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)
2170 struct dst_entry *dst = skb_dst(skb);
2171 struct net_device *dev = dst->dev;
2172 u32 hh_len = LL_RESERVED_SPACE(dev);
2173 const struct in6_addr *nexthop;
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);
2199 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2200 &ipv6_hdr(skb)->daddr);
2201 neigh = ip_neigh_gw6(dev, nexthop);
2202 if (likely(!IS_ERR(neigh))) {
2205 sock_confirm_neigh(skb, neigh);
2206 dev_xmit_recursion_inc();
2207 ret = neigh_output(neigh, skb, false);
2208 dev_xmit_recursion_dec();
2209 rcu_read_unlock_bh();
2212 rcu_read_unlock_bh();
2213 IP6_INC_STATS(dev_net(dst->dev),
2214 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2220 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev)
2222 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2223 struct net *net = dev_net(dev);
2224 int err, ret = NET_XMIT_DROP;
2225 struct dst_entry *dst;
2226 struct flowi6 fl6 = {
2227 .flowi6_flags = FLOWI_FLAG_ANYSRC,
2228 .flowi6_mark = skb->mark,
2229 .flowlabel = ip6_flowinfo(ip6h),
2230 .flowi6_oif = dev->ifindex,
2231 .flowi6_proto = ip6h->nexthdr,
2232 .daddr = ip6h->daddr,
2233 .saddr = ip6h->saddr,
2236 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2240 skb_dst_set(skb, dst);
2242 err = bpf_out_neigh_v6(net, skb);
2243 if (unlikely(net_xmit_eval(err)))
2244 dev->stats.tx_errors++;
2246 ret = NET_XMIT_SUCCESS;
2249 dev->stats.tx_errors++;
2255 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev)
2258 return NET_XMIT_DROP;
2260 #endif /* CONFIG_IPV6 */
2262 #if IS_ENABLED(CONFIG_INET)
2263 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb)
2265 struct dst_entry *dst = skb_dst(skb);
2266 struct rtable *rt = container_of(dst, struct rtable, dst);
2267 struct net_device *dev = dst->dev;
2268 u32 hh_len = LL_RESERVED_SPACE(dev);
2269 struct neighbour *neigh;
2270 bool is_v6gw = false;
2272 if (dev_xmit_recursion()) {
2273 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2280 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2281 struct sk_buff *skb2;
2283 skb2 = skb_realloc_headroom(skb, hh_len);
2284 if (unlikely(!skb2)) {
2289 skb_set_owner_w(skb2, skb->sk);
2295 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2296 if (likely(!IS_ERR(neigh))) {
2299 sock_confirm_neigh(skb, neigh);
2300 dev_xmit_recursion_inc();
2301 ret = neigh_output(neigh, skb, is_v6gw);
2302 dev_xmit_recursion_dec();
2303 rcu_read_unlock_bh();
2306 rcu_read_unlock_bh();
2312 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev)
2314 const struct iphdr *ip4h = ip_hdr(skb);
2315 struct net *net = dev_net(dev);
2316 int err, ret = NET_XMIT_DROP;
2318 struct flowi4 fl4 = {
2319 .flowi4_flags = FLOWI_FLAG_ANYSRC,
2320 .flowi4_mark = skb->mark,
2321 .flowi4_tos = RT_TOS(ip4h->tos),
2322 .flowi4_oif = dev->ifindex,
2323 .flowi4_proto = ip4h->protocol,
2324 .daddr = ip4h->daddr,
2325 .saddr = ip4h->saddr,
2328 rt = ip_route_output_flow(net, &fl4, NULL);
2331 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2336 skb_dst_set(skb, &rt->dst);
2338 err = bpf_out_neigh_v4(net, skb);
2339 if (unlikely(net_xmit_eval(err)))
2340 dev->stats.tx_errors++;
2342 ret = NET_XMIT_SUCCESS;
2345 dev->stats.tx_errors++;
2351 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev)
2354 return NET_XMIT_DROP;
2356 #endif /* CONFIG_INET */
2358 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev)
2360 struct ethhdr *ethh = eth_hdr(skb);
2362 if (unlikely(skb->mac_header >= skb->network_header))
2364 bpf_push_mac_rcsum(skb);
2365 if (is_multicast_ether_addr(ethh->h_dest))
2368 skb_pull(skb, sizeof(*ethh));
2369 skb_unset_mac_header(skb);
2370 skb_reset_network_header(skb);
2372 if (skb->protocol == htons(ETH_P_IP))
2373 return __bpf_redirect_neigh_v4(skb, dev);
2374 else if (skb->protocol == htons(ETH_P_IPV6))
2375 return __bpf_redirect_neigh_v6(skb, dev);
2381 /* Internal, non-exposed redirect flags. */
2383 BPF_F_NEIGH = (1ULL << 1),
2384 BPF_F_PEER = (1ULL << 2),
2385 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER)
2388 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2390 struct net_device *dev;
2391 struct sk_buff *clone;
2394 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2397 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2401 clone = skb_clone(skb, GFP_ATOMIC);
2402 if (unlikely(!clone))
2405 /* For direct write, we need to keep the invariant that the skbs
2406 * we're dealing with need to be uncloned. Should uncloning fail
2407 * here, we need to free the just generated clone to unclone once
2410 ret = bpf_try_make_head_writable(skb);
2411 if (unlikely(ret)) {
2416 return __bpf_redirect(clone, dev, flags);
2419 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2420 .func = bpf_clone_redirect,
2422 .ret_type = RET_INTEGER,
2423 .arg1_type = ARG_PTR_TO_CTX,
2424 .arg2_type = ARG_ANYTHING,
2425 .arg3_type = ARG_ANYTHING,
2428 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2429 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2431 int skb_do_redirect(struct sk_buff *skb)
2433 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2434 struct net *net = dev_net(skb->dev);
2435 struct net_device *dev;
2436 u32 flags = ri->flags;
2438 dev = dev_get_by_index_rcu(net, ri->tgt_index);
2443 if (flags & BPF_F_PEER) {
2444 const struct net_device_ops *ops = dev->netdev_ops;
2446 if (unlikely(!ops->ndo_get_peer_dev ||
2447 !skb_at_tc_ingress(skb)))
2449 dev = ops->ndo_get_peer_dev(dev);
2450 if (unlikely(!dev ||
2451 !is_skb_forwardable(dev, skb) ||
2452 net_eq(net, dev_net(dev))))
2457 return flags & BPF_F_NEIGH ?
2458 __bpf_redirect_neigh(skb, dev) :
2459 __bpf_redirect(skb, dev, flags);
2465 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2467 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2469 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2473 ri->tgt_index = ifindex;
2475 return TC_ACT_REDIRECT;
2478 static const struct bpf_func_proto bpf_redirect_proto = {
2479 .func = bpf_redirect,
2481 .ret_type = RET_INTEGER,
2482 .arg1_type = ARG_ANYTHING,
2483 .arg2_type = ARG_ANYTHING,
2486 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2488 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2490 if (unlikely(flags))
2493 ri->flags = BPF_F_PEER;
2494 ri->tgt_index = ifindex;
2496 return TC_ACT_REDIRECT;
2499 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2500 .func = bpf_redirect_peer,
2502 .ret_type = RET_INTEGER,
2503 .arg1_type = ARG_ANYTHING,
2504 .arg2_type = ARG_ANYTHING,
2507 BPF_CALL_2(bpf_redirect_neigh, u32, ifindex, u64, flags)
2509 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2511 if (unlikely(flags))
2514 ri->flags = BPF_F_NEIGH;
2515 ri->tgt_index = ifindex;
2517 return TC_ACT_REDIRECT;
2520 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2521 .func = bpf_redirect_neigh,
2523 .ret_type = RET_INTEGER,
2524 .arg1_type = ARG_ANYTHING,
2525 .arg2_type = ARG_ANYTHING,
2528 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2530 msg->apply_bytes = bytes;
2534 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2535 .func = bpf_msg_apply_bytes,
2537 .ret_type = RET_INTEGER,
2538 .arg1_type = ARG_PTR_TO_CTX,
2539 .arg2_type = ARG_ANYTHING,
2542 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2544 msg->cork_bytes = bytes;
2548 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2549 .func = bpf_msg_cork_bytes,
2551 .ret_type = RET_INTEGER,
2552 .arg1_type = ARG_PTR_TO_CTX,
2553 .arg2_type = ARG_ANYTHING,
2556 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2557 u32, end, u64, flags)
2559 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2560 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2561 struct scatterlist *sge;
2562 u8 *raw, *to, *from;
2565 if (unlikely(flags || end <= start))
2568 /* First find the starting scatterlist element */
2572 len = sk_msg_elem(msg, i)->length;
2573 if (start < offset + len)
2575 sk_msg_iter_var_next(i);
2576 } while (i != msg->sg.end);
2578 if (unlikely(start >= offset + len))
2582 /* The start may point into the sg element so we need to also
2583 * account for the headroom.
2585 bytes_sg_total = start - offset + bytes;
2586 if (!test_bit(i, &msg->sg.copy) && bytes_sg_total <= len)
2589 /* At this point we need to linearize multiple scatterlist
2590 * elements or a single shared page. Either way we need to
2591 * copy into a linear buffer exclusively owned by BPF. Then
2592 * place the buffer in the scatterlist and fixup the original
2593 * entries by removing the entries now in the linear buffer
2594 * and shifting the remaining entries. For now we do not try
2595 * to copy partial entries to avoid complexity of running out
2596 * of sg_entry slots. The downside is reading a single byte
2597 * will copy the entire sg entry.
2600 copy += sk_msg_elem(msg, i)->length;
2601 sk_msg_iter_var_next(i);
2602 if (bytes_sg_total <= copy)
2604 } while (i != msg->sg.end);
2607 if (unlikely(bytes_sg_total > copy))
2610 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2612 if (unlikely(!page))
2615 raw = page_address(page);
2618 sge = sk_msg_elem(msg, i);
2619 from = sg_virt(sge);
2623 memcpy(to, from, len);
2626 put_page(sg_page(sge));
2628 sk_msg_iter_var_next(i);
2629 } while (i != last_sge);
2631 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2633 /* To repair sg ring we need to shift entries. If we only
2634 * had a single entry though we can just replace it and
2635 * be done. Otherwise walk the ring and shift the entries.
2637 WARN_ON_ONCE(last_sge == first_sge);
2638 shift = last_sge > first_sge ?
2639 last_sge - first_sge - 1 :
2640 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2645 sk_msg_iter_var_next(i);
2649 if (i + shift >= NR_MSG_FRAG_IDS)
2650 move_from = i + shift - NR_MSG_FRAG_IDS;
2652 move_from = i + shift;
2653 if (move_from == msg->sg.end)
2656 msg->sg.data[i] = msg->sg.data[move_from];
2657 msg->sg.data[move_from].length = 0;
2658 msg->sg.data[move_from].page_link = 0;
2659 msg->sg.data[move_from].offset = 0;
2660 sk_msg_iter_var_next(i);
2663 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2664 msg->sg.end - shift + NR_MSG_FRAG_IDS :
2665 msg->sg.end - shift;
2667 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2668 msg->data_end = msg->data + bytes;
2672 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2673 .func = bpf_msg_pull_data,
2675 .ret_type = RET_INTEGER,
2676 .arg1_type = ARG_PTR_TO_CTX,
2677 .arg2_type = ARG_ANYTHING,
2678 .arg3_type = ARG_ANYTHING,
2679 .arg4_type = ARG_ANYTHING,
2682 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2683 u32, len, u64, flags)
2685 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2686 u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2687 u8 *raw, *to, *from;
2690 if (unlikely(flags))
2693 /* First find the starting scatterlist element */
2697 l = sk_msg_elem(msg, i)->length;
2699 if (start < offset + l)
2701 sk_msg_iter_var_next(i);
2702 } while (i != msg->sg.end);
2704 if (start >= offset + l)
2707 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2709 /* If no space available will fallback to copy, we need at
2710 * least one scatterlist elem available to push data into
2711 * when start aligns to the beginning of an element or two
2712 * when it falls inside an element. We handle the start equals
2713 * offset case because its the common case for inserting a
2716 if (!space || (space == 1 && start != offset))
2717 copy = msg->sg.data[i].length;
2719 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2720 get_order(copy + len));
2721 if (unlikely(!page))
2727 raw = page_address(page);
2729 psge = sk_msg_elem(msg, i);
2730 front = start - offset;
2731 back = psge->length - front;
2732 from = sg_virt(psge);
2735 memcpy(raw, from, front);
2739 to = raw + front + len;
2741 memcpy(to, from, back);
2744 put_page(sg_page(psge));
2745 } else if (start - offset) {
2746 psge = sk_msg_elem(msg, i);
2747 rsge = sk_msg_elem_cpy(msg, i);
2749 psge->length = start - offset;
2750 rsge.length -= psge->length;
2751 rsge.offset += start;
2753 sk_msg_iter_var_next(i);
2754 sg_unmark_end(psge);
2755 sg_unmark_end(&rsge);
2756 sk_msg_iter_next(msg, end);
2759 /* Slot(s) to place newly allocated data */
2762 /* Shift one or two slots as needed */
2764 sge = sk_msg_elem_cpy(msg, i);
2766 sk_msg_iter_var_next(i);
2767 sg_unmark_end(&sge);
2768 sk_msg_iter_next(msg, end);
2770 nsge = sk_msg_elem_cpy(msg, i);
2772 sk_msg_iter_var_next(i);
2773 nnsge = sk_msg_elem_cpy(msg, i);
2776 while (i != msg->sg.end) {
2777 msg->sg.data[i] = sge;
2779 sk_msg_iter_var_next(i);
2782 nnsge = sk_msg_elem_cpy(msg, i);
2784 nsge = sk_msg_elem_cpy(msg, i);
2789 /* Place newly allocated data buffer */
2790 sk_mem_charge(msg->sk, len);
2791 msg->sg.size += len;
2792 __clear_bit(new, &msg->sg.copy);
2793 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2795 get_page(sg_page(&rsge));
2796 sk_msg_iter_var_next(new);
2797 msg->sg.data[new] = rsge;
2800 sk_msg_compute_data_pointers(msg);
2804 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2805 .func = bpf_msg_push_data,
2807 .ret_type = RET_INTEGER,
2808 .arg1_type = ARG_PTR_TO_CTX,
2809 .arg2_type = ARG_ANYTHING,
2810 .arg3_type = ARG_ANYTHING,
2811 .arg4_type = ARG_ANYTHING,
2814 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2820 sk_msg_iter_var_next(i);
2821 msg->sg.data[prev] = msg->sg.data[i];
2822 } while (i != msg->sg.end);
2824 sk_msg_iter_prev(msg, end);
2827 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2829 struct scatterlist tmp, sge;
2831 sk_msg_iter_next(msg, end);
2832 sge = sk_msg_elem_cpy(msg, i);
2833 sk_msg_iter_var_next(i);
2834 tmp = sk_msg_elem_cpy(msg, i);
2836 while (i != msg->sg.end) {
2837 msg->sg.data[i] = sge;
2838 sk_msg_iter_var_next(i);
2840 tmp = sk_msg_elem_cpy(msg, i);
2844 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2845 u32, len, u64, flags)
2847 u32 i = 0, l = 0, space, offset = 0;
2848 u64 last = start + len;
2851 if (unlikely(flags))
2854 /* First find the starting scatterlist element */
2858 l = sk_msg_elem(msg, i)->length;
2860 if (start < offset + l)
2862 sk_msg_iter_var_next(i);
2863 } while (i != msg->sg.end);
2865 /* Bounds checks: start and pop must be inside message */
2866 if (start >= offset + l || last >= msg->sg.size)
2869 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2872 /* --------------| offset
2873 * -| start |-------- len -------|
2875 * |----- a ----|-------- pop -------|----- b ----|
2876 * |______________________________________________| length
2879 * a: region at front of scatter element to save
2880 * b: region at back of scatter element to save when length > A + pop
2881 * pop: region to pop from element, same as input 'pop' here will be
2882 * decremented below per iteration.
2884 * Two top-level cases to handle when start != offset, first B is non
2885 * zero and second B is zero corresponding to when a pop includes more
2888 * Then if B is non-zero AND there is no space allocate space and
2889 * compact A, B regions into page. If there is space shift ring to
2890 * the rigth free'ing the next element in ring to place B, leaving
2891 * A untouched except to reduce length.
2893 if (start != offset) {
2894 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2896 int b = sge->length - pop - a;
2898 sk_msg_iter_var_next(i);
2900 if (pop < sge->length - a) {
2903 sk_msg_shift_right(msg, i);
2904 nsge = sk_msg_elem(msg, i);
2905 get_page(sg_page(sge));
2908 b, sge->offset + pop + a);
2910 struct page *page, *orig;
2913 page = alloc_pages(__GFP_NOWARN |
2914 __GFP_COMP | GFP_ATOMIC,
2916 if (unlikely(!page))
2920 orig = sg_page(sge);
2921 from = sg_virt(sge);
2922 to = page_address(page);
2923 memcpy(to, from, a);
2924 memcpy(to + a, from + a + pop, b);
2925 sg_set_page(sge, page, a + b, 0);
2929 } else if (pop >= sge->length - a) {
2930 pop -= (sge->length - a);
2935 /* From above the current layout _must_ be as follows,
2940 * |---- pop ---|---------------- b ------------|
2941 * |____________________________________________| length
2943 * Offset and start of the current msg elem are equal because in the
2944 * previous case we handled offset != start and either consumed the
2945 * entire element and advanced to the next element OR pop == 0.
2947 * Two cases to handle here are first pop is less than the length
2948 * leaving some remainder b above. Simply adjust the element's layout
2949 * in this case. Or pop >= length of the element so that b = 0. In this
2950 * case advance to next element decrementing pop.
2953 struct scatterlist *sge = sk_msg_elem(msg, i);
2955 if (pop < sge->length) {
2961 sk_msg_shift_left(msg, i);
2963 sk_msg_iter_var_next(i);
2966 sk_mem_uncharge(msg->sk, len - pop);
2967 msg->sg.size -= (len - pop);
2968 sk_msg_compute_data_pointers(msg);
2972 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
2973 .func = bpf_msg_pop_data,
2975 .ret_type = RET_INTEGER,
2976 .arg1_type = ARG_PTR_TO_CTX,
2977 .arg2_type = ARG_ANYTHING,
2978 .arg3_type = ARG_ANYTHING,
2979 .arg4_type = ARG_ANYTHING,
2982 #ifdef CONFIG_CGROUP_NET_CLASSID
2983 BPF_CALL_0(bpf_get_cgroup_classid_curr)
2985 return __task_get_classid(current);
2988 static const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
2989 .func = bpf_get_cgroup_classid_curr,
2991 .ret_type = RET_INTEGER,
2994 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
2996 struct sock *sk = skb_to_full_sk(skb);
2998 if (!sk || !sk_fullsock(sk))
3001 return sock_cgroup_classid(&sk->sk_cgrp_data);
3004 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3005 .func = bpf_skb_cgroup_classid,
3007 .ret_type = RET_INTEGER,
3008 .arg1_type = ARG_PTR_TO_CTX,
3012 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3014 return task_get_classid(skb);
3017 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3018 .func = bpf_get_cgroup_classid,
3020 .ret_type = RET_INTEGER,
3021 .arg1_type = ARG_PTR_TO_CTX,
3024 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3026 return dst_tclassid(skb);
3029 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3030 .func = bpf_get_route_realm,
3032 .ret_type = RET_INTEGER,
3033 .arg1_type = ARG_PTR_TO_CTX,
3036 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3038 /* If skb_clear_hash() was called due to mangling, we can
3039 * trigger SW recalculation here. Later access to hash
3040 * can then use the inline skb->hash via context directly
3041 * instead of calling this helper again.
3043 return skb_get_hash(skb);
3046 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3047 .func = bpf_get_hash_recalc,
3049 .ret_type = RET_INTEGER,
3050 .arg1_type = ARG_PTR_TO_CTX,
3053 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3055 /* After all direct packet write, this can be used once for
3056 * triggering a lazy recalc on next skb_get_hash() invocation.
3058 skb_clear_hash(skb);
3062 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3063 .func = bpf_set_hash_invalid,
3065 .ret_type = RET_INTEGER,
3066 .arg1_type = ARG_PTR_TO_CTX,
3069 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3071 /* Set user specified hash as L4(+), so that it gets returned
3072 * on skb_get_hash() call unless BPF prog later on triggers a
3075 __skb_set_sw_hash(skb, hash, true);
3079 static const struct bpf_func_proto bpf_set_hash_proto = {
3080 .func = bpf_set_hash,
3082 .ret_type = RET_INTEGER,
3083 .arg1_type = ARG_PTR_TO_CTX,
3084 .arg2_type = ARG_ANYTHING,
3087 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3092 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3093 vlan_proto != htons(ETH_P_8021AD)))
3094 vlan_proto = htons(ETH_P_8021Q);
3096 bpf_push_mac_rcsum(skb);
3097 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3098 bpf_pull_mac_rcsum(skb);
3100 bpf_compute_data_pointers(skb);
3104 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3105 .func = bpf_skb_vlan_push,
3107 .ret_type = RET_INTEGER,
3108 .arg1_type = ARG_PTR_TO_CTX,
3109 .arg2_type = ARG_ANYTHING,
3110 .arg3_type = ARG_ANYTHING,
3113 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3117 bpf_push_mac_rcsum(skb);
3118 ret = skb_vlan_pop(skb);
3119 bpf_pull_mac_rcsum(skb);
3121 bpf_compute_data_pointers(skb);
3125 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3126 .func = bpf_skb_vlan_pop,
3128 .ret_type = RET_INTEGER,
3129 .arg1_type = ARG_PTR_TO_CTX,
3132 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3134 /* Caller already did skb_cow() with len as headroom,
3135 * so no need to do it here.
3138 memmove(skb->data, skb->data + len, off);
3139 memset(skb->data + off, 0, len);
3141 /* No skb_postpush_rcsum(skb, skb->data + off, len)
3142 * needed here as it does not change the skb->csum
3143 * result for checksum complete when summing over
3149 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3151 /* skb_ensure_writable() is not needed here, as we're
3152 * already working on an uncloned skb.
3154 if (unlikely(!pskb_may_pull(skb, off + len)))
3157 skb_postpull_rcsum(skb, skb->data + off, len);
3158 memmove(skb->data + len, skb->data, off);
3159 __skb_pull(skb, len);
3164 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3166 bool trans_same = skb->transport_header == skb->network_header;
3169 /* There's no need for __skb_push()/__skb_pull() pair to
3170 * get to the start of the mac header as we're guaranteed
3171 * to always start from here under eBPF.
3173 ret = bpf_skb_generic_push(skb, off, len);
3175 skb->mac_header -= len;
3176 skb->network_header -= len;
3178 skb->transport_header = skb->network_header;
3184 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3186 bool trans_same = skb->transport_header == skb->network_header;
3189 /* Same here, __skb_push()/__skb_pull() pair not needed. */
3190 ret = bpf_skb_generic_pop(skb, off, len);
3192 skb->mac_header += len;
3193 skb->network_header += len;
3195 skb->transport_header = skb->network_header;
3201 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3203 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3204 u32 off = skb_mac_header_len(skb);
3207 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
3210 ret = skb_cow(skb, len_diff);
3211 if (unlikely(ret < 0))
3214 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3215 if (unlikely(ret < 0))
3218 if (skb_is_gso(skb)) {
3219 struct skb_shared_info *shinfo = skb_shinfo(skb);
3221 /* SKB_GSO_TCPV4 needs to be changed into
3224 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3225 shinfo->gso_type &= ~SKB_GSO_TCPV4;
3226 shinfo->gso_type |= SKB_GSO_TCPV6;
3229 /* Due to IPv6 header, MSS needs to be downgraded. */
3230 skb_decrease_gso_size(shinfo, len_diff);
3231 /* Header must be checked, and gso_segs recomputed. */
3232 shinfo->gso_type |= SKB_GSO_DODGY;
3233 shinfo->gso_segs = 0;
3236 skb->protocol = htons(ETH_P_IPV6);
3237 skb_clear_hash(skb);
3242 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3244 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3245 u32 off = skb_mac_header_len(skb);
3248 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
3251 ret = skb_unclone(skb, GFP_ATOMIC);
3252 if (unlikely(ret < 0))
3255 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3256 if (unlikely(ret < 0))
3259 if (skb_is_gso(skb)) {
3260 struct skb_shared_info *shinfo = skb_shinfo(skb);
3262 /* SKB_GSO_TCPV6 needs to be changed into
3265 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3266 shinfo->gso_type &= ~SKB_GSO_TCPV6;
3267 shinfo->gso_type |= SKB_GSO_TCPV4;
3270 /* Due to IPv4 header, MSS can be upgraded. */
3271 skb_increase_gso_size(shinfo, len_diff);
3272 /* Header must be checked, and gso_segs recomputed. */
3273 shinfo->gso_type |= SKB_GSO_DODGY;
3274 shinfo->gso_segs = 0;
3277 skb->protocol = htons(ETH_P_IP);
3278 skb_clear_hash(skb);
3283 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3285 __be16 from_proto = skb->protocol;
3287 if (from_proto == htons(ETH_P_IP) &&
3288 to_proto == htons(ETH_P_IPV6))
3289 return bpf_skb_proto_4_to_6(skb);
3291 if (from_proto == htons(ETH_P_IPV6) &&
3292 to_proto == htons(ETH_P_IP))
3293 return bpf_skb_proto_6_to_4(skb);
3298 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3303 if (unlikely(flags))
3306 /* General idea is that this helper does the basic groundwork
3307 * needed for changing the protocol, and eBPF program fills the
3308 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3309 * and other helpers, rather than passing a raw buffer here.
