1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Linux Socket Filter - Kernel level socket filtering
5 * Based on the design of the Berkeley Packet Filter. The new
6 * internal format has been designed by PLUMgrid:
8 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
12 * Jay Schulist <jschlst@samba.org>
13 * Alexei Starovoitov <ast@plumgrid.com>
14 * Daniel Borkmann <dborkman@redhat.com>
16 * Andi Kleen - Fix a few bad bugs and races.
17 * Kris Katterjohn - Added many additional checks in bpf_check_classic()
20 #include <linux/module.h>
21 #include <linux/types.h>
23 #include <linux/fcntl.h>
24 #include <linux/socket.h>
25 #include <linux/sock_diag.h>
27 #include <linux/inet.h>
28 #include <linux/netdevice.h>
29 #include <linux/if_packet.h>
30 #include <linux/if_arp.h>
31 #include <linux/gfp.h>
32 #include <net/inet_common.h>
34 #include <net/protocol.h>
35 #include <net/netlink.h>
36 #include <linux/skbuff.h>
37 #include <linux/skmsg.h>
39 #include <net/flow_dissector.h>
40 #include <linux/errno.h>
41 #include <linux/timer.h>
42 #include <linux/uaccess.h>
43 #include <asm/unaligned.h>
44 #include <asm/cmpxchg.h>
45 #include <linux/filter.h>
46 #include <linux/ratelimit.h>
47 #include <linux/seccomp.h>
48 #include <linux/if_vlan.h>
49 #include <linux/bpf.h>
50 #include <linux/btf.h>
51 #include <net/sch_generic.h>
52 #include <net/cls_cgroup.h>
53 #include <net/dst_metadata.h>
55 #include <net/sock_reuseport.h>
56 #include <net/busy_poll.h>
60 #include <linux/bpf_trace.h>
61 #include <net/xdp_sock.h>
62 #include <linux/inetdevice.h>
63 #include <net/inet_hashtables.h>
64 #include <net/inet6_hashtables.h>
65 #include <net/ip_fib.h>
66 #include <net/nexthop.h>
70 #include <net/net_namespace.h>
71 #include <linux/seg6_local.h>
73 #include <net/seg6_local.h>
74 #include <net/lwtunnel.h>
75 #include <net/ipv6_stubs.h>
76 #include <net/bpf_sk_storage.h>
77 #include <net/transp_v6.h>
78 #include <linux/btf_ids.h>
81 static const struct bpf_func_proto *
82 bpf_sk_base_func_proto(enum bpf_func_id func_id);
84 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len)
86 if (in_compat_syscall()) {
87 struct compat_sock_fprog f32;
89 if (len != sizeof(f32))
91 if (copy_from_sockptr(&f32, src, sizeof(f32)))
93 memset(dst, 0, sizeof(*dst));
95 dst->filter = compat_ptr(f32.filter);
97 if (len != sizeof(*dst))
99 if (copy_from_sockptr(dst, src, sizeof(*dst)))
105 EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user);
108 * sk_filter_trim_cap - run a packet through a socket filter
109 * @sk: sock associated with &sk_buff
110 * @skb: buffer to filter
111 * @cap: limit on how short the eBPF program may trim the packet
113 * Run the eBPF program and then cut skb->data to correct size returned by
114 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
115 * than pkt_len we keep whole skb->data. This is the socket level
116 * wrapper to BPF_PROG_RUN. It returns 0 if the packet should
117 * be accepted or -EPERM if the packet should be tossed.
120 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
123 struct sk_filter *filter;
126 * If the skb was allocated from pfmemalloc reserves, only
127 * allow SOCK_MEMALLOC sockets to use it as this socket is
128 * helping free memory
130 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
131 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
134 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
138 err = security_sock_rcv_skb(sk, skb);
143 filter = rcu_dereference(sk->sk_filter);
145 struct sock *save_sk = skb->sk;
146 unsigned int pkt_len;
149 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
151 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
157 EXPORT_SYMBOL(sk_filter_trim_cap);
159 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
161 return skb_get_poff(skb);
164 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
168 if (skb_is_nonlinear(skb))
171 if (skb->len < sizeof(struct nlattr))
174 if (a > skb->len - sizeof(struct nlattr))
177 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
179 return (void *) nla - (void *) skb->data;
184 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
188 if (skb_is_nonlinear(skb))
191 if (skb->len < sizeof(struct nlattr))
194 if (a > skb->len - sizeof(struct nlattr))
197 nla = (struct nlattr *) &skb->data[a];
198 if (nla->nla_len > skb->len - a)
201 nla = nla_find_nested(nla, x);
203 return (void *) nla - (void *) skb->data;
208 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
209 data, int, headlen, int, offset)
212 const int len = sizeof(tmp);
215 if (headlen - offset >= len)
216 return *(u8 *)(data + offset);
217 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
220 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
228 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
231 return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
235 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
236 data, int, headlen, int, offset)
239 const int len = sizeof(tmp);
242 if (headlen - offset >= len)
243 return get_unaligned_be16(data + offset);
244 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
245 return be16_to_cpu(tmp);
247 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
249 return get_unaligned_be16(ptr);
255 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
258 return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
262 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
263 data, int, headlen, int, offset)
266 const int len = sizeof(tmp);
268 if (likely(offset >= 0)) {
269 if (headlen - offset >= len)
270 return get_unaligned_be32(data + offset);
271 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
272 return be32_to_cpu(tmp);
274 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
276 return get_unaligned_be32(ptr);
282 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
285 return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
289 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
290 struct bpf_insn *insn_buf)
292 struct bpf_insn *insn = insn_buf;
296 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4);
298 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
299 offsetof(struct sk_buff, mark));
303 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
304 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
305 #ifdef __BIG_ENDIAN_BITFIELD
306 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
311 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2);
313 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
314 offsetof(struct sk_buff, queue_mapping));
317 case SKF_AD_VLAN_TAG:
318 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2);
320 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
321 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
322 offsetof(struct sk_buff, vlan_tci));
324 case SKF_AD_VLAN_TAG_PRESENT:
325 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_VLAN_PRESENT_OFFSET());
326 if (PKT_VLAN_PRESENT_BIT)
327 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, PKT_VLAN_PRESENT_BIT);
328 if (PKT_VLAN_PRESENT_BIT < 7)
329 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
333 return insn - insn_buf;
336 static bool convert_bpf_extensions(struct sock_filter *fp,
337 struct bpf_insn **insnp)
339 struct bpf_insn *insn = *insnp;
343 case SKF_AD_OFF + SKF_AD_PROTOCOL:
344 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2);
346 /* A = *(u16 *) (CTX + offsetof(protocol)) */
347 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
348 offsetof(struct sk_buff, protocol));
349 /* A = ntohs(A) [emitting a nop or swap16] */
350 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
353 case SKF_AD_OFF + SKF_AD_PKTTYPE:
354 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
358 case SKF_AD_OFF + SKF_AD_IFINDEX:
359 case SKF_AD_OFF + SKF_AD_HATYPE:
360 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4);
361 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2);
363 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
364 BPF_REG_TMP, BPF_REG_CTX,
365 offsetof(struct sk_buff, dev));
366 /* if (tmp != 0) goto pc + 1 */
367 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
368 *insn++ = BPF_EXIT_INSN();
369 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
370 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
371 offsetof(struct net_device, ifindex));
373 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
374 offsetof(struct net_device, type));
377 case SKF_AD_OFF + SKF_AD_MARK:
378 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
382 case SKF_AD_OFF + SKF_AD_RXHASH:
383 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4);
385 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
386 offsetof(struct sk_buff, hash));
389 case SKF_AD_OFF + SKF_AD_QUEUE:
390 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
394 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
395 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
396 BPF_REG_A, BPF_REG_CTX, insn);
400 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
401 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
402 BPF_REG_A, BPF_REG_CTX, insn);
406 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
407 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2);
409 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
410 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
411 offsetof(struct sk_buff, vlan_proto));
412 /* A = ntohs(A) [emitting a nop or swap16] */
413 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
416 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
417 case SKF_AD_OFF + SKF_AD_NLATTR:
418 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
419 case SKF_AD_OFF + SKF_AD_CPU:
420 case SKF_AD_OFF + SKF_AD_RANDOM:
422 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
424 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
426 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
427 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
429 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
430 *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
432 case SKF_AD_OFF + SKF_AD_NLATTR:
433 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
435 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
436 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
438 case SKF_AD_OFF + SKF_AD_CPU:
439 *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
441 case SKF_AD_OFF + SKF_AD_RANDOM:
442 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
443 bpf_user_rnd_init_once();
448 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
450 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
454 /* This is just a dummy call to avoid letting the compiler
455 * evict __bpf_call_base() as an optimization. Placed here
456 * where no-one bothers.
458 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
466 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
468 const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
469 int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
470 bool endian = BPF_SIZE(fp->code) == BPF_H ||
471 BPF_SIZE(fp->code) == BPF_W;
472 bool indirect = BPF_MODE(fp->code) == BPF_IND;
473 const int ip_align = NET_IP_ALIGN;
474 struct bpf_insn *insn = *insnp;
478 ((unaligned_ok && offset >= 0) ||
479 (!unaligned_ok && offset >= 0 &&
480 offset + ip_align >= 0 &&
481 offset + ip_align % size == 0))) {
482 bool ldx_off_ok = offset <= S16_MAX;
484 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
486 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
487 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
488 size, 2 + endian + (!ldx_off_ok * 2));
490 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
493 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
494 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
495 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
499 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
500 *insn++ = BPF_JMP_A(8);
503 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
504 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
505 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
507 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
509 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
511 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
514 switch (BPF_SIZE(fp->code)) {
516 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
519 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
522 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
528 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
529 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
530 *insn = BPF_EXIT_INSN();
537 * bpf_convert_filter - convert filter program
538 * @prog: the user passed filter program
539 * @len: the length of the user passed filter program
540 * @new_prog: allocated 'struct bpf_prog' or NULL
541 * @new_len: pointer to store length of converted program
542 * @seen_ld_abs: bool whether we've seen ld_abs/ind
544 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
545 * style extended BPF (eBPF).
546 * Conversion workflow:
548 * 1) First pass for calculating the new program length:
549 * bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
551 * 2) 2nd pass to remap in two passes: 1st pass finds new
552 * jump offsets, 2nd pass remapping:
553 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
555 static int bpf_convert_filter(struct sock_filter *prog, int len,
556 struct bpf_prog *new_prog, int *new_len,
559 int new_flen = 0, pass = 0, target, i, stack_off;
560 struct bpf_insn *new_insn, *first_insn = NULL;
561 struct sock_filter *fp;
565 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
566 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
568 if (len <= 0 || len > BPF_MAXINSNS)
572 first_insn = new_prog->insnsi;
573 addrs = kcalloc(len, sizeof(*addrs),
574 GFP_KERNEL | __GFP_NOWARN);
580 new_insn = first_insn;
583 /* Classic BPF related prologue emission. */
585 /* Classic BPF expects A and X to be reset first. These need
586 * to be guaranteed to be the first two instructions.
588 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
589 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
591 /* All programs must keep CTX in callee saved BPF_REG_CTX.
592 * In eBPF case it's done by the compiler, here we need to
593 * do this ourself. Initial CTX is present in BPF_REG_ARG1.
595 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
597 /* For packet access in classic BPF, cache skb->data
598 * in callee-saved BPF R8 and skb->len - skb->data_len
599 * (headlen) in BPF R9. Since classic BPF is read-only
600 * on CTX, we only need to cache it once.
602 *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
603 BPF_REG_D, BPF_REG_CTX,
604 offsetof(struct sk_buff, data));
605 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
606 offsetof(struct sk_buff, len));
607 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
608 offsetof(struct sk_buff, data_len));
609 *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
615 for (i = 0; i < len; fp++, i++) {
616 struct bpf_insn tmp_insns[32] = { };
617 struct bpf_insn *insn = tmp_insns;
620 addrs[i] = new_insn - first_insn;
623 /* All arithmetic insns and skb loads map as-is. */
624 case BPF_ALU | BPF_ADD | BPF_X:
625 case BPF_ALU | BPF_ADD | BPF_K:
626 case BPF_ALU | BPF_SUB | BPF_X:
627 case BPF_ALU | BPF_SUB | BPF_K:
628 case BPF_ALU | BPF_AND | BPF_X:
629 case BPF_ALU | BPF_AND | BPF_K:
630 case BPF_ALU | BPF_OR | BPF_X:
631 case BPF_ALU | BPF_OR | BPF_K:
632 case BPF_ALU | BPF_LSH | BPF_X:
633 case BPF_ALU | BPF_LSH | BPF_K:
634 case BPF_ALU | BPF_RSH | BPF_X:
635 case BPF_ALU | BPF_RSH | BPF_K:
636 case BPF_ALU | BPF_XOR | BPF_X:
637 case BPF_ALU | BPF_XOR | BPF_K:
638 case BPF_ALU | BPF_MUL | BPF_X:
639 case BPF_ALU | BPF_MUL | BPF_K:
640 case BPF_ALU | BPF_DIV | BPF_X:
641 case BPF_ALU | BPF_DIV | BPF_K:
642 case BPF_ALU | BPF_MOD | BPF_X:
643 case BPF_ALU | BPF_MOD | BPF_K:
644 case BPF_ALU | BPF_NEG:
645 case BPF_LD | BPF_ABS | BPF_W:
646 case BPF_LD | BPF_ABS | BPF_H:
647 case BPF_LD | BPF_ABS | BPF_B:
648 case BPF_LD | BPF_IND | BPF_W:
649 case BPF_LD | BPF_IND | BPF_H:
650 case BPF_LD | BPF_IND | BPF_B:
651 /* Check for overloaded BPF extension and
652 * directly convert it if found, otherwise
653 * just move on with mapping.
655 if (BPF_CLASS(fp->code) == BPF_LD &&
656 BPF_MODE(fp->code) == BPF_ABS &&
657 convert_bpf_extensions(fp, &insn))
659 if (BPF_CLASS(fp->code) == BPF_LD &&
660 convert_bpf_ld_abs(fp, &insn)) {
665 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
666 fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
667 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
668 /* Error with exception code on div/mod by 0.
669 * For cBPF programs, this was always return 0.
671 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
672 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
673 *insn++ = BPF_EXIT_INSN();
676 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
679 /* Jump transformation cannot use BPF block macros
680 * everywhere as offset calculation and target updates
681 * require a bit more work than the rest, i.e. jump
682 * opcodes map as-is, but offsets need adjustment.
685 #define BPF_EMIT_JMP \
687 const s32 off_min = S16_MIN, off_max = S16_MAX; \
690 if (target >= len || target < 0) \
692 off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
693 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
694 off -= insn - tmp_insns; \
695 /* Reject anything not fitting into insn->off. */ \
696 if (off < off_min || off > off_max) \
701 case BPF_JMP | BPF_JA:
702 target = i + fp->k + 1;
703 insn->code = fp->code;
707 case BPF_JMP | BPF_JEQ | BPF_K:
708 case BPF_JMP | BPF_JEQ | BPF_X:
709 case BPF_JMP | BPF_JSET | BPF_K:
710 case BPF_JMP | BPF_JSET | BPF_X:
711 case BPF_JMP | BPF_JGT | BPF_K:
712 case BPF_JMP | BPF_JGT | BPF_X:
713 case BPF_JMP | BPF_JGE | BPF_K:
714 case BPF_JMP | BPF_JGE | BPF_X:
715 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
716 /* BPF immediates are signed, zero extend
717 * immediate into tmp register and use it
720 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
722 insn->dst_reg = BPF_REG_A;
723 insn->src_reg = BPF_REG_TMP;
726 insn->dst_reg = BPF_REG_A;
728 bpf_src = BPF_SRC(fp->code);
729 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
732 /* Common case where 'jump_false' is next insn. */
734 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
735 target = i + fp->jt + 1;
740 /* Convert some jumps when 'jump_true' is next insn. */
742 switch (BPF_OP(fp->code)) {
744 insn->code = BPF_JMP | BPF_JNE | bpf_src;
747 insn->code = BPF_JMP | BPF_JLE | bpf_src;
750 insn->code = BPF_JMP | BPF_JLT | bpf_src;
756 target = i + fp->jf + 1;
761 /* Other jumps are mapped into two insns: Jxx and JA. */
762 target = i + fp->jt + 1;
763 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
767 insn->code = BPF_JMP | BPF_JA;
768 target = i + fp->jf + 1;
772 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
773 case BPF_LDX | BPF_MSH | BPF_B: {
774 struct sock_filter tmp = {
775 .code = BPF_LD | BPF_ABS | BPF_B,
782 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
783 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
784 convert_bpf_ld_abs(&tmp, &insn);
787 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
789 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
791 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
793 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
795 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
798 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
799 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
801 case BPF_RET | BPF_A:
802 case BPF_RET | BPF_K:
803 if (BPF_RVAL(fp->code) == BPF_K)
804 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
806 *insn = BPF_EXIT_INSN();
809 /* Store to stack. */
812 stack_off = fp->k * 4 + 4;
813 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
814 BPF_ST ? BPF_REG_A : BPF_REG_X,
816 /* check_load_and_stores() verifies that classic BPF can
817 * load from stack only after write, so tracking
818 * stack_depth for ST|STX insns is enough
820 if (new_prog && new_prog->aux->stack_depth < stack_off)
821 new_prog->aux->stack_depth = stack_off;
824 /* Load from stack. */
825 case BPF_LD | BPF_MEM:
826 case BPF_LDX | BPF_MEM:
827 stack_off = fp->k * 4 + 4;
828 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
829 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
834 case BPF_LD | BPF_IMM:
835 case BPF_LDX | BPF_IMM:
836 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
837 BPF_REG_A : BPF_REG_X, fp->k);
841 case BPF_MISC | BPF_TAX:
842 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
846 case BPF_MISC | BPF_TXA:
847 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
850 /* A = skb->len or X = skb->len */
851 case BPF_LD | BPF_W | BPF_LEN:
852 case BPF_LDX | BPF_W | BPF_LEN:
853 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
854 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
855 offsetof(struct sk_buff, len));
858 /* Access seccomp_data fields. */
859 case BPF_LDX | BPF_ABS | BPF_W:
860 /* A = *(u32 *) (ctx + K) */
861 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
864 /* Unknown instruction. */
871 memcpy(new_insn, tmp_insns,
872 sizeof(*insn) * (insn - tmp_insns));
873 new_insn += insn - tmp_insns;
877 /* Only calculating new length. */
878 *new_len = new_insn - first_insn;
880 *new_len += 4; /* Prologue bits. */
885 if (new_flen != new_insn - first_insn) {
886 new_flen = new_insn - first_insn;
893 BUG_ON(*new_len != new_flen);
902 * As we dont want to clear mem[] array for each packet going through
903 * __bpf_prog_run(), we check that filter loaded by user never try to read
904 * a cell if not previously written, and we check all branches to be sure
905 * a malicious user doesn't try to abuse us.
907 static int check_load_and_stores(const struct sock_filter *filter, int flen)
909 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
912 BUILD_BUG_ON(BPF_MEMWORDS > 16);
914 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
918 memset(masks, 0xff, flen * sizeof(*masks));
920 for (pc = 0; pc < flen; pc++) {
921 memvalid &= masks[pc];
923 switch (filter[pc].code) {
926 memvalid |= (1 << filter[pc].k);
928 case BPF_LD | BPF_MEM:
929 case BPF_LDX | BPF_MEM:
930 if (!(memvalid & (1 << filter[pc].k))) {
935 case BPF_JMP | BPF_JA:
936 /* A jump must set masks on target */
937 masks[pc + 1 + filter[pc].k] &= memvalid;
940 case BPF_JMP | BPF_JEQ | BPF_K:
941 case BPF_JMP | BPF_JEQ | BPF_X:
942 case BPF_JMP | BPF_JGE | BPF_K:
943 case BPF_JMP | BPF_JGE | BPF_X:
944 case BPF_JMP | BPF_JGT | BPF_K:
945 case BPF_JMP | BPF_JGT | BPF_X:
946 case BPF_JMP | BPF_JSET | BPF_K:
947 case BPF_JMP | BPF_JSET | BPF_X:
948 /* A jump must set masks on targets */
949 masks[pc + 1 + filter[pc].jt] &= memvalid;
950 masks[pc + 1 + filter[pc].jf] &= memvalid;
960 static bool chk_code_allowed(u16 code_to_probe)
962 static const bool codes[] = {
963 /* 32 bit ALU operations */
964 [BPF_ALU | BPF_ADD | BPF_K] = true,
965 [BPF_ALU | BPF_ADD | BPF_X] = true,
966 [BPF_ALU | BPF_SUB | BPF_K] = true,
967 [BPF_ALU | BPF_SUB | BPF_X] = true,
968 [BPF_ALU | BPF_MUL | BPF_K] = true,
969 [BPF_ALU | BPF_MUL | BPF_X] = true,
970 [BPF_ALU | BPF_DIV | BPF_K] = true,
971 [BPF_ALU | BPF_DIV | BPF_X] = true,
972 [BPF_ALU | BPF_MOD | BPF_K] = true,
973 [BPF_ALU | BPF_MOD | BPF_X] = true,
974 [BPF_ALU | BPF_AND | BPF_K] = true,
975 [BPF_ALU | BPF_AND | BPF_X] = true,
976 [BPF_ALU | BPF_OR | BPF_K] = true,
977 [BPF_ALU | BPF_OR | BPF_X] = true,
978 [BPF_ALU | BPF_XOR | BPF_K] = true,
979 [BPF_ALU | BPF_XOR | BPF_X] = true,
980 [BPF_ALU | BPF_LSH | BPF_K] = true,
981 [BPF_ALU | BPF_LSH | BPF_X] = true,
982 [BPF_ALU | BPF_RSH | BPF_K] = true,
983 [BPF_ALU | BPF_RSH | BPF_X] = true,
984 [BPF_ALU | BPF_NEG] = true,
985 /* Load instructions */
986 [BPF_LD | BPF_W | BPF_ABS] = true,
987 [BPF_LD | BPF_H | BPF_ABS] = true,
988 [BPF_LD | BPF_B | BPF_ABS] = true,
989 [BPF_LD | BPF_W | BPF_LEN] = true,
990 [BPF_LD | BPF_W | BPF_IND] = true,
991 [BPF_LD | BPF_H | BPF_IND] = true,
992 [BPF_LD | BPF_B | BPF_IND] = true,
993 [BPF_LD | BPF_IMM] = true,
994 [BPF_LD | BPF_MEM] = true,
995 [BPF_LDX | BPF_W | BPF_LEN] = true,
996 [BPF_LDX | BPF_B | BPF_MSH] = true,
997 [BPF_LDX | BPF_IMM] = true,
998 [BPF_LDX | BPF_MEM] = true,
999 /* Store instructions */
1002 /* Misc instructions */
1003 [BPF_MISC | BPF_TAX] = true,
1004 [BPF_MISC | BPF_TXA] = true,
1005 /* Return instructions */
1006 [BPF_RET | BPF_K] = true,
1007 [BPF_RET | BPF_A] = true,
1008 /* Jump instructions */
1009 [BPF_JMP | BPF_JA] = true,
1010 [BPF_JMP | BPF_JEQ | BPF_K] = true,
1011 [BPF_JMP | BPF_JEQ | BPF_X] = true,
1012 [BPF_JMP | BPF_JGE | BPF_K] = true,
1013 [BPF_JMP | BPF_JGE | BPF_X] = true,
1014 [BPF_JMP | BPF_JGT | BPF_K] = true,
1015 [BPF_JMP | BPF_JGT | BPF_X] = true,
1016 [BPF_JMP | BPF_JSET | BPF_K] = true,
1017 [BPF_JMP | BPF_JSET | BPF_X] = true,
1020 if (code_to_probe >= ARRAY_SIZE(codes))
1023 return codes[code_to_probe];
1026 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1031 if (flen == 0 || flen > BPF_MAXINSNS)
1038 * bpf_check_classic - verify socket filter code
1039 * @filter: filter to verify
1040 * @flen: length of filter
1042 * Check the user's filter code. If we let some ugly
1043 * filter code slip through kaboom! The filter must contain
1044 * no references or jumps that are out of range, no illegal
1045 * instructions, and must end with a RET instruction.
1047 * All jumps are forward as they are not signed.
1049 * Returns 0 if the rule set is legal or -EINVAL if not.
1051 static int bpf_check_classic(const struct sock_filter *filter,
1057 /* Check the filter code now */
1058 for (pc = 0; pc < flen; pc++) {
1059 const struct sock_filter *ftest = &filter[pc];
1061 /* May we actually operate on this code? */
1062 if (!chk_code_allowed(ftest->code))
1065 /* Some instructions need special checks */
1066 switch (ftest->code) {
1067 case BPF_ALU | BPF_DIV | BPF_K:
1068 case BPF_ALU | BPF_MOD | BPF_K:
1069 /* Check for division by zero */
1073 case BPF_ALU | BPF_LSH | BPF_K:
1074 case BPF_ALU | BPF_RSH | BPF_K:
1078 case BPF_LD | BPF_MEM:
1079 case BPF_LDX | BPF_MEM:
1082 /* Check for invalid memory addresses */
1083 if (ftest->k >= BPF_MEMWORDS)
1086 case BPF_JMP | BPF_JA:
1087 /* Note, the large ftest->k might cause loops.
1088 * Compare this with conditional jumps below,
1089 * where offsets are limited. --ANK (981016)
1091 if (ftest->k >= (unsigned int)(flen - pc - 1))
1094 case BPF_JMP | BPF_JEQ | BPF_K:
1095 case BPF_JMP | BPF_JEQ | BPF_X:
1096 case BPF_JMP | BPF_JGE | BPF_K:
1097 case BPF_JMP | BPF_JGE | BPF_X:
1098 case BPF_JMP | BPF_JGT | BPF_K:
1099 case BPF_JMP | BPF_JGT | BPF_X:
1100 case BPF_JMP | BPF_JSET | BPF_K:
1101 case BPF_JMP | BPF_JSET | BPF_X:
1102 /* Both conditionals must be safe */
1103 if (pc + ftest->jt + 1 >= flen ||
1104 pc + ftest->jf + 1 >= flen)
1107 case BPF_LD | BPF_W | BPF_ABS:
1108 case BPF_LD | BPF_H | BPF_ABS:
1109 case BPF_LD | BPF_B | BPF_ABS:
1111 if (bpf_anc_helper(ftest) & BPF_ANC)
1113 /* Ancillary operation unknown or unsupported */
1114 if (anc_found == false && ftest->k >= SKF_AD_OFF)
1119 /* Last instruction must be a RET code */
1120 switch (filter[flen - 1].code) {
1121 case BPF_RET | BPF_K:
1122 case BPF_RET | BPF_A:
1123 return check_load_and_stores(filter, flen);
1129 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1130 const struct sock_fprog *fprog)
1132 unsigned int fsize = bpf_classic_proglen(fprog);
1133 struct sock_fprog_kern *fkprog;
1135 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1139 fkprog = fp->orig_prog;
1140 fkprog->len = fprog->len;
1142 fkprog->filter = kmemdup(fp->insns, fsize,
1143 GFP_KERNEL | __GFP_NOWARN);
1144 if (!fkprog->filter) {
1145 kfree(fp->orig_prog);
1152 static void bpf_release_orig_filter(struct bpf_prog *fp)
1154 struct sock_fprog_kern *fprog = fp->orig_prog;
1157 kfree(fprog->filter);
1162 static void __bpf_prog_release(struct bpf_prog *prog)
1164 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1167 bpf_release_orig_filter(prog);
1168 bpf_prog_free(prog);
1172 static void __sk_filter_release(struct sk_filter *fp)
1174 __bpf_prog_release(fp->prog);
1179 * sk_filter_release_rcu - Release a socket filter by rcu_head
1180 * @rcu: rcu_head that contains the sk_filter to free
1182 static void sk_filter_release_rcu(struct rcu_head *rcu)
1184 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1186 __sk_filter_release(fp);
1190 * sk_filter_release - release a socket filter
1191 * @fp: filter to remove
1193 * Remove a filter from a socket and release its resources.
1195 static void sk_filter_release(struct sk_filter *fp)
1197 if (refcount_dec_and_test(&fp->refcnt))
1198 call_rcu(&fp->rcu, sk_filter_release_rcu);
1201 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1203 u32 filter_size = bpf_prog_size(fp->prog->len);
1205 atomic_sub(filter_size, &sk->sk_omem_alloc);
1206 sk_filter_release(fp);
1209 /* try to charge the socket memory if there is space available
1210 * return true on success
1212 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1214 u32 filter_size = bpf_prog_size(fp->prog->len);
1216 /* same check as in sock_kmalloc() */
1217 if (filter_size <= sysctl_optmem_max &&
1218 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
1219 atomic_add(filter_size, &sk->sk_omem_alloc);
1225 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1227 if (!refcount_inc_not_zero(&fp->refcnt))
1230 if (!__sk_filter_charge(sk, fp)) {
1231 sk_filter_release(fp);
1237 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1239 struct sock_filter *old_prog;
1240 struct bpf_prog *old_fp;
1241 int err, new_len, old_len = fp->len;
1242 bool seen_ld_abs = false;
1244 /* We are free to overwrite insns et al right here as it
1245 * won't be used at this point in time anymore internally
1246 * after the migration to the internal BPF instruction
1249 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1250 sizeof(struct bpf_insn));
1252 /* Conversion cannot happen on overlapping memory areas,
1253 * so we need to keep the user BPF around until the 2nd
1254 * pass. At this time, the user BPF is stored in fp->insns.
1256 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1257 GFP_KERNEL | __GFP_NOWARN);
1263 /* 1st pass: calculate the new program length. */
1264 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1269 /* Expand fp for appending the new filter representation. */
1271 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1273 /* The old_fp is still around in case we couldn't
1274 * allocate new memory, so uncharge on that one.
1283 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1284 err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1287 /* 2nd bpf_convert_filter() can fail only if it fails
1288 * to allocate memory, remapping must succeed. Note,
1289 * that at this time old_fp has already been released
1294 fp = bpf_prog_select_runtime(fp, &err);
1304 __bpf_prog_release(fp);
1305 return ERR_PTR(err);
1308 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1309 bpf_aux_classic_check_t trans)
1313 fp->bpf_func = NULL;
1316 err = bpf_check_classic(fp->insns, fp->len);
1318 __bpf_prog_release(fp);
1319 return ERR_PTR(err);
1322 /* There might be additional checks and transformations
1323 * needed on classic filters, f.e. in case of seccomp.
1326 err = trans(fp->insns, fp->len);
1328 __bpf_prog_release(fp);
1329 return ERR_PTR(err);
1333 /* Probe if we can JIT compile the filter and if so, do
1334 * the compilation of the filter.
1336 bpf_jit_compile(fp);
1338 /* JIT compiler couldn't process this filter, so do the
1339 * internal BPF translation for the optimized interpreter.
1342 fp = bpf_migrate_filter(fp);
1348 * bpf_prog_create - create an unattached filter
1349 * @pfp: the unattached filter that is created
1350 * @fprog: the filter program
1352 * Create a filter independent of any socket. We first run some
1353 * sanity checks on it to make sure it does not explode on us later.
