2 * Linux Socket Filter - Kernel level socket filtering
4 * Based on the design of the Berkeley Packet Filter. The new
5 * internal format has been designed by PLUMgrid:
7 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
11 * Jay Schulist <jschlst@samba.org>
12 * Alexei Starovoitov <ast@plumgrid.com>
13 * Daniel Borkmann <dborkman@redhat.com>
15 * This program is free software; you can redistribute it and/or
16 * modify it under the terms of the GNU General Public License
17 * as published by the Free Software Foundation; either version
18 * 2 of the License, or (at your option) any later version.
20 * Andi Kleen - Fix a few bad bugs and races.
21 * Kris Katterjohn - Added many additional checks in bpf_check_classic()
24 #include <linux/module.h>
25 #include <linux/types.h>
27 #include <linux/fcntl.h>
28 #include <linux/socket.h>
29 #include <linux/sock_diag.h>
31 #include <linux/inet.h>
32 #include <linux/netdevice.h>
33 #include <linux/if_packet.h>
34 #include <linux/if_arp.h>
35 #include <linux/gfp.h>
36 #include <net/inet_common.h>
38 #include <net/protocol.h>
39 #include <net/netlink.h>
40 #include <linux/skbuff.h>
41 #include <linux/skmsg.h>
43 #include <net/flow_dissector.h>
44 #include <linux/errno.h>
45 #include <linux/timer.h>
46 #include <linux/uaccess.h>
47 #include <asm/unaligned.h>
48 #include <asm/cmpxchg.h>
49 #include <linux/filter.h>
50 #include <linux/ratelimit.h>
51 #include <linux/seccomp.h>
52 #include <linux/if_vlan.h>
53 #include <linux/bpf.h>
54 #include <net/sch_generic.h>
55 #include <net/cls_cgroup.h>
56 #include <net/dst_metadata.h>
58 #include <net/sock_reuseport.h>
59 #include <net/busy_poll.h>
63 #include <linux/bpf_trace.h>
64 #include <net/xdp_sock.h>
65 #include <linux/inetdevice.h>
66 #include <net/inet_hashtables.h>
67 #include <net/inet6_hashtables.h>
68 #include <net/ip_fib.h>
72 #include <net/net_namespace.h>
73 #include <linux/seg6_local.h>
75 #include <net/seg6_local.h>
76 #include <net/lwtunnel.h>
79 * sk_filter_trim_cap - run a packet through a socket filter
80 * @sk: sock associated with &sk_buff
81 * @skb: buffer to filter
82 * @cap: limit on how short the eBPF program may trim the packet
84 * Run the eBPF program and then cut skb->data to correct size returned by
85 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
86 * than pkt_len we keep whole skb->data. This is the socket level
87 * wrapper to BPF_PROG_RUN. It returns 0 if the packet should
88 * be accepted or -EPERM if the packet should be tossed.
91 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
94 struct sk_filter *filter;
97 * If the skb was allocated from pfmemalloc reserves, only
98 * allow SOCK_MEMALLOC sockets to use it as this socket is
101 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
102 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
105 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
109 err = security_sock_rcv_skb(sk, skb);
114 filter = rcu_dereference(sk->sk_filter);
116 struct sock *save_sk = skb->sk;
117 unsigned int pkt_len;
120 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
122 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
128 EXPORT_SYMBOL(sk_filter_trim_cap);
130 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
132 return skb_get_poff(skb);
135 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
139 if (skb_is_nonlinear(skb))
142 if (skb->len < sizeof(struct nlattr))
145 if (a > skb->len - sizeof(struct nlattr))
148 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
150 return (void *) nla - (void *) skb->data;
155 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
159 if (skb_is_nonlinear(skb))
162 if (skb->len < sizeof(struct nlattr))
165 if (a > skb->len - sizeof(struct nlattr))
168 nla = (struct nlattr *) &skb->data[a];
169 if (nla->nla_len > skb->len - a)
172 nla = nla_find_nested(nla, x);
174 return (void *) nla - (void *) skb->data;
179 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
180 data, int, headlen, int, offset)
183 const int len = sizeof(tmp);
186 if (headlen - offset >= len)
187 return *(u8 *)(data + offset);
188 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
191 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
199 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
202 return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
206 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
207 data, int, headlen, int, offset)
210 const int len = sizeof(tmp);
213 if (headlen - offset >= len)
214 return get_unaligned_be16(data + offset);
215 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
216 return be16_to_cpu(tmp);
218 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
220 return get_unaligned_be16(ptr);
226 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
229 return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
233 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
234 data, int, headlen, int, offset)
237 const int len = sizeof(tmp);
239 if (likely(offset >= 0)) {
240 if (headlen - offset >= len)
241 return get_unaligned_be32(data + offset);
242 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
243 return be32_to_cpu(tmp);
245 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
247 return get_unaligned_be32(ptr);
253 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
256 return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
260 BPF_CALL_0(bpf_get_raw_cpu_id)
262 return raw_smp_processor_id();
265 static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = {
266 .func = bpf_get_raw_cpu_id,
268 .ret_type = RET_INTEGER,
271 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
272 struct bpf_insn *insn_buf)
274 struct bpf_insn *insn = insn_buf;
278 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
280 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
281 offsetof(struct sk_buff, mark));
285 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
286 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
287 #ifdef __BIG_ENDIAN_BITFIELD
288 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
293 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2);
295 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
296 offsetof(struct sk_buff, queue_mapping));
299 case SKF_AD_VLAN_TAG:
300 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2);
302 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
303 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
304 offsetof(struct sk_buff, vlan_tci));
306 case SKF_AD_VLAN_TAG_PRESENT:
307 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_VLAN_PRESENT_OFFSET());
308 if (PKT_VLAN_PRESENT_BIT)
309 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, PKT_VLAN_PRESENT_BIT);
310 if (PKT_VLAN_PRESENT_BIT < 7)
311 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
315 return insn - insn_buf;
318 static bool convert_bpf_extensions(struct sock_filter *fp,
319 struct bpf_insn **insnp)
321 struct bpf_insn *insn = *insnp;
325 case SKF_AD_OFF + SKF_AD_PROTOCOL:
326 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
328 /* A = *(u16 *) (CTX + offsetof(protocol)) */
329 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
330 offsetof(struct sk_buff, protocol));
331 /* A = ntohs(A) [emitting a nop or swap16] */
332 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
335 case SKF_AD_OFF + SKF_AD_PKTTYPE:
336 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
340 case SKF_AD_OFF + SKF_AD_IFINDEX:
341 case SKF_AD_OFF + SKF_AD_HATYPE:
342 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
343 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, type) != 2);
345 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
346 BPF_REG_TMP, BPF_REG_CTX,
347 offsetof(struct sk_buff, dev));
348 /* if (tmp != 0) goto pc + 1 */
349 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
350 *insn++ = BPF_EXIT_INSN();
351 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
352 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
353 offsetof(struct net_device, ifindex));
355 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
356 offsetof(struct net_device, type));
359 case SKF_AD_OFF + SKF_AD_MARK:
360 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
364 case SKF_AD_OFF + SKF_AD_RXHASH:
365 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
367 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
368 offsetof(struct sk_buff, hash));
371 case SKF_AD_OFF + SKF_AD_QUEUE:
372 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
376 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
377 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
378 BPF_REG_A, BPF_REG_CTX, insn);
382 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
383 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
384 BPF_REG_A, BPF_REG_CTX, insn);
388 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
389 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2);
391 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
392 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
393 offsetof(struct sk_buff, vlan_proto));
394 /* A = ntohs(A) [emitting a nop or swap16] */
395 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
398 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
399 case SKF_AD_OFF + SKF_AD_NLATTR:
400 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
401 case SKF_AD_OFF + SKF_AD_CPU:
402 case SKF_AD_OFF + SKF_AD_RANDOM:
404 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
406 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
408 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
409 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
411 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
412 *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
414 case SKF_AD_OFF + SKF_AD_NLATTR:
415 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
417 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
418 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
420 case SKF_AD_OFF + SKF_AD_CPU:
421 *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
423 case SKF_AD_OFF + SKF_AD_RANDOM:
424 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
425 bpf_user_rnd_init_once();
430 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
432 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
436 /* This is just a dummy call to avoid letting the compiler
437 * evict __bpf_call_base() as an optimization. Placed here
438 * where no-one bothers.
440 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
448 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
450 const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
451 int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
452 bool endian = BPF_SIZE(fp->code) == BPF_H ||
453 BPF_SIZE(fp->code) == BPF_W;
454 bool indirect = BPF_MODE(fp->code) == BPF_IND;
455 const int ip_align = NET_IP_ALIGN;
456 struct bpf_insn *insn = *insnp;
460 ((unaligned_ok && offset >= 0) ||
461 (!unaligned_ok && offset >= 0 &&
462 offset + ip_align >= 0 &&
463 offset + ip_align % size == 0))) {
464 bool ldx_off_ok = offset <= S16_MAX;
466 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
468 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
469 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
470 size, 2 + endian + (!ldx_off_ok * 2));
472 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
475 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
476 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
477 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
481 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
482 *insn++ = BPF_JMP_A(8);
485 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
486 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
487 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
489 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
491 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
493 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
496 switch (BPF_SIZE(fp->code)) {
498 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
501 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
504 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
510 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
511 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
512 *insn = BPF_EXIT_INSN();
519 * bpf_convert_filter - convert filter program
520 * @prog: the user passed filter program
521 * @len: the length of the user passed filter program
522 * @new_prog: allocated 'struct bpf_prog' or NULL
523 * @new_len: pointer to store length of converted program
524 * @seen_ld_abs: bool whether we've seen ld_abs/ind
526 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
527 * style extended BPF (eBPF).
528 * Conversion workflow:
530 * 1) First pass for calculating the new program length:
531 * bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
533 * 2) 2nd pass to remap in two passes: 1st pass finds new
534 * jump offsets, 2nd pass remapping:
535 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
537 static int bpf_convert_filter(struct sock_filter *prog, int len,
538 struct bpf_prog *new_prog, int *new_len,
541 int new_flen = 0, pass = 0, target, i, stack_off;
542 struct bpf_insn *new_insn, *first_insn = NULL;
543 struct sock_filter *fp;
547 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
548 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
550 if (len <= 0 || len > BPF_MAXINSNS)
554 first_insn = new_prog->insnsi;
555 addrs = kcalloc(len, sizeof(*addrs),
556 GFP_KERNEL | __GFP_NOWARN);
562 new_insn = first_insn;
565 /* Classic BPF related prologue emission. */
567 /* Classic BPF expects A and X to be reset first. These need
568 * to be guaranteed to be the first two instructions.
570 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
571 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
573 /* All programs must keep CTX in callee saved BPF_REG_CTX.
574 * In eBPF case it's done by the compiler, here we need to
575 * do this ourself. Initial CTX is present in BPF_REG_ARG1.
577 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
579 /* For packet access in classic BPF, cache skb->data
580 * in callee-saved BPF R8 and skb->len - skb->data_len
581 * (headlen) in BPF R9. Since classic BPF is read-only
582 * on CTX, we only need to cache it once.
584 *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
585 BPF_REG_D, BPF_REG_CTX,
586 offsetof(struct sk_buff, data));
587 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
588 offsetof(struct sk_buff, len));
589 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
590 offsetof(struct sk_buff, data_len));
591 *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
597 for (i = 0; i < len; fp++, i++) {
598 struct bpf_insn tmp_insns[32] = { };
599 struct bpf_insn *insn = tmp_insns;
602 addrs[i] = new_insn - first_insn;
605 /* All arithmetic insns and skb loads map as-is. */
606 case BPF_ALU | BPF_ADD | BPF_X:
607 case BPF_ALU | BPF_ADD | BPF_K:
608 case BPF_ALU | BPF_SUB | BPF_X:
609 case BPF_ALU | BPF_SUB | BPF_K:
610 case BPF_ALU | BPF_AND | BPF_X:
611 case BPF_ALU | BPF_AND | BPF_K:
612 case BPF_ALU | BPF_OR | BPF_X:
613 case BPF_ALU | BPF_OR | BPF_K:
614 case BPF_ALU | BPF_LSH | BPF_X:
615 case BPF_ALU | BPF_LSH | BPF_K:
616 case BPF_ALU | BPF_RSH | BPF_X:
617 case BPF_ALU | BPF_RSH | BPF_K:
618 case BPF_ALU | BPF_XOR | BPF_X:
619 case BPF_ALU | BPF_XOR | BPF_K:
620 case BPF_ALU | BPF_MUL | BPF_X:
621 case BPF_ALU | BPF_MUL | BPF_K:
622 case BPF_ALU | BPF_DIV | BPF_X:
623 case BPF_ALU | BPF_DIV | BPF_K:
624 case BPF_ALU | BPF_MOD | BPF_X:
625 case BPF_ALU | BPF_MOD | BPF_K:
626 case BPF_ALU | BPF_NEG:
627 case BPF_LD | BPF_ABS | BPF_W:
628 case BPF_LD | BPF_ABS | BPF_H:
629 case BPF_LD | BPF_ABS | BPF_B:
630 case BPF_LD | BPF_IND | BPF_W:
631 case BPF_LD | BPF_IND | BPF_H:
632 case BPF_LD | BPF_IND | BPF_B:
633 /* Check for overloaded BPF extension and
634 * directly convert it if found, otherwise
635 * just move on with mapping.
637 if (BPF_CLASS(fp->code) == BPF_LD &&
638 BPF_MODE(fp->code) == BPF_ABS &&
639 convert_bpf_extensions(fp, &insn))
641 if (BPF_CLASS(fp->code) == BPF_LD &&
642 convert_bpf_ld_abs(fp, &insn)) {
647 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
648 fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
649 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
650 /* Error with exception code on div/mod by 0.
651 * For cBPF programs, this was always return 0.
653 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
654 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
655 *insn++ = BPF_EXIT_INSN();
658 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
661 /* Jump transformation cannot use BPF block macros
662 * everywhere as offset calculation and target updates
663 * require a bit more work than the rest, i.e. jump
664 * opcodes map as-is, but offsets need adjustment.
667 #define BPF_EMIT_JMP \
669 const s32 off_min = S16_MIN, off_max = S16_MAX; \
672 if (target >= len || target < 0) \
674 off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
675 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
676 off -= insn - tmp_insns; \
677 /* Reject anything not fitting into insn->off. */ \
678 if (off < off_min || off > off_max) \
683 case BPF_JMP | BPF_JA:
684 target = i + fp->k + 1;
685 insn->code = fp->code;
689 case BPF_JMP | BPF_JEQ | BPF_K:
690 case BPF_JMP | BPF_JEQ | BPF_X:
691 case BPF_JMP | BPF_JSET | BPF_K:
692 case BPF_JMP | BPF_JSET | BPF_X:
693 case BPF_JMP | BPF_JGT | BPF_K:
694 case BPF_JMP | BPF_JGT | BPF_X:
695 case BPF_JMP | BPF_JGE | BPF_K:
696 case BPF_JMP | BPF_JGE | BPF_X:
697 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
698 /* BPF immediates are signed, zero extend
699 * immediate into tmp register and use it
702 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
704 insn->dst_reg = BPF_REG_A;
705 insn->src_reg = BPF_REG_TMP;
708 insn->dst_reg = BPF_REG_A;
710 bpf_src = BPF_SRC(fp->code);
711 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
714 /* Common case where 'jump_false' is next insn. */
716 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
717 target = i + fp->jt + 1;
722 /* Convert some jumps when 'jump_true' is next insn. */
724 switch (BPF_OP(fp->code)) {
726 insn->code = BPF_JMP | BPF_JNE | bpf_src;
729 insn->code = BPF_JMP | BPF_JLE | bpf_src;
732 insn->code = BPF_JMP | BPF_JLT | bpf_src;
738 target = i + fp->jf + 1;
743 /* Other jumps are mapped into two insns: Jxx and JA. */
744 target = i + fp->jt + 1;
745 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
749 insn->code = BPF_JMP | BPF_JA;
750 target = i + fp->jf + 1;
754 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
755 case BPF_LDX | BPF_MSH | BPF_B: {
756 struct sock_filter tmp = {
757 .code = BPF_LD | BPF_ABS | BPF_B,
764 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
765 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
766 convert_bpf_ld_abs(&tmp, &insn);
769 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
771 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
773 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
775 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
777 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
780 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
781 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
783 case BPF_RET | BPF_A:
784 case BPF_RET | BPF_K:
785 if (BPF_RVAL(fp->code) == BPF_K)
786 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
788 *insn = BPF_EXIT_INSN();
791 /* Store to stack. */
794 stack_off = fp->k * 4 + 4;
795 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
796 BPF_ST ? BPF_REG_A : BPF_REG_X,
798 /* check_load_and_stores() verifies that classic BPF can
799 * load from stack only after write, so tracking
800 * stack_depth for ST|STX insns is enough
802 if (new_prog && new_prog->aux->stack_depth < stack_off)
803 new_prog->aux->stack_depth = stack_off;
806 /* Load from stack. */
807 case BPF_LD | BPF_MEM:
808 case BPF_LDX | BPF_MEM:
809 stack_off = fp->k * 4 + 4;
810 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
811 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
816 case BPF_LD | BPF_IMM:
817 case BPF_LDX | BPF_IMM:
818 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
819 BPF_REG_A : BPF_REG_X, fp->k);
823 case BPF_MISC | BPF_TAX:
824 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
828 case BPF_MISC | BPF_TXA:
829 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
832 /* A = skb->len or X = skb->len */
833 case BPF_LD | BPF_W | BPF_LEN:
834 case BPF_LDX | BPF_W | BPF_LEN:
835 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
836 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
837 offsetof(struct sk_buff, len));
840 /* Access seccomp_data fields. */
841 case BPF_LDX | BPF_ABS | BPF_W:
842 /* A = *(u32 *) (ctx + K) */
843 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
846 /* Unknown instruction. */
853 memcpy(new_insn, tmp_insns,
854 sizeof(*insn) * (insn - tmp_insns));
855 new_insn += insn - tmp_insns;
859 /* Only calculating new length. */
860 *new_len = new_insn - first_insn;
862 *new_len += 4; /* Prologue bits. */
867 if (new_flen != new_insn - first_insn) {
868 new_flen = new_insn - first_insn;
875 BUG_ON(*new_len != new_flen);
884 * As we dont want to clear mem[] array for each packet going through
885 * __bpf_prog_run(), we check that filter loaded by user never try to read
886 * a cell if not previously written, and we check all branches to be sure
887 * a malicious user doesn't try to abuse us.
889 static int check_load_and_stores(const struct sock_filter *filter, int flen)
891 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
894 BUILD_BUG_ON(BPF_MEMWORDS > 16);
896 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
900 memset(masks, 0xff, flen * sizeof(*masks));
902 for (pc = 0; pc < flen; pc++) {
903 memvalid &= masks[pc];
905 switch (filter[pc].code) {
908 memvalid |= (1 << filter[pc].k);
910 case BPF_LD | BPF_MEM:
911 case BPF_LDX | BPF_MEM:
912 if (!(memvalid & (1 << filter[pc].k))) {
917 case BPF_JMP | BPF_JA:
918 /* A jump must set masks on target */
919 masks[pc + 1 + filter[pc].k] &= memvalid;
922 case BPF_JMP | BPF_JEQ | BPF_K:
923 case BPF_JMP | BPF_JEQ | BPF_X:
924 case BPF_JMP | BPF_JGE | BPF_K:
925 case BPF_JMP | BPF_JGE | BPF_X:
926 case BPF_JMP | BPF_JGT | BPF_K:
927 case BPF_JMP | BPF_JGT | BPF_X:
928 case BPF_JMP | BPF_JSET | BPF_K:
929 case BPF_JMP | BPF_JSET | BPF_X:
930 /* A jump must set masks on targets */
931 masks[pc + 1 + filter[pc].jt] &= memvalid;
932 masks[pc + 1 + filter[pc].jf] &= memvalid;
942 static bool chk_code_allowed(u16 code_to_probe)
944 static const bool codes[] = {
945 /* 32 bit ALU operations */
946 [BPF_ALU | BPF_ADD | BPF_K] = true,
947 [BPF_ALU | BPF_ADD | BPF_X] = true,
948 [BPF_ALU | BPF_SUB | BPF_K] = true,
949 [BPF_ALU | BPF_SUB | BPF_X] = true,
950 [BPF_ALU | BPF_MUL | BPF_K] = true,
951 [BPF_ALU | BPF_MUL | BPF_X] = true,
952 [BPF_ALU | BPF_DIV | BPF_K] = true,
953 [BPF_ALU | BPF_DIV | BPF_X] = true,
954 [BPF_ALU | BPF_MOD | BPF_K] = true,
955 [BPF_ALU | BPF_MOD | BPF_X] = true,
956 [BPF_ALU | BPF_AND | BPF_K] = true,
957 [BPF_ALU | BPF_AND | BPF_X] = true,
958 [BPF_ALU | BPF_OR | BPF_K] = true,
959 [BPF_ALU | BPF_OR | BPF_X] = true,
960 [BPF_ALU | BPF_XOR | BPF_K] = true,
961 [BPF_ALU | BPF_XOR | BPF_X] = true,
962 [BPF_ALU | BPF_LSH | BPF_K] = true,
963 [BPF_ALU | BPF_LSH | BPF_X] = true,
964 [BPF_ALU | BPF_RSH | BPF_K] = true,
965 [BPF_ALU | BPF_RSH | BPF_X] = true,
966 [BPF_ALU | BPF_NEG] = true,
967 /* Load instructions */
968 [BPF_LD | BPF_W | BPF_ABS] = true,
969 [BPF_LD | BPF_H | BPF_ABS] = true,
970 [BPF_LD | BPF_B | BPF_ABS] = true,
971 [BPF_LD | BPF_W | BPF_LEN] = true,
972 [BPF_LD | BPF_W | BPF_IND] = true,
973 [BPF_LD | BPF_H | BPF_IND] = true,
974 [BPF_LD | BPF_B | BPF_IND] = true,
975 [BPF_LD | BPF_IMM] = true,
976 [BPF_LD | BPF_MEM] = true,
977 [BPF_LDX | BPF_W | BPF_LEN] = true,
978 [BPF_LDX | BPF_B | BPF_MSH] = true,
979 [BPF_LDX | BPF_IMM] = true,
980 [BPF_LDX | BPF_MEM] = true,
981 /* Store instructions */
984 /* Misc instructions */
985 [BPF_MISC | BPF_TAX] = true,
986 [BPF_MISC | BPF_TXA] = true,
987 /* Return instructions */
988 [BPF_RET | BPF_K] = true,
989 [BPF_RET | BPF_A] = true,
990 /* Jump instructions */
991 [BPF_JMP | BPF_JA] = true,
992 [BPF_JMP | BPF_JEQ | BPF_K] = true,
993 [BPF_JMP | BPF_JEQ | BPF_X] = true,
994 [BPF_JMP | BPF_JGE | BPF_K] = true,
995 [BPF_JMP | BPF_JGE | BPF_X] = true,
996 [BPF_JMP | BPF_JGT | BPF_K] = true,
997 [BPF_JMP | BPF_JGT | BPF_X] = true,
998 [BPF_JMP | BPF_JSET | BPF_K] = true,
999 [BPF_JMP | BPF_JSET | BPF_X] = true,
1002 if (code_to_probe >= ARRAY_SIZE(codes))
1005 return codes[code_to_probe];
1008 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1013 if (flen == 0 || flen > BPF_MAXINSNS)
1020 * bpf_check_classic - verify socket filter code
1021 * @filter: filter to verify
1022 * @flen: length of filter
1024 * Check the user's filter code. If we let some ugly
1025 * filter code slip through kaboom! The filter must contain
1026 * no references or jumps that are out of range, no illegal
1027 * instructions, and must end with a RET instruction.
1029 * All jumps are forward as they are not signed.
1031 * Returns 0 if the rule set is legal or -EINVAL if not.
1033 static int bpf_check_classic(const struct sock_filter *filter,
1039 /* Check the filter code now */
1040 for (pc = 0; pc < flen; pc++) {
1041 const struct sock_filter *ftest = &filter[pc];
1043 /* May we actually operate on this code? */
1044 if (!chk_code_allowed(ftest->code))
1047 /* Some instructions need special checks */
1048 switch (ftest->code) {
1049 case BPF_ALU | BPF_DIV | BPF_K:
1050 case BPF_ALU | BPF_MOD | BPF_K:
1051 /* Check for division by zero */
1055 case BPF_ALU | BPF_LSH | BPF_K:
1056 case BPF_ALU | BPF_RSH | BPF_K:
1060 case BPF_LD | BPF_MEM:
1061 case BPF_LDX | BPF_MEM:
1064 /* Check for invalid memory addresses */
1065 if (ftest->k >= BPF_MEMWORDS)
1068 case BPF_JMP | BPF_JA:
1069 /* Note, the large ftest->k might cause loops.
