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 <net/sch_generic.h>
51 #include <net/cls_cgroup.h>
52 #include <net/dst_metadata.h>
54 #include <net/sock_reuseport.h>
55 #include <net/busy_poll.h>
59 #include <linux/bpf_trace.h>
60 #include <net/xdp_sock.h>
61 #include <linux/inetdevice.h>
62 #include <net/inet_hashtables.h>
63 #include <net/inet6_hashtables.h>
64 #include <net/ip_fib.h>
65 #include <net/nexthop.h>
69 #include <net/net_namespace.h>
70 #include <linux/seg6_local.h>
72 #include <net/seg6_local.h>
73 #include <net/lwtunnel.h>
74 #include <net/ipv6_stubs.h>
75 #include <net/bpf_sk_storage.h>
78 * sk_filter_trim_cap - run a packet through a socket filter
79 * @sk: sock associated with &sk_buff
80 * @skb: buffer to filter
81 * @cap: limit on how short the eBPF program may trim the packet
83 * Run the eBPF program and then cut skb->data to correct size returned by
84 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
85 * than pkt_len we keep whole skb->data. This is the socket level
86 * wrapper to BPF_PROG_RUN. It returns 0 if the packet should
87 * be accepted or -EPERM if the packet should be tossed.
90 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
93 struct sk_filter *filter;
96 * If the skb was allocated from pfmemalloc reserves, only
97 * allow SOCK_MEMALLOC sockets to use it as this socket is
100 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
101 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
104 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
108 err = security_sock_rcv_skb(sk, skb);
113 filter = rcu_dereference(sk->sk_filter);
115 struct sock *save_sk = skb->sk;
116 unsigned int pkt_len;
119 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
121 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
127 EXPORT_SYMBOL(sk_filter_trim_cap);
129 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
131 return skb_get_poff(skb);
134 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
138 if (skb_is_nonlinear(skb))
141 if (skb->len < sizeof(struct nlattr))
144 if (a > skb->len - sizeof(struct nlattr))
147 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
149 return (void *) nla - (void *) skb->data;
154 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
158 if (skb_is_nonlinear(skb))
161 if (skb->len < sizeof(struct nlattr))
164 if (a > skb->len - sizeof(struct nlattr))
167 nla = (struct nlattr *) &skb->data[a];
168 if (nla->nla_len > skb->len - a)
171 nla = nla_find_nested(nla, x);
173 return (void *) nla - (void *) skb->data;
178 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
179 data, int, headlen, int, offset)
182 const int len = sizeof(tmp);
185 if (headlen - offset >= len)
186 return *(u8 *)(data + offset);
187 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
190 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
198 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
201 return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
205 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
206 data, int, headlen, int, offset)
209 const int len = sizeof(tmp);
212 if (headlen - offset >= len)
213 return get_unaligned_be16(data + offset);
214 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
215 return be16_to_cpu(tmp);
217 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
219 return get_unaligned_be16(ptr);
225 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
228 return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
232 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
233 data, int, headlen, int, offset)
236 const int len = sizeof(tmp);
238 if (likely(offset >= 0)) {
239 if (headlen - offset >= len)
240 return get_unaligned_be32(data + offset);
241 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
242 return be32_to_cpu(tmp);
244 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
246 return get_unaligned_be32(ptr);
252 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
255 return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
259 BPF_CALL_0(bpf_get_raw_cpu_id)
261 return raw_smp_processor_id();
264 static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = {
265 .func = bpf_get_raw_cpu_id,
267 .ret_type = RET_INTEGER,
270 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
271 struct bpf_insn *insn_buf)
273 struct bpf_insn *insn = insn_buf;
277 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4);
279 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
280 offsetof(struct sk_buff, mark));
284 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
285 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
286 #ifdef __BIG_ENDIAN_BITFIELD
287 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
292 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2);
294 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
295 offsetof(struct sk_buff, queue_mapping));
298 case SKF_AD_VLAN_TAG:
299 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2);
301 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
302 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
303 offsetof(struct sk_buff, vlan_tci));
305 case SKF_AD_VLAN_TAG_PRESENT:
306 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_VLAN_PRESENT_OFFSET());
307 if (PKT_VLAN_PRESENT_BIT)
308 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, PKT_VLAN_PRESENT_BIT);
309 if (PKT_VLAN_PRESENT_BIT < 7)
310 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
314 return insn - insn_buf;
317 static bool convert_bpf_extensions(struct sock_filter *fp,
318 struct bpf_insn **insnp)
320 struct bpf_insn *insn = *insnp;
324 case SKF_AD_OFF + SKF_AD_PROTOCOL:
325 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2);
327 /* A = *(u16 *) (CTX + offsetof(protocol)) */
328 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
329 offsetof(struct sk_buff, protocol));
330 /* A = ntohs(A) [emitting a nop or swap16] */
331 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
334 case SKF_AD_OFF + SKF_AD_PKTTYPE:
335 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
339 case SKF_AD_OFF + SKF_AD_IFINDEX:
340 case SKF_AD_OFF + SKF_AD_HATYPE:
341 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4);
342 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2);
344 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
345 BPF_REG_TMP, BPF_REG_CTX,
346 offsetof(struct sk_buff, dev));
347 /* if (tmp != 0) goto pc + 1 */
348 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
349 *insn++ = BPF_EXIT_INSN();
350 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
351 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
352 offsetof(struct net_device, ifindex));
354 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
355 offsetof(struct net_device, type));
358 case SKF_AD_OFF + SKF_AD_MARK:
359 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
363 case SKF_AD_OFF + SKF_AD_RXHASH:
364 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4);
366 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
367 offsetof(struct sk_buff, hash));
370 case SKF_AD_OFF + SKF_AD_QUEUE:
371 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
375 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
376 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
377 BPF_REG_A, BPF_REG_CTX, insn);
381 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
382 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
383 BPF_REG_A, BPF_REG_CTX, insn);
387 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
388 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2);
390 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
391 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
392 offsetof(struct sk_buff, vlan_proto));
393 /* A = ntohs(A) [emitting a nop or swap16] */
394 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
397 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
398 case SKF_AD_OFF + SKF_AD_NLATTR:
399 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
400 case SKF_AD_OFF + SKF_AD_CPU:
401 case SKF_AD_OFF + SKF_AD_RANDOM:
403 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
405 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
407 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
408 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
410 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
411 *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
413 case SKF_AD_OFF + SKF_AD_NLATTR:
414 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
416 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
417 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
419 case SKF_AD_OFF + SKF_AD_CPU:
420 *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
422 case SKF_AD_OFF + SKF_AD_RANDOM:
423 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
424 bpf_user_rnd_init_once();
429 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
431 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
435 /* This is just a dummy call to avoid letting the compiler
436 * evict __bpf_call_base() as an optimization. Placed here
437 * where no-one bothers.
439 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
447 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
449 const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
450 int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
451 bool endian = BPF_SIZE(fp->code) == BPF_H ||
452 BPF_SIZE(fp->code) == BPF_W;
453 bool indirect = BPF_MODE(fp->code) == BPF_IND;
454 const int ip_align = NET_IP_ALIGN;
455 struct bpf_insn *insn = *insnp;
459 ((unaligned_ok && offset >= 0) ||
460 (!unaligned_ok && offset >= 0 &&
461 offset + ip_align >= 0 &&
462 offset + ip_align % size == 0))) {
463 bool ldx_off_ok = offset <= S16_MAX;
465 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
467 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
468 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
469 size, 2 + endian + (!ldx_off_ok * 2));
471 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
474 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
475 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
476 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
480 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
481 *insn++ = BPF_JMP_A(8);
484 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
485 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
486 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
488 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
490 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
492 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
495 switch (BPF_SIZE(fp->code)) {
497 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
500 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
503 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
509 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
510 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
511 *insn = BPF_EXIT_INSN();
518 * bpf_convert_filter - convert filter program
519 * @prog: the user passed filter program
520 * @len: the length of the user passed filter program
521 * @new_prog: allocated 'struct bpf_prog' or NULL
522 * @new_len: pointer to store length of converted program
523 * @seen_ld_abs: bool whether we've seen ld_abs/ind
525 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
526 * style extended BPF (eBPF).
527 * Conversion workflow:
529 * 1) First pass for calculating the new program length:
530 * bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
532 * 2) 2nd pass to remap in two passes: 1st pass finds new
533 * jump offsets, 2nd pass remapping:
534 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
536 static int bpf_convert_filter(struct sock_filter *prog, int len,
537 struct bpf_prog *new_prog, int *new_len,
540 int new_flen = 0, pass = 0, target, i, stack_off;
541 struct bpf_insn *new_insn, *first_insn = NULL;
542 struct sock_filter *fp;
546 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
547 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
549 if (len <= 0 || len > BPF_MAXINSNS)
553 first_insn = new_prog->insnsi;
554 addrs = kcalloc(len, sizeof(*addrs),
555 GFP_KERNEL | __GFP_NOWARN);
561 new_insn = first_insn;
564 /* Classic BPF related prologue emission. */
566 /* Classic BPF expects A and X to be reset first. These need
567 * to be guaranteed to be the first two instructions.
569 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
570 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
572 /* All programs must keep CTX in callee saved BPF_REG_CTX.
573 * In eBPF case it's done by the compiler, here we need to
574 * do this ourself. Initial CTX is present in BPF_REG_ARG1.
576 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
578 /* For packet access in classic BPF, cache skb->data
579 * in callee-saved BPF R8 and skb->len - skb->data_len
580 * (headlen) in BPF R9. Since classic BPF is read-only
581 * on CTX, we only need to cache it once.
583 *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
584 BPF_REG_D, BPF_REG_CTX,
585 offsetof(struct sk_buff, data));
586 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
587 offsetof(struct sk_buff, len));
588 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
589 offsetof(struct sk_buff, data_len));
590 *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
596 for (i = 0; i < len; fp++, i++) {
597 struct bpf_insn tmp_insns[32] = { };
598 struct bpf_insn *insn = tmp_insns;
601 addrs[i] = new_insn - first_insn;
604 /* All arithmetic insns and skb loads map as-is. */
605 case BPF_ALU | BPF_ADD | BPF_X:
606 case BPF_ALU | BPF_ADD | BPF_K:
607 case BPF_ALU | BPF_SUB | BPF_X:
608 case BPF_ALU | BPF_SUB | BPF_K:
609 case BPF_ALU | BPF_AND | BPF_X:
610 case BPF_ALU | BPF_AND | BPF_K:
611 case BPF_ALU | BPF_OR | BPF_X:
612 case BPF_ALU | BPF_OR | BPF_K:
613 case BPF_ALU | BPF_LSH | BPF_X:
614 case BPF_ALU | BPF_LSH | BPF_K:
615 case BPF_ALU | BPF_RSH | BPF_X:
616 case BPF_ALU | BPF_RSH | BPF_K:
617 case BPF_ALU | BPF_XOR | BPF_X:
618 case BPF_ALU | BPF_XOR | BPF_K:
619 case BPF_ALU | BPF_MUL | BPF_X:
620 case BPF_ALU | BPF_MUL | BPF_K:
621 case BPF_ALU | BPF_DIV | BPF_X:
622 case BPF_ALU | BPF_DIV | BPF_K:
623 case BPF_ALU | BPF_MOD | BPF_X:
624 case BPF_ALU | BPF_MOD | BPF_K:
625 case BPF_ALU | BPF_NEG:
626 case BPF_LD | BPF_ABS | BPF_W:
627 case BPF_LD | BPF_ABS | BPF_H:
628 case BPF_LD | BPF_ABS | BPF_B:
629 case BPF_LD | BPF_IND | BPF_W:
630 case BPF_LD | BPF_IND | BPF_H:
631 case BPF_LD | BPF_IND | BPF_B:
632 /* Check for overloaded BPF extension and
633 * directly convert it if found, otherwise
634 * just move on with mapping.
636 if (BPF_CLASS(fp->code) == BPF_LD &&
637 BPF_MODE(fp->code) == BPF_ABS &&
638 convert_bpf_extensions(fp, &insn))
640 if (BPF_CLASS(fp->code) == BPF_LD &&
641 convert_bpf_ld_abs(fp, &insn)) {
646 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
647 fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
648 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
649 /* Error with exception code on div/mod by 0.
650 * For cBPF programs, this was always return 0.
652 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
653 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
654 *insn++ = BPF_EXIT_INSN();
657 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
660 /* Jump transformation cannot use BPF block macros
661 * everywhere as offset calculation and target updates
662 * require a bit more work than the rest, i.e. jump
663 * opcodes map as-is, but offsets need adjustment.
666 #define BPF_EMIT_JMP \
668 const s32 off_min = S16_MIN, off_max = S16_MAX; \
671 if (target >= len || target < 0) \
673 off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
674 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
675 off -= insn - tmp_insns; \
676 /* Reject anything not fitting into insn->off. */ \
677 if (off < off_min || off > off_max) \
682 case BPF_JMP | BPF_JA:
683 target = i + fp->k + 1;
684 insn->code = fp->code;
688 case BPF_JMP | BPF_JEQ | BPF_K:
689 case BPF_JMP | BPF_JEQ | BPF_X:
690 case BPF_JMP | BPF_JSET | BPF_K:
691 case BPF_JMP | BPF_JSET | BPF_X:
692 case BPF_JMP | BPF_JGT | BPF_K:
693 case BPF_JMP | BPF_JGT | BPF_X:
694 case BPF_JMP | BPF_JGE | BPF_K:
695 case BPF_JMP | BPF_JGE | BPF_X:
696 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
697 /* BPF immediates are signed, zero extend
698 * immediate into tmp register and use it
701 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
703 insn->dst_reg = BPF_REG_A;
704 insn->src_reg = BPF_REG_TMP;
707 insn->dst_reg = BPF_REG_A;
709 bpf_src = BPF_SRC(fp->code);
710 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
713 /* Common case where 'jump_false' is next insn. */
715 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
716 target = i + fp->jt + 1;
721 /* Convert some jumps when 'jump_true' is next insn. */
723 switch (BPF_OP(fp->code)) {
725 insn->code = BPF_JMP | BPF_JNE | bpf_src;
728 insn->code = BPF_JMP | BPF_JLE | bpf_src;
731 insn->code = BPF_JMP | BPF_JLT | bpf_src;
737 target = i + fp->jf + 1;
742 /* Other jumps are mapped into two insns: Jxx and JA. */
743 target = i + fp->jt + 1;
744 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
748 insn->code = BPF_JMP | BPF_JA;
749 target = i + fp->jf + 1;
753 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
754 case BPF_LDX | BPF_MSH | BPF_B: {
755 struct sock_filter tmp = {
756 .code = BPF_LD | BPF_ABS | BPF_B,
763 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
764 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
765 convert_bpf_ld_abs(&tmp, &insn);
768 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
770 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
772 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
774 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
776 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
779 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
780 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
782 case BPF_RET | BPF_A:
783 case BPF_RET | BPF_K:
784 if (BPF_RVAL(fp->code) == BPF_K)
785 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
787 *insn = BPF_EXIT_INSN();
790 /* Store to stack. */
793 stack_off = fp->k * 4 + 4;
794 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
795 BPF_ST ? BPF_REG_A : BPF_REG_X,
797 /* check_load_and_stores() verifies that classic BPF can
798 * load from stack only after write, so tracking
799 * stack_depth for ST|STX insns is enough
801 if (new_prog && new_prog->aux->stack_depth < stack_off)
802 new_prog->aux->stack_depth = stack_off;
805 /* Load from stack. */
806 case BPF_LD | BPF_MEM:
807 case BPF_LDX | BPF_MEM:
808 stack_off = fp->k * 4 + 4;
809 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
810 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
815 case BPF_LD | BPF_IMM:
816 case BPF_LDX | BPF_IMM:
817 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
818 BPF_REG_A : BPF_REG_X, fp->k);
822 case BPF_MISC | BPF_TAX:
823 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
827 case BPF_MISC | BPF_TXA:
828 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
831 /* A = skb->len or X = skb->len */
832 case BPF_LD | BPF_W | BPF_LEN:
833 case BPF_LDX | BPF_W | BPF_LEN:
834 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
835 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
836 offsetof(struct sk_buff, len));
839 /* Access seccomp_data fields. */
840 case BPF_LDX | BPF_ABS | BPF_W:
841 /* A = *(u32 *) (ctx + K) */
842 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
845 /* Unknown instruction. */
852 memcpy(new_insn, tmp_insns,
853 sizeof(*insn) * (insn - tmp_insns));
854 new_insn += insn - tmp_insns;
858 /* Only calculating new length. */
859 *new_len = new_insn - first_insn;
861 *new_len += 4; /* Prologue bits. */
866 if (new_flen != new_insn - first_insn) {
867 new_flen = new_insn - first_insn;
874 BUG_ON(*new_len != new_flen);
883 * As we dont want to clear mem[] array for each packet going through
884 * __bpf_prog_run(), we check that filter loaded by user never try to read
885 * a cell if not previously written, and we check all branches to be sure
886 * a malicious user doesn't try to abuse us.
888 static int check_load_and_stores(const struct sock_filter *filter, int flen)
890 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
893 BUILD_BUG_ON(BPF_MEMWORDS > 16);
895 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
899 memset(masks, 0xff, flen * sizeof(*masks));
901 for (pc = 0; pc < flen; pc++) {
902 memvalid &= masks[pc];
904 switch (filter[pc].code) {
907 memvalid |= (1 << filter[pc].k);
909 case BPF_LD | BPF_MEM:
910 case BPF_LDX | BPF_MEM:
911 if (!(memvalid & (1 << filter[pc].k))) {
916 case BPF_JMP | BPF_JA:
917 /* A jump must set masks on target */
918 masks[pc + 1 + filter[pc].k] &= memvalid;
921 case BPF_JMP | BPF_JEQ | BPF_K:
922 case BPF_JMP | BPF_JEQ | BPF_X:
923 case BPF_JMP | BPF_JGE | BPF_K:
924 case BPF_JMP | BPF_JGE | BPF_X:
925 case BPF_JMP | BPF_JGT | BPF_K:
926 case BPF_JMP | BPF_JGT | BPF_X:
927 case BPF_JMP | BPF_JSET | BPF_K:
928 case BPF_JMP | BPF_JSET | BPF_X:
929 /* A jump must set masks on targets */
930 masks[pc + 1 + filter[pc].jt] &= memvalid;
931 masks[pc + 1 + filter[pc].jf] &= memvalid;
941 static bool chk_code_allowed(u16 code_to_probe)
943 static const bool codes[] = {
944 /* 32 bit ALU operations */
945 [BPF_ALU | BPF_ADD | BPF_K] = true,
946 [BPF_ALU | BPF_ADD | BPF_X] = true,
947 [BPF_ALU | BPF_SUB | BPF_K] = true,
948 [BPF_ALU | BPF_SUB | BPF_X] = true,
949 [BPF_ALU | BPF_MUL | BPF_K] = true,
950 [BPF_ALU | BPF_MUL | BPF_X] = true,
951 [BPF_ALU | BPF_DIV | BPF_K] = true,
952 [BPF_ALU | BPF_DIV | BPF_X] = true,
953 [BPF_ALU | BPF_MOD | BPF_K] = true,
954 [BPF_ALU | BPF_MOD | BPF_X] = true,
955 [BPF_ALU | BPF_AND | BPF_K] = true,
956 [BPF_ALU | BPF_AND | BPF_X] = true,
957 [BPF_ALU | BPF_OR | BPF_K] = true,
958 [BPF_ALU | BPF_OR | BPF_X] = true,
959 [BPF_ALU | BPF_XOR | BPF_K] = true,
960 [BPF_ALU | BPF_XOR | BPF_X] = true,
961 [BPF_ALU | BPF_LSH | BPF_K] = true,
962 [BPF_ALU | BPF_LSH | BPF_X] = true,
963 [BPF_ALU | BPF_RSH | BPF_K] = true,
964 [BPF_ALU | BPF_RSH | BPF_X] = true,
965 [BPF_ALU | BPF_NEG] = true,
966 /* Load instructions */
967 [BPF_LD | BPF_W | BPF_ABS] = true,
968 [BPF_LD | BPF_H | BPF_ABS] = true,
969 [BPF_LD | BPF_B | BPF_ABS] = true,
970 [BPF_LD | BPF_W | BPF_LEN] = true,
971 [BPF_LD | BPF_W | BPF_IND] = true,
972 [BPF_LD | BPF_H | BPF_IND] = true,
973 [BPF_LD | BPF_B | BPF_IND] = true,
974 [BPF_LD | BPF_IMM] = true,
975 [BPF_LD | BPF_MEM] = true,
976 [BPF_LDX | BPF_W | BPF_LEN] = true,
977 [BPF_LDX | BPF_B | BPF_MSH] = true,
978 [BPF_LDX | BPF_IMM] = true,
979 [BPF_LDX | BPF_MEM] = true,
980 /* Store instructions */
983 /* Misc instructions */
984 [BPF_MISC | BPF_TAX] = true,
985 [BPF_MISC | BPF_TXA] = true,
986 /* Return instructions */
987 [BPF_RET | BPF_K] = true,
988 [BPF_RET | BPF_A] = true,
989 /* Jump instructions */
990 [BPF_JMP | BPF_JA] = true,
991 [BPF_JMP | BPF_JEQ | BPF_K] = true,
992 [BPF_JMP | BPF_JEQ | BPF_X] = true,
993 [BPF_JMP | BPF_JGE | BPF_K] = true,
994 [BPF_JMP | BPF_JGE | BPF_X] = true,
995 [BPF_JMP | BPF_JGT | BPF_K] = true,
996 [BPF_JMP | BPF_JGT | BPF_X] = true,
997 [BPF_JMP | BPF_JSET | BPF_K] = true,
998 [BPF_JMP | BPF_JSET | BPF_X] = true,
1001 if (code_to_probe >= ARRAY_SIZE(codes))
1004 return codes[code_to_probe];
1007 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1012 if (flen == 0 || flen > BPF_MAXINSNS)
1019 * bpf_check_classic - verify socket filter code
1020 * @filter: filter to verify
1021 * @flen: length of filter
1023 * Check the user's filter code. If we let some ugly
1024 * filter code slip through kaboom! The filter must contain
1025 * no references or jumps that are out of range, no illegal
1026 * instructions, and must end with a RET instruction.
1028 * All jumps are forward as they are not signed.
1030 * Returns 0 if the rule set is legal or -EINVAL if not.
1032 static int bpf_check_classic(const struct sock_filter *filter,
1038 /* Check the filter code now */
1039 for (pc = 0; pc < flen; pc++) {
1040 const struct sock_filter *ftest = &filter[pc];
1042 /* May we actually operate on this code? */
1043 if (!chk_code_allowed(ftest->code))
1046 /* Some instructions need special checks */
1047 switch (ftest->code) {
1048 case BPF_ALU | BPF_DIV | BPF_K:
1049 case BPF_ALU | BPF_MOD | BPF_K:
1050 /* Check for division by zero */
1054 case BPF_ALU | BPF_LSH | BPF_K:
1055 case BPF_ALU | BPF_RSH | BPF_K:
1059 case BPF_LD | BPF_MEM:
1060 case BPF_LDX | BPF_MEM:
1063 /* Check for invalid memory addresses */
1064 if (ftest->k >= BPF_MEMWORDS)
1067 case BPF_JMP | BPF_JA:
1068 /* Note, the large ftest->k might cause loops.
1069 * Compare this with conditional jumps below,
1070 * where offsets are limited. --ANK (981016)
1072 if (ftest->k >= (unsigned int)(flen - pc - 1))
1075 case BPF_JMP | BPF_JEQ | BPF_K:
1076 case BPF_JMP | BPF_JEQ | BPF_X:
1077 case BPF_JMP | BPF_JGE | BPF_K:
1078 case BPF_JMP | BPF_JGE | BPF_X:
1079 case BPF_JMP | BPF_JGT | BPF_K:
1080 case BPF_JMP | BPF_JGT | BPF_X:
1081 case BPF_JMP | BPF_JSET | BPF_K:
1082 case BPF_JMP | BPF_JSET | BPF_X:
1083 /* Both conditionals must be safe */
1084 if (pc + ftest->jt + 1 >= flen ||
1085 pc + ftest->jf + 1 >= flen)
1088 case BPF_LD | BPF_W | BPF_ABS:
1089 case BPF_LD | BPF_H | BPF_ABS:
1090 case BPF_LD | BPF_B | BPF_ABS:
1092 if (bpf_anc_helper(ftest) & BPF_ANC)
1094 /* Ancillary operation unknown or unsupported */
1095 if (anc_found == false && ftest->k >= SKF_AD_OFF)
1100 /* Last instruction must be a RET code */
1101 switch (filter[flen - 1].code) {
1102 case BPF_RET | BPF_K:
1103 case BPF_RET | BPF_A:
1104 return check_load_and_stores(filter, flen);
1110 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1111 const struct sock_fprog *fprog)
1113 unsigned int fsize = bpf_classic_proglen(fprog);
1114 struct sock_fprog_kern *fkprog;
1116 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1120 fkprog = fp->orig_prog;
1121 fkprog->len = fprog->len;
1123 fkprog->filter = kmemdup(fp->insns, fsize,
1124 GFP_KERNEL | __GFP_NOWARN);
1125 if (!fkprog->filter) {
1126 kfree(fp->orig_prog);
1133 static void bpf_release_orig_filter(struct bpf_prog *fp)
1135 struct sock_fprog_kern *fprog = fp->orig_prog;
1138 kfree(fprog->filter);
1143 static void __bpf_prog_release(struct bpf_prog *prog)
1145 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1148 bpf_release_orig_filter(prog);
1149 bpf_prog_free(prog);
1153 static void __sk_filter_release(struct sk_filter *fp)
1155 __bpf_prog_release(fp->prog);
1160 * sk_filter_release_rcu - Release a socket filter by rcu_head
1161 * @rcu: rcu_head that contains the sk_filter to free
1163 static void sk_filter_release_rcu(struct rcu_head *rcu)
1165 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1167 __sk_filter_release(fp);
1171 * sk_filter_release - release a socket filter
1172 * @fp: filter to remove
1174 * Remove a filter from a socket and release its resources.
