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);
1215 int optmem_max = READ_ONCE(sysctl_optmem_max);
1217 /* same check as in sock_kmalloc() */
1218 if (filter_size <= optmem_max &&
1219 atomic_read(&sk->sk_omem_alloc) + filter_size < optmem_max) {
1220 atomic_add(filter_size, &sk->sk_omem_alloc);
1226 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1228 if (!refcount_inc_not_zero(&fp->refcnt))
1231 if (!__sk_filter_charge(sk, fp)) {
1232 sk_filter_release(fp);
1238 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1240 struct sock_filter *old_prog;
1241 struct bpf_prog *old_fp;
1242 int err, new_len, old_len = fp->len;
1243 bool seen_ld_abs = false;
1245 /* We are free to overwrite insns et al right here as it
1246 * won't be used at this point in time anymore internally
1247 * after the migration to the internal BPF instruction
1250 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1251 sizeof(struct bpf_insn));
1253 /* Conversion cannot happen on overlapping memory areas,
1254 * so we need to keep the user BPF around until the 2nd
1255 * pass. At this time, the user BPF is stored in fp->insns.
1257 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1258 GFP_KERNEL | __GFP_NOWARN);
1264 /* 1st pass: calculate the new program length. */
1265 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1270 /* Expand fp for appending the new filter representation. */
1272 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1274 /* The old_fp is still around in case we couldn't
1275 * allocate new memory, so uncharge on that one.
1284 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1285 err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1288 /* 2nd bpf_convert_filter() can fail only if it fails
1289 * to allocate memory, remapping must succeed. Note,
1290 * that at this time old_fp has already been released
1295 fp = bpf_prog_select_runtime(fp, &err);
1305 __bpf_prog_release(fp);
1306 return ERR_PTR(err);
1309 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1310 bpf_aux_classic_check_t trans)
1314 fp->bpf_func = NULL;
1317 err = bpf_check_classic(fp->insns, fp->len);
1319 __bpf_prog_release(fp);
1320 return ERR_PTR(err);
1323 /* There might be additional checks and transformations
1324 * needed on classic filters, f.e. in case of seccomp.
1327 err = trans(fp->insns, fp->len);
1329 __bpf_prog_release(fp);
1330 return ERR_PTR(err);
1334 /* Probe if we can JIT compile the filter and if so, do
1335 * the compilation of the filter.
1337 bpf_jit_compile(fp);
1339 /* JIT compiler couldn't process this filter, so do the
1340 * internal BPF translation for the optimized interpreter.
1343 fp = bpf_migrate_filter(fp);
1349 * bpf_prog_create - create an unattached filter
1350 * @pfp: the unattached filter that is created
1351 * @fprog: the filter program
1353 * Create a filter independent of any socket. We first run some
1354 * sanity checks on it to make sure it does not explode on us later.
1355 * If an error occurs or there is insufficient memory for the filter
1356 * a negative errno code is returned. On success the return is zero.
1358 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1360 unsigned int fsize = bpf_classic_proglen(fprog);
1361 struct bpf_prog *fp;
1363 /* Make sure new filter is there and in the right amounts. */
1364 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1367 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1371 memcpy(fp->insns, fprog->filter, fsize);
1373 fp->len = fprog->len;
1374 /* Since unattached filters are not copied back to user
1375 * space through sk_get_filter(), we do not need to hold
1376 * a copy here, and can spare us the work.
1378 fp->orig_prog = NULL;
1380 /* bpf_prepare_filter() already takes care of freeing
1381 * memory in case something goes wrong.
1383 fp = bpf_prepare_filter(fp, NULL);
1390 EXPORT_SYMBOL_GPL(bpf_prog_create);
1393 * bpf_prog_create_from_user - create an unattached filter from user buffer
1394 * @pfp: the unattached filter that is created
1395 * @fprog: the filter program
1396 * @trans: post-classic verifier transformation handler
1397 * @save_orig: save classic BPF program
1399 * This function effectively does the same as bpf_prog_create(), only
1400 * that it builds up its insns buffer from user space provided buffer.
1401 * It also allows for passing a bpf_aux_classic_check_t handler.
1403 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1404 bpf_aux_classic_check_t trans, bool save_orig)
1406 unsigned int fsize = bpf_classic_proglen(fprog);
1407 struct bpf_prog *fp;
1410 /* Make sure new filter is there and in the right amounts. */
1411 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1414 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1418 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1419 __bpf_prog_free(fp);
1423 fp->len = fprog->len;
1424 fp->orig_prog = NULL;
1427 err = bpf_prog_store_orig_filter(fp, fprog);
1429 __bpf_prog_free(fp);
1434 /* bpf_prepare_filter() already takes care of freeing
1435 * memory in case something goes wrong.
1437 fp = bpf_prepare_filter(fp, trans);
1444 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1446 void bpf_prog_destroy(struct bpf_prog *fp)
1448 __bpf_prog_release(fp);
1450 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1452 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1454 struct sk_filter *fp, *old_fp;
1456 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1462 if (!__sk_filter_charge(sk, fp)) {
1466 refcount_set(&fp->refcnt, 1);
1468 old_fp = rcu_dereference_protected(sk->sk_filter,
1469 lockdep_sock_is_held(sk));
1470 rcu_assign_pointer(sk->sk_filter, fp);
1473 sk_filter_uncharge(sk, old_fp);
1479 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1481 unsigned int fsize = bpf_classic_proglen(fprog);
1482 struct bpf_prog *prog;
1485 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1486 return ERR_PTR(-EPERM);
1488 /* Make sure new filter is there and in the right amounts. */
1489 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1490 return ERR_PTR(-EINVAL);
1492 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1494 return ERR_PTR(-ENOMEM);
1496 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1497 __bpf_prog_free(prog);
1498 return ERR_PTR(-EFAULT);
1501 prog->len = fprog->len;
1503 err = bpf_prog_store_orig_filter(prog, fprog);
1505 __bpf_prog_free(prog);
1506 return ERR_PTR(-ENOMEM);
1509 /* bpf_prepare_filter() already takes care of freeing
1510 * memory in case something goes wrong.
1512 return bpf_prepare_filter(prog, NULL);
1516 * sk_attach_filter - attach a socket filter
1517 * @fprog: the filter program
1518 * @sk: the socket to use
1520 * Attach the user's filter code. We first run some sanity checks on
1521 * it to make sure it does not explode on us later. If an error
1522 * occurs or there is insufficient memory for the filter a negative
1523 * errno code is returned. On success the return is zero.
1525 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1527 struct bpf_prog *prog = __get_filter(fprog, sk);
1531 return PTR_ERR(prog);
1533 err = __sk_attach_prog(prog, sk);
1535 __bpf_prog_release(prog);
1541 EXPORT_SYMBOL_GPL(sk_attach_filter);
1543 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1545 struct bpf_prog *prog = __get_filter(fprog, sk);
1549 return PTR_ERR(prog);
1551 if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max))
1554 err = reuseport_attach_prog(sk, prog);
1557 __bpf_prog_release(prog);
1562 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1564 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1565 return ERR_PTR(-EPERM);
1567 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1570 int sk_attach_bpf(u32 ufd, struct sock *sk)
1572 struct bpf_prog *prog = __get_bpf(ufd, sk);
1576 return PTR_ERR(prog);
1578 err = __sk_attach_prog(prog, sk);
1587 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1589 struct bpf_prog *prog;
1592 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1595 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1596 if (PTR_ERR(prog) == -EINVAL)
1597 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1599 return PTR_ERR(prog);
1601 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1602 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1603 * bpf prog (e.g. sockmap). It depends on the
1604 * limitation imposed by bpf_prog_load().
1605 * Hence, sysctl_optmem_max is not checked.
1607 if ((sk->sk_type != SOCK_STREAM &&
1608 sk->sk_type != SOCK_DGRAM) ||
1609 (sk->sk_protocol != IPPROTO_UDP &&
1610 sk->sk_protocol != IPPROTO_TCP) ||
1611 (sk->sk_family != AF_INET &&
1612 sk->sk_family != AF_INET6)) {
1617 /* BPF_PROG_TYPE_SOCKET_FILTER */
1618 if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max)) {
1624 err = reuseport_attach_prog(sk, prog);
1632 void sk_reuseport_prog_free(struct bpf_prog *prog)
1637 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1640 bpf_prog_destroy(prog);
1643 struct bpf_scratchpad {
1645 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1646 u8 buff[MAX_BPF_STACK];
1650 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1652 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1653 unsigned int write_len)
1655 return skb_ensure_writable(skb, write_len);
1658 static inline int bpf_try_make_writable(struct sk_buff *skb,
1659 unsigned int write_len)
1661 int err = __bpf_try_make_writable(skb, write_len);
1663 bpf_compute_data_pointers(skb);
1667 static int bpf_try_make_head_writable(struct sk_buff *skb)
1669 return bpf_try_make_writable(skb, skb_headlen(skb));
1672 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1674 if (skb_at_tc_ingress(skb))
1675 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1678 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1680 if (skb_at_tc_ingress(skb))
1681 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1684 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1685 const void *, from, u32, len, u64, flags)
1689 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1691 if (unlikely(offset > INT_MAX))
1693 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1696 ptr = skb->data + offset;
1697 if (flags & BPF_F_RECOMPUTE_CSUM)
1698 __skb_postpull_rcsum(skb, ptr, len, offset);
1700 memcpy(ptr, from, len);
1702 if (flags & BPF_F_RECOMPUTE_CSUM)
1703 __skb_postpush_rcsum(skb, ptr, len, offset);
1704 if (flags & BPF_F_INVALIDATE_HASH)
1705 skb_clear_hash(skb);
1710 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1711 .func = bpf_skb_store_bytes,
1713 .ret_type = RET_INTEGER,
1714 .arg1_type = ARG_PTR_TO_CTX,
1715 .arg2_type = ARG_ANYTHING,
1716 .arg3_type = ARG_PTR_TO_MEM,
1717 .arg4_type = ARG_CONST_SIZE,
1718 .arg5_type = ARG_ANYTHING,
1721 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1722 void *, to, u32, len)
1726 if (unlikely(offset > INT_MAX))
1729 ptr = skb_header_pointer(skb, offset, len, to);
1733 memcpy(to, ptr, len);
1741 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1742 .func = bpf_skb_load_bytes,
1744 .ret_type = RET_INTEGER,
1745 .arg1_type = ARG_PTR_TO_CTX,
1746 .arg2_type = ARG_ANYTHING,
1747 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1748 .arg4_type = ARG_CONST_SIZE,
1751 BPF_CALL_4(bpf_flow_dissector_load_bytes,
1752 const struct bpf_flow_dissector *, ctx, u32, offset,
1753 void *, to, u32, len)
1757 if (unlikely(offset > 0xffff))
1760 if (unlikely(!ctx->skb))
1763 ptr = skb_header_pointer(ctx->skb, offset, len, to);
1767 memcpy(to, ptr, len);
1775 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1776 .func = bpf_flow_dissector_load_bytes,
1778 .ret_type = RET_INTEGER,
1779 .arg1_type = ARG_PTR_TO_CTX,
1780 .arg2_type = ARG_ANYTHING,
1781 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1782 .arg4_type = ARG_CONST_SIZE,
1785 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1786 u32, offset, void *, to, u32, len, u32, start_header)
1788 u8 *end = skb_tail_pointer(skb);
1791 if (unlikely(offset > 0xffff))
1794 switch (start_header) {
1795 case BPF_HDR_START_MAC:
1796 if (unlikely(!skb_mac_header_was_set(skb)))
1798 start = skb_mac_header(skb);
1800 case BPF_HDR_START_NET:
1801 start = skb_network_header(skb);
1807 ptr = start + offset;
1809 if (likely(ptr + len <= end)) {
1810 memcpy(to, ptr, len);
1819 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1820 .func = bpf_skb_load_bytes_relative,
1822 .ret_type = RET_INTEGER,
1823 .arg1_type = ARG_PTR_TO_CTX,
1824 .arg2_type = ARG_ANYTHING,
1825 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1826 .arg4_type = ARG_CONST_SIZE,
1827 .arg5_type = ARG_ANYTHING,
1830 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1832 /* Idea is the following: should the needed direct read/write
1833 * test fail during runtime, we can pull in more data and redo
1834 * again, since implicitly, we invalidate previous checks here.
1836 * Or, since we know how much we need to make read/writeable,
1837 * this can be done once at the program beginning for direct
1838 * access case. By this we overcome limitations of only current
1839 * headroom being accessible.
1841 return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1844 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1845 .func = bpf_skb_pull_data,
1847 .ret_type = RET_INTEGER,
1848 .arg1_type = ARG_PTR_TO_CTX,
1849 .arg2_type = ARG_ANYTHING,
1852 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1854 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1857 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1858 .func = bpf_sk_fullsock,
1860 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
1861 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
1864 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1865 unsigned int write_len)
1867 int err = __bpf_try_make_writable(skb, write_len);
1869 bpf_compute_data_end_sk_skb(skb);
1873 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1875 /* Idea is the following: should the needed direct read/write
1876 * test fail during runtime, we can pull in more data and redo
1877 * again, since implicitly, we invalidate previous checks here.
1879 * Or, since we know how much we need to make read/writeable,
1880 * this can be done once at the program beginning for direct
1881 * access case. By this we overcome limitations of only current
1882 * headroom being accessible.
1884 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1887 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1888 .func = sk_skb_pull_data,
1890 .ret_type = RET_INTEGER,
1891 .arg1_type = ARG_PTR_TO_CTX,
1892 .arg2_type = ARG_ANYTHING,
1895 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1896 u64, from, u64, to, u64, flags)
1900 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1902 if (unlikely(offset > 0xffff || offset & 1))
1904 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1907 ptr = (__sum16 *)(skb->data + offset);
1908 switch (flags & BPF_F_HDR_FIELD_MASK) {
1910 if (unlikely(from != 0))
1913 csum_replace_by_diff(ptr, to);
1916 csum_replace2(ptr, from, to);
1919 csum_replace4(ptr, from, to);
1928 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1929 .func = bpf_l3_csum_replace,
1931 .ret_type = RET_INTEGER,
1932 .arg1_type = ARG_PTR_TO_CTX,
1933 .arg2_type = ARG_ANYTHING,
1934 .arg3_type = ARG_ANYTHING,
1935 .arg4_type = ARG_ANYTHING,
1936 .arg5_type = ARG_ANYTHING,
1939 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1940 u64, from, u64, to, u64, flags)
1942 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1943 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1944 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1947 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1948 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1950 if (unlikely(offset > 0xffff || offset & 1))
1952 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1955 ptr = (__sum16 *)(skb->data + offset);
1956 if (is_mmzero && !do_mforce && !*ptr)
1959 switch (flags & BPF_F_HDR_FIELD_MASK) {
1961 if (unlikely(from != 0))
1964 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1967 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1970 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1976 if (is_mmzero && !*ptr)
1977 *ptr = CSUM_MANGLED_0;
1981 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1982 .func = bpf_l4_csum_replace,
1984 .ret_type = RET_INTEGER,
1985 .arg1_type = ARG_PTR_TO_CTX,
1986 .arg2_type = ARG_ANYTHING,
1987 .arg3_type = ARG_ANYTHING,
1988 .arg4_type = ARG_ANYTHING,
1989 .arg5_type = ARG_ANYTHING,
1992 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1993 __be32 *, to, u32, to_size, __wsum, seed)
1995 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1996 u32 diff_size = from_size + to_size;
1999 /* This is quite flexible, some examples:
2001 * from_size == 0, to_size > 0, seed := csum --> pushing data
2002 * from_size > 0, to_size == 0, seed := csum --> pulling data
2003 * from_size > 0, to_size > 0, seed := 0 --> diffing data
2005 * Even for diffing, from_size and to_size don't need to be equal.
2007 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
2008 diff_size > sizeof(sp->diff)))
2011 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
2012 sp->diff[j] = ~from[i];
2013 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
2014 sp->diff[j] = to[i];
2016 return csum_partial(sp->diff, diff_size, seed);
2019 static const struct bpf_func_proto bpf_csum_diff_proto = {
2020 .func = bpf_csum_diff,
2023 .ret_type = RET_INTEGER,
2024 .arg1_type = ARG_PTR_TO_MEM_OR_NULL,
2025 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
2026 .arg3_type = ARG_PTR_TO_MEM_OR_NULL,
2027 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
2028 .arg5_type = ARG_ANYTHING,
2031 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2033 /* The interface is to be used in combination with bpf_csum_diff()
2034 * for direct packet writes. csum rotation for alignment as well
2035 * as emulating csum_sub() can be done from the eBPF program.
2037 if (skb->ip_summed == CHECKSUM_COMPLETE)
2038 return (skb->csum = csum_add(skb->csum, csum));
2043 static const struct bpf_func_proto bpf_csum_update_proto = {
2044 .func = bpf_csum_update,
2046 .ret_type = RET_INTEGER,
2047 .arg1_type = ARG_PTR_TO_CTX,
2048 .arg2_type = ARG_ANYTHING,
2051 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
2053 /* The interface is to be used in combination with bpf_skb_adjust_room()
2054 * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET
2055 * is passed as flags, for example.
2058 case BPF_CSUM_LEVEL_INC:
2059 __skb_incr_checksum_unnecessary(skb);
2061 case BPF_CSUM_LEVEL_DEC:
2062 __skb_decr_checksum_unnecessary(skb);
2064 case BPF_CSUM_LEVEL_RESET:
2065 __skb_reset_checksum_unnecessary(skb);
2067 case BPF_CSUM_LEVEL_QUERY:
2068 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2069 skb->csum_level : -EACCES;
2077 static const struct bpf_func_proto bpf_csum_level_proto = {
2078 .func = bpf_csum_level,
2080 .ret_type = RET_INTEGER,
2081 .arg1_type = ARG_PTR_TO_CTX,
2082 .arg2_type = ARG_ANYTHING,
2085 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2087 return dev_forward_skb(dev, skb);
2090 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2091 struct sk_buff *skb)
2093 int ret = ____dev_forward_skb(dev, skb);
2097 ret = netif_rx(skb);
2103 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2107 if (dev_xmit_recursion()) {
2108 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2116 dev_xmit_recursion_inc();
2117 ret = dev_queue_xmit(skb);
2118 dev_xmit_recursion_dec();
2123 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2126 unsigned int mlen = skb_network_offset(skb);
2129 __skb_pull(skb, mlen);
2131 /* At ingress, the mac header has already been pulled once.
2132 * At egress, skb_pospull_rcsum has to be done in case that
2133 * the skb is originated from ingress (i.e. a forwarded skb)
2134 * to ensure that rcsum starts at net header.
2136 if (!skb_at_tc_ingress(skb))
2137 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2139 skb_pop_mac_header(skb);
2140 skb_reset_mac_len(skb);
2141 return flags & BPF_F_INGRESS ?
2142 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2145 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2148 /* Verify that a link layer header is carried */
2149 if (unlikely(skb->mac_header >= skb->network_header)) {
2154 bpf_push_mac_rcsum(skb);
2155 return flags & BPF_F_INGRESS ?
2156 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2159 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2162 if (dev_is_mac_header_xmit(dev))
2163 return __bpf_redirect_common(skb, dev, flags);
2165 return __bpf_redirect_no_mac(skb, dev, flags);
2168 #if IS_ENABLED(CONFIG_IPV6)
2169 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2170 struct net_device *dev, struct bpf_nh_params *nh)
2172 u32 hh_len = LL_RESERVED_SPACE(dev);
2173 const struct in6_addr *nexthop;
2174 struct dst_entry *dst = NULL;
2175 struct neighbour *neigh;
2177 if (dev_xmit_recursion()) {
2178 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2185 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2186 struct sk_buff *skb2;
2188 skb2 = skb_realloc_headroom(skb, hh_len);
2189 if (unlikely(!skb2)) {
2194 skb_set_owner_w(skb2, skb->sk);
2202 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2203 &ipv6_hdr(skb)->daddr);
2205 nexthop = &nh->ipv6_nh;
2207 neigh = ip_neigh_gw6(dev, nexthop);
2208 if (likely(!IS_ERR(neigh))) {
2211 sock_confirm_neigh(skb, neigh);
2212 dev_xmit_recursion_inc();
2213 ret = neigh_output(neigh, skb, false);
2214 dev_xmit_recursion_dec();
2215 rcu_read_unlock_bh();
2218 rcu_read_unlock_bh();
2220 IP6_INC_STATS(dev_net(dst->dev),
2221 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2227 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2228 struct bpf_nh_params *nh)
2230 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2231 struct net *net = dev_net(dev);
2232 int err, ret = NET_XMIT_DROP;
2235 struct dst_entry *dst;
2236 struct flowi6 fl6 = {
2237 .flowi6_flags = FLOWI_FLAG_ANYSRC,
2238 .flowi6_mark = skb->mark,
2239 .flowlabel = ip6_flowinfo(ip6h),
2240 .flowi6_oif = dev->ifindex,
2241 .flowi6_proto = ip6h->nexthdr,
2242 .daddr = ip6h->daddr,
2243 .saddr = ip6h->saddr,
2246 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2250 skb_dst_set(skb, dst);
2251 } else if (nh->nh_family != AF_INET6) {
2255 err = bpf_out_neigh_v6(net, skb, dev, nh);
2256 if (unlikely(net_xmit_eval(err)))
2257 dev->stats.tx_errors++;
2259 ret = NET_XMIT_SUCCESS;
2262 dev->stats.tx_errors++;
2268 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2269 struct bpf_nh_params *nh)
2272 return NET_XMIT_DROP;
2274 #endif /* CONFIG_IPV6 */
2276 #if IS_ENABLED(CONFIG_INET)
2277 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2278 struct net_device *dev, struct bpf_nh_params *nh)
2280 u32 hh_len = LL_RESERVED_SPACE(dev);
2281 struct neighbour *neigh;
2282 bool is_v6gw = false;
2284 if (dev_xmit_recursion()) {
2285 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2292 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2293 struct sk_buff *skb2;
2295 skb2 = skb_realloc_headroom(skb, hh_len);
2296 if (unlikely(!skb2)) {
2301 skb_set_owner_w(skb2, skb->sk);
2308 struct dst_entry *dst = skb_dst(skb);
2309 struct rtable *rt = container_of(dst, struct rtable, dst);
2311 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2312 } else if (nh->nh_family == AF_INET6) {
2313 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2315 } else if (nh->nh_family == AF_INET) {
2316 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2318 rcu_read_unlock_bh();
2322 if (likely(!IS_ERR(neigh))) {
2325 sock_confirm_neigh(skb, neigh);
2326 dev_xmit_recursion_inc();
2327 ret = neigh_output(neigh, skb, is_v6gw);
2328 dev_xmit_recursion_dec();
2329 rcu_read_unlock_bh();
2332 rcu_read_unlock_bh();
2338 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2339 struct bpf_nh_params *nh)
2341 const struct iphdr *ip4h = ip_hdr(skb);
2342 struct net *net = dev_net(dev);
2343 int err, ret = NET_XMIT_DROP;
2346 struct flowi4 fl4 = {
2347 .flowi4_flags = FLOWI_FLAG_ANYSRC,
2348 .flowi4_mark = skb->mark,
2349 .flowi4_tos = RT_TOS(ip4h->tos),
2350 .flowi4_oif = dev->ifindex,
2351 .flowi4_proto = ip4h->protocol,
2352 .daddr = ip4h->daddr,
2353 .saddr = ip4h->saddr,
2357 rt = ip_route_output_flow(net, &fl4, NULL);
2360 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2365 skb_dst_set(skb, &rt->dst);
2368 err = bpf_out_neigh_v4(net, skb, dev, nh);
2369 if (unlikely(net_xmit_eval(err)))
2370 dev->stats.tx_errors++;
2372 ret = NET_XMIT_SUCCESS;
2375 dev->stats.tx_errors++;
2381 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2382 struct bpf_nh_params *nh)
2385 return NET_XMIT_DROP;
2387 #endif /* CONFIG_INET */
2389 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2390 struct bpf_nh_params *nh)
2392 struct ethhdr *ethh = eth_hdr(skb);
2394 if (unlikely(skb->mac_header >= skb->network_header))
2396 bpf_push_mac_rcsum(skb);
2397 if (is_multicast_ether_addr(ethh->h_dest))
2400 skb_pull(skb, sizeof(*ethh));
2401 skb_unset_mac_header(skb);
2402 skb_reset_network_header(skb);
2404 if (skb->protocol == htons(ETH_P_IP))
2405 return __bpf_redirect_neigh_v4(skb, dev, nh);
2406 else if (skb->protocol == htons(ETH_P_IPV6))
2407 return __bpf_redirect_neigh_v6(skb, dev, nh);
2413 /* Internal, non-exposed redirect flags. */
2415 BPF_F_NEIGH = (1ULL << 1),
2416 BPF_F_PEER = (1ULL << 2),
2417 BPF_F_NEXTHOP = (1ULL << 3),
2418 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2421 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2423 struct net_device *dev;
2424 struct sk_buff *clone;
2427 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2430 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2434 clone = skb_clone(skb, GFP_ATOMIC);
2435 if (unlikely(!clone))
2438 /* For direct write, we need to keep the invariant that the skbs
2439 * we're dealing with need to be uncloned. Should uncloning fail
2440 * here, we need to free the just generated clone to unclone once
2443 ret = bpf_try_make_head_writable(skb);
2444 if (unlikely(ret)) {
2449 return __bpf_redirect(clone, dev, flags);
2452 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2453 .func = bpf_clone_redirect,
2455 .ret_type = RET_INTEGER,
2456 .arg1_type = ARG_PTR_TO_CTX,
2457 .arg2_type = ARG_ANYTHING,
2458 .arg3_type = ARG_ANYTHING,
2461 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2462 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2464 int skb_do_redirect(struct sk_buff *skb)
2466 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2467 struct net *net = dev_net(skb->dev);
2468 struct net_device *dev;
2469 u32 flags = ri->flags;
2471 dev = dev_get_by_index_rcu(net, ri->tgt_index);
2476 if (flags & BPF_F_PEER) {
2477 const struct net_device_ops *ops = dev->netdev_ops;
2479 if (unlikely(!ops->ndo_get_peer_dev ||
2480 !skb_at_tc_ingress(skb)))
2482 dev = ops->ndo_get_peer_dev(dev);
2483 if (unlikely(!dev ||
2484 !is_skb_forwardable(dev, skb) ||
2485 net_eq(net, dev_net(dev))))
2490 return flags & BPF_F_NEIGH ?
2491 __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2493 __bpf_redirect(skb, dev, flags);
2499 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2501 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2503 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2507 ri->tgt_index = ifindex;
2509 return TC_ACT_REDIRECT;
2512 static const struct bpf_func_proto bpf_redirect_proto = {
2513 .func = bpf_redirect,
2515 .ret_type = RET_INTEGER,
2516 .arg1_type = ARG_ANYTHING,
2517 .arg2_type = ARG_ANYTHING,
2520 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2522 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2524 if (unlikely(flags))
2527 ri->flags = BPF_F_PEER;
2528 ri->tgt_index = ifindex;
2530 return TC_ACT_REDIRECT;
2533 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2534 .func = bpf_redirect_peer,
2536 .ret_type = RET_INTEGER,
2537 .arg1_type = ARG_ANYTHING,
2538 .arg2_type = ARG_ANYTHING,
2541 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2542 int, plen, u64, flags)
2544 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2546 if (unlikely((plen && plen < sizeof(*params)) || flags))
2549 ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2550 ri->tgt_index = ifindex;
2552 BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2554 memcpy(&ri->nh, params, sizeof(ri->nh));
2556 return TC_ACT_REDIRECT;
2559 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2560 .func = bpf_redirect_neigh,
2562 .ret_type = RET_INTEGER,
2563 .arg1_type = ARG_ANYTHING,
2564 .arg2_type = ARG_PTR_TO_MEM_OR_NULL,
2565 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
2566 .arg4_type = ARG_ANYTHING,
2569 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2571 msg->apply_bytes = bytes;
2575 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2576 .func = bpf_msg_apply_bytes,
2578 .ret_type = RET_INTEGER,
2579 .arg1_type = ARG_PTR_TO_CTX,
2580 .arg2_type = ARG_ANYTHING,
2583 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2585 msg->cork_bytes = bytes;
2589 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2590 .func = bpf_msg_cork_bytes,
2592 .ret_type = RET_INTEGER,
2593 .arg1_type = ARG_PTR_TO_CTX,
2594 .arg2_type = ARG_ANYTHING,
2597 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2598 u32, end, u64, flags)
2600 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2601 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2602 struct scatterlist *sge;
2603 u8 *raw, *to, *from;
2606 if (unlikely(flags || end <= start))
2609 /* First find the starting scatterlist element */
2613 len = sk_msg_elem(msg, i)->length;
2614 if (start < offset + len)
2616 sk_msg_iter_var_next(i);
2617 } while (i != msg->sg.end);
2619 if (unlikely(start >= offset + len))
2623 /* The start may point into the sg element so we need to also
2624 * account for the headroom.