1070 * Compare this with conditional jumps below,
1071 * where offsets are limited. --ANK (981016)
1073 if (ftest->k >= (unsigned int)(flen - pc - 1))
1076 case BPF_JMP | BPF_JEQ | BPF_K:
1077 case BPF_JMP | BPF_JEQ | BPF_X:
1078 case BPF_JMP | BPF_JGE | BPF_K:
1079 case BPF_JMP | BPF_JGE | BPF_X:
1080 case BPF_JMP | BPF_JGT | BPF_K:
1081 case BPF_JMP | BPF_JGT | BPF_X:
1082 case BPF_JMP | BPF_JSET | BPF_K:
1083 case BPF_JMP | BPF_JSET | BPF_X:
1084 /* Both conditionals must be safe */
1085 if (pc + ftest->jt + 1 >= flen ||
1086 pc + ftest->jf + 1 >= flen)
1089 case BPF_LD | BPF_W | BPF_ABS:
1090 case BPF_LD | BPF_H | BPF_ABS:
1091 case BPF_LD | BPF_B | BPF_ABS:
1093 if (bpf_anc_helper(ftest) & BPF_ANC)
1095 /* Ancillary operation unknown or unsupported */
1096 if (anc_found == false && ftest->k >= SKF_AD_OFF)
1101 /* Last instruction must be a RET code */
1102 switch (filter[flen - 1].code) {
1103 case BPF_RET | BPF_K:
1104 case BPF_RET | BPF_A:
1105 return check_load_and_stores(filter, flen);
1111 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1112 const struct sock_fprog *fprog)
1114 unsigned int fsize = bpf_classic_proglen(fprog);
1115 struct sock_fprog_kern *fkprog;
1117 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1121 fkprog = fp->orig_prog;
1122 fkprog->len = fprog->len;
1124 fkprog->filter = kmemdup(fp->insns, fsize,
1125 GFP_KERNEL | __GFP_NOWARN);
1126 if (!fkprog->filter) {
1127 kfree(fp->orig_prog);
1134 static void bpf_release_orig_filter(struct bpf_prog *fp)
1136 struct sock_fprog_kern *fprog = fp->orig_prog;
1139 kfree(fprog->filter);
1144 static void __bpf_prog_release(struct bpf_prog *prog)
1146 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1149 bpf_release_orig_filter(prog);
1150 bpf_prog_free(prog);
1154 static void __sk_filter_release(struct sk_filter *fp)
1156 __bpf_prog_release(fp->prog);
1161 * sk_filter_release_rcu - Release a socket filter by rcu_head
1162 * @rcu: rcu_head that contains the sk_filter to free
1164 static void sk_filter_release_rcu(struct rcu_head *rcu)
1166 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1168 __sk_filter_release(fp);
1172 * sk_filter_release - release a socket filter
1173 * @fp: filter to remove
1175 * Remove a filter from a socket and release its resources.
1177 static void sk_filter_release(struct sk_filter *fp)
1179 if (refcount_dec_and_test(&fp->refcnt))
1180 call_rcu(&fp->rcu, sk_filter_release_rcu);
1183 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1185 u32 filter_size = bpf_prog_size(fp->prog->len);
1187 atomic_sub(filter_size, &sk->sk_omem_alloc);
1188 sk_filter_release(fp);
1191 /* try to charge the socket memory if there is space available
1192 * return true on success
1194 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1196 u32 filter_size = bpf_prog_size(fp->prog->len);
1198 /* same check as in sock_kmalloc() */
1199 if (filter_size <= sysctl_optmem_max &&
1200 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
1201 atomic_add(filter_size, &sk->sk_omem_alloc);
1207 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1209 if (!refcount_inc_not_zero(&fp->refcnt))
1212 if (!__sk_filter_charge(sk, fp)) {
1213 sk_filter_release(fp);
1219 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1221 struct sock_filter *old_prog;
1222 struct bpf_prog *old_fp;
1223 int err, new_len, old_len = fp->len;
1224 bool seen_ld_abs = false;
1226 /* We are free to overwrite insns et al right here as it
1227 * won't be used at this point in time anymore internally
1228 * after the migration to the internal BPF instruction
1231 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1232 sizeof(struct bpf_insn));
1234 /* Conversion cannot happen on overlapping memory areas,
1235 * so we need to keep the user BPF around until the 2nd
1236 * pass. At this time, the user BPF is stored in fp->insns.
1238 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1239 GFP_KERNEL | __GFP_NOWARN);
1245 /* 1st pass: calculate the new program length. */
1246 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1251 /* Expand fp for appending the new filter representation. */
1253 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1255 /* The old_fp is still around in case we couldn't
1256 * allocate new memory, so uncharge on that one.
1265 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1266 err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1269 /* 2nd bpf_convert_filter() can fail only if it fails
1270 * to allocate memory, remapping must succeed. Note,
1271 * that at this time old_fp has already been released
1276 fp = bpf_prog_select_runtime(fp, &err);
1286 __bpf_prog_release(fp);
1287 return ERR_PTR(err);
1290 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1291 bpf_aux_classic_check_t trans)
1295 fp->bpf_func = NULL;
1298 err = bpf_check_classic(fp->insns, fp->len);
1300 __bpf_prog_release(fp);
1301 return ERR_PTR(err);
1304 /* There might be additional checks and transformations
1305 * needed on classic filters, f.e. in case of seccomp.
1308 err = trans(fp->insns, fp->len);
1310 __bpf_prog_release(fp);
1311 return ERR_PTR(err);
1315 /* Probe if we can JIT compile the filter and if so, do
1316 * the compilation of the filter.
1318 bpf_jit_compile(fp);
1320 /* JIT compiler couldn't process this filter, so do the
1321 * internal BPF translation for the optimized interpreter.
1324 fp = bpf_migrate_filter(fp);
1330 * bpf_prog_create - create an unattached filter
1331 * @pfp: the unattached filter that is created
1332 * @fprog: the filter program
1334 * Create a filter independent of any socket. We first run some
1335 * sanity checks on it to make sure it does not explode on us later.
1336 * If an error occurs or there is insufficient memory for the filter
1337 * a negative errno code is returned. On success the return is zero.
1339 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1341 unsigned int fsize = bpf_classic_proglen(fprog);
1342 struct bpf_prog *fp;
1344 /* Make sure new filter is there and in the right amounts. */
1345 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1348 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1352 memcpy(fp->insns, fprog->filter, fsize);
1354 fp->len = fprog->len;
1355 /* Since unattached filters are not copied back to user
1356 * space through sk_get_filter(), we do not need to hold
1357 * a copy here, and can spare us the work.
1359 fp->orig_prog = NULL;
1361 /* bpf_prepare_filter() already takes care of freeing
1362 * memory in case something goes wrong.
1364 fp = bpf_prepare_filter(fp, NULL);
1371 EXPORT_SYMBOL_GPL(bpf_prog_create);
1374 * bpf_prog_create_from_user - create an unattached filter from user buffer
1375 * @pfp: the unattached filter that is created
1376 * @fprog: the filter program
1377 * @trans: post-classic verifier transformation handler
1378 * @save_orig: save classic BPF program
1380 * This function effectively does the same as bpf_prog_create(), only
1381 * that it builds up its insns buffer from user space provided buffer.
1382 * It also allows for passing a bpf_aux_classic_check_t handler.
1384 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1385 bpf_aux_classic_check_t trans, bool save_orig)
1387 unsigned int fsize = bpf_classic_proglen(fprog);
1388 struct bpf_prog *fp;
1391 /* Make sure new filter is there and in the right amounts. */
1392 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1395 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1399 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1400 __bpf_prog_free(fp);
1404 fp->len = fprog->len;
1405 fp->orig_prog = NULL;
1408 err = bpf_prog_store_orig_filter(fp, fprog);
1410 __bpf_prog_free(fp);
1415 /* bpf_prepare_filter() already takes care of freeing
1416 * memory in case something goes wrong.
1418 fp = bpf_prepare_filter(fp, trans);
1425 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1427 void bpf_prog_destroy(struct bpf_prog *fp)
1429 __bpf_prog_release(fp);
1431 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1433 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1435 struct sk_filter *fp, *old_fp;
1437 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1443 if (!__sk_filter_charge(sk, fp)) {
1447 refcount_set(&fp->refcnt, 1);
1449 old_fp = rcu_dereference_protected(sk->sk_filter,
1450 lockdep_sock_is_held(sk));
1451 rcu_assign_pointer(sk->sk_filter, fp);
1454 sk_filter_uncharge(sk, old_fp);
1460 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1462 unsigned int fsize = bpf_classic_proglen(fprog);
1463 struct bpf_prog *prog;
1466 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1467 return ERR_PTR(-EPERM);
1469 /* Make sure new filter is there and in the right amounts. */
1470 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1471 return ERR_PTR(-EINVAL);
1473 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1475 return ERR_PTR(-ENOMEM);
1477 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1478 __bpf_prog_free(prog);
1479 return ERR_PTR(-EFAULT);
1482 prog->len = fprog->len;
1484 err = bpf_prog_store_orig_filter(prog, fprog);
1486 __bpf_prog_free(prog);
1487 return ERR_PTR(-ENOMEM);
1490 /* bpf_prepare_filter() already takes care of freeing
1491 * memory in case something goes wrong.
1493 return bpf_prepare_filter(prog, NULL);
1497 * sk_attach_filter - attach a socket filter
1498 * @fprog: the filter program
1499 * @sk: the socket to use
1501 * Attach the user's filter code. We first run some sanity checks on
1502 * it to make sure it does not explode on us later. If an error
1503 * occurs or there is insufficient memory for the filter a negative
1504 * errno code is returned. On success the return is zero.
1506 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1508 struct bpf_prog *prog = __get_filter(fprog, sk);
1512 return PTR_ERR(prog);
1514 err = __sk_attach_prog(prog, sk);
1516 __bpf_prog_release(prog);
1522 EXPORT_SYMBOL_GPL(sk_attach_filter);
1524 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1526 struct bpf_prog *prog = __get_filter(fprog, sk);
1530 return PTR_ERR(prog);
1532 if (bpf_prog_size(prog->len) > sysctl_optmem_max)
1535 err = reuseport_attach_prog(sk, prog);
1538 __bpf_prog_release(prog);
1543 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1545 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1546 return ERR_PTR(-EPERM);
1548 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1551 int sk_attach_bpf(u32 ufd, struct sock *sk)
1553 struct bpf_prog *prog = __get_bpf(ufd, sk);
1557 return PTR_ERR(prog);
1559 err = __sk_attach_prog(prog, sk);
1568 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1570 struct bpf_prog *prog;
1573 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1576 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1577 if (IS_ERR(prog) && PTR_ERR(prog) == -EINVAL)
1578 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1580 return PTR_ERR(prog);
1582 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1583 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1584 * bpf prog (e.g. sockmap). It depends on the
1585 * limitation imposed by bpf_prog_load().
1586 * Hence, sysctl_optmem_max is not checked.
1588 if ((sk->sk_type != SOCK_STREAM &&
1589 sk->sk_type != SOCK_DGRAM) ||
1590 (sk->sk_protocol != IPPROTO_UDP &&
1591 sk->sk_protocol != IPPROTO_TCP) ||
1592 (sk->sk_family != AF_INET &&
1593 sk->sk_family != AF_INET6)) {
1598 /* BPF_PROG_TYPE_SOCKET_FILTER */
1599 if (bpf_prog_size(prog->len) > sysctl_optmem_max) {
1605 err = reuseport_attach_prog(sk, prog);
1613 void sk_reuseport_prog_free(struct bpf_prog *prog)
1618 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1621 bpf_prog_destroy(prog);
1624 struct bpf_scratchpad {
1626 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1627 u8 buff[MAX_BPF_STACK];
1631 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1633 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1634 unsigned int write_len)
1636 return skb_ensure_writable(skb, write_len);
1639 static inline int bpf_try_make_writable(struct sk_buff *skb,
1640 unsigned int write_len)
1642 int err = __bpf_try_make_writable(skb, write_len);
1644 bpf_compute_data_pointers(skb);
1648 static int bpf_try_make_head_writable(struct sk_buff *skb)
1650 return bpf_try_make_writable(skb, skb_headlen(skb));
1653 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1655 if (skb_at_tc_ingress(skb))
1656 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1659 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1661 if (skb_at_tc_ingress(skb))
1662 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1665 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1666 const void *, from, u32, len, u64, flags)
1670 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1672 if (unlikely(offset > 0xffff))
1674 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1677 ptr = skb->data + offset;
1678 if (flags & BPF_F_RECOMPUTE_CSUM)
1679 __skb_postpull_rcsum(skb, ptr, len, offset);
1681 memcpy(ptr, from, len);
1683 if (flags & BPF_F_RECOMPUTE_CSUM)
1684 __skb_postpush_rcsum(skb, ptr, len, offset);
1685 if (flags & BPF_F_INVALIDATE_HASH)
1686 skb_clear_hash(skb);
1691 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1692 .func = bpf_skb_store_bytes,
1694 .ret_type = RET_INTEGER,
1695 .arg1_type = ARG_PTR_TO_CTX,
1696 .arg2_type = ARG_ANYTHING,
1697 .arg3_type = ARG_PTR_TO_MEM,
1698 .arg4_type = ARG_CONST_SIZE,
1699 .arg5_type = ARG_ANYTHING,
1702 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1703 void *, to, u32, len)
1707 if (unlikely(offset > 0xffff))
1710 ptr = skb_header_pointer(skb, offset, len, to);
1714 memcpy(to, ptr, len);
1722 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1723 .func = bpf_skb_load_bytes,
1725 .ret_type = RET_INTEGER,
1726 .arg1_type = ARG_PTR_TO_CTX,
1727 .arg2_type = ARG_ANYTHING,
1728 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1729 .arg4_type = ARG_CONST_SIZE,
1732 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1733 u32, offset, void *, to, u32, len, u32, start_header)
1735 u8 *end = skb_tail_pointer(skb);
1736 u8 *net = skb_network_header(skb);
1737 u8 *mac = skb_mac_header(skb);
1740 if (unlikely(offset > 0xffff || len > (end - mac)))
1743 switch (start_header) {
1744 case BPF_HDR_START_MAC:
1747 case BPF_HDR_START_NET:
1754 if (likely(ptr >= mac && ptr + len <= end)) {
1755 memcpy(to, ptr, len);
1764 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1765 .func = bpf_skb_load_bytes_relative,
1767 .ret_type = RET_INTEGER,
1768 .arg1_type = ARG_PTR_TO_CTX,
1769 .arg2_type = ARG_ANYTHING,
1770 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1771 .arg4_type = ARG_CONST_SIZE,
1772 .arg5_type = ARG_ANYTHING,
1775 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1777 /* Idea is the following: should the needed direct read/write
1778 * test fail during runtime, we can pull in more data and redo
1779 * again, since implicitly, we invalidate previous checks here.
1781 * Or, since we know how much we need to make read/writeable,
1782 * this can be done once at the program beginning for direct
1783 * access case. By this we overcome limitations of only current
1784 * headroom being accessible.
1786 return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1789 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1790 .func = bpf_skb_pull_data,
1792 .ret_type = RET_INTEGER,
1793 .arg1_type = ARG_PTR_TO_CTX,
1794 .arg2_type = ARG_ANYTHING,
1797 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1799 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1802 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1803 .func = bpf_sk_fullsock,
1805 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
1806 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
1809 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1810 unsigned int write_len)
1812 int err = __bpf_try_make_writable(skb, write_len);
1814 bpf_compute_data_end_sk_skb(skb);
1818 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1820 /* Idea is the following: should the needed direct read/write
1821 * test fail during runtime, we can pull in more data and redo
1822 * again, since implicitly, we invalidate previous checks here.
1824 * Or, since we know how much we need to make read/writeable,
1825 * this can be done once at the program beginning for direct
1826 * access case. By this we overcome limitations of only current
1827 * headroom being accessible.
1829 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1832 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1833 .func = sk_skb_pull_data,
1835 .ret_type = RET_INTEGER,
1836 .arg1_type = ARG_PTR_TO_CTX,
1837 .arg2_type = ARG_ANYTHING,
1840 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1841 u64, from, u64, to, u64, flags)
1845 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1847 if (unlikely(offset > 0xffff || offset & 1))
1849 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1852 ptr = (__sum16 *)(skb->data + offset);
1853 switch (flags & BPF_F_HDR_FIELD_MASK) {
1855 if (unlikely(from != 0))
1858 csum_replace_by_diff(ptr, to);
1861 csum_replace2(ptr, from, to);
1864 csum_replace4(ptr, from, to);
1873 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1874 .func = bpf_l3_csum_replace,
1876 .ret_type = RET_INTEGER,
1877 .arg1_type = ARG_PTR_TO_CTX,
1878 .arg2_type = ARG_ANYTHING,
1879 .arg3_type = ARG_ANYTHING,
1880 .arg4_type = ARG_ANYTHING,
1881 .arg5_type = ARG_ANYTHING,
1884 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1885 u64, from, u64, to, u64, flags)
1887 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1888 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1889 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1892 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1893 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1895 if (unlikely(offset > 0xffff || offset & 1))
1897 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1900 ptr = (__sum16 *)(skb->data + offset);
1901 if (is_mmzero && !do_mforce && !*ptr)
1904 switch (flags & BPF_F_HDR_FIELD_MASK) {
1906 if (unlikely(from != 0))
1909 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1912 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1915 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1921 if (is_mmzero && !*ptr)
1922 *ptr = CSUM_MANGLED_0;
1926 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1927 .func = bpf_l4_csum_replace,
1929 .ret_type = RET_INTEGER,
1930 .arg1_type = ARG_PTR_TO_CTX,
1931 .arg2_type = ARG_ANYTHING,
1932 .arg3_type = ARG_ANYTHING,
1933 .arg4_type = ARG_ANYTHING,
1934 .arg5_type = ARG_ANYTHING,
1937 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1938 __be32 *, to, u32, to_size, __wsum, seed)
1940 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1941 u32 diff_size = from_size + to_size;
1944 /* This is quite flexible, some examples:
1946 * from_size == 0, to_size > 0, seed := csum --> pushing data
1947 * from_size > 0, to_size == 0, seed := csum --> pulling data
1948 * from_size > 0, to_size > 0, seed := 0 --> diffing data
1950 * Even for diffing, from_size and to_size don't need to be equal.
1952 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
1953 diff_size > sizeof(sp->diff)))
1956 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
1957 sp->diff[j] = ~from[i];
1958 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
1959 sp->diff[j] = to[i];
1961 return csum_partial(sp->diff, diff_size, seed);
1964 static const struct bpf_func_proto bpf_csum_diff_proto = {
1965 .func = bpf_csum_diff,
1968 .ret_type = RET_INTEGER,
1969 .arg1_type = ARG_PTR_TO_MEM_OR_NULL,
1970 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
1971 .arg3_type = ARG_PTR_TO_MEM_OR_NULL,
1972 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
1973 .arg5_type = ARG_ANYTHING,
1976 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
1978 /* The interface is to be used in combination with bpf_csum_diff()
1979 * for direct packet writes. csum rotation for alignment as well
1980 * as emulating csum_sub() can be done from the eBPF program.
1982 if (skb->ip_summed == CHECKSUM_COMPLETE)
1983 return (skb->csum = csum_add(skb->csum, csum));
1988 static const struct bpf_func_proto bpf_csum_update_proto = {
1989 .func = bpf_csum_update,
1991 .ret_type = RET_INTEGER,
1992 .arg1_type = ARG_PTR_TO_CTX,
1993 .arg2_type = ARG_ANYTHING,
1996 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
1998 return dev_forward_skb(dev, skb);
2001 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2002 struct sk_buff *skb)
2004 int ret = ____dev_forward_skb(dev, skb);
2008 ret = netif_rx(skb);
2014 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2018 if (unlikely(__this_cpu_read(xmit_recursion) > XMIT_RECURSION_LIMIT)) {
2019 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2026 __this_cpu_inc(xmit_recursion);
2027 ret = dev_queue_xmit(skb);
2028 __this_cpu_dec(xmit_recursion);
2033 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2036 unsigned int mlen = skb_network_offset(skb);
2039 __skb_pull(skb, mlen);
2041 /* At ingress, the mac header has already been pulled once.
2042 * At egress, skb_pospull_rcsum has to be done in case that
2043 * the skb is originated from ingress (i.e. a forwarded skb)
2044 * to ensure that rcsum starts at net header.
2046 if (!skb_at_tc_ingress(skb))
2047 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2049 skb_pop_mac_header(skb);
2050 skb_reset_mac_len(skb);
2051 return flags & BPF_F_INGRESS ?
2052 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2055 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2058 /* Verify that a link layer header is carried */
2059 if (unlikely(skb->mac_header >= skb->network_header)) {
2064 bpf_push_mac_rcsum(skb);
2065 return flags & BPF_F_INGRESS ?
2066 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2069 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2072 if (dev_is_mac_header_xmit(dev))
2073 return __bpf_redirect_common(skb, dev, flags);
2075 return __bpf_redirect_no_mac(skb, dev, flags);
2078 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2080 struct net_device *dev;
2081 struct sk_buff *clone;
2084 if (unlikely(flags & ~(BPF_F_INGRESS)))
2087 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2091 clone = skb_clone(skb, GFP_ATOMIC);
2092 if (unlikely(!clone))
2095 /* For direct write, we need to keep the invariant that the skbs
2096 * we're dealing with need to be uncloned. Should uncloning fail
2097 * here, we need to free the just generated clone to unclone once
2100 ret = bpf_try_make_head_writable(skb);
2101 if (unlikely(ret)) {
2106 return __bpf_redirect(clone, dev, flags);
2109 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2110 .func = bpf_clone_redirect,
2112 .ret_type = RET_INTEGER,
2113 .arg1_type = ARG_PTR_TO_CTX,
2114 .arg2_type = ARG_ANYTHING,
2115 .arg3_type = ARG_ANYTHING,
2118 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2119 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2121 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2123 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2125 if (unlikely(flags & ~(BPF_F_INGRESS)))
2128 ri->ifindex = ifindex;
2131 return TC_ACT_REDIRECT;
2134 int skb_do_redirect(struct sk_buff *skb)
2136 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2137 struct net_device *dev;
2139 dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->ifindex);
2141 if (unlikely(!dev)) {
2146 return __bpf_redirect(skb, dev, ri->flags);
2149 static const struct bpf_func_proto bpf_redirect_proto = {
2150 .func = bpf_redirect,
2152 .ret_type = RET_INTEGER,
2153 .arg1_type = ARG_ANYTHING,
2154 .arg2_type = ARG_ANYTHING,
2157 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2159 msg->apply_bytes = bytes;
2163 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2164 .func = bpf_msg_apply_bytes,
2166 .ret_type = RET_INTEGER,
2167 .arg1_type = ARG_PTR_TO_CTX,
2168 .arg2_type = ARG_ANYTHING,
2171 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2173 msg->cork_bytes = bytes;
2177 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2178 .func = bpf_msg_cork_bytes,
2180 .ret_type = RET_INTEGER,
2181 .arg1_type = ARG_PTR_TO_CTX,
2182 .arg2_type = ARG_ANYTHING,
2185 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2186 u32, end, u64, flags)
2188 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2189 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2190 struct scatterlist *sge;
2191 u8 *raw, *to, *from;
2194 if (unlikely(flags || end <= start))
2197 /* First find the starting scatterlist element */
2200 len = sk_msg_elem(msg, i)->length;
2201 if (start < offset + len)
2204 sk_msg_iter_var_next(i);
2205 } while (i != msg->sg.end);
2207 if (unlikely(start >= offset + len))
2211 /* The start may point into the sg element so we need to also
2212 * account for the headroom.
2214 bytes_sg_total = start - offset + bytes;
2215 if (!msg->sg.copy[i] && bytes_sg_total <= len)
2218 /* At this point we need to linearize multiple scatterlist
2219 * elements or a single shared page. Either way we need to
2220 * copy into a linear buffer exclusively owned by BPF. Then
2221 * place the buffer in the scatterlist and fixup the original
2222 * entries by removing the entries now in the linear buffer
2223 * and shifting the remaining entries. For now we do not try
2224 * to copy partial entries to avoid complexity of running out
2225 * of sg_entry slots. The downside is reading a single byte
2226 * will copy the entire sg entry.
2229 copy += sk_msg_elem(msg, i)->length;
2230 sk_msg_iter_var_next(i);
2231 if (bytes_sg_total <= copy)
2233 } while (i != msg->sg.end);
2236 if (unlikely(bytes_sg_total > copy))
2239 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2241 if (unlikely(!page))
2244 raw = page_address(page);
2247 sge = sk_msg_elem(msg, i);
2248 from = sg_virt(sge);
2252 memcpy(to, from, len);
2255 put_page(sg_page(sge));
2257 sk_msg_iter_var_next(i);
2258 } while (i != last_sge);
2260 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2262 /* To repair sg ring we need to shift entries. If we only
2263 * had a single entry though we can just replace it and
2264 * be done. Otherwise walk the ring and shift the entries.
2266 WARN_ON_ONCE(last_sge == first_sge);
2267 shift = last_sge > first_sge ?