1176 static void sk_filter_release(struct sk_filter *fp)
1178 if (refcount_dec_and_test(&fp->refcnt))
1179 call_rcu(&fp->rcu, sk_filter_release_rcu);
1182 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1184 u32 filter_size = bpf_prog_size(fp->prog->len);
1186 atomic_sub(filter_size, &sk->sk_omem_alloc);
1187 sk_filter_release(fp);
1190 /* try to charge the socket memory if there is space available
1191 * return true on success
1193 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1195 u32 filter_size = bpf_prog_size(fp->prog->len);
1197 /* same check as in sock_kmalloc() */
1198 if (filter_size <= sysctl_optmem_max &&
1199 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
1200 atomic_add(filter_size, &sk->sk_omem_alloc);
1206 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1208 if (!refcount_inc_not_zero(&fp->refcnt))
1211 if (!__sk_filter_charge(sk, fp)) {
1212 sk_filter_release(fp);
1218 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1220 struct sock_filter *old_prog;
1221 struct bpf_prog *old_fp;
1222 int err, new_len, old_len = fp->len;
1223 bool seen_ld_abs = false;
1225 /* We are free to overwrite insns et al right here as it
1226 * won't be used at this point in time anymore internally
1227 * after the migration to the internal BPF instruction
1230 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1231 sizeof(struct bpf_insn));
1233 /* Conversion cannot happen on overlapping memory areas,
1234 * so we need to keep the user BPF around until the 2nd
1235 * pass. At this time, the user BPF is stored in fp->insns.
1237 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1238 GFP_KERNEL | __GFP_NOWARN);
1244 /* 1st pass: calculate the new program length. */
1245 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1250 /* Expand fp for appending the new filter representation. */
1252 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1254 /* The old_fp is still around in case we couldn't
1255 * allocate new memory, so uncharge on that one.
1264 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1265 err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1268 /* 2nd bpf_convert_filter() can fail only if it fails
1269 * to allocate memory, remapping must succeed. Note,
1270 * that at this time old_fp has already been released
1275 fp = bpf_prog_select_runtime(fp, &err);
1285 __bpf_prog_release(fp);
1286 return ERR_PTR(err);
1289 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1290 bpf_aux_classic_check_t trans)
1294 fp->bpf_func = NULL;
1297 err = bpf_check_classic(fp->insns, fp->len);
1299 __bpf_prog_release(fp);
1300 return ERR_PTR(err);
1303 /* There might be additional checks and transformations
1304 * needed on classic filters, f.e. in case of seccomp.
1307 err = trans(fp->insns, fp->len);
1309 __bpf_prog_release(fp);
1310 return ERR_PTR(err);
1314 /* Probe if we can JIT compile the filter and if so, do
1315 * the compilation of the filter.
1317 bpf_jit_compile(fp);
1319 /* JIT compiler couldn't process this filter, so do the
1320 * internal BPF translation for the optimized interpreter.
1323 fp = bpf_migrate_filter(fp);
1329 * bpf_prog_create - create an unattached filter
1330 * @pfp: the unattached filter that is created
1331 * @fprog: the filter program
1333 * Create a filter independent of any socket. We first run some
1334 * sanity checks on it to make sure it does not explode on us later.
1335 * If an error occurs or there is insufficient memory for the filter
1336 * a negative errno code is returned. On success the return is zero.
1338 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1340 unsigned int fsize = bpf_classic_proglen(fprog);
1341 struct bpf_prog *fp;
1343 /* Make sure new filter is there and in the right amounts. */
1344 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1347 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1351 memcpy(fp->insns, fprog->filter, fsize);
1353 fp->len = fprog->len;
1354 /* Since unattached filters are not copied back to user
1355 * space through sk_get_filter(), we do not need to hold
1356 * a copy here, and can spare us the work.
1358 fp->orig_prog = NULL;
1360 /* bpf_prepare_filter() already takes care of freeing
1361 * memory in case something goes wrong.
1363 fp = bpf_prepare_filter(fp, NULL);
1370 EXPORT_SYMBOL_GPL(bpf_prog_create);
1373 * bpf_prog_create_from_user - create an unattached filter from user buffer
1374 * @pfp: the unattached filter that is created
1375 * @fprog: the filter program
1376 * @trans: post-classic verifier transformation handler
1377 * @save_orig: save classic BPF program
1379 * This function effectively does the same as bpf_prog_create(), only
1380 * that it builds up its insns buffer from user space provided buffer.
1381 * It also allows for passing a bpf_aux_classic_check_t handler.
1383 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1384 bpf_aux_classic_check_t trans, bool save_orig)
1386 unsigned int fsize = bpf_classic_proglen(fprog);
1387 struct bpf_prog *fp;
1390 /* Make sure new filter is there and in the right amounts. */
1391 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1394 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1398 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1399 __bpf_prog_free(fp);
1403 fp->len = fprog->len;
1404 fp->orig_prog = NULL;
1407 err = bpf_prog_store_orig_filter(fp, fprog);
1409 __bpf_prog_free(fp);
1414 /* bpf_prepare_filter() already takes care of freeing
1415 * memory in case something goes wrong.
1417 fp = bpf_prepare_filter(fp, trans);
1424 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1426 void bpf_prog_destroy(struct bpf_prog *fp)
1428 __bpf_prog_release(fp);
1430 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1432 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1434 struct sk_filter *fp, *old_fp;
1436 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1442 if (!__sk_filter_charge(sk, fp)) {
1446 refcount_set(&fp->refcnt, 1);
1448 old_fp = rcu_dereference_protected(sk->sk_filter,
1449 lockdep_sock_is_held(sk));
1450 rcu_assign_pointer(sk->sk_filter, fp);
1453 sk_filter_uncharge(sk, old_fp);
1459 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1461 unsigned int fsize = bpf_classic_proglen(fprog);
1462 struct bpf_prog *prog;
1465 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1466 return ERR_PTR(-EPERM);
1468 /* Make sure new filter is there and in the right amounts. */
1469 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1470 return ERR_PTR(-EINVAL);
1472 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1474 return ERR_PTR(-ENOMEM);
1476 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1477 __bpf_prog_free(prog);
1478 return ERR_PTR(-EFAULT);
1481 prog->len = fprog->len;
1483 err = bpf_prog_store_orig_filter(prog, fprog);
1485 __bpf_prog_free(prog);
1486 return ERR_PTR(-ENOMEM);
1489 /* bpf_prepare_filter() already takes care of freeing
1490 * memory in case something goes wrong.
1492 return bpf_prepare_filter(prog, NULL);
1496 * sk_attach_filter - attach a socket filter
1497 * @fprog: the filter program
1498 * @sk: the socket to use
1500 * Attach the user's filter code. We first run some sanity checks on
1501 * it to make sure it does not explode on us later. If an error
1502 * occurs or there is insufficient memory for the filter a negative
1503 * errno code is returned. On success the return is zero.
1505 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1507 struct bpf_prog *prog = __get_filter(fprog, sk);
1511 return PTR_ERR(prog);
1513 err = __sk_attach_prog(prog, sk);
1515 __bpf_prog_release(prog);
1521 EXPORT_SYMBOL_GPL(sk_attach_filter);
1523 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1525 struct bpf_prog *prog = __get_filter(fprog, sk);
1529 return PTR_ERR(prog);
1531 if (bpf_prog_size(prog->len) > sysctl_optmem_max)
1534 err = reuseport_attach_prog(sk, prog);
1537 __bpf_prog_release(prog);
1542 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1544 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1545 return ERR_PTR(-EPERM);
1547 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1550 int sk_attach_bpf(u32 ufd, struct sock *sk)
1552 struct bpf_prog *prog = __get_bpf(ufd, sk);
1556 return PTR_ERR(prog);
1558 err = __sk_attach_prog(prog, sk);
1567 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1569 struct bpf_prog *prog;
1572 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1575 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1576 if (PTR_ERR(prog) == -EINVAL)
1577 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1579 return PTR_ERR(prog);
1581 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1582 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1583 * bpf prog (e.g. sockmap). It depends on the
1584 * limitation imposed by bpf_prog_load().
1585 * Hence, sysctl_optmem_max is not checked.
1587 if ((sk->sk_type != SOCK_STREAM &&
1588 sk->sk_type != SOCK_DGRAM) ||
1589 (sk->sk_protocol != IPPROTO_UDP &&
1590 sk->sk_protocol != IPPROTO_TCP) ||
1591 (sk->sk_family != AF_INET &&
1592 sk->sk_family != AF_INET6)) {
1597 /* BPF_PROG_TYPE_SOCKET_FILTER */
1598 if (bpf_prog_size(prog->len) > sysctl_optmem_max) {
1604 err = reuseport_attach_prog(sk, prog);
1612 void sk_reuseport_prog_free(struct bpf_prog *prog)
1617 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1620 bpf_prog_destroy(prog);
1623 struct bpf_scratchpad {
1625 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1626 u8 buff[MAX_BPF_STACK];
1630 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1632 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1633 unsigned int write_len)
1635 return skb_ensure_writable(skb, write_len);
1638 static inline int bpf_try_make_writable(struct sk_buff *skb,
1639 unsigned int write_len)
1641 int err = __bpf_try_make_writable(skb, write_len);
1643 bpf_compute_data_pointers(skb);
1647 static int bpf_try_make_head_writable(struct sk_buff *skb)
1649 return bpf_try_make_writable(skb, skb_headlen(skb));
1652 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1654 if (skb_at_tc_ingress(skb))
1655 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1658 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1660 if (skb_at_tc_ingress(skb))
1661 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1664 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1665 const void *, from, u32, len, u64, flags)
1669 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1671 if (unlikely(offset > 0xffff))
1673 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1676 ptr = skb->data + offset;
1677 if (flags & BPF_F_RECOMPUTE_CSUM)
1678 __skb_postpull_rcsum(skb, ptr, len, offset);
1680 memcpy(ptr, from, len);
1682 if (flags & BPF_F_RECOMPUTE_CSUM)
1683 __skb_postpush_rcsum(skb, ptr, len, offset);
1684 if (flags & BPF_F_INVALIDATE_HASH)
1685 skb_clear_hash(skb);
1690 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1691 .func = bpf_skb_store_bytes,
1693 .ret_type = RET_INTEGER,
1694 .arg1_type = ARG_PTR_TO_CTX,
1695 .arg2_type = ARG_ANYTHING,
1696 .arg3_type = ARG_PTR_TO_MEM,
1697 .arg4_type = ARG_CONST_SIZE,
1698 .arg5_type = ARG_ANYTHING,
1701 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1702 void *, to, u32, len)
1706 if (unlikely(offset > 0xffff))
1709 ptr = skb_header_pointer(skb, offset, len, to);
1713 memcpy(to, ptr, len);
1721 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1722 .func = bpf_skb_load_bytes,
1724 .ret_type = RET_INTEGER,
1725 .arg1_type = ARG_PTR_TO_CTX,
1726 .arg2_type = ARG_ANYTHING,
1727 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1728 .arg4_type = ARG_CONST_SIZE,
1731 BPF_CALL_4(bpf_flow_dissector_load_bytes,
1732 const struct bpf_flow_dissector *, ctx, u32, offset,
1733 void *, to, u32, len)
1737 if (unlikely(offset > 0xffff))
1740 if (unlikely(!ctx->skb))
1743 ptr = skb_header_pointer(ctx->skb, offset, len, to);
1747 memcpy(to, ptr, len);
1755 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1756 .func = bpf_flow_dissector_load_bytes,
1758 .ret_type = RET_INTEGER,
1759 .arg1_type = ARG_PTR_TO_CTX,
1760 .arg2_type = ARG_ANYTHING,
1761 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1762 .arg4_type = ARG_CONST_SIZE,
1765 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1766 u32, offset, void *, to, u32, len, u32, start_header)
1768 u8 *end = skb_tail_pointer(skb);
1769 u8 *net = skb_network_header(skb);
1770 u8 *mac = skb_mac_header(skb);
1773 if (unlikely(offset > 0xffff || len > (end - mac)))
1776 switch (start_header) {
1777 case BPF_HDR_START_MAC:
1780 case BPF_HDR_START_NET:
1787 if (likely(ptr >= mac && ptr + len <= end)) {
1788 memcpy(to, ptr, len);
1797 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1798 .func = bpf_skb_load_bytes_relative,
1800 .ret_type = RET_INTEGER,
1801 .arg1_type = ARG_PTR_TO_CTX,
1802 .arg2_type = ARG_ANYTHING,
1803 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1804 .arg4_type = ARG_CONST_SIZE,
1805 .arg5_type = ARG_ANYTHING,
1808 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1810 /* Idea is the following: should the needed direct read/write
1811 * test fail during runtime, we can pull in more data and redo
1812 * again, since implicitly, we invalidate previous checks here.
1814 * Or, since we know how much we need to make read/writeable,
1815 * this can be done once at the program beginning for direct
1816 * access case. By this we overcome limitations of only current
1817 * headroom being accessible.
1819 return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1822 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1823 .func = bpf_skb_pull_data,
1825 .ret_type = RET_INTEGER,
1826 .arg1_type = ARG_PTR_TO_CTX,
1827 .arg2_type = ARG_ANYTHING,
1830 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1832 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1835 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1836 .func = bpf_sk_fullsock,
1838 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
1839 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
1842 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1843 unsigned int write_len)
1845 int err = __bpf_try_make_writable(skb, write_len);
1847 bpf_compute_data_end_sk_skb(skb);
1851 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1853 /* Idea is the following: should the needed direct read/write
1854 * test fail during runtime, we can pull in more data and redo
1855 * again, since implicitly, we invalidate previous checks here.
1857 * Or, since we know how much we need to make read/writeable,
1858 * this can be done once at the program beginning for direct
1859 * access case. By this we overcome limitations of only current
1860 * headroom being accessible.
1862 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1865 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1866 .func = sk_skb_pull_data,
1868 .ret_type = RET_INTEGER,
1869 .arg1_type = ARG_PTR_TO_CTX,
1870 .arg2_type = ARG_ANYTHING,
1873 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1874 u64, from, u64, to, u64, flags)
1878 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1880 if (unlikely(offset > 0xffff || offset & 1))
1882 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1885 ptr = (__sum16 *)(skb->data + offset);
1886 switch (flags & BPF_F_HDR_FIELD_MASK) {
1888 if (unlikely(from != 0))
1891 csum_replace_by_diff(ptr, to);
1894 csum_replace2(ptr, from, to);
1897 csum_replace4(ptr, from, to);
1906 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1907 .func = bpf_l3_csum_replace,
1909 .ret_type = RET_INTEGER,
1910 .arg1_type = ARG_PTR_TO_CTX,
1911 .arg2_type = ARG_ANYTHING,
1912 .arg3_type = ARG_ANYTHING,
1913 .arg4_type = ARG_ANYTHING,
1914 .arg5_type = ARG_ANYTHING,
1917 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1918 u64, from, u64, to, u64, flags)
1920 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1921 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1922 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1925 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1926 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1928 if (unlikely(offset > 0xffff || offset & 1))
1930 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1933 ptr = (__sum16 *)(skb->data + offset);
1934 if (is_mmzero && !do_mforce && !*ptr)
1937 switch (flags & BPF_F_HDR_FIELD_MASK) {
1939 if (unlikely(from != 0))
1942 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1945 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1948 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1954 if (is_mmzero && !*ptr)
1955 *ptr = CSUM_MANGLED_0;
1959 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1960 .func = bpf_l4_csum_replace,
1962 .ret_type = RET_INTEGER,
1963 .arg1_type = ARG_PTR_TO_CTX,
1964 .arg2_type = ARG_ANYTHING,
1965 .arg3_type = ARG_ANYTHING,
1966 .arg4_type = ARG_ANYTHING,
1967 .arg5_type = ARG_ANYTHING,
1970 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1971 __be32 *, to, u32, to_size, __wsum, seed)
1973 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1974 u32 diff_size = from_size + to_size;
1977 /* This is quite flexible, some examples:
1979 * from_size == 0, to_size > 0, seed := csum --> pushing data
1980 * from_size > 0, to_size == 0, seed := csum --> pulling data
1981 * from_size > 0, to_size > 0, seed := 0 --> diffing data
1983 * Even for diffing, from_size and to_size don't need to be equal.
1985 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
1986 diff_size > sizeof(sp->diff)))
1989 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
1990 sp->diff[j] = ~from[i];
1991 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
1992 sp->diff[j] = to[i];
1994 return csum_partial(sp->diff, diff_size, seed);
1997 static const struct bpf_func_proto bpf_csum_diff_proto = {
1998 .func = bpf_csum_diff,
2001 .ret_type = RET_INTEGER,
2002 .arg1_type = ARG_PTR_TO_MEM_OR_NULL,
2003 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
2004 .arg3_type = ARG_PTR_TO_MEM_OR_NULL,
2005 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
2006 .arg5_type = ARG_ANYTHING,
2009 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2011 /* The interface is to be used in combination with bpf_csum_diff()
2012 * for direct packet writes. csum rotation for alignment as well
2013 * as emulating csum_sub() can be done from the eBPF program.
2015 if (skb->ip_summed == CHECKSUM_COMPLETE)
2016 return (skb->csum = csum_add(skb->csum, csum));
2021 static const struct bpf_func_proto bpf_csum_update_proto = {
2022 .func = bpf_csum_update,
2024 .ret_type = RET_INTEGER,
2025 .arg1_type = ARG_PTR_TO_CTX,
2026 .arg2_type = ARG_ANYTHING,
2029 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2031 return dev_forward_skb(dev, skb);
2034 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2035 struct sk_buff *skb)
2037 int ret = ____dev_forward_skb(dev, skb);
2041 ret = netif_rx(skb);
2047 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2051 if (dev_xmit_recursion()) {
2052 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2060 dev_xmit_recursion_inc();
2061 ret = dev_queue_xmit(skb);
2062 dev_xmit_recursion_dec();
2067 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2070 unsigned int mlen = skb_network_offset(skb);
2073 __skb_pull(skb, mlen);
2075 /* At ingress, the mac header has already been pulled once.
2076 * At egress, skb_pospull_rcsum has to be done in case that
2077 * the skb is originated from ingress (i.e. a forwarded skb)
2078 * to ensure that rcsum starts at net header.
2080 if (!skb_at_tc_ingress(skb))
2081 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2083 skb_pop_mac_header(skb);
2084 skb_reset_mac_len(skb);
2085 return flags & BPF_F_INGRESS ?
2086 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2089 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2092 /* Verify that a link layer header is carried */
2093 if (unlikely(skb->mac_header >= skb->network_header)) {
2098 bpf_push_mac_rcsum(skb);
2099 return flags & BPF_F_INGRESS ?
2100 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2103 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2106 if (dev_is_mac_header_xmit(dev))
2107 return __bpf_redirect_common(skb, dev, flags);
2109 return __bpf_redirect_no_mac(skb, dev, flags);
2112 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2114 struct net_device *dev;
2115 struct sk_buff *clone;
2118 if (unlikely(flags & ~(BPF_F_INGRESS)))
2121 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2125 clone = skb_clone(skb, GFP_ATOMIC);
2126 if (unlikely(!clone))
2129 /* For direct write, we need to keep the invariant that the skbs
2130 * we're dealing with need to be uncloned. Should uncloning fail
2131 * here, we need to free the just generated clone to unclone once
2134 ret = bpf_try_make_head_writable(skb);
2135 if (unlikely(ret)) {
2140 return __bpf_redirect(clone, dev, flags);
2143 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2144 .func = bpf_clone_redirect,
2146 .ret_type = RET_INTEGER,
2147 .arg1_type = ARG_PTR_TO_CTX,
2148 .arg2_type = ARG_ANYTHING,
2149 .arg3_type = ARG_ANYTHING,
2152 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2153 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2155 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2157 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2159 if (unlikely(flags & ~(BPF_F_INGRESS)))
2163 ri->tgt_index = ifindex;
2165 return TC_ACT_REDIRECT;
2168 int skb_do_redirect(struct sk_buff *skb)
2170 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2171 struct net_device *dev;
2173 dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->tgt_index);
2175 if (unlikely(!dev)) {
2180 return __bpf_redirect(skb, dev, ri->flags);
2183 static const struct bpf_func_proto bpf_redirect_proto = {
2184 .func = bpf_redirect,
2186 .ret_type = RET_INTEGER,
2187 .arg1_type = ARG_ANYTHING,
2188 .arg2_type = ARG_ANYTHING,
2191 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2193 msg->apply_bytes = bytes;
2197 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2198 .func = bpf_msg_apply_bytes,
2200 .ret_type = RET_INTEGER,
2201 .arg1_type = ARG_PTR_TO_CTX,
2202 .arg2_type = ARG_ANYTHING,
2205 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2207 msg->cork_bytes = bytes;
2211 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2212 .func = bpf_msg_cork_bytes,
2214 .ret_type = RET_INTEGER,
2215 .arg1_type = ARG_PTR_TO_CTX,
2216 .arg2_type = ARG_ANYTHING,
2219 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2220 u32, end, u64, flags)
2222 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2223 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2224 struct scatterlist *sge;
2225 u8 *raw, *to, *from;
2228 if (unlikely(flags || end <= start))
2231 /* First find the starting scatterlist element */
2235 len = sk_msg_elem(msg, i)->length;
2236 if (start < offset + len)
2238 sk_msg_iter_var_next(i);
2239 } while (i != msg->sg.end);
2241 if (unlikely(start >= offset + len))
2245 /* The start may point into the sg element so we need to also
2246 * account for the headroom.
2248 bytes_sg_total = start - offset + bytes;
2249 if (!test_bit(i, &msg->sg.copy) && bytes_sg_total <= len)
2252 /* At this point we need to linearize multiple scatterlist
2253 * elements or a single shared page. Either way we need to
2254 * copy into a linear buffer exclusively owned by BPF. Then
2255 * place the buffer in the scatterlist and fixup the original
2256 * entries by removing the entries now in the linear buffer
2257 * and shifting the remaining entries. For now we do not try
2258 * to copy partial entries to avoid complexity of running out
2259 * of sg_entry slots. The downside is reading a single byte
2260 * will copy the entire sg entry.
2263 copy += sk_msg_elem(msg, i)->length;
2264 sk_msg_iter_var_next(i);
2265 if (bytes_sg_total <= copy)
2267 } while (i != msg->sg.end);
2270 if (unlikely(bytes_sg_total > copy))
2273 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2275 if (unlikely(!page))
2278 raw = page_address(page);
2281 sge = sk_msg_elem(msg, i);
2282 from = sg_virt(sge);
2286 memcpy(to, from, len);
2289 put_page(sg_page(sge));
2291 sk_msg_iter_var_next(i);
2292 } while (i != last_sge);
2294 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2296 /* To repair sg ring we need to shift entries. If we only
2297 * had a single entry though we can just replace it and
2298 * be done. Otherwise walk the ring and shift the entries.
2300 WARN_ON_ONCE(last_sge == first_sge);
2301 shift = last_sge > first_sge ?
2302 last_sge - first_sge - 1 :
2303 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2308 sk_msg_iter_var_next(i);
2312 if (i + shift >= NR_MSG_FRAG_IDS)
2313 move_from = i + shift - NR_MSG_FRAG_IDS;
2315 move_from = i + shift;
2316 if (move_from == msg->sg.end)
2319 msg->sg.data[i] = msg->sg.data[move_from];
2320 msg->sg.data[move_from].length = 0;
2321 msg->sg.data[move_from].page_link = 0;
2322 msg->sg.data[move_from].offset = 0;
2323 sk_msg_iter_var_next(i);
2326 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2327 msg->sg.end - shift + NR_MSG_FRAG_IDS :
2328 msg->sg.end - shift;
2330 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2331 msg->data_end = msg->data + bytes;
2335 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2336 .func = bpf_msg_pull_data,
2338 .ret_type = RET_INTEGER,
2339 .arg1_type = ARG_PTR_TO_CTX,
2340 .arg2_type = ARG_ANYTHING,
2341 .arg3_type = ARG_ANYTHING,
2342 .arg4_type = ARG_ANYTHING,
2345 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2346 u32, len, u64, flags)
2348 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2349 u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2350 u8 *raw, *to, *from;
2353 if (unlikely(flags))
2356 /* First find the starting scatterlist element */
2360 l = sk_msg_elem(msg, i)->length;
2362 if (start < offset + l)
2364 sk_msg_iter_var_next(i);
2365 } while (i != msg->sg.end);
2367 if (start >= offset + l)
2370 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2372 /* If no space available will fallback to copy, we need at
2373 * least one scatterlist elem available to push data into
2374 * when start aligns to the beginning of an element or two
2375 * when it falls inside an element. We handle the start equals
2376 * offset case because its the common case for inserting a
2379 if (!space || (space == 1 && start != offset))
2380 copy = msg->sg.data[i].length;
2382 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2383 get_order(copy + len));
2384 if (unlikely(!page))
2390 raw = page_address(page);
2392 psge = sk_msg_elem(msg, i);
2393 front = start - offset;
2394 back = psge->length - front;
2395 from = sg_virt(psge);
2398 memcpy(raw, from, front);
2402 to = raw + front + len;
2404 memcpy(to, from, back);
2407 put_page(sg_page(psge));
2408 } else if (start - offset) {
2409 psge = sk_msg_elem(msg, i);
2410 rsge = sk_msg_elem_cpy(msg, i);
2412 psge->length = start - offset;
2413 rsge.length -= psge->length;
2414 rsge.offset += start;
2416 sk_msg_iter_var_next(i);
2417 sg_unmark_end(psge);
2418 sg_unmark_end(&rsge);
2419 sk_msg_iter_next(msg, end);
2422 /* Slot(s) to place newly allocated data */
2425 /* Shift one or two slots as needed */
2427 sge = sk_msg_elem_cpy(msg, i);
2429 sk_msg_iter_var_next(i);
2430 sg_unmark_end(&sge);
2431 sk_msg_iter_next(msg, end);
2433 nsge = sk_msg_elem_cpy(msg, i);
2435 sk_msg_iter_var_next(i);
2436 nnsge = sk_msg_elem_cpy(msg, i);
2439 while (i != msg->sg.end) {
2440 msg->sg.data[i] = sge;
2442 sk_msg_iter_var_next(i);
2445 nnsge = sk_msg_elem_cpy(msg, i);
2447 nsge = sk_msg_elem_cpy(msg, i);
2452 /* Place newly allocated data buffer */
2453 sk_mem_charge(msg->sk, len);
2454 msg->sg.size += len;
2455 __clear_bit(new, &msg->sg.copy);
2456 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2458 get_page(sg_page(&rsge));
2459 sk_msg_iter_var_next(new);
2460 msg->sg.data[new] = rsge;
2463 sk_msg_compute_data_pointers(msg);
2467 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2468 .func = bpf_msg_push_data,
2470 .ret_type = RET_INTEGER,
2471 .arg1_type = ARG_PTR_TO_CTX,
2472 .arg2_type = ARG_ANYTHING,
2473 .arg3_type = ARG_ANYTHING,
2474 .arg4_type = ARG_ANYTHING,
2477 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2483 sk_msg_iter_var_next(i);
2484 msg->sg.data[prev] = msg->sg.data[i];
2485 } while (i != msg->sg.end);
2487 sk_msg_iter_prev(msg, end);
2490 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2492 struct scatterlist tmp, sge;
2494 sk_msg_iter_next(msg, end);
2495 sge = sk_msg_elem_cpy(msg, i);
2496 sk_msg_iter_var_next(i);
2497 tmp = sk_msg_elem_cpy(msg, i);
2499 while (i != msg->sg.end) {
2500 msg->sg.data[i] = sge;
2501 sk_msg_iter_var_next(i);
2503 tmp = sk_msg_elem_cpy(msg, i);
2507 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2508 u32, len, u64, flags)
2510 u32 i = 0, l = 0, space, offset = 0;
2511 u64 last = start + len;
2514 if (unlikely(flags))
2517 /* First find the starting scatterlist element */
2521 l = sk_msg_elem(msg, i)->length;
2523 if (start < offset + l)
2525 sk_msg_iter_var_next(i);
2526 } while (i != msg->sg.end);
2528 /* Bounds checks: start and pop must be inside message */
2529 if (start >= offset + l || last >= msg->sg.size)
2532 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2535 /* --------------| offset
2536 * -| start |-------- len -------|
2538 * |----- a ----|-------- pop -------|----- b ----|
2539 * |______________________________________________| length
2542 * a: region at front of scatter element to save
2543 * b: region at back of scatter element to save when length > A + pop
2544 * pop: region to pop from element, same as input 'pop' here will be
2545 * decremented below per iteration.