3311 * The rationale is to keep this minimal and without a need to
3312 * deal with raw packet data. F.e. even if we would pass buffers
3313 * here, the program still needs to call the bpf_lX_csum_replace()
3314 * helpers anyway. Plus, this way we keep also separation of
3315 * concerns, since f.e. bpf_skb_store_bytes() should only take
3318 * Currently, additional options and extension header space are
3319 * not supported, but flags register is reserved so we can adapt
3320 * that. For offloads, we mark packet as dodgy, so that headers
3321 * need to be verified first.
3323 ret = bpf_skb_proto_xlat(skb, proto);
3324 bpf_compute_data_pointers(skb);
3328 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3329 .func = bpf_skb_change_proto,
3331 .ret_type = RET_INTEGER,
3332 .arg1_type = ARG_PTR_TO_CTX,
3333 .arg2_type = ARG_ANYTHING,
3334 .arg3_type = ARG_ANYTHING,
3337 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3339 /* We only allow a restricted subset to be changed for now. */
3340 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3341 !skb_pkt_type_ok(pkt_type)))
3344 skb->pkt_type = pkt_type;
3348 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3349 .func = bpf_skb_change_type,
3351 .ret_type = RET_INTEGER,
3352 .arg1_type = ARG_PTR_TO_CTX,
3353 .arg2_type = ARG_ANYTHING,
3356 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3358 switch (skb->protocol) {
3359 case htons(ETH_P_IP):
3360 return sizeof(struct iphdr);
3361 case htons(ETH_P_IPV6):
3362 return sizeof(struct ipv6hdr);
3368 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3369 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3371 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \
3372 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3373 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3374 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3375 BPF_F_ADJ_ROOM_ENCAP_L2( \
3376 BPF_ADJ_ROOM_ENCAP_L2_MASK))
3378 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3381 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3382 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3383 u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3384 unsigned int gso_type = SKB_GSO_DODGY;
3387 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3388 /* udp gso_size delineates datagrams, only allow if fixed */
3389 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3390 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3394 ret = skb_cow_head(skb, len_diff);
3395 if (unlikely(ret < 0))
3399 if (skb->protocol != htons(ETH_P_IP) &&
3400 skb->protocol != htons(ETH_P_IPV6))
3403 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3404 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3407 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3408 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3411 if (skb->encapsulation)
3414 mac_len = skb->network_header - skb->mac_header;
3415 inner_net = skb->network_header;
3416 if (inner_mac_len > len_diff)
3418 inner_trans = skb->transport_header;
3421 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3422 if (unlikely(ret < 0))
3426 skb->inner_mac_header = inner_net - inner_mac_len;
3427 skb->inner_network_header = inner_net;
3428 skb->inner_transport_header = inner_trans;
3429 skb_set_inner_protocol(skb, skb->protocol);
3431 skb->encapsulation = 1;
3432 skb_set_network_header(skb, mac_len);
3434 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3435 gso_type |= SKB_GSO_UDP_TUNNEL;
3436 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3437 gso_type |= SKB_GSO_GRE;
3438 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3439 gso_type |= SKB_GSO_IPXIP6;
3440 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3441 gso_type |= SKB_GSO_IPXIP4;
3443 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3444 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3445 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3446 sizeof(struct ipv6hdr) :
3447 sizeof(struct iphdr);
3449 skb_set_transport_header(skb, mac_len + nh_len);
3452 /* Match skb->protocol to new outer l3 protocol */
3453 if (skb->protocol == htons(ETH_P_IP) &&
3454 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3455 skb->protocol = htons(ETH_P_IPV6);
3456 else if (skb->protocol == htons(ETH_P_IPV6) &&
3457 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3458 skb->protocol = htons(ETH_P_IP);
3461 if (skb_is_gso(skb)) {
3462 struct skb_shared_info *shinfo = skb_shinfo(skb);
3464 /* Due to header grow, MSS needs to be downgraded. */
3465 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3466 skb_decrease_gso_size(shinfo, len_diff);
3468 /* Header must be checked, and gso_segs recomputed. */
3469 shinfo->gso_type |= gso_type;
3470 shinfo->gso_segs = 0;
3476 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3481 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3482 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3485 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3486 /* udp gso_size delineates datagrams, only allow if fixed */
3487 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3488 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3492 ret = skb_unclone(skb, GFP_ATOMIC);
3493 if (unlikely(ret < 0))
3496 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3497 if (unlikely(ret < 0))
3500 if (skb_is_gso(skb)) {
3501 struct skb_shared_info *shinfo = skb_shinfo(skb);
3503 /* Due to header shrink, MSS can be upgraded. */
3504 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3505 skb_increase_gso_size(shinfo, len_diff);
3507 /* Header must be checked, and gso_segs recomputed. */
3508 shinfo->gso_type |= SKB_GSO_DODGY;
3509 shinfo->gso_segs = 0;
3515 static u32 __bpf_skb_max_len(const struct sk_buff *skb)
3517 return skb->dev ? skb->dev->mtu + skb->dev->hard_header_len :
3521 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3522 u32, mode, u64, flags)
3524 u32 len_diff_abs = abs(len_diff);
3525 bool shrink = len_diff < 0;
3528 if (unlikely(flags || mode))
3530 if (unlikely(len_diff_abs > 0xfffU))
3534 ret = skb_cow(skb, len_diff);
3535 if (unlikely(ret < 0))
3537 __skb_push(skb, len_diff_abs);
3538 memset(skb->data, 0, len_diff_abs);
3540 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3542 __skb_pull(skb, len_diff_abs);
3544 bpf_compute_data_end_sk_skb(skb);
3545 if (tls_sw_has_ctx_rx(skb->sk)) {
3546 struct strp_msg *rxm = strp_msg(skb);
3548 rxm->full_len += len_diff;
3553 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3554 .func = sk_skb_adjust_room,
3556 .ret_type = RET_INTEGER,
3557 .arg1_type = ARG_PTR_TO_CTX,
3558 .arg2_type = ARG_ANYTHING,
3559 .arg3_type = ARG_ANYTHING,
3560 .arg4_type = ARG_ANYTHING,
3563 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3564 u32, mode, u64, flags)
3566 u32 len_cur, len_diff_abs = abs(len_diff);
3567 u32 len_min = bpf_skb_net_base_len(skb);
3568 u32 len_max = __bpf_skb_max_len(skb);
3569 __be16 proto = skb->protocol;
3570 bool shrink = len_diff < 0;
3574 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3575 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3577 if (unlikely(len_diff_abs > 0xfffU))
3579 if (unlikely(proto != htons(ETH_P_IP) &&
3580 proto != htons(ETH_P_IPV6)))
3583 off = skb_mac_header_len(skb);
3585 case BPF_ADJ_ROOM_NET:
3586 off += bpf_skb_net_base_len(skb);
3588 case BPF_ADJ_ROOM_MAC:
3594 len_cur = skb->len - skb_network_offset(skb);
3595 if ((shrink && (len_diff_abs >= len_cur ||
3596 len_cur - len_diff_abs < len_min)) ||
3597 (!shrink && (skb->len + len_diff_abs > len_max &&
3601 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3602 bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3603 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3604 __skb_reset_checksum_unnecessary(skb);
3606 bpf_compute_data_pointers(skb);
3610 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3611 .func = bpf_skb_adjust_room,
3613 .ret_type = RET_INTEGER,
3614 .arg1_type = ARG_PTR_TO_CTX,
3615 .arg2_type = ARG_ANYTHING,
3616 .arg3_type = ARG_ANYTHING,
3617 .arg4_type = ARG_ANYTHING,
3620 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3622 u32 min_len = skb_network_offset(skb);
3624 if (skb_transport_header_was_set(skb))
3625 min_len = skb_transport_offset(skb);
3626 if (skb->ip_summed == CHECKSUM_PARTIAL)
3627 min_len = skb_checksum_start_offset(skb) +
3628 skb->csum_offset + sizeof(__sum16);
3632 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3634 unsigned int old_len = skb->len;
3637 ret = __skb_grow_rcsum(skb, new_len);
3639 memset(skb->data + old_len, 0, new_len - old_len);
3643 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3645 return __skb_trim_rcsum(skb, new_len);
3648 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3651 u32 max_len = __bpf_skb_max_len(skb);
3652 u32 min_len = __bpf_skb_min_len(skb);
3655 if (unlikely(flags || new_len > max_len || new_len < min_len))
3657 if (skb->encapsulation)
3660 /* The basic idea of this helper is that it's performing the
3661 * needed work to either grow or trim an skb, and eBPF program
3662 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3663 * bpf_lX_csum_replace() and others rather than passing a raw
3664 * buffer here. This one is a slow path helper and intended
3665 * for replies with control messages.
3667 * Like in bpf_skb_change_proto(), we want to keep this rather
3668 * minimal and without protocol specifics so that we are able
3669 * to separate concerns as in bpf_skb_store_bytes() should only
3670 * be the one responsible for writing buffers.
3672 * It's really expected to be a slow path operation here for
3673 * control message replies, so we're implicitly linearizing,
3674 * uncloning and drop offloads from the skb by this.
3676 ret = __bpf_try_make_writable(skb, skb->len);
3678 if (new_len > skb->len)
3679 ret = bpf_skb_grow_rcsum(skb, new_len);
3680 else if (new_len < skb->len)
3681 ret = bpf_skb_trim_rcsum(skb, new_len);
3682 if (!ret && skb_is_gso(skb))
3688 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3691 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3693 bpf_compute_data_pointers(skb);
3697 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3698 .func = bpf_skb_change_tail,
3700 .ret_type = RET_INTEGER,
3701 .arg1_type = ARG_PTR_TO_CTX,
3702 .arg2_type = ARG_ANYTHING,
3703 .arg3_type = ARG_ANYTHING,
3706 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3709 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3711 bpf_compute_data_end_sk_skb(skb);
3715 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3716 .func = sk_skb_change_tail,
3718 .ret_type = RET_INTEGER,
3719 .arg1_type = ARG_PTR_TO_CTX,
3720 .arg2_type = ARG_ANYTHING,
3721 .arg3_type = ARG_ANYTHING,
3724 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3727 u32 max_len = __bpf_skb_max_len(skb);
3728 u32 new_len = skb->len + head_room;
3731 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3732 new_len < skb->len))
3735 ret = skb_cow(skb, head_room);
3737 /* Idea for this helper is that we currently only
3738 * allow to expand on mac header. This means that
3739 * skb->protocol network header, etc, stay as is.
3740 * Compared to bpf_skb_change_tail(), we're more
3741 * flexible due to not needing to linearize or
3742 * reset GSO. Intention for this helper is to be
3743 * used by an L3 skb that needs to push mac header
3744 * for redirection into L2 device.
3746 __skb_push(skb, head_room);
3747 memset(skb->data, 0, head_room);
3748 skb_reset_mac_header(skb);
3754 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3757 int ret = __bpf_skb_change_head(skb, head_room, flags);
3759 bpf_compute_data_pointers(skb);
3763 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3764 .func = bpf_skb_change_head,
3766 .ret_type = RET_INTEGER,
3767 .arg1_type = ARG_PTR_TO_CTX,
3768 .arg2_type = ARG_ANYTHING,
3769 .arg3_type = ARG_ANYTHING,
3772 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3775 int ret = __bpf_skb_change_head(skb, head_room, flags);
3777 bpf_compute_data_end_sk_skb(skb);
3781 static const struct bpf_func_proto sk_skb_change_head_proto = {
3782 .func = sk_skb_change_head,
3784 .ret_type = RET_INTEGER,
3785 .arg1_type = ARG_PTR_TO_CTX,
3786 .arg2_type = ARG_ANYTHING,
3787 .arg3_type = ARG_ANYTHING,
3789 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3791 return xdp_data_meta_unsupported(xdp) ? 0 :
3792 xdp->data - xdp->data_meta;
3795 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3797 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3798 unsigned long metalen = xdp_get_metalen(xdp);
3799 void *data_start = xdp_frame_end + metalen;
3800 void *data = xdp->data + offset;
3802 if (unlikely(data < data_start ||
3803 data > xdp->data_end - ETH_HLEN))
3807 memmove(xdp->data_meta + offset,
3808 xdp->data_meta, metalen);
3809 xdp->data_meta += offset;
3815 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3816 .func = bpf_xdp_adjust_head,
3818 .ret_type = RET_INTEGER,
3819 .arg1_type = ARG_PTR_TO_CTX,
3820 .arg2_type = ARG_ANYTHING,
3823 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3825 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
3826 void *data_end = xdp->data_end + offset;
3828 /* Notice that xdp_data_hard_end have reserved some tailroom */
3829 if (unlikely(data_end > data_hard_end))
3832 /* ALL drivers MUST init xdp->frame_sz, chicken check below */
3833 if (unlikely(xdp->frame_sz > PAGE_SIZE)) {
3834 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz);
3838 if (unlikely(data_end < xdp->data + ETH_HLEN))
3841 /* Clear memory area on grow, can contain uninit kernel memory */
3843 memset(xdp->data_end, 0, offset);
3845 xdp->data_end = data_end;
3850 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3851 .func = bpf_xdp_adjust_tail,
3853 .ret_type = RET_INTEGER,
3854 .arg1_type = ARG_PTR_TO_CTX,
3855 .arg2_type = ARG_ANYTHING,
3858 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3860 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3861 void *meta = xdp->data_meta + offset;
3862 unsigned long metalen = xdp->data - meta;
3864 if (xdp_data_meta_unsupported(xdp))
3866 if (unlikely(meta < xdp_frame_end ||
3869 if (unlikely((metalen & (sizeof(__u32) - 1)) ||
3873 xdp->data_meta = meta;
3878 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3879 .func = bpf_xdp_adjust_meta,
3881 .ret_type = RET_INTEGER,
3882 .arg1_type = ARG_PTR_TO_CTX,
3883 .arg2_type = ARG_ANYTHING,
3886 static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd,
3887 struct bpf_map *map, struct xdp_buff *xdp)
3889 switch (map->map_type) {
3890 case BPF_MAP_TYPE_DEVMAP:
3891 case BPF_MAP_TYPE_DEVMAP_HASH:
3892 return dev_map_enqueue(fwd, xdp, dev_rx);
3893 case BPF_MAP_TYPE_CPUMAP:
3894 return cpu_map_enqueue(fwd, xdp, dev_rx);
3895 case BPF_MAP_TYPE_XSKMAP:
3896 return __xsk_map_redirect(fwd, xdp);
3903 void xdp_do_flush(void)
3909 EXPORT_SYMBOL_GPL(xdp_do_flush);
3911 static inline void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index)
3913 switch (map->map_type) {
3914 case BPF_MAP_TYPE_DEVMAP:
3915 return __dev_map_lookup_elem(map, index);
3916 case BPF_MAP_TYPE_DEVMAP_HASH:
3917 return __dev_map_hash_lookup_elem(map, index);
3918 case BPF_MAP_TYPE_CPUMAP:
3919 return __cpu_map_lookup_elem(map, index);
3920 case BPF_MAP_TYPE_XSKMAP:
3921 return __xsk_map_lookup_elem(map, index);
3927 void bpf_clear_redirect_map(struct bpf_map *map)
3929 struct bpf_redirect_info *ri;
3932 for_each_possible_cpu(cpu) {
3933 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3934 /* Avoid polluting remote cacheline due to writes if
3935 * not needed. Once we pass this test, we need the
3936 * cmpxchg() to make sure it hasn't been changed in
3937 * the meantime by remote CPU.
3939 if (unlikely(READ_ONCE(ri->map) == map))
3940 cmpxchg(&ri->map, map, NULL);
3944 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3945 struct bpf_prog *xdp_prog)
3947 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3948 struct bpf_map *map = READ_ONCE(ri->map);
3949 u32 index = ri->tgt_index;
3950 void *fwd = ri->tgt_value;
3954 ri->tgt_value = NULL;
3955 WRITE_ONCE(ri->map, NULL);
3957 if (unlikely(!map)) {
3958 fwd = dev_get_by_index_rcu(dev_net(dev), index);
3959 if (unlikely(!fwd)) {
3964 err = dev_xdp_enqueue(fwd, xdp, dev);
3966 err = __bpf_tx_xdp_map(dev, fwd, map, xdp);
3972 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3975 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3978 EXPORT_SYMBOL_GPL(xdp_do_redirect);
3980 static int xdp_do_generic_redirect_map(struct net_device *dev,
3981 struct sk_buff *skb,
3982 struct xdp_buff *xdp,
3983 struct bpf_prog *xdp_prog,
3984 struct bpf_map *map)
3986 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3987 u32 index = ri->tgt_index;
3988 void *fwd = ri->tgt_value;
3992 ri->tgt_value = NULL;
3993 WRITE_ONCE(ri->map, NULL);
3995 if (map->map_type == BPF_MAP_TYPE_DEVMAP ||
3996 map->map_type == BPF_MAP_TYPE_DEVMAP_HASH) {
3997 struct bpf_dtab_netdev *dst = fwd;
3999 err = dev_map_generic_redirect(dst, skb, xdp_prog);
4002 } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) {
4003 struct xdp_sock *xs = fwd;
4005 err = xsk_generic_rcv(xs, xdp);
4010 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
4015 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
4018 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
4022 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4023 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4025 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4026 struct bpf_map *map = READ_ONCE(ri->map);
4027 u32 index = ri->tgt_index;
4028 struct net_device *fwd;
4032 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog,
4035 fwd = dev_get_by_index_rcu(dev_net(dev), index);
4036 if (unlikely(!fwd)) {
4041 err = xdp_ok_fwd_dev(fwd, skb->len);
4046 _trace_xdp_redirect(dev, xdp_prog, index);
4047 generic_xdp_tx(skb, xdp_prog);
4050 _trace_xdp_redirect_err(dev, xdp_prog, index, err);
4054 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4056 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4058 if (unlikely(flags))
4062 ri->tgt_index = ifindex;
4063 ri->tgt_value = NULL;
4064 WRITE_ONCE(ri->map, NULL);
4066 return XDP_REDIRECT;
4069 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4070 .func = bpf_xdp_redirect,
4072 .ret_type = RET_INTEGER,
4073 .arg1_type = ARG_ANYTHING,
4074 .arg2_type = ARG_ANYTHING,
4077 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
4080 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4082 /* Lower bits of the flags are used as return code on lookup failure */
4083 if (unlikely(flags > XDP_TX))
4086 ri->tgt_value = __xdp_map_lookup_elem(map, ifindex);
4087 if (unlikely(!ri->tgt_value)) {
4088 /* If the lookup fails we want to clear out the state in the
4089 * redirect_info struct completely, so that if an eBPF program
4090 * performs multiple lookups, the last one always takes
4093 WRITE_ONCE(ri->map, NULL);
4098 ri->tgt_index = ifindex;
4099 WRITE_ONCE(ri->map, map);
4101 return XDP_REDIRECT;
4104 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4105 .func = bpf_xdp_redirect_map,
4107 .ret_type = RET_INTEGER,
4108 .arg1_type = ARG_CONST_MAP_PTR,
4109 .arg2_type = ARG_ANYTHING,
4110 .arg3_type = ARG_ANYTHING,
4113 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4114 unsigned long off, unsigned long len)
4116 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4120 if (ptr != dst_buff)
4121 memcpy(dst_buff, ptr, len);
4126 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4127 u64, flags, void *, meta, u64, meta_size)
4129 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4131 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4133 if (unlikely(!skb || skb_size > skb->len))
4136 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4140 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4141 .func = bpf_skb_event_output,
4143 .ret_type = RET_INTEGER,
4144 .arg1_type = ARG_PTR_TO_CTX,
4145 .arg2_type = ARG_CONST_MAP_PTR,
4146 .arg3_type = ARG_ANYTHING,
4147 .arg4_type = ARG_PTR_TO_MEM,
4148 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4151 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4153 const struct bpf_func_proto bpf_skb_output_proto = {
4154 .func = bpf_skb_event_output,
4156 .ret_type = RET_INTEGER,
4157 .arg1_type = ARG_PTR_TO_BTF_ID,
4158 .arg1_btf_id = &bpf_skb_output_btf_ids[0],
4159 .arg2_type = ARG_CONST_MAP_PTR,
4160 .arg3_type = ARG_ANYTHING,
4161 .arg4_type = ARG_PTR_TO_MEM,
4162 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4165 static unsigned short bpf_tunnel_key_af(u64 flags)
4167 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4170 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4171 u32, size, u64, flags)
4173 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4174 u8 compat[sizeof(struct bpf_tunnel_key)];
4178 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
4182 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4186 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4189 case offsetof(struct bpf_tunnel_key, tunnel_label):
4190 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4192 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4193 /* Fixup deprecated structure layouts here, so we have
4194 * a common path later on.
4196 if (ip_tunnel_info_af(info) != AF_INET)
4199 to = (struct bpf_tunnel_key *)compat;
4206 to->tunnel_id = be64_to_cpu(info->key.tun_id);
4207 to->tunnel_tos = info->key.tos;
4208 to->tunnel_ttl = info->key.ttl;
4211 if (flags & BPF_F_TUNINFO_IPV6) {
4212 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4213 sizeof(to->remote_ipv6));
4214 to->tunnel_label = be32_to_cpu(info->key.label);
4216 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4217 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4218 to->tunnel_label = 0;
4221 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4222 memcpy(to_orig, to, size);
4226 memset(to_orig, 0, size);
4230 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4231 .func = bpf_skb_get_tunnel_key,
4233 .ret_type = RET_INTEGER,
4234 .arg1_type = ARG_PTR_TO_CTX,
4235 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4236 .arg3_type = ARG_CONST_SIZE,
4237 .arg4_type = ARG_ANYTHING,
4240 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4242 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4245 if (unlikely(!info ||
4246 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4250 if (unlikely(size < info->options_len)) {
4255 ip_tunnel_info_opts_get(to, info);
4256 if (size > info->options_len)
4257 memset(to + info->options_len, 0, size - info->options_len);
4259 return info->options_len;
4261 memset(to, 0, size);
4265 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4266 .func = bpf_skb_get_tunnel_opt,
4268 .ret_type = RET_INTEGER,
4269 .arg1_type = ARG_PTR_TO_CTX,
4270 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4271 .arg3_type = ARG_CONST_SIZE,
4274 static struct metadata_dst __percpu *md_dst;
4276 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4277 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4279 struct metadata_dst *md = this_cpu_ptr(md_dst);
4280 u8 compat[sizeof(struct bpf_tunnel_key)];
4281 struct ip_tunnel_info *info;
4283 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4284 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
4286 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4288 case offsetof(struct bpf_tunnel_key, tunnel_label):
4289 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4290 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4291 /* Fixup deprecated structure layouts here, so we have
4292 * a common path later on.