2268 last_sge - first_sge - 1 :
2269 MAX_SKB_FRAGS - first_sge + last_sge - 1;
2274 sk_msg_iter_var_next(i);
2278 if (i + shift >= MAX_MSG_FRAGS)
2279 move_from = i + shift - MAX_MSG_FRAGS;
2281 move_from = i + shift;
2282 if (move_from == msg->sg.end)
2285 msg->sg.data[i] = msg->sg.data[move_from];
2286 msg->sg.data[move_from].length = 0;
2287 msg->sg.data[move_from].page_link = 0;
2288 msg->sg.data[move_from].offset = 0;
2289 sk_msg_iter_var_next(i);
2292 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2293 msg->sg.end - shift + MAX_MSG_FRAGS :
2294 msg->sg.end - shift;
2296 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2297 msg->data_end = msg->data + bytes;
2301 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2302 .func = bpf_msg_pull_data,
2304 .ret_type = RET_INTEGER,
2305 .arg1_type = ARG_PTR_TO_CTX,
2306 .arg2_type = ARG_ANYTHING,
2307 .arg3_type = ARG_ANYTHING,
2308 .arg4_type = ARG_ANYTHING,
2311 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2312 u32, len, u64, flags)
2314 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2315 u32 new, i = 0, l, space, copy = 0, offset = 0;
2316 u8 *raw, *to, *from;
2319 if (unlikely(flags))
2322 /* First find the starting scatterlist element */
2325 l = sk_msg_elem(msg, i)->length;
2327 if (start < offset + l)
2330 sk_msg_iter_var_next(i);
2331 } while (i != msg->sg.end);
2333 if (start >= offset + l)
2336 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2338 /* If no space available will fallback to copy, we need at
2339 * least one scatterlist elem available to push data into
2340 * when start aligns to the beginning of an element or two
2341 * when it falls inside an element. We handle the start equals
2342 * offset case because its the common case for inserting a
2345 if (!space || (space == 1 && start != offset))
2346 copy = msg->sg.data[i].length;
2348 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2349 get_order(copy + len));
2350 if (unlikely(!page))
2356 raw = page_address(page);
2358 psge = sk_msg_elem(msg, i);
2359 front = start - offset;
2360 back = psge->length - front;
2361 from = sg_virt(psge);
2364 memcpy(raw, from, front);
2368 to = raw + front + len;
2370 memcpy(to, from, back);
2373 put_page(sg_page(psge));
2374 } else if (start - offset) {
2375 psge = sk_msg_elem(msg, i);
2376 rsge = sk_msg_elem_cpy(msg, i);
2378 psge->length = start - offset;
2379 rsge.length -= psge->length;
2380 rsge.offset += start;
2382 sk_msg_iter_var_next(i);
2383 sg_unmark_end(psge);
2384 sk_msg_iter_next(msg, end);
2387 /* Slot(s) to place newly allocated data */
2390 /* Shift one or two slots as needed */
2392 sge = sk_msg_elem_cpy(msg, i);
2394 sk_msg_iter_var_next(i);
2395 sg_unmark_end(&sge);
2396 sk_msg_iter_next(msg, end);
2398 nsge = sk_msg_elem_cpy(msg, i);
2400 sk_msg_iter_var_next(i);
2401 nnsge = sk_msg_elem_cpy(msg, i);
2404 while (i != msg->sg.end) {
2405 msg->sg.data[i] = sge;
2407 sk_msg_iter_var_next(i);
2410 nnsge = sk_msg_elem_cpy(msg, i);
2412 nsge = sk_msg_elem_cpy(msg, i);
2417 /* Place newly allocated data buffer */
2418 sk_mem_charge(msg->sk, len);
2419 msg->sg.size += len;
2420 msg->sg.copy[new] = false;
2421 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2423 get_page(sg_page(&rsge));
2424 sk_msg_iter_var_next(new);
2425 msg->sg.data[new] = rsge;
2428 sk_msg_compute_data_pointers(msg);
2432 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2433 .func = bpf_msg_push_data,
2435 .ret_type = RET_INTEGER,
2436 .arg1_type = ARG_PTR_TO_CTX,
2437 .arg2_type = ARG_ANYTHING,
2438 .arg3_type = ARG_ANYTHING,
2439 .arg4_type = ARG_ANYTHING,
2442 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2448 sk_msg_iter_var_next(i);
2449 msg->sg.data[prev] = msg->sg.data[i];
2450 } while (i != msg->sg.end);
2452 sk_msg_iter_prev(msg, end);
2455 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2457 struct scatterlist tmp, sge;
2459 sk_msg_iter_next(msg, end);
2460 sge = sk_msg_elem_cpy(msg, i);
2461 sk_msg_iter_var_next(i);
2462 tmp = sk_msg_elem_cpy(msg, i);
2464 while (i != msg->sg.end) {
2465 msg->sg.data[i] = sge;
2466 sk_msg_iter_var_next(i);
2468 tmp = sk_msg_elem_cpy(msg, i);
2472 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2473 u32, len, u64, flags)
2475 u32 i = 0, l, space, offset = 0;
2476 u64 last = start + len;
2479 if (unlikely(flags))
2482 /* First find the starting scatterlist element */
2485 l = sk_msg_elem(msg, i)->length;
2487 if (start < offset + l)
2490 sk_msg_iter_var_next(i);
2491 } while (i != msg->sg.end);
2493 /* Bounds checks: start and pop must be inside message */
2494 if (start >= offset + l || last >= msg->sg.size)
2497 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2500 /* --------------| offset
2501 * -| start |-------- len -------|
2503 * |----- a ----|-------- pop -------|----- b ----|
2504 * |______________________________________________| length
2507 * a: region at front of scatter element to save
2508 * b: region at back of scatter element to save when length > A + pop
2509 * pop: region to pop from element, same as input 'pop' here will be
2510 * decremented below per iteration.
2512 * Two top-level cases to handle when start != offset, first B is non
2513 * zero and second B is zero corresponding to when a pop includes more
2516 * Then if B is non-zero AND there is no space allocate space and
2517 * compact A, B regions into page. If there is space shift ring to
2518 * the rigth free'ing the next element in ring to place B, leaving
2519 * A untouched except to reduce length.
2521 if (start != offset) {
2522 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2524 int b = sge->length - pop - a;
2526 sk_msg_iter_var_next(i);
2528 if (pop < sge->length - a) {
2531 sk_msg_shift_right(msg, i);
2532 nsge = sk_msg_elem(msg, i);
2533 get_page(sg_page(sge));
2536 b, sge->offset + pop + a);
2538 struct page *page, *orig;
2541 page = alloc_pages(__GFP_NOWARN |
2542 __GFP_COMP | GFP_ATOMIC,
2544 if (unlikely(!page))
2548 orig = sg_page(sge);
2549 from = sg_virt(sge);
2550 to = page_address(page);
2551 memcpy(to, from, a);
2552 memcpy(to + a, from + a + pop, b);
2553 sg_set_page(sge, page, a + b, 0);
2557 } else if (pop >= sge->length - a) {
2559 pop -= (sge->length - a);
2563 /* From above the current layout _must_ be as follows,
2568 * |---- pop ---|---------------- b ------------|
2569 * |____________________________________________| length
2571 * Offset and start of the current msg elem are equal because in the
2572 * previous case we handled offset != start and either consumed the
2573 * entire element and advanced to the next element OR pop == 0.
2575 * Two cases to handle here are first pop is less than the length
2576 * leaving some remainder b above. Simply adjust the element's layout
2577 * in this case. Or pop >= length of the element so that b = 0. In this
2578 * case advance to next element decrementing pop.
2581 struct scatterlist *sge = sk_msg_elem(msg, i);
2583 if (pop < sge->length) {
2589 sk_msg_shift_left(msg, i);
2591 sk_msg_iter_var_next(i);
2594 sk_mem_uncharge(msg->sk, len - pop);
2595 msg->sg.size -= (len - pop);
2596 sk_msg_compute_data_pointers(msg);
2600 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
2601 .func = bpf_msg_pop_data,
2603 .ret_type = RET_INTEGER,
2604 .arg1_type = ARG_PTR_TO_CTX,
2605 .arg2_type = ARG_ANYTHING,
2606 .arg3_type = ARG_ANYTHING,
2607 .arg4_type = ARG_ANYTHING,
2610 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
2612 return task_get_classid(skb);
2615 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
2616 .func = bpf_get_cgroup_classid,
2618 .ret_type = RET_INTEGER,
2619 .arg1_type = ARG_PTR_TO_CTX,
2622 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
2624 return dst_tclassid(skb);
2627 static const struct bpf_func_proto bpf_get_route_realm_proto = {
2628 .func = bpf_get_route_realm,
2630 .ret_type = RET_INTEGER,
2631 .arg1_type = ARG_PTR_TO_CTX,
2634 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
2636 /* If skb_clear_hash() was called due to mangling, we can
2637 * trigger SW recalculation here. Later access to hash
2638 * can then use the inline skb->hash via context directly
2639 * instead of calling this helper again.
2641 return skb_get_hash(skb);
2644 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
2645 .func = bpf_get_hash_recalc,
2647 .ret_type = RET_INTEGER,
2648 .arg1_type = ARG_PTR_TO_CTX,
2651 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
2653 /* After all direct packet write, this can be used once for
2654 * triggering a lazy recalc on next skb_get_hash() invocation.
2656 skb_clear_hash(skb);
2660 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
2661 .func = bpf_set_hash_invalid,
2663 .ret_type = RET_INTEGER,
2664 .arg1_type = ARG_PTR_TO_CTX,
2667 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
2669 /* Set user specified hash as L4(+), so that it gets returned
2670 * on skb_get_hash() call unless BPF prog later on triggers a
2673 __skb_set_sw_hash(skb, hash, true);
2677 static const struct bpf_func_proto bpf_set_hash_proto = {
2678 .func = bpf_set_hash,
2680 .ret_type = RET_INTEGER,
2681 .arg1_type = ARG_PTR_TO_CTX,
2682 .arg2_type = ARG_ANYTHING,
2685 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
2690 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
2691 vlan_proto != htons(ETH_P_8021AD)))
2692 vlan_proto = htons(ETH_P_8021Q);
2694 bpf_push_mac_rcsum(skb);
2695 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
2696 bpf_pull_mac_rcsum(skb);
2698 bpf_compute_data_pointers(skb);
2702 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
2703 .func = bpf_skb_vlan_push,
2705 .ret_type = RET_INTEGER,
2706 .arg1_type = ARG_PTR_TO_CTX,
2707 .arg2_type = ARG_ANYTHING,
2708 .arg3_type = ARG_ANYTHING,
2711 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
2715 bpf_push_mac_rcsum(skb);
2716 ret = skb_vlan_pop(skb);
2717 bpf_pull_mac_rcsum(skb);
2719 bpf_compute_data_pointers(skb);
2723 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
2724 .func = bpf_skb_vlan_pop,
2726 .ret_type = RET_INTEGER,
2727 .arg1_type = ARG_PTR_TO_CTX,
2730 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
2732 /* Caller already did skb_cow() with len as headroom,
2733 * so no need to do it here.
2736 memmove(skb->data, skb->data + len, off);
2737 memset(skb->data + off, 0, len);
2739 /* No skb_postpush_rcsum(skb, skb->data + off, len)
2740 * needed here as it does not change the skb->csum
2741 * result for checksum complete when summing over
2747 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
2749 /* skb_ensure_writable() is not needed here, as we're
2750 * already working on an uncloned skb.
2752 if (unlikely(!pskb_may_pull(skb, off + len)))
2755 skb_postpull_rcsum(skb, skb->data + off, len);
2756 memmove(skb->data + len, skb->data, off);
2757 __skb_pull(skb, len);
2762 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
2764 bool trans_same = skb->transport_header == skb->network_header;
2767 /* There's no need for __skb_push()/__skb_pull() pair to
2768 * get to the start of the mac header as we're guaranteed
2769 * to always start from here under eBPF.
2771 ret = bpf_skb_generic_push(skb, off, len);
2773 skb->mac_header -= len;
2774 skb->network_header -= len;
2776 skb->transport_header = skb->network_header;
2782 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
2784 bool trans_same = skb->transport_header == skb->network_header;
2787 /* Same here, __skb_push()/__skb_pull() pair not needed. */
2788 ret = bpf_skb_generic_pop(skb, off, len);
2790 skb->mac_header += len;
2791 skb->network_header += len;
2793 skb->transport_header = skb->network_header;
2799 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
2801 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2802 u32 off = skb_mac_header_len(skb);
2805 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
2808 ret = skb_cow(skb, len_diff);
2809 if (unlikely(ret < 0))
2812 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2813 if (unlikely(ret < 0))
2816 if (skb_is_gso(skb)) {
2817 struct skb_shared_info *shinfo = skb_shinfo(skb);
2819 /* SKB_GSO_TCPV4 needs to be changed into
2822 if (shinfo->gso_type & SKB_GSO_TCPV4) {
2823 shinfo->gso_type &= ~SKB_GSO_TCPV4;
2824 shinfo->gso_type |= SKB_GSO_TCPV6;
2827 /* Due to IPv6 header, MSS needs to be downgraded. */
2828 skb_decrease_gso_size(shinfo, len_diff);
2829 /* Header must be checked, and gso_segs recomputed. */
2830 shinfo->gso_type |= SKB_GSO_DODGY;
2831 shinfo->gso_segs = 0;
2834 skb->protocol = htons(ETH_P_IPV6);
2835 skb_clear_hash(skb);
2840 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
2842 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2843 u32 off = skb_mac_header_len(skb);
2846 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
2849 ret = skb_unclone(skb, GFP_ATOMIC);
2850 if (unlikely(ret < 0))
2853 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2854 if (unlikely(ret < 0))
2857 if (skb_is_gso(skb)) {
2858 struct skb_shared_info *shinfo = skb_shinfo(skb);
2860 /* SKB_GSO_TCPV6 needs to be changed into
2863 if (shinfo->gso_type & SKB_GSO_TCPV6) {
2864 shinfo->gso_type &= ~SKB_GSO_TCPV6;
2865 shinfo->gso_type |= SKB_GSO_TCPV4;
2868 /* Due to IPv4 header, MSS can be upgraded. */
2869 skb_increase_gso_size(shinfo, len_diff);
2870 /* Header must be checked, and gso_segs recomputed. */
2871 shinfo->gso_type |= SKB_GSO_DODGY;
2872 shinfo->gso_segs = 0;
2875 skb->protocol = htons(ETH_P_IP);
2876 skb_clear_hash(skb);
2881 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
2883 __be16 from_proto = skb->protocol;
2885 if (from_proto == htons(ETH_P_IP) &&
2886 to_proto == htons(ETH_P_IPV6))
2887 return bpf_skb_proto_4_to_6(skb);
2889 if (from_proto == htons(ETH_P_IPV6) &&
2890 to_proto == htons(ETH_P_IP))
2891 return bpf_skb_proto_6_to_4(skb);
2896 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
2901 if (unlikely(flags))
2904 /* General idea is that this helper does the basic groundwork
2905 * needed for changing the protocol, and eBPF program fills the
2906 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
2907 * and other helpers, rather than passing a raw buffer here.
2909 * The rationale is to keep this minimal and without a need to
2910 * deal with raw packet data. F.e. even if we would pass buffers
2911 * here, the program still needs to call the bpf_lX_csum_replace()
2912 * helpers anyway. Plus, this way we keep also separation of
2913 * concerns, since f.e. bpf_skb_store_bytes() should only take
2916 * Currently, additional options and extension header space are
2917 * not supported, but flags register is reserved so we can adapt
2918 * that. For offloads, we mark packet as dodgy, so that headers
2919 * need to be verified first.
2921 ret = bpf_skb_proto_xlat(skb, proto);
2922 bpf_compute_data_pointers(skb);
2926 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
2927 .func = bpf_skb_change_proto,
2929 .ret_type = RET_INTEGER,
2930 .arg1_type = ARG_PTR_TO_CTX,
2931 .arg2_type = ARG_ANYTHING,
2932 .arg3_type = ARG_ANYTHING,
2935 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
2937 /* We only allow a restricted subset to be changed for now. */
2938 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
2939 !skb_pkt_type_ok(pkt_type)))
2942 skb->pkt_type = pkt_type;
2946 static const struct bpf_func_proto bpf_skb_change_type_proto = {
2947 .func = bpf_skb_change_type,
2949 .ret_type = RET_INTEGER,
2950 .arg1_type = ARG_PTR_TO_CTX,
2951 .arg2_type = ARG_ANYTHING,
2954 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
2956 switch (skb->protocol) {
2957 case htons(ETH_P_IP):
2958 return sizeof(struct iphdr);
2959 case htons(ETH_P_IPV6):
2960 return sizeof(struct ipv6hdr);
2966 static int bpf_skb_net_grow(struct sk_buff *skb, u32 len_diff)
2968 u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
2971 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
2974 ret = skb_cow(skb, len_diff);
2975 if (unlikely(ret < 0))
2978 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2979 if (unlikely(ret < 0))
2982 if (skb_is_gso(skb)) {
2983 struct skb_shared_info *shinfo = skb_shinfo(skb);
2985 /* Due to header grow, MSS needs to be downgraded. */
2986 skb_decrease_gso_size(shinfo, len_diff);
2987 /* Header must be checked, and gso_segs recomputed. */
2988 shinfo->gso_type |= SKB_GSO_DODGY;
2989 shinfo->gso_segs = 0;
2995 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 len_diff)
2997 u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
3000 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
3003 ret = skb_unclone(skb, GFP_ATOMIC);
3004 if (unlikely(ret < 0))
3007 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3008 if (unlikely(ret < 0))
3011 if (skb_is_gso(skb)) {
3012 struct skb_shared_info *shinfo = skb_shinfo(skb);
3014 /* Due to header shrink, MSS can be upgraded. */
3015 skb_increase_gso_size(shinfo, len_diff);
3016 /* Header must be checked, and gso_segs recomputed. */
3017 shinfo->gso_type |= SKB_GSO_DODGY;
3018 shinfo->gso_segs = 0;
3024 static u32 __bpf_skb_max_len(const struct sk_buff *skb)
3026 return skb->dev ? skb->dev->mtu + skb->dev->hard_header_len :
3030 static int bpf_skb_adjust_net(struct sk_buff *skb, s32 len_diff)
3032 bool trans_same = skb->transport_header == skb->network_header;
3033 u32 len_cur, len_diff_abs = abs(len_diff);
3034 u32 len_min = bpf_skb_net_base_len(skb);
3035 u32 len_max = __bpf_skb_max_len(skb);
3036 __be16 proto = skb->protocol;
3037 bool shrink = len_diff < 0;
3040 if (unlikely(len_diff_abs > 0xfffU))
3042 if (unlikely(proto != htons(ETH_P_IP) &&
3043 proto != htons(ETH_P_IPV6)))
3046 len_cur = skb->len - skb_network_offset(skb);
3047 if (skb_transport_header_was_set(skb) && !trans_same)
3048 len_cur = skb_network_header_len(skb);
3049 if ((shrink && (len_diff_abs >= len_cur ||
3050 len_cur - len_diff_abs < len_min)) ||
3051 (!shrink && (skb->len + len_diff_abs > len_max &&
3055 ret = shrink ? bpf_skb_net_shrink(skb, len_diff_abs) :
3056 bpf_skb_net_grow(skb, len_diff_abs);
3058 bpf_compute_data_pointers(skb);
3062 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3063 u32, mode, u64, flags)
3065 if (unlikely(flags))
3067 if (likely(mode == BPF_ADJ_ROOM_NET))
3068 return bpf_skb_adjust_net(skb, len_diff);
3073 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3074 .func = bpf_skb_adjust_room,
3076 .ret_type = RET_INTEGER,
3077 .arg1_type = ARG_PTR_TO_CTX,
3078 .arg2_type = ARG_ANYTHING,
3079 .arg3_type = ARG_ANYTHING,
3080 .arg4_type = ARG_ANYTHING,
3083 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3085 u32 min_len = skb_network_offset(skb);
3087 if (skb_transport_header_was_set(skb))
3088 min_len = skb_transport_offset(skb);
3089 if (skb->ip_summed == CHECKSUM_PARTIAL)
3090 min_len = skb_checksum_start_offset(skb) +
3091 skb->csum_offset + sizeof(__sum16);
3095 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3097 unsigned int old_len = skb->len;
3100 ret = __skb_grow_rcsum(skb, new_len);
3102 memset(skb->data + old_len, 0, new_len - old_len);
3106 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3108 return __skb_trim_rcsum(skb, new_len);
3111 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3114 u32 max_len = __bpf_skb_max_len(skb);
3115 u32 min_len = __bpf_skb_min_len(skb);
3118 if (unlikely(flags || new_len > max_len || new_len < min_len))
3120 if (skb->encapsulation)
3123 /* The basic idea of this helper is that it's performing the
3124 * needed work to either grow or trim an skb, and eBPF program
3125 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3126 * bpf_lX_csum_replace() and others rather than passing a raw
3127 * buffer here. This one is a slow path helper and intended
3128 * for replies with control messages.
3130 * Like in bpf_skb_change_proto(), we want to keep this rather
3131 * minimal and without protocol specifics so that we are able
3132 * to separate concerns as in bpf_skb_store_bytes() should only
3133 * be the one responsible for writing buffers.
3135 * It's really expected to be a slow path operation here for
3136 * control message replies, so we're implicitly linearizing,
3137 * uncloning and drop offloads from the skb by this.
3139 ret = __bpf_try_make_writable(skb, skb->len);
3141 if (new_len > skb->len)
3142 ret = bpf_skb_grow_rcsum(skb, new_len);
3143 else if (new_len < skb->len)
3144 ret = bpf_skb_trim_rcsum(skb, new_len);
3145 if (!ret && skb_is_gso(skb))
3151 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3154 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3156 bpf_compute_data_pointers(skb);
3160 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3161 .func = bpf_skb_change_tail,
3163 .ret_type = RET_INTEGER,
3164 .arg1_type = ARG_PTR_TO_CTX,
3165 .arg2_type = ARG_ANYTHING,
3166 .arg3_type = ARG_ANYTHING,
3169 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3172 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3174 bpf_compute_data_end_sk_skb(skb);
3178 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3179 .func = sk_skb_change_tail,
3181 .ret_type = RET_INTEGER,
3182 .arg1_type = ARG_PTR_TO_CTX,
3183 .arg2_type = ARG_ANYTHING,
3184 .arg3_type = ARG_ANYTHING,
3187 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3190 u32 max_len = __bpf_skb_max_len(skb);
3191 u32 new_len = skb->len + head_room;
3194 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3195 new_len < skb->len))
3198 ret = skb_cow(skb, head_room);
3200 /* Idea for this helper is that we currently only
3201 * allow to expand on mac header. This means that
3202 * skb->protocol network header, etc, stay as is.
3203 * Compared to bpf_skb_change_tail(), we're more
3204 * flexible due to not needing to linearize or
3205 * reset GSO. Intention for this helper is to be
3206 * used by an L3 skb that needs to push mac header
3207 * for redirection into L2 device.
