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 sk = sk_to_full_sk(sk);
1801 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1804 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1805 .func = bpf_sk_fullsock,
1807 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
1808 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
1811 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1812 unsigned int write_len)
1814 int err = __bpf_try_make_writable(skb, write_len);
1816 bpf_compute_data_end_sk_skb(skb);
1820 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1822 /* Idea is the following: should the needed direct read/write
1823 * test fail during runtime, we can pull in more data and redo
1824 * again, since implicitly, we invalidate previous checks here.
1826 * Or, since we know how much we need to make read/writeable,
1827 * this can be done once at the program beginning for direct
1828 * access case. By this we overcome limitations of only current
1829 * headroom being accessible.
1831 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1834 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1835 .func = sk_skb_pull_data,
1837 .ret_type = RET_INTEGER,
1838 .arg1_type = ARG_PTR_TO_CTX,
1839 .arg2_type = ARG_ANYTHING,
1842 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1843 u64, from, u64, to, u64, flags)
1847 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1849 if (unlikely(offset > 0xffff || offset & 1))
1851 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1854 ptr = (__sum16 *)(skb->data + offset);
1855 switch (flags & BPF_F_HDR_FIELD_MASK) {
1857 if (unlikely(from != 0))
1860 csum_replace_by_diff(ptr, to);
1863 csum_replace2(ptr, from, to);
1866 csum_replace4(ptr, from, to);
1875 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1876 .func = bpf_l3_csum_replace,
1878 .ret_type = RET_INTEGER,
1879 .arg1_type = ARG_PTR_TO_CTX,
1880 .arg2_type = ARG_ANYTHING,
1881 .arg3_type = ARG_ANYTHING,
1882 .arg4_type = ARG_ANYTHING,
1883 .arg5_type = ARG_ANYTHING,
1886 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1887 u64, from, u64, to, u64, flags)
1889 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1890 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1891 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1894 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1895 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1897 if (unlikely(offset > 0xffff || offset & 1))
1899 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1902 ptr = (__sum16 *)(skb->data + offset);
1903 if (is_mmzero && !do_mforce && !*ptr)
1906 switch (flags & BPF_F_HDR_FIELD_MASK) {
1908 if (unlikely(from != 0))
1911 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1914 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1917 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1923 if (is_mmzero && !*ptr)
1924 *ptr = CSUM_MANGLED_0;
1928 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1929 .func = bpf_l4_csum_replace,
1931 .ret_type = RET_INTEGER,
1932 .arg1_type = ARG_PTR_TO_CTX,
1933 .arg2_type = ARG_ANYTHING,
1934 .arg3_type = ARG_ANYTHING,
1935 .arg4_type = ARG_ANYTHING,
1936 .arg5_type = ARG_ANYTHING,
1939 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1940 __be32 *, to, u32, to_size, __wsum, seed)
1942 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1943 u32 diff_size = from_size + to_size;
1946 /* This is quite flexible, some examples:
1948 * from_size == 0, to_size > 0, seed := csum --> pushing data
1949 * from_size > 0, to_size == 0, seed := csum --> pulling data
1950 * from_size > 0, to_size > 0, seed := 0 --> diffing data
1952 * Even for diffing, from_size and to_size don't need to be equal.
1954 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
1955 diff_size > sizeof(sp->diff)))
1958 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
1959 sp->diff[j] = ~from[i];
1960 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
1961 sp->diff[j] = to[i];
1963 return csum_partial(sp->diff, diff_size, seed);
1966 static const struct bpf_func_proto bpf_csum_diff_proto = {
1967 .func = bpf_csum_diff,
1970 .ret_type = RET_INTEGER,
1971 .arg1_type = ARG_PTR_TO_MEM_OR_NULL,
1972 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
1973 .arg3_type = ARG_PTR_TO_MEM_OR_NULL,
1974 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
1975 .arg5_type = ARG_ANYTHING,
1978 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
1980 /* The interface is to be used in combination with bpf_csum_diff()
1981 * for direct packet writes. csum rotation for alignment as well
1982 * as emulating csum_sub() can be done from the eBPF program.
1984 if (skb->ip_summed == CHECKSUM_COMPLETE)
1985 return (skb->csum = csum_add(skb->csum, csum));
1990 static const struct bpf_func_proto bpf_csum_update_proto = {
1991 .func = bpf_csum_update,
1993 .ret_type = RET_INTEGER,
1994 .arg1_type = ARG_PTR_TO_CTX,
1995 .arg2_type = ARG_ANYTHING,
1998 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2000 return dev_forward_skb(dev, skb);
2003 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2004 struct sk_buff *skb)
2006 int ret = ____dev_forward_skb(dev, skb);
2010 ret = netif_rx(skb);
2016 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2020 if (unlikely(__this_cpu_read(xmit_recursion) > XMIT_RECURSION_LIMIT)) {
2021 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2028 __this_cpu_inc(xmit_recursion);
2029 ret = dev_queue_xmit(skb);
2030 __this_cpu_dec(xmit_recursion);
2035 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2038 unsigned int mlen = skb_network_offset(skb);
2041 __skb_pull(skb, mlen);
2043 /* At ingress, the mac header has already been pulled once.
2044 * At egress, skb_pospull_rcsum has to be done in case that
2045 * the skb is originated from ingress (i.e. a forwarded skb)
2046 * to ensure that rcsum starts at net header.
2048 if (!skb_at_tc_ingress(skb))
2049 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2051 skb_pop_mac_header(skb);
2052 skb_reset_mac_len(skb);
2053 return flags & BPF_F_INGRESS ?
2054 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2057 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2060 /* Verify that a link layer header is carried */
2061 if (unlikely(skb->mac_header >= skb->network_header)) {
2066 bpf_push_mac_rcsum(skb);
2067 return flags & BPF_F_INGRESS ?
2068 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2071 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2074 if (dev_is_mac_header_xmit(dev))
2075 return __bpf_redirect_common(skb, dev, flags);
2077 return __bpf_redirect_no_mac(skb, dev, flags);
2080 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2082 struct net_device *dev;
2083 struct sk_buff *clone;
2086 if (unlikely(flags & ~(BPF_F_INGRESS)))
2089 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2093 clone = skb_clone(skb, GFP_ATOMIC);
2094 if (unlikely(!clone))
2097 /* For direct write, we need to keep the invariant that the skbs
2098 * we're dealing with need to be uncloned. Should uncloning fail
2099 * here, we need to free the just generated clone to unclone once
2102 ret = bpf_try_make_head_writable(skb);
2103 if (unlikely(ret)) {
2108 return __bpf_redirect(clone, dev, flags);
2111 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2112 .func = bpf_clone_redirect,
2114 .ret_type = RET_INTEGER,
2115 .arg1_type = ARG_PTR_TO_CTX,
2116 .arg2_type = ARG_ANYTHING,
2117 .arg3_type = ARG_ANYTHING,
2120 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2121 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2123 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2125 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2127 if (unlikely(flags & ~(BPF_F_INGRESS)))
2130 ri->ifindex = ifindex;
2133 return TC_ACT_REDIRECT;
2136 int skb_do_redirect(struct sk_buff *skb)
2138 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2139 struct net_device *dev;
2141 dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->ifindex);
2143 if (unlikely(!dev)) {
2148 return __bpf_redirect(skb, dev, ri->flags);
2151 static const struct bpf_func_proto bpf_redirect_proto = {
2152 .func = bpf_redirect,
2154 .ret_type = RET_INTEGER,
2155 .arg1_type = ARG_ANYTHING,
2156 .arg2_type = ARG_ANYTHING,
2159 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2161 msg->apply_bytes = bytes;
2165 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2166 .func = bpf_msg_apply_bytes,
2168 .ret_type = RET_INTEGER,
2169 .arg1_type = ARG_PTR_TO_CTX,
2170 .arg2_type = ARG_ANYTHING,
2173 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2175 msg->cork_bytes = bytes;
2179 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2180 .func = bpf_msg_cork_bytes,
2182 .ret_type = RET_INTEGER,
2183 .arg1_type = ARG_PTR_TO_CTX,
2184 .arg2_type = ARG_ANYTHING,
2187 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2188 u32, end, u64, flags)
2190 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2191 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2192 struct scatterlist *sge;
2193 u8 *raw, *to, *from;
2196 if (unlikely(flags || end <= start))
2199 /* First find the starting scatterlist element */
2202 len = sk_msg_elem(msg, i)->length;
2203 if (start < offset + len)
2206 sk_msg_iter_var_next(i);
2207 } while (i != msg->sg.end);
2209 if (unlikely(start >= offset + len))
2213 /* The start may point into the sg element so we need to also
2214 * account for the headroom.
2216 bytes_sg_total = start - offset + bytes;
2217 if (!msg->sg.copy[i] && bytes_sg_total <= len)
2220 /* At this point we need to linearize multiple scatterlist
2221 * elements or a single shared page. Either way we need to
2222 * copy into a linear buffer exclusively owned by BPF. Then
2223 * place the buffer in the scatterlist and fixup the original
2224 * entries by removing the entries now in the linear buffer
2225 * and shifting the remaining entries. For now we do not try
2226 * to copy partial entries to avoid complexity of running out
2227 * of sg_entry slots. The downside is reading a single byte
2228 * will copy the entire sg entry.
2231 copy += sk_msg_elem(msg, i)->length;
2232 sk_msg_iter_var_next(i);
2233 if (bytes_sg_total <= copy)
2235 } while (i != msg->sg.end);
2238 if (unlikely(bytes_sg_total > copy))
2241 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2243 if (unlikely(!page))
2246 raw = page_address(page);
2249 sge = sk_msg_elem(msg, i);
2250 from = sg_virt(sge);
2254 memcpy(to, from, len);
2257 put_page(sg_page(sge));
2259 sk_msg_iter_var_next(i);
2260 } while (i != last_sge);
2262 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2264 /* To repair sg ring we need to shift entries. If we only
2265 * had a single entry though we can just replace it and
2266 * be done. Otherwise walk the ring and shift the entries.
2268 WARN_ON_ONCE(last_sge == first_sge);
2269 shift = last_sge > first_sge ?