2547 * Two top-level cases to handle when start != offset, first B is non
2548 * zero and second B is zero corresponding to when a pop includes more
2551 * Then if B is non-zero AND there is no space allocate space and
2552 * compact A, B regions into page. If there is space shift ring to
2553 * the rigth free'ing the next element in ring to place B, leaving
2554 * A untouched except to reduce length.
2556 if (start != offset) {
2557 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2559 int b = sge->length - pop - a;
2561 sk_msg_iter_var_next(i);
2563 if (pop < sge->length - a) {
2566 sk_msg_shift_right(msg, i);
2567 nsge = sk_msg_elem(msg, i);
2568 get_page(sg_page(sge));
2571 b, sge->offset + pop + a);
2573 struct page *page, *orig;
2576 page = alloc_pages(__GFP_NOWARN |
2577 __GFP_COMP | GFP_ATOMIC,
2579 if (unlikely(!page))
2583 orig = sg_page(sge);
2584 from = sg_virt(sge);
2585 to = page_address(page);
2586 memcpy(to, from, a);
2587 memcpy(to + a, from + a + pop, b);
2588 sg_set_page(sge, page, a + b, 0);
2592 } else if (pop >= sge->length - a) {
2594 pop -= (sge->length - a);
2598 /* From above the current layout _must_ be as follows,
2603 * |---- pop ---|---------------- b ------------|
2604 * |____________________________________________| length
2606 * Offset and start of the current msg elem are equal because in the
2607 * previous case we handled offset != start and either consumed the
2608 * entire element and advanced to the next element OR pop == 0.
2610 * Two cases to handle here are first pop is less than the length
2611 * leaving some remainder b above. Simply adjust the element's layout
2612 * in this case. Or pop >= length of the element so that b = 0. In this
2613 * case advance to next element decrementing pop.
2616 struct scatterlist *sge = sk_msg_elem(msg, i);
2618 if (pop < sge->length) {
2624 sk_msg_shift_left(msg, i);
2626 sk_msg_iter_var_next(i);
2629 sk_mem_uncharge(msg->sk, len - pop);
2630 msg->sg.size -= (len - pop);
2631 sk_msg_compute_data_pointers(msg);
2635 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
2636 .func = bpf_msg_pop_data,
2638 .ret_type = RET_INTEGER,
2639 .arg1_type = ARG_PTR_TO_CTX,
2640 .arg2_type = ARG_ANYTHING,
2641 .arg3_type = ARG_ANYTHING,
2642 .arg4_type = ARG_ANYTHING,
2645 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
2647 return task_get_classid(skb);
2650 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
2651 .func = bpf_get_cgroup_classid,
2653 .ret_type = RET_INTEGER,
2654 .arg1_type = ARG_PTR_TO_CTX,
2657 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
2659 return dst_tclassid(skb);
2662 static const struct bpf_func_proto bpf_get_route_realm_proto = {
2663 .func = bpf_get_route_realm,
2665 .ret_type = RET_INTEGER,
2666 .arg1_type = ARG_PTR_TO_CTX,
2669 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
2671 /* If skb_clear_hash() was called due to mangling, we can
2672 * trigger SW recalculation here. Later access to hash
2673 * can then use the inline skb->hash via context directly
2674 * instead of calling this helper again.
2676 return skb_get_hash(skb);
2679 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
2680 .func = bpf_get_hash_recalc,
2682 .ret_type = RET_INTEGER,
2683 .arg1_type = ARG_PTR_TO_CTX,
2686 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
2688 /* After all direct packet write, this can be used once for
2689 * triggering a lazy recalc on next skb_get_hash() invocation.
2691 skb_clear_hash(skb);
2695 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
2696 .func = bpf_set_hash_invalid,
2698 .ret_type = RET_INTEGER,
2699 .arg1_type = ARG_PTR_TO_CTX,
2702 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
2704 /* Set user specified hash as L4(+), so that it gets returned
2705 * on skb_get_hash() call unless BPF prog later on triggers a
2708 __skb_set_sw_hash(skb, hash, true);
2712 static const struct bpf_func_proto bpf_set_hash_proto = {
2713 .func = bpf_set_hash,
2715 .ret_type = RET_INTEGER,
2716 .arg1_type = ARG_PTR_TO_CTX,
2717 .arg2_type = ARG_ANYTHING,
2720 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
2725 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
2726 vlan_proto != htons(ETH_P_8021AD)))
2727 vlan_proto = htons(ETH_P_8021Q);
2729 bpf_push_mac_rcsum(skb);
2730 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
2731 bpf_pull_mac_rcsum(skb);
2733 bpf_compute_data_pointers(skb);
2737 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
2738 .func = bpf_skb_vlan_push,
2740 .ret_type = RET_INTEGER,
2741 .arg1_type = ARG_PTR_TO_CTX,
2742 .arg2_type = ARG_ANYTHING,
2743 .arg3_type = ARG_ANYTHING,
2746 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
2750 bpf_push_mac_rcsum(skb);
2751 ret = skb_vlan_pop(skb);
2752 bpf_pull_mac_rcsum(skb);
2754 bpf_compute_data_pointers(skb);
2758 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
2759 .func = bpf_skb_vlan_pop,
2761 .ret_type = RET_INTEGER,
2762 .arg1_type = ARG_PTR_TO_CTX,
2765 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
2767 /* Caller already did skb_cow() with len as headroom,
2768 * so no need to do it here.
2771 memmove(skb->data, skb->data + len, off);
2772 memset(skb->data + off, 0, len);
2774 /* No skb_postpush_rcsum(skb, skb->data + off, len)
2775 * needed here as it does not change the skb->csum
2776 * result for checksum complete when summing over
2782 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
2784 /* skb_ensure_writable() is not needed here, as we're
2785 * already working on an uncloned skb.
2787 if (unlikely(!pskb_may_pull(skb, off + len)))
2790 skb_postpull_rcsum(skb, skb->data + off, len);
2791 memmove(skb->data + len, skb->data, off);
2792 __skb_pull(skb, len);
2797 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
2799 bool trans_same = skb->transport_header == skb->network_header;
2802 /* There's no need for __skb_push()/__skb_pull() pair to
2803 * get to the start of the mac header as we're guaranteed
2804 * to always start from here under eBPF.
2806 ret = bpf_skb_generic_push(skb, off, len);
2808 skb->mac_header -= len;
2809 skb->network_header -= len;
2811 skb->transport_header = skb->network_header;
2817 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
2819 bool trans_same = skb->transport_header == skb->network_header;
2822 /* Same here, __skb_push()/__skb_pull() pair not needed. */
2823 ret = bpf_skb_generic_pop(skb, off, len);
2825 skb->mac_header += len;
2826 skb->network_header += len;
2828 skb->transport_header = skb->network_header;
2834 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
2836 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2837 u32 off = skb_mac_header_len(skb);
2840 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
2843 ret = skb_cow(skb, len_diff);
2844 if (unlikely(ret < 0))
2847 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2848 if (unlikely(ret < 0))
2851 if (skb_is_gso(skb)) {
2852 struct skb_shared_info *shinfo = skb_shinfo(skb);
2854 /* SKB_GSO_TCPV4 needs to be changed into
2857 if (shinfo->gso_type & SKB_GSO_TCPV4) {
2858 shinfo->gso_type &= ~SKB_GSO_TCPV4;
2859 shinfo->gso_type |= SKB_GSO_TCPV6;
2862 /* Due to IPv6 header, MSS needs to be downgraded. */
2863 skb_decrease_gso_size(shinfo, len_diff);
2864 /* Header must be checked, and gso_segs recomputed. */
2865 shinfo->gso_type |= SKB_GSO_DODGY;
2866 shinfo->gso_segs = 0;
2869 skb->protocol = htons(ETH_P_IPV6);
2870 skb_clear_hash(skb);
2875 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
2877 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2878 u32 off = skb_mac_header_len(skb);
2881 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
2884 ret = skb_unclone(skb, GFP_ATOMIC);
2885 if (unlikely(ret < 0))
2888 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2889 if (unlikely(ret < 0))
2892 if (skb_is_gso(skb)) {
2893 struct skb_shared_info *shinfo = skb_shinfo(skb);
2895 /* SKB_GSO_TCPV6 needs to be changed into
2898 if (shinfo->gso_type & SKB_GSO_TCPV6) {
2899 shinfo->gso_type &= ~SKB_GSO_TCPV6;
2900 shinfo->gso_type |= SKB_GSO_TCPV4;
2903 /* Due to IPv4 header, MSS can be upgraded. */
2904 skb_increase_gso_size(shinfo, len_diff);
2905 /* Header must be checked, and gso_segs recomputed. */
2906 shinfo->gso_type |= SKB_GSO_DODGY;
2907 shinfo->gso_segs = 0;
2910 skb->protocol = htons(ETH_P_IP);
2911 skb_clear_hash(skb);
2916 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
2918 __be16 from_proto = skb->protocol;
2920 if (from_proto == htons(ETH_P_IP) &&
2921 to_proto == htons(ETH_P_IPV6))
2922 return bpf_skb_proto_4_to_6(skb);
2924 if (from_proto == htons(ETH_P_IPV6) &&
2925 to_proto == htons(ETH_P_IP))
2926 return bpf_skb_proto_6_to_4(skb);
2931 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
2936 if (unlikely(flags))
2939 /* General idea is that this helper does the basic groundwork
2940 * needed for changing the protocol, and eBPF program fills the
2941 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
2942 * and other helpers, rather than passing a raw buffer here.
2944 * The rationale is to keep this minimal and without a need to
2945 * deal with raw packet data. F.e. even if we would pass buffers
2946 * here, the program still needs to call the bpf_lX_csum_replace()
2947 * helpers anyway. Plus, this way we keep also separation of
2948 * concerns, since f.e. bpf_skb_store_bytes() should only take
2951 * Currently, additional options and extension header space are
2952 * not supported, but flags register is reserved so we can adapt
2953 * that. For offloads, we mark packet as dodgy, so that headers
2954 * need to be verified first.
2956 ret = bpf_skb_proto_xlat(skb, proto);
2957 bpf_compute_data_pointers(skb);
2961 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
2962 .func = bpf_skb_change_proto,
2964 .ret_type = RET_INTEGER,
2965 .arg1_type = ARG_PTR_TO_CTX,
2966 .arg2_type = ARG_ANYTHING,
2967 .arg3_type = ARG_ANYTHING,
2970 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
2972 /* We only allow a restricted subset to be changed for now. */
2973 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
2974 !skb_pkt_type_ok(pkt_type)))
2977 skb->pkt_type = pkt_type;
2981 static const struct bpf_func_proto bpf_skb_change_type_proto = {
2982 .func = bpf_skb_change_type,
2984 .ret_type = RET_INTEGER,
2985 .arg1_type = ARG_PTR_TO_CTX,
2986 .arg2_type = ARG_ANYTHING,
2989 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
2991 switch (skb->protocol) {
2992 case htons(ETH_P_IP):
2993 return sizeof(struct iphdr);
2994 case htons(ETH_P_IPV6):
2995 return sizeof(struct ipv6hdr);
3001 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3002 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3004 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \
3005 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3006 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3007 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3008 BPF_F_ADJ_ROOM_ENCAP_L2( \
3009 BPF_ADJ_ROOM_ENCAP_L2_MASK))
3011 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3014 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3015 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3016 u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3017 unsigned int gso_type = SKB_GSO_DODGY;
3020 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3021 /* udp gso_size delineates datagrams, only allow if fixed */
3022 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3023 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3027 ret = skb_cow_head(skb, len_diff);
3028 if (unlikely(ret < 0))
3032 if (skb->protocol != htons(ETH_P_IP) &&
3033 skb->protocol != htons(ETH_P_IPV6))
3036 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3037 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3040 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3041 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3044 if (skb->encapsulation)
3047 mac_len = skb->network_header - skb->mac_header;
3048 inner_net = skb->network_header;
3049 if (inner_mac_len > len_diff)
3051 inner_trans = skb->transport_header;
3054 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3055 if (unlikely(ret < 0))
3059 skb->inner_mac_header = inner_net - inner_mac_len;
3060 skb->inner_network_header = inner_net;
3061 skb->inner_transport_header = inner_trans;
3062 skb_set_inner_protocol(skb, skb->protocol);
3064 skb->encapsulation = 1;
3065 skb_set_network_header(skb, mac_len);
3067 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3068 gso_type |= SKB_GSO_UDP_TUNNEL;
3069 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3070 gso_type |= SKB_GSO_GRE;
3071 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3072 gso_type |= SKB_GSO_IPXIP6;
3073 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3074 gso_type |= SKB_GSO_IPXIP4;
3076 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3077 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3078 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3079 sizeof(struct ipv6hdr) :
3080 sizeof(struct iphdr);
3082 skb_set_transport_header(skb, mac_len + nh_len);
3085 /* Match skb->protocol to new outer l3 protocol */
3086 if (skb->protocol == htons(ETH_P_IP) &&
3087 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3088 skb->protocol = htons(ETH_P_IPV6);
3089 else if (skb->protocol == htons(ETH_P_IPV6) &&
3090 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3091 skb->protocol = htons(ETH_P_IP);
3094 if (skb_is_gso(skb)) {
3095 struct skb_shared_info *shinfo = skb_shinfo(skb);
3097 /* Due to header grow, MSS needs to be downgraded. */
3098 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3099 skb_decrease_gso_size(shinfo, len_diff);
3101 /* Header must be checked, and gso_segs recomputed. */
3102 shinfo->gso_type |= gso_type;
3103 shinfo->gso_segs = 0;
3109 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3114 if (flags & ~BPF_F_ADJ_ROOM_FIXED_GSO)
3117 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3118 /* udp gso_size delineates datagrams, only allow if fixed */
3119 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3120 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3124 ret = skb_unclone(skb, GFP_ATOMIC);
3125 if (unlikely(ret < 0))
3128 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3129 if (unlikely(ret < 0))
3132 if (skb_is_gso(skb)) {
3133 struct skb_shared_info *shinfo = skb_shinfo(skb);
3135 /* Due to header shrink, MSS can be upgraded. */
3136 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3137 skb_increase_gso_size(shinfo, len_diff);
3139 /* Header must be checked, and gso_segs recomputed. */
3140 shinfo->gso_type |= SKB_GSO_DODGY;
3141 shinfo->gso_segs = 0;
3147 static u32 __bpf_skb_max_len(const struct sk_buff *skb)
3149 return skb->dev ? skb->dev->mtu + skb->dev->hard_header_len :
3153 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3154 u32, mode, u64, flags)
3156 u32 len_cur, len_diff_abs = abs(len_diff);
3157 u32 len_min = bpf_skb_net_base_len(skb);
3158 u32 len_max = __bpf_skb_max_len(skb);
3159 __be16 proto = skb->protocol;
3160 bool shrink = len_diff < 0;
3164 if (unlikely(flags & ~BPF_F_ADJ_ROOM_MASK))
3166 if (unlikely(len_diff_abs > 0xfffU))
3168 if (unlikely(proto != htons(ETH_P_IP) &&
3169 proto != htons(ETH_P_IPV6)))
3172 off = skb_mac_header_len(skb);
3174 case BPF_ADJ_ROOM_NET:
3175 off += bpf_skb_net_base_len(skb);
3177 case BPF_ADJ_ROOM_MAC:
3183 len_cur = skb->len - skb_network_offset(skb);
3184 if ((shrink && (len_diff_abs >= len_cur ||
3185 len_cur - len_diff_abs < len_min)) ||
3186 (!shrink && (skb->len + len_diff_abs > len_max &&
3190 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3191 bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3193 bpf_compute_data_pointers(skb);
3197 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3198 .func = bpf_skb_adjust_room,
3200 .ret_type = RET_INTEGER,
3201 .arg1_type = ARG_PTR_TO_CTX,
3202 .arg2_type = ARG_ANYTHING,
3203 .arg3_type = ARG_ANYTHING,
3204 .arg4_type = ARG_ANYTHING,
3207 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3209 u32 min_len = skb_network_offset(skb);
3211 if (skb_transport_header_was_set(skb))
3212 min_len = skb_transport_offset(skb);
3213 if (skb->ip_summed == CHECKSUM_PARTIAL)
3214 min_len = skb_checksum_start_offset(skb) +
3215 skb->csum_offset + sizeof(__sum16);
3219 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3221 unsigned int old_len = skb->len;
3224 ret = __skb_grow_rcsum(skb, new_len);
3226 memset(skb->data + old_len, 0, new_len - old_len);
3230 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3232 return __skb_trim_rcsum(skb, new_len);
3235 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3238 u32 max_len = __bpf_skb_max_len(skb);
3239 u32 min_len = __bpf_skb_min_len(skb);
3242 if (unlikely(flags || new_len > max_len || new_len < min_len))
3244 if (skb->encapsulation)
3247 /* The basic idea of this helper is that it's performing the
3248 * needed work to either grow or trim an skb, and eBPF program
3249 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3250 * bpf_lX_csum_replace() and others rather than passing a raw
3251 * buffer here. This one is a slow path helper and intended
3252 * for replies with control messages.
3254 * Like in bpf_skb_change_proto(), we want to keep this rather
3255 * minimal and without protocol specifics so that we are able
3256 * to separate concerns as in bpf_skb_store_bytes() should only
3257 * be the one responsible for writing buffers.
3259 * It's really expected to be a slow path operation here for
3260 * control message replies, so we're implicitly linearizing,
3261 * uncloning and drop offloads from the skb by this.
3263 ret = __bpf_try_make_writable(skb, skb->len);
3265 if (new_len > skb->len)
3266 ret = bpf_skb_grow_rcsum(skb, new_len);
3267 else if (new_len < skb->len)
3268 ret = bpf_skb_trim_rcsum(skb, new_len);
3269 if (!ret && skb_is_gso(skb))
3275 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3278 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3280 bpf_compute_data_pointers(skb);
3284 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3285 .func = bpf_skb_change_tail,
3287 .ret_type = RET_INTEGER,
3288 .arg1_type = ARG_PTR_TO_CTX,
3289 .arg2_type = ARG_ANYTHING,
3290 .arg3_type = ARG_ANYTHING,
3293 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3296 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3298 bpf_compute_data_end_sk_skb(skb);
3302 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3303 .func = sk_skb_change_tail,
3305 .ret_type = RET_INTEGER,
3306 .arg1_type = ARG_PTR_TO_CTX,
3307 .arg2_type = ARG_ANYTHING,
3308 .arg3_type = ARG_ANYTHING,
3311 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3314 u32 max_len = __bpf_skb_max_len(skb);
3315 u32 new_len = skb->len + head_room;
3318 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3319 new_len < skb->len))
3322 ret = skb_cow(skb, head_room);
3324 /* Idea for this helper is that we currently only
3325 * allow to expand on mac header. This means that
3326 * skb->protocol network header, etc, stay as is.
3327 * Compared to bpf_skb_change_tail(), we're more
3328 * flexible due to not needing to linearize or
3329 * reset GSO. Intention for this helper is to be
3330 * used by an L3 skb that needs to push mac header
3331 * for redirection into L2 device.
3333 __skb_push(skb, head_room);
3334 memset(skb->data, 0, head_room);
3335 skb_reset_mac_header(skb);
3341 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3344 int ret = __bpf_skb_change_head(skb, head_room, flags);
3346 bpf_compute_data_pointers(skb);
3350 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3351 .func = bpf_skb_change_head,
3353 .ret_type = RET_INTEGER,
3354 .arg1_type = ARG_PTR_TO_CTX,
3355 .arg2_type = ARG_ANYTHING,
3356 .arg3_type = ARG_ANYTHING,
3359 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3362 int ret = __bpf_skb_change_head(skb, head_room, flags);
3364 bpf_compute_data_end_sk_skb(skb);
3368 static const struct bpf_func_proto sk_skb_change_head_proto = {
3369 .func = sk_skb_change_head,
3371 .ret_type = RET_INTEGER,
3372 .arg1_type = ARG_PTR_TO_CTX,
3373 .arg2_type = ARG_ANYTHING,
3374 .arg3_type = ARG_ANYTHING,
3376 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3378 return xdp_data_meta_unsupported(xdp) ? 0 :
3379 xdp->data - xdp->data_meta;
3382 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3384 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3385 unsigned long metalen = xdp_get_metalen(xdp);
3386 void *data_start = xdp_frame_end + metalen;
3387 void *data = xdp->data + offset;
3389 if (unlikely(data < data_start ||
3390 data > xdp->data_end - ETH_HLEN))
3394 memmove(xdp->data_meta + offset,
3395 xdp->data_meta, metalen);
3396 xdp->data_meta += offset;
3402 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3403 .func = bpf_xdp_adjust_head,
3405 .ret_type = RET_INTEGER,
3406 .arg1_type = ARG_PTR_TO_CTX,
3407 .arg2_type = ARG_ANYTHING,
3410 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3412 void *data_end = xdp->data_end + offset;
3414 /* only shrinking is allowed for now. */
3415 if (unlikely(offset >= 0))
3418 if (unlikely(data_end < xdp->data + ETH_HLEN))
3421 xdp->data_end = data_end;
3426 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3427 .func = bpf_xdp_adjust_tail,
3429 .ret_type = RET_INTEGER,
3430 .arg1_type = ARG_PTR_TO_CTX,
3431 .arg2_type = ARG_ANYTHING,
3434 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3436 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3437 void *meta = xdp->data_meta + offset;
3438 unsigned long metalen = xdp->data - meta;
3440 if (xdp_data_meta_unsupported(xdp))
3442 if (unlikely(meta < xdp_frame_end ||
3445 if (unlikely((metalen & (sizeof(__u32) - 1)) ||
3449 xdp->data_meta = meta;
3454 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3455 .func = bpf_xdp_adjust_meta,
3457 .ret_type = RET_INTEGER,
3458 .arg1_type = ARG_PTR_TO_CTX,
3459 .arg2_type = ARG_ANYTHING,
3462 static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd,
3463 struct bpf_map *map, struct xdp_buff *xdp)
3465 switch (map->map_type) {
3466 case BPF_MAP_TYPE_DEVMAP:
3467 case BPF_MAP_TYPE_DEVMAP_HASH:
3468 return dev_map_enqueue(fwd, xdp, dev_rx);
3469 case BPF_MAP_TYPE_CPUMAP:
3470 return cpu_map_enqueue(fwd, xdp, dev_rx);
3471 case BPF_MAP_TYPE_XSKMAP:
3472 return __xsk_map_redirect(fwd, xdp);
3479 void xdp_do_flush(void)
3485 EXPORT_SYMBOL_GPL(xdp_do_flush);
3487 static inline void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index)
3489 switch (map->map_type) {
3490 case BPF_MAP_TYPE_DEVMAP:
3491 return __dev_map_lookup_elem(map, index);
3492 case BPF_MAP_TYPE_DEVMAP_HASH:
3493 return __dev_map_hash_lookup_elem(map, index);
3494 case BPF_MAP_TYPE_CPUMAP:
3495 return __cpu_map_lookup_elem(map, index);
3496 case BPF_MAP_TYPE_XSKMAP:
3497 return __xsk_map_lookup_elem(map, index);
3503 void bpf_clear_redirect_map(struct bpf_map *map)
3505 struct bpf_redirect_info *ri;
3508 for_each_possible_cpu(cpu) {
3509 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3510 /* Avoid polluting remote cacheline due to writes if
3511 * not needed. Once we pass this test, we need the
3512 * cmpxchg() to make sure it hasn't been changed in
3513 * the meantime by remote CPU.
