1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Linux Socket Filter - Kernel level socket filtering
5 * Based on the design of the Berkeley Packet Filter. The new
6 * internal format has been designed by PLUMgrid:
8 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
12 * Jay Schulist <jschlst@samba.org>
13 * Alexei Starovoitov <ast@plumgrid.com>
14 * Daniel Borkmann <dborkman@redhat.com>
16 * Andi Kleen - Fix a few bad bugs and races.
17 * Kris Katterjohn - Added many additional checks in bpf_check_classic()
20 #include <linux/module.h>
21 #include <linux/types.h>
23 #include <linux/fcntl.h>
24 #include <linux/socket.h>
25 #include <linux/sock_diag.h>
27 #include <linux/inet.h>
28 #include <linux/netdevice.h>
29 #include <linux/if_packet.h>
30 #include <linux/if_arp.h>
31 #include <linux/gfp.h>
32 #include <net/inet_common.h>
34 #include <net/protocol.h>
35 #include <net/netlink.h>
36 #include <linux/skbuff.h>
37 #include <linux/skmsg.h>
39 #include <net/flow_dissector.h>
40 #include <linux/errno.h>
41 #include <linux/timer.h>
42 #include <linux/uaccess.h>
43 #include <asm/unaligned.h>
44 #include <asm/cmpxchg.h>
45 #include <linux/filter.h>
46 #include <linux/ratelimit.h>
47 #include <linux/seccomp.h>
48 #include <linux/if_vlan.h>
49 #include <linux/bpf.h>
50 #include <linux/btf.h>
51 #include <net/sch_generic.h>
52 #include <net/cls_cgroup.h>
53 #include <net/dst_metadata.h>
55 #include <net/sock_reuseport.h>
56 #include <net/busy_poll.h>
60 #include <linux/bpf_trace.h>
61 #include <net/xdp_sock.h>
62 #include <linux/inetdevice.h>
63 #include <net/inet_hashtables.h>
64 #include <net/inet6_hashtables.h>
65 #include <net/ip_fib.h>
66 #include <net/nexthop.h>
70 #include <net/net_namespace.h>
71 #include <linux/seg6_local.h>
73 #include <net/seg6_local.h>
74 #include <net/lwtunnel.h>
75 #include <net/ipv6_stubs.h>
76 #include <net/bpf_sk_storage.h>
77 #include <net/transp_v6.h>
78 #include <linux/btf_ids.h>
81 static const struct bpf_func_proto *
82 bpf_sk_base_func_proto(enum bpf_func_id func_id);
84 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len)
86 if (in_compat_syscall()) {
87 struct compat_sock_fprog f32;
89 if (len != sizeof(f32))
91 if (copy_from_sockptr(&f32, src, sizeof(f32)))
93 memset(dst, 0, sizeof(*dst));
95 dst->filter = compat_ptr(f32.filter);
97 if (len != sizeof(*dst))
99 if (copy_from_sockptr(dst, src, sizeof(*dst)))
105 EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user);
108 * sk_filter_trim_cap - run a packet through a socket filter
109 * @sk: sock associated with &sk_buff
110 * @skb: buffer to filter
111 * @cap: limit on how short the eBPF program may trim the packet
113 * Run the eBPF program and then cut skb->data to correct size returned by
114 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
115 * than pkt_len we keep whole skb->data. This is the socket level
116 * wrapper to BPF_PROG_RUN. It returns 0 if the packet should
117 * be accepted or -EPERM if the packet should be tossed.
120 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
123 struct sk_filter *filter;
126 * If the skb was allocated from pfmemalloc reserves, only
127 * allow SOCK_MEMALLOC sockets to use it as this socket is
128 * helping free memory
130 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
131 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
134 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
138 err = security_sock_rcv_skb(sk, skb);
143 filter = rcu_dereference(sk->sk_filter);
145 struct sock *save_sk = skb->sk;
146 unsigned int pkt_len;
149 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
151 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
157 EXPORT_SYMBOL(sk_filter_trim_cap);
159 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
161 return skb_get_poff(skb);
164 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
168 if (skb_is_nonlinear(skb))
171 if (skb->len < sizeof(struct nlattr))
174 if (a > skb->len - sizeof(struct nlattr))
177 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
179 return (void *) nla - (void *) skb->data;
184 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
188 if (skb_is_nonlinear(skb))
191 if (skb->len < sizeof(struct nlattr))
194 if (a > skb->len - sizeof(struct nlattr))
197 nla = (struct nlattr *) &skb->data[a];
198 if (nla->nla_len > skb->len - a)
201 nla = nla_find_nested(nla, x);
203 return (void *) nla - (void *) skb->data;
208 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
209 data, int, headlen, int, offset)
212 const int len = sizeof(tmp);
215 if (headlen - offset >= len)
216 return *(u8 *)(data + offset);
217 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
220 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
228 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
231 return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
235 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
236 data, int, headlen, int, offset)
239 const int len = sizeof(tmp);
242 if (headlen - offset >= len)
243 return get_unaligned_be16(data + offset);
244 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
245 return be16_to_cpu(tmp);
247 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
249 return get_unaligned_be16(ptr);
255 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
258 return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
262 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
263 data, int, headlen, int, offset)
266 const int len = sizeof(tmp);
268 if (likely(offset >= 0)) {
269 if (headlen - offset >= len)
270 return get_unaligned_be32(data + offset);
271 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
272 return be32_to_cpu(tmp);
274 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
276 return get_unaligned_be32(ptr);
282 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
285 return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
289 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
290 struct bpf_insn *insn_buf)
292 struct bpf_insn *insn = insn_buf;
296 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4);
298 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
299 offsetof(struct sk_buff, mark));
303 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
304 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
305 #ifdef __BIG_ENDIAN_BITFIELD
306 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
311 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2);
313 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
314 offsetof(struct sk_buff, queue_mapping));
317 case SKF_AD_VLAN_TAG:
318 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2);
320 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
321 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
322 offsetof(struct sk_buff, vlan_tci));
324 case SKF_AD_VLAN_TAG_PRESENT:
325 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_VLAN_PRESENT_OFFSET());
326 if (PKT_VLAN_PRESENT_BIT)
327 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, PKT_VLAN_PRESENT_BIT);
328 if (PKT_VLAN_PRESENT_BIT < 7)
329 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
333 return insn - insn_buf;
336 static bool convert_bpf_extensions(struct sock_filter *fp,
337 struct bpf_insn **insnp)
339 struct bpf_insn *insn = *insnp;
343 case SKF_AD_OFF + SKF_AD_PROTOCOL:
344 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2);
346 /* A = *(u16 *) (CTX + offsetof(protocol)) */
347 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
348 offsetof(struct sk_buff, protocol));
349 /* A = ntohs(A) [emitting a nop or swap16] */
350 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
353 case SKF_AD_OFF + SKF_AD_PKTTYPE:
354 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
358 case SKF_AD_OFF + SKF_AD_IFINDEX:
359 case SKF_AD_OFF + SKF_AD_HATYPE:
360 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4);
361 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2);
363 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
364 BPF_REG_TMP, BPF_REG_CTX,
365 offsetof(struct sk_buff, dev));
366 /* if (tmp != 0) goto pc + 1 */
367 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
368 *insn++ = BPF_EXIT_INSN();
369 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
370 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
371 offsetof(struct net_device, ifindex));
373 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
374 offsetof(struct net_device, type));
377 case SKF_AD_OFF + SKF_AD_MARK:
378 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
382 case SKF_AD_OFF + SKF_AD_RXHASH:
383 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4);
385 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
386 offsetof(struct sk_buff, hash));
389 case SKF_AD_OFF + SKF_AD_QUEUE:
390 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
394 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
395 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
396 BPF_REG_A, BPF_REG_CTX, insn);
400 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
401 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
402 BPF_REG_A, BPF_REG_CTX, insn);
406 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
407 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2);
409 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
410 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
411 offsetof(struct sk_buff, vlan_proto));
412 /* A = ntohs(A) [emitting a nop or swap16] */
413 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
416 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
417 case SKF_AD_OFF + SKF_AD_NLATTR:
418 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
419 case SKF_AD_OFF + SKF_AD_CPU:
420 case SKF_AD_OFF + SKF_AD_RANDOM:
422 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
424 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
426 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
427 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
429 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
430 *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
432 case SKF_AD_OFF + SKF_AD_NLATTR:
433 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
435 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
436 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
438 case SKF_AD_OFF + SKF_AD_CPU:
439 *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
441 case SKF_AD_OFF + SKF_AD_RANDOM:
442 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
443 bpf_user_rnd_init_once();
448 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
450 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
454 /* This is just a dummy call to avoid letting the compiler
455 * evict __bpf_call_base() as an optimization. Placed here
456 * where no-one bothers.
458 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
466 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
468 const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
469 int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
470 bool endian = BPF_SIZE(fp->code) == BPF_H ||
471 BPF_SIZE(fp->code) == BPF_W;
472 bool indirect = BPF_MODE(fp->code) == BPF_IND;
473 const int ip_align = NET_IP_ALIGN;
474 struct bpf_insn *insn = *insnp;
478 ((unaligned_ok && offset >= 0) ||
479 (!unaligned_ok && offset >= 0 &&
480 offset + ip_align >= 0 &&
481 offset + ip_align % size == 0))) {
482 bool ldx_off_ok = offset <= S16_MAX;
484 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
486 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
487 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
488 size, 2 + endian + (!ldx_off_ok * 2));
490 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
493 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
494 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
495 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
499 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
500 *insn++ = BPF_JMP_A(8);
503 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
504 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
505 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
507 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
509 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
511 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
514 switch (BPF_SIZE(fp->code)) {
516 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
519 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
522 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
528 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
529 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
530 *insn = BPF_EXIT_INSN();
537 * bpf_convert_filter - convert filter program
538 * @prog: the user passed filter program
539 * @len: the length of the user passed filter program
540 * @new_prog: allocated 'struct bpf_prog' or NULL
541 * @new_len: pointer to store length of converted program
542 * @seen_ld_abs: bool whether we've seen ld_abs/ind
544 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
545 * style extended BPF (eBPF).
546 * Conversion workflow:
548 * 1) First pass for calculating the new program length:
549 * bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
551 * 2) 2nd pass to remap in two passes: 1st pass finds new
552 * jump offsets, 2nd pass remapping:
553 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
555 static int bpf_convert_filter(struct sock_filter *prog, int len,
556 struct bpf_prog *new_prog, int *new_len,
559 int new_flen = 0, pass = 0, target, i, stack_off;
560 struct bpf_insn *new_insn, *first_insn = NULL;
561 struct sock_filter *fp;
565 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
566 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
568 if (len <= 0 || len > BPF_MAXINSNS)
572 first_insn = new_prog->insnsi;
573 addrs = kcalloc(len, sizeof(*addrs),
574 GFP_KERNEL | __GFP_NOWARN);
580 new_insn = first_insn;
583 /* Classic BPF related prologue emission. */
585 /* Classic BPF expects A and X to be reset first. These need
586 * to be guaranteed to be the first two instructions.
588 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
589 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
591 /* All programs must keep CTX in callee saved BPF_REG_CTX.
592 * In eBPF case it's done by the compiler, here we need to
593 * do this ourself. Initial CTX is present in BPF_REG_ARG1.
595 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
597 /* For packet access in classic BPF, cache skb->data
598 * in callee-saved BPF R8 and skb->len - skb->data_len
599 * (headlen) in BPF R9. Since classic BPF is read-only
600 * on CTX, we only need to cache it once.
602 *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
603 BPF_REG_D, BPF_REG_CTX,
604 offsetof(struct sk_buff, data));
605 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
606 offsetof(struct sk_buff, len));
607 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
608 offsetof(struct sk_buff, data_len));
609 *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
615 for (i = 0; i < len; fp++, i++) {
616 struct bpf_insn tmp_insns[32] = { };
617 struct bpf_insn *insn = tmp_insns;
620 addrs[i] = new_insn - first_insn;
623 /* All arithmetic insns and skb loads map as-is. */
624 case BPF_ALU | BPF_ADD | BPF_X:
625 case BPF_ALU | BPF_ADD | BPF_K:
626 case BPF_ALU | BPF_SUB | BPF_X:
627 case BPF_ALU | BPF_SUB | BPF_K:
628 case BPF_ALU | BPF_AND | BPF_X:
629 case BPF_ALU | BPF_AND | BPF_K:
630 case BPF_ALU | BPF_OR | BPF_X:
631 case BPF_ALU | BPF_OR | BPF_K:
632 case BPF_ALU | BPF_LSH | BPF_X:
633 case BPF_ALU | BPF_LSH | BPF_K:
634 case BPF_ALU | BPF_RSH | BPF_X:
635 case BPF_ALU | BPF_RSH | BPF_K:
636 case BPF_ALU | BPF_XOR | BPF_X:
637 case BPF_ALU | BPF_XOR | BPF_K:
638 case BPF_ALU | BPF_MUL | BPF_X:
639 case BPF_ALU | BPF_MUL | BPF_K:
640 case BPF_ALU | BPF_DIV | BPF_X:
641 case BPF_ALU | BPF_DIV | BPF_K:
642 case BPF_ALU | BPF_MOD | BPF_X:
643 case BPF_ALU | BPF_MOD | BPF_K:
644 case BPF_ALU | BPF_NEG:
645 case BPF_LD | BPF_ABS | BPF_W:
646 case BPF_LD | BPF_ABS | BPF_H:
647 case BPF_LD | BPF_ABS | BPF_B:
648 case BPF_LD | BPF_IND | BPF_W:
649 case BPF_LD | BPF_IND | BPF_H:
650 case BPF_LD | BPF_IND | BPF_B:
651 /* Check for overloaded BPF extension and
652 * directly convert it if found, otherwise
653 * just move on with mapping.
655 if (BPF_CLASS(fp->code) == BPF_LD &&
656 BPF_MODE(fp->code) == BPF_ABS &&
657 convert_bpf_extensions(fp, &insn))
659 if (BPF_CLASS(fp->code) == BPF_LD &&
660 convert_bpf_ld_abs(fp, &insn)) {
665 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
666 fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
667 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
668 /* Error with exception code on div/mod by 0.
669 * For cBPF programs, this was always return 0.
671 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
672 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
673 *insn++ = BPF_EXIT_INSN();
676 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
679 /* Jump transformation cannot use BPF block macros
680 * everywhere as offset calculation and target updates
681 * require a bit more work than the rest, i.e. jump
682 * opcodes map as-is, but offsets need adjustment.
685 #define BPF_EMIT_JMP \
687 const s32 off_min = S16_MIN, off_max = S16_MAX; \
690 if (target >= len || target < 0) \
692 off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
693 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
694 off -= insn - tmp_insns; \
695 /* Reject anything not fitting into insn->off. */ \
696 if (off < off_min || off > off_max) \
701 case BPF_JMP | BPF_JA:
702 target = i + fp->k + 1;
703 insn->code = fp->code;
707 case BPF_JMP | BPF_JEQ | BPF_K:
708 case BPF_JMP | BPF_JEQ | BPF_X:
709 case BPF_JMP | BPF_JSET | BPF_K:
710 case BPF_JMP | BPF_JSET | BPF_X:
711 case BPF_JMP | BPF_JGT | BPF_K:
712 case BPF_JMP | BPF_JGT | BPF_X:
713 case BPF_JMP | BPF_JGE | BPF_K:
714 case BPF_JMP | BPF_JGE | BPF_X:
715 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
716 /* BPF immediates are signed, zero extend
717 * immediate into tmp register and use it
720 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
722 insn->dst_reg = BPF_REG_A;
723 insn->src_reg = BPF_REG_TMP;
726 insn->dst_reg = BPF_REG_A;
728 bpf_src = BPF_SRC(fp->code);
729 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
732 /* Common case where 'jump_false' is next insn. */
734 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
735 target = i + fp->jt + 1;
740 /* Convert some jumps when 'jump_true' is next insn. */
742 switch (BPF_OP(fp->code)) {
744 insn->code = BPF_JMP | BPF_JNE | bpf_src;
747 insn->code = BPF_JMP | BPF_JLE | bpf_src;
750 insn->code = BPF_JMP | BPF_JLT | bpf_src;
756 target = i + fp->jf + 1;
761 /* Other jumps are mapped into two insns: Jxx and JA. */
762 target = i + fp->jt + 1;
763 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
767 insn->code = BPF_JMP | BPF_JA;
768 target = i + fp->jf + 1;
772 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
773 case BPF_LDX | BPF_MSH | BPF_B: {
774 struct sock_filter tmp = {
775 .code = BPF_LD | BPF_ABS | BPF_B,
782 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
783 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
784 convert_bpf_ld_abs(&tmp, &insn);
787 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
789 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
791 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
793 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
795 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
798 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
799 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
801 case BPF_RET | BPF_A:
802 case BPF_RET | BPF_K:
803 if (BPF_RVAL(fp->code) == BPF_K)
804 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
806 *insn = BPF_EXIT_INSN();
809 /* Store to stack. */
812 stack_off = fp->k * 4 + 4;
813 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
814 BPF_ST ? BPF_REG_A : BPF_REG_X,
816 /* check_load_and_stores() verifies that classic BPF can
817 * load from stack only after write, so tracking
818 * stack_depth for ST|STX insns is enough
820 if (new_prog && new_prog->aux->stack_depth < stack_off)
821 new_prog->aux->stack_depth = stack_off;
824 /* Load from stack. */
825 case BPF_LD | BPF_MEM:
826 case BPF_LDX | BPF_MEM:
827 stack_off = fp->k * 4 + 4;
828 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
829 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
834 case BPF_LD | BPF_IMM:
835 case BPF_LDX | BPF_IMM:
836 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
837 BPF_REG_A : BPF_REG_X, fp->k);
841 case BPF_MISC | BPF_TAX:
842 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
846 case BPF_MISC | BPF_TXA:
847 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
850 /* A = skb->len or X = skb->len */
851 case BPF_LD | BPF_W | BPF_LEN:
852 case BPF_LDX | BPF_W | BPF_LEN:
853 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
854 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
855 offsetof(struct sk_buff, len));
858 /* Access seccomp_data fields. */
859 case BPF_LDX | BPF_ABS | BPF_W:
860 /* A = *(u32 *) (ctx + K) */
861 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
864 /* Unknown instruction. */
871 memcpy(new_insn, tmp_insns,
872 sizeof(*insn) * (insn - tmp_insns));
873 new_insn += insn - tmp_insns;
877 /* Only calculating new length. */
878 *new_len = new_insn - first_insn;
880 *new_len += 4; /* Prologue bits. */
885 if (new_flen != new_insn - first_insn) {
886 new_flen = new_insn - first_insn;
893 BUG_ON(*new_len != new_flen);
902 * As we dont want to clear mem[] array for each packet going through
903 * __bpf_prog_run(), we check that filter loaded by user never try to read
904 * a cell if not previously written, and we check all branches to be sure
905 * a malicious user doesn't try to abuse us.
907 static int check_load_and_stores(const struct sock_filter *filter, int flen)
909 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
912 BUILD_BUG_ON(BPF_MEMWORDS > 16);
914 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
918 memset(masks, 0xff, flen * sizeof(*masks));
920 for (pc = 0; pc < flen; pc++) {
921 memvalid &= masks[pc];
923 switch (filter[pc].code) {
926 memvalid |= (1 << filter[pc].k);
928 case BPF_LD | BPF_MEM:
929 case BPF_LDX | BPF_MEM:
930 if (!(memvalid & (1 << filter[pc].k))) {
935 case BPF_JMP | BPF_JA:
936 /* A jump must set masks on target */
937 masks[pc + 1 + filter[pc].k] &= memvalid;
940 case BPF_JMP | BPF_JEQ | BPF_K:
941 case BPF_JMP | BPF_JEQ | BPF_X:
942 case BPF_JMP | BPF_JGE | BPF_K:
943 case BPF_JMP | BPF_JGE | BPF_X:
944 case BPF_JMP | BPF_JGT | BPF_K:
945 case BPF_JMP | BPF_JGT | BPF_X:
946 case BPF_JMP | BPF_JSET | BPF_K:
947 case BPF_JMP | BPF_JSET | BPF_X:
948 /* A jump must set masks on targets */
949 masks[pc + 1 + filter[pc].jt] &= memvalid;
950 masks[pc + 1 + filter[pc].jf] &= memvalid;
960 static bool chk_code_allowed(u16 code_to_probe)
962 static const bool codes[] = {
963 /* 32 bit ALU operations */
964 [BPF_ALU | BPF_ADD | BPF_K] = true,
965 [BPF_ALU | BPF_ADD | BPF_X] = true,
966 [BPF_ALU | BPF_SUB | BPF_K] = true,
967 [BPF_ALU | BPF_SUB | BPF_X] = true,
968 [BPF_ALU | BPF_MUL | BPF_K] = true,
969 [BPF_ALU | BPF_MUL | BPF_X] = true,
970 [BPF_ALU | BPF_DIV | BPF_K] = true,
971 [BPF_ALU | BPF_DIV | BPF_X] = true,
972 [BPF_ALU | BPF_MOD | BPF_K] = true,
973 [BPF_ALU | BPF_MOD | BPF_X] = true,
974 [BPF_ALU | BPF_AND | BPF_K] = true,
975 [BPF_ALU | BPF_AND | BPF_X] = true,
976 [BPF_ALU | BPF_OR | BPF_K] = true,
977 [BPF_ALU | BPF_OR | BPF_X] = true,
978 [BPF_ALU | BPF_XOR | BPF_K] = true,
979 [BPF_ALU | BPF_XOR | BPF_X] = true,
980 [BPF_ALU | BPF_LSH | BPF_K] = true,
981 [BPF_ALU | BPF_LSH | BPF_X] = true,
982 [BPF_ALU | BPF_RSH | BPF_K] = true,
983 [BPF_ALU | BPF_RSH | BPF_X] = true,
984 [BPF_ALU | BPF_NEG] = true,
985 /* Load instructions */
986 [BPF_LD | BPF_W | BPF_ABS] = true,
987 [BPF_LD | BPF_H | BPF_ABS] = true,
988 [BPF_LD | BPF_B | BPF_ABS] = true,
989 [BPF_LD | BPF_W | BPF_LEN] = true,
990 [BPF_LD | BPF_W | BPF_IND] = true,
991 [BPF_LD | BPF_H | BPF_IND] = true,
992 [BPF_LD | BPF_B | BPF_IND] = true,
993 [BPF_LD | BPF_IMM] = true,
994 [BPF_LD | BPF_MEM] = true,
995 [BPF_LDX | BPF_W | BPF_LEN] = true,
996 [BPF_LDX | BPF_B | BPF_MSH] = true,
997 [BPF_LDX | BPF_IMM] = true,
998 [BPF_LDX | BPF_MEM] = true,
999 /* Store instructions */
1002 /* Misc instructions */
1003 [BPF_MISC | BPF_TAX] = true,
1004 [BPF_MISC | BPF_TXA] = true,
1005 /* Return instructions */
1006 [BPF_RET | BPF_K] = true,
1007 [BPF_RET | BPF_A] = true,
1008 /* Jump instructions */
1009 [BPF_JMP | BPF_JA] = true,
1010 [BPF_JMP | BPF_JEQ | BPF_K] = true,
1011 [BPF_JMP | BPF_JEQ | BPF_X] = true,
1012 [BPF_JMP | BPF_JGE | BPF_K] = true,
1013 [BPF_JMP | BPF_JGE | BPF_X] = true,
1014 [BPF_JMP | BPF_JGT | BPF_K] = true,
1015 [BPF_JMP | BPF_JGT | BPF_X] = true,
1016 [BPF_JMP | BPF_JSET | BPF_K] = true,
1017 [BPF_JMP | BPF_JSET | BPF_X] = true,
1020 if (code_to_probe >= ARRAY_SIZE(codes))
1023 return codes[code_to_probe];
1026 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1031 if (flen == 0 || flen > BPF_MAXINSNS)
1038 * bpf_check_classic - verify socket filter code
1039 * @filter: filter to verify
1040 * @flen: length of filter
1042 * Check the user's filter code. If we let some ugly
1043 * filter code slip through kaboom! The filter must contain
1044 * no references or jumps that are out of range, no illegal
1045 * instructions, and must end with a RET instruction.
1047 * All jumps are forward as they are not signed.
1049 * Returns 0 if the rule set is legal or -EINVAL if not.
1051 static int bpf_check_classic(const struct sock_filter *filter,
1057 /* Check the filter code now */
1058 for (pc = 0; pc < flen; pc++) {
1059 const struct sock_filter *ftest = &filter[pc];
1061 /* May we actually operate on this code? */
1062 if (!chk_code_allowed(ftest->code))
1065 /* Some instructions need special checks */
1066 switch (ftest->code) {
1067 case BPF_ALU | BPF_DIV | BPF_K:
1068 case BPF_ALU | BPF_MOD | BPF_K:
1069 /* Check for division by zero */
1073 case BPF_ALU | BPF_LSH | BPF_K:
1074 case BPF_ALU | BPF_RSH | BPF_K:
1078 case BPF_LD | BPF_MEM:
1079 case BPF_LDX | BPF_MEM:
1082 /* Check for invalid memory addresses */
1083 if (ftest->k >= BPF_MEMWORDS)
1086 case BPF_JMP | BPF_JA:
1087 /* Note, the large ftest->k might cause loops.
1088 * Compare this with conditional jumps below,
1089 * where offsets are limited. --ANK (981016)
1091 if (ftest->k >= (unsigned int)(flen - pc - 1))
1094 case BPF_JMP | BPF_JEQ | BPF_K:
1095 case BPF_JMP | BPF_JEQ | BPF_X:
1096 case BPF_JMP | BPF_JGE | BPF_K:
1097 case BPF_JMP | BPF_JGE | BPF_X:
1098 case BPF_JMP | BPF_JGT | BPF_K:
1099 case BPF_JMP | BPF_JGT | BPF_X:
1100 case BPF_JMP | BPF_JSET | BPF_K:
1101 case BPF_JMP | BPF_JSET | BPF_X:
1102 /* Both conditionals must be safe */
1103 if (pc + ftest->jt + 1 >= flen ||
1104 pc + ftest->jf + 1 >= flen)
1107 case BPF_LD | BPF_W | BPF_ABS:
1108 case BPF_LD | BPF_H | BPF_ABS:
1109 case BPF_LD | BPF_B | BPF_ABS:
1111 if (bpf_anc_helper(ftest) & BPF_ANC)
1113 /* Ancillary operation unknown or unsupported */
1114 if (anc_found == false && ftest->k >= SKF_AD_OFF)
1119 /* Last instruction must be a RET code */
1120 switch (filter[flen - 1].code) {
1121 case BPF_RET | BPF_K:
1122 case BPF_RET | BPF_A:
1123 return check_load_and_stores(filter, flen);
1129 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1130 const struct sock_fprog *fprog)
1132 unsigned int fsize = bpf_classic_proglen(fprog);
1133 struct sock_fprog_kern *fkprog;
1135 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1139 fkprog = fp->orig_prog;
1140 fkprog->len = fprog->len;
1142 fkprog->filter = kmemdup(fp->insns, fsize,
1143 GFP_KERNEL | __GFP_NOWARN);
1144 if (!fkprog->filter) {
1145 kfree(fp->orig_prog);
1152 static void bpf_release_orig_filter(struct bpf_prog *fp)
1154 struct sock_fprog_kern *fprog = fp->orig_prog;
1157 kfree(fprog->filter);
1162 static void __bpf_prog_release(struct bpf_prog *prog)
1164 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1167 bpf_release_orig_filter(prog);
1168 bpf_prog_free(prog);
1172 static void __sk_filter_release(struct sk_filter *fp)
1174 __bpf_prog_release(fp->prog);
1179 * sk_filter_release_rcu - Release a socket filter by rcu_head
1180 * @rcu: rcu_head that contains the sk_filter to free
1182 static void sk_filter_release_rcu(struct rcu_head *rcu)
1184 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1186 __sk_filter_release(fp);
1190 * sk_filter_release - release a socket filter
1191 * @fp: filter to remove
1193 * Remove a filter from a socket and release its resources.
1195 static void sk_filter_release(struct sk_filter *fp)
1197 if (refcount_dec_and_test(&fp->refcnt))
1198 call_rcu(&fp->rcu, sk_filter_release_rcu);
1201 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1203 u32 filter_size = bpf_prog_size(fp->prog->len);
1205 atomic_sub(filter_size, &sk->sk_omem_alloc);
1206 sk_filter_release(fp);
1209 /* try to charge the socket memory if there is space available
1210 * return true on success
1212 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1214 u32 filter_size = bpf_prog_size(fp->prog->len);
1216 /* same check as in sock_kmalloc() */
1217 if (filter_size <= sysctl_optmem_max &&
1218 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
1219 atomic_add(filter_size, &sk->sk_omem_alloc);
1225 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1227 if (!refcount_inc_not_zero(&fp->refcnt))
1230 if (!__sk_filter_charge(sk, fp)) {
1231 sk_filter_release(fp);
1237 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1239 struct sock_filter *old_prog;
1240 struct bpf_prog *old_fp;
1241 int err, new_len, old_len = fp->len;
1242 bool seen_ld_abs = false;
1244 /* We are free to overwrite insns et al right here as it
1245 * won't be used at this point in time anymore internally
1246 * after the migration to the internal BPF instruction
1249 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1250 sizeof(struct bpf_insn));
1252 /* Conversion cannot happen on overlapping memory areas,
1253 * so we need to keep the user BPF around until the 2nd
1254 * pass. At this time, the user BPF is stored in fp->insns.
1256 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1257 GFP_KERNEL | __GFP_NOWARN);
1263 /* 1st pass: calculate the new program length. */
1264 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1269 /* Expand fp for appending the new filter representation. */
1271 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1273 /* The old_fp is still around in case we couldn't
1274 * allocate new memory, so uncharge on that one.
1283 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1284 err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1287 /* 2nd bpf_convert_filter() can fail only if it fails
1288 * to allocate memory, remapping must succeed. Note,
1289 * that at this time old_fp has already been released
1294 fp = bpf_prog_select_runtime(fp, &err);
1304 __bpf_prog_release(fp);
1305 return ERR_PTR(err);
1308 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1309 bpf_aux_classic_check_t trans)
1313 fp->bpf_func = NULL;
1316 err = bpf_check_classic(fp->insns, fp->len);
1318 __bpf_prog_release(fp);
1319 return ERR_PTR(err);
1322 /* There might be additional checks and transformations
1323 * needed on classic filters, f.e. in case of seccomp.
1326 err = trans(fp->insns, fp->len);
1328 __bpf_prog_release(fp);
1329 return ERR_PTR(err);
1333 /* Probe if we can JIT compile the filter and if so, do
1334 * the compilation of the filter.
1336 bpf_jit_compile(fp);
1338 /* JIT compiler couldn't process this filter, so do the
1339 * internal BPF translation for the optimized interpreter.
1342 fp = bpf_migrate_filter(fp);
1348 * bpf_prog_create - create an unattached filter
1349 * @pfp: the unattached filter that is created
1350 * @fprog: the filter program
1352 * Create a filter independent of any socket. We first run some
1353 * sanity checks on it to make sure it does not explode on us later.
1354 * If an error occurs or there is insufficient memory for the filter
1355 * a negative errno code is returned. On success the return is zero.