4294 memcpy(compat, from, size);
4295 memset(compat + size, 0, sizeof(compat) - size);
4296 from = (const struct bpf_tunnel_key *) compat;
4302 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4307 dst_hold((struct dst_entry *) md);
4308 skb_dst_set(skb, (struct dst_entry *) md);
4310 info = &md->u.tun_info;
4311 memset(info, 0, sizeof(*info));
4312 info->mode = IP_TUNNEL_INFO_TX;
4314 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4315 if (flags & BPF_F_DONT_FRAGMENT)
4316 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4317 if (flags & BPF_F_ZERO_CSUM_TX)
4318 info->key.tun_flags &= ~TUNNEL_CSUM;
4319 if (flags & BPF_F_SEQ_NUMBER)
4320 info->key.tun_flags |= TUNNEL_SEQ;
4322 info->key.tun_id = cpu_to_be64(from->tunnel_id);
4323 info->key.tos = from->tunnel_tos;
4324 info->key.ttl = from->tunnel_ttl;
4326 if (flags & BPF_F_TUNINFO_IPV6) {
4327 info->mode |= IP_TUNNEL_INFO_IPV6;
4328 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4329 sizeof(from->remote_ipv6));
4330 info->key.label = cpu_to_be32(from->tunnel_label) &
4331 IPV6_FLOWLABEL_MASK;
4333 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4339 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4340 .func = bpf_skb_set_tunnel_key,
4342 .ret_type = RET_INTEGER,
4343 .arg1_type = ARG_PTR_TO_CTX,
4344 .arg2_type = ARG_PTR_TO_MEM,
4345 .arg3_type = ARG_CONST_SIZE,
4346 .arg4_type = ARG_ANYTHING,
4349 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4350 const u8 *, from, u32, size)
4352 struct ip_tunnel_info *info = skb_tunnel_info(skb);
4353 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4355 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4357 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4360 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4365 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4366 .func = bpf_skb_set_tunnel_opt,
4368 .ret_type = RET_INTEGER,
4369 .arg1_type = ARG_PTR_TO_CTX,
4370 .arg2_type = ARG_PTR_TO_MEM,
4371 .arg3_type = ARG_CONST_SIZE,
4374 static const struct bpf_func_proto *
4375 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4378 struct metadata_dst __percpu *tmp;
4380 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4385 if (cmpxchg(&md_dst, NULL, tmp))
4386 metadata_dst_free_percpu(tmp);
4390 case BPF_FUNC_skb_set_tunnel_key:
4391 return &bpf_skb_set_tunnel_key_proto;
4392 case BPF_FUNC_skb_set_tunnel_opt:
4393 return &bpf_skb_set_tunnel_opt_proto;
4399 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4402 struct bpf_array *array = container_of(map, struct bpf_array, map);
4403 struct cgroup *cgrp;
4406 sk = skb_to_full_sk(skb);
4407 if (!sk || !sk_fullsock(sk))
4409 if (unlikely(idx >= array->map.max_entries))
4412 cgrp = READ_ONCE(array->ptrs[idx]);
4413 if (unlikely(!cgrp))
4416 return sk_under_cgroup_hierarchy(sk, cgrp);
4419 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4420 .func = bpf_skb_under_cgroup,
4422 .ret_type = RET_INTEGER,
4423 .arg1_type = ARG_PTR_TO_CTX,
4424 .arg2_type = ARG_CONST_MAP_PTR,
4425 .arg3_type = ARG_ANYTHING,
4428 #ifdef CONFIG_SOCK_CGROUP_DATA
4429 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4431 struct cgroup *cgrp;
4433 sk = sk_to_full_sk(sk);
4434 if (!sk || !sk_fullsock(sk))
4437 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4438 return cgroup_id(cgrp);
4441 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4443 return __bpf_sk_cgroup_id(skb->sk);
4446 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4447 .func = bpf_skb_cgroup_id,
4449 .ret_type = RET_INTEGER,
4450 .arg1_type = ARG_PTR_TO_CTX,
4453 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4456 struct cgroup *ancestor;
4457 struct cgroup *cgrp;
4459 sk = sk_to_full_sk(sk);
4460 if (!sk || !sk_fullsock(sk))
4463 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4464 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4468 return cgroup_id(ancestor);
4471 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4474 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4477 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4478 .func = bpf_skb_ancestor_cgroup_id,
4480 .ret_type = RET_INTEGER,
4481 .arg1_type = ARG_PTR_TO_CTX,
4482 .arg2_type = ARG_ANYTHING,
4485 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4487 return __bpf_sk_cgroup_id(sk);
4490 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4491 .func = bpf_sk_cgroup_id,
4493 .ret_type = RET_INTEGER,
4494 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4497 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4499 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4502 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4503 .func = bpf_sk_ancestor_cgroup_id,
4505 .ret_type = RET_INTEGER,
4506 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4507 .arg2_type = ARG_ANYTHING,
4511 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4512 unsigned long off, unsigned long len)
4514 memcpy(dst_buff, src_buff + off, len);
4518 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4519 u64, flags, void *, meta, u64, meta_size)
4521 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4523 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4525 if (unlikely(!xdp ||
4526 xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4529 return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4530 xdp_size, bpf_xdp_copy);
4533 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4534 .func = bpf_xdp_event_output,
4536 .ret_type = RET_INTEGER,
4537 .arg1_type = ARG_PTR_TO_CTX,
4538 .arg2_type = ARG_CONST_MAP_PTR,
4539 .arg3_type = ARG_ANYTHING,
4540 .arg4_type = ARG_PTR_TO_MEM,
4541 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4544 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4546 const struct bpf_func_proto bpf_xdp_output_proto = {
4547 .func = bpf_xdp_event_output,
4549 .ret_type = RET_INTEGER,
4550 .arg1_type = ARG_PTR_TO_BTF_ID,
4551 .arg1_btf_id = &bpf_xdp_output_btf_ids[0],
4552 .arg2_type = ARG_CONST_MAP_PTR,
4553 .arg3_type = ARG_ANYTHING,
4554 .arg4_type = ARG_PTR_TO_MEM,
4555 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4558 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4560 return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4563 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4564 .func = bpf_get_socket_cookie,
4566 .ret_type = RET_INTEGER,
4567 .arg1_type = ARG_PTR_TO_CTX,
4570 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4572 return __sock_gen_cookie(ctx->sk);
4575 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4576 .func = bpf_get_socket_cookie_sock_addr,
4578 .ret_type = RET_INTEGER,
4579 .arg1_type = ARG_PTR_TO_CTX,
4582 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4584 return __sock_gen_cookie(ctx);
4587 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4588 .func = bpf_get_socket_cookie_sock,
4590 .ret_type = RET_INTEGER,
4591 .arg1_type = ARG_PTR_TO_CTX,
4594 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4596 return __sock_gen_cookie(ctx->sk);
4599 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4600 .func = bpf_get_socket_cookie_sock_ops,
4602 .ret_type = RET_INTEGER,
4603 .arg1_type = ARG_PTR_TO_CTX,
4606 static u64 __bpf_get_netns_cookie(struct sock *sk)
4608 #ifdef CONFIG_NET_NS
4609 return __net_gen_cookie(sk ? sk->sk_net.net : &init_net);
4615 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
4617 return __bpf_get_netns_cookie(ctx);
4620 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
4621 .func = bpf_get_netns_cookie_sock,
4623 .ret_type = RET_INTEGER,
4624 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4627 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4629 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4632 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
4633 .func = bpf_get_netns_cookie_sock_addr,
4635 .ret_type = RET_INTEGER,
4636 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4639 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4641 struct sock *sk = sk_to_full_sk(skb->sk);
4644 if (!sk || !sk_fullsock(sk))
4646 kuid = sock_net_uid(sock_net(sk), sk);
4647 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4650 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4651 .func = bpf_get_socket_uid,
4653 .ret_type = RET_INTEGER,
4654 .arg1_type = ARG_PTR_TO_CTX,
4657 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
4658 char *optval, int optlen)
4660 char devname[IFNAMSIZ];
4666 if (!sk_fullsock(sk))
4669 sock_owned_by_me(sk);
4671 if (level == SOL_SOCKET) {
4672 if (optlen != sizeof(int) && optname != SO_BINDTODEVICE)
4674 val = *((int *)optval);
4675 valbool = val ? 1 : 0;
4677 /* Only some socketops are supported */
4680 val = min_t(u32, val, sysctl_rmem_max);
4681 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4682 WRITE_ONCE(sk->sk_rcvbuf,
4683 max_t(int, val * 2, SOCK_MIN_RCVBUF));
4686 val = min_t(u32, val, sysctl_wmem_max);
4687 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4688 WRITE_ONCE(sk->sk_sndbuf,
4689 max_t(int, val * 2, SOCK_MIN_SNDBUF));
4691 case SO_MAX_PACING_RATE: /* 32bit version */
4693 cmpxchg(&sk->sk_pacing_status,
4696 sk->sk_max_pacing_rate = (val == ~0U) ? ~0UL : val;
4697 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4698 sk->sk_max_pacing_rate);
4701 sk->sk_priority = val;
4706 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
4709 if (sk->sk_mark != val) {
4714 case SO_BINDTODEVICE:
4715 optlen = min_t(long, optlen, IFNAMSIZ - 1);
4716 strncpy(devname, optval, optlen);
4717 devname[optlen] = 0;
4720 if (devname[0] != '\0') {
4721 struct net_device *dev;
4726 dev = dev_get_by_name(net, devname);
4729 ifindex = dev->ifindex;
4732 ret = sock_bindtoindex(sk, ifindex, false);
4735 if (sk->sk_prot->keepalive)
4736 sk->sk_prot->keepalive(sk, valbool);
4737 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
4743 } else if (level == SOL_IP) {
4744 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4747 val = *((int *)optval);
4748 /* Only some options are supported */
4751 if (val < -1 || val > 0xff) {
4754 struct inet_sock *inet = inet_sk(sk);
4764 #if IS_ENABLED(CONFIG_IPV6)
4765 } else if (level == SOL_IPV6) {
4766 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4769 val = *((int *)optval);
4770 /* Only some options are supported */
4773 if (val < -1 || val > 0xff) {
4776 struct ipv6_pinfo *np = inet6_sk(sk);
4787 } else if (level == SOL_TCP &&
4788 sk->sk_prot->setsockopt == tcp_setsockopt) {
4789 if (optname == TCP_CONGESTION) {
4790 char name[TCP_CA_NAME_MAX];
4792 strncpy(name, optval, min_t(long, optlen,
4793 TCP_CA_NAME_MAX-1));
4794 name[TCP_CA_NAME_MAX-1] = 0;
4795 ret = tcp_set_congestion_control(sk, name, false, true);
4797 struct inet_connection_sock *icsk = inet_csk(sk);
4798 struct tcp_sock *tp = tcp_sk(sk);
4799 unsigned long timeout;
4801 if (optlen != sizeof(int))
4804 val = *((int *)optval);
4805 /* Only some options are supported */
4808 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4813 case TCP_BPF_SNDCWND_CLAMP:
4817 tp->snd_cwnd_clamp = val;
4818 tp->snd_ssthresh = val;
4821 case TCP_BPF_DELACK_MAX:
4822 timeout = usecs_to_jiffies(val);
4823 if (timeout > TCP_DELACK_MAX ||
4824 timeout < TCP_TIMEOUT_MIN)
4826 inet_csk(sk)->icsk_delack_max = timeout;
4828 case TCP_BPF_RTO_MIN:
4829 timeout = usecs_to_jiffies(val);
4830 if (timeout > TCP_RTO_MIN ||
4831 timeout < TCP_TIMEOUT_MIN)
4833 inet_csk(sk)->icsk_rto_min = timeout;
4836 if (val < 0 || val > 1)
4842 ret = tcp_sock_set_keepidle_locked(sk, val);
4845 if (val < 1 || val > MAX_TCP_KEEPINTVL)
4848 tp->keepalive_intvl = val * HZ;
4851 if (val < 1 || val > MAX_TCP_KEEPCNT)
4854 tp->keepalive_probes = val;
4857 if (val < 1 || val > MAX_TCP_SYNCNT)
4860 icsk->icsk_syn_retries = val;
4862 case TCP_USER_TIMEOUT:
4866 icsk->icsk_user_timeout = val;
4868 case TCP_NOTSENT_LOWAT:
4869 tp->notsent_lowat = val;
4870 sk->sk_write_space(sk);
4883 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
4884 char *optval, int optlen)
4886 if (!sk_fullsock(sk))
4889 sock_owned_by_me(sk);
4892 if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4893 struct inet_connection_sock *icsk;
4894 struct tcp_sock *tp;
4897 case TCP_CONGESTION:
4898 icsk = inet_csk(sk);
4900 if (!icsk->icsk_ca_ops || optlen <= 1)
4902 strncpy(optval, icsk->icsk_ca_ops->name, optlen);
4903 optval[optlen - 1] = 0;
4908 if (optlen <= 0 || !tp->saved_syn ||
4909 optlen > tcp_saved_syn_len(tp->saved_syn))
4911 memcpy(optval, tp->saved_syn->data, optlen);
4916 } else if (level == SOL_IP) {
4917 struct inet_sock *inet = inet_sk(sk);
4919 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4922 /* Only some options are supported */
4925 *((int *)optval) = (int)inet->tos;
4930 #if IS_ENABLED(CONFIG_IPV6)
4931 } else if (level == SOL_IPV6) {
4932 struct ipv6_pinfo *np = inet6_sk(sk);
4934 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4937 /* Only some options are supported */
4940 *((int *)optval) = (int)np->tclass;
4952 memset(optval, 0, optlen);
4956 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
4957 int, level, int, optname, char *, optval, int, optlen)
4959 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
4962 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
4963 .func = bpf_sock_addr_setsockopt,
4965 .ret_type = RET_INTEGER,
4966 .arg1_type = ARG_PTR_TO_CTX,
4967 .arg2_type = ARG_ANYTHING,
4968 .arg3_type = ARG_ANYTHING,
4969 .arg4_type = ARG_PTR_TO_MEM,
4970 .arg5_type = ARG_CONST_SIZE,
4973 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
4974 int, level, int, optname, char *, optval, int, optlen)
4976 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
4979 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
4980 .func = bpf_sock_addr_getsockopt,
4982 .ret_type = RET_INTEGER,
4983 .arg1_type = ARG_PTR_TO_CTX,
4984 .arg2_type = ARG_ANYTHING,
4985 .arg3_type = ARG_ANYTHING,
4986 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
4987 .arg5_type = ARG_CONST_SIZE,
4990 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
4991 int, level, int, optname, char *, optval, int, optlen)
4993 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
4996 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
4997 .func = bpf_sock_ops_setsockopt,
4999 .ret_type = RET_INTEGER,
5000 .arg1_type = ARG_PTR_TO_CTX,
5001 .arg2_type = ARG_ANYTHING,
5002 .arg3_type = ARG_ANYTHING,
5003 .arg4_type = ARG_PTR_TO_MEM,
5004 .arg5_type = ARG_CONST_SIZE,
5007 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5008 int optname, const u8 **start)
5010 struct sk_buff *syn_skb = bpf_sock->syn_skb;
5011 const u8 *hdr_start;
5015 /* sk is a request_sock here */
5017 if (optname == TCP_BPF_SYN) {
5018 hdr_start = syn_skb->data;
5019 ret = tcp_hdrlen(syn_skb);
5020 } else if (optname == TCP_BPF_SYN_IP) {
5021 hdr_start = skb_network_header(syn_skb);
5022 ret = skb_network_header_len(syn_skb) +
5023 tcp_hdrlen(syn_skb);
5025 /* optname == TCP_BPF_SYN_MAC */
5026 hdr_start = skb_mac_header(syn_skb);
5027 ret = skb_mac_header_len(syn_skb) +
5028 skb_network_header_len(syn_skb) +
5029 tcp_hdrlen(syn_skb);
5032 struct sock *sk = bpf_sock->sk;
5033 struct saved_syn *saved_syn;
5035 if (sk->sk_state == TCP_NEW_SYN_RECV)
5036 /* synack retransmit. bpf_sock->syn_skb will
5037 * not be available. It has to resort to
5038 * saved_syn (if it is saved).
5040 saved_syn = inet_reqsk(sk)->saved_syn;
5042 saved_syn = tcp_sk(sk)->saved_syn;
5047 if (optname == TCP_BPF_SYN) {
5048 hdr_start = saved_syn->data +
5049 saved_syn->mac_hdrlen +
5050 saved_syn->network_hdrlen;
5051 ret = saved_syn->tcp_hdrlen;
5052 } else if (optname == TCP_BPF_SYN_IP) {
5053 hdr_start = saved_syn->data +
5054 saved_syn->mac_hdrlen;
5055 ret = saved_syn->network_hdrlen +
5056 saved_syn->tcp_hdrlen;
5058 /* optname == TCP_BPF_SYN_MAC */
5060 /* TCP_SAVE_SYN may not have saved the mac hdr */
5061 if (!saved_syn->mac_hdrlen)
5064 hdr_start = saved_syn->data;
5065 ret = saved_syn->mac_hdrlen +
5066 saved_syn->network_hdrlen +
5067 saved_syn->tcp_hdrlen;
5075 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5076 int, level, int, optname, char *, optval, int, optlen)
5078 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5079 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5080 int ret, copy_len = 0;
5083 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5086 if (optlen < copy_len) {
5091 memcpy(optval, start, copy_len);
5094 /* Zero out unused buffer at the end */
5095 memset(optval + copy_len, 0, optlen - copy_len);
5100 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5103 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5104 .func = bpf_sock_ops_getsockopt,
5106 .ret_type = RET_INTEGER,
5107 .arg1_type = ARG_PTR_TO_CTX,
5108 .arg2_type = ARG_ANYTHING,
5109 .arg3_type = ARG_ANYTHING,
5110 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5111 .arg5_type = ARG_CONST_SIZE,
5114 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5117 struct sock *sk = bpf_sock->sk;
5118 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5120 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5123 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5125 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5128 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5129 .func = bpf_sock_ops_cb_flags_set,
5131 .ret_type = RET_INTEGER,
5132 .arg1_type = ARG_PTR_TO_CTX,
5133 .arg2_type = ARG_ANYTHING,
5136 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5137 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5139 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5143 struct sock *sk = ctx->sk;
5144 u32 flags = BIND_FROM_BPF;
5148 if (addr_len < offsetofend(struct sockaddr, sa_family))
5150 if (addr->sa_family == AF_INET) {
5151 if (addr_len < sizeof(struct sockaddr_in))
5153 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5154 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5155 return __inet_bind(sk, addr, addr_len, flags);
5156 #if IS_ENABLED(CONFIG_IPV6)
5157 } else if (addr->sa_family == AF_INET6) {
5158 if (addr_len < SIN6_LEN_RFC2133)
5160 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5161 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5162 /* ipv6_bpf_stub cannot be NULL, since it's called from
5163 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5165 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5166 #endif /* CONFIG_IPV6 */
5168 #endif /* CONFIG_INET */
5170 return -EAFNOSUPPORT;
5173 static const struct bpf_func_proto bpf_bind_proto = {
5176 .ret_type = RET_INTEGER,
5177 .arg1_type = ARG_PTR_TO_CTX,
5178 .arg2_type = ARG_PTR_TO_MEM,
5179 .arg3_type = ARG_CONST_SIZE,
5183 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5184 struct bpf_xfrm_state *, to, u32, size, u64, flags)
5186 const struct sec_path *sp = skb_sec_path(skb);
5187 const struct xfrm_state *x;
5189 if (!sp || unlikely(index >= sp->len || flags))
5192 x = sp->xvec[index];
5194 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5197 to->reqid = x->props.reqid;
5198 to->spi = x->id.spi;
5199 to->family = x->props.family;
5202 if (to->family == AF_INET6) {
5203 memcpy(to->remote_ipv6, x->props.saddr.a6,
5204 sizeof(to->remote_ipv6));
5206 to->remote_ipv4 = x->props.saddr.a4;
5207 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5212 memset(to, 0, size);
5216 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5217 .func = bpf_skb_get_xfrm_state,
5219 .ret_type = RET_INTEGER,
5220 .arg1_type = ARG_PTR_TO_CTX,
5221 .arg2_type = ARG_ANYTHING,
5222 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
5223 .arg4_type = ARG_CONST_SIZE,
5224 .arg5_type = ARG_ANYTHING,
5228 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5229 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
5230 const struct neighbour *neigh,
5231 const struct net_device *dev)
5233 memcpy(params->dmac, neigh->ha, ETH_ALEN);
5234 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5235 params->h_vlan_TCI = 0;
5236 params->h_vlan_proto = 0;
5242 #if IS_ENABLED(CONFIG_INET)
5243 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5244 u32 flags, bool check_mtu)
5246 struct fib_nh_common *nhc;
5247 struct in_device *in_dev;
5248 struct neighbour *neigh;
5249 struct net_device *dev;
5250 struct fib_result res;
5255 dev = dev_get_by_index_rcu(net, params->ifindex);
5259 /* verify forwarding is enabled on this interface */
5260 in_dev = __in_dev_get_rcu(dev);
5261 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5262 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5264 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5266 fl4.flowi4_oif = params->ifindex;
5268 fl4.flowi4_iif = params->ifindex;
5271 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5272 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5273 fl4.flowi4_flags = 0;
5275 fl4.flowi4_proto = params->l4_protocol;
5276 fl4.daddr = params->ipv4_dst;
5277 fl4.saddr = params->ipv4_src;
5278 fl4.fl4_sport = params->sport;
5279 fl4.fl4_dport = params->dport;
5280 fl4.flowi4_multipath_hash = 0;
5282 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5283 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5284 struct fib_table *tb;
5286 tb = fib_get_table(net, tbid);
5288 return BPF_FIB_LKUP_RET_NOT_FWDED;
5290 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5292 fl4.flowi4_mark = 0;
5293 fl4.flowi4_secid = 0;
5294 fl4.flowi4_tun_key.tun_id = 0;
5295 fl4.flowi4_uid = sock_net_uid(net, NULL);
5297 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5301 /* map fib lookup errors to RTN_ type */
5303 return BPF_FIB_LKUP_RET_BLACKHOLE;
5304 if (err == -EHOSTUNREACH)
5305 return BPF_FIB_LKUP_RET_UNREACHABLE;
5307 return BPF_FIB_LKUP_RET_PROHIBIT;
5309 return BPF_FIB_LKUP_RET_NOT_FWDED;
5312 if (res.type != RTN_UNICAST)
5313 return BPF_FIB_LKUP_RET_NOT_FWDED;
5315 if (fib_info_num_path(res.fi) > 1)
5316 fib_select_path(net, &res, &fl4, NULL);
5319 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5320 if (params->tot_len > mtu)
5321 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5326 /* do not handle lwt encaps right now */
5327 if (nhc->nhc_lwtstate)
5328 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5332 params->rt_metric = res.fi->fib_priority;
5333 params->ifindex = dev->ifindex;
5335 /* xdp and cls_bpf programs are run in RCU-bh so
5336 * rcu_read_lock_bh is not needed here
5338 if (likely(nhc->nhc_gw_family != AF_INET6)) {
5339 if (nhc->nhc_gw_family)
5340 params->ipv4_dst = nhc->nhc_gw.ipv4;
5342 neigh = __ipv4_neigh_lookup_noref(dev,
5343 (__force u32)params->ipv4_dst);
5345 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5347 params->family = AF_INET6;
5348 *dst = nhc->nhc_gw.ipv6;
5349 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5353 return BPF_FIB_LKUP_RET_NO_NEIGH;
5355 return bpf_fib_set_fwd_params(params, neigh, dev);
5359 #if IS_ENABLED(CONFIG_IPV6)
5360 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5361 u32 flags, bool check_mtu)
5363 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5364 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5365 struct fib6_result res = {};
5366 struct neighbour *neigh;
5367 struct net_device *dev;
5368 struct inet6_dev *idev;
5374 /* link local addresses are never forwarded */
5375 if (rt6_need_strict(dst) || rt6_need_strict(src))
5376 return BPF_FIB_LKUP_RET_NOT_FWDED;
5378 dev = dev_get_by_index_rcu(net, params->ifindex);
5382 idev = __in6_dev_get_safely(dev);
5383 if (unlikely(!idev || !idev->cnf.forwarding))
5384 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5386 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5388 oif = fl6.flowi6_oif = params->ifindex;
5390 oif = fl6.flowi6_iif = params->ifindex;
5392 strict = RT6_LOOKUP_F_HAS_SADDR;
5394 fl6.flowlabel = params->flowinfo;
5395 fl6.flowi6_scope = 0;
5396 fl6.flowi6_flags = 0;
5399 fl6.flowi6_proto = params->l4_protocol;
5402 fl6.fl6_sport = params->sport;
5403 fl6.fl6_dport = params->dport;
5405 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5406 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5407 struct fib6_table *tb;
5409 tb = ipv6_stub->fib6_get_table(net, tbid);
5411 return BPF_FIB_LKUP_RET_NOT_FWDED;
5413 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5416 fl6.flowi6_mark = 0;
5417 fl6.flowi6_secid = 0;
5418 fl6.flowi6_tun_key.tun_id = 0;
5419 fl6.flowi6_uid = sock_net_uid(net, NULL);
5421 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5424 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5425 res.f6i == net->ipv6.fib6_null_entry))
5426 return BPF_FIB_LKUP_RET_NOT_FWDED;
5428 switch (res.fib6_type) {
5429 /* only unicast is forwarded */
5433 return BPF_FIB_LKUP_RET_BLACKHOLE;
5434 case RTN_UNREACHABLE:
5435 return BPF_FIB_LKUP_RET_UNREACHABLE;
5437 return BPF_FIB_LKUP_RET_PROHIBIT;
5439 return BPF_FIB_LKUP_RET_NOT_FWDED;
5442 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5443 fl6.flowi6_oif != 0, NULL, strict);
5446 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5447 if (params->tot_len > mtu)
5448 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5451 if (res.nh->fib_nh_lws)
5452 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5454 if (res.nh->fib_nh_gw_family)
5455 *dst = res.nh->fib_nh_gw6;
5457 dev = res.nh->fib_nh_dev;
5458 params->rt_metric = res.f6i->fib6_metric;
5459 params->ifindex = dev->ifindex;
5461 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
5464 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5466 return BPF_FIB_LKUP_RET_NO_NEIGH;
5468 return bpf_fib_set_fwd_params(params, neigh, dev);
5472 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
5473 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5475 if (plen < sizeof(*params))
5478 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5481 switch (params->family) {
5482 #if IS_ENABLED(CONFIG_INET)
5484 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
5487 #if IS_ENABLED(CONFIG_IPV6)
5489 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
5493 return -EAFNOSUPPORT;
5496 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
5497 .func = bpf_xdp_fib_lookup,
5499 .ret_type = RET_INTEGER,
5500 .arg1_type = ARG_PTR_TO_CTX,
5501 .arg2_type = ARG_PTR_TO_MEM,
5502 .arg3_type = ARG_CONST_SIZE,
5503 .arg4_type = ARG_ANYTHING,
5506 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
5507 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5509 struct net *net = dev_net(skb->dev);
5510 int rc = -EAFNOSUPPORT;
5512 if (plen < sizeof(*params))
5515 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5518 switch (params->family) {
5519 #if IS_ENABLED(CONFIG_INET)
5521 rc = bpf_ipv4_fib_lookup(net, params, flags, false);
5524 #if IS_ENABLED(CONFIG_IPV6)
5526 rc = bpf_ipv6_fib_lookup(net, params, flags, false);
5532 struct net_device *dev;
5534 dev = dev_get_by_index_rcu(net, params->ifindex);
5535 if (!is_skb_forwardable(dev, skb))
5536 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
5542 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
5543 .func = bpf_skb_fib_lookup,
5545 .ret_type = RET_INTEGER,
5546 .arg1_type = ARG_PTR_TO_CTX,
5547 .arg2_type = ARG_PTR_TO_MEM,
5548 .arg3_type = ARG_CONST_SIZE,
5549 .arg4_type = ARG_ANYTHING,