2270 last_sge - first_sge - 1 :
2271 MAX_SKB_FRAGS - first_sge + last_sge - 1;
2276 sk_msg_iter_var_next(i);
2280 if (i + shift >= MAX_MSG_FRAGS)
2281 move_from = i + shift - MAX_MSG_FRAGS;
2283 move_from = i + shift;
2284 if (move_from == msg->sg.end)
2287 msg->sg.data[i] = msg->sg.data[move_from];
2288 msg->sg.data[move_from].length = 0;
2289 msg->sg.data[move_from].page_link = 0;
2290 msg->sg.data[move_from].offset = 0;
2291 sk_msg_iter_var_next(i);
2294 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2295 msg->sg.end - shift + MAX_MSG_FRAGS :
2296 msg->sg.end - shift;
2298 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2299 msg->data_end = msg->data + bytes;
2303 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2304 .func = bpf_msg_pull_data,
2306 .ret_type = RET_INTEGER,
2307 .arg1_type = ARG_PTR_TO_CTX,
2308 .arg2_type = ARG_ANYTHING,
2309 .arg3_type = ARG_ANYTHING,
2310 .arg4_type = ARG_ANYTHING,
2313 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2314 u32, len, u64, flags)
2316 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2317 u32 new, i = 0, l, space, copy = 0, offset = 0;
2318 u8 *raw, *to, *from;
2321 if (unlikely(flags))
2324 /* First find the starting scatterlist element */
2327 l = sk_msg_elem(msg, i)->length;
2329 if (start < offset + l)
2332 sk_msg_iter_var_next(i);
2333 } while (i != msg->sg.end);
2335 if (start >= offset + l)
2338 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2340 /* If no space available will fallback to copy, we need at
2341 * least one scatterlist elem available to push data into
2342 * when start aligns to the beginning of an element or two
2343 * when it falls inside an element. We handle the start equals
2344 * offset case because its the common case for inserting a
2347 if (!space || (space == 1 && start != offset))
2348 copy = msg->sg.data[i].length;
2350 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2351 get_order(copy + len));
2352 if (unlikely(!page))
2358 raw = page_address(page);
2360 psge = sk_msg_elem(msg, i);
2361 front = start - offset;
2362 back = psge->length - front;
2363 from = sg_virt(psge);
2366 memcpy(raw, from, front);
2370 to = raw + front + len;
2372 memcpy(to, from, back);
2375 put_page(sg_page(psge));
2376 } else if (start - offset) {
2377 psge = sk_msg_elem(msg, i);
2378 rsge = sk_msg_elem_cpy(msg, i);
2380 psge->length = start - offset;
2381 rsge.length -= psge->length;
2382 rsge.offset += start;
2384 sk_msg_iter_var_next(i);
2385 sg_unmark_end(psge);
2386 sk_msg_iter_next(msg, end);
2389 /* Slot(s) to place newly allocated data */
2392 /* Shift one or two slots as needed */
2394 sge = sk_msg_elem_cpy(msg, i);
2396 sk_msg_iter_var_next(i);
2397 sg_unmark_end(&sge);
2398 sk_msg_iter_next(msg, end);
2400 nsge = sk_msg_elem_cpy(msg, i);
2402 sk_msg_iter_var_next(i);
2403 nnsge = sk_msg_elem_cpy(msg, i);
2406 while (i != msg->sg.end) {
2407 msg->sg.data[i] = sge;
2409 sk_msg_iter_var_next(i);
2412 nnsge = sk_msg_elem_cpy(msg, i);
2414 nsge = sk_msg_elem_cpy(msg, i);
2419 /* Place newly allocated data buffer */
2420 sk_mem_charge(msg->sk, len);
2421 msg->sg.size += len;
2422 msg->sg.copy[new] = false;
2423 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2425 get_page(sg_page(&rsge));
2426 sk_msg_iter_var_next(new);
2427 msg->sg.data[new] = rsge;
2430 sk_msg_compute_data_pointers(msg);
2434 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2435 .func = bpf_msg_push_data,
2437 .ret_type = RET_INTEGER,
2438 .arg1_type = ARG_PTR_TO_CTX,
2439 .arg2_type = ARG_ANYTHING,
2440 .arg3_type = ARG_ANYTHING,
2441 .arg4_type = ARG_ANYTHING,
2444 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2450 sk_msg_iter_var_next(i);
2451 msg->sg.data[prev] = msg->sg.data[i];
2452 } while (i != msg->sg.end);
2454 sk_msg_iter_prev(msg, end);
2457 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2459 struct scatterlist tmp, sge;
2461 sk_msg_iter_next(msg, end);
2462 sge = sk_msg_elem_cpy(msg, i);
2463 sk_msg_iter_var_next(i);
2464 tmp = sk_msg_elem_cpy(msg, i);
2466 while (i != msg->sg.end) {
2467 msg->sg.data[i] = sge;
2468 sk_msg_iter_var_next(i);
2470 tmp = sk_msg_elem_cpy(msg, i);
2474 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2475 u32, len, u64, flags)
2477 u32 i = 0, l, space, offset = 0;
2478 u64 last = start + len;
2481 if (unlikely(flags))
2484 /* First find the starting scatterlist element */
2487 l = sk_msg_elem(msg, i)->length;
2489 if (start < offset + l)
2492 sk_msg_iter_var_next(i);
2493 } while (i != msg->sg.end);
2495 /* Bounds checks: start and pop must be inside message */
2496 if (start >= offset + l || last >= msg->sg.size)
2499 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2502 /* --------------| offset
2503 * -| start |-------- len -------|
2505 * |----- a ----|-------- pop -------|----- b ----|
2506 * |______________________________________________| length
2509 * a: region at front of scatter element to save
2510 * b: region at back of scatter element to save when length > A + pop
2511 * pop: region to pop from element, same as input 'pop' here will be
2512 * decremented below per iteration.
2514 * Two top-level cases to handle when start != offset, first B is non
2515 * zero and second B is zero corresponding to when a pop includes more
2518 * Then if B is non-zero AND there is no space allocate space and
2519 * compact A, B regions into page. If there is space shift ring to
2520 * the rigth free'ing the next element in ring to place B, leaving
2521 * A untouched except to reduce length.
2523 if (start != offset) {
2524 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2526 int b = sge->length - pop - a;
2528 sk_msg_iter_var_next(i);
2530 if (pop < sge->length - a) {
2533 sk_msg_shift_right(msg, i);
2534 nsge = sk_msg_elem(msg, i);
2535 get_page(sg_page(sge));
2538 b, sge->offset + pop + a);
2540 struct page *page, *orig;
2543 page = alloc_pages(__GFP_NOWARN |
2544 __GFP_COMP | GFP_ATOMIC,
2546 if (unlikely(!page))
2550 orig = sg_page(sge);
2551 from = sg_virt(sge);
2552 to = page_address(page);
2553 memcpy(to, from, a);
2554 memcpy(to + a, from + a + pop, b);
2555 sg_set_page(sge, page, a + b, 0);
2559 } else if (pop >= sge->length - a) {
2561 pop -= (sge->length - a);
2565 /* From above the current layout _must_ be as follows,
2570 * |---- pop ---|---------------- b ------------|
2571 * |____________________________________________| length
2573 * Offset and start of the current msg elem are equal because in the
2574 * previous case we handled offset != start and either consumed the
2575 * entire element and advanced to the next element OR pop == 0.
2577 * Two cases to handle here are first pop is less than the length
2578 * leaving some remainder b above. Simply adjust the element's layout
2579 * in this case. Or pop >= length of the element so that b = 0. In this
2580 * case advance to next element decrementing pop.
2583 struct scatterlist *sge = sk_msg_elem(msg, i);
2585 if (pop < sge->length) {
2591 sk_msg_shift_left(msg, i);
2593 sk_msg_iter_var_next(i);
2596 sk_mem_uncharge(msg->sk, len - pop);
2597 msg->sg.size -= (len - pop);
2598 sk_msg_compute_data_pointers(msg);
2602 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
2603 .func = bpf_msg_pop_data,
2605 .ret_type = RET_INTEGER,
2606 .arg1_type = ARG_PTR_TO_CTX,
2607 .arg2_type = ARG_ANYTHING,
2608 .arg3_type = ARG_ANYTHING,
2609 .arg4_type = ARG_ANYTHING,
2612 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
2614 return task_get_classid(skb);
2617 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
2618 .func = bpf_get_cgroup_classid,
2620 .ret_type = RET_INTEGER,
2621 .arg1_type = ARG_PTR_TO_CTX,
2624 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
2626 return dst_tclassid(skb);
2629 static const struct bpf_func_proto bpf_get_route_realm_proto = {
2630 .func = bpf_get_route_realm,
2632 .ret_type = RET_INTEGER,
2633 .arg1_type = ARG_PTR_TO_CTX,
2636 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
2638 /* If skb_clear_hash() was called due to mangling, we can
2639 * trigger SW recalculation here. Later access to hash
2640 * can then use the inline skb->hash via context directly
2641 * instead of calling this helper again.
2643 return skb_get_hash(skb);
2646 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
2647 .func = bpf_get_hash_recalc,
2649 .ret_type = RET_INTEGER,
2650 .arg1_type = ARG_PTR_TO_CTX,
2653 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
2655 /* After all direct packet write, this can be used once for
2656 * triggering a lazy recalc on next skb_get_hash() invocation.
2658 skb_clear_hash(skb);
2662 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
2663 .func = bpf_set_hash_invalid,
2665 .ret_type = RET_INTEGER,
2666 .arg1_type = ARG_PTR_TO_CTX,
2669 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
2671 /* Set user specified hash as L4(+), so that it gets returned
2672 * on skb_get_hash() call unless BPF prog later on triggers a
2675 __skb_set_sw_hash(skb, hash, true);
2679 static const struct bpf_func_proto bpf_set_hash_proto = {
2680 .func = bpf_set_hash,
2682 .ret_type = RET_INTEGER,
2683 .arg1_type = ARG_PTR_TO_CTX,
2684 .arg2_type = ARG_ANYTHING,
2687 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
2692 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
2693 vlan_proto != htons(ETH_P_8021AD)))
2694 vlan_proto = htons(ETH_P_8021Q);
2696 bpf_push_mac_rcsum(skb);
2697 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
2698 bpf_pull_mac_rcsum(skb);
2700 bpf_compute_data_pointers(skb);
2704 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
2705 .func = bpf_skb_vlan_push,
2707 .ret_type = RET_INTEGER,
2708 .arg1_type = ARG_PTR_TO_CTX,
2709 .arg2_type = ARG_ANYTHING,
2710 .arg3_type = ARG_ANYTHING,
2713 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
2717 bpf_push_mac_rcsum(skb);
2718 ret = skb_vlan_pop(skb);
2719 bpf_pull_mac_rcsum(skb);
2721 bpf_compute_data_pointers(skb);
2725 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
2726 .func = bpf_skb_vlan_pop,
2728 .ret_type = RET_INTEGER,
2729 .arg1_type = ARG_PTR_TO_CTX,
2732 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
2734 /* Caller already did skb_cow() with len as headroom,
2735 * so no need to do it here.
2738 memmove(skb->data, skb->data + len, off);
2739 memset(skb->data + off, 0, len);
2741 /* No skb_postpush_rcsum(skb, skb->data + off, len)
2742 * needed here as it does not change the skb->csum
2743 * result for checksum complete when summing over
2749 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
2751 /* skb_ensure_writable() is not needed here, as we're
2752 * already working on an uncloned skb.
2754 if (unlikely(!pskb_may_pull(skb, off + len)))
2757 skb_postpull_rcsum(skb, skb->data + off, len);
2758 memmove(skb->data + len, skb->data, off);
2759 __skb_pull(skb, len);
2764 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
2766 bool trans_same = skb->transport_header == skb->network_header;
2769 /* There's no need for __skb_push()/__skb_pull() pair to
2770 * get to the start of the mac header as we're guaranteed
2771 * to always start from here under eBPF.
2773 ret = bpf_skb_generic_push(skb, off, len);
2775 skb->mac_header -= len;
2776 skb->network_header -= len;
2778 skb->transport_header = skb->network_header;
2784 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
2786 bool trans_same = skb->transport_header == skb->network_header;
2789 /* Same here, __skb_push()/__skb_pull() pair not needed. */
2790 ret = bpf_skb_generic_pop(skb, off, len);
2792 skb->mac_header += len;
2793 skb->network_header += len;
2795 skb->transport_header = skb->network_header;
2801 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
2803 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2804 u32 off = skb_mac_header_len(skb);
2807 if (!skb_is_gso_tcp(skb))
2810 ret = skb_cow(skb, len_diff);
2811 if (unlikely(ret < 0))
2814 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2815 if (unlikely(ret < 0))
2818 if (skb_is_gso(skb)) {
2819 struct skb_shared_info *shinfo = skb_shinfo(skb);
2821 /* SKB_GSO_TCPV4 needs to be changed into
2824 if (shinfo->gso_type & SKB_GSO_TCPV4) {
2825 shinfo->gso_type &= ~SKB_GSO_TCPV4;
2826 shinfo->gso_type |= SKB_GSO_TCPV6;
2829 /* Due to IPv6 header, MSS needs to be downgraded. */
2830 skb_decrease_gso_size(shinfo, len_diff);
2831 /* Header must be checked, and gso_segs recomputed. */
2832 shinfo->gso_type |= SKB_GSO_DODGY;
2833 shinfo->gso_segs = 0;
2836 skb->protocol = htons(ETH_P_IPV6);
2837 skb_clear_hash(skb);
2842 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
2844 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2845 u32 off = skb_mac_header_len(skb);
2848 if (!skb_is_gso_tcp(skb))
2851 ret = skb_unclone(skb, GFP_ATOMIC);
2852 if (unlikely(ret < 0))
2855 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2856 if (unlikely(ret < 0))
2859 if (skb_is_gso(skb)) {
2860 struct skb_shared_info *shinfo = skb_shinfo(skb);
2862 /* SKB_GSO_TCPV6 needs to be changed into
2865 if (shinfo->gso_type & SKB_GSO_TCPV6) {
2866 shinfo->gso_type &= ~SKB_GSO_TCPV6;
2867 shinfo->gso_type |= SKB_GSO_TCPV4;
2870 /* Due to IPv4 header, MSS can be upgraded. */
2871 skb_increase_gso_size(shinfo, len_diff);
2872 /* Header must be checked, and gso_segs recomputed. */
2873 shinfo->gso_type |= SKB_GSO_DODGY;
2874 shinfo->gso_segs = 0;
2877 skb->protocol = htons(ETH_P_IP);
2878 skb_clear_hash(skb);
2883 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
2885 __be16 from_proto = skb->protocol;
2887 if (from_proto == htons(ETH_P_IP) &&
2888 to_proto == htons(ETH_P_IPV6))
2889 return bpf_skb_proto_4_to_6(skb);
2891 if (from_proto == htons(ETH_P_IPV6) &&
2892 to_proto == htons(ETH_P_IP))
2893 return bpf_skb_proto_6_to_4(skb);
2898 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
2903 if (unlikely(flags))
2906 /* General idea is that this helper does the basic groundwork
2907 * needed for changing the protocol, and eBPF program fills the
2908 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
2909 * and other helpers, rather than passing a raw buffer here.