3515 if (unlikely(READ_ONCE(ri->map) == map))
3516 cmpxchg(&ri->map, map, NULL);
3520 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3521 struct bpf_prog *xdp_prog)
3523 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3524 struct bpf_map *map = READ_ONCE(ri->map);
3525 u32 index = ri->tgt_index;
3526 void *fwd = ri->tgt_value;
3530 ri->tgt_value = NULL;
3531 WRITE_ONCE(ri->map, NULL);
3533 if (unlikely(!map)) {
3534 fwd = dev_get_by_index_rcu(dev_net(dev), index);
3535 if (unlikely(!fwd)) {
3540 err = dev_xdp_enqueue(fwd, xdp, dev);
3542 err = __bpf_tx_xdp_map(dev, fwd, map, xdp);
3548 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3551 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3554 EXPORT_SYMBOL_GPL(xdp_do_redirect);
3556 static int xdp_do_generic_redirect_map(struct net_device *dev,
3557 struct sk_buff *skb,
3558 struct xdp_buff *xdp,
3559 struct bpf_prog *xdp_prog,
3560 struct bpf_map *map)
3562 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3563 u32 index = ri->tgt_index;
3564 void *fwd = ri->tgt_value;
3568 ri->tgt_value = NULL;
3569 WRITE_ONCE(ri->map, NULL);
3571 if (map->map_type == BPF_MAP_TYPE_DEVMAP ||
3572 map->map_type == BPF_MAP_TYPE_DEVMAP_HASH) {
3573 struct bpf_dtab_netdev *dst = fwd;
3575 err = dev_map_generic_redirect(dst, skb, xdp_prog);
3578 } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) {
3579 struct xdp_sock *xs = fwd;
3581 err = xsk_generic_rcv(xs, xdp);
3586 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
3591 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3594 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3598 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
3599 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
3601 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3602 struct bpf_map *map = READ_ONCE(ri->map);
3603 u32 index = ri->tgt_index;
3604 struct net_device *fwd;
3608 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog,
3611 fwd = dev_get_by_index_rcu(dev_net(dev), index);
3612 if (unlikely(!fwd)) {
3617 err = xdp_ok_fwd_dev(fwd, skb->len);
3622 _trace_xdp_redirect(dev, xdp_prog, index);
3623 generic_xdp_tx(skb, xdp_prog);
3626 _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3629 EXPORT_SYMBOL_GPL(xdp_do_generic_redirect);
3631 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
3633 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3635 if (unlikely(flags))
3639 ri->tgt_index = ifindex;
3640 ri->tgt_value = NULL;
3641 WRITE_ONCE(ri->map, NULL);
3643 return XDP_REDIRECT;
3646 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
3647 .func = bpf_xdp_redirect,
3649 .ret_type = RET_INTEGER,
3650 .arg1_type = ARG_ANYTHING,
3651 .arg2_type = ARG_ANYTHING,
3654 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
3657 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3659 /* Lower bits of the flags are used as return code on lookup failure */
3660 if (unlikely(flags > XDP_TX))
3663 ri->tgt_value = __xdp_map_lookup_elem(map, ifindex);
3664 if (unlikely(!ri->tgt_value)) {
3665 /* If the lookup fails we want to clear out the state in the
3666 * redirect_info struct completely, so that if an eBPF program
3667 * performs multiple lookups, the last one always takes
3670 WRITE_ONCE(ri->map, NULL);
3675 ri->tgt_index = ifindex;
3676 WRITE_ONCE(ri->map, map);
3678 return XDP_REDIRECT;
3681 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
3682 .func = bpf_xdp_redirect_map,
3684 .ret_type = RET_INTEGER,
3685 .arg1_type = ARG_CONST_MAP_PTR,
3686 .arg2_type = ARG_ANYTHING,
3687 .arg3_type = ARG_ANYTHING,
3690 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
3691 unsigned long off, unsigned long len)
3693 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
3697 if (ptr != dst_buff)
3698 memcpy(dst_buff, ptr, len);
3703 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
3704 u64, flags, void *, meta, u64, meta_size)
3706 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
3708 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
3710 if (unlikely(!skb || skb_size > skb->len))
3713 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
3717 static const struct bpf_func_proto bpf_skb_event_output_proto = {
3718 .func = bpf_skb_event_output,
3720 .ret_type = RET_INTEGER,
3721 .arg1_type = ARG_PTR_TO_CTX,
3722 .arg2_type = ARG_CONST_MAP_PTR,
3723 .arg3_type = ARG_ANYTHING,
3724 .arg4_type = ARG_PTR_TO_MEM,
3725 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
3728 static int bpf_skb_output_btf_ids[5];
3729 const struct bpf_func_proto bpf_skb_output_proto = {
3730 .func = bpf_skb_event_output,
3732 .ret_type = RET_INTEGER,
3733 .arg1_type = ARG_PTR_TO_BTF_ID,
3734 .arg2_type = ARG_CONST_MAP_PTR,
3735 .arg3_type = ARG_ANYTHING,
3736 .arg4_type = ARG_PTR_TO_MEM,
3737 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
3738 .btf_id = bpf_skb_output_btf_ids,
3741 static unsigned short bpf_tunnel_key_af(u64 flags)
3743 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
3746 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
3747 u32, size, u64, flags)
3749 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3750 u8 compat[sizeof(struct bpf_tunnel_key)];
3754 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
3758 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
3762 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3765 case offsetof(struct bpf_tunnel_key, tunnel_label):
3766 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3768 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3769 /* Fixup deprecated structure layouts here, so we have
3770 * a common path later on.
3772 if (ip_tunnel_info_af(info) != AF_INET)
3775 to = (struct bpf_tunnel_key *)compat;
3782 to->tunnel_id = be64_to_cpu(info->key.tun_id);
3783 to->tunnel_tos = info->key.tos;
3784 to->tunnel_ttl = info->key.ttl;
3787 if (flags & BPF_F_TUNINFO_IPV6) {
3788 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
3789 sizeof(to->remote_ipv6));
3790 to->tunnel_label = be32_to_cpu(info->key.label);
3792 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
3793 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
3794 to->tunnel_label = 0;
3797 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
3798 memcpy(to_orig, to, size);
3802 memset(to_orig, 0, size);
3806 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
3807 .func = bpf_skb_get_tunnel_key,
3809 .ret_type = RET_INTEGER,
3810 .arg1_type = ARG_PTR_TO_CTX,
3811 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
3812 .arg3_type = ARG_CONST_SIZE,
3813 .arg4_type = ARG_ANYTHING,
3816 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
3818 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3821 if (unlikely(!info ||
3822 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
3826 if (unlikely(size < info->options_len)) {
3831 ip_tunnel_info_opts_get(to, info);
3832 if (size > info->options_len)
3833 memset(to + info->options_len, 0, size - info->options_len);
3835 return info->options_len;
3837 memset(to, 0, size);
3841 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
3842 .func = bpf_skb_get_tunnel_opt,
3844 .ret_type = RET_INTEGER,
3845 .arg1_type = ARG_PTR_TO_CTX,
3846 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
3847 .arg3_type = ARG_CONST_SIZE,
3850 static struct metadata_dst __percpu *md_dst;
3852 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
3853 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
3855 struct metadata_dst *md = this_cpu_ptr(md_dst);
3856 u8 compat[sizeof(struct bpf_tunnel_key)];
3857 struct ip_tunnel_info *info;
3859 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
3860 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
3862 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3864 case offsetof(struct bpf_tunnel_key, tunnel_label):
3865 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3866 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3867 /* Fixup deprecated structure layouts here, so we have
3868 * a common path later on.
3870 memcpy(compat, from, size);
3871 memset(compat + size, 0, sizeof(compat) - size);
3872 from = (const struct bpf_tunnel_key *) compat;
3878 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
3883 dst_hold((struct dst_entry *) md);
3884 skb_dst_set(skb, (struct dst_entry *) md);
3886 info = &md->u.tun_info;
3887 memset(info, 0, sizeof(*info));
3888 info->mode = IP_TUNNEL_INFO_TX;
3890 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
3891 if (flags & BPF_F_DONT_FRAGMENT)
3892 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
3893 if (flags & BPF_F_ZERO_CSUM_TX)
3894 info->key.tun_flags &= ~TUNNEL_CSUM;
3895 if (flags & BPF_F_SEQ_NUMBER)
3896 info->key.tun_flags |= TUNNEL_SEQ;
3898 info->key.tun_id = cpu_to_be64(from->tunnel_id);
3899 info->key.tos = from->tunnel_tos;
3900 info->key.ttl = from->tunnel_ttl;
3902 if (flags & BPF_F_TUNINFO_IPV6) {
3903 info->mode |= IP_TUNNEL_INFO_IPV6;
3904 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
3905 sizeof(from->remote_ipv6));
3906 info->key.label = cpu_to_be32(from->tunnel_label) &
3907 IPV6_FLOWLABEL_MASK;
3909 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
3915 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
3916 .func = bpf_skb_set_tunnel_key,
3918 .ret_type = RET_INTEGER,
3919 .arg1_type = ARG_PTR_TO_CTX,
3920 .arg2_type = ARG_PTR_TO_MEM,
3921 .arg3_type = ARG_CONST_SIZE,
3922 .arg4_type = ARG_ANYTHING,
3925 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
3926 const u8 *, from, u32, size)
3928 struct ip_tunnel_info *info = skb_tunnel_info(skb);
3929 const struct metadata_dst *md = this_cpu_ptr(md_dst);
3931 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
3933 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
3936 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
3941 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
3942 .func = bpf_skb_set_tunnel_opt,
3944 .ret_type = RET_INTEGER,
3945 .arg1_type = ARG_PTR_TO_CTX,
3946 .arg2_type = ARG_PTR_TO_MEM,
3947 .arg3_type = ARG_CONST_SIZE,
3950 static const struct bpf_func_proto *
3951 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
3954 struct metadata_dst __percpu *tmp;
3956 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
3961 if (cmpxchg(&md_dst, NULL, tmp))
3962 metadata_dst_free_percpu(tmp);
3966 case BPF_FUNC_skb_set_tunnel_key:
3967 return &bpf_skb_set_tunnel_key_proto;
3968 case BPF_FUNC_skb_set_tunnel_opt:
3969 return &bpf_skb_set_tunnel_opt_proto;
3975 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
3978 struct bpf_array *array = container_of(map, struct bpf_array, map);
3979 struct cgroup *cgrp;
3982 sk = skb_to_full_sk(skb);
3983 if (!sk || !sk_fullsock(sk))
3985 if (unlikely(idx >= array->map.max_entries))
3988 cgrp = READ_ONCE(array->ptrs[idx]);
3989 if (unlikely(!cgrp))
3992 return sk_under_cgroup_hierarchy(sk, cgrp);
3995 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
3996 .func = bpf_skb_under_cgroup,
3998 .ret_type = RET_INTEGER,
3999 .arg1_type = ARG_PTR_TO_CTX,
4000 .arg2_type = ARG_CONST_MAP_PTR,
4001 .arg3_type = ARG_ANYTHING,
4004 #ifdef CONFIG_SOCK_CGROUP_DATA
4005 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4007 struct sock *sk = skb_to_full_sk(skb);
4008 struct cgroup *cgrp;
4010 if (!sk || !sk_fullsock(sk))
4013 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4014 return cgroup_id(cgrp);
4017 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4018 .func = bpf_skb_cgroup_id,
4020 .ret_type = RET_INTEGER,
4021 .arg1_type = ARG_PTR_TO_CTX,
4024 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4027 struct sock *sk = skb_to_full_sk(skb);
4028 struct cgroup *ancestor;
4029 struct cgroup *cgrp;
4031 if (!sk || !sk_fullsock(sk))
4034 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4035 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4039 return cgroup_id(ancestor);
4042 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4043 .func = bpf_skb_ancestor_cgroup_id,
4045 .ret_type = RET_INTEGER,
4046 .arg1_type = ARG_PTR_TO_CTX,
4047 .arg2_type = ARG_ANYTHING,
4051 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4052 unsigned long off, unsigned long len)
4054 memcpy(dst_buff, src_buff + off, len);
4058 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4059 u64, flags, void *, meta, u64, meta_size)
4061 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4063 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4065 if (unlikely(xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4068 return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4069 xdp_size, bpf_xdp_copy);
4072 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4073 .func = bpf_xdp_event_output,
4075 .ret_type = RET_INTEGER,
4076 .arg1_type = ARG_PTR_TO_CTX,
4077 .arg2_type = ARG_CONST_MAP_PTR,
4078 .arg3_type = ARG_ANYTHING,
4079 .arg4_type = ARG_PTR_TO_MEM,
4080 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4083 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4085 return skb->sk ? sock_gen_cookie(skb->sk) : 0;
4088 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4089 .func = bpf_get_socket_cookie,
4091 .ret_type = RET_INTEGER,
4092 .arg1_type = ARG_PTR_TO_CTX,
4095 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4097 return sock_gen_cookie(ctx->sk);
4100 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4101 .func = bpf_get_socket_cookie_sock_addr,
4103 .ret_type = RET_INTEGER,
4104 .arg1_type = ARG_PTR_TO_CTX,
4107 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4109 return sock_gen_cookie(ctx->sk);
4112 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4113 .func = bpf_get_socket_cookie_sock_ops,
4115 .ret_type = RET_INTEGER,
4116 .arg1_type = ARG_PTR_TO_CTX,
4119 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4121 struct sock *sk = sk_to_full_sk(skb->sk);
4124 if (!sk || !sk_fullsock(sk))
4126 kuid = sock_net_uid(sock_net(sk), sk);
4127 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4130 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4131 .func = bpf_get_socket_uid,
4133 .ret_type = RET_INTEGER,
4134 .arg1_type = ARG_PTR_TO_CTX,
4137 BPF_CALL_5(bpf_sockopt_event_output, struct bpf_sock_ops_kern *, bpf_sock,
4138 struct bpf_map *, map, u64, flags, void *, data, u64, size)
4140 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
4143 return bpf_event_output(map, flags, data, size, NULL, 0, NULL);
4146 static const struct bpf_func_proto bpf_sockopt_event_output_proto = {
4147 .func = bpf_sockopt_event_output,
4149 .ret_type = RET_INTEGER,
4150 .arg1_type = ARG_PTR_TO_CTX,
4151 .arg2_type = ARG_CONST_MAP_PTR,
4152 .arg3_type = ARG_ANYTHING,
4153 .arg4_type = ARG_PTR_TO_MEM,
4154 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4157 BPF_CALL_5(bpf_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
4158 int, level, int, optname, char *, optval, int, optlen)
4160 struct sock *sk = bpf_sock->sk;
4164 if (!sk_fullsock(sk))
4167 if (level == SOL_SOCKET) {
4168 if (optlen != sizeof(int))
4170 val = *((int *)optval);
4172 /* Only some socketops are supported */
4175 val = min_t(u32, val, sysctl_rmem_max);
4176 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4177 WRITE_ONCE(sk->sk_rcvbuf,
4178 max_t(int, val * 2, SOCK_MIN_RCVBUF));
4181 val = min_t(u32, val, sysctl_wmem_max);
4182 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4183 WRITE_ONCE(sk->sk_sndbuf,
4184 max_t(int, val * 2, SOCK_MIN_SNDBUF));
4186 case SO_MAX_PACING_RATE: /* 32bit version */
4188 cmpxchg(&sk->sk_pacing_status,
4191 sk->sk_max_pacing_rate = (val == ~0U) ? ~0UL : val;
4192 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4193 sk->sk_max_pacing_rate);
4196 sk->sk_priority = val;
4201 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
4204 if (sk->sk_mark != val) {
4213 } else if (level == SOL_IP) {
4214 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4217 val = *((int *)optval);
4218 /* Only some options are supported */
4221 if (val < -1 || val > 0xff) {
4224 struct inet_sock *inet = inet_sk(sk);
4234 #if IS_ENABLED(CONFIG_IPV6)
4235 } else if (level == SOL_IPV6) {
4236 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4239 val = *((int *)optval);
4240 /* Only some options are supported */
4243 if (val < -1 || val > 0xff) {
4246 struct ipv6_pinfo *np = inet6_sk(sk);
4257 } else if (level == SOL_TCP &&
4258 sk->sk_prot->setsockopt == tcp_setsockopt) {
4259 if (optname == TCP_CONGESTION) {
4260 char name[TCP_CA_NAME_MAX];
4261 bool reinit = bpf_sock->op > BPF_SOCK_OPS_NEEDS_ECN;
4263 strncpy(name, optval, min_t(long, optlen,
4264 TCP_CA_NAME_MAX-1));
4265 name[TCP_CA_NAME_MAX-1] = 0;
4266 ret = tcp_set_congestion_control(sk, name, false,
4269 struct tcp_sock *tp = tcp_sk(sk);
4271 if (optlen != sizeof(int))
4274 val = *((int *)optval);
4275 /* Only some options are supported */
4278 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4283 case TCP_BPF_SNDCWND_CLAMP:
4287 tp->snd_cwnd_clamp = val;
4288 tp->snd_ssthresh = val;
4292 if (val < 0 || val > 1)
4308 static const struct bpf_func_proto bpf_setsockopt_proto = {
4309 .func = bpf_setsockopt,
4311 .ret_type = RET_INTEGER,
4312 .arg1_type = ARG_PTR_TO_CTX,
4313 .arg2_type = ARG_ANYTHING,
4314 .arg3_type = ARG_ANYTHING,
4315 .arg4_type = ARG_PTR_TO_MEM,
4316 .arg5_type = ARG_CONST_SIZE,
4319 BPF_CALL_5(bpf_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
4320 int, level, int, optname, char *, optval, int, optlen)
4322 struct sock *sk = bpf_sock->sk;
4324 if (!sk_fullsock(sk))
4327 if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4328 struct inet_connection_sock *icsk;
4329 struct tcp_sock *tp;
4332 case TCP_CONGESTION:
4333 icsk = inet_csk(sk);
4335 if (!icsk->icsk_ca_ops || optlen <= 1)
4337 strncpy(optval, icsk->icsk_ca_ops->name, optlen);
4338 optval[optlen - 1] = 0;
4343 if (optlen <= 0 || !tp->saved_syn ||
4344 optlen > tp->saved_syn[0])
4346 memcpy(optval, tp->saved_syn + 1, optlen);
4351 } else if (level == SOL_IP) {
4352 struct inet_sock *inet = inet_sk(sk);
4354 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4357 /* Only some options are supported */
4360 *((int *)optval) = (int)inet->tos;
4365 #if IS_ENABLED(CONFIG_IPV6)
4366 } else if (level == SOL_IPV6) {
4367 struct ipv6_pinfo *np = inet6_sk(sk);
4369 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4372 /* Only some options are supported */
4375 *((int *)optval) = (int)np->tclass;
4387 memset(optval, 0, optlen);
4391 static const struct bpf_func_proto bpf_getsockopt_proto = {
4392 .func = bpf_getsockopt,
4394 .ret_type = RET_INTEGER,
4395 .arg1_type = ARG_PTR_TO_CTX,
4396 .arg2_type = ARG_ANYTHING,
4397 .arg3_type = ARG_ANYTHING,
4398 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
4399 .arg5_type = ARG_CONST_SIZE,
4402 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
4405 struct sock *sk = bpf_sock->sk;
4406 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
4408 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
4411 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
4413 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
4416 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
4417 .func = bpf_sock_ops_cb_flags_set,
4419 .ret_type = RET_INTEGER,
4420 .arg1_type = ARG_PTR_TO_CTX,
4421 .arg2_type = ARG_ANYTHING,
4424 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
4425 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
4427 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
4431 struct sock *sk = ctx->sk;
4434 /* Binding to port can be expensive so it's prohibited in the helper.
4435 * Only binding to IP is supported.
4438 if (addr_len < offsetofend(struct sockaddr, sa_family))
4440 if (addr->sa_family == AF_INET) {
4441 if (addr_len < sizeof(struct sockaddr_in))
4443 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
4445 return __inet_bind(sk, addr, addr_len, true, false);
4446 #if IS_ENABLED(CONFIG_IPV6)
4447 } else if (addr->sa_family == AF_INET6) {
4448 if (addr_len < SIN6_LEN_RFC2133)
4450 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
4452 /* ipv6_bpf_stub cannot be NULL, since it's called from
4453 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
4455 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, true, false);
4456 #endif /* CONFIG_IPV6 */
4458 #endif /* CONFIG_INET */
4460 return -EAFNOSUPPORT;
4463 static const struct bpf_func_proto bpf_bind_proto = {
4466 .ret_type = RET_INTEGER,
4467 .arg1_type = ARG_PTR_TO_CTX,
4468 .arg2_type = ARG_PTR_TO_MEM,
4469 .arg3_type = ARG_CONST_SIZE,
4473 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
4474 struct bpf_xfrm_state *, to, u32, size, u64, flags)
4476 const struct sec_path *sp = skb_sec_path(skb);
4477 const struct xfrm_state *x;
4479 if (!sp || unlikely(index >= sp->len || flags))
4482 x = sp->xvec[index];
4484 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
4487 to->reqid = x->props.reqid;
4488 to->spi = x->id.spi;
4489 to->family = x->props.family;
4492 if (to->family == AF_INET6) {
4493 memcpy(to->remote_ipv6, x->props.saddr.a6,
4494 sizeof(to->remote_ipv6));
4496 to->remote_ipv4 = x->props.saddr.a4;
4497 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4502 memset(to, 0, size);
4506 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
4507 .func = bpf_skb_get_xfrm_state,
4509 .ret_type = RET_INTEGER,
4510 .arg1_type = ARG_PTR_TO_CTX,
4511 .arg2_type = ARG_ANYTHING,
4512 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
4513 .arg4_type = ARG_CONST_SIZE,
4514 .arg5_type = ARG_ANYTHING,
4518 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
4519 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
4520 const struct neighbour *neigh,
4521 const struct net_device *dev)
4523 memcpy(params->dmac, neigh->ha, ETH_ALEN);
4524 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
4525 params->h_vlan_TCI = 0;
4526 params->h_vlan_proto = 0;
4527 params->ifindex = dev->ifindex;
4533 #if IS_ENABLED(CONFIG_INET)
4534 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4535 u32 flags, bool check_mtu)
4537 struct fib_nh_common *nhc;
4538 struct in_device *in_dev;
4539 struct neighbour *neigh;
4540 struct net_device *dev;
4541 struct fib_result res;
4546 dev = dev_get_by_index_rcu(net, params->ifindex);
4550 /* verify forwarding is enabled on this interface */
4551 in_dev = __in_dev_get_rcu(dev);
4552 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
4553 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4555 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4557 fl4.flowi4_oif = params->ifindex;
4559 fl4.flowi4_iif = params->ifindex;
4562 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
4563 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
4564 fl4.flowi4_flags = 0;
4566 fl4.flowi4_proto = params->l4_protocol;
4567 fl4.daddr = params->ipv4_dst;
4568 fl4.saddr = params->ipv4_src;
4569 fl4.fl4_sport = params->sport;
4570 fl4.fl4_dport = params->dport;
4572 if (flags & BPF_FIB_LOOKUP_DIRECT) {
4573 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4574 struct fib_table *tb;
4576 tb = fib_get_table(net, tbid);
4578 return BPF_FIB_LKUP_RET_NOT_FWDED;
4580 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
4582 fl4.flowi4_mark = 0;
4583 fl4.flowi4_secid = 0;
4584 fl4.flowi4_tun_key.tun_id = 0;
4585 fl4.flowi4_uid = sock_net_uid(net, NULL);
4587 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
4591 /* map fib lookup errors to RTN_ type */
4593 return BPF_FIB_LKUP_RET_BLACKHOLE;
4594 if (err == -EHOSTUNREACH)
4595 return BPF_FIB_LKUP_RET_UNREACHABLE;
4597 return BPF_FIB_LKUP_RET_PROHIBIT;
4599 return BPF_FIB_LKUP_RET_NOT_FWDED;
4602 if (res.type != RTN_UNICAST)
4603 return BPF_FIB_LKUP_RET_NOT_FWDED;
4605 if (fib_info_num_path(res.fi) > 1)
4606 fib_select_path(net, &res, &fl4, NULL);
4609 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
4610 if (params->tot_len > mtu)
4611 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4616 /* do not handle lwt encaps right now */
4617 if (nhc->nhc_lwtstate)
4618 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4622 params->rt_metric = res.fi->fib_priority;
4624 /* xdp and cls_bpf programs are run in RCU-bh so
4625 * rcu_read_lock_bh is not needed here
4627 if (likely(nhc->nhc_gw_family != AF_INET6)) {
4628 if (nhc->nhc_gw_family)
4629 params->ipv4_dst = nhc->nhc_gw.ipv4;
4631 neigh = __ipv4_neigh_lookup_noref(dev,
4632 (__force u32)params->ipv4_dst);
4634 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
4636 params->family = AF_INET6;
4637 *dst = nhc->nhc_gw.ipv6;
4638 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
4642 return BPF_FIB_LKUP_RET_NO_NEIGH;
4644 return bpf_fib_set_fwd_params(params, neigh, dev);
4648 #if IS_ENABLED(CONFIG_IPV6)
4649 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4650 u32 flags, bool check_mtu)
4652 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
4653 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
4654 struct fib6_result res = {};
4655 struct neighbour *neigh;
4656 struct net_device *dev;
4657 struct inet6_dev *idev;
4663 /* link local addresses are never forwarded */
4664 if (rt6_need_strict(dst) || rt6_need_strict(src))
4665 return BPF_FIB_LKUP_RET_NOT_FWDED;
4667 dev = dev_get_by_index_rcu(net, params->ifindex);
4671 idev = __in6_dev_get_safely(dev);
4672 if (unlikely(!idev || !idev->cnf.forwarding))
4673 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4675 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4677 oif = fl6.flowi6_oif = params->ifindex;
4679 oif = fl6.flowi6_iif = params->ifindex;
4681 strict = RT6_LOOKUP_F_HAS_SADDR;
4683 fl6.flowlabel = params->flowinfo;
4684 fl6.flowi6_scope = 0;
4685 fl6.flowi6_flags = 0;
4688 fl6.flowi6_proto = params->l4_protocol;
4691 fl6.fl6_sport = params->sport;
4692 fl6.fl6_dport = params->dport;
4694 if (flags & BPF_FIB_LOOKUP_DIRECT) {
4695 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4696 struct fib6_table *tb;
4698 tb = ipv6_stub->fib6_get_table(net, tbid);
4700 return BPF_FIB_LKUP_RET_NOT_FWDED;
4702 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
