 struct net_device *fwd;
4072 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog,
4075 fwd = dev_get_by_index_rcu(dev_net(dev), index);
4076 if (unlikely(!fwd)) {
4081 err = xdp_ok_fwd_dev(fwd, skb->len);
4086 _trace_xdp_redirect(dev, xdp_prog, index);
4087 generic_xdp_tx(skb, xdp_prog);
4090 _trace_xdp_redirect_err(dev, xdp_prog, index, err);
4094 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4096 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4098 if (unlikely(flags))
4102 ri->tgt_index = ifindex;
4103 ri->tgt_value = NULL;
4104 WRITE_ONCE(ri->map, NULL);
4106 return XDP_REDIRECT;
4109 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4110 .func = bpf_xdp_redirect,
4112 .ret_type = RET_INTEGER,
4113 .arg1_type = ARG_ANYTHING,
4114 .arg2_type = ARG_ANYTHING,
4117 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
4120 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4122 /* Lower bits of the flags are used as return code on lookup failure */
4123 if (unlikely(flags > XDP_TX))
4126 ri->tgt_value = __xdp_map_lookup_elem(map, ifindex);
4127 if (unlikely(!ri->tgt_value)) {
4128 /* If the lookup fails we want to clear out the state in the
4129 * redirect_info struct completely, so that if an eBPF program
4130 * performs multiple lookups, the last one always takes
4133 WRITE_ONCE(ri->map, NULL);
4138 ri->tgt_index = ifindex;
4139 WRITE_ONCE(ri->map, map);
4141 return XDP_REDIRECT;
4144 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4145 .func = bpf_xdp_redirect_map,
4147 .ret_type = RET_INTEGER,
4148 .arg1_type = ARG_CONST_MAP_PTR,
4149 .arg2_type = ARG_ANYTHING,
4150 .arg3_type = ARG_ANYTHING,
4153 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4154 unsigned long off, unsigned long len)
4156 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4160 if (ptr != dst_buff)
4161 memcpy(dst_buff, ptr, len);
4166 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4167 u64, flags, void *, meta, u64, meta_size)
4169 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4171 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4173 if (unlikely(!skb || skb_size > skb->len))
4176 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4180 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4181 .func = bpf_skb_event_output,
4183 .ret_type = RET_INTEGER,
4184 .arg1_type = ARG_PTR_TO_CTX,
4185 .arg2_type = ARG_CONST_MAP_PTR,
4186 .arg3_type = ARG_ANYTHING,
4187 .arg4_type = ARG_PTR_TO_MEM,
4188 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4191 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4193 const struct bpf_func_proto bpf_skb_output_proto = {
4194 .func = bpf_skb_event_output,
4196 .ret_type = RET_INTEGER,
4197 .arg1_type = ARG_PTR_TO_BTF_ID,
4198 .arg1_btf_id = &bpf_skb_output_btf_ids[0],
4199 .arg2_type = ARG_CONST_MAP_PTR,
4200 .arg3_type = ARG_ANYTHING,
4201 .arg4_type = ARG_PTR_TO_MEM,
4202 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4205 static unsigned short bpf_tunnel_key_af(u64 flags)
4207 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4210 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4211 u32, size, u64, flags)
4213 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4214 u8 compat[sizeof(struct bpf_tunnel_key)];
4218 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
4222 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4226 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4229 case offsetof(struct bpf_tunnel_key, tunnel_label):
4230 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4232 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4233 /* Fixup deprecated structure layouts here, so we have
4234 * a common path later on.
4236 if (ip_tunnel_info_af(info) != AF_INET)
4239 to = (struct bpf_tunnel_key *)compat;
4246 to->tunnel_id = be64_to_cpu(info->key.tun_id);
4247 to->tunnel_tos = info->key.tos;
4248 to->tunnel_ttl = info->key.ttl;
4251 if (flags & BPF_F_TUNINFO_IPV6) {
4252 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4253 sizeof(to->remote_ipv6));
4254 to->tunnel_label = be32_to_cpu(info->key.label);
4256 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4257 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4258 to->tunnel_label = 0;
4261 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4262 memcpy(to_orig, to, size);
4266 memset(to_orig, 0, size);
4270 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4271 .func = bpf_skb_get_tunnel_key,
4273 .ret_type = RET_INTEGER,
4274 .arg1_type = ARG_PTR_TO_CTX,
4275 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4276 .arg3_type = ARG_CONST_SIZE,
4277 .arg4_type = ARG_ANYTHING,
4280 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4282 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4285 if (unlikely(!info ||
4286 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4290 if (unlikely(size < info->options_len)) {
4295 ip_tunnel_info_opts_get(to, info);
4296 if (size > info->options_len)
4297 memset(to + info->options_len, 0, size - info->options_len);
4299 return info->options_len;
4301 memset(to, 0, size);
4305 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4306 .func = bpf_skb_get_tunnel_opt,
4308 .ret_type = RET_INTEGER,
4309 .arg1_type = ARG_PTR_TO_CTX,
4310 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4311 .arg3_type = ARG_CONST_SIZE,
4314 static struct metadata_dst __percpu *md_dst;
4316 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4317 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4319 struct metadata_dst *md = this_cpu_ptr(md_dst);
4320 u8 compat[sizeof(struct bpf_tunnel_key)];
4321 struct ip_tunnel_info *info;
4323 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4324 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
4326 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4328 case offsetof(struct bpf_tunnel_key, tunnel_label):
4329 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4330 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4331 /* Fixup deprecated structure layouts here, so we have
4332 * a common path later on.
4334 memcpy(compat, from, size);
4335 memset(compat + size, 0, sizeof(compat) - size);
4336 from = (const struct bpf_tunnel_key *) compat;
4342 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4347 dst_hold((struct dst_entry *) md);
4348 skb_dst_set(skb, (struct dst_entry *) md);
4350 info = &md->u.tun_info;
4351 memset(info, 0, sizeof(*info));
4352 info->mode = IP_TUNNEL_INFO_TX;
4354 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4355 if (flags & BPF_F_DONT_FRAGMENT)
4356 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4357 if (flags & BPF_F_ZERO_CSUM_TX)
4358 info->key.tun_flags &= ~TUNNEL_CSUM;
4359 if (flags & BPF_F_SEQ_NUMBER)
4360 info->key.tun_flags |= TUNNEL_SEQ;
4362 info->key.tun_id = cpu_to_be64(from->tunnel_id);
4363 info->key.tos = from->tunnel_tos;
4364 info->key.ttl = from->tunnel_ttl;
4366 if (flags & BPF_F_TUNINFO_IPV6) {
4367 info->mode |= IP_TUNNEL_INFO_IPV6;
4368 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4369 sizeof(from->remote_ipv6));
4370 info->key.label = cpu_to_be32(from->tunnel_label) &
4371 IPV6_FLOWLABEL_MASK;
4373 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4379 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4380 .func = bpf_skb_set_tunnel_key,
4382 .ret_type = RET_INTEGER,
4383 .arg1_type = ARG_PTR_TO_CTX,
4384 .arg2_type = ARG_PTR_TO_MEM,
4385 .arg3_type = ARG_CONST_SIZE,
4386 .arg4_type = ARG_ANYTHING,
4389 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4390 const u8 *, from, u32, size)
4392 struct ip_tunnel_info *info = skb_tunnel_info(skb);
4393 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4395 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4397 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4400 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4405 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4406 .func = bpf_skb_set_tunnel_opt,
4408 .ret_type = RET_INTEGER,
4409 .arg1_type = ARG_PTR_TO_CTX,
4410 .arg2_type = ARG_PTR_TO_MEM,
4411 .arg3_type = ARG_CONST_SIZE,
4414 static const struct bpf_func_proto *
4415 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4418 struct metadata_dst __percpu *tmp;
4420 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4425 if (cmpxchg(&md_dst, NULL, tmp))
4426 metadata_dst_free_percpu(tmp);
4430 case BPF_FUNC_skb_set_tunnel_key:
4431 return &bpf_skb_set_tunnel_key_proto;
4432 case BPF_FUNC_skb_set_tunnel_opt:
4433 return &bpf_skb_set_tunnel_opt_proto;
4439 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4442 struct bpf_array *array = container_of(map, struct bpf_array, map);
4443 struct cgroup *cgrp;
4446 sk = skb_to_full_sk(skb);
4447 if (!sk || !sk_fullsock(sk))
4449 if (unlikely(idx >= array->map.max_entries))
4452 cgrp = READ_ONCE(array->ptrs[idx]);
4453 if (unlikely(!cgrp))
4456 return sk_under_cgroup_hierarchy(sk, cgrp);
4459 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4460 .func = bpf_skb_under_cgroup,
4462 .ret_type = RET_INTEGER,
4463 .arg1_type = ARG_PTR_TO_CTX,
4464 .arg2_type = ARG_CONST_MAP_PTR,
4465 .arg3_type = ARG_ANYTHING,
4468 #ifdef CONFIG_SOCK_CGROUP_DATA
4469 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4471 struct cgroup *cgrp;
4473 sk = sk_to_full_sk(sk);
4474 if (!sk || !sk_fullsock(sk))
4477 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4478 return cgroup_id(cgrp);
4481 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4483 return __bpf_sk_cgroup_id(skb->sk);
4486 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4487 .func = bpf_skb_cgroup_id,
4489 .ret_type = RET_INTEGER,
4490 .arg1_type = ARG_PTR_TO_CTX,
4493 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4496 struct cgroup *ancestor;
4497 struct cgroup *cgrp;
4499 sk = sk_to_full_sk(sk);
4500 if (!sk || !sk_fullsock(sk))
4503 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4504 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4508 return cgroup_id(ancestor);
4511 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4514 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4517 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4518 .func = bpf_skb_ancestor_cgroup_id,
4520 .ret_type = RET_INTEGER,
4521 .arg1_type = ARG_PTR_TO_CTX,
4522 .arg2_type = ARG_ANYTHING,
4525 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4527 return __bpf_sk_cgroup_id(sk);
4530 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4531 .func = bpf_sk_cgroup_id,
4533 .ret_type = RET_INTEGER,
4534 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4537 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4539 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4542 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4543 .func = bpf_sk_ancestor_cgroup_id,
4545 .ret_type = RET_INTEGER,
4546 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4547 .arg2_type = ARG_ANYTHING,
4551 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4552 unsigned long off, unsigned long len)
4554 memcpy(dst_buff, src_buff + off, len);
4558 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4559 u64, flags, void *, meta, u64, meta_size)
4561 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4563 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4565 if (unlikely(!xdp ||
4566 xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4569 return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4570 xdp_size, bpf_xdp_copy);
4573 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4574 .func = bpf_xdp_event_output,
4576 .ret_type = RET_INTEGER,
4577 .arg1_type = ARG_PTR_TO_CTX,
4578 .arg2_type = ARG_CONST_MAP_PTR,
4579 .arg3_type = ARG_ANYTHING,
4580 .arg4_type = ARG_PTR_TO_MEM,
4581 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4584 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4586 const struct bpf_func_proto bpf_xdp_output_proto = {
4587 .func = bpf_xdp_event_output,
4589 .ret_type = RET_INTEGER,
4590 .arg1_type = ARG_PTR_TO_BTF_ID,
4591 .arg1_btf_id = &bpf_xdp_output_btf_ids[0],
4592 .arg2_type = ARG_CONST_MAP_PTR,
4593 .arg3_type = ARG_ANYTHING,
4594 .arg4_type = ARG_PTR_TO_MEM,
4595 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4598 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4600 return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4603 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4604 .func = bpf_get_socket_cookie,
4606 .ret_type = RET_INTEGER,
4607 .arg1_type = ARG_PTR_TO_CTX,
4610 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4612 return __sock_gen_cookie(ctx->sk);
4615 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4616 .func = bpf_get_socket_cookie_sock_addr,
4618 .ret_type = RET_INTEGER,
4619 .arg1_type = ARG_PTR_TO_CTX,
4622 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4624 return __sock_gen_cookie(ctx);
4627 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4628 .func = bpf_get_socket_cookie_sock,
4630 .ret_type = RET_INTEGER,
4631 .arg1_type = ARG_PTR_TO_CTX,
4634 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4636 return __sock_gen_cookie(ctx->sk);
4639 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4640 .func = bpf_get_socket_cookie_sock_ops,
4642 .ret_type = RET_INTEGER,
4643 .arg1_type = ARG_PTR_TO_CTX,
4646 static u64 __bpf_get_netns_cookie(struct sock *sk)
4648 #ifdef CONFIG_NET_NS
4649 return __net_gen_cookie(sk ? sk->sk_net.net : &init_net);
4655 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
4657 return __bpf_get_netns_cookie(ctx);
4660 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
4661 .func = bpf_get_netns_cookie_sock,
4663 .ret_type = RET_INTEGER,
4664 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4667 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4669 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4672 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
4673 .func = bpf_get_netns_cookie_sock_addr,
4675 .ret_type = RET_INTEGER,
4676 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4679 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4681 struct sock *sk = sk_to_full_sk(skb->sk);
4684 if (!sk || !sk_fullsock(sk))
4686 kuid = sock_net_uid(sock_net(sk), sk);
4687 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4690 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4691 .func = bpf_get_socket_uid,
4693 .ret_type = RET_INTEGER,
4694 .arg1_type = ARG_PTR_TO_CTX,
4697 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
4698 char *optval, int optlen)
4700 char devname[IFNAMSIZ];
4706 if (!sk_fullsock(sk))
4709 sock_owned_by_me(sk);
4711 if (level == SOL_SOCKET) {
4712 if (optlen != sizeof(int) && optname != SO_BINDTODEVICE)
4714 val = *((int *)optval);
4715 valbool = val ? 1 : 0;
4717 /* Only some socketops are supported */
4720 val = min_t(u32, val, sysctl_rmem_max);
4721 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4722 WRITE_ONCE(sk->sk_rcvbuf,
4723 max_t(int, val * 2, SOCK_MIN_RCVBUF));
4726 val = min_t(u32, val, sysctl_wmem_max);
4727 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4728 WRITE_ONCE(sk->sk_sndbuf,
4729 max_t(int, val * 2, SOCK_MIN_SNDBUF));
4731 case SO_MAX_PACING_RATE: /* 32bit version */
4733 cmpxchg(&sk->sk_pacing_status,
4736 sk->sk_max_pacing_rate = (val == ~0U) ?
4737 ~0UL : (unsigned int)val;
4738 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4739 sk->sk_max_pacing_rate);
4742 sk->sk_priority = val;
4747 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
4750 if (sk->sk_mark != val) {
4755 case SO_BINDTODEVICE:
4756 optlen = min_t(long, optlen, IFNAMSIZ - 1);
4757 strncpy(devname, optval, optlen);
4758 devname[optlen] = 0;
4761 if (devname[0] != '\0') {
4762 struct net_device *dev;
4767 dev = dev_get_by_name(net, devname);
4770 ifindex = dev->ifindex;
4774 case SO_BINDTOIFINDEX:
4775 if (optname == SO_BINDTOIFINDEX)
4777 ret = sock_bindtoindex(sk, ifindex, false);
4780 if (sk->sk_prot->keepalive)
4781 sk->sk_prot->keepalive(sk, valbool);
4782 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
4788 } else if (level == SOL_IP) {
4789 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4792 val = *((int *)optval);
4793 /* Only some options are supported */
4796 if (val < -1 || val > 0xff) {
4799 struct inet_sock *inet = inet_sk(sk);
4809 #if IS_ENABLED(CONFIG_IPV6)
4810 } else if (level == SOL_IPV6) {
4811 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4814 val = *((int *)optval);
4815 /* Only some options are supported */
4818 if (val < -1 || val > 0xff) {
4821 struct ipv6_pinfo *np = inet6_sk(sk);
4832 } else if (level == SOL_TCP &&
4833 sk->sk_prot->setsockopt == tcp_setsockopt) {
4834 if (optname == TCP_CONGESTION) {
4835 char name[TCP_CA_NAME_MAX];
4837 strncpy(name, optval, min_t(long, optlen,
4838 TCP_CA_NAME_MAX-1));
4839 name[TCP_CA_NAME_MAX-1] = 0;
4840 ret = tcp_set_congestion_control(sk, name, false, true);
4842 struct inet_connection_sock *icsk = inet_csk(sk);
4843 struct tcp_sock *tp = tcp_sk(sk);
4844 unsigned long timeout;
4846 if (optlen != sizeof(int))
4849 val = *((int *)optval);
4850 /* Only some options are supported */
4853 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4858 case TCP_BPF_SNDCWND_CLAMP:
4862 tp->snd_cwnd_clamp = val;
4863 tp->snd_ssthresh = val;
4866 case TCP_BPF_DELACK_MAX:
4867 timeout = usecs_to_jiffies(val);
4868 if (timeout > TCP_DELACK_MAX ||
4869 timeout < TCP_TIMEOUT_MIN)
4871 inet_csk(sk)->icsk_delack_max = timeout;
4873 case TCP_BPF_RTO_MIN:
4874 timeout = usecs_to_jiffies(val);
4875 if (timeout > TCP_RTO_MIN ||
4876 timeout < TCP_TIMEOUT_MIN)
4878 inet_csk(sk)->icsk_rto_min = timeout;
4881 if (val < 0 || val > 1)
4887 ret = tcp_sock_set_keepidle_locked(sk, val);
4890 if (val < 1 || val > MAX_TCP_KEEPINTVL)
4893 tp->keepalive_intvl = val * HZ;
4896 if (val < 1 || val > MAX_TCP_KEEPCNT)
4899 tp->keepalive_probes = val;
4902 if (val < 1 || val > MAX_TCP_SYNCNT)
4905 icsk->icsk_syn_retries = val;
4907 case TCP_USER_TIMEOUT:
4911 icsk->icsk_user_timeout = val;
4913 case TCP_NOTSENT_LOWAT:
4914 tp->notsent_lowat = val;
4915 sk->sk_write_space(sk);
4917 case TCP_WINDOW_CLAMP:
4918 ret = tcp_set_window_clamp(sk, val);
4931 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
4932 char *optval, int optlen)
4934 if (!sk_fullsock(sk))
4937 sock_owned_by_me(sk);
4939 if (level == SOL_SOCKET) {
4940 if (optlen != sizeof(int))
4945 *((int *)optval) = sk->sk_mark;
4948 *((int *)optval) = sk->sk_priority;
4950 case SO_BINDTOIFINDEX:
4951 *((int *)optval) = sk->sk_bound_dev_if;
4957 } else if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4958 struct inet_connection_sock *icsk;
4959 struct tcp_sock *tp;
4962 case TCP_CONGESTION:
4963 icsk = inet_csk(sk);
4965 if (!icsk->icsk_ca_ops || optlen <= 1)
4967 strncpy(optval, icsk->icsk_ca_ops->name, optlen);
4968 optval[optlen - 1] = 0;
4973 if (optlen <= 0 || !tp->saved_syn ||
4974 optlen > tcp_saved_syn_len(tp->saved_syn))
4976 memcpy(optval, tp->saved_syn->data, optlen);
4981 } else if (level == SOL_IP) {
4982 struct inet_sock *inet = inet_sk(sk);
4984 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4987 /* Only some options are supported */
4990 *((int *)optval) = (int)inet->tos;
4995 #if IS_ENABLED(CONFIG_IPV6)
4996 } else if (level == SOL_IPV6) {
4997 struct ipv6_pinfo *np = inet6_sk(sk);
4999 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
5002 /* Only some options are supported */
5005 *((int *)optval) = (int)np->tclass;
5017 memset(optval, 0, optlen);
5021 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5022 int, level, int, optname, char *, optval, int, optlen)
5024 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5027 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5028 .func = bpf_sock_addr_setsockopt,
5030 .ret_type = RET_INTEGER,
5031 .arg1_type = ARG_PTR_TO_CTX,
5032 .arg2_type = ARG_ANYTHING,
5033 .arg3_type = ARG_ANYTHING,
5034 .arg4_type = ARG_PTR_TO_MEM,
5035 .arg5_type = ARG_CONST_SIZE,
5038 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5039 int, level, int, optname, char *, optval, int, optlen)
5041 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5044 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5045 .func = bpf_sock_addr_getsockopt,
5047 .ret_type = RET_INTEGER,
5048 .arg1_type = ARG_PTR_TO_CTX,
5049 .arg2_type = ARG_ANYTHING,
5050 .arg3_type = ARG_ANYTHING,
5051 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5052 .arg5_type = ARG_CONST_SIZE,
5055 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5056 int, level, int, optname, char *, optval, int, optlen)
5058 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5061 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5062 .func = bpf_sock_ops_setsockopt,
5064 .ret_type = RET_INTEGER,
5065 .arg1_type = ARG_PTR_TO_CTX,
5066 .arg2_type = ARG_ANYTHING,
5067 .arg3_type = ARG_ANYTHING,
5068 .arg4_type = ARG_PTR_TO_MEM,
5069 .arg5_type = ARG_CONST_SIZE,
5072 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5073 int optname, const u8 **start)
5075 struct sk_buff *syn_skb = bpf_sock->syn_skb;
5076 const u8 *hdr_start;
5080 /* sk is a request_sock here */
5082 if (optname == TCP_BPF_SYN) {
5083 hdr_start = syn_skb->data;
5084 ret = tcp_hdrlen(syn_skb);
5085 } else if (optname == TCP_BPF_SYN_IP) {
5086 hdr_start = skb_network_header(syn_skb);
5087 ret = skb_network_header_len(syn_skb) +
5088 tcp_hdrlen(syn_skb);
5090 /* optname == TCP_BPF_SYN_MAC */
5091 hdr_start = skb_mac_header(syn_skb);
5092 ret = skb_mac_header_len(syn_skb) +
5093 skb_network_header_len(syn_skb) +
5094 tcp_hdrlen(syn_skb);
5097 struct sock *sk = bpf_sock->sk;
5098 struct saved_syn *saved_syn;
5100 if (sk->sk_state == TCP_NEW_SYN_RECV)
5101 /* synack retransmit. bpf_sock->syn_skb will
5102 * not be available. It has to resort to
5103 * saved_syn (if it is saved).