2911 * The rationale is to keep this minimal and without a need to
2912 * deal with raw packet data. F.e. even if we would pass buffers
2913 * here, the program still needs to call the bpf_lX_csum_replace()
2914 * helpers anyway. Plus, this way we keep also separation of
2915 * concerns, since f.e. bpf_skb_store_bytes() should only take
2918 * Currently, additional options and extension header space are
2919 * not supported, but flags register is reserved so we can adapt
2920 * that. For offloads, we mark packet as dodgy, so that headers
2921 * need to be verified first.
2923 ret = bpf_skb_proto_xlat(skb, proto);
2924 bpf_compute_data_pointers(skb);
2928 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
2929 .func = bpf_skb_change_proto,
2931 .ret_type = RET_INTEGER,
2932 .arg1_type = ARG_PTR_TO_CTX,
2933 .arg2_type = ARG_ANYTHING,
2934 .arg3_type = ARG_ANYTHING,
2937 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
2939 /* We only allow a restricted subset to be changed for now. */
2940 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
2941 !skb_pkt_type_ok(pkt_type)))
2944 skb->pkt_type = pkt_type;
2948 static const struct bpf_func_proto bpf_skb_change_type_proto = {
2949 .func = bpf_skb_change_type,
2951 .ret_type = RET_INTEGER,
2952 .arg1_type = ARG_PTR_TO_CTX,
2953 .arg2_type = ARG_ANYTHING,
2956 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
2958 switch (skb->protocol) {
2959 case htons(ETH_P_IP):
2960 return sizeof(struct iphdr);
2961 case htons(ETH_P_IPV6):
2962 return sizeof(struct ipv6hdr);
2968 static int bpf_skb_net_grow(struct sk_buff *skb, u32 len_diff)
2970 u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
2973 if (!skb_is_gso_tcp(skb))
2976 ret = skb_cow(skb, len_diff);
2977 if (unlikely(ret < 0))
2980 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2981 if (unlikely(ret < 0))
2984 if (skb_is_gso(skb)) {
2985 struct skb_shared_info *shinfo = skb_shinfo(skb);
2987 /* Due to header grow, MSS needs to be downgraded. */
2988 skb_decrease_gso_size(shinfo, len_diff);
2989 /* Header must be checked, and gso_segs recomputed. */
2990 shinfo->gso_type |= SKB_GSO_DODGY;
2991 shinfo->gso_segs = 0;
2997 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 len_diff)
2999 u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
3002 if (!skb_is_gso_tcp(skb))
3005 ret = skb_unclone(skb, GFP_ATOMIC);
3006 if (unlikely(ret < 0))
3009 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3010 if (unlikely(ret < 0))
3013 if (skb_is_gso(skb)) {
3014 struct skb_shared_info *shinfo = skb_shinfo(skb);
3016 /* Due to header shrink, MSS can be upgraded. */
3017 skb_increase_gso_size(shinfo, len_diff);
3018 /* Header must be checked, and gso_segs recomputed. */
3019 shinfo->gso_type |= SKB_GSO_DODGY;
3020 shinfo->gso_segs = 0;
3026 static u32 __bpf_skb_max_len(const struct sk_buff *skb)
3028 return skb->dev ? skb->dev->mtu + skb->dev->hard_header_len :
3032 static int bpf_skb_adjust_net(struct sk_buff *skb, s32 len_diff)
3034 bool trans_same = skb->transport_header == skb->network_header;
3035 u32 len_cur, len_diff_abs = abs(len_diff);
3036 u32 len_min = bpf_skb_net_base_len(skb);
3037 u32 len_max = __bpf_skb_max_len(skb);
3038 __be16 proto = skb->protocol;
3039 bool shrink = len_diff < 0;
3042 if (unlikely(len_diff_abs > 0xfffU))
3044 if (unlikely(proto != htons(ETH_P_IP) &&
3045 proto != htons(ETH_P_IPV6)))
3048 len_cur = skb->len - skb_network_offset(skb);
3049 if (skb_transport_header_was_set(skb) && !trans_same)
3050 len_cur = skb_network_header_len(skb);
3051 if ((shrink && (len_diff_abs >= len_cur ||
3052 len_cur - len_diff_abs < len_min)) ||
3053 (!shrink && (skb->len + len_diff_abs > len_max &&
3057 ret = shrink ? bpf_skb_net_shrink(skb, len_diff_abs) :
3058 bpf_skb_net_grow(skb, len_diff_abs);
3060 bpf_compute_data_pointers(skb);
3064 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3065 u32, mode, u64, flags)
3067 if (unlikely(flags))
3069 if (likely(mode == BPF_ADJ_ROOM_NET))
3070 return bpf_skb_adjust_net(skb, len_diff);
3075 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3076 .func = bpf_skb_adjust_room,
3078 .ret_type = RET_INTEGER,
3079 .arg1_type = ARG_PTR_TO_CTX,
3080 .arg2_type = ARG_ANYTHING,
3081 .arg3_type = ARG_ANYTHING,
3082 .arg4_type = ARG_ANYTHING,
3085 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3087 u32 min_len = skb_network_offset(skb);
3089 if (skb_transport_header_was_set(skb))
3090 min_len = skb_transport_offset(skb);
3091 if (skb->ip_summed == CHECKSUM_PARTIAL)
3092 min_len = skb_checksum_start_offset(skb) +
3093 skb->csum_offset + sizeof(__sum16);
3097 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3099 unsigned int old_len = skb->len;
3102 ret = __skb_grow_rcsum(skb, new_len);
3104 memset(skb->data + old_len, 0, new_len - old_len);
3108 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3110 return __skb_trim_rcsum(skb, new_len);
3113 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3116 u32 max_len = __bpf_skb_max_len(skb);
3117 u32 min_len = __bpf_skb_min_len(skb);
3120 if (unlikely(flags || new_len > max_len || new_len < min_len))
3122 if (skb->encapsulation)
3125 /* The basic idea of this helper is that it's performing the
3126 * needed work to either grow or trim an skb, and eBPF program
3127 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3128 * bpf_lX_csum_replace() and others rather than passing a raw
3129 * buffer here. This one is a slow path helper and intended
3130 * for replies with control messages.
3132 * Like in bpf_skb_change_proto(), we want to keep this rather
3133 * minimal and without protocol specifics so that we are able
3134 * to separate concerns as in bpf_skb_store_bytes() should only
3135 * be the one responsible for writing buffers.
3137 * It's really expected to be a slow path operation here for
3138 * control message replies, so we're implicitly linearizing,
3139 * uncloning and drop offloads from the skb by this.
3141 ret = __bpf_try_make_writable(skb, skb->len);
3143 if (new_len > skb->len)
3144 ret = bpf_skb_grow_rcsum(skb, new_len);
3145 else if (new_len < skb->len)
3146 ret = bpf_skb_trim_rcsum(skb, new_len);
3147 if (!ret && skb_is_gso(skb))
3153 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3156 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3158 bpf_compute_data_pointers(skb);
3162 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3163 .func = bpf_skb_change_tail,
3165 .ret_type = RET_INTEGER,
3166 .arg1_type = ARG_PTR_TO_CTX,
3167 .arg2_type = ARG_ANYTHING,
3168 .arg3_type = ARG_ANYTHING,
3171 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3174 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3176 bpf_compute_data_end_sk_skb(skb);
3180 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3181 .func = sk_skb_change_tail,
3183 .ret_type = RET_INTEGER,
3184 .arg1_type = ARG_PTR_TO_CTX,
3185 .arg2_type = ARG_ANYTHING,
3186 .arg3_type = ARG_ANYTHING,
3189 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3192 u32 max_len = __bpf_skb_max_len(skb);
3193 u32 new_len = skb->len + head_room;
3196 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3197 new_len < skb->len))
3200 ret = skb_cow(skb, head_room);
3202 /* Idea for this helper is that we currently only
3203 * allow to expand on mac header. This means that
3204 * skb->protocol network header, etc, stay as is.
3205 * Compared to bpf_skb_change_tail(), we're more
3206 * flexible due to not needing to linearize or
3207 * reset GSO. Intention for this helper is to be
3208 * used by an L3 skb that needs to push mac header
3209 * for redirection into L2 device.
