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(FIELD_SIZEOF(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(FIELD_SIZEOF(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(FIELD_SIZEOF(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(FIELD_SIZEOF(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(FIELD_SIZEOF(struct net_device, ifindex) != 4);
342 BUILD_BUG_ON(FIELD_SIZEOF(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(FIELD_SIZEOF(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(FIELD_SIZEOF(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 (IS_ERR(prog) && 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");
2059 dev_xmit_recursion_inc();
2060 ret = dev_queue_xmit(skb);
2061 dev_xmit_recursion_dec();
2066 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2069 unsigned int mlen = skb_network_offset(skb);
2072 __skb_pull(skb, mlen);
2074 /* At ingress, the mac header has already been pulled once.
2075 * At egress, skb_pospull_rcsum has to be done in case that
2076 * the skb is originated from ingress (i.e. a forwarded skb)
2077 * to ensure that rcsum starts at net header.
2079 if (!skb_at_tc_ingress(skb))
2080 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2082 skb_pop_mac_header(skb);
2083 skb_reset_mac_len(skb);
2084 return flags & BPF_F_INGRESS ?
2085 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2088 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2091 /* Verify that a link layer header is carried */
2092 if (unlikely(skb->mac_header >= skb->network_header)) {
2097 bpf_push_mac_rcsum(skb);
2098 return flags & BPF_F_INGRESS ?
2099 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2102 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2105 if (dev_is_mac_header_xmit(dev))
2106 return __bpf_redirect_common(skb, dev, flags);
2108 return __bpf_redirect_no_mac(skb, dev, flags);
2111 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2113 struct net_device *dev;
2114 struct sk_buff *clone;
2117 if (unlikely(flags & ~(BPF_F_INGRESS)))
2120 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2124 clone = skb_clone(skb, GFP_ATOMIC);
2125 if (unlikely(!clone))
2128 /* For direct write, we need to keep the invariant that the skbs
2129 * we're dealing with need to be uncloned. Should uncloning fail
2130 * here, we need to free the just generated clone to unclone once
2133 ret = bpf_try_make_head_writable(skb);
2134 if (unlikely(ret)) {
2139 return __bpf_redirect(clone, dev, flags);
2142 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2143 .func = bpf_clone_redirect,
2145 .ret_type = RET_INTEGER,
2146 .arg1_type = ARG_PTR_TO_CTX,
2147 .arg2_type = ARG_ANYTHING,
2148 .arg3_type = ARG_ANYTHING,
2151 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2152 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2154 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2156 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2158 if (unlikely(flags & ~(BPF_F_INGRESS)))
2162 ri->tgt_index = ifindex;
2164 return TC_ACT_REDIRECT;
2167 int skb_do_redirect(struct sk_buff *skb)
2169 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2170 struct net_device *dev;
2172 dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->tgt_index);
2174 if (unlikely(!dev)) {
2179 return __bpf_redirect(skb, dev, ri->flags);
2182 static const struct bpf_func_proto bpf_redirect_proto = {
2183 .func = bpf_redirect,
2185 .ret_type = RET_INTEGER,
2186 .arg1_type = ARG_ANYTHING,
2187 .arg2_type = ARG_ANYTHING,
2190 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2192 msg->apply_bytes = bytes;
2196 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2197 .func = bpf_msg_apply_bytes,
2199 .ret_type = RET_INTEGER,
2200 .arg1_type = ARG_PTR_TO_CTX,
2201 .arg2_type = ARG_ANYTHING,
2204 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2206 msg->cork_bytes = bytes;
2210 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2211 .func = bpf_msg_cork_bytes,
2213 .ret_type = RET_INTEGER,
2214 .arg1_type = ARG_PTR_TO_CTX,
2215 .arg2_type = ARG_ANYTHING,
2218 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2219 u32, end, u64, flags)
2221 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2222 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2223 struct scatterlist *sge;
2224 u8 *raw, *to, *from;
2227 if (unlikely(flags || end <= start))
2230 /* First find the starting scatterlist element */
2233 len = sk_msg_elem(msg, i)->length;
2234 if (start < offset + len)
2237 sk_msg_iter_var_next(i);
2238 } while (i != msg->sg.end);
2240 if (unlikely(start >= offset + len))
2244 /* The start may point into the sg element so we need to also
2245 * account for the headroom.
2247 bytes_sg_total = start - offset + bytes;
2248 if (!msg->sg.copy[i] && bytes_sg_total <= len)
2251 /* At this point we need to linearize multiple scatterlist
2252 * elements or a single shared page. Either way we need to
2253 * copy into a linear buffer exclusively owned by BPF. Then
2254 * place the buffer in the scatterlist and fixup the original
2255 * entries by removing the entries now in the linear buffer
2256 * and shifting the remaining entries. For now we do not try
2257 * to copy partial entries to avoid complexity of running out
2258 * of sg_entry slots. The downside is reading a single byte
2259 * will copy the entire sg entry.
2262 copy += sk_msg_elem(msg, i)->length;
2263 sk_msg_iter_var_next(i);
2264 if (bytes_sg_total <= copy)
2266 } while (i != msg->sg.end);
2269 if (unlikely(bytes_sg_total > copy))
2272 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2274 if (unlikely(!page))
2277 raw = page_address(page);
2280 sge = sk_msg_elem(msg, i);
2281 from = sg_virt(sge);
2285 memcpy(to, from, len);
2288 put_page(sg_page(sge));
2290 sk_msg_iter_var_next(i);
2291 } while (i != last_sge);
2293 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2295 /* To repair sg ring we need to shift entries. If we only
2296 * had a single entry though we can just replace it and
2297 * be done. Otherwise walk the ring and shift the entries.
2299 WARN_ON_ONCE(last_sge == first_sge);
2300 shift = last_sge > first_sge ?
2301 last_sge - first_sge - 1 :
2302 MAX_SKB_FRAGS - first_sge + last_sge - 1;
2307 sk_msg_iter_var_next(i);
2311 if (i + shift >= MAX_MSG_FRAGS)
2312 move_from = i + shift - MAX_MSG_FRAGS;
2314 move_from = i + shift;
2315 if (move_from == msg->sg.end)
2318 msg->sg.data[i] = msg->sg.data[move_from];
2319 msg->sg.data[move_from].length = 0;
2320 msg->sg.data[move_from].page_link = 0;
2321 msg->sg.data[move_from].offset = 0;
2322 sk_msg_iter_var_next(i);
2325 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2326 msg->sg.end - shift + MAX_MSG_FRAGS :
2327 msg->sg.end - shift;
2329 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2330 msg->data_end = msg->data + bytes;
2334 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2335 .func = bpf_msg_pull_data,
2337 .ret_type = RET_INTEGER,
2338 .arg1_type = ARG_PTR_TO_CTX,
2339 .arg2_type = ARG_ANYTHING,
2340 .arg3_type = ARG_ANYTHING,
2341 .arg4_type = ARG_ANYTHING,
2344 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2345 u32, len, u64, flags)
2347 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2348 u32 new, i = 0, l, space, copy = 0, offset = 0;
2349 u8 *raw, *to, *from;
2352 if (unlikely(flags))
2355 /* First find the starting scatterlist element */
2358 l = sk_msg_elem(msg, i)->length;
2360 if (start < offset + l)
2363 sk_msg_iter_var_next(i);
2364 } while (i != msg->sg.end);
2366 if (start >= offset + l)
2369 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2371 /* If no space available will fallback to copy, we need at
2372 * least one scatterlist elem available to push data into
2373 * when start aligns to the beginning of an element or two
2374 * when it falls inside an element. We handle the start equals
2375 * offset case because its the common case for inserting a
2378 if (!space || (space == 1 && start != offset))
2379 copy = msg->sg.data[i].length;
2381 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2382 get_order(copy + len));
2383 if (unlikely(!page))
2389 raw = page_address(page);
2391 psge = sk_msg_elem(msg, i);
2392 front = start - offset;
2393 back = psge->length - front;
2394 from = sg_virt(psge);
2397 memcpy(raw, from, front);
2401 to = raw + front + len;
2403 memcpy(to, from, back);
2406 put_page(sg_page(psge));
2407 } else if (start - offset) {
2408 psge = sk_msg_elem(msg, i);
2409 rsge = sk_msg_elem_cpy(msg, i);
2411 psge->length = start - offset;
2412 rsge.length -= psge->length;
2413 rsge.offset += start;
2415 sk_msg_iter_var_next(i);
2416 sg_unmark_end(psge);
2417 sk_msg_iter_next(msg, end);
2420 /* Slot(s) to place newly allocated data */
2423 /* Shift one or two slots as needed */
2425 sge = sk_msg_elem_cpy(msg, i);
2427 sk_msg_iter_var_next(i);
2428 sg_unmark_end(&sge);
2429 sk_msg_iter_next(msg, end);
2431 nsge = sk_msg_elem_cpy(msg, i);
2433 sk_msg_iter_var_next(i);
2434 nnsge = sk_msg_elem_cpy(msg, i);
2437 while (i != msg->sg.end) {
2438 msg->sg.data[i] = sge;
2440 sk_msg_iter_var_next(i);
2443 nnsge = sk_msg_elem_cpy(msg, i);
2445 nsge = sk_msg_elem_cpy(msg, i);
2450 /* Place newly allocated data buffer */
2451 sk_mem_charge(msg->sk, len);
2452 msg->sg.size += len;
2453 msg->sg.copy[new] = false;
2454 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2456 get_page(sg_page(&rsge));
2457 sk_msg_iter_var_next(new);
2458 msg->sg.data[new] = rsge;
2461 sk_msg_compute_data_pointers(msg);
2465 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2466 .func = bpf_msg_push_data,
2468 .ret_type = RET_INTEGER,
2469 .arg1_type = ARG_PTR_TO_CTX,
2470 .arg2_type = ARG_ANYTHING,
2471 .arg3_type = ARG_ANYTHING,
2472 .arg4_type = ARG_ANYTHING,
2475 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2481 sk_msg_iter_var_next(i);
2482 msg->sg.data[prev] = msg->sg.data[i];
2483 } while (i != msg->sg.end);
2485 sk_msg_iter_prev(msg, end);
2488 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2490 struct scatterlist tmp, sge;
2492 sk_msg_iter_next(msg, end);
2493 sge = sk_msg_elem_cpy(msg, i);
2494 sk_msg_iter_var_next(i);
2495 tmp = sk_msg_elem_cpy(msg, i);
2497 while (i != msg->sg.end) {
2498 msg->sg.data[i] = sge;
2499 sk_msg_iter_var_next(i);
2501 tmp = sk_msg_elem_cpy(msg, i);
2505 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2506 u32, len, u64, flags)
2508 u32 i = 0, l, space, offset = 0;
2509 u64 last = start + len;
2512 if (unlikely(flags))
2515 /* First find the starting scatterlist element */
2518 l = sk_msg_elem(msg, i)->length;
2520 if (start < offset + l)
2523 sk_msg_iter_var_next(i);
2524 } while (i != msg->sg.end);
2526 /* Bounds checks: start and pop must be inside message */
2527 if (start >= offset + l || last >= msg->sg.size)
2530 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2533 /* --------------| offset
2534 * -| start |-------- len -------|
2536 * |----- a ----|-------- pop -------|----- b ----|
2537 * |______________________________________________| length
2540 * a: region at front of scatter element to save
2541 * b: region at back of scatter element to save when length > A + pop
2542 * pop: region to pop from element, same as input 'pop' here will be
2543 * decremented below per iteration.
2545 * Two top-level cases to handle when start != offset, first B is non
2546 * zero and second B is zero corresponding to when a pop includes more
2549 * Then if B is non-zero AND there is no space allocate space and
2550 * compact A, B regions into page. If there is space shift ring to
2551 * the rigth free'ing the next element in ring to place B, leaving
2552 * A untouched except to reduce length.
2554 if (start != offset) {
2555 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2557 int b = sge->length - pop - a;
2559 sk_msg_iter_var_next(i);
2561 if (pop < sge->length - a) {
2564 sk_msg_shift_right(msg, i);
2565 nsge = sk_msg_elem(msg, i);
2566 get_page(sg_page(sge));
2569 b, sge->offset + pop + a);
2571 struct page *page, *orig;
2574 page = alloc_pages(__GFP_NOWARN |
2575 __GFP_COMP | GFP_ATOMIC,
2577 if (unlikely(!page))
2581 orig = sg_page(sge);
2582 from = sg_virt(sge);
2583 to = page_address(page);
2584 memcpy(to, from, a);
2585 memcpy(to + a, from + a + pop, b);
2586 sg_set_page(sge, page, a + b, 0);
2590 } else if (pop >= sge->length - a) {
2592 pop -= (sge->length - a);
2596 /* From above the current layout _must_ be as follows,
2601 * |---- pop ---|---------------- b ------------|
2602 * |____________________________________________| length
2604 * Offset and start of the current msg elem are equal because in the
2605 * previous case we handled offset != start and either consumed the
2606 * entire element and advanced to the next element OR pop == 0.
2608 * Two cases to handle here are first pop is less than the length
2609 * leaving some remainder b above. Simply adjust the element's layout
2610 * in this case. Or pop >= length of the element so that b = 0. In this
2611 * case advance to next element decrementing pop.
2614 struct scatterlist *sge = sk_msg_elem(msg, i);
2616 if (pop < sge->length) {
2622 sk_msg_shift_left(msg, i);
2624 sk_msg_iter_var_next(i);
2627 sk_mem_uncharge(msg->sk, len - pop);
2628 msg->sg.size -= (len - pop);
2629 sk_msg_compute_data_pointers(msg);
2633 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
2634 .func = bpf_msg_pop_data,
2636 .ret_type = RET_INTEGER,
2637 .arg1_type = ARG_PTR_TO_CTX,
2638 .arg2_type = ARG_ANYTHING,
2639 .arg3_type = ARG_ANYTHING,
2640 .arg4_type = ARG_ANYTHING,
2643 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
2645 return task_get_classid(skb);
2648 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
2649 .func = bpf_get_cgroup_classid,
2651 .ret_type = RET_INTEGER,
2652 .arg1_type = ARG_PTR_TO_CTX,
2655 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
2657 return dst_tclassid(skb);
2660 static const struct bpf_func_proto bpf_get_route_realm_proto = {
2661 .func = bpf_get_route_realm,
2663 .ret_type = RET_INTEGER,
2664 .arg1_type = ARG_PTR_TO_CTX,
2667 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
2669 /* If skb_clear_hash() was called due to mangling, we can
2670 * trigger SW recalculation here. Later access to hash
2671 * can then use the inline skb->hash via context directly
2672 * instead of calling this helper again.
2674 return skb_get_hash(skb);
2677 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
2678 .func = bpf_get_hash_recalc,
2680 .ret_type = RET_INTEGER,
2681 .arg1_type = ARG_PTR_TO_CTX,
2684 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
2686 /* After all direct packet write, this can be used once for
2687 * triggering a lazy recalc on next skb_get_hash() invocation.
2689 skb_clear_hash(skb);
2693 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
2694 .func = bpf_set_hash_invalid,
2696 .ret_type = RET_INTEGER,
2697 .arg1_type = ARG_PTR_TO_CTX,
2700 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
2702 /* Set user specified hash as L4(+), so that it gets returned
2703 * on skb_get_hash() call unless BPF prog later on triggers a
2706 __skb_set_sw_hash(skb, hash, true);
2710 static const struct bpf_func_proto bpf_set_hash_proto = {
2711 .func = bpf_set_hash,
2713 .ret_type = RET_INTEGER,
2714 .arg1_type = ARG_PTR_TO_CTX,
2715 .arg2_type = ARG_ANYTHING,
2718 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
2723 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
2724 vlan_proto != htons(ETH_P_8021AD)))
2725 vlan_proto = htons(ETH_P_8021Q);
2727 bpf_push_mac_rcsum(skb);
2728 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
2729 bpf_pull_mac_rcsum(skb);
2731 bpf_compute_data_pointers(skb);
2735 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
2736 .func = bpf_skb_vlan_push,
2738 .ret_type = RET_INTEGER,
2739 .arg1_type = ARG_PTR_TO_CTX,
2740 .arg2_type = ARG_ANYTHING,
2741 .arg3_type = ARG_ANYTHING,
2744 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
2748 bpf_push_mac_rcsum(skb);
2749 ret = skb_vlan_pop(skb);
2750 bpf_pull_mac_rcsum(skb);
2752 bpf_compute_data_pointers(skb);
2756 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
2757 .func = bpf_skb_vlan_pop,
2759 .ret_type = RET_INTEGER,
2760 .arg1_type = ARG_PTR_TO_CTX,
2763 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
2765 /* Caller already did skb_cow() with len as headroom,
2766 * so no need to do it here.
2769 memmove(skb->data, skb->data + len, off);
2770 memset(skb->data + off, 0, len);
2772 /* No skb_postpush_rcsum(skb, skb->data + off, len)
2773 * needed here as it does not change the skb->csum
2774 * result for checksum complete when summing over
2780 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
2782 /* skb_ensure_writable() is not needed here, as we're
2783 * already working on an uncloned skb.
2785 if (unlikely(!pskb_may_pull(skb, off + len)))
2788 skb_postpull_rcsum(skb, skb->data + off, len);
2789 memmove(skb->data + len, skb->data, off);
2790 __skb_pull(skb, len);
2795 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
2797 bool trans_same = skb->transport_header == skb->network_header;
2800 /* There's no need for __skb_push()/__skb_pull() pair to
2801 * get to the start of the mac header as we're guaranteed
2802 * to always start from here under eBPF.
2804 ret = bpf_skb_generic_push(skb, off, len);
2806 skb->mac_header -= len;
2807 skb->network_header -= len;
2809 skb->transport_header = skb->network_header;
2815 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
2817 bool trans_same = skb->transport_header == skb->network_header;
2820 /* Same here, __skb_push()/__skb_pull() pair not needed. */
2821 ret = bpf_skb_generic_pop(skb, off, len);
2823 skb->mac_header += len;
2824 skb->network_header += len;
2826 skb->transport_header = skb->network_header;
2832 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
2834 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2835 u32 off = skb_mac_header_len(skb);
2838 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
2841 ret = skb_cow(skb, len_diff);
2842 if (unlikely(ret < 0))
2845 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2846 if (unlikely(ret < 0))
2849 if (skb_is_gso(skb)) {
2850 struct skb_shared_info *shinfo = skb_shinfo(skb);
2852 /* SKB_GSO_TCPV4 needs to be changed into
2855 if (shinfo->gso_type & SKB_GSO_TCPV4) {
2856 shinfo->gso_type &= ~SKB_GSO_TCPV4;
2857 shinfo->gso_type |= SKB_GSO_TCPV6;
2860 /* Due to IPv6 header, MSS needs to be downgraded. */
2861 skb_decrease_gso_size(shinfo, len_diff);
2862 /* Header must be checked, and gso_segs recomputed. */
2863 shinfo->gso_type |= SKB_GSO_DODGY;
2864 shinfo->gso_segs = 0;
2867 skb->protocol = htons(ETH_P_IPV6);
2868 skb_clear_hash(skb);
2873 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
2875 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2876 u32 off = skb_mac_header_len(skb);
2879 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
2882 ret = skb_unclone(skb, GFP_ATOMIC);
2883 if (unlikely(ret < 0))
2886 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2887 if (unlikely(ret < 0))
2890 if (skb_is_gso(skb)) {
2891 struct skb_shared_info *shinfo = skb_shinfo(skb);
2893 /* SKB_GSO_TCPV6 needs to be changed into
2896 if (shinfo->gso_type & SKB_GSO_TCPV6) {
2897 shinfo->gso_type &= ~SKB_GSO_TCPV6;
2898 shinfo->gso_type |= SKB_GSO_TCPV4;
2901 /* Due to IPv4 header, MSS can be upgraded. */
2902 skb_increase_gso_size(shinfo, len_diff);
2903 /* Header must be checked, and gso_segs recomputed. */
2904 shinfo->gso_type |= SKB_GSO_DODGY;
2905 shinfo->gso_segs = 0;
2908 skb->protocol = htons(ETH_P_IP);
2909 skb_clear_hash(skb);
2914 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
2916 __be16 from_proto = skb->protocol;
2918 if (from_proto == htons(ETH_P_IP) &&
2919 to_proto == htons(ETH_P_IPV6))
2920 return bpf_skb_proto_4_to_6(skb);
2922 if (from_proto == htons(ETH_P_IPV6) &&
2923 to_proto == htons(ETH_P_IP))
2924 return bpf_skb_proto_6_to_4(skb);
2929 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
2934 if (unlikely(flags))
2937 /* General idea is that this helper does the basic groundwork
2938 * needed for changing the protocol, and eBPF program fills the
2939 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
2940 * and other helpers, rather than passing a raw buffer here.
2942 * The rationale is to keep this minimal and without a need to
2943 * deal with raw packet data. F.e. even if we would pass buffers
2944 * here, the program still needs to call the bpf_lX_csum_replace()
2945 * helpers anyway. Plus, this way we keep also separation of
2946 * concerns, since f.e. bpf_skb_store_bytes() should only take
2949 * Currently, additional options and extension header space are
2950 * not supported, but flags register is reserved so we can adapt
2951 * that. For offloads, we mark packet as dodgy, so that headers
2952 * need to be verified first.
