1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Linux Socket Filter - Kernel level socket filtering
5 * Based on the design of the Berkeley Packet Filter. The new
6 * internal format has been designed by PLUMgrid:
8 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
12 * Jay Schulist <jschlst@samba.org>
13 * Alexei Starovoitov <ast@plumgrid.com>
14 * Daniel Borkmann <dborkman@redhat.com>
16 * Andi Kleen - Fix a few bad bugs and races.
17 * Kris Katterjohn - Added many additional checks in bpf_check_classic()
20 #include <linux/atomic.h>
21 #include <linux/module.h>
22 #include <linux/types.h>
24 #include <linux/fcntl.h>
25 #include <linux/socket.h>
26 #include <linux/sock_diag.h>
28 #include <linux/inet.h>
29 #include <linux/netdevice.h>
30 #include <linux/if_packet.h>
31 #include <linux/if_arp.h>
32 #include <linux/gfp.h>
33 #include <net/inet_common.h>
35 #include <net/protocol.h>
36 #include <net/netlink.h>
37 #include <linux/skbuff.h>
38 #include <linux/skmsg.h>
40 #include <net/flow_dissector.h>
41 #include <linux/errno.h>
42 #include <linux/timer.h>
43 #include <linux/uaccess.h>
44 #include <asm/unaligned.h>
45 #include <linux/filter.h>
46 #include <linux/ratelimit.h>
47 #include <linux/seccomp.h>
48 #include <linux/if_vlan.h>
49 #include <linux/bpf.h>
50 #include <linux/btf.h>
51 #include <net/sch_generic.h>
52 #include <net/cls_cgroup.h>
53 #include <net/dst_metadata.h>
55 #include <net/sock_reuseport.h>
56 #include <net/busy_poll.h>
60 #include <linux/bpf_trace.h>
61 #include <net/xdp_sock.h>
62 #include <linux/inetdevice.h>
63 #include <net/inet_hashtables.h>
64 #include <net/inet6_hashtables.h>
65 #include <net/ip_fib.h>
66 #include <net/nexthop.h>
70 #include <net/net_namespace.h>
71 #include <linux/seg6_local.h>
73 #include <net/seg6_local.h>
74 #include <net/lwtunnel.h>
75 #include <net/ipv6_stubs.h>
76 #include <net/bpf_sk_storage.h>
77 #include <net/transp_v6.h>
78 #include <linux/btf_ids.h>
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 return __bpf_try_make_writable(skb, write_len);
1869 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1871 /* Idea is the following: should the needed direct read/write
1872 * test fail during runtime, we can pull in more data and redo
1873 * again, since implicitly, we invalidate previous checks here.
1875 * Or, since we know how much we need to make read/writeable,
1876 * this can be done once at the program beginning for direct
1877 * access case. By this we overcome limitations of only current
1878 * headroom being accessible.
1880 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1883 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1884 .func = sk_skb_pull_data,
1886 .ret_type = RET_INTEGER,
1887 .arg1_type = ARG_PTR_TO_CTX,
1888 .arg2_type = ARG_ANYTHING,
1891 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1892 u64, from, u64, to, u64, flags)
1896 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1898 if (unlikely(offset > 0xffff || offset & 1))
1900 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1903 ptr = (__sum16 *)(skb->data + offset);
1904 switch (flags & BPF_F_HDR_FIELD_MASK) {
1906 if (unlikely(from != 0))
1909 csum_replace_by_diff(ptr, to);
1912 csum_replace2(ptr, from, to);
1915 csum_replace4(ptr, from, to);
1924 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1925 .func = bpf_l3_csum_replace,
1927 .ret_type = RET_INTEGER,
1928 .arg1_type = ARG_PTR_TO_CTX,
1929 .arg2_type = ARG_ANYTHING,
1930 .arg3_type = ARG_ANYTHING,
1931 .arg4_type = ARG_ANYTHING,
1932 .arg5_type = ARG_ANYTHING,
1935 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1936 u64, from, u64, to, u64, flags)
1938 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1939 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1940 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1943 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1944 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1946 if (unlikely(offset > 0xffff || offset & 1))
1948 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1951 ptr = (__sum16 *)(skb->data + offset);
1952 if (is_mmzero && !do_mforce && !*ptr)
1955 switch (flags & BPF_F_HDR_FIELD_MASK) {
1957 if (unlikely(from != 0))
1960 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1963 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1966 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1972 if (is_mmzero && !*ptr)
1973 *ptr = CSUM_MANGLED_0;
1977 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1978 .func = bpf_l4_csum_replace,
1980 .ret_type = RET_INTEGER,
1981 .arg1_type = ARG_PTR_TO_CTX,
1982 .arg2_type = ARG_ANYTHING,
1983 .arg3_type = ARG_ANYTHING,
1984 .arg4_type = ARG_ANYTHING,
1985 .arg5_type = ARG_ANYTHING,
1988 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1989 __be32 *, to, u32, to_size, __wsum, seed)
1991 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1992 u32 diff_size = from_size + to_size;
1995 /* This is quite flexible, some examples:
1997 * from_size == 0, to_size > 0, seed := csum --> pushing data
1998 * from_size > 0, to_size == 0, seed := csum --> pulling data
1999 * from_size > 0, to_size > 0, seed := 0 --> diffing data
2001 * Even for diffing, from_size and to_size don't need to be equal.
2003 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
2004 diff_size > sizeof(sp->diff)))
2007 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
2008 sp->diff[j] = ~from[i];
2009 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
2010 sp->diff[j] = to[i];
2012 return csum_partial(sp->diff, diff_size, seed);
2015 static const struct bpf_func_proto bpf_csum_diff_proto = {
2016 .func = bpf_csum_diff,
2019 .ret_type = RET_INTEGER,
2020 .arg1_type = ARG_PTR_TO_MEM_OR_NULL,
2021 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
2022 .arg3_type = ARG_PTR_TO_MEM_OR_NULL,
2023 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
2024 .arg5_type = ARG_ANYTHING,
2027 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2029 /* The interface is to be used in combination with bpf_csum_diff()
2030 * for direct packet writes. csum rotation for alignment as well
2031 * as emulating csum_sub() can be done from the eBPF program.
2033 if (skb->ip_summed == CHECKSUM_COMPLETE)
2034 return (skb->csum = csum_add(skb->csum, csum));
2039 static const struct bpf_func_proto bpf_csum_update_proto = {
2040 .func = bpf_csum_update,
2042 .ret_type = RET_INTEGER,
2043 .arg1_type = ARG_PTR_TO_CTX,
2044 .arg2_type = ARG_ANYTHING,
2047 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
2049 /* The interface is to be used in combination with bpf_skb_adjust_room()
2050 * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET
2051 * is passed as flags, for example.
2054 case BPF_CSUM_LEVEL_INC:
2055 __skb_incr_checksum_unnecessary(skb);
2057 case BPF_CSUM_LEVEL_DEC:
2058 __skb_decr_checksum_unnecessary(skb);
2060 case BPF_CSUM_LEVEL_RESET:
2061 __skb_reset_checksum_unnecessary(skb);
2063 case BPF_CSUM_LEVEL_QUERY:
2064 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2065 skb->csum_level : -EACCES;
2073 static const struct bpf_func_proto bpf_csum_level_proto = {
2074 .func = bpf_csum_level,
2076 .ret_type = RET_INTEGER,
2077 .arg1_type = ARG_PTR_TO_CTX,
2078 .arg2_type = ARG_ANYTHING,
2081 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2083 return dev_forward_skb_nomtu(dev, skb);
2086 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2087 struct sk_buff *skb)
2089 int ret = ____dev_forward_skb(dev, skb, false);
2093 ret = netif_rx(skb);
2099 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2103 if (dev_xmit_recursion()) {
2104 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2112 dev_xmit_recursion_inc();
2113 ret = dev_queue_xmit(skb);
2114 dev_xmit_recursion_dec();
2119 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2122 unsigned int mlen = skb_network_offset(skb);
2125 __skb_pull(skb, mlen);
2127 /* At ingress, the mac header has already been pulled once.
2128 * At egress, skb_pospull_rcsum has to be done in case that
2129 * the skb is originated from ingress (i.e. a forwarded skb)
2130 * to ensure that rcsum starts at net header.
2132 if (!skb_at_tc_ingress(skb))
2133 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2135 skb_pop_mac_header(skb);
2136 skb_reset_mac_len(skb);
2137 return flags & BPF_F_INGRESS ?
2138 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2141 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2144 /* Verify that a link layer header is carried */
2145 if (unlikely(skb->mac_header >= skb->network_header)) {
2150 bpf_push_mac_rcsum(skb);
2151 return flags & BPF_F_INGRESS ?
2152 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2155 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2158 if (dev_is_mac_header_xmit(dev))
2159 return __bpf_redirect_common(skb, dev, flags);
2161 return __bpf_redirect_no_mac(skb, dev, flags);
2164 #if IS_ENABLED(CONFIG_IPV6)
2165 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2166 struct net_device *dev, struct bpf_nh_params *nh)
2168 u32 hh_len = LL_RESERVED_SPACE(dev);
2169 const struct in6_addr *nexthop;
2170 struct dst_entry *dst = NULL;
2171 struct neighbour *neigh;
2173 if (dev_xmit_recursion()) {
2174 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2181 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2182 struct sk_buff *skb2;
2184 skb2 = skb_realloc_headroom(skb, hh_len);
2185 if (unlikely(!skb2)) {
2190 skb_set_owner_w(skb2, skb->sk);
2198 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2199 &ipv6_hdr(skb)->daddr);
2201 nexthop = &nh->ipv6_nh;
2203 neigh = ip_neigh_gw6(dev, nexthop);
2204 if (likely(!IS_ERR(neigh))) {
2207 sock_confirm_neigh(skb, neigh);
2208 dev_xmit_recursion_inc();
2209 ret = neigh_output(neigh, skb, false);
2210 dev_xmit_recursion_dec();
2211 rcu_read_unlock_bh();
2214 rcu_read_unlock_bh();
2216 IP6_INC_STATS(dev_net(dst->dev),
2217 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2223 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2224 struct bpf_nh_params *nh)
2226 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2227 struct net *net = dev_net(dev);
2228 int err, ret = NET_XMIT_DROP;
2231 struct dst_entry *dst;
2232 struct flowi6 fl6 = {
2233 .flowi6_flags = FLOWI_FLAG_ANYSRC,
2234 .flowi6_mark = skb->mark,
2235 .flowlabel = ip6_flowinfo(ip6h),
2236 .flowi6_oif = dev->ifindex,
2237 .flowi6_proto = ip6h->nexthdr,
2238 .daddr = ip6h->daddr,
2239 .saddr = ip6h->saddr,
2242 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2246 skb_dst_set(skb, dst);
2247 } else if (nh->nh_family != AF_INET6) {
2251 err = bpf_out_neigh_v6(net, skb, dev, nh);
2252 if (unlikely(net_xmit_eval(err)))
2253 dev->stats.tx_errors++;
2255 ret = NET_XMIT_SUCCESS;
2258 dev->stats.tx_errors++;
2264 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2265 struct bpf_nh_params *nh)
2268 return NET_XMIT_DROP;
2270 #endif /* CONFIG_IPV6 */
2272 #if IS_ENABLED(CONFIG_INET)
2273 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2274 struct net_device *dev, struct bpf_nh_params *nh)
2276 u32 hh_len = LL_RESERVED_SPACE(dev);
2277 struct neighbour *neigh;
2278 bool is_v6gw = false;
2280 if (dev_xmit_recursion()) {
2281 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2288 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2289 struct sk_buff *skb2;
2291 skb2 = skb_realloc_headroom(skb, hh_len);
2292 if (unlikely(!skb2)) {
2297 skb_set_owner_w(skb2, skb->sk);
2304 struct dst_entry *dst = skb_dst(skb);
2305 struct rtable *rt = container_of(dst, struct rtable, dst);
2307 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2308 } else if (nh->nh_family == AF_INET6) {
2309 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2311 } else if (nh->nh_family == AF_INET) {
2312 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2314 rcu_read_unlock_bh();
2318 if (likely(!IS_ERR(neigh))) {
2321 sock_confirm_neigh(skb, neigh);
2322 dev_xmit_recursion_inc();
2323 ret = neigh_output(neigh, skb, is_v6gw);
2324 dev_xmit_recursion_dec();
2325 rcu_read_unlock_bh();
2328 rcu_read_unlock_bh();
2334 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2335 struct bpf_nh_params *nh)
2337 const struct iphdr *ip4h = ip_hdr(skb);
2338 struct net *net = dev_net(dev);
2339 int err, ret = NET_XMIT_DROP;
2342 struct flowi4 fl4 = {
2343 .flowi4_flags = FLOWI_FLAG_ANYSRC,
2344 .flowi4_mark = skb->mark,
2345 .flowi4_tos = RT_TOS(ip4h->tos),
2346 .flowi4_oif = dev->ifindex,
2347 .flowi4_proto = ip4h->protocol,
2348 .daddr = ip4h->daddr,
2349 .saddr = ip4h->saddr,
2353 rt = ip_route_output_flow(net, &fl4, NULL);
2356 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2361 skb_dst_set(skb, &rt->dst);
2364 err = bpf_out_neigh_v4(net, skb, dev, nh);
2365 if (unlikely(net_xmit_eval(err)))
2366 dev->stats.tx_errors++;
2368 ret = NET_XMIT_SUCCESS;
2371 dev->stats.tx_errors++;
2377 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2378 struct bpf_nh_params *nh)
2381 return NET_XMIT_DROP;
2383 #endif /* CONFIG_INET */
2385 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2386 struct bpf_nh_params *nh)
2388 struct ethhdr *ethh = eth_hdr(skb);
2390 if (unlikely(skb->mac_header >= skb->network_header))
2392 bpf_push_mac_rcsum(skb);
2393 if (is_multicast_ether_addr(ethh->h_dest))
2396 skb_pull(skb, sizeof(*ethh));
2397 skb_unset_mac_header(skb);
2398 skb_reset_network_header(skb);
2400 if (skb->protocol == htons(ETH_P_IP))
2401 return __bpf_redirect_neigh_v4(skb, dev, nh);
2402 else if (skb->protocol == htons(ETH_P_IPV6))
2403 return __bpf_redirect_neigh_v6(skb, dev, nh);
2409 /* Internal, non-exposed redirect flags. */
2411 BPF_F_NEIGH = (1ULL << 1),
2412 BPF_F_PEER = (1ULL << 2),
2413 BPF_F_NEXTHOP = (1ULL << 3),
2414 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2417 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2419 struct net_device *dev;
2420 struct sk_buff *clone;
2423 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2426 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2430 clone = skb_clone(skb, GFP_ATOMIC);
2431 if (unlikely(!clone))
2434 /* For direct write, we need to keep the invariant that the skbs
2435 * we're dealing with need to be uncloned. Should uncloning fail
2436 * here, we need to free the just generated clone to unclone once
2439 ret = bpf_try_make_head_writable(skb);
2440 if (unlikely(ret)) {
2445 return __bpf_redirect(clone, dev, flags);
2448 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2449 .func = bpf_clone_redirect,
2451 .ret_type = RET_INTEGER,
2452 .arg1_type = ARG_PTR_TO_CTX,
2453 .arg2_type = ARG_ANYTHING,
2454 .arg3_type = ARG_ANYTHING,
2457 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2458 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2460 int skb_do_redirect(struct sk_buff *skb)
2462 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2463 struct net *net = dev_net(skb->dev);
2464 struct net_device *dev;
2465 u32 flags = ri->flags;
2467 dev = dev_get_by_index_rcu(net, ri->tgt_index);
2472 if (flags & BPF_F_PEER) {
2473 const struct net_device_ops *ops = dev->netdev_ops;
2475 if (unlikely(!ops->ndo_get_peer_dev ||
2476 !skb_at_tc_ingress(skb)))
2478 dev = ops->ndo_get_peer_dev(dev);
2479 if (unlikely(!dev ||
2480 !(dev->flags & IFF_UP) ||
2481 net_eq(net, dev_net(dev))))
2486 return flags & BPF_F_NEIGH ?
2487 __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2489 __bpf_redirect(skb, dev, flags);
2495 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2497 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2499 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2503 ri->tgt_index = ifindex;
2505 return TC_ACT_REDIRECT;
2508 static const struct bpf_func_proto bpf_redirect_proto = {
2509 .func = bpf_redirect,
2511 .ret_type = RET_INTEGER,
2512 .arg1_type = ARG_ANYTHING,
2513 .arg2_type = ARG_ANYTHING,
2516 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2518 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2520 if (unlikely(flags))
2523 ri->flags = BPF_F_PEER;
2524 ri->tgt_index = ifindex;
2526 return TC_ACT_REDIRECT;
2529 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2530 .func = bpf_redirect_peer,
2532 .ret_type = RET_INTEGER,
2533 .arg1_type = ARG_ANYTHING,
2534 .arg2_type = ARG_ANYTHING,
2537 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2538 int, plen, u64, flags)
2540 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2542 if (unlikely((plen && plen < sizeof(*params)) || flags))
2545 ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2546 ri->tgt_index = ifindex;
2548 BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2550 memcpy(&ri->nh, params, sizeof(ri->nh));
2552 return TC_ACT_REDIRECT;
2555 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2556 .func = bpf_redirect_neigh,
2558 .ret_type = RET_INTEGER,
2559 .arg1_type = ARG_ANYTHING,
2560 .arg2_type = ARG_PTR_TO_MEM_OR_NULL,
2561 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
2562 .arg4_type = ARG_ANYTHING,
2565 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2567 msg->apply_bytes = bytes;
2571 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2572 .func = bpf_msg_apply_bytes,
2574 .ret_type = RET_INTEGER,
2575 .arg1_type = ARG_PTR_TO_CTX,
2576 .arg2_type = ARG_ANYTHING,
2579 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2581 msg->cork_bytes = bytes;
2585 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2586 .func = bpf_msg_cork_bytes,
2588 .ret_type = RET_INTEGER,
2589 .arg1_type = ARG_PTR_TO_CTX,
2590 .arg2_type = ARG_ANYTHING,
2593 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2594 u32, end, u64, flags)
2596 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2597 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2598 struct scatterlist *sge;
2599 u8 *raw, *to, *from;
2602 if (unlikely(flags || end <= start))
2605 /* First find the starting scatterlist element */
2609 len = sk_msg_elem(msg, i)->length;
2610 if (start < offset + len)
2612 sk_msg_iter_var_next(i);
2613 } while (i != msg->sg.end);
2615 if (unlikely(start >= offset + len))
2619 /* The start may point into the sg element so we need to also
2620 * account for the headroom.
2622 bytes_sg_total = start - offset + bytes;
2623 if (!test_bit(i, &msg->sg.copy) && bytes_sg_total <= len)
2626 /* At this point we need to linearize multiple scatterlist
2627 * elements or a single shared page. Either way we need to
2628 * copy into a linear buffer exclusively owned by BPF. Then
2629 * place the buffer in the scatterlist and fixup the original
2630 * entries by removing the entries now in the linear buffer
2631 * and shifting the remaining entries. For now we do not try
2632 * to copy partial entries to avoid complexity of running out
2633 * of sg_entry slots. The downside is reading a single byte
2634 * will copy the entire sg entry.
2637 copy += sk_msg_elem(msg, i)->length;
2638 sk_msg_iter_var_next(i);
2639 if (bytes_sg_total <= copy)
2641 } while (i != msg->sg.end);
2644 if (unlikely(bytes_sg_total > copy))
2647 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2649 if (unlikely(!page))
2652 raw = page_address(page);
2655 sge = sk_msg_elem(msg, i);
2656 from = sg_virt(sge);
2660 memcpy(to, from, len);
2663 put_page(sg_page(sge));
2665 sk_msg_iter_var_next(i);
2666 } while (i != last_sge);
2668 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2670 /* To repair sg ring we need to shift entries. If we only
2671 * had a single entry though we can just replace it and
2672 * be done. Otherwise walk the ring and shift the entries.
2674 WARN_ON_ONCE(last_sge == first_sge);
2675 shift = last_sge > first_sge ?
2676 last_sge - first_sge - 1 :
2677 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2682 sk_msg_iter_var_next(i);
2686 if (i + shift >= NR_MSG_FRAG_IDS)
2687 move_from = i + shift - NR_MSG_FRAG_IDS;
2689 move_from = i + shift;
2690 if (move_from == msg->sg.end)
2693 msg->sg.data[i] = msg->sg.data[move_from];
2694 msg->sg.data[move_from].length = 0;
2695 msg->sg.data[move_from].page_link = 0;
2696 msg->sg.data[move_from].offset = 0;
2697 sk_msg_iter_var_next(i);
2700 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2701 msg->sg.end - shift + NR_MSG_FRAG_IDS :
2702 msg->sg.end - shift;
2704 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2705 msg->data_end = msg->data + bytes;
2709 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2710 .func = bpf_msg_pull_data,
2712 .ret_type = RET_INTEGER,
2713 .arg1_type = ARG_PTR_TO_CTX,
2714 .arg2_type = ARG_ANYTHING,
2715 .arg3_type = ARG_ANYTHING,
2716 .arg4_type = ARG_ANYTHING,
2719 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2720 u32, len, u64, flags)
2722 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2723 u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2724 u8 *raw, *to, *from;
2727 if (unlikely(flags))
2730 /* First find the starting scatterlist element */
2734 l = sk_msg_elem(msg, i)->length;
2736 if (start < offset + l)
2738 sk_msg_iter_var_next(i);
2739 } while (i != msg->sg.end);
2741 if (start >= offset + l)
2744 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2746 /* If no space available will fallback to copy, we need at
2747 * least one scatterlist elem available to push data into
2748 * when start aligns to the beginning of an element or two
2749 * when it falls inside an element. We handle the start equals
2750 * offset case because its the common case for inserting a
2753 if (!space || (space == 1 && start != offset))
2754 copy = msg->sg.data[i].length;
2756 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2757 get_order(copy + len));
2758 if (unlikely(!page))
2764 raw = page_address(page);
2766 psge = sk_msg_elem(msg, i);
2767 front = start - offset;
2768 back = psge->length - front;
2769 from = sg_virt(psge);
2772 memcpy(raw, from, front);
2776 to = raw + front + len;
2778 memcpy(to, from, back);
2781 put_page(sg_page(psge));
2782 } else if (start - offset) {
2783 psge = sk_msg_elem(msg, i);
2784 rsge = sk_msg_elem_cpy(msg, i);
2786 psge->length = start - offset;
2787 rsge.length -= psge->length;
2788 rsge.offset += start;
2790 sk_msg_iter_var_next(i);
2791 sg_unmark_end(psge);
2792 sg_unmark_end(&rsge);
2793 sk_msg_iter_next(msg, end);
2796 /* Slot(s) to place newly allocated data */
2799 /* Shift one or two slots as needed */
2801 sge = sk_msg_elem_cpy(msg, i);
2803 sk_msg_iter_var_next(i);
2804 sg_unmark_end(&sge);
2805 sk_msg_iter_next(msg, end);
2807 nsge = sk_msg_elem_cpy(msg, i);
2809 sk_msg_iter_var_next(i);
2810 nnsge = sk_msg_elem_cpy(msg, i);
2813 while (i != msg->sg.end) {
2814 msg->sg.data[i] = sge;
2816 sk_msg_iter_var_next(i);
2819 nnsge = sk_msg_elem_cpy(msg, i);
2821 nsge = sk_msg_elem_cpy(msg, i);
2826 /* Place newly allocated data buffer */
2827 sk_mem_charge(msg->sk, len);
2828 msg->sg.size += len;
2829 __clear_bit(new, &msg->sg.copy);
2830 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2832 get_page(sg_page(&rsge));
2833 sk_msg_iter_var_next(new);
2834 msg->sg.data[new] = rsge;
2837 sk_msg_compute_data_pointers(msg);
2841 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2842 .func = bpf_msg_push_data,
2844 .ret_type = RET_INTEGER,
2845 .arg1_type = ARG_PTR_TO_CTX,
2846 .arg2_type = ARG_ANYTHING,
2847 .arg3_type = ARG_ANYTHING,
2848 .arg4_type = ARG_ANYTHING,
2851 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2857 sk_msg_iter_var_next(i);
2858 msg->sg.data[prev] = msg->sg.data[i];
2859 } while (i != msg->sg.end);
2861 sk_msg_iter_prev(msg, end);
2864 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2866 struct scatterlist tmp, sge;
2868 sk_msg_iter_next(msg, end);
2869 sge = sk_msg_elem_cpy(msg, i);
2870 sk_msg_iter_var_next(i);
2871 tmp = sk_msg_elem_cpy(msg, i);
2873 while (i != msg->sg.end) {
2874 msg->sg.data[i] = sge;
2875 sk_msg_iter_var_next(i);
2877 tmp = sk_msg_elem_cpy(msg, i);
2881 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2882 u32, len, u64, flags)
2884 u32 i = 0, l = 0, space, offset = 0;
2885 u64 last = start + len;
2888 if (unlikely(flags))
2891 /* First find the starting scatterlist element */
2895 l = sk_msg_elem(msg, i)->length;
2897 if (start < offset + l)
2899 sk_msg_iter_var_next(i);
2900 } while (i != msg->sg.end);
2902 /* Bounds checks: start and pop must be inside message */
2903 if (start >= offset + l || last >= msg->sg.size)
2906 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2909 /* --------------| offset
2910 * -| start |-------- len -------|
2912 * |----- a ----|-------- pop -------|----- b ----|
2913 * |______________________________________________| length
2916 * a: region at front of scatter element to save
2917 * b: region at back of scatter element to save when length > A + pop
2918 * pop: region to pop from element, same as input 'pop' here will be
2919 * decremented below per iteration.
2921 * Two top-level cases to handle when start != offset, first B is non
2922 * zero and second B is zero corresponding to when a pop includes more
2925 * Then if B is non-zero AND there is no space allocate space and
2926 * compact A, B regions into page. If there is space shift ring to
2927 * the rigth free'ing the next element in ring to place B, leaving
2928 * A untouched except to reduce length.
2930 if (start != offset) {
2931 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2933 int b = sge->length - pop - a;
2935 sk_msg_iter_var_next(i);
2937 if (pop < sge->length - a) {
2940 sk_msg_shift_right(msg, i);
2941 nsge = sk_msg_elem(msg, i);
2942 get_page(sg_page(sge));
2945 b, sge->offset + pop + a);
2947 struct page *page, *orig;
2950 page = alloc_pages(__GFP_NOWARN |
2951 __GFP_COMP | GFP_ATOMIC,
2953 if (unlikely(!page))
2957 orig = sg_page(sge);
2958 from = sg_virt(sge);
2959 to = page_address(page);
2960 memcpy(to, from, a);
2961 memcpy(to + a, from + a + pop, b);
2962 sg_set_page(sge, page, a + b, 0);
2966 } else if (pop >= sge->length - a) {
2967 pop -= (sge->length - a);
2972 /* From above the current layout _must_ be as follows,
2977 * |---- pop ---|---------------- b ------------|
2978 * |____________________________________________| length
2980 * Offset and start of the current msg elem are equal because in the
2981 * previous case we handled offset != start and either consumed the
2982 * entire element and advanced to the next element OR pop == 0.