5105 saved_syn = inet_reqsk(sk)->saved_syn;
5107 saved_syn = tcp_sk(sk)->saved_syn;
5112 if (optname == TCP_BPF_SYN) {
5113 hdr_start = saved_syn->data +
5114 saved_syn->mac_hdrlen +
5115 saved_syn->network_hdrlen;
5116 ret = saved_syn->tcp_hdrlen;
5117 } else if (optname == TCP_BPF_SYN_IP) {
5118 hdr_start = saved_syn->data +
5119 saved_syn->mac_hdrlen;
5120 ret = saved_syn->network_hdrlen +
5121 saved_syn->tcp_hdrlen;
5123 /* optname == TCP_BPF_SYN_MAC */
5125 /* TCP_SAVE_SYN may not have saved the mac hdr */
5126 if (!saved_syn->mac_hdrlen)
5129 hdr_start = saved_syn->data;
5130 ret = saved_syn->mac_hdrlen +
5131 saved_syn->network_hdrlen +
5132 saved_syn->tcp_hdrlen;
5140 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5141 int, level, int, optname, char *, optval, int, optlen)
5143 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5144 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5145 int ret, copy_len = 0;
5148 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5151 if (optlen < copy_len) {
5156 memcpy(optval, start, copy_len);
5159 /* Zero out unused buffer at the end */
5160 memset(optval + copy_len, 0, optlen - copy_len);
5165 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5168 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5169 .func = bpf_sock_ops_getsockopt,
5171 .ret_type = RET_INTEGER,
5172 .arg1_type = ARG_PTR_TO_CTX,
5173 .arg2_type = ARG_ANYTHING,
5174 .arg3_type = ARG_ANYTHING,
5175 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5176 .arg5_type = ARG_CONST_SIZE,
5179 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5182 struct sock *sk = bpf_sock->sk;
5183 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5185 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5188 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5190 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5193 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5194 .func = bpf_sock_ops_cb_flags_set,
5196 .ret_type = RET_INTEGER,
5197 .arg1_type = ARG_PTR_TO_CTX,
5198 .arg2_type = ARG_ANYTHING,
5201 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5202 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5204 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5208 struct sock *sk = ctx->sk;
5209 u32 flags = BIND_FROM_BPF;
5213 if (addr_len < offsetofend(struct sockaddr, sa_family))
5215 if (addr->sa_family == AF_INET) {
5216 if (addr_len < sizeof(struct sockaddr_in))
5218 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5219 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5220 return __inet_bind(sk, addr, addr_len, flags);
5221 #if IS_ENABLED(CONFIG_IPV6)
5222 } else if (addr->sa_family == AF_INET6) {
5223 if (addr_len < SIN6_LEN_RFC2133)
5225 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5226 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5227 /* ipv6_bpf_stub cannot be NULL, since it's called from
5228 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5230 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5231 #endif /* CONFIG_IPV6 */
5233 #endif /* CONFIG_INET */
5235 return -EAFNOSUPPORT;
5238 static const struct bpf_func_proto bpf_bind_proto = {
5241 .ret_type = RET_INTEGER,
5242 .arg1_type = ARG_PTR_TO_CTX,
5243 .arg2_type = ARG_PTR_TO_MEM,
5244 .arg3_type = ARG_CONST_SIZE,
5248 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5249 struct bpf_xfrm_state *, to, u32, size, u64, flags)
5251 const struct sec_path *sp = skb_sec_path(skb);
5252 const struct xfrm_state *x;
5254 if (!sp || unlikely(index >= sp->len || flags))
5257 x = sp->xvec[index];
5259 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5262 to->reqid = x->props.reqid;
5263 to->spi = x->id.spi;
5264 to->family = x->props.family;
5267 if (to->family == AF_INET6) {
5268 memcpy(to->remote_ipv6, x->props.saddr.a6,
5269 sizeof(to->remote_ipv6));
5271 to->remote_ipv4 = x->props.saddr.a4;
5272 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5277 memset(to, 0, size);
5281 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5282 .func = bpf_skb_get_xfrm_state,
5284 .ret_type = RET_INTEGER,
5285 .arg1_type = ARG_PTR_TO_CTX,
5286 .arg2_type = ARG_ANYTHING,
5287 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
5288 .arg4_type = ARG_CONST_SIZE,
5289 .arg5_type = ARG_ANYTHING,
5293 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5294 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
5295 const struct neighbour *neigh,
5296 const struct net_device *dev)
5298 memcpy(params->dmac, neigh->ha, ETH_ALEN);
5299 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5300 params->h_vlan_TCI = 0;
5301 params->h_vlan_proto = 0;
5307 #if IS_ENABLED(CONFIG_INET)
5308 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5309 u32 flags, bool check_mtu)
5311 struct fib_nh_common *nhc;
5312 struct in_device *in_dev;
5313 struct neighbour *neigh;
5314 struct net_device *dev;
5315 struct fib_result res;
5320 dev = dev_get_by_index_rcu(net, params->ifindex);
5324 /* verify forwarding is enabled on this interface */
5325 in_dev = __in_dev_get_rcu(dev);
5326 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5327 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5329 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5331 fl4.flowi4_oif = params->ifindex;
5333 fl4.flowi4_iif = params->ifindex;
5336 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5337 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5338 fl4.flowi4_flags = 0;
5340 fl4.flowi4_proto = params->l4_protocol;
5341 fl4.daddr = params->ipv4_dst;
5342 fl4.saddr = params->ipv4_src;
5343 fl4.fl4_sport = params->sport;
5344 fl4.fl4_dport = params->dport;
5345 fl4.flowi4_multipath_hash = 0;
5347 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5348 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5349 struct fib_table *tb;
5351 tb = fib_get_table(net, tbid);
5353 return BPF_FIB_LKUP_RET_NOT_FWDED;
5355 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5357 fl4.flowi4_mark = 0;
5358 fl4.flowi4_secid = 0;
5359 fl4.flowi4_tun_key.tun_id = 0;
5360 fl4.flowi4_uid = sock_net_uid(net, NULL);
5362 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5366 /* map fib lookup errors to RTN_ type */
5368 return BPF_FIB_LKUP_RET_BLACKHOLE;
5369 if (err == -EHOSTUNREACH)
5370 return BPF_FIB_LKUP_RET_UNREACHABLE;
5372 return BPF_FIB_LKUP_RET_PROHIBIT;
5374 return BPF_FIB_LKUP_RET_NOT_FWDED;
5377 if (res.type != RTN_UNICAST)
5378 return BPF_FIB_LKUP_RET_NOT_FWDED;
5380 if (fib_info_num_path(res.fi) > 1)
5381 fib_select_path(net, &res, &fl4, NULL);
5384 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5385 if (params->tot_len > mtu)
5386 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5391 /* do not handle lwt encaps right now */
5392 if (nhc->nhc_lwtstate)
5393 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5397 params->rt_metric = res.fi->fib_priority;
5398 params->ifindex = dev->ifindex;
5400 /* xdp and cls_bpf programs are run in RCU-bh so
5401 * rcu_read_lock_bh is not needed here
5403 if (likely(nhc->nhc_gw_family != AF_INET6)) {
5404 if (nhc->nhc_gw_family)
5405 params->ipv4_dst = nhc->nhc_gw.ipv4;
5407 neigh = __ipv4_neigh_lookup_noref(dev,
5408 (__force u32)params->ipv4_dst);
5410 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5412 params->family = AF_INET6;
5413 *dst = nhc->nhc_gw.ipv6;
5414 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5418 return BPF_FIB_LKUP_RET_NO_NEIGH;
5420 return bpf_fib_set_fwd_params(params, neigh, dev);
5424 #if IS_ENABLED(CONFIG_IPV6)
5425 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5426 u32 flags, bool check_mtu)
5428 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5429 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5430 struct fib6_result res = {};
5431 struct neighbour *neigh;
5432 struct net_device *dev;
5433 struct inet6_dev *idev;
5439 /* link local addresses are never forwarded */
5440 if (rt6_need_strict(dst) || rt6_need_strict(src))
5441 return BPF_FIB_LKUP_RET_NOT_FWDED;
5443 dev = dev_get_by_index_rcu(net, params->ifindex);
5447 idev = __in6_dev_get_safely(dev);
5448 if (unlikely(!idev || !idev->cnf.forwarding))
5449 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5451 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5453 oif = fl6.flowi6_oif = params->ifindex;
5455 oif = fl6.flowi6_iif = params->ifindex;
5457 strict = RT6_LOOKUP_F_HAS_SADDR;
5459 fl6.flowlabel = params->flowinfo;
5460 fl6.flowi6_scope = 0;
5461 fl6.flowi6_flags = 0;
5464 fl6.flowi6_proto = params->l4_protocol;
5467 fl6.fl6_sport = params->sport;
5468 fl6.fl6_dport = params->dport;
5470 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5471 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5472 struct fib6_table *tb;
5474 tb = ipv6_stub->fib6_get_table(net, tbid);
5476 return BPF_FIB_LKUP_RET_NOT_FWDED;
5478 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5481 fl6.flowi6_mark = 0;
5482 fl6.flowi6_secid = 0;
5483 fl6.flowi6_tun_key.tun_id = 0;
5484 fl6.flowi6_uid = sock_net_uid(net, NULL);
5486 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5489 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5490 res.f6i == net->ipv6.fib6_null_entry))
5491 return BPF_FIB_LKUP_RET_NOT_FWDED;
5493 switch (res.fib6_type) {
5494 /* only unicast is forwarded */
5498 return BPF_FIB_LKUP_RET_BLACKHOLE;
5499 case RTN_UNREACHABLE:
5500 return BPF_FIB_LKUP_RET_UNREACHABLE;
5502 return BPF_FIB_LKUP_RET_PROHIBIT;
5504 return BPF_FIB_LKUP_RET_NOT_FWDED;
5507 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5508 fl6.flowi6_oif != 0, NULL, strict);
5511 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5512 if (params->tot_len > mtu)
5513 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5516 if (res.nh->fib_nh_lws)
5517 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5519 if (res.nh->fib_nh_gw_family)
5520 *dst = res.nh->fib_nh_gw6;
5522 dev = res.nh->fib_nh_dev;
5523 params->rt_metric = res.f6i->fib6_metric;
5524 params->ifindex = dev->ifindex;
5526 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
5529 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5531 return BPF_FIB_LKUP_RET_NO_NEIGH;
5533 return bpf_fib_set_fwd_params(params, neigh, dev);
5537 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
5538 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5540 if (plen < sizeof(*params))
5543 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5546 switch (params->family) {
5547 #if IS_ENABLED(CONFIG_INET)
5549 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
5552 #if IS_ENABLED(CONFIG_IPV6)
5554 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
5558 return -EAFNOSUPPORT;
5561 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
5562 .func = bpf_xdp_fib_lookup,
5564 .ret_type = RET_INTEGER,
5565 .arg1_type = ARG_PTR_TO_CTX,
5566 .arg2_type = ARG_PTR_TO_MEM,
5567 .arg3_type = ARG_CONST_SIZE,
5568 .arg4_type = ARG_ANYTHING,
5571 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
5572 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5574 struct net *net = dev_net(skb->dev);
5575 int rc = -EAFNOSUPPORT;
5577 if (plen < sizeof(*params))
5580 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5583 switch (params->family) {
5584 #if IS_ENABLED(CONFIG_INET)
5586 rc = bpf_ipv4_fib_lookup(net, params, flags, false);
5589 #if IS_ENABLED(CONFIG_IPV6)
5591 rc = bpf_ipv6_fib_lookup(net, params, flags, false);
5597 struct net_device *dev;
5599 dev = dev_get_by_index_rcu(net, params->ifindex);
5600 if (!is_skb_forwardable(dev, skb))
5601 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
5607 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
5608 .func = bpf_skb_fib_lookup,
5610 .ret_type = RET_INTEGER,
5611 .arg1_type = ARG_PTR_TO_CTX,
5612 .arg2_type = ARG_PTR_TO_MEM,
5613 .arg3_type = ARG_CONST_SIZE,
5614 .arg4_type = ARG_ANYTHING,
5617 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5618 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
5621 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
5623 if (!seg6_validate_srh(srh, len, false))
5627 case BPF_LWT_ENCAP_SEG6_INLINE:
5628 if (skb->protocol != htons(ETH_P_IPV6))
5631 err = seg6_do_srh_inline(skb, srh);
5633 case BPF_LWT_ENCAP_SEG6:
5634 skb_reset_inner_headers(skb);
5635 skb->encapsulation = 1;
5636 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
5642 bpf_compute_data_pointers(skb);
5646 ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5647 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
5649 return seg6_lookup_nexthop(skb, NULL, 0);
5651 #endif /* CONFIG_IPV6_SEG6_BPF */
5653 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5654 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
5657 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
5661 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
5665 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5666 case BPF_LWT_ENCAP_SEG6:
5667 case BPF_LWT_ENCAP_SEG6_INLINE:
5668 return bpf_push_seg6_encap(skb, type, hdr, len);
5670 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5671 case BPF_LWT_ENCAP_IP:
5672 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
5679 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
5680 void *, hdr, u32, len)
5683 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5684 case BPF_LWT_ENCAP_IP:
5685 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
5692 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
5693 .func = bpf_lwt_in_push_encap,
5695 .ret_type = RET_INTEGER,
5696 .arg1_type = ARG_PTR_TO_CTX,
5697 .arg2_type = ARG_ANYTHING,
5698 .arg3_type = ARG_PTR_TO_MEM,
5699 .arg4_type = ARG_CONST_SIZE
5702 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
5703 .func = bpf_lwt_xmit_push_encap,
5705 .ret_type = RET_INTEGER,
5706 .arg1_type = ARG_PTR_TO_CTX,
5707 .arg2_type = ARG_ANYTHING,
5708 .arg3_type = ARG_PTR_TO_MEM,
5709 .arg4_type = ARG_CONST_SIZE
5712 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5713 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
5714 const void *, from, u32, len)
5716 struct seg6_bpf_srh_state *srh_state =
5717 this_cpu_ptr(&seg6_bpf_srh_states);
5718 struct ipv6_sr_hdr *srh = srh_state->srh;
5719 void *srh_tlvs, *srh_end, *ptr;
5725 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
5726 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
5728 ptr = skb->data + offset;
5729 if (ptr >= srh_tlvs && ptr + len <= srh_end)
5730 srh_state->valid = false;
5731 else if (ptr < (void *)&srh->flags ||
5732 ptr + len > (void *)&srh->segments)
5735 if (unlikely(bpf_try_make_writable(skb, offset + len)))
5737 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5739 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5741 memcpy(skb->data + offset, from, len);
5745 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
5746 .func = bpf_lwt_seg6_store_bytes,
5748 .ret_type = RET_INTEGER,
5749 .arg1_type = ARG_PTR_TO_CTX,
5750 .arg2_type = ARG_ANYTHING,
5751 .arg3_type = ARG_PTR_TO_MEM,
5752 .arg4_type = ARG_CONST_SIZE
5755 static void bpf_update_srh_state(struct sk_buff *skb)
5757 struct seg6_bpf_srh_state *srh_state =
5758 this_cpu_ptr(&seg6_bpf_srh_states);
5761 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
5762 srh_state->srh = NULL;
5764 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5765 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
5766 srh_state->valid = true;
5770 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
5771 u32, action, void *, param, u32, param_len)
5773 struct seg6_bpf_srh_state *srh_state =
5774 this_cpu_ptr(&seg6_bpf_srh_states);
5779 case SEG6_LOCAL_ACTION_END_X:
5780 if (!seg6_bpf_has_valid_srh(skb))
5782 if (param_len != sizeof(struct in6_addr))
5784 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
5785 case SEG6_LOCAL_ACTION_END_T:
5786 if (!seg6_bpf_has_valid_srh(skb))
5788 if (param_len != sizeof(int))
5790 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5791 case SEG6_LOCAL_ACTION_END_DT6:
5792 if (!seg6_bpf_has_valid_srh(skb))
5794 if (param_len != sizeof(int))
5797 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
5799 if (!pskb_pull(skb, hdroff))
5802 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
5803 skb_reset_network_header(skb);
5804 skb_reset_transport_header(skb);
5805 skb->encapsulation = 0;
5807 bpf_compute_data_pointers(skb);
5808 bpf_update_srh_state(skb);
5809 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5810 case SEG6_LOCAL_ACTION_END_B6:
5811 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5813 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
5816 bpf_update_srh_state(skb);
5819 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
5820 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5822 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
5825 bpf_update_srh_state(skb);
5833 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
5834 .func = bpf_lwt_seg6_action,
5836 .ret_type = RET_INTEGER,
5837 .arg1_type = ARG_PTR_TO_CTX,
5838 .arg2_type = ARG_ANYTHING,
5839 .arg3_type = ARG_PTR_TO_MEM,
5840 .arg4_type = ARG_CONST_SIZE
5843 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
5846 struct seg6_bpf_srh_state *srh_state =
5847 this_cpu_ptr(&seg6_bpf_srh_states);
5848 struct ipv6_sr_hdr *srh = srh_state->srh;
5849 void *srh_end, *srh_tlvs, *ptr;
5850 struct ipv6hdr *hdr;
5854 if (unlikely(srh == NULL))
5857 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
5858 ((srh->first_segment + 1) << 4));
5859 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
5861 ptr = skb->data + offset;
5863 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
5865 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
5869 ret = skb_cow_head(skb, len);
5870 if (unlikely(ret < 0))
5873 ret = bpf_skb_net_hdr_push(skb, offset, len);
5875 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
5878 bpf_compute_data_pointers(skb);
5879 if (unlikely(ret < 0))
5882 hdr = (struct ipv6hdr *)skb->data;
5883 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5885 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5887 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5888 srh_state->hdrlen += len;
5889 srh_state->valid = false;
5893 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
5894 .func = bpf_lwt_seg6_adjust_srh,
5896 .ret_type = RET_INTEGER,
5897 .arg1_type = ARG_PTR_TO_CTX,
5898 .arg2_type = ARG_ANYTHING,
5899 .arg3_type = ARG_ANYTHING,
5901 #endif /* CONFIG_IPV6_SEG6_BPF */
5904 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
5905 int dif, int sdif, u8 family, u8 proto)
5907 bool refcounted = false;
5908 struct sock *sk = NULL;
5910 if (family == AF_INET) {
5911 __be32 src4 = tuple->ipv4.saddr;
5912 __be32 dst4 = tuple->ipv4.daddr;
5914 if (proto == IPPROTO_TCP)
5915 sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
5916 src4, tuple->ipv4.sport,
5917 dst4, tuple->ipv4.dport,
5918 dif, sdif, &refcounted);
5920 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
5921 dst4, tuple->ipv4.dport,
5922 dif, sdif, &udp_table, NULL);
5923 #if IS_ENABLED(CONFIG_IPV6)
5925 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
5926 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
5928 if (proto == IPPROTO_TCP)
5929 sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
5930 src6, tuple->ipv6.sport,
5931 dst6, ntohs(tuple->ipv6.dport),
5932 dif, sdif, &refcounted);
5933 else if (likely(ipv6_bpf_stub))
5934 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
5935 src6, tuple->ipv6.sport,
5936 dst6, tuple->ipv6.dport,
5942 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
5943 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
5949 /* bpf_skc_lookup performs the core lookup for different types of sockets,
5950 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
5951 * Returns the socket as an 'unsigned long' to simplify the casting in the
5952 * callers to satisfy BPF_CALL declarations.