3209 __skb_push(skb, head_room);
3210 memset(skb->data, 0, head_room);
3211 skb_reset_mac_header(skb);
3217 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3220 int ret = __bpf_skb_change_head(skb, head_room, flags);
3222 bpf_compute_data_pointers(skb);
3226 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3227 .func = bpf_skb_change_head,
3229 .ret_type = RET_INTEGER,
3230 .arg1_type = ARG_PTR_TO_CTX,
3231 .arg2_type = ARG_ANYTHING,
3232 .arg3_type = ARG_ANYTHING,
3235 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3238 int ret = __bpf_skb_change_head(skb, head_room, flags);
3240 bpf_compute_data_end_sk_skb(skb);
3244 static const struct bpf_func_proto sk_skb_change_head_proto = {
3245 .func = sk_skb_change_head,
3247 .ret_type = RET_INTEGER,
3248 .arg1_type = ARG_PTR_TO_CTX,
3249 .arg2_type = ARG_ANYTHING,
3250 .arg3_type = ARG_ANYTHING,
3252 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3254 return xdp_data_meta_unsupported(xdp) ? 0 :
3255 xdp->data - xdp->data_meta;
3258 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3260 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3261 unsigned long metalen = xdp_get_metalen(xdp);
3262 void *data_start = xdp_frame_end + metalen;
3263 void *data = xdp->data + offset;
3265 if (unlikely(data < data_start ||
3266 data > xdp->data_end - ETH_HLEN))
3270 memmove(xdp->data_meta + offset,
3271 xdp->data_meta, metalen);
3272 xdp->data_meta += offset;
3278 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3279 .func = bpf_xdp_adjust_head,
3281 .ret_type = RET_INTEGER,
3282 .arg1_type = ARG_PTR_TO_CTX,
3283 .arg2_type = ARG_ANYTHING,
3286 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3288 void *data_end = xdp->data_end + offset;
3290 /* only shrinking is allowed for now. */
3291 if (unlikely(offset >= 0))
3294 if (unlikely(data_end < xdp->data + ETH_HLEN))
3297 xdp->data_end = data_end;
3302 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3303 .func = bpf_xdp_adjust_tail,
3305 .ret_type = RET_INTEGER,
3306 .arg1_type = ARG_PTR_TO_CTX,
3307 .arg2_type = ARG_ANYTHING,
3310 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3312 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3313 void *meta = xdp->data_meta + offset;
3314 unsigned long metalen = xdp->data - meta;
3316 if (xdp_data_meta_unsupported(xdp))
3318 if (unlikely(meta < xdp_frame_end ||
3321 if (unlikely((metalen & (sizeof(__u32) - 1)) ||
3325 xdp->data_meta = meta;
3330 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3331 .func = bpf_xdp_adjust_meta,
3333 .ret_type = RET_INTEGER,
3334 .arg1_type = ARG_PTR_TO_CTX,
3335 .arg2_type = ARG_ANYTHING,
3338 static int __bpf_tx_xdp(struct net_device *dev,
3339 struct bpf_map *map,
3340 struct xdp_buff *xdp,
3343 struct xdp_frame *xdpf;
3346 if (!dev->netdev_ops->ndo_xdp_xmit) {
3350 err = xdp_ok_fwd_dev(dev, xdp->data_end - xdp->data);
3354 xdpf = convert_to_xdp_frame(xdp);
3355 if (unlikely(!xdpf))
3358 sent = dev->netdev_ops->ndo_xdp_xmit(dev, 1, &xdpf, XDP_XMIT_FLUSH);
3365 xdp_do_redirect_slow(struct net_device *dev, struct xdp_buff *xdp,
3366 struct bpf_prog *xdp_prog, struct bpf_redirect_info *ri)
3368 struct net_device *fwd;
3369 u32 index = ri->ifindex;
3372 fwd = dev_get_by_index_rcu(dev_net(dev), index);
3374 if (unlikely(!fwd)) {
3379 err = __bpf_tx_xdp(fwd, NULL, xdp, 0);
3383 _trace_xdp_redirect(dev, xdp_prog, index);
3386 _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3390 static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd,
3391 struct bpf_map *map,
3392 struct xdp_buff *xdp,
3397 switch (map->map_type) {
3398 case BPF_MAP_TYPE_DEVMAP: {
3399 struct bpf_dtab_netdev *dst = fwd;
3401 err = dev_map_enqueue(dst, xdp, dev_rx);
3404 __dev_map_insert_ctx(map, index);
3407 case BPF_MAP_TYPE_CPUMAP: {
3408 struct bpf_cpu_map_entry *rcpu = fwd;
3410 err = cpu_map_enqueue(rcpu, xdp, dev_rx);
3413 __cpu_map_insert_ctx(map, index);
3416 case BPF_MAP_TYPE_XSKMAP: {
3417 struct xdp_sock *xs = fwd;
3419 err = __xsk_map_redirect(map, xdp, xs);
3428 void xdp_do_flush_map(void)
3430 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3431 struct bpf_map *map = ri->map_to_flush;
3433 ri->map_to_flush = NULL;
3435 switch (map->map_type) {
3436 case BPF_MAP_TYPE_DEVMAP:
3437 __dev_map_flush(map);
3439 case BPF_MAP_TYPE_CPUMAP:
3440 __cpu_map_flush(map);
3442 case BPF_MAP_TYPE_XSKMAP:
3443 __xsk_map_flush(map);
3450 EXPORT_SYMBOL_GPL(xdp_do_flush_map);
3452 static inline void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index)
3454 switch (map->map_type) {
3455 case BPF_MAP_TYPE_DEVMAP:
3456 return __dev_map_lookup_elem(map, index);
3457 case BPF_MAP_TYPE_CPUMAP:
3458 return __cpu_map_lookup_elem(map, index);
3459 case BPF_MAP_TYPE_XSKMAP:
3460 return __xsk_map_lookup_elem(map, index);
3466 void bpf_clear_redirect_map(struct bpf_map *map)
3468 struct bpf_redirect_info *ri;
3471 for_each_possible_cpu(cpu) {
3472 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3473 /* Avoid polluting remote cacheline due to writes if
3474 * not needed. Once we pass this test, we need the
3475 * cmpxchg() to make sure it hasn't been changed in
3476 * the meantime by remote CPU.
3478 if (unlikely(READ_ONCE(ri->map) == map))
3479 cmpxchg(&ri->map, map, NULL);
3483 static int xdp_do_redirect_map(struct net_device *dev, struct xdp_buff *xdp,
3484 struct bpf_prog *xdp_prog, struct bpf_map *map,
3485 struct bpf_redirect_info *ri)
3487 u32 index = ri->ifindex;
3492 WRITE_ONCE(ri->map, NULL);
3494 fwd = __xdp_map_lookup_elem(map, index);
3495 if (unlikely(!fwd)) {
3499 if (ri->map_to_flush && unlikely(ri->map_to_flush != map))
3502 err = __bpf_tx_xdp_map(dev, fwd, map, xdp, index);
3506 ri->map_to_flush = map;
3507 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3510 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3514 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3515 struct bpf_prog *xdp_prog)
3517 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3518 struct bpf_map *map = READ_ONCE(ri->map);
3521 return xdp_do_redirect_map(dev, xdp, xdp_prog, map, ri);
3523 return xdp_do_redirect_slow(dev, xdp, xdp_prog, ri);
3525 EXPORT_SYMBOL_GPL(xdp_do_redirect);
3527 static int xdp_do_generic_redirect_map(struct net_device *dev,
3528 struct sk_buff *skb,
3529 struct xdp_buff *xdp,
3530 struct bpf_prog *xdp_prog,
3531 struct bpf_map *map)
3533 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3534 u32 index = ri->ifindex;
3539 WRITE_ONCE(ri->map, NULL);
3541 fwd = __xdp_map_lookup_elem(map, index);
3542 if (unlikely(!fwd)) {
3547 if (map->map_type == BPF_MAP_TYPE_DEVMAP) {
3548 struct bpf_dtab_netdev *dst = fwd;
3550 err = dev_map_generic_redirect(dst, skb, xdp_prog);
3553 } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) {
3554 struct xdp_sock *xs = fwd;
3556 err = xsk_generic_rcv(xs, xdp);
3561 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
3566 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3569 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3573 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
3574 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
3576 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3577 struct bpf_map *map = READ_ONCE(ri->map);
3578 u32 index = ri->ifindex;
3579 struct net_device *fwd;
3583 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog,
3586 fwd = dev_get_by_index_rcu(dev_net(dev), index);
3587 if (unlikely(!fwd)) {
3592 err = xdp_ok_fwd_dev(fwd, skb->len);
3597 _trace_xdp_redirect(dev, xdp_prog, index);
3598 generic_xdp_tx(skb, xdp_prog);
3601 _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3604 EXPORT_SYMBOL_GPL(xdp_do_generic_redirect);
3606 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
3608 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3610 if (unlikely(flags))
3613 ri->ifindex = ifindex;
3615 WRITE_ONCE(ri->map, NULL);
3617 return XDP_REDIRECT;
3620 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
3621 .func = bpf_xdp_redirect,
3623 .ret_type = RET_INTEGER,
3624 .arg1_type = ARG_ANYTHING,
3625 .arg2_type = ARG_ANYTHING,
3628 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
3631 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3633 if (unlikely(flags))
3636 ri->ifindex = ifindex;
3638 WRITE_ONCE(ri->map, map);
3640 return XDP_REDIRECT;
3643 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
3644 .func = bpf_xdp_redirect_map,
3646 .ret_type = RET_INTEGER,
3647 .arg1_type = ARG_CONST_MAP_PTR,
3648 .arg2_type = ARG_ANYTHING,
3649 .arg3_type = ARG_ANYTHING,
3652 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
3653 unsigned long off, unsigned long len)
3655 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
3659 if (ptr != dst_buff)
3660 memcpy(dst_buff, ptr, len);
3665 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
3666 u64, flags, void *, meta, u64, meta_size)
3668 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
3670 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
3672 if (unlikely(skb_size > skb->len))
3675 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
3679 static const struct bpf_func_proto bpf_skb_event_output_proto = {
3680 .func = bpf_skb_event_output,
3682 .ret_type = RET_INTEGER,
3683 .arg1_type = ARG_PTR_TO_CTX,
3684 .arg2_type = ARG_CONST_MAP_PTR,
3685 .arg3_type = ARG_ANYTHING,
3686 .arg4_type = ARG_PTR_TO_MEM,
3687 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
3690 static unsigned short bpf_tunnel_key_af(u64 flags)
3692 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
3695 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
3696 u32, size, u64, flags)
3698 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3699 u8 compat[sizeof(struct bpf_tunnel_key)];
3703 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
3707 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
3711 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3714 case offsetof(struct bpf_tunnel_key, tunnel_label):
3715 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3717 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3718 /* Fixup deprecated structure layouts here, so we have
3719 * a common path later on.
3721 if (ip_tunnel_info_af(info) != AF_INET)
3724 to = (struct bpf_tunnel_key *)compat;
3731 to->tunnel_id = be64_to_cpu(info->key.tun_id);
3732 to->tunnel_tos = info->key.tos;
3733 to->tunnel_ttl = info->key.ttl;
3736 if (flags & BPF_F_TUNINFO_IPV6) {
3737 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
3738 sizeof(to->remote_ipv6));
3739 to->tunnel_label = be32_to_cpu(info->key.label);
3741 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
3742 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
3743 to->tunnel_label = 0;
3746 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
3747 memcpy(to_orig, to, size);
3751 memset(to_orig, 0, size);
3755 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
3756 .func = bpf_skb_get_tunnel_key,
3758 .ret_type = RET_INTEGER,
3759 .arg1_type = ARG_PTR_TO_CTX,
3760 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
3761 .arg3_type = ARG_CONST_SIZE,
3762 .arg4_type = ARG_ANYTHING,
3765 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
3767 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3770 if (unlikely(!info ||
3771 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
3775 if (unlikely(size < info->options_len)) {
3780 ip_tunnel_info_opts_get(to, info);
3781 if (size > info->options_len)
3782 memset(to + info->options_len, 0, size - info->options_len);
3784 return info->options_len;
3786 memset(to, 0, size);
3790 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
3791 .func = bpf_skb_get_tunnel_opt,
3793 .ret_type = RET_INTEGER,
3794 .arg1_type = ARG_PTR_TO_CTX,
3795 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
3796 .arg3_type = ARG_CONST_SIZE,
3799 static struct metadata_dst __percpu *md_dst;
3801 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
3802 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
3804 struct metadata_dst *md = this_cpu_ptr(md_dst);
3805 u8 compat[sizeof(struct bpf_tunnel_key)];
3806 struct ip_tunnel_info *info;
3808 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
3809 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
3811 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3813 case offsetof(struct bpf_tunnel_key, tunnel_label):
3814 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3815 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3816 /* Fixup deprecated structure layouts here, so we have
3817 * a common path later on.
3819 memcpy(compat, from, size);
3820 memset(compat + size, 0, sizeof(compat) - size);
3821 from = (const struct bpf_tunnel_key *) compat;
3827 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
3832 dst_hold((struct dst_entry *) md);
3833 skb_dst_set(skb, (struct dst_entry *) md);
3835 info = &md->u.tun_info;
3836 memset(info, 0, sizeof(*info));
3837 info->mode = IP_TUNNEL_INFO_TX;
3839 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
3840 if (flags & BPF_F_DONT_FRAGMENT)
3841 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
3842 if (flags & BPF_F_ZERO_CSUM_TX)
3843 info->key.tun_flags &= ~TUNNEL_CSUM;
3844 if (flags & BPF_F_SEQ_NUMBER)
3845 info->key.tun_flags |= TUNNEL_SEQ;
3847 info->key.tun_id = cpu_to_be64(from->tunnel_id);
3848 info->key.tos = from->tunnel_tos;
3849 info->key.ttl = from->tunnel_ttl;
3851 if (flags & BPF_F_TUNINFO_IPV6) {
3852 info->mode |= IP_TUNNEL_INFO_IPV6;
3853 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
3854 sizeof(from->remote_ipv6));
3855 info->key.label = cpu_to_be32(from->tunnel_label) &
3856 IPV6_FLOWLABEL_MASK;
3858 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
3864 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
3865 .func = bpf_skb_set_tunnel_key,
3867 .ret_type = RET_INTEGER,
3868 .arg1_type = ARG_PTR_TO_CTX,
3869 .arg2_type = ARG_PTR_TO_MEM,
3870 .arg3_type = ARG_CONST_SIZE,
3871 .arg4_type = ARG_ANYTHING,
3874 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
3875 const u8 *, from, u32, size)
3877 struct ip_tunnel_info *info = skb_tunnel_info(skb);
3878 const struct metadata_dst *md = this_cpu_ptr(md_dst);
3880 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
3882 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
3885 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
3890 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
3891 .func = bpf_skb_set_tunnel_opt,
3893 .ret_type = RET_INTEGER,
3894 .arg1_type = ARG_PTR_TO_CTX,
3895 .arg2_type = ARG_PTR_TO_MEM,
3896 .arg3_type = ARG_CONST_SIZE,
3899 static const struct bpf_func_proto *
3900 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
3903 struct metadata_dst __percpu *tmp;
3905 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
3910 if (cmpxchg(&md_dst, NULL, tmp))
3911 metadata_dst_free_percpu(tmp);
3915 case BPF_FUNC_skb_set_tunnel_key:
3916 return &bpf_skb_set_tunnel_key_proto;
3917 case BPF_FUNC_skb_set_tunnel_opt:
3918 return &bpf_skb_set_tunnel_opt_proto;
3924 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
3927 struct bpf_array *array = container_of(map, struct bpf_array, map);
3928 struct cgroup *cgrp;
3931 sk = skb_to_full_sk(skb);
3932 if (!sk || !sk_fullsock(sk))
3934 if (unlikely(idx >= array->map.max_entries))
3937 cgrp = READ_ONCE(array->ptrs[idx]);
3938 if (unlikely(!cgrp))
3941 return sk_under_cgroup_hierarchy(sk, cgrp);
3944 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
3945 .func = bpf_skb_under_cgroup,
3947 .ret_type = RET_INTEGER,
3948 .arg1_type = ARG_PTR_TO_CTX,
3949 .arg2_type = ARG_CONST_MAP_PTR,
3950 .arg3_type = ARG_ANYTHING,
3953 #ifdef CONFIG_SOCK_CGROUP_DATA
3954 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
3956 struct sock *sk = skb_to_full_sk(skb);
3957 struct cgroup *cgrp;
3959 if (!sk || !sk_fullsock(sk))
3962 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
3963 return cgrp->kn->id.id;
3966 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
3967 .func = bpf_skb_cgroup_id,
3969 .ret_type = RET_INTEGER,
3970 .arg1_type = ARG_PTR_TO_CTX,
3973 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
3976 struct sock *sk = skb_to_full_sk(skb);
3977 struct cgroup *ancestor;
3978 struct cgroup *cgrp;
3980 if (!sk || !sk_fullsock(sk))
3983 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
3984 ancestor = cgroup_ancestor(cgrp, ancestor_level);
3988 return ancestor->kn->id.id;
3991 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
3992 .func = bpf_skb_ancestor_cgroup_id,
3994 .ret_type = RET_INTEGER,
3995 .arg1_type = ARG_PTR_TO_CTX,
3996 .arg2_type = ARG_ANYTHING,
4000 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4001 unsigned long off, unsigned long len)
4003 memcpy(dst_buff, src_buff + off, len);
4007 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4008 u64, flags, void *, meta, u64, meta_size)
4010 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4012 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4014 if (unlikely(xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4017 return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4018 xdp_size, bpf_xdp_copy);
4021 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4022 .func = bpf_xdp_event_output,
4024 .ret_type = RET_INTEGER,
4025 .arg1_type = ARG_PTR_TO_CTX,
4026 .arg2_type = ARG_CONST_MAP_PTR,
4027 .arg3_type = ARG_ANYTHING,
4028 .arg4_type = ARG_PTR_TO_MEM,
4029 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4032 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4034 return skb->sk ? sock_gen_cookie(skb->sk) : 0;
4037 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4038 .func = bpf_get_socket_cookie,
4040 .ret_type = RET_INTEGER,
4041 .arg1_type = ARG_PTR_TO_CTX,
4044 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4046 return sock_gen_cookie(ctx->sk);
4049 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4050 .func = bpf_get_socket_cookie_sock_addr,
4052 .ret_type = RET_INTEGER,
4053 .arg1_type = ARG_PTR_TO_CTX,
4056 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4058 return sock_gen_cookie(ctx->sk);
4061 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4062 .func = bpf_get_socket_cookie_sock_ops,
4064 .ret_type = RET_INTEGER,
4065 .arg1_type = ARG_PTR_TO_CTX,
4068 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4070 struct sock *sk = sk_to_full_sk(skb->sk);
4073 if (!sk || !sk_fullsock(sk))
4075 kuid = sock_net_uid(sock_net(sk), sk);
4076 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4079 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4080 .func = bpf_get_socket_uid,
4082 .ret_type = RET_INTEGER,
4083 .arg1_type = ARG_PTR_TO_CTX,
4086 BPF_CALL_5(bpf_sockopt_event_output, struct bpf_sock_ops_kern *, bpf_sock,
4087 struct bpf_map *, map, u64, flags, void *, data, u64, size)
4089 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
4092 return bpf_event_output(map, flags, data, size, NULL, 0, NULL);
4095 static const struct bpf_func_proto bpf_sockopt_event_output_proto = {
4096 .func = bpf_sockopt_event_output,
4098 .ret_type = RET_INTEGER,
4099 .arg1_type = ARG_PTR_TO_CTX,
4100 .arg2_type = ARG_CONST_MAP_PTR,
4101 .arg3_type = ARG_ANYTHING,
4102 .arg4_type = ARG_PTR_TO_MEM,
4103 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4106 BPF_CALL_5(bpf_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
4107 int, level, int, optname, char *, optval, int, optlen)
4109 struct sock *sk = bpf_sock->sk;
4113 if (!sk_fullsock(sk))
4116 if (level == SOL_SOCKET) {
4117 if (optlen != sizeof(int))
4119 val = *((int *)optval);
4121 /* Only some socketops are supported */
4124 val = min_t(u32, val, sysctl_rmem_max);
4125 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4126 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
4129 val = min_t(u32, val, sysctl_wmem_max);
4130 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4131 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
4133 case SO_MAX_PACING_RATE: /* 32bit version */
4135 cmpxchg(&sk->sk_pacing_status,
4138 sk->sk_max_pacing_rate = (val == ~0U) ? ~0UL : val;
4139 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4140 sk->sk_max_pacing_rate);
4143 sk->sk_priority = val;
4148 sk->sk_rcvlowat = val ? : 1;
4151 if (sk->sk_mark != val) {
4160 } else if (level == SOL_IP) {
4161 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4164 val = *((int *)optval);
4165 /* Only some options are supported */
4168 if (val < -1 || val > 0xff) {
4171 struct inet_sock *inet = inet_sk(sk);
4181 #if IS_ENABLED(CONFIG_IPV6)
4182 } else if (level == SOL_IPV6) {
4183 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4186 val = *((int *)optval);
4187 /* Only some options are supported */
4190 if (val < -1 || val > 0xff) {
4193 struct ipv6_pinfo *np = inet6_sk(sk);
4204 } else if (level == SOL_TCP &&
4205 sk->sk_prot->setsockopt == tcp_setsockopt) {
4206 if (optname == TCP_CONGESTION) {
4207 char name[TCP_CA_NAME_MAX];
4208 bool reinit = bpf_sock->op > BPF_SOCK_OPS_NEEDS_ECN;
4210 strncpy(name, optval, min_t(long, optlen,
4211 TCP_CA_NAME_MAX-1));
4212 name[TCP_CA_NAME_MAX-1] = 0;
4213 ret = tcp_set_congestion_control(sk, name, false,
4216 struct tcp_sock *tp = tcp_sk(sk);
4218 if (optlen != sizeof(int))
4221 val = *((int *)optval);
4222 /* Only some options are supported */
4225 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4230 case TCP_BPF_SNDCWND_CLAMP:
4234 tp->snd_cwnd_clamp = val;
4235 tp->snd_ssthresh = val;
4239 if (val < 0 || val > 1)
4255 static const struct bpf_func_proto bpf_setsockopt_proto = {
4256 .func = bpf_setsockopt,
4258 .ret_type = RET_INTEGER,
4259 .arg1_type = ARG_PTR_TO_CTX,
4260 .arg2_type = ARG_ANYTHING,
4261 .arg3_type = ARG_ANYTHING,
4262 .arg4_type = ARG_PTR_TO_MEM,
4263 .arg5_type = ARG_CONST_SIZE,
4266 BPF_CALL_5(bpf_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
4267 int, level, int, optname, char *, optval, int, optlen)
4269 struct sock *sk = bpf_sock->sk;
4271 if (!sk_fullsock(sk))
4274 if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4275 struct inet_connection_sock *icsk;
4276 struct tcp_sock *tp;
4279 case TCP_CONGESTION:
4280 icsk = inet_csk(sk);
4282 if (!icsk->icsk_ca_ops || optlen <= 1)
4284 strncpy(optval, icsk->icsk_ca_ops->name, optlen);
4285 optval[optlen - 1] = 0;
4290 if (optlen <= 0 || !tp->saved_syn ||
4291 optlen > tp->saved_syn[0])
4293 memcpy(optval, tp->saved_syn + 1, optlen);
4298 } else if (level == SOL_IP) {
4299 struct inet_sock *inet = inet_sk(sk);
4301 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4304 /* Only some options are supported */
4307 *((int *)optval) = (int)inet->tos;
4312 #if IS_ENABLED(CONFIG_IPV6)
4313 } else if (level == SOL_IPV6) {
4314 struct ipv6_pinfo *np = inet6_sk(sk);
4316 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4319 /* Only some options are supported */
4322 *((int *)optval) = (int)np->tclass;
4334 memset(optval, 0, optlen);
4338 static const struct bpf_func_proto bpf_getsockopt_proto = {
4339 .func = bpf_getsockopt,
4341 .ret_type = RET_INTEGER,
4342 .arg1_type = ARG_PTR_TO_CTX,
4343 .arg2_type = ARG_ANYTHING,
4344 .arg3_type = ARG_ANYTHING,
4345 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
4346 .arg5_type = ARG_CONST_SIZE,
4349 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
4352 struct sock *sk = bpf_sock->sk;
4353 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
4355 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
4359 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
4361 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
4364 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
4365 .func = bpf_sock_ops_cb_flags_set,
4367 .ret_type = RET_INTEGER,
4368 .arg1_type = ARG_PTR_TO_CTX,
4369 .arg2_type = ARG_ANYTHING,
4372 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
4373 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
4375 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
4379 struct sock *sk = ctx->sk;
4382 /* Binding to port can be expensive so it's prohibited in the helper.
4383 * Only binding to IP is supported.