2984 * Two cases to handle here are first pop is less than the length
2985 * leaving some remainder b above. Simply adjust the element's layout
2986 * in this case. Or pop >= length of the element so that b = 0. In this
2987 * case advance to next element decrementing pop.
2990 struct scatterlist *sge = sk_msg_elem(msg, i);
2992 if (pop < sge->length) {
2998 sk_msg_shift_left(msg, i);
3000 sk_msg_iter_var_next(i);
3003 sk_mem_uncharge(msg->sk, len - pop);
3004 msg->sg.size -= (len - pop);
3005 sk_msg_compute_data_pointers(msg);
3009 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
3010 .func = bpf_msg_pop_data,
3012 .ret_type = RET_INTEGER,
3013 .arg1_type = ARG_PTR_TO_CTX,
3014 .arg2_type = ARG_ANYTHING,
3015 .arg3_type = ARG_ANYTHING,
3016 .arg4_type = ARG_ANYTHING,
3019 #ifdef CONFIG_CGROUP_NET_CLASSID
3020 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3022 return __task_get_classid(current);
3025 static const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3026 .func = bpf_get_cgroup_classid_curr,
3028 .ret_type = RET_INTEGER,
3031 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3033 struct sock *sk = skb_to_full_sk(skb);
3035 if (!sk || !sk_fullsock(sk))
3038 return sock_cgroup_classid(&sk->sk_cgrp_data);
3041 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3042 .func = bpf_skb_cgroup_classid,
3044 .ret_type = RET_INTEGER,
3045 .arg1_type = ARG_PTR_TO_CTX,
3049 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3051 return task_get_classid(skb);
3054 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3055 .func = bpf_get_cgroup_classid,
3057 .ret_type = RET_INTEGER,
3058 .arg1_type = ARG_PTR_TO_CTX,
3061 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3063 return dst_tclassid(skb);
3066 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3067 .func = bpf_get_route_realm,
3069 .ret_type = RET_INTEGER,
3070 .arg1_type = ARG_PTR_TO_CTX,
3073 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3075 /* If skb_clear_hash() was called due to mangling, we can
3076 * trigger SW recalculation here. Later access to hash
3077 * can then use the inline skb->hash via context directly
3078 * instead of calling this helper again.
3080 return skb_get_hash(skb);
3083 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3084 .func = bpf_get_hash_recalc,
3086 .ret_type = RET_INTEGER,
3087 .arg1_type = ARG_PTR_TO_CTX,
3090 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3092 /* After all direct packet write, this can be used once for
3093 * triggering a lazy recalc on next skb_get_hash() invocation.
3095 skb_clear_hash(skb);
3099 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3100 .func = bpf_set_hash_invalid,
3102 .ret_type = RET_INTEGER,
3103 .arg1_type = ARG_PTR_TO_CTX,
3106 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3108 /* Set user specified hash as L4(+), so that it gets returned
3109 * on skb_get_hash() call unless BPF prog later on triggers a
3112 __skb_set_sw_hash(skb, hash, true);
3116 static const struct bpf_func_proto bpf_set_hash_proto = {
3117 .func = bpf_set_hash,
3119 .ret_type = RET_INTEGER,
3120 .arg1_type = ARG_PTR_TO_CTX,
3121 .arg2_type = ARG_ANYTHING,
3124 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3129 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3130 vlan_proto != htons(ETH_P_8021AD)))
3131 vlan_proto = htons(ETH_P_8021Q);
3133 bpf_push_mac_rcsum(skb);
3134 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3135 bpf_pull_mac_rcsum(skb);
3137 bpf_compute_data_pointers(skb);
3141 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3142 .func = bpf_skb_vlan_push,
3144 .ret_type = RET_INTEGER,
3145 .arg1_type = ARG_PTR_TO_CTX,
3146 .arg2_type = ARG_ANYTHING,
3147 .arg3_type = ARG_ANYTHING,
3150 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3154 bpf_push_mac_rcsum(skb);
3155 ret = skb_vlan_pop(skb);
3156 bpf_pull_mac_rcsum(skb);
3158 bpf_compute_data_pointers(skb);
3162 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3163 .func = bpf_skb_vlan_pop,
3165 .ret_type = RET_INTEGER,
3166 .arg1_type = ARG_PTR_TO_CTX,
3169 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3171 /* Caller already did skb_cow() with len as headroom,
3172 * so no need to do it here.
3175 memmove(skb->data, skb->data + len, off);
3176 memset(skb->data + off, 0, len);
3178 /* No skb_postpush_rcsum(skb, skb->data + off, len)
3179 * needed here as it does not change the skb->csum
3180 * result for checksum complete when summing over
3186 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3188 /* skb_ensure_writable() is not needed here, as we're
3189 * already working on an uncloned skb.
3191 if (unlikely(!pskb_may_pull(skb, off + len)))
3194 skb_postpull_rcsum(skb, skb->data + off, len);
3195 memmove(skb->data + len, skb->data, off);
3196 __skb_pull(skb, len);
3201 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3203 bool trans_same = skb->transport_header == skb->network_header;
3206 /* There's no need for __skb_push()/__skb_pull() pair to
3207 * get to the start of the mac header as we're guaranteed
3208 * to always start from here under eBPF.
3210 ret = bpf_skb_generic_push(skb, off, len);
3212 skb->mac_header -= len;
3213 skb->network_header -= len;
3215 skb->transport_header = skb->network_header;
3221 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3223 bool trans_same = skb->transport_header == skb->network_header;
3226 /* Same here, __skb_push()/__skb_pull() pair not needed. */
3227 ret = bpf_skb_generic_pop(skb, off, len);
3229 skb->mac_header += len;
3230 skb->network_header += len;
3232 skb->transport_header = skb->network_header;
3238 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3240 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3241 u32 off = skb_mac_header_len(skb);
3244 ret = skb_cow(skb, len_diff);
3245 if (unlikely(ret < 0))
3248 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3249 if (unlikely(ret < 0))
3252 if (skb_is_gso(skb)) {
3253 struct skb_shared_info *shinfo = skb_shinfo(skb);
3255 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */
3256 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3257 shinfo->gso_type &= ~SKB_GSO_TCPV4;
3258 shinfo->gso_type |= SKB_GSO_TCPV6;
3262 skb->protocol = htons(ETH_P_IPV6);
3263 skb_clear_hash(skb);
3268 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3270 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3271 u32 off = skb_mac_header_len(skb);
3274 ret = skb_unclone(skb, GFP_ATOMIC);
3275 if (unlikely(ret < 0))
3278 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3279 if (unlikely(ret < 0))
3282 if (skb_is_gso(skb)) {
3283 struct skb_shared_info *shinfo = skb_shinfo(skb);
3285 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */
3286 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3287 shinfo->gso_type &= ~SKB_GSO_TCPV6;
3288 shinfo->gso_type |= SKB_GSO_TCPV4;
3292 skb->protocol = htons(ETH_P_IP);
3293 skb_clear_hash(skb);
3298 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3300 __be16 from_proto = skb->protocol;
3302 if (from_proto == htons(ETH_P_IP) &&
3303 to_proto == htons(ETH_P_IPV6))
3304 return bpf_skb_proto_4_to_6(skb);
3306 if (from_proto == htons(ETH_P_IPV6) &&
3307 to_proto == htons(ETH_P_IP))
3308 return bpf_skb_proto_6_to_4(skb);
3313 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3318 if (unlikely(flags))
3321 /* General idea is that this helper does the basic groundwork
3322 * needed for changing the protocol, and eBPF program fills the
3323 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3324 * and other helpers, rather than passing a raw buffer here.
3326 * The rationale is to keep this minimal and without a need to
3327 * deal with raw packet data. F.e. even if we would pass buffers
3328 * here, the program still needs to call the bpf_lX_csum_replace()
3329 * helpers anyway. Plus, this way we keep also separation of
3330 * concerns, since f.e. bpf_skb_store_bytes() should only take
3333 * Currently, additional options and extension header space are
3334 * not supported, but flags register is reserved so we can adapt
3335 * that. For offloads, we mark packet as dodgy, so that headers
3336 * need to be verified first.
3338 ret = bpf_skb_proto_xlat(skb, proto);
3339 bpf_compute_data_pointers(skb);
3343 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3344 .func = bpf_skb_change_proto,
3346 .ret_type = RET_INTEGER,
3347 .arg1_type = ARG_PTR_TO_CTX,
3348 .arg2_type = ARG_ANYTHING,
3349 .arg3_type = ARG_ANYTHING,
3352 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3354 /* We only allow a restricted subset to be changed for now. */
3355 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3356 !skb_pkt_type_ok(pkt_type)))
3359 skb->pkt_type = pkt_type;
3363 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3364 .func = bpf_skb_change_type,
3366 .ret_type = RET_INTEGER,
3367 .arg1_type = ARG_PTR_TO_CTX,
3368 .arg2_type = ARG_ANYTHING,
3371 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3373 switch (skb->protocol) {
3374 case htons(ETH_P_IP):
3375 return sizeof(struct iphdr);
3376 case htons(ETH_P_IPV6):
3377 return sizeof(struct ipv6hdr);
3383 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3384 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3386 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \
3387 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3388 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3389 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3390 BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \
3391 BPF_F_ADJ_ROOM_ENCAP_L2( \
3392 BPF_ADJ_ROOM_ENCAP_L2_MASK))
3394 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3397 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3398 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3399 u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3400 unsigned int gso_type = SKB_GSO_DODGY;
3403 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3404 /* udp gso_size delineates datagrams, only allow if fixed */
3405 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3406 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3410 ret = skb_cow_head(skb, len_diff);
3411 if (unlikely(ret < 0))
3415 if (skb->protocol != htons(ETH_P_IP) &&
3416 skb->protocol != htons(ETH_P_IPV6))
3419 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3420 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3423 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3424 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3427 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH &&
3428 inner_mac_len < ETH_HLEN)
3431 if (skb->encapsulation)
3434 mac_len = skb->network_header - skb->mac_header;
3435 inner_net = skb->network_header;
3436 if (inner_mac_len > len_diff)
3438 inner_trans = skb->transport_header;
3441 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3442 if (unlikely(ret < 0))
3446 skb->inner_mac_header = inner_net - inner_mac_len;
3447 skb->inner_network_header = inner_net;
3448 skb->inner_transport_header = inner_trans;
3450 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH)
3451 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
3453 skb_set_inner_protocol(skb, skb->protocol);
3455 skb->encapsulation = 1;
3456 skb_set_network_header(skb, mac_len);
3458 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3459 gso_type |= SKB_GSO_UDP_TUNNEL;
3460 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3461 gso_type |= SKB_GSO_GRE;
3462 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3463 gso_type |= SKB_GSO_IPXIP6;
3464 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3465 gso_type |= SKB_GSO_IPXIP4;
3467 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3468 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3469 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3470 sizeof(struct ipv6hdr) :
3471 sizeof(struct iphdr);
3473 skb_set_transport_header(skb, mac_len + nh_len);
3476 /* Match skb->protocol to new outer l3 protocol */
3477 if (skb->protocol == htons(ETH_P_IP) &&
3478 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3479 skb->protocol = htons(ETH_P_IPV6);
3480 else if (skb->protocol == htons(ETH_P_IPV6) &&
3481 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3482 skb->protocol = htons(ETH_P_IP);
3485 if (skb_is_gso(skb)) {
3486 struct skb_shared_info *shinfo = skb_shinfo(skb);
3488 /* Due to header grow, MSS needs to be downgraded. */
3489 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3490 skb_decrease_gso_size(shinfo, len_diff);
3492 /* Header must be checked, and gso_segs recomputed. */
3493 shinfo->gso_type |= gso_type;
3494 shinfo->gso_segs = 0;
3500 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3505 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3506 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3509 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3510 /* udp gso_size delineates datagrams, only allow if fixed */
3511 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3512 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3516 ret = skb_unclone(skb, GFP_ATOMIC);
3517 if (unlikely(ret < 0))
3520 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3521 if (unlikely(ret < 0))
3524 if (skb_is_gso(skb)) {
3525 struct skb_shared_info *shinfo = skb_shinfo(skb);
3527 /* Due to header shrink, MSS can be upgraded. */
3528 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3529 skb_increase_gso_size(shinfo, len_diff);
3531 /* Header must be checked, and gso_segs recomputed. */
3532 shinfo->gso_type |= SKB_GSO_DODGY;
3533 shinfo->gso_segs = 0;
3539 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3541 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3542 u32, mode, u64, flags)
3544 u32 len_diff_abs = abs(len_diff);
3545 bool shrink = len_diff < 0;
3548 if (unlikely(flags || mode))
3550 if (unlikely(len_diff_abs > 0xfffU))
3554 ret = skb_cow(skb, len_diff);
3555 if (unlikely(ret < 0))
3557 __skb_push(skb, len_diff_abs);
3558 memset(skb->data, 0, len_diff_abs);
3560 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3562 __skb_pull(skb, len_diff_abs);
3564 if (tls_sw_has_ctx_rx(skb->sk)) {
3565 struct strp_msg *rxm = strp_msg(skb);
3567 rxm->full_len += len_diff;
3572 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3573 .func = sk_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 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3583 u32, mode, u64, flags)
3585 u32 len_cur, len_diff_abs = abs(len_diff);
3586 u32 len_min = bpf_skb_net_base_len(skb);
3587 u32 len_max = BPF_SKB_MAX_LEN;
3588 __be16 proto = skb->protocol;
3589 bool shrink = len_diff < 0;
3593 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3594 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3596 if (unlikely(len_diff_abs > 0xfffU))
3598 if (unlikely(proto != htons(ETH_P_IP) &&
3599 proto != htons(ETH_P_IPV6)))
3602 off = skb_mac_header_len(skb);
3604 case BPF_ADJ_ROOM_NET:
3605 off += bpf_skb_net_base_len(skb);
3607 case BPF_ADJ_ROOM_MAC:
3613 len_cur = skb->len - skb_network_offset(skb);
3614 if ((shrink && (len_diff_abs >= len_cur ||
3615 len_cur - len_diff_abs < len_min)) ||
3616 (!shrink && (skb->len + len_diff_abs > len_max &&
3620 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3621 bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3622 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3623 __skb_reset_checksum_unnecessary(skb);
3625 bpf_compute_data_pointers(skb);
3629 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3630 .func = bpf_skb_adjust_room,
3632 .ret_type = RET_INTEGER,
3633 .arg1_type = ARG_PTR_TO_CTX,
3634 .arg2_type = ARG_ANYTHING,
3635 .arg3_type = ARG_ANYTHING,
3636 .arg4_type = ARG_ANYTHING,
3639 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3641 u32 min_len = skb_network_offset(skb);
3643 if (skb_transport_header_was_set(skb))
3644 min_len = skb_transport_offset(skb);
3645 if (skb->ip_summed == CHECKSUM_PARTIAL)
3646 min_len = skb_checksum_start_offset(skb) +
3647 skb->csum_offset + sizeof(__sum16);
3651 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3653 unsigned int old_len = skb->len;
3656 ret = __skb_grow_rcsum(skb, new_len);
3658 memset(skb->data + old_len, 0, new_len - old_len);
3662 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3664 return __skb_trim_rcsum(skb, new_len);
3667 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3670 u32 max_len = BPF_SKB_MAX_LEN;
3671 u32 min_len = __bpf_skb_min_len(skb);
3674 if (unlikely(flags || new_len > max_len || new_len < min_len))
3676 if (skb->encapsulation)
3679 /* The basic idea of this helper is that it's performing the
3680 * needed work to either grow or trim an skb, and eBPF program
3681 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3682 * bpf_lX_csum_replace() and others rather than passing a raw
3683 * buffer here. This one is a slow path helper and intended
3684 * for replies with control messages.
3686 * Like in bpf_skb_change_proto(), we want to keep this rather
3687 * minimal and without protocol specifics so that we are able
3688 * to separate concerns as in bpf_skb_store_bytes() should only
3689 * be the one responsible for writing buffers.
3691 * It's really expected to be a slow path operation here for
3692 * control message replies, so we're implicitly linearizing,
3693 * uncloning and drop offloads from the skb by this.
3695 ret = __bpf_try_make_writable(skb, skb->len);
3697 if (new_len > skb->len)
3698 ret = bpf_skb_grow_rcsum(skb, new_len);
3699 else if (new_len < skb->len)
3700 ret = bpf_skb_trim_rcsum(skb, new_len);
3701 if (!ret && skb_is_gso(skb))
3707 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3710 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3712 bpf_compute_data_pointers(skb);
3716 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3717 .func = bpf_skb_change_tail,
3719 .ret_type = RET_INTEGER,
3720 .arg1_type = ARG_PTR_TO_CTX,
3721 .arg2_type = ARG_ANYTHING,
3722 .arg3_type = ARG_ANYTHING,
3725 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3728 return __bpf_skb_change_tail(skb, new_len, flags);
3731 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3732 .func = sk_skb_change_tail,
3734 .ret_type = RET_INTEGER,
3735 .arg1_type = ARG_PTR_TO_CTX,
3736 .arg2_type = ARG_ANYTHING,
3737 .arg3_type = ARG_ANYTHING,
3740 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3743 u32 max_len = BPF_SKB_MAX_LEN;
3744 u32 new_len = skb->len + head_room;
3747 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3748 new_len < skb->len))
3751 ret = skb_cow(skb, head_room);
3753 /* Idea for this helper is that we currently only
3754 * allow to expand on mac header. This means that
3755 * skb->protocol network header, etc, stay as is.
3756 * Compared to bpf_skb_change_tail(), we're more
3757 * flexible due to not needing to linearize or
3758 * reset GSO. Intention for this helper is to be
3759 * used by an L3 skb that needs to push mac header
3760 * for redirection into L2 device.
3762 __skb_push(skb, head_room);
3763 memset(skb->data, 0, head_room);
3764 skb_reset_mac_header(skb);
3765 skb_reset_mac_len(skb);
3771 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3774 int ret = __bpf_skb_change_head(skb, head_room, flags);
3776 bpf_compute_data_pointers(skb);
3780 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3781 .func = bpf_skb_change_head,
3783 .ret_type = RET_INTEGER,
3784 .arg1_type = ARG_PTR_TO_CTX,
3785 .arg2_type = ARG_ANYTHING,
3786 .arg3_type = ARG_ANYTHING,
3789 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3792 return __bpf_skb_change_head(skb, head_room, flags);
3795 static const struct bpf_func_proto sk_skb_change_head_proto = {
3796 .func = sk_skb_change_head,
3798 .ret_type = RET_INTEGER,
3799 .arg1_type = ARG_PTR_TO_CTX,
3800 .arg2_type = ARG_ANYTHING,
3801 .arg3_type = ARG_ANYTHING,
3803 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3805 return xdp_data_meta_unsupported(xdp) ? 0 :
3806 xdp->data - xdp->data_meta;
3809 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3811 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3812 unsigned long metalen = xdp_get_metalen(xdp);
3813 void *data_start = xdp_frame_end + metalen;
3814 void *data = xdp->data + offset;
3816 if (unlikely(data < data_start ||
3817 data > xdp->data_end - ETH_HLEN))
3821 memmove(xdp->data_meta + offset,
3822 xdp->data_meta, metalen);
3823 xdp->data_meta += offset;
3829 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3830 .func = bpf_xdp_adjust_head,
3832 .ret_type = RET_INTEGER,
3833 .arg1_type = ARG_PTR_TO_CTX,
3834 .arg2_type = ARG_ANYTHING,
3837 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3839 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
3840 void *data_end = xdp->data_end + offset;
3842 /* Notice that xdp_data_hard_end have reserved some tailroom */
3843 if (unlikely(data_end > data_hard_end))
3846 /* ALL drivers MUST init xdp->frame_sz, chicken check below */
3847 if (unlikely(xdp->frame_sz > PAGE_SIZE)) {
3848 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz);
3852 if (unlikely(data_end < xdp->data + ETH_HLEN))
3855 /* Clear memory area on grow, can contain uninit kernel memory */
3857 memset(xdp->data_end, 0, offset);
3859 xdp->data_end = data_end;
3864 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3865 .func = bpf_xdp_adjust_tail,
3867 .ret_type = RET_INTEGER,
3868 .arg1_type = ARG_PTR_TO_CTX,
3869 .arg2_type = ARG_ANYTHING,
3872 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3874 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3875 void *meta = xdp->data_meta + offset;
3876 unsigned long metalen = xdp->data - meta;
3878 if (xdp_data_meta_unsupported(xdp))
3880 if (unlikely(meta < xdp_frame_end ||
3883 if (unlikely((metalen & (sizeof(__u32) - 1)) ||
3887 xdp->data_meta = meta;
3892 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3893 .func = bpf_xdp_adjust_meta,
3895 .ret_type = RET_INTEGER,
3896 .arg1_type = ARG_PTR_TO_CTX,
3897 .arg2_type = ARG_ANYTHING,
3900 /* XDP_REDIRECT works by a three-step process, implemented in the functions
3903 * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target
3904 * of the redirect and store it (along with some other metadata) in a per-CPU
3905 * struct bpf_redirect_info.
3907 * 2. When the program returns the XDP_REDIRECT return code, the driver will
3908 * call xdp_do_redirect() which will use the information in struct
3909 * bpf_redirect_info to actually enqueue the frame into a map type-specific
3910 * bulk queue structure.
3912 * 3. Before exiting its NAPI poll loop, the driver will call xdp_do_flush(),
3913 * which will flush all the different bulk queues, thus completing the
3916 * Pointers to the map entries will be kept around for this whole sequence of
3917 * steps, protected by RCU. However, there is no top-level rcu_read_lock() in
3918 * the core code; instead, the RCU protection relies on everything happening
3919 * inside a single NAPI poll sequence, which means it's between a pair of calls
3920 * to local_bh_disable()/local_bh_enable().
3922 * The map entries are marked as __rcu and the map code makes sure to
3923 * dereference those pointers with rcu_dereference_check() in a way that works
3924 * for both sections that to hold an rcu_read_lock() and sections that are
3925 * called from NAPI without a separate rcu_read_lock(). The code below does not
3926 * use RCU annotations, but relies on those in the map code.
3928 void xdp_do_flush(void)
3934 EXPORT_SYMBOL_GPL(xdp_do_flush);
3936 void bpf_clear_redirect_map(struct bpf_map *map)
3938 struct bpf_redirect_info *ri;
3941 for_each_possible_cpu(cpu) {
3942 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3943 /* Avoid polluting remote cacheline due to writes if
3944 * not needed. Once we pass this test, we need the
3945 * cmpxchg() to make sure it hasn't been changed in
3946 * the meantime by remote CPU.
3948 if (unlikely(READ_ONCE(ri->map) == map))
3949 cmpxchg(&ri->map, map, NULL);
3953 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3954 struct bpf_prog *xdp_prog)
3956 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3957 enum bpf_map_type map_type = ri->map_type;
3958 void *fwd = ri->tgt_value;
3959 u32 map_id = ri->map_id;
3960 struct bpf_map *map;
3963 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
3964 ri->map_type = BPF_MAP_TYPE_UNSPEC;
3967 case BPF_MAP_TYPE_DEVMAP:
3969 case BPF_MAP_TYPE_DEVMAP_HASH:
3970 map = READ_ONCE(ri->map);
3971 if (unlikely(map)) {
3972 WRITE_ONCE(ri->map, NULL);
3973 err = dev_map_enqueue_multi(xdp, dev, map,
3974 ri->flags & BPF_F_EXCLUDE_INGRESS);
3976 err = dev_map_enqueue(fwd, xdp, dev);
3979 case BPF_MAP_TYPE_CPUMAP:
3980 err = cpu_map_enqueue(fwd, xdp, dev);
3982 case BPF_MAP_TYPE_XSKMAP:
3983 err = __xsk_map_redirect(fwd, xdp);
3985 case BPF_MAP_TYPE_UNSPEC:
3986 if (map_id == INT_MAX) {
3987 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
3988 if (unlikely(!fwd)) {
3992 err = dev_xdp_enqueue(fwd, xdp, dev);
4003 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4006 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4009 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4011 static int xdp_do_generic_redirect_map(struct net_device *dev,
4012 struct sk_buff *skb,
4013 struct xdp_buff *xdp,
4014 struct bpf_prog *xdp_prog,
4016 enum bpf_map_type map_type, u32 map_id)
4018 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4019 struct bpf_map *map;
4023 case BPF_MAP_TYPE_DEVMAP:
4025 case BPF_MAP_TYPE_DEVMAP_HASH:
4026 map = READ_ONCE(ri->map);
4027 if (unlikely(map)) {
4028 WRITE_ONCE(ri->map, NULL);
4029 err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4030 ri->flags & BPF_F_EXCLUDE_INGRESS);
4032 err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4037 case BPF_MAP_TYPE_XSKMAP:
4038 err = xsk_generic_rcv(fwd, xdp);
4044 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
4049 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4052 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4056 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4057 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4059 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4060 enum bpf_map_type map_type = ri->map_type;
4061 void *fwd = ri->tgt_value;
4062 u32 map_id = ri->map_id;
4065 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4066 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4068 if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4069 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4070 if (unlikely(!fwd)) {
4075 err = xdp_ok_fwd_dev(fwd, skb->len);
4080 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4081 generic_xdp_tx(skb, xdp_prog);
4085 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id);
4087 _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4091 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4093 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4095 if (unlikely(flags))
4098 /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4099 * by map_idr) is used for ifindex based XDP redirect.
4101 ri->tgt_index = ifindex;
4102 ri->map_id = INT_MAX;
4103 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4105 return XDP_REDIRECT;
4108 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4109 .func = bpf_xdp_redirect,
4111 .ret_type = RET_INTEGER,
4112 .arg1_type = ARG_ANYTHING,
4113 .arg2_type = ARG_ANYTHING,
4116 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
4119 return map->ops->map_redirect(map, ifindex, flags);
4122 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4123 .func = bpf_xdp_redirect_map,
4125 .ret_type = RET_INTEGER,
4126 .arg1_type = ARG_CONST_MAP_PTR,
4127 .arg2_type = ARG_ANYTHING,
4128 .arg3_type = ARG_ANYTHING,
4131 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4132 unsigned long off, unsigned long len)
4134 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4138 if (ptr != dst_buff)
4139 memcpy(dst_buff, ptr, len);
4144 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4145 u64, flags, void *, meta, u64, meta_size)
4147 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4149 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4151 if (unlikely(!skb || skb_size > skb->len))
4154 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4158 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4159 .func = bpf_skb_event_output,
4161 .ret_type = RET_INTEGER,
4162 .arg1_type = ARG_PTR_TO_CTX,
4163 .arg2_type = ARG_CONST_MAP_PTR,
4164 .arg3_type = ARG_ANYTHING,
4165 .arg4_type = ARG_PTR_TO_MEM,
4166 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4169 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4171 const struct bpf_func_proto bpf_skb_output_proto = {
4172 .func = bpf_skb_event_output,
4174 .ret_type = RET_INTEGER,
4175 .arg1_type = ARG_PTR_TO_BTF_ID,
4176 .arg1_btf_id = &bpf_skb_output_btf_ids[0],
4177 .arg2_type = ARG_CONST_MAP_PTR,
4178 .arg3_type = ARG_ANYTHING,
4179 .arg4_type = ARG_PTR_TO_MEM,
4180 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4183 static unsigned short bpf_tunnel_key_af(u64 flags)
4185 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4188 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4189 u32, size, u64, flags)
4191 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4192 u8 compat[sizeof(struct bpf_tunnel_key)];
4196 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
4200 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4204 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4207 case offsetof(struct bpf_tunnel_key, tunnel_label):
4208 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4210 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4211 /* Fixup deprecated structure layouts here, so we have
4212 * a common path later on.