3211 __skb_push(skb, head_room);
3212 memset(skb->data, 0, head_room);
3213 skb_reset_mac_header(skb);
3219 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3222 int ret = __bpf_skb_change_head(skb, head_room, flags);
3224 bpf_compute_data_pointers(skb);
3228 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3229 .func = bpf_skb_change_head,
3231 .ret_type = RET_INTEGER,
3232 .arg1_type = ARG_PTR_TO_CTX,
3233 .arg2_type = ARG_ANYTHING,
3234 .arg3_type = ARG_ANYTHING,
3237 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3240 int ret = __bpf_skb_change_head(skb, head_room, flags);
3242 bpf_compute_data_end_sk_skb(skb);
3246 static const struct bpf_func_proto sk_skb_change_head_proto = {
3247 .func = sk_skb_change_head,
3249 .ret_type = RET_INTEGER,
3250 .arg1_type = ARG_PTR_TO_CTX,
3251 .arg2_type = ARG_ANYTHING,
3252 .arg3_type = ARG_ANYTHING,
3254 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3256 return xdp_data_meta_unsupported(xdp) ? 0 :
3257 xdp->data - xdp->data_meta;
3260 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3262 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3263 unsigned long metalen = xdp_get_metalen(xdp);
3264 void *data_start = xdp_frame_end + metalen;
3265 void *data = xdp->data + offset;
3267 if (unlikely(data < data_start ||
3268 data > xdp->data_end - ETH_HLEN))
3272 memmove(xdp->data_meta + offset,
3273 xdp->data_meta, metalen);
3274 xdp->data_meta += offset;
3280 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3281 .func = bpf_xdp_adjust_head,
3283 .ret_type = RET_INTEGER,
3284 .arg1_type = ARG_PTR_TO_CTX,
3285 .arg2_type = ARG_ANYTHING,
3288 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3290 void *data_end = xdp->data_end + offset;
3292 /* only shrinking is allowed for now. */
3293 if (unlikely(offset >= 0))
3296 if (unlikely(data_end < xdp->data + ETH_HLEN))
3299 xdp->data_end = data_end;
3304 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3305 .func = bpf_xdp_adjust_tail,
3307 .ret_type = RET_INTEGER,
3308 .arg1_type = ARG_PTR_TO_CTX,
3309 .arg2_type = ARG_ANYTHING,
3312 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3314 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3315 void *meta = xdp->data_meta + offset;
3316 unsigned long metalen = xdp->data - meta;
3318 if (xdp_data_meta_unsupported(xdp))
3320 if (unlikely(meta < xdp_frame_end ||
3323 if (unlikely((metalen & (sizeof(__u32) - 1)) ||
3327 xdp->data_meta = meta;
3332 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3333 .func = bpf_xdp_adjust_meta,
3335 .ret_type = RET_INTEGER,
3336 .arg1_type = ARG_PTR_TO_CTX,
3337 .arg2_type = ARG_ANYTHING,
3340 static int __bpf_tx_xdp(struct net_device *dev,
3341 struct bpf_map *map,
3342 struct xdp_buff *xdp,
3345 struct xdp_frame *xdpf;
3348 if (!dev->netdev_ops->ndo_xdp_xmit) {
3352 err = xdp_ok_fwd_dev(dev, xdp->data_end - xdp->data);
3356 xdpf = convert_to_xdp_frame(xdp);
3357 if (unlikely(!xdpf))
3360 sent = dev->netdev_ops->ndo_xdp_xmit(dev, 1, &xdpf, XDP_XMIT_FLUSH);
3367 xdp_do_redirect_slow(struct net_device *dev, struct xdp_buff *xdp,
3368 struct bpf_prog *xdp_prog, struct bpf_redirect_info *ri)
3370 struct net_device *fwd;
3371 u32 index = ri->ifindex;
3374 fwd = dev_get_by_index_rcu(dev_net(dev), index);
3376 if (unlikely(!fwd)) {
3381 err = __bpf_tx_xdp(fwd, NULL, xdp, 0);
3385 _trace_xdp_redirect(dev, xdp_prog, index);
3388 _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3392 static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd,
3393 struct bpf_map *map,
3394 struct xdp_buff *xdp,
3399 switch (map->map_type) {
3400 case BPF_MAP_TYPE_DEVMAP: {
3401 struct bpf_dtab_netdev *dst = fwd;
3403 err = dev_map_enqueue(dst, xdp, dev_rx);
3406 __dev_map_insert_ctx(map, index);
3409 case BPF_MAP_TYPE_CPUMAP: {
3410 struct bpf_cpu_map_entry *rcpu = fwd;
3412 err = cpu_map_enqueue(rcpu, xdp, dev_rx);
3415 __cpu_map_insert_ctx(map, index);
3418 case BPF_MAP_TYPE_XSKMAP: {
3419 struct xdp_sock *xs = fwd;
3421 err = __xsk_map_redirect(map, xdp, xs);
3430 void xdp_do_flush_map(void)
3432 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3433 struct bpf_map *map = ri->map_to_flush;
3435 ri->map_to_flush = NULL;
3437 switch (map->map_type) {
3438 case BPF_MAP_TYPE_DEVMAP:
3439 __dev_map_flush(map);
3441 case BPF_MAP_TYPE_CPUMAP:
3442 __cpu_map_flush(map);
3444 case BPF_MAP_TYPE_XSKMAP:
3445 __xsk_map_flush(map);
3452 EXPORT_SYMBOL_GPL(xdp_do_flush_map);
3454 static inline void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index)
3456 switch (map->map_type) {
3457 case BPF_MAP_TYPE_DEVMAP:
3458 return __dev_map_lookup_elem(map, index);
3459 case BPF_MAP_TYPE_CPUMAP:
3460 return __cpu_map_lookup_elem(map, index);
3461 case BPF_MAP_TYPE_XSKMAP:
3462 return __xsk_map_lookup_elem(map, index);
3468 void bpf_clear_redirect_map(struct bpf_map *map)
3470 struct bpf_redirect_info *ri;
3473 for_each_possible_cpu(cpu) {
3474 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3475 /* Avoid polluting remote cacheline due to writes if
3476 * not needed. Once we pass this test, we need the
3477 * cmpxchg() to make sure it hasn't been changed in
3478 * the meantime by remote CPU.
3480 if (unlikely(READ_ONCE(ri->map) == map))
3481 cmpxchg(&ri->map, map, NULL);
3485 static int xdp_do_redirect_map(struct net_device *dev, struct xdp_buff *xdp,
3486 struct bpf_prog *xdp_prog, struct bpf_map *map,
3487 struct bpf_redirect_info *ri)
3489 u32 index = ri->ifindex;
3494 WRITE_ONCE(ri->map, NULL);
3496 fwd = __xdp_map_lookup_elem(map, index);
3497 if (unlikely(!fwd)) {
3501 if (ri->map_to_flush && unlikely(ri->map_to_flush != map))
3504 err = __bpf_tx_xdp_map(dev, fwd, map, xdp, index);
3508 ri->map_to_flush = map;
3509 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3512 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3516 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3517 struct bpf_prog *xdp_prog)
3519 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3520 struct bpf_map *map = READ_ONCE(ri->map);
3523 return xdp_do_redirect_map(dev, xdp, xdp_prog, map, ri);
3525 return xdp_do_redirect_slow(dev, xdp, xdp_prog, ri);
3527 EXPORT_SYMBOL_GPL(xdp_do_redirect);
3529 static int xdp_do_generic_redirect_map(struct net_device *dev,
3530 struct sk_buff *skb,
3531 struct xdp_buff *xdp,
3532 struct bpf_prog *xdp_prog,
3533 struct bpf_map *map)
3535 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3536 u32 index = ri->ifindex;
3541 WRITE_ONCE(ri->map, NULL);
3543 fwd = __xdp_map_lookup_elem(map, index);
3544 if (unlikely(!fwd)) {
3549 if (map->map_type == BPF_MAP_TYPE_DEVMAP) {
3550 struct bpf_dtab_netdev *dst = fwd;
3552 err = dev_map_generic_redirect(dst, skb, xdp_prog);
3555 } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) {
3556 struct xdp_sock *xs = fwd;
3558 err = xsk_generic_rcv(xs, xdp);
3563 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
3568 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3571 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3575 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
3576 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
3578 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3579 struct bpf_map *map = READ_ONCE(ri->map);
3580 u32 index = ri->ifindex;
3581 struct net_device *fwd;
3585 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog,
3588 fwd = dev_get_by_index_rcu(dev_net(dev), index);
3589 if (unlikely(!fwd)) {
3594 err = xdp_ok_fwd_dev(fwd, skb->len);
3599 _trace_xdp_redirect(dev, xdp_prog, index);
3600 generic_xdp_tx(skb, xdp_prog);
3603 _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3606 EXPORT_SYMBOL_GPL(xdp_do_generic_redirect);
3608 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
3610 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3612 if (unlikely(flags))
3615 ri->ifindex = ifindex;
3617 WRITE_ONCE(ri->map, NULL);
3619 return XDP_REDIRECT;
3622 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
3623 .func = bpf_xdp_redirect,
3625 .ret_type = RET_INTEGER,
3626 .arg1_type = ARG_ANYTHING,
3627 .arg2_type = ARG_ANYTHING,
3630 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
3633 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3635 if (unlikely(flags))
3638 ri->ifindex = ifindex;
3640 WRITE_ONCE(ri->map, map);
3642 return XDP_REDIRECT;
3645 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
3646 .func = bpf_xdp_redirect_map,
3648 .ret_type = RET_INTEGER,
3649 .arg1_type = ARG_CONST_MAP_PTR,
3650 .arg2_type = ARG_ANYTHING,
3651 .arg3_type = ARG_ANYTHING,
3654 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
3655 unsigned long off, unsigned long len)
3657 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
3661 if (ptr != dst_buff)
3662 memcpy(dst_buff, ptr, len);
3667 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
3668 u64, flags, void *, meta, u64, meta_size)
3670 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
3672 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
3674 if (unlikely(skb_size > skb->len))
3677 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
3681 static const struct bpf_func_proto bpf_skb_event_output_proto = {
3682 .func = bpf_skb_event_output,
3684 .ret_type = RET_INTEGER,
3685 .arg1_type = ARG_PTR_TO_CTX,
3686 .arg2_type = ARG_CONST_MAP_PTR,
3687 .arg3_type = ARG_ANYTHING,
3688 .arg4_type = ARG_PTR_TO_MEM,
3689 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
3692 static unsigned short bpf_tunnel_key_af(u64 flags)
3694 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
3697 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
3698 u32, size, u64, flags)
3700 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3701 u8 compat[sizeof(struct bpf_tunnel_key)];
3705 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
3709 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
3713 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3716 case offsetof(struct bpf_tunnel_key, tunnel_label):
3717 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3719 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3720 /* Fixup deprecated structure layouts here, so we have
3721 * a common path later on.
3723 if (ip_tunnel_info_af(info) != AF_INET)
3726 to = (struct bpf_tunnel_key *)compat;
3733 to->tunnel_id = be64_to_cpu(info->key.tun_id);
3734 to->tunnel_tos = info->key.tos;
3735 to->tunnel_ttl = info->key.ttl;
3738 if (flags & BPF_F_TUNINFO_IPV6) {
3739 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
3740 sizeof(to->remote_ipv6));
3741 to->tunnel_label = be32_to_cpu(info->key.label);
3743 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
3744 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
3745 to->tunnel_label = 0;
3748 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
3749 memcpy(to_orig, to, size);
3753 memset(to_orig, 0, size);
3757 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
3758 .func = bpf_skb_get_tunnel_key,
3760 .ret_type = RET_INTEGER,
3761 .arg1_type = ARG_PTR_TO_CTX,
3762 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
3763 .arg3_type = ARG_CONST_SIZE,
3764 .arg4_type = ARG_ANYTHING,
3767 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
3769 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3772 if (unlikely(!info ||
3773 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
3777 if (unlikely(size < info->options_len)) {
3782 ip_tunnel_info_opts_get(to, info);
3783 if (size > info->options_len)
3784 memset(to + info->options_len, 0, size - info->options_len);
3786 return info->options_len;
3788 memset(to, 0, size);
3792 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
3793 .func = bpf_skb_get_tunnel_opt,
3795 .ret_type = RET_INTEGER,
3796 .arg1_type = ARG_PTR_TO_CTX,
3797 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
3798 .arg3_type = ARG_CONST_SIZE,
3801 static struct metadata_dst __percpu *md_dst;
3803 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
3804 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
3806 struct metadata_dst *md = this_cpu_ptr(md_dst);
3807 u8 compat[sizeof(struct bpf_tunnel_key)];
3808 struct ip_tunnel_info *info;
3810 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
3811 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
3813 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3815 case offsetof(struct bpf_tunnel_key, tunnel_label):
3816 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3817 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3818 /* Fixup deprecated structure layouts here, so we have
3819 * a common path later on.