2954 ret = bpf_skb_proto_xlat(skb, proto);
2955 bpf_compute_data_pointers(skb);
2959 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
2960 .func = bpf_skb_change_proto,
2962 .ret_type = RET_INTEGER,
2963 .arg1_type = ARG_PTR_TO_CTX,
2964 .arg2_type = ARG_ANYTHING,
2965 .arg3_type = ARG_ANYTHING,
2968 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
2970 /* We only allow a restricted subset to be changed for now. */
2971 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
2972 !skb_pkt_type_ok(pkt_type)))
2975 skb->pkt_type = pkt_type;
2979 static const struct bpf_func_proto bpf_skb_change_type_proto = {
2980 .func = bpf_skb_change_type,
2982 .ret_type = RET_INTEGER,
2983 .arg1_type = ARG_PTR_TO_CTX,
2984 .arg2_type = ARG_ANYTHING,
2987 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
2989 switch (skb->protocol) {
2990 case htons(ETH_P_IP):
2991 return sizeof(struct iphdr);
2992 case htons(ETH_P_IPV6):
2993 return sizeof(struct ipv6hdr);
2999 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3000 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3002 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \
3003 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3004 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3005 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3006 BPF_F_ADJ_ROOM_ENCAP_L2( \
3007 BPF_ADJ_ROOM_ENCAP_L2_MASK))
3009 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3012 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3013 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3014 u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3015 unsigned int gso_type = SKB_GSO_DODGY;
3018 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3019 /* udp gso_size delineates datagrams, only allow if fixed */
3020 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3021 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3025 ret = skb_cow_head(skb, len_diff);
3026 if (unlikely(ret < 0))
3030 if (skb->protocol != htons(ETH_P_IP) &&
3031 skb->protocol != htons(ETH_P_IPV6))
3034 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3035 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3038 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3039 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3042 if (skb->encapsulation)
3045 mac_len = skb->network_header - skb->mac_header;
3046 inner_net = skb->network_header;
3047 if (inner_mac_len > len_diff)
3049 inner_trans = skb->transport_header;
3052 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3053 if (unlikely(ret < 0))
3057 skb->inner_mac_header = inner_net - inner_mac_len;
3058 skb->inner_network_header = inner_net;
3059 skb->inner_transport_header = inner_trans;
3060 skb_set_inner_protocol(skb, skb->protocol);
3062 skb->encapsulation = 1;
3063 skb_set_network_header(skb, mac_len);
3065 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3066 gso_type |= SKB_GSO_UDP_TUNNEL;
3067 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3068 gso_type |= SKB_GSO_GRE;
3069 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3070 gso_type |= SKB_GSO_IPXIP6;
3071 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3072 gso_type |= SKB_GSO_IPXIP4;
3074 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3075 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3076 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3077 sizeof(struct ipv6hdr) :
3078 sizeof(struct iphdr);
3080 skb_set_transport_header(skb, mac_len + nh_len);
3083 /* Match skb->protocol to new outer l3 protocol */
3084 if (skb->protocol == htons(ETH_P_IP) &&
3085 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3086 skb->protocol = htons(ETH_P_IPV6);
3087 else if (skb->protocol == htons(ETH_P_IPV6) &&
3088 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3089 skb->protocol = htons(ETH_P_IP);
3092 if (skb_is_gso(skb)) {
3093 struct skb_shared_info *shinfo = skb_shinfo(skb);
3095 /* Due to header grow, MSS needs to be downgraded. */
3096 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3097 skb_decrease_gso_size(shinfo, len_diff);
3099 /* Header must be checked, and gso_segs recomputed. */
3100 shinfo->gso_type |= gso_type;
3101 shinfo->gso_segs = 0;
3107 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3112 if (flags & ~BPF_F_ADJ_ROOM_FIXED_GSO)
3115 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3116 /* udp gso_size delineates datagrams, only allow if fixed */
3117 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3118 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3122 ret = skb_unclone(skb, GFP_ATOMIC);
3123 if (unlikely(ret < 0))
3126 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3127 if (unlikely(ret < 0))
3130 if (skb_is_gso(skb)) {
3131 struct skb_shared_info *shinfo = skb_shinfo(skb);
3133 /* Due to header shrink, MSS can be upgraded. */
3134 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3135 skb_increase_gso_size(shinfo, len_diff);
3137 /* Header must be checked, and gso_segs recomputed. */
3138 shinfo->gso_type |= SKB_GSO_DODGY;
3139 shinfo->gso_segs = 0;
3145 static u32 __bpf_skb_max_len(const struct sk_buff *skb)
3147 return skb->dev ? skb->dev->mtu + skb->dev->hard_header_len :
3151 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3152 u32, mode, u64, flags)
3154 u32 len_cur, len_diff_abs = abs(len_diff);
3155 u32 len_min = bpf_skb_net_base_len(skb);
3156 u32 len_max = __bpf_skb_max_len(skb);
3157 __be16 proto = skb->protocol;
3158 bool shrink = len_diff < 0;
3162 if (unlikely(flags & ~BPF_F_ADJ_ROOM_MASK))
3164 if (unlikely(len_diff_abs > 0xfffU))
3166 if (unlikely(proto != htons(ETH_P_IP) &&
3167 proto != htons(ETH_P_IPV6)))
3170 off = skb_mac_header_len(skb);
3172 case BPF_ADJ_ROOM_NET:
3173 off += bpf_skb_net_base_len(skb);
3175 case BPF_ADJ_ROOM_MAC:
3181 len_cur = skb->len - skb_network_offset(skb);
3182 if ((shrink && (len_diff_abs >= len_cur ||
3183 len_cur - len_diff_abs < len_min)) ||
3184 (!shrink && (skb->len + len_diff_abs > len_max &&
3188 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3189 bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3191 bpf_compute_data_pointers(skb);
3195 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3196 .func = bpf_skb_adjust_room,
3198 .ret_type = RET_INTEGER,
3199 .arg1_type = ARG_PTR_TO_CTX,
3200 .arg2_type = ARG_ANYTHING,
3201 .arg3_type = ARG_ANYTHING,
3202 .arg4_type = ARG_ANYTHING,
3205 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3207 u32 min_len = skb_network_offset(skb);
3209 if (skb_transport_header_was_set(skb))
3210 min_len = skb_transport_offset(skb);
3211 if (skb->ip_summed == CHECKSUM_PARTIAL)
3212 min_len = skb_checksum_start_offset(skb) +
3213 skb->csum_offset + sizeof(__sum16);
3217 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3219 unsigned int old_len = skb->len;
3222 ret = __skb_grow_rcsum(skb, new_len);
3224 memset(skb->data + old_len, 0, new_len - old_len);
3228 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3230 return __skb_trim_rcsum(skb, new_len);
3233 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3236 u32 max_len = __bpf_skb_max_len(skb);
3237 u32 min_len = __bpf_skb_min_len(skb);
3240 if (unlikely(flags || new_len > max_len || new_len < min_len))
3242 if (skb->encapsulation)
3245 /* The basic idea of this helper is that it's performing the
3246 * needed work to either grow or trim an skb, and eBPF program
3247 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3248 * bpf_lX_csum_replace() and others rather than passing a raw
3249 * buffer here. This one is a slow path helper and intended
3250 * for replies with control messages.
3252 * Like in bpf_skb_change_proto(), we want to keep this rather
3253 * minimal and without protocol specifics so that we are able
3254 * to separate concerns as in bpf_skb_store_bytes() should only
3255 * be the one responsible for writing buffers.
3257 * It's really expected to be a slow path operation here for
3258 * control message replies, so we're implicitly linearizing,
3259 * uncloning and drop offloads from the skb by this.
3261 ret = __bpf_try_make_writable(skb, skb->len);
3263 if (new_len > skb->len)
3264 ret = bpf_skb_grow_rcsum(skb, new_len);
3265 else if (new_len < skb->len)
3266 ret = bpf_skb_trim_rcsum(skb, new_len);
3267 if (!ret && skb_is_gso(skb))
3273 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3276 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3278 bpf_compute_data_pointers(skb);
3282 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3283 .func = bpf_skb_change_tail,
3285 .ret_type = RET_INTEGER,
3286 .arg1_type = ARG_PTR_TO_CTX,
3287 .arg2_type = ARG_ANYTHING,
3288 .arg3_type = ARG_ANYTHING,
3291 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3294 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3296 bpf_compute_data_end_sk_skb(skb);
3300 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3301 .func = sk_skb_change_tail,
3303 .ret_type = RET_INTEGER,
3304 .arg1_type = ARG_PTR_TO_CTX,
3305 .arg2_type = ARG_ANYTHING,
3306 .arg3_type = ARG_ANYTHING,
3309 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3312 u32 max_len = __bpf_skb_max_len(skb);
3313 u32 new_len = skb->len + head_room;
3316 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3317 new_len < skb->len))
3320 ret = skb_cow(skb, head_room);
3322 /* Idea for this helper is that we currently only
3323 * allow to expand on mac header. This means that
3324 * skb->protocol network header, etc, stay as is.
3325 * Compared to bpf_skb_change_tail(), we're more
3326 * flexible due to not needing to linearize or
3327 * reset GSO. Intention for this helper is to be
3328 * used by an L3 skb that needs to push mac header
3329 * for redirection into L2 device.
3331 __skb_push(skb, head_room);
3332 memset(skb->data, 0, head_room);
3333 skb_reset_mac_header(skb);
3339 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3342 int ret = __bpf_skb_change_head(skb, head_room, flags);
3344 bpf_compute_data_pointers(skb);
3348 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3349 .func = bpf_skb_change_head,
3351 .ret_type = RET_INTEGER,
3352 .arg1_type = ARG_PTR_TO_CTX,
3353 .arg2_type = ARG_ANYTHING,
3354 .arg3_type = ARG_ANYTHING,
3357 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3360 int ret = __bpf_skb_change_head(skb, head_room, flags);
3362 bpf_compute_data_end_sk_skb(skb);
3366 static const struct bpf_func_proto sk_skb_change_head_proto = {
3367 .func = sk_skb_change_head,
3369 .ret_type = RET_INTEGER,
3370 .arg1_type = ARG_PTR_TO_CTX,
3371 .arg2_type = ARG_ANYTHING,
3372 .arg3_type = ARG_ANYTHING,
3374 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3376 return xdp_data_meta_unsupported(xdp) ? 0 :
3377 xdp->data - xdp->data_meta;
3380 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3382 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3383 unsigned long metalen = xdp_get_metalen(xdp);
3384 void *data_start = xdp_frame_end + metalen;
3385 void *data = xdp->data + offset;
3387 if (unlikely(data < data_start ||
3388 data > xdp->data_end - ETH_HLEN))
3392 memmove(xdp->data_meta + offset,
3393 xdp->data_meta, metalen);
3394 xdp->data_meta += offset;
3400 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3401 .func = bpf_xdp_adjust_head,
3403 .ret_type = RET_INTEGER,
3404 .arg1_type = ARG_PTR_TO_CTX,
3405 .arg2_type = ARG_ANYTHING,
3408 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3410 void *data_end = xdp->data_end + offset;
3412 /* only shrinking is allowed for now. */
3413 if (unlikely(offset >= 0))
3416 if (unlikely(data_end < xdp->data + ETH_HLEN))
3419 xdp->data_end = data_end;
3424 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3425 .func = bpf_xdp_adjust_tail,
3427 .ret_type = RET_INTEGER,
3428 .arg1_type = ARG_PTR_TO_CTX,
3429 .arg2_type = ARG_ANYTHING,
3432 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3434 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3435 void *meta = xdp->data_meta + offset;
3436 unsigned long metalen = xdp->data - meta;
3438 if (xdp_data_meta_unsupported(xdp))
3440 if (unlikely(meta < xdp_frame_end ||
3443 if (unlikely((metalen & (sizeof(__u32) - 1)) ||
3447 xdp->data_meta = meta;
3452 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3453 .func = bpf_xdp_adjust_meta,
3455 .ret_type = RET_INTEGER,
3456 .arg1_type = ARG_PTR_TO_CTX,
3457 .arg2_type = ARG_ANYTHING,
3460 static int __bpf_tx_xdp(struct net_device *dev,
3461 struct bpf_map *map,
3462 struct xdp_buff *xdp,
3465 struct xdp_frame *xdpf;
3468 if (!dev->netdev_ops->ndo_xdp_xmit) {
3472 err = xdp_ok_fwd_dev(dev, xdp->data_end - xdp->data);
3476 xdpf = convert_to_xdp_frame(xdp);
3477 if (unlikely(!xdpf))
3480 sent = dev->netdev_ops->ndo_xdp_xmit(dev, 1, &xdpf, XDP_XMIT_FLUSH);
3487 xdp_do_redirect_slow(struct net_device *dev, struct xdp_buff *xdp,
3488 struct bpf_prog *xdp_prog, struct bpf_redirect_info *ri)
3490 struct net_device *fwd;
3491 u32 index = ri->tgt_index;
3494 fwd = dev_get_by_index_rcu(dev_net(dev), index);
3496 if (unlikely(!fwd)) {
3501 err = __bpf_tx_xdp(fwd, NULL, xdp, 0);
3505 _trace_xdp_redirect(dev, xdp_prog, index);
3508 _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3512 static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd,
3513 struct bpf_map *map,
3514 struct xdp_buff *xdp,
3519 switch (map->map_type) {
3520 case BPF_MAP_TYPE_DEVMAP: {
3521 struct bpf_dtab_netdev *dst = fwd;
3523 err = dev_map_enqueue(dst, xdp, dev_rx);
3528 case BPF_MAP_TYPE_CPUMAP: {
3529 struct bpf_cpu_map_entry *rcpu = fwd;
3531 err = cpu_map_enqueue(rcpu, xdp, dev_rx);
3536 case BPF_MAP_TYPE_XSKMAP: {
3537 struct xdp_sock *xs = fwd;
3539 err = __xsk_map_redirect(map, xdp, xs);
3548 void xdp_do_flush_map(void)
3550 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3551 struct bpf_map *map = ri->map_to_flush;
3553 ri->map_to_flush = NULL;
3555 switch (map->map_type) {
3556 case BPF_MAP_TYPE_DEVMAP:
3557 __dev_map_flush(map);
3559 case BPF_MAP_TYPE_CPUMAP:
3560 __cpu_map_flush(map);
3562 case BPF_MAP_TYPE_XSKMAP:
3563 __xsk_map_flush(map);
3570 EXPORT_SYMBOL_GPL(xdp_do_flush_map);
3572 static inline void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index)
3574 switch (map->map_type) {
3575 case BPF_MAP_TYPE_DEVMAP:
3576 return __dev_map_lookup_elem(map, index);
3577 case BPF_MAP_TYPE_CPUMAP:
3578 return __cpu_map_lookup_elem(map, index);
3579 case BPF_MAP_TYPE_XSKMAP:
3580 return __xsk_map_lookup_elem(map, index);
3586 void bpf_clear_redirect_map(struct bpf_map *map)
3588 struct bpf_redirect_info *ri;
3591 for_each_possible_cpu(cpu) {
3592 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3593 /* Avoid polluting remote cacheline due to writes if
3594 * not needed. Once we pass this test, we need the
3595 * cmpxchg() to make sure it hasn't been changed in
3596 * the meantime by remote CPU.
3598 if (unlikely(READ_ONCE(ri->map) == map))
3599 cmpxchg(&ri->map, map, NULL);
3603 static int xdp_do_redirect_map(struct net_device *dev, struct xdp_buff *xdp,
3604 struct bpf_prog *xdp_prog, struct bpf_map *map,
3605 struct bpf_redirect_info *ri)
3607 u32 index = ri->tgt_index;
3608 void *fwd = ri->tgt_value;
3612 ri->tgt_value = NULL;
3613 WRITE_ONCE(ri->map, NULL);
3615 if (ri->map_to_flush && unlikely(ri->map_to_flush != map))
3618 err = __bpf_tx_xdp_map(dev, fwd, map, xdp, index);
3622 ri->map_to_flush = map;
3623 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3626 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3630 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3631 struct bpf_prog *xdp_prog)
3633 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3634 struct bpf_map *map = READ_ONCE(ri->map);
3637 return xdp_do_redirect_map(dev, xdp, xdp_prog, map, ri);
3639 return xdp_do_redirect_slow(dev, xdp, xdp_prog, ri);
3641 EXPORT_SYMBOL_GPL(xdp_do_redirect);
3643 static int xdp_do_generic_redirect_map(struct net_device *dev,
3644 struct sk_buff *skb,
3645 struct xdp_buff *xdp,
3646 struct bpf_prog *xdp_prog,
3647 struct bpf_map *map)
3649 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3650 u32 index = ri->tgt_index;
3651 void *fwd = ri->tgt_value;
3655 ri->tgt_value = NULL;
3656 WRITE_ONCE(ri->map, NULL);
3658 if (map->map_type == BPF_MAP_TYPE_DEVMAP) {
3659 struct bpf_dtab_netdev *dst = fwd;
3661 err = dev_map_generic_redirect(dst, skb, xdp_prog);
3664 } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) {
3665 struct xdp_sock *xs = fwd;
3667 err = xsk_generic_rcv(xs, xdp);
3672 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
3677 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3680 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3684 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
3685 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
3687 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3688 struct bpf_map *map = READ_ONCE(ri->map);
3689 u32 index = ri->tgt_index;
3690 struct net_device *fwd;
3694 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog,
3697 fwd = dev_get_by_index_rcu(dev_net(dev), index);
3698 if (unlikely(!fwd)) {
3703 err = xdp_ok_fwd_dev(fwd, skb->len);
3708 _trace_xdp_redirect(dev, xdp_prog, index);
3709 generic_xdp_tx(skb, xdp_prog);
3712 _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3715 EXPORT_SYMBOL_GPL(xdp_do_generic_redirect);
3717 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
3719 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3721 if (unlikely(flags))
3725 ri->tgt_index = ifindex;
3726 ri->tgt_value = NULL;
3727 WRITE_ONCE(ri->map, NULL);
3729 return XDP_REDIRECT;
3732 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
3733 .func = bpf_xdp_redirect,
3735 .ret_type = RET_INTEGER,
3736 .arg1_type = ARG_ANYTHING,
3737 .arg2_type = ARG_ANYTHING,
3740 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
3743 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3745 /* Lower bits of the flags are used as return code on lookup failure */
3746 if (unlikely(flags > XDP_TX))
3749 ri->tgt_value = __xdp_map_lookup_elem(map, ifindex);
3750 if (unlikely(!ri->tgt_value)) {
3751 /* If the lookup fails we want to clear out the state in the
3752 * redirect_info struct completely, so that if an eBPF program
3753 * performs multiple lookups, the last one always takes
3756 WRITE_ONCE(ri->map, NULL);
3761 ri->tgt_index = ifindex;
3762 WRITE_ONCE(ri->map, map);
3764 return XDP_REDIRECT;
3767 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
3768 .func = bpf_xdp_redirect_map,
3770 .ret_type = RET_INTEGER,
3771 .arg1_type = ARG_CONST_MAP_PTR,
3772 .arg2_type = ARG_ANYTHING,
3773 .arg3_type = ARG_ANYTHING,
3776 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
3777 unsigned long off, unsigned long len)
3779 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
3783 if (ptr != dst_buff)
3784 memcpy(dst_buff, ptr, len);
3789 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
3790 u64, flags, void *, meta, u64, meta_size)
3792 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
3794 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
3796 if (unlikely(skb_size > skb->len))
3799 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
3803 static const struct bpf_func_proto bpf_skb_event_output_proto = {
3804 .func = bpf_skb_event_output,
3806 .ret_type = RET_INTEGER,
3807 .arg1_type = ARG_PTR_TO_CTX,
3808 .arg2_type = ARG_CONST_MAP_PTR,
3809 .arg3_type = ARG_ANYTHING,
3810 .arg4_type = ARG_PTR_TO_MEM,
3811 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
3814 static unsigned short bpf_tunnel_key_af(u64 flags)
3816 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
3819 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
3820 u32, size, u64, flags)
3822 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3823 u8 compat[sizeof(struct bpf_tunnel_key)];
3827 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
3831 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
3835 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3838 case offsetof(struct bpf_tunnel_key, tunnel_label):
3839 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3841 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3842 /* Fixup deprecated structure layouts here, so we have
3843 * a common path later on.