4386 if (addr_len < offsetofend(struct sockaddr, sa_family))
4388 if (addr->sa_family == AF_INET) {
4389 if (addr_len < sizeof(struct sockaddr_in))
4391 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
4393 return __inet_bind(sk, addr, addr_len, true, false);
4394 #if IS_ENABLED(CONFIG_IPV6)
4395 } else if (addr->sa_family == AF_INET6) {
4396 if (addr_len < SIN6_LEN_RFC2133)
4398 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
4400 /* ipv6_bpf_stub cannot be NULL, since it's called from
4401 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
4403 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, true, false);
4404 #endif /* CONFIG_IPV6 */
4406 #endif /* CONFIG_INET */
4408 return -EAFNOSUPPORT;
4411 static const struct bpf_func_proto bpf_bind_proto = {
4414 .ret_type = RET_INTEGER,
4415 .arg1_type = ARG_PTR_TO_CTX,
4416 .arg2_type = ARG_PTR_TO_MEM,
4417 .arg3_type = ARG_CONST_SIZE,
4421 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
4422 struct bpf_xfrm_state *, to, u32, size, u64, flags)
4424 const struct sec_path *sp = skb_sec_path(skb);
4425 const struct xfrm_state *x;
4427 if (!sp || unlikely(index >= sp->len || flags))
4430 x = sp->xvec[index];
4432 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
4435 to->reqid = x->props.reqid;
4436 to->spi = x->id.spi;
4437 to->family = x->props.family;
4440 if (to->family == AF_INET6) {
4441 memcpy(to->remote_ipv6, x->props.saddr.a6,
4442 sizeof(to->remote_ipv6));
4444 to->remote_ipv4 = x->props.saddr.a4;
4445 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4450 memset(to, 0, size);
4454 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
4455 .func = bpf_skb_get_xfrm_state,
4457 .ret_type = RET_INTEGER,
4458 .arg1_type = ARG_PTR_TO_CTX,
4459 .arg2_type = ARG_ANYTHING,
4460 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
4461 .arg4_type = ARG_CONST_SIZE,
4462 .arg5_type = ARG_ANYTHING,
4466 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
4467 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
4468 const struct neighbour *neigh,
4469 const struct net_device *dev)
4471 memcpy(params->dmac, neigh->ha, ETH_ALEN);
4472 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
4473 params->h_vlan_TCI = 0;
4474 params->h_vlan_proto = 0;
4475 params->ifindex = dev->ifindex;
4481 #if IS_ENABLED(CONFIG_INET)
4482 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4483 u32 flags, bool check_mtu)
4485 struct in_device *in_dev;
4486 struct neighbour *neigh;
4487 struct net_device *dev;
4488 struct fib_result res;
4494 dev = dev_get_by_index_rcu(net, params->ifindex);
4498 /* verify forwarding is enabled on this interface */
4499 in_dev = __in_dev_get_rcu(dev);
4500 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
4501 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4503 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4505 fl4.flowi4_oif = params->ifindex;
4507 fl4.flowi4_iif = params->ifindex;
4510 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
4511 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
4512 fl4.flowi4_flags = 0;
4514 fl4.flowi4_proto = params->l4_protocol;
4515 fl4.daddr = params->ipv4_dst;
4516 fl4.saddr = params->ipv4_src;
4517 fl4.fl4_sport = params->sport;
4518 fl4.fl4_dport = params->dport;
4520 if (flags & BPF_FIB_LOOKUP_DIRECT) {
4521 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4522 struct fib_table *tb;
4524 tb = fib_get_table(net, tbid);
4526 return BPF_FIB_LKUP_RET_NOT_FWDED;
4528 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
4530 fl4.flowi4_mark = 0;
4531 fl4.flowi4_secid = 0;
4532 fl4.flowi4_tun_key.tun_id = 0;
4533 fl4.flowi4_uid = sock_net_uid(net, NULL);
4535 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
4539 /* map fib lookup errors to RTN_ type */
4541 return BPF_FIB_LKUP_RET_BLACKHOLE;
4542 if (err == -EHOSTUNREACH)
4543 return BPF_FIB_LKUP_RET_UNREACHABLE;
4545 return BPF_FIB_LKUP_RET_PROHIBIT;
4547 return BPF_FIB_LKUP_RET_NOT_FWDED;
4550 if (res.type != RTN_UNICAST)
4551 return BPF_FIB_LKUP_RET_NOT_FWDED;
4553 if (res.fi->fib_nhs > 1)
4554 fib_select_path(net, &res, &fl4, NULL);
4557 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
4558 if (params->tot_len > mtu)
4559 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4562 nh = &res.fi->fib_nh[res.nh_sel];
4564 /* do not handle lwt encaps right now */
4565 if (nh->nh_lwtstate)
4566 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4570 params->ipv4_dst = nh->nh_gw;
4572 params->rt_metric = res.fi->fib_priority;
4574 /* xdp and cls_bpf programs are run in RCU-bh so
4575 * rcu_read_lock_bh is not needed here
4577 neigh = __ipv4_neigh_lookup_noref(dev, (__force u32)params->ipv4_dst);
4579 return BPF_FIB_LKUP_RET_NO_NEIGH;
4581 return bpf_fib_set_fwd_params(params, neigh, dev);
4585 #if IS_ENABLED(CONFIG_IPV6)
4586 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4587 u32 flags, bool check_mtu)
4589 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
4590 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
4591 struct neighbour *neigh;
4592 struct net_device *dev;
4593 struct inet6_dev *idev;
4594 struct fib6_info *f6i;
4600 /* link local addresses are never forwarded */
4601 if (rt6_need_strict(dst) || rt6_need_strict(src))
4602 return BPF_FIB_LKUP_RET_NOT_FWDED;
4604 dev = dev_get_by_index_rcu(net, params->ifindex);
4608 idev = __in6_dev_get_safely(dev);
4609 if (unlikely(!idev || !net->ipv6.devconf_all->forwarding))
4610 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4612 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4614 oif = fl6.flowi6_oif = params->ifindex;
4616 oif = fl6.flowi6_iif = params->ifindex;
4618 strict = RT6_LOOKUP_F_HAS_SADDR;
4620 fl6.flowlabel = params->flowinfo;
4621 fl6.flowi6_scope = 0;
4622 fl6.flowi6_flags = 0;
4625 fl6.flowi6_proto = params->l4_protocol;
4628 fl6.fl6_sport = params->sport;
4629 fl6.fl6_dport = params->dport;
4631 if (flags & BPF_FIB_LOOKUP_DIRECT) {
4632 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4633 struct fib6_table *tb;
4635 tb = ipv6_stub->fib6_get_table(net, tbid);
4637 return BPF_FIB_LKUP_RET_NOT_FWDED;
4639 f6i = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, strict);
4641 fl6.flowi6_mark = 0;
4642 fl6.flowi6_secid = 0;
4643 fl6.flowi6_tun_key.tun_id = 0;
4644 fl6.flowi6_uid = sock_net_uid(net, NULL);
4646 f6i = ipv6_stub->fib6_lookup(net, oif, &fl6, strict);
4649 if (unlikely(IS_ERR_OR_NULL(f6i) || f6i == net->ipv6.fib6_null_entry))
4650 return BPF_FIB_LKUP_RET_NOT_FWDED;
4652 if (unlikely(f6i->fib6_flags & RTF_REJECT)) {
4653 switch (f6i->fib6_type) {
4655 return BPF_FIB_LKUP_RET_BLACKHOLE;
4656 case RTN_UNREACHABLE:
4657 return BPF_FIB_LKUP_RET_UNREACHABLE;
4659 return BPF_FIB_LKUP_RET_PROHIBIT;
4661 return BPF_FIB_LKUP_RET_NOT_FWDED;
4665 if (f6i->fib6_type != RTN_UNICAST)
4666 return BPF_FIB_LKUP_RET_NOT_FWDED;
4668 if (f6i->fib6_nsiblings && fl6.flowi6_oif == 0)
4669 f6i = ipv6_stub->fib6_multipath_select(net, f6i, &fl6,
4670 fl6.flowi6_oif, NULL,
4674 mtu = ipv6_stub->ip6_mtu_from_fib6(f6i, dst, src);
4675 if (params->tot_len > mtu)
4676 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4679 if (f6i->fib6_nh.nh_lwtstate)
4680 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4682 if (f6i->fib6_flags & RTF_GATEWAY)
4683 *dst = f6i->fib6_nh.nh_gw;
4685 dev = f6i->fib6_nh.nh_dev;
4686 params->rt_metric = f6i->fib6_metric;
4688 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
4689 * not needed here. Can not use __ipv6_neigh_lookup_noref here
4690 * because we need to get nd_tbl via the stub
4692 neigh = ___neigh_lookup_noref(ipv6_stub->nd_tbl, neigh_key_eq128,
4693 ndisc_hashfn, dst, dev);
4695 return BPF_FIB_LKUP_RET_NO_NEIGH;
4697 return bpf_fib_set_fwd_params(params, neigh, dev);
4701 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
4702 struct bpf_fib_lookup *, params, int, plen, u32, flags)
4704 if (plen < sizeof(*params))
4707 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4710 switch (params->family) {
4711 #if IS_ENABLED(CONFIG_INET)
4713 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
4716 #if IS_ENABLED(CONFIG_IPV6)
4718 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
4722 return -EAFNOSUPPORT;
4725 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
4726 .func = bpf_xdp_fib_lookup,
4728 .ret_type = RET_INTEGER,
4729 .arg1_type = ARG_PTR_TO_CTX,
4730 .arg2_type = ARG_PTR_TO_MEM,
4731 .arg3_type = ARG_CONST_SIZE,
4732 .arg4_type = ARG_ANYTHING,
4735 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
4736 struct bpf_fib_lookup *, params, int, plen, u32, flags)
4738 struct net *net = dev_net(skb->dev);
4739 int rc = -EAFNOSUPPORT;
4741 if (plen < sizeof(*params))
4744 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4747 switch (params->family) {
4748 #if IS_ENABLED(CONFIG_INET)
4750 rc = bpf_ipv4_fib_lookup(net, params, flags, false);
4753 #if IS_ENABLED(CONFIG_IPV6)
4755 rc = bpf_ipv6_fib_lookup(net, params, flags, false);
4761 struct net_device *dev;
4763 dev = dev_get_by_index_rcu(net, params->ifindex);
4764 if (!is_skb_forwardable(dev, skb))
4765 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
4771 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
4772 .func = bpf_skb_fib_lookup,
4774 .ret_type = RET_INTEGER,
4775 .arg1_type = ARG_PTR_TO_CTX,
4776 .arg2_type = ARG_PTR_TO_MEM,
4777 .arg3_type = ARG_CONST_SIZE,
4778 .arg4_type = ARG_ANYTHING,
4781 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4782 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
4785 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
4787 if (!seg6_validate_srh(srh, len))
4791 case BPF_LWT_ENCAP_SEG6_INLINE:
4792 if (skb->protocol != htons(ETH_P_IPV6))
4795 err = seg6_do_srh_inline(skb, srh);
4797 case BPF_LWT_ENCAP_SEG6:
4798 skb_reset_inner_headers(skb);
4799 skb->encapsulation = 1;
4800 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
4806 bpf_compute_data_pointers(skb);
4810 ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
4811 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
4813 return seg6_lookup_nexthop(skb, NULL, 0);
4815 #endif /* CONFIG_IPV6_SEG6_BPF */
4817 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
4818 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
4821 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
4825 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
4829 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4830 case BPF_LWT_ENCAP_SEG6:
4831 case BPF_LWT_ENCAP_SEG6_INLINE:
4832 return bpf_push_seg6_encap(skb, type, hdr, len);
4834 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
4835 case BPF_LWT_ENCAP_IP:
4836 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
4843 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
4844 void *, hdr, u32, len)
4847 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
4848 case BPF_LWT_ENCAP_IP:
4849 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
4856 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
4857 .func = bpf_lwt_in_push_encap,
4859 .ret_type = RET_INTEGER,
4860 .arg1_type = ARG_PTR_TO_CTX,
4861 .arg2_type = ARG_ANYTHING,
4862 .arg3_type = ARG_PTR_TO_MEM,
4863 .arg4_type = ARG_CONST_SIZE
4866 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
4867 .func = bpf_lwt_xmit_push_encap,
4869 .ret_type = RET_INTEGER,
4870 .arg1_type = ARG_PTR_TO_CTX,
4871 .arg2_type = ARG_ANYTHING,
4872 .arg3_type = ARG_PTR_TO_MEM,
4873 .arg4_type = ARG_CONST_SIZE
4876 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4877 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
4878 const void *, from, u32, len)
4880 struct seg6_bpf_srh_state *srh_state =
4881 this_cpu_ptr(&seg6_bpf_srh_states);
4882 struct ipv6_sr_hdr *srh = srh_state->srh;
4883 void *srh_tlvs, *srh_end, *ptr;
4889 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
4890 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
4892 ptr = skb->data + offset;
4893 if (ptr >= srh_tlvs && ptr + len <= srh_end)
4894 srh_state->valid = false;
4895 else if (ptr < (void *)&srh->flags ||
4896 ptr + len > (void *)&srh->segments)
4899 if (unlikely(bpf_try_make_writable(skb, offset + len)))
4901 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
4903 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4905 memcpy(skb->data + offset, from, len);
4909 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
4910 .func = bpf_lwt_seg6_store_bytes,
4912 .ret_type = RET_INTEGER,
4913 .arg1_type = ARG_PTR_TO_CTX,
4914 .arg2_type = ARG_ANYTHING,
4915 .arg3_type = ARG_PTR_TO_MEM,
4916 .arg4_type = ARG_CONST_SIZE
4919 static void bpf_update_srh_state(struct sk_buff *skb)
4921 struct seg6_bpf_srh_state *srh_state =
4922 this_cpu_ptr(&seg6_bpf_srh_states);
4925 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
4926 srh_state->srh = NULL;
4928 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4929 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
4930 srh_state->valid = true;
4934 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
4935 u32, action, void *, param, u32, param_len)
4937 struct seg6_bpf_srh_state *srh_state =
4938 this_cpu_ptr(&seg6_bpf_srh_states);
4943 case SEG6_LOCAL_ACTION_END_X:
4944 if (!seg6_bpf_has_valid_srh(skb))
4946 if (param_len != sizeof(struct in6_addr))
4948 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
4949 case SEG6_LOCAL_ACTION_END_T:
4950 if (!seg6_bpf_has_valid_srh(skb))
4952 if (param_len != sizeof(int))
4954 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
4955 case SEG6_LOCAL_ACTION_END_DT6:
4956 if (!seg6_bpf_has_valid_srh(skb))
4958 if (param_len != sizeof(int))
4961 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
4963 if (!pskb_pull(skb, hdroff))
4966 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
4967 skb_reset_network_header(skb);
4968 skb_reset_transport_header(skb);
4969 skb->encapsulation = 0;
4971 bpf_compute_data_pointers(skb);
4972 bpf_update_srh_state(skb);
4973 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
4974 case SEG6_LOCAL_ACTION_END_B6:
4975 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
4977 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
4980 bpf_update_srh_state(skb);
4983 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
4984 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
4986 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
4989 bpf_update_srh_state(skb);
4997 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
4998 .func = bpf_lwt_seg6_action,
5000 .ret_type = RET_INTEGER,
5001 .arg1_type = ARG_PTR_TO_CTX,
5002 .arg2_type = ARG_ANYTHING,
5003 .arg3_type = ARG_PTR_TO_MEM,
5004 .arg4_type = ARG_CONST_SIZE
5007 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
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_end, *srh_tlvs, *ptr;
5014 struct ipv6hdr *hdr;
5018 if (unlikely(srh == NULL))
5021 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
5022 ((srh->first_segment + 1) << 4));
5023 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
5025 ptr = skb->data + offset;
5027 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
5029 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
5033 ret = skb_cow_head(skb, len);
5034 if (unlikely(ret < 0))
5037 ret = bpf_skb_net_hdr_push(skb, offset, len);
5039 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
5042 bpf_compute_data_pointers(skb);
5043 if (unlikely(ret < 0))
5046 hdr = (struct ipv6hdr *)skb->data;
5047 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5049 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5051 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5052 srh_state->hdrlen += len;
5053 srh_state->valid = false;
5057 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
5058 .func = bpf_lwt_seg6_adjust_srh,
5060 .ret_type = RET_INTEGER,
5061 .arg1_type = ARG_PTR_TO_CTX,
5062 .arg2_type = ARG_ANYTHING,
5063 .arg3_type = ARG_ANYTHING,
5065 #endif /* CONFIG_IPV6_SEG6_BPF */
5067 #define CONVERT_COMMON_TCP_SOCK_FIELDS(md_type, CONVERT) \
5069 switch (si->off) { \
5070 case offsetof(md_type, snd_cwnd): \
5071 CONVERT(snd_cwnd); break; \
5072 case offsetof(md_type, srtt_us): \
5073 CONVERT(srtt_us); break; \
5074 case offsetof(md_type, snd_ssthresh): \
5075 CONVERT(snd_ssthresh); break; \
5076 case offsetof(md_type, rcv_nxt): \
5077 CONVERT(rcv_nxt); break; \
5078 case offsetof(md_type, snd_nxt): \
5079 CONVERT(snd_nxt); break; \
5080 case offsetof(md_type, snd_una): \
5081 CONVERT(snd_una); break; \
5082 case offsetof(md_type, mss_cache): \
5083 CONVERT(mss_cache); break; \
5084 case offsetof(md_type, ecn_flags): \
5085 CONVERT(ecn_flags); break; \
5086 case offsetof(md_type, rate_delivered): \
5087 CONVERT(rate_delivered); break; \
5088 case offsetof(md_type, rate_interval_us): \
5089 CONVERT(rate_interval_us); break; \
5090 case offsetof(md_type, packets_out): \
5091 CONVERT(packets_out); break; \
5092 case offsetof(md_type, retrans_out): \
5093 CONVERT(retrans_out); break; \
5094 case offsetof(md_type, total_retrans): \
5095 CONVERT(total_retrans); break; \
5096 case offsetof(md_type, segs_in): \
5097 CONVERT(segs_in); break; \
5098 case offsetof(md_type, data_segs_in): \
5099 CONVERT(data_segs_in); break; \
5100 case offsetof(md_type, segs_out): \
5101 CONVERT(segs_out); break; \
5102 case offsetof(md_type, data_segs_out): \
5103 CONVERT(data_segs_out); break; \
5104 case offsetof(md_type, lost_out): \
5105 CONVERT(lost_out); break; \
5106 case offsetof(md_type, sacked_out): \
5107 CONVERT(sacked_out); break; \
5108 case offsetof(md_type, bytes_received): \
5109 CONVERT(bytes_received); break; \
5110 case offsetof(md_type, bytes_acked): \
5111 CONVERT(bytes_acked); break; \
5116 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
5117 int dif, int sdif, u8 family, u8 proto)
5119 bool refcounted = false;
5120 struct sock *sk = NULL;
5122 if (family == AF_INET) {
5123 __be32 src4 = tuple->ipv4.saddr;
5124 __be32 dst4 = tuple->ipv4.daddr;
5126 if (proto == IPPROTO_TCP)
5127 sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
5128 src4, tuple->ipv4.sport,
5129 dst4, tuple->ipv4.dport,
5130 dif, sdif, &refcounted);
5132 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
5133 dst4, tuple->ipv4.dport,
5134 dif, sdif, &udp_table, NULL);
5135 #if IS_ENABLED(CONFIG_IPV6)
5137 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
5138 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
5140 if (proto == IPPROTO_TCP)
5141 sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
5142 src6, tuple->ipv6.sport,
5143 dst6, ntohs(tuple->ipv6.dport),
5144 dif, sdif, &refcounted);
5145 else if (likely(ipv6_bpf_stub))
5146 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
5147 src6, tuple->ipv6.sport,
5148 dst6, tuple->ipv6.dport,
5154 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
5155 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
5161 /* bpf_sk_lookup performs the core lookup for different types of sockets,
5162 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
5163 * Returns the socket as an 'unsigned long' to simplify the casting in the
5164 * callers to satisfy BPF_CALL declarations.
5166 static unsigned long
5167 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5168 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
5171 struct sock *sk = NULL;
5172 u8 family = AF_UNSPEC;
5176 family = len == sizeof(tuple->ipv4) ? AF_INET : AF_INET6;
5177 if (unlikely(family == AF_UNSPEC || flags ||
5178 !((s32)netns_id < 0 || netns_id <= S32_MAX)))
5181 if (family == AF_INET)
5182 sdif = inet_sdif(skb);
5184 sdif = inet6_sdif(skb);
5186 if ((s32)netns_id < 0) {
5188 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
5190 net = get_net_ns_by_id(caller_net, netns_id);
5193 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
5198 sk = sk_to_full_sk(sk);
5200 return (unsigned long) sk;
5203 static unsigned long
5204 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5205 u8 proto, u64 netns_id, u64 flags)
5207 struct net *caller_net;
5211 caller_net = dev_net(skb->dev);
5212 ifindex = skb->dev->ifindex;
5214 caller_net = sock_net(skb->sk);
5218 return __bpf_sk_lookup(skb, tuple, len, caller_net, ifindex,
5219 proto, netns_id, flags);
5222 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
5223 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5225 return bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP, netns_id, flags);
5228 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
5229 .func = bpf_sk_lookup_tcp,
5232 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5233 .arg1_type = ARG_PTR_TO_CTX,
5234 .arg2_type = ARG_PTR_TO_MEM,
5235 .arg3_type = ARG_CONST_SIZE,
5236 .arg4_type = ARG_ANYTHING,
5237 .arg5_type = ARG_ANYTHING,
5240 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
5241 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5243 return bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP, netns_id, flags);
5246 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
5247 .func = bpf_sk_lookup_udp,
5250 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5251 .arg1_type = ARG_PTR_TO_CTX,
5252 .arg2_type = ARG_PTR_TO_MEM,
5253 .arg3_type = ARG_CONST_SIZE,
5254 .arg4_type = ARG_ANYTHING,
5255 .arg5_type = ARG_ANYTHING,
5258 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
5260 if (!sock_flag(sk, SOCK_RCU_FREE))
5265 static const struct bpf_func_proto bpf_sk_release_proto = {
5266 .func = bpf_sk_release,
5268 .ret_type = RET_INTEGER,
5269 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
5272 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
5273 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
5275 struct net *caller_net = dev_net(ctx->rxq->dev);
5276 int ifindex = ctx->rxq->dev->ifindex;
5278 return __bpf_sk_lookup(NULL, tuple, len, caller_net, ifindex,
5279 IPPROTO_UDP, netns_id, flags);
5282 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
5283 .func = bpf_xdp_sk_lookup_udp,
5286 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5287 .arg1_type = ARG_PTR_TO_CTX,
5288 .arg2_type = ARG_PTR_TO_MEM,
5289 .arg3_type = ARG_CONST_SIZE,
5290 .arg4_type = ARG_ANYTHING,
5291 .arg5_type = ARG_ANYTHING,
5294 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
5295 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
5297 struct net *caller_net = dev_net(ctx->rxq->dev);
5298 int ifindex = ctx->rxq->dev->ifindex;
5300 return __bpf_sk_lookup(NULL, tuple, len, caller_net, ifindex,
5301 IPPROTO_TCP, netns_id, flags);
5304 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
5305 .func = bpf_xdp_sk_lookup_tcp,
5308 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5309 .arg1_type = ARG_PTR_TO_CTX,
5310 .arg2_type = ARG_PTR_TO_MEM,
5311 .arg3_type = ARG_CONST_SIZE,
5312 .arg4_type = ARG_ANYTHING,
5313 .arg5_type = ARG_ANYTHING,
5316 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
5317 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5319 return __bpf_sk_lookup(NULL, tuple, len, sock_net(ctx->sk), 0,
5320 IPPROTO_TCP, netns_id, flags);
5323 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
5324 .func = bpf_sock_addr_sk_lookup_tcp,
5326 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5327 .arg1_type = ARG_PTR_TO_CTX,
5328 .arg2_type = ARG_PTR_TO_MEM,
5329 .arg3_type = ARG_CONST_SIZE,
5330 .arg4_type = ARG_ANYTHING,
5331 .arg5_type = ARG_ANYTHING,
5334 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
5335 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5337 return __bpf_sk_lookup(NULL, tuple, len, sock_net(ctx->sk), 0,
5338 IPPROTO_UDP, netns_id, flags);
5341 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
5342 .func = bpf_sock_addr_sk_lookup_udp,
5344 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5345 .arg1_type = ARG_PTR_TO_CTX,
5346 .arg2_type = ARG_PTR_TO_MEM,
5347 .arg3_type = ARG_CONST_SIZE,
5348 .arg4_type = ARG_ANYTHING,
5349 .arg5_type = ARG_ANYTHING,
5352 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
5353 struct bpf_insn_access_aux *info)
5355 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock, bytes_acked))