2626 bytes_sg_total = start - offset + bytes;
2627 if (!test_bit(i, &msg->sg.copy) && bytes_sg_total <= len)
2630 /* At this point we need to linearize multiple scatterlist
2631 * elements or a single shared page. Either way we need to
2632 * copy into a linear buffer exclusively owned by BPF. Then
2633 * place the buffer in the scatterlist and fixup the original
2634 * entries by removing the entries now in the linear buffer
2635 * and shifting the remaining entries. For now we do not try
2636 * to copy partial entries to avoid complexity of running out
2637 * of sg_entry slots. The downside is reading a single byte
2638 * will copy the entire sg entry.
2641 copy += sk_msg_elem(msg, i)->length;
2642 sk_msg_iter_var_next(i);
2643 if (bytes_sg_total <= copy)
2645 } while (i != msg->sg.end);
2648 if (unlikely(bytes_sg_total > copy))
2651 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2653 if (unlikely(!page))
2656 raw = page_address(page);
2659 sge = sk_msg_elem(msg, i);
2660 from = sg_virt(sge);
2664 memcpy(to, from, len);
2667 put_page(sg_page(sge));
2669 sk_msg_iter_var_next(i);
2670 } while (i != last_sge);
2672 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2674 /* To repair sg ring we need to shift entries. If we only
2675 * had a single entry though we can just replace it and
2676 * be done. Otherwise walk the ring and shift the entries.
2678 WARN_ON_ONCE(last_sge == first_sge);
2679 shift = last_sge > first_sge ?
2680 last_sge - first_sge - 1 :
2681 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2686 sk_msg_iter_var_next(i);
2690 if (i + shift >= NR_MSG_FRAG_IDS)
2691 move_from = i + shift - NR_MSG_FRAG_IDS;
2693 move_from = i + shift;
2694 if (move_from == msg->sg.end)
2697 msg->sg.data[i] = msg->sg.data[move_from];
2698 msg->sg.data[move_from].length = 0;
2699 msg->sg.data[move_from].page_link = 0;
2700 msg->sg.data[move_from].offset = 0;
2701 sk_msg_iter_var_next(i);
2704 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2705 msg->sg.end - shift + NR_MSG_FRAG_IDS :
2706 msg->sg.end - shift;
2708 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2709 msg->data_end = msg->data + bytes;
2713 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2714 .func = bpf_msg_pull_data,
2716 .ret_type = RET_INTEGER,
2717 .arg1_type = ARG_PTR_TO_CTX,
2718 .arg2_type = ARG_ANYTHING,
2719 .arg3_type = ARG_ANYTHING,
2720 .arg4_type = ARG_ANYTHING,
2723 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2724 u32, len, u64, flags)
2726 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2727 u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2728 u8 *raw, *to, *from;
2731 if (unlikely(flags))
2734 if (unlikely(len == 0))
2737 /* First find the starting scatterlist element */
2741 l = sk_msg_elem(msg, i)->length;
2743 if (start < offset + l)
2745 sk_msg_iter_var_next(i);
2746 } while (i != msg->sg.end);
2748 if (start >= offset + l)
2751 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2753 /* If no space available will fallback to copy, we need at
2754 * least one scatterlist elem available to push data into
2755 * when start aligns to the beginning of an element or two
2756 * when it falls inside an element. We handle the start equals
2757 * offset case because its the common case for inserting a
2760 if (!space || (space == 1 && start != offset))
2761 copy = msg->sg.data[i].length;
2763 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2764 get_order(copy + len));
2765 if (unlikely(!page))
2771 raw = page_address(page);
2773 psge = sk_msg_elem(msg, i);
2774 front = start - offset;
2775 back = psge->length - front;
2776 from = sg_virt(psge);
2779 memcpy(raw, from, front);
2783 to = raw + front + len;
2785 memcpy(to, from, back);
2788 put_page(sg_page(psge));
2789 } else if (start - offset) {
2790 psge = sk_msg_elem(msg, i);
2791 rsge = sk_msg_elem_cpy(msg, i);
2793 psge->length = start - offset;
2794 rsge.length -= psge->length;
2795 rsge.offset += start;
2797 sk_msg_iter_var_next(i);
2798 sg_unmark_end(psge);
2799 sg_unmark_end(&rsge);
2800 sk_msg_iter_next(msg, end);
2803 /* Slot(s) to place newly allocated data */
2806 /* Shift one or two slots as needed */
2808 sge = sk_msg_elem_cpy(msg, i);
2810 sk_msg_iter_var_next(i);
2811 sg_unmark_end(&sge);
2812 sk_msg_iter_next(msg, end);
2814 nsge = sk_msg_elem_cpy(msg, i);
2816 sk_msg_iter_var_next(i);
2817 nnsge = sk_msg_elem_cpy(msg, i);
2820 while (i != msg->sg.end) {
2821 msg->sg.data[i] = sge;
2823 sk_msg_iter_var_next(i);
2826 nnsge = sk_msg_elem_cpy(msg, i);
2828 nsge = sk_msg_elem_cpy(msg, i);
2833 /* Place newly allocated data buffer */
2834 sk_mem_charge(msg->sk, len);
2835 msg->sg.size += len;
2836 __clear_bit(new, &msg->sg.copy);
2837 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2839 get_page(sg_page(&rsge));
2840 sk_msg_iter_var_next(new);
2841 msg->sg.data[new] = rsge;
2844 sk_msg_compute_data_pointers(msg);
2848 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2849 .func = bpf_msg_push_data,
2851 .ret_type = RET_INTEGER,
2852 .arg1_type = ARG_PTR_TO_CTX,
2853 .arg2_type = ARG_ANYTHING,
2854 .arg3_type = ARG_ANYTHING,
2855 .arg4_type = ARG_ANYTHING,
2858 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2864 sk_msg_iter_var_next(i);
2865 msg->sg.data[prev] = msg->sg.data[i];
2866 } while (i != msg->sg.end);
2868 sk_msg_iter_prev(msg, end);
2871 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2873 struct scatterlist tmp, sge;
2875 sk_msg_iter_next(msg, end);
2876 sge = sk_msg_elem_cpy(msg, i);
2877 sk_msg_iter_var_next(i);
2878 tmp = sk_msg_elem_cpy(msg, i);
2880 while (i != msg->sg.end) {
2881 msg->sg.data[i] = sge;
2882 sk_msg_iter_var_next(i);
2884 tmp = sk_msg_elem_cpy(msg, i);
2888 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2889 u32, len, u64, flags)
2891 u32 i = 0, l = 0, space, offset = 0;
2892 u64 last = start + len;
2895 if (unlikely(flags))
2898 /* First find the starting scatterlist element */
2902 l = sk_msg_elem(msg, i)->length;
2904 if (start < offset + l)
2906 sk_msg_iter_var_next(i);
2907 } while (i != msg->sg.end);
2909 /* Bounds checks: start and pop must be inside message */
2910 if (start >= offset + l || last >= msg->sg.size)
2913 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2916 /* --------------| offset
2917 * -| start |-------- len -------|
2919 * |----- a ----|-------- pop -------|----- b ----|
2920 * |______________________________________________| length
2923 * a: region at front of scatter element to save
2924 * b: region at back of scatter element to save when length > A + pop
2925 * pop: region to pop from element, same as input 'pop' here will be
2926 * decremented below per iteration.
2928 * Two top-level cases to handle when start != offset, first B is non
2929 * zero and second B is zero corresponding to when a pop includes more
2932 * Then if B is non-zero AND there is no space allocate space and
2933 * compact A, B regions into page. If there is space shift ring to
2934 * the rigth free'ing the next element in ring to place B, leaving
2935 * A untouched except to reduce length.
2937 if (start != offset) {
2938 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2940 int b = sge->length - pop - a;
2942 sk_msg_iter_var_next(i);
2944 if (pop < sge->length - a) {
2947 sk_msg_shift_right(msg, i);
2948 nsge = sk_msg_elem(msg, i);
2949 get_page(sg_page(sge));
2952 b, sge->offset + pop + a);
2954 struct page *page, *orig;
2957 page = alloc_pages(__GFP_NOWARN |
2958 __GFP_COMP | GFP_ATOMIC,
2960 if (unlikely(!page))
2964 orig = sg_page(sge);
2965 from = sg_virt(sge);
2966 to = page_address(page);
2967 memcpy(to, from, a);
2968 memcpy(to + a, from + a + pop, b);
2969 sg_set_page(sge, page, a + b, 0);
2973 } else if (pop >= sge->length - a) {
2974 pop -= (sge->length - a);
2979 /* From above the current layout _must_ be as follows,
2984 * |---- pop ---|---------------- b ------------|
2985 * |____________________________________________| length
2987 * Offset and start of the current msg elem are equal because in the
2988 * previous case we handled offset != start and either consumed the
2989 * entire element and advanced to the next element OR pop == 0.
2991 * Two cases to handle here are first pop is less than the length
2992 * leaving some remainder b above. Simply adjust the element's layout
2993 * in this case. Or pop >= length of the element so that b = 0. In this
2994 * case advance to next element decrementing pop.
2997 struct scatterlist *sge = sk_msg_elem(msg, i);
2999 if (pop < sge->length) {
3005 sk_msg_shift_left(msg, i);
3007 sk_msg_iter_var_next(i);
3010 sk_mem_uncharge(msg->sk, len - pop);
3011 msg->sg.size -= (len - pop);
3012 sk_msg_compute_data_pointers(msg);
3016 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
3017 .func = bpf_msg_pop_data,
3019 .ret_type = RET_INTEGER,
3020 .arg1_type = ARG_PTR_TO_CTX,
3021 .arg2_type = ARG_ANYTHING,
3022 .arg3_type = ARG_ANYTHING,
3023 .arg4_type = ARG_ANYTHING,
3026 #ifdef CONFIG_CGROUP_NET_CLASSID
3027 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3029 return __task_get_classid(current);
3032 static const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3033 .func = bpf_get_cgroup_classid_curr,
3035 .ret_type = RET_INTEGER,
3038 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3040 struct sock *sk = skb_to_full_sk(skb);
3042 if (!sk || !sk_fullsock(sk))
3045 return sock_cgroup_classid(&sk->sk_cgrp_data);
3048 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3049 .func = bpf_skb_cgroup_classid,
3051 .ret_type = RET_INTEGER,
3052 .arg1_type = ARG_PTR_TO_CTX,
3056 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3058 return task_get_classid(skb);
3061 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3062 .func = bpf_get_cgroup_classid,
3064 .ret_type = RET_INTEGER,
3065 .arg1_type = ARG_PTR_TO_CTX,
3068 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3070 return dst_tclassid(skb);
3073 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3074 .func = bpf_get_route_realm,
3076 .ret_type = RET_INTEGER,
3077 .arg1_type = ARG_PTR_TO_CTX,
3080 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3082 /* If skb_clear_hash() was called due to mangling, we can
3083 * trigger SW recalculation here. Later access to hash
3084 * can then use the inline skb->hash via context directly
3085 * instead of calling this helper again.
3087 return skb_get_hash(skb);
3090 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3091 .func = bpf_get_hash_recalc,
3093 .ret_type = RET_INTEGER,
3094 .arg1_type = ARG_PTR_TO_CTX,
3097 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3099 /* After all direct packet write, this can be used once for
3100 * triggering a lazy recalc on next skb_get_hash() invocation.
3102 skb_clear_hash(skb);
3106 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3107 .func = bpf_set_hash_invalid,
3109 .ret_type = RET_INTEGER,
3110 .arg1_type = ARG_PTR_TO_CTX,
3113 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3115 /* Set user specified hash as L4(+), so that it gets returned
3116 * on skb_get_hash() call unless BPF prog later on triggers a
3119 __skb_set_sw_hash(skb, hash, true);
3123 static const struct bpf_func_proto bpf_set_hash_proto = {
3124 .func = bpf_set_hash,
3126 .ret_type = RET_INTEGER,
3127 .arg1_type = ARG_PTR_TO_CTX,
3128 .arg2_type = ARG_ANYTHING,
3131 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3136 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3137 vlan_proto != htons(ETH_P_8021AD)))
3138 vlan_proto = htons(ETH_P_8021Q);
3140 bpf_push_mac_rcsum(skb);
3141 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3142 bpf_pull_mac_rcsum(skb);
3144 bpf_compute_data_pointers(skb);
3148 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3149 .func = bpf_skb_vlan_push,
3151 .ret_type = RET_INTEGER,
3152 .arg1_type = ARG_PTR_TO_CTX,
3153 .arg2_type = ARG_ANYTHING,
3154 .arg3_type = ARG_ANYTHING,
3157 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3161 bpf_push_mac_rcsum(skb);
3162 ret = skb_vlan_pop(skb);
3163 bpf_pull_mac_rcsum(skb);
3165 bpf_compute_data_pointers(skb);
3169 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3170 .func = bpf_skb_vlan_pop,
3172 .ret_type = RET_INTEGER,
3173 .arg1_type = ARG_PTR_TO_CTX,
3176 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3178 /* Caller already did skb_cow() with len as headroom,
3179 * so no need to do it here.
3182 memmove(skb->data, skb->data + len, off);
3183 memset(skb->data + off, 0, len);
3185 /* No skb_postpush_rcsum(skb, skb->data + off, len)
3186 * needed here as it does not change the skb->csum
3187 * result for checksum complete when summing over
3193 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3195 /* skb_ensure_writable() is not needed here, as we're
3196 * already working on an uncloned skb.
3198 if (unlikely(!pskb_may_pull(skb, off + len)))
3201 skb_postpull_rcsum(skb, skb->data + off, len);
3202 memmove(skb->data + len, skb->data, off);
3203 __skb_pull(skb, len);
3208 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3210 bool trans_same = skb->transport_header == skb->network_header;
3213 /* There's no need for __skb_push()/__skb_pull() pair to
3214 * get to the start of the mac header as we're guaranteed
3215 * to always start from here under eBPF.
3217 ret = bpf_skb_generic_push(skb, off, len);
3219 skb->mac_header -= len;
3220 skb->network_header -= len;
3222 skb->transport_header = skb->network_header;
3228 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3230 bool trans_same = skb->transport_header == skb->network_header;
3233 /* Same here, __skb_push()/__skb_pull() pair not needed. */
3234 ret = bpf_skb_generic_pop(skb, off, len);
3236 skb->mac_header += len;
3237 skb->network_header += len;
3239 skb->transport_header = skb->network_header;
3245 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3247 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3248 u32 off = skb_mac_header_len(skb);
3251 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
3254 ret = skb_cow(skb, len_diff);
3255 if (unlikely(ret < 0))
3258 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3259 if (unlikely(ret < 0))
3262 if (skb_is_gso(skb)) {
3263 struct skb_shared_info *shinfo = skb_shinfo(skb);
3265 /* SKB_GSO_TCPV4 needs to be changed into
3268 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3269 shinfo->gso_type &= ~SKB_GSO_TCPV4;
3270 shinfo->gso_type |= SKB_GSO_TCPV6;
3273 /* Header must be checked, and gso_segs recomputed. */
3274 shinfo->gso_type |= SKB_GSO_DODGY;
3275 shinfo->gso_segs = 0;
3278 skb->protocol = htons(ETH_P_IPV6);
3279 skb_clear_hash(skb);
3284 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3286 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3287 u32 off = skb_mac_header_len(skb);
3290 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
3293 ret = skb_unclone(skb, GFP_ATOMIC);
3294 if (unlikely(ret < 0))
3297 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3298 if (unlikely(ret < 0))
3301 if (skb_is_gso(skb)) {
3302 struct skb_shared_info *shinfo = skb_shinfo(skb);
3304 /* SKB_GSO_TCPV6 needs to be changed into
3307 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3308 shinfo->gso_type &= ~SKB_GSO_TCPV6;
3309 shinfo->gso_type |= SKB_GSO_TCPV4;
3312 /* Header must be checked, and gso_segs recomputed. */
3313 shinfo->gso_type |= SKB_GSO_DODGY;
3314 shinfo->gso_segs = 0;
3317 skb->protocol = htons(ETH_P_IP);
3318 skb_clear_hash(skb);
3323 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3325 __be16 from_proto = skb->protocol;
3327 if (from_proto == htons(ETH_P_IP) &&
3328 to_proto == htons(ETH_P_IPV6))
3329 return bpf_skb_proto_4_to_6(skb);
3331 if (from_proto == htons(ETH_P_IPV6) &&
3332 to_proto == htons(ETH_P_IP))
3333 return bpf_skb_proto_6_to_4(skb);
3338 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3343 if (unlikely(flags))
3346 /* General idea is that this helper does the basic groundwork
3347 * needed for changing the protocol, and eBPF program fills the
3348 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3349 * and other helpers, rather than passing a raw buffer here.
3351 * The rationale is to keep this minimal and without a need to
3352 * deal with raw packet data. F.e. even if we would pass buffers
3353 * here, the program still needs to call the bpf_lX_csum_replace()
3354 * helpers anyway. Plus, this way we keep also separation of
3355 * concerns, since f.e. bpf_skb_store_bytes() should only take
3358 * Currently, additional options and extension header space are
3359 * not supported, but flags register is reserved so we can adapt
3360 * that. For offloads, we mark packet as dodgy, so that headers
3361 * need to be verified first.
3363 ret = bpf_skb_proto_xlat(skb, proto);
3364 bpf_compute_data_pointers(skb);
3368 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3369 .func = bpf_skb_change_proto,
3371 .ret_type = RET_INTEGER,
3372 .arg1_type = ARG_PTR_TO_CTX,
3373 .arg2_type = ARG_ANYTHING,
3374 .arg3_type = ARG_ANYTHING,
3377 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3379 /* We only allow a restricted subset to be changed for now. */
3380 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3381 !skb_pkt_type_ok(pkt_type)))
3384 skb->pkt_type = pkt_type;
3388 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3389 .func = bpf_skb_change_type,
3391 .ret_type = RET_INTEGER,
3392 .arg1_type = ARG_PTR_TO_CTX,
3393 .arg2_type = ARG_ANYTHING,
3396 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3398 switch (skb->protocol) {
3399 case htons(ETH_P_IP):
3400 return sizeof(struct iphdr);
3401 case htons(ETH_P_IPV6):
3402 return sizeof(struct ipv6hdr);
3408 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3409 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3411 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \
3412 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3413 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3414 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3415 BPF_F_ADJ_ROOM_ENCAP_L2( \
3416 BPF_ADJ_ROOM_ENCAP_L2_MASK))
3418 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3421 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3422 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3423 u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3424 unsigned int gso_type = SKB_GSO_DODGY;
3427 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3428 /* udp gso_size delineates datagrams, only allow if fixed */
3429 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3430 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3434 ret = skb_cow_head(skb, len_diff);
3435 if (unlikely(ret < 0))
3439 if (skb->protocol != htons(ETH_P_IP) &&
3440 skb->protocol != htons(ETH_P_IPV6))
3443 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3444 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3447 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3448 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3451 if (skb->encapsulation)
3454 mac_len = skb->network_header - skb->mac_header;
3455 inner_net = skb->network_header;
3456 if (inner_mac_len > len_diff)
3458 inner_trans = skb->transport_header;
3461 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3462 if (unlikely(ret < 0))
3466 skb->inner_mac_header = inner_net - inner_mac_len;
3467 skb->inner_network_header = inner_net;
3468 skb->inner_transport_header = inner_trans;
3469 skb_set_inner_protocol(skb, skb->protocol);
3471 skb->encapsulation = 1;
3472 skb_set_network_header(skb, mac_len);
3474 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3475 gso_type |= SKB_GSO_UDP_TUNNEL;
3476 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3477 gso_type |= SKB_GSO_GRE;
3478 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3479 gso_type |= SKB_GSO_IPXIP6;
3480 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3481 gso_type |= SKB_GSO_IPXIP4;
3483 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3484 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3485 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3486 sizeof(struct ipv6hdr) :
3487 sizeof(struct iphdr);
3489 skb_set_transport_header(skb, mac_len + nh_len);
3492 /* Match skb->protocol to new outer l3 protocol */
3493 if (skb->protocol == htons(ETH_P_IP) &&
3494 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3495 skb->protocol = htons(ETH_P_IPV6);
3496 else if (skb->protocol == htons(ETH_P_IPV6) &&
3497 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3498 skb->protocol = htons(ETH_P_IP);
3501 if (skb_is_gso(skb)) {
3502 struct skb_shared_info *shinfo = skb_shinfo(skb);
3504 /* Due to header grow, MSS needs to be downgraded. */
3505 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3506 skb_decrease_gso_size(shinfo, len_diff);
3508 /* Header must be checked, and gso_segs recomputed. */
3509 shinfo->gso_type |= gso_type;
3510 shinfo->gso_segs = 0;
3516 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3521 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3522 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3525 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3526 /* udp gso_size delineates datagrams, only allow if fixed */
3527 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3528 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3532 ret = skb_unclone(skb, GFP_ATOMIC);
3533 if (unlikely(ret < 0))
3536 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3537 if (unlikely(ret < 0))
3540 if (skb_is_gso(skb)) {
3541 struct skb_shared_info *shinfo = skb_shinfo(skb);
3543 /* Due to header shrink, MSS can be upgraded. */
3544 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3545 skb_increase_gso_size(shinfo, len_diff);
3547 /* Header must be checked, and gso_segs recomputed. */
3548 shinfo->gso_type |= SKB_GSO_DODGY;
3549 shinfo->gso_segs = 0;
3555 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3557 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3558 u32, mode, u64, flags)
3560 u32 len_diff_abs = abs(len_diff);
3561 bool shrink = len_diff < 0;
3564 if (unlikely(flags || mode))
3566 if (unlikely(len_diff_abs > 0xfffU))
3570 ret = skb_cow(skb, len_diff);
3571 if (unlikely(ret < 0))
3573 __skb_push(skb, len_diff_abs);
3574 memset(skb->data, 0, len_diff_abs);
3576 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3578 __skb_pull(skb, len_diff_abs);
3580 bpf_compute_data_end_sk_skb(skb);
3581 if (tls_sw_has_ctx_rx(skb->sk)) {
3582 struct strp_msg *rxm = strp_msg(skb);
3584 rxm->full_len += len_diff;
3589 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3590 .func = sk_skb_adjust_room,
3592 .ret_type = RET_INTEGER,
3593 .arg1_type = ARG_PTR_TO_CTX,
3594 .arg2_type = ARG_ANYTHING,
3595 .arg3_type = ARG_ANYTHING,
3596 .arg4_type = ARG_ANYTHING,
3599 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3600 u32, mode, u64, flags)
3602 u32 len_cur, len_diff_abs = abs(len_diff);
3603 u32 len_min = bpf_skb_net_base_len(skb);
3604 u32 len_max = BPF_SKB_MAX_LEN;
3605 __be16 proto = skb->protocol;
3606 bool shrink = len_diff < 0;
3610 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3611 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3613 if (unlikely(len_diff_abs > 0xfffU))
3615 if (unlikely(proto != htons(ETH_P_IP) &&
3616 proto != htons(ETH_P_IPV6)))
3619 off = skb_mac_header_len(skb);
3621 case BPF_ADJ_ROOM_NET:
3622 off += bpf_skb_net_base_len(skb);
3624 case BPF_ADJ_ROOM_MAC:
3630 len_cur = skb->len - skb_network_offset(skb);
3631 if ((shrink && (len_diff_abs >= len_cur ||
3632 len_cur - len_diff_abs < len_min)) ||
3633 (!shrink && (skb->len + len_diff_abs > len_max &&
3637 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3638 bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3639 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3640 __skb_reset_checksum_unnecessary(skb);
3642 bpf_compute_data_pointers(skb);
3646 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3647 .func = bpf_skb_adjust_room,
3649 .ret_type = RET_INTEGER,
3650 .arg1_type = ARG_PTR_TO_CTX,
3651 .arg2_type = ARG_ANYTHING,
3652 .arg3_type = ARG_ANYTHING,
3653 .arg4_type = ARG_ANYTHING,
3656 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3658 u32 min_len = skb_network_offset(skb);
3660 if (skb_transport_header_was_set(skb))
3661 min_len = skb_transport_offset(skb);
3662 if (skb->ip_summed == CHECKSUM_PARTIAL)
3663 min_len = skb_checksum_start_offset(skb) +
3664 skb->csum_offset + sizeof(__sum16);
3668 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3670 unsigned int old_len = skb->len;
3673 ret = __skb_grow_rcsum(skb, new_len);
3675 memset(skb->data + old_len, 0, new_len - old_len);
3679 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3681 return __skb_trim_rcsum(skb, new_len);
3684 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3687 u32 max_len = BPF_SKB_MAX_LEN;
3688 u32 min_len = __bpf_skb_min_len(skb);
3691 if (unlikely(flags || new_len > max_len || new_len < min_len))
3693 if (skb->encapsulation)
3696 /* The basic idea of this helper is that it's performing the
3697 * needed work to either grow or trim an skb, and eBPF program
3698 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3699 * bpf_lX_csum_replace() and others rather than passing a raw
3700 * buffer here. This one is a slow path helper and intended
3701 * for replies with control messages.
3703 * Like in bpf_skb_change_proto(), we want to keep this rather
3704 * minimal and without protocol specifics so that we are able
3705 * to separate concerns as in bpf_skb_store_bytes() should only
3706 * be the one responsible for writing buffers.
3708 * It's really expected to be a slow path operation here for
3709 * control message replies, so we're implicitly linearizing,
3710 * uncloning and drop offloads from the skb by this.
3712 ret = __bpf_try_make_writable(skb, skb->len);
3714 if (new_len > skb->len)
3715 ret = bpf_skb_grow_rcsum(skb, new_len);
3716 else if (new_len < skb->len)
3717 ret = bpf_skb_trim_rcsum(skb, new_len);
3718 if (!ret && skb_is_gso(skb))
3724 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3727 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3729 bpf_compute_data_pointers(skb);
3733 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3734 .func = bpf_skb_change_tail,
3736 .ret_type = RET_INTEGER,
3737 .arg1_type = ARG_PTR_TO_CTX,
3738 .arg2_type = ARG_ANYTHING,
3739 .arg3_type = ARG_ANYTHING,
3742 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3745 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3747 bpf_compute_data_end_sk_skb(skb);
3751 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3752 .func = sk_skb_change_tail,
3754 .ret_type = RET_INTEGER,
3755 .arg1_type = ARG_PTR_TO_CTX,
3756 .arg2_type = ARG_ANYTHING,
3757 .arg3_type = ARG_ANYTHING,
3760 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3763 u32 max_len = BPF_SKB_MAX_LEN;
3764 u32 new_len = skb->len + head_room;
3767 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3768 new_len < skb->len))
3771 ret = skb_cow(skb, head_room);
3773 /* Idea for this helper is that we currently only
3774 * allow to expand on mac header. This means that
3775 * skb->protocol network header, etc, stay as is.