5954 static struct sock *
5955 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5956 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
5959 struct sock *sk = NULL;
5960 u8 family = AF_UNSPEC;
5964 if (len == sizeof(tuple->ipv4))
5966 else if (len == sizeof(tuple->ipv6))
5971 if (unlikely(family == AF_UNSPEC || flags ||
5972 !((s32)netns_id < 0 || netns_id <= S32_MAX)))
5975 if (family == AF_INET)
5976 sdif = inet_sdif(skb);
5978 sdif = inet6_sdif(skb);
5980 if ((s32)netns_id < 0) {
5982 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
5984 net = get_net_ns_by_id(caller_net, netns_id);
5987 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
5995 static struct sock *
5996 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5997 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6000 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6001 ifindex, proto, netns_id, flags);
6004 sk = sk_to_full_sk(sk);
6005 if (!sk_fullsock(sk)) {
6014 static struct sock *
6015 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6016 u8 proto, u64 netns_id, u64 flags)
6018 struct net *caller_net;
6022 caller_net = dev_net(skb->dev);
6023 ifindex = skb->dev->ifindex;
6025 caller_net = sock_net(skb->sk);
6029 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6033 static struct sock *
6034 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6035 u8 proto, u64 netns_id, u64 flags)
6037 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6041 sk = sk_to_full_sk(sk);
6042 if (!sk_fullsock(sk)) {
6051 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6052 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6054 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6058 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6059 .func = bpf_skc_lookup_tcp,
6062 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6063 .arg1_type = ARG_PTR_TO_CTX,
6064 .arg2_type = ARG_PTR_TO_MEM,
6065 .arg3_type = ARG_CONST_SIZE,
6066 .arg4_type = ARG_ANYTHING,
6067 .arg5_type = ARG_ANYTHING,
6070 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6071 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6073 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6077 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6078 .func = bpf_sk_lookup_tcp,
6081 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6082 .arg1_type = ARG_PTR_TO_CTX,
6083 .arg2_type = ARG_PTR_TO_MEM,
6084 .arg3_type = ARG_CONST_SIZE,
6085 .arg4_type = ARG_ANYTHING,
6086 .arg5_type = ARG_ANYTHING,
6089 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6090 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6092 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6096 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6097 .func = bpf_sk_lookup_udp,
6100 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6101 .arg1_type = ARG_PTR_TO_CTX,
6102 .arg2_type = ARG_PTR_TO_MEM,
6103 .arg3_type = ARG_CONST_SIZE,
6104 .arg4_type = ARG_ANYTHING,
6105 .arg5_type = ARG_ANYTHING,
6108 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6110 if (sk && sk_is_refcounted(sk))
6115 static const struct bpf_func_proto bpf_sk_release_proto = {
6116 .func = bpf_sk_release,
6118 .ret_type = RET_INTEGER,
6119 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6122 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6123 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6125 struct net *caller_net = dev_net(ctx->rxq->dev);
6126 int ifindex = ctx->rxq->dev->ifindex;
6128 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6129 ifindex, IPPROTO_UDP, netns_id,
6133 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6134 .func = bpf_xdp_sk_lookup_udp,
6137 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6138 .arg1_type = ARG_PTR_TO_CTX,
6139 .arg2_type = ARG_PTR_TO_MEM,
6140 .arg3_type = ARG_CONST_SIZE,
6141 .arg4_type = ARG_ANYTHING,
6142 .arg5_type = ARG_ANYTHING,
6145 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6146 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6148 struct net *caller_net = dev_net(ctx->rxq->dev);
6149 int ifindex = ctx->rxq->dev->ifindex;
6151 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6152 ifindex, IPPROTO_TCP, netns_id,
6156 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6157 .func = bpf_xdp_skc_lookup_tcp,
6160 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6161 .arg1_type = ARG_PTR_TO_CTX,
6162 .arg2_type = ARG_PTR_TO_MEM,
6163 .arg3_type = ARG_CONST_SIZE,
6164 .arg4_type = ARG_ANYTHING,
6165 .arg5_type = ARG_ANYTHING,
6168 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6169 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6171 struct net *caller_net = dev_net(ctx->rxq->dev);
6172 int ifindex = ctx->rxq->dev->ifindex;
6174 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6175 ifindex, IPPROTO_TCP, netns_id,
6179 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6180 .func = bpf_xdp_sk_lookup_tcp,
6183 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6184 .arg1_type = ARG_PTR_TO_CTX,
6185 .arg2_type = ARG_PTR_TO_MEM,
6186 .arg3_type = ARG_CONST_SIZE,
6187 .arg4_type = ARG_ANYTHING,
6188 .arg5_type = ARG_ANYTHING,
6191 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6192 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6194 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6195 sock_net(ctx->sk), 0,
6196 IPPROTO_TCP, netns_id, flags);
6199 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6200 .func = bpf_sock_addr_skc_lookup_tcp,
6202 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6203 .arg1_type = ARG_PTR_TO_CTX,
6204 .arg2_type = ARG_PTR_TO_MEM,
6205 .arg3_type = ARG_CONST_SIZE,
6206 .arg4_type = ARG_ANYTHING,
6207 .arg5_type = ARG_ANYTHING,
6210 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6211 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6213 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6214 sock_net(ctx->sk), 0, IPPROTO_TCP,
6218 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6219 .func = bpf_sock_addr_sk_lookup_tcp,
6221 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6222 .arg1_type = ARG_PTR_TO_CTX,
6223 .arg2_type = ARG_PTR_TO_MEM,
6224 .arg3_type = ARG_CONST_SIZE,
6225 .arg4_type = ARG_ANYTHING,
6226 .arg5_type = ARG_ANYTHING,
6229 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6230 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6232 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6233 sock_net(ctx->sk), 0, IPPROTO_UDP,
6237 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6238 .func = bpf_sock_addr_sk_lookup_udp,
6240 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6241 .arg1_type = ARG_PTR_TO_CTX,
6242 .arg2_type = ARG_PTR_TO_MEM,
6243 .arg3_type = ARG_CONST_SIZE,
6244 .arg4_type = ARG_ANYTHING,
6245 .arg5_type = ARG_ANYTHING,
6248 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6249 struct bpf_insn_access_aux *info)
6251 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6255 if (off % size != 0)
6259 case offsetof(struct bpf_tcp_sock, bytes_received):
6260 case offsetof(struct bpf_tcp_sock, bytes_acked):
6261 return size == sizeof(__u64);
6263 return size == sizeof(__u32);
6267 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6268 const struct bpf_insn *si,
6269 struct bpf_insn *insn_buf,
6270 struct bpf_prog *prog, u32 *target_size)
6272 struct bpf_insn *insn = insn_buf;
6274 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \
6276 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \
6277 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6278 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6279 si->dst_reg, si->src_reg, \
6280 offsetof(struct tcp_sock, FIELD)); \
6283 #define BPF_INET_SOCK_GET_COMMON(FIELD) \
6285 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \
6287 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6288 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6289 struct inet_connection_sock, \
6291 si->dst_reg, si->src_reg, \
6293 struct inet_connection_sock, \
6297 if (insn > insn_buf)
6298 return insn - insn_buf;
6301 case offsetof(struct bpf_tcp_sock, rtt_min):
6302 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
6303 sizeof(struct minmax));
6304 BUILD_BUG_ON(sizeof(struct minmax) <
6305 sizeof(struct minmax_sample));
6307 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6308 offsetof(struct tcp_sock, rtt_min) +
6309 offsetof(struct minmax_sample, v));
6311 case offsetof(struct bpf_tcp_sock, snd_cwnd):
6312 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
6314 case offsetof(struct bpf_tcp_sock, srtt_us):
6315 BPF_TCP_SOCK_GET_COMMON(srtt_us);
6317 case offsetof(struct bpf_tcp_sock, snd_ssthresh):
6318 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
6320 case offsetof(struct bpf_tcp_sock, rcv_nxt):
6321 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
6323 case offsetof(struct bpf_tcp_sock, snd_nxt):
6324 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
6326 case offsetof(struct bpf_tcp_sock, snd_una):
6327 BPF_TCP_SOCK_GET_COMMON(snd_una);
6329 case offsetof(struct bpf_tcp_sock, mss_cache):
6330 BPF_TCP_SOCK_GET_COMMON(mss_cache);
6332 case offsetof(struct bpf_tcp_sock, ecn_flags):
6333 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
6335 case offsetof(struct bpf_tcp_sock, rate_delivered):
6336 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
6338 case offsetof(struct bpf_tcp_sock, rate_interval_us):
6339 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
6341 case offsetof(struct bpf_tcp_sock, packets_out):
6342 BPF_TCP_SOCK_GET_COMMON(packets_out);
6344 case offsetof(struct bpf_tcp_sock, retrans_out):
6345 BPF_TCP_SOCK_GET_COMMON(retrans_out);
6347 case offsetof(struct bpf_tcp_sock, total_retrans):
6348 BPF_TCP_SOCK_GET_COMMON(total_retrans);
6350 case offsetof(struct bpf_tcp_sock, segs_in):
6351 BPF_TCP_SOCK_GET_COMMON(segs_in);
6353 case offsetof(struct bpf_tcp_sock, data_segs_in):
6354 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
6356 case offsetof(struct bpf_tcp_sock, segs_out):
6357 BPF_TCP_SOCK_GET_COMMON(segs_out);
6359 case offsetof(struct bpf_tcp_sock, data_segs_out):
6360 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
6362 case offsetof(struct bpf_tcp_sock, lost_out):
6363 BPF_TCP_SOCK_GET_COMMON(lost_out);
6365 case offsetof(struct bpf_tcp_sock, sacked_out):
6366 BPF_TCP_SOCK_GET_COMMON(sacked_out);
6368 case offsetof(struct bpf_tcp_sock, bytes_received):
6369 BPF_TCP_SOCK_GET_COMMON(bytes_received);
6371 case offsetof(struct bpf_tcp_sock, bytes_acked):
6372 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
6374 case offsetof(struct bpf_tcp_sock, dsack_dups):
6375 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
6377 case offsetof(struct bpf_tcp_sock, delivered):
6378 BPF_TCP_SOCK_GET_COMMON(delivered);
6380 case offsetof(struct bpf_tcp_sock, delivered_ce):
6381 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
6383 case offsetof(struct bpf_tcp_sock, icsk_retransmits):
6384 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
6388 return insn - insn_buf;
6391 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
6393 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
6394 return (unsigned long)sk;
6396 return (unsigned long)NULL;
6399 const struct bpf_func_proto bpf_tcp_sock_proto = {
6400 .func = bpf_tcp_sock,
6402 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
6403 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6406 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
6408 sk = sk_to_full_sk(sk);
6410 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
6411 return (unsigned long)sk;
6413 return (unsigned long)NULL;
6416 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
6417 .func = bpf_get_listener_sock,
6419 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6420 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6423 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
6425 unsigned int iphdr_len;
6427 switch (skb_protocol(skb, true)) {
6428 case cpu_to_be16(ETH_P_IP):
6429 iphdr_len = sizeof(struct iphdr);
6431 case cpu_to_be16(ETH_P_IPV6):
6432 iphdr_len = sizeof(struct ipv6hdr);
6438 if (skb_headlen(skb) < iphdr_len)
6441 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
6444 return INET_ECN_set_ce(skb);
6447 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6448 struct bpf_insn_access_aux *info)
6450 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
6453 if (off % size != 0)
6458 return size == sizeof(__u32);
6462 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
6463 const struct bpf_insn *si,
6464 struct bpf_insn *insn_buf,
6465 struct bpf_prog *prog, u32 *target_size)
6467 struct bpf_insn *insn = insn_buf;
6469 #define BPF_XDP_SOCK_GET(FIELD) \
6471 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \
6472 sizeof_field(struct bpf_xdp_sock, FIELD)); \
6473 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
6474 si->dst_reg, si->src_reg, \
6475 offsetof(struct xdp_sock, FIELD)); \
6479 case offsetof(struct bpf_xdp_sock, queue_id):
6480 BPF_XDP_SOCK_GET(queue_id);
6484 return insn - insn_buf;
6487 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
6488 .func = bpf_skb_ecn_set_ce,
6490 .ret_type = RET_INTEGER,
6491 .arg1_type = ARG_PTR_TO_CTX,
6494 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6495 struct tcphdr *, th, u32, th_len)
6497 #ifdef CONFIG_SYN_COOKIES
6501 if (unlikely(!sk || th_len < sizeof(*th)))
6504 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
6505 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6508 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
6511 if (!th->ack || th->rst || th->syn)
6514 if (tcp_synq_no_recent_overflow(sk))
6517 cookie = ntohl(th->ack_seq) - 1;
6519 switch (sk->sk_family) {
6521 if (unlikely(iph_len < sizeof(struct iphdr)))
6524 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
6527 #if IS_BUILTIN(CONFIG_IPV6)
6529 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6532 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
6534 #endif /* CONFIG_IPV6 */
6537 return -EPROTONOSUPPORT;
6549 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
6550 .func = bpf_tcp_check_syncookie,
6553 .ret_type = RET_INTEGER,
6554 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6555 .arg2_type = ARG_PTR_TO_MEM,
6556 .arg3_type = ARG_CONST_SIZE,
6557 .arg4_type = ARG_PTR_TO_MEM,
6558 .arg5_type = ARG_CONST_SIZE,
6561 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6562 struct tcphdr *, th, u32, th_len)
6564 #ifdef CONFIG_SYN_COOKIES
6568 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
6571 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6574 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
6577 if (!th->syn || th->ack || th->fin || th->rst)
6580 if (unlikely(iph_len < sizeof(struct iphdr)))
6583 /* Both struct iphdr and struct ipv6hdr have the version field at the
6584 * same offset so we can cast to the shorter header (struct iphdr).
6586 switch (((struct iphdr *)iph)->version) {
6588 if (sk->sk_family == AF_INET6 && sk->sk_ipv6only)
6591 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
6594 #if IS_BUILTIN(CONFIG_IPV6)
6596 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6599 if (sk->sk_family != AF_INET6)
6602 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
6604 #endif /* CONFIG_IPV6 */
6607 return -EPROTONOSUPPORT;
6612 return cookie | ((u64)mss << 32);
6615 #endif /* CONFIG_SYN_COOKIES */
6618 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
6619 .func = bpf_tcp_gen_syncookie,
6620 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */
6622 .ret_type = RET_INTEGER,
6623 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6624 .arg2_type = ARG_PTR_TO_MEM,
6625 .arg3_type = ARG_CONST_SIZE,
6626 .arg4_type = ARG_PTR_TO_MEM,
6627 .arg5_type = ARG_CONST_SIZE,
6630 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
6632 if (!sk || flags != 0)
6634 if (!skb_at_tc_ingress(skb))
6636 if (unlikely(dev_net(skb->dev) != sock_net(sk)))
6637 return -ENETUNREACH;
6638 if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
6639 return -ESOCKTNOSUPPORT;
6640 if (sk_is_refcounted(sk) &&
6641 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
6646 skb->destructor = sock_pfree;
6651 static const struct bpf_func_proto bpf_sk_assign_proto = {
6652 .func = bpf_sk_assign,
6654 .ret_type = RET_INTEGER,
6655 .arg1_type = ARG_PTR_TO_CTX,
6656 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6657 .arg3_type = ARG_ANYTHING,
6660 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
6661 u8 search_kind, const u8 *magic,
6662 u8 magic_len, bool *eol)
6668 while (op < opend) {
6671 if (kind == TCPOPT_EOL) {
6673 return ERR_PTR(-ENOMSG);
6674 } else if (kind == TCPOPT_NOP) {
6679 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
6680 /* Something is wrong in the received header.
6681 * Follow the TCP stack's tcp_parse_options()
6682 * and just bail here.
6684 return ERR_PTR(-EFAULT);
6687 if (search_kind == kind) {
6691 if (magic_len > kind_len - 2)
6692 return ERR_PTR(-ENOMSG);
6694 if (!memcmp(&op[2], magic, magic_len))
6701 return ERR_PTR(-ENOMSG);
6704 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6705 void *, search_res, u32, len, u64, flags)
6707 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
6708 const u8 *op, *opend, *magic, *search = search_res;
6709 u8 search_kind, search_len, copy_len, magic_len;
6712 /* 2 byte is the minimal option len except TCPOPT_NOP and
6713 * TCPOPT_EOL which are useless for the bpf prog to learn
6714 * and this helper disallow loading them also.
6716 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
6719 search_kind = search[0];
6720 search_len = search[1];
6722 if (search_len > len || search_kind == TCPOPT_NOP ||
6723 search_kind == TCPOPT_EOL)
6726 if (search_kind == TCPOPT_EXP || search_kind == 253) {
6727 /* 16 or 32 bit magic. +2 for kind and kind length */
6728 if (search_len != 4 && search_len != 6)
6731 magic_len = search_len - 2;
6740 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
6745 op += sizeof(struct tcphdr);
6747 if (!bpf_sock->skb ||
6748 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6749 /* This bpf_sock->op cannot call this helper */
6752 opend = bpf_sock->skb_data_end;
6753 op = bpf_sock->skb->data + sizeof(struct tcphdr);
6756 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
6763 if (copy_len > len) {
6768 memcpy(search_res, op, copy_len);
6772 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
6773 .func = bpf_sock_ops_load_hdr_opt,
6775 .ret_type = RET_INTEGER,
6776 .arg1_type = ARG_PTR_TO_CTX,
6777 .arg2_type = ARG_PTR_TO_MEM,
6778 .arg3_type = ARG_CONST_SIZE,
6779 .arg4_type = ARG_ANYTHING,
6782 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6783 const void *, from, u32, len, u64, flags)
6785 u8 new_kind, new_kind_len, magic_len = 0, *opend;
6786 const u8 *op, *new_op, *magic = NULL;
6787 struct sk_buff *skb;
6790 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
6793 if (len < 2 || flags)
6797 new_kind = new_op[0];
6798 new_kind_len = new_op[1];
6800 if (new_kind_len > len || new_kind == TCPOPT_NOP ||
6801 new_kind == TCPOPT_EOL)
6804 if (new_kind_len > bpf_sock->remaining_opt_len)
6807 /* 253 is another experimental kind */
6808 if (new_kind == TCPOPT_EXP || new_kind == 253) {
6809 if (new_kind_len < 4)
6811 /* Match for the 2 byte magic also.
6812 * RFC 6994: the magic could be 2 or 4 bytes.
6813 * Hence, matching by 2 byte only is on the
6814 * conservative side but it is the right
6815 * thing to do for the 'search-for-duplication'
6822 /* Check for duplication */
6823 skb = bpf_sock->skb;
6824 op = skb->data + sizeof(struct tcphdr);
6825 opend = bpf_sock->skb_data_end;
6827 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
6832 if (PTR_ERR(op) != -ENOMSG)
6836 /* The option has been ended. Treat it as no more
6837 * header option can be written.