3821 memcpy(compat, from, size);
3822 memset(compat + size, 0, sizeof(compat) - size);
3823 from = (const struct bpf_tunnel_key *) compat;
3829 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
3834 dst_hold((struct dst_entry *) md);
3835 skb_dst_set(skb, (struct dst_entry *) md);
3837 info = &md->u.tun_info;
3838 memset(info, 0, sizeof(*info));
3839 info->mode = IP_TUNNEL_INFO_TX;
3841 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
3842 if (flags & BPF_F_DONT_FRAGMENT)
3843 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
3844 if (flags & BPF_F_ZERO_CSUM_TX)
3845 info->key.tun_flags &= ~TUNNEL_CSUM;
3846 if (flags & BPF_F_SEQ_NUMBER)
3847 info->key.tun_flags |= TUNNEL_SEQ;
3849 info->key.tun_id = cpu_to_be64(from->tunnel_id);
3850 info->key.tos = from->tunnel_tos;
3851 info->key.ttl = from->tunnel_ttl;
3853 if (flags & BPF_F_TUNINFO_IPV6) {
3854 info->mode |= IP_TUNNEL_INFO_IPV6;
3855 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
3856 sizeof(from->remote_ipv6));
3857 info->key.label = cpu_to_be32(from->tunnel_label) &
3858 IPV6_FLOWLABEL_MASK;
3860 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
3866 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
3867 .func = bpf_skb_set_tunnel_key,
3869 .ret_type = RET_INTEGER,
3870 .arg1_type = ARG_PTR_TO_CTX,
3871 .arg2_type = ARG_PTR_TO_MEM,
3872 .arg3_type = ARG_CONST_SIZE,
3873 .arg4_type = ARG_ANYTHING,
3876 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
3877 const u8 *, from, u32, size)
3879 struct ip_tunnel_info *info = skb_tunnel_info(skb);
3880 const struct metadata_dst *md = this_cpu_ptr(md_dst);
3882 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
3884 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
3887 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
3892 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
3893 .func = bpf_skb_set_tunnel_opt,
3895 .ret_type = RET_INTEGER,
3896 .arg1_type = ARG_PTR_TO_CTX,
3897 .arg2_type = ARG_PTR_TO_MEM,
3898 .arg3_type = ARG_CONST_SIZE,
3901 static const struct bpf_func_proto *
3902 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
3905 struct metadata_dst __percpu *tmp;
3907 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
3912 if (cmpxchg(&md_dst, NULL, tmp))
3913 metadata_dst_free_percpu(tmp);
3917 case BPF_FUNC_skb_set_tunnel_key:
3918 return &bpf_skb_set_tunnel_key_proto;
3919 case BPF_FUNC_skb_set_tunnel_opt:
3920 return &bpf_skb_set_tunnel_opt_proto;
3926 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
3929 struct bpf_array *array = container_of(map, struct bpf_array, map);
3930 struct cgroup *cgrp;
3933 sk = skb_to_full_sk(skb);
3934 if (!sk || !sk_fullsock(sk))
3936 if (unlikely(idx >= array->map.max_entries))
3939 cgrp = READ_ONCE(array->ptrs[idx]);
3940 if (unlikely(!cgrp))
3943 return sk_under_cgroup_hierarchy(sk, cgrp);
3946 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
3947 .func = bpf_skb_under_cgroup,
3949 .ret_type = RET_INTEGER,
3950 .arg1_type = ARG_PTR_TO_CTX,
3951 .arg2_type = ARG_CONST_MAP_PTR,
3952 .arg3_type = ARG_ANYTHING,
3955 #ifdef CONFIG_SOCK_CGROUP_DATA
3956 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
3958 struct sock *sk = skb_to_full_sk(skb);
3959 struct cgroup *cgrp;
3961 if (!sk || !sk_fullsock(sk))
3964 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
3965 return cgrp->kn->id.id;
3968 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
3969 .func = bpf_skb_cgroup_id,
3971 .ret_type = RET_INTEGER,
3972 .arg1_type = ARG_PTR_TO_CTX,
3975 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
3978 struct sock *sk = skb_to_full_sk(skb);
3979 struct cgroup *ancestor;
3980 struct cgroup *cgrp;
3982 if (!sk || !sk_fullsock(sk))
3985 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
3986 ancestor = cgroup_ancestor(cgrp, ancestor_level);
3990 return ancestor->kn->id.id;
3993 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
3994 .func = bpf_skb_ancestor_cgroup_id,
3996 .ret_type = RET_INTEGER,
3997 .arg1_type = ARG_PTR_TO_CTX,
3998 .arg2_type = ARG_ANYTHING,
4002 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4003 unsigned long off, unsigned long len)
4005 memcpy(dst_buff, src_buff + off, len);
4009 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4010 u64, flags, void *, meta, u64, meta_size)
4012 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4014 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4016 if (unlikely(xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4019 return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4020 xdp_size, bpf_xdp_copy);
4023 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4024 .func = bpf_xdp_event_output,
4026 .ret_type = RET_INTEGER,
4027 .arg1_type = ARG_PTR_TO_CTX,
4028 .arg2_type = ARG_CONST_MAP_PTR,
4029 .arg3_type = ARG_ANYTHING,
4030 .arg4_type = ARG_PTR_TO_MEM,
4031 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4034 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4036 return skb->sk ? sock_gen_cookie(skb->sk) : 0;
4039 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4040 .func = bpf_get_socket_cookie,
4042 .ret_type = RET_INTEGER,
4043 .arg1_type = ARG_PTR_TO_CTX,
4046 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4048 return sock_gen_cookie(ctx->sk);
4051 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4052 .func = bpf_get_socket_cookie_sock_addr,
4054 .ret_type = RET_INTEGER,
4055 .arg1_type = ARG_PTR_TO_CTX,
4058 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4060 return sock_gen_cookie(ctx->sk);
4063 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4064 .func = bpf_get_socket_cookie_sock_ops,
4066 .ret_type = RET_INTEGER,
4067 .arg1_type = ARG_PTR_TO_CTX,
4070 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4072 struct sock *sk = sk_to_full_sk(skb->sk);
4075 if (!sk || !sk_fullsock(sk))
4077 kuid = sock_net_uid(sock_net(sk), sk);
4078 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4081 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4082 .func = bpf_get_socket_uid,
4084 .ret_type = RET_INTEGER,
4085 .arg1_type = ARG_PTR_TO_CTX,
4088 BPF_CALL_5(bpf_sockopt_event_output, struct bpf_sock_ops_kern *, bpf_sock,
4089 struct bpf_map *, map, u64, flags, void *, data, u64, size)
4091 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
4094 return bpf_event_output(map, flags, data, size, NULL, 0, NULL);
4097 static const struct bpf_func_proto bpf_sockopt_event_output_proto = {
4098 .func = bpf_sockopt_event_output,
4100 .ret_type = RET_INTEGER,
4101 .arg1_type = ARG_PTR_TO_CTX,
4102 .arg2_type = ARG_CONST_MAP_PTR,
4103 .arg3_type = ARG_ANYTHING,
4104 .arg4_type = ARG_PTR_TO_MEM,
4105 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4108 BPF_CALL_5(bpf_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
4109 int, level, int, optname, char *, optval, int, optlen)
4111 struct sock *sk = bpf_sock->sk;
4115 if (!sk_fullsock(sk))
4118 if (level == SOL_SOCKET) {
4119 if (optlen != sizeof(int))
4121 val = *((int *)optval);
4123 /* Only some socketops are supported */
4126 val = min_t(u32, val, sysctl_rmem_max);
4127 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4128 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
4131 val = min_t(u32, val, sysctl_wmem_max);
4132 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4133 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
4135 case SO_MAX_PACING_RATE: /* 32bit version */
4137 cmpxchg(&sk->sk_pacing_status,
4140 sk->sk_max_pacing_rate = (val == ~0U) ? ~0UL : val;
4141 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4142 sk->sk_max_pacing_rate);
4145 sk->sk_priority = val;
4150 sk->sk_rcvlowat = val ? : 1;
4153 if (sk->sk_mark != val) {
4162 } else if (level == SOL_IP) {
4163 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4166 val = *((int *)optval);
4167 /* Only some options are supported */
4170 if (val < -1 || val > 0xff) {
4173 struct inet_sock *inet = inet_sk(sk);
4183 #if IS_ENABLED(CONFIG_IPV6)
4184 } else if (level == SOL_IPV6) {
4185 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4188 val = *((int *)optval);
4189 /* Only some options are supported */
4192 if (val < -1 || val > 0xff) {
4195 struct ipv6_pinfo *np = inet6_sk(sk);
4206 } else if (level == SOL_TCP &&
4207 sk->sk_prot->setsockopt == tcp_setsockopt) {
4208 if (optname == TCP_CONGESTION) {
4209 char name[TCP_CA_NAME_MAX];
4210 bool reinit = bpf_sock->op > BPF_SOCK_OPS_NEEDS_ECN;
4212 strncpy(name, optval, min_t(long, optlen,
4213 TCP_CA_NAME_MAX-1));
4214 name[TCP_CA_NAME_MAX-1] = 0;
4215 ret = tcp_set_congestion_control(sk, name, false,
4218 struct tcp_sock *tp = tcp_sk(sk);
4220 if (optlen != sizeof(int))
4223 val = *((int *)optval);
4224 /* Only some options are supported */
4227 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4232 case TCP_BPF_SNDCWND_CLAMP:
4236 tp->snd_cwnd_clamp = val;
4237 tp->snd_ssthresh = val;
4241 if (val < 0 || val > 1)
4257 static const struct bpf_func_proto bpf_setsockopt_proto = {
4258 .func = bpf_setsockopt,
4260 .ret_type = RET_INTEGER,
4261 .arg1_type = ARG_PTR_TO_CTX,
4262 .arg2_type = ARG_ANYTHING,
4263 .arg3_type = ARG_ANYTHING,
4264 .arg4_type = ARG_PTR_TO_MEM,
4265 .arg5_type = ARG_CONST_SIZE,
4268 BPF_CALL_5(bpf_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
4269 int, level, int, optname, char *, optval, int, optlen)
4271 struct sock *sk = bpf_sock->sk;
4273 if (!sk_fullsock(sk))
4276 if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4277 struct inet_connection_sock *icsk;
4278 struct tcp_sock *tp;
4281 case TCP_CONGESTION:
4282 icsk = inet_csk(sk);
4284 if (!icsk->icsk_ca_ops || optlen <= 1)
4286 strncpy(optval, icsk->icsk_ca_ops->name, optlen);
4287 optval[optlen - 1] = 0;
4292 if (optlen <= 0 || !tp->saved_syn ||
4293 optlen > tp->saved_syn[0])
4295 memcpy(optval, tp->saved_syn + 1, optlen);
4300 } else if (level == SOL_IP) {
4301 struct inet_sock *inet = inet_sk(sk);
4303 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4306 /* Only some options are supported */
4309 *((int *)optval) = (int)inet->tos;
4314 #if IS_ENABLED(CONFIG_IPV6)
4315 } else if (level == SOL_IPV6) {
4316 struct ipv6_pinfo *np = inet6_sk(sk);
4318 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4321 /* Only some options are supported */
4324 *((int *)optval) = (int)np->tclass;
4336 memset(optval, 0, optlen);
4340 static const struct bpf_func_proto bpf_getsockopt_proto = {
4341 .func = bpf_getsockopt,
4343 .ret_type = RET_INTEGER,
4344 .arg1_type = ARG_PTR_TO_CTX,
4345 .arg2_type = ARG_ANYTHING,
4346 .arg3_type = ARG_ANYTHING,
4347 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
4348 .arg5_type = ARG_CONST_SIZE,
4351 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
4354 struct sock *sk = bpf_sock->sk;
4355 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
4357 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
4361 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
4363 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
4366 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
4367 .func = bpf_sock_ops_cb_flags_set,
4369 .ret_type = RET_INTEGER,
4370 .arg1_type = ARG_PTR_TO_CTX,
4371 .arg2_type = ARG_ANYTHING,
4374 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
4375 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
4377 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
4381 struct sock *sk = ctx->sk;
4384 /* Binding to port can be expensive so it's prohibited in the helper.
4385 * Only binding to IP is supported.