4214 if (ip_tunnel_info_af(info) != AF_INET)
4217 to = (struct bpf_tunnel_key *)compat;
4224 to->tunnel_id = be64_to_cpu(info->key.tun_id);
4225 to->tunnel_tos = info->key.tos;
4226 to->tunnel_ttl = info->key.ttl;
4229 if (flags & BPF_F_TUNINFO_IPV6) {
4230 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4231 sizeof(to->remote_ipv6));
4232 to->tunnel_label = be32_to_cpu(info->key.label);
4234 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4235 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4236 to->tunnel_label = 0;
4239 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4240 memcpy(to_orig, to, size);
4244 memset(to_orig, 0, size);
4248 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4249 .func = bpf_skb_get_tunnel_key,
4251 .ret_type = RET_INTEGER,
4252 .arg1_type = ARG_PTR_TO_CTX,
4253 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4254 .arg3_type = ARG_CONST_SIZE,
4255 .arg4_type = ARG_ANYTHING,
4258 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4260 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4263 if (unlikely(!info ||
4264 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4268 if (unlikely(size < info->options_len)) {
4273 ip_tunnel_info_opts_get(to, info);
4274 if (size > info->options_len)
4275 memset(to + info->options_len, 0, size - info->options_len);
4277 return info->options_len;
4279 memset(to, 0, size);
4283 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4284 .func = bpf_skb_get_tunnel_opt,
4286 .ret_type = RET_INTEGER,
4287 .arg1_type = ARG_PTR_TO_CTX,
4288 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4289 .arg3_type = ARG_CONST_SIZE,
4292 static struct metadata_dst __percpu *md_dst;
4294 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4295 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4297 struct metadata_dst *md = this_cpu_ptr(md_dst);
4298 u8 compat[sizeof(struct bpf_tunnel_key)];
4299 struct ip_tunnel_info *info;
4301 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4302 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
4304 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4306 case offsetof(struct bpf_tunnel_key, tunnel_label):
4307 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4308 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4309 /* Fixup deprecated structure layouts here, so we have
4310 * a common path later on.
4312 memcpy(compat, from, size);
4313 memset(compat + size, 0, sizeof(compat) - size);
4314 from = (const struct bpf_tunnel_key *) compat;
4320 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4325 dst_hold((struct dst_entry *) md);
4326 skb_dst_set(skb, (struct dst_entry *) md);
4328 info = &md->u.tun_info;
4329 memset(info, 0, sizeof(*info));
4330 info->mode = IP_TUNNEL_INFO_TX;
4332 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4333 if (flags & BPF_F_DONT_FRAGMENT)
4334 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4335 if (flags & BPF_F_ZERO_CSUM_TX)
4336 info->key.tun_flags &= ~TUNNEL_CSUM;
4337 if (flags & BPF_F_SEQ_NUMBER)
4338 info->key.tun_flags |= TUNNEL_SEQ;
4340 info->key.tun_id = cpu_to_be64(from->tunnel_id);
4341 info->key.tos = from->tunnel_tos;
4342 info->key.ttl = from->tunnel_ttl;
4344 if (flags & BPF_F_TUNINFO_IPV6) {
4345 info->mode |= IP_TUNNEL_INFO_IPV6;
4346 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4347 sizeof(from->remote_ipv6));
4348 info->key.label = cpu_to_be32(from->tunnel_label) &
4349 IPV6_FLOWLABEL_MASK;
4351 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4357 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4358 .func = bpf_skb_set_tunnel_key,
4360 .ret_type = RET_INTEGER,
4361 .arg1_type = ARG_PTR_TO_CTX,
4362 .arg2_type = ARG_PTR_TO_MEM,
4363 .arg3_type = ARG_CONST_SIZE,
4364 .arg4_type = ARG_ANYTHING,
4367 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4368 const u8 *, from, u32, size)
4370 struct ip_tunnel_info *info = skb_tunnel_info(skb);
4371 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4373 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4375 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4378 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4383 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4384 .func = bpf_skb_set_tunnel_opt,
4386 .ret_type = RET_INTEGER,
4387 .arg1_type = ARG_PTR_TO_CTX,
4388 .arg2_type = ARG_PTR_TO_MEM,
4389 .arg3_type = ARG_CONST_SIZE,
4392 static const struct bpf_func_proto *
4393 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4396 struct metadata_dst __percpu *tmp;
4398 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4403 if (cmpxchg(&md_dst, NULL, tmp))
4404 metadata_dst_free_percpu(tmp);
4408 case BPF_FUNC_skb_set_tunnel_key:
4409 return &bpf_skb_set_tunnel_key_proto;
4410 case BPF_FUNC_skb_set_tunnel_opt:
4411 return &bpf_skb_set_tunnel_opt_proto;
4417 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4420 struct bpf_array *array = container_of(map, struct bpf_array, map);
4421 struct cgroup *cgrp;
4424 sk = skb_to_full_sk(skb);
4425 if (!sk || !sk_fullsock(sk))
4427 if (unlikely(idx >= array->map.max_entries))
4430 cgrp = READ_ONCE(array->ptrs[idx]);
4431 if (unlikely(!cgrp))
4434 return sk_under_cgroup_hierarchy(sk, cgrp);
4437 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4438 .func = bpf_skb_under_cgroup,
4440 .ret_type = RET_INTEGER,
4441 .arg1_type = ARG_PTR_TO_CTX,
4442 .arg2_type = ARG_CONST_MAP_PTR,
4443 .arg3_type = ARG_ANYTHING,
4446 #ifdef CONFIG_SOCK_CGROUP_DATA
4447 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4449 struct cgroup *cgrp;
4451 sk = sk_to_full_sk(sk);
4452 if (!sk || !sk_fullsock(sk))
4455 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4456 return cgroup_id(cgrp);
4459 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4461 return __bpf_sk_cgroup_id(skb->sk);
4464 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4465 .func = bpf_skb_cgroup_id,
4467 .ret_type = RET_INTEGER,
4468 .arg1_type = ARG_PTR_TO_CTX,
4471 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4474 struct cgroup *ancestor;
4475 struct cgroup *cgrp;
4477 sk = sk_to_full_sk(sk);
4478 if (!sk || !sk_fullsock(sk))
4481 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4482 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4486 return cgroup_id(ancestor);
4489 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4492 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4495 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4496 .func = bpf_skb_ancestor_cgroup_id,
4498 .ret_type = RET_INTEGER,
4499 .arg1_type = ARG_PTR_TO_CTX,
4500 .arg2_type = ARG_ANYTHING,
4503 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4505 return __bpf_sk_cgroup_id(sk);
4508 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4509 .func = bpf_sk_cgroup_id,
4511 .ret_type = RET_INTEGER,
4512 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4515 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4517 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4520 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4521 .func = bpf_sk_ancestor_cgroup_id,
4523 .ret_type = RET_INTEGER,
4524 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4525 .arg2_type = ARG_ANYTHING,
4529 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4530 unsigned long off, unsigned long len)
4532 memcpy(dst_buff, src_buff + off, len);
4536 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4537 u64, flags, void *, meta, u64, meta_size)
4539 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4541 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4543 if (unlikely(!xdp ||
4544 xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4547 return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4548 xdp_size, bpf_xdp_copy);
4551 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4552 .func = bpf_xdp_event_output,
4554 .ret_type = RET_INTEGER,
4555 .arg1_type = ARG_PTR_TO_CTX,
4556 .arg2_type = ARG_CONST_MAP_PTR,
4557 .arg3_type = ARG_ANYTHING,
4558 .arg4_type = ARG_PTR_TO_MEM,
4559 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4562 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4564 const struct bpf_func_proto bpf_xdp_output_proto = {
4565 .func = bpf_xdp_event_output,
4567 .ret_type = RET_INTEGER,
4568 .arg1_type = ARG_PTR_TO_BTF_ID,
4569 .arg1_btf_id = &bpf_xdp_output_btf_ids[0],
4570 .arg2_type = ARG_CONST_MAP_PTR,
4571 .arg3_type = ARG_ANYTHING,
4572 .arg4_type = ARG_PTR_TO_MEM,
4573 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4576 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4578 return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4581 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4582 .func = bpf_get_socket_cookie,
4584 .ret_type = RET_INTEGER,
4585 .arg1_type = ARG_PTR_TO_CTX,
4588 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4590 return __sock_gen_cookie(ctx->sk);
4593 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4594 .func = bpf_get_socket_cookie_sock_addr,
4596 .ret_type = RET_INTEGER,
4597 .arg1_type = ARG_PTR_TO_CTX,
4600 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4602 return __sock_gen_cookie(ctx);
4605 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4606 .func = bpf_get_socket_cookie_sock,
4608 .ret_type = RET_INTEGER,
4609 .arg1_type = ARG_PTR_TO_CTX,
4612 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
4614 return sk ? sock_gen_cookie(sk) : 0;
4617 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
4618 .func = bpf_get_socket_ptr_cookie,
4620 .ret_type = RET_INTEGER,
4621 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4624 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4626 return __sock_gen_cookie(ctx->sk);
4629 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4630 .func = bpf_get_socket_cookie_sock_ops,
4632 .ret_type = RET_INTEGER,
4633 .arg1_type = ARG_PTR_TO_CTX,
4636 static u64 __bpf_get_netns_cookie(struct sock *sk)
4638 const struct net *net = sk ? sock_net(sk) : &init_net;
4640 return net->net_cookie;
4643 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
4645 return __bpf_get_netns_cookie(ctx);
4648 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
4649 .func = bpf_get_netns_cookie_sock,
4651 .ret_type = RET_INTEGER,
4652 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4655 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4657 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4660 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
4661 .func = bpf_get_netns_cookie_sock_addr,
4663 .ret_type = RET_INTEGER,
4664 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4667 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4669 struct sock *sk = sk_to_full_sk(skb->sk);
4672 if (!sk || !sk_fullsock(sk))
4674 kuid = sock_net_uid(sock_net(sk), sk);
4675 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4678 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4679 .func = bpf_get_socket_uid,
4681 .ret_type = RET_INTEGER,
4682 .arg1_type = ARG_PTR_TO_CTX,
4685 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
4686 char *optval, int optlen)
4688 char devname[IFNAMSIZ];
4694 if (!sk_fullsock(sk))
4697 sock_owned_by_me(sk);
4699 if (level == SOL_SOCKET) {
4700 if (optlen != sizeof(int) && optname != SO_BINDTODEVICE)
4702 val = *((int *)optval);
4703 valbool = val ? 1 : 0;
4705 /* Only some socketops are supported */
4708 val = min_t(u32, val, sysctl_rmem_max);
4709 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4710 WRITE_ONCE(sk->sk_rcvbuf,
4711 max_t(int, val * 2, SOCK_MIN_RCVBUF));
4714 val = min_t(u32, val, sysctl_wmem_max);
4715 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4716 WRITE_ONCE(sk->sk_sndbuf,
4717 max_t(int, val * 2, SOCK_MIN_SNDBUF));
4719 case SO_MAX_PACING_RATE: /* 32bit version */
4721 cmpxchg(&sk->sk_pacing_status,
4724 sk->sk_max_pacing_rate = (val == ~0U) ?
4725 ~0UL : (unsigned int)val;
4726 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4727 sk->sk_max_pacing_rate);
4730 sk->sk_priority = val;
4735 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
4738 if (sk->sk_mark != val) {
4743 case SO_BINDTODEVICE:
4744 optlen = min_t(long, optlen, IFNAMSIZ - 1);
4745 strncpy(devname, optval, optlen);
4746 devname[optlen] = 0;
4749 if (devname[0] != '\0') {
4750 struct net_device *dev;
4755 dev = dev_get_by_name(net, devname);
4758 ifindex = dev->ifindex;
4762 case SO_BINDTOIFINDEX:
4763 if (optname == SO_BINDTOIFINDEX)
4765 ret = sock_bindtoindex(sk, ifindex, false);
4768 if (sk->sk_prot->keepalive)
4769 sk->sk_prot->keepalive(sk, valbool);
4770 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
4773 sk->sk_reuseport = valbool;
4779 } else if (level == SOL_IP) {
4780 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4783 val = *((int *)optval);
4784 /* Only some options are supported */
4787 if (val < -1 || val > 0xff) {
4790 struct inet_sock *inet = inet_sk(sk);
4800 #if IS_ENABLED(CONFIG_IPV6)
4801 } else if (level == SOL_IPV6) {
4802 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4805 val = *((int *)optval);
4806 /* Only some options are supported */
4809 if (val < -1 || val > 0xff) {
4812 struct ipv6_pinfo *np = inet6_sk(sk);
4823 } else if (level == SOL_TCP &&
4824 sk->sk_prot->setsockopt == tcp_setsockopt) {
4825 if (optname == TCP_CONGESTION) {
4826 char name[TCP_CA_NAME_MAX];
4828 strncpy(name, optval, min_t(long, optlen,
4829 TCP_CA_NAME_MAX-1));
4830 name[TCP_CA_NAME_MAX-1] = 0;
4831 ret = tcp_set_congestion_control(sk, name, false, true);
4833 struct inet_connection_sock *icsk = inet_csk(sk);
4834 struct tcp_sock *tp = tcp_sk(sk);
4835 unsigned long timeout;
4837 if (optlen != sizeof(int))
4840 val = *((int *)optval);
4841 /* Only some options are supported */
4844 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4849 case TCP_BPF_SNDCWND_CLAMP:
4853 tp->snd_cwnd_clamp = val;
4854 tp->snd_ssthresh = val;
4857 case TCP_BPF_DELACK_MAX:
4858 timeout = usecs_to_jiffies(val);
4859 if (timeout > TCP_DELACK_MAX ||
4860 timeout < TCP_TIMEOUT_MIN)
4862 inet_csk(sk)->icsk_delack_max = timeout;
4864 case TCP_BPF_RTO_MIN:
4865 timeout = usecs_to_jiffies(val);
4866 if (timeout > TCP_RTO_MIN ||
4867 timeout < TCP_TIMEOUT_MIN)
4869 inet_csk(sk)->icsk_rto_min = timeout;
4872 if (val < 0 || val > 1)
4878 ret = tcp_sock_set_keepidle_locked(sk, val);
4881 if (val < 1 || val > MAX_TCP_KEEPINTVL)
4884 tp->keepalive_intvl = val * HZ;
4887 if (val < 1 || val > MAX_TCP_KEEPCNT)
4890 tp->keepalive_probes = val;
4893 if (val < 1 || val > MAX_TCP_SYNCNT)
4896 icsk->icsk_syn_retries = val;
4898 case TCP_USER_TIMEOUT:
4902 icsk->icsk_user_timeout = val;
4904 case TCP_NOTSENT_LOWAT:
4905 tp->notsent_lowat = val;
4906 sk->sk_write_space(sk);
4908 case TCP_WINDOW_CLAMP:
4909 ret = tcp_set_window_clamp(sk, val);
4922 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
4923 char *optval, int optlen)
4925 if (!sk_fullsock(sk))
4928 sock_owned_by_me(sk);
4930 if (level == SOL_SOCKET) {
4931 if (optlen != sizeof(int))
4936 *((int *)optval) = sk->sk_mark;
4939 *((int *)optval) = sk->sk_priority;
4941 case SO_BINDTOIFINDEX:
4942 *((int *)optval) = sk->sk_bound_dev_if;
4945 *((int *)optval) = sk->sk_reuseport;
4951 } else if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4952 struct inet_connection_sock *icsk;
4953 struct tcp_sock *tp;
4956 case TCP_CONGESTION:
4957 icsk = inet_csk(sk);
4959 if (!icsk->icsk_ca_ops || optlen <= 1)
4961 strncpy(optval, icsk->icsk_ca_ops->name, optlen);
4962 optval[optlen - 1] = 0;
4967 if (optlen <= 0 || !tp->saved_syn ||
4968 optlen > tcp_saved_syn_len(tp->saved_syn))
4970 memcpy(optval, tp->saved_syn->data, optlen);
4975 } else if (level == SOL_IP) {
4976 struct inet_sock *inet = inet_sk(sk);
4978 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4981 /* Only some options are supported */
4984 *((int *)optval) = (int)inet->tos;
4989 #if IS_ENABLED(CONFIG_IPV6)
4990 } else if (level == SOL_IPV6) {
4991 struct ipv6_pinfo *np = inet6_sk(sk);
4993 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4996 /* Only some options are supported */
4999 *((int *)optval) = (int)np->tclass;
5011 memset(optval, 0, optlen);
5015 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5016 int, level, int, optname, char *, optval, int, optlen)
5018 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5021 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5022 .func = bpf_sock_addr_setsockopt,
5024 .ret_type = RET_INTEGER,
5025 .arg1_type = ARG_PTR_TO_CTX,
5026 .arg2_type = ARG_ANYTHING,
5027 .arg3_type = ARG_ANYTHING,
5028 .arg4_type = ARG_PTR_TO_MEM,
5029 .arg5_type = ARG_CONST_SIZE,
5032 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5033 int, level, int, optname, char *, optval, int, optlen)
5035 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5038 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5039 .func = bpf_sock_addr_getsockopt,
5041 .ret_type = RET_INTEGER,
5042 .arg1_type = ARG_PTR_TO_CTX,
5043 .arg2_type = ARG_ANYTHING,
5044 .arg3_type = ARG_ANYTHING,
5045 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5046 .arg5_type = ARG_CONST_SIZE,
5049 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5050 int, level, int, optname, char *, optval, int, optlen)
5052 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5055 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5056 .func = bpf_sock_ops_setsockopt,
5058 .ret_type = RET_INTEGER,
5059 .arg1_type = ARG_PTR_TO_CTX,
5060 .arg2_type = ARG_ANYTHING,
5061 .arg3_type = ARG_ANYTHING,
5062 .arg4_type = ARG_PTR_TO_MEM,
5063 .arg5_type = ARG_CONST_SIZE,
5066 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5067 int optname, const u8 **start)
5069 struct sk_buff *syn_skb = bpf_sock->syn_skb;
5070 const u8 *hdr_start;
5074 /* sk is a request_sock here */
5076 if (optname == TCP_BPF_SYN) {
5077 hdr_start = syn_skb->data;
5078 ret = tcp_hdrlen(syn_skb);
5079 } else if (optname == TCP_BPF_SYN_IP) {
5080 hdr_start = skb_network_header(syn_skb);
5081 ret = skb_network_header_len(syn_skb) +
5082 tcp_hdrlen(syn_skb);
5084 /* optname == TCP_BPF_SYN_MAC */
5085 hdr_start = skb_mac_header(syn_skb);
5086 ret = skb_mac_header_len(syn_skb) +
5087 skb_network_header_len(syn_skb) +
5088 tcp_hdrlen(syn_skb);
5091 struct sock *sk = bpf_sock->sk;
5092 struct saved_syn *saved_syn;
5094 if (sk->sk_state == TCP_NEW_SYN_RECV)
5095 /* synack retransmit. bpf_sock->syn_skb will
5096 * not be available. It has to resort to
5097 * saved_syn (if it is saved).
5099 saved_syn = inet_reqsk(sk)->saved_syn;
5101 saved_syn = tcp_sk(sk)->saved_syn;
5106 if (optname == TCP_BPF_SYN) {
5107 hdr_start = saved_syn->data +
5108 saved_syn->mac_hdrlen +
5109 saved_syn->network_hdrlen;
5110 ret = saved_syn->tcp_hdrlen;
5111 } else if (optname == TCP_BPF_SYN_IP) {
5112 hdr_start = saved_syn->data +
5113 saved_syn->mac_hdrlen;
5114 ret = saved_syn->network_hdrlen +
5115 saved_syn->tcp_hdrlen;
5117 /* optname == TCP_BPF_SYN_MAC */
5119 /* TCP_SAVE_SYN may not have saved the mac hdr */
5120 if (!saved_syn->mac_hdrlen)
5123 hdr_start = saved_syn->data;
5124 ret = saved_syn->mac_hdrlen +
5125 saved_syn->network_hdrlen +
5126 saved_syn->tcp_hdrlen;
5134 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5135 int, level, int, optname, char *, optval, int, optlen)
5137 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5138 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5139 int ret, copy_len = 0;
5142 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5145 if (optlen < copy_len) {
5150 memcpy(optval, start, copy_len);
5153 /* Zero out unused buffer at the end */
5154 memset(optval + copy_len, 0, optlen - copy_len);
5159 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5162 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5163 .func = bpf_sock_ops_getsockopt,
5165 .ret_type = RET_INTEGER,
5166 .arg1_type = ARG_PTR_TO_CTX,
5167 .arg2_type = ARG_ANYTHING,
5168 .arg3_type = ARG_ANYTHING,
5169 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5170 .arg5_type = ARG_CONST_SIZE,
5173 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5176 struct sock *sk = bpf_sock->sk;
5177 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5179 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5182 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5184 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5187 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5188 .func = bpf_sock_ops_cb_flags_set,
5190 .ret_type = RET_INTEGER,
5191 .arg1_type = ARG_PTR_TO_CTX,
5192 .arg2_type = ARG_ANYTHING,
5195 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5196 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5198 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5202 struct sock *sk = ctx->sk;
5203 u32 flags = BIND_FROM_BPF;
5207 if (addr_len < offsetofend(struct sockaddr, sa_family))
5209 if (addr->sa_family == AF_INET) {
5210 if (addr_len < sizeof(struct sockaddr_in))
5212 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5213 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5214 return __inet_bind(sk, addr, addr_len, flags);
5215 #if IS_ENABLED(CONFIG_IPV6)
5216 } else if (addr->sa_family == AF_INET6) {
5217 if (addr_len < SIN6_LEN_RFC2133)
5219 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5220 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5221 /* ipv6_bpf_stub cannot be NULL, since it's called from
5222 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5224 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5225 #endif /* CONFIG_IPV6 */
5227 #endif /* CONFIG_INET */
5229 return -EAFNOSUPPORT;
5232 static const struct bpf_func_proto bpf_bind_proto = {
5235 .ret_type = RET_INTEGER,
5236 .arg1_type = ARG_PTR_TO_CTX,
5237 .arg2_type = ARG_PTR_TO_MEM,
5238 .arg3_type = ARG_CONST_SIZE,
5242 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5243 struct bpf_xfrm_state *, to, u32, size, u64, flags)
5245 const struct sec_path *sp = skb_sec_path(skb);
5246 const struct xfrm_state *x;
5248 if (!sp || unlikely(index >= sp->len || flags))
5251 x = sp->xvec[index];
5253 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5256 to->reqid = x->props.reqid;
5257 to->spi = x->id.spi;
5258 to->family = x->props.family;
5261 if (to->family == AF_INET6) {
5262 memcpy(to->remote_ipv6, x->props.saddr.a6,
5263 sizeof(to->remote_ipv6));
5265 to->remote_ipv4 = x->props.saddr.a4;
5266 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5271 memset(to, 0, size);
5275 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5276 .func = bpf_skb_get_xfrm_state,
5278 .ret_type = RET_INTEGER,
5279 .arg1_type = ARG_PTR_TO_CTX,
5280 .arg2_type = ARG_ANYTHING,
5281 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
5282 .arg4_type = ARG_CONST_SIZE,
5283 .arg5_type = ARG_ANYTHING,
5287 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5288 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
5289 const struct neighbour *neigh,
5290 const struct net_device *dev, u32 mtu)
5292 memcpy(params->dmac, neigh->ha, ETH_ALEN);
5293 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5294 params->h_vlan_TCI = 0;
5295 params->h_vlan_proto = 0;
5297 params->mtu_result = mtu; /* union with tot_len */
5303 #if IS_ENABLED(CONFIG_INET)
5304 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5305 u32 flags, bool check_mtu)
5307 struct fib_nh_common *nhc;
5308 struct in_device *in_dev;
5309 struct neighbour *neigh;
5310 struct net_device *dev;
5311 struct fib_result res;
5316 dev = dev_get_by_index_rcu(net, params->ifindex);
5320 /* verify forwarding is enabled on this interface */
5321 in_dev = __in_dev_get_rcu(dev);
5322 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5323 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5325 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5327 fl4.flowi4_oif = params->ifindex;
5329 fl4.flowi4_iif = params->ifindex;
5332 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5333 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5334 fl4.flowi4_flags = 0;
5336 fl4.flowi4_proto = params->l4_protocol;
5337 fl4.daddr = params->ipv4_dst;
5338 fl4.saddr = params->ipv4_src;
5339 fl4.fl4_sport = params->sport;
5340 fl4.fl4_dport = params->dport;
5341 fl4.flowi4_multipath_hash = 0;
5343 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5344 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5345 struct fib_table *tb;
5347 tb = fib_get_table(net, tbid);
5349 return BPF_FIB_LKUP_RET_NOT_FWDED;
5351 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5353 fl4.flowi4_mark = 0;
5354 fl4.flowi4_secid = 0;
5355 fl4.flowi4_tun_key.tun_id = 0;
5356 fl4.flowi4_uid = sock_net_uid(net, NULL);
5358 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5362 /* map fib lookup errors to RTN_ type */
5364 return BPF_FIB_LKUP_RET_BLACKHOLE;
5365 if (err == -EHOSTUNREACH)
5366 return BPF_FIB_LKUP_RET_UNREACHABLE;
5368 return BPF_FIB_LKUP_RET_PROHIBIT;
5370 return BPF_FIB_LKUP_RET_NOT_FWDED;
5373 if (res.type != RTN_UNICAST)
5374 return BPF_FIB_LKUP_RET_NOT_FWDED;
5376 if (fib_info_num_path(res.fi) > 1)
5377 fib_select_path(net, &res, &fl4, NULL);
5380 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5381 if (params->tot_len > mtu) {
5382 params->mtu_result = mtu; /* union with tot_len */
5383 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5389 /* do not handle lwt encaps right now */
5390 if (nhc->nhc_lwtstate)
5391 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5395 params->rt_metric = res.fi->fib_priority;
5396 params->ifindex = dev->ifindex;
5398 /* xdp and cls_bpf programs are run in RCU-bh so
5399 * rcu_read_lock_bh is not needed here
5401 if (likely(nhc->nhc_gw_family != AF_INET6)) {
5402 if (nhc->nhc_gw_family)
5403 params->ipv4_dst = nhc->nhc_gw.ipv4;
5405 neigh = __ipv4_neigh_lookup_noref(dev,
5406 (__force u32)params->ipv4_dst);
5408 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5410 params->family = AF_INET6;
5411 *dst = nhc->nhc_gw.ipv6;
5412 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5416 return BPF_FIB_LKUP_RET_NO_NEIGH;
5418 return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5422 #if IS_ENABLED(CONFIG_IPV6)
5423 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5424 u32 flags, bool check_mtu)
5426 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5427 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5428 struct fib6_result res = {};
5429 struct neighbour *neigh;
5430 struct net_device *dev;
5431 struct inet6_dev *idev;
5437 /* link local addresses are never forwarded */
5438 if (rt6_need_strict(dst) || rt6_need_strict(src))
5439 return BPF_FIB_LKUP_RET_NOT_FWDED;
5441 dev = dev_get_by_index_rcu(net, params->ifindex);
5445 idev = __in6_dev_get_safely(dev);
5446 if (unlikely(!idev || !idev->cnf.forwarding))
5447 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5449 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5451 oif = fl6.flowi6_oif = params->ifindex;
5453 oif = fl6.flowi6_iif = params->ifindex;
5455 strict = RT6_LOOKUP_F_HAS_SADDR;
5457 fl6.flowlabel = params->flowinfo;
5458 fl6.flowi6_scope = 0;
5459 fl6.flowi6_flags = 0;
5462 fl6.flowi6_proto = params->l4_protocol;
5465 fl6.fl6_sport = params->sport;
5466 fl6.fl6_dport = params->dport;
5468 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5469 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5470 struct fib6_table *tb;
5472 tb = ipv6_stub->fib6_get_table(net, tbid);
5474 return BPF_FIB_LKUP_RET_NOT_FWDED;
5476 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5479 fl6.flowi6_mark = 0;
5480 fl6.flowi6_secid = 0;
5481 fl6.flowi6_tun_key.tun_id = 0;
5482 fl6.flowi6_uid = sock_net_uid(net, NULL);
5484 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5487 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5488 res.f6i == net->ipv6.fib6_null_entry))
5489 return BPF_FIB_LKUP_RET_NOT_FWDED;
5491 switch (res.fib6_type) {
5492 /* only unicast is forwarded */
5496 return BPF_FIB_LKUP_RET_BLACKHOLE;
5497 case RTN_UNREACHABLE:
5498 return BPF_FIB_LKUP_RET_UNREACHABLE;
5500 return BPF_FIB_LKUP_RET_PROHIBIT;
5502 return BPF_FIB_LKUP_RET_NOT_FWDED;
5505 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5506 fl6.flowi6_oif != 0, NULL, strict);
5509 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5510 if (params->tot_len > mtu) {
5511 params->mtu_result = mtu; /* union with tot_len */
5512 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5516 if (res.nh->fib_nh_lws)
5517 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5519 if (res.nh->fib_nh_gw_family)
5520 *dst = res.nh->fib_nh_gw6;
5522 dev = res.nh->fib_nh_dev;
5523 params->rt_metric = res.f6i->fib6_metric;
5524 params->ifindex = dev->ifindex;
5526 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
5529 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5531 return BPF_FIB_LKUP_RET_NO_NEIGH;
5533 return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5537 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
5538 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5540 if (plen < sizeof(*params))
5543 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5546 switch (params->family) {
5547 #if IS_ENABLED(CONFIG_INET)
5549 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
5552 #if IS_ENABLED(CONFIG_IPV6)
5554 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
5558 return -EAFNOSUPPORT;
5561 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
5562 .func = bpf_xdp_fib_lookup,
5564 .ret_type = RET_INTEGER,
5565 .arg1_type = ARG_PTR_TO_CTX,
5566 .arg2_type = ARG_PTR_TO_MEM,
5567 .arg3_type = ARG_CONST_SIZE,
5568 .arg4_type = ARG_ANYTHING,
5571 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
5572 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5574 struct net *net = dev_net(skb->dev);
5575 int rc = -EAFNOSUPPORT;
5576 bool check_mtu = false;
5578 if (plen < sizeof(*params))
5581 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5584 if (params->tot_len)
5587 switch (params->family) {
5588 #if IS_ENABLED(CONFIG_INET)
5590 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
5593 #if IS_ENABLED(CONFIG_IPV6)
5595 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
5600 if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
5601 struct net_device *dev;
5603 /* When tot_len isn't provided by user, check skb
5604 * against MTU of FIB lookup resulting net_device
5606 dev = dev_get_by_index_rcu(net, params->ifindex);
5607 if (!is_skb_forwardable(dev, skb))
5608 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
5610 params->mtu_result = dev->mtu; /* union with tot_len */
5616 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
5617 .func = bpf_skb_fib_lookup,
5619 .ret_type = RET_INTEGER,
5620 .arg1_type = ARG_PTR_TO_CTX,
5621 .arg2_type = ARG_PTR_TO_MEM,
5622 .arg3_type = ARG_CONST_SIZE,
5623 .arg4_type = ARG_ANYTHING,
5626 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
5629 struct net *netns = dev_net(dev_curr);
5631 /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
5635 return dev_get_by_index_rcu(netns, ifindex);
5638 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
5639 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5641 int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5642 struct net_device *dev = skb->dev;
5643 int skb_len, dev_len;
5646 if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
5649 if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
5652 dev = __dev_via_ifindex(dev, ifindex);
5656 mtu = READ_ONCE(dev->mtu);
5658 dev_len = mtu + dev->hard_header_len;
5660 /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
5661 skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
5663 skb_len += len_diff; /* minus result pass check */
5664 if (skb_len <= dev_len) {
5665 ret = BPF_MTU_CHK_RET_SUCCESS;
5668 /* At this point, skb->len exceed MTU, but as it include length of all
5669 * segments, it can still be below MTU. The SKB can possibly get
5670 * re-segmented in transmit path (see validate_xmit_skb). Thus, user
5671 * must choose if segs are to be MTU checked.