6841 /* No duplication found. Store the header option. */
6842 memcpy(opend, from, new_kind_len);
6844 bpf_sock->remaining_opt_len -= new_kind_len;
6845 bpf_sock->skb_data_end += new_kind_len;
6850 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
6851 .func = bpf_sock_ops_store_hdr_opt,
6853 .ret_type = RET_INTEGER,
6854 .arg1_type = ARG_PTR_TO_CTX,
6855 .arg2_type = ARG_PTR_TO_MEM,
6856 .arg3_type = ARG_CONST_SIZE,
6857 .arg4_type = ARG_ANYTHING,
6860 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6861 u32, len, u64, flags)
6863 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6866 if (flags || len < 2)
6869 if (len > bpf_sock->remaining_opt_len)
6872 bpf_sock->remaining_opt_len -= len;
6877 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
6878 .func = bpf_sock_ops_reserve_hdr_opt,
6880 .ret_type = RET_INTEGER,
6881 .arg1_type = ARG_PTR_TO_CTX,
6882 .arg2_type = ARG_ANYTHING,
6883 .arg3_type = ARG_ANYTHING,
6886 #endif /* CONFIG_INET */
6888 bool bpf_helper_changes_pkt_data(void *func)
6890 if (func == bpf_skb_vlan_push ||
6891 func == bpf_skb_vlan_pop ||
6892 func == bpf_skb_store_bytes ||
6893 func == bpf_skb_change_proto ||
6894 func == bpf_skb_change_head ||
6895 func == sk_skb_change_head ||
6896 func == bpf_skb_change_tail ||
6897 func == sk_skb_change_tail ||
6898 func == bpf_skb_adjust_room ||
6899 func == sk_skb_adjust_room ||
6900 func == bpf_skb_pull_data ||
6901 func == sk_skb_pull_data ||
6902 func == bpf_clone_redirect ||
6903 func == bpf_l3_csum_replace ||
6904 func == bpf_l4_csum_replace ||
6905 func == bpf_xdp_adjust_head ||
6906 func == bpf_xdp_adjust_meta ||
6907 func == bpf_msg_pull_data ||
6908 func == bpf_msg_push_data ||
6909 func == bpf_msg_pop_data ||
6910 func == bpf_xdp_adjust_tail ||
6911 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6912 func == bpf_lwt_seg6_store_bytes ||
6913 func == bpf_lwt_seg6_adjust_srh ||
6914 func == bpf_lwt_seg6_action ||
6917 func == bpf_sock_ops_store_hdr_opt ||
6919 func == bpf_lwt_in_push_encap ||
6920 func == bpf_lwt_xmit_push_encap)
6926 const struct bpf_func_proto bpf_event_output_data_proto __weak;
6927 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
6929 static const struct bpf_func_proto *
6930 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6933 /* inet and inet6 sockets are created in a process
6934 * context so there is always a valid uid/gid
6936 case BPF_FUNC_get_current_uid_gid:
6937 return &bpf_get_current_uid_gid_proto;
6938 case BPF_FUNC_get_local_storage:
6939 return &bpf_get_local_storage_proto;
6940 case BPF_FUNC_get_socket_cookie:
6941 return &bpf_get_socket_cookie_sock_proto;
6942 case BPF_FUNC_get_netns_cookie:
6943 return &bpf_get_netns_cookie_sock_proto;
6944 case BPF_FUNC_perf_event_output:
6945 return &bpf_event_output_data_proto;
6946 case BPF_FUNC_get_current_pid_tgid:
6947 return &bpf_get_current_pid_tgid_proto;
6948 case BPF_FUNC_get_current_comm:
6949 return &bpf_get_current_comm_proto;
6950 #ifdef CONFIG_CGROUPS
6951 case BPF_FUNC_get_current_cgroup_id:
6952 return &bpf_get_current_cgroup_id_proto;
6953 case BPF_FUNC_get_current_ancestor_cgroup_id:
6954 return &bpf_get_current_ancestor_cgroup_id_proto;
6956 #ifdef CONFIG_CGROUP_NET_CLASSID
6957 case BPF_FUNC_get_cgroup_classid:
6958 return &bpf_get_cgroup_classid_curr_proto;
6960 case BPF_FUNC_sk_storage_get:
6961 return &bpf_sk_storage_get_cg_sock_proto;
6963 return bpf_base_func_proto(func_id);
6967 static const struct bpf_func_proto *
6968 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6971 /* inet and inet6 sockets are created in a process
6972 * context so there is always a valid uid/gid
6974 case BPF_FUNC_get_current_uid_gid:
6975 return &bpf_get_current_uid_gid_proto;
6977 switch (prog->expected_attach_type) {
6978 case BPF_CGROUP_INET4_CONNECT:
6979 case BPF_CGROUP_INET6_CONNECT:
6980 return &bpf_bind_proto;
6984 case BPF_FUNC_get_socket_cookie:
6985 return &bpf_get_socket_cookie_sock_addr_proto;
6986 case BPF_FUNC_get_netns_cookie:
6987 return &bpf_get_netns_cookie_sock_addr_proto;
6988 case BPF_FUNC_get_local_storage:
6989 return &bpf_get_local_storage_proto;
6990 case BPF_FUNC_perf_event_output:
6991 return &bpf_event_output_data_proto;
6992 case BPF_FUNC_get_current_pid_tgid:
6993 return &bpf_get_current_pid_tgid_proto;
6994 case BPF_FUNC_get_current_comm:
6995 return &bpf_get_current_comm_proto;
6996 #ifdef CONFIG_CGROUPS
6997 case BPF_FUNC_get_current_cgroup_id:
6998 return &bpf_get_current_cgroup_id_proto;
6999 case BPF_FUNC_get_current_ancestor_cgroup_id:
7000 return &bpf_get_current_ancestor_cgroup_id_proto;
7002 #ifdef CONFIG_CGROUP_NET_CLASSID
7003 case BPF_FUNC_get_cgroup_classid:
7004 return &bpf_get_cgroup_classid_curr_proto;
7007 case BPF_FUNC_sk_lookup_tcp:
7008 return &bpf_sock_addr_sk_lookup_tcp_proto;
7009 case BPF_FUNC_sk_lookup_udp:
7010 return &bpf_sock_addr_sk_lookup_udp_proto;
7011 case BPF_FUNC_sk_release:
7012 return &bpf_sk_release_proto;
7013 case BPF_FUNC_skc_lookup_tcp:
7014 return &bpf_sock_addr_skc_lookup_tcp_proto;
7015 #endif /* CONFIG_INET */
7016 case BPF_FUNC_sk_storage_get:
7017 return &bpf_sk_storage_get_proto;
7018 case BPF_FUNC_sk_storage_delete:
7019 return &bpf_sk_storage_delete_proto;
7020 case BPF_FUNC_setsockopt:
7021 switch (prog->expected_attach_type) {
7022 case BPF_CGROUP_INET4_BIND:
7023 case BPF_CGROUP_INET6_BIND:
7024 case BPF_CGROUP_INET4_CONNECT:
7025 case BPF_CGROUP_INET6_CONNECT:
7026 return &bpf_sock_addr_setsockopt_proto;
7030 case BPF_FUNC_getsockopt:
7031 switch (prog->expected_attach_type) {
7032 case BPF_CGROUP_INET4_BIND:
7033 case BPF_CGROUP_INET6_BIND:
7034 case BPF_CGROUP_INET4_CONNECT:
7035 case BPF_CGROUP_INET6_CONNECT:
7036 return &bpf_sock_addr_getsockopt_proto;
7041 return bpf_sk_base_func_proto(func_id);
7045 static const struct bpf_func_proto *
7046 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7049 case BPF_FUNC_skb_load_bytes:
7050 return &bpf_skb_load_bytes_proto;
7051 case BPF_FUNC_skb_load_bytes_relative:
7052 return &bpf_skb_load_bytes_relative_proto;
7053 case BPF_FUNC_get_socket_cookie:
7054 return &bpf_get_socket_cookie_proto;
7055 case BPF_FUNC_get_socket_uid:
7056 return &bpf_get_socket_uid_proto;
7057 case BPF_FUNC_perf_event_output:
7058 return &bpf_skb_event_output_proto;
7060 return bpf_sk_base_func_proto(func_id);
7064 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7065 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7067 static const struct bpf_func_proto *
7068 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7071 case BPF_FUNC_get_local_storage:
7072 return &bpf_get_local_storage_proto;
7073 case BPF_FUNC_sk_fullsock:
7074 return &bpf_sk_fullsock_proto;
7075 case BPF_FUNC_sk_storage_get:
7076 return &bpf_sk_storage_get_proto;
7077 case BPF_FUNC_sk_storage_delete:
7078 return &bpf_sk_storage_delete_proto;
7079 case BPF_FUNC_perf_event_output:
7080 return &bpf_skb_event_output_proto;
7081 #ifdef CONFIG_SOCK_CGROUP_DATA
7082 case BPF_FUNC_skb_cgroup_id:
7083 return &bpf_skb_cgroup_id_proto;
7084 case BPF_FUNC_skb_ancestor_cgroup_id:
7085 return &bpf_skb_ancestor_cgroup_id_proto;
7086 case BPF_FUNC_sk_cgroup_id:
7087 return &bpf_sk_cgroup_id_proto;
7088 case BPF_FUNC_sk_ancestor_cgroup_id:
7089 return &bpf_sk_ancestor_cgroup_id_proto;
7092 case BPF_FUNC_sk_lookup_tcp:
7093 return &bpf_sk_lookup_tcp_proto;
7094 case BPF_FUNC_sk_lookup_udp:
7095 return &bpf_sk_lookup_udp_proto;
7096 case BPF_FUNC_sk_release:
7097 return &bpf_sk_release_proto;
7098 case BPF_FUNC_skc_lookup_tcp:
7099 return &bpf_skc_lookup_tcp_proto;
7100 case BPF_FUNC_tcp_sock:
7101 return &bpf_tcp_sock_proto;
7102 case BPF_FUNC_get_listener_sock:
7103 return &bpf_get_listener_sock_proto;
7104 case BPF_FUNC_skb_ecn_set_ce:
7105 return &bpf_skb_ecn_set_ce_proto;
7108 return sk_filter_func_proto(func_id, prog);
7112 static const struct bpf_func_proto *
7113 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7116 case BPF_FUNC_skb_store_bytes:
7117 return &bpf_skb_store_bytes_proto;
7118 case BPF_FUNC_skb_load_bytes:
7119 return &bpf_skb_load_bytes_proto;
7120 case BPF_FUNC_skb_load_bytes_relative:
7121 return &bpf_skb_load_bytes_relative_proto;
7122 case BPF_FUNC_skb_pull_data:
7123 return &bpf_skb_pull_data_proto;
7124 case BPF_FUNC_csum_diff:
7125 return &bpf_csum_diff_proto;
7126 case BPF_FUNC_csum_update:
7127 return &bpf_csum_update_proto;
7128 case BPF_FUNC_csum_level:
7129 return &bpf_csum_level_proto;
7130 case BPF_FUNC_l3_csum_replace:
7131 return &bpf_l3_csum_replace_proto;
7132 case BPF_FUNC_l4_csum_replace:
7133 return &bpf_l4_csum_replace_proto;
7134 case BPF_FUNC_clone_redirect:
7135 return &bpf_clone_redirect_proto;
7136 case BPF_FUNC_get_cgroup_classid:
7137 return &bpf_get_cgroup_classid_proto;
7138 case BPF_FUNC_skb_vlan_push:
7139 return &bpf_skb_vlan_push_proto;
7140 case BPF_FUNC_skb_vlan_pop:
7141 return &bpf_skb_vlan_pop_proto;
7142 case BPF_FUNC_skb_change_proto:
7143 return &bpf_skb_change_proto_proto;
7144 case BPF_FUNC_skb_change_type:
7145 return &bpf_skb_change_type_proto;
7146 case BPF_FUNC_skb_adjust_room:
7147 return &bpf_skb_adjust_room_proto;
7148 case BPF_FUNC_skb_change_tail:
7149 return &bpf_skb_change_tail_proto;
7150 case BPF_FUNC_skb_change_head:
7151 return &bpf_skb_change_head_proto;
7152 case BPF_FUNC_skb_get_tunnel_key:
7153 return &bpf_skb_get_tunnel_key_proto;
7154 case BPF_FUNC_skb_set_tunnel_key:
7155 return bpf_get_skb_set_tunnel_proto(func_id);
7156 case BPF_FUNC_skb_get_tunnel_opt:
7157 return &bpf_skb_get_tunnel_opt_proto;
7158 case BPF_FUNC_skb_set_tunnel_opt:
7159 return bpf_get_skb_set_tunnel_proto(func_id);
7160 case BPF_FUNC_redirect:
7161 return &bpf_redirect_proto;
7162 case BPF_FUNC_redirect_neigh:
7163 return &bpf_redirect_neigh_proto;
7164 case BPF_FUNC_redirect_peer:
7165 return &bpf_redirect_peer_proto;
7166 case BPF_FUNC_get_route_realm:
7167 return &bpf_get_route_realm_proto;
7168 case BPF_FUNC_get_hash_recalc:
7169 return &bpf_get_hash_recalc_proto;
7170 case BPF_FUNC_set_hash_invalid:
7171 return &bpf_set_hash_invalid_proto;
7172 case BPF_FUNC_set_hash:
7173 return &bpf_set_hash_proto;
7174 case BPF_FUNC_perf_event_output:
7175 return &bpf_skb_event_output_proto;
7176 case BPF_FUNC_get_smp_processor_id:
7177 return &bpf_get_smp_processor_id_proto;
7178 case BPF_FUNC_skb_under_cgroup:
7179 return &bpf_skb_under_cgroup_proto;
7180 case BPF_FUNC_get_socket_cookie:
7181 return &bpf_get_socket_cookie_proto;
7182 case BPF_FUNC_get_socket_uid:
7183 return &bpf_get_socket_uid_proto;
7184 case BPF_FUNC_fib_lookup:
7185 return &bpf_skb_fib_lookup_proto;
7186 case BPF_FUNC_sk_fullsock:
7187 return &bpf_sk_fullsock_proto;
7188 case BPF_FUNC_sk_storage_get:
7189 return &bpf_sk_storage_get_proto;
7190 case BPF_FUNC_sk_storage_delete:
7191 return &bpf_sk_storage_delete_proto;
7193 case BPF_FUNC_skb_get_xfrm_state:
7194 return &bpf_skb_get_xfrm_state_proto;
7196 #ifdef CONFIG_CGROUP_NET_CLASSID
7197 case BPF_FUNC_skb_cgroup_classid:
7198 return &bpf_skb_cgroup_classid_proto;
7200 #ifdef CONFIG_SOCK_CGROUP_DATA
7201 case BPF_FUNC_skb_cgroup_id:
7202 return &bpf_skb_cgroup_id_proto;
7203 case BPF_FUNC_skb_ancestor_cgroup_id:
7204 return &bpf_skb_ancestor_cgroup_id_proto;
7207 case BPF_FUNC_sk_lookup_tcp:
7208 return &bpf_sk_lookup_tcp_proto;
7209 case BPF_FUNC_sk_lookup_udp:
7210 return &bpf_sk_lookup_udp_proto;
7211 case BPF_FUNC_sk_release:
7212 return &bpf_sk_release_proto;
7213 case BPF_FUNC_tcp_sock:
7214 return &bpf_tcp_sock_proto;
7215 case BPF_FUNC_get_listener_sock:
7216 return &bpf_get_listener_sock_proto;
7217 case BPF_FUNC_skc_lookup_tcp:
7218 return &bpf_skc_lookup_tcp_proto;
7219 case BPF_FUNC_tcp_check_syncookie:
7220 return &bpf_tcp_check_syncookie_proto;
7221 case BPF_FUNC_skb_ecn_set_ce:
7222 return &bpf_skb_ecn_set_ce_proto;
7223 case BPF_FUNC_tcp_gen_syncookie:
7224 return &bpf_tcp_gen_syncookie_proto;
7225 case BPF_FUNC_sk_assign:
7226 return &bpf_sk_assign_proto;
7229 return bpf_sk_base_func_proto(func_id);
7233 static const struct bpf_func_proto *
7234 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7237 case BPF_FUNC_perf_event_output:
7238 return &bpf_xdp_event_output_proto;
7239 case BPF_FUNC_get_smp_processor_id:
7240 return &bpf_get_smp_processor_id_proto;
7241 case BPF_FUNC_csum_diff:
7242 return &bpf_csum_diff_proto;
7243 case BPF_FUNC_xdp_adjust_head:
7244 return &bpf_xdp_adjust_head_proto;
7245 case BPF_FUNC_xdp_adjust_meta:
7246 return &bpf_xdp_adjust_meta_proto;
7247 case BPF_FUNC_redirect:
7248 return &bpf_xdp_redirect_proto;
7249 case BPF_FUNC_redirect_map:
7250 return &bpf_xdp_redirect_map_proto;
7251 case BPF_FUNC_xdp_adjust_tail:
7252 return &bpf_xdp_adjust_tail_proto;
7253 case BPF_FUNC_fib_lookup:
7254 return &bpf_xdp_fib_lookup_proto;
7256 case BPF_FUNC_sk_lookup_udp:
7257 return &bpf_xdp_sk_lookup_udp_proto;
7258 case BPF_FUNC_sk_lookup_tcp:
7259 return &bpf_xdp_sk_lookup_tcp_proto;
7260 case BPF_FUNC_sk_release:
7261 return &bpf_sk_release_proto;
7262 case BPF_FUNC_skc_lookup_tcp:
7263 return &bpf_xdp_skc_lookup_tcp_proto;
7264 case BPF_FUNC_tcp_check_syncookie:
7265 return &bpf_tcp_check_syncookie_proto;
7266 case BPF_FUNC_tcp_gen_syncookie:
7267 return &bpf_tcp_gen_syncookie_proto;
7270 return bpf_sk_base_func_proto(func_id);
7274 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
7275 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
7277 static const struct bpf_func_proto *
7278 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7281 case BPF_FUNC_setsockopt:
7282 return &bpf_sock_ops_setsockopt_proto;
7283 case BPF_FUNC_getsockopt:
7284 return &bpf_sock_ops_getsockopt_proto;
7285 case BPF_FUNC_sock_ops_cb_flags_set:
7286 return &bpf_sock_ops_cb_flags_set_proto;
7287 case BPF_FUNC_sock_map_update:
7288 return &bpf_sock_map_update_proto;
7289 case BPF_FUNC_sock_hash_update:
7290 return &bpf_sock_hash_update_proto;
7291 case BPF_FUNC_get_socket_cookie:
7292 return &bpf_get_socket_cookie_sock_ops_proto;
7293 case BPF_FUNC_get_local_storage:
7294 return &bpf_get_local_storage_proto;
7295 case BPF_FUNC_perf_event_output:
7296 return &bpf_event_output_data_proto;
7297 case BPF_FUNC_sk_storage_get:
7298 return &bpf_sk_storage_get_proto;
7299 case BPF_FUNC_sk_storage_delete:
7300 return &bpf_sk_storage_delete_proto;
7302 case BPF_FUNC_load_hdr_opt:
7303 return &bpf_sock_ops_load_hdr_opt_proto;
7304 case BPF_FUNC_store_hdr_opt:
7305 return &bpf_sock_ops_store_hdr_opt_proto;
7306 case BPF_FUNC_reserve_hdr_opt:
7307 return &bpf_sock_ops_reserve_hdr_opt_proto;
7308 case BPF_FUNC_tcp_sock:
7309 return &bpf_tcp_sock_proto;
7310 #endif /* CONFIG_INET */
7312 return bpf_sk_base_func_proto(func_id);
7316 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
7317 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
7319 static const struct bpf_func_proto *
7320 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7323 case BPF_FUNC_msg_redirect_map:
7324 return &bpf_msg_redirect_map_proto;
7325 case BPF_FUNC_msg_redirect_hash:
7326 return &bpf_msg_redirect_hash_proto;
7327 case BPF_FUNC_msg_apply_bytes:
7328 return &bpf_msg_apply_bytes_proto;
7329 case BPF_FUNC_msg_cork_bytes:
7330 return &bpf_msg_cork_bytes_proto;
7331 case BPF_FUNC_msg_pull_data:
7332 return &bpf_msg_pull_data_proto;
7333 case BPF_FUNC_msg_push_data:
7334 return &bpf_msg_push_data_proto;
7335 case BPF_FUNC_msg_pop_data:
7336 return &bpf_msg_pop_data_proto;
7337 case BPF_FUNC_perf_event_output:
7338 return &bpf_event_output_data_proto;
7339 case BPF_FUNC_get_current_uid_gid:
7340 return &bpf_get_current_uid_gid_proto;
7341 case BPF_FUNC_get_current_pid_tgid:
7342 return &bpf_get_current_pid_tgid_proto;
7343 case BPF_FUNC_sk_storage_get:
7344 return &bpf_sk_storage_get_proto;
7345 case BPF_FUNC_sk_storage_delete:
7346 return &bpf_sk_storage_delete_proto;
7347 #ifdef CONFIG_CGROUPS
7348 case BPF_FUNC_get_current_cgroup_id:
7349 return &bpf_get_current_cgroup_id_proto;
7350 case BPF_FUNC_get_current_ancestor_cgroup_id:
7351 return &bpf_get_current_ancestor_cgroup_id_proto;
7353 #ifdef CONFIG_CGROUP_NET_CLASSID
7354 case BPF_FUNC_get_cgroup_classid:
7355 return &bpf_get_cgroup_classid_curr_proto;
7358 return bpf_sk_base_func_proto(func_id);
7362 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
7363 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
7365 static const struct bpf_func_proto *
7366 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7369 case BPF_FUNC_skb_store_bytes:
7370 return &bpf_skb_store_bytes_proto;
7371 case BPF_FUNC_skb_load_bytes:
7372 return &bpf_skb_load_bytes_proto;
7373 case BPF_FUNC_skb_pull_data:
7374 return &sk_skb_pull_data_proto;
7375 case BPF_FUNC_skb_change_tail:
7376 return &sk_skb_change_tail_proto;
7377 case BPF_FUNC_skb_change_head:
7378 return &sk_skb_change_head_proto;
7379 case BPF_FUNC_skb_adjust_room:
7380 return &sk_skb_adjust_room_proto;
7381 case BPF_FUNC_get_socket_cookie:
7382 return &bpf_get_socket_cookie_proto;
7383 case BPF_FUNC_get_socket_uid:
7384 return &bpf_get_socket_uid_proto;
7385 case BPF_FUNC_sk_redirect_map:
7386 return &bpf_sk_redirect_map_proto;
7387 case BPF_FUNC_sk_redirect_hash:
7388 return &bpf_sk_redirect_hash_proto;
7389 case BPF_FUNC_perf_event_output:
7390 return &bpf_skb_event_output_proto;
7392 case BPF_FUNC_sk_lookup_tcp:
7393 return &bpf_sk_lookup_tcp_proto;
7394 case BPF_FUNC_sk_lookup_udp:
7395 return &bpf_sk_lookup_udp_proto;
7396 case BPF_FUNC_sk_release:
7397 return &bpf_sk_release_proto;
7398 case BPF_FUNC_skc_lookup_tcp:
7399 return &bpf_skc_lookup_tcp_proto;
7402 return bpf_sk_base_func_proto(func_id);
7406 static const struct bpf_func_proto *
7407 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7410 case BPF_FUNC_skb_load_bytes:
7411 return &bpf_flow_dissector_load_bytes_proto;
7413 return bpf_sk_base_func_proto(func_id);
7417 static const struct bpf_func_proto *
7418 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7421 case BPF_FUNC_skb_load_bytes:
7422 return &bpf_skb_load_bytes_proto;
7423 case BPF_FUNC_skb_pull_data:
7424 return &bpf_skb_pull_data_proto;
7425 case BPF_FUNC_csum_diff:
7426 return &bpf_csum_diff_proto;
7427 case BPF_FUNC_get_cgroup_classid:
7428 return &bpf_get_cgroup_classid_proto;
7429 case BPF_FUNC_get_route_realm:
7430 return &bpf_get_route_realm_proto;
7431 case BPF_FUNC_get_hash_recalc:
7432 return &bpf_get_hash_recalc_proto;
7433 case BPF_FUNC_perf_event_output:
7434 return &bpf_skb_event_output_proto;
7435 case BPF_FUNC_get_smp_processor_id:
7436 return &bpf_get_smp_processor_id_proto;