5552 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5553 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
5556 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
5558 if (!seg6_validate_srh(srh, len, false))
5562 case BPF_LWT_ENCAP_SEG6_INLINE:
5563 if (skb->protocol != htons(ETH_P_IPV6))
5566 err = seg6_do_srh_inline(skb, srh);
5568 case BPF_LWT_ENCAP_SEG6:
5569 skb_reset_inner_headers(skb);
5570 skb->encapsulation = 1;
5571 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
5577 bpf_compute_data_pointers(skb);
5581 ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5582 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
5584 return seg6_lookup_nexthop(skb, NULL, 0);
5586 #endif /* CONFIG_IPV6_SEG6_BPF */
5588 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5589 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
5592 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
5596 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
5600 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5601 case BPF_LWT_ENCAP_SEG6:
5602 case BPF_LWT_ENCAP_SEG6_INLINE:
5603 return bpf_push_seg6_encap(skb, type, hdr, len);
5605 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5606 case BPF_LWT_ENCAP_IP:
5607 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
5614 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
5615 void *, hdr, u32, len)
5618 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5619 case BPF_LWT_ENCAP_IP:
5620 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
5627 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
5628 .func = bpf_lwt_in_push_encap,
5630 .ret_type = RET_INTEGER,
5631 .arg1_type = ARG_PTR_TO_CTX,
5632 .arg2_type = ARG_ANYTHING,
5633 .arg3_type = ARG_PTR_TO_MEM,
5634 .arg4_type = ARG_CONST_SIZE
5637 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
5638 .func = bpf_lwt_xmit_push_encap,
5640 .ret_type = RET_INTEGER,
5641 .arg1_type = ARG_PTR_TO_CTX,
5642 .arg2_type = ARG_ANYTHING,
5643 .arg3_type = ARG_PTR_TO_MEM,
5644 .arg4_type = ARG_CONST_SIZE
5647 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5648 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
5649 const void *, from, u32, len)
5651 struct seg6_bpf_srh_state *srh_state =
5652 this_cpu_ptr(&seg6_bpf_srh_states);
5653 struct ipv6_sr_hdr *srh = srh_state->srh;
5654 void *srh_tlvs, *srh_end, *ptr;
5660 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
5661 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
5663 ptr = skb->data + offset;
5664 if (ptr >= srh_tlvs && ptr + len <= srh_end)
5665 srh_state->valid = false;
5666 else if (ptr < (void *)&srh->flags ||
5667 ptr + len > (void *)&srh->segments)
5670 if (unlikely(bpf_try_make_writable(skb, offset + len)))
5672 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5674 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5676 memcpy(skb->data + offset, from, len);
5680 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
5681 .func = bpf_lwt_seg6_store_bytes,
5683 .ret_type = RET_INTEGER,
5684 .arg1_type = ARG_PTR_TO_CTX,
5685 .arg2_type = ARG_ANYTHING,
5686 .arg3_type = ARG_PTR_TO_MEM,
5687 .arg4_type = ARG_CONST_SIZE
5690 static void bpf_update_srh_state(struct sk_buff *skb)
5692 struct seg6_bpf_srh_state *srh_state =
5693 this_cpu_ptr(&seg6_bpf_srh_states);
5696 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
5697 srh_state->srh = NULL;
5699 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5700 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
5701 srh_state->valid = true;
5705 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
5706 u32, action, void *, param, u32, param_len)
5708 struct seg6_bpf_srh_state *srh_state =
5709 this_cpu_ptr(&seg6_bpf_srh_states);
5714 case SEG6_LOCAL_ACTION_END_X:
5715 if (!seg6_bpf_has_valid_srh(skb))
5717 if (param_len != sizeof(struct in6_addr))
5719 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
5720 case SEG6_LOCAL_ACTION_END_T:
5721 if (!seg6_bpf_has_valid_srh(skb))
5723 if (param_len != sizeof(int))
5725 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5726 case SEG6_LOCAL_ACTION_END_DT6:
5727 if (!seg6_bpf_has_valid_srh(skb))
5729 if (param_len != sizeof(int))
5732 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
5734 if (!pskb_pull(skb, hdroff))
5737 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
5738 skb_reset_network_header(skb);
5739 skb_reset_transport_header(skb);
5740 skb->encapsulation = 0;
5742 bpf_compute_data_pointers(skb);
5743 bpf_update_srh_state(skb);
5744 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5745 case SEG6_LOCAL_ACTION_END_B6:
5746 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5748 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
5751 bpf_update_srh_state(skb);
5754 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
5755 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5757 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
5760 bpf_update_srh_state(skb);
5768 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
5769 .func = bpf_lwt_seg6_action,
5771 .ret_type = RET_INTEGER,
5772 .arg1_type = ARG_PTR_TO_CTX,
5773 .arg2_type = ARG_ANYTHING,
5774 .arg3_type = ARG_PTR_TO_MEM,
5775 .arg4_type = ARG_CONST_SIZE
5778 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
5781 struct seg6_bpf_srh_state *srh_state =
5782 this_cpu_ptr(&seg6_bpf_srh_states);
5783 struct ipv6_sr_hdr *srh = srh_state->srh;
5784 void *srh_end, *srh_tlvs, *ptr;
5785 struct ipv6hdr *hdr;
5789 if (unlikely(srh == NULL))
5792 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
5793 ((srh->first_segment + 1) << 4));
5794 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
5796 ptr = skb->data + offset;
5798 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
5800 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
5804 ret = skb_cow_head(skb, len);
5805 if (unlikely(ret < 0))
5808 ret = bpf_skb_net_hdr_push(skb, offset, len);
5810 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
5813 bpf_compute_data_pointers(skb);
5814 if (unlikely(ret < 0))
5817 hdr = (struct ipv6hdr *)skb->data;
5818 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5820 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5822 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5823 srh_state->hdrlen += len;
5824 srh_state->valid = false;
5828 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
5829 .func = bpf_lwt_seg6_adjust_srh,
5831 .ret_type = RET_INTEGER,
5832 .arg1_type = ARG_PTR_TO_CTX,
5833 .arg2_type = ARG_ANYTHING,
5834 .arg3_type = ARG_ANYTHING,
5836 #endif /* CONFIG_IPV6_SEG6_BPF */
5839 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
5840 int dif, int sdif, u8 family, u8 proto)
5842 bool refcounted = false;
5843 struct sock *sk = NULL;
5845 if (family == AF_INET) {
5846 __be32 src4 = tuple->ipv4.saddr;
5847 __be32 dst4 = tuple->ipv4.daddr;
5849 if (proto == IPPROTO_TCP)
5850 sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
5851 src4, tuple->ipv4.sport,
5852 dst4, tuple->ipv4.dport,
5853 dif, sdif, &refcounted);
5855 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
5856 dst4, tuple->ipv4.dport,
5857 dif, sdif, &udp_table, NULL);
5858 #if IS_ENABLED(CONFIG_IPV6)
5860 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
5861 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
5863 if (proto == IPPROTO_TCP)
5864 sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
5865 src6, tuple->ipv6.sport,
5866 dst6, ntohs(tuple->ipv6.dport),
5867 dif, sdif, &refcounted);
5868 else if (likely(ipv6_bpf_stub))
5869 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
5870 src6, tuple->ipv6.sport,
5871 dst6, tuple->ipv6.dport,
5877 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
5878 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
5884 /* bpf_skc_lookup performs the core lookup for different types of sockets,
5885 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
5886 * Returns the socket as an 'unsigned long' to simplify the casting in the
5887 * callers to satisfy BPF_CALL declarations.
5889 static struct sock *
5890 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5891 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
5894 struct sock *sk = NULL;
5895 u8 family = AF_UNSPEC;
5899 if (len == sizeof(tuple->ipv4))
5901 else if (len == sizeof(tuple->ipv6))
5906 if (unlikely(family == AF_UNSPEC || flags ||
5907 !((s32)netns_id < 0 || netns_id <= S32_MAX)))
5910 if (family == AF_INET)
5911 sdif = inet_sdif(skb);
5913 sdif = inet6_sdif(skb);
5915 if ((s32)netns_id < 0) {
5917 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
5919 net = get_net_ns_by_id(caller_net, netns_id);
5922 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
5930 static struct sock *
5931 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5932 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
5935 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
5936 ifindex, proto, netns_id, flags);
5939 sk = sk_to_full_sk(sk);
5940 if (!sk_fullsock(sk)) {
5949 static struct sock *
5950 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5951 u8 proto, u64 netns_id, u64 flags)
5953 struct net *caller_net;
5957 caller_net = dev_net(skb->dev);
5958 ifindex = skb->dev->ifindex;
5960 caller_net = sock_net(skb->sk);
5964 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
5968 static struct sock *
5969 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5970 u8 proto, u64 netns_id, u64 flags)
5972 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
5976 sk = sk_to_full_sk(sk);
5977 if (!sk_fullsock(sk)) {
5986 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
5987 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5989 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
5993 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
5994 .func = bpf_skc_lookup_tcp,
5997 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
5998 .arg1_type = ARG_PTR_TO_CTX,
5999 .arg2_type = ARG_PTR_TO_MEM,
6000 .arg3_type = ARG_CONST_SIZE,
6001 .arg4_type = ARG_ANYTHING,
6002 .arg5_type = ARG_ANYTHING,
6005 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6006 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6008 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6012 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6013 .func = bpf_sk_lookup_tcp,
6016 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6017 .arg1_type = ARG_PTR_TO_CTX,
6018 .arg2_type = ARG_PTR_TO_MEM,
6019 .arg3_type = ARG_CONST_SIZE,
6020 .arg4_type = ARG_ANYTHING,
6021 .arg5_type = ARG_ANYTHING,
6024 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6025 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6027 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6031 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6032 .func = bpf_sk_lookup_udp,
6035 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6036 .arg1_type = ARG_PTR_TO_CTX,
6037 .arg2_type = ARG_PTR_TO_MEM,
6038 .arg3_type = ARG_CONST_SIZE,
6039 .arg4_type = ARG_ANYTHING,
6040 .arg5_type = ARG_ANYTHING,
6043 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6045 if (sk && sk_is_refcounted(sk))
6050 static const struct bpf_func_proto bpf_sk_release_proto = {
6051 .func = bpf_sk_release,
6053 .ret_type = RET_INTEGER,
6054 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6057 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6058 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6060 struct net *caller_net = dev_net(ctx->rxq->dev);
6061 int ifindex = ctx->rxq->dev->ifindex;
6063 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6064 ifindex, IPPROTO_UDP, netns_id,
6068 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6069 .func = bpf_xdp_sk_lookup_udp,
6072 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6073 .arg1_type = ARG_PTR_TO_CTX,
6074 .arg2_type = ARG_PTR_TO_MEM,
6075 .arg3_type = ARG_CONST_SIZE,
6076 .arg4_type = ARG_ANYTHING,
6077 .arg5_type = ARG_ANYTHING,
6080 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6081 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6083 struct net *caller_net = dev_net(ctx->rxq->dev);
6084 int ifindex = ctx->rxq->dev->ifindex;
6086 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6087 ifindex, IPPROTO_TCP, netns_id,
6091 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6092 .func = bpf_xdp_skc_lookup_tcp,
6095 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6096 .arg1_type = ARG_PTR_TO_CTX,
6097 .arg2_type = ARG_PTR_TO_MEM,
6098 .arg3_type = ARG_CONST_SIZE,
6099 .arg4_type = ARG_ANYTHING,
6100 .arg5_type = ARG_ANYTHING,
6103 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6104 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6106 struct net *caller_net = dev_net(ctx->rxq->dev);
6107 int ifindex = ctx->rxq->dev->ifindex;
6109 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6110 ifindex, IPPROTO_TCP, netns_id,
6114 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6115 .func = bpf_xdp_sk_lookup_tcp,
6118 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6119 .arg1_type = ARG_PTR_TO_CTX,
6120 .arg2_type = ARG_PTR_TO_MEM,
6121 .arg3_type = ARG_CONST_SIZE,
6122 .arg4_type = ARG_ANYTHING,
6123 .arg5_type = ARG_ANYTHING,
6126 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6127 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6129 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6130 sock_net(ctx->sk), 0,
6131 IPPROTO_TCP, netns_id, flags);
6134 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6135 .func = bpf_sock_addr_skc_lookup_tcp,
6137 .ret_type = RET_PTR_TO_SOCK_COMMON_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_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6146 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6148 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6149 sock_net(ctx->sk), 0, IPPROTO_TCP,
6153 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6154 .func = bpf_sock_addr_sk_lookup_tcp,
6156 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6157 .arg1_type = ARG_PTR_TO_CTX,
6158 .arg2_type = ARG_PTR_TO_MEM,
6159 .arg3_type = ARG_CONST_SIZE,
6160 .arg4_type = ARG_ANYTHING,
6161 .arg5_type = ARG_ANYTHING,
6164 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6165 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6167 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6168 sock_net(ctx->sk), 0, IPPROTO_UDP,
6172 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6173 .func = bpf_sock_addr_sk_lookup_udp,
6175 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6176 .arg1_type = ARG_PTR_TO_CTX,
6177 .arg2_type = ARG_PTR_TO_MEM,
6178 .arg3_type = ARG_CONST_SIZE,
6179 .arg4_type = ARG_ANYTHING,
6180 .arg5_type = ARG_ANYTHING,
6183 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6184 struct bpf_insn_access_aux *info)
6186 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6190 if (off % size != 0)
6194 case offsetof(struct bpf_tcp_sock, bytes_received):
6195 case offsetof(struct bpf_tcp_sock, bytes_acked):
6196 return size == sizeof(__u64);
6198 return size == sizeof(__u32);
6202 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6203 const struct bpf_insn *si,
6204 struct bpf_insn *insn_buf,
6205 struct bpf_prog *prog, u32 *target_size)
6207 struct bpf_insn *insn = insn_buf;
6209 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \
6211 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \
6212 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6213 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6214 si->dst_reg, si->src_reg, \
6215 offsetof(struct tcp_sock, FIELD)); \
6218 #define BPF_INET_SOCK_GET_COMMON(FIELD) \
6220 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \
6222 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6223 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6224 struct inet_connection_sock, \
6226 si->dst_reg, si->src_reg, \
6228 struct inet_connection_sock, \
6232 if (insn > insn_buf)
6233 return insn - insn_buf;
6236 case offsetof(struct bpf_tcp_sock, rtt_min):
6237 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
6238 sizeof(struct minmax));
6239 BUILD_BUG_ON(sizeof(struct minmax) <
6240 sizeof(struct minmax_sample));
6242 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6243 offsetof(struct tcp_sock, rtt_min) +
6244 offsetof(struct minmax_sample, v));
6246 case offsetof(struct bpf_tcp_sock, snd_cwnd):
6247 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
6249 case offsetof(struct bpf_tcp_sock, srtt_us):
6250 BPF_TCP_SOCK_GET_COMMON(srtt_us);
6252 case offsetof(struct bpf_tcp_sock, snd_ssthresh):
6253 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
6255 case offsetof(struct bpf_tcp_sock, rcv_nxt):
6256 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
6258 case offsetof(struct bpf_tcp_sock, snd_nxt):
6259 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
6261 case offsetof(struct bpf_tcp_sock, snd_una):
6262 BPF_TCP_SOCK_GET_COMMON(snd_una);
6264 case offsetof(struct bpf_tcp_sock, mss_cache):
6265 BPF_TCP_SOCK_GET_COMMON(mss_cache);
6267 case offsetof(struct bpf_tcp_sock, ecn_flags):
6268 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
6270 case offsetof(struct bpf_tcp_sock, rate_delivered):
6271 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
6273 case offsetof(struct bpf_tcp_sock, rate_interval_us):
6274 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
6276 case offsetof(struct bpf_tcp_sock, packets_out):
6277 BPF_TCP_SOCK_GET_COMMON(packets_out);
6279 case offsetof(struct bpf_tcp_sock, retrans_out):
6280 BPF_TCP_SOCK_GET_COMMON(retrans_out);
6282 case offsetof(struct bpf_tcp_sock, total_retrans):
6283 BPF_TCP_SOCK_GET_COMMON(total_retrans);
6285 case offsetof(struct bpf_tcp_sock, segs_in):
6286 BPF_TCP_SOCK_GET_COMMON(segs_in);
6288 case offsetof(struct bpf_tcp_sock, data_segs_in):
6289 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
6291 case offsetof(struct bpf_tcp_sock, segs_out):
6292 BPF_TCP_SOCK_GET_COMMON(segs_out);
6294 case offsetof(struct bpf_tcp_sock, data_segs_out):
6295 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
6297 case offsetof(struct bpf_tcp_sock, lost_out):
6298 BPF_TCP_SOCK_GET_COMMON(lost_out);
6300 case offsetof(struct bpf_tcp_sock, sacked_out):
6301 BPF_TCP_SOCK_GET_COMMON(sacked_out);
6303 case offsetof(struct bpf_tcp_sock, bytes_received):
6304 BPF_TCP_SOCK_GET_COMMON(bytes_received);
6306 case offsetof(struct bpf_tcp_sock, bytes_acked):
6307 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
6309 case offsetof(struct bpf_tcp_sock, dsack_dups):
6310 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
6312 case offsetof(struct bpf_tcp_sock, delivered):
6313 BPF_TCP_SOCK_GET_COMMON(delivered);
6315 case offsetof(struct bpf_tcp_sock, delivered_ce):
6316 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
6318 case offsetof(struct bpf_tcp_sock, icsk_retransmits):
6319 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
6323 return insn - insn_buf;
6326 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
6328 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
6329 return (unsigned long)sk;
6331 return (unsigned long)NULL;
6334 const struct bpf_func_proto bpf_tcp_sock_proto = {
6335 .func = bpf_tcp_sock,
6337 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
6338 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6341 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
6343 sk = sk_to_full_sk(sk);
6345 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
6346 return (unsigned long)sk;
6348 return (unsigned long)NULL;
6351 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
6352 .func = bpf_get_listener_sock,
6354 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6355 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6358 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
6360 unsigned int iphdr_len;
6362 switch (skb_protocol(skb, true)) {
6363 case cpu_to_be16(ETH_P_IP):
6364 iphdr_len = sizeof(struct iphdr);
6366 case cpu_to_be16(ETH_P_IPV6):
6367 iphdr_len = sizeof(struct ipv6hdr);
6373 if (skb_headlen(skb) < iphdr_len)
6376 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
6379 return INET_ECN_set_ce(skb);
6382 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6383 struct bpf_insn_access_aux *info)
6385 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
6388 if (off % size != 0)
6393 return size == sizeof(__u32);
6397 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
6398 const struct bpf_insn *si,
6399 struct bpf_insn *insn_buf,
6400 struct bpf_prog *prog, u32 *target_size)
6402 struct bpf_insn *insn = insn_buf;
6404 #define BPF_XDP_SOCK_GET(FIELD) \
6406 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \
6407 sizeof_field(struct bpf_xdp_sock, FIELD)); \
6408 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
6409 si->dst_reg, si->src_reg, \
6410 offsetof(struct xdp_sock, FIELD)); \
6414 case offsetof(struct bpf_xdp_sock, queue_id):
6415 BPF_XDP_SOCK_GET(queue_id);
6419 return insn - insn_buf;
6422 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
6423 .func = bpf_skb_ecn_set_ce,
6425 .ret_type = RET_INTEGER,
6426 .arg1_type = ARG_PTR_TO_CTX,
6429 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6430 struct tcphdr *, th, u32, th_len)
6432 #ifdef CONFIG_SYN_COOKIES
6436 if (unlikely(!sk || th_len < sizeof(*th)))
6439 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
6440 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6443 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
6446 if (!th->ack || th->rst || th->syn)
6449 if (tcp_synq_no_recent_overflow(sk))
6452 cookie = ntohl(th->ack_seq) - 1;
6454 switch (sk->sk_family) {
6456 if (unlikely(iph_len < sizeof(struct iphdr)))
6459 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
6462 #if IS_BUILTIN(CONFIG_IPV6)
6464 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6467 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
6469 #endif /* CONFIG_IPV6 */
6472 return -EPROTONOSUPPORT;
6484 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
6485 .func = bpf_tcp_check_syncookie,
6488 .ret_type = RET_INTEGER,
6489 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6490 .arg2_type = ARG_PTR_TO_MEM,
6491 .arg3_type = ARG_CONST_SIZE,
6492 .arg4_type = ARG_PTR_TO_MEM,
6493 .arg5_type = ARG_CONST_SIZE,
6496 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6497 struct tcphdr *, th, u32, th_len)
6499 #ifdef CONFIG_SYN_COOKIES
6503 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
6506 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6509 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
6512 if (!th->syn || th->ack || th->fin || th->rst)
6515 if (unlikely(iph_len < sizeof(struct iphdr)))
6518 /* Both struct iphdr and struct ipv6hdr have the version field at the
6519 * same offset so we can cast to the shorter header (struct iphdr).
6521 switch (((struct iphdr *)iph)->version) {
6523 if (sk->sk_family == AF_INET6 && sk->sk_ipv6only)
6526 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
6529 #if IS_BUILTIN(CONFIG_IPV6)
6531 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6534 if (sk->sk_family != AF_INET6)
6537 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
6539 #endif /* CONFIG_IPV6 */
6542 return -EPROTONOSUPPORT;
6547 return cookie | ((u64)mss << 32);
6550 #endif /* CONFIG_SYN_COOKIES */
6553 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
6554 .func = bpf_tcp_gen_syncookie,
6555 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */
6557 .ret_type = RET_INTEGER,
6558 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6559 .arg2_type = ARG_PTR_TO_MEM,
6560 .arg3_type = ARG_CONST_SIZE,
6561 .arg4_type = ARG_PTR_TO_MEM,
6562 .arg5_type = ARG_CONST_SIZE,
6565 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
6567 if (!sk || flags != 0)
6569 if (!skb_at_tc_ingress(skb))
6571 if (unlikely(dev_net(skb->dev) != sock_net(sk)))
6572 return -ENETUNREACH;
6573 if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
6574 return -ESOCKTNOSUPPORT;
6575 if (sk_is_refcounted(sk) &&
6576 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
6581 skb->destructor = sock_pfree;
6586 static const struct bpf_func_proto bpf_sk_assign_proto = {
6587 .func = bpf_sk_assign,
6589 .ret_type = RET_INTEGER,
6590 .arg1_type = ARG_PTR_TO_CTX,
6591 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6592 .arg3_type = ARG_ANYTHING,
6595 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
6596 u8 search_kind, const u8 *magic,
6597 u8 magic_len, bool *eol)
6603 while (op < opend) {
6606 if (kind == TCPOPT_EOL) {
6608 return ERR_PTR(-ENOMSG);
6609 } else if (kind == TCPOPT_NOP) {
6614 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
6615 /* Something is wrong in the received header.
6616 * Follow the TCP stack's tcp_parse_options()
6617 * and just bail here.
6619 return ERR_PTR(-EFAULT);
6622 if (search_kind == kind) {
6626 if (magic_len > kind_len - 2)
6627 return ERR_PTR(-ENOMSG);
6629 if (!memcmp(&op[2], magic, magic_len))
6636 return ERR_PTR(-ENOMSG);
6639 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6640 void *, search_res, u32, len, u64, flags)
6642 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
6643 const u8 *op, *opend, *magic, *search = search_res;
6644 u8 search_kind, search_len, copy_len, magic_len;
6647 /* 2 byte is the minimal option len except TCPOPT_NOP and
6648 * TCPOPT_EOL which are useless for the bpf prog to learn
6649 * and this helper disallow loading them also.
6651 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
6654 search_kind = search[0];
6655 search_len = search[1];
6657 if (search_len > len || search_kind == TCPOPT_NOP ||
6658 search_kind == TCPOPT_EOL)
6661 if (search_kind == TCPOPT_EXP || search_kind == 253) {
6662 /* 16 or 32 bit magic. +2 for kind and kind length */
6663 if (search_len != 4 && search_len != 6)
6666 magic_len = search_len - 2;
6675 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
6680 op += sizeof(struct tcphdr);
6682 if (!bpf_sock->skb ||
6683 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6684 /* This bpf_sock->op cannot call this helper */
6687 opend = bpf_sock->skb_data_end;
6688 op = bpf_sock->skb->data + sizeof(struct tcphdr);
6691 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
6698 if (copy_len > len) {
6703 memcpy(search_res, op, copy_len);
6707 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
6708 .func = bpf_sock_ops_load_hdr_opt,
6710 .ret_type = RET_INTEGER,
6711 .arg1_type = ARG_PTR_TO_CTX,
6712 .arg2_type = ARG_PTR_TO_MEM,
6713 .arg3_type = ARG_CONST_SIZE,
6714 .arg4_type = ARG_ANYTHING,
6717 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6718 const void *, from, u32, len, u64, flags)
6720 u8 new_kind, new_kind_len, magic_len = 0, *opend;
6721 const u8 *op, *new_op, *magic = NULL;
6722 struct sk_buff *skb;
6725 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
6728 if (len < 2 || flags)
6732 new_kind = new_op[0];
6733 new_kind_len = new_op[1];
6735 if (new_kind_len > len || new_kind == TCPOPT_NOP ||
6736 new_kind == TCPOPT_EOL)
6739 if (new_kind_len > bpf_sock->remaining_opt_len)
6742 /* 253 is another experimental kind */
6743 if (new_kind == TCPOPT_EXP || new_kind == 253) {
6744 if (new_kind_len < 4)
6746 /* Match for the 2 byte magic also.