5673 if (skb_is_gso(skb)) {
5674 ret = BPF_MTU_CHK_RET_SUCCESS;
5676 if (flags & BPF_MTU_CHK_SEGS &&
5677 !skb_gso_validate_network_len(skb, mtu))
5678 ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
5681 /* BPF verifier guarantees valid pointer */
5687 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
5688 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5690 struct net_device *dev = xdp->rxq->dev;
5691 int xdp_len = xdp->data_end - xdp->data;
5692 int ret = BPF_MTU_CHK_RET_SUCCESS;
5695 /* XDP variant doesn't support multi-buffer segment check (yet) */
5696 if (unlikely(flags))
5699 dev = __dev_via_ifindex(dev, ifindex);
5703 mtu = READ_ONCE(dev->mtu);
5705 /* Add L2-header as dev MTU is L3 size */
5706 dev_len = mtu + dev->hard_header_len;
5708 /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
5710 xdp_len = *mtu_len + dev->hard_header_len;
5712 xdp_len += len_diff; /* minus result pass check */
5713 if (xdp_len > dev_len)
5714 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5716 /* BPF verifier guarantees valid pointer */
5722 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
5723 .func = bpf_skb_check_mtu,
5725 .ret_type = RET_INTEGER,
5726 .arg1_type = ARG_PTR_TO_CTX,
5727 .arg2_type = ARG_ANYTHING,
5728 .arg3_type = ARG_PTR_TO_INT,
5729 .arg4_type = ARG_ANYTHING,
5730 .arg5_type = ARG_ANYTHING,
5733 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
5734 .func = bpf_xdp_check_mtu,
5736 .ret_type = RET_INTEGER,
5737 .arg1_type = ARG_PTR_TO_CTX,
5738 .arg2_type = ARG_ANYTHING,
5739 .arg3_type = ARG_PTR_TO_INT,
5740 .arg4_type = ARG_ANYTHING,
5741 .arg5_type = ARG_ANYTHING,
5744 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5745 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
5748 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
5750 if (!seg6_validate_srh(srh, len, false))
5754 case BPF_LWT_ENCAP_SEG6_INLINE:
5755 if (skb->protocol != htons(ETH_P_IPV6))
5758 err = seg6_do_srh_inline(skb, srh);
5760 case BPF_LWT_ENCAP_SEG6:
5761 skb_reset_inner_headers(skb);
5762 skb->encapsulation = 1;
5763 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
5769 bpf_compute_data_pointers(skb);
5773 ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5774 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
5776 return seg6_lookup_nexthop(skb, NULL, 0);
5778 #endif /* CONFIG_IPV6_SEG6_BPF */
5780 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5781 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
5784 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
5788 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
5792 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5793 case BPF_LWT_ENCAP_SEG6:
5794 case BPF_LWT_ENCAP_SEG6_INLINE:
5795 return bpf_push_seg6_encap(skb, type, hdr, len);
5797 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5798 case BPF_LWT_ENCAP_IP:
5799 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
5806 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
5807 void *, hdr, u32, len)
5810 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5811 case BPF_LWT_ENCAP_IP:
5812 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
5819 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
5820 .func = bpf_lwt_in_push_encap,
5822 .ret_type = RET_INTEGER,
5823 .arg1_type = ARG_PTR_TO_CTX,
5824 .arg2_type = ARG_ANYTHING,
5825 .arg3_type = ARG_PTR_TO_MEM,
5826 .arg4_type = ARG_CONST_SIZE
5829 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
5830 .func = bpf_lwt_xmit_push_encap,
5832 .ret_type = RET_INTEGER,
5833 .arg1_type = ARG_PTR_TO_CTX,
5834 .arg2_type = ARG_ANYTHING,
5835 .arg3_type = ARG_PTR_TO_MEM,
5836 .arg4_type = ARG_CONST_SIZE
5839 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5840 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
5841 const void *, from, u32, len)
5843 struct seg6_bpf_srh_state *srh_state =
5844 this_cpu_ptr(&seg6_bpf_srh_states);
5845 struct ipv6_sr_hdr *srh = srh_state->srh;
5846 void *srh_tlvs, *srh_end, *ptr;
5852 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
5853 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
5855 ptr = skb->data + offset;
5856 if (ptr >= srh_tlvs && ptr + len <= srh_end)
5857 srh_state->valid = false;
5858 else if (ptr < (void *)&srh->flags ||
5859 ptr + len > (void *)&srh->segments)
5862 if (unlikely(bpf_try_make_writable(skb, offset + len)))
5864 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5866 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5868 memcpy(skb->data + offset, from, len);
5872 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
5873 .func = bpf_lwt_seg6_store_bytes,
5875 .ret_type = RET_INTEGER,
5876 .arg1_type = ARG_PTR_TO_CTX,
5877 .arg2_type = ARG_ANYTHING,
5878 .arg3_type = ARG_PTR_TO_MEM,
5879 .arg4_type = ARG_CONST_SIZE
5882 static void bpf_update_srh_state(struct sk_buff *skb)
5884 struct seg6_bpf_srh_state *srh_state =
5885 this_cpu_ptr(&seg6_bpf_srh_states);
5888 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
5889 srh_state->srh = NULL;
5891 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5892 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
5893 srh_state->valid = true;
5897 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
5898 u32, action, void *, param, u32, param_len)
5900 struct seg6_bpf_srh_state *srh_state =
5901 this_cpu_ptr(&seg6_bpf_srh_states);
5906 case SEG6_LOCAL_ACTION_END_X:
5907 if (!seg6_bpf_has_valid_srh(skb))
5909 if (param_len != sizeof(struct in6_addr))
5911 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
5912 case SEG6_LOCAL_ACTION_END_T:
5913 if (!seg6_bpf_has_valid_srh(skb))
5915 if (param_len != sizeof(int))
5917 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5918 case SEG6_LOCAL_ACTION_END_DT6:
5919 if (!seg6_bpf_has_valid_srh(skb))
5921 if (param_len != sizeof(int))
5924 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
5926 if (!pskb_pull(skb, hdroff))
5929 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
5930 skb_reset_network_header(skb);
5931 skb_reset_transport_header(skb);
5932 skb->encapsulation = 0;
5934 bpf_compute_data_pointers(skb);
5935 bpf_update_srh_state(skb);
5936 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5937 case SEG6_LOCAL_ACTION_END_B6:
5938 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5940 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
5943 bpf_update_srh_state(skb);
5946 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
5947 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5949 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
5952 bpf_update_srh_state(skb);
5960 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
5961 .func = bpf_lwt_seg6_action,
5963 .ret_type = RET_INTEGER,
5964 .arg1_type = ARG_PTR_TO_CTX,
5965 .arg2_type = ARG_ANYTHING,
5966 .arg3_type = ARG_PTR_TO_MEM,
5967 .arg4_type = ARG_CONST_SIZE
5970 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
5973 struct seg6_bpf_srh_state *srh_state =
5974 this_cpu_ptr(&seg6_bpf_srh_states);
5975 struct ipv6_sr_hdr *srh = srh_state->srh;
5976 void *srh_end, *srh_tlvs, *ptr;
5977 struct ipv6hdr *hdr;
5981 if (unlikely(srh == NULL))
5984 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
5985 ((srh->first_segment + 1) << 4));
5986 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
5988 ptr = skb->data + offset;
5990 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
5992 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
5996 ret = skb_cow_head(skb, len);
5997 if (unlikely(ret < 0))
6000 ret = bpf_skb_net_hdr_push(skb, offset, len);
6002 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6005 bpf_compute_data_pointers(skb);
6006 if (unlikely(ret < 0))
6009 hdr = (struct ipv6hdr *)skb->data;
6010 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6012 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6014 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6015 srh_state->hdrlen += len;
6016 srh_state->valid = false;
6020 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6021 .func = bpf_lwt_seg6_adjust_srh,
6023 .ret_type = RET_INTEGER,
6024 .arg1_type = ARG_PTR_TO_CTX,
6025 .arg2_type = ARG_ANYTHING,
6026 .arg3_type = ARG_ANYTHING,
6028 #endif /* CONFIG_IPV6_SEG6_BPF */
6031 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6032 int dif, int sdif, u8 family, u8 proto)
6034 bool refcounted = false;
6035 struct sock *sk = NULL;
6037 if (family == AF_INET) {
6038 __be32 src4 = tuple->ipv4.saddr;
6039 __be32 dst4 = tuple->ipv4.daddr;
6041 if (proto == IPPROTO_TCP)
6042 sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
6043 src4, tuple->ipv4.sport,
6044 dst4, tuple->ipv4.dport,
6045 dif, sdif, &refcounted);
6047 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6048 dst4, tuple->ipv4.dport,
6049 dif, sdif, &udp_table, NULL);
6050 #if IS_ENABLED(CONFIG_IPV6)
6052 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6053 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6055 if (proto == IPPROTO_TCP)
6056 sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
6057 src6, tuple->ipv6.sport,
6058 dst6, ntohs(tuple->ipv6.dport),
6059 dif, sdif, &refcounted);
6060 else if (likely(ipv6_bpf_stub))
6061 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6062 src6, tuple->ipv6.sport,
6063 dst6, tuple->ipv6.dport,
6069 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6070 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6076 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6077 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6078 * Returns the socket as an 'unsigned long' to simplify the casting in the
6079 * callers to satisfy BPF_CALL declarations.
6081 static struct sock *
6082 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6083 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6086 struct sock *sk = NULL;
6087 u8 family = AF_UNSPEC;
6091 if (len == sizeof(tuple->ipv4))
6093 else if (len == sizeof(tuple->ipv6))
6098 if (unlikely(family == AF_UNSPEC || flags ||
6099 !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6102 if (family == AF_INET)
6103 sdif = inet_sdif(skb);
6105 sdif = inet6_sdif(skb);
6107 if ((s32)netns_id < 0) {
6109 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6111 net = get_net_ns_by_id(caller_net, netns_id);
6114 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6122 static struct sock *
6123 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6124 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6127 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6128 ifindex, proto, netns_id, flags);
6131 sk = sk_to_full_sk(sk);
6132 if (!sk_fullsock(sk)) {
6141 static struct sock *
6142 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6143 u8 proto, u64 netns_id, u64 flags)
6145 struct net *caller_net;
6149 caller_net = dev_net(skb->dev);
6150 ifindex = skb->dev->ifindex;
6152 caller_net = sock_net(skb->sk);
6156 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6160 static struct sock *
6161 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6162 u8 proto, u64 netns_id, u64 flags)
6164 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6168 sk = sk_to_full_sk(sk);
6169 if (!sk_fullsock(sk)) {
6178 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6179 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6181 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6185 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6186 .func = bpf_skc_lookup_tcp,
6189 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6190 .arg1_type = ARG_PTR_TO_CTX,
6191 .arg2_type = ARG_PTR_TO_MEM,
6192 .arg3_type = ARG_CONST_SIZE,
6193 .arg4_type = ARG_ANYTHING,
6194 .arg5_type = ARG_ANYTHING,
6197 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6198 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6200 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6204 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6205 .func = bpf_sk_lookup_tcp,
6208 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6209 .arg1_type = ARG_PTR_TO_CTX,
6210 .arg2_type = ARG_PTR_TO_MEM,
6211 .arg3_type = ARG_CONST_SIZE,
6212 .arg4_type = ARG_ANYTHING,
6213 .arg5_type = ARG_ANYTHING,
6216 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6217 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6219 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6223 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6224 .func = bpf_sk_lookup_udp,
6227 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6228 .arg1_type = ARG_PTR_TO_CTX,
6229 .arg2_type = ARG_PTR_TO_MEM,
6230 .arg3_type = ARG_CONST_SIZE,
6231 .arg4_type = ARG_ANYTHING,
6232 .arg5_type = ARG_ANYTHING,
6235 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6237 if (sk && sk_is_refcounted(sk))
6242 static const struct bpf_func_proto bpf_sk_release_proto = {
6243 .func = bpf_sk_release,
6245 .ret_type = RET_INTEGER,
6246 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6249 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6250 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6252 struct net *caller_net = dev_net(ctx->rxq->dev);
6253 int ifindex = ctx->rxq->dev->ifindex;
6255 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6256 ifindex, IPPROTO_UDP, netns_id,
6260 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6261 .func = bpf_xdp_sk_lookup_udp,
6264 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6265 .arg1_type = ARG_PTR_TO_CTX,
6266 .arg2_type = ARG_PTR_TO_MEM,
6267 .arg3_type = ARG_CONST_SIZE,
6268 .arg4_type = ARG_ANYTHING,
6269 .arg5_type = ARG_ANYTHING,
6272 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6273 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6275 struct net *caller_net = dev_net(ctx->rxq->dev);
6276 int ifindex = ctx->rxq->dev->ifindex;
6278 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6279 ifindex, IPPROTO_TCP, netns_id,
6283 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6284 .func = bpf_xdp_skc_lookup_tcp,
6287 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6288 .arg1_type = ARG_PTR_TO_CTX,
6289 .arg2_type = ARG_PTR_TO_MEM,
6290 .arg3_type = ARG_CONST_SIZE,
6291 .arg4_type = ARG_ANYTHING,
6292 .arg5_type = ARG_ANYTHING,
6295 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6296 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6298 struct net *caller_net = dev_net(ctx->rxq->dev);
6299 int ifindex = ctx->rxq->dev->ifindex;
6301 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6302 ifindex, IPPROTO_TCP, netns_id,
6306 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6307 .func = bpf_xdp_sk_lookup_tcp,
6310 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6311 .arg1_type = ARG_PTR_TO_CTX,
6312 .arg2_type = ARG_PTR_TO_MEM,
6313 .arg3_type = ARG_CONST_SIZE,
6314 .arg4_type = ARG_ANYTHING,
6315 .arg5_type = ARG_ANYTHING,
6318 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6319 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6321 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6322 sock_net(ctx->sk), 0,
6323 IPPROTO_TCP, netns_id, flags);
6326 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6327 .func = bpf_sock_addr_skc_lookup_tcp,
6329 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6330 .arg1_type = ARG_PTR_TO_CTX,
6331 .arg2_type = ARG_PTR_TO_MEM,
6332 .arg3_type = ARG_CONST_SIZE,
6333 .arg4_type = ARG_ANYTHING,
6334 .arg5_type = ARG_ANYTHING,
6337 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6338 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6340 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6341 sock_net(ctx->sk), 0, IPPROTO_TCP,
6345 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6346 .func = bpf_sock_addr_sk_lookup_tcp,
6348 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6349 .arg1_type = ARG_PTR_TO_CTX,
6350 .arg2_type = ARG_PTR_TO_MEM,
6351 .arg3_type = ARG_CONST_SIZE,
6352 .arg4_type = ARG_ANYTHING,
6353 .arg5_type = ARG_ANYTHING,
6356 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6357 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6359 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6360 sock_net(ctx->sk), 0, IPPROTO_UDP,
6364 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6365 .func = bpf_sock_addr_sk_lookup_udp,
6367 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6368 .arg1_type = ARG_PTR_TO_CTX,
6369 .arg2_type = ARG_PTR_TO_MEM,
6370 .arg3_type = ARG_CONST_SIZE,
6371 .arg4_type = ARG_ANYTHING,
6372 .arg5_type = ARG_ANYTHING,
6375 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6376 struct bpf_insn_access_aux *info)
6378 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6382 if (off % size != 0)
6386 case offsetof(struct bpf_tcp_sock, bytes_received):
6387 case offsetof(struct bpf_tcp_sock, bytes_acked):
6388 return size == sizeof(__u64);
6390 return size == sizeof(__u32);
6394 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6395 const struct bpf_insn *si,
6396 struct bpf_insn *insn_buf,
6397 struct bpf_prog *prog, u32 *target_size)
6399 struct bpf_insn *insn = insn_buf;
6401 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \
6403 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \
6404 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6405 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6406 si->dst_reg, si->src_reg, \
6407 offsetof(struct tcp_sock, FIELD)); \
6410 #define BPF_INET_SOCK_GET_COMMON(FIELD) \
6412 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \
6414 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6415 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6416 struct inet_connection_sock, \
6418 si->dst_reg, si->src_reg, \
6420 struct inet_connection_sock, \
6424 if (insn > insn_buf)
6425 return insn - insn_buf;
6428 case offsetof(struct bpf_tcp_sock, rtt_min):
6429 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
6430 sizeof(struct minmax));
6431 BUILD_BUG_ON(sizeof(struct minmax) <
6432 sizeof(struct minmax_sample));
6434 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6435 offsetof(struct tcp_sock, rtt_min) +
6436 offsetof(struct minmax_sample, v));
6438 case offsetof(struct bpf_tcp_sock, snd_cwnd):
6439 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
6441 case offsetof(struct bpf_tcp_sock, srtt_us):
6442 BPF_TCP_SOCK_GET_COMMON(srtt_us);
6444 case offsetof(struct bpf_tcp_sock, snd_ssthresh):
6445 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
6447 case offsetof(struct bpf_tcp_sock, rcv_nxt):
6448 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
6450 case offsetof(struct bpf_tcp_sock, snd_nxt):
6451 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
6453 case offsetof(struct bpf_tcp_sock, snd_una):
6454 BPF_TCP_SOCK_GET_COMMON(snd_una);
6456 case offsetof(struct bpf_tcp_sock, mss_cache):
6457 BPF_TCP_SOCK_GET_COMMON(mss_cache);
6459 case offsetof(struct bpf_tcp_sock, ecn_flags):
6460 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
6462 case offsetof(struct bpf_tcp_sock, rate_delivered):
6463 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
6465 case offsetof(struct bpf_tcp_sock, rate_interval_us):
6466 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
6468 case offsetof(struct bpf_tcp_sock, packets_out):
6469 BPF_TCP_SOCK_GET_COMMON(packets_out);
6471 case offsetof(struct bpf_tcp_sock, retrans_out):
6472 BPF_TCP_SOCK_GET_COMMON(retrans_out);
6474 case offsetof(struct bpf_tcp_sock, total_retrans):
6475 BPF_TCP_SOCK_GET_COMMON(total_retrans);
6477 case offsetof(struct bpf_tcp_sock, segs_in):
6478 BPF_TCP_SOCK_GET_COMMON(segs_in);
6480 case offsetof(struct bpf_tcp_sock, data_segs_in):
6481 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
6483 case offsetof(struct bpf_tcp_sock, segs_out):
6484 BPF_TCP_SOCK_GET_COMMON(segs_out);
6486 case offsetof(struct bpf_tcp_sock, data_segs_out):
6487 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
6489 case offsetof(struct bpf_tcp_sock, lost_out):
6490 BPF_TCP_SOCK_GET_COMMON(lost_out);
6492 case offsetof(struct bpf_tcp_sock, sacked_out):
6493 BPF_TCP_SOCK_GET_COMMON(sacked_out);
6495 case offsetof(struct bpf_tcp_sock, bytes_received):
6496 BPF_TCP_SOCK_GET_COMMON(bytes_received);
6498 case offsetof(struct bpf_tcp_sock, bytes_acked):
6499 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
6501 case offsetof(struct bpf_tcp_sock, dsack_dups):
6502 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
6504 case offsetof(struct bpf_tcp_sock, delivered):
6505 BPF_TCP_SOCK_GET_COMMON(delivered);
6507 case offsetof(struct bpf_tcp_sock, delivered_ce):
6508 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
6510 case offsetof(struct bpf_tcp_sock, icsk_retransmits):
6511 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
6515 return insn - insn_buf;
6518 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
6520 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
6521 return (unsigned long)sk;
6523 return (unsigned long)NULL;
6526 const struct bpf_func_proto bpf_tcp_sock_proto = {
6527 .func = bpf_tcp_sock,
6529 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
6530 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6533 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
6535 sk = sk_to_full_sk(sk);
6537 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
6538 return (unsigned long)sk;
6540 return (unsigned long)NULL;
6543 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
6544 .func = bpf_get_listener_sock,
6546 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6547 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6550 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
6552 unsigned int iphdr_len;
6554 switch (skb_protocol(skb, true)) {
6555 case cpu_to_be16(ETH_P_IP):
6556 iphdr_len = sizeof(struct iphdr);
6558 case cpu_to_be16(ETH_P_IPV6):
6559 iphdr_len = sizeof(struct ipv6hdr);
6565 if (skb_headlen(skb) < iphdr_len)
6568 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
6571 return INET_ECN_set_ce(skb);
6574 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6575 struct bpf_insn_access_aux *info)
6577 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
6580 if (off % size != 0)
6585 return size == sizeof(__u32);
6589 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
6590 const struct bpf_insn *si,
6591 struct bpf_insn *insn_buf,
6592 struct bpf_prog *prog, u32 *target_size)
6594 struct bpf_insn *insn = insn_buf;
6596 #define BPF_XDP_SOCK_GET(FIELD) \
6598 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \
6599 sizeof_field(struct bpf_xdp_sock, FIELD)); \
6600 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
6601 si->dst_reg, si->src_reg, \
6602 offsetof(struct xdp_sock, FIELD)); \
6606 case offsetof(struct bpf_xdp_sock, queue_id):
6607 BPF_XDP_SOCK_GET(queue_id);
6611 return insn - insn_buf;
6614 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
6615 .func = bpf_skb_ecn_set_ce,
6617 .ret_type = RET_INTEGER,
6618 .arg1_type = ARG_PTR_TO_CTX,
6621 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6622 struct tcphdr *, th, u32, th_len)
6624 #ifdef CONFIG_SYN_COOKIES
6628 if (unlikely(!sk || th_len < sizeof(*th)))
6631 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
6632 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6635 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
6638 if (!th->ack || th->rst || th->syn)
6641 if (tcp_synq_no_recent_overflow(sk))
6644 cookie = ntohl(th->ack_seq) - 1;
6646 switch (sk->sk_family) {
6648 if (unlikely(iph_len < sizeof(struct iphdr)))
6651 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
6654 #if IS_BUILTIN(CONFIG_IPV6)
6656 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6659 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
6661 #endif /* CONFIG_IPV6 */
6664 return -EPROTONOSUPPORT;
6676 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
6677 .func = bpf_tcp_check_syncookie,
6680 .ret_type = RET_INTEGER,
6681 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6682 .arg2_type = ARG_PTR_TO_MEM,
6683 .arg3_type = ARG_CONST_SIZE,
6684 .arg4_type = ARG_PTR_TO_MEM,
6685 .arg5_type = ARG_CONST_SIZE,
6688 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6689 struct tcphdr *, th, u32, th_len)
6691 #ifdef CONFIG_SYN_COOKIES
6695 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
6698 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6701 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
6704 if (!th->syn || th->ack || th->fin || th->rst)
6707 if (unlikely(iph_len < sizeof(struct iphdr)))
6710 /* Both struct iphdr and struct ipv6hdr have the version field at the
6711 * same offset so we can cast to the shorter header (struct iphdr).