1357 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1359 unsigned int fsize = bpf_classic_proglen(fprog);
1360 struct bpf_prog *fp;
1362 /* Make sure new filter is there and in the right amounts. */
1363 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1366 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1370 memcpy(fp->insns, fprog->filter, fsize);
1372 fp->len = fprog->len;
1373 /* Since unattached filters are not copied back to user
1374 * space through sk_get_filter(), we do not need to hold
1375 * a copy here, and can spare us the work.
1377 fp->orig_prog = NULL;
1379 /* bpf_prepare_filter() already takes care of freeing
1380 * memory in case something goes wrong.
1382 fp = bpf_prepare_filter(fp, NULL);
1389 EXPORT_SYMBOL_GPL(bpf_prog_create);
1392 * bpf_prog_create_from_user - create an unattached filter from user buffer
1393 * @pfp: the unattached filter that is created
1394 * @fprog: the filter program
1395 * @trans: post-classic verifier transformation handler
1396 * @save_orig: save classic BPF program
1398 * This function effectively does the same as bpf_prog_create(), only
1399 * that it builds up its insns buffer from user space provided buffer.
1400 * It also allows for passing a bpf_aux_classic_check_t handler.
1402 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1403 bpf_aux_classic_check_t trans, bool save_orig)
1405 unsigned int fsize = bpf_classic_proglen(fprog);
1406 struct bpf_prog *fp;
1409 /* Make sure new filter is there and in the right amounts. */
1410 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1413 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1417 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1418 __bpf_prog_free(fp);
1422 fp->len = fprog->len;
1423 fp->orig_prog = NULL;
1426 err = bpf_prog_store_orig_filter(fp, fprog);
1428 __bpf_prog_free(fp);
1433 /* bpf_prepare_filter() already takes care of freeing
1434 * memory in case something goes wrong.
1436 fp = bpf_prepare_filter(fp, trans);
1443 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1445 void bpf_prog_destroy(struct bpf_prog *fp)
1447 __bpf_prog_release(fp);
1449 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1451 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1453 struct sk_filter *fp, *old_fp;
1455 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1461 if (!__sk_filter_charge(sk, fp)) {
1465 refcount_set(&fp->refcnt, 1);
1467 old_fp = rcu_dereference_protected(sk->sk_filter,
1468 lockdep_sock_is_held(sk));
1469 rcu_assign_pointer(sk->sk_filter, fp);
1472 sk_filter_uncharge(sk, old_fp);
1478 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1480 unsigned int fsize = bpf_classic_proglen(fprog);
1481 struct bpf_prog *prog;
1484 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1485 return ERR_PTR(-EPERM);
1487 /* Make sure new filter is there and in the right amounts. */
1488 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1489 return ERR_PTR(-EINVAL);
1491 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1493 return ERR_PTR(-ENOMEM);
1495 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1496 __bpf_prog_free(prog);
1497 return ERR_PTR(-EFAULT);
1500 prog->len = fprog->len;
1502 err = bpf_prog_store_orig_filter(prog, fprog);
1504 __bpf_prog_free(prog);
1505 return ERR_PTR(-ENOMEM);
1508 /* bpf_prepare_filter() already takes care of freeing
1509 * memory in case something goes wrong.
1511 return bpf_prepare_filter(prog, NULL);
1515 * sk_attach_filter - attach a socket filter
1516 * @fprog: the filter program
1517 * @sk: the socket to use
1519 * Attach the user's filter code. We first run some sanity checks on
1520 * it to make sure it does not explode on us later. If an error
1521 * occurs or there is insufficient memory for the filter a negative
1522 * errno code is returned. On success the return is zero.
1524 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1526 struct bpf_prog *prog = __get_filter(fprog, sk);
1530 return PTR_ERR(prog);
1532 err = __sk_attach_prog(prog, sk);
1534 __bpf_prog_release(prog);
1540 EXPORT_SYMBOL_GPL(sk_attach_filter);
1542 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1544 struct bpf_prog *prog = __get_filter(fprog, sk);
1548 return PTR_ERR(prog);
1550 if (bpf_prog_size(prog->len) > sysctl_optmem_max)
1553 err = reuseport_attach_prog(sk, prog);
1556 __bpf_prog_release(prog);
1561 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1563 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1564 return ERR_PTR(-EPERM);
1566 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1569 int sk_attach_bpf(u32 ufd, struct sock *sk)
1571 struct bpf_prog *prog = __get_bpf(ufd, sk);
1575 return PTR_ERR(prog);
1577 err = __sk_attach_prog(prog, sk);
1586 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1588 struct bpf_prog *prog;
1591 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1594 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1595 if (PTR_ERR(prog) == -EINVAL)
1596 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1598 return PTR_ERR(prog);
1600 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1601 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1602 * bpf prog (e.g. sockmap). It depends on the
1603 * limitation imposed by bpf_prog_load().
1604 * Hence, sysctl_optmem_max is not checked.
1606 if ((sk->sk_type != SOCK_STREAM &&
1607 sk->sk_type != SOCK_DGRAM) ||
1608 (sk->sk_protocol != IPPROTO_UDP &&
1609 sk->sk_protocol != IPPROTO_TCP) ||
1610 (sk->sk_family != AF_INET &&
1611 sk->sk_family != AF_INET6)) {
1616 /* BPF_PROG_TYPE_SOCKET_FILTER */
1617 if (bpf_prog_size(prog->len) > sysctl_optmem_max) {
1623 err = reuseport_attach_prog(sk, prog);
1631 void sk_reuseport_prog_free(struct bpf_prog *prog)
1636 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1639 bpf_prog_destroy(prog);
1642 struct bpf_scratchpad {
1644 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1645 u8 buff[MAX_BPF_STACK];
1649 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1651 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1652 unsigned int write_len)
1654 return skb_ensure_writable(skb, write_len);
1657 static inline int bpf_try_make_writable(struct sk_buff *skb,
1658 unsigned int write_len)
1660 int err = __bpf_try_make_writable(skb, write_len);
1662 bpf_compute_data_pointers(skb);
1666 static int bpf_try_make_head_writable(struct sk_buff *skb)
1668 return bpf_try_make_writable(skb, skb_headlen(skb));
1671 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1673 if (skb_at_tc_ingress(skb))
1674 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1677 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1679 if (skb_at_tc_ingress(skb))
1680 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1683 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1684 const void *, from, u32, len, u64, flags)
1688 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1690 if (unlikely(offset > 0xffff))
1692 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1695 ptr = skb->data + offset;
1696 if (flags & BPF_F_RECOMPUTE_CSUM)
1697 __skb_postpull_rcsum(skb, ptr, len, offset);
1699 memcpy(ptr, from, len);
1701 if (flags & BPF_F_RECOMPUTE_CSUM)
1702 __skb_postpush_rcsum(skb, ptr, len, offset);
1703 if (flags & BPF_F_INVALIDATE_HASH)
1704 skb_clear_hash(skb);
1709 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1710 .func = bpf_skb_store_bytes,
1712 .ret_type = RET_INTEGER,
1713 .arg1_type = ARG_PTR_TO_CTX,
1714 .arg2_type = ARG_ANYTHING,
1715 .arg3_type = ARG_PTR_TO_MEM,
1716 .arg4_type = ARG_CONST_SIZE,
1717 .arg5_type = ARG_ANYTHING,
1720 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1721 void *, to, u32, len)
1725 if (unlikely(offset > 0xffff))
1728 ptr = skb_header_pointer(skb, offset, len, to);
1732 memcpy(to, ptr, len);
1740 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1741 .func = bpf_skb_load_bytes,
1743 .ret_type = RET_INTEGER,
1744 .arg1_type = ARG_PTR_TO_CTX,
1745 .arg2_type = ARG_ANYTHING,
1746 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1747 .arg4_type = ARG_CONST_SIZE,
1750 BPF_CALL_4(bpf_flow_dissector_load_bytes,
1751 const struct bpf_flow_dissector *, ctx, u32, offset,
1752 void *, to, u32, len)
1756 if (unlikely(offset > 0xffff))
1759 if (unlikely(!ctx->skb))
1762 ptr = skb_header_pointer(ctx->skb, offset, len, to);
1766 memcpy(to, ptr, len);
1774 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1775 .func = bpf_flow_dissector_load_bytes,
1777 .ret_type = RET_INTEGER,
1778 .arg1_type = ARG_PTR_TO_CTX,
1779 .arg2_type = ARG_ANYTHING,
1780 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1781 .arg4_type = ARG_CONST_SIZE,
1784 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1785 u32, offset, void *, to, u32, len, u32, start_header)
1787 u8 *end = skb_tail_pointer(skb);
1790 if (unlikely(offset > 0xffff))
1793 switch (start_header) {
1794 case BPF_HDR_START_MAC:
1795 if (unlikely(!skb_mac_header_was_set(skb)))
1797 start = skb_mac_header(skb);
1799 case BPF_HDR_START_NET:
1800 start = skb_network_header(skb);
1806 ptr = start + offset;
1808 if (likely(ptr + len <= end)) {
1809 memcpy(to, ptr, len);
1818 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1819 .func = bpf_skb_load_bytes_relative,
1821 .ret_type = RET_INTEGER,
1822 .arg1_type = ARG_PTR_TO_CTX,
1823 .arg2_type = ARG_ANYTHING,
1824 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1825 .arg4_type = ARG_CONST_SIZE,
1826 .arg5_type = ARG_ANYTHING,
1829 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1831 /* Idea is the following: should the needed direct read/write
1832 * test fail during runtime, we can pull in more data and redo
1833 * again, since implicitly, we invalidate previous checks here.
1835 * Or, since we know how much we need to make read/writeable,
1836 * this can be done once at the program beginning for direct
1837 * access case. By this we overcome limitations of only current
1838 * headroom being accessible.
1840 return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1843 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1844 .func = bpf_skb_pull_data,
1846 .ret_type = RET_INTEGER,
1847 .arg1_type = ARG_PTR_TO_CTX,
1848 .arg2_type = ARG_ANYTHING,
1851 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1853 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1856 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1857 .func = bpf_sk_fullsock,
1859 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
1860 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
1863 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1864 unsigned int write_len)
1866 int err = __bpf_try_make_writable(skb, write_len);
1868 bpf_compute_data_end_sk_skb(skb);
1872 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1874 /* Idea is the following: should the needed direct read/write
1875 * test fail during runtime, we can pull in more data and redo
1876 * again, since implicitly, we invalidate previous checks here.
1878 * Or, since we know how much we need to make read/writeable,
1879 * this can be done once at the program beginning for direct
1880 * access case. By this we overcome limitations of only current
1881 * headroom being accessible.
1883 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1886 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1887 .func = sk_skb_pull_data,
1889 .ret_type = RET_INTEGER,
1890 .arg1_type = ARG_PTR_TO_CTX,
1891 .arg2_type = ARG_ANYTHING,
1894 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1895 u64, from, u64, to, u64, flags)
1899 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1901 if (unlikely(offset > 0xffff || offset & 1))
1903 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1906 ptr = (__sum16 *)(skb->data + offset);
1907 switch (flags & BPF_F_HDR_FIELD_MASK) {
1909 if (unlikely(from != 0))
1912 csum_replace_by_diff(ptr, to);
1915 csum_replace2(ptr, from, to);
1918 csum_replace4(ptr, from, to);
1927 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1928 .func = bpf_l3_csum_replace,
1930 .ret_type = RET_INTEGER,
1931 .arg1_type = ARG_PTR_TO_CTX,
1932 .arg2_type = ARG_ANYTHING,
1933 .arg3_type = ARG_ANYTHING,
1934 .arg4_type = ARG_ANYTHING,
1935 .arg5_type = ARG_ANYTHING,
1938 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1939 u64, from, u64, to, u64, flags)
1941 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1942 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1943 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1946 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1947 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1949 if (unlikely(offset > 0xffff || offset & 1))
1951 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1954 ptr = (__sum16 *)(skb->data + offset);
1955 if (is_mmzero && !do_mforce && !*ptr)
1958 switch (flags & BPF_F_HDR_FIELD_MASK) {
1960 if (unlikely(from != 0))
1963 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1966 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1969 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1975 if (is_mmzero && !*ptr)
1976 *ptr = CSUM_MANGLED_0;
1980 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1981 .func = bpf_l4_csum_replace,
1983 .ret_type = RET_INTEGER,
1984 .arg1_type = ARG_PTR_TO_CTX,
1985 .arg2_type = ARG_ANYTHING,
1986 .arg3_type = ARG_ANYTHING,
1987 .arg4_type = ARG_ANYTHING,
1988 .arg5_type = ARG_ANYTHING,
1991 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1992 __be32 *, to, u32, to_size, __wsum, seed)
1994 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1995 u32 diff_size = from_size + to_size;
1998 /* This is quite flexible, some examples:
2000 * from_size == 0, to_size > 0, seed := csum --> pushing data
2001 * from_size > 0, to_size == 0, seed := csum --> pulling data
2002 * from_size > 0, to_size > 0, seed := 0 --> diffing data
2004 * Even for diffing, from_size and to_size don't need to be equal.
2006 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
2007 diff_size > sizeof(sp->diff)))
2010 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
2011 sp->diff[j] = ~from[i];
2012 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
2013 sp->diff[j] = to[i];
2015 return csum_partial(sp->diff, diff_size, seed);
2018 static const struct bpf_func_proto bpf_csum_diff_proto = {
2019 .func = bpf_csum_diff,
2022 .ret_type = RET_INTEGER,
2023 .arg1_type = ARG_PTR_TO_MEM_OR_NULL,
2024 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
2025 .arg3_type = ARG_PTR_TO_MEM_OR_NULL,
2026 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
2027 .arg5_type = ARG_ANYTHING,
2030 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2032 /* The interface is to be used in combination with bpf_csum_diff()
2033 * for direct packet writes. csum rotation for alignment as well
2034 * as emulating csum_sub() can be done from the eBPF program.
2036 if (skb->ip_summed == CHECKSUM_COMPLETE)
2037 return (skb->csum = csum_add(skb->csum, csum));
2042 static const struct bpf_func_proto bpf_csum_update_proto = {
2043 .func = bpf_csum_update,
2045 .ret_type = RET_INTEGER,
2046 .arg1_type = ARG_PTR_TO_CTX,
2047 .arg2_type = ARG_ANYTHING,
2050 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
2052 /* The interface is to be used in combination with bpf_skb_adjust_room()
2053 * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET
2054 * is passed as flags, for example.
2057 case BPF_CSUM_LEVEL_INC:
2058 __skb_incr_checksum_unnecessary(skb);
2060 case BPF_CSUM_LEVEL_DEC:
2061 __skb_decr_checksum_unnecessary(skb);
2063 case BPF_CSUM_LEVEL_RESET:
2064 __skb_reset_checksum_unnecessary(skb);
2066 case BPF_CSUM_LEVEL_QUERY:
2067 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2068 skb->csum_level : -EACCES;
2076 static const struct bpf_func_proto bpf_csum_level_proto = {
2077 .func = bpf_csum_level,
2079 .ret_type = RET_INTEGER,
2080 .arg1_type = ARG_PTR_TO_CTX,
2081 .arg2_type = ARG_ANYTHING,
2084 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2086 return dev_forward_skb(dev, skb);
2089 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2090 struct sk_buff *skb)
2092 int ret = ____dev_forward_skb(dev, skb);
2096 ret = netif_rx(skb);
2102 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2106 if (dev_xmit_recursion()) {
2107 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2115 dev_xmit_recursion_inc();
2116 ret = dev_queue_xmit(skb);
2117 dev_xmit_recursion_dec();
2122 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2125 unsigned int mlen = skb_network_offset(skb);
2128 __skb_pull(skb, mlen);
2130 /* At ingress, the mac header has already been pulled once.
2131 * At egress, skb_pospull_rcsum has to be done in case that
2132 * the skb is originated from ingress (i.e. a forwarded skb)
2133 * to ensure that rcsum starts at net header.
2135 if (!skb_at_tc_ingress(skb))
2136 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2138 skb_pop_mac_header(skb);
2139 skb_reset_mac_len(skb);
2140 return flags & BPF_F_INGRESS ?
2141 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2144 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2147 /* Verify that a link layer header is carried */
2148 if (unlikely(skb->mac_header >= skb->network_header)) {
2153 bpf_push_mac_rcsum(skb);
2154 return flags & BPF_F_INGRESS ?
2155 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2158 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2161 if (dev_is_mac_header_xmit(dev))
2162 return __bpf_redirect_common(skb, dev, flags);
2164 return __bpf_redirect_no_mac(skb, dev, flags);
2167 #if IS_ENABLED(CONFIG_IPV6)
2168 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2169 struct net_device *dev, struct bpf_nh_params *nh)
2171 u32 hh_len = LL_RESERVED_SPACE(dev);
2172 const struct in6_addr *nexthop;
2173 struct dst_entry *dst = NULL;
2174 struct neighbour *neigh;
2176 if (dev_xmit_recursion()) {
2177 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2184 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2185 struct sk_buff *skb2;
2187 skb2 = skb_realloc_headroom(skb, hh_len);
2188 if (unlikely(!skb2)) {
2193 skb_set_owner_w(skb2, skb->sk);
2201 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2202 &ipv6_hdr(skb)->daddr);
2204 nexthop = &nh->ipv6_nh;
2206 neigh = ip_neigh_gw6(dev, nexthop);
2207 if (likely(!IS_ERR(neigh))) {
2210 sock_confirm_neigh(skb, neigh);
2211 dev_xmit_recursion_inc();
2212 ret = neigh_output(neigh, skb, false);
2213 dev_xmit_recursion_dec();
2214 rcu_read_unlock_bh();
2217 rcu_read_unlock_bh();
2219 IP6_INC_STATS(dev_net(dst->dev),
2220 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2226 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2227 struct bpf_nh_params *nh)
2229 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2230 struct net *net = dev_net(dev);
2231 int err, ret = NET_XMIT_DROP;
2234 struct dst_entry *dst;
2235 struct flowi6 fl6 = {
2236 .flowi6_flags = FLOWI_FLAG_ANYSRC,
2237 .flowi6_mark = skb->mark,
2238 .flowlabel = ip6_flowinfo(ip6h),
2239 .flowi6_oif = dev->ifindex,
2240 .flowi6_proto = ip6h->nexthdr,
2241 .daddr = ip6h->daddr,
2242 .saddr = ip6h->saddr,
2245 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2249 skb_dst_set(skb, dst);
2250 } else if (nh->nh_family != AF_INET6) {
2254 err = bpf_out_neigh_v6(net, skb, dev, nh);
2255 if (unlikely(net_xmit_eval(err)))
2256 dev->stats.tx_errors++;
2258 ret = NET_XMIT_SUCCESS;
2261 dev->stats.tx_errors++;
2267 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2268 struct bpf_nh_params *nh)
2271 return NET_XMIT_DROP;
2273 #endif /* CONFIG_IPV6 */
2275 #if IS_ENABLED(CONFIG_INET)
2276 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2277 struct net_device *dev, struct bpf_nh_params *nh)
2279 u32 hh_len = LL_RESERVED_SPACE(dev);
2280 struct neighbour *neigh;
2281 bool is_v6gw = false;
2283 if (dev_xmit_recursion()) {
2284 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2291 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2292 struct sk_buff *skb2;
2294 skb2 = skb_realloc_headroom(skb, hh_len);
2295 if (unlikely(!skb2)) {
2300 skb_set_owner_w(skb2, skb->sk);
2307 struct dst_entry *dst = skb_dst(skb);
2308 struct rtable *rt = container_of(dst, struct rtable, dst);
2310 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2311 } else if (nh->nh_family == AF_INET6) {
2312 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2314 } else if (nh->nh_family == AF_INET) {
2315 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2317 rcu_read_unlock_bh();
2321 if (likely(!IS_ERR(neigh))) {
2324 sock_confirm_neigh(skb, neigh);
2325 dev_xmit_recursion_inc();
2326 ret = neigh_output(neigh, skb, is_v6gw);
2327 dev_xmit_recursion_dec();
2328 rcu_read_unlock_bh();
2331 rcu_read_unlock_bh();
2337 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2338 struct bpf_nh_params *nh)
2340 const struct iphdr *ip4h = ip_hdr(skb);
2341 struct net *net = dev_net(dev);
2342 int err, ret = NET_XMIT_DROP;
2345 struct flowi4 fl4 = {
2346 .flowi4_flags = FLOWI_FLAG_ANYSRC,
2347 .flowi4_mark = skb->mark,
2348 .flowi4_tos = RT_TOS(ip4h->tos),
2349 .flowi4_oif = dev->ifindex,
2350 .flowi4_proto = ip4h->protocol,
2351 .daddr = ip4h->daddr,
2352 .saddr = ip4h->saddr,
2356 rt = ip_route_output_flow(net, &fl4, NULL);
2359 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2364 skb_dst_set(skb, &rt->dst);
2367 err = bpf_out_neigh_v4(net, skb, dev, nh);
2368 if (unlikely(net_xmit_eval(err)))
2369 dev->stats.tx_errors++;
2371 ret = NET_XMIT_SUCCESS;
2374 dev->stats.tx_errors++;
2380 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2381 struct bpf_nh_params *nh)
2384 return NET_XMIT_DROP;
2386 #endif /* CONFIG_INET */
2388 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2389 struct bpf_nh_params *nh)
2391 struct ethhdr *ethh = eth_hdr(skb);
2393 if (unlikely(skb->mac_header >= skb->network_header))
2395 bpf_push_mac_rcsum(skb);
2396 if (is_multicast_ether_addr(ethh->h_dest))
2399 skb_pull(skb, sizeof(*ethh));
2400 skb_unset_mac_header(skb);
2401 skb_reset_network_header(skb);
2403 if (skb->protocol == htons(ETH_P_IP))
2404 return __bpf_redirect_neigh_v4(skb, dev, nh);
2405 else if (skb->protocol == htons(ETH_P_IPV6))
2406 return __bpf_redirect_neigh_v6(skb, dev, nh);
2412 /* Internal, non-exposed redirect flags. */
2414 BPF_F_NEIGH = (1ULL << 1),
2415 BPF_F_PEER = (1ULL << 2),
2416 BPF_F_NEXTHOP = (1ULL << 3),
2417 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2420 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2422 struct net_device *dev;
2423 struct sk_buff *clone;
2426 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2429 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2433 clone = skb_clone(skb, GFP_ATOMIC);
2434 if (unlikely(!clone))
2437 /* For direct write, we need to keep the invariant that the skbs
2438 * we're dealing with need to be uncloned. Should uncloning fail
2439 * here, we need to free the just generated clone to unclone once
2442 ret = bpf_try_make_head_writable(skb);
2443 if (unlikely(ret)) {
2448 return __bpf_redirect(clone, dev, flags);
2451 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2452 .func = bpf_clone_redirect,
2454 .ret_type = RET_INTEGER,
2455 .arg1_type = ARG_PTR_TO_CTX,
2456 .arg2_type = ARG_ANYTHING,
2457 .arg3_type = ARG_ANYTHING,
2460 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2461 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2463 int skb_do_redirect(struct sk_buff *skb)
2465 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2466 struct net *net = dev_net(skb->dev);
2467 struct net_device *dev;
2468 u32 flags = ri->flags;
2470 dev = dev_get_by_index_rcu(net, ri->tgt_index);
2475 if (flags & BPF_F_PEER) {
2476 const struct net_device_ops *ops = dev->netdev_ops;
2478 if (unlikely(!ops->ndo_get_peer_dev ||
2479 !skb_at_tc_ingress(skb)))
2481 dev = ops->ndo_get_peer_dev(dev);
2482 if (unlikely(!dev ||
2483 !is_skb_forwardable(dev, skb) ||
2484 net_eq(net, dev_net(dev))))
2489 return flags & BPF_F_NEIGH ?
2490 __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2492 __bpf_redirect(skb, dev, flags);
2498 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2500 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2502 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2506 ri->tgt_index = ifindex;
2508 return TC_ACT_REDIRECT;
2511 static const struct bpf_func_proto bpf_redirect_proto = {
2512 .func = bpf_redirect,
2514 .ret_type = RET_INTEGER,
2515 .arg1_type = ARG_ANYTHING,
2516 .arg2_type = ARG_ANYTHING,
2519 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2521 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2523 if (unlikely(flags))
2526 ri->flags = BPF_F_PEER;
2527 ri->tgt_index = ifindex;
2529 return TC_ACT_REDIRECT;
2532 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2533 .func = bpf_redirect_peer,
2535 .ret_type = RET_INTEGER,
2536 .arg1_type = ARG_ANYTHING,
2537 .arg2_type = ARG_ANYTHING,
2540 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2541 int, plen, u64, flags)
2543 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2545 if (unlikely((plen && plen < sizeof(*params)) || flags))
2548 ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2549 ri->tgt_index = ifindex;
2551 BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2553 memcpy(&ri->nh, params, sizeof(ri->nh));
2555 return TC_ACT_REDIRECT;
2558 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2559 .func = bpf_redirect_neigh,
2561 .ret_type = RET_INTEGER,
2562 .arg1_type = ARG_ANYTHING,
2563 .arg2_type = ARG_PTR_TO_MEM_OR_NULL,
2564 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
2565 .arg4_type = ARG_ANYTHING,
2568 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2570 msg->apply_bytes = bytes;
2574 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2575 .func = bpf_msg_apply_bytes,
2577 .ret_type = RET_INTEGER,
2578 .arg1_type = ARG_PTR_TO_CTX,
2579 .arg2_type = ARG_ANYTHING,
2582 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2584 msg->cork_bytes = bytes;
2588 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2589 .func = bpf_msg_cork_bytes,
2591 .ret_type = RET_INTEGER,
2592 .arg1_type = ARG_PTR_TO_CTX,
2593 .arg2_type = ARG_ANYTHING,
2596 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2597 u32, end, u64, flags)
2599 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2600 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2601 struct scatterlist *sge;
2602 u8 *raw, *to, *from;
2605 if (unlikely(flags || end <= start))
2608 /* First find the starting scatterlist element */
2612 len = sk_msg_elem(msg, i)->length;
2613 if (start < offset + len)
2615 sk_msg_iter_var_next(i);
2616 } while (i != msg->sg.end);
2618 if (unlikely(start >= offset + len))
2622 /* The start may point into the sg element so we need to also
2623 * account for the headroom.
2625 bytes_sg_total = start - offset + bytes;
2626 if (!test_bit(i, &msg->sg.copy) && bytes_sg_total <= len)
2629 /* At this point we need to linearize multiple scatterlist
2630 * elements or a single shared page. Either way we need to
2631 * copy into a linear buffer exclusively owned by BPF. Then
2632 * place the buffer in the scatterlist and fixup the original
2633 * entries by removing the entries now in the linear buffer
2634 * and shifting the remaining entries. For now we do not try
2635 * to copy partial entries to avoid complexity of running out
2636 * of sg_entry slots. The downside is reading a single byte
2637 * will copy the entire sg entry.
2640 copy += sk_msg_elem(msg, i)->length;
2641 sk_msg_iter_var_next(i);
2642 if (bytes_sg_total <= copy)
2644 } while (i != msg->sg.end);
2647 if (unlikely(bytes_sg_total > copy))
2650 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2652 if (unlikely(!page))
2655 raw = page_address(page);
2658 sge = sk_msg_elem(msg, i);
2659 from = sg_virt(sge);
2663 memcpy(to, from, len);
2666 put_page(sg_page(sge));
2668 sk_msg_iter_var_next(i);
2669 } while (i != last_sge);
2671 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2673 /* To repair sg ring we need to shift entries. If we only
2674 * had a single entry though we can just replace it and
2675 * be done. Otherwise walk the ring and shift the entries.
2677 WARN_ON_ONCE(last_sge == first_sge);
2678 shift = last_sge > first_sge ?
2679 last_sge - first_sge - 1 :
2680 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2685 sk_msg_iter_var_next(i);
2689 if (i + shift >= NR_MSG_FRAG_IDS)
2690 move_from = i + shift - NR_MSG_FRAG_IDS;
2692 move_from = i + shift;
2693 if (move_from == msg->sg.end)
2696 msg->sg.data[i] = msg->sg.data[move_from];
2697 msg->sg.data[move_from].length = 0;
2698 msg->sg.data[move_from].page_link = 0;
2699 msg->sg.data[move_from].offset = 0;
2700 sk_msg_iter_var_next(i);
2703 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2704 msg->sg.end - shift + NR_MSG_FRAG_IDS :
2705 msg->sg.end - shift;
2707 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2708 msg->data_end = msg->data + bytes;
2712 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2713 .func = bpf_msg_pull_data,
2715 .ret_type = RET_INTEGER,
2716 .arg1_type = ARG_PTR_TO_CTX,
2717 .arg2_type = ARG_ANYTHING,
2718 .arg3_type = ARG_ANYTHING,
2719 .arg4_type = ARG_ANYTHING,
2722 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2723 u32, len, u64, flags)
2725 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2726 u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2727 u8 *raw, *to, *from;
2730 if (unlikely(flags))
2733 /* First find the starting scatterlist element */
2737 l = sk_msg_elem(msg, i)->length;
2739 if (start < offset + l)
2741 sk_msg_iter_var_next(i);
2742 } while (i != msg->sg.end);
2744 if (start >= offset + l)
2747 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2749 /* If no space available will fallback to copy, we need at
2750 * least one scatterlist elem available to push data into
2751 * when start aligns to the beginning of an element or two
2752 * when it falls inside an element. We handle the start equals
2753 * offset case because its the common case for inserting a
2756 if (!space || (space == 1 && start != offset))
2757 copy = msg->sg.data[i].length;
2759 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2760 get_order(copy + len));
2761 if (unlikely(!page))
2767 raw = page_address(page);
2769 psge = sk_msg_elem(msg, i);
2770 front = start - offset;
2771 back = psge->length - front;
2772 from = sg_virt(psge);
2775 memcpy(raw, from, front);
2779 to = raw + front + len;
2781 memcpy(to, from, back);
2784 put_page(sg_page(psge));
2785 } else if (start - offset) {
2786 psge = sk_msg_elem(msg, i);
2787 rsge = sk_msg_elem_cpy(msg, i);
2789 psge->length = start - offset;
2790 rsge.length -= psge->length;
2791 rsge.offset += start;
2793 sk_msg_iter_var_next(i);
2794 sg_unmark_end(psge);
2795 sg_unmark_end(&rsge);
2796 sk_msg_iter_next(msg, end);
2799 /* Slot(s) to place newly allocated data */
2802 /* Shift one or two slots as needed */
2804 sge = sk_msg_elem_cpy(msg, i);
2806 sk_msg_iter_var_next(i);
2807 sg_unmark_end(&sge);
2808 sk_msg_iter_next(msg, end);
2810 nsge = sk_msg_elem_cpy(msg, i);
2812 sk_msg_iter_var_next(i);
2813 nnsge = sk_msg_elem_cpy(msg, i);
2816 while (i != msg->sg.end) {
2817 msg->sg.data[i] = sge;
2819 sk_msg_iter_var_next(i);
2822 nnsge = sk_msg_elem_cpy(msg, i);
2824 nsge = sk_msg_elem_cpy(msg, i);
2829 /* Place newly allocated data buffer */
2830 sk_mem_charge(msg->sk, len);
2831 msg->sg.size += len;
2832 __clear_bit(new, &msg->sg.copy);
2833 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2835 get_page(sg_page(&rsge));
2836 sk_msg_iter_var_next(new);
2837 msg->sg.data[new] = rsge;
2840 sk_msg_compute_data_pointers(msg);
2844 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2845 .func = bpf_msg_push_data,
2847 .ret_type = RET_INTEGER,
2848 .arg1_type = ARG_PTR_TO_CTX,
2849 .arg2_type = ARG_ANYTHING,
2850 .arg3_type = ARG_ANYTHING,
2851 .arg4_type = ARG_ANYTHING,
2854 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2860 sk_msg_iter_var_next(i);
2861 msg->sg.data[prev] = msg->sg.data[i];
2862 } while (i != msg->sg.end);
2864 sk_msg_iter_prev(msg, end);
2867 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2869 struct scatterlist tmp, sge;
2871 sk_msg_iter_next(msg, end);
2872 sge = sk_msg_elem_cpy(msg, i);
2873 sk_msg_iter_var_next(i);
2874 tmp = sk_msg_elem_cpy(msg, i);
2876 while (i != msg->sg.end) {
2877 msg->sg.data[i] = sge;
2878 sk_msg_iter_var_next(i);
2880 tmp = sk_msg_elem_cpy(msg, i);
2884 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2885 u32, len, u64, flags)
2887 u32 i = 0, l = 0, space, offset = 0;
2888 u64 last = start + len;
2891 if (unlikely(flags))
2894 /* First find the starting scatterlist element */
2898 l = sk_msg_elem(msg, i)->length;
2900 if (start < offset + l)
2902 sk_msg_iter_var_next(i);
2903 } while (i != msg->sg.end);
2905 /* Bounds checks: start and pop must be inside message */
2906 if (start >= offset + l || last >= msg->sg.size)
2909 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2912 /* --------------| offset
2913 * -| start |-------- len -------|
2915 * |----- a ----|-------- pop -------|----- b ----|
2916 * |______________________________________________| length
2919 * a: region at front of scatter element to save
2920 * b: region at back of scatter element to save when length > A + pop
2921 * pop: region to pop from element, same as input 'pop' here will be
2922 * decremented below per iteration.