5358 if (off % size != 0)
5362 case offsetof(struct bpf_tcp_sock, bytes_received):
5363 case offsetof(struct bpf_tcp_sock, bytes_acked):
5364 return size == sizeof(__u64);
5366 return size == sizeof(__u32);
5370 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
5371 const struct bpf_insn *si,
5372 struct bpf_insn *insn_buf,
5373 struct bpf_prog *prog, u32 *target_size)
5375 struct bpf_insn *insn = insn_buf;
5377 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \
5379 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, FIELD) > \
5380 FIELD_SIZEOF(struct bpf_tcp_sock, FIELD)); \
5381 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
5382 si->dst_reg, si->src_reg, \
5383 offsetof(struct tcp_sock, FIELD)); \
5386 CONVERT_COMMON_TCP_SOCK_FIELDS(struct bpf_tcp_sock,
5387 BPF_TCP_SOCK_GET_COMMON);
5389 if (insn > insn_buf)
5390 return insn - insn_buf;
5393 case offsetof(struct bpf_tcp_sock, rtt_min):
5394 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, rtt_min) !=
5395 sizeof(struct minmax));
5396 BUILD_BUG_ON(sizeof(struct minmax) <
5397 sizeof(struct minmax_sample));
5399 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5400 offsetof(struct tcp_sock, rtt_min) +
5401 offsetof(struct minmax_sample, v));
5405 return insn - insn_buf;
5408 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
5410 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
5411 return (unsigned long)sk;
5413 return (unsigned long)NULL;
5416 static const struct bpf_func_proto bpf_tcp_sock_proto = {
5417 .func = bpf_tcp_sock,
5419 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
5420 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
5423 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
5425 sk = sk_to_full_sk(sk);
5427 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
5428 return (unsigned long)sk;
5430 return (unsigned long)NULL;
5433 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
5434 .func = bpf_get_listener_sock,
5436 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
5437 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
5440 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
5442 unsigned int iphdr_len;
5444 if (skb->protocol == cpu_to_be16(ETH_P_IP))
5445 iphdr_len = sizeof(struct iphdr);
5446 else if (skb->protocol == cpu_to_be16(ETH_P_IPV6))
5447 iphdr_len = sizeof(struct ipv6hdr);
5451 if (skb_headlen(skb) < iphdr_len)
5454 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
5457 return INET_ECN_set_ce(skb);
5460 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
5461 .func = bpf_skb_ecn_set_ce,
5463 .ret_type = RET_INTEGER,
5464 .arg1_type = ARG_PTR_TO_CTX,
5466 #endif /* CONFIG_INET */
5468 bool bpf_helper_changes_pkt_data(void *func)
5470 if (func == bpf_skb_vlan_push ||
5471 func == bpf_skb_vlan_pop ||
5472 func == bpf_skb_store_bytes ||
5473 func == bpf_skb_change_proto ||
5474 func == bpf_skb_change_head ||
5475 func == sk_skb_change_head ||
5476 func == bpf_skb_change_tail ||
5477 func == sk_skb_change_tail ||
5478 func == bpf_skb_adjust_room ||
5479 func == bpf_skb_pull_data ||
5480 func == sk_skb_pull_data ||
5481 func == bpf_clone_redirect ||
5482 func == bpf_l3_csum_replace ||
5483 func == bpf_l4_csum_replace ||
5484 func == bpf_xdp_adjust_head ||
5485 func == bpf_xdp_adjust_meta ||
5486 func == bpf_msg_pull_data ||
5487 func == bpf_msg_push_data ||
5488 func == bpf_msg_pop_data ||
5489 func == bpf_xdp_adjust_tail ||
5490 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5491 func == bpf_lwt_seg6_store_bytes ||
5492 func == bpf_lwt_seg6_adjust_srh ||
5493 func == bpf_lwt_seg6_action ||
5495 func == bpf_lwt_in_push_encap ||
5496 func == bpf_lwt_xmit_push_encap)
5502 static const struct bpf_func_proto *
5503 bpf_base_func_proto(enum bpf_func_id func_id)
5506 case BPF_FUNC_map_lookup_elem:
5507 return &bpf_map_lookup_elem_proto;
5508 case BPF_FUNC_map_update_elem:
5509 return &bpf_map_update_elem_proto;
5510 case BPF_FUNC_map_delete_elem:
5511 return &bpf_map_delete_elem_proto;
5512 case BPF_FUNC_map_push_elem:
5513 return &bpf_map_push_elem_proto;
5514 case BPF_FUNC_map_pop_elem:
5515 return &bpf_map_pop_elem_proto;
5516 case BPF_FUNC_map_peek_elem:
5517 return &bpf_map_peek_elem_proto;
5518 case BPF_FUNC_get_prandom_u32:
5519 return &bpf_get_prandom_u32_proto;
5520 case BPF_FUNC_get_smp_processor_id:
5521 return &bpf_get_raw_smp_processor_id_proto;
5522 case BPF_FUNC_get_numa_node_id:
5523 return &bpf_get_numa_node_id_proto;
5524 case BPF_FUNC_tail_call:
5525 return &bpf_tail_call_proto;
5526 case BPF_FUNC_ktime_get_ns:
5527 return &bpf_ktime_get_ns_proto;
5532 if (!capable(CAP_SYS_ADMIN))
5536 case BPF_FUNC_spin_lock:
5537 return &bpf_spin_lock_proto;
5538 case BPF_FUNC_spin_unlock:
5539 return &bpf_spin_unlock_proto;
5540 case BPF_FUNC_trace_printk:
5541 return bpf_get_trace_printk_proto();
5547 static const struct bpf_func_proto *
5548 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5551 /* inet and inet6 sockets are created in a process
5552 * context so there is always a valid uid/gid
5554 case BPF_FUNC_get_current_uid_gid:
5555 return &bpf_get_current_uid_gid_proto;
5556 case BPF_FUNC_get_local_storage:
5557 return &bpf_get_local_storage_proto;
5559 return bpf_base_func_proto(func_id);
5563 static const struct bpf_func_proto *
5564 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5567 /* inet and inet6 sockets are created in a process
5568 * context so there is always a valid uid/gid
5570 case BPF_FUNC_get_current_uid_gid:
5571 return &bpf_get_current_uid_gid_proto;
5573 switch (prog->expected_attach_type) {
5574 case BPF_CGROUP_INET4_CONNECT:
5575 case BPF_CGROUP_INET6_CONNECT:
5576 return &bpf_bind_proto;
5580 case BPF_FUNC_get_socket_cookie:
5581 return &bpf_get_socket_cookie_sock_addr_proto;
5582 case BPF_FUNC_get_local_storage:
5583 return &bpf_get_local_storage_proto;
5585 case BPF_FUNC_sk_lookup_tcp:
5586 return &bpf_sock_addr_sk_lookup_tcp_proto;
5587 case BPF_FUNC_sk_lookup_udp:
5588 return &bpf_sock_addr_sk_lookup_udp_proto;
5589 case BPF_FUNC_sk_release:
5590 return &bpf_sk_release_proto;
5591 #endif /* CONFIG_INET */
5593 return bpf_base_func_proto(func_id);
5597 static const struct bpf_func_proto *
5598 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5601 case BPF_FUNC_skb_load_bytes:
5602 return &bpf_skb_load_bytes_proto;
5603 case BPF_FUNC_skb_load_bytes_relative:
5604 return &bpf_skb_load_bytes_relative_proto;
5605 case BPF_FUNC_get_socket_cookie:
5606 return &bpf_get_socket_cookie_proto;
5607 case BPF_FUNC_get_socket_uid:
5608 return &bpf_get_socket_uid_proto;
5610 return bpf_base_func_proto(func_id);
5614 static const struct bpf_func_proto *
5615 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5618 case BPF_FUNC_get_local_storage:
5619 return &bpf_get_local_storage_proto;
5620 case BPF_FUNC_sk_fullsock:
5621 return &bpf_sk_fullsock_proto;
5623 case BPF_FUNC_tcp_sock:
5624 return &bpf_tcp_sock_proto;
5625 case BPF_FUNC_get_listener_sock:
5626 return &bpf_get_listener_sock_proto;
5627 case BPF_FUNC_skb_ecn_set_ce:
5628 return &bpf_skb_ecn_set_ce_proto;
5631 return sk_filter_func_proto(func_id, prog);
5635 static const struct bpf_func_proto *
5636 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5639 case BPF_FUNC_skb_store_bytes:
5640 return &bpf_skb_store_bytes_proto;
5641 case BPF_FUNC_skb_load_bytes:
5642 return &bpf_skb_load_bytes_proto;
5643 case BPF_FUNC_skb_load_bytes_relative:
5644 return &bpf_skb_load_bytes_relative_proto;
5645 case BPF_FUNC_skb_pull_data:
5646 return &bpf_skb_pull_data_proto;
5647 case BPF_FUNC_csum_diff:
5648 return &bpf_csum_diff_proto;
5649 case BPF_FUNC_csum_update:
5650 return &bpf_csum_update_proto;
5651 case BPF_FUNC_l3_csum_replace:
5652 return &bpf_l3_csum_replace_proto;
5653 case BPF_FUNC_l4_csum_replace:
5654 return &bpf_l4_csum_replace_proto;
5655 case BPF_FUNC_clone_redirect:
5656 return &bpf_clone_redirect_proto;
5657 case BPF_FUNC_get_cgroup_classid:
5658 return &bpf_get_cgroup_classid_proto;
5659 case BPF_FUNC_skb_vlan_push:
5660 return &bpf_skb_vlan_push_proto;
5661 case BPF_FUNC_skb_vlan_pop:
5662 return &bpf_skb_vlan_pop_proto;
5663 case BPF_FUNC_skb_change_proto:
5664 return &bpf_skb_change_proto_proto;
5665 case BPF_FUNC_skb_change_type:
5666 return &bpf_skb_change_type_proto;
5667 case BPF_FUNC_skb_adjust_room:
5668 return &bpf_skb_adjust_room_proto;
5669 case BPF_FUNC_skb_change_tail:
5670 return &bpf_skb_change_tail_proto;
5671 case BPF_FUNC_skb_get_tunnel_key:
5672 return &bpf_skb_get_tunnel_key_proto;
5673 case BPF_FUNC_skb_set_tunnel_key:
5674 return bpf_get_skb_set_tunnel_proto(func_id);
5675 case BPF_FUNC_skb_get_tunnel_opt:
5676 return &bpf_skb_get_tunnel_opt_proto;
5677 case BPF_FUNC_skb_set_tunnel_opt:
5678 return bpf_get_skb_set_tunnel_proto(func_id);
5679 case BPF_FUNC_redirect:
5680 return &bpf_redirect_proto;
5681 case BPF_FUNC_get_route_realm:
5682 return &bpf_get_route_realm_proto;
5683 case BPF_FUNC_get_hash_recalc:
5684 return &bpf_get_hash_recalc_proto;
5685 case BPF_FUNC_set_hash_invalid:
5686 return &bpf_set_hash_invalid_proto;
5687 case BPF_FUNC_set_hash:
5688 return &bpf_set_hash_proto;
5689 case BPF_FUNC_perf_event_output:
5690 return &bpf_skb_event_output_proto;
5691 case BPF_FUNC_get_smp_processor_id:
5692 return &bpf_get_smp_processor_id_proto;
5693 case BPF_FUNC_skb_under_cgroup:
5694 return &bpf_skb_under_cgroup_proto;
5695 case BPF_FUNC_get_socket_cookie:
5696 return &bpf_get_socket_cookie_proto;
5697 case BPF_FUNC_get_socket_uid:
5698 return &bpf_get_socket_uid_proto;
5699 case BPF_FUNC_fib_lookup:
5700 return &bpf_skb_fib_lookup_proto;
5701 case BPF_FUNC_sk_fullsock:
5702 return &bpf_sk_fullsock_proto;
5704 case BPF_FUNC_skb_get_xfrm_state:
5705 return &bpf_skb_get_xfrm_state_proto;
5707 #ifdef CONFIG_SOCK_CGROUP_DATA
5708 case BPF_FUNC_skb_cgroup_id:
5709 return &bpf_skb_cgroup_id_proto;
5710 case BPF_FUNC_skb_ancestor_cgroup_id:
5711 return &bpf_skb_ancestor_cgroup_id_proto;
5714 case BPF_FUNC_sk_lookup_tcp:
5715 return &bpf_sk_lookup_tcp_proto;
5716 case BPF_FUNC_sk_lookup_udp:
5717 return &bpf_sk_lookup_udp_proto;
5718 case BPF_FUNC_sk_release:
5719 return &bpf_sk_release_proto;
5720 case BPF_FUNC_tcp_sock:
5721 return &bpf_tcp_sock_proto;
5722 case BPF_FUNC_get_listener_sock:
5723 return &bpf_get_listener_sock_proto;
5726 return bpf_base_func_proto(func_id);
5730 static const struct bpf_func_proto *
5731 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5734 case BPF_FUNC_perf_event_output:
5735 return &bpf_xdp_event_output_proto;
5736 case BPF_FUNC_get_smp_processor_id:
5737 return &bpf_get_smp_processor_id_proto;
5738 case BPF_FUNC_csum_diff:
5739 return &bpf_csum_diff_proto;
5740 case BPF_FUNC_xdp_adjust_head:
5741 return &bpf_xdp_adjust_head_proto;
5742 case BPF_FUNC_xdp_adjust_meta:
5743 return &bpf_xdp_adjust_meta_proto;
5744 case BPF_FUNC_redirect:
5745 return &bpf_xdp_redirect_proto;
5746 case BPF_FUNC_redirect_map:
5747 return &bpf_xdp_redirect_map_proto;
5748 case BPF_FUNC_xdp_adjust_tail:
5749 return &bpf_xdp_adjust_tail_proto;
5750 case BPF_FUNC_fib_lookup:
5751 return &bpf_xdp_fib_lookup_proto;
5753 case BPF_FUNC_sk_lookup_udp:
5754 return &bpf_xdp_sk_lookup_udp_proto;
5755 case BPF_FUNC_sk_lookup_tcp:
5756 return &bpf_xdp_sk_lookup_tcp_proto;
5757 case BPF_FUNC_sk_release:
5758 return &bpf_sk_release_proto;
5761 return bpf_base_func_proto(func_id);
5765 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
5766 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
5768 static const struct bpf_func_proto *
5769 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5772 case BPF_FUNC_setsockopt:
5773 return &bpf_setsockopt_proto;
5774 case BPF_FUNC_getsockopt:
5775 return &bpf_getsockopt_proto;
5776 case BPF_FUNC_sock_ops_cb_flags_set:
5777 return &bpf_sock_ops_cb_flags_set_proto;
5778 case BPF_FUNC_sock_map_update:
5779 return &bpf_sock_map_update_proto;
5780 case BPF_FUNC_sock_hash_update:
5781 return &bpf_sock_hash_update_proto;
5782 case BPF_FUNC_get_socket_cookie:
5783 return &bpf_get_socket_cookie_sock_ops_proto;
5784 case BPF_FUNC_get_local_storage:
5785 return &bpf_get_local_storage_proto;
5786 case BPF_FUNC_perf_event_output:
5787 return &bpf_sockopt_event_output_proto;
5789 return bpf_base_func_proto(func_id);
5793 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
5794 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
5796 static const struct bpf_func_proto *
5797 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5800 case BPF_FUNC_msg_redirect_map:
5801 return &bpf_msg_redirect_map_proto;
5802 case BPF_FUNC_msg_redirect_hash:
5803 return &bpf_msg_redirect_hash_proto;
5804 case BPF_FUNC_msg_apply_bytes:
5805 return &bpf_msg_apply_bytes_proto;
5806 case BPF_FUNC_msg_cork_bytes:
5807 return &bpf_msg_cork_bytes_proto;
5808 case BPF_FUNC_msg_pull_data:
5809 return &bpf_msg_pull_data_proto;
5810 case BPF_FUNC_msg_push_data:
5811 return &bpf_msg_push_data_proto;
5812 case BPF_FUNC_msg_pop_data:
5813 return &bpf_msg_pop_data_proto;
5815 return bpf_base_func_proto(func_id);
5819 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
5820 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
5822 static const struct bpf_func_proto *
5823 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5826 case BPF_FUNC_skb_store_bytes:
5827 return &bpf_skb_store_bytes_proto;
5828 case BPF_FUNC_skb_load_bytes:
5829 return &bpf_skb_load_bytes_proto;
5830 case BPF_FUNC_skb_pull_data:
5831 return &sk_skb_pull_data_proto;
5832 case BPF_FUNC_skb_change_tail:
5833 return &sk_skb_change_tail_proto;
5834 case BPF_FUNC_skb_change_head:
5835 return &sk_skb_change_head_proto;
5836 case BPF_FUNC_get_socket_cookie:
5837 return &bpf_get_socket_cookie_proto;
5838 case BPF_FUNC_get_socket_uid:
5839 return &bpf_get_socket_uid_proto;
5840 case BPF_FUNC_sk_redirect_map:
5841 return &bpf_sk_redirect_map_proto;
5842 case BPF_FUNC_sk_redirect_hash:
5843 return &bpf_sk_redirect_hash_proto;
5845 case BPF_FUNC_sk_lookup_tcp:
5846 return &bpf_sk_lookup_tcp_proto;
5847 case BPF_FUNC_sk_lookup_udp:
5848 return &bpf_sk_lookup_udp_proto;
5849 case BPF_FUNC_sk_release:
5850 return &bpf_sk_release_proto;
5853 return bpf_base_func_proto(func_id);
5857 static const struct bpf_func_proto *
5858 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5861 case BPF_FUNC_skb_load_bytes:
5862 return &bpf_skb_load_bytes_proto;
5864 return bpf_base_func_proto(func_id);
5868 static const struct bpf_func_proto *
5869 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5872 case BPF_FUNC_skb_load_bytes:
5873 return &bpf_skb_load_bytes_proto;
5874 case BPF_FUNC_skb_pull_data:
5875 return &bpf_skb_pull_data_proto;
5876 case BPF_FUNC_csum_diff:
5877 return &bpf_csum_diff_proto;
5878 case BPF_FUNC_get_cgroup_classid:
5879 return &bpf_get_cgroup_classid_proto;
5880 case BPF_FUNC_get_route_realm:
5881 return &bpf_get_route_realm_proto;
5882 case BPF_FUNC_get_hash_recalc:
5883 return &bpf_get_hash_recalc_proto;
5884 case BPF_FUNC_perf_event_output:
5885 return &bpf_skb_event_output_proto;
5886 case BPF_FUNC_get_smp_processor_id:
5887 return &bpf_get_smp_processor_id_proto;
5888 case BPF_FUNC_skb_under_cgroup:
5889 return &bpf_skb_under_cgroup_proto;
5891 return bpf_base_func_proto(func_id);
5895 static const struct bpf_func_proto *
5896 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5899 case BPF_FUNC_lwt_push_encap:
5900 return &bpf_lwt_in_push_encap_proto;
5902 return lwt_out_func_proto(func_id, prog);
5906 static const struct bpf_func_proto *
5907 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5910 case BPF_FUNC_skb_get_tunnel_key:
5911 return &bpf_skb_get_tunnel_key_proto;
5912 case BPF_FUNC_skb_set_tunnel_key:
5913 return bpf_get_skb_set_tunnel_proto(func_id);
5914 case BPF_FUNC_skb_get_tunnel_opt:
5915 return &bpf_skb_get_tunnel_opt_proto;
5916 case BPF_FUNC_skb_set_tunnel_opt:
5917 return bpf_get_skb_set_tunnel_proto(func_id);
5918 case BPF_FUNC_redirect:
5919 return &bpf_redirect_proto;
5920 case BPF_FUNC_clone_redirect:
5921 return &bpf_clone_redirect_proto;
5922 case BPF_FUNC_skb_change_tail:
5923 return &bpf_skb_change_tail_proto;
5924 case BPF_FUNC_skb_change_head:
5925 return &bpf_skb_change_head_proto;
5926 case BPF_FUNC_skb_store_bytes:
5927 return &bpf_skb_store_bytes_proto;
5928 case BPF_FUNC_csum_update:
5929 return &bpf_csum_update_proto;
5930 case BPF_FUNC_l3_csum_replace:
5931 return &bpf_l3_csum_replace_proto;
5932 case BPF_FUNC_l4_csum_replace:
5933 return &bpf_l4_csum_replace_proto;
5934 case BPF_FUNC_set_hash_invalid:
5935 return &bpf_set_hash_invalid_proto;
5936 case BPF_FUNC_lwt_push_encap:
5937 return &bpf_lwt_xmit_push_encap_proto;
5939 return lwt_out_func_proto(func_id, prog);
5943 static const struct bpf_func_proto *
5944 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5947 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5948 case BPF_FUNC_lwt_seg6_store_bytes:
5949 return &bpf_lwt_seg6_store_bytes_proto;
5950 case BPF_FUNC_lwt_seg6_action:
5951 return &bpf_lwt_seg6_action_proto;
5952 case BPF_FUNC_lwt_seg6_adjust_srh:
5953 return &bpf_lwt_seg6_adjust_srh_proto;
5956 return lwt_out_func_proto(func_id, prog);
5960 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
5961 const struct bpf_prog *prog,
5962 struct bpf_insn_access_aux *info)
5964 const int size_default = sizeof(__u32);
5966 if (off < 0 || off >= sizeof(struct __sk_buff))
5969 /* The verifier guarantees that size > 0. */
5970 if (off % size != 0)
5974 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5975 if (off + size > offsetofend(struct __sk_buff, cb[4]))
5978 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
5979 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
5980 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
5981 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
5982 case bpf_ctx_range(struct __sk_buff, data):
5983 case bpf_ctx_range(struct __sk_buff, data_meta):
5984 case bpf_ctx_range(struct __sk_buff, data_end):
5985 if (size != size_default)
5988 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
5989 if (size != sizeof(__u64))
5992 case bpf_ctx_range(struct __sk_buff, tstamp):
5993 if (size != sizeof(__u64))
5996 case offsetof(struct __sk_buff, sk):
5997 if (type == BPF_WRITE || size != sizeof(__u64))
5999 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
6002 /* Only narrow read access allowed for now. */
6003 if (type == BPF_WRITE) {
6004 if (size != size_default)
6007 bpf_ctx_record_field_size(info, size_default);
6008 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
6016 static bool sk_filter_is_valid_access(int off, int size,
6017 enum bpf_access_type type,
6018 const struct bpf_prog *prog,
6019 struct bpf_insn_access_aux *info)
6022 case bpf_ctx_range(struct __sk_buff, tc_classid):
6023 case bpf_ctx_range(struct __sk_buff, data):
6024 case bpf_ctx_range(struct __sk_buff, data_meta):
6025 case bpf_ctx_range(struct __sk_buff, data_end):
6026 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
6027 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6028 case bpf_ctx_range(struct __sk_buff, tstamp):
6029 case bpf_ctx_range(struct __sk_buff, wire_len):
6033 if (type == BPF_WRITE) {
6035 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6042 return bpf_skb_is_valid_access(off, size, type, prog, info);
6045 static bool cg_skb_is_valid_access(int off, int size,
6046 enum bpf_access_type type,
6047 const struct bpf_prog *prog,
6048 struct bpf_insn_access_aux *info)
6051 case bpf_ctx_range(struct __sk_buff, tc_classid):
6052 case bpf_ctx_range(struct __sk_buff, data_meta):
6053 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
6054 case bpf_ctx_range(struct __sk_buff, wire_len):
6056 case bpf_ctx_range(struct __sk_buff, data):
6057 case bpf_ctx_range(struct __sk_buff, data_end):
6058 if (!capable(CAP_SYS_ADMIN))
6063 if (type == BPF_WRITE) {
6065 case bpf_ctx_range(struct __sk_buff, mark):
6066 case bpf_ctx_range(struct __sk_buff, priority):
6067 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6069 case bpf_ctx_range(struct __sk_buff, tstamp):
6070 if (!capable(CAP_SYS_ADMIN))
6079 case bpf_ctx_range(struct __sk_buff, data):
6080 info->reg_type = PTR_TO_PACKET;
6082 case bpf_ctx_range(struct __sk_buff, data_end):
6083 info->reg_type = PTR_TO_PACKET_END;
6087 return bpf_skb_is_valid_access(off, size, type, prog, info);
6090 static bool lwt_is_valid_access(int off, int size,
6091 enum bpf_access_type type,
6092 const struct bpf_prog *prog,
6093 struct bpf_insn_access_aux *info)
6096 case bpf_ctx_range(struct __sk_buff, tc_classid):
6097 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6098 case bpf_ctx_range(struct __sk_buff, data_meta):
6099 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
6100 case bpf_ctx_range(struct __sk_buff, tstamp):
6101 case bpf_ctx_range(struct __sk_buff, wire_len):
6105 if (type == BPF_WRITE) {
6107 case bpf_ctx_range(struct __sk_buff, mark):
6108 case bpf_ctx_range(struct __sk_buff, priority):
6109 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6117 case bpf_ctx_range(struct __sk_buff, data):
6118 info->reg_type = PTR_TO_PACKET;
6120 case bpf_ctx_range(struct __sk_buff, data_end):
6121 info->reg_type = PTR_TO_PACKET_END;
6125 return bpf_skb_is_valid_access(off, size, type, prog, info);
6128 /* Attach type specific accesses */
6129 static bool __sock_filter_check_attach_type(int off,
6130 enum bpf_access_type access_type,
6131 enum bpf_attach_type attach_type)
6134 case offsetof(struct bpf_sock, bound_dev_if):
6135 case offsetof(struct bpf_sock, mark):
6136 case offsetof(struct bpf_sock, priority):
6137 switch (attach_type) {
6138 case BPF_CGROUP_INET_SOCK_CREATE:
6143 case bpf_ctx_range(struct bpf_sock, src_ip4):
6144 switch (attach_type) {
6145 case BPF_CGROUP_INET4_POST_BIND:
6150 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
6151 switch (attach_type) {
6152 case BPF_CGROUP_INET6_POST_BIND:
6157 case bpf_ctx_range(struct bpf_sock, src_port):
6158 switch (attach_type) {
6159 case BPF_CGROUP_INET4_POST_BIND:
6160 case BPF_CGROUP_INET6_POST_BIND:
6167 return access_type == BPF_READ;
6172 bool bpf_sock_common_is_valid_access(int off, int size,
6173 enum bpf_access_type type,
6174 struct bpf_insn_access_aux *info)
6177 case bpf_ctx_range_till(struct bpf_sock, type, priority):
6180 return bpf_sock_is_valid_access(off, size, type, info);
6184 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6185 struct bpf_insn_access_aux *info)
6187 const int size_default = sizeof(__u32);
6189 if (off < 0 || off >= sizeof(struct bpf_sock))
6191 if (off % size != 0)
6195 case offsetof(struct bpf_sock, state):
6196 case offsetof(struct bpf_sock, family):
6197 case offsetof(struct bpf_sock, type):
6198 case offsetof(struct bpf_sock, protocol):
6199 case offsetof(struct bpf_sock, dst_port):
6200 case offsetof(struct bpf_sock, src_port):
6201 case bpf_ctx_range(struct bpf_sock, src_ip4):
6202 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
6203 case bpf_ctx_range(struct bpf_sock, dst_ip4):
6204 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
6205 bpf_ctx_record_field_size(info, size_default);
6206 return bpf_ctx_narrow_access_ok(off, size, size_default);
6209 return size == size_default;
6212 static bool sock_filter_is_valid_access(int off, int size,
6213 enum bpf_access_type type,
6214 const struct bpf_prog *prog,
6215 struct bpf_insn_access_aux *info)
6217 if (!bpf_sock_is_valid_access(off, size, type, info))
6219 return __sock_filter_check_attach_type(off, type,
6220 prog->expected_attach_type);
6223 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
6224 const struct bpf_prog *prog)
6226 /* Neither direct read nor direct write requires any preliminary
6232 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
6233 const struct bpf_prog *prog, int drop_verdict)
6235 struct bpf_insn *insn = insn_buf;
6240 /* if (!skb->cloned)
6243 * (Fast-path, otherwise approximation that we might be
6244 * a clone, do the rest in helper.)