6713 switch (((struct iphdr *)iph)->version) {
6715 if (sk->sk_family == AF_INET6 && sk->sk_ipv6only)
6718 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
6721 #if IS_BUILTIN(CONFIG_IPV6)
6723 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6726 if (sk->sk_family != AF_INET6)
6729 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
6731 #endif /* CONFIG_IPV6 */
6734 return -EPROTONOSUPPORT;
6739 return cookie | ((u64)mss << 32);
6742 #endif /* CONFIG_SYN_COOKIES */
6745 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
6746 .func = bpf_tcp_gen_syncookie,
6747 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */
6749 .ret_type = RET_INTEGER,
6750 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6751 .arg2_type = ARG_PTR_TO_MEM,
6752 .arg3_type = ARG_CONST_SIZE,
6753 .arg4_type = ARG_PTR_TO_MEM,
6754 .arg5_type = ARG_CONST_SIZE,
6757 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
6759 if (!sk || flags != 0)
6761 if (!skb_at_tc_ingress(skb))
6763 if (unlikely(dev_net(skb->dev) != sock_net(sk)))
6764 return -ENETUNREACH;
6765 if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
6766 return -ESOCKTNOSUPPORT;
6767 if (sk_is_refcounted(sk) &&
6768 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
6773 skb->destructor = sock_pfree;
6778 static const struct bpf_func_proto bpf_sk_assign_proto = {
6779 .func = bpf_sk_assign,
6781 .ret_type = RET_INTEGER,
6782 .arg1_type = ARG_PTR_TO_CTX,
6783 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6784 .arg3_type = ARG_ANYTHING,
6787 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
6788 u8 search_kind, const u8 *magic,
6789 u8 magic_len, bool *eol)
6795 while (op < opend) {
6798 if (kind == TCPOPT_EOL) {
6800 return ERR_PTR(-ENOMSG);
6801 } else if (kind == TCPOPT_NOP) {
6806 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
6807 /* Something is wrong in the received header.
6808 * Follow the TCP stack's tcp_parse_options()
6809 * and just bail here.
6811 return ERR_PTR(-EFAULT);
6814 if (search_kind == kind) {
6818 if (magic_len > kind_len - 2)
6819 return ERR_PTR(-ENOMSG);
6821 if (!memcmp(&op[2], magic, magic_len))
6828 return ERR_PTR(-ENOMSG);
6831 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6832 void *, search_res, u32, len, u64, flags)
6834 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
6835 const u8 *op, *opend, *magic, *search = search_res;
6836 u8 search_kind, search_len, copy_len, magic_len;
6839 /* 2 byte is the minimal option len except TCPOPT_NOP and
6840 * TCPOPT_EOL which are useless for the bpf prog to learn
6841 * and this helper disallow loading them also.
6843 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
6846 search_kind = search[0];
6847 search_len = search[1];
6849 if (search_len > len || search_kind == TCPOPT_NOP ||
6850 search_kind == TCPOPT_EOL)
6853 if (search_kind == TCPOPT_EXP || search_kind == 253) {
6854 /* 16 or 32 bit magic. +2 for kind and kind length */
6855 if (search_len != 4 && search_len != 6)
6858 magic_len = search_len - 2;
6867 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
6872 op += sizeof(struct tcphdr);
6874 if (!bpf_sock->skb ||
6875 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6876 /* This bpf_sock->op cannot call this helper */
6879 opend = bpf_sock->skb_data_end;
6880 op = bpf_sock->skb->data + sizeof(struct tcphdr);
6883 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
6890 if (copy_len > len) {
6895 memcpy(search_res, op, copy_len);
6899 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
6900 .func = bpf_sock_ops_load_hdr_opt,
6902 .ret_type = RET_INTEGER,
6903 .arg1_type = ARG_PTR_TO_CTX,
6904 .arg2_type = ARG_PTR_TO_MEM,
6905 .arg3_type = ARG_CONST_SIZE,
6906 .arg4_type = ARG_ANYTHING,
6909 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6910 const void *, from, u32, len, u64, flags)
6912 u8 new_kind, new_kind_len, magic_len = 0, *opend;
6913 const u8 *op, *new_op, *magic = NULL;
6914 struct sk_buff *skb;
6917 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
6920 if (len < 2 || flags)
6924 new_kind = new_op[0];
6925 new_kind_len = new_op[1];
6927 if (new_kind_len > len || new_kind == TCPOPT_NOP ||
6928 new_kind == TCPOPT_EOL)
6931 if (new_kind_len > bpf_sock->remaining_opt_len)
6934 /* 253 is another experimental kind */
6935 if (new_kind == TCPOPT_EXP || new_kind == 253) {
6936 if (new_kind_len < 4)
6938 /* Match for the 2 byte magic also.
6939 * RFC 6994: the magic could be 2 or 4 bytes.
6940 * Hence, matching by 2 byte only is on the
6941 * conservative side but it is the right
6942 * thing to do for the 'search-for-duplication'
6949 /* Check for duplication */
6950 skb = bpf_sock->skb;
6951 op = skb->data + sizeof(struct tcphdr);
6952 opend = bpf_sock->skb_data_end;
6954 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
6959 if (PTR_ERR(op) != -ENOMSG)
6963 /* The option has been ended. Treat it as no more
6964 * header option can be written.
6968 /* No duplication found. Store the header option. */
6969 memcpy(opend, from, new_kind_len);
6971 bpf_sock->remaining_opt_len -= new_kind_len;
6972 bpf_sock->skb_data_end += new_kind_len;
6977 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
6978 .func = bpf_sock_ops_store_hdr_opt,
6980 .ret_type = RET_INTEGER,
6981 .arg1_type = ARG_PTR_TO_CTX,
6982 .arg2_type = ARG_PTR_TO_MEM,
6983 .arg3_type = ARG_CONST_SIZE,
6984 .arg4_type = ARG_ANYTHING,
6987 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6988 u32, len, u64, flags)
6990 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6993 if (flags || len < 2)
6996 if (len > bpf_sock->remaining_opt_len)
6999 bpf_sock->remaining_opt_len -= len;
7004 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7005 .func = bpf_sock_ops_reserve_hdr_opt,
7007 .ret_type = RET_INTEGER,
7008 .arg1_type = ARG_PTR_TO_CTX,
7009 .arg2_type = ARG_ANYTHING,
7010 .arg3_type = ARG_ANYTHING,
7013 #endif /* CONFIG_INET */
7015 bool bpf_helper_changes_pkt_data(void *func)
7017 if (func == bpf_skb_vlan_push ||
7018 func == bpf_skb_vlan_pop ||
7019 func == bpf_skb_store_bytes ||
7020 func == bpf_skb_change_proto ||
7021 func == bpf_skb_change_head ||
7022 func == sk_skb_change_head ||
7023 func == bpf_skb_change_tail ||
7024 func == sk_skb_change_tail ||
7025 func == bpf_skb_adjust_room ||
7026 func == sk_skb_adjust_room ||
7027 func == bpf_skb_pull_data ||
7028 func == sk_skb_pull_data ||
7029 func == bpf_clone_redirect ||
7030 func == bpf_l3_csum_replace ||
7031 func == bpf_l4_csum_replace ||
7032 func == bpf_xdp_adjust_head ||
7033 func == bpf_xdp_adjust_meta ||
7034 func == bpf_msg_pull_data ||
7035 func == bpf_msg_push_data ||
7036 func == bpf_msg_pop_data ||
7037 func == bpf_xdp_adjust_tail ||
7038 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7039 func == bpf_lwt_seg6_store_bytes ||
7040 func == bpf_lwt_seg6_adjust_srh ||
7041 func == bpf_lwt_seg6_action ||
7044 func == bpf_sock_ops_store_hdr_opt ||
7046 func == bpf_lwt_in_push_encap ||
7047 func == bpf_lwt_xmit_push_encap)
7053 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7054 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7056 static const struct bpf_func_proto *
7057 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7060 /* inet and inet6 sockets are created in a process
7061 * context so there is always a valid uid/gid
7063 case BPF_FUNC_get_current_uid_gid:
7064 return &bpf_get_current_uid_gid_proto;
7065 case BPF_FUNC_get_local_storage:
7066 return &bpf_get_local_storage_proto;
7067 case BPF_FUNC_get_socket_cookie:
7068 return &bpf_get_socket_cookie_sock_proto;
7069 case BPF_FUNC_get_netns_cookie:
7070 return &bpf_get_netns_cookie_sock_proto;
7071 case BPF_FUNC_perf_event_output:
7072 return &bpf_event_output_data_proto;
7073 case BPF_FUNC_get_current_pid_tgid:
7074 return &bpf_get_current_pid_tgid_proto;
7075 case BPF_FUNC_get_current_comm:
7076 return &bpf_get_current_comm_proto;
7077 #ifdef CONFIG_CGROUPS
7078 case BPF_FUNC_get_current_cgroup_id:
7079 return &bpf_get_current_cgroup_id_proto;
7080 case BPF_FUNC_get_current_ancestor_cgroup_id:
7081 return &bpf_get_current_ancestor_cgroup_id_proto;
7083 #ifdef CONFIG_CGROUP_NET_CLASSID
7084 case BPF_FUNC_get_cgroup_classid:
7085 return &bpf_get_cgroup_classid_curr_proto;
7087 case BPF_FUNC_sk_storage_get:
7088 return &bpf_sk_storage_get_cg_sock_proto;
7090 return bpf_base_func_proto(func_id);
7094 static const struct bpf_func_proto *
7095 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7098 /* inet and inet6 sockets are created in a process
7099 * context so there is always a valid uid/gid
7101 case BPF_FUNC_get_current_uid_gid:
7102 return &bpf_get_current_uid_gid_proto;
7104 switch (prog->expected_attach_type) {
7105 case BPF_CGROUP_INET4_CONNECT:
7106 case BPF_CGROUP_INET6_CONNECT:
7107 return &bpf_bind_proto;
7111 case BPF_FUNC_get_socket_cookie:
7112 return &bpf_get_socket_cookie_sock_addr_proto;
7113 case BPF_FUNC_get_netns_cookie:
7114 return &bpf_get_netns_cookie_sock_addr_proto;
7115 case BPF_FUNC_get_local_storage:
7116 return &bpf_get_local_storage_proto;
7117 case BPF_FUNC_perf_event_output:
7118 return &bpf_event_output_data_proto;
7119 case BPF_FUNC_get_current_pid_tgid:
7120 return &bpf_get_current_pid_tgid_proto;
7121 case BPF_FUNC_get_current_comm:
7122 return &bpf_get_current_comm_proto;
7123 #ifdef CONFIG_CGROUPS
7124 case BPF_FUNC_get_current_cgroup_id:
7125 return &bpf_get_current_cgroup_id_proto;
7126 case BPF_FUNC_get_current_ancestor_cgroup_id:
7127 return &bpf_get_current_ancestor_cgroup_id_proto;
7129 #ifdef CONFIG_CGROUP_NET_CLASSID
7130 case BPF_FUNC_get_cgroup_classid:
7131 return &bpf_get_cgroup_classid_curr_proto;
7134 case BPF_FUNC_sk_lookup_tcp:
7135 return &bpf_sock_addr_sk_lookup_tcp_proto;
7136 case BPF_FUNC_sk_lookup_udp:
7137 return &bpf_sock_addr_sk_lookup_udp_proto;
7138 case BPF_FUNC_sk_release:
7139 return &bpf_sk_release_proto;
7140 case BPF_FUNC_skc_lookup_tcp:
7141 return &bpf_sock_addr_skc_lookup_tcp_proto;
7142 #endif /* CONFIG_INET */
7143 case BPF_FUNC_sk_storage_get:
7144 return &bpf_sk_storage_get_proto;
7145 case BPF_FUNC_sk_storage_delete:
7146 return &bpf_sk_storage_delete_proto;
7147 case BPF_FUNC_setsockopt:
7148 switch (prog->expected_attach_type) {
7149 case BPF_CGROUP_INET4_BIND:
7150 case BPF_CGROUP_INET6_BIND:
7151 case BPF_CGROUP_INET4_CONNECT:
7152 case BPF_CGROUP_INET6_CONNECT:
7153 case BPF_CGROUP_UDP4_RECVMSG:
7154 case BPF_CGROUP_UDP6_RECVMSG:
7155 case BPF_CGROUP_UDP4_SENDMSG:
7156 case BPF_CGROUP_UDP6_SENDMSG:
7157 case BPF_CGROUP_INET4_GETPEERNAME:
7158 case BPF_CGROUP_INET6_GETPEERNAME:
7159 case BPF_CGROUP_INET4_GETSOCKNAME:
7160 case BPF_CGROUP_INET6_GETSOCKNAME:
7161 return &bpf_sock_addr_setsockopt_proto;
7165 case BPF_FUNC_getsockopt:
7166 switch (prog->expected_attach_type) {
7167 case BPF_CGROUP_INET4_BIND:
7168 case BPF_CGROUP_INET6_BIND:
7169 case BPF_CGROUP_INET4_CONNECT:
7170 case BPF_CGROUP_INET6_CONNECT:
7171 case BPF_CGROUP_UDP4_RECVMSG:
7172 case BPF_CGROUP_UDP6_RECVMSG:
7173 case BPF_CGROUP_UDP4_SENDMSG:
7174 case BPF_CGROUP_UDP6_SENDMSG:
7175 case BPF_CGROUP_INET4_GETPEERNAME:
7176 case BPF_CGROUP_INET6_GETPEERNAME:
7177 case BPF_CGROUP_INET4_GETSOCKNAME:
7178 case BPF_CGROUP_INET6_GETSOCKNAME:
7179 return &bpf_sock_addr_getsockopt_proto;
7184 return bpf_sk_base_func_proto(func_id);
7188 static const struct bpf_func_proto *
7189 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7192 case BPF_FUNC_skb_load_bytes:
7193 return &bpf_skb_load_bytes_proto;
7194 case BPF_FUNC_skb_load_bytes_relative:
7195 return &bpf_skb_load_bytes_relative_proto;
7196 case BPF_FUNC_get_socket_cookie:
7197 return &bpf_get_socket_cookie_proto;
7198 case BPF_FUNC_get_socket_uid:
7199 return &bpf_get_socket_uid_proto;
7200 case BPF_FUNC_perf_event_output:
7201 return &bpf_skb_event_output_proto;
7203 return bpf_sk_base_func_proto(func_id);
7207 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7208 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7210 static const struct bpf_func_proto *
7211 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7214 case BPF_FUNC_get_local_storage:
7215 return &bpf_get_local_storage_proto;
7216 case BPF_FUNC_sk_fullsock:
7217 return &bpf_sk_fullsock_proto;
7218 case BPF_FUNC_sk_storage_get:
7219 return &bpf_sk_storage_get_proto;
7220 case BPF_FUNC_sk_storage_delete:
7221 return &bpf_sk_storage_delete_proto;
7222 case BPF_FUNC_perf_event_output:
7223 return &bpf_skb_event_output_proto;
7224 #ifdef CONFIG_SOCK_CGROUP_DATA
7225 case BPF_FUNC_skb_cgroup_id:
7226 return &bpf_skb_cgroup_id_proto;
7227 case BPF_FUNC_skb_ancestor_cgroup_id:
7228 return &bpf_skb_ancestor_cgroup_id_proto;
7229 case BPF_FUNC_sk_cgroup_id:
7230 return &bpf_sk_cgroup_id_proto;
7231 case BPF_FUNC_sk_ancestor_cgroup_id:
7232 return &bpf_sk_ancestor_cgroup_id_proto;
7235 case BPF_FUNC_sk_lookup_tcp:
7236 return &bpf_sk_lookup_tcp_proto;
7237 case BPF_FUNC_sk_lookup_udp:
7238 return &bpf_sk_lookup_udp_proto;
7239 case BPF_FUNC_sk_release:
7240 return &bpf_sk_release_proto;
7241 case BPF_FUNC_skc_lookup_tcp:
7242 return &bpf_skc_lookup_tcp_proto;
7243 case BPF_FUNC_tcp_sock:
7244 return &bpf_tcp_sock_proto;
7245 case BPF_FUNC_get_listener_sock:
7246 return &bpf_get_listener_sock_proto;
7247 case BPF_FUNC_skb_ecn_set_ce:
7248 return &bpf_skb_ecn_set_ce_proto;
7251 return sk_filter_func_proto(func_id, prog);
7255 static const struct bpf_func_proto *
7256 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7259 case BPF_FUNC_skb_store_bytes:
7260 return &bpf_skb_store_bytes_proto;
7261 case BPF_FUNC_skb_load_bytes:
7262 return &bpf_skb_load_bytes_proto;
7263 case BPF_FUNC_skb_load_bytes_relative:
7264 return &bpf_skb_load_bytes_relative_proto;
7265 case BPF_FUNC_skb_pull_data:
7266 return &bpf_skb_pull_data_proto;
7267 case BPF_FUNC_csum_diff:
7268 return &bpf_csum_diff_proto;
7269 case BPF_FUNC_csum_update:
7270 return &bpf_csum_update_proto;
7271 case BPF_FUNC_csum_level:
7272 return &bpf_csum_level_proto;
7273 case BPF_FUNC_l3_csum_replace:
7274 return &bpf_l3_csum_replace_proto;
7275 case BPF_FUNC_l4_csum_replace:
7276 return &bpf_l4_csum_replace_proto;
7277 case BPF_FUNC_clone_redirect:
7278 return &bpf_clone_redirect_proto;
7279 case BPF_FUNC_get_cgroup_classid:
7280 return &bpf_get_cgroup_classid_proto;
7281 case BPF_FUNC_skb_vlan_push:
7282 return &bpf_skb_vlan_push_proto;
7283 case BPF_FUNC_skb_vlan_pop:
7284 return &bpf_skb_vlan_pop_proto;
7285 case BPF_FUNC_skb_change_proto:
7286 return &bpf_skb_change_proto_proto;
7287 case BPF_FUNC_skb_change_type:
7288 return &bpf_skb_change_type_proto;
7289 case BPF_FUNC_skb_adjust_room:
7290 return &bpf_skb_adjust_room_proto;
7291 case BPF_FUNC_skb_change_tail:
7292 return &bpf_skb_change_tail_proto;
7293 case BPF_FUNC_skb_change_head:
7294 return &bpf_skb_change_head_proto;
7295 case BPF_FUNC_skb_get_tunnel_key:
7296 return &bpf_skb_get_tunnel_key_proto;
7297 case BPF_FUNC_skb_set_tunnel_key:
7298 return bpf_get_skb_set_tunnel_proto(func_id);
7299 case BPF_FUNC_skb_get_tunnel_opt:
7300 return &bpf_skb_get_tunnel_opt_proto;
7301 case BPF_FUNC_skb_set_tunnel_opt:
7302 return bpf_get_skb_set_tunnel_proto(func_id);
7303 case BPF_FUNC_redirect:
7304 return &bpf_redirect_proto;
7305 case BPF_FUNC_redirect_neigh:
7306 return &bpf_redirect_neigh_proto;
7307 case BPF_FUNC_redirect_peer:
7308 return &bpf_redirect_peer_proto;
7309 case BPF_FUNC_get_route_realm:
7310 return &bpf_get_route_realm_proto;
7311 case BPF_FUNC_get_hash_recalc:
7312 return &bpf_get_hash_recalc_proto;
7313 case BPF_FUNC_set_hash_invalid:
7314 return &bpf_set_hash_invalid_proto;
7315 case BPF_FUNC_set_hash:
7316 return &bpf_set_hash_proto;
7317 case BPF_FUNC_perf_event_output:
7318 return &bpf_skb_event_output_proto;
7319 case BPF_FUNC_get_smp_processor_id:
7320 return &bpf_get_smp_processor_id_proto;
7321 case BPF_FUNC_skb_under_cgroup:
7322 return &bpf_skb_under_cgroup_proto;
7323 case BPF_FUNC_get_socket_cookie:
7324 return &bpf_get_socket_cookie_proto;
7325 case BPF_FUNC_get_socket_uid:
7326 return &bpf_get_socket_uid_proto;
7327 case BPF_FUNC_fib_lookup:
7328 return &bpf_skb_fib_lookup_proto;
7329 case BPF_FUNC_check_mtu:
7330 return &bpf_skb_check_mtu_proto;
7331 case BPF_FUNC_sk_fullsock:
7332 return &bpf_sk_fullsock_proto;
7333 case BPF_FUNC_sk_storage_get:
7334 return &bpf_sk_storage_get_proto;
7335 case BPF_FUNC_sk_storage_delete:
7336 return &bpf_sk_storage_delete_proto;
7338 case BPF_FUNC_skb_get_xfrm_state:
7339 return &bpf_skb_get_xfrm_state_proto;
7341 #ifdef CONFIG_CGROUP_NET_CLASSID
7342 case BPF_FUNC_skb_cgroup_classid:
7343 return &bpf_skb_cgroup_classid_proto;
7345 #ifdef CONFIG_SOCK_CGROUP_DATA
7346 case BPF_FUNC_skb_cgroup_id:
7347 return &bpf_skb_cgroup_id_proto;
7348 case BPF_FUNC_skb_ancestor_cgroup_id:
7349 return &bpf_skb_ancestor_cgroup_id_proto;
7352 case BPF_FUNC_sk_lookup_tcp:
7353 return &bpf_sk_lookup_tcp_proto;
7354 case BPF_FUNC_sk_lookup_udp:
7355 return &bpf_sk_lookup_udp_proto;
7356 case BPF_FUNC_sk_release:
7357 return &bpf_sk_release_proto;
7358 case BPF_FUNC_tcp_sock:
7359 return &bpf_tcp_sock_proto;
7360 case BPF_FUNC_get_listener_sock:
7361 return &bpf_get_listener_sock_proto;
7362 case BPF_FUNC_skc_lookup_tcp:
7363 return &bpf_skc_lookup_tcp_proto;
7364 case BPF_FUNC_tcp_check_syncookie:
7365 return &bpf_tcp_check_syncookie_proto;
7366 case BPF_FUNC_skb_ecn_set_ce:
7367 return &bpf_skb_ecn_set_ce_proto;
7368 case BPF_FUNC_tcp_gen_syncookie:
7369 return &bpf_tcp_gen_syncookie_proto;
7370 case BPF_FUNC_sk_assign:
7371 return &bpf_sk_assign_proto;
7374 return bpf_sk_base_func_proto(func_id);
7378 static const struct bpf_func_proto *
7379 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7382 case BPF_FUNC_perf_event_output:
7383 return &bpf_xdp_event_output_proto;
7384 case BPF_FUNC_get_smp_processor_id:
7385 return &bpf_get_smp_processor_id_proto;
7386 case BPF_FUNC_csum_diff:
7387 return &bpf_csum_diff_proto;
7388 case BPF_FUNC_xdp_adjust_head:
7389 return &bpf_xdp_adjust_head_proto;
7390 case BPF_FUNC_xdp_adjust_meta:
7391 return &bpf_xdp_adjust_meta_proto;
7392 case BPF_FUNC_redirect:
7393 return &bpf_xdp_redirect_proto;
7394 case BPF_FUNC_redirect_map:
7395 return &bpf_xdp_redirect_map_proto;
7396 case BPF_FUNC_xdp_adjust_tail:
7397 return &bpf_xdp_adjust_tail_proto;
7398 case BPF_FUNC_fib_lookup:
7399 return &bpf_xdp_fib_lookup_proto;
7400 case BPF_FUNC_check_mtu:
7401 return &bpf_xdp_check_mtu_proto;
7403 case BPF_FUNC_sk_lookup_udp:
7404 return &bpf_xdp_sk_lookup_udp_proto;
7405 case BPF_FUNC_sk_lookup_tcp:
7406 return &bpf_xdp_sk_lookup_tcp_proto;
7407 case BPF_FUNC_sk_release:
7408 return &bpf_sk_release_proto;
7409 case BPF_FUNC_skc_lookup_tcp:
7410 return &bpf_xdp_skc_lookup_tcp_proto;
7411 case BPF_FUNC_tcp_check_syncookie:
7412 return &bpf_tcp_check_syncookie_proto;
7413 case BPF_FUNC_tcp_gen_syncookie:
7414 return &bpf_tcp_gen_syncookie_proto;
7417 return bpf_sk_base_func_proto(func_id);
7421 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
7422 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
7424 static const struct bpf_func_proto *
7425 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7428 case BPF_FUNC_setsockopt:
7429 return &bpf_sock_ops_setsockopt_proto;
7430 case BPF_FUNC_getsockopt:
7431 return &bpf_sock_ops_getsockopt_proto;
7432 case BPF_FUNC_sock_ops_cb_flags_set:
7433 return &bpf_sock_ops_cb_flags_set_proto;
7434 case BPF_FUNC_sock_map_update:
7435 return &bpf_sock_map_update_proto;
7436 case BPF_FUNC_sock_hash_update:
7437 return &bpf_sock_hash_update_proto;
7438 case BPF_FUNC_get_socket_cookie:
7439 return &bpf_get_socket_cookie_sock_ops_proto;
7440 case BPF_FUNC_get_local_storage:
7441 return &bpf_get_local_storage_proto;
7442 case BPF_FUNC_perf_event_output:
7443 return &bpf_event_output_data_proto;
7444 case BPF_FUNC_sk_storage_get:
7445 return &bpf_sk_storage_get_proto;
7446 case BPF_FUNC_sk_storage_delete:
7447 return &bpf_sk_storage_delete_proto;
7449 case BPF_FUNC_load_hdr_opt:
7450 return &bpf_sock_ops_load_hdr_opt_proto;
7451 case BPF_FUNC_store_hdr_opt:
7452 return &bpf_sock_ops_store_hdr_opt_proto;
7453 case BPF_FUNC_reserve_hdr_opt:
7454 return &bpf_sock_ops_reserve_hdr_opt_proto;
7455 case BPF_FUNC_tcp_sock:
7456 return &bpf_tcp_sock_proto;
7457 #endif /* CONFIG_INET */
7459 return bpf_sk_base_func_proto(func_id);
7463 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
7464 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
7466 static const struct bpf_func_proto *
7467 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7470 case BPF_FUNC_msg_redirect_map:
7471 return &bpf_msg_redirect_map_proto;
7472 case BPF_FUNC_msg_redirect_hash:
7473 return &bpf_msg_redirect_hash_proto;
7474 case BPF_FUNC_msg_apply_bytes:
7475 return &bpf_msg_apply_bytes_proto;
7476 case BPF_FUNC_msg_cork_bytes:
7477 return &bpf_msg_cork_bytes_proto;
7478 case BPF_FUNC_msg_pull_data:
7479 return &bpf_msg_pull_data_proto;
7480 case BPF_FUNC_msg_push_data:
7481 return &bpf_msg_push_data_proto;
7482 case BPF_FUNC_msg_pop_data:
7483 return &bpf_msg_pop_data_proto;
7484 case BPF_FUNC_perf_event_output:
7485 return &bpf_event_output_data_proto;
7486 case BPF_FUNC_get_current_uid_gid:
7487 return &bpf_get_current_uid_gid_proto;
7488 case BPF_FUNC_get_current_pid_tgid:
7489 return &bpf_get_current_pid_tgid_proto;
7490 case BPF_FUNC_sk_storage_get:
7491 return &bpf_sk_storage_get_proto;
7492 case BPF_FUNC_sk_storage_delete:
7493 return &bpf_sk_storage_delete_proto;
7494 #ifdef CONFIG_CGROUPS
7495 case BPF_FUNC_get_current_cgroup_id:
7496 return &bpf_get_current_cgroup_id_proto;
7497 case BPF_FUNC_get_current_ancestor_cgroup_id:
7498 return &bpf_get_current_ancestor_cgroup_id_proto;
7500 #ifdef CONFIG_CGROUP_NET_CLASSID
7501 case BPF_FUNC_get_cgroup_classid:
7502 return &bpf_get_cgroup_classid_curr_proto;
7505 return bpf_sk_base_func_proto(func_id);
7509 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
7510 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
7512 static const struct bpf_func_proto *
7513 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7516 case BPF_FUNC_skb_store_bytes:
7517 return &bpf_skb_store_bytes_proto;
7518 case BPF_FUNC_skb_load_bytes:
7519 return &bpf_skb_load_bytes_proto;
7520 case BPF_FUNC_skb_pull_data:
7521 return &sk_skb_pull_data_proto;
7522 case BPF_FUNC_skb_change_tail:
7523 return &sk_skb_change_tail_proto;
7524 case BPF_FUNC_skb_change_head:
7525 return &sk_skb_change_head_proto;
7526 case BPF_FUNC_skb_adjust_room:
7527 return &sk_skb_adjust_room_proto;
7528 case BPF_FUNC_get_socket_cookie:
7529 return &bpf_get_socket_cookie_proto;
7530 case BPF_FUNC_get_socket_uid:
7531 return &bpf_get_socket_uid_proto;
7532 case BPF_FUNC_sk_redirect_map:
7533 return &bpf_sk_redirect_map_proto;
7534 case BPF_FUNC_sk_redirect_hash:
7535 return &bpf_sk_redirect_hash_proto;
7536 case BPF_FUNC_perf_event_output:
7537 return &bpf_skb_event_output_proto;
7539 case BPF_FUNC_sk_lookup_tcp:
7540 return &bpf_sk_lookup_tcp_proto;
7541 case BPF_FUNC_sk_lookup_udp:
7542 return &bpf_sk_lookup_udp_proto;
7543 case BPF_FUNC_sk_release:
7544 return &bpf_sk_release_proto;
7545 case BPF_FUNC_skc_lookup_tcp:
7546 return &bpf_skc_lookup_tcp_proto;
7549 return bpf_sk_base_func_proto(func_id);
7553 static const struct bpf_func_proto *
7554 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7557 case BPF_FUNC_skb_load_bytes:
7558 return &bpf_flow_dissector_load_bytes_proto;
7560 return bpf_sk_base_func_proto(func_id);
7564 static const struct bpf_func_proto *
7565 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7568 case BPF_FUNC_skb_load_bytes:
7569 return &bpf_skb_load_bytes_proto;