7437 case BPF_FUNC_skb_under_cgroup:
7438 return &bpf_skb_under_cgroup_proto;
7440 return bpf_sk_base_func_proto(func_id);
7444 static const struct bpf_func_proto *
7445 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7448 case BPF_FUNC_lwt_push_encap:
7449 return &bpf_lwt_in_push_encap_proto;
7451 return lwt_out_func_proto(func_id, prog);
7455 static const struct bpf_func_proto *
7456 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7459 case BPF_FUNC_skb_get_tunnel_key:
7460 return &bpf_skb_get_tunnel_key_proto;
7461 case BPF_FUNC_skb_set_tunnel_key:
7462 return bpf_get_skb_set_tunnel_proto(func_id);
7463 case BPF_FUNC_skb_get_tunnel_opt:
7464 return &bpf_skb_get_tunnel_opt_proto;
7465 case BPF_FUNC_skb_set_tunnel_opt:
7466 return bpf_get_skb_set_tunnel_proto(func_id);
7467 case BPF_FUNC_redirect:
7468 return &bpf_redirect_proto;
7469 case BPF_FUNC_clone_redirect:
7470 return &bpf_clone_redirect_proto;
7471 case BPF_FUNC_skb_change_tail:
7472 return &bpf_skb_change_tail_proto;
7473 case BPF_FUNC_skb_change_head:
7474 return &bpf_skb_change_head_proto;
7475 case BPF_FUNC_skb_store_bytes:
7476 return &bpf_skb_store_bytes_proto;
7477 case BPF_FUNC_csum_update:
7478 return &bpf_csum_update_proto;
7479 case BPF_FUNC_csum_level:
7480 return &bpf_csum_level_proto;
7481 case BPF_FUNC_l3_csum_replace:
7482 return &bpf_l3_csum_replace_proto;
7483 case BPF_FUNC_l4_csum_replace:
7484 return &bpf_l4_csum_replace_proto;
7485 case BPF_FUNC_set_hash_invalid:
7486 return &bpf_set_hash_invalid_proto;
7487 case BPF_FUNC_lwt_push_encap:
7488 return &bpf_lwt_xmit_push_encap_proto;
7490 return lwt_out_func_proto(func_id, prog);
7494 static const struct bpf_func_proto *
7495 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7498 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7499 case BPF_FUNC_lwt_seg6_store_bytes:
7500 return &bpf_lwt_seg6_store_bytes_proto;
7501 case BPF_FUNC_lwt_seg6_action:
7502 return &bpf_lwt_seg6_action_proto;
7503 case BPF_FUNC_lwt_seg6_adjust_srh:
7504 return &bpf_lwt_seg6_adjust_srh_proto;
7507 return lwt_out_func_proto(func_id, prog);
7511 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
7512 const struct bpf_prog *prog,
7513 struct bpf_insn_access_aux *info)
7515 const int size_default = sizeof(__u32);
7517 if (off < 0 || off >= sizeof(struct __sk_buff))
7520 /* The verifier guarantees that size > 0. */
7521 if (off % size != 0)
7525 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7526 if (off + size > offsetofend(struct __sk_buff, cb[4]))
7529 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
7530 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
7531 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
7532 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
7533 case bpf_ctx_range(struct __sk_buff, data):
7534 case bpf_ctx_range(struct __sk_buff, data_meta):
7535 case bpf_ctx_range(struct __sk_buff, data_end):
7536 if (size != size_default)
7539 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
7541 case bpf_ctx_range(struct __sk_buff, tstamp):
7542 if (size != sizeof(__u64))
7545 case offsetof(struct __sk_buff, sk):
7546 if (type == BPF_WRITE || size != sizeof(__u64))
7548 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
7551 /* Only narrow read access allowed for now. */
7552 if (type == BPF_WRITE) {
7553 if (size != size_default)
7556 bpf_ctx_record_field_size(info, size_default);
7557 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
7565 static bool sk_filter_is_valid_access(int off, int size,
7566 enum bpf_access_type type,
7567 const struct bpf_prog *prog,
7568 struct bpf_insn_access_aux *info)
7571 case bpf_ctx_range(struct __sk_buff, tc_classid):
7572 case bpf_ctx_range(struct __sk_buff, data):
7573 case bpf_ctx_range(struct __sk_buff, data_meta):
7574 case bpf_ctx_range(struct __sk_buff, data_end):
7575 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7576 case bpf_ctx_range(struct __sk_buff, tstamp):
7577 case bpf_ctx_range(struct __sk_buff, wire_len):
7581 if (type == BPF_WRITE) {
7583 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7590 return bpf_skb_is_valid_access(off, size, type, prog, info);
7593 static bool cg_skb_is_valid_access(int off, int size,
7594 enum bpf_access_type type,
7595 const struct bpf_prog *prog,
7596 struct bpf_insn_access_aux *info)
7599 case bpf_ctx_range(struct __sk_buff, tc_classid):
7600 case bpf_ctx_range(struct __sk_buff, data_meta):
7601 case bpf_ctx_range(struct __sk_buff, wire_len):
7603 case bpf_ctx_range(struct __sk_buff, data):
7604 case bpf_ctx_range(struct __sk_buff, data_end):
7610 if (type == BPF_WRITE) {
7612 case bpf_ctx_range(struct __sk_buff, mark):
7613 case bpf_ctx_range(struct __sk_buff, priority):
7614 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7616 case bpf_ctx_range(struct __sk_buff, tstamp):
7626 case bpf_ctx_range(struct __sk_buff, data):
7627 info->reg_type = PTR_TO_PACKET;
7629 case bpf_ctx_range(struct __sk_buff, data_end):
7630 info->reg_type = PTR_TO_PACKET_END;
7634 return bpf_skb_is_valid_access(off, size, type, prog, info);
7637 static bool lwt_is_valid_access(int off, int size,
7638 enum bpf_access_type type,
7639 const struct bpf_prog *prog,
7640 struct bpf_insn_access_aux *info)
7643 case bpf_ctx_range(struct __sk_buff, tc_classid):
7644 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7645 case bpf_ctx_range(struct __sk_buff, data_meta):
7646 case bpf_ctx_range(struct __sk_buff, tstamp):
7647 case bpf_ctx_range(struct __sk_buff, wire_len):
7651 if (type == BPF_WRITE) {
7653 case bpf_ctx_range(struct __sk_buff, mark):
7654 case bpf_ctx_range(struct __sk_buff, priority):
7655 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7663 case bpf_ctx_range(struct __sk_buff, data):
7664 info->reg_type = PTR_TO_PACKET;
7666 case bpf_ctx_range(struct __sk_buff, data_end):
7667 info->reg_type = PTR_TO_PACKET_END;
7671 return bpf_skb_is_valid_access(off, size, type, prog, info);
7674 /* Attach type specific accesses */
7675 static bool __sock_filter_check_attach_type(int off,
7676 enum bpf_access_type access_type,
7677 enum bpf_attach_type attach_type)
7680 case offsetof(struct bpf_sock, bound_dev_if):
7681 case offsetof(struct bpf_sock, mark):
7682 case offsetof(struct bpf_sock, priority):
7683 switch (attach_type) {
7684 case BPF_CGROUP_INET_SOCK_CREATE:
7685 case BPF_CGROUP_INET_SOCK_RELEASE:
7690 case bpf_ctx_range(struct bpf_sock, src_ip4):
7691 switch (attach_type) {
7692 case BPF_CGROUP_INET4_POST_BIND:
7697 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7698 switch (attach_type) {
7699 case BPF_CGROUP_INET6_POST_BIND:
7704 case bpf_ctx_range(struct bpf_sock, src_port):
7705 switch (attach_type) {
7706 case BPF_CGROUP_INET4_POST_BIND:
7707 case BPF_CGROUP_INET6_POST_BIND:
7714 return access_type == BPF_READ;
7719 bool bpf_sock_common_is_valid_access(int off, int size,
7720 enum bpf_access_type type,
7721 struct bpf_insn_access_aux *info)
7724 case bpf_ctx_range_till(struct bpf_sock, type, priority):
7727 return bpf_sock_is_valid_access(off, size, type, info);
7731 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7732 struct bpf_insn_access_aux *info)
7734 const int size_default = sizeof(__u32);
7736 if (off < 0 || off >= sizeof(struct bpf_sock))
7738 if (off % size != 0)
7742 case offsetof(struct bpf_sock, state):
7743 case offsetof(struct bpf_sock, family):
7744 case offsetof(struct bpf_sock, type):
7745 case offsetof(struct bpf_sock, protocol):
7746 case offsetof(struct bpf_sock, dst_port):
7747 case offsetof(struct bpf_sock, src_port):
7748 case offsetof(struct bpf_sock, rx_queue_mapping):
7749 case bpf_ctx_range(struct bpf_sock, src_ip4):
7750 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7751 case bpf_ctx_range(struct bpf_sock, dst_ip4):
7752 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
7753 bpf_ctx_record_field_size(info, size_default);
7754 return bpf_ctx_narrow_access_ok(off, size, size_default);
7757 return size == size_default;
7760 static bool sock_filter_is_valid_access(int off, int size,
7761 enum bpf_access_type type,
7762 const struct bpf_prog *prog,
7763 struct bpf_insn_access_aux *info)
7765 if (!bpf_sock_is_valid_access(off, size, type, info))
7767 return __sock_filter_check_attach_type(off, type,
7768 prog->expected_attach_type);
7771 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
7772 const struct bpf_prog *prog)
7774 /* Neither direct read nor direct write requires any preliminary
7780 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
7781 const struct bpf_prog *prog, int drop_verdict)
7783 struct bpf_insn *insn = insn_buf;
7788 /* if (!skb->cloned)
7791 * (Fast-path, otherwise approximation that we might be
7792 * a clone, do the rest in helper.)
7794 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
7795 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
7796 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
7798 /* ret = bpf_skb_pull_data(skb, 0); */
7799 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
7800 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
7801 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
7802 BPF_FUNC_skb_pull_data);
7805 * return TC_ACT_SHOT;
7807 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
7808 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
7809 *insn++ = BPF_EXIT_INSN();
7812 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
7814 *insn++ = prog->insnsi[0];
7816 return insn - insn_buf;
7819 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
7820 struct bpf_insn *insn_buf)
7822 bool indirect = BPF_MODE(orig->code) == BPF_IND;
7823 struct bpf_insn *insn = insn_buf;
7826 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
7828 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
7830 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
7832 /* We're guaranteed here that CTX is in R6. */
7833 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
7835 switch (BPF_SIZE(orig->code)) {
7837 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
7840 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
7843 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
7847 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
7848 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
7849 *insn++ = BPF_EXIT_INSN();
7851 return insn - insn_buf;
7854 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
7855 const struct bpf_prog *prog)
7857 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
7860 static bool tc_cls_act_is_valid_access(int off, int size,
7861 enum bpf_access_type type,
7862 const struct bpf_prog *prog,
7863 struct bpf_insn_access_aux *info)
7865 if (type == BPF_WRITE) {
7867 case bpf_ctx_range(struct __sk_buff, mark):
7868 case bpf_ctx_range(struct __sk_buff, tc_index):
7869 case bpf_ctx_range(struct __sk_buff, priority):
7870 case bpf_ctx_range(struct __sk_buff, tc_classid):
7871 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7872 case bpf_ctx_range(struct __sk_buff, tstamp):
7873 case bpf_ctx_range(struct __sk_buff, queue_mapping):
7881 case bpf_ctx_range(struct __sk_buff, data):
7882 info->reg_type = PTR_TO_PACKET;
7884 case bpf_ctx_range(struct __sk_buff, data_meta):
7885 info->reg_type = PTR_TO_PACKET_META;
7887 case bpf_ctx_range(struct __sk_buff, data_end):
7888 info->reg_type = PTR_TO_PACKET_END;
7890 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7894 return bpf_skb_is_valid_access(off, size, type, prog, info);
7897 static bool __is_valid_xdp_access(int off, int size)
7899 if (off < 0 || off >= sizeof(struct xdp_md))
7901 if (off % size != 0)
7903 if (size != sizeof(__u32))
7909 static bool xdp_is_valid_access(int off, int size,
7910 enum bpf_access_type type,
7911 const struct bpf_prog *prog,
7912 struct bpf_insn_access_aux *info)
7914 if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
7916 case offsetof(struct xdp_md, egress_ifindex):
7921 if (type == BPF_WRITE) {
7922 if (bpf_prog_is_dev_bound(prog->aux)) {
7924 case offsetof(struct xdp_md, rx_queue_index):
7925 return __is_valid_xdp_access(off, size);
7932 case offsetof(struct xdp_md, data):
7933 info->reg_type = PTR_TO_PACKET;
7935 case offsetof(struct xdp_md, data_meta):
7936 info->reg_type = PTR_TO_PACKET_META;
7938 case offsetof(struct xdp_md, data_end):
7939 info->reg_type = PTR_TO_PACKET_END;
7943 return __is_valid_xdp_access(off, size);
7946 void bpf_warn_invalid_xdp_action(u32 act)
7948 const u32 act_max = XDP_REDIRECT;
7950 WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
7951 act > act_max ? "Illegal" : "Driver unsupported",
7954 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
7956 static bool sock_addr_is_valid_access(int off, int size,
7957 enum bpf_access_type type,
7958 const struct bpf_prog *prog,
7959 struct bpf_insn_access_aux *info)
7961 const int size_default = sizeof(__u32);
7963 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
7965 if (off % size != 0)
7968 /* Disallow access to IPv6 fields from IPv4 contex and vise
7972 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
7973 switch (prog->expected_attach_type) {
7974 case BPF_CGROUP_INET4_BIND:
7975 case BPF_CGROUP_INET4_CONNECT:
7976 case BPF_CGROUP_INET4_GETPEERNAME:
7977 case BPF_CGROUP_INET4_GETSOCKNAME:
7978 case BPF_CGROUP_UDP4_SENDMSG:
7979 case BPF_CGROUP_UDP4_RECVMSG:
7985 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
7986 switch (prog->expected_attach_type) {
7987 case BPF_CGROUP_INET6_BIND:
7988 case BPF_CGROUP_INET6_CONNECT:
7989 case BPF_CGROUP_INET6_GETPEERNAME:
7990 case BPF_CGROUP_INET6_GETSOCKNAME:
7991 case BPF_CGROUP_UDP6_SENDMSG:
7992 case BPF_CGROUP_UDP6_RECVMSG:
7998 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
7999 switch (prog->expected_attach_type) {
8000 case BPF_CGROUP_UDP4_SENDMSG:
8006 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8008 switch (prog->expected_attach_type) {
8009 case BPF_CGROUP_UDP6_SENDMSG:
8018 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8019 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8020 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8021 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8023 case bpf_ctx_range(struct bpf_sock_addr, user_port):
8024 if (type == BPF_READ) {
8025 bpf_ctx_record_field_size(info, size_default);
8027 if (bpf_ctx_wide_access_ok(off, size,
8028 struct bpf_sock_addr,
8032 if (bpf_ctx_wide_access_ok(off, size,
8033 struct bpf_sock_addr,
8037 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8040 if (bpf_ctx_wide_access_ok(off, size,
8041 struct bpf_sock_addr,
8045 if (bpf_ctx_wide_access_ok(off, size,
8046 struct bpf_sock_addr,
8050 if (size != size_default)
8054 case offsetof(struct bpf_sock_addr, sk):
8055 if (type != BPF_READ)
8057 if (size != sizeof(__u64))
8059 info->reg_type = PTR_TO_SOCKET;
8062 if (type == BPF_READ) {
8063 if (size != size_default)
8073 static bool sock_ops_is_valid_access(int off, int size,
8074 enum bpf_access_type type,
8075 const struct bpf_prog *prog,
8076 struct bpf_insn_access_aux *info)
8078 const int size_default = sizeof(__u32);
8080 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
8083 /* The verifier guarantees that size > 0. */
8084 if (off % size != 0)
8087 if (type == BPF_WRITE) {
8089 case offsetof(struct bpf_sock_ops, reply):
8090 case offsetof(struct bpf_sock_ops, sk_txhash):
8091 if (size != size_default)
8099 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
8101 if (size != sizeof(__u64))
8104 case offsetof(struct bpf_sock_ops, sk):
8105 if (size != sizeof(__u64))
8107 info->reg_type = PTR_TO_SOCKET_OR_NULL;
8109 case offsetof(struct bpf_sock_ops, skb_data):
8110 if (size != sizeof(__u64))
8112 info->reg_type = PTR_TO_PACKET;
8114 case offsetof(struct bpf_sock_ops, skb_data_end):
8115 if (size != sizeof(__u64))
8117 info->reg_type = PTR_TO_PACKET_END;
8119 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
8120 bpf_ctx_record_field_size(info, size_default);
8121 return bpf_ctx_narrow_access_ok(off, size,
8124 if (size != size_default)
8133 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
8134 const struct bpf_prog *prog)
8136 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
8139 static bool sk_skb_is_valid_access(int off, int size,
8140 enum bpf_access_type type,
8141 const struct bpf_prog *prog,
8142 struct bpf_insn_access_aux *info)
8145 case bpf_ctx_range(struct __sk_buff, tc_classid):
8146 case bpf_ctx_range(struct __sk_buff, data_meta):
8147 case bpf_ctx_range(struct __sk_buff, tstamp):
8148 case bpf_ctx_range(struct __sk_buff, wire_len):
8152 if (type == BPF_WRITE) {
8154 case bpf_ctx_range(struct __sk_buff, tc_index):
8155 case bpf_ctx_range(struct __sk_buff, priority):
8163 case bpf_ctx_range(struct __sk_buff, mark):
8165 case bpf_ctx_range(struct __sk_buff, data):
8166 info->reg_type = PTR_TO_PACKET;
8168 case bpf_ctx_range(struct __sk_buff, data_end):
8169 info->reg_type = PTR_TO_PACKET_END;
8173 return bpf_skb_is_valid_access(off, size, type, prog, info);
8176 static bool sk_msg_is_valid_access(int off, int size,
8177 enum bpf_access_type type,
8178 const struct bpf_prog *prog,
8179 struct bpf_insn_access_aux *info)
8181 if (type == BPF_WRITE)
8184 if (off % size != 0)
8188 case offsetof(struct sk_msg_md, data):
8189 info->reg_type = PTR_TO_PACKET;
8190 if (size != sizeof(__u64))
8193 case offsetof(struct sk_msg_md, data_end):
8194 info->reg_type = PTR_TO_PACKET_END;
8195 if (size != sizeof(__u64))
8198 case offsetof(struct sk_msg_md, sk):
8199 if (size != sizeof(__u64))
8201 info->reg_type = PTR_TO_SOCKET;
8203 case bpf_ctx_range(struct sk_msg_md, family):
8204 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
8205 case bpf_ctx_range(struct sk_msg_md, local_ip4):
8206 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
8207 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
8208 case bpf_ctx_range(struct sk_msg_md, remote_port):
8209 case bpf_ctx_range(struct sk_msg_md, local_port):
8210 case bpf_ctx_range(struct sk_msg_md, size):
8211 if (size != sizeof(__u32))
8220 static bool flow_dissector_is_valid_access(int off, int size,
8221 enum bpf_access_type type,
8222 const struct bpf_prog *prog,
8223 struct bpf_insn_access_aux *info)
8225 const int size_default = sizeof(__u32);
8227 if (off < 0 || off >= sizeof(struct __sk_buff))
8230 if (type == BPF_WRITE)
8234 case bpf_ctx_range(struct __sk_buff, data):
8235 if (size != size_default)
8237 info->reg_type = PTR_TO_PACKET;
8239 case bpf_ctx_range(struct __sk_buff, data_end):