1354 * If an error occurs or there is insufficient memory for the filter
1355 * a negative errno code is returned. On success the return is zero.
1357 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1359 unsigned int fsize = bpf_classic_proglen(fprog);
1360 struct bpf_prog *fp;
1362 /* Make sure new filter is there and in the right amounts. */
1363 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1366 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1370 memcpy(fp->insns, fprog->filter, fsize);
1372 fp->len = fprog->len;
1373 /* Since unattached filters are not copied back to user
1374 * space through sk_get_filter(), we do not need to hold
1375 * a copy here, and can spare us the work.
1377 fp->orig_prog = NULL;
1379 /* bpf_prepare_filter() already takes care of freeing
1380 * memory in case something goes wrong.
1382 fp = bpf_prepare_filter(fp, NULL);
1389 EXPORT_SYMBOL_GPL(bpf_prog_create);
1392 * bpf_prog_create_from_user - create an unattached filter from user buffer
1393 * @pfp: the unattached filter that is created
1394 * @fprog: the filter program
1395 * @trans: post-classic verifier transformation handler
1396 * @save_orig: save classic BPF program
1398 * This function effectively does the same as bpf_prog_create(), only
1399 * that it builds up its insns buffer from user space provided buffer.
1400 * It also allows for passing a bpf_aux_classic_check_t handler.
1402 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1403 bpf_aux_classic_check_t trans, bool save_orig)
1405 unsigned int fsize = bpf_classic_proglen(fprog);
1406 struct bpf_prog *fp;
1409 /* Make sure new filter is there and in the right amounts. */
1410 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1413 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1417 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1418 __bpf_prog_free(fp);
1422 fp->len = fprog->len;
1423 fp->orig_prog = NULL;
1426 err = bpf_prog_store_orig_filter(fp, fprog);
1428 __bpf_prog_free(fp);
1433 /* bpf_prepare_filter() already takes care of freeing
1434 * memory in case something goes wrong.
1436 fp = bpf_prepare_filter(fp, trans);
1443 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1445 void bpf_prog_destroy(struct bpf_prog *fp)
1447 __bpf_prog_release(fp);
1449 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1451 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1453 struct sk_filter *fp, *old_fp;
1455 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1461 if (!__sk_filter_charge(sk, fp)) {
1465 refcount_set(&fp->refcnt, 1);
1467 old_fp = rcu_dereference_protected(sk->sk_filter,
1468 lockdep_sock_is_held(sk));
1469 rcu_assign_pointer(sk->sk_filter, fp);
1472 sk_filter_uncharge(sk, old_fp);
1478 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1480 unsigned int fsize = bpf_classic_proglen(fprog);
1481 struct bpf_prog *prog;
1484 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1485 return ERR_PTR(-EPERM);
1487 /* Make sure new filter is there and in the right amounts. */
1488 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1489 return ERR_PTR(-EINVAL);
1491 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1493 return ERR_PTR(-ENOMEM);
1495 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1496 __bpf_prog_free(prog);
1497 return ERR_PTR(-EFAULT);
1500 prog->len = fprog->len;
1502 err = bpf_prog_store_orig_filter(prog, fprog);
1504 __bpf_prog_free(prog);
1505 return ERR_PTR(-ENOMEM);
1508 /* bpf_prepare_filter() already takes care of freeing
1509 * memory in case something goes wrong.
1511 return bpf_prepare_filter(prog, NULL);
1515 * sk_attach_filter - attach a socket filter
1516 * @fprog: the filter program
1517 * @sk: the socket to use
1519 * Attach the user's filter code. We first run some sanity checks on
1520 * it to make sure it does not explode on us later. If an error
1521 * occurs or there is insufficient memory for the filter a negative
1522 * errno code is returned. On success the return is zero.
1524 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1526 struct bpf_prog *prog = __get_filter(fprog, sk);
1530 return PTR_ERR(prog);
1532 err = __sk_attach_prog(prog, sk);
1534 __bpf_prog_release(prog);
1540 EXPORT_SYMBOL_GPL(sk_attach_filter);
1542 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1544 struct bpf_prog *prog = __get_filter(fprog, sk);
1548 return PTR_ERR(prog);
1550 if (bpf_prog_size(prog->len) > sysctl_optmem_max)
1553 err = reuseport_attach_prog(sk, prog);
1556 __bpf_prog_release(prog);
1561 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1563 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1564 return ERR_PTR(-EPERM);
1566 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1569 int sk_attach_bpf(u32 ufd, struct sock *sk)
1571 struct bpf_prog *prog = __get_bpf(ufd, sk);
1575 return PTR_ERR(prog);
1577 err = __sk_attach_prog(prog, sk);
1586 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1588 struct bpf_prog *prog;
1591 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1594 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1595 if (PTR_ERR(prog) == -EINVAL)
1596 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1598 return PTR_ERR(prog);
1600 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1601 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1602 * bpf prog (e.g. sockmap). It depends on the
1603 * limitation imposed by bpf_prog_load().
1604 * Hence, sysctl_optmem_max is not checked.
1606 if ((sk->sk_type != SOCK_STREAM &&
1607 sk->sk_type != SOCK_DGRAM) ||
1608 (sk->sk_protocol != IPPROTO_UDP &&
1609 sk->sk_protocol != IPPROTO_TCP) ||
1610 (sk->sk_family != AF_INET &&
1611 sk->sk_family != AF_INET6)) {
1616 /* BPF_PROG_TYPE_SOCKET_FILTER */
1617 if (bpf_prog_size(prog->len) > sysctl_optmem_max) {
1623 err = reuseport_attach_prog(sk, prog);
1631 void sk_reuseport_prog_free(struct bpf_prog *prog)
1636 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1639 bpf_prog_destroy(prog);
1642 struct bpf_scratchpad {
1644 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1645 u8 buff[MAX_BPF_STACK];
1649 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1651 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1652 unsigned int write_len)
1654 return skb_ensure_writable(skb, write_len);
1657 static inline int bpf_try_make_writable(struct sk_buff *skb,
1658 unsigned int write_len)
1660 int err = __bpf_try_make_writable(skb, write_len);
1662 bpf_compute_data_pointers(skb);
1666 static int bpf_try_make_head_writable(struct sk_buff *skb)
1668 return bpf_try_make_writable(skb, skb_headlen(skb));
1671 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1673 if (skb_at_tc_ingress(skb))
1674 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1677 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1679 if (skb_at_tc_ingress(skb))
1680 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1683 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1684 const void *, from, u32, len, u64, flags)
1688 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1690 if (unlikely(offset > 0xffff))
1692 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1695 ptr = skb->data + offset;
1696 if (flags & BPF_F_RECOMPUTE_CSUM)
1697 __skb_postpull_rcsum(skb, ptr, len, offset);
1699 memcpy(ptr, from, len);
1701 if (flags & BPF_F_RECOMPUTE_CSUM)
1702 __skb_postpush_rcsum(skb, ptr, len, offset);
1703 if (flags & BPF_F_INVALIDATE_HASH)
1704 skb_clear_hash(skb);
1709 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1710 .func = bpf_skb_store_bytes,
1712 .ret_type = RET_INTEGER,
1713 .arg1_type = ARG_PTR_TO_CTX,
1714 .arg2_type = ARG_ANYTHING,
1715 .arg3_type = ARG_PTR_TO_MEM,
1716 .arg4_type = ARG_CONST_SIZE,
1717 .arg5_type = ARG_ANYTHING,
1720 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1721 void *, to, u32, len)
1725 if (unlikely(offset > 0xffff))
1728 ptr = skb_header_pointer(skb, offset, len, to);
1732 memcpy(to, ptr, len);
1740 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1741 .func = bpf_skb_load_bytes,
1743 .ret_type = RET_INTEGER,
1744 .arg1_type = ARG_PTR_TO_CTX,
1745 .arg2_type = ARG_ANYTHING,
1746 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1747 .arg4_type = ARG_CONST_SIZE,
1750 BPF_CALL_4(bpf_flow_dissector_load_bytes,
1751 const struct bpf_flow_dissector *, ctx, u32, offset,
1752 void *, to, u32, len)
1756 if (unlikely(offset > 0xffff))
1759 if (unlikely(!ctx->skb))
1762 ptr = skb_header_pointer(ctx->skb, offset, len, to);
1766 memcpy(to, ptr, len);
1774 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1775 .func = bpf_flow_dissector_load_bytes,
1777 .ret_type = RET_INTEGER,
1778 .arg1_type = ARG_PTR_TO_CTX,
1779 .arg2_type = ARG_ANYTHING,
1780 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1781 .arg4_type = ARG_CONST_SIZE,
1784 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1785 u32, offset, void *, to, u32, len, u32, start_header)
1787 u8 *end = skb_tail_pointer(skb);
1790 if (unlikely(offset > 0xffff))
1793 switch (start_header) {
1794 case BPF_HDR_START_MAC:
1795 if (unlikely(!skb_mac_header_was_set(skb)))
1797 start = skb_mac_header(skb);
1799 case BPF_HDR_START_NET:
1800 start = skb_network_header(skb);
1806 ptr = start + offset;
1808 if (likely(ptr + len <= end)) {
1809 memcpy(to, ptr, len);
1818 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1819 .func = bpf_skb_load_bytes_relative,
1821 .ret_type = RET_INTEGER,
1822 .arg1_type = ARG_PTR_TO_CTX,
1823 .arg2_type = ARG_ANYTHING,
1824 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1825 .arg4_type = ARG_CONST_SIZE,
1826 .arg5_type = ARG_ANYTHING,
1829 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1831 /* Idea is the following: should the needed direct read/write
1832 * test fail during runtime, we can pull in more data and redo
1833 * again, since implicitly, we invalidate previous checks here.
1835 * Or, since we know how much we need to make read/writeable,
1836 * this can be done once at the program beginning for direct
1837 * access case. By this we overcome limitations of only current
1838 * headroom being accessible.
1840 return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1843 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1844 .func = bpf_skb_pull_data,
1846 .ret_type = RET_INTEGER,
1847 .arg1_type = ARG_PTR_TO_CTX,
1848 .arg2_type = ARG_ANYTHING,
1851 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1853 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1856 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1857 .func = bpf_sk_fullsock,
1859 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
1860 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
1863 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1864 unsigned int write_len)
1866 int err = __bpf_try_make_writable(skb, write_len);
1868 bpf_compute_data_end_sk_skb(skb);
1872 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1874 /* Idea is the following: should the needed direct read/write
1875 * test fail during runtime, we can pull in more data and redo
1876 * again, since implicitly, we invalidate previous checks here.
1878 * Or, since we know how much we need to make read/writeable,
1879 * this can be done once at the program beginning for direct
1880 * access case. By this we overcome limitations of only current
1881 * headroom being accessible.
1883 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1886 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1887 .func = sk_skb_pull_data,
1889 .ret_type = RET_INTEGER,
1890 .arg1_type = ARG_PTR_TO_CTX,
1891 .arg2_type = ARG_ANYTHING,
1894 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1895 u64, from, u64, to, u64, flags)
1899 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1901 if (unlikely(offset > 0xffff || offset & 1))
1903 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1906 ptr = (__sum16 *)(skb->data + offset);
1907 switch (flags & BPF_F_HDR_FIELD_MASK) {
1909 if (unlikely(from != 0))
1912 csum_replace_by_diff(ptr, to);
1915 csum_replace2(ptr, from, to);
1918 csum_replace4(ptr, from, to);
1927 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1928 .func = bpf_l3_csum_replace,
1930 .ret_type = RET_INTEGER,
1931 .arg1_type = ARG_PTR_TO_CTX,
1932 .arg2_type = ARG_ANYTHING,
1933 .arg3_type = ARG_ANYTHING,
1934 .arg4_type = ARG_ANYTHING,
1935 .arg5_type = ARG_ANYTHING,
1938 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1939 u64, from, u64, to, u64, flags)
1941 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1942 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1943 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1946 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1947 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1949 if (unlikely(offset > 0xffff || offset & 1))
1951 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1954 ptr = (__sum16 *)(skb->data + offset);
1955 if (is_mmzero && !do_mforce && !*ptr)
1958 switch (flags & BPF_F_HDR_FIELD_MASK) {
1960 if (unlikely(from != 0))
1963 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1966 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1969 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1975 if (is_mmzero && !*ptr)
1976 *ptr = CSUM_MANGLED_0;
1980 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1981 .func = bpf_l4_csum_replace,
1983 .ret_type = RET_INTEGER,
1984 .arg1_type = ARG_PTR_TO_CTX,
1985 .arg2_type = ARG_ANYTHING,
1986 .arg3_type = ARG_ANYTHING,
1987 .arg4_type = ARG_ANYTHING,
1988 .arg5_type = ARG_ANYTHING,
1991 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1992 __be32 *, to, u32, to_size, __wsum, seed)
1994 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1995 u32 diff_size = from_size + to_size;
1998 /* This is quite flexible, some examples:
2000 * from_size == 0, to_size > 0, seed := csum --> pushing data
2001 * from_size > 0, to_size == 0, seed := csum --> pulling data
2002 * from_size > 0, to_size > 0, seed := 0 --> diffing data
2004 * Even for diffing, from_size and to_size don't need to be equal.
2006 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
2007 diff_size > sizeof(sp->diff)))
2010 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
2011 sp->diff[j] = ~from[i];
2012 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
2013 sp->diff[j] = to[i];
2015 return csum_partial(sp->diff, diff_size, seed);
2018 static const struct bpf_func_proto bpf_csum_diff_proto = {
2019 .func = bpf_csum_diff,
2022 .ret_type = RET_INTEGER,
2023 .arg1_type = ARG_PTR_TO_MEM_OR_NULL,
2024 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
2025 .arg3_type = ARG_PTR_TO_MEM_OR_NULL,
2026 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
2027 .arg5_type = ARG_ANYTHING,
2030 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2032 /* The interface is to be used in combination with bpf_csum_diff()
2033 * for direct packet writes. csum rotation for alignment as well
2034 * as emulating csum_sub() can be done from the eBPF program.
2036 if (skb->ip_summed == CHECKSUM_COMPLETE)
2037 return (skb->csum = csum_add(skb->csum, csum));
2042 static const struct bpf_func_proto bpf_csum_update_proto = {
2043 .func = bpf_csum_update,
2045 .ret_type = RET_INTEGER,
2046 .arg1_type = ARG_PTR_TO_CTX,
2047 .arg2_type = ARG_ANYTHING,
2050 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
2052 /* The interface is to be used in combination with bpf_skb_adjust_room()
2053 * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET
2054 * is passed as flags, for example.
2057 case BPF_CSUM_LEVEL_INC:
2058 __skb_incr_checksum_unnecessary(skb);
2060 case BPF_CSUM_LEVEL_DEC:
2061 __skb_decr_checksum_unnecessary(skb);
2063 case BPF_CSUM_LEVEL_RESET:
2064 __skb_reset_checksum_unnecessary(skb);
2066 case BPF_CSUM_LEVEL_QUERY:
2067 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2068 skb->csum_level : -EACCES;
2076 static const struct bpf_func_proto bpf_csum_level_proto = {
2077 .func = bpf_csum_level,
2079 .ret_type = RET_INTEGER,
2080 .arg1_type = ARG_PTR_TO_CTX,
2081 .arg2_type = ARG_ANYTHING,
2084 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2086 return dev_forward_skb(dev, skb);
2089 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2090 struct sk_buff *skb)
2092 int ret = ____dev_forward_skb(dev, skb);
2096 ret = netif_rx(skb);
2102 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2106 if (dev_xmit_recursion()) {
2107 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2115 dev_xmit_recursion_inc();
2116 ret = dev_queue_xmit(skb);
2117 dev_xmit_recursion_dec();
2122 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2125 unsigned int mlen = skb_network_offset(skb);
2128 __skb_pull(skb, mlen);
2130 /* At ingress, the mac header has already been pulled once.
2131 * At egress, skb_pospull_rcsum has to be done in case that
2132 * the skb is originated from ingress (i.e. a forwarded skb)
2133 * to ensure that rcsum starts at net header.
2135 if (!skb_at_tc_ingress(skb))
2136 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2138 skb_pop_mac_header(skb);
2139 skb_reset_mac_len(skb);
2140 return flags & BPF_F_INGRESS ?
2141 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2144 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2147 /* Verify that a link layer header is carried */
2148 if (unlikely(skb->mac_header >= skb->network_header)) {
2153 bpf_push_mac_rcsum(skb);
2154 return flags & BPF_F_INGRESS ?
2155 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2158 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2161 if (dev_is_mac_header_xmit(dev))
2162 return __bpf_redirect_common(skb, dev, flags);
2164 return __bpf_redirect_no_mac(skb, dev, flags);
2167 #if IS_ENABLED(CONFIG_IPV6)
2168 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb)
2170 struct dst_entry *dst = skb_dst(skb);
2171 struct net_device *dev = dst->dev;
2172 u32 hh_len = LL_RESERVED_SPACE(dev);
2173 const struct in6_addr *nexthop;
2174 struct neighbour *neigh;
2176 if (dev_xmit_recursion()) {
2177 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2184 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2185 struct sk_buff *skb2;
2187 skb2 = skb_realloc_headroom(skb, hh_len);
2188 if (unlikely(!skb2)) {
2193 skb_set_owner_w(skb2, skb->sk);
2199 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2200 &ipv6_hdr(skb)->daddr);
2201 neigh = ip_neigh_gw6(dev, nexthop);
2202 if (likely(!IS_ERR(neigh))) {
2205 sock_confirm_neigh(skb, neigh);
2206 dev_xmit_recursion_inc();
2207 ret = neigh_output(neigh, skb, false);
2208 dev_xmit_recursion_dec();
2209 rcu_read_unlock_bh();
2212 rcu_read_unlock_bh();
2213 IP6_INC_STATS(dev_net(dst->dev),
2214 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2220 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev)
2222 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2223 struct net *net = dev_net(dev);
2224 int err, ret = NET_XMIT_DROP;
2225 struct dst_entry *dst;
2226 struct flowi6 fl6 = {
2227 .flowi6_flags = FLOWI_FLAG_ANYSRC,
2228 .flowi6_mark = skb->mark,
2229 .flowlabel = ip6_flowinfo(ip6h),
2230 .flowi6_oif = dev->ifindex,
2231 .flowi6_proto = ip6h->nexthdr,
2232 .daddr = ip6h->daddr,
2233 .saddr = ip6h->saddr,
2236 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2240 skb_dst_set(skb, dst);
2242 err = bpf_out_neigh_v6(net, skb);
2243 if (unlikely(net_xmit_eval(err)))
2244 dev->stats.tx_errors++;
2246 ret = NET_XMIT_SUCCESS;
2249 dev->stats.tx_errors++;
2255 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev)
2258 return NET_XMIT_DROP;
2260 #endif /* CONFIG_IPV6 */
2262 #if IS_ENABLED(CONFIG_INET)
2263 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb)
2265 struct dst_entry *dst = skb_dst(skb);
2266 struct rtable *rt = container_of(dst, struct rtable, dst);
2267 struct net_device *dev = dst->dev;
2268 u32 hh_len = LL_RESERVED_SPACE(dev);
2269 struct neighbour *neigh;
2270 bool is_v6gw = false;
2272 if (dev_xmit_recursion()) {
2273 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2280 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2281 struct sk_buff *skb2;
2283 skb2 = skb_realloc_headroom(skb, hh_len);
2284 if (unlikely(!skb2)) {
2289 skb_set_owner_w(skb2, skb->sk);
2295 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2296 if (likely(!IS_ERR(neigh))) {
2299 sock_confirm_neigh(skb, neigh);
2300 dev_xmit_recursion_inc();
2301 ret = neigh_output(neigh, skb, is_v6gw);
2302 dev_xmit_recursion_dec();
2303 rcu_read_unlock_bh();
2306 rcu_read_unlock_bh();
2312 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev)
2314 const struct iphdr *ip4h = ip_hdr(skb);
2315 struct net *net = dev_net(dev);
2316 int err, ret = NET_XMIT_DROP;
2318 struct flowi4 fl4 = {
2319 .flowi4_flags = FLOWI_FLAG_ANYSRC,
2320 .flowi4_mark = skb->mark,
2321 .flowi4_tos = RT_TOS(ip4h->tos),
2322 .flowi4_oif = dev->ifindex,
2323 .flowi4_proto = ip4h->protocol,
2324 .daddr = ip4h->daddr,
2325 .saddr = ip4h->saddr,
2328 rt = ip_route_output_flow(net, &fl4, NULL);
2331 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2336 skb_dst_set(skb, &rt->dst);
2338 err = bpf_out_neigh_v4(net, skb);
2339 if (unlikely(net_xmit_eval(err)))
2340 dev->stats.tx_errors++;
2342 ret = NET_XMIT_SUCCESS;
2345 dev->stats.tx_errors++;
2351 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev)
2354 return NET_XMIT_DROP;
2356 #endif /* CONFIG_INET */
2358 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev)
2360 struct ethhdr *ethh = eth_hdr(skb);
2362 if (unlikely(skb->mac_header >= skb->network_header))
2364 bpf_push_mac_rcsum(skb);
2365 if (is_multicast_ether_addr(ethh->h_dest))
2368 skb_pull(skb, sizeof(*ethh));
2369 skb_unset_mac_header(skb);
2370 skb_reset_network_header(skb);
2372 if (skb->protocol == htons(ETH_P_IP))
2373 return __bpf_redirect_neigh_v4(skb, dev);
2374 else if (skb->protocol == htons(ETH_P_IPV6))
2375 return __bpf_redirect_neigh_v6(skb, dev);
2381 /* Internal, non-exposed redirect flags. */
2383 BPF_F_NEIGH = (1ULL << 1),
2384 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH)
2387 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2389 struct net_device *dev;
2390 struct sk_buff *clone;
2393 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2396 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2400 clone = skb_clone(skb, GFP_ATOMIC);
2401 if (unlikely(!clone))
2404 /* For direct write, we need to keep the invariant that the skbs
2405 * we're dealing with need to be uncloned. Should uncloning fail
2406 * here, we need to free the just generated clone to unclone once
2409 ret = bpf_try_make_head_writable(skb);
2410 if (unlikely(ret)) {
2415 return __bpf_redirect(clone, dev, flags);
2418 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2419 .func = bpf_clone_redirect,
2421 .ret_type = RET_INTEGER,
2422 .arg1_type = ARG_PTR_TO_CTX,
2423 .arg2_type = ARG_ANYTHING,
2424 .arg3_type = ARG_ANYTHING,
2427 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2428 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2430 int skb_do_redirect(struct sk_buff *skb)
2432 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2433 struct net_device *dev;
2434 u32 flags = ri->flags;
2436 dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->tgt_index);
2438 if (unlikely(!dev)) {
2443 return flags & BPF_F_NEIGH ?
2444 __bpf_redirect_neigh(skb, dev) :
2445 __bpf_redirect(skb, dev, flags);
2448 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2450 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2452 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2456 ri->tgt_index = ifindex;
2458 return TC_ACT_REDIRECT;
2461 static const struct bpf_func_proto bpf_redirect_proto = {
2462 .func = bpf_redirect,
2464 .ret_type = RET_INTEGER,
2465 .arg1_type = ARG_ANYTHING,
2466 .arg2_type = ARG_ANYTHING,
2469 BPF_CALL_2(bpf_redirect_neigh, u32, ifindex, u64, flags)
2471 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2473 if (unlikely(flags))
2476 ri->flags = BPF_F_NEIGH;
2477 ri->tgt_index = ifindex;
2479 return TC_ACT_REDIRECT;
2482 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2483 .func = bpf_redirect_neigh,
2485 .ret_type = RET_INTEGER,
2486 .arg1_type = ARG_ANYTHING,
2487 .arg2_type = ARG_ANYTHING,
2490 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2492 msg->apply_bytes = bytes;
2496 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2497 .func = bpf_msg_apply_bytes,
2499 .ret_type = RET_INTEGER,
2500 .arg1_type = ARG_PTR_TO_CTX,
2501 .arg2_type = ARG_ANYTHING,
2504 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2506 msg->cork_bytes = bytes;
2510 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2511 .func = bpf_msg_cork_bytes,
2513 .ret_type = RET_INTEGER,
2514 .arg1_type = ARG_PTR_TO_CTX,
2515 .arg2_type = ARG_ANYTHING,
2518 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2519 u32, end, u64, flags)
2521 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2522 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2523 struct scatterlist *sge;
2524 u8 *raw, *to, *from;
2527 if (unlikely(flags || end <= start))
2530 /* First find the starting scatterlist element */
2534 len = sk_msg_elem(msg, i)->length;
2535 if (start < offset + len)
2537 sk_msg_iter_var_next(i);
2538 } while (i != msg->sg.end);
2540 if (unlikely(start >= offset + len))
2544 /* The start may point into the sg element so we need to also
2545 * account for the headroom.
2547 bytes_sg_total = start - offset + bytes;
2548 if (!test_bit(i, &msg->sg.copy) && bytes_sg_total <= len)
2551 /* At this point we need to linearize multiple scatterlist
2552 * elements or a single shared page. Either way we need to
2553 * copy into a linear buffer exclusively owned by BPF. Then
2554 * place the buffer in the scatterlist and fixup the original
2555 * entries by removing the entries now in the linear buffer
2556 * and shifting the remaining entries. For now we do not try
2557 * to copy partial entries to avoid complexity of running out
2558 * of sg_entry slots. The downside is reading a single byte
2559 * will copy the entire sg entry.
2562 copy += sk_msg_elem(msg, i)->length;
2563 sk_msg_iter_var_next(i);
2564 if (bytes_sg_total <= copy)
2566 } while (i != msg->sg.end);
2569 if (unlikely(bytes_sg_total > copy))
2572 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2574 if (unlikely(!page))
2577 raw = page_address(page);
2580 sge = sk_msg_elem(msg, i);
2581 from = sg_virt(sge);
2585 memcpy(to, from, len);
2588 put_page(sg_page(sge));
2590 sk_msg_iter_var_next(i);
2591 } while (i != last_sge);
2593 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2595 /* To repair sg ring we need to shift entries. If we only
2596 * had a single entry though we can just replace it and
2597 * be done. Otherwise walk the ring and shift the entries.
2599 WARN_ON_ONCE(last_sge == first_sge);
2600 shift = last_sge > first_sge ?
2601 last_sge - first_sge - 1 :
2602 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2607 sk_msg_iter_var_next(i);
2611 if (i + shift >= NR_MSG_FRAG_IDS)
2612 move_from = i + shift - NR_MSG_FRAG_IDS;
2614 move_from = i + shift;
2615 if (move_from == msg->sg.end)
2618 msg->sg.data[i] = msg->sg.data[move_from];
2619 msg->sg.data[move_from].length = 0;
2620 msg->sg.data[move_from].page_link = 0;
2621 msg->sg.data[move_from].offset = 0;
2622 sk_msg_iter_var_next(i);
2625 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2626 msg->sg.end - shift + NR_MSG_FRAG_IDS :
2627 msg->sg.end - shift;
2629 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2630 msg->data_end = msg->data + bytes;
2634 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2635 .func = bpf_msg_pull_data,
2637 .ret_type = RET_INTEGER,
2638 .arg1_type = ARG_PTR_TO_CTX,
2639 .arg2_type = ARG_ANYTHING,
2640 .arg3_type = ARG_ANYTHING,
2641 .arg4_type = ARG_ANYTHING,
2644 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2645 u32, len, u64, flags)
2647 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2648 u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2649 u8 *raw, *to, *from;
2652 if (unlikely(flags))
2655 /* First find the starting scatterlist element */
2659 l = sk_msg_elem(msg, i)->length;
2661 if (start < offset + l)
2663 sk_msg_iter_var_next(i);
2664 } while (i != msg->sg.end);
2666 if (start >= offset + l)
2669 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2671 /* If no space available will fallback to copy, we need at
2672 * least one scatterlist elem available to push data into
2673 * when start aligns to the beginning of an element or two
2674 * when it falls inside an element. We handle the start equals
2675 * offset case because its the common case for inserting a
2678 if (!space || (space == 1 && start != offset))
2679 copy = msg->sg.data[i].length;
2681 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2682 get_order(copy + len));
2683 if (unlikely(!page))
2689 raw = page_address(page);
2691 psge = sk_msg_elem(msg, i);
2692 front = start - offset;
2693 back = psge->length - front;
2694 from = sg_virt(psge);
2697 memcpy(raw, from, front);
2701 to = raw + front + len;
2703 memcpy(to, from, back);
2706 put_page(sg_page(psge));
2707 } else if (start - offset) {
2708 psge = sk_msg_elem(msg, i);
2709 rsge = sk_msg_elem_cpy(msg, i);
2711 psge->length = start - offset;
2712 rsge.length -= psge->length;
2713 rsge.offset += start;
2715 sk_msg_iter_var_next(i);
2716 sg_unmark_end(psge);
2717 sg_unmark_end(&rsge);
2718 sk_msg_iter_next(msg, end);
2721 /* Slot(s) to place newly allocated data */
2724 /* Shift one or two slots as needed */
2726 sge = sk_msg_elem_cpy(msg, i);
2728 sk_msg_iter_var_next(i);
2729 sg_unmark_end(&sge);
2730 sk_msg_iter_next(msg, end);
2732 nsge = sk_msg_elem_cpy(msg, i);
2734 sk_msg_iter_var_next(i);
2735 nnsge = sk_msg_elem_cpy(msg, i);
2738 while (i != msg->sg.end) {
2739 msg->sg.data[i] = sge;
2741 sk_msg_iter_var_next(i);
2744 nnsge = sk_msg_elem_cpy(msg, i);
2746 nsge = sk_msg_elem_cpy(msg, i);
2751 /* Place newly allocated data buffer */
2752 sk_mem_charge(msg->sk, len);
2753 msg->sg.size += len;
2754 __clear_bit(new, &msg->sg.copy);
2755 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2757 get_page(sg_page(&rsge));
2758 sk_msg_iter_var_next(new);
2759 msg->sg.data[new] = rsge;
2762 sk_msg_compute_data_pointers(msg);
2766 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2767 .func = bpf_msg_push_data,
2769 .ret_type = RET_INTEGER,
2770 .arg1_type = ARG_PTR_TO_CTX,
2771 .arg2_type = ARG_ANYTHING,
2772 .arg3_type = ARG_ANYTHING,
2773 .arg4_type = ARG_ANYTHING,
2776 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2782 sk_msg_iter_var_next(i);
2783 msg->sg.data[prev] = msg->sg.data[i];
2784 } while (i != msg->sg.end);
2786 sk_msg_iter_prev(msg, end);
2789 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2791 struct scatterlist tmp, sge;
2793 sk_msg_iter_next(msg, end);
2794 sge = sk_msg_elem_cpy(msg, i);
2795 sk_msg_iter_var_next(i);
2796 tmp = sk_msg_elem_cpy(msg, i);
2798 while (i != msg->sg.end) {
2799 msg->sg.data[i] = sge;
2800 sk_msg_iter_var_next(i);
2802 tmp = sk_msg_elem_cpy(msg, i);
2806 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2807 u32, len, u64, flags)
2809 u32 i = 0, l = 0, space, offset = 0;
2810 u64 last = start + len;
2813 if (unlikely(flags))
2816 /* First find the starting scatterlist element */
2820 l = sk_msg_elem(msg, i)->length;
2822 if (start < offset + l)
2824 sk_msg_iter_var_next(i);
2825 } while (i != msg->sg.end);
2827 /* Bounds checks: start and pop must be inside message */
2828 if (start >= offset + l || last >= msg->sg.size)
2831 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2834 /* --------------| offset
2835 * -| start |-------- len -------|
2837 * |----- a ----|-------- pop -------|----- b ----|
2838 * |______________________________________________| length
2841 * a: region at front of scatter element to save
2842 * b: region at back of scatter element to save when length > A + pop
2843 * pop: region to pop from element, same as input 'pop' here will be
2844 * decremented below per iteration.