2924 * Two top-level cases to handle when start != offset, first B is non
2925 * zero and second B is zero corresponding to when a pop includes more
2928 * Then if B is non-zero AND there is no space allocate space and
2929 * compact A, B regions into page. If there is space shift ring to
2930 * the rigth free'ing the next element in ring to place B, leaving
2931 * A untouched except to reduce length.
2933 if (start != offset) {
2934 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2936 int b = sge->length - pop - a;
2938 sk_msg_iter_var_next(i);
2940 if (pop < sge->length - a) {
2943 sk_msg_shift_right(msg, i);
2944 nsge = sk_msg_elem(msg, i);
2945 get_page(sg_page(sge));
2948 b, sge->offset + pop + a);
2950 struct page *page, *orig;
2953 page = alloc_pages(__GFP_NOWARN |
2954 __GFP_COMP | GFP_ATOMIC,
2956 if (unlikely(!page))
2960 orig = sg_page(sge);
2961 from = sg_virt(sge);
2962 to = page_address(page);
2963 memcpy(to, from, a);
2964 memcpy(to + a, from + a + pop, b);
2965 sg_set_page(sge, page, a + b, 0);
2969 } else if (pop >= sge->length - a) {
2970 pop -= (sge->length - a);
2975 /* From above the current layout _must_ be as follows,
2980 * |---- pop ---|---------------- b ------------|
2981 * |____________________________________________| length
2983 * Offset and start of the current msg elem are equal because in the
2984 * previous case we handled offset != start and either consumed the
2985 * entire element and advanced to the next element OR pop == 0.
2987 * Two cases to handle here are first pop is less than the length
2988 * leaving some remainder b above. Simply adjust the element's layout
2989 * in this case. Or pop >= length of the element so that b = 0. In this
2990 * case advance to next element decrementing pop.
2993 struct scatterlist *sge = sk_msg_elem(msg, i);
2995 if (pop < sge->length) {
3001 sk_msg_shift_left(msg, i);
3003 sk_msg_iter_var_next(i);
3006 sk_mem_uncharge(msg->sk, len - pop);
3007 msg->sg.size -= (len - pop);
3008 sk_msg_compute_data_pointers(msg);
3012 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
3013 .func = bpf_msg_pop_data,
3015 .ret_type = RET_INTEGER,
3016 .arg1_type = ARG_PTR_TO_CTX,
3017 .arg2_type = ARG_ANYTHING,
3018 .arg3_type = ARG_ANYTHING,
3019 .arg4_type = ARG_ANYTHING,
3022 #ifdef CONFIG_CGROUP_NET_CLASSID
3023 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3025 return __task_get_classid(current);
3028 static const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3029 .func = bpf_get_cgroup_classid_curr,
3031 .ret_type = RET_INTEGER,
3034 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3036 struct sock *sk = skb_to_full_sk(skb);
3038 if (!sk || !sk_fullsock(sk))
3041 return sock_cgroup_classid(&sk->sk_cgrp_data);
3044 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3045 .func = bpf_skb_cgroup_classid,
3047 .ret_type = RET_INTEGER,
3048 .arg1_type = ARG_PTR_TO_CTX,
3052 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3054 return task_get_classid(skb);
3057 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3058 .func = bpf_get_cgroup_classid,
3060 .ret_type = RET_INTEGER,
3061 .arg1_type = ARG_PTR_TO_CTX,
3064 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3066 return dst_tclassid(skb);
3069 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3070 .func = bpf_get_route_realm,
3072 .ret_type = RET_INTEGER,
3073 .arg1_type = ARG_PTR_TO_CTX,
3076 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3078 /* If skb_clear_hash() was called due to mangling, we can
3079 * trigger SW recalculation here. Later access to hash
3080 * can then use the inline skb->hash via context directly
3081 * instead of calling this helper again.
3083 return skb_get_hash(skb);
3086 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3087 .func = bpf_get_hash_recalc,
3089 .ret_type = RET_INTEGER,
3090 .arg1_type = ARG_PTR_TO_CTX,
3093 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3095 /* After all direct packet write, this can be used once for
3096 * triggering a lazy recalc on next skb_get_hash() invocation.
3098 skb_clear_hash(skb);
3102 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3103 .func = bpf_set_hash_invalid,
3105 .ret_type = RET_INTEGER,
3106 .arg1_type = ARG_PTR_TO_CTX,
3109 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3111 /* Set user specified hash as L4(+), so that it gets returned
3112 * on skb_get_hash() call unless BPF prog later on triggers a
3115 __skb_set_sw_hash(skb, hash, true);
3119 static const struct bpf_func_proto bpf_set_hash_proto = {
3120 .func = bpf_set_hash,
3122 .ret_type = RET_INTEGER,
3123 .arg1_type = ARG_PTR_TO_CTX,
3124 .arg2_type = ARG_ANYTHING,
3127 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3132 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3133 vlan_proto != htons(ETH_P_8021AD)))
3134 vlan_proto = htons(ETH_P_8021Q);
3136 bpf_push_mac_rcsum(skb);
3137 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3138 bpf_pull_mac_rcsum(skb);
3140 bpf_compute_data_pointers(skb);
3144 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3145 .func = bpf_skb_vlan_push,
3147 .ret_type = RET_INTEGER,
3148 .arg1_type = ARG_PTR_TO_CTX,
3149 .arg2_type = ARG_ANYTHING,
3150 .arg3_type = ARG_ANYTHING,
3153 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3157 bpf_push_mac_rcsum(skb);
3158 ret = skb_vlan_pop(skb);
3159 bpf_pull_mac_rcsum(skb);
3161 bpf_compute_data_pointers(skb);
3165 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3166 .func = bpf_skb_vlan_pop,
3168 .ret_type = RET_INTEGER,
3169 .arg1_type = ARG_PTR_TO_CTX,
3172 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3174 /* Caller already did skb_cow() with len as headroom,
3175 * so no need to do it here.
3178 memmove(skb->data, skb->data + len, off);
3179 memset(skb->data + off, 0, len);
3181 /* No skb_postpush_rcsum(skb, skb->data + off, len)
3182 * needed here as it does not change the skb->csum
3183 * result for checksum complete when summing over
3189 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3191 /* skb_ensure_writable() is not needed here, as we're
3192 * already working on an uncloned skb.
3194 if (unlikely(!pskb_may_pull(skb, off + len)))
3197 skb_postpull_rcsum(skb, skb->data + off, len);
3198 memmove(skb->data + len, skb->data, off);
3199 __skb_pull(skb, len);
3204 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3206 bool trans_same = skb->transport_header == skb->network_header;
3209 /* There's no need for __skb_push()/__skb_pull() pair to
3210 * get to the start of the mac header as we're guaranteed
3211 * to always start from here under eBPF.
3213 ret = bpf_skb_generic_push(skb, off, len);
3215 skb->mac_header -= len;
3216 skb->network_header -= len;
3218 skb->transport_header = skb->network_header;
3224 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3226 bool trans_same = skb->transport_header == skb->network_header;
3229 /* Same here, __skb_push()/__skb_pull() pair not needed. */
3230 ret = bpf_skb_generic_pop(skb, off, len);
3232 skb->mac_header += len;
3233 skb->network_header += len;
3235 skb->transport_header = skb->network_header;
3241 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3243 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3244 u32 off = skb_mac_header_len(skb);
3247 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
3250 ret = skb_cow(skb, len_diff);
3251 if (unlikely(ret < 0))
3254 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3255 if (unlikely(ret < 0))
3258 if (skb_is_gso(skb)) {
3259 struct skb_shared_info *shinfo = skb_shinfo(skb);
3261 /* SKB_GSO_TCPV4 needs to be changed into
3264 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3265 shinfo->gso_type &= ~SKB_GSO_TCPV4;
3266 shinfo->gso_type |= SKB_GSO_TCPV6;
3269 /* Due to IPv6 header, MSS needs to be downgraded. */
3270 skb_decrease_gso_size(shinfo, len_diff);
3271 /* Header must be checked, and gso_segs recomputed. */
3272 shinfo->gso_type |= SKB_GSO_DODGY;
3273 shinfo->gso_segs = 0;
3276 skb->protocol = htons(ETH_P_IPV6);
3277 skb_clear_hash(skb);
3282 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3284 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3285 u32 off = skb_mac_header_len(skb);
3288 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
3291 ret = skb_unclone(skb, GFP_ATOMIC);
3292 if (unlikely(ret < 0))
3295 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3296 if (unlikely(ret < 0))
3299 if (skb_is_gso(skb)) {
3300 struct skb_shared_info *shinfo = skb_shinfo(skb);
3302 /* SKB_GSO_TCPV6 needs to be changed into
3305 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3306 shinfo->gso_type &= ~SKB_GSO_TCPV6;
3307 shinfo->gso_type |= SKB_GSO_TCPV4;
3310 /* Due to IPv4 header, MSS can be upgraded. */
3311 skb_increase_gso_size(shinfo, len_diff);
3312 /* Header must be checked, and gso_segs recomputed. */
3313 shinfo->gso_type |= SKB_GSO_DODGY;
3314 shinfo->gso_segs = 0;
3317 skb->protocol = htons(ETH_P_IP);
3318 skb_clear_hash(skb);
3323 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3325 __be16 from_proto = skb->protocol;
3327 if (from_proto == htons(ETH_P_IP) &&
3328 to_proto == htons(ETH_P_IPV6))
3329 return bpf_skb_proto_4_to_6(skb);
3331 if (from_proto == htons(ETH_P_IPV6) &&
3332 to_proto == htons(ETH_P_IP))
3333 return bpf_skb_proto_6_to_4(skb);
3338 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3343 if (unlikely(flags))
3346 /* General idea is that this helper does the basic groundwork
3347 * needed for changing the protocol, and eBPF program fills the
3348 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3349 * and other helpers, rather than passing a raw buffer here.
3351 * The rationale is to keep this minimal and without a need to
3352 * deal with raw packet data. F.e. even if we would pass buffers
3353 * here, the program still needs to call the bpf_lX_csum_replace()
3354 * helpers anyway. Plus, this way we keep also separation of
3355 * concerns, since f.e. bpf_skb_store_bytes() should only take
3358 * Currently, additional options and extension header space are
3359 * not supported, but flags register is reserved so we can adapt
3360 * that. For offloads, we mark packet as dodgy, so that headers
3361 * need to be verified first.
3363 ret = bpf_skb_proto_xlat(skb, proto);
3364 bpf_compute_data_pointers(skb);
3368 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3369 .func = bpf_skb_change_proto,
3371 .ret_type = RET_INTEGER,
3372 .arg1_type = ARG_PTR_TO_CTX,
3373 .arg2_type = ARG_ANYTHING,
3374 .arg3_type = ARG_ANYTHING,
3377 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3379 /* We only allow a restricted subset to be changed for now. */
3380 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3381 !skb_pkt_type_ok(pkt_type)))
3384 skb->pkt_type = pkt_type;
3388 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3389 .func = bpf_skb_change_type,
3391 .ret_type = RET_INTEGER,
3392 .arg1_type = ARG_PTR_TO_CTX,
3393 .arg2_type = ARG_ANYTHING,
3396 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3398 switch (skb->protocol) {
3399 case htons(ETH_P_IP):
3400 return sizeof(struct iphdr);
3401 case htons(ETH_P_IPV6):
3402 return sizeof(struct ipv6hdr);
3408 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3409 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3411 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \
3412 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3413 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3414 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3415 BPF_F_ADJ_ROOM_ENCAP_L2( \
3416 BPF_ADJ_ROOM_ENCAP_L2_MASK))
3418 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3421 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3422 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3423 u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3424 unsigned int gso_type = SKB_GSO_DODGY;
3427 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3428 /* udp gso_size delineates datagrams, only allow if fixed */
3429 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3430 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3434 ret = skb_cow_head(skb, len_diff);
3435 if (unlikely(ret < 0))
3439 if (skb->protocol != htons(ETH_P_IP) &&
3440 skb->protocol != htons(ETH_P_IPV6))
3443 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3444 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3447 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3448 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3451 if (skb->encapsulation)
3454 mac_len = skb->network_header - skb->mac_header;
3455 inner_net = skb->network_header;
3456 if (inner_mac_len > len_diff)
3458 inner_trans = skb->transport_header;
3461 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3462 if (unlikely(ret < 0))
3466 skb->inner_mac_header = inner_net - inner_mac_len;
3467 skb->inner_network_header = inner_net;
3468 skb->inner_transport_header = inner_trans;
3469 skb_set_inner_protocol(skb, skb->protocol);
3471 skb->encapsulation = 1;
3472 skb_set_network_header(skb, mac_len);
3474 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3475 gso_type |= SKB_GSO_UDP_TUNNEL;
3476 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3477 gso_type |= SKB_GSO_GRE;
3478 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3479 gso_type |= SKB_GSO_IPXIP6;
3480 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3481 gso_type |= SKB_GSO_IPXIP4;
3483 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3484 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3485 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3486 sizeof(struct ipv6hdr) :
3487 sizeof(struct iphdr);
3489 skb_set_transport_header(skb, mac_len + nh_len);
3492 /* Match skb->protocol to new outer l3 protocol */
3493 if (skb->protocol == htons(ETH_P_IP) &&
3494 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3495 skb->protocol = htons(ETH_P_IPV6);
3496 else if (skb->protocol == htons(ETH_P_IPV6) &&
3497 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3498 skb->protocol = htons(ETH_P_IP);
3501 if (skb_is_gso(skb)) {
3502 struct skb_shared_info *shinfo = skb_shinfo(skb);
3504 /* Due to header grow, MSS needs to be downgraded. */
3505 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3506 skb_decrease_gso_size(shinfo, len_diff);
3508 /* Header must be checked, and gso_segs recomputed. */
3509 shinfo->gso_type |= gso_type;
3510 shinfo->gso_segs = 0;
3516 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3521 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3522 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3525 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3526 /* udp gso_size delineates datagrams, only allow if fixed */
3527 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3528 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3532 ret = skb_unclone(skb, GFP_ATOMIC);
3533 if (unlikely(ret < 0))
3536 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3537 if (unlikely(ret < 0))
3540 if (skb_is_gso(skb)) {
3541 struct skb_shared_info *shinfo = skb_shinfo(skb);
3543 /* Due to header shrink, MSS can be upgraded. */
3544 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3545 skb_increase_gso_size(shinfo, len_diff);
3547 /* Header must be checked, and gso_segs recomputed. */
3548 shinfo->gso_type |= SKB_GSO_DODGY;
3549 shinfo->gso_segs = 0;
3555 static u32 __bpf_skb_max_len(const struct sk_buff *skb)
3557 return skb->dev ? skb->dev->mtu + skb->dev->hard_header_len :
3561 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3562 u32, mode, u64, flags)
3564 u32 len_diff_abs = abs(len_diff);
3565 bool shrink = len_diff < 0;
3568 if (unlikely(flags || mode))
3570 if (unlikely(len_diff_abs > 0xfffU))
3574 ret = skb_cow(skb, len_diff);
3575 if (unlikely(ret < 0))
3577 __skb_push(skb, len_diff_abs);
3578 memset(skb->data, 0, len_diff_abs);
3580 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3582 __skb_pull(skb, len_diff_abs);
3584 bpf_compute_data_end_sk_skb(skb);
3585 if (tls_sw_has_ctx_rx(skb->sk)) {
3586 struct strp_msg *rxm = strp_msg(skb);
3588 rxm->full_len += len_diff;
3593 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3594 .func = sk_skb_adjust_room,
3596 .ret_type = RET_INTEGER,
3597 .arg1_type = ARG_PTR_TO_CTX,
3598 .arg2_type = ARG_ANYTHING,
3599 .arg3_type = ARG_ANYTHING,
3600 .arg4_type = ARG_ANYTHING,
3603 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3604 u32, mode, u64, flags)
3606 u32 len_cur, len_diff_abs = abs(len_diff);
3607 u32 len_min = bpf_skb_net_base_len(skb);
3608 u32 len_max = __bpf_skb_max_len(skb);
3609 __be16 proto = skb->protocol;
3610 bool shrink = len_diff < 0;
3614 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3615 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3617 if (unlikely(len_diff_abs > 0xfffU))
3619 if (unlikely(proto != htons(ETH_P_IP) &&
3620 proto != htons(ETH_P_IPV6)))
3623 off = skb_mac_header_len(skb);
3625 case BPF_ADJ_ROOM_NET:
3626 off += bpf_skb_net_base_len(skb);
3628 case BPF_ADJ_ROOM_MAC:
3634 len_cur = skb->len - skb_network_offset(skb);
3635 if ((shrink && (len_diff_abs >= len_cur ||
3636 len_cur - len_diff_abs < len_min)) ||
3637 (!shrink && (skb->len + len_diff_abs > len_max &&
3641 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3642 bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3643 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3644 __skb_reset_checksum_unnecessary(skb);
3646 bpf_compute_data_pointers(skb);
3650 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3651 .func = bpf_skb_adjust_room,
3653 .ret_type = RET_INTEGER,
3654 .arg1_type = ARG_PTR_TO_CTX,
3655 .arg2_type = ARG_ANYTHING,
3656 .arg3_type = ARG_ANYTHING,
3657 .arg4_type = ARG_ANYTHING,
3660 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3662 u32 min_len = skb_network_offset(skb);
3664 if (skb_transport_header_was_set(skb))
3665 min_len = skb_transport_offset(skb);
3666 if (skb->ip_summed == CHECKSUM_PARTIAL)
3667 min_len = skb_checksum_start_offset(skb) +
3668 skb->csum_offset + sizeof(__sum16);
3672 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3674 unsigned int old_len = skb->len;
3677 ret = __skb_grow_rcsum(skb, new_len);
3679 memset(skb->data + old_len, 0, new_len - old_len);
3683 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3685 return __skb_trim_rcsum(skb, new_len);
3688 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3691 u32 max_len = __bpf_skb_max_len(skb);
3692 u32 min_len = __bpf_skb_min_len(skb);
3695 if (unlikely(flags || new_len > max_len || new_len < min_len))
3697 if (skb->encapsulation)
3700 /* The basic idea of this helper is that it's performing the
3701 * needed work to either grow or trim an skb, and eBPF program
3702 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3703 * bpf_lX_csum_replace() and others rather than passing a raw
3704 * buffer here. This one is a slow path helper and intended
3705 * for replies with control messages.
3707 * Like in bpf_skb_change_proto(), we want to keep this rather
3708 * minimal and without protocol specifics so that we are able
3709 * to separate concerns as in bpf_skb_store_bytes() should only
3710 * be the one responsible for writing buffers.
3712 * It's really expected to be a slow path operation here for
3713 * control message replies, so we're implicitly linearizing,
3714 * uncloning and drop offloads from the skb by this.
3716 ret = __bpf_try_make_writable(skb, skb->len);
3718 if (new_len > skb->len)
3719 ret = bpf_skb_grow_rcsum(skb, new_len);
3720 else if (new_len < skb->len)
3721 ret = bpf_skb_trim_rcsum(skb, new_len);
3722 if (!ret && skb_is_gso(skb))
3728 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3731 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3733 bpf_compute_data_pointers(skb);
3737 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3738 .func = bpf_skb_change_tail,
3740 .ret_type = RET_INTEGER,
3741 .arg1_type = ARG_PTR_TO_CTX,
3742 .arg2_type = ARG_ANYTHING,
3743 .arg3_type = ARG_ANYTHING,
3746 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3749 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3751 bpf_compute_data_end_sk_skb(skb);
3755 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3756 .func = sk_skb_change_tail,
3758 .ret_type = RET_INTEGER,
3759 .arg1_type = ARG_PTR_TO_CTX,
3760 .arg2_type = ARG_ANYTHING,
3761 .arg3_type = ARG_ANYTHING,
3764 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3767 u32 max_len = __bpf_skb_max_len(skb);
3768 u32 new_len = skb->len + head_room;
3771 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3772 new_len < skb->len))
3775 ret = skb_cow(skb, head_room);
3777 /* Idea for this helper is that we currently only
3778 * allow to expand on mac header. This means that
3779 * skb->protocol network header, etc, stay as is.
3780 * Compared to bpf_skb_change_tail(), we're more
3781 * flexible due to not needing to linearize or
3782 * reset GSO. Intention for this helper is to be
3783 * used by an L3 skb that needs to push mac header
3784 * for redirection into L2 device.
3786 __skb_push(skb, head_room);
3787 memset(skb->data, 0, head_room);
3788 skb_reset_mac_header(skb);
3794 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3797 int ret = __bpf_skb_change_head(skb, head_room, flags);
3799 bpf_compute_data_pointers(skb);
3803 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3804 .func = bpf_skb_change_head,
3806 .ret_type = RET_INTEGER,
3807 .arg1_type = ARG_PTR_TO_CTX,
3808 .arg2_type = ARG_ANYTHING,
3809 .arg3_type = ARG_ANYTHING,
3812 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3815 int ret = __bpf_skb_change_head(skb, head_room, flags);
3817 bpf_compute_data_end_sk_skb(skb);
3821 static const struct bpf_func_proto sk_skb_change_head_proto = {
3822 .func = sk_skb_change_head,
3824 .ret_type = RET_INTEGER,
3825 .arg1_type = ARG_PTR_TO_CTX,
3826 .arg2_type = ARG_ANYTHING,
3827 .arg3_type = ARG_ANYTHING,
3829 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3831 return xdp_data_meta_unsupported(xdp) ? 0 :
3832 xdp->data - xdp->data_meta;
3835 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3837 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3838 unsigned long metalen = xdp_get_metalen(xdp);
3839 void *data_start = xdp_frame_end + metalen;
3840 void *data = xdp->data + offset;
3842 if (unlikely(data < data_start ||
3843 data > xdp->data_end - ETH_HLEN))
3847 memmove(xdp->data_meta + offset,
3848 xdp->data_meta, metalen);
3849 xdp->data_meta += offset;
3855 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3856 .func = bpf_xdp_adjust_head,
3858 .ret_type = RET_INTEGER,
3859 .arg1_type = ARG_PTR_TO_CTX,
3860 .arg2_type = ARG_ANYTHING,
3863 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3865 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
3866 void *data_end = xdp->data_end + offset;
3868 /* Notice that xdp_data_hard_end have reserved some tailroom */
3869 if (unlikely(data_end > data_hard_end))
3872 /* ALL drivers MUST init xdp->frame_sz, chicken check below */
3873 if (unlikely(xdp->frame_sz > PAGE_SIZE)) {
3874 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz);
3878 if (unlikely(data_end < xdp->data + ETH_HLEN))
3881 /* Clear memory area on grow, can contain uninit kernel memory */
3883 memset(xdp->data_end, 0, offset);
3885 xdp->data_end = data_end;
3890 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3891 .func = bpf_xdp_adjust_tail,
3893 .ret_type = RET_INTEGER,
3894 .arg1_type = ARG_PTR_TO_CTX,
3895 .arg2_type = ARG_ANYTHING,
3898 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3900 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3901 void *meta = xdp->data_meta + offset;
3902 unsigned long metalen = xdp->data - meta;
3904 if (xdp_data_meta_unsupported(xdp))
3906 if (unlikely(meta < xdp_frame_end ||
3909 if (unlikely((metalen & (sizeof(__u32) - 1)) ||
3913 xdp->data_meta = meta;
3918 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3919 .func = bpf_xdp_adjust_meta,
3921 .ret_type = RET_INTEGER,
3922 .arg1_type = ARG_PTR_TO_CTX,
3923 .arg2_type = ARG_ANYTHING,
3926 static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd,
3927 struct bpf_map *map, struct xdp_buff *xdp)
3929 switch (map->map_type) {
3930 case BPF_MAP_TYPE_DEVMAP:
3931 case BPF_MAP_TYPE_DEVMAP_HASH:
3932 return dev_map_enqueue(fwd, xdp, dev_rx);
3933 case BPF_MAP_TYPE_CPUMAP:
3934 return cpu_map_enqueue(fwd, xdp, dev_rx);
3935 case BPF_MAP_TYPE_XSKMAP:
3936 return __xsk_map_redirect(fwd, xdp);
3943 void xdp_do_flush(void)
3949 EXPORT_SYMBOL_GPL(xdp_do_flush);
3951 static inline void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index)
3953 switch (map->map_type) {
3954 case BPF_MAP_TYPE_DEVMAP:
3955 return __dev_map_lookup_elem(map, index);
3956 case BPF_MAP_TYPE_DEVMAP_HASH:
3957 return __dev_map_hash_lookup_elem(map, index);
3958 case BPF_MAP_TYPE_CPUMAP:
3959 return __cpu_map_lookup_elem(map, index);
3960 case BPF_MAP_TYPE_XSKMAP:
3961 return __xsk_map_lookup_elem(map, index);
3967 void bpf_clear_redirect_map(struct bpf_map *map)
3969 struct bpf_redirect_info *ri;
3972 for_each_possible_cpu(cpu) {
3973 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3974 /* Avoid polluting remote cacheline due to writes if
3975 * not needed. Once we pass this test, we need the
3976 * cmpxchg() to make sure it hasn't been changed in
3977 * the meantime by remote CPU.
3979 if (unlikely(READ_ONCE(ri->map) == map))
3980 cmpxchg(&ri->map, map, NULL);
3984 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3985 struct bpf_prog *xdp_prog)
3987 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3988 struct bpf_map *map = READ_ONCE(ri->map);
3989 u32 index = ri->tgt_index;
3990 void *fwd = ri->tgt_value;
3994 ri->tgt_value = NULL;
3995 WRITE_ONCE(ri->map, NULL);
3997 if (unlikely(!map)) {
3998 fwd = dev_get_by_index_rcu(dev_net(dev), index);
3999 if (unlikely(!fwd)) {
4004 err = dev_xdp_enqueue(fwd, xdp, dev);
4006 err = __bpf_tx_xdp_map(dev, fwd, map, xdp);
4012 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
4015 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
4018 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4020 static int xdp_do_generic_redirect_map(struct net_device *dev,
4021 struct sk_buff *skb,
4022 struct xdp_buff *xdp,
4023 struct bpf_prog *xdp_prog,
4024 struct bpf_map *map)
4026 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4027 u32 index = ri->tgt_index;
4028 void *fwd = ri->tgt_value;
4032 ri->tgt_value = NULL;
4033 WRITE_ONCE(ri->map, NULL);
4035 if (map->map_type == BPF_MAP_TYPE_DEVMAP ||
4036 map->map_type == BPF_MAP_TYPE_DEVMAP_HASH) {
4037 struct bpf_dtab_netdev *dst = fwd;
4039 err = dev_map_generic_redirect(dst, skb, xdp_prog);
4042 } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) {
4043 struct xdp_sock *xs = fwd;
4045 err = xsk_generic_rcv(xs, xdp);
4050 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
4055 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
4058 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
4062 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4063 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4065 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4066 struct bpf_map *map = READ_ONCE(ri->map);
4067 u32 index = ri->tgt_index;
4068 struct net_device *fwd;
4072 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog,
4075 fwd = dev_get_by_index_rcu(dev_net(dev), index);
4076 if (unlikely(!fwd)) {
4081 err = xdp_ok_fwd_dev(fwd, skb->len);
4086 _trace_xdp_redirect(dev, xdp_prog, index);
4087 generic_xdp_tx(skb, xdp_prog);
4090 _trace_xdp_redirect_err(dev, xdp_prog, index, err);
4094 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4096 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4098 if (unlikely(flags))
4102 ri->tgt_index = ifindex;
4103 ri->tgt_value = NULL;
4104 WRITE_ONCE(ri->map, NULL);
4106 return XDP_REDIRECT;
4109 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4110 .func = bpf_xdp_redirect,
4112 .ret_type = RET_INTEGER,
4113 .arg1_type = ARG_ANYTHING,
4114 .arg2_type = ARG_ANYTHING,
4117 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
4120 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4122 /* Lower bits of the flags are used as return code on lookup failure */
4123 if (unlikely(flags > XDP_TX))
4126 ri->tgt_value = __xdp_map_lookup_elem(map, ifindex);
4127 if (unlikely(!ri->tgt_value)) {
4128 /* If the lookup fails we want to clear out the state in the
4129 * redirect_info struct completely, so that if an eBPF program
4130 * performs multiple lookups, the last one always takes
4133 WRITE_ONCE(ri->map, NULL);
4138 ri->tgt_index = ifindex;
4139 WRITE_ONCE(ri->map, map);
4141 return XDP_REDIRECT;
4144 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4145 .func = bpf_xdp_redirect_map,
4147 .ret_type = RET_INTEGER,
4148 .arg1_type = ARG_CONST_MAP_PTR,
4149 .arg2_type = ARG_ANYTHING,
4150 .arg3_type = ARG_ANYTHING,
4153 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4154 unsigned long off, unsigned long len)
4156 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4160 if (ptr != dst_buff)
4161 memcpy(dst_buff, ptr, len);
4166 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4167 u64, flags, void *, meta, u64, meta_size)
4169 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4171 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4173 if (unlikely(!skb || skb_size > skb->len))
4176 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4180 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4181 .func = bpf_skb_event_output,
4183 .ret_type = RET_INTEGER,
4184 .arg1_type = ARG_PTR_TO_CTX,
4185 .arg2_type = ARG_CONST_MAP_PTR,
4186 .arg3_type = ARG_ANYTHING,
4187 .arg4_type = ARG_PTR_TO_MEM,
4188 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4191 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4193 const struct bpf_func_proto bpf_skb_output_proto = {
4194 .func = bpf_skb_event_output,
4196 .ret_type = RET_INTEGER,
4197 .arg1_type = ARG_PTR_TO_BTF_ID,
4198 .arg1_btf_id = &bpf_skb_output_btf_ids[0],
4199 .arg2_type = ARG_CONST_MAP_PTR,
4200 .arg3_type = ARG_ANYTHING,
4201 .arg4_type = ARG_PTR_TO_MEM,
4202 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4205 static unsigned short bpf_tunnel_key_af(u64 flags)
4207 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4210 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4211 u32, size, u64, flags)
4213 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4214 u8 compat[sizeof(struct bpf_tunnel_key)];
4218 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
4222 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4226 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4229 case offsetof(struct bpf_tunnel_key, tunnel_label):
4230 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4232 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4233 /* Fixup deprecated structure layouts here, so we have
4234 * a common path later on.