8240 if (size != size_default)
8242 info->reg_type = PTR_TO_PACKET_END;
8244 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8245 if (size != sizeof(__u64))
8247 info->reg_type = PTR_TO_FLOW_KEYS;
8254 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
8255 const struct bpf_insn *si,
8256 struct bpf_insn *insn_buf,
8257 struct bpf_prog *prog,
8261 struct bpf_insn *insn = insn_buf;
8264 case offsetof(struct __sk_buff, data):
8265 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
8266 si->dst_reg, si->src_reg,
8267 offsetof(struct bpf_flow_dissector, data));
8270 case offsetof(struct __sk_buff, data_end):
8271 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
8272 si->dst_reg, si->src_reg,
8273 offsetof(struct bpf_flow_dissector, data_end));
8276 case offsetof(struct __sk_buff, flow_keys):
8277 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
8278 si->dst_reg, si->src_reg,
8279 offsetof(struct bpf_flow_dissector, flow_keys));
8283 return insn - insn_buf;
8286 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si,
8287 struct bpf_insn *insn)
8289 /* si->dst_reg = skb_shinfo(SKB); */
8290 #ifdef NET_SKBUFF_DATA_USES_OFFSET
8291 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8292 BPF_REG_AX, si->src_reg,
8293 offsetof(struct sk_buff, end));
8294 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
8295 si->dst_reg, si->src_reg,
8296 offsetof(struct sk_buff, head));
8297 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
8299 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8300 si->dst_reg, si->src_reg,
8301 offsetof(struct sk_buff, end));
8307 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
8308 const struct bpf_insn *si,
8309 struct bpf_insn *insn_buf,
8310 struct bpf_prog *prog, u32 *target_size)
8312 struct bpf_insn *insn = insn_buf;
8316 case offsetof(struct __sk_buff, len):
8317 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8318 bpf_target_off(struct sk_buff, len, 4,
8322 case offsetof(struct __sk_buff, protocol):
8323 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8324 bpf_target_off(struct sk_buff, protocol, 2,
8328 case offsetof(struct __sk_buff, vlan_proto):
8329 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8330 bpf_target_off(struct sk_buff, vlan_proto, 2,
8334 case offsetof(struct __sk_buff, priority):
8335 if (type == BPF_WRITE)
8336 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8337 bpf_target_off(struct sk_buff, priority, 4,
8340 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8341 bpf_target_off(struct sk_buff, priority, 4,
8345 case offsetof(struct __sk_buff, ingress_ifindex):
8346 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8347 bpf_target_off(struct sk_buff, skb_iif, 4,
8351 case offsetof(struct __sk_buff, ifindex):
8352 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8353 si->dst_reg, si->src_reg,
8354 offsetof(struct sk_buff, dev));
8355 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
8356 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8357 bpf_target_off(struct net_device, ifindex, 4,
8361 case offsetof(struct __sk_buff, hash):
8362 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8363 bpf_target_off(struct sk_buff, hash, 4,
8367 case offsetof(struct __sk_buff, mark):
8368 if (type == BPF_WRITE)
8369 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8370 bpf_target_off(struct sk_buff, mark, 4,
8373 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8374 bpf_target_off(struct sk_buff, mark, 4,
8378 case offsetof(struct __sk_buff, pkt_type):
8380 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8382 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
8383 #ifdef __BIG_ENDIAN_BITFIELD
8384 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
8388 case offsetof(struct __sk_buff, queue_mapping):
8389 if (type == BPF_WRITE) {
8390 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
8391 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8392 bpf_target_off(struct sk_buff,
8396 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8397 bpf_target_off(struct sk_buff,
8403 case offsetof(struct __sk_buff, vlan_present):
8405 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8406 PKT_VLAN_PRESENT_OFFSET());
8407 if (PKT_VLAN_PRESENT_BIT)
8408 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
8409 if (PKT_VLAN_PRESENT_BIT < 7)
8410 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
8413 case offsetof(struct __sk_buff, vlan_tci):
8414 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8415 bpf_target_off(struct sk_buff, vlan_tci, 2,
8419 case offsetof(struct __sk_buff, cb[0]) ...
8420 offsetofend(struct __sk_buff, cb[4]) - 1:
8421 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
8422 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
8423 offsetof(struct qdisc_skb_cb, data)) %
8426 prog->cb_access = 1;
8428 off -= offsetof(struct __sk_buff, cb[0]);
8429 off += offsetof(struct sk_buff, cb);
8430 off += offsetof(struct qdisc_skb_cb, data);
8431 if (type == BPF_WRITE)
8432 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
8435 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
8439 case offsetof(struct __sk_buff, tc_classid):
8440 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
8443 off -= offsetof(struct __sk_buff, tc_classid);
8444 off += offsetof(struct sk_buff, cb);
8445 off += offsetof(struct qdisc_skb_cb, tc_classid);
8447 if (type == BPF_WRITE)
8448 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
8451 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
8455 case offsetof(struct __sk_buff, data):
8456 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
8457 si->dst_reg, si->src_reg,
8458 offsetof(struct sk_buff, data));
8461 case offsetof(struct __sk_buff, data_meta):
8463 off -= offsetof(struct __sk_buff, data_meta);
8464 off += offsetof(struct sk_buff, cb);
8465 off += offsetof(struct bpf_skb_data_end, data_meta);
8466 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8470 case offsetof(struct __sk_buff, data_end):
8472 off -= offsetof(struct __sk_buff, data_end);
8473 off += offsetof(struct sk_buff, cb);
8474 off += offsetof(struct bpf_skb_data_end, data_end);
8475 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8479 case offsetof(struct __sk_buff, tc_index):
8480 #ifdef CONFIG_NET_SCHED
8481 if (type == BPF_WRITE)
8482 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8483 bpf_target_off(struct sk_buff, tc_index, 2,
8486 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8487 bpf_target_off(struct sk_buff, tc_index, 2,
8491 if (type == BPF_WRITE)
8492 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
8494 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8498 case offsetof(struct __sk_buff, napi_id):
8499 #if defined(CONFIG_NET_RX_BUSY_POLL)
8500 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8501 bpf_target_off(struct sk_buff, napi_id, 4,
8503 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
8504 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8507 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8510 case offsetof(struct __sk_buff, family):
8511 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
8513 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8514 si->dst_reg, si->src_reg,
8515 offsetof(struct sk_buff, sk));
8516 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8517 bpf_target_off(struct sock_common,
8521 case offsetof(struct __sk_buff, remote_ip4):
8522 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
8524 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8525 si->dst_reg, si->src_reg,
8526 offsetof(struct sk_buff, sk));
8527 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8528 bpf_target_off(struct sock_common,
8532 case offsetof(struct __sk_buff, local_ip4):
8533 BUILD_BUG_ON(sizeof_field(struct sock_common,
8534 skc_rcv_saddr) != 4);
8536 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8537 si->dst_reg, si->src_reg,
8538 offsetof(struct sk_buff, sk));
8539 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8540 bpf_target_off(struct sock_common,
8544 case offsetof(struct __sk_buff, remote_ip6[0]) ...
8545 offsetof(struct __sk_buff, remote_ip6[3]):
8546 #if IS_ENABLED(CONFIG_IPV6)
8547 BUILD_BUG_ON(sizeof_field(struct sock_common,
8548 skc_v6_daddr.s6_addr32[0]) != 4);
8551 off -= offsetof(struct __sk_buff, remote_ip6[0]);
8553 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8554 si->dst_reg, si->src_reg,
8555 offsetof(struct sk_buff, sk));
8556 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8557 offsetof(struct sock_common,
8558 skc_v6_daddr.s6_addr32[0]) +
8561 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8564 case offsetof(struct __sk_buff, local_ip6[0]) ...
8565 offsetof(struct __sk_buff, local_ip6[3]):
8566 #if IS_ENABLED(CONFIG_IPV6)
8567 BUILD_BUG_ON(sizeof_field(struct sock_common,
8568 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8571 off -= offsetof(struct __sk_buff, local_ip6[0]);
8573 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8574 si->dst_reg, si->src_reg,
8575 offsetof(struct sk_buff, sk));
8576 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8577 offsetof(struct sock_common,
8578 skc_v6_rcv_saddr.s6_addr32[0]) +
8581 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8585 case offsetof(struct __sk_buff, remote_port):
8586 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
8588 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8589 si->dst_reg, si->src_reg,
8590 offsetof(struct sk_buff, sk));
8591 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8592 bpf_target_off(struct sock_common,
8595 #ifndef __BIG_ENDIAN_BITFIELD
8596 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8600 case offsetof(struct __sk_buff, local_port):
8601 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
8603 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8604 si->dst_reg, si->src_reg,
8605 offsetof(struct sk_buff, sk));
8606 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8607 bpf_target_off(struct sock_common,
8608 skc_num, 2, target_size));
8611 case offsetof(struct __sk_buff, tstamp):
8612 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
8614 if (type == BPF_WRITE)
8615 *insn++ = BPF_STX_MEM(BPF_DW,
8616 si->dst_reg, si->src_reg,
8617 bpf_target_off(struct sk_buff,
8621 *insn++ = BPF_LDX_MEM(BPF_DW,
8622 si->dst_reg, si->src_reg,
8623 bpf_target_off(struct sk_buff,
8628 case offsetof(struct __sk_buff, gso_segs):
8629 insn = bpf_convert_shinfo_access(si, insn);
8630 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
8631 si->dst_reg, si->dst_reg,
8632 bpf_target_off(struct skb_shared_info,
8636 case offsetof(struct __sk_buff, gso_size):
8637 insn = bpf_convert_shinfo_access(si, insn);
8638 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
8639 si->dst_reg, si->dst_reg,
8640 bpf_target_off(struct skb_shared_info,
8644 case offsetof(struct __sk_buff, wire_len):
8645 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
8648 off -= offsetof(struct __sk_buff, wire_len);
8649 off += offsetof(struct sk_buff, cb);
8650 off += offsetof(struct qdisc_skb_cb, pkt_len);
8652 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
8655 case offsetof(struct __sk_buff, sk):
8656 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8657 si->dst_reg, si->src_reg,
8658 offsetof(struct sk_buff, sk));
8662 return insn - insn_buf;
8665 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
8666 const struct bpf_insn *si,
8667 struct bpf_insn *insn_buf,
8668 struct bpf_prog *prog, u32 *target_size)
8670 struct bpf_insn *insn = insn_buf;
8674 case offsetof(struct bpf_sock, bound_dev_if):
8675 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
8677 if (type == BPF_WRITE)
8678 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8679 offsetof(struct sock, sk_bound_dev_if));
8681 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8682 offsetof(struct sock, sk_bound_dev_if));
8685 case offsetof(struct bpf_sock, mark):
8686 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
8688 if (type == BPF_WRITE)
8689 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8690 offsetof(struct sock, sk_mark));
8692 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8693 offsetof(struct sock, sk_mark));
8696 case offsetof(struct bpf_sock, priority):
8697 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
8699 if (type == BPF_WRITE)
8700 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8701 offsetof(struct sock, sk_priority));
8703 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8704 offsetof(struct sock, sk_priority));
8707 case offsetof(struct bpf_sock, family):
8708 *insn++ = BPF_LDX_MEM(
8709 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
8710 si->dst_reg, si->src_reg,
8711 bpf_target_off(struct sock_common,
8713 sizeof_field(struct sock_common,
8718 case offsetof(struct bpf_sock, type):
8719 *insn++ = BPF_LDX_MEM(
8720 BPF_FIELD_SIZEOF(struct sock, sk_type),
8721 si->dst_reg, si->src_reg,
8722 bpf_target_off(struct sock, sk_type,
8723 sizeof_field(struct sock, sk_type),
8727 case offsetof(struct bpf_sock, protocol):
8728 *insn++ = BPF_LDX_MEM(
8729 BPF_FIELD_SIZEOF(struct sock, sk_protocol),
8730 si->dst_reg, si->src_reg,
8731 bpf_target_off(struct sock, sk_protocol,
8732 sizeof_field(struct sock, sk_protocol),
8736 case offsetof(struct bpf_sock, src_ip4):
8737 *insn++ = BPF_LDX_MEM(
8738 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8739 bpf_target_off(struct sock_common, skc_rcv_saddr,
8740 sizeof_field(struct sock_common,
8745 case offsetof(struct bpf_sock, dst_ip4):
8746 *insn++ = BPF_LDX_MEM(
8747 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8748 bpf_target_off(struct sock_common, skc_daddr,
8749 sizeof_field(struct sock_common,
8754 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8755 #if IS_ENABLED(CONFIG_IPV6)
8757 off -= offsetof(struct bpf_sock, src_ip6[0]);
8758 *insn++ = BPF_LDX_MEM(
8759 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8762 skc_v6_rcv_saddr.s6_addr32[0],
8763 sizeof_field(struct sock_common,
8764 skc_v6_rcv_saddr.s6_addr32[0]),
8765 target_size) + off);
8768 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8772 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8773 #if IS_ENABLED(CONFIG_IPV6)
8775 off -= offsetof(struct bpf_sock, dst_ip6[0]);
8776 *insn++ = BPF_LDX_MEM(
8777 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8778 bpf_target_off(struct sock_common,
8779 skc_v6_daddr.s6_addr32[0],
8780 sizeof_field(struct sock_common,
8781 skc_v6_daddr.s6_addr32[0]),
8782 target_size) + off);
8784 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8789 case offsetof(struct bpf_sock, src_port):
8790 *insn++ = BPF_LDX_MEM(
8791 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
8792 si->dst_reg, si->src_reg,
8793 bpf_target_off(struct sock_common, skc_num,
8794 sizeof_field(struct sock_common,
8799 case offsetof(struct bpf_sock, dst_port):
8800 *insn++ = BPF_LDX_MEM(
8801 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
8802 si->dst_reg, si->src_reg,
8803 bpf_target_off(struct sock_common, skc_dport,
8804 sizeof_field(struct sock_common,
8809 case offsetof(struct bpf_sock, state):
8810 *insn++ = BPF_LDX_MEM(
8811 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
8812 si->dst_reg, si->src_reg,
8813 bpf_target_off(struct sock_common, skc_state,
8814 sizeof_field(struct sock_common,
8818 case offsetof(struct bpf_sock, rx_queue_mapping):
8820 *insn++ = BPF_LDX_MEM(
8821 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
8822 si->dst_reg, si->src_reg,
8823 bpf_target_off(struct sock, sk_rx_queue_mapping,
8824 sizeof_field(struct sock,
8825 sk_rx_queue_mapping),
8827 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
8829 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
8831 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
8837 return insn - insn_buf;
8840 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
8841 const struct bpf_insn *si,
8842 struct bpf_insn *insn_buf,
8843 struct bpf_prog *prog, u32 *target_size)
8845 struct bpf_insn *insn = insn_buf;
8848 case offsetof(struct __sk_buff, ifindex):
8849 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8850 si->dst_reg, si->src_reg,
8851 offsetof(struct sk_buff, dev));
8852 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8853 bpf_target_off(struct net_device, ifindex, 4,
8857 return bpf_convert_ctx_access(type, si, insn_buf, prog,
8861 return insn - insn_buf;
8864 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
8865 const struct bpf_insn *si,
8866 struct bpf_insn *insn_buf,
8867 struct bpf_prog *prog, u32 *target_size)
8869 struct bpf_insn *insn = insn_buf;
8872 case offsetof(struct xdp_md, data):
8873 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
8874 si->dst_reg, si->src_reg,
8875 offsetof(struct xdp_buff, data));
8877 case offsetof(struct xdp_md, data_meta):
8878 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
8879 si->dst_reg, si->src_reg,
8880 offsetof(struct xdp_buff, data_meta));
8882 case offsetof(struct xdp_md, data_end):
8883 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
8884 si->dst_reg, si->src_reg,
8885 offsetof(struct xdp_buff, data_end));
8887 case offsetof(struct xdp_md, ingress_ifindex):
8888 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
8889 si->dst_reg, si->src_reg,
8890 offsetof(struct xdp_buff, rxq));
8891 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
8892 si->dst_reg, si->dst_reg,
8893 offsetof(struct xdp_rxq_info, dev));
8894 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8895 offsetof(struct net_device, ifindex));
8897 case offsetof(struct xdp_md, rx_queue_index):
8898 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
8899 si->dst_reg, si->src_reg,
8900 offsetof(struct xdp_buff, rxq));
8901 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8902 offsetof(struct xdp_rxq_info,
8905 case offsetof(struct xdp_md, egress_ifindex):
8906 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
8907 si->dst_reg, si->src_reg,
8908 offsetof(struct xdp_buff, txq));
8909 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
8910 si->dst_reg, si->dst_reg,
8911 offsetof(struct xdp_txq_info, dev));
8912 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8913 offsetof(struct net_device, ifindex));
8917 return insn - insn_buf;
8920 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
8921 * context Structure, F is Field in context structure that contains a pointer
8922 * to Nested Structure of type NS that has the field NF.