3776 * Compared to bpf_skb_change_tail(), we're more
3777 * flexible due to not needing to linearize or
3778 * reset GSO. Intention for this helper is to be
3779 * used by an L3 skb that needs to push mac header
3780 * for redirection into L2 device.
3782 __skb_push(skb, head_room);
3783 memset(skb->data, 0, head_room);
3784 skb_reset_mac_header(skb);
3785 skb_reset_mac_len(skb);
3791 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3794 int ret = __bpf_skb_change_head(skb, head_room, flags);
3796 bpf_compute_data_pointers(skb);
3800 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3801 .func = bpf_skb_change_head,
3803 .ret_type = RET_INTEGER,
3804 .arg1_type = ARG_PTR_TO_CTX,
3805 .arg2_type = ARG_ANYTHING,
3806 .arg3_type = ARG_ANYTHING,
3809 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3812 int ret = __bpf_skb_change_head(skb, head_room, flags);
3814 bpf_compute_data_end_sk_skb(skb);
3818 static const struct bpf_func_proto sk_skb_change_head_proto = {
3819 .func = sk_skb_change_head,
3821 .ret_type = RET_INTEGER,
3822 .arg1_type = ARG_PTR_TO_CTX,
3823 .arg2_type = ARG_ANYTHING,
3824 .arg3_type = ARG_ANYTHING,
3826 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3828 return xdp_data_meta_unsupported(xdp) ? 0 :
3829 xdp->data - xdp->data_meta;
3832 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3834 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3835 unsigned long metalen = xdp_get_metalen(xdp);
3836 void *data_start = xdp_frame_end + metalen;
3837 void *data = xdp->data + offset;
3839 if (unlikely(data < data_start ||
3840 data > xdp->data_end - ETH_HLEN))
3844 memmove(xdp->data_meta + offset,
3845 xdp->data_meta, metalen);
3846 xdp->data_meta += offset;
3852 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3853 .func = bpf_xdp_adjust_head,
3855 .ret_type = RET_INTEGER,
3856 .arg1_type = ARG_PTR_TO_CTX,
3857 .arg2_type = ARG_ANYTHING,
3860 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3862 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
3863 void *data_end = xdp->data_end + offset;
3865 /* Notice that xdp_data_hard_end have reserved some tailroom */
3866 if (unlikely(data_end > data_hard_end))
3869 /* ALL drivers MUST init xdp->frame_sz, chicken check below */
3870 if (unlikely(xdp->frame_sz > PAGE_SIZE)) {
3871 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz);
3875 if (unlikely(data_end < xdp->data + ETH_HLEN))
3878 /* Clear memory area on grow, can contain uninit kernel memory */
3880 memset(xdp->data_end, 0, offset);
3882 xdp->data_end = data_end;
3887 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3888 .func = bpf_xdp_adjust_tail,
3890 .ret_type = RET_INTEGER,
3891 .arg1_type = ARG_PTR_TO_CTX,
3892 .arg2_type = ARG_ANYTHING,
3895 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3897 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3898 void *meta = xdp->data_meta + offset;
3899 unsigned long metalen = xdp->data - meta;
3901 if (xdp_data_meta_unsupported(xdp))
3903 if (unlikely(meta < xdp_frame_end ||
3906 if (unlikely((metalen & (sizeof(__u32) - 1)) ||
3910 xdp->data_meta = meta;
3915 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3916 .func = bpf_xdp_adjust_meta,
3918 .ret_type = RET_INTEGER,
3919 .arg1_type = ARG_PTR_TO_CTX,
3920 .arg2_type = ARG_ANYTHING,
3923 static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd,
3924 struct bpf_map *map, struct xdp_buff *xdp)
3926 switch (map->map_type) {
3927 case BPF_MAP_TYPE_DEVMAP:
3928 case BPF_MAP_TYPE_DEVMAP_HASH:
3929 return dev_map_enqueue(fwd, xdp, dev_rx);
3930 case BPF_MAP_TYPE_CPUMAP:
3931 return cpu_map_enqueue(fwd, xdp, dev_rx);
3932 case BPF_MAP_TYPE_XSKMAP:
3933 return __xsk_map_redirect(fwd, xdp);
3940 void xdp_do_flush(void)
3946 EXPORT_SYMBOL_GPL(xdp_do_flush);
3948 static inline void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index)
3950 switch (map->map_type) {
3951 case BPF_MAP_TYPE_DEVMAP:
3952 return __dev_map_lookup_elem(map, index);
3953 case BPF_MAP_TYPE_DEVMAP_HASH:
3954 return __dev_map_hash_lookup_elem(map, index);
3955 case BPF_MAP_TYPE_CPUMAP:
3956 return __cpu_map_lookup_elem(map, index);
3957 case BPF_MAP_TYPE_XSKMAP:
3958 return __xsk_map_lookup_elem(map, index);
3964 void bpf_clear_redirect_map(struct bpf_map *map)
3966 struct bpf_redirect_info *ri;
3969 for_each_possible_cpu(cpu) {
3970 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3971 /* Avoid polluting remote cacheline due to writes if
3972 * not needed. Once we pass this test, we need the
3973 * cmpxchg() to make sure it hasn't been changed in
3974 * the meantime by remote CPU.
3976 if (unlikely(READ_ONCE(ri->map) == map))
3977 cmpxchg(&ri->map, map, NULL);
3981 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3982 struct bpf_prog *xdp_prog)
3984 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3985 struct bpf_map *map = READ_ONCE(ri->map);
3986 u32 index = ri->tgt_index;
3987 void *fwd = ri->tgt_value;
3991 ri->tgt_value = NULL;
3992 WRITE_ONCE(ri->map, NULL);
3994 if (unlikely(!map)) {
3995 fwd = dev_get_by_index_rcu(dev_net(dev), index);
3996 if (unlikely(!fwd)) {
4001 err = dev_xdp_enqueue(fwd, xdp, dev);
4003 err = __bpf_tx_xdp_map(dev, fwd, map, xdp);
4009 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
4012 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
4015 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4017 static int xdp_do_generic_redirect_map(struct net_device *dev,
4018 struct sk_buff *skb,
4019 struct xdp_buff *xdp,
4020 struct bpf_prog *xdp_prog,
4021 struct bpf_map *map)
4023 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4024 u32 index = ri->tgt_index;
4025 void *fwd = ri->tgt_value;
4029 ri->tgt_value = NULL;
4030 WRITE_ONCE(ri->map, NULL);
4032 if (map->map_type == BPF_MAP_TYPE_DEVMAP ||
4033 map->map_type == BPF_MAP_TYPE_DEVMAP_HASH) {
4034 struct bpf_dtab_netdev *dst = fwd;
4036 err = dev_map_generic_redirect(dst, skb, xdp_prog);
4039 } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) {
4040 struct xdp_sock *xs = fwd;
4042 err = xsk_generic_rcv(xs, xdp);
4047 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
4052 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
4055 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
4059 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4060 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4062 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4063 struct bpf_map *map = READ_ONCE(ri->map);
4064 u32 index = ri->tgt_index;
4065 struct net_device *fwd;
4069 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog,
4072 fwd = dev_get_by_index_rcu(dev_net(dev), index);
4073 if (unlikely(!fwd)) {
4078 err = xdp_ok_fwd_dev(fwd, skb->len);
4083 _trace_xdp_redirect(dev, xdp_prog, index);
4084 generic_xdp_tx(skb, xdp_prog);
4087 _trace_xdp_redirect_err(dev, xdp_prog, index, err);
4091 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4093 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4095 if (unlikely(flags))
4099 ri->tgt_index = ifindex;
4100 ri->tgt_value = NULL;
4101 WRITE_ONCE(ri->map, NULL);
4103 return XDP_REDIRECT;
4106 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4107 .func = bpf_xdp_redirect,
4109 .ret_type = RET_INTEGER,
4110 .arg1_type = ARG_ANYTHING,
4111 .arg2_type = ARG_ANYTHING,
4114 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
4117 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4119 /* Lower bits of the flags are used as return code on lookup failure */
4120 if (unlikely(flags > XDP_TX))
4123 ri->tgt_value = __xdp_map_lookup_elem(map, ifindex);
4124 if (unlikely(!ri->tgt_value)) {
4125 /* If the lookup fails we want to clear out the state in the
4126 * redirect_info struct completely, so that if an eBPF program
4127 * performs multiple lookups, the last one always takes
4130 WRITE_ONCE(ri->map, NULL);
4135 ri->tgt_index = ifindex;
4136 WRITE_ONCE(ri->map, map);
4138 return XDP_REDIRECT;
4141 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4142 .func = bpf_xdp_redirect_map,
4144 .ret_type = RET_INTEGER,
4145 .arg1_type = ARG_CONST_MAP_PTR,
4146 .arg2_type = ARG_ANYTHING,
4147 .arg3_type = ARG_ANYTHING,
4150 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4151 unsigned long off, unsigned long len)
4153 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4157 if (ptr != dst_buff)
4158 memcpy(dst_buff, ptr, len);
4163 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4164 u64, flags, void *, meta, u64, meta_size)
4166 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4168 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4170 if (unlikely(!skb || skb_size > skb->len))
4173 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4177 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4178 .func = bpf_skb_event_output,
4180 .ret_type = RET_INTEGER,
4181 .arg1_type = ARG_PTR_TO_CTX,
4182 .arg2_type = ARG_CONST_MAP_PTR,
4183 .arg3_type = ARG_ANYTHING,
4184 .arg4_type = ARG_PTR_TO_MEM,
4185 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4188 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4190 const struct bpf_func_proto bpf_skb_output_proto = {
4191 .func = bpf_skb_event_output,
4193 .ret_type = RET_INTEGER,
4194 .arg1_type = ARG_PTR_TO_BTF_ID,
4195 .arg1_btf_id = &bpf_skb_output_btf_ids[0],
4196 .arg2_type = ARG_CONST_MAP_PTR,
4197 .arg3_type = ARG_ANYTHING,
4198 .arg4_type = ARG_PTR_TO_MEM,
4199 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4202 static unsigned short bpf_tunnel_key_af(u64 flags)
4204 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4207 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4208 u32, size, u64, flags)
4210 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4211 u8 compat[sizeof(struct bpf_tunnel_key)];
4215 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
4219 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4223 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4226 case offsetof(struct bpf_tunnel_key, tunnel_label):
4227 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4229 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4230 /* Fixup deprecated structure layouts here, so we have
4231 * a common path later on.
4233 if (ip_tunnel_info_af(info) != AF_INET)
4236 to = (struct bpf_tunnel_key *)compat;
4243 to->tunnel_id = be64_to_cpu(info->key.tun_id);
4244 to->tunnel_tos = info->key.tos;
4245 to->tunnel_ttl = info->key.ttl;
4248 if (flags & BPF_F_TUNINFO_IPV6) {
4249 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4250 sizeof(to->remote_ipv6));
4251 to->tunnel_label = be32_to_cpu(info->key.label);
4253 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4254 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4255 to->tunnel_label = 0;
4258 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4259 memcpy(to_orig, to, size);
4263 memset(to_orig, 0, size);
4267 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4268 .func = bpf_skb_get_tunnel_key,
4270 .ret_type = RET_INTEGER,
4271 .arg1_type = ARG_PTR_TO_CTX,
4272 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4273 .arg3_type = ARG_CONST_SIZE,
4274 .arg4_type = ARG_ANYTHING,
4277 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4279 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4282 if (unlikely(!info ||
4283 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4287 if (unlikely(size < info->options_len)) {
4292 ip_tunnel_info_opts_get(to, info);
4293 if (size > info->options_len)
4294 memset(to + info->options_len, 0, size - info->options_len);
4296 return info->options_len;
4298 memset(to, 0, size);
4302 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4303 .func = bpf_skb_get_tunnel_opt,
4305 .ret_type = RET_INTEGER,
4306 .arg1_type = ARG_PTR_TO_CTX,
4307 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4308 .arg3_type = ARG_CONST_SIZE,
4311 static struct metadata_dst __percpu *md_dst;
4313 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4314 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4316 struct metadata_dst *md = this_cpu_ptr(md_dst);
4317 u8 compat[sizeof(struct bpf_tunnel_key)];
4318 struct ip_tunnel_info *info;
4320 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4321 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
4323 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4325 case offsetof(struct bpf_tunnel_key, tunnel_label):
4326 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4327 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4328 /* Fixup deprecated structure layouts here, so we have
4329 * a common path later on.
4331 memcpy(compat, from, size);
4332 memset(compat + size, 0, sizeof(compat) - size);
4333 from = (const struct bpf_tunnel_key *) compat;
4339 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4344 dst_hold((struct dst_entry *) md);
4345 skb_dst_set(skb, (struct dst_entry *) md);
4347 info = &md->u.tun_info;
4348 memset(info, 0, sizeof(*info));
4349 info->mode = IP_TUNNEL_INFO_TX;
4351 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4352 if (flags & BPF_F_DONT_FRAGMENT)
4353 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4354 if (flags & BPF_F_ZERO_CSUM_TX)
4355 info->key.tun_flags &= ~TUNNEL_CSUM;
4356 if (flags & BPF_F_SEQ_NUMBER)
4357 info->key.tun_flags |= TUNNEL_SEQ;
4359 info->key.tun_id = cpu_to_be64(from->tunnel_id);
4360 info->key.tos = from->tunnel_tos;
4361 info->key.ttl = from->tunnel_ttl;
4363 if (flags & BPF_F_TUNINFO_IPV6) {
4364 info->mode |= IP_TUNNEL_INFO_IPV6;
4365 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4366 sizeof(from->remote_ipv6));
4367 info->key.label = cpu_to_be32(from->tunnel_label) &
4368 IPV6_FLOWLABEL_MASK;
4370 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4376 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4377 .func = bpf_skb_set_tunnel_key,
4379 .ret_type = RET_INTEGER,
4380 .arg1_type = ARG_PTR_TO_CTX,
4381 .arg2_type = ARG_PTR_TO_MEM,
4382 .arg3_type = ARG_CONST_SIZE,
4383 .arg4_type = ARG_ANYTHING,
4386 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4387 const u8 *, from, u32, size)
4389 struct ip_tunnel_info *info = skb_tunnel_info(skb);
4390 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4392 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4394 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4397 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4402 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4403 .func = bpf_skb_set_tunnel_opt,
4405 .ret_type = RET_INTEGER,
4406 .arg1_type = ARG_PTR_TO_CTX,
4407 .arg2_type = ARG_PTR_TO_MEM,
4408 .arg3_type = ARG_CONST_SIZE,
4411 static const struct bpf_func_proto *
4412 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4415 struct metadata_dst __percpu *tmp;
4417 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4422 if (cmpxchg(&md_dst, NULL, tmp))
4423 metadata_dst_free_percpu(tmp);
4427 case BPF_FUNC_skb_set_tunnel_key:
4428 return &bpf_skb_set_tunnel_key_proto;
4429 case BPF_FUNC_skb_set_tunnel_opt:
4430 return &bpf_skb_set_tunnel_opt_proto;
4436 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4439 struct bpf_array *array = container_of(map, struct bpf_array, map);
4440 struct cgroup *cgrp;
4443 sk = skb_to_full_sk(skb);
4444 if (!sk || !sk_fullsock(sk))
4446 if (unlikely(idx >= array->map.max_entries))
4449 cgrp = READ_ONCE(array->ptrs[idx]);
4450 if (unlikely(!cgrp))
4453 return sk_under_cgroup_hierarchy(sk, cgrp);
4456 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4457 .func = bpf_skb_under_cgroup,
4459 .ret_type = RET_INTEGER,
4460 .arg1_type = ARG_PTR_TO_CTX,
4461 .arg2_type = ARG_CONST_MAP_PTR,
4462 .arg3_type = ARG_ANYTHING,
4465 #ifdef CONFIG_SOCK_CGROUP_DATA
4466 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4468 struct cgroup *cgrp;
4470 sk = sk_to_full_sk(sk);
4471 if (!sk || !sk_fullsock(sk))
4474 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4475 return cgroup_id(cgrp);
4478 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4480 return __bpf_sk_cgroup_id(skb->sk);
4483 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4484 .func = bpf_skb_cgroup_id,
4486 .ret_type = RET_INTEGER,
4487 .arg1_type = ARG_PTR_TO_CTX,
4490 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4493 struct cgroup *ancestor;
4494 struct cgroup *cgrp;
4496 sk = sk_to_full_sk(sk);
4497 if (!sk || !sk_fullsock(sk))
4500 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4501 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4505 return cgroup_id(ancestor);
4508 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4511 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4514 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4515 .func = bpf_skb_ancestor_cgroup_id,
4517 .ret_type = RET_INTEGER,
4518 .arg1_type = ARG_PTR_TO_CTX,
4519 .arg2_type = ARG_ANYTHING,
4522 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4524 return __bpf_sk_cgroup_id(sk);
4527 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4528 .func = bpf_sk_cgroup_id,
4530 .ret_type = RET_INTEGER,
4531 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4534 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4536 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4539 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4540 .func = bpf_sk_ancestor_cgroup_id,
4542 .ret_type = RET_INTEGER,
4543 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4544 .arg2_type = ARG_ANYTHING,
4548 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4549 unsigned long off, unsigned long len)
4551 memcpy(dst_buff, src_buff + off, len);
4555 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4556 u64, flags, void *, meta, u64, meta_size)
4558 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4560 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4562 if (unlikely(!xdp ||
4563 xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4566 return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4567 xdp_size, bpf_xdp_copy);
4570 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4571 .func = bpf_xdp_event_output,
4573 .ret_type = RET_INTEGER,
4574 .arg1_type = ARG_PTR_TO_CTX,
4575 .arg2_type = ARG_CONST_MAP_PTR,
4576 .arg3_type = ARG_ANYTHING,
4577 .arg4_type = ARG_PTR_TO_MEM,
4578 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4581 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4583 const struct bpf_func_proto bpf_xdp_output_proto = {
4584 .func = bpf_xdp_event_output,
4586 .ret_type = RET_INTEGER,
4587 .arg1_type = ARG_PTR_TO_BTF_ID,
4588 .arg1_btf_id = &bpf_xdp_output_btf_ids[0],
4589 .arg2_type = ARG_CONST_MAP_PTR,
4590 .arg3_type = ARG_ANYTHING,
4591 .arg4_type = ARG_PTR_TO_MEM,
4592 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4595 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4597 return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4600 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4601 .func = bpf_get_socket_cookie,
4603 .ret_type = RET_INTEGER,
4604 .arg1_type = ARG_PTR_TO_CTX,
4607 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4609 return __sock_gen_cookie(ctx->sk);
4612 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4613 .func = bpf_get_socket_cookie_sock_addr,
4615 .ret_type = RET_INTEGER,
4616 .arg1_type = ARG_PTR_TO_CTX,
4619 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4621 return __sock_gen_cookie(ctx);
4624 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4625 .func = bpf_get_socket_cookie_sock,
4627 .ret_type = RET_INTEGER,
4628 .arg1_type = ARG_PTR_TO_CTX,
4631 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4633 return __sock_gen_cookie(ctx->sk);
4636 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4637 .func = bpf_get_socket_cookie_sock_ops,
4639 .ret_type = RET_INTEGER,
4640 .arg1_type = ARG_PTR_TO_CTX,
4643 static u64 __bpf_get_netns_cookie(struct sock *sk)
4645 #ifdef CONFIG_NET_NS
4646 return __net_gen_cookie(sk ? sk->sk_net.net : &init_net);
4652 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
4654 return __bpf_get_netns_cookie(ctx);
4657 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
4658 .func = bpf_get_netns_cookie_sock,
4660 .ret_type = RET_INTEGER,
4661 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4664 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4666 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4669 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
4670 .func = bpf_get_netns_cookie_sock_addr,
4672 .ret_type = RET_INTEGER,
4673 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4676 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4678 struct sock *sk = sk_to_full_sk(skb->sk);
4681 if (!sk || !sk_fullsock(sk))
4683 kuid = sock_net_uid(sock_net(sk), sk);
4684 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4687 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4688 .func = bpf_get_socket_uid,
4690 .ret_type = RET_INTEGER,
4691 .arg1_type = ARG_PTR_TO_CTX,
4694 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
4695 char *optval, int optlen)
4697 char devname[IFNAMSIZ];
4703 if (!sk_fullsock(sk))
4706 sock_owned_by_me(sk);
4708 if (level == SOL_SOCKET) {
4709 if (optlen != sizeof(int) && optname != SO_BINDTODEVICE)
4711 val = *((int *)optval);
4712 valbool = val ? 1 : 0;
4714 /* Only some socketops are supported */
4717 val = min_t(u32, val, READ_ONCE(sysctl_rmem_max));
4718 val = min_t(int, val, INT_MAX / 2);
4719 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4720 WRITE_ONCE(sk->sk_rcvbuf,
4721 max_t(int, val * 2, SOCK_MIN_RCVBUF));
4724 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
4725 val = min_t(int, val, INT_MAX / 2);
4726 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4727 WRITE_ONCE(sk->sk_sndbuf,
4728 max_t(int, val * 2, SOCK_MIN_SNDBUF));
4730 case SO_MAX_PACING_RATE: /* 32bit version */
4732 cmpxchg(&sk->sk_pacing_status,
4735 sk->sk_max_pacing_rate = (val == ~0U) ?
4736 ~0UL : (unsigned int)val;
4737 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4738 sk->sk_max_pacing_rate);
4741 sk->sk_priority = val;
4746 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
4749 if (sk->sk_mark != val) {
4754 case SO_BINDTODEVICE:
4755 optlen = min_t(long, optlen, IFNAMSIZ - 1);
4756 strncpy(devname, optval, optlen);
4757 devname[optlen] = 0;
4760 if (devname[0] != '\0') {
4761 struct net_device *dev;
4766 dev = dev_get_by_name(net, devname);
4769 ifindex = dev->ifindex;
4772 ret = sock_bindtoindex(sk, ifindex, false);
4775 if (sk->sk_prot->keepalive)
4776 sk->sk_prot->keepalive(sk, valbool);
4777 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
4783 } else if (level == SOL_IP) {
4784 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4787 val = *((int *)optval);
4788 /* Only some options are supported */
4791 if (val < -1 || val > 0xff) {
4794 struct inet_sock *inet = inet_sk(sk);
4804 #if IS_ENABLED(CONFIG_IPV6)
4805 } else if (level == SOL_IPV6) {
4806 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4809 val = *((int *)optval);
4810 /* Only some options are supported */
4813 if (val < -1 || val > 0xff) {
4816 struct ipv6_pinfo *np = inet6_sk(sk);
4827 } else if (level == SOL_TCP &&
4828 sk->sk_prot->setsockopt == tcp_setsockopt) {
4829 if (optname == TCP_CONGESTION) {
4830 char name[TCP_CA_NAME_MAX];
4832 strncpy(name, optval, min_t(long, optlen,
4833 TCP_CA_NAME_MAX-1));
4834 name[TCP_CA_NAME_MAX-1] = 0;
4835 ret = tcp_set_congestion_control(sk, name, false, true);
4837 struct inet_connection_sock *icsk = inet_csk(sk);
4838 struct tcp_sock *tp = tcp_sk(sk);
4839 unsigned long timeout;
4841 if (optlen != sizeof(int))
4844 val = *((int *)optval);
4845 /* Only some options are supported */
4848 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4853 case TCP_BPF_SNDCWND_CLAMP:
4857 tp->snd_cwnd_clamp = val;
4858 tp->snd_ssthresh = val;
4861 case TCP_BPF_DELACK_MAX:
4862 timeout = usecs_to_jiffies(val);
4863 if (timeout > TCP_DELACK_MAX ||
4864 timeout < TCP_TIMEOUT_MIN)
4866 inet_csk(sk)->icsk_delack_max = timeout;
4868 case TCP_BPF_RTO_MIN:
4869 timeout = usecs_to_jiffies(val);
4870 if (timeout > TCP_RTO_MIN ||
4871 timeout < TCP_TIMEOUT_MIN)
4873 inet_csk(sk)->icsk_rto_min = timeout;
4876 if (val < 0 || val > 1)
4882 ret = tcp_sock_set_keepidle_locked(sk, val);
4885 if (val < 1 || val > MAX_TCP_KEEPINTVL)
4888 tp->keepalive_intvl = val * HZ;
4891 if (val < 1 || val > MAX_TCP_KEEPCNT)
4894 tp->keepalive_probes = val;
4897 if (val < 1 || val > MAX_TCP_SYNCNT)
4900 icsk->icsk_syn_retries = val;
4902 case TCP_USER_TIMEOUT:
4906 icsk->icsk_user_timeout = val;
4908 case TCP_NOTSENT_LOWAT:
4909 tp->notsent_lowat = val;
4910 sk->sk_write_space(sk);
4923 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
4924 char *optval, int optlen)
4926 if (!sk_fullsock(sk))
4929 sock_owned_by_me(sk);
4932 if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4933 struct inet_connection_sock *icsk;
4934 struct tcp_sock *tp;
4937 case TCP_CONGESTION:
4938 icsk = inet_csk(sk);
4940 if (!icsk->icsk_ca_ops || optlen <= 1)
4942 strncpy(optval, icsk->icsk_ca_ops->name, optlen);
4943 optval[optlen - 1] = 0;
4948 if (optlen <= 0 || !tp->saved_syn ||
4949 optlen > tcp_saved_syn_len(tp->saved_syn))
4951 memcpy(optval, tp->saved_syn->data, optlen);
4956 } else if (level == SOL_IP) {
4957 struct inet_sock *inet = inet_sk(sk);
4959 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4962 /* Only some options are supported */
4965 *((int *)optval) = (int)inet->tos;
4970 #if IS_ENABLED(CONFIG_IPV6)
4971 } else if (level == SOL_IPV6) {
4972 struct ipv6_pinfo *np = inet6_sk(sk);
4974 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4977 /* Only some options are supported */
4980 *((int *)optval) = (int)np->tclass;
4992 memset(optval, 0, optlen);
4996 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
4997 int, level, int, optname, char *, optval, int, optlen)
4999 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5002 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5003 .func = bpf_sock_addr_setsockopt,
5005 .ret_type = RET_INTEGER,
5006 .arg1_type = ARG_PTR_TO_CTX,
5007 .arg2_type = ARG_ANYTHING,
5008 .arg3_type = ARG_ANYTHING,
5009 .arg4_type = ARG_PTR_TO_MEM,
5010 .arg5_type = ARG_CONST_SIZE,
5013 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5014 int, level, int, optname, char *, optval, int, optlen)
5016 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5019 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5020 .func = bpf_sock_addr_getsockopt,
5022 .ret_type = RET_INTEGER,
5023 .arg1_type = ARG_PTR_TO_CTX,
5024 .arg2_type = ARG_ANYTHING,
5025 .arg3_type = ARG_ANYTHING,
5026 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5027 .arg5_type = ARG_CONST_SIZE,
5030 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5031 int, level, int, optname, char *, optval, int, optlen)
5033 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5036 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5037 .func = bpf_sock_ops_setsockopt,
5039 .ret_type = RET_INTEGER,
5040 .arg1_type = ARG_PTR_TO_CTX,
5041 .arg2_type = ARG_ANYTHING,
5042 .arg3_type = ARG_ANYTHING,
5043 .arg4_type = ARG_PTR_TO_MEM,
5044 .arg5_type = ARG_CONST_SIZE,
5047 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5048 int optname, const u8 **start)
5050 struct sk_buff *syn_skb = bpf_sock->syn_skb;
5051 const u8 *hdr_start;
5055 /* sk is a request_sock here */
5057 if (optname == TCP_BPF_SYN) {
5058 hdr_start = syn_skb->data;
5059 ret = tcp_hdrlen(syn_skb);
5060 } else if (optname == TCP_BPF_SYN_IP) {
5061 hdr_start = skb_network_header(syn_skb);
5062 ret = skb_network_header_len(syn_skb) +
5063 tcp_hdrlen(syn_skb);
5065 /* optname == TCP_BPF_SYN_MAC */
5066 hdr_start = skb_mac_header(syn_skb);
5067 ret = skb_mac_header_len(syn_skb) +
5068 skb_network_header_len(syn_skb) +
5069 tcp_hdrlen(syn_skb);
5072 struct sock *sk = bpf_sock->sk;
5073 struct saved_syn *saved_syn;
5075 if (sk->sk_state == TCP_NEW_SYN_RECV)
5076 /* synack retransmit. bpf_sock->syn_skb will
5077 * not be available. It has to resort to
5078 * saved_syn (if it is saved).