4705 fl6.flowi6_mark = 0;
4706 fl6.flowi6_secid = 0;
4707 fl6.flowi6_tun_key.tun_id = 0;
4708 fl6.flowi6_uid = sock_net_uid(net, NULL);
4710 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
4713 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
4714 res.f6i == net->ipv6.fib6_null_entry))
4715 return BPF_FIB_LKUP_RET_NOT_FWDED;
4717 switch (res.fib6_type) {
4718 /* only unicast is forwarded */
4722 return BPF_FIB_LKUP_RET_BLACKHOLE;
4723 case RTN_UNREACHABLE:
4724 return BPF_FIB_LKUP_RET_UNREACHABLE;
4726 return BPF_FIB_LKUP_RET_PROHIBIT;
4728 return BPF_FIB_LKUP_RET_NOT_FWDED;
4731 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
4732 fl6.flowi6_oif != 0, NULL, strict);
4735 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
4736 if (params->tot_len > mtu)
4737 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4740 if (res.nh->fib_nh_lws)
4741 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4743 if (res.nh->fib_nh_gw_family)
4744 *dst = res.nh->fib_nh_gw6;
4746 dev = res.nh->fib_nh_dev;
4747 params->rt_metric = res.f6i->fib6_metric;
4749 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
4752 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
4754 return BPF_FIB_LKUP_RET_NO_NEIGH;
4756 return bpf_fib_set_fwd_params(params, neigh, dev);
4760 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
4761 struct bpf_fib_lookup *, params, int, plen, u32, flags)
4763 if (plen < sizeof(*params))
4766 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4769 switch (params->family) {
4770 #if IS_ENABLED(CONFIG_INET)
4772 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
4775 #if IS_ENABLED(CONFIG_IPV6)
4777 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
4781 return -EAFNOSUPPORT;
4784 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
4785 .func = bpf_xdp_fib_lookup,
4787 .ret_type = RET_INTEGER,
4788 .arg1_type = ARG_PTR_TO_CTX,
4789 .arg2_type = ARG_PTR_TO_MEM,
4790 .arg3_type = ARG_CONST_SIZE,
4791 .arg4_type = ARG_ANYTHING,
4794 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
4795 struct bpf_fib_lookup *, params, int, plen, u32, flags)
4797 struct net *net = dev_net(skb->dev);
4798 int rc = -EAFNOSUPPORT;
4800 if (plen < sizeof(*params))
4803 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4806 switch (params->family) {
4807 #if IS_ENABLED(CONFIG_INET)
4809 rc = bpf_ipv4_fib_lookup(net, params, flags, false);
4812 #if IS_ENABLED(CONFIG_IPV6)
4814 rc = bpf_ipv6_fib_lookup(net, params, flags, false);
4820 struct net_device *dev;
4822 dev = dev_get_by_index_rcu(net, params->ifindex);
4823 if (!is_skb_forwardable(dev, skb))
4824 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
4830 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
4831 .func = bpf_skb_fib_lookup,
4833 .ret_type = RET_INTEGER,
4834 .arg1_type = ARG_PTR_TO_CTX,
4835 .arg2_type = ARG_PTR_TO_MEM,
4836 .arg3_type = ARG_CONST_SIZE,
4837 .arg4_type = ARG_ANYTHING,
4840 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4841 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
4844 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
4846 if (!seg6_validate_srh(srh, len))
4850 case BPF_LWT_ENCAP_SEG6_INLINE:
4851 if (skb->protocol != htons(ETH_P_IPV6))
4854 err = seg6_do_srh_inline(skb, srh);
4856 case BPF_LWT_ENCAP_SEG6:
4857 skb_reset_inner_headers(skb);
4858 skb->encapsulation = 1;
4859 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
4865 bpf_compute_data_pointers(skb);
4869 ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
4870 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
4872 return seg6_lookup_nexthop(skb, NULL, 0);
4874 #endif /* CONFIG_IPV6_SEG6_BPF */
4876 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
4877 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
4880 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
4884 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
4888 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4889 case BPF_LWT_ENCAP_SEG6:
4890 case BPF_LWT_ENCAP_SEG6_INLINE:
4891 return bpf_push_seg6_encap(skb, type, hdr, len);
4893 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
4894 case BPF_LWT_ENCAP_IP:
4895 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
4902 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
4903 void *, hdr, u32, len)
4906 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
4907 case BPF_LWT_ENCAP_IP:
4908 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
4915 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
4916 .func = bpf_lwt_in_push_encap,
4918 .ret_type = RET_INTEGER,
4919 .arg1_type = ARG_PTR_TO_CTX,
4920 .arg2_type = ARG_ANYTHING,
4921 .arg3_type = ARG_PTR_TO_MEM,
4922 .arg4_type = ARG_CONST_SIZE
4925 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
4926 .func = bpf_lwt_xmit_push_encap,
4928 .ret_type = RET_INTEGER,
4929 .arg1_type = ARG_PTR_TO_CTX,
4930 .arg2_type = ARG_ANYTHING,
4931 .arg3_type = ARG_PTR_TO_MEM,
4932 .arg4_type = ARG_CONST_SIZE
4935 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4936 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
4937 const void *, from, u32, len)
4939 struct seg6_bpf_srh_state *srh_state =
4940 this_cpu_ptr(&seg6_bpf_srh_states);
4941 struct ipv6_sr_hdr *srh = srh_state->srh;
4942 void *srh_tlvs, *srh_end, *ptr;
4948 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
4949 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
4951 ptr = skb->data + offset;
4952 if (ptr >= srh_tlvs && ptr + len <= srh_end)
4953 srh_state->valid = false;
4954 else if (ptr < (void *)&srh->flags ||
4955 ptr + len > (void *)&srh->segments)
4958 if (unlikely(bpf_try_make_writable(skb, offset + len)))
4960 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
4962 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4964 memcpy(skb->data + offset, from, len);
4968 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
4969 .func = bpf_lwt_seg6_store_bytes,
4971 .ret_type = RET_INTEGER,
4972 .arg1_type = ARG_PTR_TO_CTX,
4973 .arg2_type = ARG_ANYTHING,
4974 .arg3_type = ARG_PTR_TO_MEM,
4975 .arg4_type = ARG_CONST_SIZE
4978 static void bpf_update_srh_state(struct sk_buff *skb)
4980 struct seg6_bpf_srh_state *srh_state =
4981 this_cpu_ptr(&seg6_bpf_srh_states);
4984 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
4985 srh_state->srh = NULL;
4987 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4988 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
4989 srh_state->valid = true;
4993 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
4994 u32, action, void *, param, u32, param_len)
4996 struct seg6_bpf_srh_state *srh_state =
4997 this_cpu_ptr(&seg6_bpf_srh_states);
5002 case SEG6_LOCAL_ACTION_END_X:
5003 if (!seg6_bpf_has_valid_srh(skb))
5005 if (param_len != sizeof(struct in6_addr))
5007 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
5008 case SEG6_LOCAL_ACTION_END_T:
5009 if (!seg6_bpf_has_valid_srh(skb))
5011 if (param_len != sizeof(int))
5013 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5014 case SEG6_LOCAL_ACTION_END_DT6:
5015 if (!seg6_bpf_has_valid_srh(skb))
5017 if (param_len != sizeof(int))
5020 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
5022 if (!pskb_pull(skb, hdroff))
5025 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
5026 skb_reset_network_header(skb);
5027 skb_reset_transport_header(skb);
5028 skb->encapsulation = 0;
5030 bpf_compute_data_pointers(skb);
5031 bpf_update_srh_state(skb);
5032 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5033 case SEG6_LOCAL_ACTION_END_B6:
5034 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5036 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
5039 bpf_update_srh_state(skb);
5042 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
5043 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5045 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
5048 bpf_update_srh_state(skb);
5056 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
5057 .func = bpf_lwt_seg6_action,
5059 .ret_type = RET_INTEGER,
5060 .arg1_type = ARG_PTR_TO_CTX,
5061 .arg2_type = ARG_ANYTHING,
5062 .arg3_type = ARG_PTR_TO_MEM,
5063 .arg4_type = ARG_CONST_SIZE
5066 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
5069 struct seg6_bpf_srh_state *srh_state =
5070 this_cpu_ptr(&seg6_bpf_srh_states);
5071 struct ipv6_sr_hdr *srh = srh_state->srh;
5072 void *srh_end, *srh_tlvs, *ptr;
5073 struct ipv6hdr *hdr;
5077 if (unlikely(srh == NULL))
5080 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
5081 ((srh->first_segment + 1) << 4));
5082 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
5084 ptr = skb->data + offset;
5086 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
5088 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
5092 ret = skb_cow_head(skb, len);
5093 if (unlikely(ret < 0))
5096 ret = bpf_skb_net_hdr_push(skb, offset, len);
5098 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
5101 bpf_compute_data_pointers(skb);
5102 if (unlikely(ret < 0))
5105 hdr = (struct ipv6hdr *)skb->data;
5106 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5108 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5110 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5111 srh_state->hdrlen += len;
5112 srh_state->valid = false;
5116 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
5117 .func = bpf_lwt_seg6_adjust_srh,
5119 .ret_type = RET_INTEGER,
5120 .arg1_type = ARG_PTR_TO_CTX,
5121 .arg2_type = ARG_ANYTHING,
5122 .arg3_type = ARG_ANYTHING,
5124 #endif /* CONFIG_IPV6_SEG6_BPF */
5127 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
5128 int dif, int sdif, u8 family, u8 proto)
5130 bool refcounted = false;
5131 struct sock *sk = NULL;
5133 if (family == AF_INET) {
5134 __be32 src4 = tuple->ipv4.saddr;
5135 __be32 dst4 = tuple->ipv4.daddr;
5137 if (proto == IPPROTO_TCP)
5138 sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
5139 src4, tuple->ipv4.sport,
5140 dst4, tuple->ipv4.dport,
5141 dif, sdif, &refcounted);
5143 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
5144 dst4, tuple->ipv4.dport,
5145 dif, sdif, &udp_table, NULL);
5146 #if IS_ENABLED(CONFIG_IPV6)
5148 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
5149 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
5151 if (proto == IPPROTO_TCP)
5152 sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
5153 src6, tuple->ipv6.sport,
5154 dst6, ntohs(tuple->ipv6.dport),
5155 dif, sdif, &refcounted);
5156 else if (likely(ipv6_bpf_stub))
5157 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
5158 src6, tuple->ipv6.sport,
5159 dst6, tuple->ipv6.dport,
5165 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
5166 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
5172 /* bpf_skc_lookup performs the core lookup for different types of sockets,
5173 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
5174 * Returns the socket as an 'unsigned long' to simplify the casting in the
5175 * callers to satisfy BPF_CALL declarations.
5177 static struct sock *
5178 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5179 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
5182 struct sock *sk = NULL;
5183 u8 family = AF_UNSPEC;
5187 if (len == sizeof(tuple->ipv4))
5189 else if (len == sizeof(tuple->ipv6))
5194 if (unlikely(family == AF_UNSPEC || flags ||
5195 !((s32)netns_id < 0 || netns_id <= S32_MAX)))
5198 if (family == AF_INET)
5199 sdif = inet_sdif(skb);
5201 sdif = inet6_sdif(skb);
5203 if ((s32)netns_id < 0) {
5205 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
5207 net = get_net_ns_by_id(caller_net, netns_id);
5210 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
5218 static struct sock *
5219 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5220 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
5223 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
5224 ifindex, proto, netns_id, flags);
5227 sk = sk_to_full_sk(sk);
5228 if (!sk_fullsock(sk)) {
5237 static struct sock *
5238 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5239 u8 proto, u64 netns_id, u64 flags)
5241 struct net *caller_net;
5245 caller_net = dev_net(skb->dev);
5246 ifindex = skb->dev->ifindex;
5248 caller_net = sock_net(skb->sk);
5252 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
5256 static struct sock *
5257 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5258 u8 proto, u64 netns_id, u64 flags)
5260 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
5264 sk = sk_to_full_sk(sk);
5265 if (!sk_fullsock(sk)) {
5274 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
5275 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5277 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
5281 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
5282 .func = bpf_skc_lookup_tcp,
5285 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
5286 .arg1_type = ARG_PTR_TO_CTX,
5287 .arg2_type = ARG_PTR_TO_MEM,
5288 .arg3_type = ARG_CONST_SIZE,
5289 .arg4_type = ARG_ANYTHING,
5290 .arg5_type = ARG_ANYTHING,
5293 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
5294 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5296 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
5300 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
5301 .func = bpf_sk_lookup_tcp,
5304 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5305 .arg1_type = ARG_PTR_TO_CTX,
5306 .arg2_type = ARG_PTR_TO_MEM,
5307 .arg3_type = ARG_CONST_SIZE,
5308 .arg4_type = ARG_ANYTHING,
5309 .arg5_type = ARG_ANYTHING,
5312 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
5313 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5315 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
5319 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
5320 .func = bpf_sk_lookup_udp,
5323 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5324 .arg1_type = ARG_PTR_TO_CTX,
5325 .arg2_type = ARG_PTR_TO_MEM,
5326 .arg3_type = ARG_CONST_SIZE,
5327 .arg4_type = ARG_ANYTHING,
5328 .arg5_type = ARG_ANYTHING,
5331 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
5333 /* Only full sockets have sk->sk_flags. */
5334 if (!sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE))
5339 static const struct bpf_func_proto bpf_sk_release_proto = {
5340 .func = bpf_sk_release,
5342 .ret_type = RET_INTEGER,
5343 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
5346 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
5347 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
5349 struct net *caller_net = dev_net(ctx->rxq->dev);
5350 int ifindex = ctx->rxq->dev->ifindex;
5352 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
5353 ifindex, IPPROTO_UDP, netns_id,
5357 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
5358 .func = bpf_xdp_sk_lookup_udp,
5361 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5362 .arg1_type = ARG_PTR_TO_CTX,
5363 .arg2_type = ARG_PTR_TO_MEM,
5364 .arg3_type = ARG_CONST_SIZE,
5365 .arg4_type = ARG_ANYTHING,
5366 .arg5_type = ARG_ANYTHING,
5369 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
5370 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
5372 struct net *caller_net = dev_net(ctx->rxq->dev);
5373 int ifindex = ctx->rxq->dev->ifindex;
5375 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
5376 ifindex, IPPROTO_TCP, netns_id,
5380 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
5381 .func = bpf_xdp_skc_lookup_tcp,
5384 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
5385 .arg1_type = ARG_PTR_TO_CTX,
5386 .arg2_type = ARG_PTR_TO_MEM,
5387 .arg3_type = ARG_CONST_SIZE,
5388 .arg4_type = ARG_ANYTHING,
5389 .arg5_type = ARG_ANYTHING,
5392 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
5393 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
5395 struct net *caller_net = dev_net(ctx->rxq->dev);
5396 int ifindex = ctx->rxq->dev->ifindex;
5398 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
5399 ifindex, IPPROTO_TCP, netns_id,
5403 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
5404 .func = bpf_xdp_sk_lookup_tcp,
5407 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5408 .arg1_type = ARG_PTR_TO_CTX,
5409 .arg2_type = ARG_PTR_TO_MEM,
5410 .arg3_type = ARG_CONST_SIZE,
5411 .arg4_type = ARG_ANYTHING,
5412 .arg5_type = ARG_ANYTHING,
5415 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
5416 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5418 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
5419 sock_net(ctx->sk), 0,
5420 IPPROTO_TCP, netns_id, flags);
5423 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
5424 .func = bpf_sock_addr_skc_lookup_tcp,
5426 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
5427 .arg1_type = ARG_PTR_TO_CTX,
5428 .arg2_type = ARG_PTR_TO_MEM,
5429 .arg3_type = ARG_CONST_SIZE,
5430 .arg4_type = ARG_ANYTHING,
5431 .arg5_type = ARG_ANYTHING,
5434 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
5435 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5437 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
5438 sock_net(ctx->sk), 0, IPPROTO_TCP,
5442 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
5443 .func = bpf_sock_addr_sk_lookup_tcp,
5445 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5446 .arg1_type = ARG_PTR_TO_CTX,
5447 .arg2_type = ARG_PTR_TO_MEM,
5448 .arg3_type = ARG_CONST_SIZE,
5449 .arg4_type = ARG_ANYTHING,
5450 .arg5_type = ARG_ANYTHING,
5453 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
5454 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5456 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
5457 sock_net(ctx->sk), 0, IPPROTO_UDP,
5461 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
5462 .func = bpf_sock_addr_sk_lookup_udp,
5464 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5465 .arg1_type = ARG_PTR_TO_CTX,
5466 .arg2_type = ARG_PTR_TO_MEM,
5467 .arg3_type = ARG_CONST_SIZE,
5468 .arg4_type = ARG_ANYTHING,
5469 .arg5_type = ARG_ANYTHING,
5472 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
5473 struct bpf_insn_access_aux *info)
5475 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
5479 if (off % size != 0)
5483 case offsetof(struct bpf_tcp_sock, bytes_received):
5484 case offsetof(struct bpf_tcp_sock, bytes_acked):
5485 return size == sizeof(__u64);
5487 return size == sizeof(__u32);
5491 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
5492 const struct bpf_insn *si,
5493 struct bpf_insn *insn_buf,
5494 struct bpf_prog *prog, u32 *target_size)
5496 struct bpf_insn *insn = insn_buf;
5498 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \
5500 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \
5501 sizeof_field(struct bpf_tcp_sock, FIELD)); \
5502 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
5503 si->dst_reg, si->src_reg, \
5504 offsetof(struct tcp_sock, FIELD)); \
5507 #define BPF_INET_SOCK_GET_COMMON(FIELD) \
5509 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \
5511 sizeof_field(struct bpf_tcp_sock, FIELD)); \
5512 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
5513 struct inet_connection_sock, \
5515 si->dst_reg, si->src_reg, \
5517 struct inet_connection_sock, \
5521 if (insn > insn_buf)
5522 return insn - insn_buf;
5525 case offsetof(struct bpf_tcp_sock, rtt_min):
5526 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
5527 sizeof(struct minmax));
5528 BUILD_BUG_ON(sizeof(struct minmax) <
5529 sizeof(struct minmax_sample));
5531 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5532 offsetof(struct tcp_sock, rtt_min) +
5533 offsetof(struct minmax_sample, v));
5535 case offsetof(struct bpf_tcp_sock, snd_cwnd):
5536 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
5538 case offsetof(struct bpf_tcp_sock, srtt_us):
5539 BPF_TCP_SOCK_GET_COMMON(srtt_us);
5541 case offsetof(struct bpf_tcp_sock, snd_ssthresh):
5542 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
5544 case offsetof(struct bpf_tcp_sock, rcv_nxt):
5545 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
5547 case offsetof(struct bpf_tcp_sock, snd_nxt):
5548 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
5550 case offsetof(struct bpf_tcp_sock, snd_una):
5551 BPF_TCP_SOCK_GET_COMMON(snd_una);
5553 case offsetof(struct bpf_tcp_sock, mss_cache):
5554 BPF_TCP_SOCK_GET_COMMON(mss_cache);
5556 case offsetof(struct bpf_tcp_sock, ecn_flags):
5557 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
5559 case offsetof(struct bpf_tcp_sock, rate_delivered):
5560 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
5562 case offsetof(struct bpf_tcp_sock, rate_interval_us):
5563 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
5565 case offsetof(struct bpf_tcp_sock, packets_out):
5566 BPF_TCP_SOCK_GET_COMMON(packets_out);
5568 case offsetof(struct bpf_tcp_sock, retrans_out):
5569 BPF_TCP_SOCK_GET_COMMON(retrans_out);
5571 case offsetof(struct bpf_tcp_sock, total_retrans):
5572 BPF_TCP_SOCK_GET_COMMON(total_retrans);
5574 case offsetof(struct bpf_tcp_sock, segs_in):
5575 BPF_TCP_SOCK_GET_COMMON(segs_in);
5577 case offsetof(struct bpf_tcp_sock, data_segs_in):
5578 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
5580 case offsetof(struct bpf_tcp_sock, segs_out):
5581 BPF_TCP_SOCK_GET_COMMON(segs_out);
5583 case offsetof(struct bpf_tcp_sock, data_segs_out):
5584 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
5586 case offsetof(struct bpf_tcp_sock, lost_out):
5587 BPF_TCP_SOCK_GET_COMMON(lost_out);
5589 case offsetof(struct bpf_tcp_sock, sacked_out):
5590 BPF_TCP_SOCK_GET_COMMON(sacked_out);
5592 case offsetof(struct bpf_tcp_sock, bytes_received):
5593 BPF_TCP_SOCK_GET_COMMON(bytes_received);
5595 case offsetof(struct bpf_tcp_sock, bytes_acked):
5596 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
5598 case offsetof(struct bpf_tcp_sock, dsack_dups):
5599 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
5601 case offsetof(struct bpf_tcp_sock, delivered):
5602 BPF_TCP_SOCK_GET_COMMON(delivered);
5604 case offsetof(struct bpf_tcp_sock, delivered_ce):
5605 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
5607 case offsetof(struct bpf_tcp_sock, icsk_retransmits):
5608 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
5612 return insn - insn_buf;
5615 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
5617 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
5618 return (unsigned long)sk;
5620 return (unsigned long)NULL;
5623 const struct bpf_func_proto bpf_tcp_sock_proto = {
5624 .func = bpf_tcp_sock,
5626 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
5627 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
5630 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
5632 sk = sk_to_full_sk(sk);
5634 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
5635 return (unsigned long)sk;
5637 return (unsigned long)NULL;
5640 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
5641 .func = bpf_get_listener_sock,
5643 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5644 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
5647 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
5649 unsigned int iphdr_len;
5651 if (skb->protocol == cpu_to_be16(ETH_P_IP))
5652 iphdr_len = sizeof(struct iphdr);
5653 else if (skb->protocol == cpu_to_be16(ETH_P_IPV6))
5654 iphdr_len = sizeof(struct ipv6hdr);
5658 if (skb_headlen(skb) < iphdr_len)
5661 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
5664 return INET_ECN_set_ce(skb);
5667 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
5668 struct bpf_insn_access_aux *info)
5670 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
5673 if (off % size != 0)
5678 return size == sizeof(__u32);
5682 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
5683 const struct bpf_insn *si,
5684 struct bpf_insn *insn_buf,
5685 struct bpf_prog *prog, u32 *target_size)
5687 struct bpf_insn *insn = insn_buf;
5689 #define BPF_XDP_SOCK_GET(FIELD) \
5691 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \
5692 sizeof_field(struct bpf_xdp_sock, FIELD)); \
5693 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
5694 si->dst_reg, si->src_reg, \
5695 offsetof(struct xdp_sock, FIELD)); \
5699 case offsetof(struct bpf_xdp_sock, queue_id):
5700 BPF_XDP_SOCK_GET(queue_id);
5704 return insn - insn_buf;
5707 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
5708 .func = bpf_skb_ecn_set_ce,
5710 .ret_type = RET_INTEGER,
5711 .arg1_type = ARG_PTR_TO_CTX,
5714 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
5715 struct tcphdr *, th, u32, th_len)
5717 #ifdef CONFIG_SYN_COOKIES
5721 if (unlikely(th_len < sizeof(*th)))
5724 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
5725 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
5728 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
5731 if (!th->ack || th->rst || th->syn)
5734 if (tcp_synq_no_recent_overflow(sk))
5737 cookie = ntohl(th->ack_seq) - 1;
5739 switch (sk->sk_family) {
5741 if (unlikely(iph_len < sizeof(struct iphdr)))
5744 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
5747 #if IS_BUILTIN(CONFIG_IPV6)
5749 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
5752 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
5754 #endif /* CONFIG_IPV6 */
5757 return -EPROTONOSUPPORT;
5769 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
5770 .func = bpf_tcp_check_syncookie,
5773 .ret_type = RET_INTEGER,
5774 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
5775 .arg2_type = ARG_PTR_TO_MEM,
5776 .arg3_type = ARG_CONST_SIZE,
5777 .arg4_type = ARG_PTR_TO_MEM,
5778 .arg5_type = ARG_CONST_SIZE,
5781 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
5782 struct tcphdr *, th, u32, th_len)
5784 #ifdef CONFIG_SYN_COOKIES
5788 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
5791 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
5794 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
5797 if (!th->syn || th->ack || th->fin || th->rst)
5800 if (unlikely(iph_len < sizeof(struct iphdr)))
5803 /* Both struct iphdr and struct ipv6hdr have the version field at the
5804 * same offset so we can cast to the shorter header (struct iphdr).