4236 if (ip_tunnel_info_af(info) != AF_INET)
4239 to = (struct bpf_tunnel_key *)compat;
4246 to->tunnel_id = be64_to_cpu(info->key.tun_id);
4247 to->tunnel_tos = info->key.tos;
4248 to->tunnel_ttl = info->key.ttl;
4251 if (flags & BPF_F_TUNINFO_IPV6) {
4252 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4253 sizeof(to->remote_ipv6));
4254 to->tunnel_label = be32_to_cpu(info->key.label);
4256 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4257 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4258 to->tunnel_label = 0;
4261 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4262 memcpy(to_orig, to, size);
4266 memset(to_orig, 0, size);
4270 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4271 .func = bpf_skb_get_tunnel_key,
4273 .ret_type = RET_INTEGER,
4274 .arg1_type = ARG_PTR_TO_CTX,
4275 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4276 .arg3_type = ARG_CONST_SIZE,
4277 .arg4_type = ARG_ANYTHING,
4280 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4282 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4285 if (unlikely(!info ||
4286 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4290 if (unlikely(size < info->options_len)) {
4295 ip_tunnel_info_opts_get(to, info);
4296 if (size > info->options_len)
4297 memset(to + info->options_len, 0, size - info->options_len);
4299 return info->options_len;
4301 memset(to, 0, size);
4305 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4306 .func = bpf_skb_get_tunnel_opt,
4308 .ret_type = RET_INTEGER,
4309 .arg1_type = ARG_PTR_TO_CTX,
4310 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4311 .arg3_type = ARG_CONST_SIZE,
4314 static struct metadata_dst __percpu *md_dst;
4316 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4317 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4319 struct metadata_dst *md = this_cpu_ptr(md_dst);
4320 u8 compat[sizeof(struct bpf_tunnel_key)];
4321 struct ip_tunnel_info *info;
4323 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4324 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
4326 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4328 case offsetof(struct bpf_tunnel_key, tunnel_label):
4329 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4330 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4331 /* Fixup deprecated structure layouts here, so we have
4332 * a common path later on.
4334 memcpy(compat, from, size);
4335 memset(compat + size, 0, sizeof(compat) - size);
4336 from = (const struct bpf_tunnel_key *) compat;
4342 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4347 dst_hold((struct dst_entry *) md);
4348 skb_dst_set(skb, (struct dst_entry *) md);
4350 info = &md->u.tun_info;
4351 memset(info, 0, sizeof(*info));
4352 info->mode = IP_TUNNEL_INFO_TX;
4354 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4355 if (flags & BPF_F_DONT_FRAGMENT)
4356 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4357 if (flags & BPF_F_ZERO_CSUM_TX)
4358 info->key.tun_flags &= ~TUNNEL_CSUM;
4359 if (flags & BPF_F_SEQ_NUMBER)
4360 info->key.tun_flags |= TUNNEL_SEQ;
4362 info->key.tun_id = cpu_to_be64(from->tunnel_id);
4363 info->key.tos = from->tunnel_tos;
4364 info->key.ttl = from->tunnel_ttl;
4366 if (flags & BPF_F_TUNINFO_IPV6) {
4367 info->mode |= IP_TUNNEL_INFO_IPV6;
4368 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4369 sizeof(from->remote_ipv6));
4370 info->key.label = cpu_to_be32(from->tunnel_label) &
4371 IPV6_FLOWLABEL_MASK;
4373 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4379 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4380 .func = bpf_skb_set_tunnel_key,
4382 .ret_type = RET_INTEGER,
4383 .arg1_type = ARG_PTR_TO_CTX,
4384 .arg2_type = ARG_PTR_TO_MEM,
4385 .arg3_type = ARG_CONST_SIZE,
4386 .arg4_type = ARG_ANYTHING,
4389 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4390 const u8 *, from, u32, size)
4392 struct ip_tunnel_info *info = skb_tunnel_info(skb);
4393 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4395 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4397 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4400 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4405 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4406 .func = bpf_skb_set_tunnel_opt,
4408 .ret_type = RET_INTEGER,
4409 .arg1_type = ARG_PTR_TO_CTX,
4410 .arg2_type = ARG_PTR_TO_MEM,
4411 .arg3_type = ARG_CONST_SIZE,
4414 static const struct bpf_func_proto *
4415 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4418 struct metadata_dst __percpu *tmp;
4420 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4425 if (cmpxchg(&md_dst, NULL, tmp))
4426 metadata_dst_free_percpu(tmp);
4430 case BPF_FUNC_skb_set_tunnel_key:
4431 return &bpf_skb_set_tunnel_key_proto;
4432 case BPF_FUNC_skb_set_tunnel_opt:
4433 return &bpf_skb_set_tunnel_opt_proto;
4439 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4442 struct bpf_array *array = container_of(map, struct bpf_array, map);
4443 struct cgroup *cgrp;
4446 sk = skb_to_full_sk(skb);
4447 if (!sk || !sk_fullsock(sk))
4449 if (unlikely(idx >= array->map.max_entries))
4452 cgrp = READ_ONCE(array->ptrs[idx]);
4453 if (unlikely(!cgrp))
4456 return sk_under_cgroup_hierarchy(sk, cgrp);
4459 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4460 .func = bpf_skb_under_cgroup,
4462 .ret_type = RET_INTEGER,
4463 .arg1_type = ARG_PTR_TO_CTX,
4464 .arg2_type = ARG_CONST_MAP_PTR,
4465 .arg3_type = ARG_ANYTHING,
4468 #ifdef CONFIG_SOCK_CGROUP_DATA
4469 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4471 struct cgroup *cgrp;
4473 sk = sk_to_full_sk(sk);
4474 if (!sk || !sk_fullsock(sk))
4477 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4478 return cgroup_id(cgrp);
4481 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4483 return __bpf_sk_cgroup_id(skb->sk);
4486 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4487 .func = bpf_skb_cgroup_id,
4489 .ret_type = RET_INTEGER,
4490 .arg1_type = ARG_PTR_TO_CTX,
4493 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4496 struct cgroup *ancestor;
4497 struct cgroup *cgrp;
4499 sk = sk_to_full_sk(sk);
4500 if (!sk || !sk_fullsock(sk))
4503 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4504 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4508 return cgroup_id(ancestor);
4511 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4514 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4517 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4518 .func = bpf_skb_ancestor_cgroup_id,
4520 .ret_type = RET_INTEGER,
4521 .arg1_type = ARG_PTR_TO_CTX,
4522 .arg2_type = ARG_ANYTHING,
4525 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4527 return __bpf_sk_cgroup_id(sk);
4530 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4531 .func = bpf_sk_cgroup_id,
4533 .ret_type = RET_INTEGER,
4534 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4537 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4539 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4542 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4543 .func = bpf_sk_ancestor_cgroup_id,
4545 .ret_type = RET_INTEGER,
4546 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4547 .arg2_type = ARG_ANYTHING,
4551 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4552 unsigned long off, unsigned long len)
4554 memcpy(dst_buff, src_buff + off, len);
4558 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4559 u64, flags, void *, meta, u64, meta_size)
4561 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4563 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4565 if (unlikely(!xdp ||
4566 xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4569 return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4570 xdp_size, bpf_xdp_copy);
4573 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4574 .func = bpf_xdp_event_output,
4576 .ret_type = RET_INTEGER,
4577 .arg1_type = ARG_PTR_TO_CTX,
4578 .arg2_type = ARG_CONST_MAP_PTR,
4579 .arg3_type = ARG_ANYTHING,
4580 .arg4_type = ARG_PTR_TO_MEM,
4581 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4584 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4586 const struct bpf_func_proto bpf_xdp_output_proto = {
4587 .func = bpf_xdp_event_output,
4589 .ret_type = RET_INTEGER,
4590 .arg1_type = ARG_PTR_TO_BTF_ID,
4591 .arg1_btf_id = &bpf_xdp_output_btf_ids[0],
4592 .arg2_type = ARG_CONST_MAP_PTR,
4593 .arg3_type = ARG_ANYTHING,
4594 .arg4_type = ARG_PTR_TO_MEM,
4595 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4598 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4600 return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4603 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4604 .func = bpf_get_socket_cookie,
4606 .ret_type = RET_INTEGER,
4607 .arg1_type = ARG_PTR_TO_CTX,
4610 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4612 return __sock_gen_cookie(ctx->sk);
4615 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4616 .func = bpf_get_socket_cookie_sock_addr,
4618 .ret_type = RET_INTEGER,
4619 .arg1_type = ARG_PTR_TO_CTX,
4622 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4624 return __sock_gen_cookie(ctx);
4627 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4628 .func = bpf_get_socket_cookie_sock,
4630 .ret_type = RET_INTEGER,
4631 .arg1_type = ARG_PTR_TO_CTX,
4634 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4636 return __sock_gen_cookie(ctx->sk);
4639 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4640 .func = bpf_get_socket_cookie_sock_ops,
4642 .ret_type = RET_INTEGER,
4643 .arg1_type = ARG_PTR_TO_CTX,
4646 static u64 __bpf_get_netns_cookie(struct sock *sk)
4648 #ifdef CONFIG_NET_NS
4649 return __net_gen_cookie(sk ? sk->sk_net.net : &init_net);
4655 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
4657 return __bpf_get_netns_cookie(ctx);
4660 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
4661 .func = bpf_get_netns_cookie_sock,
4663 .ret_type = RET_INTEGER,
4664 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4667 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4669 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4672 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
4673 .func = bpf_get_netns_cookie_sock_addr,
4675 .ret_type = RET_INTEGER,
4676 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4679 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4681 struct sock *sk = sk_to_full_sk(skb->sk);
4684 if (!sk || !sk_fullsock(sk))
4686 kuid = sock_net_uid(sock_net(sk), sk);
4687 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4690 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4691 .func = bpf_get_socket_uid,
4693 .ret_type = RET_INTEGER,
4694 .arg1_type = ARG_PTR_TO_CTX,
4697 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
4698 char *optval, int optlen)
4700 char devname[IFNAMSIZ];
4706 if (!sk_fullsock(sk))
4709 sock_owned_by_me(sk);
4711 if (level == SOL_SOCKET) {
4712 if (optlen != sizeof(int) && optname != SO_BINDTODEVICE)
4714 val = *((int *)optval);
4715 valbool = val ? 1 : 0;
4717 /* Only some socketops are supported */
4720 val = min_t(u32, val, sysctl_rmem_max);
4721 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4722 WRITE_ONCE(sk->sk_rcvbuf,
4723 max_t(int, val * 2, SOCK_MIN_RCVBUF));
4726 val = min_t(u32, val, sysctl_wmem_max);
4727 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4728 WRITE_ONCE(sk->sk_sndbuf,
4729 max_t(int, val * 2, SOCK_MIN_SNDBUF));
4731 case SO_MAX_PACING_RATE: /* 32bit version */
4733 cmpxchg(&sk->sk_pacing_status,
4736 sk->sk_max_pacing_rate = (val == ~0U) ?
4737 ~0UL : (unsigned int)val;
4738 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4739 sk->sk_max_pacing_rate);
4742 sk->sk_priority = val;
4747 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
4750 if (sk->sk_mark != val) {
4755 case SO_BINDTODEVICE:
4756 optlen = min_t(long, optlen, IFNAMSIZ - 1);
4757 strncpy(devname, optval, optlen);
4758 devname[optlen] = 0;
4761 if (devname[0] != '\0') {
4762 struct net_device *dev;
4767 dev = dev_get_by_name(net, devname);
4770 ifindex = dev->ifindex;
4773 ret = sock_bindtoindex(sk, ifindex, false);
4776 if (sk->sk_prot->keepalive)
4777 sk->sk_prot->keepalive(sk, valbool);
4778 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
4784 } else if (level == SOL_IP) {
4785 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4788 val = *((int *)optval);
4789 /* Only some options are supported */
4792 if (val < -1 || val > 0xff) {
4795 struct inet_sock *inet = inet_sk(sk);
4805 #if IS_ENABLED(CONFIG_IPV6)
4806 } else if (level == SOL_IPV6) {
4807 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4810 val = *((int *)optval);
4811 /* Only some options are supported */
4814 if (val < -1 || val > 0xff) {
4817 struct ipv6_pinfo *np = inet6_sk(sk);
4828 } else if (level == SOL_TCP &&
4829 sk->sk_prot->setsockopt == tcp_setsockopt) {
4830 if (optname == TCP_CONGESTION) {
4831 char name[TCP_CA_NAME_MAX];
4833 strncpy(name, optval, min_t(long, optlen,
4834 TCP_CA_NAME_MAX-1));
4835 name[TCP_CA_NAME_MAX-1] = 0;
4836 ret = tcp_set_congestion_control(sk, name, false, true);
4838 struct inet_connection_sock *icsk = inet_csk(sk);
4839 struct tcp_sock *tp = tcp_sk(sk);
4840 unsigned long timeout;
4842 if (optlen != sizeof(int))
4845 val = *((int *)optval);
4846 /* Only some options are supported */
4849 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4854 case TCP_BPF_SNDCWND_CLAMP:
4858 tp->snd_cwnd_clamp = val;
4859 tp->snd_ssthresh = val;
4862 case TCP_BPF_DELACK_MAX:
4863 timeout = usecs_to_jiffies(val);
4864 if (timeout > TCP_DELACK_MAX ||
4865 timeout < TCP_TIMEOUT_MIN)
4867 inet_csk(sk)->icsk_delack_max = timeout;
4869 case TCP_BPF_RTO_MIN:
4870 timeout = usecs_to_jiffies(val);
4871 if (timeout > TCP_RTO_MIN ||
4872 timeout < TCP_TIMEOUT_MIN)
4874 inet_csk(sk)->icsk_rto_min = timeout;
4877 if (val < 0 || val > 1)
4883 ret = tcp_sock_set_keepidle_locked(sk, val);
4886 if (val < 1 || val > MAX_TCP_KEEPINTVL)
4889 tp->keepalive_intvl = val * HZ;
4892 if (val < 1 || val > MAX_TCP_KEEPCNT)
4895 tp->keepalive_probes = val;
4898 if (val < 1 || val > MAX_TCP_SYNCNT)
4901 icsk->icsk_syn_retries = val;
4903 case TCP_USER_TIMEOUT:
4907 icsk->icsk_user_timeout = val;
4909 case TCP_NOTSENT_LOWAT:
4910 tp->notsent_lowat = val;
4911 sk->sk_write_space(sk);
4913 case TCP_WINDOW_CLAMP:
4914 ret = tcp_set_window_clamp(sk, val);
4927 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
4928 char *optval, int optlen)
4930 if (!sk_fullsock(sk))
4933 sock_owned_by_me(sk);
4936 if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4937 struct inet_connection_sock *icsk;
4938 struct tcp_sock *tp;
4941 case TCP_CONGESTION:
4942 icsk = inet_csk(sk);
4944 if (!icsk->icsk_ca_ops || optlen <= 1)
4946 strncpy(optval, icsk->icsk_ca_ops->name, optlen);
4947 optval[optlen - 1] = 0;
4952 if (optlen <= 0 || !tp->saved_syn ||
4953 optlen > tcp_saved_syn_len(tp->saved_syn))
4955 memcpy(optval, tp->saved_syn->data, optlen);
4960 } else if (level == SOL_IP) {
4961 struct inet_sock *inet = inet_sk(sk);
4963 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4966 /* Only some options are supported */
4969 *((int *)optval) = (int)inet->tos;
4974 #if IS_ENABLED(CONFIG_IPV6)
4975 } else if (level == SOL_IPV6) {
4976 struct ipv6_pinfo *np = inet6_sk(sk);
4978 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4981 /* Only some options are supported */
4984 *((int *)optval) = (int)np->tclass;
4996 memset(optval, 0, optlen);
5000 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5001 int, level, int, optname, char *, optval, int, optlen)
5003 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5006 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5007 .func = bpf_sock_addr_setsockopt,
5009 .ret_type = RET_INTEGER,
5010 .arg1_type = ARG_PTR_TO_CTX,
5011 .arg2_type = ARG_ANYTHING,
5012 .arg3_type = ARG_ANYTHING,
5013 .arg4_type = ARG_PTR_TO_MEM,
5014 .arg5_type = ARG_CONST_SIZE,
5017 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5018 int, level, int, optname, char *, optval, int, optlen)
5020 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5023 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5024 .func = bpf_sock_addr_getsockopt,
5026 .ret_type = RET_INTEGER,
5027 .arg1_type = ARG_PTR_TO_CTX,
5028 .arg2_type = ARG_ANYTHING,
5029 .arg3_type = ARG_ANYTHING,
5030 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5031 .arg5_type = ARG_CONST_SIZE,
5034 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5035 int, level, int, optname, char *, optval, int, optlen)
5037 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5040 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5041 .func = bpf_sock_ops_setsockopt,
5043 .ret_type = RET_INTEGER,
5044 .arg1_type = ARG_PTR_TO_CTX,
5045 .arg2_type = ARG_ANYTHING,
5046 .arg3_type = ARG_ANYTHING,
5047 .arg4_type = ARG_PTR_TO_MEM,
5048 .arg5_type = ARG_CONST_SIZE,
5051 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5052 int optname, const u8 **start)
5054 struct sk_buff *syn_skb = bpf_sock->syn_skb;
5055 const u8 *hdr_start;
5059 /* sk is a request_sock here */
5061 if (optname == TCP_BPF_SYN) {
5062 hdr_start = syn_skb->data;
5063 ret = tcp_hdrlen(syn_skb);
5064 } else if (optname == TCP_BPF_SYN_IP) {
5065 hdr_start = skb_network_header(syn_skb);
5066 ret = skb_network_header_len(syn_skb) +
5067 tcp_hdrlen(syn_skb);
5069 /* optname == TCP_BPF_SYN_MAC */
5070 hdr_start = skb_mac_header(syn_skb);
5071 ret = skb_mac_header_len(syn_skb) +
5072 skb_network_header_len(syn_skb) +
5073 tcp_hdrlen(syn_skb);
5076 struct sock *sk = bpf_sock->sk;
5077 struct saved_syn *saved_syn;
5079 if (sk->sk_state == TCP_NEW_SYN_RECV)
5080 /* synack retransmit. bpf_sock->syn_skb will
5081 * not be available. It has to resort to
5082 * saved_syn (if it is saved).
5084 saved_syn = inet_reqsk(sk)->saved_syn;
5086 saved_syn = tcp_sk(sk)->saved_syn;
5091 if (optname == TCP_BPF_SYN) {
5092 hdr_start = saved_syn->data +
5093 saved_syn->mac_hdrlen +
5094 saved_syn->network_hdrlen;
5095 ret = saved_syn->tcp_hdrlen;
5096 } else if (optname == TCP_BPF_SYN_IP) {
5097 hdr_start = saved_syn->data +
5098 saved_syn->mac_hdrlen;
5099 ret = saved_syn->network_hdrlen +
5100 saved_syn->tcp_hdrlen;
5102 /* optname == TCP_BPF_SYN_MAC */
5104 /* TCP_SAVE_SYN may not have saved the mac hdr */
5105 if (!saved_syn->mac_hdrlen)
5108 hdr_start = saved_syn->data;
5109 ret = saved_syn->mac_hdrlen +
5110 saved_syn->network_hdrlen +
5111 saved_syn->tcp_hdrlen;
5119 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5120 int, level, int, optname, char *, optval, int, optlen)
5122 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5123 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5124 int ret, copy_len = 0;
5127 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5130 if (optlen < copy_len) {
5135 memcpy(optval, start, copy_len);
5138 /* Zero out unused buffer at the end */
5139 memset(optval + copy_len, 0, optlen - copy_len);
5144 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5147 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5148 .func = bpf_sock_ops_getsockopt,
5150 .ret_type = RET_INTEGER,
5151 .arg1_type = ARG_PTR_TO_CTX,
5152 .arg2_type = ARG_ANYTHING,
5153 .arg3_type = ARG_ANYTHING,
5154 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5155 .arg5_type = ARG_CONST_SIZE,
5158 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5161 struct sock *sk = bpf_sock->sk;
5162 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5164 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5167 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5169 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5172 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5173 .func = bpf_sock_ops_cb_flags_set,
5175 .ret_type = RET_INTEGER,
5176 .arg1_type = ARG_PTR_TO_CTX,
5177 .arg2_type = ARG_ANYTHING,
5180 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5181 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5183 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5187 struct sock *sk = ctx->sk;
5188 u32 flags = BIND_FROM_BPF;
5192 if (addr_len < offsetofend(struct sockaddr, sa_family))
5194 if (addr->sa_family == AF_INET) {
5195 if (addr_len < sizeof(struct sockaddr_in))
5197 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5198 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5199 return __inet_bind(sk, addr, addr_len, flags);
5200 #if IS_ENABLED(CONFIG_IPV6)
5201 } else if (addr->sa_family == AF_INET6) {
5202 if (addr_len < SIN6_LEN_RFC2133)
5204 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5205 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5206 /* ipv6_bpf_stub cannot be NULL, since it's called from
5207 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5209 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5210 #endif /* CONFIG_IPV6 */
5212 #endif /* CONFIG_INET */
5214 return -EAFNOSUPPORT;
5217 static const struct bpf_func_proto bpf_bind_proto = {
5220 .ret_type = RET_INTEGER,
5221 .arg1_type = ARG_PTR_TO_CTX,
5222 .arg2_type = ARG_PTR_TO_MEM,
5223 .arg3_type = ARG_CONST_SIZE,
5227 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5228 struct bpf_xfrm_state *, to, u32, size, u64, flags)
5230 const struct sec_path *sp = skb_sec_path(skb);
5231 const struct xfrm_state *x;
5233 if (!sp || unlikely(index >= sp->len || flags))
5236 x = sp->xvec[index];
5238 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5241 to->reqid = x->props.reqid;
5242 to->spi = x->id.spi;
5243 to->family = x->props.family;
5246 if (to->family == AF_INET6) {
5247 memcpy(to->remote_ipv6, x->props.saddr.a6,
5248 sizeof(to->remote_ipv6));
5250 to->remote_ipv4 = x->props.saddr.a4;
5251 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5256 memset(to, 0, size);
5260 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5261 .func = bpf_skb_get_xfrm_state,
5263 .ret_type = RET_INTEGER,
5264 .arg1_type = ARG_PTR_TO_CTX,
5265 .arg2_type = ARG_ANYTHING,
5266 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
5267 .arg4_type = ARG_CONST_SIZE,
5268 .arg5_type = ARG_ANYTHING,
5272 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5273 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
5274 const struct neighbour *neigh,
5275 const struct net_device *dev)
5277 memcpy(params->dmac, neigh->ha, ETH_ALEN);
5278 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5279 params->h_vlan_TCI = 0;
5280 params->h_vlan_proto = 0;
5286 #if IS_ENABLED(CONFIG_INET)
5287 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5288 u32 flags, bool check_mtu)
5290 struct fib_nh_common *nhc;
5291 struct in_device *in_dev;
5292 struct neighbour *neigh;
5293 struct net_device *dev;
5294 struct fib_result res;
5299 dev = dev_get_by_index_rcu(net, params->ifindex);
5303 /* verify forwarding is enabled on this interface */
5304 in_dev = __in_dev_get_rcu(dev);
5305 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5306 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5308 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5310 fl4.flowi4_oif = params->ifindex;
5312 fl4.flowi4_iif = params->ifindex;
5315 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5316 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5317 fl4.flowi4_flags = 0;
5319 fl4.flowi4_proto = params->l4_protocol;
5320 fl4.daddr = params->ipv4_dst;
5321 fl4.saddr = params->ipv4_src;
5322 fl4.fl4_sport = params->sport;
5323 fl4.fl4_dport = params->dport;
5324 fl4.flowi4_multipath_hash = 0;
5326 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5327 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5328 struct fib_table *tb;
5330 tb = fib_get_table(net, tbid);
5332 return BPF_FIB_LKUP_RET_NOT_FWDED;
5334 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5336 fl4.flowi4_mark = 0;
5337 fl4.flowi4_secid = 0;
5338 fl4.flowi4_tun_key.tun_id = 0;
5339 fl4.flowi4_uid = sock_net_uid(net, NULL);
5341 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5345 /* map fib lookup errors to RTN_ type */
5347 return BPF_FIB_LKUP_RET_BLACKHOLE;
5348 if (err == -EHOSTUNREACH)
5349 return BPF_FIB_LKUP_RET_UNREACHABLE;
5351 return BPF_FIB_LKUP_RET_PROHIBIT;
5353 return BPF_FIB_LKUP_RET_NOT_FWDED;
5356 if (res.type != RTN_UNICAST)
5357 return BPF_FIB_LKUP_RET_NOT_FWDED;
5359 if (fib_info_num_path(res.fi) > 1)
5360 fib_select_path(net, &res, &fl4, NULL);
5363 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5364 if (params->tot_len > mtu)
5365 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5370 /* do not handle lwt encaps right now */
5371 if (nhc->nhc_lwtstate)
5372 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5376 params->rt_metric = res.fi->fib_priority;
5377 params->ifindex = dev->ifindex;
5379 /* xdp and cls_bpf programs are run in RCU-bh so
5380 * rcu_read_lock_bh is not needed here
5382 if (likely(nhc->nhc_gw_family != AF_INET6)) {
5383 if (nhc->nhc_gw_family)
5384 params->ipv4_dst = nhc->nhc_gw.ipv4;
5386 neigh = __ipv4_neigh_lookup_noref(dev,
5387 (__force u32)params->ipv4_dst);
5389 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5391 params->family = AF_INET6;
5392 *dst = nhc->nhc_gw.ipv6;
5393 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5397 return BPF_FIB_LKUP_RET_NO_NEIGH;
5399 return bpf_fib_set_fwd_params(params, neigh, dev);
5403 #if IS_ENABLED(CONFIG_IPV6)
5404 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5405 u32 flags, bool check_mtu)
5407 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5408 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5409 struct fib6_result res = {};
5410 struct neighbour *neigh;
5411 struct net_device *dev;
5412 struct inet6_dev *idev;
5418 /* link local addresses are never forwarded */
5419 if (rt6_need_strict(dst) || rt6_need_strict(src))
5420 return BPF_FIB_LKUP_RET_NOT_FWDED;
5422 dev = dev_get_by_index_rcu(net, params->ifindex);
5426 idev = __in6_dev_get_safely(dev);
5427 if (unlikely(!idev || !idev->cnf.forwarding))
5428 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5430 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5432 oif = fl6.flowi6_oif = params->ifindex;
5434 oif = fl6.flowi6_iif = params->ifindex;
5436 strict = RT6_LOOKUP_F_HAS_SADDR;
5438 fl6.flowlabel = params->flowinfo;
5439 fl6.flowi6_scope = 0;
5440 fl6.flowi6_flags = 0;
5443 fl6.flowi6_proto = params->l4_protocol;
5446 fl6.fl6_sport = params->sport;
5447 fl6.fl6_dport = params->dport;
5449 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5450 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5451 struct fib6_table *tb;
5453 tb = ipv6_stub->fib6_get_table(net, tbid);
5455 return BPF_FIB_LKUP_RET_NOT_FWDED;
5457 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5460 fl6.flowi6_mark = 0;
5461 fl6.flowi6_secid = 0;
5462 fl6.flowi6_tun_key.tun_id = 0;
5463 fl6.flowi6_uid = sock_net_uid(net, NULL);
5465 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5468 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5469 res.f6i == net->ipv6.fib6_null_entry))
5470 return BPF_FIB_LKUP_RET_NOT_FWDED;
5472 switch (res.fib6_type) {
5473 /* only unicast is forwarded */
5477 return BPF_FIB_LKUP_RET_BLACKHOLE;
5478 case RTN_UNREACHABLE:
5479 return BPF_FIB_LKUP_RET_UNREACHABLE;
5481 return BPF_FIB_LKUP_RET_PROHIBIT;
5483 return BPF_FIB_LKUP_RET_NOT_FWDED;
5486 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5487 fl6.flowi6_oif != 0, NULL, strict);
5490 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5491 if (params->tot_len > mtu)
5492 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5495 if (res.nh->fib_nh_lws)
5496 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5498 if (res.nh->fib_nh_gw_family)
5499 *dst = res.nh->fib_nh_gw6;
5501 dev = res.nh->fib_nh_dev;
5502 params->rt_metric = res.f6i->fib6_metric;
5503 params->ifindex = dev->ifindex;
5505 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
5508 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5510 return BPF_FIB_LKUP_RET_NO_NEIGH;
5512 return bpf_fib_set_fwd_params(params, neigh, dev);
5516 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
5517 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5519 if (plen < sizeof(*params))
5522 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5525 switch (params->family) {
5526 #if IS_ENABLED(CONFIG_INET)
5528 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
5531 #if IS_ENABLED(CONFIG_IPV6)
5533 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
5537 return -EAFNOSUPPORT;
5540 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
5541 .func = bpf_xdp_fib_lookup,
5543 .ret_type = RET_INTEGER,
5544 .arg1_type = ARG_PTR_TO_CTX,
5545 .arg2_type = ARG_PTR_TO_MEM,
5546 .arg3_type = ARG_CONST_SIZE,
5547 .arg4_type = ARG_ANYTHING,
5550 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
5551 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5553 struct net *net = dev_net(skb->dev);
5554 int rc = -EAFNOSUPPORT;
5556 if (plen < sizeof(*params))
5559 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5562 switch (params->family) {
5563 #if IS_ENABLED(CONFIG_INET)
5565 rc = bpf_ipv4_fib_lookup(net, params, flags, false);
5568 #if IS_ENABLED(CONFIG_IPV6)
5570 rc = bpf_ipv6_fib_lookup(net, params, flags, false);
5576 struct net_device *dev;
5578 dev = dev_get_by_index_rcu(net, params->ifindex);
5579 if (!is_skb_forwardable(dev, skb))
5580 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
5586 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
5587 .func = bpf_skb_fib_lookup,
5589 .ret_type = RET_INTEGER,
5590 .arg1_type = ARG_PTR_TO_CTX,
5591 .arg2_type = ARG_PTR_TO_MEM,
5592 .arg3_type = ARG_CONST_SIZE,
5593 .arg4_type = ARG_ANYTHING,
5596 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5597 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
5600 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
5602 if (!seg6_validate_srh(srh, len, false))
5606 case BPF_LWT_ENCAP_SEG6_INLINE:
5607 if (skb->protocol != htons(ETH_P_IPV6))
5610 err = seg6_do_srh_inline(skb, srh);
5612 case BPF_LWT_ENCAP_SEG6:
5613 skb_reset_inner_headers(skb);
5614 skb->encapsulation = 1;
5615 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
5621 bpf_compute_data_pointers(skb);
5625 ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5626 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
5628 return seg6_lookup_nexthop(skb, NULL, 0);
5630 #endif /* CONFIG_IPV6_SEG6_BPF */
5632 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5633 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
5636 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
5640 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
5644 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5645 case BPF_LWT_ENCAP_SEG6:
5646 case BPF_LWT_ENCAP_SEG6_INLINE:
5647 return bpf_push_seg6_encap(skb, type, hdr, len);
5649 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5650 case BPF_LWT_ENCAP_IP:
5651 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
5658 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
5659 void *, hdr, u32, len)
5662 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5663 case BPF_LWT_ENCAP_IP:
5664 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
5671 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
5672 .func = bpf_lwt_in_push_encap,
5674 .ret_type = RET_INTEGER,
5675 .arg1_type = ARG_PTR_TO_CTX,
5676 .arg2_type = ARG_ANYTHING,
5677 .arg3_type = ARG_PTR_TO_MEM,
5678 .arg4_type = ARG_CONST_SIZE
5681 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
5682 .func = bpf_lwt_xmit_push_encap,
5684 .ret_type = RET_INTEGER,
5685 .arg1_type = ARG_PTR_TO_CTX,
5686 .arg2_type = ARG_ANYTHING,
5687 .arg3_type = ARG_PTR_TO_MEM,
5688 .arg4_type = ARG_CONST_SIZE
5691 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5692 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
5693 const void *, from, u32, len)
5695 struct seg6_bpf_srh_state *srh_state =
5696 this_cpu_ptr(&seg6_bpf_srh_states);
5697 struct ipv6_sr_hdr *srh = srh_state->srh;
5698 void *srh_tlvs, *srh_end, *ptr;
5704 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
5705 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
5707 ptr = skb->data + offset;
5708 if (ptr >= srh_tlvs && ptr + len <= srh_end)
5709 srh_state->valid = false;
5710 else if (ptr < (void *)&srh->flags ||
5711 ptr + len > (void *)&srh->segments)
5714 if (unlikely(bpf_try_make_writable(skb, offset + len)))
5716 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5718 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5720 memcpy(skb->data + offset, from, len);
5724 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
5725 .func = bpf_lwt_seg6_store_bytes,
5727 .ret_type = RET_INTEGER,
5728 .arg1_type = ARG_PTR_TO_CTX,
5729 .arg2_type = ARG_ANYTHING,
5730 .arg3_type = ARG_PTR_TO_MEM,
5731 .arg4_type = ARG_CONST_SIZE
5734 static void bpf_update_srh_state(struct sk_buff *skb)
5736 struct seg6_bpf_srh_state *srh_state =
5737 this_cpu_ptr(&seg6_bpf_srh_states);
5740 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
5741 srh_state->srh = NULL;
5743 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5744 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
5745 srh_state->valid = true;
5749 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
5750 u32, action, void *, param, u32, param_len)
5752 struct seg6_bpf_srh_state *srh_state =
5753 this_cpu_ptr(&seg6_bpf_srh_states);
5758 case SEG6_LOCAL_ACTION_END_X:
5759 if (!seg6_bpf_has_valid_srh(skb))
5761 if (param_len != sizeof(struct in6_addr))
5763 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
5764 case SEG6_LOCAL_ACTION_END_T:
5765 if (!seg6_bpf_has_valid_srh(skb))
5767 if (param_len != sizeof(int))
5769 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5770 case SEG6_LOCAL_ACTION_END_DT6:
5771 if (!seg6_bpf_has_valid_srh(skb))
5773 if (param_len != sizeof(int))
5776 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
5778 if (!pskb_pull(skb, hdroff))
5781 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
5782 skb_reset_network_header(skb);
5783 skb_reset_transport_header(skb);
5784 skb->encapsulation = 0;
5786 bpf_compute_data_pointers(skb);
5787 bpf_update_srh_state(skb);
5788 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5789 case SEG6_LOCAL_ACTION_END_B6:
5790 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5792 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
5795 bpf_update_srh_state(skb);
5798 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
5799 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5801 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
5804 bpf_update_srh_state(skb);
5812 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
5813 .func = bpf_lwt_seg6_action,
5815 .ret_type = RET_INTEGER,
5816 .arg1_type = ARG_PTR_TO_CTX,
5817 .arg2_type = ARG_ANYTHING,
5818 .arg3_type = ARG_PTR_TO_MEM,
5819 .arg4_type = ARG_CONST_SIZE
5822 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
5825 struct seg6_bpf_srh_state *srh_state =
5826 this_cpu_ptr(&seg6_bpf_srh_states);
5827 struct ipv6_sr_hdr *srh = srh_state->srh;
5828 void *srh_end, *srh_tlvs, *ptr;
5829 struct ipv6hdr *hdr;
5833 if (unlikely(srh == NULL))
5836 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
5837 ((srh->first_segment + 1) << 4));
5838 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
5840 ptr = skb->data + offset;
5842 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
5844 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
5848 ret = skb_cow_head(skb, len);
5849 if (unlikely(ret < 0))
5852 ret = bpf_skb_net_hdr_push(skb, offset, len);
5854 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
5857 bpf_compute_data_pointers(skb);
5858 if (unlikely(ret < 0))
5861 hdr = (struct ipv6hdr *)skb->data;
5862 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5864 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5866 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5867 srh_state->hdrlen += len;
5868 srh_state->valid = false;
5872 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
5873 .func = bpf_lwt_seg6_adjust_srh,
5875 .ret_type = RET_INTEGER,
5876 .arg1_type = ARG_PTR_TO_CTX,
5877 .arg2_type = ARG_ANYTHING,
5878 .arg3_type = ARG_ANYTHING,
5880 #endif /* CONFIG_IPV6_SEG6_BPF */
5883 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
5884 int dif, int sdif, u8 family, u8 proto)
5886 bool refcounted = false;
5887 struct sock *sk = NULL;
5889 if (family == AF_INET) {
5890 __be32 src4 = tuple->ipv4.saddr;
5891 __be32 dst4 = tuple->ipv4.daddr;
5893 if (proto == IPPROTO_TCP)
5894 sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
5895 src4, tuple->ipv4.sport,
5896 dst4, tuple->ipv4.dport,
5897 dif, sdif, &refcounted);
5899 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
5900 dst4, tuple->ipv4.dport,
5901 dif, sdif, &udp_table, NULL);
5902 #if IS_ENABLED(CONFIG_IPV6)
5904 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
5905 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
5907 if (proto == IPPROTO_TCP)
5908 sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
5909 src6, tuple->ipv6.sport,
5910 dst6, ntohs(tuple->ipv6.dport),
5911 dif, sdif, &refcounted);
5912 else if (likely(ipv6_bpf_stub))
5913 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
5914 src6, tuple->ipv6.sport,
5915 dst6, tuple->ipv6.dport,
5921 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
5922 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
5928 /* bpf_skc_lookup performs the core lookup for different types of sockets,
5929 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
5930 * Returns the socket as an 'unsigned long' to simplify the casting in the
5931 * callers to satisfy BPF_CALL declarations.