5080 saved_syn = inet_reqsk(sk)->saved_syn;
5082 saved_syn = tcp_sk(sk)->saved_syn;
5087 if (optname == TCP_BPF_SYN) {
5088 hdr_start = saved_syn->data +
5089 saved_syn->mac_hdrlen +
5090 saved_syn->network_hdrlen;
5091 ret = saved_syn->tcp_hdrlen;
5092 } else if (optname == TCP_BPF_SYN_IP) {
5093 hdr_start = saved_syn->data +
5094 saved_syn->mac_hdrlen;
5095 ret = saved_syn->network_hdrlen +
5096 saved_syn->tcp_hdrlen;
5098 /* optname == TCP_BPF_SYN_MAC */
5100 /* TCP_SAVE_SYN may not have saved the mac hdr */
5101 if (!saved_syn->mac_hdrlen)
5104 hdr_start = saved_syn->data;
5105 ret = saved_syn->mac_hdrlen +
5106 saved_syn->network_hdrlen +
5107 saved_syn->tcp_hdrlen;
5115 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5116 int, level, int, optname, char *, optval, int, optlen)
5118 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5119 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5120 int ret, copy_len = 0;
5123 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5126 if (optlen < copy_len) {
5131 memcpy(optval, start, copy_len);
5134 /* Zero out unused buffer at the end */
5135 memset(optval + copy_len, 0, optlen - copy_len);
5140 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5143 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5144 .func = bpf_sock_ops_getsockopt,
5146 .ret_type = RET_INTEGER,
5147 .arg1_type = ARG_PTR_TO_CTX,
5148 .arg2_type = ARG_ANYTHING,
5149 .arg3_type = ARG_ANYTHING,
5150 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5151 .arg5_type = ARG_CONST_SIZE,
5154 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5157 struct sock *sk = bpf_sock->sk;
5158 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5160 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5163 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5165 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5168 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5169 .func = bpf_sock_ops_cb_flags_set,
5171 .ret_type = RET_INTEGER,
5172 .arg1_type = ARG_PTR_TO_CTX,
5173 .arg2_type = ARG_ANYTHING,
5176 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5177 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5179 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5183 struct sock *sk = ctx->sk;
5184 u32 flags = BIND_FROM_BPF;
5188 if (addr_len < offsetofend(struct sockaddr, sa_family))
5190 if (addr->sa_family == AF_INET) {
5191 if (addr_len < sizeof(struct sockaddr_in))
5193 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5194 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5195 return __inet_bind(sk, addr, addr_len, flags);
5196 #if IS_ENABLED(CONFIG_IPV6)
5197 } else if (addr->sa_family == AF_INET6) {
5198 if (addr_len < SIN6_LEN_RFC2133)
5200 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5201 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5202 /* ipv6_bpf_stub cannot be NULL, since it's called from
5203 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5205 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5206 #endif /* CONFIG_IPV6 */
5208 #endif /* CONFIG_INET */
5210 return -EAFNOSUPPORT;
5213 static const struct bpf_func_proto bpf_bind_proto = {
5216 .ret_type = RET_INTEGER,
5217 .arg1_type = ARG_PTR_TO_CTX,
5218 .arg2_type = ARG_PTR_TO_MEM,
5219 .arg3_type = ARG_CONST_SIZE,
5223 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5224 struct bpf_xfrm_state *, to, u32, size, u64, flags)
5226 const struct sec_path *sp = skb_sec_path(skb);
5227 const struct xfrm_state *x;
5229 if (!sp || unlikely(index >= sp->len || flags))
5232 x = sp->xvec[index];
5234 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5237 to->reqid = x->props.reqid;
5238 to->spi = x->id.spi;
5239 to->family = x->props.family;
5242 if (to->family == AF_INET6) {
5243 memcpy(to->remote_ipv6, x->props.saddr.a6,
5244 sizeof(to->remote_ipv6));
5246 to->remote_ipv4 = x->props.saddr.a4;
5247 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5252 memset(to, 0, size);
5256 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5257 .func = bpf_skb_get_xfrm_state,
5259 .ret_type = RET_INTEGER,
5260 .arg1_type = ARG_PTR_TO_CTX,
5261 .arg2_type = ARG_ANYTHING,
5262 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
5263 .arg4_type = ARG_CONST_SIZE,
5264 .arg5_type = ARG_ANYTHING,
5268 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5269 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
5270 const struct neighbour *neigh,
5271 const struct net_device *dev)
5273 memcpy(params->dmac, neigh->ha, ETH_ALEN);
5274 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5275 params->h_vlan_TCI = 0;
5276 params->h_vlan_proto = 0;
5282 #if IS_ENABLED(CONFIG_INET)
5283 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5284 u32 flags, bool check_mtu)
5286 struct fib_nh_common *nhc;
5287 struct in_device *in_dev;
5288 struct neighbour *neigh;
5289 struct net_device *dev;
5290 struct fib_result res;
5295 dev = dev_get_by_index_rcu(net, params->ifindex);
5299 /* verify forwarding is enabled on this interface */
5300 in_dev = __in_dev_get_rcu(dev);
5301 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5302 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5304 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5306 fl4.flowi4_oif = params->ifindex;
5308 fl4.flowi4_iif = params->ifindex;
5311 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5312 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5313 fl4.flowi4_flags = 0;
5315 fl4.flowi4_proto = params->l4_protocol;
5316 fl4.daddr = params->ipv4_dst;
5317 fl4.saddr = params->ipv4_src;
5318 fl4.fl4_sport = params->sport;
5319 fl4.fl4_dport = params->dport;
5320 fl4.flowi4_multipath_hash = 0;
5322 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5323 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5324 struct fib_table *tb;
5326 tb = fib_get_table(net, tbid);
5328 return BPF_FIB_LKUP_RET_NOT_FWDED;
5330 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5332 fl4.flowi4_mark = 0;
5333 fl4.flowi4_secid = 0;
5334 fl4.flowi4_tun_key.tun_id = 0;
5335 fl4.flowi4_uid = sock_net_uid(net, NULL);
5337 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5341 /* map fib lookup errors to RTN_ type */
5343 return BPF_FIB_LKUP_RET_BLACKHOLE;
5344 if (err == -EHOSTUNREACH)
5345 return BPF_FIB_LKUP_RET_UNREACHABLE;
5347 return BPF_FIB_LKUP_RET_PROHIBIT;
5349 return BPF_FIB_LKUP_RET_NOT_FWDED;
5352 if (res.type != RTN_UNICAST)
5353 return BPF_FIB_LKUP_RET_NOT_FWDED;
5355 if (fib_info_num_path(res.fi) > 1)
5356 fib_select_path(net, &res, &fl4, NULL);
5359 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5360 if (params->tot_len > mtu)
5361 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5366 /* do not handle lwt encaps right now */
5367 if (nhc->nhc_lwtstate)
5368 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5372 params->rt_metric = res.fi->fib_priority;
5373 params->ifindex = dev->ifindex;
5375 /* xdp and cls_bpf programs are run in RCU-bh so
5376 * rcu_read_lock_bh is not needed here
5378 if (likely(nhc->nhc_gw_family != AF_INET6)) {
5379 if (nhc->nhc_gw_family)
5380 params->ipv4_dst = nhc->nhc_gw.ipv4;
5382 neigh = __ipv4_neigh_lookup_noref(dev,
5383 (__force u32)params->ipv4_dst);
5385 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5387 params->family = AF_INET6;
5388 *dst = nhc->nhc_gw.ipv6;
5389 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5393 return BPF_FIB_LKUP_RET_NO_NEIGH;
5395 return bpf_fib_set_fwd_params(params, neigh, dev);
5399 #if IS_ENABLED(CONFIG_IPV6)
5400 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5401 u32 flags, bool check_mtu)
5403 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5404 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5405 struct fib6_result res = {};
5406 struct neighbour *neigh;
5407 struct net_device *dev;
5408 struct inet6_dev *idev;
5414 /* link local addresses are never forwarded */
5415 if (rt6_need_strict(dst) || rt6_need_strict(src))
5416 return BPF_FIB_LKUP_RET_NOT_FWDED;
5418 dev = dev_get_by_index_rcu(net, params->ifindex);
5422 idev = __in6_dev_get_safely(dev);
5423 if (unlikely(!idev || !idev->cnf.forwarding))
5424 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5426 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5428 oif = fl6.flowi6_oif = params->ifindex;
5430 oif = fl6.flowi6_iif = params->ifindex;
5432 strict = RT6_LOOKUP_F_HAS_SADDR;
5434 fl6.flowlabel = params->flowinfo;
5435 fl6.flowi6_scope = 0;
5436 fl6.flowi6_flags = 0;
5439 fl6.flowi6_proto = params->l4_protocol;
5442 fl6.fl6_sport = params->sport;
5443 fl6.fl6_dport = params->dport;
5445 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5446 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5447 struct fib6_table *tb;
5449 tb = ipv6_stub->fib6_get_table(net, tbid);
5451 return BPF_FIB_LKUP_RET_NOT_FWDED;
5453 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5456 fl6.flowi6_mark = 0;
5457 fl6.flowi6_secid = 0;
5458 fl6.flowi6_tun_key.tun_id = 0;
5459 fl6.flowi6_uid = sock_net_uid(net, NULL);
5461 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5464 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5465 res.f6i == net->ipv6.fib6_null_entry))
5466 return BPF_FIB_LKUP_RET_NOT_FWDED;
5468 switch (res.fib6_type) {
5469 /* only unicast is forwarded */
5473 return BPF_FIB_LKUP_RET_BLACKHOLE;
5474 case RTN_UNREACHABLE:
5475 return BPF_FIB_LKUP_RET_UNREACHABLE;
5477 return BPF_FIB_LKUP_RET_PROHIBIT;
5479 return BPF_FIB_LKUP_RET_NOT_FWDED;
5482 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5483 fl6.flowi6_oif != 0, NULL, strict);
5486 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5487 if (params->tot_len > mtu)
5488 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5491 if (res.nh->fib_nh_lws)
5492 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5494 if (res.nh->fib_nh_gw_family)
5495 *dst = res.nh->fib_nh_gw6;
5497 dev = res.nh->fib_nh_dev;
5498 params->rt_metric = res.f6i->fib6_metric;
5499 params->ifindex = dev->ifindex;
5501 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
5504 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5506 return BPF_FIB_LKUP_RET_NO_NEIGH;
5508 return bpf_fib_set_fwd_params(params, neigh, dev);
5512 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
5513 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5515 if (plen < sizeof(*params))
5518 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5521 switch (params->family) {
5522 #if IS_ENABLED(CONFIG_INET)
5524 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
5527 #if IS_ENABLED(CONFIG_IPV6)
5529 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
5533 return -EAFNOSUPPORT;
5536 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
5537 .func = bpf_xdp_fib_lookup,
5539 .ret_type = RET_INTEGER,
5540 .arg1_type = ARG_PTR_TO_CTX,
5541 .arg2_type = ARG_PTR_TO_MEM,
5542 .arg3_type = ARG_CONST_SIZE,
5543 .arg4_type = ARG_ANYTHING,
5546 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
5547 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5549 struct net *net = dev_net(skb->dev);
5550 int rc = -EAFNOSUPPORT;
5551 bool check_mtu = false;
5553 if (plen < sizeof(*params))
5556 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5559 if (params->tot_len)
5562 switch (params->family) {
5563 #if IS_ENABLED(CONFIG_INET)
5565 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
5568 #if IS_ENABLED(CONFIG_IPV6)
5570 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
5575 if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
5576 struct net_device *dev;
5578 /* When tot_len isn't provided by user, check skb
5579 * against MTU of FIB lookup resulting net_device
5581 dev = dev_get_by_index_rcu(net, params->ifindex);
5582 if (!is_skb_forwardable(dev, skb))
5583 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
5589 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
5590 .func = bpf_skb_fib_lookup,
5592 .ret_type = RET_INTEGER,
5593 .arg1_type = ARG_PTR_TO_CTX,
5594 .arg2_type = ARG_PTR_TO_MEM,
5595 .arg3_type = ARG_CONST_SIZE,
5596 .arg4_type = ARG_ANYTHING,
5599 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5600 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
5603 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
5605 if (!seg6_validate_srh(srh, len, false))
5609 case BPF_LWT_ENCAP_SEG6_INLINE:
5610 if (skb->protocol != htons(ETH_P_IPV6))
5613 err = seg6_do_srh_inline(skb, srh);
5615 case BPF_LWT_ENCAP_SEG6:
5616 skb_reset_inner_headers(skb);
5617 skb->encapsulation = 1;
5618 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
5624 bpf_compute_data_pointers(skb);
5628 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
5630 return seg6_lookup_nexthop(skb, NULL, 0);
5632 #endif /* CONFIG_IPV6_SEG6_BPF */
5634 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5635 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
5638 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
5642 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
5646 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5647 case BPF_LWT_ENCAP_SEG6:
5648 case BPF_LWT_ENCAP_SEG6_INLINE:
5649 return bpf_push_seg6_encap(skb, type, hdr, len);
5651 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5652 case BPF_LWT_ENCAP_IP:
5653 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
5660 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
5661 void *, hdr, u32, len)
5664 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5665 case BPF_LWT_ENCAP_IP:
5666 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
5673 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
5674 .func = bpf_lwt_in_push_encap,
5676 .ret_type = RET_INTEGER,
5677 .arg1_type = ARG_PTR_TO_CTX,
5678 .arg2_type = ARG_ANYTHING,
5679 .arg3_type = ARG_PTR_TO_MEM,
5680 .arg4_type = ARG_CONST_SIZE
5683 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
5684 .func = bpf_lwt_xmit_push_encap,
5686 .ret_type = RET_INTEGER,
5687 .arg1_type = ARG_PTR_TO_CTX,
5688 .arg2_type = ARG_ANYTHING,
5689 .arg3_type = ARG_PTR_TO_MEM,
5690 .arg4_type = ARG_CONST_SIZE
5693 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5694 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
5695 const void *, from, u32, len)
5697 struct seg6_bpf_srh_state *srh_state =
5698 this_cpu_ptr(&seg6_bpf_srh_states);
5699 struct ipv6_sr_hdr *srh = srh_state->srh;
5700 void *srh_tlvs, *srh_end, *ptr;
5706 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
5707 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
5709 ptr = skb->data + offset;
5710 if (ptr >= srh_tlvs && ptr + len <= srh_end)
5711 srh_state->valid = false;
5712 else if (ptr < (void *)&srh->flags ||
5713 ptr + len > (void *)&srh->segments)
5716 if (unlikely(bpf_try_make_writable(skb, offset + len)))
5718 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5720 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5722 memcpy(skb->data + offset, from, len);
5726 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
5727 .func = bpf_lwt_seg6_store_bytes,
5729 .ret_type = RET_INTEGER,
5730 .arg1_type = ARG_PTR_TO_CTX,
5731 .arg2_type = ARG_ANYTHING,
5732 .arg3_type = ARG_PTR_TO_MEM,
5733 .arg4_type = ARG_CONST_SIZE
5736 static void bpf_update_srh_state(struct sk_buff *skb)
5738 struct seg6_bpf_srh_state *srh_state =
5739 this_cpu_ptr(&seg6_bpf_srh_states);
5742 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
5743 srh_state->srh = NULL;
5745 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5746 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
5747 srh_state->valid = true;
5751 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
5752 u32, action, void *, param, u32, param_len)
5754 struct seg6_bpf_srh_state *srh_state =
5755 this_cpu_ptr(&seg6_bpf_srh_states);
5760 case SEG6_LOCAL_ACTION_END_X:
5761 if (!seg6_bpf_has_valid_srh(skb))
5763 if (param_len != sizeof(struct in6_addr))
5765 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
5766 case SEG6_LOCAL_ACTION_END_T:
5767 if (!seg6_bpf_has_valid_srh(skb))
5769 if (param_len != sizeof(int))
5771 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5772 case SEG6_LOCAL_ACTION_END_DT6:
5773 if (!seg6_bpf_has_valid_srh(skb))
5775 if (param_len != sizeof(int))
5778 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
5780 if (!pskb_pull(skb, hdroff))
5783 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
5784 skb_reset_network_header(skb);
5785 skb_reset_transport_header(skb);
5786 skb->encapsulation = 0;
5788 bpf_compute_data_pointers(skb);
5789 bpf_update_srh_state(skb);
5790 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5791 case SEG6_LOCAL_ACTION_END_B6:
5792 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5794 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
5797 bpf_update_srh_state(skb);
5800 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
5801 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5803 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
5806 bpf_update_srh_state(skb);
5814 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
5815 .func = bpf_lwt_seg6_action,
5817 .ret_type = RET_INTEGER,
5818 .arg1_type = ARG_PTR_TO_CTX,
5819 .arg2_type = ARG_ANYTHING,
5820 .arg3_type = ARG_PTR_TO_MEM,
5821 .arg4_type = ARG_CONST_SIZE
5824 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
5827 struct seg6_bpf_srh_state *srh_state =
5828 this_cpu_ptr(&seg6_bpf_srh_states);
5829 struct ipv6_sr_hdr *srh = srh_state->srh;
5830 void *srh_end, *srh_tlvs, *ptr;
5831 struct ipv6hdr *hdr;
5835 if (unlikely(srh == NULL))
5838 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
5839 ((srh->first_segment + 1) << 4));
5840 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
5842 ptr = skb->data + offset;
5844 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
5846 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
5850 ret = skb_cow_head(skb, len);
5851 if (unlikely(ret < 0))
5854 ret = bpf_skb_net_hdr_push(skb, offset, len);
5856 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
5859 bpf_compute_data_pointers(skb);
5860 if (unlikely(ret < 0))
5863 hdr = (struct ipv6hdr *)skb->data;
5864 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5866 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5868 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5869 srh_state->hdrlen += len;
5870 srh_state->valid = false;
5874 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
5875 .func = bpf_lwt_seg6_adjust_srh,
5877 .ret_type = RET_INTEGER,
5878 .arg1_type = ARG_PTR_TO_CTX,
5879 .arg2_type = ARG_ANYTHING,
5880 .arg3_type = ARG_ANYTHING,
5882 #endif /* CONFIG_IPV6_SEG6_BPF */
5885 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
5886 int dif, int sdif, u8 family, u8 proto)
5888 bool refcounted = false;
5889 struct sock *sk = NULL;
5891 if (family == AF_INET) {
5892 __be32 src4 = tuple->ipv4.saddr;
5893 __be32 dst4 = tuple->ipv4.daddr;
5895 if (proto == IPPROTO_TCP)
5896 sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
5897 src4, tuple->ipv4.sport,
5898 dst4, tuple->ipv4.dport,
5899 dif, sdif, &refcounted);
5901 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
5902 dst4, tuple->ipv4.dport,
5903 dif, sdif, &udp_table, NULL);
5904 #if IS_ENABLED(CONFIG_IPV6)
5906 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
5907 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
5909 if (proto == IPPROTO_TCP)
5910 sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
5911 src6, tuple->ipv6.sport,
5912 dst6, ntohs(tuple->ipv6.dport),
5913 dif, sdif, &refcounted);
5914 else if (likely(ipv6_bpf_stub))
5915 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
5916 src6, tuple->ipv6.sport,
5917 dst6, tuple->ipv6.dport,
5923 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
5924 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
5930 /* bpf_skc_lookup performs the core lookup for different types of sockets,
5931 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
5932 * Returns the socket as an 'unsigned long' to simplify the casting in the
5933 * callers to satisfy BPF_CALL declarations.
5935 static struct sock *
5936 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5937 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
5940 struct sock *sk = NULL;
5941 u8 family = AF_UNSPEC;
5945 if (len == sizeof(tuple->ipv4))
5947 else if (len == sizeof(tuple->ipv6))
5952 if (unlikely(family == AF_UNSPEC || flags ||
5953 !((s32)netns_id < 0 || netns_id <= S32_MAX)))
5956 if (family == AF_INET)
5957 sdif = inet_sdif(skb);
5959 sdif = inet6_sdif(skb);
5961 if ((s32)netns_id < 0) {
5963 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
5965 net = get_net_ns_by_id(caller_net, netns_id);
5968 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
5976 static struct sock *
5977 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5978 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
5981 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
5982 ifindex, proto, netns_id, flags);
5985 struct sock *sk2 = sk_to_full_sk(sk);
5987 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
5988 * sock refcnt is decremented to prevent a request_sock leak.
5990 if (!sk_fullsock(sk2))
5994 /* Ensure there is no need to bump sk2 refcnt */
5995 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
5996 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6006 static struct sock *
6007 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6008 u8 proto, u64 netns_id, u64 flags)
6010 struct net *caller_net;
6014 caller_net = dev_net(skb->dev);
6015 ifindex = skb->dev->ifindex;
6017 caller_net = sock_net(skb->sk);
6021 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6025 static struct sock *
6026 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6027 u8 proto, u64 netns_id, u64 flags)
6029 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6033 struct sock *sk2 = sk_to_full_sk(sk);
6035 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6036 * sock refcnt is decremented to prevent a request_sock leak.
6038 if (!sk_fullsock(sk2))
6042 /* Ensure there is no need to bump sk2 refcnt */
6043 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6044 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6054 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6055 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6057 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6061 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6062 .func = bpf_skc_lookup_tcp,
6065 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6066 .arg1_type = ARG_PTR_TO_CTX,
6067 .arg2_type = ARG_PTR_TO_MEM,
6068 .arg3_type = ARG_CONST_SIZE,
6069 .arg4_type = ARG_ANYTHING,
6070 .arg5_type = ARG_ANYTHING,
6073 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6074 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6076 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6080 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6081 .func = bpf_sk_lookup_tcp,
6084 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6085 .arg1_type = ARG_PTR_TO_CTX,
6086 .arg2_type = ARG_PTR_TO_MEM,
6087 .arg3_type = ARG_CONST_SIZE,
6088 .arg4_type = ARG_ANYTHING,
6089 .arg5_type = ARG_ANYTHING,
6092 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6093 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6095 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6099 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6100 .func = bpf_sk_lookup_udp,
6103 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6104 .arg1_type = ARG_PTR_TO_CTX,
6105 .arg2_type = ARG_PTR_TO_MEM,
6106 .arg3_type = ARG_CONST_SIZE,
6107 .arg4_type = ARG_ANYTHING,
6108 .arg5_type = ARG_ANYTHING,
6111 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6113 if (sk && sk_is_refcounted(sk))
6118 static const struct bpf_func_proto bpf_sk_release_proto = {
6119 .func = bpf_sk_release,
6121 .ret_type = RET_INTEGER,
6122 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6125 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6126 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6128 struct net *caller_net = dev_net(ctx->rxq->dev);
6129 int ifindex = ctx->rxq->dev->ifindex;
6131 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6132 ifindex, IPPROTO_UDP, netns_id,
6136 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6137 .func = bpf_xdp_sk_lookup_udp,
6140 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6141 .arg1_type = ARG_PTR_TO_CTX,
6142 .arg2_type = ARG_PTR_TO_MEM,
6143 .arg3_type = ARG_CONST_SIZE,
6144 .arg4_type = ARG_ANYTHING,
6145 .arg5_type = ARG_ANYTHING,
6148 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6149 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6151 struct net *caller_net = dev_net(ctx->rxq->dev);
6152 int ifindex = ctx->rxq->dev->ifindex;
6154 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6155 ifindex, IPPROTO_TCP, netns_id,
6159 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6160 .func = bpf_xdp_skc_lookup_tcp,
6163 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6164 .arg1_type = ARG_PTR_TO_CTX,
6165 .arg2_type = ARG_PTR_TO_MEM,
6166 .arg3_type = ARG_CONST_SIZE,
6167 .arg4_type = ARG_ANYTHING,
6168 .arg5_type = ARG_ANYTHING,
6171 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6172 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6174 struct net *caller_net = dev_net(ctx->rxq->dev);
6175 int ifindex = ctx->rxq->dev->ifindex;
6177 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6178 ifindex, IPPROTO_TCP, netns_id,
6182 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6183 .func = bpf_xdp_sk_lookup_tcp,
6186 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6187 .arg1_type = ARG_PTR_TO_CTX,
6188 .arg2_type = ARG_PTR_TO_MEM,
6189 .arg3_type = ARG_CONST_SIZE,
6190 .arg4_type = ARG_ANYTHING,
6191 .arg5_type = ARG_ANYTHING,
6194 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6195 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6197 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6198 sock_net(ctx->sk), 0,
6199 IPPROTO_TCP, netns_id, flags);
6202 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6203 .func = bpf_sock_addr_skc_lookup_tcp,
6205 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6206 .arg1_type = ARG_PTR_TO_CTX,
6207 .arg2_type = ARG_PTR_TO_MEM,
6208 .arg3_type = ARG_CONST_SIZE,
6209 .arg4_type = ARG_ANYTHING,
6210 .arg5_type = ARG_ANYTHING,
6213 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6214 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6216 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6217 sock_net(ctx->sk), 0, IPPROTO_TCP,
6221 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6222 .func = bpf_sock_addr_sk_lookup_tcp,
6224 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6225 .arg1_type = ARG_PTR_TO_CTX,
6226 .arg2_type = ARG_PTR_TO_MEM,
6227 .arg3_type = ARG_CONST_SIZE,
6228 .arg4_type = ARG_ANYTHING,
6229 .arg5_type = ARG_ANYTHING,
6232 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6233 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6235 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6236 sock_net(ctx->sk), 0, IPPROTO_UDP,
6240 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6241 .func = bpf_sock_addr_sk_lookup_udp,
6243 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6244 .arg1_type = ARG_PTR_TO_CTX,
6245 .arg2_type = ARG_PTR_TO_MEM,
6246 .arg3_type = ARG_CONST_SIZE,
6247 .arg4_type = ARG_ANYTHING,
6248 .arg5_type = ARG_ANYTHING,
6251 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6252 struct bpf_insn_access_aux *info)
6254 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6258 if (off % size != 0)
6262 case offsetof(struct bpf_tcp_sock, bytes_received):
6263 case offsetof(struct bpf_tcp_sock, bytes_acked):
6264 return size == sizeof(__u64);
6266 return size == sizeof(__u32);
6270 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6271 const struct bpf_insn *si,
6272 struct bpf_insn *insn_buf,
6273 struct bpf_prog *prog, u32 *target_size)
6275 struct bpf_insn *insn = insn_buf;
6277 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \
6279 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \
6280 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6281 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6282 si->dst_reg, si->src_reg, \
6283 offsetof(struct tcp_sock, FIELD)); \
6286 #define BPF_INET_SOCK_GET_COMMON(FIELD) \
6288 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \
6290 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6291 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6292 struct inet_connection_sock, \
6294 si->dst_reg, si->src_reg, \
6296 struct inet_connection_sock, \
6300 if (insn > insn_buf)
6301 return insn - insn_buf;
6304 case offsetof(struct bpf_tcp_sock, rtt_min):
6305 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
6306 sizeof(struct minmax));
6307 BUILD_BUG_ON(sizeof(struct minmax) <
6308 sizeof(struct minmax_sample));
6310 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6311 offsetof(struct tcp_sock, rtt_min) +
6312 offsetof(struct minmax_sample, v));
6314 case offsetof(struct bpf_tcp_sock, snd_cwnd):
6315 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
6317 case offsetof(struct bpf_tcp_sock, srtt_us):
6318 BPF_TCP_SOCK_GET_COMMON(srtt_us);
6320 case offsetof(struct bpf_tcp_sock, snd_ssthresh):
6321 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
6323 case offsetof(struct bpf_tcp_sock, rcv_nxt):
6324 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
6326 case offsetof(struct bpf_tcp_sock, snd_nxt):
6327 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
6329 case offsetof(struct bpf_tcp_sock, snd_una):
6330 BPF_TCP_SOCK_GET_COMMON(snd_una);
6332 case offsetof(struct bpf_tcp_sock, mss_cache):
6333 BPF_TCP_SOCK_GET_COMMON(mss_cache);
6335 case offsetof(struct bpf_tcp_sock, ecn_flags):
6336 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
6338 case offsetof(struct bpf_tcp_sock, rate_delivered):
6339 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
6341 case offsetof(struct bpf_tcp_sock, rate_interval_us):
6342 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
6344 case offsetof(struct bpf_tcp_sock, packets_out):
6345 BPF_TCP_SOCK_GET_COMMON(packets_out);
6347 case offsetof(struct bpf_tcp_sock, retrans_out):
6348 BPF_TCP_SOCK_GET_COMMON(retrans_out);
6350 case offsetof(struct bpf_tcp_sock, total_retrans):
6351 BPF_TCP_SOCK_GET_COMMON(total_retrans);
6353 case offsetof(struct bpf_tcp_sock, segs_in):
6354 BPF_TCP_SOCK_GET_COMMON(segs_in);
6356 case offsetof(struct bpf_tcp_sock, data_segs_in):
6357 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
6359 case offsetof(struct bpf_tcp_sock, segs_out):
6360 BPF_TCP_SOCK_GET_COMMON(segs_out);
6362 case offsetof(struct bpf_tcp_sock, data_segs_out):
6363 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
6365 case offsetof(struct bpf_tcp_sock, lost_out):
6366 BPF_TCP_SOCK_GET_COMMON(lost_out);
6368 case offsetof(struct bpf_tcp_sock, sacked_out):
6369 BPF_TCP_SOCK_GET_COMMON(sacked_out);
6371 case offsetof(struct bpf_tcp_sock, bytes_received):
6372 BPF_TCP_SOCK_GET_COMMON(bytes_received);
6374 case offsetof(struct bpf_tcp_sock, bytes_acked):
6375 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
6377 case offsetof(struct bpf_tcp_sock, dsack_dups):
6378 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
6380 case offsetof(struct bpf_tcp_sock, delivered):
6381 BPF_TCP_SOCK_GET_COMMON(delivered);
6383 case offsetof(struct bpf_tcp_sock, delivered_ce):
6384 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
6386 case offsetof(struct bpf_tcp_sock, icsk_retransmits):
6387 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
6391 return insn - insn_buf;
6394 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
6396 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
6397 return (unsigned long)sk;
6399 return (unsigned long)NULL;
6402 const struct bpf_func_proto bpf_tcp_sock_proto = {
6403 .func = bpf_tcp_sock,
6405 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
6406 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6409 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
6411 sk = sk_to_full_sk(sk);
6413 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
6414 return (unsigned long)sk;
6416 return (unsigned long)NULL;
6419 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
6420 .func = bpf_get_listener_sock,
6422 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6423 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6426 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
6428 unsigned int iphdr_len;
6430 switch (skb_protocol(skb, true)) {
6431 case cpu_to_be16(ETH_P_IP):
6432 iphdr_len = sizeof(struct iphdr);
6434 case cpu_to_be16(ETH_P_IPV6):
6435 iphdr_len = sizeof(struct ipv6hdr);
6441 if (skb_headlen(skb) < iphdr_len)
6444 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
6447 return INET_ECN_set_ce(skb);
6450 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6451 struct bpf_insn_access_aux *info)
6453 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
6456 if (off % size != 0)
6461 return size == sizeof(__u32);
6465 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
6466 const struct bpf_insn *si,
6467 struct bpf_insn *insn_buf,
6468 struct bpf_prog *prog, u32 *target_size)
6470 struct bpf_insn *insn = insn_buf;
6472 #define BPF_XDP_SOCK_GET(FIELD) \
6474 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \
6475 sizeof_field(struct bpf_xdp_sock, FIELD)); \
6476 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
6477 si->dst_reg, si->src_reg, \
6478 offsetof(struct xdp_sock, FIELD)); \
6482 case offsetof(struct bpf_xdp_sock, queue_id):
6483 BPF_XDP_SOCK_GET(queue_id);
6487 return insn - insn_buf;
6490 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
6491 .func = bpf_skb_ecn_set_ce,
6493 .ret_type = RET_INTEGER,
6494 .arg1_type = ARG_PTR_TO_CTX,
6497 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6498 struct tcphdr *, th, u32, th_len)
6500 #ifdef CONFIG_SYN_COOKIES
6504 if (unlikely(!sk || th_len < sizeof(*th)))
6507 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
6508 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6511 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
6514 if (!th->ack || th->rst || th->syn)
6517 if (unlikely(iph_len < sizeof(struct iphdr)))
6520 if (tcp_synq_no_recent_overflow(sk))
6523 cookie = ntohl(th->ack_seq) - 1;
6525 /* Both struct iphdr and struct ipv6hdr have the version field at the
6526 * same offset so we can cast to the shorter header (struct iphdr).