2846 * Two top-level cases to handle when start != offset, first B is non
2847 * zero and second B is zero corresponding to when a pop includes more
2850 * Then if B is non-zero AND there is no space allocate space and
2851 * compact A, B regions into page. If there is space shift ring to
2852 * the rigth free'ing the next element in ring to place B, leaving
2853 * A untouched except to reduce length.
2855 if (start != offset) {
2856 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2858 int b = sge->length - pop - a;
2860 sk_msg_iter_var_next(i);
2862 if (pop < sge->length - a) {
2865 sk_msg_shift_right(msg, i);
2866 nsge = sk_msg_elem(msg, i);
2867 get_page(sg_page(sge));
2870 b, sge->offset + pop + a);
2872 struct page *page, *orig;
2875 page = alloc_pages(__GFP_NOWARN |
2876 __GFP_COMP | GFP_ATOMIC,
2878 if (unlikely(!page))
2882 orig = sg_page(sge);
2883 from = sg_virt(sge);
2884 to = page_address(page);
2885 memcpy(to, from, a);
2886 memcpy(to + a, from + a + pop, b);
2887 sg_set_page(sge, page, a + b, 0);
2891 } else if (pop >= sge->length - a) {
2892 pop -= (sge->length - a);
2897 /* From above the current layout _must_ be as follows,
2902 * |---- pop ---|---------------- b ------------|
2903 * |____________________________________________| length
2905 * Offset and start of the current msg elem are equal because in the
2906 * previous case we handled offset != start and either consumed the
2907 * entire element and advanced to the next element OR pop == 0.
2909 * Two cases to handle here are first pop is less than the length
2910 * leaving some remainder b above. Simply adjust the element's layout
2911 * in this case. Or pop >= length of the element so that b = 0. In this
2912 * case advance to next element decrementing pop.
2915 struct scatterlist *sge = sk_msg_elem(msg, i);
2917 if (pop < sge->length) {
2923 sk_msg_shift_left(msg, i);
2925 sk_msg_iter_var_next(i);
2928 sk_mem_uncharge(msg->sk, len - pop);
2929 msg->sg.size -= (len - pop);
2930 sk_msg_compute_data_pointers(msg);
2934 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
2935 .func = bpf_msg_pop_data,
2937 .ret_type = RET_INTEGER,
2938 .arg1_type = ARG_PTR_TO_CTX,
2939 .arg2_type = ARG_ANYTHING,
2940 .arg3_type = ARG_ANYTHING,
2941 .arg4_type = ARG_ANYTHING,
2944 #ifdef CONFIG_CGROUP_NET_CLASSID
2945 BPF_CALL_0(bpf_get_cgroup_classid_curr)
2947 return __task_get_classid(current);
2950 static const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
2951 .func = bpf_get_cgroup_classid_curr,
2953 .ret_type = RET_INTEGER,
2956 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
2958 struct sock *sk = skb_to_full_sk(skb);
2960 if (!sk || !sk_fullsock(sk))
2963 return sock_cgroup_classid(&sk->sk_cgrp_data);
2966 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
2967 .func = bpf_skb_cgroup_classid,
2969 .ret_type = RET_INTEGER,
2970 .arg1_type = ARG_PTR_TO_CTX,
2974 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
2976 return task_get_classid(skb);
2979 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
2980 .func = bpf_get_cgroup_classid,
2982 .ret_type = RET_INTEGER,
2983 .arg1_type = ARG_PTR_TO_CTX,
2986 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
2988 return dst_tclassid(skb);
2991 static const struct bpf_func_proto bpf_get_route_realm_proto = {
2992 .func = bpf_get_route_realm,
2994 .ret_type = RET_INTEGER,
2995 .arg1_type = ARG_PTR_TO_CTX,
2998 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3000 /* If skb_clear_hash() was called due to mangling, we can
3001 * trigger SW recalculation here. Later access to hash
3002 * can then use the inline skb->hash via context directly
3003 * instead of calling this helper again.
3005 return skb_get_hash(skb);
3008 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3009 .func = bpf_get_hash_recalc,
3011 .ret_type = RET_INTEGER,
3012 .arg1_type = ARG_PTR_TO_CTX,
3015 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3017 /* After all direct packet write, this can be used once for
3018 * triggering a lazy recalc on next skb_get_hash() invocation.
3020 skb_clear_hash(skb);
3024 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3025 .func = bpf_set_hash_invalid,
3027 .ret_type = RET_INTEGER,
3028 .arg1_type = ARG_PTR_TO_CTX,
3031 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3033 /* Set user specified hash as L4(+), so that it gets returned
3034 * on skb_get_hash() call unless BPF prog later on triggers a
3037 __skb_set_sw_hash(skb, hash, true);
3041 static const struct bpf_func_proto bpf_set_hash_proto = {
3042 .func = bpf_set_hash,
3044 .ret_type = RET_INTEGER,
3045 .arg1_type = ARG_PTR_TO_CTX,
3046 .arg2_type = ARG_ANYTHING,
3049 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3054 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3055 vlan_proto != htons(ETH_P_8021AD)))
3056 vlan_proto = htons(ETH_P_8021Q);
3058 bpf_push_mac_rcsum(skb);
3059 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3060 bpf_pull_mac_rcsum(skb);
3062 bpf_compute_data_pointers(skb);
3066 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3067 .func = bpf_skb_vlan_push,
3069 .ret_type = RET_INTEGER,
3070 .arg1_type = ARG_PTR_TO_CTX,
3071 .arg2_type = ARG_ANYTHING,
3072 .arg3_type = ARG_ANYTHING,
3075 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3079 bpf_push_mac_rcsum(skb);
3080 ret = skb_vlan_pop(skb);
3081 bpf_pull_mac_rcsum(skb);
3083 bpf_compute_data_pointers(skb);
3087 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3088 .func = bpf_skb_vlan_pop,
3090 .ret_type = RET_INTEGER,
3091 .arg1_type = ARG_PTR_TO_CTX,
3094 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3096 /* Caller already did skb_cow() with len as headroom,
3097 * so no need to do it here.
3100 memmove(skb->data, skb->data + len, off);
3101 memset(skb->data + off, 0, len);
3103 /* No skb_postpush_rcsum(skb, skb->data + off, len)
3104 * needed here as it does not change the skb->csum
3105 * result for checksum complete when summing over
3111 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3113 /* skb_ensure_writable() is not needed here, as we're
3114 * already working on an uncloned skb.
3116 if (unlikely(!pskb_may_pull(skb, off + len)))
3119 skb_postpull_rcsum(skb, skb->data + off, len);
3120 memmove(skb->data + len, skb->data, off);
3121 __skb_pull(skb, len);
3126 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3128 bool trans_same = skb->transport_header == skb->network_header;
3131 /* There's no need for __skb_push()/__skb_pull() pair to
3132 * get to the start of the mac header as we're guaranteed
3133 * to always start from here under eBPF.
3135 ret = bpf_skb_generic_push(skb, off, len);
3137 skb->mac_header -= len;
3138 skb->network_header -= len;
3140 skb->transport_header = skb->network_header;
3146 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3148 bool trans_same = skb->transport_header == skb->network_header;
3151 /* Same here, __skb_push()/__skb_pull() pair not needed. */
3152 ret = bpf_skb_generic_pop(skb, off, len);
3154 skb->mac_header += len;
3155 skb->network_header += len;
3157 skb->transport_header = skb->network_header;
3163 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3165 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3166 u32 off = skb_mac_header_len(skb);
3169 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
3172 ret = skb_cow(skb, len_diff);
3173 if (unlikely(ret < 0))
3176 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3177 if (unlikely(ret < 0))
3180 if (skb_is_gso(skb)) {
3181 struct skb_shared_info *shinfo = skb_shinfo(skb);
3183 /* SKB_GSO_TCPV4 needs to be changed into
3186 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3187 shinfo->gso_type &= ~SKB_GSO_TCPV4;
3188 shinfo->gso_type |= SKB_GSO_TCPV6;
3191 /* Due to IPv6 header, MSS needs to be downgraded. */
3192 skb_decrease_gso_size(shinfo, len_diff);
3193 /* Header must be checked, and gso_segs recomputed. */
3194 shinfo->gso_type |= SKB_GSO_DODGY;
3195 shinfo->gso_segs = 0;
3198 skb->protocol = htons(ETH_P_IPV6);
3199 skb_clear_hash(skb);
3204 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3206 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3207 u32 off = skb_mac_header_len(skb);
3210 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
3213 ret = skb_unclone(skb, GFP_ATOMIC);
3214 if (unlikely(ret < 0))
3217 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3218 if (unlikely(ret < 0))
3221 if (skb_is_gso(skb)) {
3222 struct skb_shared_info *shinfo = skb_shinfo(skb);
3224 /* SKB_GSO_TCPV6 needs to be changed into
3227 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3228 shinfo->gso_type &= ~SKB_GSO_TCPV6;
3229 shinfo->gso_type |= SKB_GSO_TCPV4;
3232 /* Due to IPv4 header, MSS can be upgraded. */
3233 skb_increase_gso_size(shinfo, len_diff);
3234 /* Header must be checked, and gso_segs recomputed. */
3235 shinfo->gso_type |= SKB_GSO_DODGY;
3236 shinfo->gso_segs = 0;
3239 skb->protocol = htons(ETH_P_IP);
3240 skb_clear_hash(skb);
3245 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3247 __be16 from_proto = skb->protocol;
3249 if (from_proto == htons(ETH_P_IP) &&
3250 to_proto == htons(ETH_P_IPV6))
3251 return bpf_skb_proto_4_to_6(skb);
3253 if (from_proto == htons(ETH_P_IPV6) &&
3254 to_proto == htons(ETH_P_IP))
3255 return bpf_skb_proto_6_to_4(skb);
3260 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3265 if (unlikely(flags))
3268 /* General idea is that this helper does the basic groundwork
3269 * needed for changing the protocol, and eBPF program fills the
3270 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3271 * and other helpers, rather than passing a raw buffer here.
3273 * The rationale is to keep this minimal and without a need to
3274 * deal with raw packet data. F.e. even if we would pass buffers
3275 * here, the program still needs to call the bpf_lX_csum_replace()
3276 * helpers anyway. Plus, this way we keep also separation of
3277 * concerns, since f.e. bpf_skb_store_bytes() should only take
3280 * Currently, additional options and extension header space are
3281 * not supported, but flags register is reserved so we can adapt
3282 * that. For offloads, we mark packet as dodgy, so that headers
3283 * need to be verified first.
3285 ret = bpf_skb_proto_xlat(skb, proto);
3286 bpf_compute_data_pointers(skb);
3290 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3291 .func = bpf_skb_change_proto,
3293 .ret_type = RET_INTEGER,
3294 .arg1_type = ARG_PTR_TO_CTX,
3295 .arg2_type = ARG_ANYTHING,
3296 .arg3_type = ARG_ANYTHING,
3299 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3301 /* We only allow a restricted subset to be changed for now. */
3302 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3303 !skb_pkt_type_ok(pkt_type)))
3306 skb->pkt_type = pkt_type;
3310 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3311 .func = bpf_skb_change_type,
3313 .ret_type = RET_INTEGER,
3314 .arg1_type = ARG_PTR_TO_CTX,
3315 .arg2_type = ARG_ANYTHING,
3318 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3320 switch (skb->protocol) {
3321 case htons(ETH_P_IP):
3322 return sizeof(struct iphdr);
3323 case htons(ETH_P_IPV6):
3324 return sizeof(struct ipv6hdr);
3330 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3331 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3333 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \
3334 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3335 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3336 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3337 BPF_F_ADJ_ROOM_ENCAP_L2( \
3338 BPF_ADJ_ROOM_ENCAP_L2_MASK))
3340 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3343 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3344 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3345 u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3346 unsigned int gso_type = SKB_GSO_DODGY;
3349 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3350 /* udp gso_size delineates datagrams, only allow if fixed */
3351 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3352 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3356 ret = skb_cow_head(skb, len_diff);
3357 if (unlikely(ret < 0))
3361 if (skb->protocol != htons(ETH_P_IP) &&
3362 skb->protocol != htons(ETH_P_IPV6))
3365 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3366 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3369 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3370 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3373 if (skb->encapsulation)
3376 mac_len = skb->network_header - skb->mac_header;
3377 inner_net = skb->network_header;
3378 if (inner_mac_len > len_diff)
3380 inner_trans = skb->transport_header;
3383 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3384 if (unlikely(ret < 0))
3388 skb->inner_mac_header = inner_net - inner_mac_len;
3389 skb->inner_network_header = inner_net;
3390 skb->inner_transport_header = inner_trans;
3391 skb_set_inner_protocol(skb, skb->protocol);
3393 skb->encapsulation = 1;
3394 skb_set_network_header(skb, mac_len);
3396 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3397 gso_type |= SKB_GSO_UDP_TUNNEL;
3398 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3399 gso_type |= SKB_GSO_GRE;
3400 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3401 gso_type |= SKB_GSO_IPXIP6;
3402 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3403 gso_type |= SKB_GSO_IPXIP4;
3405 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3406 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3407 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3408 sizeof(struct ipv6hdr) :
3409 sizeof(struct iphdr);
3411 skb_set_transport_header(skb, mac_len + nh_len);
3414 /* Match skb->protocol to new outer l3 protocol */
3415 if (skb->protocol == htons(ETH_P_IP) &&
3416 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3417 skb->protocol = htons(ETH_P_IPV6);
3418 else if (skb->protocol == htons(ETH_P_IPV6) &&
3419 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3420 skb->protocol = htons(ETH_P_IP);
3423 if (skb_is_gso(skb)) {
3424 struct skb_shared_info *shinfo = skb_shinfo(skb);
3426 /* Due to header grow, MSS needs to be downgraded. */
3427 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3428 skb_decrease_gso_size(shinfo, len_diff);
3430 /* Header must be checked, and gso_segs recomputed. */
3431 shinfo->gso_type |= gso_type;
3432 shinfo->gso_segs = 0;
3438 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3443 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3444 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3447 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3448 /* udp gso_size delineates datagrams, only allow if fixed */
3449 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3450 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3454 ret = skb_unclone(skb, GFP_ATOMIC);
3455 if (unlikely(ret < 0))
3458 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3459 if (unlikely(ret < 0))
3462 if (skb_is_gso(skb)) {
3463 struct skb_shared_info *shinfo = skb_shinfo(skb);
3465 /* Due to header shrink, MSS can be upgraded. */
3466 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3467 skb_increase_gso_size(shinfo, len_diff);
3469 /* Header must be checked, and gso_segs recomputed. */
3470 shinfo->gso_type |= SKB_GSO_DODGY;
3471 shinfo->gso_segs = 0;
3477 static u32 __bpf_skb_max_len(const struct sk_buff *skb)
3479 return skb->dev ? skb->dev->mtu + skb->dev->hard_header_len :
3483 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3484 u32, mode, u64, flags)
3486 u32 len_diff_abs = abs(len_diff);
3487 bool shrink = len_diff < 0;
3490 if (unlikely(flags || mode))
3492 if (unlikely(len_diff_abs > 0xfffU))
3496 ret = skb_cow(skb, len_diff);
3497 if (unlikely(ret < 0))
3499 __skb_push(skb, len_diff_abs);
3500 memset(skb->data, 0, len_diff_abs);
3502 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3504 __skb_pull(skb, len_diff_abs);
3506 bpf_compute_data_end_sk_skb(skb);
3507 if (tls_sw_has_ctx_rx(skb->sk)) {
3508 struct strp_msg *rxm = strp_msg(skb);
3510 rxm->full_len += len_diff;
3515 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3516 .func = sk_skb_adjust_room,
3518 .ret_type = RET_INTEGER,
3519 .arg1_type = ARG_PTR_TO_CTX,
3520 .arg2_type = ARG_ANYTHING,
3521 .arg3_type = ARG_ANYTHING,
3522 .arg4_type = ARG_ANYTHING,
3525 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3526 u32, mode, u64, flags)
3528 u32 len_cur, len_diff_abs = abs(len_diff);
3529 u32 len_min = bpf_skb_net_base_len(skb);
3530 u32 len_max = __bpf_skb_max_len(skb);
3531 __be16 proto = skb->protocol;
3532 bool shrink = len_diff < 0;
3536 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3537 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3539 if (unlikely(len_diff_abs > 0xfffU))
3541 if (unlikely(proto != htons(ETH_P_IP) &&
3542 proto != htons(ETH_P_IPV6)))
3545 off = skb_mac_header_len(skb);
3547 case BPF_ADJ_ROOM_NET:
3548 off += bpf_skb_net_base_len(skb);
3550 case BPF_ADJ_ROOM_MAC:
3556 len_cur = skb->len - skb_network_offset(skb);
3557 if ((shrink && (len_diff_abs >= len_cur ||
3558 len_cur - len_diff_abs < len_min)) ||
3559 (!shrink && (skb->len + len_diff_abs > len_max &&
3563 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3564 bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3565 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3566 __skb_reset_checksum_unnecessary(skb);
3568 bpf_compute_data_pointers(skb);
3572 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3573 .func = bpf_skb_adjust_room,
3575 .ret_type = RET_INTEGER,
3576 .arg1_type = ARG_PTR_TO_CTX,
3577 .arg2_type = ARG_ANYTHING,
3578 .arg3_type = ARG_ANYTHING,
3579 .arg4_type = ARG_ANYTHING,
3582 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3584 u32 min_len = skb_network_offset(skb);
3586 if (skb_transport_header_was_set(skb))
3587 min_len = skb_transport_offset(skb);
3588 if (skb->ip_summed == CHECKSUM_PARTIAL)
3589 min_len = skb_checksum_start_offset(skb) +
3590 skb->csum_offset + sizeof(__sum16);
3594 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3596 unsigned int old_len = skb->len;
3599 ret = __skb_grow_rcsum(skb, new_len);
3601 memset(skb->data + old_len, 0, new_len - old_len);
3605 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3607 return __skb_trim_rcsum(skb, new_len);
3610 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3613 u32 max_len = __bpf_skb_max_len(skb);
3614 u32 min_len = __bpf_skb_min_len(skb);
3617 if (unlikely(flags || new_len > max_len || new_len < min_len))
3619 if (skb->encapsulation)
3622 /* The basic idea of this helper is that it's performing the
3623 * needed work to either grow or trim an skb, and eBPF program
3624 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3625 * bpf_lX_csum_replace() and others rather than passing a raw
3626 * buffer here. This one is a slow path helper and intended
3627 * for replies with control messages.
3629 * Like in bpf_skb_change_proto(), we want to keep this rather
3630 * minimal and without protocol specifics so that we are able
3631 * to separate concerns as in bpf_skb_store_bytes() should only
3632 * be the one responsible for writing buffers.
3634 * It's really expected to be a slow path operation here for
3635 * control message replies, so we're implicitly linearizing,
3636 * uncloning and drop offloads from the skb by this.
3638 ret = __bpf_try_make_writable(skb, skb->len);
3640 if (new_len > skb->len)
3641 ret = bpf_skb_grow_rcsum(skb, new_len);
3642 else if (new_len < skb->len)
3643 ret = bpf_skb_trim_rcsum(skb, new_len);
3644 if (!ret && skb_is_gso(skb))
3650 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3653 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3655 bpf_compute_data_pointers(skb);
3659 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3660 .func = bpf_skb_change_tail,
3662 .ret_type = RET_INTEGER,
3663 .arg1_type = ARG_PTR_TO_CTX,
3664 .arg2_type = ARG_ANYTHING,
3665 .arg3_type = ARG_ANYTHING,
3668 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3671 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3673 bpf_compute_data_end_sk_skb(skb);
3677 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3678 .func = sk_skb_change_tail,
3680 .ret_type = RET_INTEGER,
3681 .arg1_type = ARG_PTR_TO_CTX,
3682 .arg2_type = ARG_ANYTHING,
3683 .arg3_type = ARG_ANYTHING,
3686 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3689 u32 max_len = __bpf_skb_max_len(skb);
3690 u32 new_len = skb->len + head_room;
3693 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3694 new_len < skb->len))
3697 ret = skb_cow(skb, head_room);
3699 /* Idea for this helper is that we currently only
3700 * allow to expand on mac header. This means that
3701 * skb->protocol network header, etc, stay as is.
3702 * Compared to bpf_skb_change_tail(), we're more
3703 * flexible due to not needing to linearize or
3704 * reset GSO. Intention for this helper is to be
3705 * used by an L3 skb that needs to push mac header
3706 * for redirection into L2 device.
3708 __skb_push(skb, head_room);
3709 memset(skb->data, 0, head_room);
3710 skb_reset_mac_header(skb);
3716 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3719 int ret = __bpf_skb_change_head(skb, head_room, flags);
3721 bpf_compute_data_pointers(skb);
3725 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3726 .func = bpf_skb_change_head,
3728 .ret_type = RET_INTEGER,
3729 .arg1_type = ARG_PTR_TO_CTX,
3730 .arg2_type = ARG_ANYTHING,
3731 .arg3_type = ARG_ANYTHING,
3734 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3737 int ret = __bpf_skb_change_head(skb, head_room, flags);
3739 bpf_compute_data_end_sk_skb(skb);
3743 static const struct bpf_func_proto sk_skb_change_head_proto = {
3744 .func = sk_skb_change_head,
3746 .ret_type = RET_INTEGER,
3747 .arg1_type = ARG_PTR_TO_CTX,
3748 .arg2_type = ARG_ANYTHING,
3749 .arg3_type = ARG_ANYTHING,
3751 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3753 return xdp_data_meta_unsupported(xdp) ? 0 :
3754 xdp->data - xdp->data_meta;
3757 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3759 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3760 unsigned long metalen = xdp_get_metalen(xdp);
3761 void *data_start = xdp_frame_end + metalen;
3762 void *data = xdp->data + offset;
3764 if (unlikely(data < data_start ||
3765 data > xdp->data_end - ETH_HLEN))
3769 memmove(xdp->data_meta + offset,
3770 xdp->data_meta, metalen);
3771 xdp->data_meta += offset;
3777 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3778 .func = bpf_xdp_adjust_head,
3780 .ret_type = RET_INTEGER,
3781 .arg1_type = ARG_PTR_TO_CTX,
3782 .arg2_type = ARG_ANYTHING,
3785 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3787 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
3788 void *data_end = xdp->data_end + offset;
3790 /* Notice that xdp_data_hard_end have reserved some tailroom */
3791 if (unlikely(data_end > data_hard_end))
3794 /* ALL drivers MUST init xdp->frame_sz, chicken check below */
3795 if (unlikely(xdp->frame_sz > PAGE_SIZE)) {
3796 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz);
3800 if (unlikely(data_end < xdp->data + ETH_HLEN))
3803 /* Clear memory area on grow, can contain uninit kernel memory */
3805 memset(xdp->data_end, 0, offset);
3807 xdp->data_end = data_end;
3812 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3813 .func = bpf_xdp_adjust_tail,
3815 .ret_type = RET_INTEGER,
3816 .arg1_type = ARG_PTR_TO_CTX,
3817 .arg2_type = ARG_ANYTHING,
3820 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3822 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3823 void *meta = xdp->data_meta + offset;
3824 unsigned long metalen = xdp->data - meta;
3826 if (xdp_data_meta_unsupported(xdp))
3828 if (unlikely(meta < xdp_frame_end ||
3831 if (unlikely((metalen & (sizeof(__u32) - 1)) ||
3835 xdp->data_meta = meta;
3840 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3841 .func = bpf_xdp_adjust_meta,
3843 .ret_type = RET_INTEGER,
3844 .arg1_type = ARG_PTR_TO_CTX,
3845 .arg2_type = ARG_ANYTHING,
3848 static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd,
3849 struct bpf_map *map, struct xdp_buff *xdp)
3851 switch (map->map_type) {
3852 case BPF_MAP_TYPE_DEVMAP:
3853 case BPF_MAP_TYPE_DEVMAP_HASH:
3854 return dev_map_enqueue(fwd, xdp, dev_rx);
3855 case BPF_MAP_TYPE_CPUMAP:
3856 return cpu_map_enqueue(fwd, xdp, dev_rx);
3857 case BPF_MAP_TYPE_XSKMAP:
3858 return __xsk_map_redirect(fwd, xdp);
3865 void xdp_do_flush(void)
3871 EXPORT_SYMBOL_GPL(xdp_do_flush);
3873 static inline void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index)
3875 switch (map->map_type) {
3876 case BPF_MAP_TYPE_DEVMAP:
3877 return __dev_map_lookup_elem(map, index);
3878 case BPF_MAP_TYPE_DEVMAP_HASH:
3879 return __dev_map_hash_lookup_elem(map, index);
3880 case BPF_MAP_TYPE_CPUMAP:
3881 return __cpu_map_lookup_elem(map, index);
3882 case BPF_MAP_TYPE_XSKMAP:
3883 return __xsk_map_lookup_elem(map, index);
3889 void bpf_clear_redirect_map(struct bpf_map *map)
3891 struct bpf_redirect_info *ri;
3894 for_each_possible_cpu(cpu) {
3895 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3896 /* Avoid polluting remote cacheline due to writes if
3897 * not needed. Once we pass this test, we need the
3898 * cmpxchg() to make sure it hasn't been changed in
3899 * the meantime by remote CPU.
3901 if (unlikely(READ_ONCE(ri->map) == map))
3902 cmpxchg(&ri->map, map, NULL);
3906 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3907 struct bpf_prog *xdp_prog)
3909 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3910 struct bpf_map *map = READ_ONCE(ri->map);
3911 u32 index = ri->tgt_index;
3912 void *fwd = ri->tgt_value;
3916 ri->tgt_value = NULL;
3917 WRITE_ONCE(ri->map, NULL);
3919 if (unlikely(!map)) {
3920 fwd = dev_get_by_index_rcu(dev_net(dev), index);
3921 if (unlikely(!fwd)) {
3926 err = dev_xdp_enqueue(fwd, xdp, dev);
3928 err = __bpf_tx_xdp_map(dev, fwd, map, xdp);
3934 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3937 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3940 EXPORT_SYMBOL_GPL(xdp_do_redirect);
3942 static int xdp_do_generic_redirect_map(struct net_device *dev,
3943 struct sk_buff *skb,
3944 struct xdp_buff *xdp,
3945 struct bpf_prog *xdp_prog,
3946 struct bpf_map *map)
3948 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3949 u32 index = ri->tgt_index;
3950 void *fwd = ri->tgt_value;
3954 ri->tgt_value = NULL;
3955 WRITE_ONCE(ri->map, NULL);
3957 if (map->map_type == BPF_MAP_TYPE_DEVMAP ||
3958 map->map_type == BPF_MAP_TYPE_DEVMAP_HASH) {
3959 struct bpf_dtab_netdev *dst = fwd;
3961 err = dev_map_generic_redirect(dst, skb, xdp_prog);
3964 } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) {
3965 struct xdp_sock *xs = fwd;
3967 err = xsk_generic_rcv(xs, xdp);
3972 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
3977 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3980 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3984 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
3985 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
3987 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3988 struct bpf_map *map = READ_ONCE(ri->map);
3989 u32 index = ri->tgt_index;
3990 struct net_device *fwd;
3994 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog,
3997 fwd = dev_get_by_index_rcu(dev_net(dev), index);
3998 if (unlikely(!fwd)) {
4003 err = xdp_ok_fwd_dev(fwd, skb->len);
4008 _trace_xdp_redirect(dev, xdp_prog, index);
4009 generic_xdp_tx(skb, xdp_prog);
4012 _trace_xdp_redirect_err(dev, xdp_prog, index, err);
4016 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4018 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4020 if (unlikely(flags))
4024 ri->tgt_index = ifindex;
4025 ri->tgt_value = NULL;
4026 WRITE_ONCE(ri->map, NULL);
4028 return XDP_REDIRECT;
4031 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4032 .func = bpf_xdp_redirect,
4034 .ret_type = RET_INTEGER,
4035 .arg1_type = ARG_ANYTHING,
4036 .arg2_type = ARG_ANYTHING,
4039 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
4042 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4044 /* Lower bits of the flags are used as return code on lookup failure */
4045 if (unlikely(flags > XDP_TX))
4048 ri->tgt_value = __xdp_map_lookup_elem(map, ifindex);
4049 if (unlikely(!ri->tgt_value)) {
4050 /* If the lookup fails we want to clear out the state in the
4051 * redirect_info struct completely, so that if an eBPF program
4052 * performs multiple lookups, the last one always takes
4055 WRITE_ONCE(ri->map, NULL);
4060 ri->tgt_index = ifindex;
4061 WRITE_ONCE(ri->map, map);
4063 return XDP_REDIRECT;
4066 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4067 .func = bpf_xdp_redirect_map,
4069 .ret_type = RET_INTEGER,
4070 .arg1_type = ARG_CONST_MAP_PTR,
4071 .arg2_type = ARG_ANYTHING,
4072 .arg3_type = ARG_ANYTHING,
4075 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4076 unsigned long off, unsigned long len)
4078 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4082 if (ptr != dst_buff)
4083 memcpy(dst_buff, ptr, len);
4088 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4089 u64, flags, void *, meta, u64, meta_size)
4091 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4093 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4095 if (unlikely(!skb || skb_size > skb->len))
4098 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4102 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4103 .func = bpf_skb_event_output,
4105 .ret_type = RET_INTEGER,
4106 .arg1_type = ARG_PTR_TO_CTX,
4107 .arg2_type = ARG_CONST_MAP_PTR,
4108 .arg3_type = ARG_ANYTHING,
4109 .arg4_type = ARG_PTR_TO_MEM,
4110 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4113 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4115 const struct bpf_func_proto bpf_skb_output_proto = {
4116 .func = bpf_skb_event_output,
4118 .ret_type = RET_INTEGER,
4119 .arg1_type = ARG_PTR_TO_BTF_ID,
4120 .arg1_btf_id = &bpf_skb_output_btf_ids[0],
4121 .arg2_type = ARG_CONST_MAP_PTR,
4122 .arg3_type = ARG_ANYTHING,
4123 .arg4_type = ARG_PTR_TO_MEM,
4124 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4127 static unsigned short bpf_tunnel_key_af(u64 flags)
4129 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4132 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4133 u32, size, u64, flags)
4135 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4136 u8 compat[sizeof(struct bpf_tunnel_key)];
4140 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
4144 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4148 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4151 case offsetof(struct bpf_tunnel_key, tunnel_label):
4152 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4154 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4155 /* Fixup deprecated structure layouts here, so we have
4156 * a common path later on.