5806 switch (((struct iphdr *)iph)->version) {
5808 if (sk->sk_family == AF_INET6 && sk->sk_ipv6only)
5811 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
5814 #if IS_BUILTIN(CONFIG_IPV6)
5816 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
5819 if (sk->sk_family != AF_INET6)
5822 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
5824 #endif /* CONFIG_IPV6 */
5827 return -EPROTONOSUPPORT;
5832 return cookie | ((u64)mss << 32);
5835 #endif /* CONFIG_SYN_COOKIES */
5838 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
5839 .func = bpf_tcp_gen_syncookie,
5840 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */
5842 .ret_type = RET_INTEGER,
5843 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
5844 .arg2_type = ARG_PTR_TO_MEM,
5845 .arg3_type = ARG_CONST_SIZE,
5846 .arg4_type = ARG_PTR_TO_MEM,
5847 .arg5_type = ARG_CONST_SIZE,
5850 #endif /* CONFIG_INET */
5852 bool bpf_helper_changes_pkt_data(void *func)
5854 if (func == bpf_skb_vlan_push ||
5855 func == bpf_skb_vlan_pop ||
5856 func == bpf_skb_store_bytes ||
5857 func == bpf_skb_change_proto ||
5858 func == bpf_skb_change_head ||
5859 func == sk_skb_change_head ||
5860 func == bpf_skb_change_tail ||
5861 func == sk_skb_change_tail ||
5862 func == bpf_skb_adjust_room ||
5863 func == bpf_skb_pull_data ||
5864 func == sk_skb_pull_data ||
5865 func == bpf_clone_redirect ||
5866 func == bpf_l3_csum_replace ||
5867 func == bpf_l4_csum_replace ||
5868 func == bpf_xdp_adjust_head ||
5869 func == bpf_xdp_adjust_meta ||
5870 func == bpf_msg_pull_data ||
5871 func == bpf_msg_push_data ||
5872 func == bpf_msg_pop_data ||
5873 func == bpf_xdp_adjust_tail ||
5874 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5875 func == bpf_lwt_seg6_store_bytes ||
5876 func == bpf_lwt_seg6_adjust_srh ||
5877 func == bpf_lwt_seg6_action ||
5879 func == bpf_lwt_in_push_encap ||
5880 func == bpf_lwt_xmit_push_encap)
5886 const struct bpf_func_proto *
5887 bpf_base_func_proto(enum bpf_func_id func_id)
5890 case BPF_FUNC_map_lookup_elem:
5891 return &bpf_map_lookup_elem_proto;
5892 case BPF_FUNC_map_update_elem:
5893 return &bpf_map_update_elem_proto;
5894 case BPF_FUNC_map_delete_elem:
5895 return &bpf_map_delete_elem_proto;
5896 case BPF_FUNC_map_push_elem:
5897 return &bpf_map_push_elem_proto;
5898 case BPF_FUNC_map_pop_elem:
5899 return &bpf_map_pop_elem_proto;
5900 case BPF_FUNC_map_peek_elem:
5901 return &bpf_map_peek_elem_proto;
5902 case BPF_FUNC_get_prandom_u32:
5903 return &bpf_get_prandom_u32_proto;
5904 case BPF_FUNC_get_smp_processor_id:
5905 return &bpf_get_raw_smp_processor_id_proto;
5906 case BPF_FUNC_get_numa_node_id:
5907 return &bpf_get_numa_node_id_proto;
5908 case BPF_FUNC_tail_call:
5909 return &bpf_tail_call_proto;
5910 case BPF_FUNC_ktime_get_ns:
5911 return &bpf_ktime_get_ns_proto;
5916 if (!capable(CAP_SYS_ADMIN))
5920 case BPF_FUNC_spin_lock:
5921 return &bpf_spin_lock_proto;
5922 case BPF_FUNC_spin_unlock:
5923 return &bpf_spin_unlock_proto;
5924 case BPF_FUNC_trace_printk:
5925 return bpf_get_trace_printk_proto();
5926 case BPF_FUNC_jiffies64:
5927 return &bpf_jiffies64_proto;
5933 static const struct bpf_func_proto *
5934 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5937 /* inet and inet6 sockets are created in a process
5938 * context so there is always a valid uid/gid
5940 case BPF_FUNC_get_current_uid_gid:
5941 return &bpf_get_current_uid_gid_proto;
5942 case BPF_FUNC_get_local_storage:
5943 return &bpf_get_local_storage_proto;
5945 return bpf_base_func_proto(func_id);
5949 static const struct bpf_func_proto *
5950 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5953 /* inet and inet6 sockets are created in a process
5954 * context so there is always a valid uid/gid
5956 case BPF_FUNC_get_current_uid_gid:
5957 return &bpf_get_current_uid_gid_proto;
5959 switch (prog->expected_attach_type) {
5960 case BPF_CGROUP_INET4_CONNECT:
5961 case BPF_CGROUP_INET6_CONNECT:
5962 return &bpf_bind_proto;
5966 case BPF_FUNC_get_socket_cookie:
5967 return &bpf_get_socket_cookie_sock_addr_proto;
5968 case BPF_FUNC_get_local_storage:
5969 return &bpf_get_local_storage_proto;
5971 case BPF_FUNC_sk_lookup_tcp:
5972 return &bpf_sock_addr_sk_lookup_tcp_proto;
5973 case BPF_FUNC_sk_lookup_udp:
5974 return &bpf_sock_addr_sk_lookup_udp_proto;
5975 case BPF_FUNC_sk_release:
5976 return &bpf_sk_release_proto;
5977 case BPF_FUNC_skc_lookup_tcp:
5978 return &bpf_sock_addr_skc_lookup_tcp_proto;
5979 #endif /* CONFIG_INET */
5980 case BPF_FUNC_sk_storage_get:
5981 return &bpf_sk_storage_get_proto;
5982 case BPF_FUNC_sk_storage_delete:
5983 return &bpf_sk_storage_delete_proto;
5985 return bpf_base_func_proto(func_id);
5989 static const struct bpf_func_proto *
5990 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5993 case BPF_FUNC_skb_load_bytes:
5994 return &bpf_skb_load_bytes_proto;
5995 case BPF_FUNC_skb_load_bytes_relative:
5996 return &bpf_skb_load_bytes_relative_proto;
5997 case BPF_FUNC_get_socket_cookie:
5998 return &bpf_get_socket_cookie_proto;
5999 case BPF_FUNC_get_socket_uid:
6000 return &bpf_get_socket_uid_proto;
6001 case BPF_FUNC_perf_event_output:
6002 return &bpf_skb_event_output_proto;
6004 return bpf_base_func_proto(func_id);
6008 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
6009 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
6011 static const struct bpf_func_proto *
6012 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6015 case BPF_FUNC_get_local_storage:
6016 return &bpf_get_local_storage_proto;
6017 case BPF_FUNC_sk_fullsock:
6018 return &bpf_sk_fullsock_proto;
6019 case BPF_FUNC_sk_storage_get:
6020 return &bpf_sk_storage_get_proto;
6021 case BPF_FUNC_sk_storage_delete:
6022 return &bpf_sk_storage_delete_proto;
6023 case BPF_FUNC_perf_event_output:
6024 return &bpf_skb_event_output_proto;
6025 #ifdef CONFIG_SOCK_CGROUP_DATA
6026 case BPF_FUNC_skb_cgroup_id:
6027 return &bpf_skb_cgroup_id_proto;
6030 case BPF_FUNC_tcp_sock:
6031 return &bpf_tcp_sock_proto;
6032 case BPF_FUNC_get_listener_sock:
6033 return &bpf_get_listener_sock_proto;
6034 case BPF_FUNC_skb_ecn_set_ce:
6035 return &bpf_skb_ecn_set_ce_proto;
6038 return sk_filter_func_proto(func_id, prog);
6042 static const struct bpf_func_proto *
6043 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6046 case BPF_FUNC_skb_store_bytes:
6047 return &bpf_skb_store_bytes_proto;
6048 case BPF_FUNC_skb_load_bytes:
6049 return &bpf_skb_load_bytes_proto;
6050 case BPF_FUNC_skb_load_bytes_relative:
6051 return &bpf_skb_load_bytes_relative_proto;
6052 case BPF_FUNC_skb_pull_data:
6053 return &bpf_skb_pull_data_proto;
6054 case BPF_FUNC_csum_diff:
6055 return &bpf_csum_diff_proto;
6056 case BPF_FUNC_csum_update:
6057 return &bpf_csum_update_proto;
6058 case BPF_FUNC_l3_csum_replace:
6059 return &bpf_l3_csum_replace_proto;
6060 case BPF_FUNC_l4_csum_replace:
6061 return &bpf_l4_csum_replace_proto;
6062 case BPF_FUNC_clone_redirect:
6063 return &bpf_clone_redirect_proto;
6064 case BPF_FUNC_get_cgroup_classid:
6065 return &bpf_get_cgroup_classid_proto;
6066 case BPF_FUNC_skb_vlan_push:
6067 return &bpf_skb_vlan_push_proto;
6068 case BPF_FUNC_skb_vlan_pop:
6069 return &bpf_skb_vlan_pop_proto;
6070 case BPF_FUNC_skb_change_proto:
6071 return &bpf_skb_change_proto_proto;
6072 case BPF_FUNC_skb_change_type:
6073 return &bpf_skb_change_type_proto;
6074 case BPF_FUNC_skb_adjust_room:
6075 return &bpf_skb_adjust_room_proto;
6076 case BPF_FUNC_skb_change_tail:
6077 return &bpf_skb_change_tail_proto;
6078 case BPF_FUNC_skb_get_tunnel_key:
6079 return &bpf_skb_get_tunnel_key_proto;
6080 case BPF_FUNC_skb_set_tunnel_key:
6081 return bpf_get_skb_set_tunnel_proto(func_id);
6082 case BPF_FUNC_skb_get_tunnel_opt:
6083 return &bpf_skb_get_tunnel_opt_proto;
6084 case BPF_FUNC_skb_set_tunnel_opt:
6085 return bpf_get_skb_set_tunnel_proto(func_id);
6086 case BPF_FUNC_redirect:
6087 return &bpf_redirect_proto;
6088 case BPF_FUNC_get_route_realm:
6089 return &bpf_get_route_realm_proto;
6090 case BPF_FUNC_get_hash_recalc:
6091 return &bpf_get_hash_recalc_proto;
6092 case BPF_FUNC_set_hash_invalid:
6093 return &bpf_set_hash_invalid_proto;
6094 case BPF_FUNC_set_hash:
6095 return &bpf_set_hash_proto;
6096 case BPF_FUNC_perf_event_output:
6097 return &bpf_skb_event_output_proto;
6098 case BPF_FUNC_get_smp_processor_id:
6099 return &bpf_get_smp_processor_id_proto;
6100 case BPF_FUNC_skb_under_cgroup:
6101 return &bpf_skb_under_cgroup_proto;
6102 case BPF_FUNC_get_socket_cookie:
6103 return &bpf_get_socket_cookie_proto;
6104 case BPF_FUNC_get_socket_uid:
6105 return &bpf_get_socket_uid_proto;
6106 case BPF_FUNC_fib_lookup:
6107 return &bpf_skb_fib_lookup_proto;
6108 case BPF_FUNC_sk_fullsock:
6109 return &bpf_sk_fullsock_proto;
6110 case BPF_FUNC_sk_storage_get:
6111 return &bpf_sk_storage_get_proto;
6112 case BPF_FUNC_sk_storage_delete:
6113 return &bpf_sk_storage_delete_proto;
6115 case BPF_FUNC_skb_get_xfrm_state:
6116 return &bpf_skb_get_xfrm_state_proto;
6118 #ifdef CONFIG_SOCK_CGROUP_DATA
6119 case BPF_FUNC_skb_cgroup_id:
6120 return &bpf_skb_cgroup_id_proto;
6121 case BPF_FUNC_skb_ancestor_cgroup_id:
6122 return &bpf_skb_ancestor_cgroup_id_proto;
6125 case BPF_FUNC_sk_lookup_tcp:
6126 return &bpf_sk_lookup_tcp_proto;
6127 case BPF_FUNC_sk_lookup_udp:
6128 return &bpf_sk_lookup_udp_proto;
6129 case BPF_FUNC_sk_release:
6130 return &bpf_sk_release_proto;
6131 case BPF_FUNC_tcp_sock:
6132 return &bpf_tcp_sock_proto;
6133 case BPF_FUNC_get_listener_sock:
6134 return &bpf_get_listener_sock_proto;
6135 case BPF_FUNC_skc_lookup_tcp:
6136 return &bpf_skc_lookup_tcp_proto;
6137 case BPF_FUNC_tcp_check_syncookie:
6138 return &bpf_tcp_check_syncookie_proto;
6139 case BPF_FUNC_skb_ecn_set_ce:
6140 return &bpf_skb_ecn_set_ce_proto;
6141 case BPF_FUNC_tcp_gen_syncookie:
6142 return &bpf_tcp_gen_syncookie_proto;
6145 return bpf_base_func_proto(func_id);
6149 static const struct bpf_func_proto *
6150 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6153 case BPF_FUNC_perf_event_output:
6154 return &bpf_xdp_event_output_proto;
6155 case BPF_FUNC_get_smp_processor_id:
6156 return &bpf_get_smp_processor_id_proto;
6157 case BPF_FUNC_csum_diff:
6158 return &bpf_csum_diff_proto;
6159 case BPF_FUNC_xdp_adjust_head:
6160 return &bpf_xdp_adjust_head_proto;
6161 case BPF_FUNC_xdp_adjust_meta:
6162 return &bpf_xdp_adjust_meta_proto;
6163 case BPF_FUNC_redirect:
6164 return &bpf_xdp_redirect_proto;
6165 case BPF_FUNC_redirect_map:
6166 return &bpf_xdp_redirect_map_proto;
6167 case BPF_FUNC_xdp_adjust_tail:
6168 return &bpf_xdp_adjust_tail_proto;
6169 case BPF_FUNC_fib_lookup:
6170 return &bpf_xdp_fib_lookup_proto;
6172 case BPF_FUNC_sk_lookup_udp:
6173 return &bpf_xdp_sk_lookup_udp_proto;
6174 case BPF_FUNC_sk_lookup_tcp:
6175 return &bpf_xdp_sk_lookup_tcp_proto;
6176 case BPF_FUNC_sk_release:
6177 return &bpf_sk_release_proto;
6178 case BPF_FUNC_skc_lookup_tcp:
6179 return &bpf_xdp_skc_lookup_tcp_proto;
6180 case BPF_FUNC_tcp_check_syncookie:
6181 return &bpf_tcp_check_syncookie_proto;
6182 case BPF_FUNC_tcp_gen_syncookie:
6183 return &bpf_tcp_gen_syncookie_proto;
6186 return bpf_base_func_proto(func_id);
6190 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
6191 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
6193 static const struct bpf_func_proto *
6194 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6197 case BPF_FUNC_setsockopt:
6198 return &bpf_setsockopt_proto;
6199 case BPF_FUNC_getsockopt:
6200 return &bpf_getsockopt_proto;
6201 case BPF_FUNC_sock_ops_cb_flags_set:
6202 return &bpf_sock_ops_cb_flags_set_proto;
6203 case BPF_FUNC_sock_map_update:
6204 return &bpf_sock_map_update_proto;
6205 case BPF_FUNC_sock_hash_update:
6206 return &bpf_sock_hash_update_proto;
6207 case BPF_FUNC_get_socket_cookie:
6208 return &bpf_get_socket_cookie_sock_ops_proto;
6209 case BPF_FUNC_get_local_storage:
6210 return &bpf_get_local_storage_proto;
6211 case BPF_FUNC_perf_event_output:
6212 return &bpf_sockopt_event_output_proto;
6213 case BPF_FUNC_sk_storage_get:
6214 return &bpf_sk_storage_get_proto;
6215 case BPF_FUNC_sk_storage_delete:
6216 return &bpf_sk_storage_delete_proto;
6218 case BPF_FUNC_tcp_sock:
6219 return &bpf_tcp_sock_proto;
6220 #endif /* CONFIG_INET */
6222 return bpf_base_func_proto(func_id);
6226 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
6227 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
6229 static const struct bpf_func_proto *
6230 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6233 case BPF_FUNC_msg_redirect_map:
6234 return &bpf_msg_redirect_map_proto;
6235 case BPF_FUNC_msg_redirect_hash:
6236 return &bpf_msg_redirect_hash_proto;
6237 case BPF_FUNC_msg_apply_bytes:
6238 return &bpf_msg_apply_bytes_proto;
6239 case BPF_FUNC_msg_cork_bytes:
6240 return &bpf_msg_cork_bytes_proto;
6241 case BPF_FUNC_msg_pull_data:
6242 return &bpf_msg_pull_data_proto;
6243 case BPF_FUNC_msg_push_data:
6244 return &bpf_msg_push_data_proto;
6245 case BPF_FUNC_msg_pop_data:
6246 return &bpf_msg_pop_data_proto;
6248 return bpf_base_func_proto(func_id);
6252 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
6253 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
6255 static const struct bpf_func_proto *
6256 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6259 case BPF_FUNC_skb_store_bytes:
6260 return &bpf_skb_store_bytes_proto;
6261 case BPF_FUNC_skb_load_bytes:
6262 return &bpf_skb_load_bytes_proto;
6263 case BPF_FUNC_skb_pull_data:
6264 return &sk_skb_pull_data_proto;
6265 case BPF_FUNC_skb_change_tail:
6266 return &sk_skb_change_tail_proto;
6267 case BPF_FUNC_skb_change_head:
6268 return &sk_skb_change_head_proto;
6269 case BPF_FUNC_get_socket_cookie:
6270 return &bpf_get_socket_cookie_proto;
6271 case BPF_FUNC_get_socket_uid:
6272 return &bpf_get_socket_uid_proto;
6273 case BPF_FUNC_sk_redirect_map:
6274 return &bpf_sk_redirect_map_proto;
6275 case BPF_FUNC_sk_redirect_hash:
6276 return &bpf_sk_redirect_hash_proto;
6277 case BPF_FUNC_perf_event_output:
6278 return &bpf_skb_event_output_proto;
6280 case BPF_FUNC_sk_lookup_tcp:
6281 return &bpf_sk_lookup_tcp_proto;
6282 case BPF_FUNC_sk_lookup_udp:
6283 return &bpf_sk_lookup_udp_proto;
6284 case BPF_FUNC_sk_release:
6285 return &bpf_sk_release_proto;
6286 case BPF_FUNC_skc_lookup_tcp:
6287 return &bpf_skc_lookup_tcp_proto;
6290 return bpf_base_func_proto(func_id);
6294 static const struct bpf_func_proto *
6295 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6298 case BPF_FUNC_skb_load_bytes:
6299 return &bpf_flow_dissector_load_bytes_proto;
6301 return bpf_base_func_proto(func_id);
6305 static const struct bpf_func_proto *
6306 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6309 case BPF_FUNC_skb_load_bytes:
6310 return &bpf_skb_load_bytes_proto;
6311 case BPF_FUNC_skb_pull_data:
6312 return &bpf_skb_pull_data_proto;
6313 case BPF_FUNC_csum_diff:
6314 return &bpf_csum_diff_proto;
6315 case BPF_FUNC_get_cgroup_classid:
6316 return &bpf_get_cgroup_classid_proto;
6317 case BPF_FUNC_get_route_realm:
6318 return &bpf_get_route_realm_proto;
6319 case BPF_FUNC_get_hash_recalc:
6320 return &bpf_get_hash_recalc_proto;
6321 case BPF_FUNC_perf_event_output:
6322 return &bpf_skb_event_output_proto;
6323 case BPF_FUNC_get_smp_processor_id:
6324 return &bpf_get_smp_processor_id_proto;
6325 case BPF_FUNC_skb_under_cgroup:
6326 return &bpf_skb_under_cgroup_proto;
6328 return bpf_base_func_proto(func_id);
6332 static const struct bpf_func_proto *
6333 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6336 case BPF_FUNC_lwt_push_encap:
6337 return &bpf_lwt_in_push_encap_proto;
6339 return lwt_out_func_proto(func_id, prog);
6343 static const struct bpf_func_proto *
6344 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6347 case BPF_FUNC_skb_get_tunnel_key:
6348 return &bpf_skb_get_tunnel_key_proto;
6349 case BPF_FUNC_skb_set_tunnel_key:
6350 return bpf_get_skb_set_tunnel_proto(func_id);
6351 case BPF_FUNC_skb_get_tunnel_opt:
6352 return &bpf_skb_get_tunnel_opt_proto;
6353 case BPF_FUNC_skb_set_tunnel_opt:
6354 return bpf_get_skb_set_tunnel_proto(func_id);
6355 case BPF_FUNC_redirect:
6356 return &bpf_redirect_proto;
6357 case BPF_FUNC_clone_redirect:
6358 return &bpf_clone_redirect_proto;
6359 case BPF_FUNC_skb_change_tail:
6360 return &bpf_skb_change_tail_proto;
6361 case BPF_FUNC_skb_change_head:
6362 return &bpf_skb_change_head_proto;
6363 case BPF_FUNC_skb_store_bytes:
6364 return &bpf_skb_store_bytes_proto;
6365 case BPF_FUNC_csum_update:
6366 return &bpf_csum_update_proto;
6367 case BPF_FUNC_l3_csum_replace:
6368 return &bpf_l3_csum_replace_proto;
6369 case BPF_FUNC_l4_csum_replace:
6370 return &bpf_l4_csum_replace_proto;
6371 case BPF_FUNC_set_hash_invalid:
6372 return &bpf_set_hash_invalid_proto;
6373 case BPF_FUNC_lwt_push_encap:
6374 return &bpf_lwt_xmit_push_encap_proto;
6376 return lwt_out_func_proto(func_id, prog);
6380 static const struct bpf_func_proto *
6381 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6384 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6385 case BPF_FUNC_lwt_seg6_store_bytes:
6386 return &bpf_lwt_seg6_store_bytes_proto;
6387 case BPF_FUNC_lwt_seg6_action:
6388 return &bpf_lwt_seg6_action_proto;
6389 case BPF_FUNC_lwt_seg6_adjust_srh:
6390 return &bpf_lwt_seg6_adjust_srh_proto;
6393 return lwt_out_func_proto(func_id, prog);
6397 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
6398 const struct bpf_prog *prog,
6399 struct bpf_insn_access_aux *info)
6401 const int size_default = sizeof(__u32);
6403 if (off < 0 || off >= sizeof(struct __sk_buff))
6406 /* The verifier guarantees that size > 0. */
6407 if (off % size != 0)
6411 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6412 if (off + size > offsetofend(struct __sk_buff, cb[4]))
6415 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
6416 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
6417 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
6418 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
6419 case bpf_ctx_range(struct __sk_buff, data):
6420 case bpf_ctx_range(struct __sk_buff, data_meta):
6421 case bpf_ctx_range(struct __sk_buff, data_end):
6422 if (size != size_default)
6425 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
6427 case bpf_ctx_range(struct __sk_buff, tstamp):
6428 if (size != sizeof(__u64))
6431 case offsetof(struct __sk_buff, sk):
6432 if (type == BPF_WRITE || size != sizeof(__u64))
6434 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
6437 /* Only narrow read access allowed for now. */
6438 if (type == BPF_WRITE) {
6439 if (size != size_default)
6442 bpf_ctx_record_field_size(info, size_default);
6443 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
6451 static bool sk_filter_is_valid_access(int off, int size,
6452 enum bpf_access_type type,
6453 const struct bpf_prog *prog,
6454 struct bpf_insn_access_aux *info)
6457 case bpf_ctx_range(struct __sk_buff, tc_classid):
6458 case bpf_ctx_range(struct __sk_buff, data):
6459 case bpf_ctx_range(struct __sk_buff, data_meta):
6460 case bpf_ctx_range(struct __sk_buff, data_end):
6461 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6462 case bpf_ctx_range(struct __sk_buff, tstamp):
6463 case bpf_ctx_range(struct __sk_buff, wire_len):
6467 if (type == BPF_WRITE) {
6469 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6476 return bpf_skb_is_valid_access(off, size, type, prog, info);
6479 static bool cg_skb_is_valid_access(int off, int size,
6480 enum bpf_access_type type,
6481 const struct bpf_prog *prog,
6482 struct bpf_insn_access_aux *info)
6485 case bpf_ctx_range(struct __sk_buff, tc_classid):
6486 case bpf_ctx_range(struct __sk_buff, data_meta):
6487 case bpf_ctx_range(struct __sk_buff, wire_len):
6489 case bpf_ctx_range(struct __sk_buff, data):
6490 case bpf_ctx_range(struct __sk_buff, data_end):
6491 if (!capable(CAP_SYS_ADMIN))
6496 if (type == BPF_WRITE) {
6498 case bpf_ctx_range(struct __sk_buff, mark):
6499 case bpf_ctx_range(struct __sk_buff, priority):
6500 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6502 case bpf_ctx_range(struct __sk_buff, tstamp):
6503 if (!capable(CAP_SYS_ADMIN))
6512 case bpf_ctx_range(struct __sk_buff, data):
6513 info->reg_type = PTR_TO_PACKET;
6515 case bpf_ctx_range(struct __sk_buff, data_end):
6516 info->reg_type = PTR_TO_PACKET_END;
6520 return bpf_skb_is_valid_access(off, size, type, prog, info);
6523 static bool lwt_is_valid_access(int off, int size,
6524 enum bpf_access_type type,
6525 const struct bpf_prog *prog,
6526 struct bpf_insn_access_aux *info)
6529 case bpf_ctx_range(struct __sk_buff, tc_classid):
6530 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6531 case bpf_ctx_range(struct __sk_buff, data_meta):
6532 case bpf_ctx_range(struct __sk_buff, tstamp):
6533 case bpf_ctx_range(struct __sk_buff, wire_len):
6537 if (type == BPF_WRITE) {
6539 case bpf_ctx_range(struct __sk_buff, mark):
6540 case bpf_ctx_range(struct __sk_buff, priority):
6541 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6549 case bpf_ctx_range(struct __sk_buff, data):
6550 info->reg_type = PTR_TO_PACKET;
6552 case bpf_ctx_range(struct __sk_buff, data_end):
6553 info->reg_type = PTR_TO_PACKET_END;
6557 return bpf_skb_is_valid_access(off, size, type, prog, info);
6560 /* Attach type specific accesses */
6561 static bool __sock_filter_check_attach_type(int off,
6562 enum bpf_access_type access_type,
6563 enum bpf_attach_type attach_type)
6566 case offsetof(struct bpf_sock, bound_dev_if):
6567 case offsetof(struct bpf_sock, mark):
6568 case offsetof(struct bpf_sock, priority):
6569 switch (attach_type) {
6570 case BPF_CGROUP_INET_SOCK_CREATE:
6575 case bpf_ctx_range(struct bpf_sock, src_ip4):
6576 switch (attach_type) {
6577 case BPF_CGROUP_INET4_POST_BIND:
6582 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
6583 switch (attach_type) {
6584 case BPF_CGROUP_INET6_POST_BIND:
6589 case bpf_ctx_range(struct bpf_sock, src_port):
6590 switch (attach_type) {
6591 case BPF_CGROUP_INET4_POST_BIND:
6592 case BPF_CGROUP_INET6_POST_BIND:
6599 return access_type == BPF_READ;
6604 bool bpf_sock_common_is_valid_access(int off, int size,
6605 enum bpf_access_type type,
6606 struct bpf_insn_access_aux *info)
6609 case bpf_ctx_range_till(struct bpf_sock, type, priority):
6612 return bpf_sock_is_valid_access(off, size, type, info);
6616 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6617 struct bpf_insn_access_aux *info)
6619 const int size_default = sizeof(__u32);
6621 if (off < 0 || off >= sizeof(struct bpf_sock))
6623 if (off % size != 0)
6627 case offsetof(struct bpf_sock, state):
6628 case offsetof(struct bpf_sock, family):
6629 case offsetof(struct bpf_sock, type):
6630 case offsetof(struct bpf_sock, protocol):
6631 case offsetof(struct bpf_sock, dst_port):
6632 case offsetof(struct bpf_sock, src_port):
6633 case bpf_ctx_range(struct bpf_sock, src_ip4):
6634 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
6635 case bpf_ctx_range(struct bpf_sock, dst_ip4):
6636 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
6637 bpf_ctx_record_field_size(info, size_default);
6638 return bpf_ctx_narrow_access_ok(off, size, size_default);
6641 return size == size_default;
6644 static bool sock_filter_is_valid_access(int off, int size,
6645 enum bpf_access_type type,
6646 const struct bpf_prog *prog,
6647 struct bpf_insn_access_aux *info)
6649 if (!bpf_sock_is_valid_access(off, size, type, info))
6651 return __sock_filter_check_attach_type(off, type,
6652 prog->expected_attach_type);
6655 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
6656 const struct bpf_prog *prog)
6658 /* Neither direct read nor direct write requires any preliminary
6664 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
6665 const struct bpf_prog *prog, int drop_verdict)
6667 struct bpf_insn *insn = insn_buf;
6672 /* if (!skb->cloned)
6675 * (Fast-path, otherwise approximation that we might be
6676 * a clone, do the rest in helper.)