8924 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
8925 * sure that SIZE is not greater than actual size of S.F.NF.
8927 * If offset OFF is provided, the load happens from that offset relative to
8930 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
8932 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
8933 si->src_reg, offsetof(S, F)); \
8934 *insn++ = BPF_LDX_MEM( \
8935 SIZE, si->dst_reg, si->dst_reg, \
8936 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
8941 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
8942 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
8943 BPF_FIELD_SIZEOF(NS, NF), 0)
8945 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
8946 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
8948 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
8949 * "register" since two registers available in convert_ctx_access are not
8950 * enough: we can't override neither SRC, since it contains value to store, nor
8951 * DST since it contains pointer to context that may be used by later
8952 * instructions. But we need a temporary place to save pointer to nested
8953 * structure whose field we want to store to.
8955 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
8957 int tmp_reg = BPF_REG_9; \
8958 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
8960 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
8962 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
8964 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
8965 si->dst_reg, offsetof(S, F)); \
8966 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \
8967 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
8970 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
8974 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
8977 if (type == BPF_WRITE) { \
8978 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
8981 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
8982 S, NS, F, NF, SIZE, OFF); \
8986 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
8987 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
8988 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
8990 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
8991 const struct bpf_insn *si,
8992 struct bpf_insn *insn_buf,
8993 struct bpf_prog *prog, u32 *target_size)
8995 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
8996 struct bpf_insn *insn = insn_buf;
8999 case offsetof(struct bpf_sock_addr, user_family):
9000 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9001 struct sockaddr, uaddr, sa_family);
9004 case offsetof(struct bpf_sock_addr, user_ip4):
9005 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9006 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
9007 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
9010 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9012 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
9013 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9014 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9015 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
9019 case offsetof(struct bpf_sock_addr, user_port):
9020 /* To get port we need to know sa_family first and then treat
9021 * sockaddr as either sockaddr_in or sockaddr_in6.
9022 * Though we can simplify since port field has same offset and
9023 * size in both structures.
9024 * Here we check this invariant and use just one of the
9025 * structures if it's true.
9027 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
9028 offsetof(struct sockaddr_in6, sin6_port));
9029 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
9030 sizeof_field(struct sockaddr_in6, sin6_port));
9031 /* Account for sin6_port being smaller than user_port. */
9032 port_size = min(port_size, BPF_LDST_BYTES(si));
9033 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9034 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9035 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
9038 case offsetof(struct bpf_sock_addr, family):
9039 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9040 struct sock, sk, sk_family);
9043 case offsetof(struct bpf_sock_addr, type):
9044 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9045 struct sock, sk, sk_type);
9048 case offsetof(struct bpf_sock_addr, protocol):
9049 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9050 struct sock, sk, sk_protocol);
9053 case offsetof(struct bpf_sock_addr, msg_src_ip4):
9054 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
9055 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9056 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
9057 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
9060 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9063 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
9064 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
9065 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9066 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
9067 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
9069 case offsetof(struct bpf_sock_addr, sk):
9070 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
9071 si->dst_reg, si->src_reg,
9072 offsetof(struct bpf_sock_addr_kern, sk));
9076 return insn - insn_buf;
9079 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
9080 const struct bpf_insn *si,
9081 struct bpf_insn *insn_buf,
9082 struct bpf_prog *prog,
9085 struct bpf_insn *insn = insn_buf;
9088 /* Helper macro for adding read access to tcp_sock or sock fields. */
9089 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
9091 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \
9092 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
9093 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
9094 if (si->dst_reg == reg || si->src_reg == reg) \
9096 if (si->dst_reg == reg || si->src_reg == reg) \
9098 if (si->dst_reg == si->src_reg) { \
9099 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
9100 offsetof(struct bpf_sock_ops_kern, \
9102 fullsock_reg = reg; \
9105 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9106 struct bpf_sock_ops_kern, \
9108 fullsock_reg, si->src_reg, \
9109 offsetof(struct bpf_sock_ops_kern, \
9111 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
9112 if (si->dst_reg == si->src_reg) \
9113 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9114 offsetof(struct bpf_sock_ops_kern, \
9116 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9117 struct bpf_sock_ops_kern, sk),\
9118 si->dst_reg, si->src_reg, \
9119 offsetof(struct bpf_sock_ops_kern, sk));\
9120 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
9122 si->dst_reg, si->dst_reg, \
9123 offsetof(OBJ, OBJ_FIELD)); \
9124 if (si->dst_reg == si->src_reg) { \
9125 *insn++ = BPF_JMP_A(1); \
9126 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9127 offsetof(struct bpf_sock_ops_kern, \
9132 #define SOCK_OPS_GET_SK() \
9134 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \
9135 if (si->dst_reg == reg || si->src_reg == reg) \
9137 if (si->dst_reg == reg || si->src_reg == reg) \
9139 if (si->dst_reg == si->src_reg) { \
9140 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
9141 offsetof(struct bpf_sock_ops_kern, \
9143 fullsock_reg = reg; \
9146 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9147 struct bpf_sock_ops_kern, \
9149 fullsock_reg, si->src_reg, \
9150 offsetof(struct bpf_sock_ops_kern, \
9152 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
9153 if (si->dst_reg == si->src_reg) \
9154 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9155 offsetof(struct bpf_sock_ops_kern, \
9157 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9158 struct bpf_sock_ops_kern, sk),\
9159 si->dst_reg, si->src_reg, \
9160 offsetof(struct bpf_sock_ops_kern, sk));\
9161 if (si->dst_reg == si->src_reg) { \
9162 *insn++ = BPF_JMP_A(1); \
9163 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9164 offsetof(struct bpf_sock_ops_kern, \
9169 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
9170 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
9172 /* Helper macro for adding write access to tcp_sock or sock fields.
9173 * The macro is called with two registers, dst_reg which contains a pointer
9174 * to ctx (context) and src_reg which contains the value that should be
9175 * stored. However, we need an additional register since we cannot overwrite
9176 * dst_reg because it may be used later in the program.
9177 * Instead we "borrow" one of the other register. We first save its value
9178 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
9179 * it at the end of the macro.
9181 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
9183 int reg = BPF_REG_9; \
9184 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
9185 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
9186 if (si->dst_reg == reg || si->src_reg == reg) \
9188 if (si->dst_reg == reg || si->src_reg == reg) \
9190 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
9191 offsetof(struct bpf_sock_ops_kern, \
9193 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9194 struct bpf_sock_ops_kern, \
9197 offsetof(struct bpf_sock_ops_kern, \
9199 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
9200 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9201 struct bpf_sock_ops_kern, sk),\
9203 offsetof(struct bpf_sock_ops_kern, sk));\
9204 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \
9206 offsetof(OBJ, OBJ_FIELD)); \
9207 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
9208 offsetof(struct bpf_sock_ops_kern, \
9212 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
9214 if (TYPE == BPF_WRITE) \
9215 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
9217 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
9220 if (insn > insn_buf)
9221 return insn - insn_buf;
9224 case offsetof(struct bpf_sock_ops, op):
9225 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9227 si->dst_reg, si->src_reg,
9228 offsetof(struct bpf_sock_ops_kern, op));
9231 case offsetof(struct bpf_sock_ops, replylong[0]) ...
9232 offsetof(struct bpf_sock_ops, replylong[3]):
9233 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
9234 sizeof_field(struct bpf_sock_ops_kern, reply));
9235 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
9236 sizeof_field(struct bpf_sock_ops_kern, replylong));
9238 off -= offsetof(struct bpf_sock_ops, replylong[0]);
9239 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
9240 if (type == BPF_WRITE)
9241 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9244 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9248 case offsetof(struct bpf_sock_ops, family):
9249 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9251 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9252 struct bpf_sock_ops_kern, sk),
9253 si->dst_reg, si->src_reg,
9254 offsetof(struct bpf_sock_ops_kern, sk));
9255 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9256 offsetof(struct sock_common, skc_family));
9259 case offsetof(struct bpf_sock_ops, remote_ip4):
9260 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9262 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9263 struct bpf_sock_ops_kern, sk),
9264 si->dst_reg, si->src_reg,
9265 offsetof(struct bpf_sock_ops_kern, sk));
9266 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9267 offsetof(struct sock_common, skc_daddr));
9270 case offsetof(struct bpf_sock_ops, local_ip4):
9271 BUILD_BUG_ON(sizeof_field(struct sock_common,
9272 skc_rcv_saddr) != 4);
9274 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9275 struct bpf_sock_ops_kern, sk),
9276 si->dst_reg, si->src_reg,
9277 offsetof(struct bpf_sock_ops_kern, sk));
9278 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9279 offsetof(struct sock_common,
9283 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
9284 offsetof(struct bpf_sock_ops, remote_ip6[3]):
9285 #if IS_ENABLED(CONFIG_IPV6)
9286 BUILD_BUG_ON(sizeof_field(struct sock_common,
9287 skc_v6_daddr.s6_addr32[0]) != 4);
9290 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
9291 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9292 struct bpf_sock_ops_kern, sk),
9293 si->dst_reg, si->src_reg,
9294 offsetof(struct bpf_sock_ops_kern, sk));
9295 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9296 offsetof(struct sock_common,
9297 skc_v6_daddr.s6_addr32[0]) +
9300 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9304 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
9305 offsetof(struct bpf_sock_ops, local_ip6[3]):
9306 #if IS_ENABLED(CONFIG_IPV6)
9307 BUILD_BUG_ON(sizeof_field(struct sock_common,
9308 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9311 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
9312 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9313 struct bpf_sock_ops_kern, sk),
9314 si->dst_reg, si->src_reg,
9315 offsetof(struct bpf_sock_ops_kern, sk));
9316 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9317 offsetof(struct sock_common,
9318 skc_v6_rcv_saddr.s6_addr32[0]) +
9321 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9325 case offsetof(struct bpf_sock_ops, remote_port):
9326 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9328 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9329 struct bpf_sock_ops_kern, sk),
9330 si->dst_reg, si->src_reg,
9331 offsetof(struct bpf_sock_ops_kern, sk));
9332 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9333 offsetof(struct sock_common, skc_dport));
9334 #ifndef __BIG_ENDIAN_BITFIELD
9335 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9339 case offsetof(struct bpf_sock_ops, local_port):
9340 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9342 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9343 struct bpf_sock_ops_kern, sk),
9344 si->dst_reg, si->src_reg,
9345 offsetof(struct bpf_sock_ops_kern, sk));
9346 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9347 offsetof(struct sock_common, skc_num));
9350 case offsetof(struct bpf_sock_ops, is_fullsock):
9351 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9352 struct bpf_sock_ops_kern,
9354 si->dst_reg, si->src_reg,
9355 offsetof(struct bpf_sock_ops_kern,
9359 case offsetof(struct bpf_sock_ops, state):
9360 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
9362 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9363 struct bpf_sock_ops_kern, sk),
9364 si->dst_reg, si->src_reg,
9365 offsetof(struct bpf_sock_ops_kern, sk));
9366 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
9367 offsetof(struct sock_common, skc_state));
9370 case offsetof(struct bpf_sock_ops, rtt_min):
9371 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
9372 sizeof(struct minmax));
9373 BUILD_BUG_ON(sizeof(struct minmax) <
9374 sizeof(struct minmax_sample));
9376 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9377 struct bpf_sock_ops_kern, sk),
9378 si->dst_reg, si->src_reg,
9379 offsetof(struct bpf_sock_ops_kern, sk));
9380 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9381 offsetof(struct tcp_sock, rtt_min) +
9382 sizeof_field(struct minmax_sample, t));
9385 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
9386 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
9390 case offsetof(struct bpf_sock_ops, sk_txhash):
9391 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
9394 case offsetof(struct bpf_sock_ops, snd_cwnd):
9395 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
9397 case offsetof(struct bpf_sock_ops, srtt_us):
9398 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
9400 case offsetof(struct bpf_sock_ops, snd_ssthresh):
9401 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
9403 case offsetof(struct bpf_sock_ops, rcv_nxt):
9404 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
9406 case offsetof(struct bpf_sock_ops, snd_nxt):
9407 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
9409 case offsetof(struct bpf_sock_ops, snd_una):
9410 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
9412 case offsetof(struct bpf_sock_ops, mss_cache):
9413 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
9415 case offsetof(struct bpf_sock_ops, ecn_flags):
9416 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
9418 case offsetof(struct bpf_sock_ops, rate_delivered):
9419 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
9421 case offsetof(struct bpf_sock_ops, rate_interval_us):
9422 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
9424 case offsetof(struct bpf_sock_ops, packets_out):
9425 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
9427 case offsetof(struct bpf_sock_ops, retrans_out):
9428 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
9430 case offsetof(struct bpf_sock_ops, total_retrans):
9431 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
9433 case offsetof(struct bpf_sock_ops, segs_in):
9434 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
9436 case offsetof(struct bpf_sock_ops, data_segs_in):
9437 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
9439 case offsetof(struct bpf_sock_ops, segs_out):
9440 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
9442 case offsetof(struct bpf_sock_ops, data_segs_out):
9443 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
9445 case offsetof(struct bpf_sock_ops, lost_out):
9446 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
9448 case offsetof(struct bpf_sock_ops, sacked_out):
9449 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
9451 case offsetof(struct bpf_sock_ops, bytes_received):
9452 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
9454 case offsetof(struct bpf_sock_ops, bytes_acked):
9455 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
9457 case offsetof(struct bpf_sock_ops, sk):
9460 case offsetof(struct bpf_sock_ops, skb_data_end):
9461 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9463 si->dst_reg, si->src_reg,
9464 offsetof(struct bpf_sock_ops_kern,
9467 case offsetof(struct bpf_sock_ops, skb_data):
9468 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9470 si->dst_reg, si->src_reg,
9471 offsetof(struct bpf_sock_ops_kern,
9473 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9474 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9475 si->dst_reg, si->dst_reg,
9476 offsetof(struct sk_buff, data));
9478 case offsetof(struct bpf_sock_ops, skb_len):
9479 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9481 si->dst_reg, si->src_reg,
9482 offsetof(struct bpf_sock_ops_kern,
9484 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9485 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9486 si->dst_reg, si->dst_reg,
9487 offsetof(struct sk_buff, len));
9489 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9490 off = offsetof(struct sk_buff, cb);
9491 off += offsetof(struct tcp_skb_cb, tcp_flags);
9492 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
9493 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9495 si->dst_reg, si->src_reg,
9496 offsetof(struct bpf_sock_ops_kern,
9498 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9499 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
9501 si->dst_reg, si->dst_reg, off);
9504 return insn - insn_buf;
9507 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
9508 const struct bpf_insn *si,
9509 struct bpf_insn *insn_buf,
9510 struct bpf_prog *prog, u32 *target_size)
9512 struct bpf_insn *insn = insn_buf;
9516 case offsetof(struct __sk_buff, data_end):
9518 off -= offsetof(struct __sk_buff, data_end);
9519 off += offsetof(struct sk_buff, cb);
9520 off += offsetof(struct tcp_skb_cb, bpf.data_end);
9521 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9525 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9529 return insn - insn_buf;
9532 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
9533 const struct bpf_insn *si,
9534 struct bpf_insn *insn_buf,
9535 struct bpf_prog *prog, u32 *target_size)
9537 struct bpf_insn *insn = insn_buf;
9538 #if IS_ENABLED(CONFIG_IPV6)
9542 /* convert ctx uses the fact sg element is first in struct */
9543 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
9546 case offsetof(struct sk_msg_md, data):
9547 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
9548 si->dst_reg, si->src_reg,
9549 offsetof(struct sk_msg, data));
9551 case offsetof(struct sk_msg_md, data_end):
9552 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
9553 si->dst_reg, si->src_reg,
9554 offsetof(struct sk_msg, data_end));
9556 case offsetof(struct sk_msg_md, family):
9557 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9559 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9561 si->dst_reg, si->src_reg,
9562 offsetof(struct sk_msg, sk));
9563 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9564 offsetof(struct sock_common, skc_family));
9567 case offsetof(struct sk_msg_md, remote_ip4):
9568 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9570 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9572 si->dst_reg, si->src_reg,
9573 offsetof(struct sk_msg, sk));
9574 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9575 offsetof(struct sock_common, skc_daddr));
9578 case offsetof(struct sk_msg_md, local_ip4):
9579 BUILD_BUG_ON(sizeof_field(struct sock_common,
9580 skc_rcv_saddr) != 4);
9582 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9584 si->dst_reg, si->src_reg,
9585 offsetof(struct sk_msg, sk));
9586 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9587 offsetof(struct sock_common,
9591 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
9592 offsetof(struct sk_msg_md, remote_ip6[3]):
9593 #if IS_ENABLED(CONFIG_IPV6)
9594 BUILD_BUG_ON(sizeof_field(struct sock_common,
9595 skc_v6_daddr.s6_addr32[0]) != 4);
9598 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
9599 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9601 si->dst_reg, si->src_reg,
9602 offsetof(struct sk_msg, sk));
9603 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9604 offsetof(struct sock_common,
9605 skc_v6_daddr.s6_addr32[0]) +
9608 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9612 case offsetof(struct sk_msg_md, local_ip6[0]) ...