4158 if (ip_tunnel_info_af(info) != AF_INET)
4161 to = (struct bpf_tunnel_key *)compat;
4168 to->tunnel_id = be64_to_cpu(info->key.tun_id);
4169 to->tunnel_tos = info->key.tos;
4170 to->tunnel_ttl = info->key.ttl;
4173 if (flags & BPF_F_TUNINFO_IPV6) {
4174 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4175 sizeof(to->remote_ipv6));
4176 to->tunnel_label = be32_to_cpu(info->key.label);
4178 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4179 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4180 to->tunnel_label = 0;
4183 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4184 memcpy(to_orig, to, size);
4188 memset(to_orig, 0, size);
4192 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4193 .func = bpf_skb_get_tunnel_key,
4195 .ret_type = RET_INTEGER,
4196 .arg1_type = ARG_PTR_TO_CTX,
4197 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4198 .arg3_type = ARG_CONST_SIZE,
4199 .arg4_type = ARG_ANYTHING,
4202 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4204 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4207 if (unlikely(!info ||
4208 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4212 if (unlikely(size < info->options_len)) {
4217 ip_tunnel_info_opts_get(to, info);
4218 if (size > info->options_len)
4219 memset(to + info->options_len, 0, size - info->options_len);
4221 return info->options_len;
4223 memset(to, 0, size);
4227 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4228 .func = bpf_skb_get_tunnel_opt,
4230 .ret_type = RET_INTEGER,
4231 .arg1_type = ARG_PTR_TO_CTX,
4232 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4233 .arg3_type = ARG_CONST_SIZE,
4236 static struct metadata_dst __percpu *md_dst;
4238 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4239 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4241 struct metadata_dst *md = this_cpu_ptr(md_dst);
4242 u8 compat[sizeof(struct bpf_tunnel_key)];
4243 struct ip_tunnel_info *info;
4245 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4246 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
4248 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4250 case offsetof(struct bpf_tunnel_key, tunnel_label):
4251 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4252 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4253 /* Fixup deprecated structure layouts here, so we have
4254 * a common path later on.
4256 memcpy(compat, from, size);
4257 memset(compat + size, 0, sizeof(compat) - size);
4258 from = (const struct bpf_tunnel_key *) compat;
4264 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4269 dst_hold((struct dst_entry *) md);
4270 skb_dst_set(skb, (struct dst_entry *) md);
4272 info = &md->u.tun_info;
4273 memset(info, 0, sizeof(*info));
4274 info->mode = IP_TUNNEL_INFO_TX;
4276 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4277 if (flags & BPF_F_DONT_FRAGMENT)
4278 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4279 if (flags & BPF_F_ZERO_CSUM_TX)
4280 info->key.tun_flags &= ~TUNNEL_CSUM;
4281 if (flags & BPF_F_SEQ_NUMBER)
4282 info->key.tun_flags |= TUNNEL_SEQ;
4284 info->key.tun_id = cpu_to_be64(from->tunnel_id);
4285 info->key.tos = from->tunnel_tos;
4286 info->key.ttl = from->tunnel_ttl;
4288 if (flags & BPF_F_TUNINFO_IPV6) {
4289 info->mode |= IP_TUNNEL_INFO_IPV6;
4290 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4291 sizeof(from->remote_ipv6));
4292 info->key.label = cpu_to_be32(from->tunnel_label) &
4293 IPV6_FLOWLABEL_MASK;
4295 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4301 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4302 .func = bpf_skb_set_tunnel_key,
4304 .ret_type = RET_INTEGER,
4305 .arg1_type = ARG_PTR_TO_CTX,
4306 .arg2_type = ARG_PTR_TO_MEM,
4307 .arg3_type = ARG_CONST_SIZE,
4308 .arg4_type = ARG_ANYTHING,
4311 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4312 const u8 *, from, u32, size)
4314 struct ip_tunnel_info *info = skb_tunnel_info(skb);
4315 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4317 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4319 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4322 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4327 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4328 .func = bpf_skb_set_tunnel_opt,
4330 .ret_type = RET_INTEGER,
4331 .arg1_type = ARG_PTR_TO_CTX,
4332 .arg2_type = ARG_PTR_TO_MEM,
4333 .arg3_type = ARG_CONST_SIZE,
4336 static const struct bpf_func_proto *
4337 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4340 struct metadata_dst __percpu *tmp;
4342 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4347 if (cmpxchg(&md_dst, NULL, tmp))
4348 metadata_dst_free_percpu(tmp);
4352 case BPF_FUNC_skb_set_tunnel_key:
4353 return &bpf_skb_set_tunnel_key_proto;
4354 case BPF_FUNC_skb_set_tunnel_opt:
4355 return &bpf_skb_set_tunnel_opt_proto;
4361 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4364 struct bpf_array *array = container_of(map, struct bpf_array, map);
4365 struct cgroup *cgrp;
4368 sk = skb_to_full_sk(skb);
4369 if (!sk || !sk_fullsock(sk))
4371 if (unlikely(idx >= array->map.max_entries))
4374 cgrp = READ_ONCE(array->ptrs[idx]);
4375 if (unlikely(!cgrp))
4378 return sk_under_cgroup_hierarchy(sk, cgrp);
4381 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4382 .func = bpf_skb_under_cgroup,
4384 .ret_type = RET_INTEGER,
4385 .arg1_type = ARG_PTR_TO_CTX,
4386 .arg2_type = ARG_CONST_MAP_PTR,
4387 .arg3_type = ARG_ANYTHING,
4390 #ifdef CONFIG_SOCK_CGROUP_DATA
4391 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4393 struct cgroup *cgrp;
4395 sk = sk_to_full_sk(sk);
4396 if (!sk || !sk_fullsock(sk))
4399 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4400 return cgroup_id(cgrp);
4403 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4405 return __bpf_sk_cgroup_id(skb->sk);
4408 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4409 .func = bpf_skb_cgroup_id,
4411 .ret_type = RET_INTEGER,
4412 .arg1_type = ARG_PTR_TO_CTX,
4415 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4418 struct cgroup *ancestor;
4419 struct cgroup *cgrp;
4421 sk = sk_to_full_sk(sk);
4422 if (!sk || !sk_fullsock(sk))
4425 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4426 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4430 return cgroup_id(ancestor);
4433 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4436 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4439 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4440 .func = bpf_skb_ancestor_cgroup_id,
4442 .ret_type = RET_INTEGER,
4443 .arg1_type = ARG_PTR_TO_CTX,
4444 .arg2_type = ARG_ANYTHING,
4447 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4449 return __bpf_sk_cgroup_id(sk);
4452 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4453 .func = bpf_sk_cgroup_id,
4455 .ret_type = RET_INTEGER,
4456 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4459 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4461 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4464 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4465 .func = bpf_sk_ancestor_cgroup_id,
4467 .ret_type = RET_INTEGER,
4468 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4469 .arg2_type = ARG_ANYTHING,
4473 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4474 unsigned long off, unsigned long len)
4476 memcpy(dst_buff, src_buff + off, len);
4480 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4481 u64, flags, void *, meta, u64, meta_size)
4483 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4485 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4487 if (unlikely(!xdp ||
4488 xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4491 return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4492 xdp_size, bpf_xdp_copy);
4495 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4496 .func = bpf_xdp_event_output,
4498 .ret_type = RET_INTEGER,
4499 .arg1_type = ARG_PTR_TO_CTX,
4500 .arg2_type = ARG_CONST_MAP_PTR,
4501 .arg3_type = ARG_ANYTHING,
4502 .arg4_type = ARG_PTR_TO_MEM,
4503 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4506 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4508 const struct bpf_func_proto bpf_xdp_output_proto = {
4509 .func = bpf_xdp_event_output,
4511 .ret_type = RET_INTEGER,
4512 .arg1_type = ARG_PTR_TO_BTF_ID,
4513 .arg1_btf_id = &bpf_xdp_output_btf_ids[0],
4514 .arg2_type = ARG_CONST_MAP_PTR,
4515 .arg3_type = ARG_ANYTHING,
4516 .arg4_type = ARG_PTR_TO_MEM,
4517 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4520 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4522 return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4525 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4526 .func = bpf_get_socket_cookie,
4528 .ret_type = RET_INTEGER,
4529 .arg1_type = ARG_PTR_TO_CTX,
4532 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4534 return __sock_gen_cookie(ctx->sk);
4537 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4538 .func = bpf_get_socket_cookie_sock_addr,
4540 .ret_type = RET_INTEGER,
4541 .arg1_type = ARG_PTR_TO_CTX,
4544 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4546 return __sock_gen_cookie(ctx);
4549 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4550 .func = bpf_get_socket_cookie_sock,
4552 .ret_type = RET_INTEGER,
4553 .arg1_type = ARG_PTR_TO_CTX,
4556 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4558 return __sock_gen_cookie(ctx->sk);
4561 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4562 .func = bpf_get_socket_cookie_sock_ops,
4564 .ret_type = RET_INTEGER,
4565 .arg1_type = ARG_PTR_TO_CTX,
4568 static u64 __bpf_get_netns_cookie(struct sock *sk)
4570 #ifdef CONFIG_NET_NS
4571 return __net_gen_cookie(sk ? sk->sk_net.net : &init_net);
4577 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
4579 return __bpf_get_netns_cookie(ctx);
4582 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
4583 .func = bpf_get_netns_cookie_sock,
4585 .ret_type = RET_INTEGER,
4586 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4589 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4591 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4594 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
4595 .func = bpf_get_netns_cookie_sock_addr,
4597 .ret_type = RET_INTEGER,
4598 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4601 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4603 struct sock *sk = sk_to_full_sk(skb->sk);
4606 if (!sk || !sk_fullsock(sk))
4608 kuid = sock_net_uid(sock_net(sk), sk);
4609 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4612 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4613 .func = bpf_get_socket_uid,
4615 .ret_type = RET_INTEGER,
4616 .arg1_type = ARG_PTR_TO_CTX,
4619 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
4620 char *optval, int optlen)
4622 char devname[IFNAMSIZ];
4628 if (!sk_fullsock(sk))
4631 sock_owned_by_me(sk);
4633 if (level == SOL_SOCKET) {
4634 if (optlen != sizeof(int) && optname != SO_BINDTODEVICE)
4636 val = *((int *)optval);
4637 valbool = val ? 1 : 0;
4639 /* Only some socketops are supported */
4642 val = min_t(u32, val, sysctl_rmem_max);
4643 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4644 WRITE_ONCE(sk->sk_rcvbuf,
4645 max_t(int, val * 2, SOCK_MIN_RCVBUF));
4648 val = min_t(u32, val, sysctl_wmem_max);
4649 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4650 WRITE_ONCE(sk->sk_sndbuf,
4651 max_t(int, val * 2, SOCK_MIN_SNDBUF));
4653 case SO_MAX_PACING_RATE: /* 32bit version */
4655 cmpxchg(&sk->sk_pacing_status,
4658 sk->sk_max_pacing_rate = (val == ~0U) ? ~0UL : val;
4659 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4660 sk->sk_max_pacing_rate);
4663 sk->sk_priority = val;
4668 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
4671 if (sk->sk_mark != val) {
4676 case SO_BINDTODEVICE:
4677 optlen = min_t(long, optlen, IFNAMSIZ - 1);
4678 strncpy(devname, optval, optlen);
4679 devname[optlen] = 0;
4682 if (devname[0] != '\0') {
4683 struct net_device *dev;
4688 dev = dev_get_by_name(net, devname);
4691 ifindex = dev->ifindex;
4694 ret = sock_bindtoindex(sk, ifindex, false);
4697 if (sk->sk_prot->keepalive)
4698 sk->sk_prot->keepalive(sk, valbool);
4699 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
4705 } else if (level == SOL_IP) {
4706 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4709 val = *((int *)optval);
4710 /* Only some options are supported */
4713 if (val < -1 || val > 0xff) {
4716 struct inet_sock *inet = inet_sk(sk);
4726 #if IS_ENABLED(CONFIG_IPV6)
4727 } else if (level == SOL_IPV6) {
4728 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4731 val = *((int *)optval);
4732 /* Only some options are supported */
4735 if (val < -1 || val > 0xff) {
4738 struct ipv6_pinfo *np = inet6_sk(sk);
4749 } else if (level == SOL_TCP &&
4750 sk->sk_prot->setsockopt == tcp_setsockopt) {
4751 if (optname == TCP_CONGESTION) {
4752 char name[TCP_CA_NAME_MAX];
4754 strncpy(name, optval, min_t(long, optlen,
4755 TCP_CA_NAME_MAX-1));
4756 name[TCP_CA_NAME_MAX-1] = 0;
4757 ret = tcp_set_congestion_control(sk, name, false, true);
4759 struct inet_connection_sock *icsk = inet_csk(sk);
4760 struct tcp_sock *tp = tcp_sk(sk);
4761 unsigned long timeout;
4763 if (optlen != sizeof(int))
4766 val = *((int *)optval);
4767 /* Only some options are supported */
4770 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4775 case TCP_BPF_SNDCWND_CLAMP:
4779 tp->snd_cwnd_clamp = val;
4780 tp->snd_ssthresh = val;
4783 case TCP_BPF_DELACK_MAX:
4784 timeout = usecs_to_jiffies(val);
4785 if (timeout > TCP_DELACK_MAX ||
4786 timeout < TCP_TIMEOUT_MIN)
4788 inet_csk(sk)->icsk_delack_max = timeout;
4790 case TCP_BPF_RTO_MIN:
4791 timeout = usecs_to_jiffies(val);
4792 if (timeout > TCP_RTO_MIN ||
4793 timeout < TCP_TIMEOUT_MIN)
4795 inet_csk(sk)->icsk_rto_min = timeout;
4798 if (val < 0 || val > 1)
4804 ret = tcp_sock_set_keepidle_locked(sk, val);
4807 if (val < 1 || val > MAX_TCP_KEEPINTVL)
4810 tp->keepalive_intvl = val * HZ;
4813 if (val < 1 || val > MAX_TCP_KEEPCNT)
4816 tp->keepalive_probes = val;
4819 if (val < 1 || val > MAX_TCP_SYNCNT)
4822 icsk->icsk_syn_retries = val;
4824 case TCP_USER_TIMEOUT:
4828 icsk->icsk_user_timeout = val;
4841 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
4842 char *optval, int optlen)
4844 if (!sk_fullsock(sk))
4847 sock_owned_by_me(sk);
4850 if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4851 struct inet_connection_sock *icsk;
4852 struct tcp_sock *tp;
4855 case TCP_CONGESTION:
4856 icsk = inet_csk(sk);
4858 if (!icsk->icsk_ca_ops || optlen <= 1)
4860 strncpy(optval, icsk->icsk_ca_ops->name, optlen);
4861 optval[optlen - 1] = 0;
4866 if (optlen <= 0 || !tp->saved_syn ||
4867 optlen > tcp_saved_syn_len(tp->saved_syn))
4869 memcpy(optval, tp->saved_syn->data, optlen);
4874 } else if (level == SOL_IP) {
4875 struct inet_sock *inet = inet_sk(sk);
4877 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4880 /* Only some options are supported */
4883 *((int *)optval) = (int)inet->tos;
4888 #if IS_ENABLED(CONFIG_IPV6)
4889 } else if (level == SOL_IPV6) {
4890 struct ipv6_pinfo *np = inet6_sk(sk);
4892 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4895 /* Only some options are supported */
4898 *((int *)optval) = (int)np->tclass;
4910 memset(optval, 0, optlen);
4914 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
4915 int, level, int, optname, char *, optval, int, optlen)
4917 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
4920 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
4921 .func = bpf_sock_addr_setsockopt,
4923 .ret_type = RET_INTEGER,
4924 .arg1_type = ARG_PTR_TO_CTX,
4925 .arg2_type = ARG_ANYTHING,
4926 .arg3_type = ARG_ANYTHING,
4927 .arg4_type = ARG_PTR_TO_MEM,
4928 .arg5_type = ARG_CONST_SIZE,
4931 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
4932 int, level, int, optname, char *, optval, int, optlen)
4934 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
4937 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
4938 .func = bpf_sock_addr_getsockopt,
4940 .ret_type = RET_INTEGER,
4941 .arg1_type = ARG_PTR_TO_CTX,
4942 .arg2_type = ARG_ANYTHING,
4943 .arg3_type = ARG_ANYTHING,
4944 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
4945 .arg5_type = ARG_CONST_SIZE,
4948 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
4949 int, level, int, optname, char *, optval, int, optlen)
4951 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
4954 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
4955 .func = bpf_sock_ops_setsockopt,
4957 .ret_type = RET_INTEGER,
4958 .arg1_type = ARG_PTR_TO_CTX,
4959 .arg2_type = ARG_ANYTHING,
4960 .arg3_type = ARG_ANYTHING,
4961 .arg4_type = ARG_PTR_TO_MEM,
4962 .arg5_type = ARG_CONST_SIZE,
4965 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
4966 int optname, const u8 **start)
4968 struct sk_buff *syn_skb = bpf_sock->syn_skb;
4969 const u8 *hdr_start;
4973 /* sk is a request_sock here */
4975 if (optname == TCP_BPF_SYN) {
4976 hdr_start = syn_skb->data;
4977 ret = tcp_hdrlen(syn_skb);
4978 } else if (optname == TCP_BPF_SYN_IP) {
4979 hdr_start = skb_network_header(syn_skb);
4980 ret = skb_network_header_len(syn_skb) +
4981 tcp_hdrlen(syn_skb);
4983 /* optname == TCP_BPF_SYN_MAC */
4984 hdr_start = skb_mac_header(syn_skb);
4985 ret = skb_mac_header_len(syn_skb) +
4986 skb_network_header_len(syn_skb) +
4987 tcp_hdrlen(syn_skb);
4990 struct sock *sk = bpf_sock->sk;
4991 struct saved_syn *saved_syn;
4993 if (sk->sk_state == TCP_NEW_SYN_RECV)
4994 /* synack retransmit. bpf_sock->syn_skb will
4995 * not be available. It has to resort to
4996 * saved_syn (if it is saved).
4998 saved_syn = inet_reqsk(sk)->saved_syn;
5000 saved_syn = tcp_sk(sk)->saved_syn;
5005 if (optname == TCP_BPF_SYN) {
5006 hdr_start = saved_syn->data +
5007 saved_syn->mac_hdrlen +
5008 saved_syn->network_hdrlen;
5009 ret = saved_syn->tcp_hdrlen;
5010 } else if (optname == TCP_BPF_SYN_IP) {
5011 hdr_start = saved_syn->data +
5012 saved_syn->mac_hdrlen;
5013 ret = saved_syn->network_hdrlen +
5014 saved_syn->tcp_hdrlen;
5016 /* optname == TCP_BPF_SYN_MAC */
5018 /* TCP_SAVE_SYN may not have saved the mac hdr */
5019 if (!saved_syn->mac_hdrlen)
5022 hdr_start = saved_syn->data;
5023 ret = saved_syn->mac_hdrlen +
5024 saved_syn->network_hdrlen +
5025 saved_syn->tcp_hdrlen;
5033 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5034 int, level, int, optname, char *, optval, int, optlen)
5036 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5037 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5038 int ret, copy_len = 0;
5041 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5044 if (optlen < copy_len) {
5049 memcpy(optval, start, copy_len);
5052 /* Zero out unused buffer at the end */
5053 memset(optval + copy_len, 0, optlen - copy_len);
5058 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5061 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5062 .func = bpf_sock_ops_getsockopt,
5064 .ret_type = RET_INTEGER,
5065 .arg1_type = ARG_PTR_TO_CTX,
5066 .arg2_type = ARG_ANYTHING,
5067 .arg3_type = ARG_ANYTHING,
5068 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5069 .arg5_type = ARG_CONST_SIZE,
5072 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5075 struct sock *sk = bpf_sock->sk;
5076 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5078 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5081 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5083 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5086 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5087 .func = bpf_sock_ops_cb_flags_set,
5089 .ret_type = RET_INTEGER,
5090 .arg1_type = ARG_PTR_TO_CTX,
5091 .arg2_type = ARG_ANYTHING,
5094 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5095 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5097 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5101 struct sock *sk = ctx->sk;
5102 u32 flags = BIND_FROM_BPF;
5106 if (addr_len < offsetofend(struct sockaddr, sa_family))
5108 if (addr->sa_family == AF_INET) {
5109 if (addr_len < sizeof(struct sockaddr_in))
5111 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5112 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5113 return __inet_bind(sk, addr, addr_len, flags);
5114 #if IS_ENABLED(CONFIG_IPV6)
5115 } else if (addr->sa_family == AF_INET6) {
5116 if (addr_len < SIN6_LEN_RFC2133)
5118 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5119 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5120 /* ipv6_bpf_stub cannot be NULL, since it's called from
5121 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5123 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5124 #endif /* CONFIG_IPV6 */
5126 #endif /* CONFIG_INET */
5128 return -EAFNOSUPPORT;
5131 static const struct bpf_func_proto bpf_bind_proto = {
5134 .ret_type = RET_INTEGER,
5135 .arg1_type = ARG_PTR_TO_CTX,
5136 .arg2_type = ARG_PTR_TO_MEM,
5137 .arg3_type = ARG_CONST_SIZE,
5141 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5142 struct bpf_xfrm_state *, to, u32, size, u64, flags)
5144 const struct sec_path *sp = skb_sec_path(skb);
5145 const struct xfrm_state *x;
5147 if (!sp || unlikely(index >= sp->len || flags))
5150 x = sp->xvec[index];
5152 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5155 to->reqid = x->props.reqid;
5156 to->spi = x->id.spi;
5157 to->family = x->props.family;
5160 if (to->family == AF_INET6) {
5161 memcpy(to->remote_ipv6, x->props.saddr.a6,
5162 sizeof(to->remote_ipv6));
5164 to->remote_ipv4 = x->props.saddr.a4;
5165 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5170 memset(to, 0, size);
5174 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5175 .func = bpf_skb_get_xfrm_state,
5177 .ret_type = RET_INTEGER,
5178 .arg1_type = ARG_PTR_TO_CTX,
5179 .arg2_type = ARG_ANYTHING,
5180 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
5181 .arg4_type = ARG_CONST_SIZE,
5182 .arg5_type = ARG_ANYTHING,
5186 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5187 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
5188 const struct neighbour *neigh,
5189 const struct net_device *dev)
5191 memcpy(params->dmac, neigh->ha, ETH_ALEN);
5192 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5193 params->h_vlan_TCI = 0;
5194 params->h_vlan_proto = 0;
5195 params->ifindex = dev->ifindex;
5201 #if IS_ENABLED(CONFIG_INET)
5202 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5203 u32 flags, bool check_mtu)
5205 struct fib_nh_common *nhc;
5206 struct in_device *in_dev;
5207 struct neighbour *neigh;
5208 struct net_device *dev;
5209 struct fib_result res;
5214 dev = dev_get_by_index_rcu(net, params->ifindex);
5218 /* verify forwarding is enabled on this interface */
5219 in_dev = __in_dev_get_rcu(dev);
5220 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5221 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5223 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5225 fl4.flowi4_oif = params->ifindex;
5227 fl4.flowi4_iif = params->ifindex;
5230 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5231 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5232 fl4.flowi4_flags = 0;
5234 fl4.flowi4_proto = params->l4_protocol;
5235 fl4.daddr = params->ipv4_dst;
5236 fl4.saddr = params->ipv4_src;
5237 fl4.fl4_sport = params->sport;
5238 fl4.fl4_dport = params->dport;
5239 fl4.flowi4_multipath_hash = 0;
5241 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5242 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5243 struct fib_table *tb;
5245 tb = fib_get_table(net, tbid);
5247 return BPF_FIB_LKUP_RET_NOT_FWDED;
5249 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5251 fl4.flowi4_mark = 0;
5252 fl4.flowi4_secid = 0;
5253 fl4.flowi4_tun_key.tun_id = 0;
5254 fl4.flowi4_uid = sock_net_uid(net, NULL);
5256 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5260 /* map fib lookup errors to RTN_ type */
5262 return BPF_FIB_LKUP_RET_BLACKHOLE;
5263 if (err == -EHOSTUNREACH)
5264 return BPF_FIB_LKUP_RET_UNREACHABLE;
5266 return BPF_FIB_LKUP_RET_PROHIBIT;
5268 return BPF_FIB_LKUP_RET_NOT_FWDED;
5271 if (res.type != RTN_UNICAST)
5272 return BPF_FIB_LKUP_RET_NOT_FWDED;
5274 if (fib_info_num_path(res.fi) > 1)
5275 fib_select_path(net, &res, &fl4, NULL);
5278 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5279 if (params->tot_len > mtu)
5280 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5285 /* do not handle lwt encaps right now */
5286 if (nhc->nhc_lwtstate)
5287 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5291 params->rt_metric = res.fi->fib_priority;
5293 /* xdp and cls_bpf programs are run in RCU-bh so
5294 * rcu_read_lock_bh is not needed here
5296 if (likely(nhc->nhc_gw_family != AF_INET6)) {
5297 if (nhc->nhc_gw_family)
5298 params->ipv4_dst = nhc->nhc_gw.ipv4;
5300 neigh = __ipv4_neigh_lookup_noref(dev,
5301 (__force u32)params->ipv4_dst);
5303 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5305 params->family = AF_INET6;
5306 *dst = nhc->nhc_gw.ipv6;
5307 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5311 return BPF_FIB_LKUP_RET_NO_NEIGH;
5313 return bpf_fib_set_fwd_params(params, neigh, dev);
5317 #if IS_ENABLED(CONFIG_IPV6)
5318 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5319 u32 flags, bool check_mtu)
5321 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5322 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5323 struct fib6_result res = {};
5324 struct neighbour *neigh;
5325 struct net_device *dev;
5326 struct inet6_dev *idev;
5332 /* link local addresses are never forwarded */
5333 if (rt6_need_strict(dst) || rt6_need_strict(src))
5334 return BPF_FIB_LKUP_RET_NOT_FWDED;
5336 dev = dev_get_by_index_rcu(net, params->ifindex);
5340 idev = __in6_dev_get_safely(dev);
5341 if (unlikely(!idev || !idev->cnf.forwarding))
5342 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5344 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5346 oif = fl6.flowi6_oif = params->ifindex;
5348 oif = fl6.flowi6_iif = params->ifindex;
5350 strict = RT6_LOOKUP_F_HAS_SADDR;
5352 fl6.flowlabel = params->flowinfo;
5353 fl6.flowi6_scope = 0;
5354 fl6.flowi6_flags = 0;
5357 fl6.flowi6_proto = params->l4_protocol;
5360 fl6.fl6_sport = params->sport;
5361 fl6.fl6_dport = params->dport;
5363 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5364 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5365 struct fib6_table *tb;
5367 tb = ipv6_stub->fib6_get_table(net, tbid);
5369 return BPF_FIB_LKUP_RET_NOT_FWDED;
5371 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5374 fl6.flowi6_mark = 0;
5375 fl6.flowi6_secid = 0;
5376 fl6.flowi6_tun_key.tun_id = 0;
5377 fl6.flowi6_uid = sock_net_uid(net, NULL);
5379 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5382 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5383 res.f6i == net->ipv6.fib6_null_entry))
5384 return BPF_FIB_LKUP_RET_NOT_FWDED;
5386 switch (res.fib6_type) {
5387 /* only unicast is forwarded */
5391 return BPF_FIB_LKUP_RET_BLACKHOLE;
5392 case RTN_UNREACHABLE:
5393 return BPF_FIB_LKUP_RET_UNREACHABLE;
5395 return BPF_FIB_LKUP_RET_PROHIBIT;
5397 return BPF_FIB_LKUP_RET_NOT_FWDED;
5400 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5401 fl6.flowi6_oif != 0, NULL, strict);
5404 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5405 if (params->tot_len > mtu)
5406 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5409 if (res.nh->fib_nh_lws)
5410 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5412 if (res.nh->fib_nh_gw_family)
5413 *dst = res.nh->fib_nh_gw6;
5415 dev = res.nh->fib_nh_dev;
5416 params->rt_metric = res.f6i->fib6_metric;
5418 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
5421 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5423 return BPF_FIB_LKUP_RET_NO_NEIGH;
5425 return bpf_fib_set_fwd_params(params, neigh, dev);
5429 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
5430 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5432 if (plen < sizeof(*params))
5435 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5438 switch (params->family) {
5439 #if IS_ENABLED(CONFIG_INET)
5441 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
5444 #if IS_ENABLED(CONFIG_IPV6)
5446 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
5450 return -EAFNOSUPPORT;
5453 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
5454 .func = bpf_xdp_fib_lookup,
5456 .ret_type = RET_INTEGER,
5457 .arg1_type = ARG_PTR_TO_CTX,
5458 .arg2_type = ARG_PTR_TO_MEM,
5459 .arg3_type = ARG_CONST_SIZE,
5460 .arg4_type = ARG_ANYTHING,
5463 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
5464 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5466 struct net *net = dev_net(skb->dev);
5467 int rc = -EAFNOSUPPORT;
5469 if (plen < sizeof(*params))
5472 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5475 switch (params->family) {
5476 #if IS_ENABLED(CONFIG_INET)
5478 rc = bpf_ipv4_fib_lookup(net, params, flags, false);
5481 #if IS_ENABLED(CONFIG_IPV6)
5483 rc = bpf_ipv6_fib_lookup(net, params, flags, false);
5489 struct net_device *dev;
5491 dev = dev_get_by_index_rcu(net, params->ifindex);
5492 if (!is_skb_forwardable(dev, skb))
5493 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
5499 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
5500 .func = bpf_skb_fib_lookup,
5502 .ret_type = RET_INTEGER,
5503 .arg1_type = ARG_PTR_TO_CTX,
5504 .arg2_type = ARG_PTR_TO_MEM,
5505 .arg3_type = ARG_CONST_SIZE,
5506 .arg4_type = ARG_ANYTHING,
5509 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5510 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
5513 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
5515 if (!seg6_validate_srh(srh, len, false))
5519 case BPF_LWT_ENCAP_SEG6_INLINE:
5520 if (skb->protocol != htons(ETH_P_IPV6))
5523 err = seg6_do_srh_inline(skb, srh);
5525 case BPF_LWT_ENCAP_SEG6:
5526 skb_reset_inner_headers(skb);
5527 skb->encapsulation = 1;
5528 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
5534 bpf_compute_data_pointers(skb);
5538 ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5539 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
5541 return seg6_lookup_nexthop(skb, NULL, 0);
5543 #endif /* CONFIG_IPV6_SEG6_BPF */
5545 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5546 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
5549 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
5553 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
5557 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5558 case BPF_LWT_ENCAP_SEG6:
5559 case BPF_LWT_ENCAP_SEG6_INLINE:
5560 return bpf_push_seg6_encap(skb, type, hdr, len);
5562 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5563 case BPF_LWT_ENCAP_IP:
5564 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
5571 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
5572 void *, hdr, u32, len)
5575 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5576 case BPF_LWT_ENCAP_IP:
5577 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
5584 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
5585 .func = bpf_lwt_in_push_encap,
5587 .ret_type = RET_INTEGER,
5588 .arg1_type = ARG_PTR_TO_CTX,
5589 .arg2_type = ARG_ANYTHING,
5590 .arg3_type = ARG_PTR_TO_MEM,
5591 .arg4_type = ARG_CONST_SIZE
5594 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
5595 .func = bpf_lwt_xmit_push_encap,
5597 .ret_type = RET_INTEGER,
5598 .arg1_type = ARG_PTR_TO_CTX,
5599 .arg2_type = ARG_ANYTHING,
5600 .arg3_type = ARG_PTR_TO_MEM,
5601 .arg4_type = ARG_CONST_SIZE
5604 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5605 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
5606 const void *, from, u32, len)
5608 struct seg6_bpf_srh_state *srh_state =
5609 this_cpu_ptr(&seg6_bpf_srh_states);
5610 struct ipv6_sr_hdr *srh = srh_state->srh;
5611 void *srh_tlvs, *srh_end, *ptr;
5617 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
5618 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
5620 ptr = skb->data + offset;
5621 if (ptr >= srh_tlvs && ptr + len <= srh_end)
5622 srh_state->valid = false;
5623 else if (ptr < (void *)&srh->flags ||
5624 ptr + len > (void *)&srh->segments)
5627 if (unlikely(bpf_try_make_writable(skb, offset + len)))
5629 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5631 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5633 memcpy(skb->data + offset, from, len);
5637 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
5638 .func = bpf_lwt_seg6_store_bytes,
5640 .ret_type = RET_INTEGER,
5641 .arg1_type = ARG_PTR_TO_CTX,
5642 .arg2_type = ARG_ANYTHING,
5643 .arg3_type = ARG_PTR_TO_MEM,
5644 .arg4_type = ARG_CONST_SIZE
5647 static void bpf_update_srh_state(struct sk_buff *skb)
5649 struct seg6_bpf_srh_state *srh_state =
5650 this_cpu_ptr(&seg6_bpf_srh_states);
5653 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
5654 srh_state->srh = NULL;
5656 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5657 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
5658 srh_state->valid = true;
5662 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
5663 u32, action, void *, param, u32, param_len)
5665 struct seg6_bpf_srh_state *srh_state =
5666 this_cpu_ptr(&seg6_bpf_srh_states);
5671 case SEG6_LOCAL_ACTION_END_X:
5672 if (!seg6_bpf_has_valid_srh(skb))
5674 if (param_len != sizeof(struct in6_addr))
5676 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
5677 case SEG6_LOCAL_ACTION_END_T:
5678 if (!seg6_bpf_has_valid_srh(skb))
5680 if (param_len != sizeof(int))
5682 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5683 case SEG6_LOCAL_ACTION_END_DT6:
5684 if (!seg6_bpf_has_valid_srh(skb))
5686 if (param_len != sizeof(int))
5689 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
5691 if (!pskb_pull(skb, hdroff))
5694 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
5695 skb_reset_network_header(skb);
5696 skb_reset_transport_header(skb);
5697 skb->encapsulation = 0;
5699 bpf_compute_data_pointers(skb);
5700 bpf_update_srh_state(skb);
5701 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5702 case SEG6_LOCAL_ACTION_END_B6:
5703 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5705 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
5708 bpf_update_srh_state(skb);
5711 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
5712 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5714 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
5717 bpf_update_srh_state(skb);
5725 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
5726 .func = bpf_lwt_seg6_action,
5728 .ret_type = RET_INTEGER,
5729 .arg1_type = ARG_PTR_TO_CTX,
5730 .arg2_type = ARG_ANYTHING,
5731 .arg3_type = ARG_PTR_TO_MEM,
5732 .arg4_type = ARG_CONST_SIZE
5735 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
5738 struct seg6_bpf_srh_state *srh_state =
5739 this_cpu_ptr(&seg6_bpf_srh_states);
5740 struct ipv6_sr_hdr *srh = srh_state->srh;
5741 void *srh_end, *srh_tlvs, *ptr;
5742 struct ipv6hdr *hdr;
5746 if (unlikely(srh == NULL))
5749 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
5750 ((srh->first_segment + 1) << 4));
5751 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
5753 ptr = skb->data + offset;
5755 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
5757 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
5761 ret = skb_cow_head(skb, len);
5762 if (unlikely(ret < 0))
5765 ret = bpf_skb_net_hdr_push(skb, offset, len);
5767 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
5770 bpf_compute_data_pointers(skb);
5771 if (unlikely(ret < 0))
5774 hdr = (struct ipv6hdr *)skb->data;
5775 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5777 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5779 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5780 srh_state->hdrlen += len;
5781 srh_state->valid = false;
5785 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
5786 .func = bpf_lwt_seg6_adjust_srh,
5788 .ret_type = RET_INTEGER,
5789 .arg1_type = ARG_PTR_TO_CTX,
5790 .arg2_type = ARG_ANYTHING,
5791 .arg3_type = ARG_ANYTHING,
5793 #endif /* CONFIG_IPV6_SEG6_BPF */
5796 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
5797 int dif, int sdif, u8 family, u8 proto)
5799 bool refcounted = false;
5800 struct sock *sk = NULL;
5802 if (family == AF_INET) {
5803 __be32 src4 = tuple->ipv4.saddr;
5804 __be32 dst4 = tuple->ipv4.daddr;
5806 if (proto == IPPROTO_TCP)
5807 sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
5808 src4, tuple->ipv4.sport,
5809 dst4, tuple->ipv4.dport,
5810 dif, sdif, &refcounted);
5812 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
5813 dst4, tuple->ipv4.dport,
5814 dif, sdif, &udp_table, NULL);
5815 #if IS_ENABLED(CONFIG_IPV6)
5817 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
5818 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
5820 if (proto == IPPROTO_TCP)
5821 sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
5822 src6, tuple->ipv6.sport,
5823 dst6, ntohs(tuple->ipv6.dport),
5824 dif, sdif, &refcounted);
5825 else if (likely(ipv6_bpf_stub))
5826 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
5827 src6, tuple->ipv6.sport,
5828 dst6, tuple->ipv6.dport,
5834 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
5835 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
5841 /* bpf_skc_lookup performs the core lookup for different types of sockets,
5842 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
5843 * Returns the socket as an 'unsigned long' to simplify the casting in the
5844 * callers to satisfy BPF_CALL declarations.