6747 * RFC 6994: the magic could be 2 or 4 bytes.
6748 * Hence, matching by 2 byte only is on the
6749 * conservative side but it is the right
6750 * thing to do for the 'search-for-duplication'
6757 /* Check for duplication */
6758 skb = bpf_sock->skb;
6759 op = skb->data + sizeof(struct tcphdr);
6760 opend = bpf_sock->skb_data_end;
6762 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
6767 if (PTR_ERR(op) != -ENOMSG)
6771 /* The option has been ended. Treat it as no more
6772 * header option can be written.
6776 /* No duplication found. Store the header option. */
6777 memcpy(opend, from, new_kind_len);
6779 bpf_sock->remaining_opt_len -= new_kind_len;
6780 bpf_sock->skb_data_end += new_kind_len;
6785 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
6786 .func = bpf_sock_ops_store_hdr_opt,
6788 .ret_type = RET_INTEGER,
6789 .arg1_type = ARG_PTR_TO_CTX,
6790 .arg2_type = ARG_PTR_TO_MEM,
6791 .arg3_type = ARG_CONST_SIZE,
6792 .arg4_type = ARG_ANYTHING,
6795 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6796 u32, len, u64, flags)
6798 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6801 if (flags || len < 2)
6804 if (len > bpf_sock->remaining_opt_len)
6807 bpf_sock->remaining_opt_len -= len;
6812 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
6813 .func = bpf_sock_ops_reserve_hdr_opt,
6815 .ret_type = RET_INTEGER,
6816 .arg1_type = ARG_PTR_TO_CTX,
6817 .arg2_type = ARG_ANYTHING,
6818 .arg3_type = ARG_ANYTHING,
6821 #endif /* CONFIG_INET */
6823 bool bpf_helper_changes_pkt_data(void *func)
6825 if (func == bpf_skb_vlan_push ||
6826 func == bpf_skb_vlan_pop ||
6827 func == bpf_skb_store_bytes ||
6828 func == bpf_skb_change_proto ||
6829 func == bpf_skb_change_head ||
6830 func == sk_skb_change_head ||
6831 func == bpf_skb_change_tail ||
6832 func == sk_skb_change_tail ||
6833 func == bpf_skb_adjust_room ||
6834 func == sk_skb_adjust_room ||
6835 func == bpf_skb_pull_data ||
6836 func == sk_skb_pull_data ||
6837 func == bpf_clone_redirect ||
6838 func == bpf_l3_csum_replace ||
6839 func == bpf_l4_csum_replace ||
6840 func == bpf_xdp_adjust_head ||
6841 func == bpf_xdp_adjust_meta ||
6842 func == bpf_msg_pull_data ||
6843 func == bpf_msg_push_data ||
6844 func == bpf_msg_pop_data ||
6845 func == bpf_xdp_adjust_tail ||
6846 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6847 func == bpf_lwt_seg6_store_bytes ||
6848 func == bpf_lwt_seg6_adjust_srh ||
6849 func == bpf_lwt_seg6_action ||
6852 func == bpf_sock_ops_store_hdr_opt ||
6854 func == bpf_lwt_in_push_encap ||
6855 func == bpf_lwt_xmit_push_encap)
6861 const struct bpf_func_proto bpf_event_output_data_proto __weak;
6862 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
6864 static const struct bpf_func_proto *
6865 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6868 /* inet and inet6 sockets are created in a process
6869 * context so there is always a valid uid/gid
6871 case BPF_FUNC_get_current_uid_gid:
6872 return &bpf_get_current_uid_gid_proto;
6873 case BPF_FUNC_get_local_storage:
6874 return &bpf_get_local_storage_proto;
6875 case BPF_FUNC_get_socket_cookie:
6876 return &bpf_get_socket_cookie_sock_proto;
6877 case BPF_FUNC_get_netns_cookie:
6878 return &bpf_get_netns_cookie_sock_proto;
6879 case BPF_FUNC_perf_event_output:
6880 return &bpf_event_output_data_proto;
6881 case BPF_FUNC_get_current_pid_tgid:
6882 return &bpf_get_current_pid_tgid_proto;
6883 case BPF_FUNC_get_current_comm:
6884 return &bpf_get_current_comm_proto;
6885 #ifdef CONFIG_CGROUPS
6886 case BPF_FUNC_get_current_cgroup_id:
6887 return &bpf_get_current_cgroup_id_proto;
6888 case BPF_FUNC_get_current_ancestor_cgroup_id:
6889 return &bpf_get_current_ancestor_cgroup_id_proto;
6891 #ifdef CONFIG_CGROUP_NET_CLASSID
6892 case BPF_FUNC_get_cgroup_classid:
6893 return &bpf_get_cgroup_classid_curr_proto;
6895 case BPF_FUNC_sk_storage_get:
6896 return &bpf_sk_storage_get_cg_sock_proto;
6898 return bpf_base_func_proto(func_id);
6902 static const struct bpf_func_proto *
6903 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6906 /* inet and inet6 sockets are created in a process
6907 * context so there is always a valid uid/gid
6909 case BPF_FUNC_get_current_uid_gid:
6910 return &bpf_get_current_uid_gid_proto;
6912 switch (prog->expected_attach_type) {
6913 case BPF_CGROUP_INET4_CONNECT:
6914 case BPF_CGROUP_INET6_CONNECT:
6915 return &bpf_bind_proto;
6919 case BPF_FUNC_get_socket_cookie:
6920 return &bpf_get_socket_cookie_sock_addr_proto;
6921 case BPF_FUNC_get_netns_cookie:
6922 return &bpf_get_netns_cookie_sock_addr_proto;
6923 case BPF_FUNC_get_local_storage:
6924 return &bpf_get_local_storage_proto;
6925 case BPF_FUNC_perf_event_output:
6926 return &bpf_event_output_data_proto;
6927 case BPF_FUNC_get_current_pid_tgid:
6928 return &bpf_get_current_pid_tgid_proto;
6929 case BPF_FUNC_get_current_comm:
6930 return &bpf_get_current_comm_proto;
6931 #ifdef CONFIG_CGROUPS
6932 case BPF_FUNC_get_current_cgroup_id:
6933 return &bpf_get_current_cgroup_id_proto;
6934 case BPF_FUNC_get_current_ancestor_cgroup_id:
6935 return &bpf_get_current_ancestor_cgroup_id_proto;
6937 #ifdef CONFIG_CGROUP_NET_CLASSID
6938 case BPF_FUNC_get_cgroup_classid:
6939 return &bpf_get_cgroup_classid_curr_proto;
6942 case BPF_FUNC_sk_lookup_tcp:
6943 return &bpf_sock_addr_sk_lookup_tcp_proto;
6944 case BPF_FUNC_sk_lookup_udp:
6945 return &bpf_sock_addr_sk_lookup_udp_proto;
6946 case BPF_FUNC_sk_release:
6947 return &bpf_sk_release_proto;
6948 case BPF_FUNC_skc_lookup_tcp:
6949 return &bpf_sock_addr_skc_lookup_tcp_proto;
6950 #endif /* CONFIG_INET */
6951 case BPF_FUNC_sk_storage_get:
6952 return &bpf_sk_storage_get_proto;
6953 case BPF_FUNC_sk_storage_delete:
6954 return &bpf_sk_storage_delete_proto;
6955 case BPF_FUNC_setsockopt:
6956 switch (prog->expected_attach_type) {
6957 case BPF_CGROUP_INET4_CONNECT:
6958 case BPF_CGROUP_INET6_CONNECT:
6959 return &bpf_sock_addr_setsockopt_proto;
6963 case BPF_FUNC_getsockopt:
6964 switch (prog->expected_attach_type) {
6965 case BPF_CGROUP_INET4_CONNECT:
6966 case BPF_CGROUP_INET6_CONNECT:
6967 return &bpf_sock_addr_getsockopt_proto;
6972 return bpf_sk_base_func_proto(func_id);
6976 static const struct bpf_func_proto *
6977 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6980 case BPF_FUNC_skb_load_bytes:
6981 return &bpf_skb_load_bytes_proto;
6982 case BPF_FUNC_skb_load_bytes_relative:
6983 return &bpf_skb_load_bytes_relative_proto;
6984 case BPF_FUNC_get_socket_cookie:
6985 return &bpf_get_socket_cookie_proto;
6986 case BPF_FUNC_get_socket_uid:
6987 return &bpf_get_socket_uid_proto;
6988 case BPF_FUNC_perf_event_output:
6989 return &bpf_skb_event_output_proto;
6991 return bpf_sk_base_func_proto(func_id);
6995 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
6996 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
6998 static const struct bpf_func_proto *
6999 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7002 case BPF_FUNC_get_local_storage:
7003 return &bpf_get_local_storage_proto;
7004 case BPF_FUNC_sk_fullsock:
7005 return &bpf_sk_fullsock_proto;
7006 case BPF_FUNC_sk_storage_get:
7007 return &bpf_sk_storage_get_proto;
7008 case BPF_FUNC_sk_storage_delete:
7009 return &bpf_sk_storage_delete_proto;
7010 case BPF_FUNC_perf_event_output:
7011 return &bpf_skb_event_output_proto;
7012 #ifdef CONFIG_SOCK_CGROUP_DATA
7013 case BPF_FUNC_skb_cgroup_id:
7014 return &bpf_skb_cgroup_id_proto;
7015 case BPF_FUNC_skb_ancestor_cgroup_id:
7016 return &bpf_skb_ancestor_cgroup_id_proto;
7017 case BPF_FUNC_sk_cgroup_id:
7018 return &bpf_sk_cgroup_id_proto;
7019 case BPF_FUNC_sk_ancestor_cgroup_id:
7020 return &bpf_sk_ancestor_cgroup_id_proto;
7023 case BPF_FUNC_sk_lookup_tcp:
7024 return &bpf_sk_lookup_tcp_proto;
7025 case BPF_FUNC_sk_lookup_udp:
7026 return &bpf_sk_lookup_udp_proto;
7027 case BPF_FUNC_sk_release:
7028 return &bpf_sk_release_proto;
7029 case BPF_FUNC_skc_lookup_tcp:
7030 return &bpf_skc_lookup_tcp_proto;
7031 case BPF_FUNC_tcp_sock:
7032 return &bpf_tcp_sock_proto;
7033 case BPF_FUNC_get_listener_sock:
7034 return &bpf_get_listener_sock_proto;
7035 case BPF_FUNC_skb_ecn_set_ce:
7036 return &bpf_skb_ecn_set_ce_proto;
7039 return sk_filter_func_proto(func_id, prog);
7043 static const struct bpf_func_proto *
7044 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7047 case BPF_FUNC_skb_store_bytes:
7048 return &bpf_skb_store_bytes_proto;
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_skb_pull_data:
7054 return &bpf_skb_pull_data_proto;
7055 case BPF_FUNC_csum_diff:
7056 return &bpf_csum_diff_proto;
7057 case BPF_FUNC_csum_update:
7058 return &bpf_csum_update_proto;
7059 case BPF_FUNC_csum_level:
7060 return &bpf_csum_level_proto;
7061 case BPF_FUNC_l3_csum_replace:
7062 return &bpf_l3_csum_replace_proto;
7063 case BPF_FUNC_l4_csum_replace:
7064 return &bpf_l4_csum_replace_proto;
7065 case BPF_FUNC_clone_redirect:
7066 return &bpf_clone_redirect_proto;
7067 case BPF_FUNC_get_cgroup_classid:
7068 return &bpf_get_cgroup_classid_proto;
7069 case BPF_FUNC_skb_vlan_push:
7070 return &bpf_skb_vlan_push_proto;
7071 case BPF_FUNC_skb_vlan_pop:
7072 return &bpf_skb_vlan_pop_proto;
7073 case BPF_FUNC_skb_change_proto:
7074 return &bpf_skb_change_proto_proto;
7075 case BPF_FUNC_skb_change_type:
7076 return &bpf_skb_change_type_proto;
7077 case BPF_FUNC_skb_adjust_room:
7078 return &bpf_skb_adjust_room_proto;
7079 case BPF_FUNC_skb_change_tail:
7080 return &bpf_skb_change_tail_proto;
7081 case BPF_FUNC_skb_change_head:
7082 return &bpf_skb_change_head_proto;
7083 case BPF_FUNC_skb_get_tunnel_key:
7084 return &bpf_skb_get_tunnel_key_proto;
7085 case BPF_FUNC_skb_set_tunnel_key:
7086 return bpf_get_skb_set_tunnel_proto(func_id);
7087 case BPF_FUNC_skb_get_tunnel_opt:
7088 return &bpf_skb_get_tunnel_opt_proto;
7089 case BPF_FUNC_skb_set_tunnel_opt:
7090 return bpf_get_skb_set_tunnel_proto(func_id);
7091 case BPF_FUNC_redirect:
7092 return &bpf_redirect_proto;
7093 case BPF_FUNC_redirect_neigh:
7094 return &bpf_redirect_neigh_proto;
7095 case BPF_FUNC_redirect_peer:
7096 return &bpf_redirect_peer_proto;
7097 case BPF_FUNC_get_route_realm:
7098 return &bpf_get_route_realm_proto;
7099 case BPF_FUNC_get_hash_recalc:
7100 return &bpf_get_hash_recalc_proto;
7101 case BPF_FUNC_set_hash_invalid:
7102 return &bpf_set_hash_invalid_proto;
7103 case BPF_FUNC_set_hash:
7104 return &bpf_set_hash_proto;
7105 case BPF_FUNC_perf_event_output:
7106 return &bpf_skb_event_output_proto;
7107 case BPF_FUNC_get_smp_processor_id:
7108 return &bpf_get_smp_processor_id_proto;
7109 case BPF_FUNC_skb_under_cgroup:
7110 return &bpf_skb_under_cgroup_proto;
7111 case BPF_FUNC_get_socket_cookie:
7112 return &bpf_get_socket_cookie_proto;
7113 case BPF_FUNC_get_socket_uid:
7114 return &bpf_get_socket_uid_proto;
7115 case BPF_FUNC_fib_lookup:
7116 return &bpf_skb_fib_lookup_proto;
7117 case BPF_FUNC_sk_fullsock:
7118 return &bpf_sk_fullsock_proto;
7119 case BPF_FUNC_sk_storage_get:
7120 return &bpf_sk_storage_get_proto;
7121 case BPF_FUNC_sk_storage_delete:
7122 return &bpf_sk_storage_delete_proto;
7124 case BPF_FUNC_skb_get_xfrm_state:
7125 return &bpf_skb_get_xfrm_state_proto;
7127 #ifdef CONFIG_CGROUP_NET_CLASSID
7128 case BPF_FUNC_skb_cgroup_classid:
7129 return &bpf_skb_cgroup_classid_proto;
7131 #ifdef CONFIG_SOCK_CGROUP_DATA
7132 case BPF_FUNC_skb_cgroup_id:
7133 return &bpf_skb_cgroup_id_proto;
7134 case BPF_FUNC_skb_ancestor_cgroup_id:
7135 return &bpf_skb_ancestor_cgroup_id_proto;
7138 case BPF_FUNC_sk_lookup_tcp:
7139 return &bpf_sk_lookup_tcp_proto;
7140 case BPF_FUNC_sk_lookup_udp:
7141 return &bpf_sk_lookup_udp_proto;
7142 case BPF_FUNC_sk_release:
7143 return &bpf_sk_release_proto;
7144 case BPF_FUNC_tcp_sock:
7145 return &bpf_tcp_sock_proto;
7146 case BPF_FUNC_get_listener_sock:
7147 return &bpf_get_listener_sock_proto;
7148 case BPF_FUNC_skc_lookup_tcp:
7149 return &bpf_skc_lookup_tcp_proto;
7150 case BPF_FUNC_tcp_check_syncookie:
7151 return &bpf_tcp_check_syncookie_proto;
7152 case BPF_FUNC_skb_ecn_set_ce:
7153 return &bpf_skb_ecn_set_ce_proto;
7154 case BPF_FUNC_tcp_gen_syncookie:
7155 return &bpf_tcp_gen_syncookie_proto;
7156 case BPF_FUNC_sk_assign:
7157 return &bpf_sk_assign_proto;
7160 return bpf_sk_base_func_proto(func_id);
7164 static const struct bpf_func_proto *
7165 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7168 case BPF_FUNC_perf_event_output:
7169 return &bpf_xdp_event_output_proto;
7170 case BPF_FUNC_get_smp_processor_id:
7171 return &bpf_get_smp_processor_id_proto;
7172 case BPF_FUNC_csum_diff:
7173 return &bpf_csum_diff_proto;
7174 case BPF_FUNC_xdp_adjust_head:
7175 return &bpf_xdp_adjust_head_proto;
7176 case BPF_FUNC_xdp_adjust_meta:
7177 return &bpf_xdp_adjust_meta_proto;
7178 case BPF_FUNC_redirect:
7179 return &bpf_xdp_redirect_proto;
7180 case BPF_FUNC_redirect_map:
7181 return &bpf_xdp_redirect_map_proto;
7182 case BPF_FUNC_xdp_adjust_tail:
7183 return &bpf_xdp_adjust_tail_proto;
7184 case BPF_FUNC_fib_lookup:
7185 return &bpf_xdp_fib_lookup_proto;
7187 case BPF_FUNC_sk_lookup_udp:
7188 return &bpf_xdp_sk_lookup_udp_proto;
7189 case BPF_FUNC_sk_lookup_tcp:
7190 return &bpf_xdp_sk_lookup_tcp_proto;
7191 case BPF_FUNC_sk_release:
7192 return &bpf_sk_release_proto;
7193 case BPF_FUNC_skc_lookup_tcp:
7194 return &bpf_xdp_skc_lookup_tcp_proto;
7195 case BPF_FUNC_tcp_check_syncookie:
7196 return &bpf_tcp_check_syncookie_proto;
7197 case BPF_FUNC_tcp_gen_syncookie:
7198 return &bpf_tcp_gen_syncookie_proto;
7201 return bpf_sk_base_func_proto(func_id);
7205 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
7206 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
7208 static const struct bpf_func_proto *
7209 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7212 case BPF_FUNC_setsockopt:
7213 return &bpf_sock_ops_setsockopt_proto;
7214 case BPF_FUNC_getsockopt:
7215 return &bpf_sock_ops_getsockopt_proto;
7216 case BPF_FUNC_sock_ops_cb_flags_set:
7217 return &bpf_sock_ops_cb_flags_set_proto;
7218 case BPF_FUNC_sock_map_update:
7219 return &bpf_sock_map_update_proto;
7220 case BPF_FUNC_sock_hash_update:
7221 return &bpf_sock_hash_update_proto;
7222 case BPF_FUNC_get_socket_cookie:
7223 return &bpf_get_socket_cookie_sock_ops_proto;
7224 case BPF_FUNC_get_local_storage:
7225 return &bpf_get_local_storage_proto;
7226 case BPF_FUNC_perf_event_output:
7227 return &bpf_event_output_data_proto;
7228 case BPF_FUNC_sk_storage_get:
7229 return &bpf_sk_storage_get_proto;
7230 case BPF_FUNC_sk_storage_delete:
7231 return &bpf_sk_storage_delete_proto;
7233 case BPF_FUNC_load_hdr_opt:
7234 return &bpf_sock_ops_load_hdr_opt_proto;
7235 case BPF_FUNC_store_hdr_opt:
7236 return &bpf_sock_ops_store_hdr_opt_proto;
7237 case BPF_FUNC_reserve_hdr_opt:
7238 return &bpf_sock_ops_reserve_hdr_opt_proto;
7239 case BPF_FUNC_tcp_sock:
7240 return &bpf_tcp_sock_proto;
7241 #endif /* CONFIG_INET */
7243 return bpf_sk_base_func_proto(func_id);
7247 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
7248 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
7250 static const struct bpf_func_proto *
7251 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7254 case BPF_FUNC_msg_redirect_map:
7255 return &bpf_msg_redirect_map_proto;
7256 case BPF_FUNC_msg_redirect_hash:
7257 return &bpf_msg_redirect_hash_proto;
7258 case BPF_FUNC_msg_apply_bytes:
7259 return &bpf_msg_apply_bytes_proto;
7260 case BPF_FUNC_msg_cork_bytes:
7261 return &bpf_msg_cork_bytes_proto;
7262 case BPF_FUNC_msg_pull_data:
7263 return &bpf_msg_pull_data_proto;
7264 case BPF_FUNC_msg_push_data:
7265 return &bpf_msg_push_data_proto;
7266 case BPF_FUNC_msg_pop_data:
7267 return &bpf_msg_pop_data_proto;
7268 case BPF_FUNC_perf_event_output:
7269 return &bpf_event_output_data_proto;
7270 case BPF_FUNC_get_current_uid_gid:
7271 return &bpf_get_current_uid_gid_proto;
7272 case BPF_FUNC_get_current_pid_tgid:
7273 return &bpf_get_current_pid_tgid_proto;
7274 case BPF_FUNC_sk_storage_get:
7275 return &bpf_sk_storage_get_proto;
7276 case BPF_FUNC_sk_storage_delete:
7277 return &bpf_sk_storage_delete_proto;
7278 #ifdef CONFIG_CGROUPS
7279 case BPF_FUNC_get_current_cgroup_id:
7280 return &bpf_get_current_cgroup_id_proto;
7281 case BPF_FUNC_get_current_ancestor_cgroup_id:
7282 return &bpf_get_current_ancestor_cgroup_id_proto;
7284 #ifdef CONFIG_CGROUP_NET_CLASSID
7285 case BPF_FUNC_get_cgroup_classid:
7286 return &bpf_get_cgroup_classid_curr_proto;
7289 return bpf_sk_base_func_proto(func_id);
7293 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
7294 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
7296 static const struct bpf_func_proto *
7297 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7300 case BPF_FUNC_skb_store_bytes:
7301 return &bpf_skb_store_bytes_proto;
7302 case BPF_FUNC_skb_load_bytes:
7303 return &bpf_skb_load_bytes_proto;
7304 case BPF_FUNC_skb_pull_data:
7305 return &sk_skb_pull_data_proto;
7306 case BPF_FUNC_skb_change_tail:
7307 return &sk_skb_change_tail_proto;
7308 case BPF_FUNC_skb_change_head:
7309 return &sk_skb_change_head_proto;
7310 case BPF_FUNC_skb_adjust_room:
7311 return &sk_skb_adjust_room_proto;
7312 case BPF_FUNC_get_socket_cookie:
7313 return &bpf_get_socket_cookie_proto;
7314 case BPF_FUNC_get_socket_uid:
7315 return &bpf_get_socket_uid_proto;
7316 case BPF_FUNC_sk_redirect_map:
7317 return &bpf_sk_redirect_map_proto;
7318 case BPF_FUNC_sk_redirect_hash:
7319 return &bpf_sk_redirect_hash_proto;
7320 case BPF_FUNC_perf_event_output:
7321 return &bpf_skb_event_output_proto;
7323 case BPF_FUNC_sk_lookup_tcp:
7324 return &bpf_sk_lookup_tcp_proto;
7325 case BPF_FUNC_sk_lookup_udp:
7326 return &bpf_sk_lookup_udp_proto;
7327 case BPF_FUNC_sk_release:
7328 return &bpf_sk_release_proto;
7329 case BPF_FUNC_skc_lookup_tcp:
7330 return &bpf_skc_lookup_tcp_proto;
7333 return bpf_sk_base_func_proto(func_id);
7337 static const struct bpf_func_proto *
7338 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7341 case BPF_FUNC_skb_load_bytes:
7342 return &bpf_flow_dissector_load_bytes_proto;
7344 return bpf_sk_base_func_proto(func_id);
7348 static const struct bpf_func_proto *
7349 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7352 case BPF_FUNC_skb_load_bytes:
7353 return &bpf_skb_load_bytes_proto;
7354 case BPF_FUNC_skb_pull_data:
7355 return &bpf_skb_pull_data_proto;
7356 case BPF_FUNC_csum_diff:
7357 return &bpf_csum_diff_proto;
7358 case BPF_FUNC_get_cgroup_classid:
7359 return &bpf_get_cgroup_classid_proto;
7360 case BPF_FUNC_get_route_realm:
7361 return &bpf_get_route_realm_proto;
7362 case BPF_FUNC_get_hash_recalc:
7363 return &bpf_get_hash_recalc_proto;
7364 case BPF_FUNC_perf_event_output:
7365 return &bpf_skb_event_output_proto;
7366 case BPF_FUNC_get_smp_processor_id:
7367 return &bpf_get_smp_processor_id_proto;
7368 case BPF_FUNC_skb_under_cgroup:
7369 return &bpf_skb_under_cgroup_proto;
7371 return bpf_sk_base_func_proto(func_id);
7375 static const struct bpf_func_proto *
7376 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7379 case BPF_FUNC_lwt_push_encap:
7380 return &bpf_lwt_in_push_encap_proto;
7382 return lwt_out_func_proto(func_id, prog);
7386 static const struct bpf_func_proto *
7387 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7390 case BPF_FUNC_skb_get_tunnel_key:
7391 return &bpf_skb_get_tunnel_key_proto;
7392 case BPF_FUNC_skb_set_tunnel_key:
7393 return bpf_get_skb_set_tunnel_proto(func_id);
7394 case BPF_FUNC_skb_get_tunnel_opt:
7395 return &bpf_skb_get_tunnel_opt_proto;
7396 case BPF_FUNC_skb_set_tunnel_opt:
7397 return bpf_get_skb_set_tunnel_proto(func_id);
7398 case BPF_FUNC_redirect:
7399 return &bpf_redirect_proto;
7400 case BPF_FUNC_clone_redirect:
7401 return &bpf_clone_redirect_proto;
7402 case BPF_FUNC_skb_change_tail:
7403 return &bpf_skb_change_tail_proto;
7404 case BPF_FUNC_skb_change_head:
7405 return &bpf_skb_change_head_proto;
7406 case BPF_FUNC_skb_store_bytes:
7407 return &bpf_skb_store_bytes_proto;
7408 case BPF_FUNC_csum_update:
7409 return &bpf_csum_update_proto;
7410 case BPF_FUNC_csum_level:
7411 return &bpf_csum_level_proto;
7412 case BPF_FUNC_l3_csum_replace:
7413 return &bpf_l3_csum_replace_proto;
7414 case BPF_FUNC_l4_csum_replace:
7415 return &bpf_l4_csum_replace_proto;
7416 case BPF_FUNC_set_hash_invalid:
7417 return &bpf_set_hash_invalid_proto;
7418 case BPF_FUNC_lwt_push_encap:
7419 return &bpf_lwt_xmit_push_encap_proto;
7421 return lwt_out_func_proto(func_id, prog);
7425 static const struct bpf_func_proto *
7426 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7429 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7430 case BPF_FUNC_lwt_seg6_store_bytes:
7431 return &bpf_lwt_seg6_store_bytes_proto;
7432 case BPF_FUNC_lwt_seg6_action:
7433 return &bpf_lwt_seg6_action_proto;
7434 case BPF_FUNC_lwt_seg6_adjust_srh:
7435 return &bpf_lwt_seg6_adjust_srh_proto;
7438 return lwt_out_func_proto(func_id, prog);
7442 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
7443 const struct bpf_prog *prog,
7444 struct bpf_insn_access_aux *info)
7446 const int size_default = sizeof(__u32);
7448 if (off < 0 || off >= sizeof(struct __sk_buff))
7451 /* The verifier guarantees that size > 0. */
7452 if (off % size != 0)
7456 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7457 if (off + size > offsetofend(struct __sk_buff, cb[4]))
7460 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
7461 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
7462 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
7463 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
7464 case bpf_ctx_range(struct __sk_buff, data):
7465 case bpf_ctx_range(struct __sk_buff, data_meta):
7466 case bpf_ctx_range(struct __sk_buff, data_end):
7467 if (size != size_default)
7470 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
7472 case bpf_ctx_range(struct __sk_buff, tstamp):
7473 if (size != sizeof(__u64))
7476 case offsetof(struct __sk_buff, sk):
7477 if (type == BPF_WRITE || size != sizeof(__u64))
7479 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
7482 /* Only narrow read access allowed for now. */
7483 if (type == BPF_WRITE) {
7484 if (size != size_default)
7487 bpf_ctx_record_field_size(info, size_default);
7488 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
7496 static bool sk_filter_is_valid_access(int off, int size,
7497 enum bpf_access_type type,
7498 const struct bpf_prog *prog,
7499 struct bpf_insn_access_aux *info)
7502 case bpf_ctx_range(struct __sk_buff, tc_classid):
7503 case bpf_ctx_range(struct __sk_buff, data):
7504 case bpf_ctx_range(struct __sk_buff, data_meta):
7505 case bpf_ctx_range(struct __sk_buff, data_end):
7506 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7507 case bpf_ctx_range(struct __sk_buff, tstamp):