6528 switch (((struct iphdr *)iph)->version) {
6530 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
6533 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
6536 #if IS_BUILTIN(CONFIG_IPV6)
6538 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6541 if (sk->sk_family != AF_INET6)
6544 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
6546 #endif /* CONFIG_IPV6 */
6549 return -EPROTONOSUPPORT;
6561 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
6562 .func = bpf_tcp_check_syncookie,
6565 .ret_type = RET_INTEGER,
6566 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6567 .arg2_type = ARG_PTR_TO_MEM,
6568 .arg3_type = ARG_CONST_SIZE,
6569 .arg4_type = ARG_PTR_TO_MEM,
6570 .arg5_type = ARG_CONST_SIZE,
6573 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6574 struct tcphdr *, th, u32, th_len)
6576 #ifdef CONFIG_SYN_COOKIES
6580 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
6583 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6586 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
6589 if (!th->syn || th->ack || th->fin || th->rst)
6592 if (unlikely(iph_len < sizeof(struct iphdr)))
6595 /* Both struct iphdr and struct ipv6hdr have the version field at the
6596 * same offset so we can cast to the shorter header (struct iphdr).
6598 switch (((struct iphdr *)iph)->version) {
6600 if (sk->sk_family == AF_INET6 && sk->sk_ipv6only)
6603 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
6606 #if IS_BUILTIN(CONFIG_IPV6)
6608 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6611 if (sk->sk_family != AF_INET6)
6614 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
6616 #endif /* CONFIG_IPV6 */
6619 return -EPROTONOSUPPORT;
6624 return cookie | ((u64)mss << 32);
6627 #endif /* CONFIG_SYN_COOKIES */
6630 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
6631 .func = bpf_tcp_gen_syncookie,
6632 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */
6634 .ret_type = RET_INTEGER,
6635 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6636 .arg2_type = ARG_PTR_TO_MEM,
6637 .arg3_type = ARG_CONST_SIZE,
6638 .arg4_type = ARG_PTR_TO_MEM,
6639 .arg5_type = ARG_CONST_SIZE,
6642 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
6644 if (!sk || flags != 0)
6646 if (!skb_at_tc_ingress(skb))
6648 if (unlikely(dev_net(skb->dev) != sock_net(sk)))
6649 return -ENETUNREACH;
6650 if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
6651 return -ESOCKTNOSUPPORT;
6652 if (sk_is_refcounted(sk) &&
6653 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
6658 skb->destructor = sock_pfree;
6663 static const struct bpf_func_proto bpf_sk_assign_proto = {
6664 .func = bpf_sk_assign,
6666 .ret_type = RET_INTEGER,
6667 .arg1_type = ARG_PTR_TO_CTX,
6668 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6669 .arg3_type = ARG_ANYTHING,
6672 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
6673 u8 search_kind, const u8 *magic,
6674 u8 magic_len, bool *eol)
6680 while (op < opend) {
6683 if (kind == TCPOPT_EOL) {
6685 return ERR_PTR(-ENOMSG);
6686 } else if (kind == TCPOPT_NOP) {
6691 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
6692 /* Something is wrong in the received header.
6693 * Follow the TCP stack's tcp_parse_options()
6694 * and just bail here.
6696 return ERR_PTR(-EFAULT);
6699 if (search_kind == kind) {
6703 if (magic_len > kind_len - 2)
6704 return ERR_PTR(-ENOMSG);
6706 if (!memcmp(&op[2], magic, magic_len))
6713 return ERR_PTR(-ENOMSG);
6716 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6717 void *, search_res, u32, len, u64, flags)
6719 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
6720 const u8 *op, *opend, *magic, *search = search_res;
6721 u8 search_kind, search_len, copy_len, magic_len;
6724 /* 2 byte is the minimal option len except TCPOPT_NOP and
6725 * TCPOPT_EOL which are useless for the bpf prog to learn
6726 * and this helper disallow loading them also.
6728 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
6731 search_kind = search[0];
6732 search_len = search[1];
6734 if (search_len > len || search_kind == TCPOPT_NOP ||
6735 search_kind == TCPOPT_EOL)
6738 if (search_kind == TCPOPT_EXP || search_kind == 253) {
6739 /* 16 or 32 bit magic. +2 for kind and kind length */
6740 if (search_len != 4 && search_len != 6)
6743 magic_len = search_len - 2;
6752 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
6757 op += sizeof(struct tcphdr);
6759 if (!bpf_sock->skb ||
6760 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6761 /* This bpf_sock->op cannot call this helper */
6764 opend = bpf_sock->skb_data_end;
6765 op = bpf_sock->skb->data + sizeof(struct tcphdr);
6768 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
6775 if (copy_len > len) {
6780 memcpy(search_res, op, copy_len);
6784 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
6785 .func = bpf_sock_ops_load_hdr_opt,
6787 .ret_type = RET_INTEGER,
6788 .arg1_type = ARG_PTR_TO_CTX,
6789 .arg2_type = ARG_PTR_TO_MEM,
6790 .arg3_type = ARG_CONST_SIZE,
6791 .arg4_type = ARG_ANYTHING,
6794 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6795 const void *, from, u32, len, u64, flags)
6797 u8 new_kind, new_kind_len, magic_len = 0, *opend;
6798 const u8 *op, *new_op, *magic = NULL;
6799 struct sk_buff *skb;
6802 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
6805 if (len < 2 || flags)
6809 new_kind = new_op[0];
6810 new_kind_len = new_op[1];
6812 if (new_kind_len > len || new_kind == TCPOPT_NOP ||
6813 new_kind == TCPOPT_EOL)
6816 if (new_kind_len > bpf_sock->remaining_opt_len)
6819 /* 253 is another experimental kind */
6820 if (new_kind == TCPOPT_EXP || new_kind == 253) {
6821 if (new_kind_len < 4)
6823 /* Match for the 2 byte magic also.
6824 * RFC 6994: the magic could be 2 or 4 bytes.
6825 * Hence, matching by 2 byte only is on the
6826 * conservative side but it is the right
6827 * thing to do for the 'search-for-duplication'
6834 /* Check for duplication */
6835 skb = bpf_sock->skb;
6836 op = skb->data + sizeof(struct tcphdr);
6837 opend = bpf_sock->skb_data_end;
6839 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
6844 if (PTR_ERR(op) != -ENOMSG)
6848 /* The option has been ended. Treat it as no more
6849 * header option can be written.
6853 /* No duplication found. Store the header option. */
6854 memcpy(opend, from, new_kind_len);
6856 bpf_sock->remaining_opt_len -= new_kind_len;
6857 bpf_sock->skb_data_end += new_kind_len;
6862 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
6863 .func = bpf_sock_ops_store_hdr_opt,
6865 .ret_type = RET_INTEGER,
6866 .arg1_type = ARG_PTR_TO_CTX,
6867 .arg2_type = ARG_PTR_TO_MEM,
6868 .arg3_type = ARG_CONST_SIZE,
6869 .arg4_type = ARG_ANYTHING,
6872 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6873 u32, len, u64, flags)
6875 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6878 if (flags || len < 2)
6881 if (len > bpf_sock->remaining_opt_len)
6884 bpf_sock->remaining_opt_len -= len;
6889 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
6890 .func = bpf_sock_ops_reserve_hdr_opt,
6892 .ret_type = RET_INTEGER,
6893 .arg1_type = ARG_PTR_TO_CTX,
6894 .arg2_type = ARG_ANYTHING,
6895 .arg3_type = ARG_ANYTHING,
6898 #endif /* CONFIG_INET */
6900 bool bpf_helper_changes_pkt_data(void *func)
6902 if (func == bpf_skb_vlan_push ||
6903 func == bpf_skb_vlan_pop ||
6904 func == bpf_skb_store_bytes ||
6905 func == bpf_skb_change_proto ||
6906 func == bpf_skb_change_head ||
6907 func == sk_skb_change_head ||
6908 func == bpf_skb_change_tail ||
6909 func == sk_skb_change_tail ||
6910 func == bpf_skb_adjust_room ||
6911 func == sk_skb_adjust_room ||
6912 func == bpf_skb_pull_data ||
6913 func == sk_skb_pull_data ||
6914 func == bpf_clone_redirect ||
6915 func == bpf_l3_csum_replace ||
6916 func == bpf_l4_csum_replace ||
6917 func == bpf_xdp_adjust_head ||
6918 func == bpf_xdp_adjust_meta ||
6919 func == bpf_msg_pull_data ||
6920 func == bpf_msg_push_data ||
6921 func == bpf_msg_pop_data ||
6922 func == bpf_xdp_adjust_tail ||
6923 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6924 func == bpf_lwt_seg6_store_bytes ||
6925 func == bpf_lwt_seg6_adjust_srh ||
6926 func == bpf_lwt_seg6_action ||
6929 func == bpf_sock_ops_store_hdr_opt ||
6931 func == bpf_lwt_in_push_encap ||
6932 func == bpf_lwt_xmit_push_encap)
6938 const struct bpf_func_proto bpf_event_output_data_proto __weak;
6939 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
6941 static const struct bpf_func_proto *
6942 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6945 /* inet and inet6 sockets are created in a process
6946 * context so there is always a valid uid/gid
6948 case BPF_FUNC_get_current_uid_gid:
6949 return &bpf_get_current_uid_gid_proto;
6950 case BPF_FUNC_get_local_storage:
6951 return &bpf_get_local_storage_proto;
6952 case BPF_FUNC_get_socket_cookie:
6953 return &bpf_get_socket_cookie_sock_proto;
6954 case BPF_FUNC_get_netns_cookie:
6955 return &bpf_get_netns_cookie_sock_proto;
6956 case BPF_FUNC_perf_event_output:
6957 return &bpf_event_output_data_proto;
6958 case BPF_FUNC_get_current_pid_tgid:
6959 return &bpf_get_current_pid_tgid_proto;
6960 case BPF_FUNC_get_current_comm:
6961 return &bpf_get_current_comm_proto;
6962 #ifdef CONFIG_CGROUPS
6963 case BPF_FUNC_get_current_cgroup_id:
6964 return &bpf_get_current_cgroup_id_proto;
6965 case BPF_FUNC_get_current_ancestor_cgroup_id:
6966 return &bpf_get_current_ancestor_cgroup_id_proto;
6968 #ifdef CONFIG_CGROUP_NET_CLASSID
6969 case BPF_FUNC_get_cgroup_classid:
6970 return &bpf_get_cgroup_classid_curr_proto;
6972 case BPF_FUNC_sk_storage_get:
6973 return &bpf_sk_storage_get_cg_sock_proto;
6975 return bpf_base_func_proto(func_id);
6979 static const struct bpf_func_proto *
6980 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6983 /* inet and inet6 sockets are created in a process
6984 * context so there is always a valid uid/gid
6986 case BPF_FUNC_get_current_uid_gid:
6987 return &bpf_get_current_uid_gid_proto;
6989 switch (prog->expected_attach_type) {
6990 case BPF_CGROUP_INET4_CONNECT:
6991 case BPF_CGROUP_INET6_CONNECT:
6992 return &bpf_bind_proto;
6996 case BPF_FUNC_get_socket_cookie:
6997 return &bpf_get_socket_cookie_sock_addr_proto;
6998 case BPF_FUNC_get_netns_cookie:
6999 return &bpf_get_netns_cookie_sock_addr_proto;
7000 case BPF_FUNC_get_local_storage:
7001 return &bpf_get_local_storage_proto;
7002 case BPF_FUNC_perf_event_output:
7003 return &bpf_event_output_data_proto;
7004 case BPF_FUNC_get_current_pid_tgid:
7005 return &bpf_get_current_pid_tgid_proto;
7006 case BPF_FUNC_get_current_comm:
7007 return &bpf_get_current_comm_proto;
7008 #ifdef CONFIG_CGROUPS
7009 case BPF_FUNC_get_current_cgroup_id:
7010 return &bpf_get_current_cgroup_id_proto;
7011 case BPF_FUNC_get_current_ancestor_cgroup_id:
7012 return &bpf_get_current_ancestor_cgroup_id_proto;
7014 #ifdef CONFIG_CGROUP_NET_CLASSID
7015 case BPF_FUNC_get_cgroup_classid:
7016 return &bpf_get_cgroup_classid_curr_proto;
7019 case BPF_FUNC_sk_lookup_tcp:
7020 return &bpf_sock_addr_sk_lookup_tcp_proto;
7021 case BPF_FUNC_sk_lookup_udp:
7022 return &bpf_sock_addr_sk_lookup_udp_proto;
7023 case BPF_FUNC_sk_release:
7024 return &bpf_sk_release_proto;
7025 case BPF_FUNC_skc_lookup_tcp:
7026 return &bpf_sock_addr_skc_lookup_tcp_proto;
7027 #endif /* CONFIG_INET */
7028 case BPF_FUNC_sk_storage_get:
7029 return &bpf_sk_storage_get_proto;
7030 case BPF_FUNC_sk_storage_delete:
7031 return &bpf_sk_storage_delete_proto;
7032 case BPF_FUNC_setsockopt:
7033 switch (prog->expected_attach_type) {
7034 case BPF_CGROUP_INET4_CONNECT:
7035 case BPF_CGROUP_INET6_CONNECT:
7036 return &bpf_sock_addr_setsockopt_proto;
7040 case BPF_FUNC_getsockopt:
7041 switch (prog->expected_attach_type) {
7042 case BPF_CGROUP_INET4_CONNECT:
7043 case BPF_CGROUP_INET6_CONNECT:
7044 return &bpf_sock_addr_getsockopt_proto;
7049 return bpf_sk_base_func_proto(func_id);
7053 static const struct bpf_func_proto *
7054 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7057 case BPF_FUNC_skb_load_bytes:
7058 return &bpf_skb_load_bytes_proto;
7059 case BPF_FUNC_skb_load_bytes_relative:
7060 return &bpf_skb_load_bytes_relative_proto;
7061 case BPF_FUNC_get_socket_cookie:
7062 return &bpf_get_socket_cookie_proto;
7063 case BPF_FUNC_get_socket_uid:
7064 return &bpf_get_socket_uid_proto;
7065 case BPF_FUNC_perf_event_output:
7066 return &bpf_skb_event_output_proto;
7068 return bpf_sk_base_func_proto(func_id);
7072 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7073 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7075 static const struct bpf_func_proto *
7076 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7079 case BPF_FUNC_get_local_storage:
7080 return &bpf_get_local_storage_proto;
7081 case BPF_FUNC_sk_fullsock:
7082 return &bpf_sk_fullsock_proto;
7083 case BPF_FUNC_sk_storage_get:
7084 return &bpf_sk_storage_get_proto;
7085 case BPF_FUNC_sk_storage_delete:
7086 return &bpf_sk_storage_delete_proto;
7087 case BPF_FUNC_perf_event_output:
7088 return &bpf_skb_event_output_proto;
7089 #ifdef CONFIG_SOCK_CGROUP_DATA
7090 case BPF_FUNC_skb_cgroup_id:
7091 return &bpf_skb_cgroup_id_proto;
7092 case BPF_FUNC_skb_ancestor_cgroup_id:
7093 return &bpf_skb_ancestor_cgroup_id_proto;
7094 case BPF_FUNC_sk_cgroup_id:
7095 return &bpf_sk_cgroup_id_proto;
7096 case BPF_FUNC_sk_ancestor_cgroup_id:
7097 return &bpf_sk_ancestor_cgroup_id_proto;
7100 case BPF_FUNC_sk_lookup_tcp:
7101 return &bpf_sk_lookup_tcp_proto;
7102 case BPF_FUNC_sk_lookup_udp:
7103 return &bpf_sk_lookup_udp_proto;
7104 case BPF_FUNC_sk_release:
7105 return &bpf_sk_release_proto;
7106 case BPF_FUNC_skc_lookup_tcp:
7107 return &bpf_skc_lookup_tcp_proto;
7108 case BPF_FUNC_tcp_sock:
7109 return &bpf_tcp_sock_proto;
7110 case BPF_FUNC_get_listener_sock:
7111 return &bpf_get_listener_sock_proto;
7112 case BPF_FUNC_skb_ecn_set_ce:
7113 return &bpf_skb_ecn_set_ce_proto;
7116 return sk_filter_func_proto(func_id, prog);
7120 static const struct bpf_func_proto *
7121 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7124 case BPF_FUNC_skb_store_bytes:
7125 return &bpf_skb_store_bytes_proto;
7126 case BPF_FUNC_skb_load_bytes:
7127 return &bpf_skb_load_bytes_proto;
7128 case BPF_FUNC_skb_load_bytes_relative:
7129 return &bpf_skb_load_bytes_relative_proto;
7130 case BPF_FUNC_skb_pull_data:
7131 return &bpf_skb_pull_data_proto;
7132 case BPF_FUNC_csum_diff:
7133 return &bpf_csum_diff_proto;
7134 case BPF_FUNC_csum_update:
7135 return &bpf_csum_update_proto;
7136 case BPF_FUNC_csum_level:
7137 return &bpf_csum_level_proto;
7138 case BPF_FUNC_l3_csum_replace:
7139 return &bpf_l3_csum_replace_proto;
7140 case BPF_FUNC_l4_csum_replace:
7141 return &bpf_l4_csum_replace_proto;
7142 case BPF_FUNC_clone_redirect:
7143 return &bpf_clone_redirect_proto;
7144 case BPF_FUNC_get_cgroup_classid:
7145 return &bpf_get_cgroup_classid_proto;
7146 case BPF_FUNC_skb_vlan_push:
7147 return &bpf_skb_vlan_push_proto;
7148 case BPF_FUNC_skb_vlan_pop:
7149 return &bpf_skb_vlan_pop_proto;
7150 case BPF_FUNC_skb_change_proto:
7151 return &bpf_skb_change_proto_proto;
7152 case BPF_FUNC_skb_change_type:
7153 return &bpf_skb_change_type_proto;
7154 case BPF_FUNC_skb_adjust_room:
7155 return &bpf_skb_adjust_room_proto;
7156 case BPF_FUNC_skb_change_tail:
7157 return &bpf_skb_change_tail_proto;
7158 case BPF_FUNC_skb_change_head:
7159 return &bpf_skb_change_head_proto;
7160 case BPF_FUNC_skb_get_tunnel_key:
7161 return &bpf_skb_get_tunnel_key_proto;
7162 case BPF_FUNC_skb_set_tunnel_key:
7163 return bpf_get_skb_set_tunnel_proto(func_id);
7164 case BPF_FUNC_skb_get_tunnel_opt:
7165 return &bpf_skb_get_tunnel_opt_proto;
7166 case BPF_FUNC_skb_set_tunnel_opt:
7167 return bpf_get_skb_set_tunnel_proto(func_id);
7168 case BPF_FUNC_redirect:
7169 return &bpf_redirect_proto;
7170 case BPF_FUNC_redirect_neigh:
7171 return &bpf_redirect_neigh_proto;
7172 case BPF_FUNC_redirect_peer:
7173 return &bpf_redirect_peer_proto;
7174 case BPF_FUNC_get_route_realm:
7175 return &bpf_get_route_realm_proto;
7176 case BPF_FUNC_get_hash_recalc:
7177 return &bpf_get_hash_recalc_proto;
7178 case BPF_FUNC_set_hash_invalid:
7179 return &bpf_set_hash_invalid_proto;
7180 case BPF_FUNC_set_hash:
7181 return &bpf_set_hash_proto;
7182 case BPF_FUNC_perf_event_output:
7183 return &bpf_skb_event_output_proto;
7184 case BPF_FUNC_get_smp_processor_id:
7185 return &bpf_get_smp_processor_id_proto;
7186 case BPF_FUNC_skb_under_cgroup:
7187 return &bpf_skb_under_cgroup_proto;
7188 case BPF_FUNC_get_socket_cookie:
7189 return &bpf_get_socket_cookie_proto;
7190 case BPF_FUNC_get_socket_uid:
7191 return &bpf_get_socket_uid_proto;
7192 case BPF_FUNC_fib_lookup:
7193 return &bpf_skb_fib_lookup_proto;
7194 case BPF_FUNC_sk_fullsock:
7195 return &bpf_sk_fullsock_proto;
7196 case BPF_FUNC_sk_storage_get:
7197 return &bpf_sk_storage_get_proto;
7198 case BPF_FUNC_sk_storage_delete:
7199 return &bpf_sk_storage_delete_proto;
7201 case BPF_FUNC_skb_get_xfrm_state:
7202 return &bpf_skb_get_xfrm_state_proto;
7204 #ifdef CONFIG_CGROUP_NET_CLASSID
7205 case BPF_FUNC_skb_cgroup_classid:
7206 return &bpf_skb_cgroup_classid_proto;
7208 #ifdef CONFIG_SOCK_CGROUP_DATA
7209 case BPF_FUNC_skb_cgroup_id:
7210 return &bpf_skb_cgroup_id_proto;
7211 case BPF_FUNC_skb_ancestor_cgroup_id:
7212 return &bpf_skb_ancestor_cgroup_id_proto;
7215 case BPF_FUNC_sk_lookup_tcp:
7216 return &bpf_sk_lookup_tcp_proto;
7217 case BPF_FUNC_sk_lookup_udp:
7218 return &bpf_sk_lookup_udp_proto;
7219 case BPF_FUNC_sk_release:
7220 return &bpf_sk_release_proto;
7221 case BPF_FUNC_tcp_sock:
7222 return &bpf_tcp_sock_proto;
7223 case BPF_FUNC_get_listener_sock:
7224 return &bpf_get_listener_sock_proto;
7225 case BPF_FUNC_skc_lookup_tcp:
7226 return &bpf_skc_lookup_tcp_proto;
7227 case BPF_FUNC_tcp_check_syncookie:
7228 return &bpf_tcp_check_syncookie_proto;
7229 case BPF_FUNC_skb_ecn_set_ce:
7230 return &bpf_skb_ecn_set_ce_proto;
7231 case BPF_FUNC_tcp_gen_syncookie:
7232 return &bpf_tcp_gen_syncookie_proto;
7233 case BPF_FUNC_sk_assign:
7234 return &bpf_sk_assign_proto;
7237 return bpf_sk_base_func_proto(func_id);
7241 static const struct bpf_func_proto *
7242 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7245 case BPF_FUNC_perf_event_output:
7246 return &bpf_xdp_event_output_proto;
7247 case BPF_FUNC_get_smp_processor_id:
7248 return &bpf_get_smp_processor_id_proto;
7249 case BPF_FUNC_csum_diff:
7250 return &bpf_csum_diff_proto;
7251 case BPF_FUNC_xdp_adjust_head:
7252 return &bpf_xdp_adjust_head_proto;
7253 case BPF_FUNC_xdp_adjust_meta:
7254 return &bpf_xdp_adjust_meta_proto;
7255 case BPF_FUNC_redirect:
7256 return &bpf_xdp_redirect_proto;
7257 case BPF_FUNC_redirect_map:
7258 return &bpf_xdp_redirect_map_proto;
7259 case BPF_FUNC_xdp_adjust_tail:
7260 return &bpf_xdp_adjust_tail_proto;
7261 case BPF_FUNC_fib_lookup:
7262 return &bpf_xdp_fib_lookup_proto;
7264 case BPF_FUNC_sk_lookup_udp:
7265 return &bpf_xdp_sk_lookup_udp_proto;
7266 case BPF_FUNC_sk_lookup_tcp:
7267 return &bpf_xdp_sk_lookup_tcp_proto;
7268 case BPF_FUNC_sk_release:
7269 return &bpf_sk_release_proto;
7270 case BPF_FUNC_skc_lookup_tcp:
7271 return &bpf_xdp_skc_lookup_tcp_proto;
7272 case BPF_FUNC_tcp_check_syncookie:
7273 return &bpf_tcp_check_syncookie_proto;
7274 case BPF_FUNC_tcp_gen_syncookie:
7275 return &bpf_tcp_gen_syncookie_proto;
7278 return bpf_sk_base_func_proto(func_id);
7282 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
7283 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
7285 static const struct bpf_func_proto *
7286 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7289 case BPF_FUNC_setsockopt:
7290 return &bpf_sock_ops_setsockopt_proto;
7291 case BPF_FUNC_getsockopt:
7292 return &bpf_sock_ops_getsockopt_proto;
7293 case BPF_FUNC_sock_ops_cb_flags_set:
7294 return &bpf_sock_ops_cb_flags_set_proto;
7295 case BPF_FUNC_sock_map_update:
7296 return &bpf_sock_map_update_proto;
7297 case BPF_FUNC_sock_hash_update:
7298 return &bpf_sock_hash_update_proto;
7299 case BPF_FUNC_get_socket_cookie:
7300 return &bpf_get_socket_cookie_sock_ops_proto;
7301 case BPF_FUNC_get_local_storage:
7302 return &bpf_get_local_storage_proto;
7303 case BPF_FUNC_perf_event_output:
7304 return &bpf_event_output_data_proto;
7305 case BPF_FUNC_sk_storage_get:
7306 return &bpf_sk_storage_get_proto;
7307 case BPF_FUNC_sk_storage_delete:
7308 return &bpf_sk_storage_delete_proto;
7310 case BPF_FUNC_load_hdr_opt:
7311 return &bpf_sock_ops_load_hdr_opt_proto;
7312 case BPF_FUNC_store_hdr_opt:
7313 return &bpf_sock_ops_store_hdr_opt_proto;
7314 case BPF_FUNC_reserve_hdr_opt:
7315 return &bpf_sock_ops_reserve_hdr_opt_proto;
7316 case BPF_FUNC_tcp_sock:
7317 return &bpf_tcp_sock_proto;
7318 #endif /* CONFIG_INET */
7320 return bpf_sk_base_func_proto(func_id);
7324 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
7325 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
7327 static const struct bpf_func_proto *
7328 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7331 case BPF_FUNC_msg_redirect_map:
7332 return &bpf_msg_redirect_map_proto;
7333 case BPF_FUNC_msg_redirect_hash:
7334 return &bpf_msg_redirect_hash_proto;
7335 case BPF_FUNC_msg_apply_bytes:
7336 return &bpf_msg_apply_bytes_proto;
7337 case BPF_FUNC_msg_cork_bytes:
7338 return &bpf_msg_cork_bytes_proto;
7339 case BPF_FUNC_msg_pull_data:
7340 return &bpf_msg_pull_data_proto;
7341 case BPF_FUNC_msg_push_data:
7342 return &bpf_msg_push_data_proto;
7343 case BPF_FUNC_msg_pop_data:
7344 return &bpf_msg_pop_data_proto;
7345 case BPF_FUNC_perf_event_output:
7346 return &bpf_event_output_data_proto;
7347 case BPF_FUNC_get_current_uid_gid:
7348 return &bpf_get_current_uid_gid_proto;
7349 case BPF_FUNC_get_current_pid_tgid:
7350 return &bpf_get_current_pid_tgid_proto;
7351 case BPF_FUNC_sk_storage_get:
7352 return &bpf_sk_storage_get_proto;
7353 case BPF_FUNC_sk_storage_delete:
7354 return &bpf_sk_storage_delete_proto;
7355 #ifdef CONFIG_CGROUPS
7356 case BPF_FUNC_get_current_cgroup_id:
7357 return &bpf_get_current_cgroup_id_proto;
7358 case BPF_FUNC_get_current_ancestor_cgroup_id:
7359 return &bpf_get_current_ancestor_cgroup_id_proto;
7361 #ifdef CONFIG_CGROUP_NET_CLASSID
7362 case BPF_FUNC_get_cgroup_classid:
7363 return &bpf_get_cgroup_classid_curr_proto;
7366 return bpf_sk_base_func_proto(func_id);
7370 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
7371 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
7373 static const struct bpf_func_proto *
7374 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7377 case BPF_FUNC_skb_store_bytes:
7378 return &bpf_skb_store_bytes_proto;
7379 case BPF_FUNC_skb_load_bytes:
7380 return &bpf_skb_load_bytes_proto;
7381 case BPF_FUNC_skb_pull_data:
7382 return &sk_skb_pull_data_proto;
7383 case BPF_FUNC_skb_change_tail:
7384 return &sk_skb_change_tail_proto;
7385 case BPF_FUNC_skb_change_head:
7386 return &sk_skb_change_head_proto;
7387 case BPF_FUNC_skb_adjust_room:
7388 return &sk_skb_adjust_room_proto;
7389 case BPF_FUNC_get_socket_cookie:
7390 return &bpf_get_socket_cookie_proto;
7391 case BPF_FUNC_get_socket_uid:
7392 return &bpf_get_socket_uid_proto;
7393 case BPF_FUNC_sk_redirect_map:
7394 return &bpf_sk_redirect_map_proto;
7395 case BPF_FUNC_sk_redirect_hash:
7396 return &bpf_sk_redirect_hash_proto;
7397 case BPF_FUNC_perf_event_output:
7398 return &bpf_skb_event_output_proto;
7400 case BPF_FUNC_sk_lookup_tcp:
7401 return &bpf_sk_lookup_tcp_proto;
7402 case BPF_FUNC_sk_lookup_udp:
7403 return &bpf_sk_lookup_udp_proto;
7404 case BPF_FUNC_sk_release:
7405 return &bpf_sk_release_proto;
7406 case BPF_FUNC_skc_lookup_tcp:
7407 return &bpf_skc_lookup_tcp_proto;
7410 return bpf_sk_base_func_proto(func_id);
7414 static const struct bpf_func_proto *
7415 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7418 case BPF_FUNC_skb_load_bytes:
7419 return &bpf_flow_dissector_load_bytes_proto;
7421 return bpf_sk_base_func_proto(func_id);
7425 static const struct bpf_func_proto *
7426 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7429 case BPF_FUNC_skb_load_bytes:
7430 return &bpf_skb_load_bytes_proto;
7431 case BPF_FUNC_skb_pull_data:
7432 return &bpf_skb_pull_data_proto;
7433 case BPF_FUNC_csum_diff:
7434 return &bpf_csum_diff_proto;
7435 case BPF_FUNC_get_cgroup_classid:
7436 return &bpf_get_cgroup_classid_proto;
7437 case BPF_FUNC_get_route_realm:
7438 return &bpf_get_route_realm_proto;
7439 case BPF_FUNC_get_hash_recalc:
7440 return &bpf_get_hash_recalc_proto;
7441 case BPF_FUNC_perf_event_output:
7442 return &bpf_skb_event_output_proto;
7443 case BPF_FUNC_get_smp_processor_id:
7444 return &bpf_get_smp_processor_id_proto;
7445 case BPF_FUNC_skb_under_cgroup:
7446 return &bpf_skb_under_cgroup_proto;
7448 return bpf_sk_base_func_proto(func_id);
7452 static const struct bpf_func_proto *
7453 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7456 case BPF_FUNC_lwt_push_encap:
7457 return &bpf_lwt_in_push_encap_proto;
7459 return lwt_out_func_proto(func_id, prog);
7463 static const struct bpf_func_proto *
7464 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7467 case BPF_FUNC_skb_get_tunnel_key:
7468 return &bpf_skb_get_tunnel_key_proto;
7469 case BPF_FUNC_skb_set_tunnel_key:
7470 return bpf_get_skb_set_tunnel_proto(func_id);
7471 case BPF_FUNC_skb_get_tunnel_opt:
7472 return &bpf_skb_get_tunnel_opt_proto;
7473 case BPF_FUNC_skb_set_tunnel_opt:
7474 return bpf_get_skb_set_tunnel_proto(func_id);
7475 case BPF_FUNC_redirect:
7476 return &bpf_redirect_proto;
7477 case BPF_FUNC_clone_redirect:
7478 return &bpf_clone_redirect_proto;
7479 case BPF_FUNC_skb_change_tail:
7480 return &bpf_skb_change_tail_proto;
7481 case BPF_FUNC_skb_change_head:
7482 return &bpf_skb_change_head_proto;
7483 case BPF_FUNC_skb_store_bytes:
7484 return &bpf_skb_store_bytes_proto;
7485 case BPF_FUNC_csum_update:
7486 return &bpf_csum_update_proto;
7487 case BPF_FUNC_csum_level:
7488 return &bpf_csum_level_proto;
7489 case BPF_FUNC_l3_csum_replace:
7490 return &bpf_l3_csum_replace_proto;
7491 case BPF_FUNC_l4_csum_replace:
7492 return &bpf_l4_csum_replace_proto;
7493 case BPF_FUNC_set_hash_invalid:
7494 return &bpf_set_hash_invalid_proto;
7495 case BPF_FUNC_lwt_push_encap:
7496 return &bpf_lwt_xmit_push_encap_proto;
7498 return lwt_out_func_proto(func_id, prog);
7502 static const struct bpf_func_proto *
7503 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7506 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7507 case BPF_FUNC_lwt_seg6_store_bytes:
7508 return &bpf_lwt_seg6_store_bytes_proto;
7509 case BPF_FUNC_lwt_seg6_action:
7510 return &bpf_lwt_seg6_action_proto;
7511 case BPF_FUNC_lwt_seg6_adjust_srh:
7512 return &bpf_lwt_seg6_adjust_srh_proto;
7515 return lwt_out_func_proto(func_id, prog);
7519 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
7520 const struct bpf_prog *prog,
7521 struct bpf_insn_access_aux *info)
7523 const int size_default = sizeof(__u32);
7525 if (off < 0 || off >= sizeof(struct __sk_buff))
7528 /* The verifier guarantees that size > 0. */
7529 if (off % size != 0)
7533 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7534 if (off + size > offsetofend(struct __sk_buff, cb[4]))
7537 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
7538 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
7539 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
7540 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
7541 case bpf_ctx_range(struct __sk_buff, data):
7542 case bpf_ctx_range(struct __sk_buff, data_meta):
7543 case bpf_ctx_range(struct __sk_buff, data_end):
7544 if (size != size_default)
7547 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
7549 case bpf_ctx_range(struct __sk_buff, tstamp):
7550 if (size != sizeof(__u64))
7553 case offsetof(struct __sk_buff, sk):
7554 if (type == BPF_WRITE || size != sizeof(__u64))
7556 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
7559 /* Only narrow read access allowed for now. */
7560 if (type == BPF_WRITE) {
7561 if (size != size_default)
7564 bpf_ctx_record_field_size(info, size_default);
7565 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
7573 static bool sk_filter_is_valid_access(int off, int size,
7574 enum bpf_access_type type,
7575 const struct bpf_prog *prog,
7576 struct bpf_insn_access_aux *info)
7579 case bpf_ctx_range(struct __sk_buff, tc_classid):
7580 case bpf_ctx_range(struct __sk_buff, data):
7581 case bpf_ctx_range(struct __sk_buff, data_meta):
7582 case bpf_ctx_range(struct __sk_buff, data_end):
7583 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7584 case bpf_ctx_range(struct __sk_buff, tstamp):
7585 case bpf_ctx_range(struct __sk_buff, wire_len):
7589 if (type == BPF_WRITE) {
7591 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7598 return bpf_skb_is_valid_access(off, size, type, prog, info);
7601 static bool cg_skb_is_valid_access(int off, int size,
7602 enum bpf_access_type type,
7603 const struct bpf_prog *prog,
7604 struct bpf_insn_access_aux *info)
7607 case bpf_ctx_range(struct __sk_buff, tc_classid):
7608 case bpf_ctx_range(struct __sk_buff, data_meta):
7609 case bpf_ctx_range(struct __sk_buff, wire_len):
7611 case bpf_ctx_range(struct __sk_buff, data):
7612 case bpf_ctx_range(struct __sk_buff, data_end):
7618 if (type == BPF_WRITE) {
7620 case bpf_ctx_range(struct __sk_buff, mark):
7621 case bpf_ctx_range(struct __sk_buff, priority):
7622 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7624 case bpf_ctx_range(struct __sk_buff, tstamp):
7634 case bpf_ctx_range(struct __sk_buff, data):
7635 info->reg_type = PTR_TO_PACKET;
7637 case bpf_ctx_range(struct __sk_buff, data_end):
7638 info->reg_type = PTR_TO_PACKET_END;
7642 return bpf_skb_is_valid_access(off, size, type, prog, info);
7645 static bool lwt_is_valid_access(int off, int size,
7646 enum bpf_access_type type,
7647 const struct bpf_prog *prog,
7648 struct bpf_insn_access_aux *info)
7651 case bpf_ctx_range(struct __sk_buff, tc_classid):
7652 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7653 case bpf_ctx_range(struct __sk_buff, data_meta):
7654 case bpf_ctx_range(struct __sk_buff, tstamp):
7655 case bpf_ctx_range(struct __sk_buff, wire_len):
7659 if (type == BPF_WRITE) {
7661 case bpf_ctx_range(struct __sk_buff, mark):
7662 case bpf_ctx_range(struct __sk_buff, priority):
7663 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7671 case bpf_ctx_range(struct __sk_buff, data):
7672 info->reg_type = PTR_TO_PACKET;
7674 case bpf_ctx_range(struct __sk_buff, data_end):
7675 info->reg_type = PTR_TO_PACKET_END;
7679 return bpf_skb_is_valid_access(off, size, type, prog, info);
7682 /* Attach type specific accesses */
7683 static bool __sock_filter_check_attach_type(int off,
7684 enum bpf_access_type access_type,
7685 enum bpf_attach_type attach_type)
7688 case offsetof(struct bpf_sock, bound_dev_if):
7689 case offsetof(struct bpf_sock, mark):
7690 case offsetof(struct bpf_sock, priority):
7691 switch (attach_type) {
7692 case BPF_CGROUP_INET_SOCK_CREATE:
7693 case BPF_CGROUP_INET_SOCK_RELEASE:
7698 case bpf_ctx_range(struct bpf_sock, src_ip4):
7699 switch (attach_type) {
7700 case BPF_CGROUP_INET4_POST_BIND:
7705 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7706 switch (attach_type) {
7707 case BPF_CGROUP_INET6_POST_BIND:
7712 case bpf_ctx_range(struct bpf_sock, src_port):
7713 switch (attach_type) {
7714 case BPF_CGROUP_INET4_POST_BIND:
7715 case BPF_CGROUP_INET6_POST_BIND:
7722 return access_type == BPF_READ;
7727 bool bpf_sock_common_is_valid_access(int off, int size,
7728 enum bpf_access_type type,
7729 struct bpf_insn_access_aux *info)
7732 case bpf_ctx_range_till(struct bpf_sock, type, priority):
7735 return bpf_sock_is_valid_access(off, size, type, info);
7739 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7740 struct bpf_insn_access_aux *info)
7742 const int size_default = sizeof(__u32);
7745 if (off < 0 || off >= sizeof(struct bpf_sock))
7747 if (off % size != 0)
7751 case offsetof(struct bpf_sock, state):
7752 case offsetof(struct bpf_sock, family):
7753 case offsetof(struct bpf_sock, type):
7754 case offsetof(struct bpf_sock, protocol):
7755 case offsetof(struct bpf_sock, src_port):
7756 case offsetof(struct bpf_sock, rx_queue_mapping):
7757 case bpf_ctx_range(struct bpf_sock, src_ip4):
7758 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7759 case bpf_ctx_range(struct bpf_sock, dst_ip4):
7760 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
7761 bpf_ctx_record_field_size(info, size_default);
7762 return bpf_ctx_narrow_access_ok(off, size, size_default);
7763 case bpf_ctx_range(struct bpf_sock, dst_port):
7764 field_size = size == size_default ?
7765 size_default : sizeof_field(struct bpf_sock, dst_port);
7766 bpf_ctx_record_field_size(info, field_size);
7767 return bpf_ctx_narrow_access_ok(off, size, field_size);
7768 case offsetofend(struct bpf_sock, dst_port) ...
7769 offsetof(struct bpf_sock, dst_ip4) - 1:
7773 return size == size_default;
7776 static bool sock_filter_is_valid_access(int off, int size,
7777 enum bpf_access_type type,
7778 const struct bpf_prog *prog,
7779 struct bpf_insn_access_aux *info)
7781 if (!bpf_sock_is_valid_access(off, size, type, info))
7783 return __sock_filter_check_attach_type(off, type,
7784 prog->expected_attach_type);
7787 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
7788 const struct bpf_prog *prog)
7790 /* Neither direct read nor direct write requires any preliminary
7796 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
7797 const struct bpf_prog *prog, int drop_verdict)
7799 struct bpf_insn *insn = insn_buf;
7804 /* if (!skb->cloned)
7807 * (Fast-path, otherwise approximation that we might be
7808 * a clone, do the rest in helper.)
7810 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
7811 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
7812 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
7814 /* ret = bpf_skb_pull_data(skb, 0); */
7815 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
7816 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
7817 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
7818 BPF_FUNC_skb_pull_data);
7821 * return TC_ACT_SHOT;
7823 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
7824 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
7825 *insn++ = BPF_EXIT_INSN();
7828 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
7830 *insn++ = prog->insnsi[0];
7832 return insn - insn_buf;
7835 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
7836 struct bpf_insn *insn_buf)
7838 bool indirect = BPF_MODE(orig->code) == BPF_IND;
7839 struct bpf_insn *insn = insn_buf;
7842 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
7844 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
7846 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
7848 /* We're guaranteed here that CTX is in R6. */
7849 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
7851 switch (BPF_SIZE(orig->code)) {
7853 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
7856 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
7859 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
7863 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
7864 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
7865 *insn++ = BPF_EXIT_INSN();
7867 return insn - insn_buf;
7870 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
7871 const struct bpf_prog *prog)
7873 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
7876 static bool tc_cls_act_is_valid_access(int off, int size,
7877 enum bpf_access_type type,
7878 const struct bpf_prog *prog,
7879 struct bpf_insn_access_aux *info)
7881 if (type == BPF_WRITE) {
7883 case bpf_ctx_range(struct __sk_buff, mark):
7884 case bpf_ctx_range(struct __sk_buff, tc_index):
7885 case bpf_ctx_range(struct __sk_buff, priority):
7886 case bpf_ctx_range(struct __sk_buff, tc_classid):
7887 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7888 case bpf_ctx_range(struct __sk_buff, tstamp):
7889 case bpf_ctx_range(struct __sk_buff, queue_mapping):
7897 case bpf_ctx_range(struct __sk_buff, data):
7898 info->reg_type = PTR_TO_PACKET;
7900 case bpf_ctx_range(struct __sk_buff, data_meta):
7901 info->reg_type = PTR_TO_PACKET_META;
7903 case bpf_ctx_range(struct __sk_buff, data_end):
7904 info->reg_type = PTR_TO_PACKET_END;
7906 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7910 return bpf_skb_is_valid_access(off, size, type, prog, info);
7913 static bool __is_valid_xdp_access(int off, int size)
7915 if (off < 0 || off >= sizeof(struct xdp_md))
7917 if (off % size != 0)
7919 if (size != sizeof(__u32))
7925 static bool xdp_is_valid_access(int off, int size,
7926 enum bpf_access_type type,
7927 const struct bpf_prog *prog,
7928 struct bpf_insn_access_aux *info)
7930 if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
7932 case offsetof(struct xdp_md, egress_ifindex):
7937 if (type == BPF_WRITE) {
7938 if (bpf_prog_is_dev_bound(prog->aux)) {
7940 case offsetof(struct xdp_md, rx_queue_index):
7941 return __is_valid_xdp_access(off, size);
7948 case offsetof(struct xdp_md, data):
7949 info->reg_type = PTR_TO_PACKET;
7951 case offsetof(struct xdp_md, data_meta):
7952 info->reg_type = PTR_TO_PACKET_META;
7954 case offsetof(struct xdp_md, data_end):
7955 info->reg_type = PTR_TO_PACKET_END;
7959 return __is_valid_xdp_access(off, size);
7962 void bpf_warn_invalid_xdp_action(u32 act)
7964 const u32 act_max = XDP_REDIRECT;
7966 pr_warn_once("%s XDP return value %u, expect packet loss!\n",
7967 act > act_max ? "Illegal" : "Driver unsupported",
7970 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
7972 static bool sock_addr_is_valid_access(int off, int size,
7973 enum bpf_access_type type,
7974 const struct bpf_prog *prog,
7975 struct bpf_insn_access_aux *info)
7977 const int size_default = sizeof(__u32);
7979 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
7981 if (off % size != 0)
7984 /* Disallow access to IPv6 fields from IPv4 contex and vise
7988 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
7989 switch (prog->expected_attach_type) {
7990 case BPF_CGROUP_INET4_BIND:
7991 case BPF_CGROUP_INET4_CONNECT:
7992 case BPF_CGROUP_INET4_GETPEERNAME:
7993 case BPF_CGROUP_INET4_GETSOCKNAME:
7994 case BPF_CGROUP_UDP4_SENDMSG:
7995 case BPF_CGROUP_UDP4_RECVMSG:
8001 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8002 switch (prog->expected_attach_type) {
8003 case BPF_CGROUP_INET6_BIND:
8004 case BPF_CGROUP_INET6_CONNECT:
8005 case BPF_CGROUP_INET6_GETPEERNAME:
8006 case BPF_CGROUP_INET6_GETSOCKNAME:
8007 case BPF_CGROUP_UDP6_SENDMSG:
8008 case BPF_CGROUP_UDP6_RECVMSG:
8014 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8015 switch (prog->expected_attach_type) {
8016 case BPF_CGROUP_UDP4_SENDMSG:
8022 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8024 switch (prog->expected_attach_type) {
8025 case BPF_CGROUP_UDP6_SENDMSG:
8034 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8035 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8036 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8037 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8039 case bpf_ctx_range(struct bpf_sock_addr, user_port):
8040 if (type == BPF_READ) {
8041 bpf_ctx_record_field_size(info, size_default);
8043 if (bpf_ctx_wide_access_ok(off, size,
8044 struct bpf_sock_addr,
8048 if (bpf_ctx_wide_access_ok(off, size,
8049 struct bpf_sock_addr,
8053 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8056 if (bpf_ctx_wide_access_ok(off, size,
8057 struct bpf_sock_addr,
8061 if (bpf_ctx_wide_access_ok(off, size,
8062 struct bpf_sock_addr,
8066 if (size != size_default)
8070 case offsetof(struct bpf_sock_addr, sk):
8071 if (type != BPF_READ)
8073 if (size != sizeof(__u64))
8075 info->reg_type = PTR_TO_SOCKET;
8078 if (type == BPF_READ) {
8079 if (size != size_default)
8089 static bool sock_ops_is_valid_access(int off, int size,
8090 enum bpf_access_type type,
8091 const struct bpf_prog *prog,
8092 struct bpf_insn_access_aux *info)
8094 const int size_default = sizeof(__u32);
8096 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
8099 /* The verifier guarantees that size > 0. */
8100 if (off % size != 0)
8103 if (type == BPF_WRITE) {
8105 case offsetof(struct bpf_sock_ops, reply):
8106 case offsetof(struct bpf_sock_ops, sk_txhash):
8107 if (size != size_default)
8115 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
8117 if (size != sizeof(__u64))
8120 case offsetof(struct bpf_sock_ops, sk):
8121 if (size != sizeof(__u64))
8123 info->reg_type = PTR_TO_SOCKET_OR_NULL;
8125 case offsetof(struct bpf_sock_ops, skb_data):
8126 if (size != sizeof(__u64))
8128 info->reg_type = PTR_TO_PACKET;
8130 case offsetof(struct bpf_sock_ops, skb_data_end):
8131 if (size != sizeof(__u64))
8133 info->reg_type = PTR_TO_PACKET_END;
8135 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
8136 bpf_ctx_record_field_size(info, size_default);
8137 return bpf_ctx_narrow_access_ok(off, size,
8140 if (size != size_default)
8149 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
8150 const struct bpf_prog *prog)
8152 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
8155 static bool sk_skb_is_valid_access(int off, int size,
8156 enum bpf_access_type type,
8157 const struct bpf_prog *prog,
8158 struct bpf_insn_access_aux *info)
8161 case bpf_ctx_range(struct __sk_buff, tc_classid):
8162 case bpf_ctx_range(struct __sk_buff, data_meta):
8163 case bpf_ctx_range(struct __sk_buff, tstamp):
8164 case bpf_ctx_range(struct __sk_buff, wire_len):
8168 if (type == BPF_WRITE) {
8170 case bpf_ctx_range(struct __sk_buff, tc_index):
8171 case bpf_ctx_range(struct __sk_buff, priority):
8179 case bpf_ctx_range(struct __sk_buff, mark):
8181 case bpf_ctx_range(struct __sk_buff, data):
8182 info->reg_type = PTR_TO_PACKET;
8184 case bpf_ctx_range(struct __sk_buff, data_end):
8185 info->reg_type = PTR_TO_PACKET_END;
8189 return bpf_skb_is_valid_access(off, size, type, prog, info);
8192 static bool sk_msg_is_valid_access(int off, int size,
8193 enum bpf_access_type type,
8194 const struct bpf_prog *prog,
8195 struct bpf_insn_access_aux *info)
8197 if (type == BPF_WRITE)
8200 if (off % size != 0)
8204 case offsetof(struct sk_msg_md, data):
8205 info->reg_type = PTR_TO_PACKET;
8206 if (size != sizeof(__u64))
8209 case offsetof(struct sk_msg_md, data_end):
8210 info->reg_type = PTR_TO_PACKET_END;
8211 if (size != sizeof(__u64))
8214 case offsetof(struct sk_msg_md, sk):
8215 if (size != sizeof(__u64))
8217 info->reg_type = PTR_TO_SOCKET;
8219 case bpf_ctx_range(struct sk_msg_md, family):
8220 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
8221 case bpf_ctx_range(struct sk_msg_md, local_ip4):
8222 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
8223 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
8224 case bpf_ctx_range(struct sk_msg_md, remote_port):
8225 case bpf_ctx_range(struct sk_msg_md, local_port):
8226 case bpf_ctx_range(struct sk_msg_md, size):
8227 if (size != sizeof(__u32))
8236 static bool flow_dissector_is_valid_access(int off, int size,
8237 enum bpf_access_type type,
8238 const struct bpf_prog *prog,
8239 struct bpf_insn_access_aux *info)
8241 const int size_default = sizeof(__u32);
8243 if (off < 0 || off >= sizeof(struct __sk_buff))
8246 if (type == BPF_WRITE)
8250 case bpf_ctx_range(struct __sk_buff, data):
8251 if (size != size_default)
8253 info->reg_type = PTR_TO_PACKET;
8255 case bpf_ctx_range(struct __sk_buff, data_end):
8256 if (size != size_default)
8258 info->reg_type = PTR_TO_PACKET_END;
8260 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8261 if (size != sizeof(__u64))
8263 info->reg_type = PTR_TO_FLOW_KEYS;
8270 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
8271 const struct bpf_insn *si,
8272 struct bpf_insn *insn_buf,
8273 struct bpf_prog *prog,
8277 struct bpf_insn *insn = insn_buf;
8280 case offsetof(struct __sk_buff, data):
8281 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
8282 si->dst_reg, si->src_reg,
8283 offsetof(struct bpf_flow_dissector, data));
8286 case offsetof(struct __sk_buff, data_end):
8287 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
8288 si->dst_reg, si->src_reg,
8289 offsetof(struct bpf_flow_dissector, data_end));
8292 case offsetof(struct __sk_buff, flow_keys):
8293 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
8294 si->dst_reg, si->src_reg,
8295 offsetof(struct bpf_flow_dissector, flow_keys));
8299 return insn - insn_buf;
8302 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si,
8303 struct bpf_insn *insn)
8305 /* si->dst_reg = skb_shinfo(SKB); */
8306 #ifdef NET_SKBUFF_DATA_USES_OFFSET
8307 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8308 BPF_REG_AX, si->src_reg,
8309 offsetof(struct sk_buff, end));
8310 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
8311 si->dst_reg, si->src_reg,
8312 offsetof(struct sk_buff, head));
8313 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
8315 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8316 si->dst_reg, si->src_reg,
8317 offsetof(struct sk_buff, end));
8323 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
8324 const struct bpf_insn *si,
8325 struct bpf_insn *insn_buf,
8326 struct bpf_prog *prog, u32 *target_size)
8328 struct bpf_insn *insn = insn_buf;
8332 case offsetof(struct __sk_buff, len):
8333 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8334 bpf_target_off(struct sk_buff, len, 4,
8338 case offsetof(struct __sk_buff, protocol):
8339 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8340 bpf_target_off(struct sk_buff, protocol, 2,
8344 case offsetof(struct __sk_buff, vlan_proto):
8345 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8346 bpf_target_off(struct sk_buff, vlan_proto, 2,
8350 case offsetof(struct __sk_buff, priority):
8351 if (type == BPF_WRITE)
8352 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8353 bpf_target_off(struct sk_buff, priority, 4,
8356 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8357 bpf_target_off(struct sk_buff, priority, 4,
8361 case offsetof(struct __sk_buff, ingress_ifindex):
8362 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8363 bpf_target_off(struct sk_buff, skb_iif, 4,
8367 case offsetof(struct __sk_buff, ifindex):
8368 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8369 si->dst_reg, si->src_reg,
8370 offsetof(struct sk_buff, dev));
8371 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
8372 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8373 bpf_target_off(struct net_device, ifindex, 4,
8377 case offsetof(struct __sk_buff, hash):
8378 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8379 bpf_target_off(struct sk_buff, hash, 4,
8383 case offsetof(struct __sk_buff, mark):
8384 if (type == BPF_WRITE)
8385 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8386 bpf_target_off(struct sk_buff, mark, 4,
8389 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8390 bpf_target_off(struct sk_buff, mark, 4,
8394 case offsetof(struct __sk_buff, pkt_type):
8396 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8398 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
8399 #ifdef __BIG_ENDIAN_BITFIELD
8400 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
8404 case offsetof(struct __sk_buff, queue_mapping):
8405 if (type == BPF_WRITE) {
8406 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
8407 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8408 bpf_target_off(struct sk_buff,
8412 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8413 bpf_target_off(struct sk_buff,
8419 case offsetof(struct __sk_buff, vlan_present):
8421 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8422 PKT_VLAN_PRESENT_OFFSET());
8423 if (PKT_VLAN_PRESENT_BIT)
8424 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
8425 if (PKT_VLAN_PRESENT_BIT < 7)
8426 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
8429 case offsetof(struct __sk_buff, vlan_tci):
8430 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8431 bpf_target_off(struct sk_buff, vlan_tci, 2,
8435 case offsetof(struct __sk_buff, cb[0]) ...