3845 if (ip_tunnel_info_af(info) != AF_INET)
3848 to = (struct bpf_tunnel_key *)compat;
3855 to->tunnel_id = be64_to_cpu(info->key.tun_id);
3856 to->tunnel_tos = info->key.tos;
3857 to->tunnel_ttl = info->key.ttl;
3860 if (flags & BPF_F_TUNINFO_IPV6) {
3861 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
3862 sizeof(to->remote_ipv6));
3863 to->tunnel_label = be32_to_cpu(info->key.label);
3865 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
3866 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
3867 to->tunnel_label = 0;
3870 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
3871 memcpy(to_orig, to, size);
3875 memset(to_orig, 0, size);
3879 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
3880 .func = bpf_skb_get_tunnel_key,
3882 .ret_type = RET_INTEGER,
3883 .arg1_type = ARG_PTR_TO_CTX,
3884 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
3885 .arg3_type = ARG_CONST_SIZE,
3886 .arg4_type = ARG_ANYTHING,
3889 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
3891 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3894 if (unlikely(!info ||
3895 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
3899 if (unlikely(size < info->options_len)) {
3904 ip_tunnel_info_opts_get(to, info);
3905 if (size > info->options_len)
3906 memset(to + info->options_len, 0, size - info->options_len);
3908 return info->options_len;
3910 memset(to, 0, size);
3914 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
3915 .func = bpf_skb_get_tunnel_opt,
3917 .ret_type = RET_INTEGER,
3918 .arg1_type = ARG_PTR_TO_CTX,
3919 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
3920 .arg3_type = ARG_CONST_SIZE,
3923 static struct metadata_dst __percpu *md_dst;
3925 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
3926 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
3928 struct metadata_dst *md = this_cpu_ptr(md_dst);
3929 u8 compat[sizeof(struct bpf_tunnel_key)];
3930 struct ip_tunnel_info *info;
3932 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
3933 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
3935 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3937 case offsetof(struct bpf_tunnel_key, tunnel_label):
3938 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3939 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3940 /* Fixup deprecated structure layouts here, so we have
3941 * a common path later on.
3943 memcpy(compat, from, size);
3944 memset(compat + size, 0, sizeof(compat) - size);
3945 from = (const struct bpf_tunnel_key *) compat;
3951 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
3956 dst_hold((struct dst_entry *) md);
3957 skb_dst_set(skb, (struct dst_entry *) md);
3959 info = &md->u.tun_info;
3960 memset(info, 0, sizeof(*info));
3961 info->mode = IP_TUNNEL_INFO_TX;
3963 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
3964 if (flags & BPF_F_DONT_FRAGMENT)
3965 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
3966 if (flags & BPF_F_ZERO_CSUM_TX)
3967 info->key.tun_flags &= ~TUNNEL_CSUM;
3968 if (flags & BPF_F_SEQ_NUMBER)
3969 info->key.tun_flags |= TUNNEL_SEQ;
3971 info->key.tun_id = cpu_to_be64(from->tunnel_id);
3972 info->key.tos = from->tunnel_tos;
3973 info->key.ttl = from->tunnel_ttl;
3975 if (flags & BPF_F_TUNINFO_IPV6) {
3976 info->mode |= IP_TUNNEL_INFO_IPV6;
3977 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
3978 sizeof(from->remote_ipv6));
3979 info->key.label = cpu_to_be32(from->tunnel_label) &
3980 IPV6_FLOWLABEL_MASK;
3982 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
3988 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
3989 .func = bpf_skb_set_tunnel_key,
3991 .ret_type = RET_INTEGER,
3992 .arg1_type = ARG_PTR_TO_CTX,
3993 .arg2_type = ARG_PTR_TO_MEM,
3994 .arg3_type = ARG_CONST_SIZE,
3995 .arg4_type = ARG_ANYTHING,
3998 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
3999 const u8 *, from, u32, size)
4001 struct ip_tunnel_info *info = skb_tunnel_info(skb);
4002 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4004 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4006 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4009 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4014 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4015 .func = bpf_skb_set_tunnel_opt,
4017 .ret_type = RET_INTEGER,
4018 .arg1_type = ARG_PTR_TO_CTX,
4019 .arg2_type = ARG_PTR_TO_MEM,
4020 .arg3_type = ARG_CONST_SIZE,
4023 static const struct bpf_func_proto *
4024 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4027 struct metadata_dst __percpu *tmp;
4029 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4034 if (cmpxchg(&md_dst, NULL, tmp))
4035 metadata_dst_free_percpu(tmp);
4039 case BPF_FUNC_skb_set_tunnel_key:
4040 return &bpf_skb_set_tunnel_key_proto;
4041 case BPF_FUNC_skb_set_tunnel_opt:
4042 return &bpf_skb_set_tunnel_opt_proto;
4048 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4051 struct bpf_array *array = container_of(map, struct bpf_array, map);
4052 struct cgroup *cgrp;
4055 sk = skb_to_full_sk(skb);
4056 if (!sk || !sk_fullsock(sk))
4058 if (unlikely(idx >= array->map.max_entries))
4061 cgrp = READ_ONCE(array->ptrs[idx]);
4062 if (unlikely(!cgrp))
4065 return sk_under_cgroup_hierarchy(sk, cgrp);
4068 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4069 .func = bpf_skb_under_cgroup,
4071 .ret_type = RET_INTEGER,
4072 .arg1_type = ARG_PTR_TO_CTX,
4073 .arg2_type = ARG_CONST_MAP_PTR,
4074 .arg3_type = ARG_ANYTHING,
4077 #ifdef CONFIG_SOCK_CGROUP_DATA
4078 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4080 struct sock *sk = skb_to_full_sk(skb);
4081 struct cgroup *cgrp;
4083 if (!sk || !sk_fullsock(sk))
4086 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4087 return cgrp->kn->id.id;
4090 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4091 .func = bpf_skb_cgroup_id,
4093 .ret_type = RET_INTEGER,
4094 .arg1_type = ARG_PTR_TO_CTX,
4097 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4100 struct sock *sk = skb_to_full_sk(skb);
4101 struct cgroup *ancestor;
4102 struct cgroup *cgrp;
4104 if (!sk || !sk_fullsock(sk))
4107 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4108 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4112 return ancestor->kn->id.id;
4115 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4116 .func = bpf_skb_ancestor_cgroup_id,
4118 .ret_type = RET_INTEGER,
4119 .arg1_type = ARG_PTR_TO_CTX,
4120 .arg2_type = ARG_ANYTHING,
4124 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4125 unsigned long off, unsigned long len)
4127 memcpy(dst_buff, src_buff + off, len);
4131 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4132 u64, flags, void *, meta, u64, meta_size)
4134 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4136 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4138 if (unlikely(xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4141 return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4142 xdp_size, bpf_xdp_copy);
4145 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4146 .func = bpf_xdp_event_output,
4148 .ret_type = RET_INTEGER,
4149 .arg1_type = ARG_PTR_TO_CTX,
4150 .arg2_type = ARG_CONST_MAP_PTR,
4151 .arg3_type = ARG_ANYTHING,
4152 .arg4_type = ARG_PTR_TO_MEM,
4153 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4156 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4158 return skb->sk ? sock_gen_cookie(skb->sk) : 0;
4161 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4162 .func = bpf_get_socket_cookie,
4164 .ret_type = RET_INTEGER,
4165 .arg1_type = ARG_PTR_TO_CTX,
4168 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4170 return sock_gen_cookie(ctx->sk);
4173 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4174 .func = bpf_get_socket_cookie_sock_addr,
4176 .ret_type = RET_INTEGER,
4177 .arg1_type = ARG_PTR_TO_CTX,
4180 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4182 return sock_gen_cookie(ctx->sk);
4185 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4186 .func = bpf_get_socket_cookie_sock_ops,
4188 .ret_type = RET_INTEGER,
4189 .arg1_type = ARG_PTR_TO_CTX,
4192 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4194 struct sock *sk = sk_to_full_sk(skb->sk);
4197 if (!sk || !sk_fullsock(sk))
4199 kuid = sock_net_uid(sock_net(sk), sk);
4200 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4203 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4204 .func = bpf_get_socket_uid,
4206 .ret_type = RET_INTEGER,
4207 .arg1_type = ARG_PTR_TO_CTX,
4210 BPF_CALL_5(bpf_sockopt_event_output, struct bpf_sock_ops_kern *, bpf_sock,
4211 struct bpf_map *, map, u64, flags, void *, data, u64, size)
4213 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
4216 return bpf_event_output(map, flags, data, size, NULL, 0, NULL);
4219 static const struct bpf_func_proto bpf_sockopt_event_output_proto = {
4220 .func = bpf_sockopt_event_output,
4222 .ret_type = RET_INTEGER,
4223 .arg1_type = ARG_PTR_TO_CTX,
4224 .arg2_type = ARG_CONST_MAP_PTR,
4225 .arg3_type = ARG_ANYTHING,
4226 .arg4_type = ARG_PTR_TO_MEM,
4227 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4230 BPF_CALL_5(bpf_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
4231 int, level, int, optname, char *, optval, int, optlen)
4233 struct sock *sk = bpf_sock->sk;
4237 if (!sk_fullsock(sk))
4240 if (level == SOL_SOCKET) {
4241 if (optlen != sizeof(int))
4243 val = *((int *)optval);
4245 /* Only some socketops are supported */
4248 val = min_t(u32, val, sysctl_rmem_max);
4249 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4250 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
4253 val = min_t(u32, val, sysctl_wmem_max);
4254 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4255 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
4257 case SO_MAX_PACING_RATE: /* 32bit version */
4259 cmpxchg(&sk->sk_pacing_status,
4262 sk->sk_max_pacing_rate = (val == ~0U) ? ~0UL : val;
4263 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4264 sk->sk_max_pacing_rate);
4267 sk->sk_priority = val;
4272 sk->sk_rcvlowat = val ? : 1;
4275 if (sk->sk_mark != val) {
4284 } else if (level == SOL_IP) {
4285 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4288 val = *((int *)optval);
4289 /* Only some options are supported */
4292 if (val < -1 || val > 0xff) {
4295 struct inet_sock *inet = inet_sk(sk);
4305 #if IS_ENABLED(CONFIG_IPV6)
4306 } else if (level == SOL_IPV6) {
4307 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4310 val = *((int *)optval);
4311 /* Only some options are supported */
4314 if (val < -1 || val > 0xff) {
4317 struct ipv6_pinfo *np = inet6_sk(sk);
4328 } else if (level == SOL_TCP &&
4329 sk->sk_prot->setsockopt == tcp_setsockopt) {
4330 if (optname == TCP_CONGESTION) {
4331 char name[TCP_CA_NAME_MAX];
4332 bool reinit = bpf_sock->op > BPF_SOCK_OPS_NEEDS_ECN;
4334 strncpy(name, optval, min_t(long, optlen,
4335 TCP_CA_NAME_MAX-1));
4336 name[TCP_CA_NAME_MAX-1] = 0;
4337 ret = tcp_set_congestion_control(sk, name, false,
4340 struct tcp_sock *tp = tcp_sk(sk);
4342 if (optlen != sizeof(int))
4345 val = *((int *)optval);
4346 /* Only some options are supported */
4349 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4354 case TCP_BPF_SNDCWND_CLAMP:
4358 tp->snd_cwnd_clamp = val;
4359 tp->snd_ssthresh = val;
4363 if (val < 0 || val > 1)
4379 static const struct bpf_func_proto bpf_setsockopt_proto = {
4380 .func = bpf_setsockopt,
4382 .ret_type = RET_INTEGER,
4383 .arg1_type = ARG_PTR_TO_CTX,
4384 .arg2_type = ARG_ANYTHING,
4385 .arg3_type = ARG_ANYTHING,
4386 .arg4_type = ARG_PTR_TO_MEM,
4387 .arg5_type = ARG_CONST_SIZE,
4390 BPF_CALL_5(bpf_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
4391 int, level, int, optname, char *, optval, int, optlen)
4393 struct sock *sk = bpf_sock->sk;
4395 if (!sk_fullsock(sk))
4398 if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4399 struct inet_connection_sock *icsk;
4400 struct tcp_sock *tp;
4403 case TCP_CONGESTION:
4404 icsk = inet_csk(sk);
4406 if (!icsk->icsk_ca_ops || optlen <= 1)
4408 strncpy(optval, icsk->icsk_ca_ops->name, optlen);
4409 optval[optlen - 1] = 0;
4414 if (optlen <= 0 || !tp->saved_syn ||
4415 optlen > tp->saved_syn[0])
4417 memcpy(optval, tp->saved_syn + 1, optlen);
4422 } else if (level == SOL_IP) {
4423 struct inet_sock *inet = inet_sk(sk);
4425 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4428 /* Only some options are supported */
4431 *((int *)optval) = (int)inet->tos;
4436 #if IS_ENABLED(CONFIG_IPV6)
4437 } else if (level == SOL_IPV6) {
4438 struct ipv6_pinfo *np = inet6_sk(sk);
4440 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4443 /* Only some options are supported */
4446 *((int *)optval) = (int)np->tclass;
4458 memset(optval, 0, optlen);
4462 static const struct bpf_func_proto bpf_getsockopt_proto = {
4463 .func = bpf_getsockopt,
4465 .ret_type = RET_INTEGER,
4466 .arg1_type = ARG_PTR_TO_CTX,
4467 .arg2_type = ARG_ANYTHING,
4468 .arg3_type = ARG_ANYTHING,
4469 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
4470 .arg5_type = ARG_CONST_SIZE,
4473 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
4476 struct sock *sk = bpf_sock->sk;
4477 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
4479 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
4482 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
4484 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
4487 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
4488 .func = bpf_sock_ops_cb_flags_set,
4490 .ret_type = RET_INTEGER,
4491 .arg1_type = ARG_PTR_TO_CTX,
4492 .arg2_type = ARG_ANYTHING,
4495 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
4496 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
4498 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
4502 struct sock *sk = ctx->sk;
4505 /* Binding to port can be expensive so it's prohibited in the helper.
4506 * Only binding to IP is supported.
4509 if (addr_len < offsetofend(struct sockaddr, sa_family))
4511 if (addr->sa_family == AF_INET) {
4512 if (addr_len < sizeof(struct sockaddr_in))
4514 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
4516 return __inet_bind(sk, addr, addr_len, true, false);
4517 #if IS_ENABLED(CONFIG_IPV6)
4518 } else if (addr->sa_family == AF_INET6) {
4519 if (addr_len < SIN6_LEN_RFC2133)
4521 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
4523 /* ipv6_bpf_stub cannot be NULL, since it's called from
4524 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
4526 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, true, false);
4527 #endif /* CONFIG_IPV6 */
4529 #endif /* CONFIG_INET */
4531 return -EAFNOSUPPORT;
4534 static const struct bpf_func_proto bpf_bind_proto = {
4537 .ret_type = RET_INTEGER,
4538 .arg1_type = ARG_PTR_TO_CTX,
4539 .arg2_type = ARG_PTR_TO_MEM,
4540 .arg3_type = ARG_CONST_SIZE,
4544 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
4545 struct bpf_xfrm_state *, to, u32, size, u64, flags)
4547 const struct sec_path *sp = skb_sec_path(skb);
4548 const struct xfrm_state *x;
4550 if (!sp || unlikely(index >= sp->len || flags))
4553 x = sp->xvec[index];
4555 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
4558 to->reqid = x->props.reqid;
4559 to->spi = x->id.spi;
4560 to->family = x->props.family;
4563 if (to->family == AF_INET6) {
4564 memcpy(to->remote_ipv6, x->props.saddr.a6,
4565 sizeof(to->remote_ipv6));
4567 to->remote_ipv4 = x->props.saddr.a4;
4568 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4573 memset(to, 0, size);
4577 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
4578 .func = bpf_skb_get_xfrm_state,
4580 .ret_type = RET_INTEGER,
4581 .arg1_type = ARG_PTR_TO_CTX,
4582 .arg2_type = ARG_ANYTHING,
4583 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
4584 .arg4_type = ARG_CONST_SIZE,
4585 .arg5_type = ARG_ANYTHING,
4589 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
4590 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
4591 const struct neighbour *neigh,
4592 const struct net_device *dev)
4594 memcpy(params->dmac, neigh->ha, ETH_ALEN);
4595 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
4596 params->h_vlan_TCI = 0;
4597 params->h_vlan_proto = 0;
4598 params->ifindex = dev->ifindex;
4604 #if IS_ENABLED(CONFIG_INET)
4605 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4606 u32 flags, bool check_mtu)
4608 struct fib_nh_common *nhc;
4609 struct in_device *in_dev;
4610 struct neighbour *neigh;
4611 struct net_device *dev;
4612 struct fib_result res;
4617 dev = dev_get_by_index_rcu(net, params->ifindex);
4621 /* verify forwarding is enabled on this interface */
4622 in_dev = __in_dev_get_rcu(dev);
4623 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
4624 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4626 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4628 fl4.flowi4_oif = params->ifindex;
4630 fl4.flowi4_iif = params->ifindex;
4633 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
4634 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
4635 fl4.flowi4_flags = 0;
4637 fl4.flowi4_proto = params->l4_protocol;
4638 fl4.daddr = params->ipv4_dst;
4639 fl4.saddr = params->ipv4_src;
4640 fl4.fl4_sport = params->sport;
4641 fl4.fl4_dport = params->dport;
4643 if (flags & BPF_FIB_LOOKUP_DIRECT) {
4644 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4645 struct fib_table *tb;
4647 tb = fib_get_table(net, tbid);
4649 return BPF_FIB_LKUP_RET_NOT_FWDED;
4651 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
4653 fl4.flowi4_mark = 0;
4654 fl4.flowi4_secid = 0;
4655 fl4.flowi4_tun_key.tun_id = 0;
4656 fl4.flowi4_uid = sock_net_uid(net, NULL);
4658 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
4662 /* map fib lookup errors to RTN_ type */
4664 return BPF_FIB_LKUP_RET_BLACKHOLE;
4665 if (err == -EHOSTUNREACH)
4666 return BPF_FIB_LKUP_RET_UNREACHABLE;
4668 return BPF_FIB_LKUP_RET_PROHIBIT;
4670 return BPF_FIB_LKUP_RET_NOT_FWDED;
4673 if (res.type != RTN_UNICAST)
4674 return BPF_FIB_LKUP_RET_NOT_FWDED;
4676 if (fib_info_num_path(res.fi) > 1)
4677 fib_select_path(net, &res, &fl4, NULL);
4680 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
4681 if (params->tot_len > mtu)
4682 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4687 /* do not handle lwt encaps right now */
4688 if (nhc->nhc_lwtstate)
4689 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4693 params->rt_metric = res.fi->fib_priority;
4695 /* xdp and cls_bpf programs are run in RCU-bh so
4696 * rcu_read_lock_bh is not needed here
4698 if (likely(nhc->nhc_gw_family != AF_INET6)) {
4699 if (nhc->nhc_gw_family)
4700 params->ipv4_dst = nhc->nhc_gw.ipv4;
4702 neigh = __ipv4_neigh_lookup_noref(dev,
4703 (__force u32)params->ipv4_dst);
4705 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
4707 params->family = AF_INET6;
4708 *dst = nhc->nhc_gw.ipv6;
4709 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
4713 return BPF_FIB_LKUP_RET_NO_NEIGH;
4715 return bpf_fib_set_fwd_params(params, neigh, dev);
4719 #if IS_ENABLED(CONFIG_IPV6)
4720 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4721 u32 flags, bool check_mtu)
4723 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
4724 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
4725 struct fib6_result res = {};
4726 struct neighbour *neigh;
4727 struct net_device *dev;
4728 struct inet6_dev *idev;
4734 /* link local addresses are never forwarded */
4735 if (rt6_need_strict(dst) || rt6_need_strict(src))
4736 return BPF_FIB_LKUP_RET_NOT_FWDED;
4738 dev = dev_get_by_index_rcu(net, params->ifindex);
4742 idev = __in6_dev_get_safely(dev);
4743 if (unlikely(!idev || !idev->cnf.forwarding))
4744 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4746 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4748 oif = fl6.flowi6_oif = params->ifindex;
4750 oif = fl6.flowi6_iif = params->ifindex;
4752 strict = RT6_LOOKUP_F_HAS_SADDR;
4754 fl6.flowlabel = params->flowinfo;
4755 fl6.flowi6_scope = 0;
4756 fl6.flowi6_flags = 0;
4759 fl6.flowi6_proto = params->l4_protocol;
4762 fl6.fl6_sport = params->sport;
4763 fl6.fl6_dport = params->dport;
4765 if (flags & BPF_FIB_LOOKUP_DIRECT) {
4766 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4767 struct fib6_table *tb;
4769 tb = ipv6_stub->fib6_get_table(net, tbid);
4771 return BPF_FIB_LKUP_RET_NOT_FWDED;
4773 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