5846 static struct sock *
5847 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5848 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
5851 struct sock *sk = NULL;
5852 u8 family = AF_UNSPEC;
5856 if (len == sizeof(tuple->ipv4))
5858 else if (len == sizeof(tuple->ipv6))
5863 if (unlikely(family == AF_UNSPEC || flags ||
5864 !((s32)netns_id < 0 || netns_id <= S32_MAX)))
5867 if (family == AF_INET)
5868 sdif = inet_sdif(skb);
5870 sdif = inet6_sdif(skb);
5872 if ((s32)netns_id < 0) {
5874 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
5876 net = get_net_ns_by_id(caller_net, netns_id);
5879 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
5887 static struct sock *
5888 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5889 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
5892 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
5893 ifindex, proto, netns_id, flags);
5896 sk = sk_to_full_sk(sk);
5897 if (!sk_fullsock(sk)) {
5906 static struct sock *
5907 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5908 u8 proto, u64 netns_id, u64 flags)
5910 struct net *caller_net;
5914 caller_net = dev_net(skb->dev);
5915 ifindex = skb->dev->ifindex;
5917 caller_net = sock_net(skb->sk);
5921 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
5925 static struct sock *
5926 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5927 u8 proto, u64 netns_id, u64 flags)
5929 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
5933 sk = sk_to_full_sk(sk);
5934 if (!sk_fullsock(sk)) {
5943 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
5944 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5946 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
5950 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
5951 .func = bpf_skc_lookup_tcp,
5954 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
5955 .arg1_type = ARG_PTR_TO_CTX,
5956 .arg2_type = ARG_PTR_TO_MEM,
5957 .arg3_type = ARG_CONST_SIZE,
5958 .arg4_type = ARG_ANYTHING,
5959 .arg5_type = ARG_ANYTHING,
5962 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
5963 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5965 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
5969 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
5970 .func = bpf_sk_lookup_tcp,
5973 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5974 .arg1_type = ARG_PTR_TO_CTX,
5975 .arg2_type = ARG_PTR_TO_MEM,
5976 .arg3_type = ARG_CONST_SIZE,
5977 .arg4_type = ARG_ANYTHING,
5978 .arg5_type = ARG_ANYTHING,
5981 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
5982 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5984 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
5988 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
5989 .func = bpf_sk_lookup_udp,
5992 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5993 .arg1_type = ARG_PTR_TO_CTX,
5994 .arg2_type = ARG_PTR_TO_MEM,
5995 .arg3_type = ARG_CONST_SIZE,
5996 .arg4_type = ARG_ANYTHING,
5997 .arg5_type = ARG_ANYTHING,
6000 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6002 if (sk && sk_is_refcounted(sk))
6007 static const struct bpf_func_proto bpf_sk_release_proto = {
6008 .func = bpf_sk_release,
6010 .ret_type = RET_INTEGER,
6011 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6014 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6015 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6017 struct net *caller_net = dev_net(ctx->rxq->dev);
6018 int ifindex = ctx->rxq->dev->ifindex;
6020 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6021 ifindex, IPPROTO_UDP, netns_id,
6025 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6026 .func = bpf_xdp_sk_lookup_udp,
6029 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6030 .arg1_type = ARG_PTR_TO_CTX,
6031 .arg2_type = ARG_PTR_TO_MEM,
6032 .arg3_type = ARG_CONST_SIZE,
6033 .arg4_type = ARG_ANYTHING,
6034 .arg5_type = ARG_ANYTHING,
6037 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6038 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6040 struct net *caller_net = dev_net(ctx->rxq->dev);
6041 int ifindex = ctx->rxq->dev->ifindex;
6043 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6044 ifindex, IPPROTO_TCP, netns_id,
6048 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6049 .func = bpf_xdp_skc_lookup_tcp,
6052 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6053 .arg1_type = ARG_PTR_TO_CTX,
6054 .arg2_type = ARG_PTR_TO_MEM,
6055 .arg3_type = ARG_CONST_SIZE,
6056 .arg4_type = ARG_ANYTHING,
6057 .arg5_type = ARG_ANYTHING,
6060 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6061 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6063 struct net *caller_net = dev_net(ctx->rxq->dev);
6064 int ifindex = ctx->rxq->dev->ifindex;
6066 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6067 ifindex, IPPROTO_TCP, netns_id,
6071 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6072 .func = bpf_xdp_sk_lookup_tcp,
6075 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6076 .arg1_type = ARG_PTR_TO_CTX,
6077 .arg2_type = ARG_PTR_TO_MEM,
6078 .arg3_type = ARG_CONST_SIZE,
6079 .arg4_type = ARG_ANYTHING,
6080 .arg5_type = ARG_ANYTHING,
6083 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6084 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6086 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6087 sock_net(ctx->sk), 0,
6088 IPPROTO_TCP, netns_id, flags);
6091 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6092 .func = bpf_sock_addr_skc_lookup_tcp,
6094 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6095 .arg1_type = ARG_PTR_TO_CTX,
6096 .arg2_type = ARG_PTR_TO_MEM,
6097 .arg3_type = ARG_CONST_SIZE,
6098 .arg4_type = ARG_ANYTHING,
6099 .arg5_type = ARG_ANYTHING,
6102 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6103 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6105 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6106 sock_net(ctx->sk), 0, IPPROTO_TCP,
6110 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6111 .func = bpf_sock_addr_sk_lookup_tcp,
6113 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6114 .arg1_type = ARG_PTR_TO_CTX,
6115 .arg2_type = ARG_PTR_TO_MEM,
6116 .arg3_type = ARG_CONST_SIZE,
6117 .arg4_type = ARG_ANYTHING,
6118 .arg5_type = ARG_ANYTHING,
6121 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6122 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6124 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6125 sock_net(ctx->sk), 0, IPPROTO_UDP,
6129 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6130 .func = bpf_sock_addr_sk_lookup_udp,
6132 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6133 .arg1_type = ARG_PTR_TO_CTX,
6134 .arg2_type = ARG_PTR_TO_MEM,
6135 .arg3_type = ARG_CONST_SIZE,
6136 .arg4_type = ARG_ANYTHING,
6137 .arg5_type = ARG_ANYTHING,
6140 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6141 struct bpf_insn_access_aux *info)
6143 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6147 if (off % size != 0)
6151 case offsetof(struct bpf_tcp_sock, bytes_received):
6152 case offsetof(struct bpf_tcp_sock, bytes_acked):
6153 return size == sizeof(__u64);
6155 return size == sizeof(__u32);
6159 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6160 const struct bpf_insn *si,
6161 struct bpf_insn *insn_buf,
6162 struct bpf_prog *prog, u32 *target_size)
6164 struct bpf_insn *insn = insn_buf;
6166 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \
6168 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \
6169 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6170 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6171 si->dst_reg, si->src_reg, \
6172 offsetof(struct tcp_sock, FIELD)); \
6175 #define BPF_INET_SOCK_GET_COMMON(FIELD) \
6177 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \
6179 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6180 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6181 struct inet_connection_sock, \
6183 si->dst_reg, si->src_reg, \
6185 struct inet_connection_sock, \
6189 if (insn > insn_buf)
6190 return insn - insn_buf;
6193 case offsetof(struct bpf_tcp_sock, rtt_min):
6194 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
6195 sizeof(struct minmax));
6196 BUILD_BUG_ON(sizeof(struct minmax) <
6197 sizeof(struct minmax_sample));
6199 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6200 offsetof(struct tcp_sock, rtt_min) +
6201 offsetof(struct minmax_sample, v));
6203 case offsetof(struct bpf_tcp_sock, snd_cwnd):
6204 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
6206 case offsetof(struct bpf_tcp_sock, srtt_us):
6207 BPF_TCP_SOCK_GET_COMMON(srtt_us);
6209 case offsetof(struct bpf_tcp_sock, snd_ssthresh):
6210 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
6212 case offsetof(struct bpf_tcp_sock, rcv_nxt):
6213 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
6215 case offsetof(struct bpf_tcp_sock, snd_nxt):
6216 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
6218 case offsetof(struct bpf_tcp_sock, snd_una):
6219 BPF_TCP_SOCK_GET_COMMON(snd_una);
6221 case offsetof(struct bpf_tcp_sock, mss_cache):
6222 BPF_TCP_SOCK_GET_COMMON(mss_cache);
6224 case offsetof(struct bpf_tcp_sock, ecn_flags):
6225 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
6227 case offsetof(struct bpf_tcp_sock, rate_delivered):
6228 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
6230 case offsetof(struct bpf_tcp_sock, rate_interval_us):
6231 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
6233 case offsetof(struct bpf_tcp_sock, packets_out):
6234 BPF_TCP_SOCK_GET_COMMON(packets_out);
6236 case offsetof(struct bpf_tcp_sock, retrans_out):
6237 BPF_TCP_SOCK_GET_COMMON(retrans_out);
6239 case offsetof(struct bpf_tcp_sock, total_retrans):
6240 BPF_TCP_SOCK_GET_COMMON(total_retrans);
6242 case offsetof(struct bpf_tcp_sock, segs_in):
6243 BPF_TCP_SOCK_GET_COMMON(segs_in);
6245 case offsetof(struct bpf_tcp_sock, data_segs_in):
6246 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
6248 case offsetof(struct bpf_tcp_sock, segs_out):
6249 BPF_TCP_SOCK_GET_COMMON(segs_out);
6251 case offsetof(struct bpf_tcp_sock, data_segs_out):
6252 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
6254 case offsetof(struct bpf_tcp_sock, lost_out):
6255 BPF_TCP_SOCK_GET_COMMON(lost_out);
6257 case offsetof(struct bpf_tcp_sock, sacked_out):
6258 BPF_TCP_SOCK_GET_COMMON(sacked_out);
6260 case offsetof(struct bpf_tcp_sock, bytes_received):
6261 BPF_TCP_SOCK_GET_COMMON(bytes_received);
6263 case offsetof(struct bpf_tcp_sock, bytes_acked):
6264 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
6266 case offsetof(struct bpf_tcp_sock, dsack_dups):
6267 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
6269 case offsetof(struct bpf_tcp_sock, delivered):
6270 BPF_TCP_SOCK_GET_COMMON(delivered);
6272 case offsetof(struct bpf_tcp_sock, delivered_ce):
6273 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
6275 case offsetof(struct bpf_tcp_sock, icsk_retransmits):
6276 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
6280 return insn - insn_buf;
6283 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
6285 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
6286 return (unsigned long)sk;
6288 return (unsigned long)NULL;
6291 const struct bpf_func_proto bpf_tcp_sock_proto = {
6292 .func = bpf_tcp_sock,
6294 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
6295 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6298 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
6300 sk = sk_to_full_sk(sk);
6302 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
6303 return (unsigned long)sk;
6305 return (unsigned long)NULL;
6308 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
6309 .func = bpf_get_listener_sock,
6311 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6312 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6315 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
6317 unsigned int iphdr_len;
6319 switch (skb_protocol(skb, true)) {
6320 case cpu_to_be16(ETH_P_IP):
6321 iphdr_len = sizeof(struct iphdr);
6323 case cpu_to_be16(ETH_P_IPV6):
6324 iphdr_len = sizeof(struct ipv6hdr);
6330 if (skb_headlen(skb) < iphdr_len)
6333 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
6336 return INET_ECN_set_ce(skb);
6339 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6340 struct bpf_insn_access_aux *info)
6342 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
6345 if (off % size != 0)
6350 return size == sizeof(__u32);
6354 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
6355 const struct bpf_insn *si,
6356 struct bpf_insn *insn_buf,
6357 struct bpf_prog *prog, u32 *target_size)
6359 struct bpf_insn *insn = insn_buf;
6361 #define BPF_XDP_SOCK_GET(FIELD) \
6363 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \
6364 sizeof_field(struct bpf_xdp_sock, FIELD)); \
6365 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
6366 si->dst_reg, si->src_reg, \
6367 offsetof(struct xdp_sock, FIELD)); \
6371 case offsetof(struct bpf_xdp_sock, queue_id):
6372 BPF_XDP_SOCK_GET(queue_id);
6376 return insn - insn_buf;
6379 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
6380 .func = bpf_skb_ecn_set_ce,
6382 .ret_type = RET_INTEGER,
6383 .arg1_type = ARG_PTR_TO_CTX,
6386 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6387 struct tcphdr *, th, u32, th_len)
6389 #ifdef CONFIG_SYN_COOKIES
6393 if (unlikely(!sk || th_len < sizeof(*th)))
6396 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
6397 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6400 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
6403 if (!th->ack || th->rst || th->syn)
6406 if (tcp_synq_no_recent_overflow(sk))
6409 cookie = ntohl(th->ack_seq) - 1;
6411 switch (sk->sk_family) {
6413 if (unlikely(iph_len < sizeof(struct iphdr)))
6416 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
6419 #if IS_BUILTIN(CONFIG_IPV6)
6421 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6424 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
6426 #endif /* CONFIG_IPV6 */
6429 return -EPROTONOSUPPORT;
6441 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
6442 .func = bpf_tcp_check_syncookie,
6445 .ret_type = RET_INTEGER,
6446 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6447 .arg2_type = ARG_PTR_TO_MEM,
6448 .arg3_type = ARG_CONST_SIZE,
6449 .arg4_type = ARG_PTR_TO_MEM,
6450 .arg5_type = ARG_CONST_SIZE,
6453 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6454 struct tcphdr *, th, u32, th_len)
6456 #ifdef CONFIG_SYN_COOKIES
6460 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
6463 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6466 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
6469 if (!th->syn || th->ack || th->fin || th->rst)
6472 if (unlikely(iph_len < sizeof(struct iphdr)))
6475 /* Both struct iphdr and struct ipv6hdr have the version field at the
6476 * same offset so we can cast to the shorter header (struct iphdr).
6478 switch (((struct iphdr *)iph)->version) {
6480 if (sk->sk_family == AF_INET6 && sk->sk_ipv6only)
6483 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
6486 #if IS_BUILTIN(CONFIG_IPV6)
6488 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6491 if (sk->sk_family != AF_INET6)
6494 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
6496 #endif /* CONFIG_IPV6 */
6499 return -EPROTONOSUPPORT;
6504 return cookie | ((u64)mss << 32);
6507 #endif /* CONFIG_SYN_COOKIES */
6510 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
6511 .func = bpf_tcp_gen_syncookie,
6512 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */
6514 .ret_type = RET_INTEGER,
6515 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6516 .arg2_type = ARG_PTR_TO_MEM,
6517 .arg3_type = ARG_CONST_SIZE,
6518 .arg4_type = ARG_PTR_TO_MEM,
6519 .arg5_type = ARG_CONST_SIZE,
6522 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
6524 if (!sk || flags != 0)
6526 if (!skb_at_tc_ingress(skb))
6528 if (unlikely(dev_net(skb->dev) != sock_net(sk)))
6529 return -ENETUNREACH;
6530 if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
6531 return -ESOCKTNOSUPPORT;
6532 if (sk_is_refcounted(sk) &&
6533 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
6538 skb->destructor = sock_pfree;
6543 static const struct bpf_func_proto bpf_sk_assign_proto = {
6544 .func = bpf_sk_assign,
6546 .ret_type = RET_INTEGER,
6547 .arg1_type = ARG_PTR_TO_CTX,
6548 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6549 .arg3_type = ARG_ANYTHING,
6552 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
6553 u8 search_kind, const u8 *magic,
6554 u8 magic_len, bool *eol)
6560 while (op < opend) {
6563 if (kind == TCPOPT_EOL) {
6565 return ERR_PTR(-ENOMSG);
6566 } else if (kind == TCPOPT_NOP) {
6571 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
6572 /* Something is wrong in the received header.
6573 * Follow the TCP stack's tcp_parse_options()
6574 * and just bail here.
6576 return ERR_PTR(-EFAULT);
6579 if (search_kind == kind) {
6583 if (magic_len > kind_len - 2)
6584 return ERR_PTR(-ENOMSG);
6586 if (!memcmp(&op[2], magic, magic_len))
6593 return ERR_PTR(-ENOMSG);
6596 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6597 void *, search_res, u32, len, u64, flags)
6599 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
6600 const u8 *op, *opend, *magic, *search = search_res;
6601 u8 search_kind, search_len, copy_len, magic_len;
6604 /* 2 byte is the minimal option len except TCPOPT_NOP and
6605 * TCPOPT_EOL which are useless for the bpf prog to learn
6606 * and this helper disallow loading them also.
6608 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
6611 search_kind = search[0];
6612 search_len = search[1];
6614 if (search_len > len || search_kind == TCPOPT_NOP ||
6615 search_kind == TCPOPT_EOL)
6618 if (search_kind == TCPOPT_EXP || search_kind == 253) {
6619 /* 16 or 32 bit magic. +2 for kind and kind length */
6620 if (search_len != 4 && search_len != 6)
6623 magic_len = search_len - 2;
6632 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
6637 op += sizeof(struct tcphdr);
6639 if (!bpf_sock->skb ||
6640 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6641 /* This bpf_sock->op cannot call this helper */
6644 opend = bpf_sock->skb_data_end;
6645 op = bpf_sock->skb->data + sizeof(struct tcphdr);
6648 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
6655 if (copy_len > len) {
6660 memcpy(search_res, op, copy_len);
6664 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
6665 .func = bpf_sock_ops_load_hdr_opt,
6667 .ret_type = RET_INTEGER,
6668 .arg1_type = ARG_PTR_TO_CTX,
6669 .arg2_type = ARG_PTR_TO_MEM,
6670 .arg3_type = ARG_CONST_SIZE,
6671 .arg4_type = ARG_ANYTHING,
6674 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6675 const void *, from, u32, len, u64, flags)
6677 u8 new_kind, new_kind_len, magic_len = 0, *opend;
6678 const u8 *op, *new_op, *magic = NULL;
6679 struct sk_buff *skb;
6682 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
6685 if (len < 2 || flags)
6689 new_kind = new_op[0];
6690 new_kind_len = new_op[1];
6692 if (new_kind_len > len || new_kind == TCPOPT_NOP ||
6693 new_kind == TCPOPT_EOL)
6696 if (new_kind_len > bpf_sock->remaining_opt_len)
6699 /* 253 is another experimental kind */
6700 if (new_kind == TCPOPT_EXP || new_kind == 253) {
6701 if (new_kind_len < 4)
6703 /* Match for the 2 byte magic also.
6704 * RFC 6994: the magic could be 2 or 4 bytes.
6705 * Hence, matching by 2 byte only is on the
6706 * conservative side but it is the right
6707 * thing to do for the 'search-for-duplication'
6714 /* Check for duplication */
6715 skb = bpf_sock->skb;
6716 op = skb->data + sizeof(struct tcphdr);
6717 opend = bpf_sock->skb_data_end;
6719 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
6724 if (PTR_ERR(op) != -ENOMSG)
6728 /* The option has been ended. Treat it as no more
6729 * header option can be written.