7570 case BPF_FUNC_skb_pull_data:
7571 return &bpf_skb_pull_data_proto;
7572 case BPF_FUNC_csum_diff:
7573 return &bpf_csum_diff_proto;
7574 case BPF_FUNC_get_cgroup_classid:
7575 return &bpf_get_cgroup_classid_proto;
7576 case BPF_FUNC_get_route_realm:
7577 return &bpf_get_route_realm_proto;
7578 case BPF_FUNC_get_hash_recalc:
7579 return &bpf_get_hash_recalc_proto;
7580 case BPF_FUNC_perf_event_output:
7581 return &bpf_skb_event_output_proto;
7582 case BPF_FUNC_get_smp_processor_id:
7583 return &bpf_get_smp_processor_id_proto;
7584 case BPF_FUNC_skb_under_cgroup:
7585 return &bpf_skb_under_cgroup_proto;
7587 return bpf_sk_base_func_proto(func_id);
7591 static const struct bpf_func_proto *
7592 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7595 case BPF_FUNC_lwt_push_encap:
7596 return &bpf_lwt_in_push_encap_proto;
7598 return lwt_out_func_proto(func_id, prog);
7602 static const struct bpf_func_proto *
7603 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7606 case BPF_FUNC_skb_get_tunnel_key:
7607 return &bpf_skb_get_tunnel_key_proto;
7608 case BPF_FUNC_skb_set_tunnel_key:
7609 return bpf_get_skb_set_tunnel_proto(func_id);
7610 case BPF_FUNC_skb_get_tunnel_opt:
7611 return &bpf_skb_get_tunnel_opt_proto;
7612 case BPF_FUNC_skb_set_tunnel_opt:
7613 return bpf_get_skb_set_tunnel_proto(func_id);
7614 case BPF_FUNC_redirect:
7615 return &bpf_redirect_proto;
7616 case BPF_FUNC_clone_redirect:
7617 return &bpf_clone_redirect_proto;
7618 case BPF_FUNC_skb_change_tail:
7619 return &bpf_skb_change_tail_proto;
7620 case BPF_FUNC_skb_change_head:
7621 return &bpf_skb_change_head_proto;
7622 case BPF_FUNC_skb_store_bytes:
7623 return &bpf_skb_store_bytes_proto;
7624 case BPF_FUNC_csum_update:
7625 return &bpf_csum_update_proto;
7626 case BPF_FUNC_csum_level:
7627 return &bpf_csum_level_proto;
7628 case BPF_FUNC_l3_csum_replace:
7629 return &bpf_l3_csum_replace_proto;
7630 case BPF_FUNC_l4_csum_replace:
7631 return &bpf_l4_csum_replace_proto;
7632 case BPF_FUNC_set_hash_invalid:
7633 return &bpf_set_hash_invalid_proto;
7634 case BPF_FUNC_lwt_push_encap:
7635 return &bpf_lwt_xmit_push_encap_proto;
7637 return lwt_out_func_proto(func_id, prog);
7641 static const struct bpf_func_proto *
7642 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7645 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7646 case BPF_FUNC_lwt_seg6_store_bytes:
7647 return &bpf_lwt_seg6_store_bytes_proto;
7648 case BPF_FUNC_lwt_seg6_action:
7649 return &bpf_lwt_seg6_action_proto;
7650 case BPF_FUNC_lwt_seg6_adjust_srh:
7651 return &bpf_lwt_seg6_adjust_srh_proto;
7654 return lwt_out_func_proto(func_id, prog);
7658 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
7659 const struct bpf_prog *prog,
7660 struct bpf_insn_access_aux *info)
7662 const int size_default = sizeof(__u32);
7664 if (off < 0 || off >= sizeof(struct __sk_buff))
7667 /* The verifier guarantees that size > 0. */
7668 if (off % size != 0)
7672 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7673 if (off + size > offsetofend(struct __sk_buff, cb[4]))
7676 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
7677 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
7678 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
7679 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
7680 case bpf_ctx_range(struct __sk_buff, data):
7681 case bpf_ctx_range(struct __sk_buff, data_meta):
7682 case bpf_ctx_range(struct __sk_buff, data_end):
7683 if (size != size_default)
7686 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
7688 case bpf_ctx_range(struct __sk_buff, tstamp):
7689 if (size != sizeof(__u64))
7692 case offsetof(struct __sk_buff, sk):
7693 if (type == BPF_WRITE || size != sizeof(__u64))
7695 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
7698 /* Only narrow read access allowed for now. */
7699 if (type == BPF_WRITE) {
7700 if (size != size_default)
7703 bpf_ctx_record_field_size(info, size_default);
7704 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
7712 static bool sk_filter_is_valid_access(int off, int size,
7713 enum bpf_access_type type,
7714 const struct bpf_prog *prog,
7715 struct bpf_insn_access_aux *info)
7718 case bpf_ctx_range(struct __sk_buff, tc_classid):
7719 case bpf_ctx_range(struct __sk_buff, data):
7720 case bpf_ctx_range(struct __sk_buff, data_meta):
7721 case bpf_ctx_range(struct __sk_buff, data_end):
7722 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7723 case bpf_ctx_range(struct __sk_buff, tstamp):
7724 case bpf_ctx_range(struct __sk_buff, wire_len):
7728 if (type == BPF_WRITE) {
7730 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7737 return bpf_skb_is_valid_access(off, size, type, prog, info);
7740 static bool cg_skb_is_valid_access(int off, int size,
7741 enum bpf_access_type type,
7742 const struct bpf_prog *prog,
7743 struct bpf_insn_access_aux *info)
7746 case bpf_ctx_range(struct __sk_buff, tc_classid):
7747 case bpf_ctx_range(struct __sk_buff, data_meta):
7748 case bpf_ctx_range(struct __sk_buff, wire_len):
7750 case bpf_ctx_range(struct __sk_buff, data):
7751 case bpf_ctx_range(struct __sk_buff, data_end):
7757 if (type == BPF_WRITE) {
7759 case bpf_ctx_range(struct __sk_buff, mark):
7760 case bpf_ctx_range(struct __sk_buff, priority):
7761 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7763 case bpf_ctx_range(struct __sk_buff, tstamp):
7773 case bpf_ctx_range(struct __sk_buff, data):
7774 info->reg_type = PTR_TO_PACKET;
7776 case bpf_ctx_range(struct __sk_buff, data_end):
7777 info->reg_type = PTR_TO_PACKET_END;
7781 return bpf_skb_is_valid_access(off, size, type, prog, info);
7784 static bool lwt_is_valid_access(int off, int size,
7785 enum bpf_access_type type,
7786 const struct bpf_prog *prog,
7787 struct bpf_insn_access_aux *info)
7790 case bpf_ctx_range(struct __sk_buff, tc_classid):
7791 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7792 case bpf_ctx_range(struct __sk_buff, data_meta):
7793 case bpf_ctx_range(struct __sk_buff, tstamp):
7794 case bpf_ctx_range(struct __sk_buff, wire_len):
7798 if (type == BPF_WRITE) {
7800 case bpf_ctx_range(struct __sk_buff, mark):
7801 case bpf_ctx_range(struct __sk_buff, priority):
7802 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7810 case bpf_ctx_range(struct __sk_buff, data):
7811 info->reg_type = PTR_TO_PACKET;
7813 case bpf_ctx_range(struct __sk_buff, data_end):
7814 info->reg_type = PTR_TO_PACKET_END;
7818 return bpf_skb_is_valid_access(off, size, type, prog, info);
7821 /* Attach type specific accesses */
7822 static bool __sock_filter_check_attach_type(int off,
7823 enum bpf_access_type access_type,
7824 enum bpf_attach_type attach_type)
7827 case offsetof(struct bpf_sock, bound_dev_if):
7828 case offsetof(struct bpf_sock, mark):
7829 case offsetof(struct bpf_sock, priority):
7830 switch (attach_type) {
7831 case BPF_CGROUP_INET_SOCK_CREATE:
7832 case BPF_CGROUP_INET_SOCK_RELEASE:
7837 case bpf_ctx_range(struct bpf_sock, src_ip4):
7838 switch (attach_type) {
7839 case BPF_CGROUP_INET4_POST_BIND:
7844 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7845 switch (attach_type) {
7846 case BPF_CGROUP_INET6_POST_BIND:
7851 case bpf_ctx_range(struct bpf_sock, src_port):
7852 switch (attach_type) {
7853 case BPF_CGROUP_INET4_POST_BIND:
7854 case BPF_CGROUP_INET6_POST_BIND:
7861 return access_type == BPF_READ;
7866 bool bpf_sock_common_is_valid_access(int off, int size,
7867 enum bpf_access_type type,
7868 struct bpf_insn_access_aux *info)
7871 case bpf_ctx_range_till(struct bpf_sock, type, priority):
7874 return bpf_sock_is_valid_access(off, size, type, info);
7878 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7879 struct bpf_insn_access_aux *info)
7881 const int size_default = sizeof(__u32);
7883 if (off < 0 || off >= sizeof(struct bpf_sock))
7885 if (off % size != 0)
7889 case offsetof(struct bpf_sock, state):
7890 case offsetof(struct bpf_sock, family):
7891 case offsetof(struct bpf_sock, type):
7892 case offsetof(struct bpf_sock, protocol):
7893 case offsetof(struct bpf_sock, dst_port):
7894 case offsetof(struct bpf_sock, src_port):
7895 case offsetof(struct bpf_sock, rx_queue_mapping):
7896 case bpf_ctx_range(struct bpf_sock, src_ip4):
7897 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7898 case bpf_ctx_range(struct bpf_sock, dst_ip4):
7899 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
7900 bpf_ctx_record_field_size(info, size_default);
7901 return bpf_ctx_narrow_access_ok(off, size, size_default);
7904 return size == size_default;
7907 static bool sock_filter_is_valid_access(int off, int size,
7908 enum bpf_access_type type,
7909 const struct bpf_prog *prog,
7910 struct bpf_insn_access_aux *info)
7912 if (!bpf_sock_is_valid_access(off, size, type, info))
7914 return __sock_filter_check_attach_type(off, type,
7915 prog->expected_attach_type);
7918 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
7919 const struct bpf_prog *prog)
7921 /* Neither direct read nor direct write requires any preliminary
7927 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
7928 const struct bpf_prog *prog, int drop_verdict)
7930 struct bpf_insn *insn = insn_buf;
7935 /* if (!skb->cloned)
7938 * (Fast-path, otherwise approximation that we might be
7939 * a clone, do the rest in helper.)
7941 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
7942 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
7943 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
7945 /* ret = bpf_skb_pull_data(skb, 0); */
7946 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
7947 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
7948 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
7949 BPF_FUNC_skb_pull_data);
7952 * return TC_ACT_SHOT;
7954 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
7955 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
7956 *insn++ = BPF_EXIT_INSN();
7959 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
7961 *insn++ = prog->insnsi[0];
7963 return insn - insn_buf;
7966 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
7967 struct bpf_insn *insn_buf)
7969 bool indirect = BPF_MODE(orig->code) == BPF_IND;
7970 struct bpf_insn *insn = insn_buf;
7973 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
7975 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
7977 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
7979 /* We're guaranteed here that CTX is in R6. */
7980 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
7982 switch (BPF_SIZE(orig->code)) {
7984 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
7987 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
7990 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
7994 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
7995 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
7996 *insn++ = BPF_EXIT_INSN();
7998 return insn - insn_buf;
8001 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8002 const struct bpf_prog *prog)
8004 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8007 static bool tc_cls_act_is_valid_access(int off, int size,
8008 enum bpf_access_type type,
8009 const struct bpf_prog *prog,
8010 struct bpf_insn_access_aux *info)
8012 if (type == BPF_WRITE) {
8014 case bpf_ctx_range(struct __sk_buff, mark):
8015 case bpf_ctx_range(struct __sk_buff, tc_index):
8016 case bpf_ctx_range(struct __sk_buff, priority):
8017 case bpf_ctx_range(struct __sk_buff, tc_classid):
8018 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8019 case bpf_ctx_range(struct __sk_buff, tstamp):
8020 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8028 case bpf_ctx_range(struct __sk_buff, data):
8029 info->reg_type = PTR_TO_PACKET;
8031 case bpf_ctx_range(struct __sk_buff, data_meta):
8032 info->reg_type = PTR_TO_PACKET_META;
8034 case bpf_ctx_range(struct __sk_buff, data_end):
8035 info->reg_type = PTR_TO_PACKET_END;
8037 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8041 return bpf_skb_is_valid_access(off, size, type, prog, info);
8044 static bool __is_valid_xdp_access(int off, int size)
8046 if (off < 0 || off >= sizeof(struct xdp_md))
8048 if (off % size != 0)
8050 if (size != sizeof(__u32))
8056 static bool xdp_is_valid_access(int off, int size,
8057 enum bpf_access_type type,
8058 const struct bpf_prog *prog,
8059 struct bpf_insn_access_aux *info)
8061 if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8063 case offsetof(struct xdp_md, egress_ifindex):
8068 if (type == BPF_WRITE) {
8069 if (bpf_prog_is_dev_bound(prog->aux)) {
8071 case offsetof(struct xdp_md, rx_queue_index):
8072 return __is_valid_xdp_access(off, size);
8079 case offsetof(struct xdp_md, data):
8080 info->reg_type = PTR_TO_PACKET;
8082 case offsetof(struct xdp_md, data_meta):
8083 info->reg_type = PTR_TO_PACKET_META;
8085 case offsetof(struct xdp_md, data_end):
8086 info->reg_type = PTR_TO_PACKET_END;
8090 return __is_valid_xdp_access(off, size);
8093 void bpf_warn_invalid_xdp_action(u32 act)
8095 const u32 act_max = XDP_REDIRECT;
8097 WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
8098 act > act_max ? "Illegal" : "Driver unsupported",
8101 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
8103 static bool sock_addr_is_valid_access(int off, int size,
8104 enum bpf_access_type type,
8105 const struct bpf_prog *prog,
8106 struct bpf_insn_access_aux *info)
8108 const int size_default = sizeof(__u32);
8110 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
8112 if (off % size != 0)
8115 /* Disallow access to IPv6 fields from IPv4 contex and vise
8119 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8120 switch (prog->expected_attach_type) {
8121 case BPF_CGROUP_INET4_BIND:
8122 case BPF_CGROUP_INET4_CONNECT:
8123 case BPF_CGROUP_INET4_GETPEERNAME:
8124 case BPF_CGROUP_INET4_GETSOCKNAME:
8125 case BPF_CGROUP_UDP4_SENDMSG:
8126 case BPF_CGROUP_UDP4_RECVMSG:
8132 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8133 switch (prog->expected_attach_type) {
8134 case BPF_CGROUP_INET6_BIND:
8135 case BPF_CGROUP_INET6_CONNECT:
8136 case BPF_CGROUP_INET6_GETPEERNAME:
8137 case BPF_CGROUP_INET6_GETSOCKNAME:
8138 case BPF_CGROUP_UDP6_SENDMSG:
8139 case BPF_CGROUP_UDP6_RECVMSG:
8145 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8146 switch (prog->expected_attach_type) {
8147 case BPF_CGROUP_UDP4_SENDMSG:
8153 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8155 switch (prog->expected_attach_type) {
8156 case BPF_CGROUP_UDP6_SENDMSG:
8165 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8166 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8167 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8168 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8170 case bpf_ctx_range(struct bpf_sock_addr, user_port):
8171 if (type == BPF_READ) {
8172 bpf_ctx_record_field_size(info, size_default);
8174 if (bpf_ctx_wide_access_ok(off, size,
8175 struct bpf_sock_addr,
8179 if (bpf_ctx_wide_access_ok(off, size,
8180 struct bpf_sock_addr,
8184 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8187 if (bpf_ctx_wide_access_ok(off, size,
8188 struct bpf_sock_addr,
8192 if (bpf_ctx_wide_access_ok(off, size,
8193 struct bpf_sock_addr,
8197 if (size != size_default)
8201 case offsetof(struct bpf_sock_addr, sk):
8202 if (type != BPF_READ)
8204 if (size != sizeof(__u64))
8206 info->reg_type = PTR_TO_SOCKET;
8209 if (type == BPF_READ) {
8210 if (size != size_default)
8220 static bool sock_ops_is_valid_access(int off, int size,
8221 enum bpf_access_type type,
8222 const struct bpf_prog *prog,
8223 struct bpf_insn_access_aux *info)
8225 const int size_default = sizeof(__u32);
8227 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
8230 /* The verifier guarantees that size > 0. */
8231 if (off % size != 0)
8234 if (type == BPF_WRITE) {
8236 case offsetof(struct bpf_sock_ops, reply):
8237 case offsetof(struct bpf_sock_ops, sk_txhash):
8238 if (size != size_default)
8246 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
8248 if (size != sizeof(__u64))
8251 case offsetof(struct bpf_sock_ops, sk):
8252 if (size != sizeof(__u64))
8254 info->reg_type = PTR_TO_SOCKET_OR_NULL;
8256 case offsetof(struct bpf_sock_ops, skb_data):
8257 if (size != sizeof(__u64))
8259 info->reg_type = PTR_TO_PACKET;
8261 case offsetof(struct bpf_sock_ops, skb_data_end):
8262 if (size != sizeof(__u64))
8264 info->reg_type = PTR_TO_PACKET_END;
8266 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
8267 bpf_ctx_record_field_size(info, size_default);
8268 return bpf_ctx_narrow_access_ok(off, size,
8271 if (size != size_default)
8280 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
8281 const struct bpf_prog *prog)
8283 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
8286 static bool sk_skb_is_valid_access(int off, int size,
8287 enum bpf_access_type type,
8288 const struct bpf_prog *prog,
8289 struct bpf_insn_access_aux *info)
8292 case bpf_ctx_range(struct __sk_buff, tc_classid):
8293 case bpf_ctx_range(struct __sk_buff, data_meta):
8294 case bpf_ctx_range(struct __sk_buff, tstamp):
8295 case bpf_ctx_range(struct __sk_buff, wire_len):
8299 if (type == BPF_WRITE) {
8301 case bpf_ctx_range(struct __sk_buff, tc_index):
8302 case bpf_ctx_range(struct __sk_buff, priority):
8310 case bpf_ctx_range(struct __sk_buff, mark):
8312 case bpf_ctx_range(struct __sk_buff, data):
8313 info->reg_type = PTR_TO_PACKET;
8315 case bpf_ctx_range(struct __sk_buff, data_end):
8316 info->reg_type = PTR_TO_PACKET_END;
8320 return bpf_skb_is_valid_access(off, size, type, prog, info);
8323 static bool sk_msg_is_valid_access(int off, int size,
8324 enum bpf_access_type type,
8325 const struct bpf_prog *prog,
8326 struct bpf_insn_access_aux *info)
8328 if (type == BPF_WRITE)
8331 if (off % size != 0)
8335 case offsetof(struct sk_msg_md, data):
8336 info->reg_type = PTR_TO_PACKET;
8337 if (size != sizeof(__u64))
8340 case offsetof(struct sk_msg_md, data_end):
8341 info->reg_type = PTR_TO_PACKET_END;
8342 if (size != sizeof(__u64))
8345 case offsetof(struct sk_msg_md, sk):
8346 if (size != sizeof(__u64))
8348 info->reg_type = PTR_TO_SOCKET;
8350 case bpf_ctx_range(struct sk_msg_md, family):
8351 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
8352 case bpf_ctx_range(struct sk_msg_md, local_ip4):
8353 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
8354 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
8355 case bpf_ctx_range(struct sk_msg_md, remote_port):
8356 case bpf_ctx_range(struct sk_msg_md, local_port):
8357 case bpf_ctx_range(struct sk_msg_md, size):
8358 if (size != sizeof(__u32))
8367 static bool flow_dissector_is_valid_access(int off, int size,
8368 enum bpf_access_type type,
8369 const struct bpf_prog *prog,
8370 struct bpf_insn_access_aux *info)
8372 const int size_default = sizeof(__u32);
8374 if (off < 0 || off >= sizeof(struct __sk_buff))
8377 if (type == BPF_WRITE)
8381 case bpf_ctx_range(struct __sk_buff, data):
8382 if (size != size_default)
8384 info->reg_type = PTR_TO_PACKET;
8386 case bpf_ctx_range(struct __sk_buff, data_end):
8387 if (size != size_default)
8389 info->reg_type = PTR_TO_PACKET_END;
8391 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8392 if (size != sizeof(__u64))
8394 info->reg_type = PTR_TO_FLOW_KEYS;
8401 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
8402 const struct bpf_insn *si,
8403 struct bpf_insn *insn_buf,
8404 struct bpf_prog *prog,
8408 struct bpf_insn *insn = insn_buf;
8411 case offsetof(struct __sk_buff, data):
8412 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
8413 si->dst_reg, si->src_reg,
8414 offsetof(struct bpf_flow_dissector, data));
8417 case offsetof(struct __sk_buff, data_end):
8418 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
8419 si->dst_reg, si->src_reg,
8420 offsetof(struct bpf_flow_dissector, data_end));
8423 case offsetof(struct __sk_buff, flow_keys):
8424 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
8425 si->dst_reg, si->src_reg,
8426 offsetof(struct bpf_flow_dissector, flow_keys));
8430 return insn - insn_buf;
8433 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si,
8434 struct bpf_insn *insn)
8436 /* si->dst_reg = skb_shinfo(SKB); */
8437 #ifdef NET_SKBUFF_DATA_USES_OFFSET
8438 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8439 BPF_REG_AX, si->src_reg,
8440 offsetof(struct sk_buff, end));
8441 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
8442 si->dst_reg, si->src_reg,
8443 offsetof(struct sk_buff, head));
8444 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
8446 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8447 si->dst_reg, si->src_reg,
8448 offsetof(struct sk_buff, end));
8454 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
8455 const struct bpf_insn *si,
8456 struct bpf_insn *insn_buf,
8457 struct bpf_prog *prog, u32 *target_size)
8459 struct bpf_insn *insn = insn_buf;
8463 case offsetof(struct __sk_buff, len):
8464 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8465 bpf_target_off(struct sk_buff, len, 4,
8469 case offsetof(struct __sk_buff, protocol):
8470 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8471 bpf_target_off(struct sk_buff, protocol, 2,
8475 case offsetof(struct __sk_buff, vlan_proto):
8476 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8477 bpf_target_off(struct sk_buff, vlan_proto, 2,
8481 case offsetof(struct __sk_buff, priority):
8482 if (type == BPF_WRITE)
8483 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8484 bpf_target_off(struct sk_buff, priority, 4,
8487 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8488 bpf_target_off(struct sk_buff, priority, 4,
8492 case offsetof(struct __sk_buff, ingress_ifindex):
8493 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8494 bpf_target_off(struct sk_buff, skb_iif, 4,
8498 case offsetof(struct __sk_buff, ifindex):
8499 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8500 si->dst_reg, si->src_reg,
8501 offsetof(struct sk_buff, dev));
8502 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
8503 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8504 bpf_target_off(struct net_device, ifindex, 4,
8508 case offsetof(struct __sk_buff, hash):
8509 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8510 bpf_target_off(struct sk_buff, hash, 4,
8514 case offsetof(struct __sk_buff, mark):
8515 if (type == BPF_WRITE)
8516 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8517 bpf_target_off(struct sk_buff, mark, 4,
8520 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8521 bpf_target_off(struct sk_buff, mark, 4,
8525 case offsetof(struct __sk_buff, pkt_type):
8527 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8529 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
8530 #ifdef __BIG_ENDIAN_BITFIELD
8531 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
8535 case offsetof(struct __sk_buff, queue_mapping):
8536 if (type == BPF_WRITE) {
8537 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
8538 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8539 bpf_target_off(struct sk_buff,
8543 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8544 bpf_target_off(struct sk_buff,
8550 case offsetof(struct __sk_buff, vlan_present):
8552 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8553 PKT_VLAN_PRESENT_OFFSET());
8554 if (PKT_VLAN_PRESENT_BIT)
8555 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
8556 if (PKT_VLAN_PRESENT_BIT < 7)
8557 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
8560 case offsetof(struct __sk_buff, vlan_tci):
8561 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8562 bpf_target_off(struct sk_buff, vlan_tci, 2,
8566 case offsetof(struct __sk_buff, cb[0]) ...