4776 fl6.flowi6_mark = 0;
4777 fl6.flowi6_secid = 0;
4778 fl6.flowi6_tun_key.tun_id = 0;
4779 fl6.flowi6_uid = sock_net_uid(net, NULL);
4781 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
4784 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
4785 res.f6i == net->ipv6.fib6_null_entry))
4786 return BPF_FIB_LKUP_RET_NOT_FWDED;
4788 switch (res.fib6_type) {
4789 /* only unicast is forwarded */
4793 return BPF_FIB_LKUP_RET_BLACKHOLE;
4794 case RTN_UNREACHABLE:
4795 return BPF_FIB_LKUP_RET_UNREACHABLE;
4797 return BPF_FIB_LKUP_RET_PROHIBIT;
4799 return BPF_FIB_LKUP_RET_NOT_FWDED;
4802 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
4803 fl6.flowi6_oif != 0, NULL, strict);
4806 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
4807 if (params->tot_len > mtu)
4808 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4811 if (res.nh->fib_nh_lws)
4812 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4814 if (res.nh->fib_nh_gw_family)
4815 *dst = res.nh->fib_nh_gw6;
4817 dev = res.nh->fib_nh_dev;
4818 params->rt_metric = res.f6i->fib6_metric;
4820 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
4823 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
4825 return BPF_FIB_LKUP_RET_NO_NEIGH;
4827 return bpf_fib_set_fwd_params(params, neigh, dev);
4831 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
4832 struct bpf_fib_lookup *, params, int, plen, u32, flags)
4834 if (plen < sizeof(*params))
4837 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4840 switch (params->family) {
4841 #if IS_ENABLED(CONFIG_INET)
4843 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
4846 #if IS_ENABLED(CONFIG_IPV6)
4848 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
4852 return -EAFNOSUPPORT;
4855 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
4856 .func = bpf_xdp_fib_lookup,
4858 .ret_type = RET_INTEGER,
4859 .arg1_type = ARG_PTR_TO_CTX,
4860 .arg2_type = ARG_PTR_TO_MEM,
4861 .arg3_type = ARG_CONST_SIZE,
4862 .arg4_type = ARG_ANYTHING,
4865 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
4866 struct bpf_fib_lookup *, params, int, plen, u32, flags)
4868 struct net *net = dev_net(skb->dev);
4869 int rc = -EAFNOSUPPORT;
4871 if (plen < sizeof(*params))
4874 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4877 switch (params->family) {
4878 #if IS_ENABLED(CONFIG_INET)
4880 rc = bpf_ipv4_fib_lookup(net, params, flags, false);
4883 #if IS_ENABLED(CONFIG_IPV6)
4885 rc = bpf_ipv6_fib_lookup(net, params, flags, false);
4891 struct net_device *dev;
4893 dev = dev_get_by_index_rcu(net, params->ifindex);
4894 if (!is_skb_forwardable(dev, skb))
4895 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
4901 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
4902 .func = bpf_skb_fib_lookup,
4904 .ret_type = RET_INTEGER,
4905 .arg1_type = ARG_PTR_TO_CTX,
4906 .arg2_type = ARG_PTR_TO_MEM,
4907 .arg3_type = ARG_CONST_SIZE,
4908 .arg4_type = ARG_ANYTHING,
4911 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4912 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
4915 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
4917 if (!seg6_validate_srh(srh, len))
4921 case BPF_LWT_ENCAP_SEG6_INLINE:
4922 if (skb->protocol != htons(ETH_P_IPV6))
4925 err = seg6_do_srh_inline(skb, srh);
4927 case BPF_LWT_ENCAP_SEG6:
4928 skb_reset_inner_headers(skb);
4929 skb->encapsulation = 1;
4930 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
4936 bpf_compute_data_pointers(skb);
4940 ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
4941 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
4943 return seg6_lookup_nexthop(skb, NULL, 0);
4945 #endif /* CONFIG_IPV6_SEG6_BPF */
4947 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
4948 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
4951 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
4955 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
4959 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4960 case BPF_LWT_ENCAP_SEG6:
4961 case BPF_LWT_ENCAP_SEG6_INLINE:
4962 return bpf_push_seg6_encap(skb, type, hdr, len);
4964 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
4965 case BPF_LWT_ENCAP_IP:
4966 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
4973 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
4974 void *, hdr, u32, len)
4977 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
4978 case BPF_LWT_ENCAP_IP:
4979 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
4986 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
4987 .func = bpf_lwt_in_push_encap,
4989 .ret_type = RET_INTEGER,
4990 .arg1_type = ARG_PTR_TO_CTX,
4991 .arg2_type = ARG_ANYTHING,
4992 .arg3_type = ARG_PTR_TO_MEM,
4993 .arg4_type = ARG_CONST_SIZE
4996 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
4997 .func = bpf_lwt_xmit_push_encap,
4999 .ret_type = RET_INTEGER,
5000 .arg1_type = ARG_PTR_TO_CTX,
5001 .arg2_type = ARG_ANYTHING,
5002 .arg3_type = ARG_PTR_TO_MEM,
5003 .arg4_type = ARG_CONST_SIZE
5006 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5007 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
5008 const void *, from, u32, len)
5010 struct seg6_bpf_srh_state *srh_state =
5011 this_cpu_ptr(&seg6_bpf_srh_states);
5012 struct ipv6_sr_hdr *srh = srh_state->srh;
5013 void *srh_tlvs, *srh_end, *ptr;
5019 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
5020 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
5022 ptr = skb->data + offset;
5023 if (ptr >= srh_tlvs && ptr + len <= srh_end)
5024 srh_state->valid = false;
5025 else if (ptr < (void *)&srh->flags ||
5026 ptr + len > (void *)&srh->segments)
5029 if (unlikely(bpf_try_make_writable(skb, offset + len)))
5031 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5033 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5035 memcpy(skb->data + offset, from, len);
5039 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
5040 .func = bpf_lwt_seg6_store_bytes,
5042 .ret_type = RET_INTEGER,
5043 .arg1_type = ARG_PTR_TO_CTX,
5044 .arg2_type = ARG_ANYTHING,
5045 .arg3_type = ARG_PTR_TO_MEM,
5046 .arg4_type = ARG_CONST_SIZE
5049 static void bpf_update_srh_state(struct sk_buff *skb)
5051 struct seg6_bpf_srh_state *srh_state =
5052 this_cpu_ptr(&seg6_bpf_srh_states);
5055 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
5056 srh_state->srh = NULL;
5058 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5059 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
5060 srh_state->valid = true;
5064 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
5065 u32, action, void *, param, u32, param_len)
5067 struct seg6_bpf_srh_state *srh_state =
5068 this_cpu_ptr(&seg6_bpf_srh_states);
5073 case SEG6_LOCAL_ACTION_END_X:
5074 if (!seg6_bpf_has_valid_srh(skb))
5076 if (param_len != sizeof(struct in6_addr))
5078 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
5079 case SEG6_LOCAL_ACTION_END_T:
5080 if (!seg6_bpf_has_valid_srh(skb))
5082 if (param_len != sizeof(int))
5084 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5085 case SEG6_LOCAL_ACTION_END_DT6:
5086 if (!seg6_bpf_has_valid_srh(skb))
5088 if (param_len != sizeof(int))
5091 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
5093 if (!pskb_pull(skb, hdroff))
5096 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
5097 skb_reset_network_header(skb);
5098 skb_reset_transport_header(skb);
5099 skb->encapsulation = 0;
5101 bpf_compute_data_pointers(skb);
5102 bpf_update_srh_state(skb);
5103 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5104 case SEG6_LOCAL_ACTION_END_B6:
5105 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5107 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
5110 bpf_update_srh_state(skb);
5113 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
5114 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5116 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
5119 bpf_update_srh_state(skb);
5127 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
5128 .func = bpf_lwt_seg6_action,
5130 .ret_type = RET_INTEGER,
5131 .arg1_type = ARG_PTR_TO_CTX,
5132 .arg2_type = ARG_ANYTHING,
5133 .arg3_type = ARG_PTR_TO_MEM,
5134 .arg4_type = ARG_CONST_SIZE
5137 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
5140 struct seg6_bpf_srh_state *srh_state =
5141 this_cpu_ptr(&seg6_bpf_srh_states);
5142 struct ipv6_sr_hdr *srh = srh_state->srh;
5143 void *srh_end, *srh_tlvs, *ptr;
5144 struct ipv6hdr *hdr;
5148 if (unlikely(srh == NULL))
5151 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
5152 ((srh->first_segment + 1) << 4));
5153 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
5155 ptr = skb->data + offset;
5157 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
5159 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
5163 ret = skb_cow_head(skb, len);
5164 if (unlikely(ret < 0))
5167 ret = bpf_skb_net_hdr_push(skb, offset, len);
5169 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
5172 bpf_compute_data_pointers(skb);
5173 if (unlikely(ret < 0))
5176 hdr = (struct ipv6hdr *)skb->data;
5177 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5179 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5181 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5182 srh_state->hdrlen += len;
5183 srh_state->valid = false;
5187 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
5188 .func = bpf_lwt_seg6_adjust_srh,
5190 .ret_type = RET_INTEGER,
5191 .arg1_type = ARG_PTR_TO_CTX,
5192 .arg2_type = ARG_ANYTHING,
5193 .arg3_type = ARG_ANYTHING,
5195 #endif /* CONFIG_IPV6_SEG6_BPF */
5198 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
5199 int dif, int sdif, u8 family, u8 proto)
5201 bool refcounted = false;
5202 struct sock *sk = NULL;
5204 if (family == AF_INET) {
5205 __be32 src4 = tuple->ipv4.saddr;
5206 __be32 dst4 = tuple->ipv4.daddr;
5208 if (proto == IPPROTO_TCP)
5209 sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
5210 src4, tuple->ipv4.sport,
5211 dst4, tuple->ipv4.dport,
5212 dif, sdif, &refcounted);
5214 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
5215 dst4, tuple->ipv4.dport,
5216 dif, sdif, &udp_table, NULL);
5217 #if IS_ENABLED(CONFIG_IPV6)
5219 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
5220 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
5222 if (proto == IPPROTO_TCP)
5223 sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
5224 src6, tuple->ipv6.sport,
5225 dst6, ntohs(tuple->ipv6.dport),
5226 dif, sdif, &refcounted);
5227 else if (likely(ipv6_bpf_stub))
5228 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
5229 src6, tuple->ipv6.sport,
5230 dst6, tuple->ipv6.dport,
5236 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
5237 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
5243 /* bpf_skc_lookup performs the core lookup for different types of sockets,
5244 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
5245 * Returns the socket as an 'unsigned long' to simplify the casting in the
5246 * callers to satisfy BPF_CALL declarations.
5248 static struct sock *
5249 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5250 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
5253 struct sock *sk = NULL;
5254 u8 family = AF_UNSPEC;
5258 if (len == sizeof(tuple->ipv4))
5260 else if (len == sizeof(tuple->ipv6))
5265 if (unlikely(family == AF_UNSPEC || flags ||
5266 !((s32)netns_id < 0 || netns_id <= S32_MAX)))
5269 if (family == AF_INET)
5270 sdif = inet_sdif(skb);
5272 sdif = inet6_sdif(skb);
5274 if ((s32)netns_id < 0) {
5276 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
5278 net = get_net_ns_by_id(caller_net, netns_id);
5281 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
5289 static struct sock *
5290 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5291 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
5294 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
5295 ifindex, proto, netns_id, flags);
5298 sk = sk_to_full_sk(sk);
5299 if (!sk_fullsock(sk)) {
5300 if (!sock_flag(sk, SOCK_RCU_FREE))
5309 static struct sock *
5310 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5311 u8 proto, u64 netns_id, u64 flags)
5313 struct net *caller_net;
5317 caller_net = dev_net(skb->dev);
5318 ifindex = skb->dev->ifindex;
5320 caller_net = sock_net(skb->sk);
5324 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
5328 static struct sock *
5329 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5330 u8 proto, u64 netns_id, u64 flags)
5332 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
5336 sk = sk_to_full_sk(sk);
5337 if (!sk_fullsock(sk)) {
5338 if (!sock_flag(sk, SOCK_RCU_FREE))
5347 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
5348 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5350 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
5354 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
5355 .func = bpf_skc_lookup_tcp,
5358 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
5359 .arg1_type = ARG_PTR_TO_CTX,
5360 .arg2_type = ARG_PTR_TO_MEM,
5361 .arg3_type = ARG_CONST_SIZE,
5362 .arg4_type = ARG_ANYTHING,
5363 .arg5_type = ARG_ANYTHING,
5366 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
5367 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5369 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
5373 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
5374 .func = bpf_sk_lookup_tcp,
5377 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5378 .arg1_type = ARG_PTR_TO_CTX,
5379 .arg2_type = ARG_PTR_TO_MEM,
5380 .arg3_type = ARG_CONST_SIZE,
5381 .arg4_type = ARG_ANYTHING,
5382 .arg5_type = ARG_ANYTHING,
5385 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
5386 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5388 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
5392 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
5393 .func = bpf_sk_lookup_udp,
5396 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5397 .arg1_type = ARG_PTR_TO_CTX,
5398 .arg2_type = ARG_PTR_TO_MEM,
5399 .arg3_type = ARG_CONST_SIZE,
5400 .arg4_type = ARG_ANYTHING,
5401 .arg5_type = ARG_ANYTHING,
5404 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
5406 if (!sock_flag(sk, SOCK_RCU_FREE))
5411 static const struct bpf_func_proto bpf_sk_release_proto = {
5412 .func = bpf_sk_release,
5414 .ret_type = RET_INTEGER,
5415 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
5418 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
5419 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
5421 struct net *caller_net = dev_net(ctx->rxq->dev);
5422 int ifindex = ctx->rxq->dev->ifindex;
5424 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
5425 ifindex, IPPROTO_UDP, netns_id,
5429 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
5430 .func = bpf_xdp_sk_lookup_udp,
5433 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5434 .arg1_type = ARG_PTR_TO_CTX,
5435 .arg2_type = ARG_PTR_TO_MEM,
5436 .arg3_type = ARG_CONST_SIZE,
5437 .arg4_type = ARG_ANYTHING,
5438 .arg5_type = ARG_ANYTHING,
5441 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
5442 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
5444 struct net *caller_net = dev_net(ctx->rxq->dev);
5445 int ifindex = ctx->rxq->dev->ifindex;
5447 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
5448 ifindex, IPPROTO_TCP, netns_id,
5452 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
5453 .func = bpf_xdp_skc_lookup_tcp,
5456 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
5457 .arg1_type = ARG_PTR_TO_CTX,
5458 .arg2_type = ARG_PTR_TO_MEM,
5459 .arg3_type = ARG_CONST_SIZE,
5460 .arg4_type = ARG_ANYTHING,
5461 .arg5_type = ARG_ANYTHING,
5464 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
5465 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
5467 struct net *caller_net = dev_net(ctx->rxq->dev);
5468 int ifindex = ctx->rxq->dev->ifindex;
5470 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
5471 ifindex, IPPROTO_TCP, netns_id,
5475 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
5476 .func = bpf_xdp_sk_lookup_tcp,
5479 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5480 .arg1_type = ARG_PTR_TO_CTX,
5481 .arg2_type = ARG_PTR_TO_MEM,
5482 .arg3_type = ARG_CONST_SIZE,
5483 .arg4_type = ARG_ANYTHING,
5484 .arg5_type = ARG_ANYTHING,
5487 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
5488 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5490 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
5491 sock_net(ctx->sk), 0,
5492 IPPROTO_TCP, netns_id, flags);
5495 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
5496 .func = bpf_sock_addr_skc_lookup_tcp,
5498 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
5499 .arg1_type = ARG_PTR_TO_CTX,
5500 .arg2_type = ARG_PTR_TO_MEM,
5501 .arg3_type = ARG_CONST_SIZE,
5502 .arg4_type = ARG_ANYTHING,
5503 .arg5_type = ARG_ANYTHING,
5506 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
5507 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5509 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
5510 sock_net(ctx->sk), 0, IPPROTO_TCP,
5514 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
5515 .func = bpf_sock_addr_sk_lookup_tcp,
5517 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5518 .arg1_type = ARG_PTR_TO_CTX,
5519 .arg2_type = ARG_PTR_TO_MEM,
5520 .arg3_type = ARG_CONST_SIZE,
5521 .arg4_type = ARG_ANYTHING,
5522 .arg5_type = ARG_ANYTHING,
5525 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
5526 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5528 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
5529 sock_net(ctx->sk), 0, IPPROTO_UDP,
5533 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
5534 .func = bpf_sock_addr_sk_lookup_udp,
5536 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5537 .arg1_type = ARG_PTR_TO_CTX,
5538 .arg2_type = ARG_PTR_TO_MEM,
5539 .arg3_type = ARG_CONST_SIZE,
5540 .arg4_type = ARG_ANYTHING,
5541 .arg5_type = ARG_ANYTHING,
5544 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
5545 struct bpf_insn_access_aux *info)
5547 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
5551 if (off % size != 0)
5555 case offsetof(struct bpf_tcp_sock, bytes_received):
5556 case offsetof(struct bpf_tcp_sock, bytes_acked):
5557 return size == sizeof(__u64);
5559 return size == sizeof(__u32);
5563 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
5564 const struct bpf_insn *si,
5565 struct bpf_insn *insn_buf,
5566 struct bpf_prog *prog, u32 *target_size)
5568 struct bpf_insn *insn = insn_buf;
5570 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \
5572 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, FIELD) > \
5573 FIELD_SIZEOF(struct bpf_tcp_sock, FIELD)); \
5574 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
5575 si->dst_reg, si->src_reg, \
5576 offsetof(struct tcp_sock, FIELD)); \
5579 #define BPF_INET_SOCK_GET_COMMON(FIELD) \
5581 BUILD_BUG_ON(FIELD_SIZEOF(struct inet_connection_sock, \
5583 FIELD_SIZEOF(struct bpf_tcp_sock, FIELD)); \
5584 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
5585 struct inet_connection_sock, \
5587 si->dst_reg, si->src_reg, \
5589 struct inet_connection_sock, \
5593 if (insn > insn_buf)
5594 return insn - insn_buf;
5597 case offsetof(struct bpf_tcp_sock, rtt_min):
5598 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, rtt_min) !=
5599 sizeof(struct minmax));
5600 BUILD_BUG_ON(sizeof(struct minmax) <
5601 sizeof(struct minmax_sample));
5603 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5604 offsetof(struct tcp_sock, rtt_min) +
5605 offsetof(struct minmax_sample, v));
5607 case offsetof(struct bpf_tcp_sock, snd_cwnd):
5608 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
5610 case offsetof(struct bpf_tcp_sock, srtt_us):
5611 BPF_TCP_SOCK_GET_COMMON(srtt_us);
5613 case offsetof(struct bpf_tcp_sock, snd_ssthresh):
5614 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
5616 case offsetof(struct bpf_tcp_sock, rcv_nxt):
5617 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
5619 case offsetof(struct bpf_tcp_sock, snd_nxt):
5620 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
5622 case offsetof(struct bpf_tcp_sock, snd_una):
5623 BPF_TCP_SOCK_GET_COMMON(snd_una);
5625 case offsetof(struct bpf_tcp_sock, mss_cache):
5626 BPF_TCP_SOCK_GET_COMMON(mss_cache);
5628 case offsetof(struct bpf_tcp_sock, ecn_flags):
5629 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