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_SOCKET,
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 sk = sk_to_full_sk(sk);
5412 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
5413 return (unsigned long)sk;
5415 return (unsigned long)NULL;
5418 static const struct bpf_func_proto bpf_tcp_sock_proto = {
5419 .func = bpf_tcp_sock,
5421 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
5422 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
5425 #endif /* CONFIG_INET */
5427 bool bpf_helper_changes_pkt_data(void *func)
5429 if (func == bpf_skb_vlan_push ||
5430 func == bpf_skb_vlan_pop ||
5431 func == bpf_skb_store_bytes ||
5432 func == bpf_skb_change_proto ||
5433 func == bpf_skb_change_head ||
5434 func == sk_skb_change_head ||
5435 func == bpf_skb_change_tail ||
5436 func == sk_skb_change_tail ||
5437 func == bpf_skb_adjust_room ||
5438 func == bpf_skb_pull_data ||
5439 func == sk_skb_pull_data ||
5440 func == bpf_clone_redirect ||
5441 func == bpf_l3_csum_replace ||
5442 func == bpf_l4_csum_replace ||
5443 func == bpf_xdp_adjust_head ||
5444 func == bpf_xdp_adjust_meta ||
5445 func == bpf_msg_pull_data ||
5446 func == bpf_msg_push_data ||
5447 func == bpf_msg_pop_data ||
5448 func == bpf_xdp_adjust_tail ||
5449 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5450 func == bpf_lwt_seg6_store_bytes ||
5451 func == bpf_lwt_seg6_adjust_srh ||
5452 func == bpf_lwt_seg6_action ||
5454 func == bpf_lwt_in_push_encap ||
5455 func == bpf_lwt_xmit_push_encap)
5461 static const struct bpf_func_proto *
5462 bpf_base_func_proto(enum bpf_func_id func_id)
5465 case BPF_FUNC_map_lookup_elem:
5466 return &bpf_map_lookup_elem_proto;
5467 case BPF_FUNC_map_update_elem:
5468 return &bpf_map_update_elem_proto;
5469 case BPF_FUNC_map_delete_elem:
5470 return &bpf_map_delete_elem_proto;
5471 case BPF_FUNC_map_push_elem:
5472 return &bpf_map_push_elem_proto;
5473 case BPF_FUNC_map_pop_elem:
5474 return &bpf_map_pop_elem_proto;
5475 case BPF_FUNC_map_peek_elem:
5476 return &bpf_map_peek_elem_proto;
5477 case BPF_FUNC_get_prandom_u32:
5478 return &bpf_get_prandom_u32_proto;
5479 case BPF_FUNC_get_smp_processor_id:
5480 return &bpf_get_raw_smp_processor_id_proto;
5481 case BPF_FUNC_get_numa_node_id:
5482 return &bpf_get_numa_node_id_proto;
5483 case BPF_FUNC_tail_call:
5484 return &bpf_tail_call_proto;
5485 case BPF_FUNC_ktime_get_ns:
5486 return &bpf_ktime_get_ns_proto;
5491 if (!capable(CAP_SYS_ADMIN))
5495 case BPF_FUNC_spin_lock:
5496 return &bpf_spin_lock_proto;
5497 case BPF_FUNC_spin_unlock:
5498 return &bpf_spin_unlock_proto;
5499 case BPF_FUNC_trace_printk:
5500 return bpf_get_trace_printk_proto();
5506 static const struct bpf_func_proto *
5507 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5510 /* inet and inet6 sockets are created in a process
5511 * context so there is always a valid uid/gid
5513 case BPF_FUNC_get_current_uid_gid:
5514 return &bpf_get_current_uid_gid_proto;
5515 case BPF_FUNC_get_local_storage:
5516 return &bpf_get_local_storage_proto;
5518 return bpf_base_func_proto(func_id);
5522 static const struct bpf_func_proto *
5523 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5526 /* inet and inet6 sockets are created in a process
5527 * context so there is always a valid uid/gid
5529 case BPF_FUNC_get_current_uid_gid:
5530 return &bpf_get_current_uid_gid_proto;
5532 switch (prog->expected_attach_type) {
5533 case BPF_CGROUP_INET4_CONNECT:
5534 case BPF_CGROUP_INET6_CONNECT:
5535 return &bpf_bind_proto;
5539 case BPF_FUNC_get_socket_cookie:
5540 return &bpf_get_socket_cookie_sock_addr_proto;
5541 case BPF_FUNC_get_local_storage:
5542 return &bpf_get_local_storage_proto;
5544 case BPF_FUNC_sk_lookup_tcp:
5545 return &bpf_sock_addr_sk_lookup_tcp_proto;
5546 case BPF_FUNC_sk_lookup_udp:
5547 return &bpf_sock_addr_sk_lookup_udp_proto;
5548 case BPF_FUNC_sk_release:
5549 return &bpf_sk_release_proto;
5550 #endif /* CONFIG_INET */
5552 return bpf_base_func_proto(func_id);
5556 static const struct bpf_func_proto *
5557 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5560 case BPF_FUNC_skb_load_bytes:
5561 return &bpf_skb_load_bytes_proto;
5562 case BPF_FUNC_skb_load_bytes_relative:
5563 return &bpf_skb_load_bytes_relative_proto;
5564 case BPF_FUNC_get_socket_cookie:
5565 return &bpf_get_socket_cookie_proto;
5566 case BPF_FUNC_get_socket_uid:
5567 return &bpf_get_socket_uid_proto;
5569 return bpf_base_func_proto(func_id);
5573 static const struct bpf_func_proto *
5574 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5577 case BPF_FUNC_get_local_storage:
5578 return &bpf_get_local_storage_proto;
5579 case BPF_FUNC_sk_fullsock:
5580 return &bpf_sk_fullsock_proto;
5582 case BPF_FUNC_tcp_sock:
5583 return &bpf_tcp_sock_proto;
5586 return sk_filter_func_proto(func_id, prog);
5590 static const struct bpf_func_proto *
5591 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5594 case BPF_FUNC_skb_store_bytes:
5595 return &bpf_skb_store_bytes_proto;
5596 case BPF_FUNC_skb_load_bytes:
5597 return &bpf_skb_load_bytes_proto;
5598 case BPF_FUNC_skb_load_bytes_relative:
5599 return &bpf_skb_load_bytes_relative_proto;
5600 case BPF_FUNC_skb_pull_data:
5601 return &bpf_skb_pull_data_proto;
5602 case BPF_FUNC_csum_diff:
5603 return &bpf_csum_diff_proto;
5604 case BPF_FUNC_csum_update:
5605 return &bpf_csum_update_proto;
5606 case BPF_FUNC_l3_csum_replace:
5607 return &bpf_l3_csum_replace_proto;
5608 case BPF_FUNC_l4_csum_replace:
5609 return &bpf_l4_csum_replace_proto;
5610 case BPF_FUNC_clone_redirect:
5611 return &bpf_clone_redirect_proto;
5612 case BPF_FUNC_get_cgroup_classid:
5613 return &bpf_get_cgroup_classid_proto;
5614 case BPF_FUNC_skb_vlan_push:
5615 return &bpf_skb_vlan_push_proto;
5616 case BPF_FUNC_skb_vlan_pop:
5617 return &bpf_skb_vlan_pop_proto;
5618 case BPF_FUNC_skb_change_proto:
5619 return &bpf_skb_change_proto_proto;
5620 case BPF_FUNC_skb_change_type:
5621 return &bpf_skb_change_type_proto;
5622 case BPF_FUNC_skb_adjust_room:
5623 return &bpf_skb_adjust_room_proto;
5624 case BPF_FUNC_skb_change_tail:
5625 return &bpf_skb_change_tail_proto;
5626 case BPF_FUNC_skb_get_tunnel_key:
5627 return &bpf_skb_get_tunnel_key_proto;
5628 case BPF_FUNC_skb_set_tunnel_key:
5629 return bpf_get_skb_set_tunnel_proto(func_id);
5630 case BPF_FUNC_skb_get_tunnel_opt:
5631 return &bpf_skb_get_tunnel_opt_proto;
5632 case BPF_FUNC_skb_set_tunnel_opt:
5633 return bpf_get_skb_set_tunnel_proto(func_id);
5634 case BPF_FUNC_redirect:
5635 return &bpf_redirect_proto;
5636 case BPF_FUNC_get_route_realm:
5637 return &bpf_get_route_realm_proto;
5638 case BPF_FUNC_get_hash_recalc:
5639 return &bpf_get_hash_recalc_proto;
5640 case BPF_FUNC_set_hash_invalid:
5641 return &bpf_set_hash_invalid_proto;
5642 case BPF_FUNC_set_hash:
5643 return &bpf_set_hash_proto;
5644 case BPF_FUNC_perf_event_output:
5645 return &bpf_skb_event_output_proto;
5646 case BPF_FUNC_get_smp_processor_id:
5647 return &bpf_get_smp_processor_id_proto;
5648 case BPF_FUNC_skb_under_cgroup:
5649 return &bpf_skb_under_cgroup_proto;
5650 case BPF_FUNC_get_socket_cookie:
5651 return &bpf_get_socket_cookie_proto;
5652 case BPF_FUNC_get_socket_uid:
5653 return &bpf_get_socket_uid_proto;
5654 case BPF_FUNC_fib_lookup:
5655 return &bpf_skb_fib_lookup_proto;
5656 case BPF_FUNC_sk_fullsock:
5657 return &bpf_sk_fullsock_proto;
5659 case BPF_FUNC_skb_get_xfrm_state:
5660 return &bpf_skb_get_xfrm_state_proto;
5662 #ifdef CONFIG_SOCK_CGROUP_DATA
5663 case BPF_FUNC_skb_cgroup_id:
5664 return &bpf_skb_cgroup_id_proto;
5665 case BPF_FUNC_skb_ancestor_cgroup_id:
5666 return &bpf_skb_ancestor_cgroup_id_proto;
5669 case BPF_FUNC_sk_lookup_tcp:
5670 return &bpf_sk_lookup_tcp_proto;
5671 case BPF_FUNC_sk_lookup_udp:
5672 return &bpf_sk_lookup_udp_proto;
5673 case BPF_FUNC_sk_release:
5674 return &bpf_sk_release_proto;
5675 case BPF_FUNC_tcp_sock:
5676 return &bpf_tcp_sock_proto;
5679 return bpf_base_func_proto(func_id);
5683 static const struct bpf_func_proto *
5684 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5687 case BPF_FUNC_perf_event_output:
5688 return &bpf_xdp_event_output_proto;
5689 case BPF_FUNC_get_smp_processor_id:
5690 return &bpf_get_smp_processor_id_proto;
5691 case BPF_FUNC_csum_diff:
5692 return &bpf_csum_diff_proto;
5693 case BPF_FUNC_xdp_adjust_head:
5694 return &bpf_xdp_adjust_head_proto;
5695 case BPF_FUNC_xdp_adjust_meta:
5696 return &bpf_xdp_adjust_meta_proto;
5697 case BPF_FUNC_redirect:
5698 return &bpf_xdp_redirect_proto;
5699 case BPF_FUNC_redirect_map:
5700 return &bpf_xdp_redirect_map_proto;
5701 case BPF_FUNC_xdp_adjust_tail:
5702 return &bpf_xdp_adjust_tail_proto;
5703 case BPF_FUNC_fib_lookup:
5704 return &bpf_xdp_fib_lookup_proto;
5706 case BPF_FUNC_sk_lookup_udp:
5707 return &bpf_xdp_sk_lookup_udp_proto;
5708 case BPF_FUNC_sk_lookup_tcp:
5709 return &bpf_xdp_sk_lookup_tcp_proto;
5710 case BPF_FUNC_sk_release:
5711 return &bpf_sk_release_proto;
5714 return bpf_base_func_proto(func_id);
5718 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
5719 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
5721 static const struct bpf_func_proto *
5722 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5725 case BPF_FUNC_setsockopt:
5726 return &bpf_setsockopt_proto;
5727 case BPF_FUNC_getsockopt:
5728 return &bpf_getsockopt_proto;
5729 case BPF_FUNC_sock_ops_cb_flags_set:
5730 return &bpf_sock_ops_cb_flags_set_proto;
5731 case BPF_FUNC_sock_map_update:
5732 return &bpf_sock_map_update_proto;
5733 case BPF_FUNC_sock_hash_update:
5734 return &bpf_sock_hash_update_proto;
5735 case BPF_FUNC_get_socket_cookie:
5736 return &bpf_get_socket_cookie_sock_ops_proto;
5737 case BPF_FUNC_get_local_storage:
5738 return &bpf_get_local_storage_proto;
5739 case BPF_FUNC_perf_event_output:
5740 return &bpf_sockopt_event_output_proto;
5742 return bpf_base_func_proto(func_id);
5746 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
5747 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
5749 static const struct bpf_func_proto *
5750 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5753 case BPF_FUNC_msg_redirect_map:
5754 return &bpf_msg_redirect_map_proto;
5755 case BPF_FUNC_msg_redirect_hash:
5756 return &bpf_msg_redirect_hash_proto;
5757 case BPF_FUNC_msg_apply_bytes:
5758 return &bpf_msg_apply_bytes_proto;
5759 case BPF_FUNC_msg_cork_bytes:
5760 return &bpf_msg_cork_bytes_proto;
5761 case BPF_FUNC_msg_pull_data:
5762 return &bpf_msg_pull_data_proto;
5763 case BPF_FUNC_msg_push_data:
5764 return &bpf_msg_push_data_proto;
5765 case BPF_FUNC_msg_pop_data:
5766 return &bpf_msg_pop_data_proto;
5768 return bpf_base_func_proto(func_id);
5772 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
5773 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
5775 static const struct bpf_func_proto *