9613 offsetof(struct sk_msg_md, local_ip6[3]):
9614 #if IS_ENABLED(CONFIG_IPV6)
9615 BUILD_BUG_ON(sizeof_field(struct sock_common,
9616 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9619 off -= offsetof(struct sk_msg_md, local_ip6[0]);
9620 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9622 si->dst_reg, si->src_reg,
9623 offsetof(struct sk_msg, sk));
9624 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9625 offsetof(struct sock_common,
9626 skc_v6_rcv_saddr.s6_addr32[0]) +
9629 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9633 case offsetof(struct sk_msg_md, remote_port):
9634 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9636 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9638 si->dst_reg, si->src_reg,
9639 offsetof(struct sk_msg, sk));
9640 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9641 offsetof(struct sock_common, skc_dport));
9642 #ifndef __BIG_ENDIAN_BITFIELD
9643 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9647 case offsetof(struct sk_msg_md, local_port):
9648 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9650 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9652 si->dst_reg, si->src_reg,
9653 offsetof(struct sk_msg, sk));
9654 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9655 offsetof(struct sock_common, skc_num));
9658 case offsetof(struct sk_msg_md, size):
9659 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
9660 si->dst_reg, si->src_reg,
9661 offsetof(struct sk_msg_sg, size));
9664 case offsetof(struct sk_msg_md, sk):
9665 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
9666 si->dst_reg, si->src_reg,
9667 offsetof(struct sk_msg, sk));
9671 return insn - insn_buf;
9674 const struct bpf_verifier_ops sk_filter_verifier_ops = {
9675 .get_func_proto = sk_filter_func_proto,
9676 .is_valid_access = sk_filter_is_valid_access,
9677 .convert_ctx_access = bpf_convert_ctx_access,
9678 .gen_ld_abs = bpf_gen_ld_abs,
9681 const struct bpf_prog_ops sk_filter_prog_ops = {
9682 .test_run = bpf_prog_test_run_skb,
9685 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
9686 .get_func_proto = tc_cls_act_func_proto,
9687 .is_valid_access = tc_cls_act_is_valid_access,
9688 .convert_ctx_access = tc_cls_act_convert_ctx_access,
9689 .gen_prologue = tc_cls_act_prologue,
9690 .gen_ld_abs = bpf_gen_ld_abs,
9693 const struct bpf_prog_ops tc_cls_act_prog_ops = {
9694 .test_run = bpf_prog_test_run_skb,
9697 const struct bpf_verifier_ops xdp_verifier_ops = {
9698 .get_func_proto = xdp_func_proto,
9699 .is_valid_access = xdp_is_valid_access,
9700 .convert_ctx_access = xdp_convert_ctx_access,
9701 .gen_prologue = bpf_noop_prologue,
9704 const struct bpf_prog_ops xdp_prog_ops = {
9705 .test_run = bpf_prog_test_run_xdp,
9708 const struct bpf_verifier_ops cg_skb_verifier_ops = {
9709 .get_func_proto = cg_skb_func_proto,
9710 .is_valid_access = cg_skb_is_valid_access,
9711 .convert_ctx_access = bpf_convert_ctx_access,
9714 const struct bpf_prog_ops cg_skb_prog_ops = {
9715 .test_run = bpf_prog_test_run_skb,
9718 const struct bpf_verifier_ops lwt_in_verifier_ops = {
9719 .get_func_proto = lwt_in_func_proto,
9720 .is_valid_access = lwt_is_valid_access,
9721 .convert_ctx_access = bpf_convert_ctx_access,
9724 const struct bpf_prog_ops lwt_in_prog_ops = {
9725 .test_run = bpf_prog_test_run_skb,
9728 const struct bpf_verifier_ops lwt_out_verifier_ops = {
9729 .get_func_proto = lwt_out_func_proto,
9730 .is_valid_access = lwt_is_valid_access,
9731 .convert_ctx_access = bpf_convert_ctx_access,
9734 const struct bpf_prog_ops lwt_out_prog_ops = {
9735 .test_run = bpf_prog_test_run_skb,
9738 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
9739 .get_func_proto = lwt_xmit_func_proto,
9740 .is_valid_access = lwt_is_valid_access,
9741 .convert_ctx_access = bpf_convert_ctx_access,
9742 .gen_prologue = tc_cls_act_prologue,
9745 const struct bpf_prog_ops lwt_xmit_prog_ops = {
9746 .test_run = bpf_prog_test_run_skb,
9749 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
9750 .get_func_proto = lwt_seg6local_func_proto,
9751 .is_valid_access = lwt_is_valid_access,
9752 .convert_ctx_access = bpf_convert_ctx_access,
9755 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
9756 .test_run = bpf_prog_test_run_skb,
9759 const struct bpf_verifier_ops cg_sock_verifier_ops = {
9760 .get_func_proto = sock_filter_func_proto,
9761 .is_valid_access = sock_filter_is_valid_access,
9762 .convert_ctx_access = bpf_sock_convert_ctx_access,
9765 const struct bpf_prog_ops cg_sock_prog_ops = {
9768 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
9769 .get_func_proto = sock_addr_func_proto,
9770 .is_valid_access = sock_addr_is_valid_access,
9771 .convert_ctx_access = sock_addr_convert_ctx_access,
9774 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
9777 const struct bpf_verifier_ops sock_ops_verifier_ops = {
9778 .get_func_proto = sock_ops_func_proto,
9779 .is_valid_access = sock_ops_is_valid_access,
9780 .convert_ctx_access = sock_ops_convert_ctx_access,
9783 const struct bpf_prog_ops sock_ops_prog_ops = {
9786 const struct bpf_verifier_ops sk_skb_verifier_ops = {
9787 .get_func_proto = sk_skb_func_proto,
9788 .is_valid_access = sk_skb_is_valid_access,
9789 .convert_ctx_access = sk_skb_convert_ctx_access,
9790 .gen_prologue = sk_skb_prologue,
9793 const struct bpf_prog_ops sk_skb_prog_ops = {
9796 const struct bpf_verifier_ops sk_msg_verifier_ops = {
9797 .get_func_proto = sk_msg_func_proto,
9798 .is_valid_access = sk_msg_is_valid_access,
9799 .convert_ctx_access = sk_msg_convert_ctx_access,
9800 .gen_prologue = bpf_noop_prologue,
9803 const struct bpf_prog_ops sk_msg_prog_ops = {
9806 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
9807 .get_func_proto = flow_dissector_func_proto,
9808 .is_valid_access = flow_dissector_is_valid_access,
9809 .convert_ctx_access = flow_dissector_convert_ctx_access,
9812 const struct bpf_prog_ops flow_dissector_prog_ops = {
9813 .test_run = bpf_prog_test_run_flow_dissector,
9816 int sk_detach_filter(struct sock *sk)
9819 struct sk_filter *filter;
9821 if (sock_flag(sk, SOCK_FILTER_LOCKED))
9824 filter = rcu_dereference_protected(sk->sk_filter,
9825 lockdep_sock_is_held(sk));
9827 RCU_INIT_POINTER(sk->sk_filter, NULL);
9828 sk_filter_uncharge(sk, filter);
9834 EXPORT_SYMBOL_GPL(sk_detach_filter);
9836 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
9839 struct sock_fprog_kern *fprog;
9840 struct sk_filter *filter;
9844 filter = rcu_dereference_protected(sk->sk_filter,
9845 lockdep_sock_is_held(sk));
9849 /* We're copying the filter that has been originally attached,
9850 * so no conversion/decode needed anymore. eBPF programs that
9851 * have no original program cannot be dumped through this.
9854 fprog = filter->prog->orig_prog;
9860 /* User space only enquires number of filter blocks. */
9864 if (len < fprog->len)
9868 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
9871 /* Instead of bytes, the API requests to return the number
9881 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
9882 struct sock_reuseport *reuse,
9883 struct sock *sk, struct sk_buff *skb,
9886 reuse_kern->skb = skb;
9887 reuse_kern->sk = sk;
9888 reuse_kern->selected_sk = NULL;
9889 reuse_kern->data_end = skb->data + skb_headlen(skb);
9890 reuse_kern->hash = hash;
9891 reuse_kern->reuseport_id = reuse->reuseport_id;
9892 reuse_kern->bind_inany = reuse->bind_inany;
9895 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
9896 struct bpf_prog *prog, struct sk_buff *skb,
9899 struct sk_reuseport_kern reuse_kern;
9900 enum sk_action action;
9902 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, hash);
9903 action = BPF_PROG_RUN(prog, &reuse_kern);
9905 if (action == SK_PASS)
9906 return reuse_kern.selected_sk;
9908 return ERR_PTR(-ECONNREFUSED);
9911 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
9912 struct bpf_map *, map, void *, key, u32, flags)
9914 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
9915 struct sock_reuseport *reuse;
9916 struct sock *selected_sk;
9918 selected_sk = map->ops->map_lookup_elem(map, key);
9922 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
9924 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
9925 if (sk_is_refcounted(selected_sk))
9926 sock_put(selected_sk);
9928 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
9929 * The only (!reuse) case here is - the sk has already been
9930 * unhashed (e.g. by close()), so treat it as -ENOENT.
9932 * Other maps (e.g. sock_map) do not provide this guarantee and
9933 * the sk may never be in the reuseport group to begin with.
9935 return is_sockarray ? -ENOENT : -EINVAL;
9938 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
9939 struct sock *sk = reuse_kern->sk;
9941 if (sk->sk_protocol != selected_sk->sk_protocol)
9943 else if (sk->sk_family != selected_sk->sk_family)
9944 return -EAFNOSUPPORT;
9946 /* Catch all. Likely bound to a different sockaddr. */
9950 reuse_kern->selected_sk = selected_sk;
9955 static const struct bpf_func_proto sk_select_reuseport_proto = {
9956 .func = sk_select_reuseport,
9958 .ret_type = RET_INTEGER,
9959 .arg1_type = ARG_PTR_TO_CTX,
9960 .arg2_type = ARG_CONST_MAP_PTR,
9961 .arg3_type = ARG_PTR_TO_MAP_KEY,
9962 .arg4_type = ARG_ANYTHING,
9965 BPF_CALL_4(sk_reuseport_load_bytes,
9966 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
9967 void *, to, u32, len)
9969 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
9972 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
9973 .func = sk_reuseport_load_bytes,
9975 .ret_type = RET_INTEGER,
9976 .arg1_type = ARG_PTR_TO_CTX,
9977 .arg2_type = ARG_ANYTHING,
9978 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
9979 .arg4_type = ARG_CONST_SIZE,
9982 BPF_CALL_5(sk_reuseport_load_bytes_relative,
9983 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
9984 void *, to, u32, len, u32, start_header)
9986 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
9990 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
9991 .func = sk_reuseport_load_bytes_relative,
9993 .ret_type = RET_INTEGER,
9994 .arg1_type = ARG_PTR_TO_CTX,
9995 .arg2_type = ARG_ANYTHING,
9996 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
9997 .arg4_type = ARG_CONST_SIZE,
9998 .arg5_type = ARG_ANYTHING,
10001 static const struct bpf_func_proto *
10002 sk_reuseport_func_proto(enum bpf_func_id func_id,
10003 const struct bpf_prog *prog)
10006 case BPF_FUNC_sk_select_reuseport:
10007 return &sk_select_reuseport_proto;
10008 case BPF_FUNC_skb_load_bytes:
10009 return &sk_reuseport_load_bytes_proto;
10010 case BPF_FUNC_skb_load_bytes_relative:
10011 return &sk_reuseport_load_bytes_relative_proto;
10013 return bpf_base_func_proto(func_id);
10018 sk_reuseport_is_valid_access(int off, int size,
10019 enum bpf_access_type type,
10020 const struct bpf_prog *prog,
10021 struct bpf_insn_access_aux *info)
10023 const u32 size_default = sizeof(__u32);
10025 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
10026 off % size || type != BPF_READ)
10030 case offsetof(struct sk_reuseport_md, data):
10031 info->reg_type = PTR_TO_PACKET;
10032 return size == sizeof(__u64);
10034 case offsetof(struct sk_reuseport_md, data_end):
10035 info->reg_type = PTR_TO_PACKET_END;
10036 return size == sizeof(__u64);
10038 case offsetof(struct sk_reuseport_md, hash):
10039 return size == size_default;
10041 /* Fields that allow narrowing */
10042 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
10043 if (size < sizeof_field(struct sk_buff, protocol))
10046 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
10047 case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
10048 case bpf_ctx_range(struct sk_reuseport_md, len):
10049 bpf_ctx_record_field_size(info, size_default);
10050 return bpf_ctx_narrow_access_ok(off, size, size_default);
10057 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
10058 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
10059 si->dst_reg, si->src_reg, \
10060 bpf_target_off(struct sk_reuseport_kern, F, \
10061 sizeof_field(struct sk_reuseport_kern, F), \
10065 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
10066 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
10071 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \
10072 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
10077 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
10078 const struct bpf_insn *si,
10079 struct bpf_insn *insn_buf,
10080 struct bpf_prog *prog,
10083 struct bpf_insn *insn = insn_buf;
10086 case offsetof(struct sk_reuseport_md, data):
10087 SK_REUSEPORT_LOAD_SKB_FIELD(data);
10090 case offsetof(struct sk_reuseport_md, len):
10091 SK_REUSEPORT_LOAD_SKB_FIELD(len);
10094 case offsetof(struct sk_reuseport_md, eth_protocol):
10095 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
10098 case offsetof(struct sk_reuseport_md, ip_protocol):
10099 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
10102 case offsetof(struct sk_reuseport_md, data_end):
10103 SK_REUSEPORT_LOAD_FIELD(data_end);
10106 case offsetof(struct sk_reuseport_md, hash):
10107 SK_REUSEPORT_LOAD_FIELD(hash);
10110 case offsetof(struct sk_reuseport_md, bind_inany):
10111 SK_REUSEPORT_LOAD_FIELD(bind_inany);
10115 return insn - insn_buf;
10118 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
10119 .get_func_proto = sk_reuseport_func_proto,
10120 .is_valid_access = sk_reuseport_is_valid_access,
10121 .convert_ctx_access = sk_reuseport_convert_ctx_access,
10124 const struct bpf_prog_ops sk_reuseport_prog_ops = {
10127 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
10128 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
10130 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
10131 struct sock *, sk, u64, flags)
10133 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
10134 BPF_SK_LOOKUP_F_NO_REUSEPORT)))
10136 if (unlikely(sk && sk_is_refcounted(sk)))
10137 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
10138 if (unlikely(sk && sk->sk_state == TCP_ESTABLISHED))
10139 return -ESOCKTNOSUPPORT; /* reject connected sockets */
10141 /* Check if socket is suitable for packet L3/L4 protocol */
10142 if (sk && sk->sk_protocol != ctx->protocol)
10143 return -EPROTOTYPE;
10144 if (sk && sk->sk_family != ctx->family &&
10145 (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
10146 return -EAFNOSUPPORT;
10148 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
10151 /* Select socket as lookup result */
10152 ctx->selected_sk = sk;
10153 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
10157 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
10158 .func = bpf_sk_lookup_assign,
10160 .ret_type = RET_INTEGER,
10161 .arg1_type = ARG_PTR_TO_CTX,
10162 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL,
10163 .arg3_type = ARG_ANYTHING,
10166 static const struct bpf_func_proto *
10167 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
10170 case BPF_FUNC_perf_event_output:
10171 return &bpf_event_output_data_proto;
10172 case BPF_FUNC_sk_assign:
10173 return &bpf_sk_lookup_assign_proto;
10174 case BPF_FUNC_sk_release:
10175 return &bpf_sk_release_proto;
10177 return bpf_sk_base_func_proto(func_id);
10181 static bool sk_lookup_is_valid_access(int off, int size,
10182 enum bpf_access_type type,
10183 const struct bpf_prog *prog,
10184 struct bpf_insn_access_aux *info)
10186 if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
10188 if (off % size != 0)
10190 if (type != BPF_READ)
10194 case offsetof(struct bpf_sk_lookup, sk):
10195 info->reg_type = PTR_TO_SOCKET_OR_NULL;
10196 return size == sizeof(__u64);
10198 case bpf_ctx_range(struct bpf_sk_lookup, family):
10199 case bpf_ctx_range(struct bpf_sk_lookup, protocol):
10200 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
10201 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
10202 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
10203 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
10204 case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
10205 case bpf_ctx_range(struct bpf_sk_lookup, local_port):
10206 bpf_ctx_record_field_size(info, sizeof(__u32));
10207 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
10214 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
10215 const struct bpf_insn *si,
10216 struct bpf_insn *insn_buf,
10217 struct bpf_prog *prog,
10220 struct bpf_insn *insn = insn_buf;
10223 case offsetof(struct bpf_sk_lookup, sk):
10224 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10225 offsetof(struct bpf_sk_lookup_kern, selected_sk));
10228 case offsetof(struct bpf_sk_lookup, family):
10229 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10230 bpf_target_off(struct bpf_sk_lookup_kern,
10231 family, 2, target_size));
10234 case offsetof(struct bpf_sk_lookup, protocol):
10235 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10236 bpf_target_off(struct bpf_sk_lookup_kern,
10237 protocol, 2, target_size));
10240 case offsetof(struct bpf_sk_lookup, remote_ip4):
10241 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10242 bpf_target_off(struct bpf_sk_lookup_kern,
10243 v4.saddr, 4, target_size));
10246 case offsetof(struct bpf_sk_lookup, local_ip4):
10247 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10248 bpf_target_off(struct bpf_sk_lookup_kern,
10249 v4.daddr, 4, target_size));
10252 case bpf_ctx_range_till(struct bpf_sk_lookup,
10253 remote_ip6[0], remote_ip6[3]): {
10254 #if IS_ENABLED(CONFIG_IPV6)
10257 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
10258 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10259 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10260 offsetof(struct bpf_sk_lookup_kern, v6.saddr));
10261 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10262 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10264 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10268 case bpf_ctx_range_till(struct bpf_sk_lookup,
10269 local_ip6[0], local_ip6[3]): {
10270 #if IS_ENABLED(CONFIG_IPV6)
10273 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
10274 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10275 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10276 offsetof(struct bpf_sk_lookup_kern, v6.daddr));
10277 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10278 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10280 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10284 case offsetof(struct bpf_sk_lookup, remote_port):
10285 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10286 bpf_target_off(struct bpf_sk_lookup_kern,
10287 sport, 2, target_size));
10290 case offsetof(struct bpf_sk_lookup, local_port):
10291 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10292 bpf_target_off(struct bpf_sk_lookup_kern,
10293 dport, 2, target_size));
10297 return insn - insn_buf;
10300 const struct bpf_prog_ops sk_lookup_prog_ops = {
10303 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
10304 .get_func_proto = sk_lookup_func_proto,
10305 .is_valid_access = sk_lookup_is_valid_access,
10306 .convert_ctx_access = sk_lookup_convert_ctx_access,
10309 #endif /* CONFIG_INET */
10311 DEFINE_BPF_DISPATCHER(xdp)
10313 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
10315 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
10318 #ifdef CONFIG_DEBUG_INFO_BTF
10319 BTF_ID_LIST_GLOBAL(btf_sock_ids)
10320 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
10322 #undef BTF_SOCK_TYPE
10324 u32 btf_sock_ids[MAX_BTF_SOCK_TYPE];
10327 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
10329 /* tcp6_sock type is not generated in dwarf and hence btf,
10330 * trigger an explicit type generation here.
10332 BTF_TYPE_EMIT(struct tcp6_sock);
10333 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
10334 sk->sk_family == AF_INET6)
10335 return (unsigned long)sk;
10337 return (unsigned long)NULL;
10340 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
10341 .func = bpf_skc_to_tcp6_sock,
10343 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10344 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10345 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
10348 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
10350 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
10351 return (unsigned long)sk;
10353 return (unsigned long)NULL;
10356 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
10357 .func = bpf_skc_to_tcp_sock,
10359 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10360 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10361 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
10364 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
10366 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
10367 * generated if CONFIG_INET=n. Trigger an explicit generation here.
10369 BTF_TYPE_EMIT(struct inet_timewait_sock);
10370 BTF_TYPE_EMIT(struct tcp_timewait_sock);
10373 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
10374 return (unsigned long)sk;
10377 #if IS_BUILTIN(CONFIG_IPV6)
10378 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
10379 return (unsigned long)sk;
10382 return (unsigned long)NULL;
10385 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
10386 .func = bpf_skc_to_tcp_timewait_sock,
10388 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10389 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10390 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
10393 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
10396 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10397 return (unsigned long)sk;
10400 #if IS_BUILTIN(CONFIG_IPV6)
10401 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10402 return (unsigned long)sk;
10405 return (unsigned long)NULL;
10408 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
10409 .func = bpf_skc_to_tcp_request_sock,
10411 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10412 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10413 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
10416 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
10418 /* udp6_sock type is not generated in dwarf and hence btf,
10419 * trigger an explicit type generation here.
10421 BTF_TYPE_EMIT(struct udp6_sock);
10422 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
10423 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
10424 return (unsigned long)sk;
10426 return (unsigned long)NULL;
10429 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
10430 .func = bpf_skc_to_udp6_sock,
10432 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10433 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10434 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
10437 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
10439 return (unsigned long)sock_from_file(file);
10442 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
10443 BTF_ID(struct, socket)
10444 BTF_ID(struct, file)
10446 const struct bpf_func_proto bpf_sock_from_file_proto = {
10447 .func = bpf_sock_from_file,
10449 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10450 .ret_btf_id = &bpf_sock_from_file_btf_ids[0],
10451 .arg1_type = ARG_PTR_TO_BTF_ID,
10452 .arg1_btf_id = &bpf_sock_from_file_btf_ids[1],
10455 static const struct bpf_func_proto *
10456 bpf_sk_base_func_proto(enum bpf_func_id func_id)
10458 const struct bpf_func_proto *func;
10461 case BPF_FUNC_skc_to_tcp6_sock:
10462 func = &bpf_skc_to_tcp6_sock_proto;
10464 case BPF_FUNC_skc_to_tcp_sock:
10465 func = &bpf_skc_to_tcp_sock_proto;
10467 case BPF_FUNC_skc_to_tcp_timewait_sock:
10468 func = &bpf_skc_to_tcp_timewait_sock_proto;
10470 case BPF_FUNC_skc_to_tcp_request_sock:
10471 func = &bpf_skc_to_tcp_request_sock_proto;
10473 case BPF_FUNC_skc_to_udp6_sock:
10474 func = &bpf_skc_to_udp6_sock_proto;
10477 return bpf_base_func_proto(func_id);
10480 if (!perfmon_capable())