5933 static struct sock *
5934 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5935 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
5938 struct sock *sk = NULL;
5939 u8 family = AF_UNSPEC;
5943 if (len == sizeof(tuple->ipv4))
5945 else if (len == sizeof(tuple->ipv6))
5950 if (unlikely(family == AF_UNSPEC || flags ||
5951 !((s32)netns_id < 0 || netns_id <= S32_MAX)))
5954 if (family == AF_INET)
5955 sdif = inet_sdif(skb);
5957 sdif = inet6_sdif(skb);
5959 if ((s32)netns_id < 0) {
5961 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
5963 net = get_net_ns_by_id(caller_net, netns_id);
5966 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
5974 static struct sock *
5975 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5976 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
5979 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
5980 ifindex, proto, netns_id, flags);
5983 sk = sk_to_full_sk(sk);
5984 if (!sk_fullsock(sk)) {
5993 static struct sock *
5994 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5995 u8 proto, u64 netns_id, u64 flags)
5997 struct net *caller_net;
6001 caller_net = dev_net(skb->dev);
6002 ifindex = skb->dev->ifindex;
6004 caller_net = sock_net(skb->sk);
6008 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6012 static struct sock *
6013 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6014 u8 proto, u64 netns_id, u64 flags)
6016 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6020 sk = sk_to_full_sk(sk);
6021 if (!sk_fullsock(sk)) {
6030 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6031 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6033 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6037 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6038 .func = bpf_skc_lookup_tcp,
6041 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6042 .arg1_type = ARG_PTR_TO_CTX,
6043 .arg2_type = ARG_PTR_TO_MEM,
6044 .arg3_type = ARG_CONST_SIZE,
6045 .arg4_type = ARG_ANYTHING,
6046 .arg5_type = ARG_ANYTHING,
6049 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6050 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6052 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6056 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6057 .func = bpf_sk_lookup_tcp,
6060 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6061 .arg1_type = ARG_PTR_TO_CTX,
6062 .arg2_type = ARG_PTR_TO_MEM,
6063 .arg3_type = ARG_CONST_SIZE,
6064 .arg4_type = ARG_ANYTHING,
6065 .arg5_type = ARG_ANYTHING,
6068 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6069 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6071 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6075 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6076 .func = bpf_sk_lookup_udp,
6079 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6080 .arg1_type = ARG_PTR_TO_CTX,
6081 .arg2_type = ARG_PTR_TO_MEM,
6082 .arg3_type = ARG_CONST_SIZE,
6083 .arg4_type = ARG_ANYTHING,
6084 .arg5_type = ARG_ANYTHING,
6087 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6089 if (sk && sk_is_refcounted(sk))
6094 static const struct bpf_func_proto bpf_sk_release_proto = {
6095 .func = bpf_sk_release,
6097 .ret_type = RET_INTEGER,
6098 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6101 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6102 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6104 struct net *caller_net = dev_net(ctx->rxq->dev);
6105 int ifindex = ctx->rxq->dev->ifindex;
6107 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6108 ifindex, IPPROTO_UDP, netns_id,
6112 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6113 .func = bpf_xdp_sk_lookup_udp,
6116 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6117 .arg1_type = ARG_PTR_TO_CTX,
6118 .arg2_type = ARG_PTR_TO_MEM,
6119 .arg3_type = ARG_CONST_SIZE,
6120 .arg4_type = ARG_ANYTHING,
6121 .arg5_type = ARG_ANYTHING,
6124 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6125 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6127 struct net *caller_net = dev_net(ctx->rxq->dev);
6128 int ifindex = ctx->rxq->dev->ifindex;
6130 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6131 ifindex, IPPROTO_TCP, netns_id,
6135 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6136 .func = bpf_xdp_skc_lookup_tcp,
6139 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6140 .arg1_type = ARG_PTR_TO_CTX,
6141 .arg2_type = ARG_PTR_TO_MEM,
6142 .arg3_type = ARG_CONST_SIZE,
6143 .arg4_type = ARG_ANYTHING,
6144 .arg5_type = ARG_ANYTHING,
6147 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6148 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6150 struct net *caller_net = dev_net(ctx->rxq->dev);
6151 int ifindex = ctx->rxq->dev->ifindex;
6153 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6154 ifindex, IPPROTO_TCP, netns_id,
6158 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6159 .func = bpf_xdp_sk_lookup_tcp,
6162 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6163 .arg1_type = ARG_PTR_TO_CTX,
6164 .arg2_type = ARG_PTR_TO_MEM,
6165 .arg3_type = ARG_CONST_SIZE,
6166 .arg4_type = ARG_ANYTHING,
6167 .arg5_type = ARG_ANYTHING,
6170 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6171 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6173 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6174 sock_net(ctx->sk), 0,
6175 IPPROTO_TCP, netns_id, flags);
6178 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6179 .func = bpf_sock_addr_skc_lookup_tcp,
6181 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6182 .arg1_type = ARG_PTR_TO_CTX,
6183 .arg2_type = ARG_PTR_TO_MEM,
6184 .arg3_type = ARG_CONST_SIZE,
6185 .arg4_type = ARG_ANYTHING,
6186 .arg5_type = ARG_ANYTHING,
6189 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6190 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6192 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6193 sock_net(ctx->sk), 0, IPPROTO_TCP,
6197 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6198 .func = bpf_sock_addr_sk_lookup_tcp,
6200 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6201 .arg1_type = ARG_PTR_TO_CTX,
6202 .arg2_type = ARG_PTR_TO_MEM,
6203 .arg3_type = ARG_CONST_SIZE,
6204 .arg4_type = ARG_ANYTHING,
6205 .arg5_type = ARG_ANYTHING,
6208 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6209 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6211 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6212 sock_net(ctx->sk), 0, IPPROTO_UDP,
6216 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6217 .func = bpf_sock_addr_sk_lookup_udp,
6219 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6220 .arg1_type = ARG_PTR_TO_CTX,
6221 .arg2_type = ARG_PTR_TO_MEM,
6222 .arg3_type = ARG_CONST_SIZE,
6223 .arg4_type = ARG_ANYTHING,
6224 .arg5_type = ARG_ANYTHING,
6227 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6228 struct bpf_insn_access_aux *info)
6230 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6234 if (off % size != 0)
6238 case offsetof(struct bpf_tcp_sock, bytes_received):
6239 case offsetof(struct bpf_tcp_sock, bytes_acked):
6240 return size == sizeof(__u64);
6242 return size == sizeof(__u32);
6246 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6247 const struct bpf_insn *si,
6248 struct bpf_insn *insn_buf,
6249 struct bpf_prog *prog, u32 *target_size)
6251 struct bpf_insn *insn = insn_buf;
6253 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \
6255 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \
6256 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6257 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6258 si->dst_reg, si->src_reg, \
6259 offsetof(struct tcp_sock, FIELD)); \
6262 #define BPF_INET_SOCK_GET_COMMON(FIELD) \
6264 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \
6266 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6267 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6268 struct inet_connection_sock, \
6270 si->dst_reg, si->src_reg, \
6272 struct inet_connection_sock, \
6276 if (insn > insn_buf)
6277 return insn - insn_buf;
6280 case offsetof(struct bpf_tcp_sock, rtt_min):
6281 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
6282 sizeof(struct minmax));
6283 BUILD_BUG_ON(sizeof(struct minmax) <
6284 sizeof(struct minmax_sample));
6286 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6287 offsetof(struct tcp_sock, rtt_min) +
6288 offsetof(struct minmax_sample, v));
6290 case offsetof(struct bpf_tcp_sock, snd_cwnd):
6291 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
6293 case offsetof(struct bpf_tcp_sock, srtt_us):
6294 BPF_TCP_SOCK_GET_COMMON(srtt_us);
6296 case offsetof(struct bpf_tcp_sock, snd_ssthresh):
6297 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
6299 case offsetof(struct bpf_tcp_sock, rcv_nxt):
6300 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
6302 case offsetof(struct bpf_tcp_sock, snd_nxt):
6303 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
6305 case offsetof(struct bpf_tcp_sock, snd_una):
6306 BPF_TCP_SOCK_GET_COMMON(snd_una);
6308 case offsetof(struct bpf_tcp_sock, mss_cache):
6309 BPF_TCP_SOCK_GET_COMMON(mss_cache);
6311 case offsetof(struct bpf_tcp_sock, ecn_flags):
6312 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
6314 case offsetof(struct bpf_tcp_sock, rate_delivered):
6315 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
6317 case offsetof(struct bpf_tcp_sock, rate_interval_us):
6318 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
6320 case offsetof(struct bpf_tcp_sock, packets_out):
6321 BPF_TCP_SOCK_GET_COMMON(packets_out);
6323 case offsetof(struct bpf_tcp_sock, retrans_out):
6324 BPF_TCP_SOCK_GET_COMMON(retrans_out);
6326 case offsetof(struct bpf_tcp_sock, total_retrans):
6327 BPF_TCP_SOCK_GET_COMMON(total_retrans);
6329 case offsetof(struct bpf_tcp_sock, segs_in):
6330 BPF_TCP_SOCK_GET_COMMON(segs_in);
6332 case offsetof(struct bpf_tcp_sock, data_segs_in):
6333 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
6335 case offsetof(struct bpf_tcp_sock, segs_out):
6336 BPF_TCP_SOCK_GET_COMMON(segs_out);
6338 case offsetof(struct bpf_tcp_sock, data_segs_out):
6339 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
6341 case offsetof(struct bpf_tcp_sock, lost_out):
6342 BPF_TCP_SOCK_GET_COMMON(lost_out);
6344 case offsetof(struct bpf_tcp_sock, sacked_out):
6345 BPF_TCP_SOCK_GET_COMMON(sacked_out);
6347 case offsetof(struct bpf_tcp_sock, bytes_received):
6348 BPF_TCP_SOCK_GET_COMMON(bytes_received);
6350 case offsetof(struct bpf_tcp_sock, bytes_acked):
6351 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
6353 case offsetof(struct bpf_tcp_sock, dsack_dups):
6354 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
6356 case offsetof(struct bpf_tcp_sock, delivered):
6357 BPF_TCP_SOCK_GET_COMMON(delivered);
6359 case offsetof(struct bpf_tcp_sock, delivered_ce):
6360 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
6362 case offsetof(struct bpf_tcp_sock, icsk_retransmits):
6363 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
6367 return insn - insn_buf;
6370 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
6372 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
6373 return (unsigned long)sk;
6375 return (unsigned long)NULL;
6378 const struct bpf_func_proto bpf_tcp_sock_proto = {
6379 .func = bpf_tcp_sock,
6381 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
6382 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6385 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
6387 sk = sk_to_full_sk(sk);
6389 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
6390 return (unsigned long)sk;
6392 return (unsigned long)NULL;
6395 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
6396 .func = bpf_get_listener_sock,
6398 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6399 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6402 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
6404 unsigned int iphdr_len;
6406 switch (skb_protocol(skb, true)) {
6407 case cpu_to_be16(ETH_P_IP):
6408 iphdr_len = sizeof(struct iphdr);
6410 case cpu_to_be16(ETH_P_IPV6):
6411 iphdr_len = sizeof(struct ipv6hdr);
6417 if (skb_headlen(skb) < iphdr_len)
6420 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
6423 return INET_ECN_set_ce(skb);
6426 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6427 struct bpf_insn_access_aux *info)
6429 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
6432 if (off % size != 0)
6437 return size == sizeof(__u32);
6441 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
6442 const struct bpf_insn *si,
6443 struct bpf_insn *insn_buf,
6444 struct bpf_prog *prog, u32 *target_size)
6446 struct bpf_insn *insn = insn_buf;
6448 #define BPF_XDP_SOCK_GET(FIELD) \
6450 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \
6451 sizeof_field(struct bpf_xdp_sock, FIELD)); \
6452 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
6453 si->dst_reg, si->src_reg, \
6454 offsetof(struct xdp_sock, FIELD)); \
6458 case offsetof(struct bpf_xdp_sock, queue_id):
6459 BPF_XDP_SOCK_GET(queue_id);
6463 return insn - insn_buf;
6466 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
6467 .func = bpf_skb_ecn_set_ce,
6469 .ret_type = RET_INTEGER,
6470 .arg1_type = ARG_PTR_TO_CTX,
6473 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6474 struct tcphdr *, th, u32, th_len)
6476 #ifdef CONFIG_SYN_COOKIES
6480 if (unlikely(!sk || th_len < sizeof(*th)))
6483 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
6484 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6487 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
6490 if (!th->ack || th->rst || th->syn)
6493 if (tcp_synq_no_recent_overflow(sk))
6496 cookie = ntohl(th->ack_seq) - 1;
6498 switch (sk->sk_family) {
6500 if (unlikely(iph_len < sizeof(struct iphdr)))
6503 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
6506 #if IS_BUILTIN(CONFIG_IPV6)
6508 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6511 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
6513 #endif /* CONFIG_IPV6 */
6516 return -EPROTONOSUPPORT;
6528 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
6529 .func = bpf_tcp_check_syncookie,
6532 .ret_type = RET_INTEGER,
6533 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6534 .arg2_type = ARG_PTR_TO_MEM,
6535 .arg3_type = ARG_CONST_SIZE,
6536 .arg4_type = ARG_PTR_TO_MEM,
6537 .arg5_type = ARG_CONST_SIZE,
6540 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6541 struct tcphdr *, th, u32, th_len)
6543 #ifdef CONFIG_SYN_COOKIES
6547 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
6550 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6553 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
6556 if (!th->syn || th->ack || th->fin || th->rst)
6559 if (unlikely(iph_len < sizeof(struct iphdr)))
6562 /* Both struct iphdr and struct ipv6hdr have the version field at the
6563 * same offset so we can cast to the shorter header (struct iphdr).
6565 switch (((struct iphdr *)iph)->version) {
6567 if (sk->sk_family == AF_INET6 && sk->sk_ipv6only)
6570 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
6573 #if IS_BUILTIN(CONFIG_IPV6)
6575 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6578 if (sk->sk_family != AF_INET6)
6581 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
6583 #endif /* CONFIG_IPV6 */
6586 return -EPROTONOSUPPORT;
6591 return cookie | ((u64)mss << 32);
6594 #endif /* CONFIG_SYN_COOKIES */
6597 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
6598 .func = bpf_tcp_gen_syncookie,
6599 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */
6601 .ret_type = RET_INTEGER,
6602 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6603 .arg2_type = ARG_PTR_TO_MEM,
6604 .arg3_type = ARG_CONST_SIZE,
6605 .arg4_type = ARG_PTR_TO_MEM,
6606 .arg5_type = ARG_CONST_SIZE,
6609 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
6611 if (!sk || flags != 0)
6613 if (!skb_at_tc_ingress(skb))
6615 if (unlikely(dev_net(skb->dev) != sock_net(sk)))
6616 return -ENETUNREACH;
6617 if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
6618 return -ESOCKTNOSUPPORT;
6619 if (sk_is_refcounted(sk) &&
6620 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
6625 skb->destructor = sock_pfree;
6630 static const struct bpf_func_proto bpf_sk_assign_proto = {
6631 .func = bpf_sk_assign,
6633 .ret_type = RET_INTEGER,
6634 .arg1_type = ARG_PTR_TO_CTX,
6635 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6636 .arg3_type = ARG_ANYTHING,
6639 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
6640 u8 search_kind, const u8 *magic,
6641 u8 magic_len, bool *eol)
6647 while (op < opend) {
6650 if (kind == TCPOPT_EOL) {
6652 return ERR_PTR(-ENOMSG);
6653 } else if (kind == TCPOPT_NOP) {
6658 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
6659 /* Something is wrong in the received header.
6660 * Follow the TCP stack's tcp_parse_options()
6661 * and just bail here.
6663 return ERR_PTR(-EFAULT);
6666 if (search_kind == kind) {
6670 if (magic_len > kind_len - 2)
6671 return ERR_PTR(-ENOMSG);
6673 if (!memcmp(&op[2], magic, magic_len))
6680 return ERR_PTR(-ENOMSG);
6683 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6684 void *, search_res, u32, len, u64, flags)
6686 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
6687 const u8 *op, *opend, *magic, *search = search_res;
6688 u8 search_kind, search_len, copy_len, magic_len;
6691 /* 2 byte is the minimal option len except TCPOPT_NOP and
6692 * TCPOPT_EOL which are useless for the bpf prog to learn
6693 * and this helper disallow loading them also.
6695 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
6698 search_kind = search[0];
6699 search_len = search[1];
6701 if (search_len > len || search_kind == TCPOPT_NOP ||
6702 search_kind == TCPOPT_EOL)
6705 if (search_kind == TCPOPT_EXP || search_kind == 253) {
6706 /* 16 or 32 bit magic. +2 for kind and kind length */
6707 if (search_len != 4 && search_len != 6)
6710 magic_len = search_len - 2;
6719 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
6724 op += sizeof(struct tcphdr);
6726 if (!bpf_sock->skb ||
6727 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6728 /* This bpf_sock->op cannot call this helper */
6731 opend = bpf_sock->skb_data_end;
6732 op = bpf_sock->skb->data + sizeof(struct tcphdr);
6735 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
6742 if (copy_len > len) {
6747 memcpy(search_res, op, copy_len);
6751 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
6752 .func = bpf_sock_ops_load_hdr_opt,
6754 .ret_type = RET_INTEGER,
6755 .arg1_type = ARG_PTR_TO_CTX,
6756 .arg2_type = ARG_PTR_TO_MEM,
6757 .arg3_type = ARG_CONST_SIZE,
6758 .arg4_type = ARG_ANYTHING,
6761 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6762 const void *, from, u32, len, u64, flags)
6764 u8 new_kind, new_kind_len, magic_len = 0, *opend;
6765 const u8 *op, *new_op, *magic = NULL;
6766 struct sk_buff *skb;
6769 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
6772 if (len < 2 || flags)
6776 new_kind = new_op[0];
6777 new_kind_len = new_op[1];
6779 if (new_kind_len > len || new_kind == TCPOPT_NOP ||
6780 new_kind == TCPOPT_EOL)
6783 if (new_kind_len > bpf_sock->remaining_opt_len)
6786 /* 253 is another experimental kind */
6787 if (new_kind == TCPOPT_EXP || new_kind == 253) {
6788 if (new_kind_len < 4)
6790 /* Match for the 2 byte magic also.
6791 * RFC 6994: the magic could be 2 or 4 bytes.
6792 * Hence, matching by 2 byte only is on the
6793 * conservative side but it is the right
6794 * thing to do for the 'search-for-duplication'
6801 /* Check for duplication */
6802 skb = bpf_sock->skb;
6803 op = skb->data + sizeof(struct tcphdr);
6804 opend = bpf_sock->skb_data_end;
6806 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
6811 if (PTR_ERR(op) != -ENOMSG)
6815 /* The option has been ended. Treat it as no more
6816 * header option can be written.