5776 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5779 case BPF_FUNC_skb_store_bytes:
5780 return &bpf_skb_store_bytes_proto;
5781 case BPF_FUNC_skb_load_bytes:
5782 return &bpf_skb_load_bytes_proto;
5783 case BPF_FUNC_skb_pull_data:
5784 return &sk_skb_pull_data_proto;
5785 case BPF_FUNC_skb_change_tail:
5786 return &sk_skb_change_tail_proto;
5787 case BPF_FUNC_skb_change_head:
5788 return &sk_skb_change_head_proto;
5789 case BPF_FUNC_get_socket_cookie:
5790 return &bpf_get_socket_cookie_proto;
5791 case BPF_FUNC_get_socket_uid:
5792 return &bpf_get_socket_uid_proto;
5793 case BPF_FUNC_sk_redirect_map:
5794 return &bpf_sk_redirect_map_proto;
5795 case BPF_FUNC_sk_redirect_hash:
5796 return &bpf_sk_redirect_hash_proto;
5798 case BPF_FUNC_sk_lookup_tcp:
5799 return &bpf_sk_lookup_tcp_proto;
5800 case BPF_FUNC_sk_lookup_udp:
5801 return &bpf_sk_lookup_udp_proto;
5802 case BPF_FUNC_sk_release:
5803 return &bpf_sk_release_proto;
5806 return bpf_base_func_proto(func_id);
5810 static const struct bpf_func_proto *
5811 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5814 case BPF_FUNC_skb_load_bytes:
5815 return &bpf_skb_load_bytes_proto;
5817 return bpf_base_func_proto(func_id);
5821 static const struct bpf_func_proto *
5822 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5825 case BPF_FUNC_skb_load_bytes:
5826 return &bpf_skb_load_bytes_proto;
5827 case BPF_FUNC_skb_pull_data:
5828 return &bpf_skb_pull_data_proto;
5829 case BPF_FUNC_csum_diff:
5830 return &bpf_csum_diff_proto;
5831 case BPF_FUNC_get_cgroup_classid:
5832 return &bpf_get_cgroup_classid_proto;
5833 case BPF_FUNC_get_route_realm:
5834 return &bpf_get_route_realm_proto;
5835 case BPF_FUNC_get_hash_recalc:
5836 return &bpf_get_hash_recalc_proto;
5837 case BPF_FUNC_perf_event_output:
5838 return &bpf_skb_event_output_proto;
5839 case BPF_FUNC_get_smp_processor_id:
5840 return &bpf_get_smp_processor_id_proto;
5841 case BPF_FUNC_skb_under_cgroup:
5842 return &bpf_skb_under_cgroup_proto;
5844 return bpf_base_func_proto(func_id);
5848 static const struct bpf_func_proto *
5849 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5852 case BPF_FUNC_lwt_push_encap:
5853 return &bpf_lwt_in_push_encap_proto;
5855 return lwt_out_func_proto(func_id, prog);
5859 static const struct bpf_func_proto *
5860 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5863 case BPF_FUNC_skb_get_tunnel_key:
5864 return &bpf_skb_get_tunnel_key_proto;
5865 case BPF_FUNC_skb_set_tunnel_key:
5866 return bpf_get_skb_set_tunnel_proto(func_id);
5867 case BPF_FUNC_skb_get_tunnel_opt:
5868 return &bpf_skb_get_tunnel_opt_proto;
5869 case BPF_FUNC_skb_set_tunnel_opt:
5870 return bpf_get_skb_set_tunnel_proto(func_id);
5871 case BPF_FUNC_redirect:
5872 return &bpf_redirect_proto;
5873 case BPF_FUNC_clone_redirect:
5874 return &bpf_clone_redirect_proto;
5875 case BPF_FUNC_skb_change_tail:
5876 return &bpf_skb_change_tail_proto;
5877 case BPF_FUNC_skb_change_head:
5878 return &bpf_skb_change_head_proto;
5879 case BPF_FUNC_skb_store_bytes:
5880 return &bpf_skb_store_bytes_proto;
5881 case BPF_FUNC_csum_update:
5882 return &bpf_csum_update_proto;
5883 case BPF_FUNC_l3_csum_replace:
5884 return &bpf_l3_csum_replace_proto;
5885 case BPF_FUNC_l4_csum_replace:
5886 return &bpf_l4_csum_replace_proto;
5887 case BPF_FUNC_set_hash_invalid:
5888 return &bpf_set_hash_invalid_proto;
5889 case BPF_FUNC_lwt_push_encap:
5890 return &bpf_lwt_xmit_push_encap_proto;
5892 return lwt_out_func_proto(func_id, prog);
5896 static const struct bpf_func_proto *
5897 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5900 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5901 case BPF_FUNC_lwt_seg6_store_bytes:
5902 return &bpf_lwt_seg6_store_bytes_proto;
5903 case BPF_FUNC_lwt_seg6_action:
5904 return &bpf_lwt_seg6_action_proto;
5905 case BPF_FUNC_lwt_seg6_adjust_srh:
5906 return &bpf_lwt_seg6_adjust_srh_proto;
5909 return lwt_out_func_proto(func_id, prog);
5913 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
5914 const struct bpf_prog *prog,
5915 struct bpf_insn_access_aux *info)
5917 const int size_default = sizeof(__u32);
5919 if (off < 0 || off >= sizeof(struct __sk_buff))
5922 /* The verifier guarantees that size > 0. */
5923 if (off % size != 0)
5927 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5928 if (off + size > offsetofend(struct __sk_buff, cb[4]))
5931 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
5932 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
5933 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
5934 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
5935 case bpf_ctx_range(struct __sk_buff, data):
5936 case bpf_ctx_range(struct __sk_buff, data_meta):
5937 case bpf_ctx_range(struct __sk_buff, data_end):
5938 if (size != size_default)
5941 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
5942 if (size != sizeof(__u64))
5945 case bpf_ctx_range(struct __sk_buff, tstamp):
5946 if (size != sizeof(__u64))
5949 case offsetof(struct __sk_buff, sk):
5950 if (type == BPF_WRITE || size != sizeof(__u64))
5952 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
5955 /* Only narrow read access allowed for now. */
5956 if (type == BPF_WRITE) {
5957 if (size != size_default)
5960 bpf_ctx_record_field_size(info, size_default);
5961 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
5969 static bool sk_filter_is_valid_access(int off, int size,
5970 enum bpf_access_type type,
5971 const struct bpf_prog *prog,
5972 struct bpf_insn_access_aux *info)
5975 case bpf_ctx_range(struct __sk_buff, tc_classid):
5976 case bpf_ctx_range(struct __sk_buff, data):
5977 case bpf_ctx_range(struct __sk_buff, data_meta):
5978 case bpf_ctx_range(struct __sk_buff, data_end):
5979 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
5980 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
5981 case bpf_ctx_range(struct __sk_buff, tstamp):
5982 case bpf_ctx_range(struct __sk_buff, wire_len):
5986 if (type == BPF_WRITE) {
5988 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5995 return bpf_skb_is_valid_access(off, size, type, prog, info);
5998 static bool cg_skb_is_valid_access(int off, int size,
5999 enum bpf_access_type type,
6000 const struct bpf_prog *prog,
6001 struct bpf_insn_access_aux *info)
6004 case bpf_ctx_range(struct __sk_buff, tc_classid):
6005 case bpf_ctx_range(struct __sk_buff, data_meta):
6006 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
6007 case bpf_ctx_range(struct __sk_buff, wire_len):
6009 case bpf_ctx_range(struct __sk_buff, data):
6010 case bpf_ctx_range(struct __sk_buff, data_end):
6011 if (!capable(CAP_SYS_ADMIN))
6016 if (type == BPF_WRITE) {
6018 case bpf_ctx_range(struct __sk_buff, mark):
6019 case bpf_ctx_range(struct __sk_buff, priority):
6020 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6022 case bpf_ctx_range(struct __sk_buff, tstamp):
6023 if (!capable(CAP_SYS_ADMIN))
6032 case bpf_ctx_range(struct __sk_buff, data):
6033 info->reg_type = PTR_TO_PACKET;
6035 case bpf_ctx_range(struct __sk_buff, data_end):
6036 info->reg_type = PTR_TO_PACKET_END;
6040 return bpf_skb_is_valid_access(off, size, type, prog, info);
6043 static bool lwt_is_valid_access(int off, int size,
6044 enum bpf_access_type type,
6045 const struct bpf_prog *prog,
6046 struct bpf_insn_access_aux *info)
6049 case bpf_ctx_range(struct __sk_buff, tc_classid):
6050 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6051 case bpf_ctx_range(struct __sk_buff, data_meta):
6052 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
6053 case bpf_ctx_range(struct __sk_buff, tstamp):
6054 case bpf_ctx_range(struct __sk_buff, wire_len):
6058 if (type == BPF_WRITE) {
6060 case bpf_ctx_range(struct __sk_buff, mark):
6061 case bpf_ctx_range(struct __sk_buff, priority):
6062 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6070 case bpf_ctx_range(struct __sk_buff, data):
6071 info->reg_type = PTR_TO_PACKET;
6073 case bpf_ctx_range(struct __sk_buff, data_end):
6074 info->reg_type = PTR_TO_PACKET_END;
6078 return bpf_skb_is_valid_access(off, size, type, prog, info);
6081 /* Attach type specific accesses */
6082 static bool __sock_filter_check_attach_type(int off,
6083 enum bpf_access_type access_type,
6084 enum bpf_attach_type attach_type)
6087 case offsetof(struct bpf_sock, bound_dev_if):
6088 case offsetof(struct bpf_sock, mark):
6089 case offsetof(struct bpf_sock, priority):
6090 switch (attach_type) {
6091 case BPF_CGROUP_INET_SOCK_CREATE:
6096 case bpf_ctx_range(struct bpf_sock, src_ip4):
6097 switch (attach_type) {
6098 case BPF_CGROUP_INET4_POST_BIND:
6103 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
6104 switch (attach_type) {
6105 case BPF_CGROUP_INET6_POST_BIND:
6110 case bpf_ctx_range(struct bpf_sock, src_port):
6111 switch (attach_type) {
6112 case BPF_CGROUP_INET4_POST_BIND:
6113 case BPF_CGROUP_INET6_POST_BIND:
6120 return access_type == BPF_READ;
6125 bool bpf_sock_common_is_valid_access(int off, int size,
6126 enum bpf_access_type type,
6127 struct bpf_insn_access_aux *info)
6130 case bpf_ctx_range_till(struct bpf_sock, type, priority):
6133 return bpf_sock_is_valid_access(off, size, type, info);
6137 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6138 struct bpf_insn_access_aux *info)
6140 const int size_default = sizeof(__u32);
6142 if (off < 0 || off >= sizeof(struct bpf_sock))
6144 if (off % size != 0)
6148 case offsetof(struct bpf_sock, state):
6149 case offsetof(struct bpf_sock, family):
6150 case offsetof(struct bpf_sock, type):
6151 case offsetof(struct bpf_sock, protocol):
6152 case offsetof(struct bpf_sock, dst_port):
6153 case offsetof(struct bpf_sock, src_port):
6154 case bpf_ctx_range(struct bpf_sock, src_ip4):
6155 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
6156 case bpf_ctx_range(struct bpf_sock, dst_ip4):
6157 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
6158 bpf_ctx_record_field_size(info, size_default);
6159 return bpf_ctx_narrow_access_ok(off, size, size_default);
6162 return size == size_default;
6165 static bool sock_filter_is_valid_access(int off, int size,
6166 enum bpf_access_type type,
6167 const struct bpf_prog *prog,
6168 struct bpf_insn_access_aux *info)
6170 if (!bpf_sock_is_valid_access(off, size, type, info))
6172 return __sock_filter_check_attach_type(off, type,
6173 prog->expected_attach_type);
6176 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
6177 const struct bpf_prog *prog)
6179 /* Neither direct read nor direct write requires any preliminary
6185 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
6186 const struct bpf_prog *prog, int drop_verdict)
6188 struct bpf_insn *insn = insn_buf;
6193 /* if (!skb->cloned)
6196 * (Fast-path, otherwise approximation that we might be
6197 * a clone, do the rest in helper.)