7508 case bpf_ctx_range(struct __sk_buff, wire_len):
7512 if (type == BPF_WRITE) {
7514 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7521 return bpf_skb_is_valid_access(off, size, type, prog, info);
7524 static bool cg_skb_is_valid_access(int off, int size,
7525 enum bpf_access_type type,
7526 const struct bpf_prog *prog,
7527 struct bpf_insn_access_aux *info)
7530 case bpf_ctx_range(struct __sk_buff, tc_classid):
7531 case bpf_ctx_range(struct __sk_buff, data_meta):
7532 case bpf_ctx_range(struct __sk_buff, wire_len):
7534 case bpf_ctx_range(struct __sk_buff, data):
7535 case bpf_ctx_range(struct __sk_buff, data_end):
7541 if (type == BPF_WRITE) {
7543 case bpf_ctx_range(struct __sk_buff, mark):
7544 case bpf_ctx_range(struct __sk_buff, priority):
7545 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7547 case bpf_ctx_range(struct __sk_buff, tstamp):
7557 case bpf_ctx_range(struct __sk_buff, data):
7558 info->reg_type = PTR_TO_PACKET;
7560 case bpf_ctx_range(struct __sk_buff, data_end):
7561 info->reg_type = PTR_TO_PACKET_END;
7565 return bpf_skb_is_valid_access(off, size, type, prog, info);
7568 static bool lwt_is_valid_access(int off, int size,
7569 enum bpf_access_type type,
7570 const struct bpf_prog *prog,
7571 struct bpf_insn_access_aux *info)
7574 case bpf_ctx_range(struct __sk_buff, tc_classid):
7575 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7576 case bpf_ctx_range(struct __sk_buff, data_meta):
7577 case bpf_ctx_range(struct __sk_buff, tstamp):
7578 case bpf_ctx_range(struct __sk_buff, wire_len):
7582 if (type == BPF_WRITE) {
7584 case bpf_ctx_range(struct __sk_buff, mark):
7585 case bpf_ctx_range(struct __sk_buff, priority):
7586 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7594 case bpf_ctx_range(struct __sk_buff, data):
7595 info->reg_type = PTR_TO_PACKET;
7597 case bpf_ctx_range(struct __sk_buff, data_end):
7598 info->reg_type = PTR_TO_PACKET_END;
7602 return bpf_skb_is_valid_access(off, size, type, prog, info);
7605 /* Attach type specific accesses */
7606 static bool __sock_filter_check_attach_type(int off,
7607 enum bpf_access_type access_type,
7608 enum bpf_attach_type attach_type)
7611 case offsetof(struct bpf_sock, bound_dev_if):
7612 case offsetof(struct bpf_sock, mark):
7613 case offsetof(struct bpf_sock, priority):
7614 switch (attach_type) {
7615 case BPF_CGROUP_INET_SOCK_CREATE:
7616 case BPF_CGROUP_INET_SOCK_RELEASE:
7621 case bpf_ctx_range(struct bpf_sock, src_ip4):
7622 switch (attach_type) {
7623 case BPF_CGROUP_INET4_POST_BIND:
7628 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7629 switch (attach_type) {
7630 case BPF_CGROUP_INET6_POST_BIND:
7635 case bpf_ctx_range(struct bpf_sock, src_port):
7636 switch (attach_type) {
7637 case BPF_CGROUP_INET4_POST_BIND:
7638 case BPF_CGROUP_INET6_POST_BIND:
7645 return access_type == BPF_READ;
7650 bool bpf_sock_common_is_valid_access(int off, int size,
7651 enum bpf_access_type type,
7652 struct bpf_insn_access_aux *info)
7655 case bpf_ctx_range_till(struct bpf_sock, type, priority):
7658 return bpf_sock_is_valid_access(off, size, type, info);
7662 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7663 struct bpf_insn_access_aux *info)
7665 const int size_default = sizeof(__u32);
7667 if (off < 0 || off >= sizeof(struct bpf_sock))
7669 if (off % size != 0)
7673 case offsetof(struct bpf_sock, state):
7674 case offsetof(struct bpf_sock, family):
7675 case offsetof(struct bpf_sock, type):
7676 case offsetof(struct bpf_sock, protocol):
7677 case offsetof(struct bpf_sock, dst_port):
7678 case offsetof(struct bpf_sock, src_port):
7679 case offsetof(struct bpf_sock, rx_queue_mapping):
7680 case bpf_ctx_range(struct bpf_sock, src_ip4):
7681 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7682 case bpf_ctx_range(struct bpf_sock, dst_ip4):
7683 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
7684 bpf_ctx_record_field_size(info, size_default);
7685 return bpf_ctx_narrow_access_ok(off, size, size_default);
7688 return size == size_default;
7691 static bool sock_filter_is_valid_access(int off, int size,
7692 enum bpf_access_type type,
7693 const struct bpf_prog *prog,
7694 struct bpf_insn_access_aux *info)
7696 if (!bpf_sock_is_valid_access(off, size, type, info))
7698 return __sock_filter_check_attach_type(off, type,
7699 prog->expected_attach_type);
7702 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
7703 const struct bpf_prog *prog)
7705 /* Neither direct read nor direct write requires any preliminary
7711 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
7712 const struct bpf_prog *prog, int drop_verdict)
7714 struct bpf_insn *insn = insn_buf;
7719 /* if (!skb->cloned)
7722 * (Fast-path, otherwise approximation that we might be
7723 * a clone, do the rest in helper.)
7725 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
7726 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
7727 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
7729 /* ret = bpf_skb_pull_data(skb, 0); */
7730 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
7731 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
7732 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
7733 BPF_FUNC_skb_pull_data);
7736 * return TC_ACT_SHOT;
7738 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
7739 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
7740 *insn++ = BPF_EXIT_INSN();
7743 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
7745 *insn++ = prog->insnsi[0];
7747 return insn - insn_buf;
7750 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
7751 struct bpf_insn *insn_buf)
7753 bool indirect = BPF_MODE(orig->code) == BPF_IND;
7754 struct bpf_insn *insn = insn_buf;
7757 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
7759 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
7761 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
7763 /* We're guaranteed here that CTX is in R6. */
7764 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
7766 switch (BPF_SIZE(orig->code)) {
7768 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
7771 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
7774 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
7778 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
7779 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
7780 *insn++ = BPF_EXIT_INSN();
7782 return insn - insn_buf;
7785 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
7786 const struct bpf_prog *prog)
7788 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
7791 static bool tc_cls_act_is_valid_access(int off, int size,
7792 enum bpf_access_type type,
7793 const struct bpf_prog *prog,
7794 struct bpf_insn_access_aux *info)
7796 if (type == BPF_WRITE) {
7798 case bpf_ctx_range(struct __sk_buff, mark):
7799 case bpf_ctx_range(struct __sk_buff, tc_index):
7800 case bpf_ctx_range(struct __sk_buff, priority):
7801 case bpf_ctx_range(struct __sk_buff, tc_classid):
7802 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7803 case bpf_ctx_range(struct __sk_buff, tstamp):
7804 case bpf_ctx_range(struct __sk_buff, queue_mapping):
7812 case bpf_ctx_range(struct __sk_buff, data):
7813 info->reg_type = PTR_TO_PACKET;
7815 case bpf_ctx_range(struct __sk_buff, data_meta):
7816 info->reg_type = PTR_TO_PACKET_META;
7818 case bpf_ctx_range(struct __sk_buff, data_end):
7819 info->reg_type = PTR_TO_PACKET_END;
7821 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7825 return bpf_skb_is_valid_access(off, size, type, prog, info);
7828 static bool __is_valid_xdp_access(int off, int size)
7830 if (off < 0 || off >= sizeof(struct xdp_md))
7832 if (off % size != 0)
7834 if (size != sizeof(__u32))
7840 static bool xdp_is_valid_access(int off, int size,
7841 enum bpf_access_type type,
7842 const struct bpf_prog *prog,
7843 struct bpf_insn_access_aux *info)
7845 if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
7847 case offsetof(struct xdp_md, egress_ifindex):
7852 if (type == BPF_WRITE) {
7853 if (bpf_prog_is_dev_bound(prog->aux)) {
7855 case offsetof(struct xdp_md, rx_queue_index):
7856 return __is_valid_xdp_access(off, size);
7863 case offsetof(struct xdp_md, data):
7864 info->reg_type = PTR_TO_PACKET;
7866 case offsetof(struct xdp_md, data_meta):
7867 info->reg_type = PTR_TO_PACKET_META;
7869 case offsetof(struct xdp_md, data_end):
7870 info->reg_type = PTR_TO_PACKET_END;
7874 return __is_valid_xdp_access(off, size);
7877 void bpf_warn_invalid_xdp_action(u32 act)
7879 const u32 act_max = XDP_REDIRECT;
7881 WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
7882 act > act_max ? "Illegal" : "Driver unsupported",
7885 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
7887 static bool sock_addr_is_valid_access(int off, int size,
7888 enum bpf_access_type type,
7889 const struct bpf_prog *prog,
7890 struct bpf_insn_access_aux *info)
7892 const int size_default = sizeof(__u32);
7894 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
7896 if (off % size != 0)
7899 /* Disallow access to IPv6 fields from IPv4 contex and vise
7903 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
7904 switch (prog->expected_attach_type) {
7905 case BPF_CGROUP_INET4_BIND:
7906 case BPF_CGROUP_INET4_CONNECT:
7907 case BPF_CGROUP_INET4_GETPEERNAME:
7908 case BPF_CGROUP_INET4_GETSOCKNAME:
7909 case BPF_CGROUP_UDP4_SENDMSG:
7910 case BPF_CGROUP_UDP4_RECVMSG:
7916 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
7917 switch (prog->expected_attach_type) {
7918 case BPF_CGROUP_INET6_BIND:
7919 case BPF_CGROUP_INET6_CONNECT:
7920 case BPF_CGROUP_INET6_GETPEERNAME:
7921 case BPF_CGROUP_INET6_GETSOCKNAME:
7922 case BPF_CGROUP_UDP6_SENDMSG:
7923 case BPF_CGROUP_UDP6_RECVMSG:
7929 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
7930 switch (prog->expected_attach_type) {
7931 case BPF_CGROUP_UDP4_SENDMSG:
7937 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
7939 switch (prog->expected_attach_type) {
7940 case BPF_CGROUP_UDP6_SENDMSG:
7949 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
7950 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
7951 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
7952 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
7954 case bpf_ctx_range(struct bpf_sock_addr, user_port):
7955 if (type == BPF_READ) {
7956 bpf_ctx_record_field_size(info, size_default);
7958 if (bpf_ctx_wide_access_ok(off, size,
7959 struct bpf_sock_addr,
7963 if (bpf_ctx_wide_access_ok(off, size,
7964 struct bpf_sock_addr,
7968 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
7971 if (bpf_ctx_wide_access_ok(off, size,
7972 struct bpf_sock_addr,
7976 if (bpf_ctx_wide_access_ok(off, size,
7977 struct bpf_sock_addr,
7981 if (size != size_default)
7985 case offsetof(struct bpf_sock_addr, sk):
7986 if (type != BPF_READ)
7988 if (size != sizeof(__u64))
7990 info->reg_type = PTR_TO_SOCKET;
7993 if (type == BPF_READ) {
7994 if (size != size_default)
8004 static bool sock_ops_is_valid_access(int off, int size,
8005 enum bpf_access_type type,
8006 const struct bpf_prog *prog,
8007 struct bpf_insn_access_aux *info)
8009 const int size_default = sizeof(__u32);
8011 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
8014 /* The verifier guarantees that size > 0. */
8015 if (off % size != 0)
8018 if (type == BPF_WRITE) {
8020 case offsetof(struct bpf_sock_ops, reply):
8021 case offsetof(struct bpf_sock_ops, sk_txhash):
8022 if (size != size_default)
8030 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
8032 if (size != sizeof(__u64))
8035 case offsetof(struct bpf_sock_ops, sk):
8036 if (size != sizeof(__u64))
8038 info->reg_type = PTR_TO_SOCKET_OR_NULL;
8040 case offsetof(struct bpf_sock_ops, skb_data):
8041 if (size != sizeof(__u64))
8043 info->reg_type = PTR_TO_PACKET;
8045 case offsetof(struct bpf_sock_ops, skb_data_end):
8046 if (size != sizeof(__u64))
8048 info->reg_type = PTR_TO_PACKET_END;
8050 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
8051 bpf_ctx_record_field_size(info, size_default);
8052 return bpf_ctx_narrow_access_ok(off, size,
8055 if (size != size_default)
8064 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
8065 const struct bpf_prog *prog)
8067 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
8070 static bool sk_skb_is_valid_access(int off, int size,
8071 enum bpf_access_type type,
8072 const struct bpf_prog *prog,
8073 struct bpf_insn_access_aux *info)
8076 case bpf_ctx_range(struct __sk_buff, tc_classid):
8077 case bpf_ctx_range(struct __sk_buff, data_meta):
8078 case bpf_ctx_range(struct __sk_buff, tstamp):
8079 case bpf_ctx_range(struct __sk_buff, wire_len):
8083 if (type == BPF_WRITE) {
8085 case bpf_ctx_range(struct __sk_buff, tc_index):
8086 case bpf_ctx_range(struct __sk_buff, priority):
8094 case bpf_ctx_range(struct __sk_buff, mark):
8096 case bpf_ctx_range(struct __sk_buff, data):
8097 info->reg_type = PTR_TO_PACKET;
8099 case bpf_ctx_range(struct __sk_buff, data_end):
8100 info->reg_type = PTR_TO_PACKET_END;
8104 return bpf_skb_is_valid_access(off, size, type, prog, info);
8107 static bool sk_msg_is_valid_access(int off, int size,
8108 enum bpf_access_type type,
8109 const struct bpf_prog *prog,
8110 struct bpf_insn_access_aux *info)
8112 if (type == BPF_WRITE)
8115 if (off % size != 0)
8119 case offsetof(struct sk_msg_md, data):
8120 info->reg_type = PTR_TO_PACKET;
8121 if (size != sizeof(__u64))
8124 case offsetof(struct sk_msg_md, data_end):
8125 info->reg_type = PTR_TO_PACKET_END;
8126 if (size != sizeof(__u64))
8129 case offsetof(struct sk_msg_md, sk):
8130 if (size != sizeof(__u64))
8132 info->reg_type = PTR_TO_SOCKET;
8134 case bpf_ctx_range(struct sk_msg_md, family):
8135 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
8136 case bpf_ctx_range(struct sk_msg_md, local_ip4):
8137 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
8138 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
8139 case bpf_ctx_range(struct sk_msg_md, remote_port):
8140 case bpf_ctx_range(struct sk_msg_md, local_port):
8141 case bpf_ctx_range(struct sk_msg_md, size):
8142 if (size != sizeof(__u32))
8151 static bool flow_dissector_is_valid_access(int off, int size,
8152 enum bpf_access_type type,
8153 const struct bpf_prog *prog,
8154 struct bpf_insn_access_aux *info)
8156 const int size_default = sizeof(__u32);
8158 if (off < 0 || off >= sizeof(struct __sk_buff))
8161 if (type == BPF_WRITE)
8165 case bpf_ctx_range(struct __sk_buff, data):
8166 if (size != size_default)
8168 info->reg_type = PTR_TO_PACKET;
8170 case bpf_ctx_range(struct __sk_buff, data_end):
8171 if (size != size_default)
8173 info->reg_type = PTR_TO_PACKET_END;
8175 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8176 if (size != sizeof(__u64))
8178 info->reg_type = PTR_TO_FLOW_KEYS;
8185 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
8186 const struct bpf_insn *si,
8187 struct bpf_insn *insn_buf,
8188 struct bpf_prog *prog,
8192 struct bpf_insn *insn = insn_buf;
8195 case offsetof(struct __sk_buff, data):
8196 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
8197 si->dst_reg, si->src_reg,
8198 offsetof(struct bpf_flow_dissector, data));
8201 case offsetof(struct __sk_buff, data_end):
8202 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
8203 si->dst_reg, si->src_reg,
8204 offsetof(struct bpf_flow_dissector, data_end));
8207 case offsetof(struct __sk_buff, flow_keys):
8208 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
8209 si->dst_reg, si->src_reg,
8210 offsetof(struct bpf_flow_dissector, flow_keys));
8214 return insn - insn_buf;
8217 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si,
8218 struct bpf_insn *insn)
8220 /* si->dst_reg = skb_shinfo(SKB); */
8221 #ifdef NET_SKBUFF_DATA_USES_OFFSET
8222 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8223 BPF_REG_AX, si->src_reg,
8224 offsetof(struct sk_buff, end));
8225 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
8226 si->dst_reg, si->src_reg,
8227 offsetof(struct sk_buff, head));
8228 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
8230 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8231 si->dst_reg, si->src_reg,
8232 offsetof(struct sk_buff, end));
8238 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
8239 const struct bpf_insn *si,
8240 struct bpf_insn *insn_buf,
8241 struct bpf_prog *prog, u32 *target_size)
8243 struct bpf_insn *insn = insn_buf;
8247 case offsetof(struct __sk_buff, len):
8248 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8249 bpf_target_off(struct sk_buff, len, 4,
8253 case offsetof(struct __sk_buff, protocol):
8254 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8255 bpf_target_off(struct sk_buff, protocol, 2,
8259 case offsetof(struct __sk_buff, vlan_proto):
8260 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8261 bpf_target_off(struct sk_buff, vlan_proto, 2,
8265 case offsetof(struct __sk_buff, priority):
8266 if (type == BPF_WRITE)
8267 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8268 bpf_target_off(struct sk_buff, priority, 4,
8271 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8272 bpf_target_off(struct sk_buff, priority, 4,
8276 case offsetof(struct __sk_buff, ingress_ifindex):
8277 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8278 bpf_target_off(struct sk_buff, skb_iif, 4,
8282 case offsetof(struct __sk_buff, ifindex):
8283 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8284 si->dst_reg, si->src_reg,
8285 offsetof(struct sk_buff, dev));
8286 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
8287 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8288 bpf_target_off(struct net_device, ifindex, 4,
8292 case offsetof(struct __sk_buff, hash):
8293 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8294 bpf_target_off(struct sk_buff, hash, 4,
8298 case offsetof(struct __sk_buff, mark):
8299 if (type == BPF_WRITE)
8300 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8301 bpf_target_off(struct sk_buff, mark, 4,
8304 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8305 bpf_target_off(struct sk_buff, mark, 4,
8309 case offsetof(struct __sk_buff, pkt_type):
8311 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8313 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
8314 #ifdef __BIG_ENDIAN_BITFIELD
8315 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
8319 case offsetof(struct __sk_buff, queue_mapping):
8320 if (type == BPF_WRITE) {
8321 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
8322 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8323 bpf_target_off(struct sk_buff,
8327 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8328 bpf_target_off(struct sk_buff,
8334 case offsetof(struct __sk_buff, vlan_present):
8336 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8337 PKT_VLAN_PRESENT_OFFSET());
8338 if (PKT_VLAN_PRESENT_BIT)
8339 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
8340 if (PKT_VLAN_PRESENT_BIT < 7)
8341 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
8344 case offsetof(struct __sk_buff, vlan_tci):
8345 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8346 bpf_target_off(struct sk_buff, vlan_tci, 2,
8350 case offsetof(struct __sk_buff, cb[0]) ...
8351 offsetofend(struct __sk_buff, cb[4]) - 1:
8352 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
8353 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
8354 offsetof(struct qdisc_skb_cb, data)) %
8357 prog->cb_access = 1;
8359 off -= offsetof(struct __sk_buff, cb[0]);
8360 off += offsetof(struct sk_buff, cb);
8361 off += offsetof(struct qdisc_skb_cb, data);
8362 if (type == BPF_WRITE)
8363 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
8366 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
8370 case offsetof(struct __sk_buff, tc_classid):
8371 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
8374 off -= offsetof(struct __sk_buff, tc_classid);
8375 off += offsetof(struct sk_buff, cb);
8376 off += offsetof(struct qdisc_skb_cb, tc_classid);
8378 if (type == BPF_WRITE)
8379 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
8382 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
8386 case offsetof(struct __sk_buff, data):
8387 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
8388 si->dst_reg, si->src_reg,
8389 offsetof(struct sk_buff, data));
8392 case offsetof(struct __sk_buff, data_meta):
8394 off -= offsetof(struct __sk_buff, data_meta);
8395 off += offsetof(struct sk_buff, cb);
8396 off += offsetof(struct bpf_skb_data_end, data_meta);
8397 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8401 case offsetof(struct __sk_buff, data_end):
8403 off -= offsetof(struct __sk_buff, data_end);
8404 off += offsetof(struct sk_buff, cb);
8405 off += offsetof(struct bpf_skb_data_end, data_end);
8406 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8410 case offsetof(struct __sk_buff, tc_index):
8411 #ifdef CONFIG_NET_SCHED
8412 if (type == BPF_WRITE)
8413 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8414 bpf_target_off(struct sk_buff, tc_index, 2,
8417 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8418 bpf_target_off(struct sk_buff, tc_index, 2,
8422 if (type == BPF_WRITE)
8423 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
8425 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8429 case offsetof(struct __sk_buff, napi_id):
8430 #if defined(CONFIG_NET_RX_BUSY_POLL)
8431 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8432 bpf_target_off(struct sk_buff, napi_id, 4,
8434 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
8435 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8438 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8441 case offsetof(struct __sk_buff, family):
8442 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
8444 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8445 si->dst_reg, si->src_reg,
8446 offsetof(struct sk_buff, sk));
8447 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8448 bpf_target_off(struct sock_common,
8452 case offsetof(struct __sk_buff, remote_ip4):
8453 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
8455 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8456 si->dst_reg, si->src_reg,
8457 offsetof(struct sk_buff, sk));
8458 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8459 bpf_target_off(struct sock_common,
8463 case offsetof(struct __sk_buff, local_ip4):
8464 BUILD_BUG_ON(sizeof_field(struct sock_common,
8465 skc_rcv_saddr) != 4);
8467 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8468 si->dst_reg, si->src_reg,
8469 offsetof(struct sk_buff, sk));
8470 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8471 bpf_target_off(struct sock_common,
8475 case offsetof(struct __sk_buff, remote_ip6[0]) ...