5631 case offsetof(struct bpf_tcp_sock, rate_delivered):
5632 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
5634 case offsetof(struct bpf_tcp_sock, rate_interval_us):
5635 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
5637 case offsetof(struct bpf_tcp_sock, packets_out):
5638 BPF_TCP_SOCK_GET_COMMON(packets_out);
5640 case offsetof(struct bpf_tcp_sock, retrans_out):
5641 BPF_TCP_SOCK_GET_COMMON(retrans_out);
5643 case offsetof(struct bpf_tcp_sock, total_retrans):
5644 BPF_TCP_SOCK_GET_COMMON(total_retrans);
5646 case offsetof(struct bpf_tcp_sock, segs_in):
5647 BPF_TCP_SOCK_GET_COMMON(segs_in);
5649 case offsetof(struct bpf_tcp_sock, data_segs_in):
5650 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
5652 case offsetof(struct bpf_tcp_sock, segs_out):
5653 BPF_TCP_SOCK_GET_COMMON(segs_out);
5655 case offsetof(struct bpf_tcp_sock, data_segs_out):
5656 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
5658 case offsetof(struct bpf_tcp_sock, lost_out):
5659 BPF_TCP_SOCK_GET_COMMON(lost_out);
5661 case offsetof(struct bpf_tcp_sock, sacked_out):
5662 BPF_TCP_SOCK_GET_COMMON(sacked_out);
5664 case offsetof(struct bpf_tcp_sock, bytes_received):
5665 BPF_TCP_SOCK_GET_COMMON(bytes_received);
5667 case offsetof(struct bpf_tcp_sock, bytes_acked):
5668 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
5670 case offsetof(struct bpf_tcp_sock, dsack_dups):
5671 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
5673 case offsetof(struct bpf_tcp_sock, delivered):
5674 BPF_TCP_SOCK_GET_COMMON(delivered);
5676 case offsetof(struct bpf_tcp_sock, delivered_ce):
5677 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
5679 case offsetof(struct bpf_tcp_sock, icsk_retransmits):
5680 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
5684 return insn - insn_buf;
5687 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
5689 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
5690 return (unsigned long)sk;
5692 return (unsigned long)NULL;
5695 const struct bpf_func_proto bpf_tcp_sock_proto = {
5696 .func = bpf_tcp_sock,
5698 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
5699 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
5702 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
5704 sk = sk_to_full_sk(sk);
5706 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
5707 return (unsigned long)sk;
5709 return (unsigned long)NULL;
5712 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
5713 .func = bpf_get_listener_sock,
5715 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5716 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
5719 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
5721 unsigned int iphdr_len;
5723 if (skb->protocol == cpu_to_be16(ETH_P_IP))
5724 iphdr_len = sizeof(struct iphdr);
5725 else if (skb->protocol == cpu_to_be16(ETH_P_IPV6))
5726 iphdr_len = sizeof(struct ipv6hdr);
5730 if (skb_headlen(skb) < iphdr_len)
5733 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
5736 return INET_ECN_set_ce(skb);
5739 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
5740 struct bpf_insn_access_aux *info)
5742 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
5745 if (off % size != 0)
5750 return size == sizeof(__u32);
5754 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
5755 const struct bpf_insn *si,
5756 struct bpf_insn *insn_buf,
5757 struct bpf_prog *prog, u32 *target_size)
5759 struct bpf_insn *insn = insn_buf;
5761 #define BPF_XDP_SOCK_GET(FIELD) \
5763 BUILD_BUG_ON(FIELD_SIZEOF(struct xdp_sock, FIELD) > \
5764 FIELD_SIZEOF(struct bpf_xdp_sock, FIELD)); \
5765 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
5766 si->dst_reg, si->src_reg, \
5767 offsetof(struct xdp_sock, FIELD)); \
5771 case offsetof(struct bpf_xdp_sock, queue_id):
5772 BPF_XDP_SOCK_GET(queue_id);
5776 return insn - insn_buf;
5779 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
5780 .func = bpf_skb_ecn_set_ce,
5782 .ret_type = RET_INTEGER,
5783 .arg1_type = ARG_PTR_TO_CTX,
5786 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
5787 struct tcphdr *, th, u32, th_len)
5789 #ifdef CONFIG_SYN_COOKIES
5793 if (unlikely(th_len < sizeof(*th)))
5796 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
5797 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
5800 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
5803 if (!th->ack || th->rst || th->syn)
5806 if (tcp_synq_no_recent_overflow(sk))
5809 cookie = ntohl(th->ack_seq) - 1;
5811 switch (sk->sk_family) {
5813 if (unlikely(iph_len < sizeof(struct iphdr)))
5816 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
5819 #if IS_BUILTIN(CONFIG_IPV6)
5821 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
5824 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
5826 #endif /* CONFIG_IPV6 */
5829 return -EPROTONOSUPPORT;
5841 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
5842 .func = bpf_tcp_check_syncookie,
5845 .ret_type = RET_INTEGER,
5846 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
5847 .arg2_type = ARG_PTR_TO_MEM,
5848 .arg3_type = ARG_CONST_SIZE,
5849 .arg4_type = ARG_PTR_TO_MEM,
5850 .arg5_type = ARG_CONST_SIZE,
5853 #endif /* CONFIG_INET */
5855 bool bpf_helper_changes_pkt_data(void *func)
5857 if (func == bpf_skb_vlan_push ||
5858 func == bpf_skb_vlan_pop ||
5859 func == bpf_skb_store_bytes ||
5860 func == bpf_skb_change_proto ||
5861 func == bpf_skb_change_head ||
5862 func == sk_skb_change_head ||
5863 func == bpf_skb_change_tail ||
5864 func == sk_skb_change_tail ||
5865 func == bpf_skb_adjust_room ||
5866 func == bpf_skb_pull_data ||
5867 func == sk_skb_pull_data ||
5868 func == bpf_clone_redirect ||
5869 func == bpf_l3_csum_replace ||
5870 func == bpf_l4_csum_replace ||
5871 func == bpf_xdp_adjust_head ||
5872 func == bpf_xdp_adjust_meta ||
5873 func == bpf_msg_pull_data ||
5874 func == bpf_msg_push_data ||
5875 func == bpf_msg_pop_data ||
5876 func == bpf_xdp_adjust_tail ||
5877 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5878 func == bpf_lwt_seg6_store_bytes ||
5879 func == bpf_lwt_seg6_adjust_srh ||
5880 func == bpf_lwt_seg6_action ||
5882 func == bpf_lwt_in_push_encap ||
5883 func == bpf_lwt_xmit_push_encap)
5889 static const struct bpf_func_proto *
5890 bpf_base_func_proto(enum bpf_func_id func_id)
5893 case BPF_FUNC_map_lookup_elem:
5894 return &bpf_map_lookup_elem_proto;
5895 case BPF_FUNC_map_update_elem:
5896 return &bpf_map_update_elem_proto;
5897 case BPF_FUNC_map_delete_elem:
5898 return &bpf_map_delete_elem_proto;
5899 case BPF_FUNC_map_push_elem:
5900 return &bpf_map_push_elem_proto;
5901 case BPF_FUNC_map_pop_elem:
5902 return &bpf_map_pop_elem_proto;
5903 case BPF_FUNC_map_peek_elem:
5904 return &bpf_map_peek_elem_proto;
5905 case BPF_FUNC_get_prandom_u32:
5906 return &bpf_get_prandom_u32_proto;
5907 case BPF_FUNC_get_smp_processor_id:
5908 return &bpf_get_raw_smp_processor_id_proto;
5909 case BPF_FUNC_get_numa_node_id:
5910 return &bpf_get_numa_node_id_proto;
5911 case BPF_FUNC_tail_call:
5912 return &bpf_tail_call_proto;
5913 case BPF_FUNC_ktime_get_ns:
5914 return &bpf_ktime_get_ns_proto;
5919 if (!capable(CAP_SYS_ADMIN))
5923 case BPF_FUNC_spin_lock:
5924 return &bpf_spin_lock_proto;
5925 case BPF_FUNC_spin_unlock:
5926 return &bpf_spin_unlock_proto;
5927 case BPF_FUNC_trace_printk:
5928 return bpf_get_trace_printk_proto();
5934 static const struct bpf_func_proto *
5935 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5938 /* inet and inet6 sockets are created in a process
5939 * context so there is always a valid uid/gid
5941 case BPF_FUNC_get_current_uid_gid:
5942 return &bpf_get_current_uid_gid_proto;
5943 case BPF_FUNC_get_local_storage:
5944 return &bpf_get_local_storage_proto;
5946 return bpf_base_func_proto(func_id);
5950 static const struct bpf_func_proto *
5951 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5954 /* inet and inet6 sockets are created in a process
5955 * context so there is always a valid uid/gid
5957 case BPF_FUNC_get_current_uid_gid:
5958 return &bpf_get_current_uid_gid_proto;
5960 switch (prog->expected_attach_type) {
5961 case BPF_CGROUP_INET4_CONNECT:
5962 case BPF_CGROUP_INET6_CONNECT:
5963 return &bpf_bind_proto;
5967 case BPF_FUNC_get_socket_cookie:
5968 return &bpf_get_socket_cookie_sock_addr_proto;
5969 case BPF_FUNC_get_local_storage:
5970 return &bpf_get_local_storage_proto;
5972 case BPF_FUNC_sk_lookup_tcp:
5973 return &bpf_sock_addr_sk_lookup_tcp_proto;
5974 case BPF_FUNC_sk_lookup_udp:
5975 return &bpf_sock_addr_sk_lookup_udp_proto;
5976 case BPF_FUNC_sk_release:
5977 return &bpf_sk_release_proto;
5978 case BPF_FUNC_skc_lookup_tcp:
5979 return &bpf_sock_addr_skc_lookup_tcp_proto;
5980 #endif /* CONFIG_INET */
5981 case BPF_FUNC_sk_storage_get:
5982 return &bpf_sk_storage_get_proto;
5983 case BPF_FUNC_sk_storage_delete:
5984 return &bpf_sk_storage_delete_proto;
5986 return bpf_base_func_proto(func_id);
5990 static const struct bpf_func_proto *
5991 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5994 case BPF_FUNC_skb_load_bytes:
5995 return &bpf_skb_load_bytes_proto;
5996 case BPF_FUNC_skb_load_bytes_relative:
5997 return &bpf_skb_load_bytes_relative_proto;
5998 case BPF_FUNC_get_socket_cookie:
5999 return &bpf_get_socket_cookie_proto;
6000 case BPF_FUNC_get_socket_uid:
6001 return &bpf_get_socket_uid_proto;
6003 return bpf_base_func_proto(func_id);
6007 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
6008 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
6010 static const struct bpf_func_proto *
6011 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6014 case BPF_FUNC_get_local_storage:
6015 return &bpf_get_local_storage_proto;
6016 case BPF_FUNC_sk_fullsock:
6017 return &bpf_sk_fullsock_proto;
6018 case BPF_FUNC_sk_storage_get:
6019 return &bpf_sk_storage_get_proto;
6020 case BPF_FUNC_sk_storage_delete:
6021 return &bpf_sk_storage_delete_proto;
6022 #ifdef CONFIG_SOCK_CGROUP_DATA
6023 case BPF_FUNC_skb_cgroup_id:
6024 return &bpf_skb_cgroup_id_proto;
6027 case BPF_FUNC_tcp_sock:
6028 return &bpf_tcp_sock_proto;
6029 case BPF_FUNC_get_listener_sock:
6030 return &bpf_get_listener_sock_proto;
6031 case BPF_FUNC_skb_ecn_set_ce:
6032 return &bpf_skb_ecn_set_ce_proto;
6035 return sk_filter_func_proto(func_id, prog);
6039 static const struct bpf_func_proto *
6040 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6043 case BPF_FUNC_skb_store_bytes:
6044 return &bpf_skb_store_bytes_proto;
6045 case BPF_FUNC_skb_load_bytes:
6046 return &bpf_skb_load_bytes_proto;
6047 case BPF_FUNC_skb_load_bytes_relative:
6048 return &bpf_skb_load_bytes_relative_proto;
6049 case BPF_FUNC_skb_pull_data:
6050 return &bpf_skb_pull_data_proto;
6051 case BPF_FUNC_csum_diff:
6052 return &bpf_csum_diff_proto;
6053 case BPF_FUNC_csum_update:
6054 return &bpf_csum_update_proto;
6055 case BPF_FUNC_l3_csum_replace:
6056 return &bpf_l3_csum_replace_proto;
6057 case BPF_FUNC_l4_csum_replace:
6058 return &bpf_l4_csum_replace_proto;
6059 case BPF_FUNC_clone_redirect:
6060 return &bpf_clone_redirect_proto;
6061 case BPF_FUNC_get_cgroup_classid:
6062 return &bpf_get_cgroup_classid_proto;
6063 case BPF_FUNC_skb_vlan_push:
6064 return &bpf_skb_vlan_push_proto;
6065 case BPF_FUNC_skb_vlan_pop:
6066 return &bpf_skb_vlan_pop_proto;
6067 case BPF_FUNC_skb_change_proto:
6068 return &bpf_skb_change_proto_proto;
6069 case BPF_FUNC_skb_change_type:
6070 return &bpf_skb_change_type_proto;
6071 case BPF_FUNC_skb_adjust_room:
6072 return &bpf_skb_adjust_room_proto;
6073 case BPF_FUNC_skb_change_tail:
6074 return &bpf_skb_change_tail_proto;
6075 case BPF_FUNC_skb_get_tunnel_key:
6076 return &bpf_skb_get_tunnel_key_proto;
6077 case BPF_FUNC_skb_set_tunnel_key:
6078 return bpf_get_skb_set_tunnel_proto(func_id);
6079 case BPF_FUNC_skb_get_tunnel_opt:
6080 return &bpf_skb_get_tunnel_opt_proto;
6081 case BPF_FUNC_skb_set_tunnel_opt:
6082 return bpf_get_skb_set_tunnel_proto(func_id);
6083 case BPF_FUNC_redirect:
6084 return &bpf_redirect_proto;
6085 case BPF_FUNC_get_route_realm:
6086 return &bpf_get_route_realm_proto;
6087 case BPF_FUNC_get_hash_recalc:
6088 return &bpf_get_hash_recalc_proto;
6089 case BPF_FUNC_set_hash_invalid:
6090 return &bpf_set_hash_invalid_proto;
6091 case BPF_FUNC_set_hash:
6092 return &bpf_set_hash_proto;
6093 case BPF_FUNC_perf_event_output:
6094 return &bpf_skb_event_output_proto;
6095 case BPF_FUNC_get_smp_processor_id:
6096 return &bpf_get_smp_processor_id_proto;
6097 case BPF_FUNC_skb_under_cgroup:
6098 return &bpf_skb_under_cgroup_proto;
6099 case BPF_FUNC_get_socket_cookie:
6100 return &bpf_get_socket_cookie_proto;
6101 case BPF_FUNC_get_socket_uid:
6102 return &bpf_get_socket_uid_proto;
6103 case BPF_FUNC_fib_lookup:
6104 return &bpf_skb_fib_lookup_proto;
6105 case BPF_FUNC_sk_fullsock:
6106 return &bpf_sk_fullsock_proto;
6107 case BPF_FUNC_sk_storage_get:
6108 return &bpf_sk_storage_get_proto;
6109 case BPF_FUNC_sk_storage_delete:
6110 return &bpf_sk_storage_delete_proto;
6112 case BPF_FUNC_skb_get_xfrm_state:
6113 return &bpf_skb_get_xfrm_state_proto;
6115 #ifdef CONFIG_SOCK_CGROUP_DATA
6116 case BPF_FUNC_skb_cgroup_id:
6117 return &bpf_skb_cgroup_id_proto;
6118 case BPF_FUNC_skb_ancestor_cgroup_id:
6119 return &bpf_skb_ancestor_cgroup_id_proto;
6122 case BPF_FUNC_sk_lookup_tcp:
6123 return &bpf_sk_lookup_tcp_proto;
6124 case BPF_FUNC_sk_lookup_udp:
6125 return &bpf_sk_lookup_udp_proto;
6126 case BPF_FUNC_sk_release:
6127 return &bpf_sk_release_proto;
6128 case BPF_FUNC_tcp_sock:
6129 return &bpf_tcp_sock_proto;
6130 case BPF_FUNC_get_listener_sock:
6131 return &bpf_get_listener_sock_proto;
6132 case BPF_FUNC_skc_lookup_tcp:
6133 return &bpf_skc_lookup_tcp_proto;
6134 case BPF_FUNC_tcp_check_syncookie:
6135 return &bpf_tcp_check_syncookie_proto;
6136 case BPF_FUNC_skb_ecn_set_ce:
6137 return &bpf_skb_ecn_set_ce_proto;
6140 return bpf_base_func_proto(func_id);
6144 static const struct bpf_func_proto *
6145 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6148 case BPF_FUNC_perf_event_output:
6149 return &bpf_xdp_event_output_proto;
6150 case BPF_FUNC_get_smp_processor_id:
6151 return &bpf_get_smp_processor_id_proto;
6152 case BPF_FUNC_csum_diff:
6153 return &bpf_csum_diff_proto;
6154 case BPF_FUNC_xdp_adjust_head:
6155 return &bpf_xdp_adjust_head_proto;
6156 case BPF_FUNC_xdp_adjust_meta:
6157 return &bpf_xdp_adjust_meta_proto;
6158 case BPF_FUNC_redirect:
6159 return &bpf_xdp_redirect_proto;
6160 case BPF_FUNC_redirect_map:
6161 return &bpf_xdp_redirect_map_proto;
6162 case BPF_FUNC_xdp_adjust_tail:
6163 return &bpf_xdp_adjust_tail_proto;
6164 case BPF_FUNC_fib_lookup:
6165 return &bpf_xdp_fib_lookup_proto;
6167 case BPF_FUNC_sk_lookup_udp:
6168 return &bpf_xdp_sk_lookup_udp_proto;
6169 case BPF_FUNC_sk_lookup_tcp:
6170 return &bpf_xdp_sk_lookup_tcp_proto;
6171 case BPF_FUNC_sk_release:
6172 return &bpf_sk_release_proto;
6173 case BPF_FUNC_skc_lookup_tcp:
6174 return &bpf_xdp_skc_lookup_tcp_proto;
6175 case BPF_FUNC_tcp_check_syncookie:
6176 return &bpf_tcp_check_syncookie_proto;
6179 return bpf_base_func_proto(func_id);
6183 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
6184 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
6186 static const struct bpf_func_proto *
6187 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6190 case BPF_FUNC_setsockopt:
6191 return &bpf_setsockopt_proto;
6192 case BPF_FUNC_getsockopt:
6193 return &bpf_getsockopt_proto;
6194 case BPF_FUNC_sock_ops_cb_flags_set:
6195 return &bpf_sock_ops_cb_flags_set_proto;
6196 case BPF_FUNC_sock_map_update:
6197 return &bpf_sock_map_update_proto;
6198 case BPF_FUNC_sock_hash_update:
6199 return &bpf_sock_hash_update_proto;
6200 case BPF_FUNC_get_socket_cookie:
6201 return &bpf_get_socket_cookie_sock_ops_proto;
6202 case BPF_FUNC_get_local_storage:
6203 return &bpf_get_local_storage_proto;
6204 case BPF_FUNC_perf_event_output:
6205 return &bpf_sockopt_event_output_proto;
6206 case BPF_FUNC_sk_storage_get:
6207 return &bpf_sk_storage_get_proto;
6208 case BPF_FUNC_sk_storage_delete:
6209 return &bpf_sk_storage_delete_proto;
6211 case BPF_FUNC_tcp_sock:
6212 return &bpf_tcp_sock_proto;
6213 #endif /* CONFIG_INET */
6215 return bpf_base_func_proto(func_id);
6219 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
6220 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
6222 static const struct bpf_func_proto *
6223 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6226 case BPF_FUNC_msg_redirect_map:
6227 return &bpf_msg_redirect_map_proto;
6228 case BPF_FUNC_msg_redirect_hash:
6229 return &bpf_msg_redirect_hash_proto;
6230 case BPF_FUNC_msg_apply_bytes:
6231 return &bpf_msg_apply_bytes_proto;
6232 case BPF_FUNC_msg_cork_bytes:
6233 return &bpf_msg_cork_bytes_proto;
6234 case BPF_FUNC_msg_pull_data:
6235 return &bpf_msg_pull_data_proto;
6236 case BPF_FUNC_msg_push_data:
6237 return &bpf_msg_push_data_proto;
6238 case BPF_FUNC_msg_pop_data:
6239 return &bpf_msg_pop_data_proto;
6241 return bpf_base_func_proto(func_id);
6245 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
6246 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
6248 static const struct bpf_func_proto *
6249 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6252 case BPF_FUNC_skb_store_bytes:
6253 return &bpf_skb_store_bytes_proto;
6254 case BPF_FUNC_skb_load_bytes:
6255 return &bpf_skb_load_bytes_proto;
6256 case BPF_FUNC_skb_pull_data:
6257 return &sk_skb_pull_data_proto;
6258 case BPF_FUNC_skb_change_tail:
6259 return &sk_skb_change_tail_proto;
6260 case BPF_FUNC_skb_change_head:
6261 return &sk_skb_change_head_proto;
6262 case BPF_FUNC_get_socket_cookie:
6263 return &bpf_get_socket_cookie_proto;
6264 case BPF_FUNC_get_socket_uid:
6265 return &bpf_get_socket_uid_proto;
6266 case BPF_FUNC_sk_redirect_map:
6267 return &bpf_sk_redirect_map_proto;
6268 case BPF_FUNC_sk_redirect_hash:
6269 return &bpf_sk_redirect_hash_proto;
6271 case BPF_FUNC_sk_lookup_tcp:
6272 return &bpf_sk_lookup_tcp_proto;
6273 case BPF_FUNC_sk_lookup_udp:
6274 return &bpf_sk_lookup_udp_proto;
6275 case BPF_FUNC_sk_release:
6276 return &bpf_sk_release_proto;
6277 case BPF_FUNC_skc_lookup_tcp:
6278 return &bpf_skc_lookup_tcp_proto;
6281 return bpf_base_func_proto(func_id);
6285 static const struct bpf_func_proto *
6286 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6289 case BPF_FUNC_skb_load_bytes:
6290 return &bpf_flow_dissector_load_bytes_proto;
6292 return bpf_base_func_proto(func_id);
6296 static const struct bpf_func_proto *
6297 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6300 case BPF_FUNC_skb_load_bytes:
6301 return &bpf_skb_load_bytes_proto;
6302 case BPF_FUNC_skb_pull_data:
6303 return &bpf_skb_pull_data_proto;
6304 case BPF_FUNC_csum_diff:
6305 return &bpf_csum_diff_proto;
6306 case BPF_FUNC_get_cgroup_classid:
6307 return &bpf_get_cgroup_classid_proto;
6308 case BPF_FUNC_get_route_realm:
6309 return &bpf_get_route_realm_proto;
6310 case BPF_FUNC_get_hash_recalc:
6311 return &bpf_get_hash_recalc_proto;
6312 case BPF_FUNC_perf_event_output:
6313 return &bpf_skb_event_output_proto;
6314 case BPF_FUNC_get_smp_processor_id:
6315 return &bpf_get_smp_processor_id_proto;
6316 case BPF_FUNC_skb_under_cgroup:
6317 return &bpf_skb_under_cgroup_proto;
6319 return bpf_base_func_proto(func_id);
6323 static const struct bpf_func_proto *
6324 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6327 case BPF_FUNC_lwt_push_encap:
6328 return &bpf_lwt_in_push_encap_proto;
6330 return lwt_out_func_proto(func_id, prog);
6334 static const struct bpf_func_proto *
6335 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6338 case BPF_FUNC_skb_get_tunnel_key:
6339 return &bpf_skb_get_tunnel_key_proto;
6340 case BPF_FUNC_skb_set_tunnel_key:
6341 return bpf_get_skb_set_tunnel_proto(func_id);
6342 case BPF_FUNC_skb_get_tunnel_opt:
6343 return &bpf_skb_get_tunnel_opt_proto;
6344 case BPF_FUNC_skb_set_tunnel_opt:
6345 return bpf_get_skb_set_tunnel_proto(func_id);
6346 case BPF_FUNC_redirect:
6347 return &bpf_redirect_proto;
6348 case BPF_FUNC_clone_redirect:
6349 return &bpf_clone_redirect_proto;
6350 case BPF_FUNC_skb_change_tail:
6351 return &bpf_skb_change_tail_proto;
6352 case BPF_FUNC_skb_change_head:
6353 return &bpf_skb_change_head_proto;
6354 case BPF_FUNC_skb_store_bytes:
6355 return &bpf_skb_store_bytes_proto;
6356 case BPF_FUNC_csum_update:
6357 return &bpf_csum_update_proto;
6358 case BPF_FUNC_l3_csum_replace:
6359 return &bpf_l3_csum_replace_proto;
6360 case BPF_FUNC_l4_csum_replace:
6361 return &bpf_l4_csum_replace_proto;
6362 case BPF_FUNC_set_hash_invalid:
6363 return &bpf_set_hash_invalid_proto;
6364 case BPF_FUNC_lwt_push_encap:
6365 return &bpf_lwt_xmit_push_encap_proto;
6367 return lwt_out_func_proto(func_id, prog);
6371 static const struct bpf_func_proto *
6372 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6375 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6376 case BPF_FUNC_lwt_seg6_store_bytes:
6377 return &bpf_lwt_seg6_store_bytes_proto;
6378 case BPF_FUNC_lwt_seg6_action:
6379 return &bpf_lwt_seg6_action_proto;
6380 case BPF_FUNC_lwt_seg6_adjust_srh:
6381 return &bpf_lwt_seg6_adjust_srh_proto;
6384 return lwt_out_func_proto(func_id, prog);
6388 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
6389 const struct bpf_prog *prog,
6390 struct bpf_insn_access_aux *info)
6392 const int size_default = sizeof(__u32);
6394 if (off < 0 || off >= sizeof(struct __sk_buff))
6397 /* The verifier guarantees that size > 0. */
6398 if (off % size != 0)
6402 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6403 if (off + size > offsetofend(struct __sk_buff, cb[4]))
6406 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
6407 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
6408 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
6409 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
6410 case bpf_ctx_range(struct __sk_buff, data):
6411 case bpf_ctx_range(struct __sk_buff, data_meta):
6412 case bpf_ctx_range(struct __sk_buff, data_end):
6413 if (size != size_default)
6416 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
6418 case bpf_ctx_range(struct __sk_buff, tstamp):
6419 if (size != sizeof(__u64))
6422 case offsetof(struct __sk_buff, sk):
6423 if (type == BPF_WRITE || size != sizeof(__u64))
6425 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
6428 /* Only narrow read access allowed for now. */
6429 if (type == BPF_WRITE) {
6430 if (size != size_default)
6433 bpf_ctx_record_field_size(info, size_default);
6434 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
6442 static bool sk_filter_is_valid_access(int off, int size,
6443 enum bpf_access_type type,
6444 const struct bpf_prog *prog,
6445 struct bpf_insn_access_aux *info)
6448 case bpf_ctx_range(struct __sk_buff, tc_classid):
6449 case bpf_ctx_range(struct __sk_buff, data):
6450 case bpf_ctx_range(struct __sk_buff, data_meta):
6451 case bpf_ctx_range(struct __sk_buff, data_end):
6452 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6453 case bpf_ctx_range(struct __sk_buff, tstamp):
6454 case bpf_ctx_range(struct __sk_buff, wire_len):
6458 if (type == BPF_WRITE) {
6460 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6467 return bpf_skb_is_valid_access(off, size, type, prog, info);
6470 static bool cg_skb_is_valid_access(int off, int size,
6471 enum bpf_access_type type,
6472 const struct bpf_prog *prog,
6473 struct bpf_insn_access_aux *info)
6476 case bpf_ctx_range(struct __sk_buff, tc_classid):
6477 case bpf_ctx_range(struct __sk_buff, data_meta):
6478 case bpf_ctx_range(struct __sk_buff, wire_len):
6480 case bpf_ctx_range(struct __sk_buff, data):
6481 case bpf_ctx_range(struct __sk_buff, data_end):
6482 if (!capable(CAP_SYS_ADMIN))
6487 if (type == BPF_WRITE) {
6489 case bpf_ctx_range(struct __sk_buff, mark):
6490 case bpf_ctx_range(struct __sk_buff, priority):
6491 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6493 case bpf_ctx_range(struct __sk_buff, tstamp):
6494 if (!capable(CAP_SYS_ADMIN))
6503 case bpf_ctx_range(struct __sk_buff, data):
6504 info->reg_type = PTR_TO_PACKET;
6506 case bpf_ctx_range(struct __sk_buff, data_end):
6507 info->reg_type = PTR_TO_PACKET_END;
6511 return bpf_skb_is_valid_access(off, size, type, prog, info);
6514 static bool lwt_is_valid_access(int off, int size,
6515 enum bpf_access_type type,
6516 const struct bpf_prog *prog,
6517 struct bpf_insn_access_aux *info)
6520 case bpf_ctx_range(struct __sk_buff, tc_classid):
6521 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6522 case bpf_ctx_range(struct __sk_buff, data_meta):
6523 case bpf_ctx_range(struct __sk_buff, tstamp):
6524 case bpf_ctx_range(struct __sk_buff, wire_len):
6528 if (type == BPF_WRITE) {
6530 case bpf_ctx_range(struct __sk_buff, mark):
6531 case bpf_ctx_range(struct __sk_buff, priority):
6532 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6540 case bpf_ctx_range(struct __sk_buff, data):
6541 info->reg_type = PTR_TO_PACKET;
6543 case bpf_ctx_range(struct __sk_buff, data_end):
6544 info->reg_type = PTR_TO_PACKET_END;
6548 return bpf_skb_is_valid_access(off, size, type, prog, info);
6551 /* Attach type specific accesses */
6552 static bool __sock_filter_check_attach_type(int off,
6553 enum bpf_access_type access_type,
6554 enum bpf_attach_type attach_type)
6557 case offsetof(struct bpf_sock, bound_dev_if):
6558 case offsetof(struct bpf_sock, mark):
6559 case offsetof(struct bpf_sock, priority):
6560 switch (attach_type) {
6561 case BPF_CGROUP_INET_SOCK_CREATE:
6566 case bpf_ctx_range(struct bpf_sock, src_ip4):
6567 switch (attach_type) {
6568 case BPF_CGROUP_INET4_POST_BIND:
6573 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
6574 switch (attach_type) {
6575 case BPF_CGROUP_INET6_POST_BIND:
6580 case bpf_ctx_range(struct bpf_sock, src_port):
6581 switch (attach_type) {
6582 case BPF_CGROUP_INET4_POST_BIND:
6583 case BPF_CGROUP_INET6_POST_BIND:
6590 return access_type == BPF_READ;
6595 bool bpf_sock_common_is_valid_access(int off, int size,
6596 enum bpf_access_type type,
6597 struct bpf_insn_access_aux *info)
6600 case bpf_ctx_range_till(struct bpf_sock, type, priority):
6603 return bpf_sock_is_valid_access(off, size, type, info);
6607 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6608 struct bpf_insn_access_aux *info)
6610 const int size_default = sizeof(__u32);
6612 if (off < 0 || off >= sizeof(struct bpf_sock))
6614 if (off % size != 0)
6618 case offsetof(struct bpf_sock, state):
6619 case offsetof(struct bpf_sock, family):
6620 case offsetof(struct bpf_sock, type):
6621 case offsetof(struct bpf_sock, protocol):
6622 case offsetof(struct bpf_sock, dst_port):
6623 case offsetof(struct bpf_sock, src_port):
6624 case bpf_ctx_range(struct bpf_sock, src_ip4):
6625 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
6626 case bpf_ctx_range(struct bpf_sock, dst_ip4):
6627 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
6628 bpf_ctx_record_field_size(info, size_default);
6629 return bpf_ctx_narrow_access_ok(off, size, size_default);
6632 return size == size_default;
6635 static bool sock_filter_is_valid_access(int off, int size,
6636 enum bpf_access_type type,
6637 const struct bpf_prog *prog,
6638 struct bpf_insn_access_aux *info)
6640 if (!bpf_sock_is_valid_access(off, size, type, info))
6642 return __sock_filter_check_attach_type(off, type,
6643 prog->expected_attach_type);
6646 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
6647 const struct bpf_prog *prog)
6649 /* Neither direct read nor direct write requires any preliminary
6655 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
6656 const struct bpf_prog *prog, int drop_verdict)
6658 struct bpf_insn *insn = insn_buf;
6663 /* if (!skb->cloned)
6666 * (Fast-path, otherwise approximation that we might be
6667 * a clone, do the rest in helper.)
6669 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
6670 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
6671 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
6673 /* ret = bpf_skb_pull_data(skb, 0); */
6674 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
6675 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
6676 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
6677 BPF_FUNC_skb_pull_data);
6680 * return TC_ACT_SHOT;
6682 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
6683 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
6684 *insn++ = BPF_EXIT_INSN();
6687 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
6689 *insn++ = prog->insnsi[0];
6691 return insn - insn_buf;
6694 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
6695 struct bpf_insn *insn_buf)
6697 bool indirect = BPF_MODE(orig->code) == BPF_IND;
6698 struct bpf_insn *insn = insn_buf;
6700 /* We're guaranteed here that CTX is in R6. */
6701 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
6703 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
6705 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
6707 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
6710 switch (BPF_SIZE(orig->code)) {
6712 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
6715 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
6718 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
6722 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
6723 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
6724 *insn++ = BPF_EXIT_INSN();
6726 return insn - insn_buf;
6729 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
6730 const struct bpf_prog *prog)
6732 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
6735 static bool tc_cls_act_is_valid_access(int off, int size,
6736 enum bpf_access_type type,
6737 const struct bpf_prog *prog,
6738 struct bpf_insn_access_aux *info)
6740 if (type == BPF_WRITE) {
6742 case bpf_ctx_range(struct __sk_buff, mark):
6743 case bpf_ctx_range(struct __sk_buff, tc_index):
6744 case bpf_ctx_range(struct __sk_buff, priority):
6745 case bpf_ctx_range(struct __sk_buff, tc_classid):
6746 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6747 case bpf_ctx_range(struct __sk_buff, tstamp):
6748 case bpf_ctx_range(struct __sk_buff, queue_mapping):
6756 case bpf_ctx_range(struct __sk_buff, data):
6757 info->reg_type = PTR_TO_PACKET;
6759 case bpf_ctx_range(struct __sk_buff, data_meta):
6760 info->reg_type = PTR_TO_PACKET_META;
6762 case bpf_ctx_range(struct __sk_buff, data_end):
6763 info->reg_type = PTR_TO_PACKET_END;
6765 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6769 return bpf_skb_is_valid_access(off, size, type, prog, info);
6772 static bool __is_valid_xdp_access(int off, int size)
6774 if (off < 0 || off >= sizeof(struct xdp_md))
6776 if (off % size != 0)
6778 if (size != sizeof(__u32))
6784 static bool xdp_is_valid_access(int off, int size,
6785 enum bpf_access_type type,
6786 const struct bpf_prog *prog,
6787 struct bpf_insn_access_aux *info)
6789 if (type == BPF_WRITE) {
6790 if (bpf_prog_is_dev_bound(prog->aux)) {
6792 case offsetof(struct xdp_md, rx_queue_index):
6793 return __is_valid_xdp_access(off, size);
6800 case offsetof(struct xdp_md, data):
6801 info->reg_type = PTR_TO_PACKET;
6803 case offsetof(struct xdp_md, data_meta):
6804 info->reg_type = PTR_TO_PACKET_META;
6806 case offsetof(struct xdp_md, data_end):
6807 info->reg_type = PTR_TO_PACKET_END;
6811 return __is_valid_xdp_access(off, size);
6814 void bpf_warn_invalid_xdp_action(u32 act)
6816 const u32 act_max = XDP_REDIRECT;
6818 WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
6819 act > act_max ? "Illegal" : "Driver unsupported",
6822 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
6824 static bool sock_addr_is_valid_access(int off, int size,
6825 enum bpf_access_type type,
6826 const struct bpf_prog *prog,
6827 struct bpf_insn_access_aux *info)
6829 const int size_default = sizeof(__u32);
6831 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
6833 if (off % size != 0)
6836 /* Disallow access to IPv6 fields from IPv4 contex and vise
6840 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
6841 switch (prog->expected_attach_type) {
6842 case BPF_CGROUP_INET4_BIND:
6843 case BPF_CGROUP_INET4_CONNECT:
6844 case BPF_CGROUP_UDP4_SENDMSG:
6845 case BPF_CGROUP_UDP4_RECVMSG:
6851 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
6852 switch (prog->expected_attach_type) {
6853 case BPF_CGROUP_INET6_BIND:
6854 case BPF_CGROUP_INET6_CONNECT:
6855 case BPF_CGROUP_UDP6_SENDMSG:
6856 case BPF_CGROUP_UDP6_RECVMSG:
6862 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
6863 switch (prog->expected_attach_type) {
6864 case BPF_CGROUP_UDP4_SENDMSG:
6870 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
6872 switch (prog->expected_attach_type) {
6873 case BPF_CGROUP_UDP6_SENDMSG:
6882 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
6883 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
6884 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
6885 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
6887 /* Only narrow read access allowed for now. */
6888 if (type == BPF_READ) {
6889 bpf_ctx_record_field_size(info, size_default);
6890 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
6893 if (bpf_ctx_wide_store_ok(off, size,
6894 struct bpf_sock_addr,
6898 if (bpf_ctx_wide_store_ok(off, size,
6899 struct bpf_sock_addr,
6903 if (size != size_default)
6907 case bpf_ctx_range(struct bpf_sock_addr, user_port):
6908 if (size != size_default)
6911 case offsetof(struct bpf_sock_addr, sk):
6912 if (type != BPF_READ)
6914 if (size != sizeof(__u64))
6916 info->reg_type = PTR_TO_SOCKET;
6919 if (type == BPF_READ) {
6920 if (size != size_default)
6930 static bool sock_ops_is_valid_access(int off, int size,
6931 enum bpf_access_type type,
6932 const struct bpf_prog *prog,
6933 struct bpf_insn_access_aux *info)
6935 const int size_default = sizeof(__u32);
6937 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
6940 /* The verifier guarantees that size > 0. */
6941 if (off % size != 0)
6944 if (type == BPF_WRITE) {
6946 case offsetof(struct bpf_sock_ops, reply):
6947 case offsetof(struct bpf_sock_ops, sk_txhash):
6948 if (size != size_default)
6956 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
6958 if (size != sizeof(__u64))
6961 case offsetof(struct bpf_sock_ops, sk):
6962 if (size != sizeof(__u64))
6964 info->reg_type = PTR_TO_SOCKET_OR_NULL;
6967 if (size != size_default)
6976 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
6977 const struct bpf_prog *prog)
6979 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
6982 static bool sk_skb_is_valid_access(int off, int size,
6983 enum bpf_access_type type,
6984 const struct bpf_prog *prog,
6985 struct bpf_insn_access_aux *info)
6988 case bpf_ctx_range(struct __sk_buff, tc_classid):
6989 case bpf_ctx_range(struct __sk_buff, data_meta):
6990 case bpf_ctx_range(struct __sk_buff, tstamp):
6991 case bpf_ctx_range(struct __sk_buff, wire_len):
6995 if (type == BPF_WRITE) {
6997 case bpf_ctx_range(struct __sk_buff, tc_index):
6998 case bpf_ctx_range(struct __sk_buff, priority):
7006 case bpf_ctx_range(struct __sk_buff, mark):
7008 case bpf_ctx_range(struct __sk_buff, data):
7009 info->reg_type = PTR_TO_PACKET;
7011 case bpf_ctx_range(struct __sk_buff, data_end):
7012 info->reg_type = PTR_TO_PACKET_END;
7016 return bpf_skb_is_valid_access(off, size, type, prog, info);
7019 static bool sk_msg_is_valid_access(int off, int size,
7020 enum bpf_access_type type,
7021 const struct bpf_prog *prog,
7022 struct bpf_insn_access_aux *info)
7024 if (type == BPF_WRITE)
7027 if (off % size != 0)
7031 case offsetof(struct sk_msg_md, data):
7032 info->reg_type = PTR_TO_PACKET;
7033 if (size != sizeof(__u64))
7036 case offsetof(struct sk_msg_md, data_end):
7037 info->reg_type = PTR_TO_PACKET_END;
7038 if (size != sizeof(__u64))
7041 case bpf_ctx_range(struct sk_msg_md, family):
7042 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
7043 case bpf_ctx_range(struct sk_msg_md, local_ip4):
7044 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
7045 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
7046 case bpf_ctx_range(struct sk_msg_md, remote_port):
7047 case bpf_ctx_range(struct sk_msg_md, local_port):
7048 case bpf_ctx_range(struct sk_msg_md, size):
7049 if (size != sizeof(__u32))
7058 static bool flow_dissector_is_valid_access(int off, int size,
7059 enum bpf_access_type type,
7060 const struct bpf_prog *prog,
7061 struct bpf_insn_access_aux *info)
7063 const int size_default = sizeof(__u32);
7065 if (off < 0 || off >= sizeof(struct __sk_buff))
7068 if (type == BPF_WRITE)
7072 case bpf_ctx_range(struct __sk_buff, data):
7073 if (size != size_default)
7075 info->reg_type = PTR_TO_PACKET;
7077 case bpf_ctx_range(struct __sk_buff, data_end):
7078 if (size != size_default)
7080 info->reg_type = PTR_TO_PACKET_END;
7082 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
7083 if (size != sizeof(__u64))
7085 info->reg_type = PTR_TO_FLOW_KEYS;
7092 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
7093 const struct bpf_insn *si,
7094 struct bpf_insn *insn_buf,
7095 struct bpf_prog *prog,
7099 struct bpf_insn *insn = insn_buf;
7102 case offsetof(struct __sk_buff, data):
7103 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
7104 si->dst_reg, si->src_reg,
7105 offsetof(struct bpf_flow_dissector, data));
7108 case offsetof(struct __sk_buff, data_end):
7109 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
7110 si->dst_reg, si->src_reg,
7111 offsetof(struct bpf_flow_dissector, data_end));
7114 case offsetof(struct __sk_buff, flow_keys):
7115 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
7116 si->dst_reg, si->src_reg,
7117 offsetof(struct bpf_flow_dissector, flow_keys));
7121 return insn - insn_buf;
7124 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
7125 const struct bpf_insn *si,
7126 struct bpf_insn *insn_buf,
7127 struct bpf_prog *prog, u32 *target_size)
7129 struct bpf_insn *insn = insn_buf;
7133 case offsetof(struct __sk_buff, len):
7134 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7135 bpf_target_off(struct sk_buff, len, 4,
7139 case offsetof(struct __sk_buff, protocol):
7140 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
7141 bpf_target_off(struct sk_buff, protocol, 2,
7145 case offsetof(struct __sk_buff, vlan_proto):
7146 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
7147 bpf_target_off(struct sk_buff, vlan_proto, 2,
7151 case offsetof(struct __sk_buff, priority):
7152 if (type == BPF_WRITE)
7153 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7154 bpf_target_off(struct sk_buff, priority, 4,
7157 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7158 bpf_target_off(struct sk_buff, priority, 4,
7162 case offsetof(struct __sk_buff, ingress_ifindex):
7163 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7164 bpf_target_off(struct sk_buff, skb_iif, 4,
7168 case offsetof(struct __sk_buff, ifindex):
7169 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
7170 si->dst_reg, si->src_reg,
7171 offsetof(struct sk_buff, dev));
7172 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
7173 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7174 bpf_target_off(struct net_device, ifindex, 4,
7178 case offsetof(struct __sk_buff, hash):
7179 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7180 bpf_target_off(struct sk_buff, hash, 4,
7184 case offsetof(struct __sk_buff, mark):
7185 if (type == BPF_WRITE)
7186 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7187 bpf_target_off(struct sk_buff, mark, 4,
7190 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7191 bpf_target_off(struct sk_buff, mark, 4,
7195 case offsetof(struct __sk_buff, pkt_type):
7197 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
7199 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
7200 #ifdef __BIG_ENDIAN_BITFIELD
7201 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
7205 case offsetof(struct __sk_buff, queue_mapping):
7206 if (type == BPF_WRITE) {
7207 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
7208 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
7209 bpf_target_off(struct sk_buff,
7213 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
7214 bpf_target_off(struct sk_buff,
7220 case offsetof(struct __sk_buff, vlan_present):
7222 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
7223 PKT_VLAN_PRESENT_OFFSET());
7224 if (PKT_VLAN_PRESENT_BIT)
7225 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
7226 if (PKT_VLAN_PRESENT_BIT < 7)
7227 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
7230 case offsetof(struct __sk_buff, vlan_tci):
7231 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
7232 bpf_target_off(struct sk_buff, vlan_tci, 2,
7236 case offsetof(struct __sk_buff, cb[0]) ...