8436 offsetofend(struct __sk_buff, cb[4]) - 1:
8437 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
8438 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
8439 offsetof(struct qdisc_skb_cb, data)) %
8442 prog->cb_access = 1;
8444 off -= offsetof(struct __sk_buff, cb[0]);
8445 off += offsetof(struct sk_buff, cb);
8446 off += offsetof(struct qdisc_skb_cb, data);
8447 if (type == BPF_WRITE)
8448 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
8451 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
8455 case offsetof(struct __sk_buff, tc_classid):
8456 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
8459 off -= offsetof(struct __sk_buff, tc_classid);
8460 off += offsetof(struct sk_buff, cb);
8461 off += offsetof(struct qdisc_skb_cb, tc_classid);
8463 if (type == BPF_WRITE)
8464 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
8467 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
8471 case offsetof(struct __sk_buff, data):
8472 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
8473 si->dst_reg, si->src_reg,
8474 offsetof(struct sk_buff, data));
8477 case offsetof(struct __sk_buff, data_meta):
8479 off -= offsetof(struct __sk_buff, data_meta);
8480 off += offsetof(struct sk_buff, cb);
8481 off += offsetof(struct bpf_skb_data_end, data_meta);
8482 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8486 case offsetof(struct __sk_buff, data_end):
8488 off -= offsetof(struct __sk_buff, data_end);
8489 off += offsetof(struct sk_buff, cb);
8490 off += offsetof(struct bpf_skb_data_end, data_end);
8491 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8495 case offsetof(struct __sk_buff, tc_index):
8496 #ifdef CONFIG_NET_SCHED
8497 if (type == BPF_WRITE)
8498 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8499 bpf_target_off(struct sk_buff, tc_index, 2,
8502 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8503 bpf_target_off(struct sk_buff, tc_index, 2,
8507 if (type == BPF_WRITE)
8508 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
8510 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8514 case offsetof(struct __sk_buff, napi_id):
8515 #if defined(CONFIG_NET_RX_BUSY_POLL)
8516 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8517 bpf_target_off(struct sk_buff, napi_id, 4,
8519 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
8520 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8523 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8526 case offsetof(struct __sk_buff, family):
8527 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
8529 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8530 si->dst_reg, si->src_reg,
8531 offsetof(struct sk_buff, sk));
8532 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8533 bpf_target_off(struct sock_common,
8537 case offsetof(struct __sk_buff, remote_ip4):
8538 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
8540 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8541 si->dst_reg, si->src_reg,
8542 offsetof(struct sk_buff, sk));
8543 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8544 bpf_target_off(struct sock_common,
8548 case offsetof(struct __sk_buff, local_ip4):
8549 BUILD_BUG_ON(sizeof_field(struct sock_common,
8550 skc_rcv_saddr) != 4);
8552 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8553 si->dst_reg, si->src_reg,
8554 offsetof(struct sk_buff, sk));
8555 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8556 bpf_target_off(struct sock_common,
8560 case offsetof(struct __sk_buff, remote_ip6[0]) ...
8561 offsetof(struct __sk_buff, remote_ip6[3]):
8562 #if IS_ENABLED(CONFIG_IPV6)
8563 BUILD_BUG_ON(sizeof_field(struct sock_common,
8564 skc_v6_daddr.s6_addr32[0]) != 4);
8567 off -= offsetof(struct __sk_buff, remote_ip6[0]);
8569 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8570 si->dst_reg, si->src_reg,
8571 offsetof(struct sk_buff, sk));
8572 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8573 offsetof(struct sock_common,
8574 skc_v6_daddr.s6_addr32[0]) +
8577 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8580 case offsetof(struct __sk_buff, local_ip6[0]) ...
8581 offsetof(struct __sk_buff, local_ip6[3]):
8582 #if IS_ENABLED(CONFIG_IPV6)
8583 BUILD_BUG_ON(sizeof_field(struct sock_common,
8584 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8587 off -= offsetof(struct __sk_buff, local_ip6[0]);
8589 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8590 si->dst_reg, si->src_reg,
8591 offsetof(struct sk_buff, sk));
8592 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8593 offsetof(struct sock_common,
8594 skc_v6_rcv_saddr.s6_addr32[0]) +
8597 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8601 case offsetof(struct __sk_buff, remote_port):
8602 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
8604 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8605 si->dst_reg, si->src_reg,
8606 offsetof(struct sk_buff, sk));
8607 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8608 bpf_target_off(struct sock_common,
8611 #ifndef __BIG_ENDIAN_BITFIELD
8612 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8616 case offsetof(struct __sk_buff, local_port):
8617 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
8619 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8620 si->dst_reg, si->src_reg,
8621 offsetof(struct sk_buff, sk));
8622 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8623 bpf_target_off(struct sock_common,
8624 skc_num, 2, target_size));
8627 case offsetof(struct __sk_buff, tstamp):
8628 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
8630 if (type == BPF_WRITE)
8631 *insn++ = BPF_STX_MEM(BPF_DW,
8632 si->dst_reg, si->src_reg,
8633 bpf_target_off(struct sk_buff,
8637 *insn++ = BPF_LDX_MEM(BPF_DW,
8638 si->dst_reg, si->src_reg,
8639 bpf_target_off(struct sk_buff,
8644 case offsetof(struct __sk_buff, gso_segs):
8645 insn = bpf_convert_shinfo_access(si, insn);
8646 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
8647 si->dst_reg, si->dst_reg,
8648 bpf_target_off(struct skb_shared_info,
8652 case offsetof(struct __sk_buff, gso_size):
8653 insn = bpf_convert_shinfo_access(si, insn);
8654 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
8655 si->dst_reg, si->dst_reg,
8656 bpf_target_off(struct skb_shared_info,
8660 case offsetof(struct __sk_buff, wire_len):
8661 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
8664 off -= offsetof(struct __sk_buff, wire_len);
8665 off += offsetof(struct sk_buff, cb);
8666 off += offsetof(struct qdisc_skb_cb, pkt_len);
8668 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
8671 case offsetof(struct __sk_buff, sk):
8672 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8673 si->dst_reg, si->src_reg,
8674 offsetof(struct sk_buff, sk));
8678 return insn - insn_buf;
8681 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
8682 const struct bpf_insn *si,
8683 struct bpf_insn *insn_buf,
8684 struct bpf_prog *prog, u32 *target_size)
8686 struct bpf_insn *insn = insn_buf;
8690 case offsetof(struct bpf_sock, bound_dev_if):
8691 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
8693 if (type == BPF_WRITE)
8694 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8695 offsetof(struct sock, sk_bound_dev_if));
8697 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8698 offsetof(struct sock, sk_bound_dev_if));
8701 case offsetof(struct bpf_sock, mark):
8702 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
8704 if (type == BPF_WRITE)
8705 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8706 offsetof(struct sock, sk_mark));
8708 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8709 offsetof(struct sock, sk_mark));
8712 case offsetof(struct bpf_sock, priority):
8713 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
8715 if (type == BPF_WRITE)
8716 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8717 offsetof(struct sock, sk_priority));
8719 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8720 offsetof(struct sock, sk_priority));
8723 case offsetof(struct bpf_sock, family):
8724 *insn++ = BPF_LDX_MEM(
8725 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
8726 si->dst_reg, si->src_reg,
8727 bpf_target_off(struct sock_common,
8729 sizeof_field(struct sock_common,
8734 case offsetof(struct bpf_sock, type):
8735 *insn++ = BPF_LDX_MEM(
8736 BPF_FIELD_SIZEOF(struct sock, sk_type),
8737 si->dst_reg, si->src_reg,
8738 bpf_target_off(struct sock, sk_type,
8739 sizeof_field(struct sock, sk_type),
8743 case offsetof(struct bpf_sock, protocol):
8744 *insn++ = BPF_LDX_MEM(
8745 BPF_FIELD_SIZEOF(struct sock, sk_protocol),
8746 si->dst_reg, si->src_reg,
8747 bpf_target_off(struct sock, sk_protocol,
8748 sizeof_field(struct sock, sk_protocol),
8752 case offsetof(struct bpf_sock, src_ip4):
8753 *insn++ = BPF_LDX_MEM(
8754 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8755 bpf_target_off(struct sock_common, skc_rcv_saddr,
8756 sizeof_field(struct sock_common,
8761 case offsetof(struct bpf_sock, dst_ip4):
8762 *insn++ = BPF_LDX_MEM(
8763 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8764 bpf_target_off(struct sock_common, skc_daddr,
8765 sizeof_field(struct sock_common,
8770 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8771 #if IS_ENABLED(CONFIG_IPV6)
8773 off -= offsetof(struct bpf_sock, src_ip6[0]);
8774 *insn++ = BPF_LDX_MEM(
8775 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8778 skc_v6_rcv_saddr.s6_addr32[0],
8779 sizeof_field(struct sock_common,
8780 skc_v6_rcv_saddr.s6_addr32[0]),
8781 target_size) + off);
8784 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8788 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8789 #if IS_ENABLED(CONFIG_IPV6)
8791 off -= offsetof(struct bpf_sock, dst_ip6[0]);
8792 *insn++ = BPF_LDX_MEM(
8793 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8794 bpf_target_off(struct sock_common,
8795 skc_v6_daddr.s6_addr32[0],
8796 sizeof_field(struct sock_common,
8797 skc_v6_daddr.s6_addr32[0]),
8798 target_size) + off);
8800 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8805 case offsetof(struct bpf_sock, src_port):
8806 *insn++ = BPF_LDX_MEM(
8807 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
8808 si->dst_reg, si->src_reg,
8809 bpf_target_off(struct sock_common, skc_num,
8810 sizeof_field(struct sock_common,
8815 case offsetof(struct bpf_sock, dst_port):
8816 *insn++ = BPF_LDX_MEM(
8817 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
8818 si->dst_reg, si->src_reg,
8819 bpf_target_off(struct sock_common, skc_dport,
8820 sizeof_field(struct sock_common,
8825 case offsetof(struct bpf_sock, state):
8826 *insn++ = BPF_LDX_MEM(
8827 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
8828 si->dst_reg, si->src_reg,
8829 bpf_target_off(struct sock_common, skc_state,
8830 sizeof_field(struct sock_common,
8834 case offsetof(struct bpf_sock, rx_queue_mapping):
8836 *insn++ = BPF_LDX_MEM(
8837 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
8838 si->dst_reg, si->src_reg,
8839 bpf_target_off(struct sock, sk_rx_queue_mapping,
8840 sizeof_field(struct sock,
8841 sk_rx_queue_mapping),
8843 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
8845 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
8847 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
8853 return insn - insn_buf;
8856 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
8857 const struct bpf_insn *si,
8858 struct bpf_insn *insn_buf,
8859 struct bpf_prog *prog, u32 *target_size)
8861 struct bpf_insn *insn = insn_buf;
8864 case offsetof(struct __sk_buff, ifindex):
8865 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8866 si->dst_reg, si->src_reg,
8867 offsetof(struct sk_buff, dev));
8868 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8869 bpf_target_off(struct net_device, ifindex, 4,
8873 return bpf_convert_ctx_access(type, si, insn_buf, prog,
8877 return insn - insn_buf;
8880 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
8881 const struct bpf_insn *si,
8882 struct bpf_insn *insn_buf,
8883 struct bpf_prog *prog, u32 *target_size)
8885 struct bpf_insn *insn = insn_buf;
8888 case offsetof(struct xdp_md, data):
8889 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
8890 si->dst_reg, si->src_reg,
8891 offsetof(struct xdp_buff, data));
8893 case offsetof(struct xdp_md, data_meta):
8894 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
8895 si->dst_reg, si->src_reg,
8896 offsetof(struct xdp_buff, data_meta));
8898 case offsetof(struct xdp_md, data_end):
8899 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
8900 si->dst_reg, si->src_reg,
8901 offsetof(struct xdp_buff, data_end));
8903 case offsetof(struct xdp_md, ingress_ifindex):
8904 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
8905 si->dst_reg, si->src_reg,
8906 offsetof(struct xdp_buff, rxq));
8907 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
8908 si->dst_reg, si->dst_reg,
8909 offsetof(struct xdp_rxq_info, dev));
8910 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8911 offsetof(struct net_device, ifindex));
8913 case offsetof(struct xdp_md, rx_queue_index):
8914 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
8915 si->dst_reg, si->src_reg,
8916 offsetof(struct xdp_buff, rxq));
8917 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8918 offsetof(struct xdp_rxq_info,
8921 case offsetof(struct xdp_md, egress_ifindex):
8922 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
8923 si->dst_reg, si->src_reg,
8924 offsetof(struct xdp_buff, txq));
8925 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
8926 si->dst_reg, si->dst_reg,
8927 offsetof(struct xdp_txq_info, dev));
8928 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8929 offsetof(struct net_device, ifindex));
8933 return insn - insn_buf;
8936 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
8937 * context Structure, F is Field in context structure that contains a pointer
8938 * to Nested Structure of type NS that has the field NF.
8940 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
8941 * sure that SIZE is not greater than actual size of S.F.NF.
8943 * If offset OFF is provided, the load happens from that offset relative to
8946 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
8948 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
8949 si->src_reg, offsetof(S, F)); \
8950 *insn++ = BPF_LDX_MEM( \
8951 SIZE, si->dst_reg, si->dst_reg, \
8952 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
8957 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
8958 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
8959 BPF_FIELD_SIZEOF(NS, NF), 0)
8961 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
8962 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
8964 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
8965 * "register" since two registers available in convert_ctx_access are not
8966 * enough: we can't override neither SRC, since it contains value to store, nor
8967 * DST since it contains pointer to context that may be used by later
8968 * instructions. But we need a temporary place to save pointer to nested
8969 * structure whose field we want to store to.
8971 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
8973 int tmp_reg = BPF_REG_9; \
8974 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
8976 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
8978 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
8980 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
8981 si->dst_reg, offsetof(S, F)); \
8982 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \
8983 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
8986 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
8990 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
8993 if (type == BPF_WRITE) { \
8994 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
8997 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
8998 S, NS, F, NF, SIZE, OFF); \
9002 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
9003 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
9004 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
9006 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
9007 const struct bpf_insn *si,
9008 struct bpf_insn *insn_buf,
9009 struct bpf_prog *prog, u32 *target_size)
9011 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
9012 struct bpf_insn *insn = insn_buf;
9015 case offsetof(struct bpf_sock_addr, user_family):
9016 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9017 struct sockaddr, uaddr, sa_family);
9020 case offsetof(struct bpf_sock_addr, user_ip4):
9021 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9022 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
9023 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
9026 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9028 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
9029 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9030 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9031 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
9035 case offsetof(struct bpf_sock_addr, user_port):
9036 /* To get port we need to know sa_family first and then treat
9037 * sockaddr as either sockaddr_in or sockaddr_in6.
9038 * Though we can simplify since port field has same offset and
9039 * size in both structures.
9040 * Here we check this invariant and use just one of the
9041 * structures if it's true.
9043 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
9044 offsetof(struct sockaddr_in6, sin6_port));
9045 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
9046 sizeof_field(struct sockaddr_in6, sin6_port));
9047 /* Account for sin6_port being smaller than user_port. */
9048 port_size = min(port_size, BPF_LDST_BYTES(si));
9049 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9050 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9051 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
9054 case offsetof(struct bpf_sock_addr, family):
9055 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9056 struct sock, sk, sk_family);
9059 case offsetof(struct bpf_sock_addr, type):
9060 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9061 struct sock, sk, sk_type);
9064 case offsetof(struct bpf_sock_addr, protocol):
9065 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9066 struct sock, sk, sk_protocol);
9069 case offsetof(struct bpf_sock_addr, msg_src_ip4):
9070 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
9071 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9072 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
9073 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
9076 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9079 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
9080 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
9081 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9082 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
9083 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
9085 case offsetof(struct bpf_sock_addr, sk):
9086 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
9087 si->dst_reg, si->src_reg,
9088 offsetof(struct bpf_sock_addr_kern, sk));
9092 return insn - insn_buf;
9095 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
9096 const struct bpf_insn *si,
9097 struct bpf_insn *insn_buf,
9098 struct bpf_prog *prog,
9101 struct bpf_insn *insn = insn_buf;
9104 /* Helper macro for adding read access to tcp_sock or sock fields. */
9105 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
9107 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \
9108 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
9109 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
9110 if (si->dst_reg == reg || si->src_reg == reg) \
9112 if (si->dst_reg == reg || si->src_reg == reg) \
9114 if (si->dst_reg == si->src_reg) { \
9115 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
9116 offsetof(struct bpf_sock_ops_kern, \
9118 fullsock_reg = reg; \
9121 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9122 struct bpf_sock_ops_kern, \
9124 fullsock_reg, si->src_reg, \
9125 offsetof(struct bpf_sock_ops_kern, \
9127 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
9128 if (si->dst_reg == si->src_reg) \
9129 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9130 offsetof(struct bpf_sock_ops_kern, \
9132 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9133 struct bpf_sock_ops_kern, sk),\
9134 si->dst_reg, si->src_reg, \
9135 offsetof(struct bpf_sock_ops_kern, sk));\
9136 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
9138 si->dst_reg, si->dst_reg, \
9139 offsetof(OBJ, OBJ_FIELD)); \
9140 if (si->dst_reg == si->src_reg) { \
9141 *insn++ = BPF_JMP_A(1); \
9142 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9143 offsetof(struct bpf_sock_ops_kern, \
9148 #define SOCK_OPS_GET_SK() \
9150 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \
9151 if (si->dst_reg == reg || si->src_reg == reg) \
9153 if (si->dst_reg == reg || si->src_reg == reg) \
9155 if (si->dst_reg == si->src_reg) { \
9156 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
9157 offsetof(struct bpf_sock_ops_kern, \
9159 fullsock_reg = reg; \
9162 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9163 struct bpf_sock_ops_kern, \
9165 fullsock_reg, si->src_reg, \
9166 offsetof(struct bpf_sock_ops_kern, \
9168 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
9169 if (si->dst_reg == si->src_reg) \
9170 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9171 offsetof(struct bpf_sock_ops_kern, \
9173 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9174 struct bpf_sock_ops_kern, sk),\
9175 si->dst_reg, si->src_reg, \
9176 offsetof(struct bpf_sock_ops_kern, sk));\
9177 if (si->dst_reg == si->src_reg) { \
9178 *insn++ = BPF_JMP_A(1); \
9179 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9180 offsetof(struct bpf_sock_ops_kern, \
9185 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
9186 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
9188 /* Helper macro for adding write access to tcp_sock or sock fields.
9189 * The macro is called with two registers, dst_reg which contains a pointer
9190 * to ctx (context) and src_reg which contains the value that should be
9191 * stored. However, we need an additional register since we cannot overwrite
9192 * dst_reg because it may be used later in the program.
9193 * Instead we "borrow" one of the other register. We first save its value
9194 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
9195 * it at the end of the macro.
9197 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
9199 int reg = BPF_REG_9; \
9200 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
9201 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
9202 if (si->dst_reg == reg || si->src_reg == reg) \
9204 if (si->dst_reg == reg || si->src_reg == reg) \
9206 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
9207 offsetof(struct bpf_sock_ops_kern, \
9209 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9210 struct bpf_sock_ops_kern, \
9213 offsetof(struct bpf_sock_ops_kern, \
9215 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
9216 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9217 struct bpf_sock_ops_kern, sk),\
9219 offsetof(struct bpf_sock_ops_kern, sk));\
9220 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \
9222 offsetof(OBJ, OBJ_FIELD)); \
9223 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
9224 offsetof(struct bpf_sock_ops_kern, \
9228 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
9230 if (TYPE == BPF_WRITE) \
9231 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
9233 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
9236 if (insn > insn_buf)
9237 return insn - insn_buf;
9240 case offsetof(struct bpf_sock_ops, op):
9241 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9243 si->dst_reg, si->src_reg,
9244 offsetof(struct bpf_sock_ops_kern, op));
9247 case offsetof(struct bpf_sock_ops, replylong[0]) ...
9248 offsetof(struct bpf_sock_ops, replylong[3]):
9249 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
9250 sizeof_field(struct bpf_sock_ops_kern, reply));
9251 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
9252 sizeof_field(struct bpf_sock_ops_kern, replylong));
9254 off -= offsetof(struct bpf_sock_ops, replylong[0]);
9255 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
9256 if (type == BPF_WRITE)
9257 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9260 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9264 case offsetof(struct bpf_sock_ops, family):
9265 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9267 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9268 struct bpf_sock_ops_kern, sk),
9269 si->dst_reg, si->src_reg,
9270 offsetof(struct bpf_sock_ops_kern, sk));
9271 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9272 offsetof(struct sock_common, skc_family));
9275 case offsetof(struct bpf_sock_ops, remote_ip4):
9276 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9278 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9279 struct bpf_sock_ops_kern, sk),
9280 si->dst_reg, si->src_reg,
9281 offsetof(struct bpf_sock_ops_kern, sk));
9282 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9283 offsetof(struct sock_common, skc_daddr));
9286 case offsetof(struct bpf_sock_ops, local_ip4):
9287 BUILD_BUG_ON(sizeof_field(struct sock_common,
9288 skc_rcv_saddr) != 4);
9290 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9291 struct bpf_sock_ops_kern, sk),
9292 si->dst_reg, si->src_reg,
9293 offsetof(struct bpf_sock_ops_kern, sk));
9294 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9295 offsetof(struct sock_common,
9299 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
9300 offsetof(struct bpf_sock_ops, remote_ip6[3]):
9301 #if IS_ENABLED(CONFIG_IPV6)
9302 BUILD_BUG_ON(sizeof_field(struct sock_common,
9303 skc_v6_daddr.s6_addr32[0]) != 4);
9306 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
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 sock_common,
9313 skc_v6_daddr.s6_addr32[0]) +
9316 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9320 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
9321 offsetof(struct bpf_sock_ops, local_ip6[3]):
9322 #if IS_ENABLED(CONFIG_IPV6)
9323 BUILD_BUG_ON(sizeof_field(struct sock_common,
9324 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9327 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
9328 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9329 struct bpf_sock_ops_kern, sk),
9330 si->dst_reg, si->src_reg,
9331 offsetof(struct bpf_sock_ops_kern, sk));
9332 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9333 offsetof(struct sock_common,
9334 skc_v6_rcv_saddr.s6_addr32[0]) +
9337 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9341 case offsetof(struct bpf_sock_ops, remote_port):
9342 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9344 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9345 struct bpf_sock_ops_kern, sk),
9346 si->dst_reg, si->src_reg,
9347 offsetof(struct bpf_sock_ops_kern, sk));
9348 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9349 offsetof(struct sock_common, skc_dport));
9350 #ifndef __BIG_ENDIAN_BITFIELD
9351 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9355 case offsetof(struct bpf_sock_ops, local_port):
9356 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9358 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9359 struct bpf_sock_ops_kern, sk),
9360 si->dst_reg, si->src_reg,
9361 offsetof(struct bpf_sock_ops_kern, sk));
9362 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9363 offsetof(struct sock_common, skc_num));
9366 case offsetof(struct bpf_sock_ops, is_fullsock):
9367 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9368 struct bpf_sock_ops_kern,
9370 si->dst_reg, si->src_reg,
9371 offsetof(struct bpf_sock_ops_kern,
9375 case offsetof(struct bpf_sock_ops, state):
9376 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
9378 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9379 struct bpf_sock_ops_kern, sk),
9380 si->dst_reg, si->src_reg,
9381 offsetof(struct bpf_sock_ops_kern, sk));
9382 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
9383 offsetof(struct sock_common, skc_state));
9386 case offsetof(struct bpf_sock_ops, rtt_min):
9387 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
9388 sizeof(struct minmax));
9389 BUILD_BUG_ON(sizeof(struct minmax) <
9390 sizeof(struct minmax_sample));
9392 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9393 struct bpf_sock_ops_kern, sk),
9394 si->dst_reg, si->src_reg,
9395 offsetof(struct bpf_sock_ops_kern, sk));
9396 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9397 offsetof(struct tcp_sock, rtt_min) +
9398 sizeof_field(struct minmax_sample, t));
9401 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
9402 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
9406 case offsetof(struct bpf_sock_ops, sk_txhash):
9407 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
9410 case offsetof(struct bpf_sock_ops, snd_cwnd):
9411 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
9413 case offsetof(struct bpf_sock_ops, srtt_us):
9414 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
9416 case offsetof(struct bpf_sock_ops, snd_ssthresh):
9417 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
9419 case offsetof(struct bpf_sock_ops, rcv_nxt):
9420 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
9422 case offsetof(struct bpf_sock_ops, snd_nxt):
9423 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
9425 case offsetof(struct bpf_sock_ops, snd_una):
9426 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
9428 case offsetof(struct bpf_sock_ops, mss_cache):
9429 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
9431 case offsetof(struct bpf_sock_ops, ecn_flags):
9432 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
9434 case offsetof(struct bpf_sock_ops, rate_delivered):
9435 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
9437 case offsetof(struct bpf_sock_ops, rate_interval_us):
9438 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
9440 case offsetof(struct bpf_sock_ops, packets_out):
9441 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
9443 case offsetof(struct bpf_sock_ops, retrans_out):
9444 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
9446 case offsetof(struct bpf_sock_ops, total_retrans):
9447 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
9449 case offsetof(struct bpf_sock_ops, segs_in):
9450 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
9452 case offsetof(struct bpf_sock_ops, data_segs_in):
9453 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
9455 case offsetof(struct bpf_sock_ops, segs_out):
9456 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
9458 case offsetof(struct bpf_sock_ops, data_segs_out):
9459 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
9461 case offsetof(struct bpf_sock_ops, lost_out):
9462 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
9464 case offsetof(struct bpf_sock_ops, sacked_out):
9465 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
9467 case offsetof(struct bpf_sock_ops, bytes_received):
9468 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
9470 case offsetof(struct bpf_sock_ops, bytes_acked):
9471 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
9473 case offsetof(struct bpf_sock_ops, sk):
9476 case offsetof(struct bpf_sock_ops, skb_data_end):
9477 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9479 si->dst_reg, si->src_reg,
9480 offsetof(struct bpf_sock_ops_kern,
9483 case offsetof(struct bpf_sock_ops, skb_data):
9484 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9486 si->dst_reg, si->src_reg,
9487 offsetof(struct bpf_sock_ops_kern,
9489 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9490 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9491 si->dst_reg, si->dst_reg,
9492 offsetof(struct sk_buff, data));
9494 case offsetof(struct bpf_sock_ops, skb_len):
9495 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9497 si->dst_reg, si->src_reg,
9498 offsetof(struct bpf_sock_ops_kern,
9500 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9501 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9502 si->dst_reg, si->dst_reg,
9503 offsetof(struct sk_buff, len));
9505 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9506 off = offsetof(struct sk_buff, cb);
9507 off += offsetof(struct tcp_skb_cb, tcp_flags);
9508 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
9509 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9511 si->dst_reg, si->src_reg,
9512 offsetof(struct bpf_sock_ops_kern,
9514 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9515 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
9517 si->dst_reg, si->dst_reg, off);
9520 return insn - insn_buf;
9523 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
9524 const struct bpf_insn *si,
9525 struct bpf_insn *insn_buf,
9526 struct bpf_prog *prog, u32 *target_size)
9528 struct bpf_insn *insn = insn_buf;
9532 case offsetof(struct __sk_buff, data_end):
9534 off -= offsetof(struct __sk_buff, data_end);
9535 off += offsetof(struct sk_buff, cb);
9536 off += offsetof(struct tcp_skb_cb, bpf.data_end);
9537 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9540 case offsetof(struct __sk_buff, cb[0]) ...