6246 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
6247 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
6248 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
6250 /* ret = bpf_skb_pull_data(skb, 0); */
6251 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
6252 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
6253 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
6254 BPF_FUNC_skb_pull_data);
6257 * return TC_ACT_SHOT;
6259 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
6260 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
6261 *insn++ = BPF_EXIT_INSN();
6264 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
6266 *insn++ = prog->insnsi[0];
6268 return insn - insn_buf;
6271 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
6272 struct bpf_insn *insn_buf)
6274 bool indirect = BPF_MODE(orig->code) == BPF_IND;
6275 struct bpf_insn *insn = insn_buf;
6277 /* We're guaranteed here that CTX is in R6. */
6278 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
6280 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
6282 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
6284 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
6287 switch (BPF_SIZE(orig->code)) {
6289 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
6292 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
6295 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
6299 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
6300 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
6301 *insn++ = BPF_EXIT_INSN();
6303 return insn - insn_buf;
6306 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
6307 const struct bpf_prog *prog)
6309 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
6312 static bool tc_cls_act_is_valid_access(int off, int size,
6313 enum bpf_access_type type,
6314 const struct bpf_prog *prog,
6315 struct bpf_insn_access_aux *info)
6317 if (type == BPF_WRITE) {
6319 case bpf_ctx_range(struct __sk_buff, mark):
6320 case bpf_ctx_range(struct __sk_buff, tc_index):
6321 case bpf_ctx_range(struct __sk_buff, priority):
6322 case bpf_ctx_range(struct __sk_buff, tc_classid):
6323 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6324 case bpf_ctx_range(struct __sk_buff, tstamp):
6325 case bpf_ctx_range(struct __sk_buff, queue_mapping):
6333 case bpf_ctx_range(struct __sk_buff, data):
6334 info->reg_type = PTR_TO_PACKET;
6336 case bpf_ctx_range(struct __sk_buff, data_meta):
6337 info->reg_type = PTR_TO_PACKET_META;
6339 case bpf_ctx_range(struct __sk_buff, data_end):
6340 info->reg_type = PTR_TO_PACKET_END;
6342 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
6343 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6347 return bpf_skb_is_valid_access(off, size, type, prog, info);
6350 static bool __is_valid_xdp_access(int off, int size)
6352 if (off < 0 || off >= sizeof(struct xdp_md))
6354 if (off % size != 0)
6356 if (size != sizeof(__u32))
6362 static bool xdp_is_valid_access(int off, int size,
6363 enum bpf_access_type type,
6364 const struct bpf_prog *prog,
6365 struct bpf_insn_access_aux *info)
6367 if (type == BPF_WRITE) {
6368 if (bpf_prog_is_dev_bound(prog->aux)) {
6370 case offsetof(struct xdp_md, rx_queue_index):
6371 return __is_valid_xdp_access(off, size);
6378 case offsetof(struct xdp_md, data):
6379 info->reg_type = PTR_TO_PACKET;
6381 case offsetof(struct xdp_md, data_meta):
6382 info->reg_type = PTR_TO_PACKET_META;
6384 case offsetof(struct xdp_md, data_end):
6385 info->reg_type = PTR_TO_PACKET_END;
6389 return __is_valid_xdp_access(off, size);
6392 void bpf_warn_invalid_xdp_action(u32 act)
6394 const u32 act_max = XDP_REDIRECT;
6396 WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
6397 act > act_max ? "Illegal" : "Driver unsupported",
6400 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
6402 static bool sock_addr_is_valid_access(int off, int size,
6403 enum bpf_access_type type,
6404 const struct bpf_prog *prog,
6405 struct bpf_insn_access_aux *info)
6407 const int size_default = sizeof(__u32);
6409 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
6411 if (off % size != 0)
6414 /* Disallow access to IPv6 fields from IPv4 contex and vise
6418 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
6419 switch (prog->expected_attach_type) {
6420 case BPF_CGROUP_INET4_BIND:
6421 case BPF_CGROUP_INET4_CONNECT:
6422 case BPF_CGROUP_UDP4_SENDMSG:
6428 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
6429 switch (prog->expected_attach_type) {
6430 case BPF_CGROUP_INET6_BIND:
6431 case BPF_CGROUP_INET6_CONNECT:
6432 case BPF_CGROUP_UDP6_SENDMSG:
6438 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
6439 switch (prog->expected_attach_type) {
6440 case BPF_CGROUP_UDP4_SENDMSG:
6446 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
6448 switch (prog->expected_attach_type) {
6449 case BPF_CGROUP_UDP6_SENDMSG:
6458 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
6459 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
6460 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
6461 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
6463 /* Only narrow read access allowed for now. */
6464 if (type == BPF_READ) {
6465 bpf_ctx_record_field_size(info, size_default);
6466 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
6469 if (size != size_default)
6473 case bpf_ctx_range(struct bpf_sock_addr, user_port):
6474 if (size != size_default)
6478 if (type == BPF_READ) {
6479 if (size != size_default)
6489 static bool sock_ops_is_valid_access(int off, int size,
6490 enum bpf_access_type type,
6491 const struct bpf_prog *prog,
6492 struct bpf_insn_access_aux *info)
6494 const int size_default = sizeof(__u32);
6496 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
6499 /* The verifier guarantees that size > 0. */
6500 if (off % size != 0)
6503 if (type == BPF_WRITE) {
6505 case offsetof(struct bpf_sock_ops, reply):
6506 case offsetof(struct bpf_sock_ops, sk_txhash):
6507 if (size != size_default)
6515 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
6517 if (size != sizeof(__u64))
6521 if (size != size_default)
6530 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
6531 const struct bpf_prog *prog)
6533 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
6536 static bool sk_skb_is_valid_access(int off, int size,
6537 enum bpf_access_type type,
6538 const struct bpf_prog *prog,
6539 struct bpf_insn_access_aux *info)
6542 case bpf_ctx_range(struct __sk_buff, tc_classid):
6543 case bpf_ctx_range(struct __sk_buff, data_meta):
6544 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
6545 case bpf_ctx_range(struct __sk_buff, tstamp):
6546 case bpf_ctx_range(struct __sk_buff, wire_len):
6550 if (type == BPF_WRITE) {
6552 case bpf_ctx_range(struct __sk_buff, tc_index):
6553 case bpf_ctx_range(struct __sk_buff, priority):
6561 case bpf_ctx_range(struct __sk_buff, mark):
6563 case bpf_ctx_range(struct __sk_buff, data):
6564 info->reg_type = PTR_TO_PACKET;
6566 case bpf_ctx_range(struct __sk_buff, data_end):
6567 info->reg_type = PTR_TO_PACKET_END;
6571 return bpf_skb_is_valid_access(off, size, type, prog, info);
6574 static bool sk_msg_is_valid_access(int off, int size,
6575 enum bpf_access_type type,
6576 const struct bpf_prog *prog,
6577 struct bpf_insn_access_aux *info)
6579 if (type == BPF_WRITE)
6582 if (off % size != 0)
6586 case offsetof(struct sk_msg_md, data):
6587 info->reg_type = PTR_TO_PACKET;
6588 if (size != sizeof(__u64))
6591 case offsetof(struct sk_msg_md, data_end):
6592 info->reg_type = PTR_TO_PACKET_END;
6593 if (size != sizeof(__u64))
6596 case bpf_ctx_range(struct sk_msg_md, family):
6597 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
6598 case bpf_ctx_range(struct sk_msg_md, local_ip4):
6599 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
6600 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
6601 case bpf_ctx_range(struct sk_msg_md, remote_port):
6602 case bpf_ctx_range(struct sk_msg_md, local_port):
6603 case bpf_ctx_range(struct sk_msg_md, size):
6604 if (size != sizeof(__u32))
6613 static bool flow_dissector_is_valid_access(int off, int size,
6614 enum bpf_access_type type,
6615 const struct bpf_prog *prog,
6616 struct bpf_insn_access_aux *info)
6618 if (type == BPF_WRITE)
6622 case bpf_ctx_range(struct __sk_buff, data):
6623 info->reg_type = PTR_TO_PACKET;
6625 case bpf_ctx_range(struct __sk_buff, data_end):
6626 info->reg_type = PTR_TO_PACKET_END;
6628 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
6629 info->reg_type = PTR_TO_FLOW_KEYS;
6635 return bpf_skb_is_valid_access(off, size, type, prog, info);
6638 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
6639 const struct bpf_insn *si,
6640 struct bpf_insn *insn_buf,
6641 struct bpf_prog *prog, u32 *target_size)
6643 struct bpf_insn *insn = insn_buf;
6647 case offsetof(struct __sk_buff, len):
6648 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6649 bpf_target_off(struct sk_buff, len, 4,
6653 case offsetof(struct __sk_buff, protocol):
6654 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6655 bpf_target_off(struct sk_buff, protocol, 2,
6659 case offsetof(struct __sk_buff, vlan_proto):
6660 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6661 bpf_target_off(struct sk_buff, vlan_proto, 2,
6665 case offsetof(struct __sk_buff, priority):
6666 if (type == BPF_WRITE)
6667 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6668 bpf_target_off(struct sk_buff, priority, 4,
6671 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6672 bpf_target_off(struct sk_buff, priority, 4,
6676 case offsetof(struct __sk_buff, ingress_ifindex):
6677 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6678 bpf_target_off(struct sk_buff, skb_iif, 4,
6682 case offsetof(struct __sk_buff, ifindex):
6683 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
6684 si->dst_reg, si->src_reg,
6685 offsetof(struct sk_buff, dev));
6686 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
6687 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6688 bpf_target_off(struct net_device, ifindex, 4,
6692 case offsetof(struct __sk_buff, hash):
6693 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6694 bpf_target_off(struct sk_buff, hash, 4,
6698 case offsetof(struct __sk_buff, mark):
6699 if (type == BPF_WRITE)
6700 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6701 bpf_target_off(struct sk_buff, mark, 4,
6704 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6705 bpf_target_off(struct sk_buff, mark, 4,
6709 case offsetof(struct __sk_buff, pkt_type):
6711 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
6713 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
6714 #ifdef __BIG_ENDIAN_BITFIELD
6715 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
6719 case offsetof(struct __sk_buff, queue_mapping):
6720 if (type == BPF_WRITE) {
6721 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
6722 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
6723 bpf_target_off(struct sk_buff,
6727 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6728 bpf_target_off(struct sk_buff,
6734 case offsetof(struct __sk_buff, vlan_present):
6736 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
6737 PKT_VLAN_PRESENT_OFFSET());
6738 if (PKT_VLAN_PRESENT_BIT)
6739 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
6740 if (PKT_VLAN_PRESENT_BIT < 7)
6741 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
6744 case offsetof(struct __sk_buff, vlan_tci):
6745 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6746 bpf_target_off(struct sk_buff, vlan_tci, 2,
6750 case offsetof(struct __sk_buff, cb[0]) ...
6751 offsetofend(struct __sk_buff, cb[4]) - 1:
6752 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20);
6753 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
6754 offsetof(struct qdisc_skb_cb, data)) %
6757 prog->cb_access = 1;
6759 off -= offsetof(struct __sk_buff, cb[0]);
6760 off += offsetof(struct sk_buff, cb);
6761 off += offsetof(struct qdisc_skb_cb, data);
6762 if (type == BPF_WRITE)
6763 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
6766 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
6770 case offsetof(struct __sk_buff, tc_classid):
6771 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, tc_classid) != 2);
6774 off -= offsetof(struct __sk_buff, tc_classid);
6775 off += offsetof(struct sk_buff, cb);
6776 off += offsetof(struct qdisc_skb_cb, tc_classid);
6778 if (type == BPF_WRITE)
6779 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
6782 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
6786 case offsetof(struct __sk_buff, data):
6787 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
6788 si->dst_reg, si->src_reg,
6789 offsetof(struct sk_buff, data));
6792 case offsetof(struct __sk_buff, data_meta):
6794 off -= offsetof(struct __sk_buff, data_meta);
6795 off += offsetof(struct sk_buff, cb);
6796 off += offsetof(struct bpf_skb_data_end, data_meta);
6797 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
6801 case offsetof(struct __sk_buff, data_end):
6803 off -= offsetof(struct __sk_buff, data_end);
6804 off += offsetof(struct sk_buff, cb);
6805 off += offsetof(struct bpf_skb_data_end, data_end);
6806 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
6810 case offsetof(struct __sk_buff, tc_index):
6811 #ifdef CONFIG_NET_SCHED
6812 if (type == BPF_WRITE)
6813 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
6814 bpf_target_off(struct sk_buff, tc_index, 2,
6817 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6818 bpf_target_off(struct sk_buff, tc_index, 2,
6822 if (type == BPF_WRITE)
6823 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
6825 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
6829 case offsetof(struct __sk_buff, napi_id):
6830 #if defined(CONFIG_NET_RX_BUSY_POLL)
6831 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6832 bpf_target_off(struct sk_buff, napi_id, 4,
6834 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
6835 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
6838 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
6841 case offsetof(struct __sk_buff, family):
6842 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
6844 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6845 si->dst_reg, si->src_reg,
6846 offsetof(struct sk_buff, sk));
6847 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6848 bpf_target_off(struct sock_common,
6852 case offsetof(struct __sk_buff, remote_ip4):
6853 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
6855 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6856 si->dst_reg, si->src_reg,
6857 offsetof(struct sk_buff, sk));
6858 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6859 bpf_target_off(struct sock_common,
6863 case offsetof(struct __sk_buff, local_ip4):
6864 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6865 skc_rcv_saddr) != 4);
6867 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6868 si->dst_reg, si->src_reg,
6869 offsetof(struct sk_buff, sk));
6870 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6871 bpf_target_off(struct sock_common,
6875 case offsetof(struct __sk_buff, remote_ip6[0]) ...
6876 offsetof(struct __sk_buff, remote_ip6[3]):
6877 #if IS_ENABLED(CONFIG_IPV6)
6878 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6879 skc_v6_daddr.s6_addr32[0]) != 4);
6882 off -= offsetof(struct __sk_buff, remote_ip6[0]);
6884 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6885 si->dst_reg, si->src_reg,
6886 offsetof(struct sk_buff, sk));
6887 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6888 offsetof(struct sock_common,
6889 skc_v6_daddr.s6_addr32[0]) +
6892 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6895 case offsetof(struct __sk_buff, local_ip6[0]) ...
6896 offsetof(struct __sk_buff, local_ip6[3]):
6897 #if IS_ENABLED(CONFIG_IPV6)
6898 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6899 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
6902 off -= offsetof(struct __sk_buff, local_ip6[0]);
6904 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6905 si->dst_reg, si->src_reg,
6906 offsetof(struct sk_buff, sk));
6907 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6908 offsetof(struct sock_common,
6909 skc_v6_rcv_saddr.s6_addr32[0]) +
6912 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6916 case offsetof(struct __sk_buff, remote_port):
6917 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
6919 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6920 si->dst_reg, si->src_reg,
6921 offsetof(struct sk_buff, sk));
6922 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6923 bpf_target_off(struct sock_common,
6926 #ifndef __BIG_ENDIAN_BITFIELD
6927 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
6931 case offsetof(struct __sk_buff, local_port):
6932 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
6934 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6935 si->dst_reg, si->src_reg,
6936 offsetof(struct sk_buff, sk));
6937 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6938 bpf_target_off(struct sock_common,
6939 skc_num, 2, target_size));
6942 case offsetof(struct __sk_buff, flow_keys):
6944 off -= offsetof(struct __sk_buff, flow_keys);
6945 off += offsetof(struct sk_buff, cb);
6946 off += offsetof(struct qdisc_skb_cb, flow_keys);
6947 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
6951 case offsetof(struct __sk_buff, tstamp):
6952 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, tstamp) != 8);
6954 if (type == BPF_WRITE)
6955 *insn++ = BPF_STX_MEM(BPF_DW,
6956 si->dst_reg, si->src_reg,
6957 bpf_target_off(struct sk_buff,
6961 *insn++ = BPF_LDX_MEM(BPF_DW,
6962 si->dst_reg, si->src_reg,
6963 bpf_target_off(struct sk_buff,
6968 case offsetof(struct __sk_buff, gso_segs):
6969 /* si->dst_reg = skb_shinfo(SKB); */
6970 #ifdef NET_SKBUFF_DATA_USES_OFFSET
6971 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
6972 si->dst_reg, si->src_reg,
6973 offsetof(struct sk_buff, head));
6974 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
6975 BPF_REG_AX, si->src_reg,
6976 offsetof(struct sk_buff, end));
6977 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
6979 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
6980 si->dst_reg, si->src_reg,
6981 offsetof(struct sk_buff, end));
6983 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
6984 si->dst_reg, si->dst_reg,
6985 bpf_target_off(struct skb_shared_info,
6989 case offsetof(struct __sk_buff, wire_len):
6990 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, pkt_len) != 4);
6993 off -= offsetof(struct __sk_buff, wire_len);
6994 off += offsetof(struct sk_buff, cb);
6995 off += offsetof(struct qdisc_skb_cb, pkt_len);
6997 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
7000 case offsetof(struct __sk_buff, sk):
7001 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7002 si->dst_reg, si->src_reg,
7003 offsetof(struct sk_buff, sk));
7007 return insn - insn_buf;
7010 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
7011 const struct bpf_insn *si,
7012 struct bpf_insn *insn_buf,
7013 struct bpf_prog *prog, u32 *target_size)
7015 struct bpf_insn *insn = insn_buf;
7019 case offsetof(struct bpf_sock, bound_dev_if):
7020 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_bound_dev_if) != 4);
7022 if (type == BPF_WRITE)
7023 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7024 offsetof(struct sock, sk_bound_dev_if));
7026 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7027 offsetof(struct sock, sk_bound_dev_if));
7030 case offsetof(struct bpf_sock, mark):
7031 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_mark) != 4);
7033 if (type == BPF_WRITE)
7034 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7035 offsetof(struct sock, sk_mark));
7037 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7038 offsetof(struct sock, sk_mark));
7041 case offsetof(struct bpf_sock, priority):
7042 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_priority) != 4);
7044 if (type == BPF_WRITE)
7045 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7046 offsetof(struct sock, sk_priority));
7048 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7049 offsetof(struct sock, sk_priority));
7052 case offsetof(struct bpf_sock, family):
7053 *insn++ = BPF_LDX_MEM(
7054 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
7055 si->dst_reg, si->src_reg,
7056 bpf_target_off(struct sock_common,
7058 FIELD_SIZEOF(struct sock_common,
7063 case offsetof(struct bpf_sock, type):
7064 BUILD_BUG_ON(HWEIGHT32(SK_FL_TYPE_MASK) != BITS_PER_BYTE * 2);
7065 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7066 offsetof(struct sock, __sk_flags_offset));
7067 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
7068 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
7072 case offsetof(struct bpf_sock, protocol):
7073 BUILD_BUG_ON(HWEIGHT32(SK_FL_PROTO_MASK) != BITS_PER_BYTE);
7074 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7075 offsetof(struct sock, __sk_flags_offset));
7076 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
7077 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_PROTO_SHIFT);
7081 case offsetof(struct bpf_sock, src_ip4):
7082 *insn++ = BPF_LDX_MEM(
7083 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7084 bpf_target_off(struct sock_common, skc_rcv_saddr,
7085 FIELD_SIZEOF(struct sock_common,
7090 case offsetof(struct bpf_sock, dst_ip4):
7091 *insn++ = BPF_LDX_MEM(
7092 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7093 bpf_target_off(struct sock_common, skc_daddr,
7094 FIELD_SIZEOF(struct sock_common,
7099 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7100 #if IS_ENABLED(CONFIG_IPV6)
7102 off -= offsetof(struct bpf_sock, src_ip6[0]);
7103 *insn++ = BPF_LDX_MEM(
7104 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7107 skc_v6_rcv_saddr.s6_addr32[0],
7108 FIELD_SIZEOF(struct sock_common,
7109 skc_v6_rcv_saddr.s6_addr32[0]),
7110 target_size) + off);
7113 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7117 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
7118 #if IS_ENABLED(CONFIG_IPV6)
7120 off -= offsetof(struct bpf_sock, dst_ip6[0]);
7121 *insn++ = BPF_LDX_MEM(
7122 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7123 bpf_target_off(struct sock_common,
7124 skc_v6_daddr.s6_addr32[0],
7125 FIELD_SIZEOF(struct sock_common,
7126 skc_v6_daddr.s6_addr32[0]),
7127 target_size) + off);
7129 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7134 case offsetof(struct bpf_sock, src_port):
7135 *insn++ = BPF_LDX_MEM(
7136 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
7137 si->dst_reg, si->src_reg,
7138 bpf_target_off(struct sock_common, skc_num,
7139 FIELD_SIZEOF(struct sock_common,
7144 case offsetof(struct bpf_sock, dst_port):
7145 *insn++ = BPF_LDX_MEM(
7146 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
7147 si->dst_reg, si->src_reg,
7148 bpf_target_off(struct sock_common, skc_dport,
7149 FIELD_SIZEOF(struct sock_common,
7154 case offsetof(struct bpf_sock, state):
7155 *insn++ = BPF_LDX_MEM(
7156 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
7157 si->dst_reg, si->src_reg,
7158 bpf_target_off(struct sock_common, skc_state,
7159 FIELD_SIZEOF(struct sock_common,
7165 return insn - insn_buf;
7168 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
7169 const struct bpf_insn *si,
7170 struct bpf_insn *insn_buf,
7171 struct bpf_prog *prog, u32 *target_size)
7173 struct bpf_insn *insn = insn_buf;
7176 case offsetof(struct __sk_buff, ifindex):
7177 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
7178 si->dst_reg, si->src_reg,
7179 offsetof(struct sk_buff, dev));
7180 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7181 bpf_target_off(struct net_device, ifindex, 4,
7185 return bpf_convert_ctx_access(type, si, insn_buf, prog,
7189 return insn - insn_buf;
7192 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
7193 const struct bpf_insn *si,
7194 struct bpf_insn *insn_buf,
7195 struct bpf_prog *prog, u32 *target_size)
7197 struct bpf_insn *insn = insn_buf;
7200 case offsetof(struct xdp_md, data):
7201 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
7202 si->dst_reg, si->src_reg,
7203 offsetof(struct xdp_buff, data));
7205 case offsetof(struct xdp_md, data_meta):
7206 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
7207 si->dst_reg, si->src_reg,
7208 offsetof(struct xdp_buff, data_meta));
7210 case offsetof(struct xdp_md, data_end):
7211 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
7212 si->dst_reg, si->src_reg,
7213 offsetof(struct xdp_buff, data_end));
7215 case offsetof(struct xdp_md, ingress_ifindex):
7216 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
7217 si->dst_reg, si->src_reg,
7218 offsetof(struct xdp_buff, rxq));
7219 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
7220 si->dst_reg, si->dst_reg,
7221 offsetof(struct xdp_rxq_info, dev));
7222 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7223 offsetof(struct net_device, ifindex));
7225 case offsetof(struct xdp_md, rx_queue_index):
7226 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
7227 si->dst_reg, si->src_reg,
7228 offsetof(struct xdp_buff, rxq));
7229 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7230 offsetof(struct xdp_rxq_info,
7235 return insn - insn_buf;
7238 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
7239 * context Structure, F is Field in context structure that contains a pointer
7240 * to Nested Structure of type NS that has the field NF.
7242 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
7243 * sure that SIZE is not greater than actual size of S.F.NF.
7245 * If offset OFF is provided, the load happens from that offset relative to
7248 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
7250 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
7251 si->src_reg, offsetof(S, F)); \
7252 *insn++ = BPF_LDX_MEM( \
7253 SIZE, si->dst_reg, si->dst_reg, \
7254 bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF), \
7259 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
7260 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
7261 BPF_FIELD_SIZEOF(NS, NF), 0)
7263 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
7264 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
7266 * It doesn't support SIZE argument though since narrow stores are not
7267 * supported for now.
7269 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
7270 * "register" since two registers available in convert_ctx_access are not
7271 * enough: we can't override neither SRC, since it contains value to store, nor
7272 * DST since it contains pointer to context that may be used by later
7273 * instructions. But we need a temporary place to save pointer to nested
7274 * structure whose field we want to store to.
7276 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, OFF, TF) \
7278 int tmp_reg = BPF_REG_9; \
7279 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
7281 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
7283 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
7285 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
7286 si->dst_reg, offsetof(S, F)); \
7287 *insn++ = BPF_STX_MEM( \
7288 BPF_FIELD_SIZEOF(NS, NF), tmp_reg, si->src_reg, \
7289 bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF), \
7292 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
7296 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
7299 if (type == BPF_WRITE) { \
7300 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, OFF, \
7303 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
7304 S, NS, F, NF, SIZE, OFF); \
7308 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
7309 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
7310 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
7312 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
7313 const struct bpf_insn *si,
7314 struct bpf_insn *insn_buf,
7315 struct bpf_prog *prog, u32 *target_size)
7317 struct bpf_insn *insn = insn_buf;
7321 case offsetof(struct bpf_sock_addr, user_family):
7322 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
7323 struct sockaddr, uaddr, sa_family);
7326 case offsetof(struct bpf_sock_addr, user_ip4):
7327 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7328 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
7329 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
7332 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
7334 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
7335 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7336 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
7337 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
7341 case offsetof(struct bpf_sock_addr, user_port):
7342 /* To get port we need to know sa_family first and then treat
7343 * sockaddr as either sockaddr_in or sockaddr_in6.
7344 * Though we can simplify since port field has same offset and
7345 * size in both structures.
7346 * Here we check this invariant and use just one of the
7347 * structures if it's true.