6820 /* No duplication found. Store the header option. */
6821 memcpy(opend, from, new_kind_len);
6823 bpf_sock->remaining_opt_len -= new_kind_len;
6824 bpf_sock->skb_data_end += new_kind_len;
6829 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
6830 .func = bpf_sock_ops_store_hdr_opt,
6832 .ret_type = RET_INTEGER,
6833 .arg1_type = ARG_PTR_TO_CTX,
6834 .arg2_type = ARG_PTR_TO_MEM,
6835 .arg3_type = ARG_CONST_SIZE,
6836 .arg4_type = ARG_ANYTHING,
6839 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6840 u32, len, u64, flags)
6842 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6845 if (flags || len < 2)
6848 if (len > bpf_sock->remaining_opt_len)
6851 bpf_sock->remaining_opt_len -= len;
6856 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
6857 .func = bpf_sock_ops_reserve_hdr_opt,
6859 .ret_type = RET_INTEGER,
6860 .arg1_type = ARG_PTR_TO_CTX,
6861 .arg2_type = ARG_ANYTHING,
6862 .arg3_type = ARG_ANYTHING,
6865 #endif /* CONFIG_INET */
6867 bool bpf_helper_changes_pkt_data(void *func)
6869 if (func == bpf_skb_vlan_push ||
6870 func == bpf_skb_vlan_pop ||
6871 func == bpf_skb_store_bytes ||
6872 func == bpf_skb_change_proto ||
6873 func == bpf_skb_change_head ||
6874 func == sk_skb_change_head ||
6875 func == bpf_skb_change_tail ||
6876 func == sk_skb_change_tail ||
6877 func == bpf_skb_adjust_room ||
6878 func == sk_skb_adjust_room ||
6879 func == bpf_skb_pull_data ||
6880 func == sk_skb_pull_data ||
6881 func == bpf_clone_redirect ||
6882 func == bpf_l3_csum_replace ||
6883 func == bpf_l4_csum_replace ||
6884 func == bpf_xdp_adjust_head ||
6885 func == bpf_xdp_adjust_meta ||
6886 func == bpf_msg_pull_data ||
6887 func == bpf_msg_push_data ||
6888 func == bpf_msg_pop_data ||
6889 func == bpf_xdp_adjust_tail ||
6890 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6891 func == bpf_lwt_seg6_store_bytes ||
6892 func == bpf_lwt_seg6_adjust_srh ||
6893 func == bpf_lwt_seg6_action ||
6896 func == bpf_sock_ops_store_hdr_opt ||
6898 func == bpf_lwt_in_push_encap ||
6899 func == bpf_lwt_xmit_push_encap)
6905 const struct bpf_func_proto bpf_event_output_data_proto __weak;
6906 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
6908 static const struct bpf_func_proto *
6909 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6912 /* inet and inet6 sockets are created in a process
6913 * context so there is always a valid uid/gid
6915 case BPF_FUNC_get_current_uid_gid:
6916 return &bpf_get_current_uid_gid_proto;
6917 case BPF_FUNC_get_local_storage:
6918 return &bpf_get_local_storage_proto;
6919 case BPF_FUNC_get_socket_cookie:
6920 return &bpf_get_socket_cookie_sock_proto;
6921 case BPF_FUNC_get_netns_cookie:
6922 return &bpf_get_netns_cookie_sock_proto;
6923 case BPF_FUNC_perf_event_output:
6924 return &bpf_event_output_data_proto;
6925 case BPF_FUNC_get_current_pid_tgid:
6926 return &bpf_get_current_pid_tgid_proto;
6927 case BPF_FUNC_get_current_comm:
6928 return &bpf_get_current_comm_proto;
6929 #ifdef CONFIG_CGROUPS
6930 case BPF_FUNC_get_current_cgroup_id:
6931 return &bpf_get_current_cgroup_id_proto;
6932 case BPF_FUNC_get_current_ancestor_cgroup_id:
6933 return &bpf_get_current_ancestor_cgroup_id_proto;
6935 #ifdef CONFIG_CGROUP_NET_CLASSID
6936 case BPF_FUNC_get_cgroup_classid:
6937 return &bpf_get_cgroup_classid_curr_proto;
6939 case BPF_FUNC_sk_storage_get:
6940 return &bpf_sk_storage_get_cg_sock_proto;
6942 return bpf_base_func_proto(func_id);
6946 static const struct bpf_func_proto *
6947 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6950 /* inet and inet6 sockets are created in a process
6951 * context so there is always a valid uid/gid
6953 case BPF_FUNC_get_current_uid_gid:
6954 return &bpf_get_current_uid_gid_proto;
6956 switch (prog->expected_attach_type) {
6957 case BPF_CGROUP_INET4_CONNECT:
6958 case BPF_CGROUP_INET6_CONNECT:
6959 return &bpf_bind_proto;
6963 case BPF_FUNC_get_socket_cookie:
6964 return &bpf_get_socket_cookie_sock_addr_proto;
6965 case BPF_FUNC_get_netns_cookie:
6966 return &bpf_get_netns_cookie_sock_addr_proto;
6967 case BPF_FUNC_get_local_storage:
6968 return &bpf_get_local_storage_proto;
6969 case BPF_FUNC_perf_event_output:
6970 return &bpf_event_output_data_proto;
6971 case BPF_FUNC_get_current_pid_tgid:
6972 return &bpf_get_current_pid_tgid_proto;
6973 case BPF_FUNC_get_current_comm:
6974 return &bpf_get_current_comm_proto;
6975 #ifdef CONFIG_CGROUPS
6976 case BPF_FUNC_get_current_cgroup_id:
6977 return &bpf_get_current_cgroup_id_proto;
6978 case BPF_FUNC_get_current_ancestor_cgroup_id:
6979 return &bpf_get_current_ancestor_cgroup_id_proto;
6981 #ifdef CONFIG_CGROUP_NET_CLASSID
6982 case BPF_FUNC_get_cgroup_classid:
6983 return &bpf_get_cgroup_classid_curr_proto;
6986 case BPF_FUNC_sk_lookup_tcp:
6987 return &bpf_sock_addr_sk_lookup_tcp_proto;
6988 case BPF_FUNC_sk_lookup_udp:
6989 return &bpf_sock_addr_sk_lookup_udp_proto;
6990 case BPF_FUNC_sk_release:
6991 return &bpf_sk_release_proto;
6992 case BPF_FUNC_skc_lookup_tcp:
6993 return &bpf_sock_addr_skc_lookup_tcp_proto;
6994 #endif /* CONFIG_INET */
6995 case BPF_FUNC_sk_storage_get:
6996 return &bpf_sk_storage_get_proto;
6997 case BPF_FUNC_sk_storage_delete:
6998 return &bpf_sk_storage_delete_proto;
6999 case BPF_FUNC_setsockopt:
7000 switch (prog->expected_attach_type) {
7001 case BPF_CGROUP_INET4_BIND:
7002 case BPF_CGROUP_INET6_BIND:
7003 case BPF_CGROUP_INET4_CONNECT:
7004 case BPF_CGROUP_INET6_CONNECT:
7005 return &bpf_sock_addr_setsockopt_proto;
7009 case BPF_FUNC_getsockopt:
7010 switch (prog->expected_attach_type) {
7011 case BPF_CGROUP_INET4_BIND:
7012 case BPF_CGROUP_INET6_BIND:
7013 case BPF_CGROUP_INET4_CONNECT:
7014 case BPF_CGROUP_INET6_CONNECT:
7015 return &bpf_sock_addr_getsockopt_proto;
7020 return bpf_sk_base_func_proto(func_id);
7024 static const struct bpf_func_proto *
7025 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7028 case BPF_FUNC_skb_load_bytes:
7029 return &bpf_skb_load_bytes_proto;
7030 case BPF_FUNC_skb_load_bytes_relative:
7031 return &bpf_skb_load_bytes_relative_proto;
7032 case BPF_FUNC_get_socket_cookie:
7033 return &bpf_get_socket_cookie_proto;
7034 case BPF_FUNC_get_socket_uid:
7035 return &bpf_get_socket_uid_proto;
7036 case BPF_FUNC_perf_event_output:
7037 return &bpf_skb_event_output_proto;
7039 return bpf_sk_base_func_proto(func_id);
7043 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7044 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7046 static const struct bpf_func_proto *
7047 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7050 case BPF_FUNC_get_local_storage:
7051 return &bpf_get_local_storage_proto;
7052 case BPF_FUNC_sk_fullsock:
7053 return &bpf_sk_fullsock_proto;
7054 case BPF_FUNC_sk_storage_get:
7055 return &bpf_sk_storage_get_proto;
7056 case BPF_FUNC_sk_storage_delete:
7057 return &bpf_sk_storage_delete_proto;
7058 case BPF_FUNC_perf_event_output:
7059 return &bpf_skb_event_output_proto;
7060 #ifdef CONFIG_SOCK_CGROUP_DATA
7061 case BPF_FUNC_skb_cgroup_id:
7062 return &bpf_skb_cgroup_id_proto;
7063 case BPF_FUNC_skb_ancestor_cgroup_id:
7064 return &bpf_skb_ancestor_cgroup_id_proto;
7065 case BPF_FUNC_sk_cgroup_id:
7066 return &bpf_sk_cgroup_id_proto;
7067 case BPF_FUNC_sk_ancestor_cgroup_id:
7068 return &bpf_sk_ancestor_cgroup_id_proto;
7071 case BPF_FUNC_sk_lookup_tcp:
7072 return &bpf_sk_lookup_tcp_proto;
7073 case BPF_FUNC_sk_lookup_udp:
7074 return &bpf_sk_lookup_udp_proto;
7075 case BPF_FUNC_sk_release:
7076 return &bpf_sk_release_proto;
7077 case BPF_FUNC_skc_lookup_tcp:
7078 return &bpf_skc_lookup_tcp_proto;
7079 case BPF_FUNC_tcp_sock:
7080 return &bpf_tcp_sock_proto;
7081 case BPF_FUNC_get_listener_sock:
7082 return &bpf_get_listener_sock_proto;
7083 case BPF_FUNC_skb_ecn_set_ce:
7084 return &bpf_skb_ecn_set_ce_proto;
7087 return sk_filter_func_proto(func_id, prog);
7091 static const struct bpf_func_proto *
7092 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7095 case BPF_FUNC_skb_store_bytes:
7096 return &bpf_skb_store_bytes_proto;
7097 case BPF_FUNC_skb_load_bytes:
7098 return &bpf_skb_load_bytes_proto;
7099 case BPF_FUNC_skb_load_bytes_relative:
7100 return &bpf_skb_load_bytes_relative_proto;
7101 case BPF_FUNC_skb_pull_data:
7102 return &bpf_skb_pull_data_proto;
7103 case BPF_FUNC_csum_diff:
7104 return &bpf_csum_diff_proto;
7105 case BPF_FUNC_csum_update:
7106 return &bpf_csum_update_proto;
7107 case BPF_FUNC_csum_level:
7108 return &bpf_csum_level_proto;
7109 case BPF_FUNC_l3_csum_replace:
7110 return &bpf_l3_csum_replace_proto;
7111 case BPF_FUNC_l4_csum_replace:
7112 return &bpf_l4_csum_replace_proto;
7113 case BPF_FUNC_clone_redirect:
7114 return &bpf_clone_redirect_proto;
7115 case BPF_FUNC_get_cgroup_classid:
7116 return &bpf_get_cgroup_classid_proto;
7117 case BPF_FUNC_skb_vlan_push:
7118 return &bpf_skb_vlan_push_proto;
7119 case BPF_FUNC_skb_vlan_pop:
7120 return &bpf_skb_vlan_pop_proto;
7121 case BPF_FUNC_skb_change_proto:
7122 return &bpf_skb_change_proto_proto;
7123 case BPF_FUNC_skb_change_type:
7124 return &bpf_skb_change_type_proto;
7125 case BPF_FUNC_skb_adjust_room:
7126 return &bpf_skb_adjust_room_proto;
7127 case BPF_FUNC_skb_change_tail:
7128 return &bpf_skb_change_tail_proto;
7129 case BPF_FUNC_skb_change_head:
7130 return &bpf_skb_change_head_proto;
7131 case BPF_FUNC_skb_get_tunnel_key:
7132 return &bpf_skb_get_tunnel_key_proto;
7133 case BPF_FUNC_skb_set_tunnel_key:
7134 return bpf_get_skb_set_tunnel_proto(func_id);
7135 case BPF_FUNC_skb_get_tunnel_opt:
7136 return &bpf_skb_get_tunnel_opt_proto;
7137 case BPF_FUNC_skb_set_tunnel_opt:
7138 return bpf_get_skb_set_tunnel_proto(func_id);
7139 case BPF_FUNC_redirect:
7140 return &bpf_redirect_proto;
7141 case BPF_FUNC_redirect_neigh:
7142 return &bpf_redirect_neigh_proto;
7143 case BPF_FUNC_redirect_peer:
7144 return &bpf_redirect_peer_proto;
7145 case BPF_FUNC_get_route_realm:
7146 return &bpf_get_route_realm_proto;
7147 case BPF_FUNC_get_hash_recalc:
7148 return &bpf_get_hash_recalc_proto;
7149 case BPF_FUNC_set_hash_invalid:
7150 return &bpf_set_hash_invalid_proto;
7151 case BPF_FUNC_set_hash:
7152 return &bpf_set_hash_proto;
7153 case BPF_FUNC_perf_event_output:
7154 return &bpf_skb_event_output_proto;
7155 case BPF_FUNC_get_smp_processor_id:
7156 return &bpf_get_smp_processor_id_proto;
7157 case BPF_FUNC_skb_under_cgroup:
7158 return &bpf_skb_under_cgroup_proto;
7159 case BPF_FUNC_get_socket_cookie:
7160 return &bpf_get_socket_cookie_proto;
7161 case BPF_FUNC_get_socket_uid:
7162 return &bpf_get_socket_uid_proto;
7163 case BPF_FUNC_fib_lookup:
7164 return &bpf_skb_fib_lookup_proto;
7165 case BPF_FUNC_sk_fullsock:
7166 return &bpf_sk_fullsock_proto;
7167 case BPF_FUNC_sk_storage_get:
7168 return &bpf_sk_storage_get_proto;
7169 case BPF_FUNC_sk_storage_delete:
7170 return &bpf_sk_storage_delete_proto;
7172 case BPF_FUNC_skb_get_xfrm_state:
7173 return &bpf_skb_get_xfrm_state_proto;
7175 #ifdef CONFIG_CGROUP_NET_CLASSID
7176 case BPF_FUNC_skb_cgroup_classid:
7177 return &bpf_skb_cgroup_classid_proto;
7179 #ifdef CONFIG_SOCK_CGROUP_DATA
7180 case BPF_FUNC_skb_cgroup_id:
7181 return &bpf_skb_cgroup_id_proto;
7182 case BPF_FUNC_skb_ancestor_cgroup_id:
7183 return &bpf_skb_ancestor_cgroup_id_proto;
7186 case BPF_FUNC_sk_lookup_tcp:
7187 return &bpf_sk_lookup_tcp_proto;
7188 case BPF_FUNC_sk_lookup_udp:
7189 return &bpf_sk_lookup_udp_proto;
7190 case BPF_FUNC_sk_release:
7191 return &bpf_sk_release_proto;
7192 case BPF_FUNC_tcp_sock:
7193 return &bpf_tcp_sock_proto;
7194 case BPF_FUNC_get_listener_sock:
7195 return &bpf_get_listener_sock_proto;
7196 case BPF_FUNC_skc_lookup_tcp:
7197 return &bpf_skc_lookup_tcp_proto;
7198 case BPF_FUNC_tcp_check_syncookie:
7199 return &bpf_tcp_check_syncookie_proto;
7200 case BPF_FUNC_skb_ecn_set_ce:
7201 return &bpf_skb_ecn_set_ce_proto;
7202 case BPF_FUNC_tcp_gen_syncookie:
7203 return &bpf_tcp_gen_syncookie_proto;
7204 case BPF_FUNC_sk_assign:
7205 return &bpf_sk_assign_proto;
7208 return bpf_sk_base_func_proto(func_id);
7212 static const struct bpf_func_proto *
7213 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7216 case BPF_FUNC_perf_event_output:
7217 return &bpf_xdp_event_output_proto;
7218 case BPF_FUNC_get_smp_processor_id:
7219 return &bpf_get_smp_processor_id_proto;
7220 case BPF_FUNC_csum_diff:
7221 return &bpf_csum_diff_proto;
7222 case BPF_FUNC_xdp_adjust_head:
7223 return &bpf_xdp_adjust_head_proto;
7224 case BPF_FUNC_xdp_adjust_meta:
7225 return &bpf_xdp_adjust_meta_proto;
7226 case BPF_FUNC_redirect:
7227 return &bpf_xdp_redirect_proto;
7228 case BPF_FUNC_redirect_map:
7229 return &bpf_xdp_redirect_map_proto;
7230 case BPF_FUNC_xdp_adjust_tail:
7231 return &bpf_xdp_adjust_tail_proto;
7232 case BPF_FUNC_fib_lookup:
7233 return &bpf_xdp_fib_lookup_proto;
7235 case BPF_FUNC_sk_lookup_udp:
7236 return &bpf_xdp_sk_lookup_udp_proto;
7237 case BPF_FUNC_sk_lookup_tcp:
7238 return &bpf_xdp_sk_lookup_tcp_proto;
7239 case BPF_FUNC_sk_release:
7240 return &bpf_sk_release_proto;
7241 case BPF_FUNC_skc_lookup_tcp:
7242 return &bpf_xdp_skc_lookup_tcp_proto;
7243 case BPF_FUNC_tcp_check_syncookie:
7244 return &bpf_tcp_check_syncookie_proto;
7245 case BPF_FUNC_tcp_gen_syncookie:
7246 return &bpf_tcp_gen_syncookie_proto;
7249 return bpf_sk_base_func_proto(func_id);
7253 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
7254 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
7256 static const struct bpf_func_proto *
7257 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7260 case BPF_FUNC_setsockopt:
7261 return &bpf_sock_ops_setsockopt_proto;
7262 case BPF_FUNC_getsockopt:
7263 return &bpf_sock_ops_getsockopt_proto;
7264 case BPF_FUNC_sock_ops_cb_flags_set:
7265 return &bpf_sock_ops_cb_flags_set_proto;
7266 case BPF_FUNC_sock_map_update:
7267 return &bpf_sock_map_update_proto;
7268 case BPF_FUNC_sock_hash_update:
7269 return &bpf_sock_hash_update_proto;
7270 case BPF_FUNC_get_socket_cookie:
7271 return &bpf_get_socket_cookie_sock_ops_proto;
7272 case BPF_FUNC_get_local_storage:
7273 return &bpf_get_local_storage_proto;
7274 case BPF_FUNC_perf_event_output:
7275 return &bpf_event_output_data_proto;
7276 case BPF_FUNC_sk_storage_get:
7277 return &bpf_sk_storage_get_proto;
7278 case BPF_FUNC_sk_storage_delete:
7279 return &bpf_sk_storage_delete_proto;
7281 case BPF_FUNC_load_hdr_opt:
7282 return &bpf_sock_ops_load_hdr_opt_proto;
7283 case BPF_FUNC_store_hdr_opt:
7284 return &bpf_sock_ops_store_hdr_opt_proto;
7285 case BPF_FUNC_reserve_hdr_opt:
7286 return &bpf_sock_ops_reserve_hdr_opt_proto;
7287 case BPF_FUNC_tcp_sock:
7288 return &bpf_tcp_sock_proto;
7289 #endif /* CONFIG_INET */
7291 return bpf_sk_base_func_proto(func_id);
7295 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
7296 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
7298 static const struct bpf_func_proto *
7299 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7302 case BPF_FUNC_msg_redirect_map:
7303 return &bpf_msg_redirect_map_proto;
7304 case BPF_FUNC_msg_redirect_hash:
7305 return &bpf_msg_redirect_hash_proto;
7306 case BPF_FUNC_msg_apply_bytes:
7307 return &bpf_msg_apply_bytes_proto;
7308 case BPF_FUNC_msg_cork_bytes:
7309 return &bpf_msg_cork_bytes_proto;
7310 case BPF_FUNC_msg_pull_data:
7311 return &bpf_msg_pull_data_proto;
7312 case BPF_FUNC_msg_push_data:
7313 return &bpf_msg_push_data_proto;
7314 case BPF_FUNC_msg_pop_data:
7315 return &bpf_msg_pop_data_proto;
7316 case BPF_FUNC_perf_event_output:
7317 return &bpf_event_output_data_proto;
7318 case BPF_FUNC_get_current_uid_gid:
7319 return &bpf_get_current_uid_gid_proto;
7320 case BPF_FUNC_get_current_pid_tgid:
7321 return &bpf_get_current_pid_tgid_proto;
7322 case BPF_FUNC_sk_storage_get:
7323 return &bpf_sk_storage_get_proto;
7324 case BPF_FUNC_sk_storage_delete:
7325 return &bpf_sk_storage_delete_proto;
7326 #ifdef CONFIG_CGROUPS
7327 case BPF_FUNC_get_current_cgroup_id:
7328 return &bpf_get_current_cgroup_id_proto;
7329 case BPF_FUNC_get_current_ancestor_cgroup_id:
7330 return &bpf_get_current_ancestor_cgroup_id_proto;
7332 #ifdef CONFIG_CGROUP_NET_CLASSID
7333 case BPF_FUNC_get_cgroup_classid:
7334 return &bpf_get_cgroup_classid_curr_proto;
7337 return bpf_sk_base_func_proto(func_id);
7341 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
7342 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
7344 static const struct bpf_func_proto *
7345 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7348 case BPF_FUNC_skb_store_bytes:
7349 return &bpf_skb_store_bytes_proto;
7350 case BPF_FUNC_skb_load_bytes:
7351 return &bpf_skb_load_bytes_proto;
7352 case BPF_FUNC_skb_pull_data:
7353 return &sk_skb_pull_data_proto;
7354 case BPF_FUNC_skb_change_tail:
7355 return &sk_skb_change_tail_proto;
7356 case BPF_FUNC_skb_change_head:
7357 return &sk_skb_change_head_proto;
7358 case BPF_FUNC_skb_adjust_room:
7359 return &sk_skb_adjust_room_proto;
7360 case BPF_FUNC_get_socket_cookie:
7361 return &bpf_get_socket_cookie_proto;
7362 case BPF_FUNC_get_socket_uid:
7363 return &bpf_get_socket_uid_proto;
7364 case BPF_FUNC_sk_redirect_map:
7365 return &bpf_sk_redirect_map_proto;
7366 case BPF_FUNC_sk_redirect_hash:
7367 return &bpf_sk_redirect_hash_proto;
7368 case BPF_FUNC_perf_event_output:
7369 return &bpf_skb_event_output_proto;
7371 case BPF_FUNC_sk_lookup_tcp:
7372 return &bpf_sk_lookup_tcp_proto;
7373 case BPF_FUNC_sk_lookup_udp:
7374 return &bpf_sk_lookup_udp_proto;
7375 case BPF_FUNC_sk_release:
7376 return &bpf_sk_release_proto;
7377 case BPF_FUNC_skc_lookup_tcp:
7378 return &bpf_skc_lookup_tcp_proto;
7381 return bpf_sk_base_func_proto(func_id);
7385 static const struct bpf_func_proto *
7386 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7389 case BPF_FUNC_skb_load_bytes:
7390 return &bpf_flow_dissector_load_bytes_proto;
7392 return bpf_sk_base_func_proto(func_id);
7396 static const struct bpf_func_proto *
7397 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7400 case BPF_FUNC_skb_load_bytes:
7401 return &bpf_skb_load_bytes_proto;
7402 case BPF_FUNC_skb_pull_data:
7403 return &bpf_skb_pull_data_proto;
7404 case BPF_FUNC_csum_diff:
7405 return &bpf_csum_diff_proto;
7406 case BPF_FUNC_get_cgroup_classid:
7407 return &bpf_get_cgroup_classid_proto;
7408 case BPF_FUNC_get_route_realm:
7409 return &bpf_get_route_realm_proto;
7410 case BPF_FUNC_get_hash_recalc:
7411 return &bpf_get_hash_recalc_proto;
7412 case BPF_FUNC_perf_event_output:
7413 return &bpf_skb_event_output_proto;
7414 case BPF_FUNC_get_smp_processor_id:
7415 return &bpf_get_smp_processor_id_proto;
7416 case BPF_FUNC_skb_under_cgroup:
7417 return &bpf_skb_under_cgroup_proto;
7419 return bpf_sk_base_func_proto(func_id);
7423 static const struct bpf_func_proto *
7424 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7427 case BPF_FUNC_lwt_push_encap:
7428 return &bpf_lwt_in_push_encap_proto;
7430 return lwt_out_func_proto(func_id, prog);
7434 static const struct bpf_func_proto *
7435 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7438 case BPF_FUNC_skb_get_tunnel_key:
7439 return &bpf_skb_get_tunnel_key_proto;
7440 case BPF_FUNC_skb_set_tunnel_key:
7441 return bpf_get_skb_set_tunnel_proto(func_id);
7442 case BPF_FUNC_skb_get_tunnel_opt:
7443 return &bpf_skb_get_tunnel_opt_proto;
7444 case BPF_FUNC_skb_set_tunnel_opt:
7445 return bpf_get_skb_set_tunnel_proto(func_id);
7446 case BPF_FUNC_redirect:
7447 return &bpf_redirect_proto;
7448 case BPF_FUNC_clone_redirect:
7449 return &bpf_clone_redirect_proto;
7450 case BPF_FUNC_skb_change_tail:
7451 return &bpf_skb_change_tail_proto;
7452 case BPF_FUNC_skb_change_head:
7453 return &bpf_skb_change_head_proto;
7454 case BPF_FUNC_skb_store_bytes:
7455 return &bpf_skb_store_bytes_proto;
7456 case BPF_FUNC_csum_update:
7457 return &bpf_csum_update_proto;
7458 case BPF_FUNC_csum_level:
7459 return &bpf_csum_level_proto;
7460 case BPF_FUNC_l3_csum_replace:
7461 return &bpf_l3_csum_replace_proto;
7462 case BPF_FUNC_l4_csum_replace:
7463 return &bpf_l4_csum_replace_proto;
7464 case BPF_FUNC_set_hash_invalid:
7465 return &bpf_set_hash_invalid_proto;
7466 case BPF_FUNC_lwt_push_encap:
7467 return &bpf_lwt_xmit_push_encap_proto;
7469 return lwt_out_func_proto(func_id, prog);
7473 static const struct bpf_func_proto *
7474 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7477 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7478 case BPF_FUNC_lwt_seg6_store_bytes:
7479 return &bpf_lwt_seg6_store_bytes_proto;
7480 case BPF_FUNC_lwt_seg6_action:
7481 return &bpf_lwt_seg6_action_proto;
7482 case BPF_FUNC_lwt_seg6_adjust_srh:
7483 return &bpf_lwt_seg6_adjust_srh_proto;
7486 return lwt_out_func_proto(func_id, prog);
7490 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
7491 const struct bpf_prog *prog,
7492 struct bpf_insn_access_aux *info)
7494 const int size_default = sizeof(__u32);
7496 if (off < 0 || off >= sizeof(struct __sk_buff))
7499 /* The verifier guarantees that size > 0. */
7500 if (off % size != 0)
7504 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7505 if (off + size > offsetofend(struct __sk_buff, cb[4]))
7508 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
7509 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
7510 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
7511 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
7512 case bpf_ctx_range(struct __sk_buff, data):
7513 case bpf_ctx_range(struct __sk_buff, data_meta):
7514 case bpf_ctx_range(struct __sk_buff, data_end):
7515 if (size != size_default)
7518 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
7520 case bpf_ctx_range(struct __sk_buff, tstamp):
7521 if (size != sizeof(__u64))
7524 case offsetof(struct __sk_buff, sk):
7525 if (type == BPF_WRITE || size != sizeof(__u64))
7527 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
7530 /* Only narrow read access allowed for now. */
7531 if (type == BPF_WRITE) {
7532 if (size != size_default)
7535 bpf_ctx_record_field_size(info, size_default);
7536 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
7544 static bool sk_filter_is_valid_access(int off, int size,
7545 enum bpf_access_type type,
7546 const struct bpf_prog *prog,
7547 struct bpf_insn_access_aux *info)
7550 case bpf_ctx_range(struct __sk_buff, tc_classid):
7551 case bpf_ctx_range(struct __sk_buff, data):
7552 case bpf_ctx_range(struct __sk_buff, data_meta):
7553 case bpf_ctx_range(struct __sk_buff, data_end):
7554 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7555 case bpf_ctx_range(struct __sk_buff, tstamp):
7556 case bpf_ctx_range(struct __sk_buff, wire_len):
7560 if (type == BPF_WRITE) {
7562 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7569 return bpf_skb_is_valid_access(off, size, type, prog, info);
7572 static bool cg_skb_is_valid_access(int off, int size,
7573 enum bpf_access_type type,
7574 const struct bpf_prog *prog,
7575 struct bpf_insn_access_aux *info)
7578 case bpf_ctx_range(struct __sk_buff, tc_classid):
7579 case bpf_ctx_range(struct __sk_buff, data_meta):
7580 case bpf_ctx_range(struct __sk_buff, wire_len):
7582 case bpf_ctx_range(struct __sk_buff, data):
7583 case bpf_ctx_range(struct __sk_buff, data_end):
7589 if (type == BPF_WRITE) {
7591 case bpf_ctx_range(struct __sk_buff, mark):
7592 case bpf_ctx_range(struct __sk_buff, priority):
7593 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7595 case bpf_ctx_range(struct __sk_buff, tstamp):
7605 case bpf_ctx_range(struct __sk_buff, data):
7606 info->reg_type = PTR_TO_PACKET;
7608 case bpf_ctx_range(struct __sk_buff, data_end):
7609 info->reg_type = PTR_TO_PACKET_END;
7613 return bpf_skb_is_valid_access(off, size, type, prog, info);
7616 static bool lwt_is_valid_access(int off, int size,
7617 enum bpf_access_type type,
7618 const struct bpf_prog *prog,
7619 struct bpf_insn_access_aux *info)
7622 case bpf_ctx_range(struct __sk_buff, tc_classid):
7623 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7624 case bpf_ctx_range(struct __sk_buff, data_meta):
7625 case bpf_ctx_range(struct __sk_buff, tstamp):
7626 case bpf_ctx_range(struct __sk_buff, wire_len):
7630 if (type == BPF_WRITE) {
7632 case bpf_ctx_range(struct __sk_buff, mark):
7633 case bpf_ctx_range(struct __sk_buff, priority):
7634 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7642 case bpf_ctx_range(struct __sk_buff, data):
7643 info->reg_type = PTR_TO_PACKET;
7645 case bpf_ctx_range(struct __sk_buff, data_end):
7646 info->reg_type = PTR_TO_PACKET_END;
7650 return bpf_skb_is_valid_access(off, size, type, prog, info);
7653 /* Attach type specific accesses */
7654 static bool __sock_filter_check_attach_type(int off,
7655 enum bpf_access_type access_type,
7656 enum bpf_attach_type attach_type)
7659 case offsetof(struct bpf_sock, bound_dev_if):
7660 case offsetof(struct bpf_sock, mark):
7661 case offsetof(struct bpf_sock, priority):
7662 switch (attach_type) {
7663 case BPF_CGROUP_INET_SOCK_CREATE:
7664 case BPF_CGROUP_INET_SOCK_RELEASE:
7669 case bpf_ctx_range(struct bpf_sock, src_ip4):
7670 switch (attach_type) {
7671 case BPF_CGROUP_INET4_POST_BIND:
7676 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7677 switch (attach_type) {
7678 case BPF_CGROUP_INET6_POST_BIND:
7683 case bpf_ctx_range(struct bpf_sock, src_port):
7684 switch (attach_type) {
7685 case BPF_CGROUP_INET4_POST_BIND:
7686 case BPF_CGROUP_INET6_POST_BIND:
7693 return access_type == BPF_READ;
7698 bool bpf_sock_common_is_valid_access(int off, int size,
7699 enum bpf_access_type type,
7700 struct bpf_insn_access_aux *info)
7703 case bpf_ctx_range_till(struct bpf_sock, type, priority):
7706 return bpf_sock_is_valid_access(off, size, type, info);
7710 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7711 struct bpf_insn_access_aux *info)
7713 const int size_default = sizeof(__u32);
7715 if (off < 0 || off >= sizeof(struct bpf_sock))
7717 if (off % size != 0)
7721 case offsetof(struct bpf_sock, state):
7722 case offsetof(struct bpf_sock, family):
7723 case offsetof(struct bpf_sock, type):
7724 case offsetof(struct bpf_sock, protocol):
7725 case offsetof(struct bpf_sock, dst_port):
7726 case offsetof(struct bpf_sock, src_port):
7727 case offsetof(struct bpf_sock, rx_queue_mapping):
7728 case bpf_ctx_range(struct bpf_sock, src_ip4):
7729 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7730 case bpf_ctx_range(struct bpf_sock, dst_ip4):
7731 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
7732 bpf_ctx_record_field_size(info, size_default);
7733 return bpf_ctx_narrow_access_ok(off, size, size_default);
7736 return size == size_default;
7739 static bool sock_filter_is_valid_access(int off, int size,
7740 enum bpf_access_type type,
7741 const struct bpf_prog *prog,
7742 struct bpf_insn_access_aux *info)
7744 if (!bpf_sock_is_valid_access(off, size, type, info))
7746 return __sock_filter_check_attach_type(off, type,
7747 prog->expected_attach_type);
7750 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
7751 const struct bpf_prog *prog)
7753 /* Neither direct read nor direct write requires any preliminary
7759 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
7760 const struct bpf_prog *prog, int drop_verdict)
7762 struct bpf_insn *insn = insn_buf;
7767 /* if (!skb->cloned)
7770 * (Fast-path, otherwise approximation that we might be
7771 * a clone, do the rest in helper.)