6678 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
6679 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
6680 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
6682 /* ret = bpf_skb_pull_data(skb, 0); */
6683 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
6684 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
6685 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
6686 BPF_FUNC_skb_pull_data);
6689 * return TC_ACT_SHOT;
6691 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
6692 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
6693 *insn++ = BPF_EXIT_INSN();
6696 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
6698 *insn++ = prog->insnsi[0];
6700 return insn - insn_buf;
6703 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
6704 struct bpf_insn *insn_buf)
6706 bool indirect = BPF_MODE(orig->code) == BPF_IND;
6707 struct bpf_insn *insn = insn_buf;
6709 /* We're guaranteed here that CTX is in R6. */
6710 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
6712 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
6714 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
6716 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
6719 switch (BPF_SIZE(orig->code)) {
6721 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
6724 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
6727 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
6731 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
6732 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
6733 *insn++ = BPF_EXIT_INSN();
6735 return insn - insn_buf;
6738 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
6739 const struct bpf_prog *prog)
6741 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
6744 static bool tc_cls_act_is_valid_access(int off, int size,
6745 enum bpf_access_type type,
6746 const struct bpf_prog *prog,
6747 struct bpf_insn_access_aux *info)
6749 if (type == BPF_WRITE) {
6751 case bpf_ctx_range(struct __sk_buff, mark):
6752 case bpf_ctx_range(struct __sk_buff, tc_index):
6753 case bpf_ctx_range(struct __sk_buff, priority):
6754 case bpf_ctx_range(struct __sk_buff, tc_classid):
6755 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6756 case bpf_ctx_range(struct __sk_buff, tstamp):
6757 case bpf_ctx_range(struct __sk_buff, queue_mapping):
6765 case bpf_ctx_range(struct __sk_buff, data):
6766 info->reg_type = PTR_TO_PACKET;
6768 case bpf_ctx_range(struct __sk_buff, data_meta):
6769 info->reg_type = PTR_TO_PACKET_META;
6771 case bpf_ctx_range(struct __sk_buff, data_end):
6772 info->reg_type = PTR_TO_PACKET_END;
6774 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6778 return bpf_skb_is_valid_access(off, size, type, prog, info);
6781 static bool __is_valid_xdp_access(int off, int size)
6783 if (off < 0 || off >= sizeof(struct xdp_md))
6785 if (off % size != 0)
6787 if (size != sizeof(__u32))
6793 static bool xdp_is_valid_access(int off, int size,
6794 enum bpf_access_type type,
6795 const struct bpf_prog *prog,
6796 struct bpf_insn_access_aux *info)
6798 if (type == BPF_WRITE) {
6799 if (bpf_prog_is_dev_bound(prog->aux)) {
6801 case offsetof(struct xdp_md, rx_queue_index):
6802 return __is_valid_xdp_access(off, size);
6809 case offsetof(struct xdp_md, data):
6810 info->reg_type = PTR_TO_PACKET;
6812 case offsetof(struct xdp_md, data_meta):
6813 info->reg_type = PTR_TO_PACKET_META;
6815 case offsetof(struct xdp_md, data_end):
6816 info->reg_type = PTR_TO_PACKET_END;
6820 return __is_valid_xdp_access(off, size);
6823 void bpf_warn_invalid_xdp_action(u32 act)
6825 const u32 act_max = XDP_REDIRECT;
6827 WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
6828 act > act_max ? "Illegal" : "Driver unsupported",
6831 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
6833 static bool sock_addr_is_valid_access(int off, int size,
6834 enum bpf_access_type type,
6835 const struct bpf_prog *prog,
6836 struct bpf_insn_access_aux *info)
6838 const int size_default = sizeof(__u32);
6840 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
6842 if (off % size != 0)
6845 /* Disallow access to IPv6 fields from IPv4 contex and vise
6849 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
6850 switch (prog->expected_attach_type) {
6851 case BPF_CGROUP_INET4_BIND:
6852 case BPF_CGROUP_INET4_CONNECT:
6853 case BPF_CGROUP_UDP4_SENDMSG:
6854 case BPF_CGROUP_UDP4_RECVMSG:
6860 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
6861 switch (prog->expected_attach_type) {
6862 case BPF_CGROUP_INET6_BIND:
6863 case BPF_CGROUP_INET6_CONNECT:
6864 case BPF_CGROUP_UDP6_SENDMSG:
6865 case BPF_CGROUP_UDP6_RECVMSG:
6871 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
6872 switch (prog->expected_attach_type) {
6873 case BPF_CGROUP_UDP4_SENDMSG:
6879 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
6881 switch (prog->expected_attach_type) {
6882 case BPF_CGROUP_UDP6_SENDMSG:
6891 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
6892 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
6893 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
6894 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
6896 if (type == BPF_READ) {
6897 bpf_ctx_record_field_size(info, size_default);
6899 if (bpf_ctx_wide_access_ok(off, size,
6900 struct bpf_sock_addr,
6904 if (bpf_ctx_wide_access_ok(off, size,
6905 struct bpf_sock_addr,
6909 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
6912 if (bpf_ctx_wide_access_ok(off, size,
6913 struct bpf_sock_addr,
6917 if (bpf_ctx_wide_access_ok(off, size,
6918 struct bpf_sock_addr,
6922 if (size != size_default)
6926 case bpf_ctx_range(struct bpf_sock_addr, user_port):
6927 if (size != size_default)
6930 case offsetof(struct bpf_sock_addr, sk):
6931 if (type != BPF_READ)
6933 if (size != sizeof(__u64))
6935 info->reg_type = PTR_TO_SOCKET;
6938 if (type == BPF_READ) {
6939 if (size != size_default)
6949 static bool sock_ops_is_valid_access(int off, int size,
6950 enum bpf_access_type type,
6951 const struct bpf_prog *prog,
6952 struct bpf_insn_access_aux *info)
6954 const int size_default = sizeof(__u32);
6956 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
6959 /* The verifier guarantees that size > 0. */
6960 if (off % size != 0)
6963 if (type == BPF_WRITE) {
6965 case offsetof(struct bpf_sock_ops, reply):
6966 case offsetof(struct bpf_sock_ops, sk_txhash):
6967 if (size != size_default)
6975 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
6977 if (size != sizeof(__u64))
6980 case offsetof(struct bpf_sock_ops, sk):
6981 if (size != sizeof(__u64))
6983 info->reg_type = PTR_TO_SOCKET_OR_NULL;
6986 if (size != size_default)
6995 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
6996 const struct bpf_prog *prog)
6998 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
7001 static bool sk_skb_is_valid_access(int off, int size,
7002 enum bpf_access_type type,
7003 const struct bpf_prog *prog,
7004 struct bpf_insn_access_aux *info)
7007 case bpf_ctx_range(struct __sk_buff, tc_classid):
7008 case bpf_ctx_range(struct __sk_buff, data_meta):
7009 case bpf_ctx_range(struct __sk_buff, tstamp):
7010 case bpf_ctx_range(struct __sk_buff, wire_len):
7014 if (type == BPF_WRITE) {
7016 case bpf_ctx_range(struct __sk_buff, tc_index):
7017 case bpf_ctx_range(struct __sk_buff, priority):
7025 case bpf_ctx_range(struct __sk_buff, mark):
7027 case bpf_ctx_range(struct __sk_buff, data):
7028 info->reg_type = PTR_TO_PACKET;
7030 case bpf_ctx_range(struct __sk_buff, data_end):
7031 info->reg_type = PTR_TO_PACKET_END;
7035 return bpf_skb_is_valid_access(off, size, type, prog, info);
7038 static bool sk_msg_is_valid_access(int off, int size,
7039 enum bpf_access_type type,
7040 const struct bpf_prog *prog,
7041 struct bpf_insn_access_aux *info)
7043 if (type == BPF_WRITE)
7046 if (off % size != 0)
7050 case offsetof(struct sk_msg_md, data):
7051 info->reg_type = PTR_TO_PACKET;
7052 if (size != sizeof(__u64))
7055 case offsetof(struct sk_msg_md, data_end):
7056 info->reg_type = PTR_TO_PACKET_END;
7057 if (size != sizeof(__u64))
7060 case bpf_ctx_range(struct sk_msg_md, family):
7061 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
7062 case bpf_ctx_range(struct sk_msg_md, local_ip4):
7063 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
7064 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
7065 case bpf_ctx_range(struct sk_msg_md, remote_port):
7066 case bpf_ctx_range(struct sk_msg_md, local_port):
7067 case bpf_ctx_range(struct sk_msg_md, size):
7068 if (size != sizeof(__u32))
7077 static bool flow_dissector_is_valid_access(int off, int size,
7078 enum bpf_access_type type,
7079 const struct bpf_prog *prog,
7080 struct bpf_insn_access_aux *info)
7082 const int size_default = sizeof(__u32);
7084 if (off < 0 || off >= sizeof(struct __sk_buff))
7087 if (type == BPF_WRITE)
7091 case bpf_ctx_range(struct __sk_buff, data):
7092 if (size != size_default)
7094 info->reg_type = PTR_TO_PACKET;
7096 case bpf_ctx_range(struct __sk_buff, data_end):
7097 if (size != size_default)
7099 info->reg_type = PTR_TO_PACKET_END;
7101 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
7102 if (size != sizeof(__u64))
7104 info->reg_type = PTR_TO_FLOW_KEYS;
7111 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
7112 const struct bpf_insn *si,
7113 struct bpf_insn *insn_buf,
7114 struct bpf_prog *prog,
7118 struct bpf_insn *insn = insn_buf;
7121 case offsetof(struct __sk_buff, data):
7122 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
7123 si->dst_reg, si->src_reg,
7124 offsetof(struct bpf_flow_dissector, data));
7127 case offsetof(struct __sk_buff, data_end):
7128 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
7129 si->dst_reg, si->src_reg,
7130 offsetof(struct bpf_flow_dissector, data_end));
7133 case offsetof(struct __sk_buff, flow_keys):
7134 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
7135 si->dst_reg, si->src_reg,
7136 offsetof(struct bpf_flow_dissector, flow_keys));
7140 return insn - insn_buf;
7143 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
7144 const struct bpf_insn *si,
7145 struct bpf_insn *insn_buf,
7146 struct bpf_prog *prog, u32 *target_size)
7148 struct bpf_insn *insn = insn_buf;
7152 case offsetof(struct __sk_buff, len):
7153 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7154 bpf_target_off(struct sk_buff, len, 4,
7158 case offsetof(struct __sk_buff, protocol):
7159 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
7160 bpf_target_off(struct sk_buff, protocol, 2,
7164 case offsetof(struct __sk_buff, vlan_proto):
7165 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
7166 bpf_target_off(struct sk_buff, vlan_proto, 2,
7170 case offsetof(struct __sk_buff, priority):
7171 if (type == BPF_WRITE)
7172 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7173 bpf_target_off(struct sk_buff, priority, 4,
7176 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7177 bpf_target_off(struct sk_buff, priority, 4,
7181 case offsetof(struct __sk_buff, ingress_ifindex):
7182 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7183 bpf_target_off(struct sk_buff, skb_iif, 4,
7187 case offsetof(struct __sk_buff, ifindex):
7188 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
7189 si->dst_reg, si->src_reg,
7190 offsetof(struct sk_buff, dev));
7191 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
7192 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7193 bpf_target_off(struct net_device, ifindex, 4,
7197 case offsetof(struct __sk_buff, hash):
7198 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7199 bpf_target_off(struct sk_buff, hash, 4,
7203 case offsetof(struct __sk_buff, mark):
7204 if (type == BPF_WRITE)
7205 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7206 bpf_target_off(struct sk_buff, mark, 4,
7209 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7210 bpf_target_off(struct sk_buff, mark, 4,
7214 case offsetof(struct __sk_buff, pkt_type):
7216 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
7218 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
7219 #ifdef __BIG_ENDIAN_BITFIELD
7220 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
7224 case offsetof(struct __sk_buff, queue_mapping):
7225 if (type == BPF_WRITE) {
7226 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
7227 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
7228 bpf_target_off(struct sk_buff,
7232 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
7233 bpf_target_off(struct sk_buff,
7239 case offsetof(struct __sk_buff, vlan_present):
7241 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
7242 PKT_VLAN_PRESENT_OFFSET());
7243 if (PKT_VLAN_PRESENT_BIT)
7244 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
7245 if (PKT_VLAN_PRESENT_BIT < 7)
7246 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
7249 case offsetof(struct __sk_buff, vlan_tci):
7250 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
7251 bpf_target_off(struct sk_buff, vlan_tci, 2,
7255 case offsetof(struct __sk_buff, cb[0]) ...
7256 offsetofend(struct __sk_buff, cb[4]) - 1:
7257 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
7258 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
7259 offsetof(struct qdisc_skb_cb, data)) %
7262 prog->cb_access = 1;
7264 off -= offsetof(struct __sk_buff, cb[0]);
7265 off += offsetof(struct sk_buff, cb);
7266 off += offsetof(struct qdisc_skb_cb, data);
7267 if (type == BPF_WRITE)
7268 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
7271 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
7275 case offsetof(struct __sk_buff, tc_classid):
7276 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
7279 off -= offsetof(struct __sk_buff, tc_classid);
7280 off += offsetof(struct sk_buff, cb);
7281 off += offsetof(struct qdisc_skb_cb, tc_classid);
7283 if (type == BPF_WRITE)
7284 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
7287 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
7291 case offsetof(struct __sk_buff, data):
7292 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
7293 si->dst_reg, si->src_reg,
7294 offsetof(struct sk_buff, data));
7297 case offsetof(struct __sk_buff, data_meta):
7299 off -= offsetof(struct __sk_buff, data_meta);
7300 off += offsetof(struct sk_buff, cb);
7301 off += offsetof(struct bpf_skb_data_end, data_meta);
7302 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
7306 case offsetof(struct __sk_buff, data_end):
7308 off -= offsetof(struct __sk_buff, data_end);
7309 off += offsetof(struct sk_buff, cb);
7310 off += offsetof(struct bpf_skb_data_end, data_end);
7311 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
7315 case offsetof(struct __sk_buff, tc_index):
7316 #ifdef CONFIG_NET_SCHED
7317 if (type == BPF_WRITE)
7318 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
7319 bpf_target_off(struct sk_buff, tc_index, 2,
7322 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
7323 bpf_target_off(struct sk_buff, tc_index, 2,
7327 if (type == BPF_WRITE)
7328 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
7330 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
7334 case offsetof(struct __sk_buff, napi_id):
7335 #if defined(CONFIG_NET_RX_BUSY_POLL)
7336 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7337 bpf_target_off(struct sk_buff, napi_id, 4,
7339 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
7340 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
7343 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
7346 case offsetof(struct __sk_buff, family):
7347 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
7349 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7350 si->dst_reg, si->src_reg,
7351 offsetof(struct sk_buff, sk));
7352 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7353 bpf_target_off(struct sock_common,
7357 case offsetof(struct __sk_buff, remote_ip4):
7358 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
7360 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7361 si->dst_reg, si->src_reg,
7362 offsetof(struct sk_buff, sk));
7363 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7364 bpf_target_off(struct sock_common,
7368 case offsetof(struct __sk_buff, local_ip4):
7369 BUILD_BUG_ON(sizeof_field(struct sock_common,
7370 skc_rcv_saddr) != 4);
7372 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7373 si->dst_reg, si->src_reg,
7374 offsetof(struct sk_buff, sk));
7375 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7376 bpf_target_off(struct sock_common,
7380 case offsetof(struct __sk_buff, remote_ip6[0]) ...
7381 offsetof(struct __sk_buff, remote_ip6[3]):
7382 #if IS_ENABLED(CONFIG_IPV6)
7383 BUILD_BUG_ON(sizeof_field(struct sock_common,
7384 skc_v6_daddr.s6_addr32[0]) != 4);
7387 off -= offsetof(struct __sk_buff, remote_ip6[0]);
7389 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7390 si->dst_reg, si->src_reg,
7391 offsetof(struct sk_buff, sk));
7392 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7393 offsetof(struct sock_common,
7394 skc_v6_daddr.s6_addr32[0]) +
7397 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7400 case offsetof(struct __sk_buff, local_ip6[0]) ...
7401 offsetof(struct __sk_buff, local_ip6[3]):
7402 #if IS_ENABLED(CONFIG_IPV6)
7403 BUILD_BUG_ON(sizeof_field(struct sock_common,
7404 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
7407 off -= offsetof(struct __sk_buff, local_ip6[0]);
7409 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7410 si->dst_reg, si->src_reg,
7411 offsetof(struct sk_buff, sk));
7412 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7413 offsetof(struct sock_common,
7414 skc_v6_rcv_saddr.s6_addr32[0]) +
7417 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7421 case offsetof(struct __sk_buff, remote_port):
7422 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
7424 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7425 si->dst_reg, si->src_reg,
7426 offsetof(struct sk_buff, sk));
7427 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7428 bpf_target_off(struct sock_common,
7431 #ifndef __BIG_ENDIAN_BITFIELD
7432 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
7436 case offsetof(struct __sk_buff, local_port):
7437 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
7439 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7440 si->dst_reg, si->src_reg,
7441 offsetof(struct sk_buff, sk));
7442 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7443 bpf_target_off(struct sock_common,
7444 skc_num, 2, target_size));
7447 case offsetof(struct __sk_buff, tstamp):
7448 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
7450 if (type == BPF_WRITE)
7451 *insn++ = BPF_STX_MEM(BPF_DW,
7452 si->dst_reg, si->src_reg,
7453 bpf_target_off(struct sk_buff,
7457 *insn++ = BPF_LDX_MEM(BPF_DW,
7458 si->dst_reg, si->src_reg,
7459 bpf_target_off(struct sk_buff,
7464 case offsetof(struct __sk_buff, gso_segs):
7465 /* si->dst_reg = skb_shinfo(SKB); */
7466 #ifdef NET_SKBUFF_DATA_USES_OFFSET
7467 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
7468 BPF_REG_AX, si->src_reg,
7469 offsetof(struct sk_buff, end));
7470 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
7471 si->dst_reg, si->src_reg,
7472 offsetof(struct sk_buff, head));
7473 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
7475 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
7476 si->dst_reg, si->src_reg,
7477 offsetof(struct sk_buff, end));
7479 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
7480 si->dst_reg, si->dst_reg,
7481 bpf_target_off(struct skb_shared_info,
7485 case offsetof(struct __sk_buff, wire_len):
7486 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
7489 off -= offsetof(struct __sk_buff, wire_len);
7490 off += offsetof(struct sk_buff, cb);
7491 off += offsetof(struct qdisc_skb_cb, pkt_len);
7493 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
7496 case offsetof(struct __sk_buff, sk):
7497 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7498 si->dst_reg, si->src_reg,
7499 offsetof(struct sk_buff, sk));
7503 return insn - insn_buf;
7506 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
7507 const struct bpf_insn *si,
7508 struct bpf_insn *insn_buf,
7509 struct bpf_prog *prog, u32 *target_size)
7511 struct bpf_insn *insn = insn_buf;
7515 case offsetof(struct bpf_sock, bound_dev_if):
7516 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
7518 if (type == BPF_WRITE)
7519 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7520 offsetof(struct sock, sk_bound_dev_if));
7522 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7523 offsetof(struct sock, sk_bound_dev_if));
7526 case offsetof(struct bpf_sock, mark):
7527 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
7529 if (type == BPF_WRITE)
7530 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7531 offsetof(struct sock, sk_mark));
7533 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7534 offsetof(struct sock, sk_mark));
7537 case offsetof(struct bpf_sock, priority):
7538 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
7540 if (type == BPF_WRITE)
7541 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7542 offsetof(struct sock, sk_priority));
7544 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7545 offsetof(struct sock, sk_priority));
7548 case offsetof(struct bpf_sock, family):
7549 *insn++ = BPF_LDX_MEM(
7550 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
7551 si->dst_reg, si->src_reg,
7552 bpf_target_off(struct sock_common,
7554 sizeof_field(struct sock_common,
7559 case offsetof(struct bpf_sock, type):
7560 *insn++ = BPF_LDX_MEM(
7561 BPF_FIELD_SIZEOF(struct sock, sk_type),
7562 si->dst_reg, si->src_reg,
7563 bpf_target_off(struct sock, sk_type,
7564 sizeof_field(struct sock, sk_type),
7568 case offsetof(struct bpf_sock, protocol):
7569 *insn++ = BPF_LDX_MEM(
7570 BPF_FIELD_SIZEOF(struct sock, sk_protocol),
7571 si->dst_reg, si->src_reg,
7572 bpf_target_off(struct sock, sk_protocol,
7573 sizeof_field(struct sock, sk_protocol),
7577 case offsetof(struct bpf_sock, src_ip4):
7578 *insn++ = BPF_LDX_MEM(
7579 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7580 bpf_target_off(struct sock_common, skc_rcv_saddr,
7581 sizeof_field(struct sock_common,
7586 case offsetof(struct bpf_sock, dst_ip4):
7587 *insn++ = BPF_LDX_MEM(
7588 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7589 bpf_target_off(struct sock_common, skc_daddr,
7590 sizeof_field(struct sock_common,
7595 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7596 #if IS_ENABLED(CONFIG_IPV6)
7598 off -= offsetof(struct bpf_sock, src_ip6[0]);
7599 *insn++ = BPF_LDX_MEM(
7600 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7603 skc_v6_rcv_saddr.s6_addr32[0],
7604 sizeof_field(struct sock_common,
7605 skc_v6_rcv_saddr.s6_addr32[0]),
7606 target_size) + off);
7609 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7613 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
7614 #if IS_ENABLED(CONFIG_IPV6)
7616 off -= offsetof(struct bpf_sock, dst_ip6[0]);
7617 *insn++ = BPF_LDX_MEM(
7618 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7619 bpf_target_off(struct sock_common,
7620 skc_v6_daddr.s6_addr32[0],
7621 sizeof_field(struct sock_common,
7622 skc_v6_daddr.s6_addr32[0]),
7623 target_size) + off);
7625 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7630 case offsetof(struct bpf_sock, src_port):
7631 *insn++ = BPF_LDX_MEM(
7632 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
7633 si->dst_reg, si->src_reg,
7634 bpf_target_off(struct sock_common, skc_num,
7635 sizeof_field(struct sock_common,
7640 case offsetof(struct bpf_sock, dst_port):
7641 *insn++ = BPF_LDX_MEM(
7642 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
7643 si->dst_reg, si->src_reg,
7644 bpf_target_off(struct sock_common, skc_dport,
7645 sizeof_field(struct sock_common,
7650 case offsetof(struct bpf_sock, state):
7651 *insn++ = BPF_LDX_MEM(
7652 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
7653 si->dst_reg, si->src_reg,
7654 bpf_target_off(struct sock_common, skc_state,
7655 sizeof_field(struct sock_common,
7661 return insn - insn_buf;
7664 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
7665 const struct bpf_insn *si,
7666 struct bpf_insn *insn_buf,
7667 struct bpf_prog *prog, u32 *target_size)
7669 struct bpf_insn *insn = insn_buf;
7672 case offsetof(struct __sk_buff, ifindex):
7673 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
7674 si->dst_reg, si->src_reg,
7675 offsetof(struct sk_buff, dev));
7676 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7677 bpf_target_off(struct net_device, ifindex, 4,
7681 return bpf_convert_ctx_access(type, si, insn_buf, prog,
7685 return insn - insn_buf;
7688 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
7689 const struct bpf_insn *si,
7690 struct bpf_insn *insn_buf,
7691 struct bpf_prog *prog, u32 *target_size)
7693 struct bpf_insn *insn = insn_buf;
7696 case offsetof(struct xdp_md, data):
7697 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
7698 si->dst_reg, si->src_reg,
7699 offsetof(struct xdp_buff, data));
7701 case offsetof(struct xdp_md, data_meta):
7702 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
7703 si->dst_reg, si->src_reg,
7704 offsetof(struct xdp_buff, data_meta));
7706 case offsetof(struct xdp_md, data_end):
7707 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
7708 si->dst_reg, si->src_reg,
7709 offsetof(struct xdp_buff, data_end));
7711 case offsetof(struct xdp_md, ingress_ifindex):
7712 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
7713 si->dst_reg, si->src_reg,
7714 offsetof(struct xdp_buff, rxq));
7715 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
7716 si->dst_reg, si->dst_reg,
7717 offsetof(struct xdp_rxq_info, dev));
7718 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7719 offsetof(struct net_device, ifindex));
7721 case offsetof(struct xdp_md, rx_queue_index):
7722 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
7723 si->dst_reg, si->src_reg,
7724 offsetof(struct xdp_buff, rxq));
7725 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7726 offsetof(struct xdp_rxq_info,
7731 return insn - insn_buf;
7734 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
7735 * context Structure, F is Field in context structure that contains a pointer
7736 * to Nested Structure of type NS that has the field NF.
7738 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
7739 * sure that SIZE is not greater than actual size of S.F.NF.
7741 * If offset OFF is provided, the load happens from that offset relative to
7744 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
7746 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
7747 si->src_reg, offsetof(S, F)); \
7748 *insn++ = BPF_LDX_MEM( \
7749 SIZE, si->dst_reg, si->dst_reg, \
7750 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
7755 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
7756 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
7757 BPF_FIELD_SIZEOF(NS, NF), 0)
7759 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
7760 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
7762 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
7763 * "register" since two registers available in convert_ctx_access are not
7764 * enough: we can't override neither SRC, since it contains value to store, nor
7765 * DST since it contains pointer to context that may be used by later
7766 * instructions. But we need a temporary place to save pointer to nested
7767 * structure whose field we want to store to.
7769 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
7771 int tmp_reg = BPF_REG_9; \
7772 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
7774 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
7776 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
7778 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
7779 si->dst_reg, offsetof(S, F)); \
7780 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \
7781 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
7784 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
7788 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
7791 if (type == BPF_WRITE) { \
7792 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
7795 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
7796 S, NS, F, NF, SIZE, OFF); \
7800 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
7801 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
7802 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
7804 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
7805 const struct bpf_insn *si,
7806 struct bpf_insn *insn_buf,
7807 struct bpf_prog *prog, u32 *target_size)
7809 struct bpf_insn *insn = insn_buf;
7813 case offsetof(struct bpf_sock_addr, user_family):
7814 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
7815 struct sockaddr, uaddr, sa_family);
7818 case offsetof(struct bpf_sock_addr, user_ip4):
7819 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7820 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
7821 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
7824 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
7826 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
7827 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7828 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
7829 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
7833 case offsetof(struct bpf_sock_addr, user_port):
7834 /* To get port we need to know sa_family first and then treat
7835 * sockaddr as either sockaddr_in or sockaddr_in6.
7836 * Though we can simplify since port field has same offset and
7837 * size in both structures.
7838 * Here we check this invariant and use just one of the
7839 * structures if it's true.