7349 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
7350 offsetof(struct sockaddr_in6, sin6_port));
7351 BUILD_BUG_ON(FIELD_SIZEOF(struct sockaddr_in, sin_port) !=
7352 FIELD_SIZEOF(struct sockaddr_in6, sin6_port));
7353 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(struct bpf_sock_addr_kern,
7354 struct sockaddr_in6, uaddr,
7355 sin6_port, tmp_reg);
7358 case offsetof(struct bpf_sock_addr, family):
7359 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
7360 struct sock, sk, sk_family);
7363 case offsetof(struct bpf_sock_addr, type):
7364 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(
7365 struct bpf_sock_addr_kern, struct sock, sk,
7366 __sk_flags_offset, BPF_W, 0);
7367 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
7368 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
7371 case offsetof(struct bpf_sock_addr, protocol):
7372 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(
7373 struct bpf_sock_addr_kern, struct sock, sk,
7374 __sk_flags_offset, BPF_W, 0);
7375 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
7376 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg,
7380 case offsetof(struct bpf_sock_addr, msg_src_ip4):
7381 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
7382 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7383 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
7384 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
7387 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
7390 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
7391 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
7392 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7393 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
7394 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
7398 return insn - insn_buf;
7401 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
7402 const struct bpf_insn *si,
7403 struct bpf_insn *insn_buf,
7404 struct bpf_prog *prog,
7407 struct bpf_insn *insn = insn_buf;
7410 /* Helper macro for adding read access to tcp_sock or sock fields. */
7411 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
7413 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) > \
7414 FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD)); \
7415 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
7416 struct bpf_sock_ops_kern, \
7418 si->dst_reg, si->src_reg, \
7419 offsetof(struct bpf_sock_ops_kern, \
7421 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 2); \
7422 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
7423 struct bpf_sock_ops_kern, sk),\
7424 si->dst_reg, si->src_reg, \
7425 offsetof(struct bpf_sock_ops_kern, sk));\
7426 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
7428 si->dst_reg, si->dst_reg, \
7429 offsetof(OBJ, OBJ_FIELD)); \
7432 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
7433 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
7435 /* Helper macro for adding write access to tcp_sock or sock fields.
7436 * The macro is called with two registers, dst_reg which contains a pointer
7437 * to ctx (context) and src_reg which contains the value that should be
7438 * stored. However, we need an additional register since we cannot overwrite
7439 * dst_reg because it may be used later in the program.
7440 * Instead we "borrow" one of the other register. We first save its value
7441 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
7442 * it at the end of the macro.
7444 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
7446 int reg = BPF_REG_9; \
7447 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) > \
7448 FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD)); \
7449 if (si->dst_reg == reg || si->src_reg == reg) \
7451 if (si->dst_reg == reg || si->src_reg == reg) \
7453 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
7454 offsetof(struct bpf_sock_ops_kern, \
7456 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
7457 struct bpf_sock_ops_kern, \
7460 offsetof(struct bpf_sock_ops_kern, \
7462 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
7463 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
7464 struct bpf_sock_ops_kern, sk),\
7466 offsetof(struct bpf_sock_ops_kern, sk));\
7467 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \
7469 offsetof(OBJ, OBJ_FIELD)); \
7470 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
7471 offsetof(struct bpf_sock_ops_kern, \
7475 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
7477 if (TYPE == BPF_WRITE) \
7478 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
7480 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
7483 CONVERT_COMMON_TCP_SOCK_FIELDS(struct bpf_sock_ops,
7484 SOCK_OPS_GET_TCP_SOCK_FIELD);
7486 if (insn > insn_buf)
7487 return insn - insn_buf;
7490 case offsetof(struct bpf_sock_ops, op) ...
7491 offsetof(struct bpf_sock_ops, replylong[3]):
7492 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, op) !=
7493 FIELD_SIZEOF(struct bpf_sock_ops_kern, op));
7494 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, reply) !=
7495 FIELD_SIZEOF(struct bpf_sock_ops_kern, reply));
7496 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, replylong) !=
7497 FIELD_SIZEOF(struct bpf_sock_ops_kern, replylong));
7499 off -= offsetof(struct bpf_sock_ops, op);
7500 off += offsetof(struct bpf_sock_ops_kern, op);
7501 if (type == BPF_WRITE)
7502 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7505 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7509 case offsetof(struct bpf_sock_ops, family):
7510 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
7512 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7513 struct bpf_sock_ops_kern, sk),
7514 si->dst_reg, si->src_reg,
7515 offsetof(struct bpf_sock_ops_kern, sk));
7516 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7517 offsetof(struct sock_common, skc_family));
7520 case offsetof(struct bpf_sock_ops, remote_ip4):
7521 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
7523 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7524 struct bpf_sock_ops_kern, sk),
7525 si->dst_reg, si->src_reg,
7526 offsetof(struct bpf_sock_ops_kern, sk));
7527 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7528 offsetof(struct sock_common, skc_daddr));
7531 case offsetof(struct bpf_sock_ops, local_ip4):
7532 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7533 skc_rcv_saddr) != 4);
7535 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7536 struct bpf_sock_ops_kern, sk),
7537 si->dst_reg, si->src_reg,
7538 offsetof(struct bpf_sock_ops_kern, sk));
7539 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7540 offsetof(struct sock_common,
7544 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
7545 offsetof(struct bpf_sock_ops, remote_ip6[3]):
7546 #if IS_ENABLED(CONFIG_IPV6)
7547 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7548 skc_v6_daddr.s6_addr32[0]) != 4);
7551 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
7552 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7553 struct bpf_sock_ops_kern, sk),
7554 si->dst_reg, si->src_reg,
7555 offsetof(struct bpf_sock_ops_kern, sk));
7556 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7557 offsetof(struct sock_common,
7558 skc_v6_daddr.s6_addr32[0]) +
7561 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7565 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
7566 offsetof(struct bpf_sock_ops, local_ip6[3]):
7567 #if IS_ENABLED(CONFIG_IPV6)
7568 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7569 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
7572 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
7573 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7574 struct bpf_sock_ops_kern, sk),
7575 si->dst_reg, si->src_reg,
7576 offsetof(struct bpf_sock_ops_kern, sk));
7577 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7578 offsetof(struct sock_common,
7579 skc_v6_rcv_saddr.s6_addr32[0]) +
7582 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7586 case offsetof(struct bpf_sock_ops, remote_port):
7587 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
7589 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7590 struct bpf_sock_ops_kern, sk),
7591 si->dst_reg, si->src_reg,
7592 offsetof(struct bpf_sock_ops_kern, sk));
7593 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7594 offsetof(struct sock_common, skc_dport));
7595 #ifndef __BIG_ENDIAN_BITFIELD
7596 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
7600 case offsetof(struct bpf_sock_ops, local_port):
7601 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
7603 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7604 struct bpf_sock_ops_kern, sk),
7605 si->dst_reg, si->src_reg,
7606 offsetof(struct bpf_sock_ops_kern, sk));
7607 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7608 offsetof(struct sock_common, skc_num));
7611 case offsetof(struct bpf_sock_ops, is_fullsock):
7612 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7613 struct bpf_sock_ops_kern,
7615 si->dst_reg, si->src_reg,
7616 offsetof(struct bpf_sock_ops_kern,
7620 case offsetof(struct bpf_sock_ops, state):
7621 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_state) != 1);
7623 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7624 struct bpf_sock_ops_kern, sk),
7625 si->dst_reg, si->src_reg,
7626 offsetof(struct bpf_sock_ops_kern, sk));
7627 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
7628 offsetof(struct sock_common, skc_state));
7631 case offsetof(struct bpf_sock_ops, rtt_min):
7632 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, rtt_min) !=
7633 sizeof(struct minmax));
7634 BUILD_BUG_ON(sizeof(struct minmax) <
7635 sizeof(struct minmax_sample));
7637 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7638 struct bpf_sock_ops_kern, sk),
7639 si->dst_reg, si->src_reg,
7640 offsetof(struct bpf_sock_ops_kern, sk));
7641 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7642 offsetof(struct tcp_sock, rtt_min) +
7643 FIELD_SIZEOF(struct minmax_sample, t));
7646 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
7647 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
7651 case offsetof(struct bpf_sock_ops, sk_txhash):
7652 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
7656 return insn - insn_buf;
7659 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
7660 const struct bpf_insn *si,
7661 struct bpf_insn *insn_buf,
7662 struct bpf_prog *prog, u32 *target_size)
7664 struct bpf_insn *insn = insn_buf;
7668 case offsetof(struct __sk_buff, data_end):
7670 off -= offsetof(struct __sk_buff, data_end);
7671 off += offsetof(struct sk_buff, cb);
7672 off += offsetof(struct tcp_skb_cb, bpf.data_end);
7673 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
7677 return bpf_convert_ctx_access(type, si, insn_buf, prog,
7681 return insn - insn_buf;
7684 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
7685 const struct bpf_insn *si,
7686 struct bpf_insn *insn_buf,
7687 struct bpf_prog *prog, u32 *target_size)
7689 struct bpf_insn *insn = insn_buf;
7690 #if IS_ENABLED(CONFIG_IPV6)
7694 /* convert ctx uses the fact sg element is first in struct */
7695 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
7698 case offsetof(struct sk_msg_md, data):
7699 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
7700 si->dst_reg, si->src_reg,
7701 offsetof(struct sk_msg, data));
7703 case offsetof(struct sk_msg_md, data_end):
7704 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
7705 si->dst_reg, si->src_reg,
7706 offsetof(struct sk_msg, data_end));
7708 case offsetof(struct sk_msg_md, family):
7709 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
7711 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7713 si->dst_reg, si->src_reg,
7714 offsetof(struct sk_msg, sk));
7715 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7716 offsetof(struct sock_common, skc_family));
7719 case offsetof(struct sk_msg_md, remote_ip4):
7720 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
7722 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7724 si->dst_reg, si->src_reg,
7725 offsetof(struct sk_msg, sk));
7726 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7727 offsetof(struct sock_common, skc_daddr));
7730 case offsetof(struct sk_msg_md, local_ip4):
7731 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7732 skc_rcv_saddr) != 4);
7734 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7736 si->dst_reg, si->src_reg,
7737 offsetof(struct sk_msg, sk));
7738 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7739 offsetof(struct sock_common,
7743 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
7744 offsetof(struct sk_msg_md, remote_ip6[3]):
7745 #if IS_ENABLED(CONFIG_IPV6)
7746 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7747 skc_v6_daddr.s6_addr32[0]) != 4);
7750 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
7751 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7753 si->dst_reg, si->src_reg,
7754 offsetof(struct sk_msg, sk));
7755 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7756 offsetof(struct sock_common,
7757 skc_v6_daddr.s6_addr32[0]) +
7760 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7764 case offsetof(struct sk_msg_md, local_ip6[0]) ...
7765 offsetof(struct sk_msg_md, local_ip6[3]):
7766 #if IS_ENABLED(CONFIG_IPV6)
7767 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7768 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
7771 off -= offsetof(struct sk_msg_md, local_ip6[0]);
7772 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7774 si->dst_reg, si->src_reg,
7775 offsetof(struct sk_msg, sk));
7776 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7777 offsetof(struct sock_common,
7778 skc_v6_rcv_saddr.s6_addr32[0]) +
7781 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7785 case offsetof(struct sk_msg_md, remote_port):
7786 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
7788 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7790 si->dst_reg, si->src_reg,
7791 offsetof(struct sk_msg, sk));
7792 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7793 offsetof(struct sock_common, skc_dport));
7794 #ifndef __BIG_ENDIAN_BITFIELD
7795 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
7799 case offsetof(struct sk_msg_md, local_port):
7800 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
7802 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7804 si->dst_reg, si->src_reg,
7805 offsetof(struct sk_msg, sk));
7806 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7807 offsetof(struct sock_common, skc_num));
7810 case offsetof(struct sk_msg_md, size):
7811 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
7812 si->dst_reg, si->src_reg,
7813 offsetof(struct sk_msg_sg, size));
7817 return insn - insn_buf;
7820 const struct bpf_verifier_ops sk_filter_verifier_ops = {
7821 .get_func_proto = sk_filter_func_proto,
7822 .is_valid_access = sk_filter_is_valid_access,
7823 .convert_ctx_access = bpf_convert_ctx_access,
7824 .gen_ld_abs = bpf_gen_ld_abs,
7827 const struct bpf_prog_ops sk_filter_prog_ops = {
7828 .test_run = bpf_prog_test_run_skb,
7831 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
7832 .get_func_proto = tc_cls_act_func_proto,
7833 .is_valid_access = tc_cls_act_is_valid_access,
7834 .convert_ctx_access = tc_cls_act_convert_ctx_access,
7835 .gen_prologue = tc_cls_act_prologue,
7836 .gen_ld_abs = bpf_gen_ld_abs,
7839 const struct bpf_prog_ops tc_cls_act_prog_ops = {
7840 .test_run = bpf_prog_test_run_skb,
7843 const struct bpf_verifier_ops xdp_verifier_ops = {
7844 .get_func_proto = xdp_func_proto,
7845 .is_valid_access = xdp_is_valid_access,
7846 .convert_ctx_access = xdp_convert_ctx_access,
7847 .gen_prologue = bpf_noop_prologue,
7850 const struct bpf_prog_ops xdp_prog_ops = {
7851 .test_run = bpf_prog_test_run_xdp,
7854 const struct bpf_verifier_ops cg_skb_verifier_ops = {
7855 .get_func_proto = cg_skb_func_proto,
7856 .is_valid_access = cg_skb_is_valid_access,
7857 .convert_ctx_access = bpf_convert_ctx_access,
7860 const struct bpf_prog_ops cg_skb_prog_ops = {
7861 .test_run = bpf_prog_test_run_skb,
7864 const struct bpf_verifier_ops lwt_in_verifier_ops = {
7865 .get_func_proto = lwt_in_func_proto,
7866 .is_valid_access = lwt_is_valid_access,
7867 .convert_ctx_access = bpf_convert_ctx_access,
7870 const struct bpf_prog_ops lwt_in_prog_ops = {
7871 .test_run = bpf_prog_test_run_skb,
7874 const struct bpf_verifier_ops lwt_out_verifier_ops = {
7875 .get_func_proto = lwt_out_func_proto,
7876 .is_valid_access = lwt_is_valid_access,
7877 .convert_ctx_access = bpf_convert_ctx_access,
7880 const struct bpf_prog_ops lwt_out_prog_ops = {
7881 .test_run = bpf_prog_test_run_skb,
7884 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
7885 .get_func_proto = lwt_xmit_func_proto,
7886 .is_valid_access = lwt_is_valid_access,
7887 .convert_ctx_access = bpf_convert_ctx_access,
7888 .gen_prologue = tc_cls_act_prologue,
7891 const struct bpf_prog_ops lwt_xmit_prog_ops = {
7892 .test_run = bpf_prog_test_run_skb,
7895 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
7896 .get_func_proto = lwt_seg6local_func_proto,
7897 .is_valid_access = lwt_is_valid_access,
7898 .convert_ctx_access = bpf_convert_ctx_access,
7901 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
7902 .test_run = bpf_prog_test_run_skb,
7905 const struct bpf_verifier_ops cg_sock_verifier_ops = {
7906 .get_func_proto = sock_filter_func_proto,
7907 .is_valid_access = sock_filter_is_valid_access,
7908 .convert_ctx_access = bpf_sock_convert_ctx_access,
7911 const struct bpf_prog_ops cg_sock_prog_ops = {
7914 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
7915 .get_func_proto = sock_addr_func_proto,
7916 .is_valid_access = sock_addr_is_valid_access,
7917 .convert_ctx_access = sock_addr_convert_ctx_access,
7920 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
7923 const struct bpf_verifier_ops sock_ops_verifier_ops = {
7924 .get_func_proto = sock_ops_func_proto,
7925 .is_valid_access = sock_ops_is_valid_access,
7926 .convert_ctx_access = sock_ops_convert_ctx_access,
7929 const struct bpf_prog_ops sock_ops_prog_ops = {
7932 const struct bpf_verifier_ops sk_skb_verifier_ops = {
7933 .get_func_proto = sk_skb_func_proto,
7934 .is_valid_access = sk_skb_is_valid_access,
7935 .convert_ctx_access = sk_skb_convert_ctx_access,
7936 .gen_prologue = sk_skb_prologue,
7939 const struct bpf_prog_ops sk_skb_prog_ops = {
7942 const struct bpf_verifier_ops sk_msg_verifier_ops = {
7943 .get_func_proto = sk_msg_func_proto,
7944 .is_valid_access = sk_msg_is_valid_access,
7945 .convert_ctx_access = sk_msg_convert_ctx_access,
7946 .gen_prologue = bpf_noop_prologue,
7949 const struct bpf_prog_ops sk_msg_prog_ops = {
7952 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
7953 .get_func_proto = flow_dissector_func_proto,
7954 .is_valid_access = flow_dissector_is_valid_access,
7955 .convert_ctx_access = bpf_convert_ctx_access,
7958 const struct bpf_prog_ops flow_dissector_prog_ops = {
7959 .test_run = bpf_prog_test_run_flow_dissector,
7962 int sk_detach_filter(struct sock *sk)
7965 struct sk_filter *filter;
7967 if (sock_flag(sk, SOCK_FILTER_LOCKED))
7970 filter = rcu_dereference_protected(sk->sk_filter,
7971 lockdep_sock_is_held(sk));
7973 RCU_INIT_POINTER(sk->sk_filter, NULL);
7974 sk_filter_uncharge(sk, filter);
7980 EXPORT_SYMBOL_GPL(sk_detach_filter);
7982 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
7985 struct sock_fprog_kern *fprog;
7986 struct sk_filter *filter;
7990 filter = rcu_dereference_protected(sk->sk_filter,
7991 lockdep_sock_is_held(sk));
7995 /* We're copying the filter that has been originally attached,
7996 * so no conversion/decode needed anymore. eBPF programs that
7997 * have no original program cannot be dumped through this.
8000 fprog = filter->prog->orig_prog;
8006 /* User space only enquires number of filter blocks. */
8010 if (len < fprog->len)
8014 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
8017 /* Instead of bytes, the API requests to return the number
8027 struct sk_reuseport_kern {
8028 struct sk_buff *skb;
8030 struct sock *selected_sk;
8037 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
8038 struct sock_reuseport *reuse,
8039 struct sock *sk, struct sk_buff *skb,
8042 reuse_kern->skb = skb;
8043 reuse_kern->sk = sk;
8044 reuse_kern->selected_sk = NULL;
8045 reuse_kern->data_end = skb->data + skb_headlen(skb);
8046 reuse_kern->hash = hash;
8047 reuse_kern->reuseport_id = reuse->reuseport_id;
8048 reuse_kern->bind_inany = reuse->bind_inany;
8051 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
8052 struct bpf_prog *prog, struct sk_buff *skb,
8055 struct sk_reuseport_kern reuse_kern;
8056 enum sk_action action;
8058 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, hash);
8059 action = BPF_PROG_RUN(prog, &reuse_kern);
8061 if (action == SK_PASS)
8062 return reuse_kern.selected_sk;
8064 return ERR_PTR(-ECONNREFUSED);
8067 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
8068 struct bpf_map *, map, void *, key, u32, flags)
8070 struct sock_reuseport *reuse;
8071 struct sock *selected_sk;
8073 selected_sk = map->ops->map_lookup_elem(map, key);
8077 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
8079 /* selected_sk is unhashed (e.g. by close()) after the
8080 * above map_lookup_elem(). Treat selected_sk has already
8081 * been removed from the map.
8085 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
8088 if (unlikely(!reuse_kern->reuseport_id))
8089 /* There is a small race between adding the
8090 * sk to the map and setting the
8091 * reuse_kern->reuseport_id.
8092 * Treat it as the sk has not been added to
8097 sk = reuse_kern->sk;
8098 if (sk->sk_protocol != selected_sk->sk_protocol)
8100 else if (sk->sk_family != selected_sk->sk_family)
8101 return -EAFNOSUPPORT;
8103 /* Catch all. Likely bound to a different sockaddr. */
8107 reuse_kern->selected_sk = selected_sk;
8112 static const struct bpf_func_proto sk_select_reuseport_proto = {
8113 .func = sk_select_reuseport,
8115 .ret_type = RET_INTEGER,
8116 .arg1_type = ARG_PTR_TO_CTX,
8117 .arg2_type = ARG_CONST_MAP_PTR,
8118 .arg3_type = ARG_PTR_TO_MAP_KEY,
8119 .arg4_type = ARG_ANYTHING,
8122 BPF_CALL_4(sk_reuseport_load_bytes,
8123 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
8124 void *, to, u32, len)
8126 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
8129 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
8130 .func = sk_reuseport_load_bytes,
8132 .ret_type = RET_INTEGER,
8133 .arg1_type = ARG_PTR_TO_CTX,
8134 .arg2_type = ARG_ANYTHING,
8135 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
8136 .arg4_type = ARG_CONST_SIZE,
8139 BPF_CALL_5(sk_reuseport_load_bytes_relative,
8140 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
8141 void *, to, u32, len, u32, start_header)
8143 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
8147 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
8148 .func = sk_reuseport_load_bytes_relative,
8150 .ret_type = RET_INTEGER,
8151 .arg1_type = ARG_PTR_TO_CTX,
8152 .arg2_type = ARG_ANYTHING,
8153 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
8154 .arg4_type = ARG_CONST_SIZE,
8155 .arg5_type = ARG_ANYTHING,
8158 static const struct bpf_func_proto *
8159 sk_reuseport_func_proto(enum bpf_func_id func_id,
8160 const struct bpf_prog *prog)
8163 case BPF_FUNC_sk_select_reuseport:
8164 return &sk_select_reuseport_proto;
8165 case BPF_FUNC_skb_load_bytes:
8166 return &sk_reuseport_load_bytes_proto;
8167 case BPF_FUNC_skb_load_bytes_relative:
8168 return &sk_reuseport_load_bytes_relative_proto;
8170 return bpf_base_func_proto(func_id);
8175 sk_reuseport_is_valid_access(int off, int size,
8176 enum bpf_access_type type,
8177 const struct bpf_prog *prog,
8178 struct bpf_insn_access_aux *info)
8180 const u32 size_default = sizeof(__u32);
8182 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
8183 off % size || type != BPF_READ)
8187 case offsetof(struct sk_reuseport_md, data):
8188 info->reg_type = PTR_TO_PACKET;
8189 return size == sizeof(__u64);
8191 case offsetof(struct sk_reuseport_md, data_end):
8192 info->reg_type = PTR_TO_PACKET_END;
8193 return size == sizeof(__u64);
8195 case offsetof(struct sk_reuseport_md, hash):
8196 return size == size_default;
8198 /* Fields that allow narrowing */
8199 case offsetof(struct sk_reuseport_md, eth_protocol):
8200 if (size < FIELD_SIZEOF(struct sk_buff, protocol))
8203 case offsetof(struct sk_reuseport_md, ip_protocol):
8204 case offsetof(struct sk_reuseport_md, bind_inany):
8205 case offsetof(struct sk_reuseport_md, len):
8206 bpf_ctx_record_field_size(info, size_default);
8207 return bpf_ctx_narrow_access_ok(off, size, size_default);
8214 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
8215 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
8216 si->dst_reg, si->src_reg, \
8217 bpf_target_off(struct sk_reuseport_kern, F, \
8218 FIELD_SIZEOF(struct sk_reuseport_kern, F), \
8222 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
8223 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
8228 #define SK_REUSEPORT_LOAD_SK_FIELD_SIZE_OFF(SK_FIELD, BPF_SIZE, EXTRA_OFF) \
8229 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(struct sk_reuseport_kern, \
8232 SK_FIELD, BPF_SIZE, EXTRA_OFF)
8234 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
8235 const struct bpf_insn *si,
8236 struct bpf_insn *insn_buf,
8237 struct bpf_prog *prog,
8240 struct bpf_insn *insn = insn_buf;
8243 case offsetof(struct sk_reuseport_md, data):
8244 SK_REUSEPORT_LOAD_SKB_FIELD(data);
8247 case offsetof(struct sk_reuseport_md, len):
8248 SK_REUSEPORT_LOAD_SKB_FIELD(len);
8251 case offsetof(struct sk_reuseport_md, eth_protocol):
8252 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
8255 case offsetof(struct sk_reuseport_md, ip_protocol):
8256 BUILD_BUG_ON(HWEIGHT32(SK_FL_PROTO_MASK) != BITS_PER_BYTE);
8257 SK_REUSEPORT_LOAD_SK_FIELD_SIZE_OFF(__sk_flags_offset,
8259 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
8260 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg,
8262 /* SK_FL_PROTO_MASK and SK_FL_PROTO_SHIFT are endian
8263 * aware. No further narrowing or masking is needed.
8268 case offsetof(struct sk_reuseport_md, data_end):
8269 SK_REUSEPORT_LOAD_FIELD(data_end);
8272 case offsetof(struct sk_reuseport_md, hash):
8273 SK_REUSEPORT_LOAD_FIELD(hash);
8276 case offsetof(struct sk_reuseport_md, bind_inany):
8277 SK_REUSEPORT_LOAD_FIELD(bind_inany);
8281 return insn - insn_buf;
8284 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
8285 .get_func_proto = sk_reuseport_func_proto,
8286 .is_valid_access = sk_reuseport_is_valid_access,
8287 .convert_ctx_access = sk_reuseport_convert_ctx_access,
8290 const struct bpf_prog_ops sk_reuseport_prog_ops = {
8292 #endif /* CONFIG_INET */