7773 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
7774 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
7775 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
7777 /* ret = bpf_skb_pull_data(skb, 0); */
7778 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
7779 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
7780 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
7781 BPF_FUNC_skb_pull_data);
7784 * return TC_ACT_SHOT;
7786 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
7787 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
7788 *insn++ = BPF_EXIT_INSN();
7791 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
7793 *insn++ = prog->insnsi[0];
7795 return insn - insn_buf;
7798 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
7799 struct bpf_insn *insn_buf)
7801 bool indirect = BPF_MODE(orig->code) == BPF_IND;
7802 struct bpf_insn *insn = insn_buf;
7805 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
7807 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
7809 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
7811 /* We're guaranteed here that CTX is in R6. */
7812 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
7814 switch (BPF_SIZE(orig->code)) {
7816 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
7819 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
7822 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
7826 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
7827 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
7828 *insn++ = BPF_EXIT_INSN();
7830 return insn - insn_buf;
7833 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
7834 const struct bpf_prog *prog)
7836 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
7839 static bool tc_cls_act_is_valid_access(int off, int size,
7840 enum bpf_access_type type,
7841 const struct bpf_prog *prog,
7842 struct bpf_insn_access_aux *info)
7844 if (type == BPF_WRITE) {
7846 case bpf_ctx_range(struct __sk_buff, mark):
7847 case bpf_ctx_range(struct __sk_buff, tc_index):
7848 case bpf_ctx_range(struct __sk_buff, priority):
7849 case bpf_ctx_range(struct __sk_buff, tc_classid):
7850 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7851 case bpf_ctx_range(struct __sk_buff, tstamp):
7852 case bpf_ctx_range(struct __sk_buff, queue_mapping):
7860 case bpf_ctx_range(struct __sk_buff, data):
7861 info->reg_type = PTR_TO_PACKET;
7863 case bpf_ctx_range(struct __sk_buff, data_meta):
7864 info->reg_type = PTR_TO_PACKET_META;
7866 case bpf_ctx_range(struct __sk_buff, data_end):
7867 info->reg_type = PTR_TO_PACKET_END;
7869 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7873 return bpf_skb_is_valid_access(off, size, type, prog, info);
7876 static bool __is_valid_xdp_access(int off, int size)
7878 if (off < 0 || off >= sizeof(struct xdp_md))
7880 if (off % size != 0)
7882 if (size != sizeof(__u32))
7888 static bool xdp_is_valid_access(int off, int size,
7889 enum bpf_access_type type,
7890 const struct bpf_prog *prog,
7891 struct bpf_insn_access_aux *info)
7893 if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
7895 case offsetof(struct xdp_md, egress_ifindex):
7900 if (type == BPF_WRITE) {
7901 if (bpf_prog_is_dev_bound(prog->aux)) {
7903 case offsetof(struct xdp_md, rx_queue_index):
7904 return __is_valid_xdp_access(off, size);
7911 case offsetof(struct xdp_md, data):
7912 info->reg_type = PTR_TO_PACKET;
7914 case offsetof(struct xdp_md, data_meta):
7915 info->reg_type = PTR_TO_PACKET_META;
7917 case offsetof(struct xdp_md, data_end):
7918 info->reg_type = PTR_TO_PACKET_END;
7922 return __is_valid_xdp_access(off, size);
7925 void bpf_warn_invalid_xdp_action(u32 act)
7927 const u32 act_max = XDP_REDIRECT;
7929 WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
7930 act > act_max ? "Illegal" : "Driver unsupported",
7933 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
7935 static bool sock_addr_is_valid_access(int off, int size,
7936 enum bpf_access_type type,
7937 const struct bpf_prog *prog,
7938 struct bpf_insn_access_aux *info)
7940 const int size_default = sizeof(__u32);
7942 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
7944 if (off % size != 0)
7947 /* Disallow access to IPv6 fields from IPv4 contex and vise
7951 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
7952 switch (prog->expected_attach_type) {
7953 case BPF_CGROUP_INET4_BIND:
7954 case BPF_CGROUP_INET4_CONNECT:
7955 case BPF_CGROUP_INET4_GETPEERNAME:
7956 case BPF_CGROUP_INET4_GETSOCKNAME:
7957 case BPF_CGROUP_UDP4_SENDMSG:
7958 case BPF_CGROUP_UDP4_RECVMSG:
7964 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
7965 switch (prog->expected_attach_type) {
7966 case BPF_CGROUP_INET6_BIND:
7967 case BPF_CGROUP_INET6_CONNECT:
7968 case BPF_CGROUP_INET6_GETPEERNAME:
7969 case BPF_CGROUP_INET6_GETSOCKNAME:
7970 case BPF_CGROUP_UDP6_SENDMSG:
7971 case BPF_CGROUP_UDP6_RECVMSG:
7977 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
7978 switch (prog->expected_attach_type) {
7979 case BPF_CGROUP_UDP4_SENDMSG:
7985 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
7987 switch (prog->expected_attach_type) {
7988 case BPF_CGROUP_UDP6_SENDMSG:
7997 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
7998 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
7999 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8000 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8002 case bpf_ctx_range(struct bpf_sock_addr, user_port):
8003 if (type == BPF_READ) {
8004 bpf_ctx_record_field_size(info, size_default);
8006 if (bpf_ctx_wide_access_ok(off, size,
8007 struct bpf_sock_addr,
8011 if (bpf_ctx_wide_access_ok(off, size,
8012 struct bpf_sock_addr,
8016 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8019 if (bpf_ctx_wide_access_ok(off, size,
8020 struct bpf_sock_addr,
8024 if (bpf_ctx_wide_access_ok(off, size,
8025 struct bpf_sock_addr,
8029 if (size != size_default)
8033 case offsetof(struct bpf_sock_addr, sk):
8034 if (type != BPF_READ)
8036 if (size != sizeof(__u64))
8038 info->reg_type = PTR_TO_SOCKET;
8041 if (type == BPF_READ) {
8042 if (size != size_default)
8052 static bool sock_ops_is_valid_access(int off, int size,
8053 enum bpf_access_type type,
8054 const struct bpf_prog *prog,
8055 struct bpf_insn_access_aux *info)
8057 const int size_default = sizeof(__u32);
8059 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
8062 /* The verifier guarantees that size > 0. */
8063 if (off % size != 0)
8066 if (type == BPF_WRITE) {
8068 case offsetof(struct bpf_sock_ops, reply):
8069 case offsetof(struct bpf_sock_ops, sk_txhash):
8070 if (size != size_default)
8078 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
8080 if (size != sizeof(__u64))
8083 case offsetof(struct bpf_sock_ops, sk):
8084 if (size != sizeof(__u64))
8086 info->reg_type = PTR_TO_SOCKET_OR_NULL;
8088 case offsetof(struct bpf_sock_ops, skb_data):
8089 if (size != sizeof(__u64))
8091 info->reg_type = PTR_TO_PACKET;
8093 case offsetof(struct bpf_sock_ops, skb_data_end):
8094 if (size != sizeof(__u64))
8096 info->reg_type = PTR_TO_PACKET_END;
8098 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
8099 bpf_ctx_record_field_size(info, size_default);
8100 return bpf_ctx_narrow_access_ok(off, size,
8103 if (size != size_default)
8112 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
8113 const struct bpf_prog *prog)
8115 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
8118 static bool sk_skb_is_valid_access(int off, int size,
8119 enum bpf_access_type type,
8120 const struct bpf_prog *prog,
8121 struct bpf_insn_access_aux *info)
8124 case bpf_ctx_range(struct __sk_buff, tc_classid):
8125 case bpf_ctx_range(struct __sk_buff, data_meta):
8126 case bpf_ctx_range(struct __sk_buff, tstamp):
8127 case bpf_ctx_range(struct __sk_buff, wire_len):
8131 if (type == BPF_WRITE) {
8133 case bpf_ctx_range(struct __sk_buff, tc_index):
8134 case bpf_ctx_range(struct __sk_buff, priority):
8142 case bpf_ctx_range(struct __sk_buff, mark):
8144 case bpf_ctx_range(struct __sk_buff, data):
8145 info->reg_type = PTR_TO_PACKET;
8147 case bpf_ctx_range(struct __sk_buff, data_end):
8148 info->reg_type = PTR_TO_PACKET_END;
8152 return bpf_skb_is_valid_access(off, size, type, prog, info);
8155 static bool sk_msg_is_valid_access(int off, int size,
8156 enum bpf_access_type type,
8157 const struct bpf_prog *prog,
8158 struct bpf_insn_access_aux *info)
8160 if (type == BPF_WRITE)
8163 if (off % size != 0)
8167 case offsetof(struct sk_msg_md, data):
8168 info->reg_type = PTR_TO_PACKET;
8169 if (size != sizeof(__u64))
8172 case offsetof(struct sk_msg_md, data_end):
8173 info->reg_type = PTR_TO_PACKET_END;
8174 if (size != sizeof(__u64))
8177 case offsetof(struct sk_msg_md, sk):
8178 if (size != sizeof(__u64))
8180 info->reg_type = PTR_TO_SOCKET;
8182 case bpf_ctx_range(struct sk_msg_md, family):
8183 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
8184 case bpf_ctx_range(struct sk_msg_md, local_ip4):
8185 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
8186 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
8187 case bpf_ctx_range(struct sk_msg_md, remote_port):
8188 case bpf_ctx_range(struct sk_msg_md, local_port):
8189 case bpf_ctx_range(struct sk_msg_md, size):
8190 if (size != sizeof(__u32))
8199 static bool flow_dissector_is_valid_access(int off, int size,
8200 enum bpf_access_type type,
8201 const struct bpf_prog *prog,
8202 struct bpf_insn_access_aux *info)
8204 const int size_default = sizeof(__u32);
8206 if (off < 0 || off >= sizeof(struct __sk_buff))
8209 if (type == BPF_WRITE)
8213 case bpf_ctx_range(struct __sk_buff, data):
8214 if (size != size_default)
8216 info->reg_type = PTR_TO_PACKET;
8218 case bpf_ctx_range(struct __sk_buff, data_end):
8219 if (size != size_default)
8221 info->reg_type = PTR_TO_PACKET_END;
8223 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8224 if (size != sizeof(__u64))
8226 info->reg_type = PTR_TO_FLOW_KEYS;
8233 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
8234 const struct bpf_insn *si,
8235 struct bpf_insn *insn_buf,
8236 struct bpf_prog *prog,
8240 struct bpf_insn *insn = insn_buf;
8243 case offsetof(struct __sk_buff, data):
8244 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
8245 si->dst_reg, si->src_reg,
8246 offsetof(struct bpf_flow_dissector, data));
8249 case offsetof(struct __sk_buff, data_end):
8250 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
8251 si->dst_reg, si->src_reg,
8252 offsetof(struct bpf_flow_dissector, data_end));
8255 case offsetof(struct __sk_buff, flow_keys):
8256 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
8257 si->dst_reg, si->src_reg,
8258 offsetof(struct bpf_flow_dissector, flow_keys));
8262 return insn - insn_buf;
8265 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si,
8266 struct bpf_insn *insn)
8268 /* si->dst_reg = skb_shinfo(SKB); */
8269 #ifdef NET_SKBUFF_DATA_USES_OFFSET
8270 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8271 BPF_REG_AX, si->src_reg,
8272 offsetof(struct sk_buff, end));
8273 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
8274 si->dst_reg, si->src_reg,
8275 offsetof(struct sk_buff, head));
8276 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
8278 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8279 si->dst_reg, si->src_reg,
8280 offsetof(struct sk_buff, end));
8286 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
8287 const struct bpf_insn *si,
8288 struct bpf_insn *insn_buf,
8289 struct bpf_prog *prog, u32 *target_size)
8291 struct bpf_insn *insn = insn_buf;
8295 case offsetof(struct __sk_buff, len):
8296 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8297 bpf_target_off(struct sk_buff, len, 4,
8301 case offsetof(struct __sk_buff, protocol):
8302 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8303 bpf_target_off(struct sk_buff, protocol, 2,
8307 case offsetof(struct __sk_buff, vlan_proto):
8308 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8309 bpf_target_off(struct sk_buff, vlan_proto, 2,
8313 case offsetof(struct __sk_buff, priority):
8314 if (type == BPF_WRITE)
8315 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8316 bpf_target_off(struct sk_buff, priority, 4,
8319 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8320 bpf_target_off(struct sk_buff, priority, 4,
8324 case offsetof(struct __sk_buff, ingress_ifindex):
8325 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8326 bpf_target_off(struct sk_buff, skb_iif, 4,
8330 case offsetof(struct __sk_buff, ifindex):
8331 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8332 si->dst_reg, si->src_reg,
8333 offsetof(struct sk_buff, dev));
8334 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
8335 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8336 bpf_target_off(struct net_device, ifindex, 4,
8340 case offsetof(struct __sk_buff, hash):
8341 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8342 bpf_target_off(struct sk_buff, hash, 4,
8346 case offsetof(struct __sk_buff, mark):
8347 if (type == BPF_WRITE)
8348 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8349 bpf_target_off(struct sk_buff, mark, 4,
8352 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8353 bpf_target_off(struct sk_buff, mark, 4,
8357 case offsetof(struct __sk_buff, pkt_type):
8359 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8361 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
8362 #ifdef __BIG_ENDIAN_BITFIELD
8363 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
8367 case offsetof(struct __sk_buff, queue_mapping):
8368 if (type == BPF_WRITE) {
8369 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
8370 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8371 bpf_target_off(struct sk_buff,
8375 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8376 bpf_target_off(struct sk_buff,
8382 case offsetof(struct __sk_buff, vlan_present):
8384 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8385 PKT_VLAN_PRESENT_OFFSET());
8386 if (PKT_VLAN_PRESENT_BIT)
8387 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
8388 if (PKT_VLAN_PRESENT_BIT < 7)
8389 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
8392 case offsetof(struct __sk_buff, vlan_tci):
8393 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8394 bpf_target_off(struct sk_buff, vlan_tci, 2,
8398 case offsetof(struct __sk_buff, cb[0]) ...
8399 offsetofend(struct __sk_buff, cb[4]) - 1:
8400 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
8401 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
8402 offsetof(struct qdisc_skb_cb, data)) %
8405 prog->cb_access = 1;
8407 off -= offsetof(struct __sk_buff, cb[0]);
8408 off += offsetof(struct sk_buff, cb);
8409 off += offsetof(struct qdisc_skb_cb, data);
8410 if (type == BPF_WRITE)
8411 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
8414 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
8418 case offsetof(struct __sk_buff, tc_classid):
8419 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
8422 off -= offsetof(struct __sk_buff, tc_classid);
8423 off += offsetof(struct sk_buff, cb);
8424 off += offsetof(struct qdisc_skb_cb, tc_classid);
8426 if (type == BPF_WRITE)
8427 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
8430 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
8434 case offsetof(struct __sk_buff, data):
8435 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
8436 si->dst_reg, si->src_reg,
8437 offsetof(struct sk_buff, data));
8440 case offsetof(struct __sk_buff, data_meta):
8442 off -= offsetof(struct __sk_buff, data_meta);
8443 off += offsetof(struct sk_buff, cb);
8444 off += offsetof(struct bpf_skb_data_end, data_meta);
8445 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8449 case offsetof(struct __sk_buff, data_end):
8451 off -= offsetof(struct __sk_buff, data_end);
8452 off += offsetof(struct sk_buff, cb);
8453 off += offsetof(struct bpf_skb_data_end, data_end);
8454 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8458 case offsetof(struct __sk_buff, tc_index):
8459 #ifdef CONFIG_NET_SCHED
8460 if (type == BPF_WRITE)
8461 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8462 bpf_target_off(struct sk_buff, tc_index, 2,
8465 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8466 bpf_target_off(struct sk_buff, tc_index, 2,
8470 if (type == BPF_WRITE)
8471 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
8473 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8477 case offsetof(struct __sk_buff, napi_id):
8478 #if defined(CONFIG_NET_RX_BUSY_POLL)
8479 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8480 bpf_target_off(struct sk_buff, napi_id, 4,
8482 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
8483 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8486 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8489 case offsetof(struct __sk_buff, family):
8490 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
8492 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8493 si->dst_reg, si->src_reg,
8494 offsetof(struct sk_buff, sk));
8495 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8496 bpf_target_off(struct sock_common,
8500 case offsetof(struct __sk_buff, remote_ip4):
8501 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
8503 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8504 si->dst_reg, si->src_reg,
8505 offsetof(struct sk_buff, sk));
8506 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8507 bpf_target_off(struct sock_common,
8511 case offsetof(struct __sk_buff, local_ip4):
8512 BUILD_BUG_ON(sizeof_field(struct sock_common,
8513 skc_rcv_saddr) != 4);
8515 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8516 si->dst_reg, si->src_reg,
8517 offsetof(struct sk_buff, sk));
8518 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8519 bpf_target_off(struct sock_common,
8523 case offsetof(struct __sk_buff, remote_ip6[0]) ...
8524 offsetof(struct __sk_buff, remote_ip6[3]):
8525 #if IS_ENABLED(CONFIG_IPV6)
8526 BUILD_BUG_ON(sizeof_field(struct sock_common,
8527 skc_v6_daddr.s6_addr32[0]) != 4);
8530 off -= offsetof(struct __sk_buff, remote_ip6[0]);
8532 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8533 si->dst_reg, si->src_reg,
8534 offsetof(struct sk_buff, sk));
8535 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8536 offsetof(struct sock_common,
8537 skc_v6_daddr.s6_addr32[0]) +
8540 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8543 case offsetof(struct __sk_buff, local_ip6[0]) ...
8544 offsetof(struct __sk_buff, local_ip6[3]):
8545 #if IS_ENABLED(CONFIG_IPV6)
8546 BUILD_BUG_ON(sizeof_field(struct sock_common,
8547 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8550 off -= offsetof(struct __sk_buff, local_ip6[0]);
8552 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8553 si->dst_reg, si->src_reg,
8554 offsetof(struct sk_buff, sk));
8555 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8556 offsetof(struct sock_common,
8557 skc_v6_rcv_saddr.s6_addr32[0]) +
8560 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8564 case offsetof(struct __sk_buff, remote_port):
8565 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
8567 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8568 si->dst_reg, si->src_reg,
8569 offsetof(struct sk_buff, sk));
8570 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8571 bpf_target_off(struct sock_common,
8574 #ifndef __BIG_ENDIAN_BITFIELD
8575 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8579 case offsetof(struct __sk_buff, local_port):
8580 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
8582 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8583 si->dst_reg, si->src_reg,
8584 offsetof(struct sk_buff, sk));
8585 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8586 bpf_target_off(struct sock_common,
8587 skc_num, 2, target_size));
8590 case offsetof(struct __sk_buff, tstamp):
8591 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
8593 if (type == BPF_WRITE)
8594 *insn++ = BPF_STX_MEM(BPF_DW,
8595 si->dst_reg, si->src_reg,
8596 bpf_target_off(struct sk_buff,
8600 *insn++ = BPF_LDX_MEM(BPF_DW,
8601 si->dst_reg, si->src_reg,
8602 bpf_target_off(struct sk_buff,
8607 case offsetof(struct __sk_buff, gso_segs):
8608 insn = bpf_convert_shinfo_access(si, insn);
8609 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
8610 si->dst_reg, si->dst_reg,
8611 bpf_target_off(struct skb_shared_info,
8615 case offsetof(struct __sk_buff, gso_size):
8616 insn = bpf_convert_shinfo_access(si, insn);
8617 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
8618 si->dst_reg, si->dst_reg,
8619 bpf_target_off(struct skb_shared_info,
8623 case offsetof(struct __sk_buff, wire_len):
8624 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
8627 off -= offsetof(struct __sk_buff, wire_len);
8628 off += offsetof(struct sk_buff, cb);
8629 off += offsetof(struct qdisc_skb_cb, pkt_len);
8631 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
8634 case offsetof(struct __sk_buff, sk):
8635 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8636 si->dst_reg, si->src_reg,
8637 offsetof(struct sk_buff, sk));
8641 return insn - insn_buf;
8644 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
8645 const struct bpf_insn *si,
8646 struct bpf_insn *insn_buf,
8647 struct bpf_prog *prog, u32 *target_size)
8649 struct bpf_insn *insn = insn_buf;
8653 case offsetof(struct bpf_sock, bound_dev_if):
8654 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
8656 if (type == BPF_WRITE)
8657 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8658 offsetof(struct sock, sk_bound_dev_if));
8660 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8661 offsetof(struct sock, sk_bound_dev_if));
8664 case offsetof(struct bpf_sock, mark):
8665 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
8667 if (type == BPF_WRITE)
8668 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8669 offsetof(struct sock, sk_mark));
8671 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8672 offsetof(struct sock, sk_mark));
8675 case offsetof(struct bpf_sock, priority):
8676 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
8678 if (type == BPF_WRITE)
8679 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8680 offsetof(struct sock, sk_priority));
8682 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8683 offsetof(struct sock, sk_priority));
8686 case offsetof(struct bpf_sock, family):
8687 *insn++ = BPF_LDX_MEM(
8688 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
8689 si->dst_reg, si->src_reg,
8690 bpf_target_off(struct sock_common,
8692 sizeof_field(struct sock_common,
8697 case offsetof(struct bpf_sock, type):
8698 *insn++ = BPF_LDX_MEM(
8699 BPF_FIELD_SIZEOF(struct sock, sk_type),
8700 si->dst_reg, si->src_reg,
8701 bpf_target_off(struct sock, sk_type,
8702 sizeof_field(struct sock, sk_type),
8706 case offsetof(struct bpf_sock, protocol):
8707 *insn++ = BPF_LDX_MEM(
8708 BPF_FIELD_SIZEOF(struct sock, sk_protocol),
8709 si->dst_reg, si->src_reg,
8710 bpf_target_off(struct sock, sk_protocol,
8711 sizeof_field(struct sock, sk_protocol),
8715 case offsetof(struct bpf_sock, src_ip4):
8716 *insn++ = BPF_LDX_MEM(
8717 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8718 bpf_target_off(struct sock_common, skc_rcv_saddr,
8719 sizeof_field(struct sock_common,
8724 case offsetof(struct bpf_sock, dst_ip4):
8725 *insn++ = BPF_LDX_MEM(
8726 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8727 bpf_target_off(struct sock_common, skc_daddr,
8728 sizeof_field(struct sock_common,
8733 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8734 #if IS_ENABLED(CONFIG_IPV6)
8736 off -= offsetof(struct bpf_sock, src_ip6[0]);
8737 *insn++ = BPF_LDX_MEM(
8738 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8741 skc_v6_rcv_saddr.s6_addr32[0],
8742 sizeof_field(struct sock_common,
8743 skc_v6_rcv_saddr.s6_addr32[0]),
8744 target_size) + off);
8747 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8751 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8752 #if IS_ENABLED(CONFIG_IPV6)
8754 off -= offsetof(struct bpf_sock, dst_ip6[0]);
8755 *insn++ = BPF_LDX_MEM(
8756 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8757 bpf_target_off(struct sock_common,
8758 skc_v6_daddr.s6_addr32[0],
8759 sizeof_field(struct sock_common,
8760 skc_v6_daddr.s6_addr32[0]),
8761 target_size) + off);
8763 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8768 case offsetof(struct bpf_sock, src_port):
8769 *insn++ = BPF_LDX_MEM(
8770 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
8771 si->dst_reg, si->src_reg,
8772 bpf_target_off(struct sock_common, skc_num,
8773 sizeof_field(struct sock_common,
8778 case offsetof(struct bpf_sock, dst_port):
8779 *insn++ = BPF_LDX_MEM(
8780 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
8781 si->dst_reg, si->src_reg,
8782 bpf_target_off(struct sock_common, skc_dport,
8783 sizeof_field(struct sock_common,
8788 case offsetof(struct bpf_sock, state):
8789 *insn++ = BPF_LDX_MEM(
8790 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
8791 si->dst_reg, si->src_reg,
8792 bpf_target_off(struct sock_common, skc_state,
8793 sizeof_field(struct sock_common,
8797 case offsetof(struct bpf_sock, rx_queue_mapping):
8799 *insn++ = BPF_LDX_MEM(
8800 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
8801 si->dst_reg, si->src_reg,
8802 bpf_target_off(struct sock, sk_rx_queue_mapping,
8803 sizeof_field(struct sock,
8804 sk_rx_queue_mapping),
8806 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
8808 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
8810 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
8816 return insn - insn_buf;
8819 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
8820 const struct bpf_insn *si,
8821 struct bpf_insn *insn_buf,
8822 struct bpf_prog *prog, u32 *target_size)
8824 struct bpf_insn *insn = insn_buf;
8827 case offsetof(struct __sk_buff, ifindex):
8828 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8829 si->dst_reg, si->src_reg,
8830 offsetof(struct sk_buff, dev));
8831 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8832 bpf_target_off(struct net_device, ifindex, 4,
8836 return bpf_convert_ctx_access(type, si, insn_buf, prog,
8840 return insn - insn_buf;
8843 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
8844 const struct bpf_insn *si,
8845 struct bpf_insn *insn_buf,
8846 struct bpf_prog *prog, u32 *target_size)
8848 struct bpf_insn *insn = insn_buf;
8851 case offsetof(struct xdp_md, data):
8852 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
8853 si->dst_reg, si->src_reg,
8854 offsetof(struct xdp_buff, data));
8856 case offsetof(struct xdp_md, data_meta):
8857 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
8858 si->dst_reg, si->src_reg,
8859 offsetof(struct xdp_buff, data_meta));
8861 case offsetof(struct xdp_md, data_end):
8862 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
8863 si->dst_reg, si->src_reg,
8864 offsetof(struct xdp_buff, data_end));
8866 case offsetof(struct xdp_md, ingress_ifindex):
8867 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
8868 si->dst_reg, si->src_reg,
8869 offsetof(struct xdp_buff, rxq));
8870 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
8871 si->dst_reg, si->dst_reg,
8872 offsetof(struct xdp_rxq_info, dev));
8873 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8874 offsetof(struct net_device, ifindex));
8876 case offsetof(struct xdp_md, rx_queue_index):
8877 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
8878 si->dst_reg, si->src_reg,
8879 offsetof(struct xdp_buff, rxq));
8880 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8881 offsetof(struct xdp_rxq_info,
8884 case offsetof(struct xdp_md, egress_ifindex):
8885 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
8886 si->dst_reg, si->src_reg,
8887 offsetof(struct xdp_buff, txq));
8888 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
8889 si->dst_reg, si->dst_reg,
8890 offsetof(struct xdp_txq_info, dev));
8891 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8892 offsetof(struct net_device, ifindex));
8896 return insn - insn_buf;
8899 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
8900 * context Structure, F is Field in context structure that contains a pointer
8901 * to Nested Structure of type NS that has the field NF.
8903 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
8904 * sure that SIZE is not greater than actual size of S.F.NF.
8906 * If offset OFF is provided, the load happens from that offset relative to
8909 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
8911 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
8912 si->src_reg, offsetof(S, F)); \
8913 *insn++ = BPF_LDX_MEM( \
8914 SIZE, si->dst_reg, si->dst_reg, \
8915 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
8920 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
8921 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
8922 BPF_FIELD_SIZEOF(NS, NF), 0)
8924 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
8925 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
8927 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
8928 * "register" since two registers available in convert_ctx_access are not
8929 * enough: we can't override neither SRC, since it contains value to store, nor
8930 * DST since it contains pointer to context that may be used by later
8931 * instructions. But we need a temporary place to save pointer to nested
8932 * structure whose field we want to store to.
8934 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
8936 int tmp_reg = BPF_REG_9; \
8937 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
8939 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
8941 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
8943 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
8944 si->dst_reg, offsetof(S, F)); \
8945 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \
8946 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
8949 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
8953 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
8956 if (type == BPF_WRITE) { \
8957 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
8960 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
8961 S, NS, F, NF, SIZE, OFF); \
8965 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
8966 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
8967 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
8969 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
8970 const struct bpf_insn *si,
8971 struct bpf_insn *insn_buf,
8972 struct bpf_prog *prog, u32 *target_size)
8974 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
8975 struct bpf_insn *insn = insn_buf;
8978 case offsetof(struct bpf_sock_addr, user_family):
8979 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
8980 struct sockaddr, uaddr, sa_family);
8983 case offsetof(struct bpf_sock_addr, user_ip4):
8984 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
8985 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
8986 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
8989 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8991 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
8992 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
8993 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
8994 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
8998 case offsetof(struct bpf_sock_addr, user_port):
8999 /* To get port we need to know sa_family first and then treat
9000 * sockaddr as either sockaddr_in or sockaddr_in6.
9001 * Though we can simplify since port field has same offset and
9002 * size in both structures.
9003 * Here we check this invariant and use just one of the
9004 * structures if it's true.
9006 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
9007 offsetof(struct sockaddr_in6, sin6_port));
9008 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
9009 sizeof_field(struct sockaddr_in6, sin6_port));
9010 /* Account for sin6_port being smaller than user_port. */
9011 port_size = min(port_size, BPF_LDST_BYTES(si));
9012 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9013 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9014 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
9017 case offsetof(struct bpf_sock_addr, family):
9018 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9019 struct sock, sk, sk_family);
9022 case offsetof(struct bpf_sock_addr, type):
9023 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9024 struct sock, sk, sk_type);
9027 case offsetof(struct bpf_sock_addr, protocol):
9028 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9029 struct sock, sk, sk_protocol);
9032 case offsetof(struct bpf_sock_addr, msg_src_ip4):
9033 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
9034 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9035 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
9036 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
9039 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9042 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
9043 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
9044 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9045 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
9046 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
9048 case offsetof(struct bpf_sock_addr, sk):
9049 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
9050 si->dst_reg, si->src_reg,
9051 offsetof(struct bpf_sock_addr_kern, sk));
9055 return insn - insn_buf;
9058 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
9059 const struct bpf_insn *si,
9060 struct bpf_insn *insn_buf,
9061 struct bpf_prog *prog,
9064 struct bpf_insn *insn = insn_buf;
9067 /* Helper macro for adding read access to tcp_sock or sock fields. */
9068 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
9070 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \
9071 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
9072 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
9073 if (si->dst_reg == reg || si->src_reg == reg) \
9075 if (si->dst_reg == reg || si->src_reg == reg) \
9077 if (si->dst_reg == si->src_reg) { \
9078 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
9079 offsetof(struct bpf_sock_ops_kern, \
9081 fullsock_reg = reg; \
9084 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9085 struct bpf_sock_ops_kern, \
9087 fullsock_reg, si->src_reg, \
9088 offsetof(struct bpf_sock_ops_kern, \
9090 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
9091 if (si->dst_reg == si->src_reg) \
9092 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9093 offsetof(struct bpf_sock_ops_kern, \
9095 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9096 struct bpf_sock_ops_kern, sk),\
9097 si->dst_reg, si->src_reg, \
9098 offsetof(struct bpf_sock_ops_kern, sk));\
9099 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
9101 si->dst_reg, si->dst_reg, \
9102 offsetof(OBJ, OBJ_FIELD)); \
9103 if (si->dst_reg == si->src_reg) { \
9104 *insn++ = BPF_JMP_A(1); \
9105 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9106 offsetof(struct bpf_sock_ops_kern, \
9111 #define SOCK_OPS_GET_SK() \
9113 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \
9114 if (si->dst_reg == reg || si->src_reg == reg) \
9116 if (si->dst_reg == reg || si->src_reg == reg) \
9118 if (si->dst_reg == si->src_reg) { \
9119 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
9120 offsetof(struct bpf_sock_ops_kern, \
9122 fullsock_reg = reg; \
9125 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9126 struct bpf_sock_ops_kern, \
9128 fullsock_reg, si->src_reg, \
9129 offsetof(struct bpf_sock_ops_kern, \
9131 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
9132 if (si->dst_reg == si->src_reg) \
9133 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9134 offsetof(struct bpf_sock_ops_kern, \
9136 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9137 struct bpf_sock_ops_kern, sk),\
9138 si->dst_reg, si->src_reg, \
9139 offsetof(struct bpf_sock_ops_kern, sk));\
9140 if (si->dst_reg == si->src_reg) { \
9141 *insn++ = BPF_JMP_A(1); \
9142 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
9143 offsetof(struct bpf_sock_ops_kern, \
9148 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
9149 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
9151 /* Helper macro for adding write access to tcp_sock or sock fields.
9152 * The macro is called with two registers, dst_reg which contains a pointer
9153 * to ctx (context) and src_reg which contains the value that should be
9154 * stored. However, we need an additional register since we cannot overwrite
9155 * dst_reg because it may be used later in the program.
9156 * Instead we "borrow" one of the other register. We first save its value
9157 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
9158 * it at the end of the macro.
9160 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
9162 int reg = BPF_REG_9; \
9163 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
9164 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
9165 if (si->dst_reg == reg || si->src_reg == reg) \
9167 if (si->dst_reg == reg || si->src_reg == reg) \
9169 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
9170 offsetof(struct bpf_sock_ops_kern, \
9172 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9173 struct bpf_sock_ops_kern, \
9176 offsetof(struct bpf_sock_ops_kern, \
9178 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
9179 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
9180 struct bpf_sock_ops_kern, sk),\
9182 offsetof(struct bpf_sock_ops_kern, sk));\
9183 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \
9185 offsetof(OBJ, OBJ_FIELD)); \
9186 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
9187 offsetof(struct bpf_sock_ops_kern, \
9191 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
9193 if (TYPE == BPF_WRITE) \
9194 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
9196 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
9199 if (insn > insn_buf)
9200 return insn - insn_buf;
9203 case offsetof(struct bpf_sock_ops, op):
9204 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9206 si->dst_reg, si->src_reg,
9207 offsetof(struct bpf_sock_ops_kern, op));
9210 case offsetof(struct bpf_sock_ops, replylong[0]) ...