7841 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
7842 offsetof(struct sockaddr_in6, sin6_port));
7843 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
7844 sizeof_field(struct sockaddr_in6, sin6_port));
7845 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(struct bpf_sock_addr_kern,
7846 struct sockaddr_in6, uaddr,
7847 sin6_port, tmp_reg);
7850 case offsetof(struct bpf_sock_addr, family):
7851 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
7852 struct sock, sk, sk_family);
7855 case offsetof(struct bpf_sock_addr, type):
7856 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
7857 struct sock, sk, sk_type);
7860 case offsetof(struct bpf_sock_addr, protocol):
7861 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
7862 struct sock, sk, sk_protocol);
7865 case offsetof(struct bpf_sock_addr, msg_src_ip4):
7866 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
7867 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7868 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
7869 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
7872 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
7875 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
7876 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
7877 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7878 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
7879 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
7881 case offsetof(struct bpf_sock_addr, sk):
7882 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
7883 si->dst_reg, si->src_reg,
7884 offsetof(struct bpf_sock_addr_kern, sk));
7888 return insn - insn_buf;
7891 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
7892 const struct bpf_insn *si,
7893 struct bpf_insn *insn_buf,
7894 struct bpf_prog *prog,
7897 struct bpf_insn *insn = insn_buf;
7900 /* Helper macro for adding read access to tcp_sock or sock fields. */
7901 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
7903 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
7904 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
7905 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
7906 struct bpf_sock_ops_kern, \
7908 si->dst_reg, si->src_reg, \
7909 offsetof(struct bpf_sock_ops_kern, \
7911 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 2); \
7912 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
7913 struct bpf_sock_ops_kern, sk),\
7914 si->dst_reg, si->src_reg, \
7915 offsetof(struct bpf_sock_ops_kern, sk));\
7916 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
7918 si->dst_reg, si->dst_reg, \
7919 offsetof(OBJ, OBJ_FIELD)); \
7922 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
7923 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
7925 /* Helper macro for adding write access to tcp_sock or sock fields.
7926 * The macro is called with two registers, dst_reg which contains a pointer
7927 * to ctx (context) and src_reg which contains the value that should be
7928 * stored. However, we need an additional register since we cannot overwrite
7929 * dst_reg because it may be used later in the program.
7930 * Instead we "borrow" one of the other register. We first save its value
7931 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
7932 * it at the end of the macro.
7934 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
7936 int reg = BPF_REG_9; \
7937 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
7938 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
7939 if (si->dst_reg == reg || si->src_reg == reg) \
7941 if (si->dst_reg == reg || si->src_reg == reg) \
7943 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
7944 offsetof(struct bpf_sock_ops_kern, \
7946 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
7947 struct bpf_sock_ops_kern, \
7950 offsetof(struct bpf_sock_ops_kern, \
7952 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
7953 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
7954 struct bpf_sock_ops_kern, sk),\
7956 offsetof(struct bpf_sock_ops_kern, sk));\
7957 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \
7959 offsetof(OBJ, OBJ_FIELD)); \
7960 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
7961 offsetof(struct bpf_sock_ops_kern, \
7965 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
7967 if (TYPE == BPF_WRITE) \
7968 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
7970 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
7973 if (insn > insn_buf)
7974 return insn - insn_buf;
7977 case offsetof(struct bpf_sock_ops, op) ...
7978 offsetof(struct bpf_sock_ops, replylong[3]):
7979 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, op) !=
7980 sizeof_field(struct bpf_sock_ops_kern, op));
7981 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
7982 sizeof_field(struct bpf_sock_ops_kern, reply));
7983 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
7984 sizeof_field(struct bpf_sock_ops_kern, replylong));
7986 off -= offsetof(struct bpf_sock_ops, op);
7987 off += offsetof(struct bpf_sock_ops_kern, op);
7988 if (type == BPF_WRITE)
7989 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7992 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7996 case offsetof(struct bpf_sock_ops, family):
7997 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
7999 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8000 struct bpf_sock_ops_kern, sk),
8001 si->dst_reg, si->src_reg,
8002 offsetof(struct bpf_sock_ops_kern, sk));
8003 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8004 offsetof(struct sock_common, skc_family));
8007 case offsetof(struct bpf_sock_ops, remote_ip4):
8008 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
8010 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8011 struct bpf_sock_ops_kern, sk),
8012 si->dst_reg, si->src_reg,
8013 offsetof(struct bpf_sock_ops_kern, sk));
8014 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8015 offsetof(struct sock_common, skc_daddr));
8018 case offsetof(struct bpf_sock_ops, local_ip4):
8019 BUILD_BUG_ON(sizeof_field(struct sock_common,
8020 skc_rcv_saddr) != 4);
8022 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8023 struct bpf_sock_ops_kern, sk),
8024 si->dst_reg, si->src_reg,
8025 offsetof(struct bpf_sock_ops_kern, sk));
8026 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8027 offsetof(struct sock_common,
8031 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
8032 offsetof(struct bpf_sock_ops, remote_ip6[3]):
8033 #if IS_ENABLED(CONFIG_IPV6)
8034 BUILD_BUG_ON(sizeof_field(struct sock_common,
8035 skc_v6_daddr.s6_addr32[0]) != 4);
8038 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
8039 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8040 struct bpf_sock_ops_kern, sk),
8041 si->dst_reg, si->src_reg,
8042 offsetof(struct bpf_sock_ops_kern, sk));
8043 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8044 offsetof(struct sock_common,
8045 skc_v6_daddr.s6_addr32[0]) +
8048 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8052 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
8053 offsetof(struct bpf_sock_ops, local_ip6[3]):
8054 #if IS_ENABLED(CONFIG_IPV6)
8055 BUILD_BUG_ON(sizeof_field(struct sock_common,
8056 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8059 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
8060 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8061 struct bpf_sock_ops_kern, sk),
8062 si->dst_reg, si->src_reg,
8063 offsetof(struct bpf_sock_ops_kern, sk));
8064 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8065 offsetof(struct sock_common,
8066 skc_v6_rcv_saddr.s6_addr32[0]) +
8069 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8073 case offsetof(struct bpf_sock_ops, remote_port):
8074 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
8076 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8077 struct bpf_sock_ops_kern, sk),
8078 si->dst_reg, si->src_reg,
8079 offsetof(struct bpf_sock_ops_kern, sk));
8080 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8081 offsetof(struct sock_common, skc_dport));
8082 #ifndef __BIG_ENDIAN_BITFIELD
8083 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8087 case offsetof(struct bpf_sock_ops, local_port):
8088 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
8090 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8091 struct bpf_sock_ops_kern, sk),
8092 si->dst_reg, si->src_reg,
8093 offsetof(struct bpf_sock_ops_kern, sk));
8094 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8095 offsetof(struct sock_common, skc_num));
8098 case offsetof(struct bpf_sock_ops, is_fullsock):
8099 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8100 struct bpf_sock_ops_kern,
8102 si->dst_reg, si->src_reg,
8103 offsetof(struct bpf_sock_ops_kern,
8107 case offsetof(struct bpf_sock_ops, state):
8108 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
8110 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8111 struct bpf_sock_ops_kern, sk),
8112 si->dst_reg, si->src_reg,
8113 offsetof(struct bpf_sock_ops_kern, sk));
8114 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
8115 offsetof(struct sock_common, skc_state));
8118 case offsetof(struct bpf_sock_ops, rtt_min):
8119 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
8120 sizeof(struct minmax));
8121 BUILD_BUG_ON(sizeof(struct minmax) <
8122 sizeof(struct minmax_sample));
8124 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8125 struct bpf_sock_ops_kern, sk),
8126 si->dst_reg, si->src_reg,
8127 offsetof(struct bpf_sock_ops_kern, sk));
8128 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8129 offsetof(struct tcp_sock, rtt_min) +
8130 sizeof_field(struct minmax_sample, t));
8133 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
8134 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
8138 case offsetof(struct bpf_sock_ops, sk_txhash):
8139 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
8142 case offsetof(struct bpf_sock_ops, snd_cwnd):
8143 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
8145 case offsetof(struct bpf_sock_ops, srtt_us):
8146 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
8148 case offsetof(struct bpf_sock_ops, snd_ssthresh):
8149 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
8151 case offsetof(struct bpf_sock_ops, rcv_nxt):
8152 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
8154 case offsetof(struct bpf_sock_ops, snd_nxt):
8155 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
8157 case offsetof(struct bpf_sock_ops, snd_una):
8158 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
8160 case offsetof(struct bpf_sock_ops, mss_cache):
8161 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
8163 case offsetof(struct bpf_sock_ops, ecn_flags):
8164 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
8166 case offsetof(struct bpf_sock_ops, rate_delivered):
8167 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
8169 case offsetof(struct bpf_sock_ops, rate_interval_us):
8170 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
8172 case offsetof(struct bpf_sock_ops, packets_out):
8173 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
8175 case offsetof(struct bpf_sock_ops, retrans_out):
8176 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
8178 case offsetof(struct bpf_sock_ops, total_retrans):
8179 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
8181 case offsetof(struct bpf_sock_ops, segs_in):
8182 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
8184 case offsetof(struct bpf_sock_ops, data_segs_in):
8185 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
8187 case offsetof(struct bpf_sock_ops, segs_out):
8188 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
8190 case offsetof(struct bpf_sock_ops, data_segs_out):
8191 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
8193 case offsetof(struct bpf_sock_ops, lost_out):
8194 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
8196 case offsetof(struct bpf_sock_ops, sacked_out):
8197 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
8199 case offsetof(struct bpf_sock_ops, bytes_received):
8200 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
8202 case offsetof(struct bpf_sock_ops, bytes_acked):
8203 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
8205 case offsetof(struct bpf_sock_ops, sk):
8206 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8207 struct bpf_sock_ops_kern,
8209 si->dst_reg, si->src_reg,
8210 offsetof(struct bpf_sock_ops_kern,
8212 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
8213 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8214 struct bpf_sock_ops_kern, sk),
8215 si->dst_reg, si->src_reg,
8216 offsetof(struct bpf_sock_ops_kern, sk));
8219 return insn - insn_buf;
8222 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
8223 const struct bpf_insn *si,
8224 struct bpf_insn *insn_buf,
8225 struct bpf_prog *prog, u32 *target_size)
8227 struct bpf_insn *insn = insn_buf;
8231 case offsetof(struct __sk_buff, data_end):
8233 off -= offsetof(struct __sk_buff, data_end);
8234 off += offsetof(struct sk_buff, cb);
8235 off += offsetof(struct tcp_skb_cb, bpf.data_end);
8236 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8240 return bpf_convert_ctx_access(type, si, insn_buf, prog,
8244 return insn - insn_buf;
8247 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
8248 const struct bpf_insn *si,
8249 struct bpf_insn *insn_buf,
8250 struct bpf_prog *prog, u32 *target_size)
8252 struct bpf_insn *insn = insn_buf;
8253 #if IS_ENABLED(CONFIG_IPV6)
8257 /* convert ctx uses the fact sg element is first in struct */
8258 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
8261 case offsetof(struct sk_msg_md, data):
8262 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
8263 si->dst_reg, si->src_reg,
8264 offsetof(struct sk_msg, data));
8266 case offsetof(struct sk_msg_md, data_end):
8267 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
8268 si->dst_reg, si->src_reg,
8269 offsetof(struct sk_msg, data_end));
8271 case offsetof(struct sk_msg_md, family):
8272 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
8274 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8276 si->dst_reg, si->src_reg,
8277 offsetof(struct sk_msg, sk));
8278 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8279 offsetof(struct sock_common, skc_family));
8282 case offsetof(struct sk_msg_md, remote_ip4):
8283 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
8285 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8287 si->dst_reg, si->src_reg,
8288 offsetof(struct sk_msg, sk));
8289 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8290 offsetof(struct sock_common, skc_daddr));
8293 case offsetof(struct sk_msg_md, local_ip4):
8294 BUILD_BUG_ON(sizeof_field(struct sock_common,
8295 skc_rcv_saddr) != 4);
8297 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8299 si->dst_reg, si->src_reg,
8300 offsetof(struct sk_msg, sk));
8301 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8302 offsetof(struct sock_common,
8306 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
8307 offsetof(struct sk_msg_md, remote_ip6[3]):
8308 #if IS_ENABLED(CONFIG_IPV6)
8309 BUILD_BUG_ON(sizeof_field(struct sock_common,
8310 skc_v6_daddr.s6_addr32[0]) != 4);
8313 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
8314 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8316 si->dst_reg, si->src_reg,
8317 offsetof(struct sk_msg, sk));
8318 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8319 offsetof(struct sock_common,
8320 skc_v6_daddr.s6_addr32[0]) +
8323 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8327 case offsetof(struct sk_msg_md, local_ip6[0]) ...
8328 offsetof(struct sk_msg_md, local_ip6[3]):
8329 #if IS_ENABLED(CONFIG_IPV6)
8330 BUILD_BUG_ON(sizeof_field(struct sock_common,
8331 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8334 off -= offsetof(struct sk_msg_md, local_ip6[0]);
8335 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8337 si->dst_reg, si->src_reg,
8338 offsetof(struct sk_msg, sk));
8339 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8340 offsetof(struct sock_common,
8341 skc_v6_rcv_saddr.s6_addr32[0]) +
8344 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8348 case offsetof(struct sk_msg_md, remote_port):
8349 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
8351 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8353 si->dst_reg, si->src_reg,
8354 offsetof(struct sk_msg, sk));
8355 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8356 offsetof(struct sock_common, skc_dport));
8357 #ifndef __BIG_ENDIAN_BITFIELD
8358 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8362 case offsetof(struct sk_msg_md, local_port):
8363 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
8365 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8367 si->dst_reg, si->src_reg,
8368 offsetof(struct sk_msg, sk));
8369 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8370 offsetof(struct sock_common, skc_num));
8373 case offsetof(struct sk_msg_md, size):
8374 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
8375 si->dst_reg, si->src_reg,
8376 offsetof(struct sk_msg_sg, size));
8380 return insn - insn_buf;
8383 const struct bpf_verifier_ops sk_filter_verifier_ops = {
8384 .get_func_proto = sk_filter_func_proto,
8385 .is_valid_access = sk_filter_is_valid_access,
8386 .convert_ctx_access = bpf_convert_ctx_access,
8387 .gen_ld_abs = bpf_gen_ld_abs,
8390 const struct bpf_prog_ops sk_filter_prog_ops = {
8391 .test_run = bpf_prog_test_run_skb,
8394 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
8395 .get_func_proto = tc_cls_act_func_proto,
8396 .is_valid_access = tc_cls_act_is_valid_access,
8397 .convert_ctx_access = tc_cls_act_convert_ctx_access,
8398 .gen_prologue = tc_cls_act_prologue,
8399 .gen_ld_abs = bpf_gen_ld_abs,
8402 const struct bpf_prog_ops tc_cls_act_prog_ops = {
8403 .test_run = bpf_prog_test_run_skb,
8406 const struct bpf_verifier_ops xdp_verifier_ops = {
8407 .get_func_proto = xdp_func_proto,
8408 .is_valid_access = xdp_is_valid_access,
8409 .convert_ctx_access = xdp_convert_ctx_access,
8410 .gen_prologue = bpf_noop_prologue,
8413 const struct bpf_prog_ops xdp_prog_ops = {
8414 .test_run = bpf_prog_test_run_xdp,
8417 const struct bpf_verifier_ops cg_skb_verifier_ops = {
8418 .get_func_proto = cg_skb_func_proto,
8419 .is_valid_access = cg_skb_is_valid_access,
8420 .convert_ctx_access = bpf_convert_ctx_access,
8423 const struct bpf_prog_ops cg_skb_prog_ops = {
8424 .test_run = bpf_prog_test_run_skb,
8427 const struct bpf_verifier_ops lwt_in_verifier_ops = {
8428 .get_func_proto = lwt_in_func_proto,
8429 .is_valid_access = lwt_is_valid_access,
8430 .convert_ctx_access = bpf_convert_ctx_access,
8433 const struct bpf_prog_ops lwt_in_prog_ops = {
8434 .test_run = bpf_prog_test_run_skb,
8437 const struct bpf_verifier_ops lwt_out_verifier_ops = {
8438 .get_func_proto = lwt_out_func_proto,
8439 .is_valid_access = lwt_is_valid_access,
8440 .convert_ctx_access = bpf_convert_ctx_access,
8443 const struct bpf_prog_ops lwt_out_prog_ops = {
8444 .test_run = bpf_prog_test_run_skb,
8447 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
8448 .get_func_proto = lwt_xmit_func_proto,
8449 .is_valid_access = lwt_is_valid_access,
8450 .convert_ctx_access = bpf_convert_ctx_access,
8451 .gen_prologue = tc_cls_act_prologue,
8454 const struct bpf_prog_ops lwt_xmit_prog_ops = {
8455 .test_run = bpf_prog_test_run_skb,
8458 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
8459 .get_func_proto = lwt_seg6local_func_proto,
8460 .is_valid_access = lwt_is_valid_access,
8461 .convert_ctx_access = bpf_convert_ctx_access,
8464 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
8465 .test_run = bpf_prog_test_run_skb,
8468 const struct bpf_verifier_ops cg_sock_verifier_ops = {
8469 .get_func_proto = sock_filter_func_proto,
8470 .is_valid_access = sock_filter_is_valid_access,
8471 .convert_ctx_access = bpf_sock_convert_ctx_access,
8474 const struct bpf_prog_ops cg_sock_prog_ops = {
8477 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
8478 .get_func_proto = sock_addr_func_proto,
8479 .is_valid_access = sock_addr_is_valid_access,
8480 .convert_ctx_access = sock_addr_convert_ctx_access,
8483 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
8486 const struct bpf_verifier_ops sock_ops_verifier_ops = {
8487 .get_func_proto = sock_ops_func_proto,
8488 .is_valid_access = sock_ops_is_valid_access,
8489 .convert_ctx_access = sock_ops_convert_ctx_access,
8492 const struct bpf_prog_ops sock_ops_prog_ops = {
8495 const struct bpf_verifier_ops sk_skb_verifier_ops = {
8496 .get_func_proto = sk_skb_func_proto,
8497 .is_valid_access = sk_skb_is_valid_access,
8498 .convert_ctx_access = sk_skb_convert_ctx_access,
8499 .gen_prologue = sk_skb_prologue,
8502 const struct bpf_prog_ops sk_skb_prog_ops = {
8505 const struct bpf_verifier_ops sk_msg_verifier_ops = {
8506 .get_func_proto = sk_msg_func_proto,
8507 .is_valid_access = sk_msg_is_valid_access,
8508 .convert_ctx_access = sk_msg_convert_ctx_access,
8509 .gen_prologue = bpf_noop_prologue,
8512 const struct bpf_prog_ops sk_msg_prog_ops = {
8515 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
8516 .get_func_proto = flow_dissector_func_proto,
8517 .is_valid_access = flow_dissector_is_valid_access,
8518 .convert_ctx_access = flow_dissector_convert_ctx_access,
8521 const struct bpf_prog_ops flow_dissector_prog_ops = {
8522 .test_run = bpf_prog_test_run_flow_dissector,
8525 int sk_detach_filter(struct sock *sk)
8528 struct sk_filter *filter;
8530 if (sock_flag(sk, SOCK_FILTER_LOCKED))
8533 filter = rcu_dereference_protected(sk->sk_filter,
8534 lockdep_sock_is_held(sk));
8536 RCU_INIT_POINTER(sk->sk_filter, NULL);
8537 sk_filter_uncharge(sk, filter);
8543 EXPORT_SYMBOL_GPL(sk_detach_filter);
8545 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
8548 struct sock_fprog_kern *fprog;
8549 struct sk_filter *filter;
8553 filter = rcu_dereference_protected(sk->sk_filter,
8554 lockdep_sock_is_held(sk));
8558 /* We're copying the filter that has been originally attached,
8559 * so no conversion/decode needed anymore. eBPF programs that
8560 * have no original program cannot be dumped through this.
8563 fprog = filter->prog->orig_prog;
8569 /* User space only enquires number of filter blocks. */
8573 if (len < fprog->len)
8577 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
8580 /* Instead of bytes, the API requests to return the number
8590 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
8591 struct sock_reuseport *reuse,
8592 struct sock *sk, struct sk_buff *skb,
8595 reuse_kern->skb = skb;
8596 reuse_kern->sk = sk;
8597 reuse_kern->selected_sk = NULL;
8598 reuse_kern->data_end = skb->data + skb_headlen(skb);
8599 reuse_kern->hash = hash;
8600 reuse_kern->reuseport_id = reuse->reuseport_id;
8601 reuse_kern->bind_inany = reuse->bind_inany;
8604 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
8605 struct bpf_prog *prog, struct sk_buff *skb,
8608 struct sk_reuseport_kern reuse_kern;
8609 enum sk_action action;
8611 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, hash);
8612 action = BPF_PROG_RUN(prog, &reuse_kern);
8614 if (action == SK_PASS)
8615 return reuse_kern.selected_sk;
8617 return ERR_PTR(-ECONNREFUSED);
8620 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
8621 struct bpf_map *, map, void *, key, u32, flags)
8623 struct sock_reuseport *reuse;
8624 struct sock *selected_sk;
8626 selected_sk = map->ops->map_lookup_elem(map, key);
8630 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
8632 /* selected_sk is unhashed (e.g. by close()) after the
8633 * above map_lookup_elem(). Treat selected_sk has already
8634 * been removed from the map.
8638 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
8641 if (unlikely(!reuse_kern->reuseport_id))
8642 /* There is a small race between adding the
8643 * sk to the map and setting the
8644 * reuse_kern->reuseport_id.
8645 * Treat it as the sk has not been added to
8650 sk = reuse_kern->sk;
8651 if (sk->sk_protocol != selected_sk->sk_protocol)
8653 else if (sk->sk_family != selected_sk->sk_family)
8654 return -EAFNOSUPPORT;
8656 /* Catch all. Likely bound to a different sockaddr. */
8660 reuse_kern->selected_sk = selected_sk;
8665 static const struct bpf_func_proto sk_select_reuseport_proto = {
8666 .func = sk_select_reuseport,
8668 .ret_type = RET_INTEGER,
8669 .arg1_type = ARG_PTR_TO_CTX,
8670 .arg2_type = ARG_CONST_MAP_PTR,
8671 .arg3_type = ARG_PTR_TO_MAP_KEY,
8672 .arg4_type = ARG_ANYTHING,
8675 BPF_CALL_4(sk_reuseport_load_bytes,
8676 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
8677 void *, to, u32, len)
8679 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
8682 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
8683 .func = sk_reuseport_load_bytes,
8685 .ret_type = RET_INTEGER,
8686 .arg1_type = ARG_PTR_TO_CTX,
8687 .arg2_type = ARG_ANYTHING,
8688 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
8689 .arg4_type = ARG_CONST_SIZE,
8692 BPF_CALL_5(sk_reuseport_load_bytes_relative,
8693 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
8694 void *, to, u32, len, u32, start_header)
8696 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
8700 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
8701 .func = sk_reuseport_load_bytes_relative,
8703 .ret_type = RET_INTEGER,
8704 .arg1_type = ARG_PTR_TO_CTX,
8705 .arg2_type = ARG_ANYTHING,
8706 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
8707 .arg4_type = ARG_CONST_SIZE,
8708 .arg5_type = ARG_ANYTHING,
8711 static const struct bpf_func_proto *
8712 sk_reuseport_func_proto(enum bpf_func_id func_id,
8713 const struct bpf_prog *prog)
8716 case BPF_FUNC_sk_select_reuseport:
8717 return &sk_select_reuseport_proto;
8718 case BPF_FUNC_skb_load_bytes:
8719 return &sk_reuseport_load_bytes_proto;
8720 case BPF_FUNC_skb_load_bytes_relative:
8721 return &sk_reuseport_load_bytes_relative_proto;
8723 return bpf_base_func_proto(func_id);
8728 sk_reuseport_is_valid_access(int off, int size,
8729 enum bpf_access_type type,
8730 const struct bpf_prog *prog,
8731 struct bpf_insn_access_aux *info)
8733 const u32 size_default = sizeof(__u32);
8735 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
8736 off % size || type != BPF_READ)
8740 case offsetof(struct sk_reuseport_md, data):
8741 info->reg_type = PTR_TO_PACKET;
8742 return size == sizeof(__u64);
8744 case offsetof(struct sk_reuseport_md, data_end):
8745 info->reg_type = PTR_TO_PACKET_END;
8746 return size == sizeof(__u64);
8748 case offsetof(struct sk_reuseport_md, hash):
8749 return size == size_default;
8751 /* Fields that allow narrowing */
8752 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
8753 if (size < sizeof_field(struct sk_buff, protocol))
8756 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
8757 case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
8758 case bpf_ctx_range(struct sk_reuseport_md, len):
8759 bpf_ctx_record_field_size(info, size_default);
8760 return bpf_ctx_narrow_access_ok(off, size, size_default);
8767 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
8768 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
8769 si->dst_reg, si->src_reg, \
8770 bpf_target_off(struct sk_reuseport_kern, F, \
8771 sizeof_field(struct sk_reuseport_kern, F), \
8775 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
8776 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
8781 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \
8782 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
8787 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
8788 const struct bpf_insn *si,
8789 struct bpf_insn *insn_buf,
8790 struct bpf_prog *prog,
8793 struct bpf_insn *insn = insn_buf;
8796 case offsetof(struct sk_reuseport_md, data):
8797 SK_REUSEPORT_LOAD_SKB_FIELD(data);
8800 case offsetof(struct sk_reuseport_md, len):
8801 SK_REUSEPORT_LOAD_SKB_FIELD(len);
8804 case offsetof(struct sk_reuseport_md, eth_protocol):
8805 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
8808 case offsetof(struct sk_reuseport_md, ip_protocol):
8809 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
8812 case offsetof(struct sk_reuseport_md, data_end):
8813 SK_REUSEPORT_LOAD_FIELD(data_end);
8816 case offsetof(struct sk_reuseport_md, hash):
8817 SK_REUSEPORT_LOAD_FIELD(hash);
8820 case offsetof(struct sk_reuseport_md, bind_inany):
8821 SK_REUSEPORT_LOAD_FIELD(bind_inany);
8825 return insn - insn_buf;
8828 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
8829 .get_func_proto = sk_reuseport_func_proto,
8830 .is_valid_access = sk_reuseport_is_valid_access,
8831 .convert_ctx_access = sk_reuseport_convert_ctx_access,
8834 const struct bpf_prog_ops sk_reuseport_prog_ops = {
8836 #endif /* CONFIG_INET */
8838 DEFINE_BPF_DISPATCHER(bpf_dispatcher_xdp)
8840 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
8842 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(bpf_dispatcher_xdp),