7237 offsetofend(struct __sk_buff, cb[4]) - 1:
7238 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20);
7239 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
7240 offsetof(struct qdisc_skb_cb, data)) %
7243 prog->cb_access = 1;
7245 off -= offsetof(struct __sk_buff, cb[0]);
7246 off += offsetof(struct sk_buff, cb);
7247 off += offsetof(struct qdisc_skb_cb, data);
7248 if (type == BPF_WRITE)
7249 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
7252 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
7256 case offsetof(struct __sk_buff, tc_classid):
7257 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, tc_classid) != 2);
7260 off -= offsetof(struct __sk_buff, tc_classid);
7261 off += offsetof(struct sk_buff, cb);
7262 off += offsetof(struct qdisc_skb_cb, tc_classid);
7264 if (type == BPF_WRITE)
7265 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
7268 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
7272 case offsetof(struct __sk_buff, data):
7273 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
7274 si->dst_reg, si->src_reg,
7275 offsetof(struct sk_buff, data));
7278 case offsetof(struct __sk_buff, data_meta):
7280 off -= offsetof(struct __sk_buff, data_meta);
7281 off += offsetof(struct sk_buff, cb);
7282 off += offsetof(struct bpf_skb_data_end, data_meta);
7283 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
7287 case offsetof(struct __sk_buff, data_end):
7289 off -= offsetof(struct __sk_buff, data_end);
7290 off += offsetof(struct sk_buff, cb);
7291 off += offsetof(struct bpf_skb_data_end, data_end);
7292 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
7296 case offsetof(struct __sk_buff, tc_index):
7297 #ifdef CONFIG_NET_SCHED
7298 if (type == BPF_WRITE)
7299 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
7300 bpf_target_off(struct sk_buff, tc_index, 2,
7303 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
7304 bpf_target_off(struct sk_buff, tc_index, 2,
7308 if (type == BPF_WRITE)
7309 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
7311 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
7315 case offsetof(struct __sk_buff, napi_id):
7316 #if defined(CONFIG_NET_RX_BUSY_POLL)
7317 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7318 bpf_target_off(struct sk_buff, napi_id, 4,
7320 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
7321 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
7324 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
7327 case offsetof(struct __sk_buff, family):
7328 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
7330 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7331 si->dst_reg, si->src_reg,
7332 offsetof(struct sk_buff, sk));
7333 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7334 bpf_target_off(struct sock_common,
7338 case offsetof(struct __sk_buff, remote_ip4):
7339 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
7341 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7342 si->dst_reg, si->src_reg,
7343 offsetof(struct sk_buff, sk));
7344 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7345 bpf_target_off(struct sock_common,
7349 case offsetof(struct __sk_buff, local_ip4):
7350 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7351 skc_rcv_saddr) != 4);
7353 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7354 si->dst_reg, si->src_reg,
7355 offsetof(struct sk_buff, sk));
7356 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7357 bpf_target_off(struct sock_common,
7361 case offsetof(struct __sk_buff, remote_ip6[0]) ...
7362 offsetof(struct __sk_buff, remote_ip6[3]):
7363 #if IS_ENABLED(CONFIG_IPV6)
7364 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7365 skc_v6_daddr.s6_addr32[0]) != 4);
7368 off -= offsetof(struct __sk_buff, remote_ip6[0]);
7370 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7371 si->dst_reg, si->src_reg,
7372 offsetof(struct sk_buff, sk));
7373 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7374 offsetof(struct sock_common,
7375 skc_v6_daddr.s6_addr32[0]) +
7378 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7381 case offsetof(struct __sk_buff, local_ip6[0]) ...
7382 offsetof(struct __sk_buff, local_ip6[3]):
7383 #if IS_ENABLED(CONFIG_IPV6)
7384 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7385 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
7388 off -= offsetof(struct __sk_buff, local_ip6[0]);
7390 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7391 si->dst_reg, si->src_reg,
7392 offsetof(struct sk_buff, sk));
7393 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7394 offsetof(struct sock_common,
7395 skc_v6_rcv_saddr.s6_addr32[0]) +
7398 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7402 case offsetof(struct __sk_buff, remote_port):
7403 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
7405 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7406 si->dst_reg, si->src_reg,
7407 offsetof(struct sk_buff, sk));
7408 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7409 bpf_target_off(struct sock_common,
7412 #ifndef __BIG_ENDIAN_BITFIELD
7413 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
7417 case offsetof(struct __sk_buff, local_port):
7418 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
7420 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7421 si->dst_reg, si->src_reg,
7422 offsetof(struct sk_buff, sk));
7423 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7424 bpf_target_off(struct sock_common,
7425 skc_num, 2, target_size));
7428 case offsetof(struct __sk_buff, tstamp):
7429 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, tstamp) != 8);
7431 if (type == BPF_WRITE)
7432 *insn++ = BPF_STX_MEM(BPF_DW,
7433 si->dst_reg, si->src_reg,
7434 bpf_target_off(struct sk_buff,
7438 *insn++ = BPF_LDX_MEM(BPF_DW,
7439 si->dst_reg, si->src_reg,
7440 bpf_target_off(struct sk_buff,
7445 case offsetof(struct __sk_buff, gso_segs):
7446 /* si->dst_reg = skb_shinfo(SKB); */
7447 #ifdef NET_SKBUFF_DATA_USES_OFFSET
7448 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
7449 si->dst_reg, si->src_reg,
7450 offsetof(struct sk_buff, head));
7451 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
7452 BPF_REG_AX, si->src_reg,
7453 offsetof(struct sk_buff, end));
7454 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
7456 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
7457 si->dst_reg, si->src_reg,
7458 offsetof(struct sk_buff, end));
7460 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
7461 si->dst_reg, si->dst_reg,
7462 bpf_target_off(struct skb_shared_info,
7466 case offsetof(struct __sk_buff, wire_len):
7467 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, pkt_len) != 4);
7470 off -= offsetof(struct __sk_buff, wire_len);
7471 off += offsetof(struct sk_buff, cb);
7472 off += offsetof(struct qdisc_skb_cb, pkt_len);
7474 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
7477 case offsetof(struct __sk_buff, sk):
7478 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7479 si->dst_reg, si->src_reg,
7480 offsetof(struct sk_buff, sk));
7484 return insn - insn_buf;
7487 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
7488 const struct bpf_insn *si,
7489 struct bpf_insn *insn_buf,
7490 struct bpf_prog *prog, u32 *target_size)
7492 struct bpf_insn *insn = insn_buf;
7496 case offsetof(struct bpf_sock, bound_dev_if):
7497 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_bound_dev_if) != 4);
7499 if (type == BPF_WRITE)
7500 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7501 offsetof(struct sock, sk_bound_dev_if));
7503 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7504 offsetof(struct sock, sk_bound_dev_if));
7507 case offsetof(struct bpf_sock, mark):
7508 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_mark) != 4);
7510 if (type == BPF_WRITE)
7511 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7512 offsetof(struct sock, sk_mark));
7514 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7515 offsetof(struct sock, sk_mark));
7518 case offsetof(struct bpf_sock, priority):
7519 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_priority) != 4);
7521 if (type == BPF_WRITE)
7522 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7523 offsetof(struct sock, sk_priority));
7525 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7526 offsetof(struct sock, sk_priority));
7529 case offsetof(struct bpf_sock, family):
7530 *insn++ = BPF_LDX_MEM(
7531 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
7532 si->dst_reg, si->src_reg,
7533 bpf_target_off(struct sock_common,
7535 FIELD_SIZEOF(struct sock_common,
7540 case offsetof(struct bpf_sock, type):
7541 BUILD_BUG_ON(HWEIGHT32(SK_FL_TYPE_MASK) != BITS_PER_BYTE * 2);
7542 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7543 offsetof(struct sock, __sk_flags_offset));
7544 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
7545 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
7549 case offsetof(struct bpf_sock, protocol):
7550 BUILD_BUG_ON(HWEIGHT32(SK_FL_PROTO_MASK) != BITS_PER_BYTE);
7551 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7552 offsetof(struct sock, __sk_flags_offset));
7553 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
7554 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_PROTO_SHIFT);
7558 case offsetof(struct bpf_sock, src_ip4):
7559 *insn++ = BPF_LDX_MEM(
7560 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7561 bpf_target_off(struct sock_common, skc_rcv_saddr,
7562 FIELD_SIZEOF(struct sock_common,
7567 case offsetof(struct bpf_sock, dst_ip4):
7568 *insn++ = BPF_LDX_MEM(
7569 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7570 bpf_target_off(struct sock_common, skc_daddr,
7571 FIELD_SIZEOF(struct sock_common,
7576 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7577 #if IS_ENABLED(CONFIG_IPV6)
7579 off -= offsetof(struct bpf_sock, src_ip6[0]);
7580 *insn++ = BPF_LDX_MEM(
7581 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7584 skc_v6_rcv_saddr.s6_addr32[0],
7585 FIELD_SIZEOF(struct sock_common,
7586 skc_v6_rcv_saddr.s6_addr32[0]),
7587 target_size) + off);
7590 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7594 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
7595 #if IS_ENABLED(CONFIG_IPV6)
7597 off -= offsetof(struct bpf_sock, dst_ip6[0]);
7598 *insn++ = BPF_LDX_MEM(
7599 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7600 bpf_target_off(struct sock_common,
7601 skc_v6_daddr.s6_addr32[0],
7602 FIELD_SIZEOF(struct sock_common,
7603 skc_v6_daddr.s6_addr32[0]),
7604 target_size) + off);
7606 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7611 case offsetof(struct bpf_sock, src_port):
7612 *insn++ = BPF_LDX_MEM(
7613 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
7614 si->dst_reg, si->src_reg,
7615 bpf_target_off(struct sock_common, skc_num,
7616 FIELD_SIZEOF(struct sock_common,
7621 case offsetof(struct bpf_sock, dst_port):
7622 *insn++ = BPF_LDX_MEM(
7623 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
7624 si->dst_reg, si->src_reg,
7625 bpf_target_off(struct sock_common, skc_dport,
7626 FIELD_SIZEOF(struct sock_common,
7631 case offsetof(struct bpf_sock, state):
7632 *insn++ = BPF_LDX_MEM(
7633 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
7634 si->dst_reg, si->src_reg,
7635 bpf_target_off(struct sock_common, skc_state,
7636 FIELD_SIZEOF(struct sock_common,
7642 return insn - insn_buf;
7645 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
7646 const struct bpf_insn *si,
7647 struct bpf_insn *insn_buf,
7648 struct bpf_prog *prog, u32 *target_size)
7650 struct bpf_insn *insn = insn_buf;
7653 case offsetof(struct __sk_buff, ifindex):
7654 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
7655 si->dst_reg, si->src_reg,
7656 offsetof(struct sk_buff, dev));
7657 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7658 bpf_target_off(struct net_device, ifindex, 4,
7662 return bpf_convert_ctx_access(type, si, insn_buf, prog,
7666 return insn - insn_buf;
7669 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
7670 const struct bpf_insn *si,
7671 struct bpf_insn *insn_buf,
7672 struct bpf_prog *prog, u32 *target_size)
7674 struct bpf_insn *insn = insn_buf;
7677 case offsetof(struct xdp_md, data):
7678 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
7679 si->dst_reg, si->src_reg,
7680 offsetof(struct xdp_buff, data));
7682 case offsetof(struct xdp_md, data_meta):
7683 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
7684 si->dst_reg, si->src_reg,
7685 offsetof(struct xdp_buff, data_meta));
7687 case offsetof(struct xdp_md, data_end):
7688 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
7689 si->dst_reg, si->src_reg,
7690 offsetof(struct xdp_buff, data_end));
7692 case offsetof(struct xdp_md, ingress_ifindex):
7693 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
7694 si->dst_reg, si->src_reg,
7695 offsetof(struct xdp_buff, rxq));
7696 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
7697 si->dst_reg, si->dst_reg,
7698 offsetof(struct xdp_rxq_info, dev));
7699 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7700 offsetof(struct net_device, ifindex));
7702 case offsetof(struct xdp_md, rx_queue_index):
7703 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
7704 si->dst_reg, si->src_reg,
7705 offsetof(struct xdp_buff, rxq));
7706 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7707 offsetof(struct xdp_rxq_info,
7712 return insn - insn_buf;
7715 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
7716 * context Structure, F is Field in context structure that contains a pointer
7717 * to Nested Structure of type NS that has the field NF.
7719 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
7720 * sure that SIZE is not greater than actual size of S.F.NF.
7722 * If offset OFF is provided, the load happens from that offset relative to
7725 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
7727 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
7728 si->src_reg, offsetof(S, F)); \
7729 *insn++ = BPF_LDX_MEM( \
7730 SIZE, si->dst_reg, si->dst_reg, \
7731 bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF), \
7736 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
7737 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
7738 BPF_FIELD_SIZEOF(NS, NF), 0)
7740 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
7741 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
7743 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
7744 * "register" since two registers available in convert_ctx_access are not
7745 * enough: we can't override neither SRC, since it contains value to store, nor
7746 * DST since it contains pointer to context that may be used by later
7747 * instructions. But we need a temporary place to save pointer to nested
7748 * structure whose field we want to store to.
7750 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
7752 int tmp_reg = BPF_REG_9; \
7753 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
7755 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
7757 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
7759 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
7760 si->dst_reg, offsetof(S, F)); \
7761 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \
7762 bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF), \
7765 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
7769 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
7772 if (type == BPF_WRITE) { \
7773 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
7776 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
7777 S, NS, F, NF, SIZE, OFF); \
7781 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
7782 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
7783 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
7785 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
7786 const struct bpf_insn *si,
7787 struct bpf_insn *insn_buf,
7788 struct bpf_prog *prog, u32 *target_size)
7790 struct bpf_insn *insn = insn_buf;
7794 case offsetof(struct bpf_sock_addr, user_family):
7795 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
7796 struct sockaddr, uaddr, sa_family);
7799 case offsetof(struct bpf_sock_addr, user_ip4):
7800 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7801 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
7802 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
7805 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
7807 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
7808 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7809 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
7810 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
7814 case offsetof(struct bpf_sock_addr, user_port):
7815 /* To get port we need to know sa_family first and then treat
7816 * sockaddr as either sockaddr_in or sockaddr_in6.
7817 * Though we can simplify since port field has same offset and
7818 * size in both structures.
7819 * Here we check this invariant and use just one of the
7820 * structures if it's true.
7822 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
7823 offsetof(struct sockaddr_in6, sin6_port));
7824 BUILD_BUG_ON(FIELD_SIZEOF(struct sockaddr_in, sin_port) !=
7825 FIELD_SIZEOF(struct sockaddr_in6, sin6_port));
7826 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(struct bpf_sock_addr_kern,
7827 struct sockaddr_in6, uaddr,
7828 sin6_port, tmp_reg);
7831 case offsetof(struct bpf_sock_addr, family):
7832 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
7833 struct sock, sk, sk_family);
7836 case offsetof(struct bpf_sock_addr, type):
7837 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(
7838 struct bpf_sock_addr_kern, struct sock, sk,
7839 __sk_flags_offset, BPF_W, 0);
7840 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
7841 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
7844 case offsetof(struct bpf_sock_addr, protocol):
7845 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(
7846 struct bpf_sock_addr_kern, struct sock, sk,
7847 __sk_flags_offset, BPF_W, 0);
7848 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
7849 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg,
7853 case offsetof(struct bpf_sock_addr, msg_src_ip4):
7854 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
7855 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7856 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
7857 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
7860 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
7863 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
7864 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
7865 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7866 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
7867 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
7869 case offsetof(struct bpf_sock_addr, sk):
7870 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
7871 si->dst_reg, si->src_reg,
7872 offsetof(struct bpf_sock_addr_kern, sk));
7876 return insn - insn_buf;
7879 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
7880 const struct bpf_insn *si,
7881 struct bpf_insn *insn_buf,
7882 struct bpf_prog *prog,
7885 struct bpf_insn *insn = insn_buf;
7888 /* Helper macro for adding read access to tcp_sock or sock fields. */
7889 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
7891 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) > \
7892 FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD)); \
7893 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
7894 struct bpf_sock_ops_kern, \
7896 si->dst_reg, si->src_reg, \
7897 offsetof(struct bpf_sock_ops_kern, \
7899 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 2); \
7900 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
7901 struct bpf_sock_ops_kern, sk),\
7902 si->dst_reg, si->src_reg, \
7903 offsetof(struct bpf_sock_ops_kern, sk));\
7904 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
7906 si->dst_reg, si->dst_reg, \
7907 offsetof(OBJ, OBJ_FIELD)); \
7910 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
7911 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
7913 /* Helper macro for adding write access to tcp_sock or sock fields.
7914 * The macro is called with two registers, dst_reg which contains a pointer
7915 * to ctx (context) and src_reg which contains the value that should be
7916 * stored. However, we need an additional register since we cannot overwrite
7917 * dst_reg because it may be used later in the program.
7918 * Instead we "borrow" one of the other register. We first save its value
7919 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
7920 * it at the end of the macro.
7922 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
7924 int reg = BPF_REG_9; \
7925 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) > \
7926 FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD)); \
7927 if (si->dst_reg == reg || si->src_reg == reg) \
7929 if (si->dst_reg == reg || si->src_reg == reg) \
7931 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
7932 offsetof(struct bpf_sock_ops_kern, \
7934 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
7935 struct bpf_sock_ops_kern, \
7938 offsetof(struct bpf_sock_ops_kern, \
7940 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
7941 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
7942 struct bpf_sock_ops_kern, sk),\
7944 offsetof(struct bpf_sock_ops_kern, sk));\
7945 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \
7947 offsetof(OBJ, OBJ_FIELD)); \
7948 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
7949 offsetof(struct bpf_sock_ops_kern, \
7953 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
7955 if (TYPE == BPF_WRITE) \
7956 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
7958 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
7961 if (insn > insn_buf)
7962 return insn - insn_buf;
7965 case offsetof(struct bpf_sock_ops, op) ...