9541 offsetofend(struct __sk_buff, cb[4]) - 1:
9542 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20);
9543 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
9544 offsetof(struct sk_skb_cb, data)) %
9547 prog->cb_access = 1;
9549 off -= offsetof(struct __sk_buff, cb[0]);
9550 off += offsetof(struct sk_buff, cb);
9551 off += offsetof(struct sk_skb_cb, data);
9552 if (type == BPF_WRITE)
9553 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
9556 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
9562 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9566 return insn - insn_buf;
9569 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
9570 const struct bpf_insn *si,
9571 struct bpf_insn *insn_buf,
9572 struct bpf_prog *prog, u32 *target_size)
9574 struct bpf_insn *insn = insn_buf;
9575 #if IS_ENABLED(CONFIG_IPV6)
9579 /* convert ctx uses the fact sg element is first in struct */
9580 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
9583 case offsetof(struct sk_msg_md, data):
9584 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
9585 si->dst_reg, si->src_reg,
9586 offsetof(struct sk_msg, data));
9588 case offsetof(struct sk_msg_md, data_end):
9589 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
9590 si->dst_reg, si->src_reg,
9591 offsetof(struct sk_msg, data_end));
9593 case offsetof(struct sk_msg_md, family):
9594 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9596 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9598 si->dst_reg, si->src_reg,
9599 offsetof(struct sk_msg, sk));
9600 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9601 offsetof(struct sock_common, skc_family));
9604 case offsetof(struct sk_msg_md, remote_ip4):
9605 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9607 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9609 si->dst_reg, si->src_reg,
9610 offsetof(struct sk_msg, sk));
9611 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9612 offsetof(struct sock_common, skc_daddr));
9615 case offsetof(struct sk_msg_md, local_ip4):
9616 BUILD_BUG_ON(sizeof_field(struct sock_common,
9617 skc_rcv_saddr) != 4);
9619 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9621 si->dst_reg, si->src_reg,
9622 offsetof(struct sk_msg, sk));
9623 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9624 offsetof(struct sock_common,
9628 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
9629 offsetof(struct sk_msg_md, remote_ip6[3]):
9630 #if IS_ENABLED(CONFIG_IPV6)
9631 BUILD_BUG_ON(sizeof_field(struct sock_common,
9632 skc_v6_daddr.s6_addr32[0]) != 4);
9635 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
9636 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9638 si->dst_reg, si->src_reg,
9639 offsetof(struct sk_msg, sk));
9640 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9641 offsetof(struct sock_common,
9642 skc_v6_daddr.s6_addr32[0]) +
9645 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9649 case offsetof(struct sk_msg_md, local_ip6[0]) ...
9650 offsetof(struct sk_msg_md, local_ip6[3]):
9651 #if IS_ENABLED(CONFIG_IPV6)
9652 BUILD_BUG_ON(sizeof_field(struct sock_common,
9653 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9656 off -= offsetof(struct sk_msg_md, local_ip6[0]);
9657 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9659 si->dst_reg, si->src_reg,
9660 offsetof(struct sk_msg, sk));
9661 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9662 offsetof(struct sock_common,
9663 skc_v6_rcv_saddr.s6_addr32[0]) +
9666 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9670 case offsetof(struct sk_msg_md, remote_port):
9671 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9673 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9675 si->dst_reg, si->src_reg,
9676 offsetof(struct sk_msg, sk));
9677 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9678 offsetof(struct sock_common, skc_dport));
9679 #ifndef __BIG_ENDIAN_BITFIELD
9680 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9684 case offsetof(struct sk_msg_md, local_port):
9685 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9687 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9689 si->dst_reg, si->src_reg,
9690 offsetof(struct sk_msg, sk));
9691 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9692 offsetof(struct sock_common, skc_num));
9695 case offsetof(struct sk_msg_md, size):
9696 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
9697 si->dst_reg, si->src_reg,
9698 offsetof(struct sk_msg_sg, size));
9701 case offsetof(struct sk_msg_md, sk):
9702 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
9703 si->dst_reg, si->src_reg,
9704 offsetof(struct sk_msg, sk));
9708 return insn - insn_buf;
9711 const struct bpf_verifier_ops sk_filter_verifier_ops = {
9712 .get_func_proto = sk_filter_func_proto,
9713 .is_valid_access = sk_filter_is_valid_access,
9714 .convert_ctx_access = bpf_convert_ctx_access,
9715 .gen_ld_abs = bpf_gen_ld_abs,
9718 const struct bpf_prog_ops sk_filter_prog_ops = {
9719 .test_run = bpf_prog_test_run_skb,
9722 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
9723 .get_func_proto = tc_cls_act_func_proto,
9724 .is_valid_access = tc_cls_act_is_valid_access,
9725 .convert_ctx_access = tc_cls_act_convert_ctx_access,
9726 .gen_prologue = tc_cls_act_prologue,
9727 .gen_ld_abs = bpf_gen_ld_abs,
9730 const struct bpf_prog_ops tc_cls_act_prog_ops = {
9731 .test_run = bpf_prog_test_run_skb,
9734 const struct bpf_verifier_ops xdp_verifier_ops = {
9735 .get_func_proto = xdp_func_proto,
9736 .is_valid_access = xdp_is_valid_access,
9737 .convert_ctx_access = xdp_convert_ctx_access,
9738 .gen_prologue = bpf_noop_prologue,
9741 const struct bpf_prog_ops xdp_prog_ops = {
9742 .test_run = bpf_prog_test_run_xdp,
9745 const struct bpf_verifier_ops cg_skb_verifier_ops = {
9746 .get_func_proto = cg_skb_func_proto,
9747 .is_valid_access = cg_skb_is_valid_access,
9748 .convert_ctx_access = bpf_convert_ctx_access,
9751 const struct bpf_prog_ops cg_skb_prog_ops = {
9752 .test_run = bpf_prog_test_run_skb,
9755 const struct bpf_verifier_ops lwt_in_verifier_ops = {
9756 .get_func_proto = lwt_in_func_proto,
9757 .is_valid_access = lwt_is_valid_access,
9758 .convert_ctx_access = bpf_convert_ctx_access,
9761 const struct bpf_prog_ops lwt_in_prog_ops = {
9762 .test_run = bpf_prog_test_run_skb,
9765 const struct bpf_verifier_ops lwt_out_verifier_ops = {
9766 .get_func_proto = lwt_out_func_proto,
9767 .is_valid_access = lwt_is_valid_access,
9768 .convert_ctx_access = bpf_convert_ctx_access,
9771 const struct bpf_prog_ops lwt_out_prog_ops = {
9772 .test_run = bpf_prog_test_run_skb,
9775 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
9776 .get_func_proto = lwt_xmit_func_proto,
9777 .is_valid_access = lwt_is_valid_access,
9778 .convert_ctx_access = bpf_convert_ctx_access,
9779 .gen_prologue = tc_cls_act_prologue,
9782 const struct bpf_prog_ops lwt_xmit_prog_ops = {
9783 .test_run = bpf_prog_test_run_skb,
9786 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
9787 .get_func_proto = lwt_seg6local_func_proto,
9788 .is_valid_access = lwt_is_valid_access,
9789 .convert_ctx_access = bpf_convert_ctx_access,
9792 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
9793 .test_run = bpf_prog_test_run_skb,
9796 const struct bpf_verifier_ops cg_sock_verifier_ops = {
9797 .get_func_proto = sock_filter_func_proto,
9798 .is_valid_access = sock_filter_is_valid_access,
9799 .convert_ctx_access = bpf_sock_convert_ctx_access,
9802 const struct bpf_prog_ops cg_sock_prog_ops = {
9805 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
9806 .get_func_proto = sock_addr_func_proto,
9807 .is_valid_access = sock_addr_is_valid_access,
9808 .convert_ctx_access = sock_addr_convert_ctx_access,
9811 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
9814 const struct bpf_verifier_ops sock_ops_verifier_ops = {
9815 .get_func_proto = sock_ops_func_proto,
9816 .is_valid_access = sock_ops_is_valid_access,
9817 .convert_ctx_access = sock_ops_convert_ctx_access,
9820 const struct bpf_prog_ops sock_ops_prog_ops = {
9823 const struct bpf_verifier_ops sk_skb_verifier_ops = {
9824 .get_func_proto = sk_skb_func_proto,
9825 .is_valid_access = sk_skb_is_valid_access,
9826 .convert_ctx_access = sk_skb_convert_ctx_access,
9827 .gen_prologue = sk_skb_prologue,
9830 const struct bpf_prog_ops sk_skb_prog_ops = {
9833 const struct bpf_verifier_ops sk_msg_verifier_ops = {
9834 .get_func_proto = sk_msg_func_proto,
9835 .is_valid_access = sk_msg_is_valid_access,
9836 .convert_ctx_access = sk_msg_convert_ctx_access,
9837 .gen_prologue = bpf_noop_prologue,
9840 const struct bpf_prog_ops sk_msg_prog_ops = {
9843 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
9844 .get_func_proto = flow_dissector_func_proto,
9845 .is_valid_access = flow_dissector_is_valid_access,
9846 .convert_ctx_access = flow_dissector_convert_ctx_access,
9849 const struct bpf_prog_ops flow_dissector_prog_ops = {
9850 .test_run = bpf_prog_test_run_flow_dissector,
9853 int sk_detach_filter(struct sock *sk)
9856 struct sk_filter *filter;
9858 if (sock_flag(sk, SOCK_FILTER_LOCKED))
9861 filter = rcu_dereference_protected(sk->sk_filter,
9862 lockdep_sock_is_held(sk));
9864 RCU_INIT_POINTER(sk->sk_filter, NULL);
9865 sk_filter_uncharge(sk, filter);
9871 EXPORT_SYMBOL_GPL(sk_detach_filter);
9873 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
9876 struct sock_fprog_kern *fprog;
9877 struct sk_filter *filter;
9881 filter = rcu_dereference_protected(sk->sk_filter,
9882 lockdep_sock_is_held(sk));
9886 /* We're copying the filter that has been originally attached,
9887 * so no conversion/decode needed anymore. eBPF programs that
9888 * have no original program cannot be dumped through this.
9891 fprog = filter->prog->orig_prog;
9897 /* User space only enquires number of filter blocks. */
9901 if (len < fprog->len)
9905 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
9908 /* Instead of bytes, the API requests to return the number
9918 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
9919 struct sock_reuseport *reuse,
9920 struct sock *sk, struct sk_buff *skb,
9923 reuse_kern->skb = skb;
9924 reuse_kern->sk = sk;
9925 reuse_kern->selected_sk = NULL;
9926 reuse_kern->data_end = skb->data + skb_headlen(skb);
9927 reuse_kern->hash = hash;
9928 reuse_kern->reuseport_id = reuse->reuseport_id;
9929 reuse_kern->bind_inany = reuse->bind_inany;
9932 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
9933 struct bpf_prog *prog, struct sk_buff *skb,
9936 struct sk_reuseport_kern reuse_kern;
9937 enum sk_action action;
9939 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, hash);
9940 action = BPF_PROG_RUN(prog, &reuse_kern);
9942 if (action == SK_PASS)
9943 return reuse_kern.selected_sk;
9945 return ERR_PTR(-ECONNREFUSED);
9948 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
9949 struct bpf_map *, map, void *, key, u32, flags)
9951 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
9952 struct sock_reuseport *reuse;
9953 struct sock *selected_sk;
9955 selected_sk = map->ops->map_lookup_elem(map, key);
9959 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
9961 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
9962 if (sk_is_refcounted(selected_sk))
9963 sock_put(selected_sk);
9965 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
9966 * The only (!reuse) case here is - the sk has already been
9967 * unhashed (e.g. by close()), so treat it as -ENOENT.
9969 * Other maps (e.g. sock_map) do not provide this guarantee and
9970 * the sk may never be in the reuseport group to begin with.
9972 return is_sockarray ? -ENOENT : -EINVAL;
9975 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
9976 struct sock *sk = reuse_kern->sk;
9978 if (sk->sk_protocol != selected_sk->sk_protocol)
9980 else if (sk->sk_family != selected_sk->sk_family)
9981 return -EAFNOSUPPORT;
9983 /* Catch all. Likely bound to a different sockaddr. */
9987 reuse_kern->selected_sk = selected_sk;
9992 static const struct bpf_func_proto sk_select_reuseport_proto = {
9993 .func = sk_select_reuseport,
9995 .ret_type = RET_INTEGER,
9996 .arg1_type = ARG_PTR_TO_CTX,
9997 .arg2_type = ARG_CONST_MAP_PTR,
9998 .arg3_type = ARG_PTR_TO_MAP_KEY,
9999 .arg4_type = ARG_ANYTHING,
10002 BPF_CALL_4(sk_reuseport_load_bytes,
10003 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10004 void *, to, u32, len)
10006 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
10009 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
10010 .func = sk_reuseport_load_bytes,
10012 .ret_type = RET_INTEGER,
10013 .arg1_type = ARG_PTR_TO_CTX,
10014 .arg2_type = ARG_ANYTHING,
10015 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
10016 .arg4_type = ARG_CONST_SIZE,
10019 BPF_CALL_5(sk_reuseport_load_bytes_relative,
10020 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10021 void *, to, u32, len, u32, start_header)
10023 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
10024 len, start_header);
10027 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
10028 .func = sk_reuseport_load_bytes_relative,
10030 .ret_type = RET_INTEGER,
10031 .arg1_type = ARG_PTR_TO_CTX,
10032 .arg2_type = ARG_ANYTHING,
10033 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
10034 .arg4_type = ARG_CONST_SIZE,
10035 .arg5_type = ARG_ANYTHING,
10038 static const struct bpf_func_proto *
10039 sk_reuseport_func_proto(enum bpf_func_id func_id,
10040 const struct bpf_prog *prog)
10043 case BPF_FUNC_sk_select_reuseport:
10044 return &sk_select_reuseport_proto;
10045 case BPF_FUNC_skb_load_bytes:
10046 return &sk_reuseport_load_bytes_proto;
10047 case BPF_FUNC_skb_load_bytes_relative:
10048 return &sk_reuseport_load_bytes_relative_proto;
10050 return bpf_base_func_proto(func_id);
10055 sk_reuseport_is_valid_access(int off, int size,
10056 enum bpf_access_type type,
10057 const struct bpf_prog *prog,
10058 struct bpf_insn_access_aux *info)
10060 const u32 size_default = sizeof(__u32);
10062 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
10063 off % size || type != BPF_READ)
10067 case offsetof(struct sk_reuseport_md, data):
10068 info->reg_type = PTR_TO_PACKET;
10069 return size == sizeof(__u64);
10071 case offsetof(struct sk_reuseport_md, data_end):
10072 info->reg_type = PTR_TO_PACKET_END;
10073 return size == sizeof(__u64);
10075 case offsetof(struct sk_reuseport_md, hash):
10076 return size == size_default;
10078 /* Fields that allow narrowing */
10079 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
10080 if (size < sizeof_field(struct sk_buff, protocol))
10083 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
10084 case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
10085 case bpf_ctx_range(struct sk_reuseport_md, len):
10086 bpf_ctx_record_field_size(info, size_default);
10087 return bpf_ctx_narrow_access_ok(off, size, size_default);
10094 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
10095 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
10096 si->dst_reg, si->src_reg, \
10097 bpf_target_off(struct sk_reuseport_kern, F, \
10098 sizeof_field(struct sk_reuseport_kern, F), \
10102 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
10103 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
10108 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \
10109 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
10114 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
10115 const struct bpf_insn *si,
10116 struct bpf_insn *insn_buf,
10117 struct bpf_prog *prog,
10120 struct bpf_insn *insn = insn_buf;
10123 case offsetof(struct sk_reuseport_md, data):
10124 SK_REUSEPORT_LOAD_SKB_FIELD(data);
10127 case offsetof(struct sk_reuseport_md, len):
10128 SK_REUSEPORT_LOAD_SKB_FIELD(len);
10131 case offsetof(struct sk_reuseport_md, eth_protocol):
10132 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
10135 case offsetof(struct sk_reuseport_md, ip_protocol):
10136 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
10139 case offsetof(struct sk_reuseport_md, data_end):
10140 SK_REUSEPORT_LOAD_FIELD(data_end);
10143 case offsetof(struct sk_reuseport_md, hash):
10144 SK_REUSEPORT_LOAD_FIELD(hash);
10147 case offsetof(struct sk_reuseport_md, bind_inany):
10148 SK_REUSEPORT_LOAD_FIELD(bind_inany);
10152 return insn - insn_buf;
10155 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
10156 .get_func_proto = sk_reuseport_func_proto,
10157 .is_valid_access = sk_reuseport_is_valid_access,
10158 .convert_ctx_access = sk_reuseport_convert_ctx_access,
10161 const struct bpf_prog_ops sk_reuseport_prog_ops = {
10164 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
10165 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
10167 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
10168 struct sock *, sk, u64, flags)
10170 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
10171 BPF_SK_LOOKUP_F_NO_REUSEPORT)))
10173 if (unlikely(sk && sk_is_refcounted(sk)))
10174 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
10175 if (unlikely(sk && sk->sk_state == TCP_ESTABLISHED))
10176 return -ESOCKTNOSUPPORT; /* reject connected sockets */
10178 /* Check if socket is suitable for packet L3/L4 protocol */
10179 if (sk && sk->sk_protocol != ctx->protocol)
10180 return -EPROTOTYPE;
10181 if (sk && sk->sk_family != ctx->family &&
10182 (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
10183 return -EAFNOSUPPORT;
10185 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
10188 /* Select socket as lookup result */
10189 ctx->selected_sk = sk;
10190 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
10194 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
10195 .func = bpf_sk_lookup_assign,
10197 .ret_type = RET_INTEGER,
10198 .arg1_type = ARG_PTR_TO_CTX,
10199 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL,
10200 .arg3_type = ARG_ANYTHING,
10203 static const struct bpf_func_proto *
10204 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
10207 case BPF_FUNC_perf_event_output:
10208 return &bpf_event_output_data_proto;
10209 case BPF_FUNC_sk_assign:
10210 return &bpf_sk_lookup_assign_proto;
10211 case BPF_FUNC_sk_release:
10212 return &bpf_sk_release_proto;
10214 return bpf_sk_base_func_proto(func_id);
10218 static bool sk_lookup_is_valid_access(int off, int size,
10219 enum bpf_access_type type,
10220 const struct bpf_prog *prog,
10221 struct bpf_insn_access_aux *info)
10223 if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
10225 if (off % size != 0)
10227 if (type != BPF_READ)
10231 case offsetof(struct bpf_sk_lookup, sk):
10232 info->reg_type = PTR_TO_SOCKET_OR_NULL;
10233 return size == sizeof(__u64);
10235 case bpf_ctx_range(struct bpf_sk_lookup, family):
10236 case bpf_ctx_range(struct bpf_sk_lookup, protocol):
10237 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
10238 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
10239 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
10240 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
10241 case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
10242 case bpf_ctx_range(struct bpf_sk_lookup, local_port):
10243 bpf_ctx_record_field_size(info, sizeof(__u32));
10244 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
10251 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
10252 const struct bpf_insn *si,
10253 struct bpf_insn *insn_buf,
10254 struct bpf_prog *prog,
10257 struct bpf_insn *insn = insn_buf;
10260 case offsetof(struct bpf_sk_lookup, sk):
10261 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10262 offsetof(struct bpf_sk_lookup_kern, selected_sk));
10265 case offsetof(struct bpf_sk_lookup, family):
10266 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10267 bpf_target_off(struct bpf_sk_lookup_kern,
10268 family, 2, target_size));
10271 case offsetof(struct bpf_sk_lookup, protocol):
10272 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10273 bpf_target_off(struct bpf_sk_lookup_kern,
10274 protocol, 2, target_size));
10277 case offsetof(struct bpf_sk_lookup, remote_ip4):
10278 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10279 bpf_target_off(struct bpf_sk_lookup_kern,
10280 v4.saddr, 4, target_size));
10283 case offsetof(struct bpf_sk_lookup, local_ip4):
10284 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10285 bpf_target_off(struct bpf_sk_lookup_kern,
10286 v4.daddr, 4, target_size));
10289 case bpf_ctx_range_till(struct bpf_sk_lookup,
10290 remote_ip6[0], remote_ip6[3]): {
10291 #if IS_ENABLED(CONFIG_IPV6)
10294 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
10295 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10296 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10297 offsetof(struct bpf_sk_lookup_kern, v6.saddr));
10298 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10299 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10301 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10305 case bpf_ctx_range_till(struct bpf_sk_lookup,
10306 local_ip6[0], local_ip6[3]): {
10307 #if IS_ENABLED(CONFIG_IPV6)
10310 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
10311 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10312 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10313 offsetof(struct bpf_sk_lookup_kern, v6.daddr));
10314 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10315 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10317 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10321 case offsetof(struct bpf_sk_lookup, remote_port):
10322 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10323 bpf_target_off(struct bpf_sk_lookup_kern,
10324 sport, 2, target_size));
10327 case offsetof(struct bpf_sk_lookup, local_port):
10328 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10329 bpf_target_off(struct bpf_sk_lookup_kern,
10330 dport, 2, target_size));
10334 return insn - insn_buf;
10337 const struct bpf_prog_ops sk_lookup_prog_ops = {
10338 .test_run = bpf_prog_test_run_sk_lookup,
10341 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
10342 .get_func_proto = sk_lookup_func_proto,
10343 .is_valid_access = sk_lookup_is_valid_access,
10344 .convert_ctx_access = sk_lookup_convert_ctx_access,
10347 #endif /* CONFIG_INET */
10349 DEFINE_BPF_DISPATCHER(xdp)
10351 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
10353 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
10356 #ifdef CONFIG_DEBUG_INFO_BTF
10357 BTF_ID_LIST_GLOBAL(btf_sock_ids)
10358 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
10360 #undef BTF_SOCK_TYPE
10362 u32 btf_sock_ids[MAX_BTF_SOCK_TYPE];
10365 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
10367 /* tcp6_sock type is not generated in dwarf and hence btf,
10368 * trigger an explicit type generation here.
10370 BTF_TYPE_EMIT(struct tcp6_sock);
10371 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
10372 sk->sk_family == AF_INET6)
10373 return (unsigned long)sk;
10375 return (unsigned long)NULL;
10378 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
10379 .func = bpf_skc_to_tcp6_sock,
10381 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10382 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10383 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
10386 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
10388 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
10389 return (unsigned long)sk;
10391 return (unsigned long)NULL;
10394 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
10395 .func = bpf_skc_to_tcp_sock,
10397 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10398 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10399 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
10402 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
10404 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
10405 * generated if CONFIG_INET=n. Trigger an explicit generation here.
10407 BTF_TYPE_EMIT(struct inet_timewait_sock);
10408 BTF_TYPE_EMIT(struct tcp_timewait_sock);
10411 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
10412 return (unsigned long)sk;
10415 #if IS_BUILTIN(CONFIG_IPV6)
10416 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
10417 return (unsigned long)sk;
10420 return (unsigned long)NULL;
10423 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
10424 .func = bpf_skc_to_tcp_timewait_sock,
10426 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10427 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10428 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
10431 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
10434 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10435 return (unsigned long)sk;
10438 #if IS_BUILTIN(CONFIG_IPV6)
10439 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10440 return (unsigned long)sk;
10443 return (unsigned long)NULL;
10446 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
10447 .func = bpf_skc_to_tcp_request_sock,
10449 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10450 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10451 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
10454 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
10456 /* udp6_sock type is not generated in dwarf and hence btf,
10457 * trigger an explicit type generation here.
10459 BTF_TYPE_EMIT(struct udp6_sock);
10460 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
10461 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
10462 return (unsigned long)sk;
10464 return (unsigned long)NULL;
10467 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
10468 .func = bpf_skc_to_udp6_sock,
10470 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10471 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10472 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
10475 static const struct bpf_func_proto *
10476 bpf_sk_base_func_proto(enum bpf_func_id func_id)
10478 const struct bpf_func_proto *func;
10481 case BPF_FUNC_skc_to_tcp6_sock:
10482 func = &bpf_skc_to_tcp6_sock_proto;
10484 case BPF_FUNC_skc_to_tcp_sock:
10485 func = &bpf_skc_to_tcp_sock_proto;
10487 case BPF_FUNC_skc_to_tcp_timewait_sock:
10488 func = &bpf_skc_to_tcp_timewait_sock_proto;
10490 case BPF_FUNC_skc_to_tcp_request_sock:
10491 func = &bpf_skc_to_tcp_request_sock_proto;
10493 case BPF_FUNC_skc_to_udp6_sock:
10494 func = &bpf_skc_to_udp6_sock_proto;
10497 return bpf_base_func_proto(func_id);
10500 if (!perfmon_capable())