9211 offsetof(struct bpf_sock_ops, replylong[3]):
9212 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
9213 sizeof_field(struct bpf_sock_ops_kern, reply));
9214 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
9215 sizeof_field(struct bpf_sock_ops_kern, replylong));
9217 off -= offsetof(struct bpf_sock_ops, replylong[0]);
9218 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
9219 if (type == BPF_WRITE)
9220 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9223 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9227 case offsetof(struct bpf_sock_ops, family):
9228 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9230 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9231 struct bpf_sock_ops_kern, sk),
9232 si->dst_reg, si->src_reg,
9233 offsetof(struct bpf_sock_ops_kern, sk));
9234 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9235 offsetof(struct sock_common, skc_family));
9238 case offsetof(struct bpf_sock_ops, remote_ip4):
9239 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9241 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9242 struct bpf_sock_ops_kern, sk),
9243 si->dst_reg, si->src_reg,
9244 offsetof(struct bpf_sock_ops_kern, sk));
9245 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9246 offsetof(struct sock_common, skc_daddr));
9249 case offsetof(struct bpf_sock_ops, local_ip4):
9250 BUILD_BUG_ON(sizeof_field(struct sock_common,
9251 skc_rcv_saddr) != 4);
9253 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9254 struct bpf_sock_ops_kern, sk),
9255 si->dst_reg, si->src_reg,
9256 offsetof(struct bpf_sock_ops_kern, sk));
9257 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9258 offsetof(struct sock_common,
9262 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
9263 offsetof(struct bpf_sock_ops, remote_ip6[3]):
9264 #if IS_ENABLED(CONFIG_IPV6)
9265 BUILD_BUG_ON(sizeof_field(struct sock_common,
9266 skc_v6_daddr.s6_addr32[0]) != 4);
9269 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
9270 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9271 struct bpf_sock_ops_kern, sk),
9272 si->dst_reg, si->src_reg,
9273 offsetof(struct bpf_sock_ops_kern, sk));
9274 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9275 offsetof(struct sock_common,
9276 skc_v6_daddr.s6_addr32[0]) +
9279 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9283 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
9284 offsetof(struct bpf_sock_ops, local_ip6[3]):
9285 #if IS_ENABLED(CONFIG_IPV6)
9286 BUILD_BUG_ON(sizeof_field(struct sock_common,
9287 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9290 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
9291 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9292 struct bpf_sock_ops_kern, sk),
9293 si->dst_reg, si->src_reg,
9294 offsetof(struct bpf_sock_ops_kern, sk));
9295 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9296 offsetof(struct sock_common,
9297 skc_v6_rcv_saddr.s6_addr32[0]) +
9300 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9304 case offsetof(struct bpf_sock_ops, remote_port):
9305 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9307 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9308 struct bpf_sock_ops_kern, sk),
9309 si->dst_reg, si->src_reg,
9310 offsetof(struct bpf_sock_ops_kern, sk));
9311 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9312 offsetof(struct sock_common, skc_dport));
9313 #ifndef __BIG_ENDIAN_BITFIELD
9314 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9318 case offsetof(struct bpf_sock_ops, local_port):
9319 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9321 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9322 struct bpf_sock_ops_kern, sk),
9323 si->dst_reg, si->src_reg,
9324 offsetof(struct bpf_sock_ops_kern, sk));
9325 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9326 offsetof(struct sock_common, skc_num));
9329 case offsetof(struct bpf_sock_ops, is_fullsock):
9330 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9331 struct bpf_sock_ops_kern,
9333 si->dst_reg, si->src_reg,
9334 offsetof(struct bpf_sock_ops_kern,
9338 case offsetof(struct bpf_sock_ops, state):
9339 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
9341 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9342 struct bpf_sock_ops_kern, sk),
9343 si->dst_reg, si->src_reg,
9344 offsetof(struct bpf_sock_ops_kern, sk));
9345 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
9346 offsetof(struct sock_common, skc_state));
9349 case offsetof(struct bpf_sock_ops, rtt_min):
9350 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
9351 sizeof(struct minmax));
9352 BUILD_BUG_ON(sizeof(struct minmax) <
9353 sizeof(struct minmax_sample));
9355 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9356 struct bpf_sock_ops_kern, sk),
9357 si->dst_reg, si->src_reg,
9358 offsetof(struct bpf_sock_ops_kern, sk));
9359 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9360 offsetof(struct tcp_sock, rtt_min) +
9361 sizeof_field(struct minmax_sample, t));
9364 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
9365 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
9369 case offsetof(struct bpf_sock_ops, sk_txhash):
9370 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
9373 case offsetof(struct bpf_sock_ops, snd_cwnd):
9374 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
9376 case offsetof(struct bpf_sock_ops, srtt_us):
9377 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
9379 case offsetof(struct bpf_sock_ops, snd_ssthresh):
9380 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
9382 case offsetof(struct bpf_sock_ops, rcv_nxt):
9383 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
9385 case offsetof(struct bpf_sock_ops, snd_nxt):
9386 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
9388 case offsetof(struct bpf_sock_ops, snd_una):
9389 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
9391 case offsetof(struct bpf_sock_ops, mss_cache):
9392 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
9394 case offsetof(struct bpf_sock_ops, ecn_flags):
9395 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
9397 case offsetof(struct bpf_sock_ops, rate_delivered):
9398 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
9400 case offsetof(struct bpf_sock_ops, rate_interval_us):
9401 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
9403 case offsetof(struct bpf_sock_ops, packets_out):
9404 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
9406 case offsetof(struct bpf_sock_ops, retrans_out):
9407 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
9409 case offsetof(struct bpf_sock_ops, total_retrans):
9410 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
9412 case offsetof(struct bpf_sock_ops, segs_in):
9413 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
9415 case offsetof(struct bpf_sock_ops, data_segs_in):
9416 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
9418 case offsetof(struct bpf_sock_ops, segs_out):
9419 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
9421 case offsetof(struct bpf_sock_ops, data_segs_out):
9422 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
9424 case offsetof(struct bpf_sock_ops, lost_out):
9425 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
9427 case offsetof(struct bpf_sock_ops, sacked_out):
9428 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
9430 case offsetof(struct bpf_sock_ops, bytes_received):
9431 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
9433 case offsetof(struct bpf_sock_ops, bytes_acked):
9434 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
9436 case offsetof(struct bpf_sock_ops, sk):
9439 case offsetof(struct bpf_sock_ops, skb_data_end):
9440 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9442 si->dst_reg, si->src_reg,
9443 offsetof(struct bpf_sock_ops_kern,
9446 case offsetof(struct bpf_sock_ops, skb_data):
9447 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9449 si->dst_reg, si->src_reg,
9450 offsetof(struct bpf_sock_ops_kern,
9452 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9453 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9454 si->dst_reg, si->dst_reg,
9455 offsetof(struct sk_buff, data));
9457 case offsetof(struct bpf_sock_ops, skb_len):
9458 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9460 si->dst_reg, si->src_reg,
9461 offsetof(struct bpf_sock_ops_kern,
9463 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9464 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9465 si->dst_reg, si->dst_reg,
9466 offsetof(struct sk_buff, len));
9468 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9469 off = offsetof(struct sk_buff, cb);
9470 off += offsetof(struct tcp_skb_cb, tcp_flags);
9471 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
9472 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9474 si->dst_reg, si->src_reg,
9475 offsetof(struct bpf_sock_ops_kern,
9477 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9478 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
9480 si->dst_reg, si->dst_reg, off);
9483 return insn - insn_buf;
9486 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
9487 const struct bpf_insn *si,
9488 struct bpf_insn *insn_buf,
9489 struct bpf_prog *prog, u32 *target_size)
9491 struct bpf_insn *insn = insn_buf;
9495 case offsetof(struct __sk_buff, data_end):
9497 off -= offsetof(struct __sk_buff, data_end);
9498 off += offsetof(struct sk_buff, cb);
9499 off += offsetof(struct tcp_skb_cb, bpf.data_end);
9500 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9504 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9508 return insn - insn_buf;
9511 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
9512 const struct bpf_insn *si,
9513 struct bpf_insn *insn_buf,
9514 struct bpf_prog *prog, u32 *target_size)
9516 struct bpf_insn *insn = insn_buf;
9517 #if IS_ENABLED(CONFIG_IPV6)
9521 /* convert ctx uses the fact sg element is first in struct */
9522 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
9525 case offsetof(struct sk_msg_md, data):
9526 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
9527 si->dst_reg, si->src_reg,
9528 offsetof(struct sk_msg, data));
9530 case offsetof(struct sk_msg_md, data_end):
9531 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
9532 si->dst_reg, si->src_reg,
9533 offsetof(struct sk_msg, data_end));
9535 case offsetof(struct sk_msg_md, family):
9536 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9538 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9540 si->dst_reg, si->src_reg,
9541 offsetof(struct sk_msg, sk));
9542 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9543 offsetof(struct sock_common, skc_family));
9546 case offsetof(struct sk_msg_md, remote_ip4):
9547 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9549 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9551 si->dst_reg, si->src_reg,
9552 offsetof(struct sk_msg, sk));
9553 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9554 offsetof(struct sock_common, skc_daddr));
9557 case offsetof(struct sk_msg_md, local_ip4):
9558 BUILD_BUG_ON(sizeof_field(struct sock_common,
9559 skc_rcv_saddr) != 4);
9561 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9563 si->dst_reg, si->src_reg,
9564 offsetof(struct sk_msg, sk));
9565 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9566 offsetof(struct sock_common,
9570 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
9571 offsetof(struct sk_msg_md, remote_ip6[3]):
9572 #if IS_ENABLED(CONFIG_IPV6)
9573 BUILD_BUG_ON(sizeof_field(struct sock_common,
9574 skc_v6_daddr.s6_addr32[0]) != 4);
9577 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
9578 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9580 si->dst_reg, si->src_reg,
9581 offsetof(struct sk_msg, sk));
9582 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9583 offsetof(struct sock_common,
9584 skc_v6_daddr.s6_addr32[0]) +
9587 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9591 case offsetof(struct sk_msg_md, local_ip6[0]) ...
9592 offsetof(struct sk_msg_md, local_ip6[3]):
9593 #if IS_ENABLED(CONFIG_IPV6)
9594 BUILD_BUG_ON(sizeof_field(struct sock_common,
9595 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9598 off -= offsetof(struct sk_msg_md, local_ip6[0]);
9599 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9601 si->dst_reg, si->src_reg,
9602 offsetof(struct sk_msg, sk));
9603 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9604 offsetof(struct sock_common,
9605 skc_v6_rcv_saddr.s6_addr32[0]) +
9608 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9612 case offsetof(struct sk_msg_md, remote_port):
9613 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9615 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9617 si->dst_reg, si->src_reg,
9618 offsetof(struct sk_msg, sk));
9619 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9620 offsetof(struct sock_common, skc_dport));
9621 #ifndef __BIG_ENDIAN_BITFIELD
9622 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9626 case offsetof(struct sk_msg_md, local_port):
9627 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9629 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9631 si->dst_reg, si->src_reg,
9632 offsetof(struct sk_msg, sk));
9633 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9634 offsetof(struct sock_common, skc_num));
9637 case offsetof(struct sk_msg_md, size):
9638 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
9639 si->dst_reg, si->src_reg,
9640 offsetof(struct sk_msg_sg, size));
9643 case offsetof(struct sk_msg_md, sk):
9644 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
9645 si->dst_reg, si->src_reg,
9646 offsetof(struct sk_msg, sk));
9650 return insn - insn_buf;
9653 const struct bpf_verifier_ops sk_filter_verifier_ops = {
9654 .get_func_proto = sk_filter_func_proto,
9655 .is_valid_access = sk_filter_is_valid_access,
9656 .convert_ctx_access = bpf_convert_ctx_access,
9657 .gen_ld_abs = bpf_gen_ld_abs,
9660 const struct bpf_prog_ops sk_filter_prog_ops = {
9661 .test_run = bpf_prog_test_run_skb,
9664 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
9665 .get_func_proto = tc_cls_act_func_proto,
9666 .is_valid_access = tc_cls_act_is_valid_access,
9667 .convert_ctx_access = tc_cls_act_convert_ctx_access,
9668 .gen_prologue = tc_cls_act_prologue,
9669 .gen_ld_abs = bpf_gen_ld_abs,
9672 const struct bpf_prog_ops tc_cls_act_prog_ops = {
9673 .test_run = bpf_prog_test_run_skb,
9676 const struct bpf_verifier_ops xdp_verifier_ops = {
9677 .get_func_proto = xdp_func_proto,
9678 .is_valid_access = xdp_is_valid_access,
9679 .convert_ctx_access = xdp_convert_ctx_access,
9680 .gen_prologue = bpf_noop_prologue,
9683 const struct bpf_prog_ops xdp_prog_ops = {
9684 .test_run = bpf_prog_test_run_xdp,
9687 const struct bpf_verifier_ops cg_skb_verifier_ops = {
9688 .get_func_proto = cg_skb_func_proto,
9689 .is_valid_access = cg_skb_is_valid_access,
9690 .convert_ctx_access = bpf_convert_ctx_access,
9693 const struct bpf_prog_ops cg_skb_prog_ops = {
9694 .test_run = bpf_prog_test_run_skb,
9697 const struct bpf_verifier_ops lwt_in_verifier_ops = {
9698 .get_func_proto = lwt_in_func_proto,
9699 .is_valid_access = lwt_is_valid_access,
9700 .convert_ctx_access = bpf_convert_ctx_access,
9703 const struct bpf_prog_ops lwt_in_prog_ops = {
9704 .test_run = bpf_prog_test_run_skb,
9707 const struct bpf_verifier_ops lwt_out_verifier_ops = {
9708 .get_func_proto = lwt_out_func_proto,
9709 .is_valid_access = lwt_is_valid_access,
9710 .convert_ctx_access = bpf_convert_ctx_access,
9713 const struct bpf_prog_ops lwt_out_prog_ops = {
9714 .test_run = bpf_prog_test_run_skb,
9717 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
9718 .get_func_proto = lwt_xmit_func_proto,
9719 .is_valid_access = lwt_is_valid_access,
9720 .convert_ctx_access = bpf_convert_ctx_access,
9721 .gen_prologue = tc_cls_act_prologue,
9724 const struct bpf_prog_ops lwt_xmit_prog_ops = {
9725 .test_run = bpf_prog_test_run_skb,
9728 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
9729 .get_func_proto = lwt_seg6local_func_proto,
9730 .is_valid_access = lwt_is_valid_access,
9731 .convert_ctx_access = bpf_convert_ctx_access,
9734 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
9735 .test_run = bpf_prog_test_run_skb,
9738 const struct bpf_verifier_ops cg_sock_verifier_ops = {
9739 .get_func_proto = sock_filter_func_proto,
9740 .is_valid_access = sock_filter_is_valid_access,
9741 .convert_ctx_access = bpf_sock_convert_ctx_access,
9744 const struct bpf_prog_ops cg_sock_prog_ops = {
9747 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
9748 .get_func_proto = sock_addr_func_proto,
9749 .is_valid_access = sock_addr_is_valid_access,
9750 .convert_ctx_access = sock_addr_convert_ctx_access,
9753 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
9756 const struct bpf_verifier_ops sock_ops_verifier_ops = {
9757 .get_func_proto = sock_ops_func_proto,
9758 .is_valid_access = sock_ops_is_valid_access,
9759 .convert_ctx_access = sock_ops_convert_ctx_access,
9762 const struct bpf_prog_ops sock_ops_prog_ops = {
9765 const struct bpf_verifier_ops sk_skb_verifier_ops = {
9766 .get_func_proto = sk_skb_func_proto,
9767 .is_valid_access = sk_skb_is_valid_access,
9768 .convert_ctx_access = sk_skb_convert_ctx_access,
9769 .gen_prologue = sk_skb_prologue,
9772 const struct bpf_prog_ops sk_skb_prog_ops = {
9775 const struct bpf_verifier_ops sk_msg_verifier_ops = {
9776 .get_func_proto = sk_msg_func_proto,
9777 .is_valid_access = sk_msg_is_valid_access,
9778 .convert_ctx_access = sk_msg_convert_ctx_access,
9779 .gen_prologue = bpf_noop_prologue,
9782 const struct bpf_prog_ops sk_msg_prog_ops = {
9785 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
9786 .get_func_proto = flow_dissector_func_proto,
9787 .is_valid_access = flow_dissector_is_valid_access,
9788 .convert_ctx_access = flow_dissector_convert_ctx_access,
9791 const struct bpf_prog_ops flow_dissector_prog_ops = {
9792 .test_run = bpf_prog_test_run_flow_dissector,
9795 int sk_detach_filter(struct sock *sk)
9798 struct sk_filter *filter;
9800 if (sock_flag(sk, SOCK_FILTER_LOCKED))
9803 filter = rcu_dereference_protected(sk->sk_filter,
9804 lockdep_sock_is_held(sk));
9806 RCU_INIT_POINTER(sk->sk_filter, NULL);
9807 sk_filter_uncharge(sk, filter);
9813 EXPORT_SYMBOL_GPL(sk_detach_filter);
9815 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
9818 struct sock_fprog_kern *fprog;
9819 struct sk_filter *filter;
9823 filter = rcu_dereference_protected(sk->sk_filter,
9824 lockdep_sock_is_held(sk));
9828 /* We're copying the filter that has been originally attached,
9829 * so no conversion/decode needed anymore. eBPF programs that
9830 * have no original program cannot be dumped through this.
9833 fprog = filter->prog->orig_prog;
9839 /* User space only enquires number of filter blocks. */
9843 if (len < fprog->len)
9847 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
9850 /* Instead of bytes, the API requests to return the number
9860 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
9861 struct sock_reuseport *reuse,
9862 struct sock *sk, struct sk_buff *skb,
9865 reuse_kern->skb = skb;
9866 reuse_kern->sk = sk;
9867 reuse_kern->selected_sk = NULL;
9868 reuse_kern->data_end = skb->data + skb_headlen(skb);
9869 reuse_kern->hash = hash;
9870 reuse_kern->reuseport_id = reuse->reuseport_id;
9871 reuse_kern->bind_inany = reuse->bind_inany;
9874 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
9875 struct bpf_prog *prog, struct sk_buff *skb,
9878 struct sk_reuseport_kern reuse_kern;
9879 enum sk_action action;
9881 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, hash);
9882 action = BPF_PROG_RUN(prog, &reuse_kern);
9884 if (action == SK_PASS)
9885 return reuse_kern.selected_sk;
9887 return ERR_PTR(-ECONNREFUSED);
9890 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
9891 struct bpf_map *, map, void *, key, u32, flags)
9893 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
9894 struct sock_reuseport *reuse;
9895 struct sock *selected_sk;
9897 selected_sk = map->ops->map_lookup_elem(map, key);
9901 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
9903 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
9904 if (sk_is_refcounted(selected_sk))
9905 sock_put(selected_sk);
9907 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
9908 * The only (!reuse) case here is - the sk has already been
9909 * unhashed (e.g. by close()), so treat it as -ENOENT.
9911 * Other maps (e.g. sock_map) do not provide this guarantee and
9912 * the sk may never be in the reuseport group to begin with.
9914 return is_sockarray ? -ENOENT : -EINVAL;
9917 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
9918 struct sock *sk = reuse_kern->sk;
9920 if (sk->sk_protocol != selected_sk->sk_protocol)
9922 else if (sk->sk_family != selected_sk->sk_family)
9923 return -EAFNOSUPPORT;
9925 /* Catch all. Likely bound to a different sockaddr. */
9929 reuse_kern->selected_sk = selected_sk;
9934 static const struct bpf_func_proto sk_select_reuseport_proto = {
9935 .func = sk_select_reuseport,
9937 .ret_type = RET_INTEGER,
9938 .arg1_type = ARG_PTR_TO_CTX,
9939 .arg2_type = ARG_CONST_MAP_PTR,
9940 .arg3_type = ARG_PTR_TO_MAP_KEY,
9941 .arg4_type = ARG_ANYTHING,
9944 BPF_CALL_4(sk_reuseport_load_bytes,
9945 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
9946 void *, to, u32, len)
9948 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
9951 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
9952 .func = sk_reuseport_load_bytes,
9954 .ret_type = RET_INTEGER,
9955 .arg1_type = ARG_PTR_TO_CTX,
9956 .arg2_type = ARG_ANYTHING,
9957 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
9958 .arg4_type = ARG_CONST_SIZE,
9961 BPF_CALL_5(sk_reuseport_load_bytes_relative,
9962 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
9963 void *, to, u32, len, u32, start_header)
9965 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
9969 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
9970 .func = sk_reuseport_load_bytes_relative,
9972 .ret_type = RET_INTEGER,
9973 .arg1_type = ARG_PTR_TO_CTX,
9974 .arg2_type = ARG_ANYTHING,
9975 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
9976 .arg4_type = ARG_CONST_SIZE,
9977 .arg5_type = ARG_ANYTHING,
9980 static const struct bpf_func_proto *
9981 sk_reuseport_func_proto(enum bpf_func_id func_id,
9982 const struct bpf_prog *prog)
9985 case BPF_FUNC_sk_select_reuseport:
9986 return &sk_select_reuseport_proto;
9987 case BPF_FUNC_skb_load_bytes:
9988 return &sk_reuseport_load_bytes_proto;
9989 case BPF_FUNC_skb_load_bytes_relative:
9990 return &sk_reuseport_load_bytes_relative_proto;
9992 return bpf_base_func_proto(func_id);
9997 sk_reuseport_is_valid_access(int off, int size,
9998 enum bpf_access_type type,
9999 const struct bpf_prog *prog,
10000 struct bpf_insn_access_aux *info)
10002 const u32 size_default = sizeof(__u32);
10004 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
10005 off % size || type != BPF_READ)
10009 case offsetof(struct sk_reuseport_md, data):
10010 info->reg_type = PTR_TO_PACKET;
10011 return size == sizeof(__u64);
10013 case offsetof(struct sk_reuseport_md, data_end):
10014 info->reg_type = PTR_TO_PACKET_END;
10015 return size == sizeof(__u64);
10017 case offsetof(struct sk_reuseport_md, hash):
10018 return size == size_default;
10020 /* Fields that allow narrowing */
10021 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
10022 if (size < sizeof_field(struct sk_buff, protocol))
10025 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
10026 case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
10027 case bpf_ctx_range(struct sk_reuseport_md, len):
10028 bpf_ctx_record_field_size(info, size_default);
10029 return bpf_ctx_narrow_access_ok(off, size, size_default);
10036 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
10037 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
10038 si->dst_reg, si->src_reg, \
10039 bpf_target_off(struct sk_reuseport_kern, F, \
10040 sizeof_field(struct sk_reuseport_kern, F), \
10044 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
10045 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
10050 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \
10051 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
10056 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
10057 const struct bpf_insn *si,
10058 struct bpf_insn *insn_buf,
10059 struct bpf_prog *prog,
10062 struct bpf_insn *insn = insn_buf;
10065 case offsetof(struct sk_reuseport_md, data):
10066 SK_REUSEPORT_LOAD_SKB_FIELD(data);
10069 case offsetof(struct sk_reuseport_md, len):
10070 SK_REUSEPORT_LOAD_SKB_FIELD(len);
10073 case offsetof(struct sk_reuseport_md, eth_protocol):
10074 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
10077 case offsetof(struct sk_reuseport_md, ip_protocol):
10078 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
10081 case offsetof(struct sk_reuseport_md, data_end):
10082 SK_REUSEPORT_LOAD_FIELD(data_end);
10085 case offsetof(struct sk_reuseport_md, hash):
10086 SK_REUSEPORT_LOAD_FIELD(hash);
10089 case offsetof(struct sk_reuseport_md, bind_inany):
10090 SK_REUSEPORT_LOAD_FIELD(bind_inany);
10094 return insn - insn_buf;
10097 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
10098 .get_func_proto = sk_reuseport_func_proto,
10099 .is_valid_access = sk_reuseport_is_valid_access,
10100 .convert_ctx_access = sk_reuseport_convert_ctx_access,
10103 const struct bpf_prog_ops sk_reuseport_prog_ops = {
10106 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
10107 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
10109 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
10110 struct sock *, sk, u64, flags)
10112 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
10113 BPF_SK_LOOKUP_F_NO_REUSEPORT)))
10115 if (unlikely(sk && sk_is_refcounted(sk)))
10116 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
10117 if (unlikely(sk && sk->sk_state == TCP_ESTABLISHED))
10118 return -ESOCKTNOSUPPORT; /* reject connected sockets */
10120 /* Check if socket is suitable for packet L3/L4 protocol */
10121 if (sk && sk->sk_protocol != ctx->protocol)
10122 return -EPROTOTYPE;
10123 if (sk && sk->sk_family != ctx->family &&
10124 (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
10125 return -EAFNOSUPPORT;
10127 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
10130 /* Select socket as lookup result */
10131 ctx->selected_sk = sk;
10132 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
10136 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
10137 .func = bpf_sk_lookup_assign,
10139 .ret_type = RET_INTEGER,
10140 .arg1_type = ARG_PTR_TO_CTX,
10141 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL,
10142 .arg3_type = ARG_ANYTHING,
10145 static const struct bpf_func_proto *
10146 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
10149 case BPF_FUNC_perf_event_output:
10150 return &bpf_event_output_data_proto;
10151 case BPF_FUNC_sk_assign:
10152 return &bpf_sk_lookup_assign_proto;
10153 case BPF_FUNC_sk_release:
10154 return &bpf_sk_release_proto;
10156 return bpf_sk_base_func_proto(func_id);
10160 static bool sk_lookup_is_valid_access(int off, int size,
10161 enum bpf_access_type type,
10162 const struct bpf_prog *prog,
10163 struct bpf_insn_access_aux *info)
10165 if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
10167 if (off % size != 0)
10169 if (type != BPF_READ)
10173 case offsetof(struct bpf_sk_lookup, sk):
10174 info->reg_type = PTR_TO_SOCKET_OR_NULL;
10175 return size == sizeof(__u64);
10177 case bpf_ctx_range(struct bpf_sk_lookup, family):
10178 case bpf_ctx_range(struct bpf_sk_lookup, protocol):
10179 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
10180 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
10181 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
10182 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
10183 case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
10184 case bpf_ctx_range(struct bpf_sk_lookup, local_port):
10185 bpf_ctx_record_field_size(info, sizeof(__u32));
10186 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
10193 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
10194 const struct bpf_insn *si,
10195 struct bpf_insn *insn_buf,
10196 struct bpf_prog *prog,
10199 struct bpf_insn *insn = insn_buf;
10202 case offsetof(struct bpf_sk_lookup, sk):
10203 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10204 offsetof(struct bpf_sk_lookup_kern, selected_sk));
10207 case offsetof(struct bpf_sk_lookup, family):
10208 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10209 bpf_target_off(struct bpf_sk_lookup_kern,
10210 family, 2, target_size));
10213 case offsetof(struct bpf_sk_lookup, protocol):
10214 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10215 bpf_target_off(struct bpf_sk_lookup_kern,
10216 protocol, 2, target_size));
10219 case offsetof(struct bpf_sk_lookup, remote_ip4):
10220 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10221 bpf_target_off(struct bpf_sk_lookup_kern,
10222 v4.saddr, 4, target_size));
10225 case offsetof(struct bpf_sk_lookup, local_ip4):
10226 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10227 bpf_target_off(struct bpf_sk_lookup_kern,
10228 v4.daddr, 4, target_size));
10231 case bpf_ctx_range_till(struct bpf_sk_lookup,
10232 remote_ip6[0], remote_ip6[3]): {
10233 #if IS_ENABLED(CONFIG_IPV6)
10236 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
10237 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10238 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10239 offsetof(struct bpf_sk_lookup_kern, v6.saddr));
10240 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10241 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10243 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10247 case bpf_ctx_range_till(struct bpf_sk_lookup,
10248 local_ip6[0], local_ip6[3]): {
10249 #if IS_ENABLED(CONFIG_IPV6)
10252 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
10253 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10254 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10255 offsetof(struct bpf_sk_lookup_kern, v6.daddr));
10256 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10257 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10259 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10263 case offsetof(struct bpf_sk_lookup, remote_port):
10264 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10265 bpf_target_off(struct bpf_sk_lookup_kern,
10266 sport, 2, target_size));
10269 case offsetof(struct bpf_sk_lookup, local_port):
10270 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10271 bpf_target_off(struct bpf_sk_lookup_kern,
10272 dport, 2, target_size));
10276 return insn - insn_buf;
10279 const struct bpf_prog_ops sk_lookup_prog_ops = {
10282 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
10283 .get_func_proto = sk_lookup_func_proto,
10284 .is_valid_access = sk_lookup_is_valid_access,
10285 .convert_ctx_access = sk_lookup_convert_ctx_access,
10288 #endif /* CONFIG_INET */
10290 DEFINE_BPF_DISPATCHER(xdp)
10292 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
10294 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
10297 #ifdef CONFIG_DEBUG_INFO_BTF
10298 BTF_ID_LIST_GLOBAL(btf_sock_ids)
10299 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
10301 #undef BTF_SOCK_TYPE
10303 u32 btf_sock_ids[MAX_BTF_SOCK_TYPE];
10306 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
10308 /* tcp6_sock type is not generated in dwarf and hence btf,
10309 * trigger an explicit type generation here.
10311 BTF_TYPE_EMIT(struct tcp6_sock);
10312 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
10313 sk->sk_family == AF_INET6)
10314 return (unsigned long)sk;
10316 return (unsigned long)NULL;
10319 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
10320 .func = bpf_skc_to_tcp6_sock,
10322 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10323 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10324 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
10327 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
10329 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
10330 return (unsigned long)sk;
10332 return (unsigned long)NULL;
10335 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
10336 .func = bpf_skc_to_tcp_sock,
10338 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10339 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10340 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
10343 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
10345 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
10346 * generated if CONFIG_INET=n. Trigger an explicit generation here.
10348 BTF_TYPE_EMIT(struct inet_timewait_sock);
10349 BTF_TYPE_EMIT(struct tcp_timewait_sock);
10352 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
10353 return (unsigned long)sk;
10356 #if IS_BUILTIN(CONFIG_IPV6)
10357 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
10358 return (unsigned long)sk;
10361 return (unsigned long)NULL;
10364 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
10365 .func = bpf_skc_to_tcp_timewait_sock,
10367 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10368 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10369 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
10372 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
10375 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10376 return (unsigned long)sk;
10379 #if IS_BUILTIN(CONFIG_IPV6)
10380 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10381 return (unsigned long)sk;
10384 return (unsigned long)NULL;
10387 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
10388 .func = bpf_skc_to_tcp_request_sock,
10390 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10391 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10392 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
10395 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
10397 /* udp6_sock type is not generated in dwarf and hence btf,
10398 * trigger an explicit type generation here.
10400 BTF_TYPE_EMIT(struct udp6_sock);
10401 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
10402 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
10403 return (unsigned long)sk;
10405 return (unsigned long)NULL;
10408 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
10409 .func = bpf_skc_to_udp6_sock,
10411 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10412 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10413 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
10416 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
10418 return (unsigned long)sock_from_file(file);
10421 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
10422 BTF_ID(struct, socket)
10423 BTF_ID(struct, file)
10425 const struct bpf_func_proto bpf_sock_from_file_proto = {
10426 .func = bpf_sock_from_file,
10428 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
10429 .ret_btf_id = &bpf_sock_from_file_btf_ids[0],
10430 .arg1_type = ARG_PTR_TO_BTF_ID,
10431 .arg1_btf_id = &bpf_sock_from_file_btf_ids[1],
10434 static const struct bpf_func_proto *
10435 bpf_sk_base_func_proto(enum bpf_func_id func_id)
10437 const struct bpf_func_proto *func;
10440 case BPF_FUNC_skc_to_tcp6_sock:
10441 func = &bpf_skc_to_tcp6_sock_proto;
10443 case BPF_FUNC_skc_to_tcp_sock:
10444 func = &bpf_skc_to_tcp_sock_proto;
10446 case BPF_FUNC_skc_to_tcp_timewait_sock:
10447 func = &bpf_skc_to_tcp_timewait_sock_proto;
10449 case BPF_FUNC_skc_to_tcp_request_sock:
10450 func = &bpf_skc_to_tcp_request_sock_proto;
10452 case BPF_FUNC_skc_to_udp6_sock:
10453 func = &bpf_skc_to_udp6_sock_proto;
10456 return bpf_base_func_proto(func_id);
10459 if (!perfmon_capable())