6733 /* No duplication found. Store the header option. */
6734 memcpy(opend, from, new_kind_len);
6736 bpf_sock->remaining_opt_len -= new_kind_len;
6737 bpf_sock->skb_data_end += new_kind_len;
6742 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
6743 .func = bpf_sock_ops_store_hdr_opt,
6745 .ret_type = RET_INTEGER,
6746 .arg1_type = ARG_PTR_TO_CTX,
6747 .arg2_type = ARG_PTR_TO_MEM,
6748 .arg3_type = ARG_CONST_SIZE,
6749 .arg4_type = ARG_ANYTHING,
6752 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6753 u32, len, u64, flags)
6755 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6758 if (flags || len < 2)
6761 if (len > bpf_sock->remaining_opt_len)
6764 bpf_sock->remaining_opt_len -= len;
6769 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
6770 .func = bpf_sock_ops_reserve_hdr_opt,
6772 .ret_type = RET_INTEGER,
6773 .arg1_type = ARG_PTR_TO_CTX,
6774 .arg2_type = ARG_ANYTHING,
6775 .arg3_type = ARG_ANYTHING,
6778 #endif /* CONFIG_INET */
6780 bool bpf_helper_changes_pkt_data(void *func)
6782 if (func == bpf_skb_vlan_push ||
6783 func == bpf_skb_vlan_pop ||
6784 func == bpf_skb_store_bytes ||
6785 func == bpf_skb_change_proto ||
6786 func == bpf_skb_change_head ||
6787 func == sk_skb_change_head ||
6788 func == bpf_skb_change_tail ||
6789 func == sk_skb_change_tail ||
6790 func == bpf_skb_adjust_room ||
6791 func == sk_skb_adjust_room ||
6792 func == bpf_skb_pull_data ||
6793 func == sk_skb_pull_data ||
6794 func == bpf_clone_redirect ||
6795 func == bpf_l3_csum_replace ||
6796 func == bpf_l4_csum_replace ||
6797 func == bpf_xdp_adjust_head ||
6798 func == bpf_xdp_adjust_meta ||
6799 func == bpf_msg_pull_data ||
6800 func == bpf_msg_push_data ||
6801 func == bpf_msg_pop_data ||
6802 func == bpf_xdp_adjust_tail ||
6803 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6804 func == bpf_lwt_seg6_store_bytes ||
6805 func == bpf_lwt_seg6_adjust_srh ||
6806 func == bpf_lwt_seg6_action ||
6809 func == bpf_sock_ops_store_hdr_opt ||
6811 func == bpf_lwt_in_push_encap ||
6812 func == bpf_lwt_xmit_push_encap)
6818 const struct bpf_func_proto bpf_event_output_data_proto __weak;
6819 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
6821 static const struct bpf_func_proto *
6822 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6825 /* inet and inet6 sockets are created in a process
6826 * context so there is always a valid uid/gid
6828 case BPF_FUNC_get_current_uid_gid:
6829 return &bpf_get_current_uid_gid_proto;
6830 case BPF_FUNC_get_local_storage:
6831 return &bpf_get_local_storage_proto;
6832 case BPF_FUNC_get_socket_cookie:
6833 return &bpf_get_socket_cookie_sock_proto;
6834 case BPF_FUNC_get_netns_cookie:
6835 return &bpf_get_netns_cookie_sock_proto;
6836 case BPF_FUNC_perf_event_output:
6837 return &bpf_event_output_data_proto;
6838 case BPF_FUNC_get_current_pid_tgid:
6839 return &bpf_get_current_pid_tgid_proto;
6840 case BPF_FUNC_get_current_comm:
6841 return &bpf_get_current_comm_proto;
6842 #ifdef CONFIG_CGROUPS
6843 case BPF_FUNC_get_current_cgroup_id:
6844 return &bpf_get_current_cgroup_id_proto;
6845 case BPF_FUNC_get_current_ancestor_cgroup_id:
6846 return &bpf_get_current_ancestor_cgroup_id_proto;
6848 #ifdef CONFIG_CGROUP_NET_CLASSID
6849 case BPF_FUNC_get_cgroup_classid:
6850 return &bpf_get_cgroup_classid_curr_proto;
6852 case BPF_FUNC_sk_storage_get:
6853 return &bpf_sk_storage_get_cg_sock_proto;
6855 return bpf_base_func_proto(func_id);
6859 static const struct bpf_func_proto *
6860 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6863 /* inet and inet6 sockets are created in a process
6864 * context so there is always a valid uid/gid
6866 case BPF_FUNC_get_current_uid_gid:
6867 return &bpf_get_current_uid_gid_proto;
6869 switch (prog->expected_attach_type) {
6870 case BPF_CGROUP_INET4_CONNECT:
6871 case BPF_CGROUP_INET6_CONNECT:
6872 return &bpf_bind_proto;
6876 case BPF_FUNC_get_socket_cookie:
6877 return &bpf_get_socket_cookie_sock_addr_proto;
6878 case BPF_FUNC_get_netns_cookie:
6879 return &bpf_get_netns_cookie_sock_addr_proto;
6880 case BPF_FUNC_get_local_storage:
6881 return &bpf_get_local_storage_proto;
6882 case BPF_FUNC_perf_event_output:
6883 return &bpf_event_output_data_proto;
6884 case BPF_FUNC_get_current_pid_tgid:
6885 return &bpf_get_current_pid_tgid_proto;
6886 case BPF_FUNC_get_current_comm:
6887 return &bpf_get_current_comm_proto;
6888 #ifdef CONFIG_CGROUPS
6889 case BPF_FUNC_get_current_cgroup_id:
6890 return &bpf_get_current_cgroup_id_proto;
6891 case BPF_FUNC_get_current_ancestor_cgroup_id:
6892 return &bpf_get_current_ancestor_cgroup_id_proto;
6894 #ifdef CONFIG_CGROUP_NET_CLASSID
6895 case BPF_FUNC_get_cgroup_classid:
6896 return &bpf_get_cgroup_classid_curr_proto;
6899 case BPF_FUNC_sk_lookup_tcp:
6900 return &bpf_sock_addr_sk_lookup_tcp_proto;
6901 case BPF_FUNC_sk_lookup_udp:
6902 return &bpf_sock_addr_sk_lookup_udp_proto;
6903 case BPF_FUNC_sk_release:
6904 return &bpf_sk_release_proto;
6905 case BPF_FUNC_skc_lookup_tcp:
6906 return &bpf_sock_addr_skc_lookup_tcp_proto;
6907 #endif /* CONFIG_INET */
6908 case BPF_FUNC_sk_storage_get:
6909 return &bpf_sk_storage_get_proto;
6910 case BPF_FUNC_sk_storage_delete:
6911 return &bpf_sk_storage_delete_proto;
6912 case BPF_FUNC_setsockopt:
6913 switch (prog->expected_attach_type) {
6914 case BPF_CGROUP_INET4_CONNECT:
6915 case BPF_CGROUP_INET6_CONNECT:
6916 return &bpf_sock_addr_setsockopt_proto;
6920 case BPF_FUNC_getsockopt:
6921 switch (prog->expected_attach_type) {
6922 case BPF_CGROUP_INET4_CONNECT:
6923 case BPF_CGROUP_INET6_CONNECT:
6924 return &bpf_sock_addr_getsockopt_proto;
6929 return bpf_sk_base_func_proto(func_id);
6933 static const struct bpf_func_proto *
6934 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6937 case BPF_FUNC_skb_load_bytes:
6938 return &bpf_skb_load_bytes_proto;
6939 case BPF_FUNC_skb_load_bytes_relative:
6940 return &bpf_skb_load_bytes_relative_proto;
6941 case BPF_FUNC_get_socket_cookie:
6942 return &bpf_get_socket_cookie_proto;
6943 case BPF_FUNC_get_socket_uid:
6944 return &bpf_get_socket_uid_proto;
6945 case BPF_FUNC_perf_event_output:
6946 return &bpf_skb_event_output_proto;
6948 return bpf_sk_base_func_proto(func_id);
6952 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
6953 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
6955 static const struct bpf_func_proto *
6956 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6959 case BPF_FUNC_get_local_storage:
6960 return &bpf_get_local_storage_proto;
6961 case BPF_FUNC_sk_fullsock:
6962 return &bpf_sk_fullsock_proto;
6963 case BPF_FUNC_sk_storage_get:
6964 return &bpf_sk_storage_get_proto;
6965 case BPF_FUNC_sk_storage_delete:
6966 return &bpf_sk_storage_delete_proto;
6967 case BPF_FUNC_perf_event_output:
6968 return &bpf_skb_event_output_proto;
6969 #ifdef CONFIG_SOCK_CGROUP_DATA
6970 case BPF_FUNC_skb_cgroup_id:
6971 return &bpf_skb_cgroup_id_proto;
6972 case BPF_FUNC_skb_ancestor_cgroup_id:
6973 return &bpf_skb_ancestor_cgroup_id_proto;
6974 case BPF_FUNC_sk_cgroup_id:
6975 return &bpf_sk_cgroup_id_proto;
6976 case BPF_FUNC_sk_ancestor_cgroup_id:
6977 return &bpf_sk_ancestor_cgroup_id_proto;
6980 case BPF_FUNC_sk_lookup_tcp:
6981 return &bpf_sk_lookup_tcp_proto;
6982 case BPF_FUNC_sk_lookup_udp:
6983 return &bpf_sk_lookup_udp_proto;
6984 case BPF_FUNC_sk_release:
6985 return &bpf_sk_release_proto;
6986 case BPF_FUNC_skc_lookup_tcp:
6987 return &bpf_skc_lookup_tcp_proto;
6988 case BPF_FUNC_tcp_sock:
6989 return &bpf_tcp_sock_proto;
6990 case BPF_FUNC_get_listener_sock:
6991 return &bpf_get_listener_sock_proto;
6992 case BPF_FUNC_skb_ecn_set_ce:
6993 return &bpf_skb_ecn_set_ce_proto;
6996 return sk_filter_func_proto(func_id, prog);
7000 static const struct bpf_func_proto *
7001 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7004 case BPF_FUNC_skb_store_bytes:
7005 return &bpf_skb_store_bytes_proto;
7006 case BPF_FUNC_skb_load_bytes:
7007 return &bpf_skb_load_bytes_proto;
7008 case BPF_FUNC_skb_load_bytes_relative:
7009 return &bpf_skb_load_bytes_relative_proto;
7010 case BPF_FUNC_skb_pull_data:
7011 return &bpf_skb_pull_data_proto;
7012 case BPF_FUNC_csum_diff:
7013 return &bpf_csum_diff_proto;
7014 case BPF_FUNC_csum_update:
7015 return &bpf_csum_update_proto;
7016 case BPF_FUNC_csum_level:
7017 return &bpf_csum_level_proto;
7018 case BPF_FUNC_l3_csum_replace:
7019 return &bpf_l3_csum_replace_proto;
7020 case BPF_FUNC_l4_csum_replace:
7021 return &bpf_l4_csum_replace_proto;
7022 case BPF_FUNC_clone_redirect:
7023 return &bpf_clone_redirect_proto;
7024 case BPF_FUNC_get_cgroup_classid:
7025 return &bpf_get_cgroup_classid_proto;
7026 case BPF_FUNC_skb_vlan_push:
7027 return &bpf_skb_vlan_push_proto;
7028 case BPF_FUNC_skb_vlan_pop:
7029 return &bpf_skb_vlan_pop_proto;
7030 case BPF_FUNC_skb_change_proto:
7031 return &bpf_skb_change_proto_proto;
7032 case BPF_FUNC_skb_change_type:
7033 return &bpf_skb_change_type_proto;
7034 case BPF_FUNC_skb_adjust_room:
7035 return &bpf_skb_adjust_room_proto;
7036 case BPF_FUNC_skb_change_tail:
7037 return &bpf_skb_change_tail_proto;
7038 case BPF_FUNC_skb_change_head:
7039 return &bpf_skb_change_head_proto;
7040 case BPF_FUNC_skb_get_tunnel_key:
7041 return &bpf_skb_get_tunnel_key_proto;
7042 case BPF_FUNC_skb_set_tunnel_key:
7043 return bpf_get_skb_set_tunnel_proto(func_id);
7044 case BPF_FUNC_skb_get_tunnel_opt:
7045 return &bpf_skb_get_tunnel_opt_proto;
7046 case BPF_FUNC_skb_set_tunnel_opt:
7047 return bpf_get_skb_set_tunnel_proto(func_id);
7048 case BPF_FUNC_redirect:
7049 return &bpf_redirect_proto;
7050 case BPF_FUNC_redirect_neigh:
7051 return &bpf_redirect_neigh_proto;
7052 case BPF_FUNC_get_route_realm:
7053 return &bpf_get_route_realm_proto;
7054 case BPF_FUNC_get_hash_recalc:
7055 return &bpf_get_hash_recalc_proto;
7056 case BPF_FUNC_set_hash_invalid:
7057 return &bpf_set_hash_invalid_proto;
7058 case BPF_FUNC_set_hash:
7059 return &bpf_set_hash_proto;
7060 case BPF_FUNC_perf_event_output:
7061 return &bpf_skb_event_output_proto;
7062 case BPF_FUNC_get_smp_processor_id:
7063 return &bpf_get_smp_processor_id_proto;
7064 case BPF_FUNC_skb_under_cgroup:
7065 return &bpf_skb_under_cgroup_proto;
7066 case BPF_FUNC_get_socket_cookie:
7067 return &bpf_get_socket_cookie_proto;
7068 case BPF_FUNC_get_socket_uid:
7069 return &bpf_get_socket_uid_proto;
7070 case BPF_FUNC_fib_lookup:
7071 return &bpf_skb_fib_lookup_proto;
7072 case BPF_FUNC_sk_fullsock:
7073 return &bpf_sk_fullsock_proto;
7074 case BPF_FUNC_sk_storage_get:
7075 return &bpf_sk_storage_get_proto;
7076 case BPF_FUNC_sk_storage_delete:
7077 return &bpf_sk_storage_delete_proto;
7079 case BPF_FUNC_skb_get_xfrm_state:
7080 return &bpf_skb_get_xfrm_state_proto;
7082 #ifdef CONFIG_CGROUP_NET_CLASSID
7083 case BPF_FUNC_skb_cgroup_classid:
7084 return &bpf_skb_cgroup_classid_proto;
7086 #ifdef CONFIG_SOCK_CGROUP_DATA
7087 case BPF_FUNC_skb_cgroup_id:
7088 return &bpf_skb_cgroup_id_proto;
7089 case BPF_FUNC_skb_ancestor_cgroup_id:
7090 return &bpf_skb_ancestor_cgroup_id_proto;
7093 case BPF_FUNC_sk_lookup_tcp:
7094 return &bpf_sk_lookup_tcp_proto;
7095 case BPF_FUNC_sk_lookup_udp:
7096 return &bpf_sk_lookup_udp_proto;
7097 case BPF_FUNC_sk_release:
7098 return &bpf_sk_release_proto;
7099 case BPF_FUNC_tcp_sock:
7100 return &bpf_tcp_sock_proto;
7101 case BPF_FUNC_get_listener_sock:
7102 return &bpf_get_listener_sock_proto;
7103 case BPF_FUNC_skc_lookup_tcp:
7104 return &bpf_skc_lookup_tcp_proto;
7105 case BPF_FUNC_tcp_check_syncookie:
7106 return &bpf_tcp_check_syncookie_proto;
7107 case BPF_FUNC_skb_ecn_set_ce:
7108 return &bpf_skb_ecn_set_ce_proto;
7109 case BPF_FUNC_tcp_gen_syncookie:
7110 return &bpf_tcp_gen_syncookie_proto;
7111 case BPF_FUNC_sk_assign:
7112 return &bpf_sk_assign_proto;
7115 return bpf_sk_base_func_proto(func_id);
7119 static const struct bpf_func_proto *
7120 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7123 case BPF_FUNC_perf_event_output:
7124 return &bpf_xdp_event_output_proto;
7125 case BPF_FUNC_get_smp_processor_id:
7126 return &bpf_get_smp_processor_id_proto;
7127 case BPF_FUNC_csum_diff:
7128 return &bpf_csum_diff_proto;
7129 case BPF_FUNC_xdp_adjust_head:
7130 return &bpf_xdp_adjust_head_proto;
7131 case BPF_FUNC_xdp_adjust_meta:
7132 return &bpf_xdp_adjust_meta_proto;
7133 case BPF_FUNC_redirect:
7134 return &bpf_xdp_redirect_proto;
7135 case BPF_FUNC_redirect_map:
7136 return &bpf_xdp_redirect_map_proto;
7137 case BPF_FUNC_xdp_adjust_tail:
7138 return &bpf_xdp_adjust_tail_proto;
7139 case BPF_FUNC_fib_lookup:
7140 return &bpf_xdp_fib_lookup_proto;
7142 case BPF_FUNC_sk_lookup_udp:
7143 return &bpf_xdp_sk_lookup_udp_proto;
7144 case BPF_FUNC_sk_lookup_tcp:
7145 return &bpf_xdp_sk_lookup_tcp_proto;
7146 case BPF_FUNC_sk_release:
7147 return &bpf_sk_release_proto;
7148 case BPF_FUNC_skc_lookup_tcp:
7149 return &bpf_xdp_skc_lookup_tcp_proto;
7150 case BPF_FUNC_tcp_check_syncookie:
7151 return &bpf_tcp_check_syncookie_proto;
7152 case BPF_FUNC_tcp_gen_syncookie:
7153 return &bpf_tcp_gen_syncookie_proto;
7156 return bpf_sk_base_func_proto(func_id);
7160 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
7161 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
7163 static const struct bpf_func_proto *
7164 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7167 case BPF_FUNC_setsockopt:
7168 return &bpf_sock_ops_setsockopt_proto;
7169 case BPF_FUNC_getsockopt:
7170 return &bpf_sock_ops_getsockopt_proto;
7171 case BPF_FUNC_sock_ops_cb_flags_set:
7172 return &bpf_sock_ops_cb_flags_set_proto;
7173 case BPF_FUNC_sock_map_update:
7174 return &bpf_sock_map_update_proto;
7175 case BPF_FUNC_sock_hash_update:
7176 return &bpf_sock_hash_update_proto;
7177 case BPF_FUNC_get_socket_cookie:
7178 return &bpf_get_socket_cookie_sock_ops_proto;
7179 case BPF_FUNC_get_local_storage:
7180 return &bpf_get_local_storage_proto;
7181 case BPF_FUNC_perf_event_output:
7182 return &bpf_event_output_data_proto;
7183 case BPF_FUNC_sk_storage_get:
7184 return &bpf_sk_storage_get_proto;
7185 case BPF_FUNC_sk_storage_delete:
7186 return &bpf_sk_storage_delete_proto;
7188 case BPF_FUNC_load_hdr_opt:
7189 return &bpf_sock_ops_load_hdr_opt_proto;
7190 case BPF_FUNC_store_hdr_opt:
7191 return &bpf_sock_ops_store_hdr_opt_proto;
7192 case BPF_FUNC_reserve_hdr_opt:
7193 return &bpf_sock_ops_reserve_hdr_opt_proto;
7194 case BPF_FUNC_tcp_sock:
7195 return &bpf_tcp_sock_proto;
7196 #endif /* CONFIG_INET */
7198 return bpf_sk_base_func_proto(func_id);
7202 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
7203 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
7205 static const struct bpf_func_proto *
7206 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7209 case BPF_FUNC_msg_redirect_map:
7210 return &bpf_msg_redirect_map_proto;
7211 case BPF_FUNC_msg_redirect_hash:
7212 return &bpf_msg_redirect_hash_proto;
7213 case BPF_FUNC_msg_apply_bytes:
7214 return &bpf_msg_apply_bytes_proto;
7215 case BPF_FUNC_msg_cork_bytes:
7216 return &bpf_msg_cork_bytes_proto;
7217 case BPF_FUNC_msg_pull_data:
7218 return &bpf_msg_pull_data_proto;
7219 case BPF_FUNC_msg_push_data:
7220 return &bpf_msg_push_data_proto;
7221 case BPF_FUNC_msg_pop_data:
7222 return &bpf_msg_pop_data_proto;
7223 case BPF_FUNC_perf_event_output:
7224 return &bpf_event_output_data_proto;
7225 case BPF_FUNC_get_current_uid_gid:
7226 return &bpf_get_current_uid_gid_proto;
7227 case BPF_FUNC_get_current_pid_tgid:
7228 return &bpf_get_current_pid_tgid_proto;
7229 case BPF_FUNC_sk_storage_get:
7230 return &bpf_sk_storage_get_proto;
7231 case BPF_FUNC_sk_storage_delete:
7232 return &bpf_sk_storage_delete_proto;
7233 #ifdef CONFIG_CGROUPS
7234 case BPF_FUNC_get_current_cgroup_id:
7235 return &bpf_get_current_cgroup_id_proto;
7236 case BPF_FUNC_get_current_ancestor_cgroup_id:
7237 return &bpf_get_current_ancestor_cgroup_id_proto;
7239 #ifdef CONFIG_CGROUP_NET_CLASSID
7240 case BPF_FUNC_get_cgroup_classid:
7241 return &bpf_get_cgroup_classid_curr_proto;
7244 return bpf_sk_base_func_proto(func_id);
7248 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
7249 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
7251 static const struct bpf_func_proto *
7252 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7255 case BPF_FUNC_skb_store_bytes:
7256 return &bpf_skb_store_bytes_proto;
7257 case BPF_FUNC_skb_load_bytes:
7258 return &bpf_skb_load_bytes_proto;
7259 case BPF_FUNC_skb_pull_data:
7260 return &sk_skb_pull_data_proto;
7261 case BPF_FUNC_skb_change_tail:
7262 return &sk_skb_change_tail_proto;
7263 case BPF_FUNC_skb_change_head:
7264 return &sk_skb_change_head_proto;
7265 case BPF_FUNC_skb_adjust_room:
7266 return &sk_skb_adjust_room_proto;
7267 case BPF_FUNC_get_socket_cookie:
7268 return &bpf_get_socket_cookie_proto;
7269 case BPF_FUNC_get_socket_uid:
7270 return &bpf_get_socket_uid_proto;
7271 case BPF_FUNC_sk_redirect_map:
7272 return &bpf_sk_redirect_map_proto;
7273 case BPF_FUNC_sk_redirect_hash:
7274 return &bpf_sk_redirect_hash_proto;
7275 case BPF_FUNC_perf_event_output:
7276 return &bpf_skb_event_output_proto;
7278 case BPF_FUNC_sk_lookup_tcp:
7279 return &bpf_sk_lookup_tcp_proto;
7280 case BPF_FUNC_sk_lookup_udp:
7281 return &bpf_sk_lookup_udp_proto;
7282 case BPF_FUNC_sk_release:
7283 return &bpf_sk_release_proto;
7284 case BPF_FUNC_skc_lookup_tcp:
7285 return &bpf_skc_lookup_tcp_proto;
7288 return bpf_sk_base_func_proto(func_id);
7292 static const struct bpf_func_proto *
7293 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7296 case BPF_FUNC_skb_load_bytes:
7297 return &bpf_flow_dissector_load_bytes_proto;
7299 return bpf_sk_base_func_proto(func_id);
7303 static const struct bpf_func_proto *
7304 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7307 case BPF_FUNC_skb_load_bytes:
7308 return &bpf_skb_load_bytes_proto;
7309 case BPF_FUNC_skb_pull_data:
7310 return &bpf_skb_pull_data_proto;
7311 case BPF_FUNC_csum_diff:
7312 return &bpf_csum_diff_proto;
7313 case BPF_FUNC_get_cgroup_classid:
7314 return &bpf_get_cgroup_classid_proto;
7315 case BPF_FUNC_get_route_realm:
7316 return &bpf_get_route_realm_proto;
7317 case BPF_FUNC_get_hash_recalc:
7318 return &bpf_get_hash_recalc_proto;
7319 case BPF_FUNC_perf_event_output:
7320 return &bpf_skb_event_output_proto;
7321 case BPF_FUNC_get_smp_processor_id:
7322 return &bpf_get_smp_processor_id_proto;
7323 case BPF_FUNC_skb_under_cgroup:
7324 return &bpf_skb_under_cgroup_proto;
7326 return bpf_sk_base_func_proto(func_id);
7330 static const struct bpf_func_proto *
7331 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7334 case BPF_FUNC_lwt_push_encap:
7335 return &bpf_lwt_in_push_encap_proto;
7337 return lwt_out_func_proto(func_id, prog);
7341 static const struct bpf_func_proto *
7342 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7345 case BPF_FUNC_skb_get_tunnel_key:
7346 return &bpf_skb_get_tunnel_key_proto;
7347 case BPF_FUNC_skb_set_tunnel_key:
7348 return bpf_get_skb_set_tunnel_proto(func_id);
7349 case BPF_FUNC_skb_get_tunnel_opt:
7350 return &bpf_skb_get_tunnel_opt_proto;
7351 case BPF_FUNC_skb_set_tunnel_opt:
7352 return bpf_get_skb_set_tunnel_proto(func_id);
7353 case BPF_FUNC_redirect:
7354 return &bpf_redirect_proto;
7355 case BPF_FUNC_clone_redirect:
7356 return &bpf_clone_redirect_proto;
7357 case BPF_FUNC_skb_change_tail:
7358 return &bpf_skb_change_tail_proto;
7359 case BPF_FUNC_skb_change_head:
7360 return &bpf_skb_change_head_proto;
7361 case BPF_FUNC_skb_store_bytes:
7362 return &bpf_skb_store_bytes_proto;
7363 case BPF_FUNC_csum_update:
7364 return &bpf_csum_update_proto;
7365 case BPF_FUNC_csum_level:
7366 return &bpf_csum_level_proto;
7367 case BPF_FUNC_l3_csum_replace:
7368 return &bpf_l3_csum_replace_proto;
7369 case BPF_FUNC_l4_csum_replace:
7370 return &bpf_l4_csum_replace_proto;
7371 case BPF_FUNC_set_hash_invalid:
7372 return &bpf_set_hash_invalid_proto;
7373 case BPF_FUNC_lwt_push_encap:
7374 return &bpf_lwt_xmit_push_encap_proto;
7376 return lwt_out_func_proto(func_id, prog);
7380 static const struct bpf_func_proto *
7381 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7384 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7385 case BPF_FUNC_lwt_seg6_store_bytes:
7386 return &bpf_lwt_seg6_store_bytes_proto;
7387 case BPF_FUNC_lwt_seg6_action:
7388 return &bpf_lwt_seg6_action_proto;
7389 case BPF_FUNC_lwt_seg6_adjust_srh:
7390 return &bpf_lwt_seg6_adjust_srh_proto;
7393 return lwt_out_func_proto(func_id, prog);
7397 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
7398 const struct bpf_prog *prog,
7399 struct bpf_insn_access_aux *info)
7401 const int size_default = sizeof(__u32);
7403 if (off < 0 || off >= sizeof(struct __sk_buff))
7406 /* The verifier guarantees that size > 0. */
7407 if (off % size != 0)
7411 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7412 if (off + size > offsetofend(struct __sk_buff, cb[4]))
7415 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
7416 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
7417 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
7418 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
7419 case bpf_ctx_range(struct __sk_buff, data):
7420 case bpf_ctx_range(struct __sk_buff, data_meta):
7421 case bpf_ctx_range(struct __sk_buff, data_end):
7422 if (size != size_default)
7425 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
7427 case bpf_ctx_range(struct __sk_buff, tstamp):
7428 if (size != sizeof(__u64))
7431 case offsetof(struct __sk_buff, sk):
7432 if (type == BPF_WRITE || size != sizeof(__u64))
7434 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
7437 /* Only narrow read access allowed for now. */
7438 if (type == BPF_WRITE) {
7439 if (size != size_default)
7442 bpf_ctx_record_field_size(info, size_default);
7443 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
7451 static bool sk_filter_is_valid_access(int off, int size,
7452 enum bpf_access_type type,
7453 const struct bpf_prog *prog,
7454 struct bpf_insn_access_aux *info)
7457 case bpf_ctx_range(struct __sk_buff, tc_classid):
7458 case bpf_ctx_range(struct __sk_buff, data):
7459 case bpf_ctx_range(struct __sk_buff, data_meta):
7460 case bpf_ctx_range(struct __sk_buff, data_end):
7461 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7462 case bpf_ctx_range(struct __sk_buff, tstamp):
7463 case bpf_ctx_range(struct __sk_buff, wire_len):
7467 if (type == BPF_WRITE) {
7469 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7476 return bpf_skb_is_valid_access(off, size, type, prog, info);
7479 static bool cg_skb_is_valid_access(int off, int size,
7480 enum bpf_access_type type,
7481 const struct bpf_prog *prog,
7482 struct bpf_insn_access_aux *info)
7485 case bpf_ctx_range(struct __sk_buff, tc_classid):
7486 case bpf_ctx_range(struct __sk_buff, data_meta):
7487 case bpf_ctx_range(struct __sk_buff, wire_len):
7489 case bpf_ctx_range(struct __sk_buff, data):
7490 case bpf_ctx_range(struct __sk_buff, data_end):
7496 if (type == BPF_WRITE) {
7498 case bpf_ctx_range(struct __sk_buff, mark):
7499 case bpf_ctx_range(struct __sk_buff, priority):
7500 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7502 case bpf_ctx_range(struct __sk_buff, tstamp):
7512 case bpf_ctx_range(struct __sk_buff, data):
7513 info->reg_type = PTR_TO_PACKET;
7515 case bpf_ctx_range(struct __sk_buff, data_end):
7516 info->reg_type = PTR_TO_PACKET_END;
7520 return bpf_skb_is_valid_access(off, size, type, prog, info);
7523 static bool lwt_is_valid_access(int off, int size,
7524 enum bpf_access_type type,
7525 const struct bpf_prog *prog,
7526 struct bpf_insn_access_aux *info)
7529 case bpf_ctx_range(struct __sk_buff, tc_classid):
7530 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7531 case bpf_ctx_range(struct __sk_buff, data_meta):
7532 case bpf_ctx_range(struct __sk_buff, tstamp):
7533 case bpf_ctx_range(struct __sk_buff, wire_len):
7537 if (type == BPF_WRITE) {
7539 case bpf_ctx_range(struct __sk_buff, mark):
7540 case bpf_ctx_range(struct __sk_buff, priority):
7541 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7549 case bpf_ctx_range(struct __sk_buff, data):
7550 info->reg_type = PTR_TO_PACKET;
7552 case bpf_ctx_range(struct __sk_buff, data_end):
7553 info->reg_type = PTR_TO_PACKET_END;
7557 return bpf_skb_is_valid_access(off, size, type, prog, info);
7560 /* Attach type specific accesses */
7561 static bool __sock_filter_check_attach_type(int off,
7562 enum bpf_access_type access_type,
7563 enum bpf_attach_type attach_type)
7566 case offsetof(struct bpf_sock, bound_dev_if):
7567 case offsetof(struct bpf_sock, mark):
7568 case offsetof(struct bpf_sock, priority):
7569 switch (attach_type) {
7570 case BPF_CGROUP_INET_SOCK_CREATE:
7571 case BPF_CGROUP_INET_SOCK_RELEASE:
7576 case bpf_ctx_range(struct bpf_sock, src_ip4):
7577 switch (attach_type) {
7578 case BPF_CGROUP_INET4_POST_BIND:
7583 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7584 switch (attach_type) {
7585 case BPF_CGROUP_INET6_POST_BIND:
7590 case bpf_ctx_range(struct bpf_sock, src_port):
7591 switch (attach_type) {
7592 case BPF_CGROUP_INET4_POST_BIND:
7593 case BPF_CGROUP_INET6_POST_BIND:
7600 return access_type == BPF_READ;
7605 bool bpf_sock_common_is_valid_access(int off, int size,
7606 enum bpf_access_type type,
7607 struct bpf_insn_access_aux *info)
7610 case bpf_ctx_range_till(struct bpf_sock, type, priority):
7613 return bpf_sock_is_valid_access(off, size, type, info);
7617 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7618 struct bpf_insn_access_aux *info)
7620 const int size_default = sizeof(__u32);
7622 if (off < 0 || off >= sizeof(struct bpf_sock))
7624 if (off % size != 0)
7628 case offsetof(struct bpf_sock, state):
7629 case offsetof(struct bpf_sock, family):
7630 case offsetof(struct bpf_sock, type):
7631 case offsetof(struct bpf_sock, protocol):
7632 case offsetof(struct bpf_sock, dst_port):
7633 case offsetof(struct bpf_sock, src_port):
7634 case offsetof(struct bpf_sock, rx_queue_mapping):
7635 case bpf_ctx_range(struct bpf_sock, src_ip4):
7636 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7637 case bpf_ctx_range(struct bpf_sock, dst_ip4):
7638 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
7639 bpf_ctx_record_field_size(info, size_default);
7640 return bpf_ctx_narrow_access_ok(off, size, size_default);
7643 return size == size_default;
7646 static bool sock_filter_is_valid_access(int off, int size,
7647 enum bpf_access_type type,
7648 const struct bpf_prog *prog,
7649 struct bpf_insn_access_aux *info)
7651 if (!bpf_sock_is_valid_access(off, size, type, info))
7653 return __sock_filter_check_attach_type(off, type,
7654 prog->expected_attach_type);
7657 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
7658 const struct bpf_prog *prog)
7660 /* Neither direct read nor direct write requires any preliminary
7666 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
7667 const struct bpf_prog *prog, int drop_verdict)
7669 struct bpf_insn *insn = insn_buf;
7674 /* if (!skb->cloned)
7677 * (Fast-path, otherwise approximation that we might be
7678 * a clone, do the rest in helper.)