8476 offsetof(struct __sk_buff, remote_ip6[3]):
8477 #if IS_ENABLED(CONFIG_IPV6)
8478 BUILD_BUG_ON(sizeof_field(struct sock_common,
8479 skc_v6_daddr.s6_addr32[0]) != 4);
8482 off -= offsetof(struct __sk_buff, remote_ip6[0]);
8484 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8485 si->dst_reg, si->src_reg,
8486 offsetof(struct sk_buff, sk));
8487 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8488 offsetof(struct sock_common,
8489 skc_v6_daddr.s6_addr32[0]) +
8492 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8495 case offsetof(struct __sk_buff, local_ip6[0]) ...
8496 offsetof(struct __sk_buff, local_ip6[3]):
8497 #if IS_ENABLED(CONFIG_IPV6)
8498 BUILD_BUG_ON(sizeof_field(struct sock_common,
8499 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8502 off -= offsetof(struct __sk_buff, local_ip6[0]);
8504 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8505 si->dst_reg, si->src_reg,
8506 offsetof(struct sk_buff, sk));
8507 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8508 offsetof(struct sock_common,
8509 skc_v6_rcv_saddr.s6_addr32[0]) +
8512 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8516 case offsetof(struct __sk_buff, remote_port):
8517 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
8519 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8520 si->dst_reg, si->src_reg,
8521 offsetof(struct sk_buff, sk));
8522 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8523 bpf_target_off(struct sock_common,
8526 #ifndef __BIG_ENDIAN_BITFIELD
8527 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8531 case offsetof(struct __sk_buff, local_port):
8532 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
8534 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8535 si->dst_reg, si->src_reg,
8536 offsetof(struct sk_buff, sk));
8537 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8538 bpf_target_off(struct sock_common,
8539 skc_num, 2, target_size));
8542 case offsetof(struct __sk_buff, tstamp):
8543 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
8545 if (type == BPF_WRITE)
8546 *insn++ = BPF_STX_MEM(BPF_DW,
8547 si->dst_reg, si->src_reg,
8548 bpf_target_off(struct sk_buff,
8552 *insn++ = BPF_LDX_MEM(BPF_DW,
8553 si->dst_reg, si->src_reg,
8554 bpf_target_off(struct sk_buff,
8559 case offsetof(struct __sk_buff, gso_segs):
8560 insn = bpf_convert_shinfo_access(si, insn);
8561 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
8562 si->dst_reg, si->dst_reg,
8563 bpf_target_off(struct skb_shared_info,
8567 case offsetof(struct __sk_buff, gso_size):
8568 insn = bpf_convert_shinfo_access(si, insn);
8569 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
8570 si->dst_reg, si->dst_reg,
8571 bpf_target_off(struct skb_shared_info,
8575 case offsetof(struct __sk_buff, wire_len):
8576 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
8579 off -= offsetof(struct __sk_buff, wire_len);
8580 off += offsetof(struct sk_buff, cb);
8581 off += offsetof(struct qdisc_skb_cb, pkt_len);
8583 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
8586 case offsetof(struct __sk_buff, sk):
8587 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8588 si->dst_reg, si->src_reg,
8589 offsetof(struct sk_buff, sk));
8593 return insn - insn_buf;
8596 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
8597 const struct bpf_insn *si,
8598 struct bpf_insn *insn_buf,
8599 struct bpf_prog *prog, u32 *target_size)
8601 struct bpf_insn *insn = insn_buf;
8605 case offsetof(struct bpf_sock, bound_dev_if):
8606 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
8608 if (type == BPF_WRITE)
8609 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8610 offsetof(struct sock, sk_bound_dev_if));
8612 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8613 offsetof(struct sock, sk_bound_dev_if));
8616 case offsetof(struct bpf_sock, mark):
8617 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
8619 if (type == BPF_WRITE)
8620 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8621 offsetof(struct sock, sk_mark));
8623 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8624 offsetof(struct sock, sk_mark));
8627 case offsetof(struct bpf_sock, priority):
8628 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
8630 if (type == BPF_WRITE)
8631 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8632 offsetof(struct sock, sk_priority));
8634 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8635 offsetof(struct sock, sk_priority));
8638 case offsetof(struct bpf_sock, family):
8639 *insn++ = BPF_LDX_MEM(
8640 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
8641 si->dst_reg, si->src_reg,
8642 bpf_target_off(struct sock_common,
8644 sizeof_field(struct sock_common,
8649 case offsetof(struct bpf_sock, type):
8650 *insn++ = BPF_LDX_MEM(
8651 BPF_FIELD_SIZEOF(struct sock, sk_type),
8652 si->dst_reg, si->src_reg,
8653 bpf_target_off(struct sock, sk_type,
8654 sizeof_field(struct sock, sk_type),
8658 case offsetof(struct bpf_sock, protocol):
8659 *insn++ = BPF_LDX_MEM(
8660 BPF_FIELD_SIZEOF(struct sock, sk_protocol),
8661 si->dst_reg, si->src_reg,
8662 bpf_target_off(struct sock, sk_protocol,
8663 sizeof_field(struct sock, sk_protocol),
8667 case offsetof(struct bpf_sock, src_ip4):
8668 *insn++ = BPF_LDX_MEM(
8669 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8670 bpf_target_off(struct sock_common, skc_rcv_saddr,
8671 sizeof_field(struct sock_common,
8676 case offsetof(struct bpf_sock, dst_ip4):
8677 *insn++ = BPF_LDX_MEM(
8678 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8679 bpf_target_off(struct sock_common, skc_daddr,
8680 sizeof_field(struct sock_common,
8685 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8686 #if IS_ENABLED(CONFIG_IPV6)
8688 off -= offsetof(struct bpf_sock, src_ip6[0]);
8689 *insn++ = BPF_LDX_MEM(
8690 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8693 skc_v6_rcv_saddr.s6_addr32[0],
8694 sizeof_field(struct sock_common,
8695 skc_v6_rcv_saddr.s6_addr32[0]),
8696 target_size) + off);
8699 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8703 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8704 #if IS_ENABLED(CONFIG_IPV6)
8706 off -= offsetof(struct bpf_sock, dst_ip6[0]);
8707 *insn++ = BPF_LDX_MEM(
8708 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8709 bpf_target_off(struct sock_common,
8710 skc_v6_daddr.s6_addr32[0],
8711 sizeof_field(struct sock_common,
8712 skc_v6_daddr.s6_addr32[0]),
8713 target_size) + off);
8715 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8720 case offsetof(struct bpf_sock, src_port):
8721 *insn++ = BPF_LDX_MEM(
8722 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
8723 si->dst_reg, si->src_reg,
8724 bpf_target_off(struct sock_common, skc_num,
8725 sizeof_field(struct sock_common,
8730 case offsetof(struct bpf_sock, dst_port):
8731 *insn++ = BPF_LDX_MEM(
8732 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
8733 si->dst_reg, si->src_reg,
8734 bpf_target_off(struct sock_common, skc_dport,
8735 sizeof_field(struct sock_common,
8740 case offsetof(struct bpf_sock, state):
8741 *insn++ = BPF_LDX_MEM(
8742 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
8743 si->dst_reg, si->src_reg,
8744 bpf_target_off(struct sock_common, skc_state,
8745 sizeof_field(struct sock_common,
8749 case offsetof(struct bpf_sock, rx_queue_mapping):
8751 *insn++ = BPF_LDX_MEM(
8752 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
8753 si->dst_reg, si->src_reg,
8754 bpf_target_off(struct sock, sk_rx_queue_mapping,
8755 sizeof_field(struct sock,
8756 sk_rx_queue_mapping),
8758 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
8760 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
8762 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
8768 return insn - insn_buf;
8771 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
8772 const struct bpf_insn *si,
8773 struct bpf_insn *insn_buf,
8774 struct bpf_prog *prog, u32 *target_size)
8776 struct bpf_insn *insn = insn_buf;
8779 case offsetof(struct __sk_buff, ifindex):
8780 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8781 si->dst_reg, si->src_reg,
8782 offsetof(struct sk_buff, dev));
8783 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8784 bpf_target_off(struct net_device, ifindex, 4,
8788 return bpf_convert_ctx_access(type, si, insn_buf, prog,
8792 return insn - insn_buf;
8795 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
8796 const struct bpf_insn *si,
8797 struct bpf_insn *insn_buf,
8798 struct bpf_prog *prog, u32 *target_size)
8800 struct bpf_insn *insn = insn_buf;
8803 case offsetof(struct xdp_md, data):
8804 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
8805 si->dst_reg, si->src_reg,
8806 offsetof(struct xdp_buff, data));
8808 case offsetof(struct xdp_md, data_meta):
8809 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
8810 si->dst_reg, si->src_reg,
8811 offsetof(struct xdp_buff, data_meta));
8813 case offsetof(struct xdp_md, data_end):
8814 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
8815 si->dst_reg, si->src_reg,
8816 offsetof(struct xdp_buff, data_end));
8818 case offsetof(struct xdp_md, ingress_ifindex):
8819 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
8820 si->dst_reg, si->src_reg,
8821 offsetof(struct xdp_buff, rxq));
8822 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
8823 si->dst_reg, si->dst_reg,
8824 offsetof(struct xdp_rxq_info, dev));
8825 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8826 offsetof(struct net_device, ifindex));
8828 case offsetof(struct xdp_md, rx_queue_index):
8829 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
8830 si->dst_reg, si->src_reg,
8831 offsetof(struct xdp_buff, rxq));
8832 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8833 offsetof(struct xdp_rxq_info,
8836 case offsetof(struct xdp_md, egress_ifindex):
8837 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
8838 si->dst_reg, si->src_reg,
8839 offsetof(struct xdp_buff, txq));
8840 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
8841 si->dst_reg, si->dst_reg,
8842 offsetof(struct xdp_txq_info, dev));
8843 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8844 offsetof(struct net_device, ifindex));
8848 return insn - insn_buf;
8851 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
8852 * context Structure, F is Field in context structure that contains a pointer
8853 * to Nested Structure of type NS that has the field NF.
8855 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
8856 * sure that SIZE is not greater than actual size of S.F.NF.
8858 * If offset OFF is provided, the load happens from that offset relative to
8861 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
8863 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
8864 si->src_reg, offsetof(S, F)); \
8865 *insn++ = BPF_LDX_MEM( \
8866 SIZE, si->dst_reg, si->dst_reg, \
8867 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
8872 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
8873 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
8874 BPF_FIELD_SIZEOF(NS, NF), 0)
8876 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
8877 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
8879 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
8880 * "register" since two registers available in convert_ctx_access are not
8881 * enough: we can't override neither SRC, since it contains value to store, nor
8882 * DST since it contains pointer to context that may be used by later
8883 * instructions. But we need a temporary place to save pointer to nested
8884 * structure whose field we want to store to.
8886 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
8888 int tmp_reg = BPF_REG_9; \
8889 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
8891 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
8893 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
8895 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
8896 si->dst_reg, offsetof(S, F)); \
8897 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \
8898 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
8901 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
8905 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
8908 if (type == BPF_WRITE) { \
8909 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
8912 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
8913 S, NS, F, NF, SIZE, OFF); \
8917 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
8918 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
8919 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
8921 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
8922 const struct bpf_insn *si,
8923 struct bpf_insn *insn_buf,
8924 struct bpf_prog *prog, u32 *target_size)
8926 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
8927 struct bpf_insn *insn = insn_buf;
8930 case offsetof(struct bpf_sock_addr, user_family):
8931 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
8932 struct sockaddr, uaddr, sa_family);
8935 case offsetof(struct bpf_sock_addr, user_ip4):
8936 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
8937 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
8938 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
8941 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8943 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
8944 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
8945 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
8946 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
8950 case offsetof(struct bpf_sock_addr, user_port):
8951 /* To get port we need to know sa_family first and then treat
8952 * sockaddr as either sockaddr_in or sockaddr_in6.
8953 * Though we can simplify since port field has same offset and
8954 * size in both structures.
8955 * Here we check this invariant and use just one of the
8956 * structures if it's true.
8958 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
8959 offsetof(struct sockaddr_in6, sin6_port));
8960 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
8961 sizeof_field(struct sockaddr_in6, sin6_port));
8962 /* Account for sin6_port being smaller than user_port. */
8963 port_size = min(port_size, BPF_LDST_BYTES(si));
8964 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
8965 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
8966 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
8969 case offsetof(struct bpf_sock_addr, family):
8970 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
8971 struct sock, sk, sk_family);
8974 case offsetof(struct bpf_sock_addr, type):
8975 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
8976 struct sock, sk, sk_type);
8979 case offsetof(struct bpf_sock_addr, protocol):
8980 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
8981 struct sock, sk, sk_protocol);
8984 case offsetof(struct bpf_sock_addr, msg_src_ip4):
8985 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
8986 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
8987 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
8988 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
8991 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8994 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
8995 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
8996 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
8997 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
8998 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
9000 case offsetof(struct bpf_sock_addr, sk):
9001 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
9002 si->dst_reg, si->src_reg,
9003 offsetof(struct bpf_sock_addr_kern, sk));
9007 return insn - insn_buf;
9010 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
9011 const struct bpf_insn *si,
9012 struct bpf_insn *insn_buf,
9013 struct bpf_prog *prog,
9016 struct bpf_insn *insn = insn_buf;
9019 /* Helper macro for adding read access to tcp_sock or sock fields. */
9020 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
9022 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \
9023 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
9024 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
9025 if (si->dst_reg == reg || si->src_reg == reg) \
9027 if (si->dst_reg == reg || si->src_reg == reg) \
9029 if (si->dst_reg == si->src_reg) { \
9030 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
9031 offsetof(struct bpf_sock_ops_kern, \
9033 fullsock_reg = reg; \
9036 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9037 struct bpf_sock_ops_kern, \
9039 fullsock_reg, si->src_reg, \
9040 offsetof(struct bpf_sock_ops_kern, \
9042 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
9043 if (si->dst_reg == si->src_reg) \
9044 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9045 offsetof(struct bpf_sock_ops_kern, \
9047 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9048 struct bpf_sock_ops_kern, sk),\
9049 si->dst_reg, si->src_reg, \
9050 offsetof(struct bpf_sock_ops_kern, sk));\
9051 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
9053 si->dst_reg, si->dst_reg, \
9054 offsetof(OBJ, OBJ_FIELD)); \
9055 if (si->dst_reg == si->src_reg) { \
9056 *insn++ = BPF_JMP_A(1); \
9057 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9058 offsetof(struct bpf_sock_ops_kern, \
9063 #define SOCK_OPS_GET_SK() \
9065 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \
9066 if (si->dst_reg == reg || si->src_reg == reg) \
9068 if (si->dst_reg == reg || si->src_reg == reg) \
9070 if (si->dst_reg == si->src_reg) { \
9071 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
9072 offsetof(struct bpf_sock_ops_kern, \
9074 fullsock_reg = reg; \
9077 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9078 struct bpf_sock_ops_kern, \
9080 fullsock_reg, si->src_reg, \
9081 offsetof(struct bpf_sock_ops_kern, \
9083 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
9084 if (si->dst_reg == si->src_reg) \
9085 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9086 offsetof(struct bpf_sock_ops_kern, \
9088 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9089 struct bpf_sock_ops_kern, sk),\
9090 si->dst_reg, si->src_reg, \
9091 offsetof(struct bpf_sock_ops_kern, sk));\
9092 if (si->dst_reg == si->src_reg) { \
9093 *insn++ = BPF_JMP_A(1); \
9094 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9095 offsetof(struct bpf_sock_ops_kern, \
9100 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
9101 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
9103 /* Helper macro for adding write access to tcp_sock or sock fields.
9104 * The macro is called with two registers, dst_reg which contains a pointer
9105 * to ctx (context) and src_reg which contains the value that should be
9106 * stored. However, we need an additional register since we cannot overwrite
9107 * dst_reg because it may be used later in the program.
9108 * Instead we "borrow" one of the other register. We first save its value
9109 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
9110 * it at the end of the macro.
9112 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
9114 int reg = BPF_REG_9; \
9115 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
9116 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
9117 if (si->dst_reg == reg || si->src_reg == reg) \
9119 if (si->dst_reg == reg || si->src_reg == reg) \
9121 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
9122 offsetof(struct bpf_sock_ops_kern, \
9124 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9125 struct bpf_sock_ops_kern, \
9128 offsetof(struct bpf_sock_ops_kern, \
9130 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
9131 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9132 struct bpf_sock_ops_kern, sk),\
9134 offsetof(struct bpf_sock_ops_kern, sk));\
9135 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \
9137 offsetof(OBJ, OBJ_FIELD)); \
9138 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
9139 offsetof(struct bpf_sock_ops_kern, \
9143 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
9145 if (TYPE == BPF_WRITE) \
9146 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
9148 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
9151 if (insn > insn_buf)
9152 return insn - insn_buf;
9155 case offsetof(struct bpf_sock_ops, op):
9156 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9158 si->dst_reg, si->src_reg,
9159 offsetof(struct bpf_sock_ops_kern, op));
9162 case offsetof(struct bpf_sock_ops, replylong[0]) ...
9163 offsetof(struct bpf_sock_ops, replylong[3]):
9164 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
9165 sizeof_field(struct bpf_sock_ops_kern, reply));
9166 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
9167 sizeof_field(struct bpf_sock_ops_kern, replylong));
9169 off -= offsetof(struct bpf_sock_ops, replylong[0]);
9170 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
9171 if (type == BPF_WRITE)
9172 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9175 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9179 case offsetof(struct bpf_sock_ops, family):
9180 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9182 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9183 struct bpf_sock_ops_kern, sk),
9184 si->dst_reg, si->src_reg,
9185 offsetof(struct bpf_sock_ops_kern, sk));
9186 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9187 offsetof(struct sock_common, skc_family));
9190 case offsetof(struct bpf_sock_ops, remote_ip4):
9191 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9193 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9194 struct bpf_sock_ops_kern, sk),
9195 si->dst_reg, si->src_reg,
9196 offsetof(struct bpf_sock_ops_kern, sk));
9197 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9198 offsetof(struct sock_common, skc_daddr));
9201 case offsetof(struct bpf_sock_ops, local_ip4):
9202 BUILD_BUG_ON(sizeof_field(struct sock_common,
9203 skc_rcv_saddr) != 4);
9205 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9206 struct bpf_sock_ops_kern, sk),
9207 si->dst_reg, si->src_reg,
9208 offsetof(struct bpf_sock_ops_kern, sk));
9209 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9210 offsetof(struct sock_common,
9214 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
9215 offsetof(struct bpf_sock_ops, remote_ip6[3]):
9216 #if IS_ENABLED(CONFIG_IPV6)
9217 BUILD_BUG_ON(sizeof_field(struct sock_common,
9218 skc_v6_daddr.s6_addr32[0]) != 4);
9221 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
9222 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9223 struct bpf_sock_ops_kern, sk),
9224 si->dst_reg, si->src_reg,
9225 offsetof(struct bpf_sock_ops_kern, sk));
9226 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9227 offsetof(struct sock_common,
9228 skc_v6_daddr.s6_addr32[0]) +
9231 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9235 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
9236 offsetof(struct bpf_sock_ops, local_ip6[3]):
9237 #if IS_ENABLED(CONFIG_IPV6)
9238 BUILD_BUG_ON(sizeof_field(struct sock_common,
9239 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9242 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
9243 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9244 struct bpf_sock_ops_kern, sk),
9245 si->dst_reg, si->src_reg,
9246 offsetof(struct bpf_sock_ops_kern, sk));
9247 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9248 offsetof(struct sock_common,
9249 skc_v6_rcv_saddr.s6_addr32[0]) +
9252 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9256 case offsetof(struct bpf_sock_ops, remote_port):
9257 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9259 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9260 struct bpf_sock_ops_kern, sk),
9261 si->dst_reg, si->src_reg,
9262 offsetof(struct bpf_sock_ops_kern, sk));
9263 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9264 offsetof(struct sock_common, skc_dport));
9265 #ifndef __BIG_ENDIAN_BITFIELD
9266 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9270 case offsetof(struct bpf_sock_ops, local_port):
9271 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9273 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9274 struct bpf_sock_ops_kern, sk),
9275 si->dst_reg, si->src_reg,
9276 offsetof(struct bpf_sock_ops_kern, sk));
9277 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9278 offsetof(struct sock_common, skc_num));
9281 case offsetof(struct bpf_sock_ops, is_fullsock):
9282 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9283 struct bpf_sock_ops_kern,
9285 si->dst_reg, si->src_reg,
9286 offsetof(struct bpf_sock_ops_kern,
9290 case offsetof(struct bpf_sock_ops, state):
9291 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
9293 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9294 struct bpf_sock_ops_kern, sk),
9295 si->dst_reg, si->src_reg,
9296 offsetof(struct bpf_sock_ops_kern, sk));
9297 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
9298 offsetof(struct sock_common, skc_state));
9301 case offsetof(struct bpf_sock_ops, rtt_min):
9302 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
9303 sizeof(struct minmax));
9304 BUILD_BUG_ON(sizeof(struct minmax) <
9305 sizeof(struct minmax_sample));
9307 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9308 struct bpf_sock_ops_kern, sk),
9309 si->dst_reg, si->src_reg,
9310 offsetof(struct bpf_sock_ops_kern, sk));
9311 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9312 offsetof(struct tcp_sock, rtt_min) +
9313 sizeof_field(struct minmax_sample, t));
9316 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
9317 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
9321 case offsetof(struct bpf_sock_ops, sk_txhash):
9322 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
9325 case offsetof(struct bpf_sock_ops, snd_cwnd):
9326 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
9328 case offsetof(struct bpf_sock_ops, srtt_us):
9329 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
9331 case offsetof(struct bpf_sock_ops, snd_ssthresh):
9332 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
9334 case offsetof(struct bpf_sock_ops, rcv_nxt):
9335 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
9337 case offsetof(struct bpf_sock_ops, snd_nxt):
9338 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
9340 case offsetof(struct bpf_sock_ops, snd_una):
9341 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
9343 case offsetof(struct bpf_sock_ops, mss_cache):
9344 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
9346 case offsetof(struct bpf_sock_ops, ecn_flags):
9347 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
9349 case offsetof(struct bpf_sock_ops, rate_delivered):
9350 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
9352 case offsetof(struct bpf_sock_ops, rate_interval_us):
9353 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
9355 case offsetof(struct bpf_sock_ops, packets_out):
9356 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
9358 case offsetof(struct bpf_sock_ops, retrans_out):
9359 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
9361 case offsetof(struct bpf_sock_ops, total_retrans):
9362 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
9364 case offsetof(struct bpf_sock_ops, segs_in):
9365 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
9367 case offsetof(struct bpf_sock_ops, data_segs_in):
9368 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
9370 case offsetof(struct bpf_sock_ops, segs_out):
9371 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
9373 case offsetof(struct bpf_sock_ops, data_segs_out):
9374 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
9376 case offsetof(struct bpf_sock_ops, lost_out):
9377 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
9379 case offsetof(struct bpf_sock_ops, sacked_out):
9380 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
9382 case offsetof(struct bpf_sock_ops, bytes_received):
9383 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
9385 case offsetof(struct bpf_sock_ops, bytes_acked):
9386 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
9388 case offsetof(struct bpf_sock_ops, sk):
9391 case offsetof(struct bpf_sock_ops, skb_data_end):
9392 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9394 si->dst_reg, si->src_reg,
9395 offsetof(struct bpf_sock_ops_kern,
9398 case offsetof(struct bpf_sock_ops, skb_data):
9399 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9401 si->dst_reg, si->src_reg,
9402 offsetof(struct bpf_sock_ops_kern,
9404 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9405 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9406 si->dst_reg, si->dst_reg,
9407 offsetof(struct sk_buff, data));
9409 case offsetof(struct bpf_sock_ops, skb_len):
9410 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9412 si->dst_reg, si->src_reg,
9413 offsetof(struct bpf_sock_ops_kern,
9415 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9416 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9417 si->dst_reg, si->dst_reg,
9418 offsetof(struct sk_buff, len));
9420 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9421 off = offsetof(struct sk_buff, cb);
9422 off += offsetof(struct tcp_skb_cb, tcp_flags);
9423 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
9424 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9426 si->dst_reg, si->src_reg,
9427 offsetof(struct bpf_sock_ops_kern,
9429 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9430 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
9432 si->dst_reg, si->dst_reg, off);
9435 return insn - insn_buf;
9438 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
9439 const struct bpf_insn *si,
9440 struct bpf_insn *insn_buf,
9441 struct bpf_prog *prog, u32 *target_size)
9443 struct bpf_insn *insn = insn_buf;
9447 case offsetof(struct __sk_buff, data_end):
9449 off -= offsetof(struct __sk_buff, data_end);
9450 off += offsetof(struct sk_buff, cb);
9451 off += offsetof(struct tcp_skb_cb, bpf.data_end);
9452 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9456 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9460 return insn - insn_buf;
9463 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
9464 const struct bpf_insn *si,
9465 struct bpf_insn *insn_buf,
9466 struct bpf_prog *prog, u32 *target_size)
9468 struct bpf_insn *insn = insn_buf;
9469 #if IS_ENABLED(CONFIG_IPV6)
9473 /* convert ctx uses the fact sg element is first in struct */
9474 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
9477 case offsetof(struct sk_msg_md, data):
9478 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
9479 si->dst_reg, si->src_reg,
9480 offsetof(struct sk_msg, data));
9482 case offsetof(struct sk_msg_md, data_end):
9483 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
9484 si->dst_reg, si->src_reg,
9485 offsetof(struct sk_msg, data_end));
9487 case offsetof(struct sk_msg_md, family):
9488 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9490 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9492 si->dst_reg, si->src_reg,
9493 offsetof(struct sk_msg, sk));
9494 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9495 offsetof(struct sock_common, skc_family));
9498 case offsetof(struct sk_msg_md, remote_ip4):
9499 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9501 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9503 si->dst_reg, si->src_reg,
9504 offsetof(struct sk_msg, sk));
9505 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9506 offsetof(struct sock_common, skc_daddr));
9509 case offsetof(struct sk_msg_md, local_ip4):
9510 BUILD_BUG_ON(sizeof_field(struct sock_common,
9511 skc_rcv_saddr) != 4);
9513 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9515 si->dst_reg, si->src_reg,
9516 offsetof(struct sk_msg, sk));
9517 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9518 offsetof(struct sock_common,
9522 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
9523 offsetof(struct sk_msg_md, remote_ip6[3]):
9524 #if IS_ENABLED(CONFIG_IPV6)
9525 BUILD_BUG_ON(sizeof_field(struct sock_common,
9526 skc_v6_daddr.s6_addr32[0]) != 4);
9529 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
9530 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9532 si->dst_reg, si->src_reg,
9533 offsetof(struct sk_msg, sk));
9534 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9535 offsetof(struct sock_common,
9536 skc_v6_daddr.s6_addr32[0]) +
9539 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9543 case offsetof(struct sk_msg_md, local_ip6[0]) ...