8567 offsetofend(struct __sk_buff, cb[4]) - 1:
8568 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
8569 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
8570 offsetof(struct qdisc_skb_cb, data)) %
8573 prog->cb_access = 1;
8575 off -= offsetof(struct __sk_buff, cb[0]);
8576 off += offsetof(struct sk_buff, cb);
8577 off += offsetof(struct qdisc_skb_cb, data);
8578 if (type == BPF_WRITE)
8579 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
8582 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
8586 case offsetof(struct __sk_buff, tc_classid):
8587 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
8590 off -= offsetof(struct __sk_buff, tc_classid);
8591 off += offsetof(struct sk_buff, cb);
8592 off += offsetof(struct qdisc_skb_cb, tc_classid);
8594 if (type == BPF_WRITE)
8595 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
8598 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
8602 case offsetof(struct __sk_buff, data):
8603 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
8604 si->dst_reg, si->src_reg,
8605 offsetof(struct sk_buff, data));
8608 case offsetof(struct __sk_buff, data_meta):
8610 off -= offsetof(struct __sk_buff, data_meta);
8611 off += offsetof(struct sk_buff, cb);
8612 off += offsetof(struct bpf_skb_data_end, data_meta);
8613 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8617 case offsetof(struct __sk_buff, data_end):
8619 off -= offsetof(struct __sk_buff, data_end);
8620 off += offsetof(struct sk_buff, cb);
8621 off += offsetof(struct bpf_skb_data_end, data_end);
8622 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8626 case offsetof(struct __sk_buff, tc_index):
8627 #ifdef CONFIG_NET_SCHED
8628 if (type == BPF_WRITE)
8629 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8630 bpf_target_off(struct sk_buff, tc_index, 2,
8633 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8634 bpf_target_off(struct sk_buff, tc_index, 2,
8638 if (type == BPF_WRITE)
8639 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
8641 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8645 case offsetof(struct __sk_buff, napi_id):
8646 #if defined(CONFIG_NET_RX_BUSY_POLL)
8647 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8648 bpf_target_off(struct sk_buff, napi_id, 4,
8650 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
8651 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8654 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8657 case offsetof(struct __sk_buff, family):
8658 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
8660 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8661 si->dst_reg, si->src_reg,
8662 offsetof(struct sk_buff, sk));
8663 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8664 bpf_target_off(struct sock_common,
8668 case offsetof(struct __sk_buff, remote_ip4):
8669 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
8671 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8672 si->dst_reg, si->src_reg,
8673 offsetof(struct sk_buff, sk));
8674 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8675 bpf_target_off(struct sock_common,
8679 case offsetof(struct __sk_buff, local_ip4):
8680 BUILD_BUG_ON(sizeof_field(struct sock_common,
8681 skc_rcv_saddr) != 4);
8683 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8684 si->dst_reg, si->src_reg,
8685 offsetof(struct sk_buff, sk));
8686 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8687 bpf_target_off(struct sock_common,
8691 case offsetof(struct __sk_buff, remote_ip6[0]) ...
8692 offsetof(struct __sk_buff, remote_ip6[3]):
8693 #if IS_ENABLED(CONFIG_IPV6)
8694 BUILD_BUG_ON(sizeof_field(struct sock_common,
8695 skc_v6_daddr.s6_addr32[0]) != 4);
8698 off -= offsetof(struct __sk_buff, remote_ip6[0]);
8700 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8701 si->dst_reg, si->src_reg,
8702 offsetof(struct sk_buff, sk));
8703 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8704 offsetof(struct sock_common,
8705 skc_v6_daddr.s6_addr32[0]) +
8708 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8711 case offsetof(struct __sk_buff, local_ip6[0]) ...
8712 offsetof(struct __sk_buff, local_ip6[3]):
8713 #if IS_ENABLED(CONFIG_IPV6)
8714 BUILD_BUG_ON(sizeof_field(struct sock_common,
8715 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8718 off -= offsetof(struct __sk_buff, local_ip6[0]);
8720 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8721 si->dst_reg, si->src_reg,
8722 offsetof(struct sk_buff, sk));
8723 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8724 offsetof(struct sock_common,
8725 skc_v6_rcv_saddr.s6_addr32[0]) +
8728 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8732 case offsetof(struct __sk_buff, remote_port):
8733 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
8735 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8736 si->dst_reg, si->src_reg,
8737 offsetof(struct sk_buff, sk));
8738 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8739 bpf_target_off(struct sock_common,
8742 #ifndef __BIG_ENDIAN_BITFIELD
8743 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8747 case offsetof(struct __sk_buff, local_port):
8748 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
8750 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8751 si->dst_reg, si->src_reg,
8752 offsetof(struct sk_buff, sk));
8753 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8754 bpf_target_off(struct sock_common,
8755 skc_num, 2, target_size));
8758 case offsetof(struct __sk_buff, tstamp):
8759 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
8761 if (type == BPF_WRITE)
8762 *insn++ = BPF_STX_MEM(BPF_DW,
8763 si->dst_reg, si->src_reg,
8764 bpf_target_off(struct sk_buff,
8768 *insn++ = BPF_LDX_MEM(BPF_DW,
8769 si->dst_reg, si->src_reg,
8770 bpf_target_off(struct sk_buff,
8775 case offsetof(struct __sk_buff, gso_segs):
8776 insn = bpf_convert_shinfo_access(si, insn);
8777 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
8778 si->dst_reg, si->dst_reg,
8779 bpf_target_off(struct skb_shared_info,
8783 case offsetof(struct __sk_buff, gso_size):
8784 insn = bpf_convert_shinfo_access(si, insn);
8785 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
8786 si->dst_reg, si->dst_reg,
8787 bpf_target_off(struct skb_shared_info,
8791 case offsetof(struct __sk_buff, wire_len):
8792 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
8795 off -= offsetof(struct __sk_buff, wire_len);
8796 off += offsetof(struct sk_buff, cb);
8797 off += offsetof(struct qdisc_skb_cb, pkt_len);
8799 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
8802 case offsetof(struct __sk_buff, sk):
8803 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8804 si->dst_reg, si->src_reg,
8805 offsetof(struct sk_buff, sk));
8809 return insn - insn_buf;
8812 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
8813 const struct bpf_insn *si,
8814 struct bpf_insn *insn_buf,
8815 struct bpf_prog *prog, u32 *target_size)
8817 struct bpf_insn *insn = insn_buf;
8821 case offsetof(struct bpf_sock, bound_dev_if):
8822 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
8824 if (type == BPF_WRITE)
8825 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8826 offsetof(struct sock, sk_bound_dev_if));
8828 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8829 offsetof(struct sock, sk_bound_dev_if));
8832 case offsetof(struct bpf_sock, mark):
8833 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
8835 if (type == BPF_WRITE)
8836 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8837 offsetof(struct sock, sk_mark));
8839 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8840 offsetof(struct sock, sk_mark));
8843 case offsetof(struct bpf_sock, priority):
8844 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
8846 if (type == BPF_WRITE)
8847 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8848 offsetof(struct sock, sk_priority));
8850 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8851 offsetof(struct sock, sk_priority));
8854 case offsetof(struct bpf_sock, family):
8855 *insn++ = BPF_LDX_MEM(
8856 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
8857 si->dst_reg, si->src_reg,
8858 bpf_target_off(struct sock_common,
8860 sizeof_field(struct sock_common,
8865 case offsetof(struct bpf_sock, type):
8866 *insn++ = BPF_LDX_MEM(
8867 BPF_FIELD_SIZEOF(struct sock, sk_type),
8868 si->dst_reg, si->src_reg,
8869 bpf_target_off(struct sock, sk_type,
8870 sizeof_field(struct sock, sk_type),
8874 case offsetof(struct bpf_sock, protocol):
8875 *insn++ = BPF_LDX_MEM(
8876 BPF_FIELD_SIZEOF(struct sock, sk_protocol),
8877 si->dst_reg, si->src_reg,
8878 bpf_target_off(struct sock, sk_protocol,
8879 sizeof_field(struct sock, sk_protocol),
8883 case offsetof(struct bpf_sock, src_ip4):
8884 *insn++ = BPF_LDX_MEM(
8885 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8886 bpf_target_off(struct sock_common, skc_rcv_saddr,
8887 sizeof_field(struct sock_common,
8892 case offsetof(struct bpf_sock, dst_ip4):
8893 *insn++ = BPF_LDX_MEM(
8894 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8895 bpf_target_off(struct sock_common, skc_daddr,
8896 sizeof_field(struct sock_common,
8901 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8902 #if IS_ENABLED(CONFIG_IPV6)
8904 off -= offsetof(struct bpf_sock, src_ip6[0]);
8905 *insn++ = BPF_LDX_MEM(
8906 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8909 skc_v6_rcv_saddr.s6_addr32[0],
8910 sizeof_field(struct sock_common,
8911 skc_v6_rcv_saddr.s6_addr32[0]),
8912 target_size) + off);
8915 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8919 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8920 #if IS_ENABLED(CONFIG_IPV6)
8922 off -= offsetof(struct bpf_sock, dst_ip6[0]);
8923 *insn++ = BPF_LDX_MEM(
8924 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8925 bpf_target_off(struct sock_common,
8926 skc_v6_daddr.s6_addr32[0],
8927 sizeof_field(struct sock_common,
8928 skc_v6_daddr.s6_addr32[0]),
8929 target_size) + off);
8931 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8936 case offsetof(struct bpf_sock, src_port):
8937 *insn++ = BPF_LDX_MEM(
8938 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
8939 si->dst_reg, si->src_reg,
8940 bpf_target_off(struct sock_common, skc_num,
8941 sizeof_field(struct sock_common,
8946 case offsetof(struct bpf_sock, dst_port):
8947 *insn++ = BPF_LDX_MEM(
8948 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
8949 si->dst_reg, si->src_reg,
8950 bpf_target_off(struct sock_common, skc_dport,
8951 sizeof_field(struct sock_common,
8956 case offsetof(struct bpf_sock, state):
8957 *insn++ = BPF_LDX_MEM(
8958 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
8959 si->dst_reg, si->src_reg,
8960 bpf_target_off(struct sock_common, skc_state,
8961 sizeof_field(struct sock_common,
8965 case offsetof(struct bpf_sock, rx_queue_mapping):
8966 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
8967 *insn++ = BPF_LDX_MEM(
8968 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
8969 si->dst_reg, si->src_reg,
8970 bpf_target_off(struct sock, sk_rx_queue_mapping,
8971 sizeof_field(struct sock,
8972 sk_rx_queue_mapping),
8974 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
8976 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
8978 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
8984 return insn - insn_buf;
8987 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
8988 const struct bpf_insn *si,
8989 struct bpf_insn *insn_buf,
8990 struct bpf_prog *prog, u32 *target_size)
8992 struct bpf_insn *insn = insn_buf;
8995 case offsetof(struct __sk_buff, ifindex):
8996 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8997 si->dst_reg, si->src_reg,
8998 offsetof(struct sk_buff, dev));
8999 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9000 bpf_target_off(struct net_device, ifindex, 4,
9004 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9008 return insn - insn_buf;
9011 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
9012 const struct bpf_insn *si,
9013 struct bpf_insn *insn_buf,
9014 struct bpf_prog *prog, u32 *target_size)
9016 struct bpf_insn *insn = insn_buf;
9019 case offsetof(struct xdp_md, data):
9020 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
9021 si->dst_reg, si->src_reg,
9022 offsetof(struct xdp_buff, data));
9024 case offsetof(struct xdp_md, data_meta):
9025 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
9026 si->dst_reg, si->src_reg,
9027 offsetof(struct xdp_buff, data_meta));
9029 case offsetof(struct xdp_md, data_end):
9030 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
9031 si->dst_reg, si->src_reg,
9032 offsetof(struct xdp_buff, data_end));
9034 case offsetof(struct xdp_md, ingress_ifindex):
9035 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9036 si->dst_reg, si->src_reg,
9037 offsetof(struct xdp_buff, rxq));
9038 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
9039 si->dst_reg, si->dst_reg,
9040 offsetof(struct xdp_rxq_info, dev));
9041 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9042 offsetof(struct net_device, ifindex));
9044 case offsetof(struct xdp_md, rx_queue_index):
9045 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9046 si->dst_reg, si->src_reg,
9047 offsetof(struct xdp_buff, rxq));
9048 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9049 offsetof(struct xdp_rxq_info,
9052 case offsetof(struct xdp_md, egress_ifindex):
9053 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
9054 si->dst_reg, si->src_reg,
9055 offsetof(struct xdp_buff, txq));
9056 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
9057 si->dst_reg, si->dst_reg,
9058 offsetof(struct xdp_txq_info, dev));
9059 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9060 offsetof(struct net_device, ifindex));
9064 return insn - insn_buf;
9067 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
9068 * context Structure, F is Field in context structure that contains a pointer
9069 * to Nested Structure of type NS that has the field NF.
9071 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
9072 * sure that SIZE is not greater than actual size of S.F.NF.
9074 * If offset OFF is provided, the load happens from that offset relative to
9077 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
9079 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
9080 si->src_reg, offsetof(S, F)); \
9081 *insn++ = BPF_LDX_MEM( \
9082 SIZE, si->dst_reg, si->dst_reg, \
9083 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
9088 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
9089 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
9090 BPF_FIELD_SIZEOF(NS, NF), 0)
9092 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
9093 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
9095 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
9096 * "register" since two registers available in convert_ctx_access are not
9097 * enough: we can't override neither SRC, since it contains value to store, nor
9098 * DST since it contains pointer to context that may be used by later
9099 * instructions. But we need a temporary place to save pointer to nested
9100 * structure whose field we want to store to.
9102 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
9104 int tmp_reg = BPF_REG_9; \
9105 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
9107 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
9109 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
9111 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
9112 si->dst_reg, offsetof(S, F)); \
9113 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \
9114 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
9117 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
9121 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
9124 if (type == BPF_WRITE) { \
9125 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
9128 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
9129 S, NS, F, NF, SIZE, OFF); \
9133 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
9134 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
9135 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
9137 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
9138 const struct bpf_insn *si,
9139 struct bpf_insn *insn_buf,
9140 struct bpf_prog *prog, u32 *target_size)
9142 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
9143 struct bpf_insn *insn = insn_buf;
9146 case offsetof(struct bpf_sock_addr, user_family):
9147 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9148 struct sockaddr, uaddr, sa_family);
9151 case offsetof(struct bpf_sock_addr, user_ip4):
9152 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9153 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
9154 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
9157 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9159 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
9160 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9161 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9162 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
9166 case offsetof(struct bpf_sock_addr, user_port):
9167 /* To get port we need to know sa_family first and then treat
9168 * sockaddr as either sockaddr_in or sockaddr_in6.
9169 * Though we can simplify since port field has same offset and
9170 * size in both structures.
9171 * Here we check this invariant and use just one of the
9172 * structures if it's true.
9174 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
9175 offsetof(struct sockaddr_in6, sin6_port));
9176 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
9177 sizeof_field(struct sockaddr_in6, sin6_port));
9178 /* Account for sin6_port being smaller than user_port. */
9179 port_size = min(port_size, BPF_LDST_BYTES(si));
9180 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9181 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9182 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
9185 case offsetof(struct bpf_sock_addr, family):
9186 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9187 struct sock, sk, sk_family);
9190 case offsetof(struct bpf_sock_addr, type):
9191 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9192 struct sock, sk, sk_type);
9195 case offsetof(struct bpf_sock_addr, protocol):
9196 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9197 struct sock, sk, sk_protocol);
9200 case offsetof(struct bpf_sock_addr, msg_src_ip4):
9201 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
9202 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9203 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
9204 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
9207 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9210 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
9211 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
9212 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9213 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
9214 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
9216 case offsetof(struct bpf_sock_addr, sk):
9217 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
9218 si->dst_reg, si->src_reg,
9219 offsetof(struct bpf_sock_addr_kern, sk));
9223 return insn - insn_buf;
9226 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
9227 const struct bpf_insn *si,
9228 struct bpf_insn *insn_buf,
9229 struct bpf_prog *prog,
9232 struct bpf_insn *insn = insn_buf;
9235 /* Helper macro for adding read access to tcp_sock or sock fields. */
9236 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
9238 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \
9239 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
9240 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
9241 if (si->dst_reg == reg || si->src_reg == reg) \
9243 if (si->dst_reg == reg || si->src_reg == reg) \
9245 if (si->dst_reg == si->src_reg) { \
9246 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
9247 offsetof(struct bpf_sock_ops_kern, \
9249 fullsock_reg = reg; \
9252 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9253 struct bpf_sock_ops_kern, \
9255 fullsock_reg, si->src_reg, \
9256 offsetof(struct bpf_sock_ops_kern, \
9258 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
9259 if (si->dst_reg == si->src_reg) \
9260 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9261 offsetof(struct bpf_sock_ops_kern, \
9263 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9264 struct bpf_sock_ops_kern, sk),\
9265 si->dst_reg, si->src_reg, \
9266 offsetof(struct bpf_sock_ops_kern, sk));\
9267 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
9269 si->dst_reg, si->dst_reg, \
9270 offsetof(OBJ, OBJ_FIELD)); \
9271 if (si->dst_reg == si->src_reg) { \
9272 *insn++ = BPF_JMP_A(1); \
9273 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9274 offsetof(struct bpf_sock_ops_kern, \
9279 #define SOCK_OPS_GET_SK() \
9281 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \
9282 if (si->dst_reg == reg || si->src_reg == reg) \
9284 if (si->dst_reg == reg || si->src_reg == reg) \
9286 if (si->dst_reg == si->src_reg) { \
9287 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
9288 offsetof(struct bpf_sock_ops_kern, \
9290 fullsock_reg = reg; \
9293 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9294 struct bpf_sock_ops_kern, \
9296 fullsock_reg, si->src_reg, \
9297 offsetof(struct bpf_sock_ops_kern, \
9299 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
9300 if (si->dst_reg == si->src_reg) \
9301 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9302 offsetof(struct bpf_sock_ops_kern, \
9304 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9305 struct bpf_sock_ops_kern, sk),\
9306 si->dst_reg, si->src_reg, \
9307 offsetof(struct bpf_sock_ops_kern, sk));\
9308 if (si->dst_reg == si->src_reg) { \
9309 *insn++ = BPF_JMP_A(1); \
9310 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9311 offsetof(struct bpf_sock_ops_kern, \
9316 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
9317 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
9319 /* Helper macro for adding write access to tcp_sock or sock fields.
9320 * The macro is called with two registers, dst_reg which contains a pointer
9321 * to ctx (context) and src_reg which contains the value that should be
9322 * stored. However, we need an additional register since we cannot overwrite
9323 * dst_reg because it may be used later in the program.
9324 * Instead we "borrow" one of the other register. We first save its value
9325 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
9326 * it at the end of the macro.
9328 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
9330 int reg = BPF_REG_9; \
9331 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
9332 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
9333 if (si->dst_reg == reg || si->src_reg == reg) \
9335 if (si->dst_reg == reg || si->src_reg == reg) \
9337 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
9338 offsetof(struct bpf_sock_ops_kern, \
9340 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9341 struct bpf_sock_ops_kern, \
9344 offsetof(struct bpf_sock_ops_kern, \
9346 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
9347 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9348 struct bpf_sock_ops_kern, sk),\
9350 offsetof(struct bpf_sock_ops_kern, sk));\
9351 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \
9353 offsetof(OBJ, OBJ_FIELD)); \
9354 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
9355 offsetof(struct bpf_sock_ops_kern, \
9359 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
9361 if (TYPE == BPF_WRITE) \
9362 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
9364 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
9367 if (insn > insn_buf)
9368 return insn - insn_buf;
9371 case offsetof(struct bpf_sock_ops, op):
9372 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9374 si->dst_reg, si->src_reg,
9375 offsetof(struct bpf_sock_ops_kern, op));
9378 case offsetof(struct bpf_sock_ops, replylong[0]) ...
9379 offsetof(struct bpf_sock_ops, replylong[3]):
9380 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
9381 sizeof_field(struct bpf_sock_ops_kern, reply));
9382 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
9383 sizeof_field(struct bpf_sock_ops_kern, replylong));
9385 off -= offsetof(struct bpf_sock_ops, replylong[0]);
9386 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
9387 if (type == BPF_WRITE)
9388 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9391 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9395 case offsetof(struct bpf_sock_ops, family):
9396 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9398 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9399 struct bpf_sock_ops_kern, sk),
9400 si->dst_reg, si->src_reg,
9401 offsetof(struct bpf_sock_ops_kern, sk));
9402 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9403 offsetof(struct sock_common, skc_family));
9406 case offsetof(struct bpf_sock_ops, remote_ip4):
9407 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9409 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9410 struct bpf_sock_ops_kern, sk),
9411 si->dst_reg, si->src_reg,
9412 offsetof(struct bpf_sock_ops_kern, sk));
9413 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9414 offsetof(struct sock_common, skc_daddr));
9417 case offsetof(struct bpf_sock_ops, local_ip4):
9418 BUILD_BUG_ON(sizeof_field(struct sock_common,
9419 skc_rcv_saddr) != 4);
9421 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9422 struct bpf_sock_ops_kern, sk),
9423 si->dst_reg, si->src_reg,
9424 offsetof(struct bpf_sock_ops_kern, sk));
9425 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9426 offsetof(struct sock_common,
9430 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
9431 offsetof(struct bpf_sock_ops, remote_ip6[3]):
9432 #if IS_ENABLED(CONFIG_IPV6)
9433 BUILD_BUG_ON(sizeof_field(struct sock_common,
9434 skc_v6_daddr.s6_addr32[0]) != 4);
9437 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
9438 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9439 struct bpf_sock_ops_kern, sk),
9440 si->dst_reg, si->src_reg,
9441 offsetof(struct bpf_sock_ops_kern, sk));
9442 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9443 offsetof(struct sock_common,
9444 skc_v6_daddr.s6_addr32[0]) +
9447 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9451 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
9452 offsetof(struct bpf_sock_ops, local_ip6[3]):
9453 #if IS_ENABLED(CONFIG_IPV6)
9454 BUILD_BUG_ON(sizeof_field(struct sock_common,
9455 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9458 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
9459 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9460 struct bpf_sock_ops_kern, sk),
9461 si->dst_reg, si->src_reg,
9462 offsetof(struct bpf_sock_ops_kern, sk));
9463 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9464 offsetof(struct sock_common,
9465 skc_v6_rcv_saddr.s6_addr32[0]) +
9468 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9472 case offsetof(struct bpf_sock_ops, remote_port):
9473 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9475 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9476 struct bpf_sock_ops_kern, sk),
9477 si->dst_reg, si->src_reg,
9478 offsetof(struct bpf_sock_ops_kern, sk));
9479 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9480 offsetof(struct sock_common, skc_dport));
9481 #ifndef __BIG_ENDIAN_BITFIELD
9482 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9486 case offsetof(struct bpf_sock_ops, local_port):
9487 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9489 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9490 struct bpf_sock_ops_kern, sk),
9491 si->dst_reg, si->src_reg,
9492 offsetof(struct bpf_sock_ops_kern, sk));
9493 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9494 offsetof(struct sock_common, skc_num));
9497 case offsetof(struct bpf_sock_ops, is_fullsock):
9498 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9499 struct bpf_sock_ops_kern,
9501 si->dst_reg, si->src_reg,
9502 offsetof(struct bpf_sock_ops_kern,
9506 case offsetof(struct bpf_sock_ops, state):
9507 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
9509 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9510 struct bpf_sock_ops_kern, sk),
9511 si->dst_reg, si->src_reg,
9512 offsetof(struct bpf_sock_ops_kern, sk));
9513 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
9514 offsetof(struct sock_common, skc_state));
9517 case offsetof(struct bpf_sock_ops, rtt_min):
9518 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
9519 sizeof(struct minmax));
9520 BUILD_BUG_ON(sizeof(struct minmax) <
9521 sizeof(struct minmax_sample));
9523 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9524 struct bpf_sock_ops_kern, sk),
9525 si->dst_reg, si->src_reg,
9526 offsetof(struct bpf_sock_ops_kern, sk));
9527 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9528 offsetof(struct tcp_sock, rtt_min) +
9529 sizeof_field(struct minmax_sample, t));
9532 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
9533 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
9537 case offsetof(struct bpf_sock_ops, sk_txhash):
9538 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
9541 case offsetof(struct bpf_sock_ops, snd_cwnd):
9542 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
9544 case offsetof(struct bpf_sock_ops, srtt_us):
9545 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
9547 case offsetof(struct bpf_sock_ops, snd_ssthresh):
9548 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
9550 case offsetof(struct bpf_sock_ops, rcv_nxt):
9551 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
9553 case offsetof(struct bpf_sock_ops, snd_nxt):
9554 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
9556 case offsetof(struct bpf_sock_ops, snd_una):
9557 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
9559 case offsetof(struct bpf_sock_ops, mss_cache):
9560 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
9562 case offsetof(struct bpf_sock_ops, ecn_flags):
9563 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
9565 case offsetof(struct bpf_sock_ops, rate_delivered):
9566 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
9568 case offsetof(struct bpf_sock_ops, rate_interval_us):
9569 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
9571 case offsetof(struct bpf_sock_ops, packets_out):
9572 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
9574 case offsetof(struct bpf_sock_ops, retrans_out):
9575 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
9577 case offsetof(struct bpf_sock_ops, total_retrans):
9578 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
9580 case offsetof(struct bpf_sock_ops, segs_in):
9581 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
9583 case offsetof(struct bpf_sock_ops, data_segs_in):
9584 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
9586 case offsetof(struct bpf_sock_ops, segs_out):
9587 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
9589 case offsetof(struct bpf_sock_ops, data_segs_out):
9590 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
9592 case offsetof(struct bpf_sock_ops, lost_out):
9593 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
9595 case offsetof(struct bpf_sock_ops, sacked_out):
9596 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
9598 case offsetof(struct bpf_sock_ops, bytes_received):
9599 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
9601 case offsetof(struct bpf_sock_ops, bytes_acked):
9602 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
9604 case offsetof(struct bpf_sock_ops, sk):
9607 case offsetof(struct bpf_sock_ops, skb_data_end):
9608 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9610 si->dst_reg, si->src_reg,
9611 offsetof(struct bpf_sock_ops_kern,
9614 case offsetof(struct bpf_sock_ops, skb_data):
9615 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9617 si->dst_reg, si->src_reg,
9618 offsetof(struct bpf_sock_ops_kern,
9620 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9621 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9622 si->dst_reg, si->dst_reg,
9623 offsetof(struct sk_buff, data));
9625 case offsetof(struct bpf_sock_ops, skb_len):
9626 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9628 si->dst_reg, si->src_reg,
9629 offsetof(struct bpf_sock_ops_kern,
9631 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9632 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9633 si->dst_reg, si->dst_reg,
9634 offsetof(struct sk_buff, len));
9636 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9637 off = offsetof(struct sk_buff, cb);
9638 off += offsetof(struct tcp_skb_cb, tcp_flags);
9639 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
9640 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9642 si->dst_reg, si->src_reg,
9643 offsetof(struct bpf_sock_ops_kern,
9645 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9646 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
9648 si->dst_reg, si->dst_reg, off);
9651 return insn - insn_buf;
9654 /* data_end = skb->data + skb_headlen() */
9655 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
9656 struct bpf_insn *insn)
9658 /* si->dst_reg = skb->data */
9659 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9660 si->dst_reg, si->src_reg,
9661 offsetof(struct sk_buff, data));
9663 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9664 BPF_REG_AX, si->src_reg,
9665 offsetof(struct sk_buff, len));
9666 /* si->dst_reg = skb->data + skb->len */
9667 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
9668 /* AX = skb->data_len */
9669 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
9670 BPF_REG_AX, si->src_reg,
9671 offsetof(struct sk_buff, data_len));
9672 /* si->dst_reg = skb->data + skb->len - skb->data_len */
9673 *insn++ = BPF_ALU64_REG(BPF_SUB, si->dst_reg, BPF_REG_AX);
9678 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
9679 const struct bpf_insn *si,
9680 struct bpf_insn *insn_buf,
9681 struct bpf_prog *prog, u32 *target_size)
9683 struct bpf_insn *insn = insn_buf;
9686 case offsetof(struct __sk_buff, data_end):
9687 insn = bpf_convert_data_end_access(si, insn);
9690 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9694 return insn - insn_buf;
9697 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
9698 const struct bpf_insn *si,
9699 struct bpf_insn *insn_buf,
9700 struct bpf_prog *prog, u32 *target_size)
9702 struct bpf_insn *insn = insn_buf;
9703 #if IS_ENABLED(CONFIG_IPV6)
9707 /* convert ctx uses the fact sg element is first in struct */
9708 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
9711 case offsetof(struct sk_msg_md, data):
9712 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
9713 si->dst_reg, si->src_reg,
9714 offsetof(struct sk_msg, data));
9716 case offsetof(struct sk_msg_md, data_end):
9717 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
9718 si->dst_reg, si->src_reg,
9719 offsetof(struct sk_msg, data_end));
9721 case offsetof(struct sk_msg_md, family):
9722 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9724 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9726 si->dst_reg, si->src_reg,
9727 offsetof(struct sk_msg, sk));
9728 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9729 offsetof(struct sock_common, skc_family));
9732 case offsetof(struct sk_msg_md, remote_ip4):
9733 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9735 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9737 si->dst_reg, si->src_reg,
9738 offsetof(struct sk_msg, sk));
9739 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9740 offsetof(struct sock_common, skc_daddr));
9743 case offsetof(struct sk_msg_md, local_ip4):
9744 BUILD_BUG_ON(sizeof_field(struct sock_common,
9745 skc_rcv_saddr) != 4);
9747 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9749 si->dst_reg, si->src_reg,
9750 offsetof(struct sk_msg, sk));
9751 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9752 offsetof(struct sock_common,
9756 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
9757 offsetof(struct sk_msg_md, remote_ip6[3]):
9758 #if IS_ENABLED(CONFIG_IPV6)
9759 BUILD_BUG_ON(sizeof_field(struct sock_common,
9760 skc_v6_daddr.s6_addr32[0]) != 4);
9763 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
9764 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9766 si->dst_reg, si->src_reg,
9767 offsetof(struct sk_msg, sk));
9768 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9769 offsetof(struct sock_common,
9770 skc_v6_daddr.s6_addr32[0]) +
9773 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9777 case offsetof(struct sk_msg_md, local_ip6[0]) ...