7680 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
7681 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
7682 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
7684 /* ret = bpf_skb_pull_data(skb, 0); */
7685 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
7686 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
7687 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
7688 BPF_FUNC_skb_pull_data);
7691 * return TC_ACT_SHOT;
7693 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
7694 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
7695 *insn++ = BPF_EXIT_INSN();
7698 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
7700 *insn++ = prog->insnsi[0];
7702 return insn - insn_buf;
7705 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
7706 struct bpf_insn *insn_buf)
7708 bool indirect = BPF_MODE(orig->code) == BPF_IND;
7709 struct bpf_insn *insn = insn_buf;
7712 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
7714 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
7716 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
7718 /* We're guaranteed here that CTX is in R6. */
7719 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
7721 switch (BPF_SIZE(orig->code)) {
7723 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
7726 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
7729 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
7733 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
7734 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
7735 *insn++ = BPF_EXIT_INSN();
7737 return insn - insn_buf;
7740 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
7741 const struct bpf_prog *prog)
7743 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
7746 static bool tc_cls_act_is_valid_access(int off, int size,
7747 enum bpf_access_type type,
7748 const struct bpf_prog *prog,
7749 struct bpf_insn_access_aux *info)
7751 if (type == BPF_WRITE) {
7753 case bpf_ctx_range(struct __sk_buff, mark):
7754 case bpf_ctx_range(struct __sk_buff, tc_index):
7755 case bpf_ctx_range(struct __sk_buff, priority):
7756 case bpf_ctx_range(struct __sk_buff, tc_classid):
7757 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7758 case bpf_ctx_range(struct __sk_buff, tstamp):
7759 case bpf_ctx_range(struct __sk_buff, queue_mapping):
7767 case bpf_ctx_range(struct __sk_buff, data):
7768 info->reg_type = PTR_TO_PACKET;
7770 case bpf_ctx_range(struct __sk_buff, data_meta):
7771 info->reg_type = PTR_TO_PACKET_META;
7773 case bpf_ctx_range(struct __sk_buff, data_end):
7774 info->reg_type = PTR_TO_PACKET_END;
7776 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7780 return bpf_skb_is_valid_access(off, size, type, prog, info);
7783 static bool __is_valid_xdp_access(int off, int size)
7785 if (off < 0 || off >= sizeof(struct xdp_md))
7787 if (off % size != 0)
7789 if (size != sizeof(__u32))
7795 static bool xdp_is_valid_access(int off, int size,
7796 enum bpf_access_type type,
7797 const struct bpf_prog *prog,
7798 struct bpf_insn_access_aux *info)
7800 if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
7802 case offsetof(struct xdp_md, egress_ifindex):
7807 if (type == BPF_WRITE) {
7808 if (bpf_prog_is_dev_bound(prog->aux)) {
7810 case offsetof(struct xdp_md, rx_queue_index):
7811 return __is_valid_xdp_access(off, size);
7818 case offsetof(struct xdp_md, data):
7819 info->reg_type = PTR_TO_PACKET;
7821 case offsetof(struct xdp_md, data_meta):
7822 info->reg_type = PTR_TO_PACKET_META;
7824 case offsetof(struct xdp_md, data_end):
7825 info->reg_type = PTR_TO_PACKET_END;
7829 return __is_valid_xdp_access(off, size);
7832 void bpf_warn_invalid_xdp_action(u32 act)
7834 const u32 act_max = XDP_REDIRECT;
7836 WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
7837 act > act_max ? "Illegal" : "Driver unsupported",
7840 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
7842 static bool sock_addr_is_valid_access(int off, int size,
7843 enum bpf_access_type type,
7844 const struct bpf_prog *prog,
7845 struct bpf_insn_access_aux *info)
7847 const int size_default = sizeof(__u32);
7849 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
7851 if (off % size != 0)
7854 /* Disallow access to IPv6 fields from IPv4 contex and vise
7858 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
7859 switch (prog->expected_attach_type) {
7860 case BPF_CGROUP_INET4_BIND:
7861 case BPF_CGROUP_INET4_CONNECT:
7862 case BPF_CGROUP_INET4_GETPEERNAME:
7863 case BPF_CGROUP_INET4_GETSOCKNAME:
7864 case BPF_CGROUP_UDP4_SENDMSG:
7865 case BPF_CGROUP_UDP4_RECVMSG:
7871 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
7872 switch (prog->expected_attach_type) {
7873 case BPF_CGROUP_INET6_BIND:
7874 case BPF_CGROUP_INET6_CONNECT:
7875 case BPF_CGROUP_INET6_GETPEERNAME:
7876 case BPF_CGROUP_INET6_GETSOCKNAME:
7877 case BPF_CGROUP_UDP6_SENDMSG:
7878 case BPF_CGROUP_UDP6_RECVMSG:
7884 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
7885 switch (prog->expected_attach_type) {
7886 case BPF_CGROUP_UDP4_SENDMSG:
7892 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
7894 switch (prog->expected_attach_type) {
7895 case BPF_CGROUP_UDP6_SENDMSG:
7904 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
7905 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
7906 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
7907 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
7909 case bpf_ctx_range(struct bpf_sock_addr, user_port):
7910 if (type == BPF_READ) {
7911 bpf_ctx_record_field_size(info, size_default);
7913 if (bpf_ctx_wide_access_ok(off, size,
7914 struct bpf_sock_addr,
7918 if (bpf_ctx_wide_access_ok(off, size,
7919 struct bpf_sock_addr,
7923 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
7926 if (bpf_ctx_wide_access_ok(off, size,
7927 struct bpf_sock_addr,
7931 if (bpf_ctx_wide_access_ok(off, size,
7932 struct bpf_sock_addr,
7936 if (size != size_default)
7940 case offsetof(struct bpf_sock_addr, sk):
7941 if (type != BPF_READ)
7943 if (size != sizeof(__u64))
7945 info->reg_type = PTR_TO_SOCKET;
7948 if (type == BPF_READ) {
7949 if (size != size_default)
7959 static bool sock_ops_is_valid_access(int off, int size,
7960 enum bpf_access_type type,
7961 const struct bpf_prog *prog,
7962 struct bpf_insn_access_aux *info)
7964 const int size_default = sizeof(__u32);
7966 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
7969 /* The verifier guarantees that size > 0. */
7970 if (off % size != 0)
7973 if (type == BPF_WRITE) {
7975 case offsetof(struct bpf_sock_ops, reply):
7976 case offsetof(struct bpf_sock_ops, sk_txhash):
7977 if (size != size_default)
7985 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
7987 if (size != sizeof(__u64))
7990 case offsetof(struct bpf_sock_ops, sk):
7991 if (size != sizeof(__u64))
7993 info->reg_type = PTR_TO_SOCKET_OR_NULL;
7995 case offsetof(struct bpf_sock_ops, skb_data):
7996 if (size != sizeof(__u64))
7998 info->reg_type = PTR_TO_PACKET;
8000 case offsetof(struct bpf_sock_ops, skb_data_end):
8001 if (size != sizeof(__u64))
8003 info->reg_type = PTR_TO_PACKET_END;
8005 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
8006 bpf_ctx_record_field_size(info, size_default);
8007 return bpf_ctx_narrow_access_ok(off, size,
8010 if (size != size_default)
8019 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
8020 const struct bpf_prog *prog)
8022 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
8025 static bool sk_skb_is_valid_access(int off, int size,
8026 enum bpf_access_type type,
8027 const struct bpf_prog *prog,
8028 struct bpf_insn_access_aux *info)
8031 case bpf_ctx_range(struct __sk_buff, tc_classid):
8032 case bpf_ctx_range(struct __sk_buff, data_meta):
8033 case bpf_ctx_range(struct __sk_buff, tstamp):
8034 case bpf_ctx_range(struct __sk_buff, wire_len):
8038 if (type == BPF_WRITE) {
8040 case bpf_ctx_range(struct __sk_buff, tc_index):
8041 case bpf_ctx_range(struct __sk_buff, priority):
8049 case bpf_ctx_range(struct __sk_buff, mark):
8051 case bpf_ctx_range(struct __sk_buff, data):
8052 info->reg_type = PTR_TO_PACKET;
8054 case bpf_ctx_range(struct __sk_buff, data_end):
8055 info->reg_type = PTR_TO_PACKET_END;
8059 return bpf_skb_is_valid_access(off, size, type, prog, info);
8062 static bool sk_msg_is_valid_access(int off, int size,
8063 enum bpf_access_type type,
8064 const struct bpf_prog *prog,
8065 struct bpf_insn_access_aux *info)
8067 if (type == BPF_WRITE)
8070 if (off % size != 0)
8074 case offsetof(struct sk_msg_md, data):
8075 info->reg_type = PTR_TO_PACKET;
8076 if (size != sizeof(__u64))
8079 case offsetof(struct sk_msg_md, data_end):
8080 info->reg_type = PTR_TO_PACKET_END;
8081 if (size != sizeof(__u64))
8084 case offsetof(struct sk_msg_md, sk):
8085 if (size != sizeof(__u64))
8087 info->reg_type = PTR_TO_SOCKET;
8089 case bpf_ctx_range(struct sk_msg_md, family):
8090 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
8091 case bpf_ctx_range(struct sk_msg_md, local_ip4):
8092 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
8093 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
8094 case bpf_ctx_range(struct sk_msg_md, remote_port):
8095 case bpf_ctx_range(struct sk_msg_md, local_port):
8096 case bpf_ctx_range(struct sk_msg_md, size):
8097 if (size != sizeof(__u32))
8106 static bool flow_dissector_is_valid_access(int off, int size,
8107 enum bpf_access_type type,
8108 const struct bpf_prog *prog,
8109 struct bpf_insn_access_aux *info)
8111 const int size_default = sizeof(__u32);
8113 if (off < 0 || off >= sizeof(struct __sk_buff))
8116 if (type == BPF_WRITE)
8120 case bpf_ctx_range(struct __sk_buff, data):
8121 if (size != size_default)
8123 info->reg_type = PTR_TO_PACKET;
8125 case bpf_ctx_range(struct __sk_buff, data_end):
8126 if (size != size_default)
8128 info->reg_type = PTR_TO_PACKET_END;
8130 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8131 if (size != sizeof(__u64))
8133 info->reg_type = PTR_TO_FLOW_KEYS;
8140 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
8141 const struct bpf_insn *si,
8142 struct bpf_insn *insn_buf,
8143 struct bpf_prog *prog,
8147 struct bpf_insn *insn = insn_buf;
8150 case offsetof(struct __sk_buff, data):
8151 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
8152 si->dst_reg, si->src_reg,
8153 offsetof(struct bpf_flow_dissector, data));
8156 case offsetof(struct __sk_buff, data_end):
8157 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
8158 si->dst_reg, si->src_reg,
8159 offsetof(struct bpf_flow_dissector, data_end));
8162 case offsetof(struct __sk_buff, flow_keys):
8163 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
8164 si->dst_reg, si->src_reg,
8165 offsetof(struct bpf_flow_dissector, flow_keys));
8169 return insn - insn_buf;
8172 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si,
8173 struct bpf_insn *insn)
8175 /* si->dst_reg = skb_shinfo(SKB); */
8176 #ifdef NET_SKBUFF_DATA_USES_OFFSET
8177 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8178 BPF_REG_AX, si->src_reg,
8179 offsetof(struct sk_buff, end));
8180 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
8181 si->dst_reg, si->src_reg,
8182 offsetof(struct sk_buff, head));
8183 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
8185 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8186 si->dst_reg, si->src_reg,
8187 offsetof(struct sk_buff, end));
8193 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
8194 const struct bpf_insn *si,
8195 struct bpf_insn *insn_buf,
8196 struct bpf_prog *prog, u32 *target_size)
8198 struct bpf_insn *insn = insn_buf;
8202 case offsetof(struct __sk_buff, len):
8203 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8204 bpf_target_off(struct sk_buff, len, 4,
8208 case offsetof(struct __sk_buff, protocol):
8209 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8210 bpf_target_off(struct sk_buff, protocol, 2,
8214 case offsetof(struct __sk_buff, vlan_proto):
8215 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8216 bpf_target_off(struct sk_buff, vlan_proto, 2,
8220 case offsetof(struct __sk_buff, priority):
8221 if (type == BPF_WRITE)
8222 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8223 bpf_target_off(struct sk_buff, priority, 4,
8226 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8227 bpf_target_off(struct sk_buff, priority, 4,
8231 case offsetof(struct __sk_buff, ingress_ifindex):
8232 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8233 bpf_target_off(struct sk_buff, skb_iif, 4,
8237 case offsetof(struct __sk_buff, ifindex):
8238 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8239 si->dst_reg, si->src_reg,
8240 offsetof(struct sk_buff, dev));
8241 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
8242 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8243 bpf_target_off(struct net_device, ifindex, 4,
8247 case offsetof(struct __sk_buff, hash):
8248 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8249 bpf_target_off(struct sk_buff, hash, 4,
8253 case offsetof(struct __sk_buff, mark):
8254 if (type == BPF_WRITE)
8255 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8256 bpf_target_off(struct sk_buff, mark, 4,
8259 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8260 bpf_target_off(struct sk_buff, mark, 4,
8264 case offsetof(struct __sk_buff, pkt_type):
8266 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8268 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
8269 #ifdef __BIG_ENDIAN_BITFIELD
8270 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
8274 case offsetof(struct __sk_buff, queue_mapping):
8275 if (type == BPF_WRITE) {
8276 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
8277 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8278 bpf_target_off(struct sk_buff,
8282 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8283 bpf_target_off(struct sk_buff,
8289 case offsetof(struct __sk_buff, vlan_present):
8291 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8292 PKT_VLAN_PRESENT_OFFSET());
8293 if (PKT_VLAN_PRESENT_BIT)
8294 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
8295 if (PKT_VLAN_PRESENT_BIT < 7)
8296 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
8299 case offsetof(struct __sk_buff, vlan_tci):
8300 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8301 bpf_target_off(struct sk_buff, vlan_tci, 2,
8305 case offsetof(struct __sk_buff, cb[0]) ...
8306 offsetofend(struct __sk_buff, cb[4]) - 1:
8307 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
8308 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
8309 offsetof(struct qdisc_skb_cb, data)) %
8312 prog->cb_access = 1;
8314 off -= offsetof(struct __sk_buff, cb[0]);
8315 off += offsetof(struct sk_buff, cb);
8316 off += offsetof(struct qdisc_skb_cb, data);
8317 if (type == BPF_WRITE)
8318 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
8321 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
8325 case offsetof(struct __sk_buff, tc_classid):
8326 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
8329 off -= offsetof(struct __sk_buff, tc_classid);
8330 off += offsetof(struct sk_buff, cb);
8331 off += offsetof(struct qdisc_skb_cb, tc_classid);
8333 if (type == BPF_WRITE)
8334 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
8337 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
8341 case offsetof(struct __sk_buff, data):
8342 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
8343 si->dst_reg, si->src_reg,
8344 offsetof(struct sk_buff, data));
8347 case offsetof(struct __sk_buff, data_meta):
8349 off -= offsetof(struct __sk_buff, data_meta);
8350 off += offsetof(struct sk_buff, cb);
8351 off += offsetof(struct bpf_skb_data_end, data_meta);
8352 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8356 case offsetof(struct __sk_buff, data_end):
8358 off -= offsetof(struct __sk_buff, data_end);
8359 off += offsetof(struct sk_buff, cb);
8360 off += offsetof(struct bpf_skb_data_end, data_end);
8361 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8365 case offsetof(struct __sk_buff, tc_index):
8366 #ifdef CONFIG_NET_SCHED
8367 if (type == BPF_WRITE)
8368 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8369 bpf_target_off(struct sk_buff, tc_index, 2,
8372 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8373 bpf_target_off(struct sk_buff, tc_index, 2,
8377 if (type == BPF_WRITE)
8378 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
8380 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8384 case offsetof(struct __sk_buff, napi_id):
8385 #if defined(CONFIG_NET_RX_BUSY_POLL)
8386 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8387 bpf_target_off(struct sk_buff, napi_id, 4,
8389 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
8390 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8393 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8396 case offsetof(struct __sk_buff, family):
8397 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
8399 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8400 si->dst_reg, si->src_reg,
8401 offsetof(struct sk_buff, sk));
8402 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8403 bpf_target_off(struct sock_common,
8407 case offsetof(struct __sk_buff, remote_ip4):
8408 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
8410 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8411 si->dst_reg, si->src_reg,
8412 offsetof(struct sk_buff, sk));
8413 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8414 bpf_target_off(struct sock_common,
8418 case offsetof(struct __sk_buff, local_ip4):
8419 BUILD_BUG_ON(sizeof_field(struct sock_common,
8420 skc_rcv_saddr) != 4);
8422 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8423 si->dst_reg, si->src_reg,
8424 offsetof(struct sk_buff, sk));
8425 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8426 bpf_target_off(struct sock_common,
8430 case offsetof(struct __sk_buff, remote_ip6[0]) ...
8431 offsetof(struct __sk_buff, remote_ip6[3]):
8432 #if IS_ENABLED(CONFIG_IPV6)
8433 BUILD_BUG_ON(sizeof_field(struct sock_common,
8434 skc_v6_daddr.s6_addr32[0]) != 4);
8437 off -= offsetof(struct __sk_buff, remote_ip6[0]);
8439 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8440 si->dst_reg, si->src_reg,
8441 offsetof(struct sk_buff, sk));
8442 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8443 offsetof(struct sock_common,
8444 skc_v6_daddr.s6_addr32[0]) +
8447 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8450 case offsetof(struct __sk_buff, local_ip6[0]) ...
8451 offsetof(struct __sk_buff, local_ip6[3]):
8452 #if IS_ENABLED(CONFIG_IPV6)
8453 BUILD_BUG_ON(sizeof_field(struct sock_common,
8454 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8457 off -= offsetof(struct __sk_buff, local_ip6[0]);
8459 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8460 si->dst_reg, si->src_reg,
8461 offsetof(struct sk_buff, sk));
8462 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8463 offsetof(struct sock_common,
8464 skc_v6_rcv_saddr.s6_addr32[0]) +
8467 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8471 case offsetof(struct __sk_buff, remote_port):
8472 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
8474 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8475 si->dst_reg, si->src_reg,
8476 offsetof(struct sk_buff, sk));
8477 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8478 bpf_target_off(struct sock_common,
8481 #ifndef __BIG_ENDIAN_BITFIELD
8482 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8486 case offsetof(struct __sk_buff, local_port):
8487 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
8489 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8490 si->dst_reg, si->src_reg,
8491 offsetof(struct sk_buff, sk));
8492 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8493 bpf_target_off(struct sock_common,
8494 skc_num, 2, target_size));
8497 case offsetof(struct __sk_buff, tstamp):
8498 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
8500 if (type == BPF_WRITE)
8501 *insn++ = BPF_STX_MEM(BPF_DW,
8502 si->dst_reg, si->src_reg,
8503 bpf_target_off(struct sk_buff,
8507 *insn++ = BPF_LDX_MEM(BPF_DW,
8508 si->dst_reg, si->src_reg,
8509 bpf_target_off(struct sk_buff,
8514 case offsetof(struct __sk_buff, gso_segs):
8515 insn = bpf_convert_shinfo_access(si, insn);
8516 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
8517 si->dst_reg, si->dst_reg,
8518 bpf_target_off(struct skb_shared_info,
8522 case offsetof(struct __sk_buff, gso_size):
8523 insn = bpf_convert_shinfo_access(si, insn);
8524 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
8525 si->dst_reg, si->dst_reg,
8526 bpf_target_off(struct skb_shared_info,
8530 case offsetof(struct __sk_buff, wire_len):
8531 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
8534 off -= offsetof(struct __sk_buff, wire_len);
8535 off += offsetof(struct sk_buff, cb);
8536 off += offsetof(struct qdisc_skb_cb, pkt_len);
8538 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
8541 case offsetof(struct __sk_buff, sk):
8542 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8543 si->dst_reg, si->src_reg,
8544 offsetof(struct sk_buff, sk));
8548 return insn - insn_buf;
8551 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
8552 const struct bpf_insn *si,
8553 struct bpf_insn *insn_buf,
8554 struct bpf_prog *prog, u32 *target_size)
8556 struct bpf_insn *insn = insn_buf;
8560 case offsetof(struct bpf_sock, bound_dev_if):
8561 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
8563 if (type == BPF_WRITE)
8564 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8565 offsetof(struct sock, sk_bound_dev_if));
8567 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8568 offsetof(struct sock, sk_bound_dev_if));
8571 case offsetof(struct bpf_sock, mark):
8572 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
8574 if (type == BPF_WRITE)
8575 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8576 offsetof(struct sock, sk_mark));
8578 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8579 offsetof(struct sock, sk_mark));
8582 case offsetof(struct bpf_sock, priority):
8583 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
8585 if (type == BPF_WRITE)
8586 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8587 offsetof(struct sock, sk_priority));
8589 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8590 offsetof(struct sock, sk_priority));
8593 case offsetof(struct bpf_sock, family):
8594 *insn++ = BPF_LDX_MEM(
8595 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
8596 si->dst_reg, si->src_reg,
8597 bpf_target_off(struct sock_common,
8599 sizeof_field(struct sock_common,
8604 case offsetof(struct bpf_sock, type):
8605 *insn++ = BPF_LDX_MEM(
8606 BPF_FIELD_SIZEOF(struct sock, sk_type),
8607 si->dst_reg, si->src_reg,
8608 bpf_target_off(struct sock, sk_type,
8609 sizeof_field(struct sock, sk_type),
8613 case offsetof(struct bpf_sock, protocol):
8614 *insn++ = BPF_LDX_MEM(
8615 BPF_FIELD_SIZEOF(struct sock, sk_protocol),
8616 si->dst_reg, si->src_reg,
8617 bpf_target_off(struct sock, sk_protocol,
8618 sizeof_field(struct sock, sk_protocol),
8622 case offsetof(struct bpf_sock, src_ip4):
8623 *insn++ = BPF_LDX_MEM(
8624 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8625 bpf_target_off(struct sock_common, skc_rcv_saddr,
8626 sizeof_field(struct sock_common,
8631 case offsetof(struct bpf_sock, dst_ip4):
8632 *insn++ = BPF_LDX_MEM(
8633 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8634 bpf_target_off(struct sock_common, skc_daddr,
8635 sizeof_field(struct sock_common,
8640 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8641 #if IS_ENABLED(CONFIG_IPV6)
8643 off -= offsetof(struct bpf_sock, src_ip6[0]);
8644 *insn++ = BPF_LDX_MEM(
8645 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8648 skc_v6_rcv_saddr.s6_addr32[0],
8649 sizeof_field(struct sock_common,
8650 skc_v6_rcv_saddr.s6_addr32[0]),
8651 target_size) + off);
8654 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8658 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8659 #if IS_ENABLED(CONFIG_IPV6)
8661 off -= offsetof(struct bpf_sock, dst_ip6[0]);
8662 *insn++ = BPF_LDX_MEM(
8663 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8664 bpf_target_off(struct sock_common,
8665 skc_v6_daddr.s6_addr32[0],
8666 sizeof_field(struct sock_common,
8667 skc_v6_daddr.s6_addr32[0]),
8668 target_size) + off);
8670 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8675 case offsetof(struct bpf_sock, src_port):
8676 *insn++ = BPF_LDX_MEM(
8677 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
8678 si->dst_reg, si->src_reg,
8679 bpf_target_off(struct sock_common, skc_num,
8680 sizeof_field(struct sock_common,
8685 case offsetof(struct bpf_sock, dst_port):
8686 *insn++ = BPF_LDX_MEM(
8687 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
8688 si->dst_reg, si->src_reg,
8689 bpf_target_off(struct sock_common, skc_dport,
8690 sizeof_field(struct sock_common,
8695 case offsetof(struct bpf_sock, state):
8696 *insn++ = BPF_LDX_MEM(
8697 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
8698 si->dst_reg, si->src_reg,
8699 bpf_target_off(struct sock_common, skc_state,
8700 sizeof_field(struct sock_common,
8704 case offsetof(struct bpf_sock, rx_queue_mapping):
8706 *insn++ = BPF_LDX_MEM(
8707 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
8708 si->dst_reg, si->src_reg,
8709 bpf_target_off(struct sock, sk_rx_queue_mapping,
8710 sizeof_field(struct sock,
8711 sk_rx_queue_mapping),
8713 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
8715 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
8717 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
8723 return insn - insn_buf;
8726 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
8727 const struct bpf_insn *si,
8728 struct bpf_insn *insn_buf,
8729 struct bpf_prog *prog, u32 *target_size)
8731 struct bpf_insn *insn = insn_buf;
8734 case offsetof(struct __sk_buff, ifindex):
8735 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8736 si->dst_reg, si->src_reg,
8737 offsetof(struct sk_buff, dev));
8738 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8739 bpf_target_off(struct net_device, ifindex, 4,
8743 return bpf_convert_ctx_access(type, si, insn_buf, prog,
8747 return insn - insn_buf;
8750 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
8751 const struct bpf_insn *si,
8752 struct bpf_insn *insn_buf,
8753 struct bpf_prog *prog, u32 *target_size)
8755 struct bpf_insn *insn = insn_buf;
8758 case offsetof(struct xdp_md, data):
8759 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
8760 si->dst_reg, si->src_reg,
8761 offsetof(struct xdp_buff, data));
8763 case offsetof(struct xdp_md, data_meta):
8764 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
8765 si->dst_reg, si->src_reg,
8766 offsetof(struct xdp_buff, data_meta));
8768 case offsetof(struct xdp_md, data_end):
8769 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
8770 si->dst_reg, si->src_reg,
8771 offsetof(struct xdp_buff, data_end));
8773 case offsetof(struct xdp_md, ingress_ifindex):
8774 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
8775 si->dst_reg, si->src_reg,
8776 offsetof(struct xdp_buff, rxq));
8777 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
8778 si->dst_reg, si->dst_reg,
8779 offsetof(struct xdp_rxq_info, dev));
8780 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8781 offsetof(struct net_device, ifindex));
8783 case offsetof(struct xdp_md, rx_queue_index):
8784 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
8785 si->dst_reg, si->src_reg,
8786 offsetof(struct xdp_buff, rxq));
8787 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8788 offsetof(struct xdp_rxq_info,
8791 case offsetof(struct xdp_md, egress_ifindex):
8792 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
8793 si->dst_reg, si->src_reg,
8794 offsetof(struct xdp_buff, txq));
8795 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
8796 si->dst_reg, si->dst_reg,
8797 offsetof(struct xdp_txq_info, dev));
8798 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8799 offsetof(struct net_device, ifindex));
8803 return insn - insn_buf;
8806 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
8807 * context Structure, F is Field in context structure that contains a pointer
8808 * to Nested Structure of type NS that has the field NF.
8810 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
8811 * sure that SIZE is not greater than actual size of S.F.NF.
8813 * If offset OFF is provided, the load happens from that offset relative to
8816 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
8818 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
8819 si->src_reg, offsetof(S, F)); \
8820 *insn++ = BPF_LDX_MEM( \
8821 SIZE, si->dst_reg, si->dst_reg, \
8822 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
8827 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
8828 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
8829 BPF_FIELD_SIZEOF(NS, NF), 0)
8831 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
8832 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
8834 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
8835 * "register" since two registers available in convert_ctx_access are not
8836 * enough: we can't override neither SRC, since it contains value to store, nor
8837 * DST since it contains pointer to context that may be used by later
8838 * instructions. But we need a temporary place to save pointer to nested
8839 * structure whose field we want to store to.
8841 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
8843 int tmp_reg = BPF_REG_9; \
8844 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
8846 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
8848 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
8850 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
8851 si->dst_reg, offsetof(S, F)); \
8852 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \
8853 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
8856 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
8860 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
8863 if (type == BPF_WRITE) { \
8864 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
8867 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
8868 S, NS, F, NF, SIZE, OFF); \
8872 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
8873 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
8874 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
8876 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
8877 const struct bpf_insn *si,
8878 struct bpf_insn *insn_buf,
8879 struct bpf_prog *prog, u32 *target_size)
8881 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
8882 struct bpf_insn *insn = insn_buf;
8885 case offsetof(struct bpf_sock_addr, user_family):
8886 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
8887 struct sockaddr, uaddr, sa_family);
8890 case offsetof(struct bpf_sock_addr, user_ip4):
8891 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
8892 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
8893 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
8896 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8898 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
8899 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
8900 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
8901 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
8905 case offsetof(struct bpf_sock_addr, user_port):
8906 /* To get port we need to know sa_family first and then treat
8907 * sockaddr as either sockaddr_in or sockaddr_in6.
8908 * Though we can simplify since port field has same offset and
8909 * size in both structures.
8910 * Here we check this invariant and use just one of the
8911 * structures if it's true.
8913 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
8914 offsetof(struct sockaddr_in6, sin6_port));
8915 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
8916 sizeof_field(struct sockaddr_in6, sin6_port));
8917 /* Account for sin6_port being smaller than user_port. */
8918 port_size = min(port_size, BPF_LDST_BYTES(si));
8919 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
8920 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
8921 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
8924 case offsetof(struct bpf_sock_addr, family):
8925 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
8926 struct sock, sk, sk_family);
8929 case offsetof(struct bpf_sock_addr, type):
8930 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
8931 struct sock, sk, sk_type);
8934 case offsetof(struct bpf_sock_addr, protocol):
8935 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
8936 struct sock, sk, sk_protocol);
8939 case offsetof(struct bpf_sock_addr, msg_src_ip4):
8940 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
8941 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
8942 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
8943 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
8946 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8949 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
8950 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
8951 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
8952 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
8953 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
8955 case offsetof(struct bpf_sock_addr, sk):
8956 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
8957 si->dst_reg, si->src_reg,
8958 offsetof(struct bpf_sock_addr_kern, sk));
8962 return insn - insn_buf;
8965 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
8966 const struct bpf_insn *si,
8967 struct bpf_insn *insn_buf,
8968 struct bpf_prog *prog,
8971 struct bpf_insn *insn = insn_buf;
8974 /* Helper macro for adding read access to tcp_sock or sock fields. */
8975 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
8977 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \
8978 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
8979 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
8980 if (si->dst_reg == reg || si->src_reg == reg) \
8982 if (si->dst_reg == reg || si->src_reg == reg) \
8984 if (si->dst_reg == si->src_reg) { \
8985 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
8986 offsetof(struct bpf_sock_ops_kern, \
8988 fullsock_reg = reg; \
8991 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
8992 struct bpf_sock_ops_kern, \
8994 fullsock_reg, si->src_reg, \
8995 offsetof(struct bpf_sock_ops_kern, \
8997 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
8998 if (si->dst_reg == si->src_reg) \
8999 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9000 offsetof(struct bpf_sock_ops_kern, \
9002 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9003 struct bpf_sock_ops_kern, sk),\
9004 si->dst_reg, si->src_reg, \
9005 offsetof(struct bpf_sock_ops_kern, sk));\
9006 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
9008 si->dst_reg, si->dst_reg, \
9009 offsetof(OBJ, OBJ_FIELD)); \
9010 if (si->dst_reg == si->src_reg) { \
9011 *insn++ = BPF_JMP_A(1); \
9012 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9013 offsetof(struct bpf_sock_ops_kern, \
9018 #define SOCK_OPS_GET_SK() \
9020 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \
9021 if (si->dst_reg == reg || si->src_reg == reg) \
9023 if (si->dst_reg == reg || si->src_reg == reg) \
9025 if (si->dst_reg == si->src_reg) { \
9026 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
9027 offsetof(struct bpf_sock_ops_kern, \
9029 fullsock_reg = reg; \
9032 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9033 struct bpf_sock_ops_kern, \
9035 fullsock_reg, si->src_reg, \
9036 offsetof(struct bpf_sock_ops_kern, \
9038 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
9039 if (si->dst_reg == si->src_reg) \
9040 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9041 offsetof(struct bpf_sock_ops_kern, \
9043 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9044 struct bpf_sock_ops_kern, sk),\
9045 si->dst_reg, si->src_reg, \
9046 offsetof(struct bpf_sock_ops_kern, sk));\
9047 if (si->dst_reg == si->src_reg) { \
9048 *insn++ = BPF_JMP_A(1); \
9049 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9050 offsetof(struct bpf_sock_ops_kern, \
9055 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
9056 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
9058 /* Helper macro for adding write access to tcp_sock or sock fields.
9059 * The macro is called with two registers, dst_reg which contains a pointer
9060 * to ctx (context) and src_reg which contains the value that should be
9061 * stored. However, we need an additional register since we cannot overwrite
9062 * dst_reg because it may be used later in the program.
9063 * Instead we "borrow" one of the other register. We first save its value
9064 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
9065 * it at the end of the macro.
9067 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
9069 int reg = BPF_REG_9; \
9070 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
9071 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
9072 if (si->dst_reg == reg || si->src_reg == reg) \
9074 if (si->dst_reg == reg || si->src_reg == reg) \
9076 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
9077 offsetof(struct bpf_sock_ops_kern, \
9079 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9080 struct bpf_sock_ops_kern, \
9083 offsetof(struct bpf_sock_ops_kern, \
9085 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
9086 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9087 struct bpf_sock_ops_kern, sk),\
9089 offsetof(struct bpf_sock_ops_kern, sk));\
9090 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \
9092 offsetof(OBJ, OBJ_FIELD)); \
9093 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
9094 offsetof(struct bpf_sock_ops_kern, \
9098 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
9100 if (TYPE == BPF_WRITE) \
9101 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
9103 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
9106 if (insn > insn_buf)
9107 return insn - insn_buf;
9110 case offsetof(struct bpf_sock_ops, op):
9111 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9113 si->dst_reg, si->src_reg,
9114 offsetof(struct bpf_sock_ops_kern, op));
9117 case offsetof(struct bpf_sock_ops, replylong[0]) ...