7966 offsetof(struct bpf_sock_ops, replylong[3]):
7967 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, op) !=
7968 FIELD_SIZEOF(struct bpf_sock_ops_kern, op));
7969 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, reply) !=
7970 FIELD_SIZEOF(struct bpf_sock_ops_kern, reply));
7971 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, replylong) !=
7972 FIELD_SIZEOF(struct bpf_sock_ops_kern, replylong));
7974 off -= offsetof(struct bpf_sock_ops, op);
7975 off += offsetof(struct bpf_sock_ops_kern, op);
7976 if (type == BPF_WRITE)
7977 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7980 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7984 case offsetof(struct bpf_sock_ops, family):
7985 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
7987 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7988 struct bpf_sock_ops_kern, sk),
7989 si->dst_reg, si->src_reg,
7990 offsetof(struct bpf_sock_ops_kern, sk));
7991 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7992 offsetof(struct sock_common, skc_family));
7995 case offsetof(struct bpf_sock_ops, remote_ip4):
7996 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
7998 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7999 struct bpf_sock_ops_kern, sk),
8000 si->dst_reg, si->src_reg,
8001 offsetof(struct bpf_sock_ops_kern, sk));
8002 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8003 offsetof(struct sock_common, skc_daddr));
8006 case offsetof(struct bpf_sock_ops, local_ip4):
8007 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
8008 skc_rcv_saddr) != 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,
8019 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
8020 offsetof(struct bpf_sock_ops, remote_ip6[3]):
8021 #if IS_ENABLED(CONFIG_IPV6)
8022 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
8023 skc_v6_daddr.s6_addr32[0]) != 4);
8026 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
8027 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8028 struct bpf_sock_ops_kern, sk),
8029 si->dst_reg, si->src_reg,
8030 offsetof(struct bpf_sock_ops_kern, sk));
8031 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8032 offsetof(struct sock_common,
8033 skc_v6_daddr.s6_addr32[0]) +
8036 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8040 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
8041 offsetof(struct bpf_sock_ops, local_ip6[3]):
8042 #if IS_ENABLED(CONFIG_IPV6)
8043 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
8044 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8047 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
8048 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8049 struct bpf_sock_ops_kern, sk),
8050 si->dst_reg, si->src_reg,
8051 offsetof(struct bpf_sock_ops_kern, sk));
8052 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8053 offsetof(struct sock_common,
8054 skc_v6_rcv_saddr.s6_addr32[0]) +
8057 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8061 case offsetof(struct bpf_sock_ops, remote_port):
8062 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
8064 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8065 struct bpf_sock_ops_kern, sk),
8066 si->dst_reg, si->src_reg,
8067 offsetof(struct bpf_sock_ops_kern, sk));
8068 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8069 offsetof(struct sock_common, skc_dport));
8070 #ifndef __BIG_ENDIAN_BITFIELD
8071 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8075 case offsetof(struct bpf_sock_ops, local_port):
8076 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
8078 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8079 struct bpf_sock_ops_kern, sk),
8080 si->dst_reg, si->src_reg,
8081 offsetof(struct bpf_sock_ops_kern, sk));
8082 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8083 offsetof(struct sock_common, skc_num));
8086 case offsetof(struct bpf_sock_ops, is_fullsock):
8087 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8088 struct bpf_sock_ops_kern,
8090 si->dst_reg, si->src_reg,
8091 offsetof(struct bpf_sock_ops_kern,
8095 case offsetof(struct bpf_sock_ops, state):
8096 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_state) != 1);
8098 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8099 struct bpf_sock_ops_kern, sk),
8100 si->dst_reg, si->src_reg,
8101 offsetof(struct bpf_sock_ops_kern, sk));
8102 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
8103 offsetof(struct sock_common, skc_state));
8106 case offsetof(struct bpf_sock_ops, rtt_min):
8107 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, rtt_min) !=
8108 sizeof(struct minmax));
8109 BUILD_BUG_ON(sizeof(struct minmax) <
8110 sizeof(struct minmax_sample));
8112 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8113 struct bpf_sock_ops_kern, sk),
8114 si->dst_reg, si->src_reg,
8115 offsetof(struct bpf_sock_ops_kern, sk));
8116 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8117 offsetof(struct tcp_sock, rtt_min) +
8118 FIELD_SIZEOF(struct minmax_sample, t));
8121 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
8122 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
8126 case offsetof(struct bpf_sock_ops, sk_txhash):
8127 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
8130 case offsetof(struct bpf_sock_ops, snd_cwnd):
8131 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
8133 case offsetof(struct bpf_sock_ops, srtt_us):
8134 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
8136 case offsetof(struct bpf_sock_ops, snd_ssthresh):
8137 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
8139 case offsetof(struct bpf_sock_ops, rcv_nxt):
8140 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
8142 case offsetof(struct bpf_sock_ops, snd_nxt):
8143 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
8145 case offsetof(struct bpf_sock_ops, snd_una):
8146 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
8148 case offsetof(struct bpf_sock_ops, mss_cache):
8149 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
8151 case offsetof(struct bpf_sock_ops, ecn_flags):
8152 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
8154 case offsetof(struct bpf_sock_ops, rate_delivered):
8155 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
8157 case offsetof(struct bpf_sock_ops, rate_interval_us):
8158 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
8160 case offsetof(struct bpf_sock_ops, packets_out):
8161 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
8163 case offsetof(struct bpf_sock_ops, retrans_out):
8164 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
8166 case offsetof(struct bpf_sock_ops, total_retrans):
8167 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
8169 case offsetof(struct bpf_sock_ops, segs_in):
8170 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
8172 case offsetof(struct bpf_sock_ops, data_segs_in):
8173 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
8175 case offsetof(struct bpf_sock_ops, segs_out):
8176 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
8178 case offsetof(struct bpf_sock_ops, data_segs_out):
8179 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
8181 case offsetof(struct bpf_sock_ops, lost_out):
8182 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
8184 case offsetof(struct bpf_sock_ops, sacked_out):
8185 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
8187 case offsetof(struct bpf_sock_ops, bytes_received):
8188 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
8190 case offsetof(struct bpf_sock_ops, bytes_acked):
8191 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
8193 case offsetof(struct bpf_sock_ops, sk):
8194 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8195 struct bpf_sock_ops_kern,
8197 si->dst_reg, si->src_reg,
8198 offsetof(struct bpf_sock_ops_kern,
8200 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
8201 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8202 struct bpf_sock_ops_kern, sk),
8203 si->dst_reg, si->src_reg,
8204 offsetof(struct bpf_sock_ops_kern, sk));
8207 return insn - insn_buf;
8210 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
8211 const struct bpf_insn *si,
8212 struct bpf_insn *insn_buf,
8213 struct bpf_prog *prog, u32 *target_size)
8215 struct bpf_insn *insn = insn_buf;
8219 case offsetof(struct __sk_buff, data_end):
8221 off -= offsetof(struct __sk_buff, data_end);
8222 off += offsetof(struct sk_buff, cb);
8223 off += offsetof(struct tcp_skb_cb, bpf.data_end);
8224 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8228 return bpf_convert_ctx_access(type, si, insn_buf, prog,
8232 return insn - insn_buf;
8235 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
8236 const struct bpf_insn *si,
8237 struct bpf_insn *insn_buf,
8238 struct bpf_prog *prog, u32 *target_size)
8240 struct bpf_insn *insn = insn_buf;
8241 #if IS_ENABLED(CONFIG_IPV6)
8245 /* convert ctx uses the fact sg element is first in struct */
8246 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
8249 case offsetof(struct sk_msg_md, data):
8250 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
8251 si->dst_reg, si->src_reg,
8252 offsetof(struct sk_msg, data));
8254 case offsetof(struct sk_msg_md, data_end):
8255 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
8256 si->dst_reg, si->src_reg,
8257 offsetof(struct sk_msg, data_end));
8259 case offsetof(struct sk_msg_md, family):
8260 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
8262 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8264 si->dst_reg, si->src_reg,
8265 offsetof(struct sk_msg, sk));
8266 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8267 offsetof(struct sock_common, skc_family));
8270 case offsetof(struct sk_msg_md, remote_ip4):
8271 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
8273 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8275 si->dst_reg, si->src_reg,
8276 offsetof(struct sk_msg, sk));
8277 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8278 offsetof(struct sock_common, skc_daddr));
8281 case offsetof(struct sk_msg_md, local_ip4):
8282 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
8283 skc_rcv_saddr) != 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,
8294 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
8295 offsetof(struct sk_msg_md, remote_ip6[3]):
8296 #if IS_ENABLED(CONFIG_IPV6)
8297 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
8298 skc_v6_daddr.s6_addr32[0]) != 4);
8301 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
8302 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8304 si->dst_reg, si->src_reg,
8305 offsetof(struct sk_msg, sk));
8306 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8307 offsetof(struct sock_common,
8308 skc_v6_daddr.s6_addr32[0]) +
8311 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8315 case offsetof(struct sk_msg_md, local_ip6[0]) ...
8316 offsetof(struct sk_msg_md, local_ip6[3]):
8317 #if IS_ENABLED(CONFIG_IPV6)
8318 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
8319 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8322 off -= offsetof(struct sk_msg_md, local_ip6[0]);
8323 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8325 si->dst_reg, si->src_reg,
8326 offsetof(struct sk_msg, sk));
8327 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8328 offsetof(struct sock_common,
8329 skc_v6_rcv_saddr.s6_addr32[0]) +
8332 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8336 case offsetof(struct sk_msg_md, remote_port):
8337 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
8339 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8341 si->dst_reg, si->src_reg,
8342 offsetof(struct sk_msg, sk));
8343 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8344 offsetof(struct sock_common, skc_dport));
8345 #ifndef __BIG_ENDIAN_BITFIELD
8346 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8350 case offsetof(struct sk_msg_md, local_port):
8351 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
8353 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8355 si->dst_reg, si->src_reg,
8356 offsetof(struct sk_msg, sk));
8357 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8358 offsetof(struct sock_common, skc_num));
8361 case offsetof(struct sk_msg_md, size):
8362 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
8363 si->dst_reg, si->src_reg,
8364 offsetof(struct sk_msg_sg, size));
8368 return insn - insn_buf;
8371 const struct bpf_verifier_ops sk_filter_verifier_ops = {
8372 .get_func_proto = sk_filter_func_proto,
8373 .is_valid_access = sk_filter_is_valid_access,
8374 .convert_ctx_access = bpf_convert_ctx_access,
8375 .gen_ld_abs = bpf_gen_ld_abs,
8378 const struct bpf_prog_ops sk_filter_prog_ops = {
8379 .test_run = bpf_prog_test_run_skb,
8382 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
8383 .get_func_proto = tc_cls_act_func_proto,
8384 .is_valid_access = tc_cls_act_is_valid_access,
8385 .convert_ctx_access = tc_cls_act_convert_ctx_access,
8386 .gen_prologue = tc_cls_act_prologue,
8387 .gen_ld_abs = bpf_gen_ld_abs,
8390 const struct bpf_prog_ops tc_cls_act_prog_ops = {
8391 .test_run = bpf_prog_test_run_skb,
8394 const struct bpf_verifier_ops xdp_verifier_ops = {
8395 .get_func_proto = xdp_func_proto,
8396 .is_valid_access = xdp_is_valid_access,
8397 .convert_ctx_access = xdp_convert_ctx_access,
8398 .gen_prologue = bpf_noop_prologue,
8401 const struct bpf_prog_ops xdp_prog_ops = {
8402 .test_run = bpf_prog_test_run_xdp,
8405 const struct bpf_verifier_ops cg_skb_verifier_ops = {
8406 .get_func_proto = cg_skb_func_proto,
8407 .is_valid_access = cg_skb_is_valid_access,
8408 .convert_ctx_access = bpf_convert_ctx_access,
8411 const struct bpf_prog_ops cg_skb_prog_ops = {
8412 .test_run = bpf_prog_test_run_skb,
8415 const struct bpf_verifier_ops lwt_in_verifier_ops = {
8416 .get_func_proto = lwt_in_func_proto,
8417 .is_valid_access = lwt_is_valid_access,
8418 .convert_ctx_access = bpf_convert_ctx_access,
8421 const struct bpf_prog_ops lwt_in_prog_ops = {
8422 .test_run = bpf_prog_test_run_skb,
8425 const struct bpf_verifier_ops lwt_out_verifier_ops = {
8426 .get_func_proto = lwt_out_func_proto,
8427 .is_valid_access = lwt_is_valid_access,
8428 .convert_ctx_access = bpf_convert_ctx_access,
8431 const struct bpf_prog_ops lwt_out_prog_ops = {
8432 .test_run = bpf_prog_test_run_skb,
8435 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
8436 .get_func_proto = lwt_xmit_func_proto,
8437 .is_valid_access = lwt_is_valid_access,
8438 .convert_ctx_access = bpf_convert_ctx_access,
8439 .gen_prologue = tc_cls_act_prologue,
8442 const struct bpf_prog_ops lwt_xmit_prog_ops = {
8443 .test_run = bpf_prog_test_run_skb,
8446 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
8447 .get_func_proto = lwt_seg6local_func_proto,
8448 .is_valid_access = lwt_is_valid_access,
8449 .convert_ctx_access = bpf_convert_ctx_access,
8452 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
8453 .test_run = bpf_prog_test_run_skb,
8456 const struct bpf_verifier_ops cg_sock_verifier_ops = {
8457 .get_func_proto = sock_filter_func_proto,
8458 .is_valid_access = sock_filter_is_valid_access,
8459 .convert_ctx_access = bpf_sock_convert_ctx_access,
8462 const struct bpf_prog_ops cg_sock_prog_ops = {
8465 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
8466 .get_func_proto = sock_addr_func_proto,
8467 .is_valid_access = sock_addr_is_valid_access,
8468 .convert_ctx_access = sock_addr_convert_ctx_access,
8471 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
8474 const struct bpf_verifier_ops sock_ops_verifier_ops = {
8475 .get_func_proto = sock_ops_func_proto,
8476 .is_valid_access = sock_ops_is_valid_access,
8477 .convert_ctx_access = sock_ops_convert_ctx_access,
8480 const struct bpf_prog_ops sock_ops_prog_ops = {
8483 const struct bpf_verifier_ops sk_skb_verifier_ops = {
8484 .get_func_proto = sk_skb_func_proto,
8485 .is_valid_access = sk_skb_is_valid_access,
8486 .convert_ctx_access = sk_skb_convert_ctx_access,
8487 .gen_prologue = sk_skb_prologue,
8490 const struct bpf_prog_ops sk_skb_prog_ops = {
8493 const struct bpf_verifier_ops sk_msg_verifier_ops = {
8494 .get_func_proto = sk_msg_func_proto,
8495 .is_valid_access = sk_msg_is_valid_access,
8496 .convert_ctx_access = sk_msg_convert_ctx_access,
8497 .gen_prologue = bpf_noop_prologue,
8500 const struct bpf_prog_ops sk_msg_prog_ops = {
8503 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
8504 .get_func_proto = flow_dissector_func_proto,
8505 .is_valid_access = flow_dissector_is_valid_access,
8506 .convert_ctx_access = flow_dissector_convert_ctx_access,
8509 const struct bpf_prog_ops flow_dissector_prog_ops = {
8510 .test_run = bpf_prog_test_run_flow_dissector,
8513 int sk_detach_filter(struct sock *sk)
8516 struct sk_filter *filter;
8518 if (sock_flag(sk, SOCK_FILTER_LOCKED))
8521 filter = rcu_dereference_protected(sk->sk_filter,
8522 lockdep_sock_is_held(sk));
8524 RCU_INIT_POINTER(sk->sk_filter, NULL);
8525 sk_filter_uncharge(sk, filter);
8531 EXPORT_SYMBOL_GPL(sk_detach_filter);
8533 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
8536 struct sock_fprog_kern *fprog;
8537 struct sk_filter *filter;
8541 filter = rcu_dereference_protected(sk->sk_filter,
8542 lockdep_sock_is_held(sk));
8546 /* We're copying the filter that has been originally attached,
8547 * so no conversion/decode needed anymore. eBPF programs that
8548 * have no original program cannot be dumped through this.
8551 fprog = filter->prog->orig_prog;
8557 /* User space only enquires number of filter blocks. */
8561 if (len < fprog->len)
8565 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
8568 /* Instead of bytes, the API requests to return the number
8578 struct sk_reuseport_kern {
8579 struct sk_buff *skb;
8581 struct sock *selected_sk;
8588 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
8589 struct sock_reuseport *reuse,
8590 struct sock *sk, struct sk_buff *skb,
8593 reuse_kern->skb = skb;
8594 reuse_kern->sk = sk;
8595 reuse_kern->selected_sk = NULL;
8596 reuse_kern->data_end = skb->data + skb_headlen(skb);
8597 reuse_kern->hash = hash;
8598 reuse_kern->reuseport_id = reuse->reuseport_id;
8599 reuse_kern->bind_inany = reuse->bind_inany;
8602 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
8603 struct bpf_prog *prog, struct sk_buff *skb,
8606 struct sk_reuseport_kern reuse_kern;
8607 enum sk_action action;
8609 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, hash);
8610 action = BPF_PROG_RUN(prog, &reuse_kern);
8612 if (action == SK_PASS)
8613 return reuse_kern.selected_sk;
8615 return ERR_PTR(-ECONNREFUSED);
8618 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
8619 struct bpf_map *, map, void *, key, u32, flags)
8621 struct sock_reuseport *reuse;
8622 struct sock *selected_sk;
8624 selected_sk = map->ops->map_lookup_elem(map, key);
8628 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
8630 /* selected_sk is unhashed (e.g. by close()) after the
8631 * above map_lookup_elem(). Treat selected_sk has already
8632 * been removed from the map.
8636 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
8639 if (unlikely(!reuse_kern->reuseport_id))
8640 /* There is a small race between adding the
8641 * sk to the map and setting the
8642 * reuse_kern->reuseport_id.
8643 * Treat it as the sk has not been added to
8648 sk = reuse_kern->sk;
8649 if (sk->sk_protocol != selected_sk->sk_protocol)
8651 else if (sk->sk_family != selected_sk->sk_family)
8652 return -EAFNOSUPPORT;
8654 /* Catch all. Likely bound to a different sockaddr. */
8658 reuse_kern->selected_sk = selected_sk;
8663 static const struct bpf_func_proto sk_select_reuseport_proto = {
8664 .func = sk_select_reuseport,
8666 .ret_type = RET_INTEGER,
8667 .arg1_type = ARG_PTR_TO_CTX,
8668 .arg2_type = ARG_CONST_MAP_PTR,
8669 .arg3_type = ARG_PTR_TO_MAP_KEY,
8670 .arg4_type = ARG_ANYTHING,
8673 BPF_CALL_4(sk_reuseport_load_bytes,
8674 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
8675 void *, to, u32, len)
8677 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
8680 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
8681 .func = sk_reuseport_load_bytes,
8683 .ret_type = RET_INTEGER,
8684 .arg1_type = ARG_PTR_TO_CTX,
8685 .arg2_type = ARG_ANYTHING,
8686 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
8687 .arg4_type = ARG_CONST_SIZE,
8690 BPF_CALL_5(sk_reuseport_load_bytes_relative,
8691 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
8692 void *, to, u32, len, u32, start_header)
8694 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
8698 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
8699 .func = sk_reuseport_load_bytes_relative,
8701 .ret_type = RET_INTEGER,
8702 .arg1_type = ARG_PTR_TO_CTX,
8703 .arg2_type = ARG_ANYTHING,
8704 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
8705 .arg4_type = ARG_CONST_SIZE,
8706 .arg5_type = ARG_ANYTHING,
8709 static const struct bpf_func_proto *
8710 sk_reuseport_func_proto(enum bpf_func_id func_id,
8711 const struct bpf_prog *prog)
8714 case BPF_FUNC_sk_select_reuseport:
8715 return &sk_select_reuseport_proto;
8716 case BPF_FUNC_skb_load_bytes:
8717 return &sk_reuseport_load_bytes_proto;
8718 case BPF_FUNC_skb_load_bytes_relative:
8719 return &sk_reuseport_load_bytes_relative_proto;
8721 return bpf_base_func_proto(func_id);
8726 sk_reuseport_is_valid_access(int off, int size,
8727 enum bpf_access_type type,
8728 const struct bpf_prog *prog,
8729 struct bpf_insn_access_aux *info)
8731 const u32 size_default = sizeof(__u32);
8733 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
8734 off % size || type != BPF_READ)
8738 case offsetof(struct sk_reuseport_md, data):
8739 info->reg_type = PTR_TO_PACKET;
8740 return size == sizeof(__u64);
8742 case offsetof(struct sk_reuseport_md, data_end):
8743 info->reg_type = PTR_TO_PACKET_END;
8744 return size == sizeof(__u64);
8746 case offsetof(struct sk_reuseport_md, hash):
8747 return size == size_default;
8749 /* Fields that allow narrowing */
8750 case offsetof(struct sk_reuseport_md, eth_protocol):
8751 if (size < FIELD_SIZEOF(struct sk_buff, protocol))
8754 case offsetof(struct sk_reuseport_md, ip_protocol):
8755 case offsetof(struct sk_reuseport_md, bind_inany):
8756 case offsetof(struct sk_reuseport_md, len):
8757 bpf_ctx_record_field_size(info, size_default);
8758 return bpf_ctx_narrow_access_ok(off, size, size_default);
8765 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
8766 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
8767 si->dst_reg, si->src_reg, \
8768 bpf_target_off(struct sk_reuseport_kern, F, \
8769 FIELD_SIZEOF(struct sk_reuseport_kern, F), \
8773 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
8774 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
8779 #define SK_REUSEPORT_LOAD_SK_FIELD_SIZE_OFF(SK_FIELD, BPF_SIZE, EXTRA_OFF) \
8780 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(struct sk_reuseport_kern, \
8783 SK_FIELD, BPF_SIZE, EXTRA_OFF)
8785 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
8786 const struct bpf_insn *si,
8787 struct bpf_insn *insn_buf,
8788 struct bpf_prog *prog,
8791 struct bpf_insn *insn = insn_buf;
8794 case offsetof(struct sk_reuseport_md, data):
8795 SK_REUSEPORT_LOAD_SKB_FIELD(data);
8798 case offsetof(struct sk_reuseport_md, len):
8799 SK_REUSEPORT_LOAD_SKB_FIELD(len);
8802 case offsetof(struct sk_reuseport_md, eth_protocol):
8803 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
8806 case offsetof(struct sk_reuseport_md, ip_protocol):
8807 BUILD_BUG_ON(HWEIGHT32(SK_FL_PROTO_MASK) != BITS_PER_BYTE);
8808 SK_REUSEPORT_LOAD_SK_FIELD_SIZE_OFF(__sk_flags_offset,
8810 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
8811 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg,
8813 /* SK_FL_PROTO_MASK and SK_FL_PROTO_SHIFT are endian
8814 * aware. No further narrowing or masking is needed.
8819 case offsetof(struct sk_reuseport_md, data_end):
8820 SK_REUSEPORT_LOAD_FIELD(data_end);
8823 case offsetof(struct sk_reuseport_md, hash):
8824 SK_REUSEPORT_LOAD_FIELD(hash);
8827 case offsetof(struct sk_reuseport_md, bind_inany):
8828 SK_REUSEPORT_LOAD_FIELD(bind_inany);
8832 return insn - insn_buf;
8835 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
8836 .get_func_proto = sk_reuseport_func_proto,
8837 .is_valid_access = sk_reuseport_is_valid_access,
8838 .convert_ctx_access = sk_reuseport_convert_ctx_access,
8841 const struct bpf_prog_ops sk_reuseport_prog_ops = {
8843 #endif /* CONFIG_INET */