9778 offsetof(struct sk_msg_md, local_ip6[3]):
9779 #if IS_ENABLED(CONFIG_IPV6)
9780 BUILD_BUG_ON(sizeof_field(struct sock_common,
9781 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9784 off -= offsetof(struct sk_msg_md, local_ip6[0]);
9785 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9787 si->dst_reg, si->src_reg,
9788 offsetof(struct sk_msg, sk));
9789 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9790 offsetof(struct sock_common,
9791 skc_v6_rcv_saddr.s6_addr32[0]) +
9794 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9798 case offsetof(struct sk_msg_md, remote_port):
9799 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9801 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9803 si->dst_reg, si->src_reg,
9804 offsetof(struct sk_msg, sk));
9805 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9806 offsetof(struct sock_common, skc_dport));
9807 #ifndef __BIG_ENDIAN_BITFIELD
9808 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9812 case offsetof(struct sk_msg_md, local_port):
9813 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9815 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9817 si->dst_reg, si->src_reg,
9818 offsetof(struct sk_msg, sk));
9819 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9820 offsetof(struct sock_common, skc_num));
9823 case offsetof(struct sk_msg_md, size):
9824 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
9825 si->dst_reg, si->src_reg,
9826 offsetof(struct sk_msg_sg, size));
9829 case offsetof(struct sk_msg_md, sk):
9830 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
9831 si->dst_reg, si->src_reg,
9832 offsetof(struct sk_msg, sk));
9836 return insn - insn_buf;
9839 const struct bpf_verifier_ops sk_filter_verifier_ops = {
9840 .get_func_proto = sk_filter_func_proto,
9841 .is_valid_access = sk_filter_is_valid_access,
9842 .convert_ctx_access = bpf_convert_ctx_access,
9843 .gen_ld_abs = bpf_gen_ld_abs,
9846 const struct bpf_prog_ops sk_filter_prog_ops = {
9847 .test_run = bpf_prog_test_run_skb,
9850 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
9851 .get_func_proto = tc_cls_act_func_proto,
9852 .is_valid_access = tc_cls_act_is_valid_access,
9853 .convert_ctx_access = tc_cls_act_convert_ctx_access,
9854 .gen_prologue = tc_cls_act_prologue,
9855 .gen_ld_abs = bpf_gen_ld_abs,
9856 .check_kfunc_call = bpf_prog_test_check_kfunc_call,
9859 const struct bpf_prog_ops tc_cls_act_prog_ops = {
9860 .test_run = bpf_prog_test_run_skb,
9863 const struct bpf_verifier_ops xdp_verifier_ops = {
9864 .get_func_proto = xdp_func_proto,
9865 .is_valid_access = xdp_is_valid_access,
9866 .convert_ctx_access = xdp_convert_ctx_access,
9867 .gen_prologue = bpf_noop_prologue,
9870 const struct bpf_prog_ops xdp_prog_ops = {
9871 .test_run = bpf_prog_test_run_xdp,
9874 const struct bpf_verifier_ops cg_skb_verifier_ops = {
9875 .get_func_proto = cg_skb_func_proto,
9876 .is_valid_access = cg_skb_is_valid_access,
9877 .convert_ctx_access = bpf_convert_ctx_access,
9880 const struct bpf_prog_ops cg_skb_prog_ops = {
9881 .test_run = bpf_prog_test_run_skb,
9884 const struct bpf_verifier_ops lwt_in_verifier_ops = {
9885 .get_func_proto = lwt_in_func_proto,
9886 .is_valid_access = lwt_is_valid_access,
9887 .convert_ctx_access = bpf_convert_ctx_access,
9890 const struct bpf_prog_ops lwt_in_prog_ops = {
9891 .test_run = bpf_prog_test_run_skb,
9894 const struct bpf_verifier_ops lwt_out_verifier_ops = {
9895 .get_func_proto = lwt_out_func_proto,
9896 .is_valid_access = lwt_is_valid_access,
9897 .convert_ctx_access = bpf_convert_ctx_access,
9900 const struct bpf_prog_ops lwt_out_prog_ops = {
9901 .test_run = bpf_prog_test_run_skb,
9904 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
9905 .get_func_proto = lwt_xmit_func_proto,
9906 .is_valid_access = lwt_is_valid_access,
9907 .convert_ctx_access = bpf_convert_ctx_access,
9908 .gen_prologue = tc_cls_act_prologue,
9911 const struct bpf_prog_ops lwt_xmit_prog_ops = {
9912 .test_run = bpf_prog_test_run_skb,
9915 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
9916 .get_func_proto = lwt_seg6local_func_proto,
9917 .is_valid_access = lwt_is_valid_access,
9918 .convert_ctx_access = bpf_convert_ctx_access,
9921 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
9922 .test_run = bpf_prog_test_run_skb,
9925 const struct bpf_verifier_ops cg_sock_verifier_ops = {
9926 .get_func_proto = sock_filter_func_proto,
9927 .is_valid_access = sock_filter_is_valid_access,
9928 .convert_ctx_access = bpf_sock_convert_ctx_access,
9931 const struct bpf_prog_ops cg_sock_prog_ops = {
9934 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
9935 .get_func_proto = sock_addr_func_proto,
9936 .is_valid_access = sock_addr_is_valid_access,
9937 .convert_ctx_access = sock_addr_convert_ctx_access,
9940 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
9943 const struct bpf_verifier_ops sock_ops_verifier_ops = {
9944 .get_func_proto = sock_ops_func_proto,
9945 .is_valid_access = sock_ops_is_valid_access,
9946 .convert_ctx_access = sock_ops_convert_ctx_access,
9949 const struct bpf_prog_ops sock_ops_prog_ops = {
9952 const struct bpf_verifier_ops sk_skb_verifier_ops = {
9953 .get_func_proto = sk_skb_func_proto,
9954 .is_valid_access = sk_skb_is_valid_access,
9955 .convert_ctx_access = sk_skb_convert_ctx_access,
9956 .gen_prologue = sk_skb_prologue,
9959 const struct bpf_prog_ops sk_skb_prog_ops = {
9962 const struct bpf_verifier_ops sk_msg_verifier_ops = {
9963 .get_func_proto = sk_msg_func_proto,
9964 .is_valid_access = sk_msg_is_valid_access,
9965 .convert_ctx_access = sk_msg_convert_ctx_access,
9966 .gen_prologue = bpf_noop_prologue,
9969 const struct bpf_prog_ops sk_msg_prog_ops = {
9972 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
9973 .get_func_proto = flow_dissector_func_proto,
9974 .is_valid_access = flow_dissector_is_valid_access,
9975 .convert_ctx_access = flow_dissector_convert_ctx_access,
9978 const struct bpf_prog_ops flow_dissector_prog_ops = {
9979 .test_run = bpf_prog_test_run_flow_dissector,
9982 int sk_detach_filter(struct sock *sk)
9985 struct sk_filter *filter;
9987 if (sock_flag(sk, SOCK_FILTER_LOCKED))
9990 filter = rcu_dereference_protected(sk->sk_filter,
9991 lockdep_sock_is_held(sk));
9993 RCU_INIT_POINTER(sk->sk_filter, NULL);
9994 sk_filter_uncharge(sk, filter);
10000 EXPORT_SYMBOL_GPL(sk_detach_filter);
10002 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
10005 struct sock_fprog_kern *fprog;
10006 struct sk_filter *filter;
10010 filter = rcu_dereference_protected(sk->sk_filter,
10011 lockdep_sock_is_held(sk));
10015 /* We're copying the filter that has been originally attached,
10016 * so no conversion/decode needed anymore. eBPF programs that
10017 * have no original program cannot be dumped through this.
10020 fprog = filter->prog->orig_prog;
10026 /* User space only enquires number of filter blocks. */
10030 if (len < fprog->len)
10034 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
10037 /* Instead of bytes, the API requests to return the number
10038 * of filter blocks.
10047 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
10048 struct sock_reuseport *reuse,
10049 struct sock *sk, struct sk_buff *skb,
10050 struct sock *migrating_sk,
10053 reuse_kern->skb = skb;
10054 reuse_kern->sk = sk;
10055 reuse_kern->selected_sk = NULL;
10056 reuse_kern->migrating_sk = migrating_sk;
10057 reuse_kern->data_end = skb->data + skb_headlen(skb);
10058 reuse_kern->hash = hash;
10059 reuse_kern->reuseport_id = reuse->reuseport_id;
10060 reuse_kern->bind_inany = reuse->bind_inany;
10063 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
10064 struct bpf_prog *prog, struct sk_buff *skb,
10065 struct sock *migrating_sk,
10068 struct sk_reuseport_kern reuse_kern;
10069 enum sk_action action;
10071 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
10072 action = BPF_PROG_RUN(prog, &reuse_kern);
10074 if (action == SK_PASS)
10075 return reuse_kern.selected_sk;
10077 return ERR_PTR(-ECONNREFUSED);
10080 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
10081 struct bpf_map *, map, void *, key, u32, flags)
10083 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
10084 struct sock_reuseport *reuse;
10085 struct sock *selected_sk;
10087 selected_sk = map->ops->map_lookup_elem(map, key);
10091 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
10093 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
10094 if (sk_is_refcounted(selected_sk))
10095 sock_put(selected_sk);
10097 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
10098 * The only (!reuse) case here is - the sk has already been
10099 * unhashed (e.g. by close()), so treat it as -ENOENT.
10101 * Other maps (e.g. sock_map) do not provide this guarantee and
10102 * the sk may never be in the reuseport group to begin with.
10104 return is_sockarray ? -ENOENT : -EINVAL;
10107 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
10108 struct sock *sk = reuse_kern->sk;
10110 if (sk->sk_protocol != selected_sk->sk_protocol)
10111 return -EPROTOTYPE;
10112 else if (sk->sk_family != selected_sk->sk_family)
10113 return -EAFNOSUPPORT;
10115 /* Catch all. Likely bound to a different sockaddr. */
10119 reuse_kern->selected_sk = selected_sk;
10124 static const struct bpf_func_proto sk_select_reuseport_proto = {
10125 .func = sk_select_reuseport,
10127 .ret_type = RET_INTEGER,
10128 .arg1_type = ARG_PTR_TO_CTX,
10129 .arg2_type = ARG_CONST_MAP_PTR,
10130 .arg3_type = ARG_PTR_TO_MAP_KEY,
10131 .arg4_type = ARG_ANYTHING,
10134 BPF_CALL_4(sk_reuseport_load_bytes,
10135 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10136 void *, to, u32, len)
10138 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
10141 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
10142 .func = sk_reuseport_load_bytes,
10144 .ret_type = RET_INTEGER,
10145 .arg1_type = ARG_PTR_TO_CTX,
10146 .arg2_type = ARG_ANYTHING,
10147 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
10148 .arg4_type = ARG_CONST_SIZE,
10151 BPF_CALL_5(sk_reuseport_load_bytes_relative,
10152 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10153 void *, to, u32, len, u32, start_header)
10155 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
10156 len, start_header);
10159 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
10160 .func = sk_reuseport_load_bytes_relative,
10162 .ret_type = RET_INTEGER,
10163 .arg1_type = ARG_PTR_TO_CTX,
10164 .arg2_type = ARG_ANYTHING,
10165 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
10166 .arg4_type = ARG_CONST_SIZE,
10167 .arg5_type = ARG_ANYTHING,
10170 static const struct bpf_func_proto *
10171 sk_reuseport_func_proto(enum bpf_func_id func_id,
10172 const struct bpf_prog *prog)
10175 case BPF_FUNC_sk_select_reuseport:
10176 return &sk_select_reuseport_proto;
10177 case BPF_FUNC_skb_load_bytes:
10178 return &sk_reuseport_load_bytes_proto;
10179 case BPF_FUNC_skb_load_bytes_relative:
10180 return &sk_reuseport_load_bytes_relative_proto;
10181 case BPF_FUNC_get_socket_cookie:
10182 return &bpf_get_socket_ptr_cookie_proto;
10184 return bpf_base_func_proto(func_id);
10189 sk_reuseport_is_valid_access(int off, int size,
10190 enum bpf_access_type type,
10191 const struct bpf_prog *prog,
10192 struct bpf_insn_access_aux *info)
10194 const u32 size_default = sizeof(__u32);
10196 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
10197 off % size || type != BPF_READ)
10201 case offsetof(struct sk_reuseport_md, data):
10202 info->reg_type = PTR_TO_PACKET;
10203 return size == sizeof(__u64);
10205 case offsetof(struct sk_reuseport_md, data_end):
10206 info->reg_type = PTR_TO_PACKET_END;
10207 return size == sizeof(__u64);
10209 case offsetof(struct sk_reuseport_md, hash):
10210 return size == size_default;
10212 case offsetof(struct sk_reuseport_md, sk):
10213 info->reg_type = PTR_TO_SOCKET;
10214 return size == sizeof(__u64);
10216 case offsetof(struct sk_reuseport_md, migrating_sk):
10217 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
10218 return size == sizeof(__u64);
10220 /* Fields that allow narrowing */
10221 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
10222 if (size < sizeof_field(struct sk_buff, protocol))
10225 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
10226 case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
10227 case bpf_ctx_range(struct sk_reuseport_md, len):
10228 bpf_ctx_record_field_size(info, size_default);
10229 return bpf_ctx_narrow_access_ok(off, size, size_default);
10236 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
10237 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
10238 si->dst_reg, si->src_reg, \
10239 bpf_target_off(struct sk_reuseport_kern, F, \
10240 sizeof_field(struct sk_reuseport_kern, F), \
10244 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
10245 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
10250 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \
10251 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
10256 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
10257 const struct bpf_insn *si,
10258 struct bpf_insn *insn_buf,
10259 struct bpf_prog *prog,
10262 struct bpf_insn *insn = insn_buf;
10265 case offsetof(struct sk_reuseport_md, data):
10266 SK_REUSEPORT_LOAD_SKB_FIELD(data);
10269 case offsetof(struct sk_reuseport_md, len):
10270 SK_REUSEPORT_LOAD_SKB_FIELD(len);
10273 case offsetof(struct sk_reuseport_md, eth_protocol):
10274 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
10277 case offsetof(struct sk_reuseport_md, ip_protocol):
10278 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
10281 case offsetof(struct sk_reuseport_md, data_end):
10282 SK_REUSEPORT_LOAD_FIELD(data_end);
10285 case offsetof(struct sk_reuseport_md, hash):
10286 SK_REUSEPORT_LOAD_FIELD(hash);
10289 case offsetof(struct sk_reuseport_md, bind_inany):
10290 SK_REUSEPORT_LOAD_FIELD(bind_inany);
10293 case offsetof(struct sk_reuseport_md, sk):
10294 SK_REUSEPORT_LOAD_FIELD(sk);
10297 case offsetof(struct sk_reuseport_md, migrating_sk):
10298 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
10302 return insn - insn_buf;
10305 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
10306 .get_func_proto = sk_reuseport_func_proto,
10307 .is_valid_access = sk_reuseport_is_valid_access,
10308 .convert_ctx_access = sk_reuseport_convert_ctx_access,
10311 const struct bpf_prog_ops sk_reuseport_prog_ops = {
10314 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
10315 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
10317 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
10318 struct sock *, sk, u64, flags)
10320 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
10321 BPF_SK_LOOKUP_F_NO_REUSEPORT)))
10323 if (unlikely(sk && sk_is_refcounted(sk)))
10324 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
10325 if (unlikely(sk && sk->sk_state == TCP_ESTABLISHED))
10326 return -ESOCKTNOSUPPORT; /* reject connected sockets */
10328 /* Check if socket is suitable for packet L3/L4 protocol */
10329 if (sk && sk->sk_protocol != ctx->protocol)
10330 return -EPROTOTYPE;
10331 if (sk && sk->sk_family != ctx->family &&
10332 (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
10333 return -EAFNOSUPPORT;
10335 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
10338 /* Select socket as lookup result */
10339 ctx->selected_sk = sk;
10340 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
10344 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
10345 .func = bpf_sk_lookup_assign,
10347 .ret_type = RET_INTEGER,
10348 .arg1_type = ARG_PTR_TO_CTX,
10349 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL,
10350 .arg3_type = ARG_ANYTHING,
10353 static const struct bpf_func_proto *
10354 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
10357 case BPF_FUNC_perf_event_output:
10358 return &bpf_event_output_data_proto;
10359 case BPF_FUNC_sk_assign:
10360 return &bpf_sk_lookup_assign_proto;
10361 case BPF_FUNC_sk_release:
10362 return &bpf_sk_release_proto;
10364 return bpf_sk_base_func_proto(func_id);
10368 static bool sk_lookup_is_valid_access(int off, int size,
10369 enum bpf_access_type type,
10370 const struct bpf_prog *prog,
10371 struct bpf_insn_access_aux *info)
10373 if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
10375 if (off % size != 0)
10377 if (type != BPF_READ)
10381 case offsetof(struct bpf_sk_lookup, sk):
10382 info->reg_type = PTR_TO_SOCKET_OR_NULL;
10383 return size == sizeof(__u64);
10385 case bpf_ctx_range(struct bpf_sk_lookup, family):
10386 case bpf_ctx_range(struct bpf_sk_lookup, protocol):
10387 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
10388 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
10389 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
10390 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
10391 case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
10392 case bpf_ctx_range(struct bpf_sk_lookup, local_port):
10393 bpf_ctx_record_field_size(info, sizeof(__u32));
10394 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
10401 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
10402 const struct bpf_insn *si,
10403 struct bpf_insn *insn_buf,
10404 struct bpf_prog *prog,
10407 struct bpf_insn *insn = insn_buf;
10410 case offsetof(struct bpf_sk_lookup, sk):
10411 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10412 offsetof(struct bpf_sk_lookup_kern, selected_sk));
10415 case offsetof(struct bpf_sk_lookup, family):
10416 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10417 bpf_target_off(struct bpf_sk_lookup_kern,
10418 family, 2, target_size));
10421 case offsetof(struct bpf_sk_lookup, protocol):
10422 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10423 bpf_target_off(struct bpf_sk_lookup_kern,
10424 protocol, 2, target_size));
10427 case offsetof(struct bpf_sk_lookup, remote_ip4):
10428 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10429 bpf_target_off(struct bpf_sk_lookup_kern,
10430 v4.saddr, 4, target_size));
10433 case offsetof(struct bpf_sk_lookup, local_ip4):
10434 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10435 bpf_target_off(struct bpf_sk_lookup_kern,
10436 v4.daddr, 4, target_size));
10439 case bpf_ctx_range_till(struct bpf_sk_lookup,
10440 remote_ip6[0], remote_ip6[3]): {
10441 #if IS_ENABLED(CONFIG_IPV6)
10444 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
10445 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10446 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10447 offsetof(struct bpf_sk_lookup_kern, v6.saddr));
10448 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10449 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10451 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10455 case bpf_ctx_range_till(struct bpf_sk_lookup,
10456 local_ip6[0], local_ip6[3]): {
10457 #if IS_ENABLED(CONFIG_IPV6)
10460 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
10461 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10462 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10463 offsetof(struct bpf_sk_lookup_kern, v6.daddr));
10464 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10465 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10467 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10471 case offsetof(struct bpf_sk_lookup, remote_port):
10472 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10473 bpf_target_off(struct bpf_sk_lookup_kern,
10474 sport, 2, target_size));
10477 case offsetof(struct bpf_sk_lookup, local_port):
10478 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10479 bpf_target_off(struct bpf_sk_lookup_kern,
10480 dport, 2, target_size));
10484 return insn - insn_buf;
10487 const struct bpf_prog_ops sk_lookup_prog_ops = {
10488 .test_run = bpf_prog_test_run_sk_lookup,
10491 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
10492 .get_func_proto = sk_lookup_func_proto,
10493 .is_valid_access = sk_lookup_is_valid_access,
10494 .convert_ctx_access = sk_lookup_convert_ctx_access,
10497 #endif /* CONFIG_INET */
10499 DEFINE_BPF_DISPATCHER(xdp)
10501 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
10503 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
10506 #ifdef CONFIG_DEBUG_INFO_BTF
10507 BTF_ID_LIST_GLOBAL(btf_sock_ids)
10508 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
10510 #undef BTF_SOCK_TYPE
10512 u32 btf_sock_ids[MAX_BTF_SOCK_TYPE];
10515 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
10517 /* tcp6_sock type is not generated in dwarf and hence btf,
10518 * trigger an explicit type generation here.
10520 BTF_TYPE_EMIT(struct tcp6_sock);
10521 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
10522 sk->sk_family == AF_INET6)
10523 return (unsigned long)sk;
10525 return (unsigned long)NULL;
10528 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
10529 .func = bpf_skc_to_tcp6_sock,
10531 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10532 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10533 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
10536 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
10538 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
10539 return (unsigned long)sk;
10541 return (unsigned long)NULL;
10544 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
10545 .func = bpf_skc_to_tcp_sock,
10547 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10548 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10549 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
10552 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
10554 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
10555 * generated if CONFIG_INET=n. Trigger an explicit generation here.
10557 BTF_TYPE_EMIT(struct inet_timewait_sock);
10558 BTF_TYPE_EMIT(struct tcp_timewait_sock);
10561 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
10562 return (unsigned long)sk;
10565 #if IS_BUILTIN(CONFIG_IPV6)
10566 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
10567 return (unsigned long)sk;
10570 return (unsigned long)NULL;
10573 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
10574 .func = bpf_skc_to_tcp_timewait_sock,
10576 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10577 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10578 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
10581 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
10584 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10585 return (unsigned long)sk;
10588 #if IS_BUILTIN(CONFIG_IPV6)
10589 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10590 return (unsigned long)sk;
10593 return (unsigned long)NULL;
10596 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
10597 .func = bpf_skc_to_tcp_request_sock,
10599 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10600 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10601 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
10604 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
10606 /* udp6_sock type is not generated in dwarf and hence btf,
10607 * trigger an explicit type generation here.
10609 BTF_TYPE_EMIT(struct udp6_sock);
10610 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
10611 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
10612 return (unsigned long)sk;
10614 return (unsigned long)NULL;
10617 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
10618 .func = bpf_skc_to_udp6_sock,
10620 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10621 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10622 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
10625 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
10627 return (unsigned long)sock_from_file(file);
10630 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
10631 BTF_ID(struct, socket)
10632 BTF_ID(struct, file)
10634 const struct bpf_func_proto bpf_sock_from_file_proto = {
10635 .func = bpf_sock_from_file,
10637 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10638 .ret_btf_id = &bpf_sock_from_file_btf_ids[0],
10639 .arg1_type = ARG_PTR_TO_BTF_ID,
10640 .arg1_btf_id = &bpf_sock_from_file_btf_ids[1],
10643 static const struct bpf_func_proto *
10644 bpf_sk_base_func_proto(enum bpf_func_id func_id)
10646 const struct bpf_func_proto *func;
10649 case BPF_FUNC_skc_to_tcp6_sock:
10650 func = &bpf_skc_to_tcp6_sock_proto;
10652 case BPF_FUNC_skc_to_tcp_sock:
10653 func = &bpf_skc_to_tcp_sock_proto;
10655 case BPF_FUNC_skc_to_tcp_timewait_sock:
10656 func = &bpf_skc_to_tcp_timewait_sock_proto;
10658 case BPF_FUNC_skc_to_tcp_request_sock:
10659 func = &bpf_skc_to_tcp_request_sock_proto;
10661 case BPF_FUNC_skc_to_udp6_sock:
10662 func = &bpf_skc_to_udp6_sock_proto;
10665 return bpf_base_func_proto(func_id);
10668 if (!perfmon_capable())