9118 offsetof(struct bpf_sock_ops, replylong[3]):
9119 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
9120 sizeof_field(struct bpf_sock_ops_kern, reply));
9121 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
9122 sizeof_field(struct bpf_sock_ops_kern, replylong));
9124 off -= offsetof(struct bpf_sock_ops, replylong[0]);
9125 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
9126 if (type == BPF_WRITE)
9127 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9130 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9134 case offsetof(struct bpf_sock_ops, family):
9135 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9137 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9138 struct bpf_sock_ops_kern, sk),
9139 si->dst_reg, si->src_reg,
9140 offsetof(struct bpf_sock_ops_kern, sk));
9141 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9142 offsetof(struct sock_common, skc_family));
9145 case offsetof(struct bpf_sock_ops, remote_ip4):
9146 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9148 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9149 struct bpf_sock_ops_kern, sk),
9150 si->dst_reg, si->src_reg,
9151 offsetof(struct bpf_sock_ops_kern, sk));
9152 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9153 offsetof(struct sock_common, skc_daddr));
9156 case offsetof(struct bpf_sock_ops, local_ip4):
9157 BUILD_BUG_ON(sizeof_field(struct sock_common,
9158 skc_rcv_saddr) != 4);
9160 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9161 struct bpf_sock_ops_kern, sk),
9162 si->dst_reg, si->src_reg,
9163 offsetof(struct bpf_sock_ops_kern, sk));
9164 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9165 offsetof(struct sock_common,
9169 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
9170 offsetof(struct bpf_sock_ops, remote_ip6[3]):
9171 #if IS_ENABLED(CONFIG_IPV6)
9172 BUILD_BUG_ON(sizeof_field(struct sock_common,
9173 skc_v6_daddr.s6_addr32[0]) != 4);
9176 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
9177 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9178 struct bpf_sock_ops_kern, sk),
9179 si->dst_reg, si->src_reg,
9180 offsetof(struct bpf_sock_ops_kern, sk));
9181 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9182 offsetof(struct sock_common,
9183 skc_v6_daddr.s6_addr32[0]) +
9186 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9190 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
9191 offsetof(struct bpf_sock_ops, local_ip6[3]):
9192 #if IS_ENABLED(CONFIG_IPV6)
9193 BUILD_BUG_ON(sizeof_field(struct sock_common,
9194 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9197 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
9198 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9199 struct bpf_sock_ops_kern, sk),
9200 si->dst_reg, si->src_reg,
9201 offsetof(struct bpf_sock_ops_kern, sk));
9202 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9203 offsetof(struct sock_common,
9204 skc_v6_rcv_saddr.s6_addr32[0]) +
9207 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9211 case offsetof(struct bpf_sock_ops, remote_port):
9212 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9214 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9215 struct bpf_sock_ops_kern, sk),
9216 si->dst_reg, si->src_reg,
9217 offsetof(struct bpf_sock_ops_kern, sk));
9218 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9219 offsetof(struct sock_common, skc_dport));
9220 #ifndef __BIG_ENDIAN_BITFIELD
9221 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9225 case offsetof(struct bpf_sock_ops, local_port):
9226 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9228 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9229 struct bpf_sock_ops_kern, sk),
9230 si->dst_reg, si->src_reg,
9231 offsetof(struct bpf_sock_ops_kern, sk));
9232 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9233 offsetof(struct sock_common, skc_num));
9236 case offsetof(struct bpf_sock_ops, is_fullsock):
9237 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9238 struct bpf_sock_ops_kern,
9240 si->dst_reg, si->src_reg,
9241 offsetof(struct bpf_sock_ops_kern,
9245 case offsetof(struct bpf_sock_ops, state):
9246 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
9248 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9249 struct bpf_sock_ops_kern, sk),
9250 si->dst_reg, si->src_reg,
9251 offsetof(struct bpf_sock_ops_kern, sk));
9252 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
9253 offsetof(struct sock_common, skc_state));
9256 case offsetof(struct bpf_sock_ops, rtt_min):
9257 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
9258 sizeof(struct minmax));
9259 BUILD_BUG_ON(sizeof(struct minmax) <
9260 sizeof(struct minmax_sample));
9262 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9263 struct bpf_sock_ops_kern, sk),
9264 si->dst_reg, si->src_reg,
9265 offsetof(struct bpf_sock_ops_kern, sk));
9266 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9267 offsetof(struct tcp_sock, rtt_min) +
9268 sizeof_field(struct minmax_sample, t));
9271 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
9272 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
9276 case offsetof(struct bpf_sock_ops, sk_txhash):
9277 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
9280 case offsetof(struct bpf_sock_ops, snd_cwnd):
9281 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
9283 case offsetof(struct bpf_sock_ops, srtt_us):
9284 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
9286 case offsetof(struct bpf_sock_ops, snd_ssthresh):
9287 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
9289 case offsetof(struct bpf_sock_ops, rcv_nxt):
9290 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
9292 case offsetof(struct bpf_sock_ops, snd_nxt):
9293 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
9295 case offsetof(struct bpf_sock_ops, snd_una):
9296 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
9298 case offsetof(struct bpf_sock_ops, mss_cache):
9299 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
9301 case offsetof(struct bpf_sock_ops, ecn_flags):
9302 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
9304 case offsetof(struct bpf_sock_ops, rate_delivered):
9305 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
9307 case offsetof(struct bpf_sock_ops, rate_interval_us):
9308 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
9310 case offsetof(struct bpf_sock_ops, packets_out):
9311 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
9313 case offsetof(struct bpf_sock_ops, retrans_out):
9314 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
9316 case offsetof(struct bpf_sock_ops, total_retrans):
9317 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
9319 case offsetof(struct bpf_sock_ops, segs_in):
9320 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
9322 case offsetof(struct bpf_sock_ops, data_segs_in):
9323 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
9325 case offsetof(struct bpf_sock_ops, segs_out):
9326 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
9328 case offsetof(struct bpf_sock_ops, data_segs_out):
9329 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
9331 case offsetof(struct bpf_sock_ops, lost_out):
9332 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
9334 case offsetof(struct bpf_sock_ops, sacked_out):
9335 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
9337 case offsetof(struct bpf_sock_ops, bytes_received):
9338 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
9340 case offsetof(struct bpf_sock_ops, bytes_acked):
9341 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
9343 case offsetof(struct bpf_sock_ops, sk):
9346 case offsetof(struct bpf_sock_ops, skb_data_end):
9347 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9349 si->dst_reg, si->src_reg,
9350 offsetof(struct bpf_sock_ops_kern,
9353 case offsetof(struct bpf_sock_ops, skb_data):
9354 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9356 si->dst_reg, si->src_reg,
9357 offsetof(struct bpf_sock_ops_kern,
9359 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9360 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9361 si->dst_reg, si->dst_reg,
9362 offsetof(struct sk_buff, data));
9364 case offsetof(struct bpf_sock_ops, skb_len):
9365 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9367 si->dst_reg, si->src_reg,
9368 offsetof(struct bpf_sock_ops_kern,
9370 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9371 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9372 si->dst_reg, si->dst_reg,
9373 offsetof(struct sk_buff, len));
9375 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9376 off = offsetof(struct sk_buff, cb);
9377 off += offsetof(struct tcp_skb_cb, tcp_flags);
9378 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
9379 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9381 si->dst_reg, si->src_reg,
9382 offsetof(struct bpf_sock_ops_kern,
9384 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9385 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
9387 si->dst_reg, si->dst_reg, off);
9390 return insn - insn_buf;
9393 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
9394 const struct bpf_insn *si,
9395 struct bpf_insn *insn_buf,
9396 struct bpf_prog *prog, u32 *target_size)
9398 struct bpf_insn *insn = insn_buf;
9402 case offsetof(struct __sk_buff, data_end):
9404 off -= offsetof(struct __sk_buff, data_end);
9405 off += offsetof(struct sk_buff, cb);
9406 off += offsetof(struct tcp_skb_cb, bpf.data_end);
9407 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9411 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9415 return insn - insn_buf;
9418 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
9419 const struct bpf_insn *si,
9420 struct bpf_insn *insn_buf,
9421 struct bpf_prog *prog, u32 *target_size)
9423 struct bpf_insn *insn = insn_buf;
9424 #if IS_ENABLED(CONFIG_IPV6)
9428 /* convert ctx uses the fact sg element is first in struct */
9429 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
9432 case offsetof(struct sk_msg_md, data):
9433 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
9434 si->dst_reg, si->src_reg,
9435 offsetof(struct sk_msg, data));
9437 case offsetof(struct sk_msg_md, data_end):
9438 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
9439 si->dst_reg, si->src_reg,
9440 offsetof(struct sk_msg, data_end));
9442 case offsetof(struct sk_msg_md, family):
9443 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9445 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9447 si->dst_reg, si->src_reg,
9448 offsetof(struct sk_msg, sk));
9449 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9450 offsetof(struct sock_common, skc_family));
9453 case offsetof(struct sk_msg_md, remote_ip4):
9454 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9456 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9458 si->dst_reg, si->src_reg,
9459 offsetof(struct sk_msg, sk));
9460 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9461 offsetof(struct sock_common, skc_daddr));
9464 case offsetof(struct sk_msg_md, local_ip4):
9465 BUILD_BUG_ON(sizeof_field(struct sock_common,
9466 skc_rcv_saddr) != 4);
9468 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9470 si->dst_reg, si->src_reg,
9471 offsetof(struct sk_msg, sk));
9472 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9473 offsetof(struct sock_common,
9477 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
9478 offsetof(struct sk_msg_md, remote_ip6[3]):
9479 #if IS_ENABLED(CONFIG_IPV6)
9480 BUILD_BUG_ON(sizeof_field(struct sock_common,
9481 skc_v6_daddr.s6_addr32[0]) != 4);
9484 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
9485 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9487 si->dst_reg, si->src_reg,
9488 offsetof(struct sk_msg, sk));
9489 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9490 offsetof(struct sock_common,
9491 skc_v6_daddr.s6_addr32[0]) +
9494 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9498 case offsetof(struct sk_msg_md, local_ip6[0]) ...
9499 offsetof(struct sk_msg_md, local_ip6[3]):
9500 #if IS_ENABLED(CONFIG_IPV6)
9501 BUILD_BUG_ON(sizeof_field(struct sock_common,
9502 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9505 off -= offsetof(struct sk_msg_md, local_ip6[0]);
9506 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9508 si->dst_reg, si->src_reg,
9509 offsetof(struct sk_msg, sk));
9510 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9511 offsetof(struct sock_common,
9512 skc_v6_rcv_saddr.s6_addr32[0]) +
9515 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9519 case offsetof(struct sk_msg_md, remote_port):
9520 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9522 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9524 si->dst_reg, si->src_reg,
9525 offsetof(struct sk_msg, sk));
9526 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9527 offsetof(struct sock_common, skc_dport));
9528 #ifndef __BIG_ENDIAN_BITFIELD
9529 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9533 case offsetof(struct sk_msg_md, local_port):
9534 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9536 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9538 si->dst_reg, si->src_reg,
9539 offsetof(struct sk_msg, sk));
9540 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9541 offsetof(struct sock_common, skc_num));
9544 case offsetof(struct sk_msg_md, size):
9545 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
9546 si->dst_reg, si->src_reg,
9547 offsetof(struct sk_msg_sg, size));
9550 case offsetof(struct sk_msg_md, sk):
9551 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
9552 si->dst_reg, si->src_reg,
9553 offsetof(struct sk_msg, sk));
9557 return insn - insn_buf;
9560 const struct bpf_verifier_ops sk_filter_verifier_ops = {
9561 .get_func_proto = sk_filter_func_proto,
9562 .is_valid_access = sk_filter_is_valid_access,
9563 .convert_ctx_access = bpf_convert_ctx_access,
9564 .gen_ld_abs = bpf_gen_ld_abs,
9567 const struct bpf_prog_ops sk_filter_prog_ops = {
9568 .test_run = bpf_prog_test_run_skb,
9571 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
9572 .get_func_proto = tc_cls_act_func_proto,
9573 .is_valid_access = tc_cls_act_is_valid_access,
9574 .convert_ctx_access = tc_cls_act_convert_ctx_access,
9575 .gen_prologue = tc_cls_act_prologue,
9576 .gen_ld_abs = bpf_gen_ld_abs,
9579 const struct bpf_prog_ops tc_cls_act_prog_ops = {
9580 .test_run = bpf_prog_test_run_skb,
9583 const struct bpf_verifier_ops xdp_verifier_ops = {
9584 .get_func_proto = xdp_func_proto,
9585 .is_valid_access = xdp_is_valid_access,
9586 .convert_ctx_access = xdp_convert_ctx_access,
9587 .gen_prologue = bpf_noop_prologue,
9590 const struct bpf_prog_ops xdp_prog_ops = {
9591 .test_run = bpf_prog_test_run_xdp,
9594 const struct bpf_verifier_ops cg_skb_verifier_ops = {
9595 .get_func_proto = cg_skb_func_proto,
9596 .is_valid_access = cg_skb_is_valid_access,
9597 .convert_ctx_access = bpf_convert_ctx_access,
9600 const struct bpf_prog_ops cg_skb_prog_ops = {
9601 .test_run = bpf_prog_test_run_skb,
9604 const struct bpf_verifier_ops lwt_in_verifier_ops = {
9605 .get_func_proto = lwt_in_func_proto,
9606 .is_valid_access = lwt_is_valid_access,
9607 .convert_ctx_access = bpf_convert_ctx_access,
9610 const struct bpf_prog_ops lwt_in_prog_ops = {
9611 .test_run = bpf_prog_test_run_skb,
9614 const struct bpf_verifier_ops lwt_out_verifier_ops = {
9615 .get_func_proto = lwt_out_func_proto,
9616 .is_valid_access = lwt_is_valid_access,
9617 .convert_ctx_access = bpf_convert_ctx_access,
9620 const struct bpf_prog_ops lwt_out_prog_ops = {
9621 .test_run = bpf_prog_test_run_skb,
9624 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
9625 .get_func_proto = lwt_xmit_func_proto,
9626 .is_valid_access = lwt_is_valid_access,
9627 .convert_ctx_access = bpf_convert_ctx_access,
9628 .gen_prologue = tc_cls_act_prologue,
9631 const struct bpf_prog_ops lwt_xmit_prog_ops = {
9632 .test_run = bpf_prog_test_run_skb,
9635 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
9636 .get_func_proto = lwt_seg6local_func_proto,
9637 .is_valid_access = lwt_is_valid_access,
9638 .convert_ctx_access = bpf_convert_ctx_access,
9641 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
9642 .test_run = bpf_prog_test_run_skb,
9645 const struct bpf_verifier_ops cg_sock_verifier_ops = {
9646 .get_func_proto = sock_filter_func_proto,
9647 .is_valid_access = sock_filter_is_valid_access,
9648 .convert_ctx_access = bpf_sock_convert_ctx_access,
9651 const struct bpf_prog_ops cg_sock_prog_ops = {
9654 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
9655 .get_func_proto = sock_addr_func_proto,
9656 .is_valid_access = sock_addr_is_valid_access,
9657 .convert_ctx_access = sock_addr_convert_ctx_access,
9660 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
9663 const struct bpf_verifier_ops sock_ops_verifier_ops = {
9664 .get_func_proto = sock_ops_func_proto,
9665 .is_valid_access = sock_ops_is_valid_access,
9666 .convert_ctx_access = sock_ops_convert_ctx_access,
9669 const struct bpf_prog_ops sock_ops_prog_ops = {
9672 const struct bpf_verifier_ops sk_skb_verifier_ops = {
9673 .get_func_proto = sk_skb_func_proto,
9674 .is_valid_access = sk_skb_is_valid_access,
9675 .convert_ctx_access = sk_skb_convert_ctx_access,
9676 .gen_prologue = sk_skb_prologue,
9679 const struct bpf_prog_ops sk_skb_prog_ops = {
9682 const struct bpf_verifier_ops sk_msg_verifier_ops = {
9683 .get_func_proto = sk_msg_func_proto,
9684 .is_valid_access = sk_msg_is_valid_access,
9685 .convert_ctx_access = sk_msg_convert_ctx_access,
9686 .gen_prologue = bpf_noop_prologue,
9689 const struct bpf_prog_ops sk_msg_prog_ops = {
9692 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
9693 .get_func_proto = flow_dissector_func_proto,
9694 .is_valid_access = flow_dissector_is_valid_access,
9695 .convert_ctx_access = flow_dissector_convert_ctx_access,
9698 const struct bpf_prog_ops flow_dissector_prog_ops = {
9699 .test_run = bpf_prog_test_run_flow_dissector,
9702 int sk_detach_filter(struct sock *sk)
9705 struct sk_filter *filter;
9707 if (sock_flag(sk, SOCK_FILTER_LOCKED))
9710 filter = rcu_dereference_protected(sk->sk_filter,
9711 lockdep_sock_is_held(sk));
9713 RCU_INIT_POINTER(sk->sk_filter, NULL);
9714 sk_filter_uncharge(sk, filter);
9720 EXPORT_SYMBOL_GPL(sk_detach_filter);
9722 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
9725 struct sock_fprog_kern *fprog;
9726 struct sk_filter *filter;
9730 filter = rcu_dereference_protected(sk->sk_filter,
9731 lockdep_sock_is_held(sk));
9735 /* We're copying the filter that has been originally attached,
9736 * so no conversion/decode needed anymore. eBPF programs that
9737 * have no original program cannot be dumped through this.
9740 fprog = filter->prog->orig_prog;
9746 /* User space only enquires number of filter blocks. */
9750 if (len < fprog->len)
9754 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
9757 /* Instead of bytes, the API requests to return the number
9767 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
9768 struct sock_reuseport *reuse,
9769 struct sock *sk, struct sk_buff *skb,
9772 reuse_kern->skb = skb;
9773 reuse_kern->sk = sk;
9774 reuse_kern->selected_sk = NULL;
9775 reuse_kern->data_end = skb->data + skb_headlen(skb);
9776 reuse_kern->hash = hash;
9777 reuse_kern->reuseport_id = reuse->reuseport_id;
9778 reuse_kern->bind_inany = reuse->bind_inany;
9781 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
9782 struct bpf_prog *prog, struct sk_buff *skb,
9785 struct sk_reuseport_kern reuse_kern;
9786 enum sk_action action;
9788 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, hash);
9789 action = BPF_PROG_RUN(prog, &reuse_kern);
9791 if (action == SK_PASS)
9792 return reuse_kern.selected_sk;
9794 return ERR_PTR(-ECONNREFUSED);
9797 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
9798 struct bpf_map *, map, void *, key, u32, flags)
9800 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
9801 struct sock_reuseport *reuse;
9802 struct sock *selected_sk;
9804 selected_sk = map->ops->map_lookup_elem(map, key);
9808 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
9810 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
9811 if (sk_is_refcounted(selected_sk))
9812 sock_put(selected_sk);
9814 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
9815 * The only (!reuse) case here is - the sk has already been
9816 * unhashed (e.g. by close()), so treat it as -ENOENT.
9818 * Other maps (e.g. sock_map) do not provide this guarantee and
9819 * the sk may never be in the reuseport group to begin with.
9821 return is_sockarray ? -ENOENT : -EINVAL;
9824 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
9825 struct sock *sk = reuse_kern->sk;
9827 if (sk->sk_protocol != selected_sk->sk_protocol)
9829 else if (sk->sk_family != selected_sk->sk_family)
9830 return -EAFNOSUPPORT;
9832 /* Catch all. Likely bound to a different sockaddr. */
9836 reuse_kern->selected_sk = selected_sk;
9841 static const struct bpf_func_proto sk_select_reuseport_proto = {
9842 .func = sk_select_reuseport,
9844 .ret_type = RET_INTEGER,
9845 .arg1_type = ARG_PTR_TO_CTX,
9846 .arg2_type = ARG_CONST_MAP_PTR,
9847 .arg3_type = ARG_PTR_TO_MAP_KEY,
9848 .arg4_type = ARG_ANYTHING,
9851 BPF_CALL_4(sk_reuseport_load_bytes,
9852 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
9853 void *, to, u32, len)
9855 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
9858 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
9859 .func = sk_reuseport_load_bytes,
9861 .ret_type = RET_INTEGER,
9862 .arg1_type = ARG_PTR_TO_CTX,
9863 .arg2_type = ARG_ANYTHING,
9864 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
9865 .arg4_type = ARG_CONST_SIZE,
9868 BPF_CALL_5(sk_reuseport_load_bytes_relative,
9869 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
9870 void *, to, u32, len, u32, start_header)
9872 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
9876 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
9877 .func = sk_reuseport_load_bytes_relative,
9879 .ret_type = RET_INTEGER,
9880 .arg1_type = ARG_PTR_TO_CTX,
9881 .arg2_type = ARG_ANYTHING,
9882 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
9883 .arg4_type = ARG_CONST_SIZE,
9884 .arg5_type = ARG_ANYTHING,
9887 static const struct bpf_func_proto *
9888 sk_reuseport_func_proto(enum bpf_func_id func_id,
9889 const struct bpf_prog *prog)
9892 case BPF_FUNC_sk_select_reuseport:
9893 return &sk_select_reuseport_proto;
9894 case BPF_FUNC_skb_load_bytes:
9895 return &sk_reuseport_load_bytes_proto;
9896 case BPF_FUNC_skb_load_bytes_relative:
9897 return &sk_reuseport_load_bytes_relative_proto;
9899 return bpf_base_func_proto(func_id);
9904 sk_reuseport_is_valid_access(int off, int size,
9905 enum bpf_access_type type,
9906 const struct bpf_prog *prog,
9907 struct bpf_insn_access_aux *info)
9909 const u32 size_default = sizeof(__u32);
9911 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
9912 off % size || type != BPF_READ)
9916 case offsetof(struct sk_reuseport_md, data):
9917 info->reg_type = PTR_TO_PACKET;
9918 return size == sizeof(__u64);
9920 case offsetof(struct sk_reuseport_md, data_end):
9921 info->reg_type = PTR_TO_PACKET_END;
9922 return size == sizeof(__u64);
9924 case offsetof(struct sk_reuseport_md, hash):
9925 return size == size_default;
9927 /* Fields that allow narrowing */
9928 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
9929 if (size < sizeof_field(struct sk_buff, protocol))
9932 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
9933 case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
9934 case bpf_ctx_range(struct sk_reuseport_md, len):
9935 bpf_ctx_record_field_size(info, size_default);
9936 return bpf_ctx_narrow_access_ok(off, size, size_default);
9943 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
9944 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
9945 si->dst_reg, si->src_reg, \
9946 bpf_target_off(struct sk_reuseport_kern, F, \
9947 sizeof_field(struct sk_reuseport_kern, F), \
9951 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
9952 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
9957 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \
9958 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
9963 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
9964 const struct bpf_insn *si,
9965 struct bpf_insn *insn_buf,
9966 struct bpf_prog *prog,
9969 struct bpf_insn *insn = insn_buf;
9972 case offsetof(struct sk_reuseport_md, data):
9973 SK_REUSEPORT_LOAD_SKB_FIELD(data);
9976 case offsetof(struct sk_reuseport_md, len):
9977 SK_REUSEPORT_LOAD_SKB_FIELD(len);
9980 case offsetof(struct sk_reuseport_md, eth_protocol):
9981 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
9984 case offsetof(struct sk_reuseport_md, ip_protocol):
9985 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
9988 case offsetof(struct sk_reuseport_md, data_end):
9989 SK_REUSEPORT_LOAD_FIELD(data_end);
9992 case offsetof(struct sk_reuseport_md, hash):
9993 SK_REUSEPORT_LOAD_FIELD(hash);
9996 case offsetof(struct sk_reuseport_md, bind_inany):
9997 SK_REUSEPORT_LOAD_FIELD(bind_inany);
10001 return insn - insn_buf;
10004 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
10005 .get_func_proto = sk_reuseport_func_proto,
10006 .is_valid_access = sk_reuseport_is_valid_access,
10007 .convert_ctx_access = sk_reuseport_convert_ctx_access,
10010 const struct bpf_prog_ops sk_reuseport_prog_ops = {
10013 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
10014 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
10016 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
10017 struct sock *, sk, u64, flags)
10019 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
10020 BPF_SK_LOOKUP_F_NO_REUSEPORT)))
10022 if (unlikely(sk && sk_is_refcounted(sk)))
10023 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
10024 if (unlikely(sk && sk->sk_state == TCP_ESTABLISHED))
10025 return -ESOCKTNOSUPPORT; /* reject connected sockets */
10027 /* Check if socket is suitable for packet L3/L4 protocol */
10028 if (sk && sk->sk_protocol != ctx->protocol)
10029 return -EPROTOTYPE;
10030 if (sk && sk->sk_family != ctx->family &&
10031 (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
10032 return -EAFNOSUPPORT;
10034 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
10037 /* Select socket as lookup result */
10038 ctx->selected_sk = sk;
10039 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
10043 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
10044 .func = bpf_sk_lookup_assign,
10046 .ret_type = RET_INTEGER,
10047 .arg1_type = ARG_PTR_TO_CTX,
10048 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL,
10049 .arg3_type = ARG_ANYTHING,
10052 static const struct bpf_func_proto *
10053 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
10056 case BPF_FUNC_perf_event_output:
10057 return &bpf_event_output_data_proto;
10058 case BPF_FUNC_sk_assign:
10059 return &bpf_sk_lookup_assign_proto;
10060 case BPF_FUNC_sk_release:
10061 return &bpf_sk_release_proto;
10063 return bpf_sk_base_func_proto(func_id);
10067 static bool sk_lookup_is_valid_access(int off, int size,
10068 enum bpf_access_type type,
10069 const struct bpf_prog *prog,
10070 struct bpf_insn_access_aux *info)
10072 if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
10074 if (off % size != 0)
10076 if (type != BPF_READ)
10080 case offsetof(struct bpf_sk_lookup, sk):
10081 info->reg_type = PTR_TO_SOCKET_OR_NULL;
10082 return size == sizeof(__u64);
10084 case bpf_ctx_range(struct bpf_sk_lookup, family):
10085 case bpf_ctx_range(struct bpf_sk_lookup, protocol):
10086 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
10087 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
10088 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
10089 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
10090 case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
10091 case bpf_ctx_range(struct bpf_sk_lookup, local_port):
10092 bpf_ctx_record_field_size(info, sizeof(__u32));
10093 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
10100 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
10101 const struct bpf_insn *si,
10102 struct bpf_insn *insn_buf,
10103 struct bpf_prog *prog,
10106 struct bpf_insn *insn = insn_buf;
10109 case offsetof(struct bpf_sk_lookup, sk):
10110 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10111 offsetof(struct bpf_sk_lookup_kern, selected_sk));
10114 case offsetof(struct bpf_sk_lookup, family):
10115 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10116 bpf_target_off(struct bpf_sk_lookup_kern,
10117 family, 2, target_size));
10120 case offsetof(struct bpf_sk_lookup, protocol):
10121 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10122 bpf_target_off(struct bpf_sk_lookup_kern,
10123 protocol, 2, target_size));
10126 case offsetof(struct bpf_sk_lookup, remote_ip4):
10127 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10128 bpf_target_off(struct bpf_sk_lookup_kern,
10129 v4.saddr, 4, target_size));
10132 case offsetof(struct bpf_sk_lookup, local_ip4):
10133 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10134 bpf_target_off(struct bpf_sk_lookup_kern,
10135 v4.daddr, 4, target_size));
10138 case bpf_ctx_range_till(struct bpf_sk_lookup,
10139 remote_ip6[0], remote_ip6[3]): {
10140 #if IS_ENABLED(CONFIG_IPV6)
10143 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
10144 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10145 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10146 offsetof(struct bpf_sk_lookup_kern, v6.saddr));
10147 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10148 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10150 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10154 case bpf_ctx_range_till(struct bpf_sk_lookup,
10155 local_ip6[0], local_ip6[3]): {
10156 #if IS_ENABLED(CONFIG_IPV6)
10159 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
10160 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10161 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10162 offsetof(struct bpf_sk_lookup_kern, v6.daddr));
10163 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10164 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10166 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10170 case offsetof(struct bpf_sk_lookup, remote_port):
10171 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10172 bpf_target_off(struct bpf_sk_lookup_kern,
10173 sport, 2, target_size));
10176 case offsetof(struct bpf_sk_lookup, local_port):
10177 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10178 bpf_target_off(struct bpf_sk_lookup_kern,
10179 dport, 2, target_size));
10183 return insn - insn_buf;
10186 const struct bpf_prog_ops sk_lookup_prog_ops = {
10189 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
10190 .get_func_proto = sk_lookup_func_proto,
10191 .is_valid_access = sk_lookup_is_valid_access,
10192 .convert_ctx_access = sk_lookup_convert_ctx_access,
10195 #endif /* CONFIG_INET */
10197 DEFINE_BPF_DISPATCHER(xdp)
10199 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
10201 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
10204 #ifdef CONFIG_DEBUG_INFO_BTF
10205 BTF_ID_LIST_GLOBAL(btf_sock_ids)
10206 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
10208 #undef BTF_SOCK_TYPE
10210 u32 btf_sock_ids[MAX_BTF_SOCK_TYPE];
10213 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
10215 /* tcp6_sock type is not generated in dwarf and hence btf,
10216 * trigger an explicit type generation here.
10218 BTF_TYPE_EMIT(struct tcp6_sock);
10219 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
10220 sk->sk_family == AF_INET6)
10221 return (unsigned long)sk;
10223 return (unsigned long)NULL;
10226 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
10227 .func = bpf_skc_to_tcp6_sock,
10229 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10230 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10231 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
10234 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
10236 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
10237 return (unsigned long)sk;
10239 return (unsigned long)NULL;
10242 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
10243 .func = bpf_skc_to_tcp_sock,
10245 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10246 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10247 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
10250 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
10253 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
10254 return (unsigned long)sk;
10257 #if IS_BUILTIN(CONFIG_IPV6)
10258 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
10259 return (unsigned long)sk;
10262 return (unsigned long)NULL;
10265 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
10266 .func = bpf_skc_to_tcp_timewait_sock,
10268 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10269 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10270 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
10273 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
10276 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10277 return (unsigned long)sk;
10280 #if IS_BUILTIN(CONFIG_IPV6)
10281 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10282 return (unsigned long)sk;
10285 return (unsigned long)NULL;
10288 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
10289 .func = bpf_skc_to_tcp_request_sock,
10291 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10292 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10293 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
10296 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
10298 /* udp6_sock type is not generated in dwarf and hence btf,
10299 * trigger an explicit type generation here.
10301 BTF_TYPE_EMIT(struct udp6_sock);
10302 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
10303 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
10304 return (unsigned long)sk;
10306 return (unsigned long)NULL;
10309 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
10310 .func = bpf_skc_to_udp6_sock,
10312 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10313 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10314 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
10317 static const struct bpf_func_proto *
10318 bpf_sk_base_func_proto(enum bpf_func_id func_id)
10320 const struct bpf_func_proto *func;
10323 case BPF_FUNC_skc_to_tcp6_sock:
10324 func = &bpf_skc_to_tcp6_sock_proto;
10326 case BPF_FUNC_skc_to_tcp_sock:
10327 func = &bpf_skc_to_tcp_sock_proto;
10329 case BPF_FUNC_skc_to_tcp_timewait_sock:
10330 func = &bpf_skc_to_tcp_timewait_sock_proto;
10332 case BPF_FUNC_skc_to_tcp_request_sock:
10333 func = &bpf_skc_to_tcp_request_sock_proto;
10335 case BPF_FUNC_skc_to_udp6_sock:
10336 func = &bpf_skc_to_udp6_sock_proto;
10339 return bpf_base_func_proto(func_id);
10342 if (!perfmon_capable())