9544 offsetof(struct sk_msg_md, local_ip6[3]):
9545 #if IS_ENABLED(CONFIG_IPV6)
9546 BUILD_BUG_ON(sizeof_field(struct sock_common,
9547 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9550 off -= offsetof(struct sk_msg_md, local_ip6[0]);
9551 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9553 si->dst_reg, si->src_reg,
9554 offsetof(struct sk_msg, sk));
9555 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9556 offsetof(struct sock_common,
9557 skc_v6_rcv_saddr.s6_addr32[0]) +
9560 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9564 case offsetof(struct sk_msg_md, remote_port):
9565 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9567 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9569 si->dst_reg, si->src_reg,
9570 offsetof(struct sk_msg, sk));
9571 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9572 offsetof(struct sock_common, skc_dport));
9573 #ifndef __BIG_ENDIAN_BITFIELD
9574 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9578 case offsetof(struct sk_msg_md, local_port):
9579 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9581 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9583 si->dst_reg, si->src_reg,
9584 offsetof(struct sk_msg, sk));
9585 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9586 offsetof(struct sock_common, skc_num));
9589 case offsetof(struct sk_msg_md, size):
9590 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
9591 si->dst_reg, si->src_reg,
9592 offsetof(struct sk_msg_sg, size));
9595 case offsetof(struct sk_msg_md, sk):
9596 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
9597 si->dst_reg, si->src_reg,
9598 offsetof(struct sk_msg, sk));
9602 return insn - insn_buf;
9605 const struct bpf_verifier_ops sk_filter_verifier_ops = {
9606 .get_func_proto = sk_filter_func_proto,
9607 .is_valid_access = sk_filter_is_valid_access,
9608 .convert_ctx_access = bpf_convert_ctx_access,
9609 .gen_ld_abs = bpf_gen_ld_abs,
9612 const struct bpf_prog_ops sk_filter_prog_ops = {
9613 .test_run = bpf_prog_test_run_skb,
9616 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
9617 .get_func_proto = tc_cls_act_func_proto,
9618 .is_valid_access = tc_cls_act_is_valid_access,
9619 .convert_ctx_access = tc_cls_act_convert_ctx_access,
9620 .gen_prologue = tc_cls_act_prologue,
9621 .gen_ld_abs = bpf_gen_ld_abs,
9624 const struct bpf_prog_ops tc_cls_act_prog_ops = {
9625 .test_run = bpf_prog_test_run_skb,
9628 const struct bpf_verifier_ops xdp_verifier_ops = {
9629 .get_func_proto = xdp_func_proto,
9630 .is_valid_access = xdp_is_valid_access,
9631 .convert_ctx_access = xdp_convert_ctx_access,
9632 .gen_prologue = bpf_noop_prologue,
9635 const struct bpf_prog_ops xdp_prog_ops = {
9636 .test_run = bpf_prog_test_run_xdp,
9639 const struct bpf_verifier_ops cg_skb_verifier_ops = {
9640 .get_func_proto = cg_skb_func_proto,
9641 .is_valid_access = cg_skb_is_valid_access,
9642 .convert_ctx_access = bpf_convert_ctx_access,
9645 const struct bpf_prog_ops cg_skb_prog_ops = {
9646 .test_run = bpf_prog_test_run_skb,
9649 const struct bpf_verifier_ops lwt_in_verifier_ops = {
9650 .get_func_proto = lwt_in_func_proto,
9651 .is_valid_access = lwt_is_valid_access,
9652 .convert_ctx_access = bpf_convert_ctx_access,
9655 const struct bpf_prog_ops lwt_in_prog_ops = {
9656 .test_run = bpf_prog_test_run_skb,
9659 const struct bpf_verifier_ops lwt_out_verifier_ops = {
9660 .get_func_proto = lwt_out_func_proto,
9661 .is_valid_access = lwt_is_valid_access,
9662 .convert_ctx_access = bpf_convert_ctx_access,
9665 const struct bpf_prog_ops lwt_out_prog_ops = {
9666 .test_run = bpf_prog_test_run_skb,
9669 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
9670 .get_func_proto = lwt_xmit_func_proto,
9671 .is_valid_access = lwt_is_valid_access,
9672 .convert_ctx_access = bpf_convert_ctx_access,
9673 .gen_prologue = tc_cls_act_prologue,
9676 const struct bpf_prog_ops lwt_xmit_prog_ops = {
9677 .test_run = bpf_prog_test_run_skb,
9680 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
9681 .get_func_proto = lwt_seg6local_func_proto,
9682 .is_valid_access = lwt_is_valid_access,
9683 .convert_ctx_access = bpf_convert_ctx_access,
9686 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
9687 .test_run = bpf_prog_test_run_skb,
9690 const struct bpf_verifier_ops cg_sock_verifier_ops = {
9691 .get_func_proto = sock_filter_func_proto,
9692 .is_valid_access = sock_filter_is_valid_access,
9693 .convert_ctx_access = bpf_sock_convert_ctx_access,
9696 const struct bpf_prog_ops cg_sock_prog_ops = {
9699 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
9700 .get_func_proto = sock_addr_func_proto,
9701 .is_valid_access = sock_addr_is_valid_access,
9702 .convert_ctx_access = sock_addr_convert_ctx_access,
9705 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
9708 const struct bpf_verifier_ops sock_ops_verifier_ops = {
9709 .get_func_proto = sock_ops_func_proto,
9710 .is_valid_access = sock_ops_is_valid_access,
9711 .convert_ctx_access = sock_ops_convert_ctx_access,
9714 const struct bpf_prog_ops sock_ops_prog_ops = {
9717 const struct bpf_verifier_ops sk_skb_verifier_ops = {
9718 .get_func_proto = sk_skb_func_proto,
9719 .is_valid_access = sk_skb_is_valid_access,
9720 .convert_ctx_access = sk_skb_convert_ctx_access,
9721 .gen_prologue = sk_skb_prologue,
9724 const struct bpf_prog_ops sk_skb_prog_ops = {
9727 const struct bpf_verifier_ops sk_msg_verifier_ops = {
9728 .get_func_proto = sk_msg_func_proto,
9729 .is_valid_access = sk_msg_is_valid_access,
9730 .convert_ctx_access = sk_msg_convert_ctx_access,
9731 .gen_prologue = bpf_noop_prologue,
9734 const struct bpf_prog_ops sk_msg_prog_ops = {
9737 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
9738 .get_func_proto = flow_dissector_func_proto,
9739 .is_valid_access = flow_dissector_is_valid_access,
9740 .convert_ctx_access = flow_dissector_convert_ctx_access,
9743 const struct bpf_prog_ops flow_dissector_prog_ops = {
9744 .test_run = bpf_prog_test_run_flow_dissector,
9747 int sk_detach_filter(struct sock *sk)
9750 struct sk_filter *filter;
9752 if (sock_flag(sk, SOCK_FILTER_LOCKED))
9755 filter = rcu_dereference_protected(sk->sk_filter,
9756 lockdep_sock_is_held(sk));
9758 RCU_INIT_POINTER(sk->sk_filter, NULL);
9759 sk_filter_uncharge(sk, filter);
9765 EXPORT_SYMBOL_GPL(sk_detach_filter);
9767 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
9770 struct sock_fprog_kern *fprog;
9771 struct sk_filter *filter;
9775 filter = rcu_dereference_protected(sk->sk_filter,
9776 lockdep_sock_is_held(sk));
9780 /* We're copying the filter that has been originally attached,
9781 * so no conversion/decode needed anymore. eBPF programs that
9782 * have no original program cannot be dumped through this.
9785 fprog = filter->prog->orig_prog;
9791 /* User space only enquires number of filter blocks. */
9795 if (len < fprog->len)
9799 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
9802 /* Instead of bytes, the API requests to return the number
9812 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
9813 struct sock_reuseport *reuse,
9814 struct sock *sk, struct sk_buff *skb,
9817 reuse_kern->skb = skb;
9818 reuse_kern->sk = sk;
9819 reuse_kern->selected_sk = NULL;
9820 reuse_kern->data_end = skb->data + skb_headlen(skb);
9821 reuse_kern->hash = hash;
9822 reuse_kern->reuseport_id = reuse->reuseport_id;
9823 reuse_kern->bind_inany = reuse->bind_inany;
9826 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
9827 struct bpf_prog *prog, struct sk_buff *skb,
9830 struct sk_reuseport_kern reuse_kern;
9831 enum sk_action action;
9833 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, hash);
9834 action = BPF_PROG_RUN(prog, &reuse_kern);
9836 if (action == SK_PASS)
9837 return reuse_kern.selected_sk;
9839 return ERR_PTR(-ECONNREFUSED);
9842 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
9843 struct bpf_map *, map, void *, key, u32, flags)
9845 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
9846 struct sock_reuseport *reuse;
9847 struct sock *selected_sk;
9849 selected_sk = map->ops->map_lookup_elem(map, key);
9853 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
9855 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
9856 if (sk_is_refcounted(selected_sk))
9857 sock_put(selected_sk);
9859 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
9860 * The only (!reuse) case here is - the sk has already been
9861 * unhashed (e.g. by close()), so treat it as -ENOENT.
9863 * Other maps (e.g. sock_map) do not provide this guarantee and
9864 * the sk may never be in the reuseport group to begin with.
9866 return is_sockarray ? -ENOENT : -EINVAL;
9869 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
9870 struct sock *sk = reuse_kern->sk;
9872 if (sk->sk_protocol != selected_sk->sk_protocol)
9874 else if (sk->sk_family != selected_sk->sk_family)
9875 return -EAFNOSUPPORT;
9877 /* Catch all. Likely bound to a different sockaddr. */
9881 reuse_kern->selected_sk = selected_sk;
9886 static const struct bpf_func_proto sk_select_reuseport_proto = {
9887 .func = sk_select_reuseport,
9889 .ret_type = RET_INTEGER,
9890 .arg1_type = ARG_PTR_TO_CTX,
9891 .arg2_type = ARG_CONST_MAP_PTR,
9892 .arg3_type = ARG_PTR_TO_MAP_KEY,
9893 .arg4_type = ARG_ANYTHING,
9896 BPF_CALL_4(sk_reuseport_load_bytes,
9897 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
9898 void *, to, u32, len)
9900 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
9903 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
9904 .func = sk_reuseport_load_bytes,
9906 .ret_type = RET_INTEGER,
9907 .arg1_type = ARG_PTR_TO_CTX,
9908 .arg2_type = ARG_ANYTHING,
9909 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
9910 .arg4_type = ARG_CONST_SIZE,
9913 BPF_CALL_5(sk_reuseport_load_bytes_relative,
9914 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
9915 void *, to, u32, len, u32, start_header)
9917 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
9921 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
9922 .func = sk_reuseport_load_bytes_relative,
9924 .ret_type = RET_INTEGER,
9925 .arg1_type = ARG_PTR_TO_CTX,
9926 .arg2_type = ARG_ANYTHING,
9927 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
9928 .arg4_type = ARG_CONST_SIZE,
9929 .arg5_type = ARG_ANYTHING,
9932 static const struct bpf_func_proto *
9933 sk_reuseport_func_proto(enum bpf_func_id func_id,
9934 const struct bpf_prog *prog)
9937 case BPF_FUNC_sk_select_reuseport:
9938 return &sk_select_reuseport_proto;
9939 case BPF_FUNC_skb_load_bytes:
9940 return &sk_reuseport_load_bytes_proto;
9941 case BPF_FUNC_skb_load_bytes_relative:
9942 return &sk_reuseport_load_bytes_relative_proto;
9944 return bpf_base_func_proto(func_id);
9949 sk_reuseport_is_valid_access(int off, int size,
9950 enum bpf_access_type type,
9951 const struct bpf_prog *prog,
9952 struct bpf_insn_access_aux *info)
9954 const u32 size_default = sizeof(__u32);
9956 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
9957 off % size || type != BPF_READ)
9961 case offsetof(struct sk_reuseport_md, data):
9962 info->reg_type = PTR_TO_PACKET;
9963 return size == sizeof(__u64);
9965 case offsetof(struct sk_reuseport_md, data_end):
9966 info->reg_type = PTR_TO_PACKET_END;
9967 return size == sizeof(__u64);
9969 case offsetof(struct sk_reuseport_md, hash):
9970 return size == size_default;
9972 /* Fields that allow narrowing */
9973 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
9974 if (size < sizeof_field(struct sk_buff, protocol))
9977 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
9978 case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
9979 case bpf_ctx_range(struct sk_reuseport_md, len):
9980 bpf_ctx_record_field_size(info, size_default);
9981 return bpf_ctx_narrow_access_ok(off, size, size_default);
9988 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
9989 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
9990 si->dst_reg, si->src_reg, \
9991 bpf_target_off(struct sk_reuseport_kern, F, \
9992 sizeof_field(struct sk_reuseport_kern, F), \
9996 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
9997 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
10002 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \
10003 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
10008 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
10009 const struct bpf_insn *si,
10010 struct bpf_insn *insn_buf,
10011 struct bpf_prog *prog,
10014 struct bpf_insn *insn = insn_buf;
10017 case offsetof(struct sk_reuseport_md, data):
10018 SK_REUSEPORT_LOAD_SKB_FIELD(data);
10021 case offsetof(struct sk_reuseport_md, len):
10022 SK_REUSEPORT_LOAD_SKB_FIELD(len);
10025 case offsetof(struct sk_reuseport_md, eth_protocol):
10026 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
10029 case offsetof(struct sk_reuseport_md, ip_protocol):
10030 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
10033 case offsetof(struct sk_reuseport_md, data_end):
10034 SK_REUSEPORT_LOAD_FIELD(data_end);
10037 case offsetof(struct sk_reuseport_md, hash):
10038 SK_REUSEPORT_LOAD_FIELD(hash);
10041 case offsetof(struct sk_reuseport_md, bind_inany):
10042 SK_REUSEPORT_LOAD_FIELD(bind_inany);
10046 return insn - insn_buf;
10049 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
10050 .get_func_proto = sk_reuseport_func_proto,
10051 .is_valid_access = sk_reuseport_is_valid_access,
10052 .convert_ctx_access = sk_reuseport_convert_ctx_access,
10055 const struct bpf_prog_ops sk_reuseport_prog_ops = {
10058 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
10059 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
10061 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
10062 struct sock *, sk, u64, flags)
10064 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
10065 BPF_SK_LOOKUP_F_NO_REUSEPORT)))
10067 if (unlikely(sk && sk_is_refcounted(sk)))
10068 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
10069 if (unlikely(sk && sk->sk_state == TCP_ESTABLISHED))
10070 return -ESOCKTNOSUPPORT; /* reject connected sockets */
10072 /* Check if socket is suitable for packet L3/L4 protocol */
10073 if (sk && sk->sk_protocol != ctx->protocol)
10074 return -EPROTOTYPE;
10075 if (sk && sk->sk_family != ctx->family &&
10076 (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
10077 return -EAFNOSUPPORT;
10079 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
10082 /* Select socket as lookup result */
10083 ctx->selected_sk = sk;
10084 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
10088 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
10089 .func = bpf_sk_lookup_assign,
10091 .ret_type = RET_INTEGER,
10092 .arg1_type = ARG_PTR_TO_CTX,
10093 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL,
10094 .arg3_type = ARG_ANYTHING,
10097 static const struct bpf_func_proto *
10098 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
10101 case BPF_FUNC_perf_event_output:
10102 return &bpf_event_output_data_proto;
10103 case BPF_FUNC_sk_assign:
10104 return &bpf_sk_lookup_assign_proto;
10105 case BPF_FUNC_sk_release:
10106 return &bpf_sk_release_proto;
10108 return bpf_sk_base_func_proto(func_id);
10112 static bool sk_lookup_is_valid_access(int off, int size,
10113 enum bpf_access_type type,
10114 const struct bpf_prog *prog,
10115 struct bpf_insn_access_aux *info)
10117 if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
10119 if (off % size != 0)
10121 if (type != BPF_READ)
10125 case offsetof(struct bpf_sk_lookup, sk):
10126 info->reg_type = PTR_TO_SOCKET_OR_NULL;
10127 return size == sizeof(__u64);
10129 case bpf_ctx_range(struct bpf_sk_lookup, family):
10130 case bpf_ctx_range(struct bpf_sk_lookup, protocol):
10131 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
10132 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
10133 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
10134 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
10135 case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
10136 case bpf_ctx_range(struct bpf_sk_lookup, local_port):
10137 bpf_ctx_record_field_size(info, sizeof(__u32));
10138 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
10145 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
10146 const struct bpf_insn *si,
10147 struct bpf_insn *insn_buf,
10148 struct bpf_prog *prog,
10151 struct bpf_insn *insn = insn_buf;
10154 case offsetof(struct bpf_sk_lookup, sk):
10155 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10156 offsetof(struct bpf_sk_lookup_kern, selected_sk));
10159 case offsetof(struct bpf_sk_lookup, family):
10160 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10161 bpf_target_off(struct bpf_sk_lookup_kern,
10162 family, 2, target_size));
10165 case offsetof(struct bpf_sk_lookup, protocol):
10166 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10167 bpf_target_off(struct bpf_sk_lookup_kern,
10168 protocol, 2, target_size));
10171 case offsetof(struct bpf_sk_lookup, remote_ip4):
10172 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10173 bpf_target_off(struct bpf_sk_lookup_kern,
10174 v4.saddr, 4, target_size));
10177 case offsetof(struct bpf_sk_lookup, local_ip4):
10178 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10179 bpf_target_off(struct bpf_sk_lookup_kern,
10180 v4.daddr, 4, target_size));
10183 case bpf_ctx_range_till(struct bpf_sk_lookup,
10184 remote_ip6[0], remote_ip6[3]): {
10185 #if IS_ENABLED(CONFIG_IPV6)
10188 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
10189 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10190 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10191 offsetof(struct bpf_sk_lookup_kern, v6.saddr));
10192 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10193 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10195 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10199 case bpf_ctx_range_till(struct bpf_sk_lookup,
10200 local_ip6[0], local_ip6[3]): {
10201 #if IS_ENABLED(CONFIG_IPV6)
10204 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
10205 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10206 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10207 offsetof(struct bpf_sk_lookup_kern, v6.daddr));
10208 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10209 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10211 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10215 case offsetof(struct bpf_sk_lookup, remote_port):
10216 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10217 bpf_target_off(struct bpf_sk_lookup_kern,
10218 sport, 2, target_size));
10221 case offsetof(struct bpf_sk_lookup, local_port):
10222 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10223 bpf_target_off(struct bpf_sk_lookup_kern,
10224 dport, 2, target_size));
10228 return insn - insn_buf;
10231 const struct bpf_prog_ops sk_lookup_prog_ops = {
10234 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
10235 .get_func_proto = sk_lookup_func_proto,
10236 .is_valid_access = sk_lookup_is_valid_access,
10237 .convert_ctx_access = sk_lookup_convert_ctx_access,
10240 #endif /* CONFIG_INET */
10242 DEFINE_BPF_DISPATCHER(xdp)
10244 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
10246 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
10249 #ifdef CONFIG_DEBUG_INFO_BTF
10250 BTF_ID_LIST_GLOBAL(btf_sock_ids)
10251 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
10253 #undef BTF_SOCK_TYPE
10255 u32 btf_sock_ids[MAX_BTF_SOCK_TYPE];
10258 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
10260 /* tcp6_sock type is not generated in dwarf and hence btf,
10261 * trigger an explicit type generation here.
10263 BTF_TYPE_EMIT(struct tcp6_sock);
10264 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
10265 sk->sk_family == AF_INET6)
10266 return (unsigned long)sk;
10268 return (unsigned long)NULL;
10271 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
10272 .func = bpf_skc_to_tcp6_sock,
10274 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10275 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10276 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
10279 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
10281 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
10282 return (unsigned long)sk;
10284 return (unsigned long)NULL;
10287 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
10288 .func = bpf_skc_to_tcp_sock,
10290 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10291 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10292 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
10295 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
10297 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
10298 * generated if CONFIG_INET=n. Trigger an explicit generation here.
10300 BTF_TYPE_EMIT(struct inet_timewait_sock);
10301 BTF_TYPE_EMIT(struct tcp_timewait_sock);
10304 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
10305 return (unsigned long)sk;
10308 #if IS_BUILTIN(CONFIG_IPV6)
10309 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
10310 return (unsigned long)sk;
10313 return (unsigned long)NULL;
10316 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
10317 .func = bpf_skc_to_tcp_timewait_sock,
10319 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10320 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10321 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
10324 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
10327 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10328 return (unsigned long)sk;
10331 #if IS_BUILTIN(CONFIG_IPV6)
10332 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10333 return (unsigned long)sk;
10336 return (unsigned long)NULL;
10339 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
10340 .func = bpf_skc_to_tcp_request_sock,
10342 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10343 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10344 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
10347 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
10349 /* udp6_sock type is not generated in dwarf and hence btf,
10350 * trigger an explicit type generation here.
10352 BTF_TYPE_EMIT(struct udp6_sock);
10353 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
10354 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
10355 return (unsigned long)sk;
10357 return (unsigned long)NULL;
10360 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
10361 .func = bpf_skc_to_udp6_sock,
10363 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10364 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10365 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
10368 static const struct bpf_func_proto *
10369 bpf_sk_base_func_proto(enum bpf_func_id func_id)
10371 const struct bpf_func_proto *func;
10374 case BPF_FUNC_skc_to_tcp6_sock:
10375 func = &bpf_skc_to_tcp6_sock_proto;
10377 case BPF_FUNC_skc_to_tcp_sock:
10378 func = &bpf_skc_to_tcp_sock_proto;
10380 case BPF_FUNC_skc_to_tcp_timewait_sock:
10381 func = &bpf_skc_to_tcp_timewait_sock_proto;
10383 case BPF_FUNC_skc_to_tcp_request_sock:
10384 func = &bpf_skc_to_tcp_request_sock_proto;
10386 case BPF_FUNC_skc_to_udp6_sock:
10387 func = &bpf_skc_to_udp6_sock_proto;
10390 return bpf_base_func_proto(func_id);
10393 if (!perfmon_capable())