6199 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
6200 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
6201 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
6203 /* ret = bpf_skb_pull_data(skb, 0); */
6204 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
6205 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
6206 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
6207 BPF_FUNC_skb_pull_data);
6210 * return TC_ACT_SHOT;
6212 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
6213 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
6214 *insn++ = BPF_EXIT_INSN();
6217 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
6219 *insn++ = prog->insnsi[0];
6221 return insn - insn_buf;
6224 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
6225 struct bpf_insn *insn_buf)
6227 bool indirect = BPF_MODE(orig->code) == BPF_IND;
6228 struct bpf_insn *insn = insn_buf;
6230 /* We're guaranteed here that CTX is in R6. */
6231 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
6233 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
6235 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
6237 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
6240 switch (BPF_SIZE(orig->code)) {
6242 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
6245 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
6248 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
6252 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
6253 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
6254 *insn++ = BPF_EXIT_INSN();
6256 return insn - insn_buf;
6259 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
6260 const struct bpf_prog *prog)
6262 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
6265 static bool tc_cls_act_is_valid_access(int off, int size,
6266 enum bpf_access_type type,
6267 const struct bpf_prog *prog,
6268 struct bpf_insn_access_aux *info)
6270 if (type == BPF_WRITE) {
6272 case bpf_ctx_range(struct __sk_buff, mark):
6273 case bpf_ctx_range(struct __sk_buff, tc_index):
6274 case bpf_ctx_range(struct __sk_buff, priority):
6275 case bpf_ctx_range(struct __sk_buff, tc_classid):
6276 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6277 case bpf_ctx_range(struct __sk_buff, tstamp):
6285 case bpf_ctx_range(struct __sk_buff, data):
6286 info->reg_type = PTR_TO_PACKET;
6288 case bpf_ctx_range(struct __sk_buff, data_meta):
6289 info->reg_type = PTR_TO_PACKET_META;
6291 case bpf_ctx_range(struct __sk_buff, data_end):
6292 info->reg_type = PTR_TO_PACKET_END;
6294 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
6295 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6299 return bpf_skb_is_valid_access(off, size, type, prog, info);
6302 static bool __is_valid_xdp_access(int off, int size)
6304 if (off < 0 || off >= sizeof(struct xdp_md))
6306 if (off % size != 0)
6308 if (size != sizeof(__u32))
6314 static bool xdp_is_valid_access(int off, int size,
6315 enum bpf_access_type type,
6316 const struct bpf_prog *prog,
6317 struct bpf_insn_access_aux *info)
6319 if (type == BPF_WRITE) {
6320 if (bpf_prog_is_dev_bound(prog->aux)) {
6322 case offsetof(struct xdp_md, rx_queue_index):
6323 return __is_valid_xdp_access(off, size);
6330 case offsetof(struct xdp_md, data):
6331 info->reg_type = PTR_TO_PACKET;
6333 case offsetof(struct xdp_md, data_meta):
6334 info->reg_type = PTR_TO_PACKET_META;
6336 case offsetof(struct xdp_md, data_end):
6337 info->reg_type = PTR_TO_PACKET_END;
6341 return __is_valid_xdp_access(off, size);
6344 void bpf_warn_invalid_xdp_action(u32 act)
6346 const u32 act_max = XDP_REDIRECT;
6348 WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
6349 act > act_max ? "Illegal" : "Driver unsupported",
6352 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
6354 static bool sock_addr_is_valid_access(int off, int size,
6355 enum bpf_access_type type,
6356 const struct bpf_prog *prog,
6357 struct bpf_insn_access_aux *info)
6359 const int size_default = sizeof(__u32);
6361 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
6363 if (off % size != 0)
6366 /* Disallow access to IPv6 fields from IPv4 contex and vise
6370 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
6371 switch (prog->expected_attach_type) {
6372 case BPF_CGROUP_INET4_BIND:
6373 case BPF_CGROUP_INET4_CONNECT:
6374 case BPF_CGROUP_UDP4_SENDMSG:
6380 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
6381 switch (prog->expected_attach_type) {
6382 case BPF_CGROUP_INET6_BIND:
6383 case BPF_CGROUP_INET6_CONNECT:
6384 case BPF_CGROUP_UDP6_SENDMSG:
6390 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
6391 switch (prog->expected_attach_type) {
6392 case BPF_CGROUP_UDP4_SENDMSG:
6398 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
6400 switch (prog->expected_attach_type) {
6401 case BPF_CGROUP_UDP6_SENDMSG:
6410 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
6411 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
6412 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
6413 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
6415 /* Only narrow read access allowed for now. */
6416 if (type == BPF_READ) {
6417 bpf_ctx_record_field_size(info, size_default);
6418 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
6421 if (size != size_default)
6425 case bpf_ctx_range(struct bpf_sock_addr, user_port):
6426 if (size != size_default)
6430 if (type == BPF_READ) {
6431 if (size != size_default)
6441 static bool sock_ops_is_valid_access(int off, int size,
6442 enum bpf_access_type type,
6443 const struct bpf_prog *prog,
6444 struct bpf_insn_access_aux *info)
6446 const int size_default = sizeof(__u32);
6448 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
6451 /* The verifier guarantees that size > 0. */
6452 if (off % size != 0)
6455 if (type == BPF_WRITE) {
6457 case offsetof(struct bpf_sock_ops, reply):
6458 case offsetof(struct bpf_sock_ops, sk_txhash):
6459 if (size != size_default)
6467 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
6469 if (size != sizeof(__u64))
6473 if (size != size_default)
6482 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
6483 const struct bpf_prog *prog)
6485 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
6488 static bool sk_skb_is_valid_access(int off, int size,
6489 enum bpf_access_type type,
6490 const struct bpf_prog *prog,
6491 struct bpf_insn_access_aux *info)
6494 case bpf_ctx_range(struct __sk_buff, tc_classid):
6495 case bpf_ctx_range(struct __sk_buff, data_meta):
6496 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
6497 case bpf_ctx_range(struct __sk_buff, tstamp):
6498 case bpf_ctx_range(struct __sk_buff, wire_len):
6502 if (type == BPF_WRITE) {
6504 case bpf_ctx_range(struct __sk_buff, tc_index):
6505 case bpf_ctx_range(struct __sk_buff, priority):
6513 case bpf_ctx_range(struct __sk_buff, mark):
6515 case bpf_ctx_range(struct __sk_buff, data):
6516 info->reg_type = PTR_TO_PACKET;
6518 case bpf_ctx_range(struct __sk_buff, data_end):
6519 info->reg_type = PTR_TO_PACKET_END;
6523 return bpf_skb_is_valid_access(off, size, type, prog, info);
6526 static bool sk_msg_is_valid_access(int off, int size,
6527 enum bpf_access_type type,
6528 const struct bpf_prog *prog,
6529 struct bpf_insn_access_aux *info)
6531 if (type == BPF_WRITE)
6534 if (off % size != 0)
6538 case offsetof(struct sk_msg_md, data):
6539 info->reg_type = PTR_TO_PACKET;
6540 if (size != sizeof(__u64))
6543 case offsetof(struct sk_msg_md, data_end):
6544 info->reg_type = PTR_TO_PACKET_END;
6545 if (size != sizeof(__u64))
6548 case bpf_ctx_range(struct sk_msg_md, family):
6549 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
6550 case bpf_ctx_range(struct sk_msg_md, local_ip4):
6551 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
6552 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
6553 case bpf_ctx_range(struct sk_msg_md, remote_port):
6554 case bpf_ctx_range(struct sk_msg_md, local_port):
6555 case bpf_ctx_range(struct sk_msg_md, size):
6556 if (size != sizeof(__u32))
6565 static bool flow_dissector_is_valid_access(int off, int size,
6566 enum bpf_access_type type,
6567 const struct bpf_prog *prog,
6568 struct bpf_insn_access_aux *info)
6570 if (type == BPF_WRITE) {
6572 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6580 case bpf_ctx_range(struct __sk_buff, data):
6581 info->reg_type = PTR_TO_PACKET;
6583 case bpf_ctx_range(struct __sk_buff, data_end):
6584 info->reg_type = PTR_TO_PACKET_END;
6586 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
6587 info->reg_type = PTR_TO_FLOW_KEYS;
6589 case bpf_ctx_range(struct __sk_buff, tc_classid):
6590 case bpf_ctx_range(struct __sk_buff, data_meta):
6591 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6592 case bpf_ctx_range(struct __sk_buff, tstamp):
6593 case bpf_ctx_range(struct __sk_buff, wire_len):
6597 return bpf_skb_is_valid_access(off, size, type, prog, info);
6600 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
6601 const struct bpf_insn *si,
6602 struct bpf_insn *insn_buf,
6603 struct bpf_prog *prog, u32 *target_size)
6605 struct bpf_insn *insn = insn_buf;
6609 case offsetof(struct __sk_buff, len):
6610 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6611 bpf_target_off(struct sk_buff, len, 4,
6615 case offsetof(struct __sk_buff, protocol):
6616 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6617 bpf_target_off(struct sk_buff, protocol, 2,
6621 case offsetof(struct __sk_buff, vlan_proto):
6622 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6623 bpf_target_off(struct sk_buff, vlan_proto, 2,
6627 case offsetof(struct __sk_buff, priority):
6628 if (type == BPF_WRITE)
6629 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6630 bpf_target_off(struct sk_buff, priority, 4,
6633 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6634 bpf_target_off(struct sk_buff, priority, 4,
6638 case offsetof(struct __sk_buff, ingress_ifindex):
6639 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6640 bpf_target_off(struct sk_buff, skb_iif, 4,
6644 case offsetof(struct __sk_buff, ifindex):
6645 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
6646 si->dst_reg, si->src_reg,
6647 offsetof(struct sk_buff, dev));
6648 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
6649 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6650 bpf_target_off(struct net_device, ifindex, 4,
6654 case offsetof(struct __sk_buff, hash):
6655 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6656 bpf_target_off(struct sk_buff, hash, 4,
6660 case offsetof(struct __sk_buff, mark):
6661 if (type == BPF_WRITE)
6662 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6663 bpf_target_off(struct sk_buff, mark, 4,
6666 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6667 bpf_target_off(struct sk_buff, mark, 4,
6671 case offsetof(struct __sk_buff, pkt_type):
6673 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
6675 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
6676 #ifdef __BIG_ENDIAN_BITFIELD
6677 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
6681 case offsetof(struct __sk_buff, queue_mapping):
6682 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6683 bpf_target_off(struct sk_buff, queue_mapping, 2,
6687 case offsetof(struct __sk_buff, vlan_present):
6689 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
6690 PKT_VLAN_PRESENT_OFFSET());
6691 if (PKT_VLAN_PRESENT_BIT)
6692 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
6693 if (PKT_VLAN_PRESENT_BIT < 7)
6694 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
6697 case offsetof(struct __sk_buff, vlan_tci):
6698 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6699 bpf_target_off(struct sk_buff, vlan_tci, 2,
6703 case offsetof(struct __sk_buff, cb[0]) ...
6704 offsetofend(struct __sk_buff, cb[4]) - 1:
6705 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20);
6706 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
6707 offsetof(struct qdisc_skb_cb, data)) %
6710 prog->cb_access = 1;
6712 off -= offsetof(struct __sk_buff, cb[0]);
6713 off += offsetof(struct sk_buff, cb);
6714 off += offsetof(struct qdisc_skb_cb, data);
6715 if (type == BPF_WRITE)
6716 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
6719 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
6723 case offsetof(struct __sk_buff, tc_classid):
6724 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, tc_classid) != 2);
6727 off -= offsetof(struct __sk_buff, tc_classid);
6728 off += offsetof(struct sk_buff, cb);
6729 off += offsetof(struct qdisc_skb_cb, tc_classid);
6731 if (type == BPF_WRITE)
6732 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
6735 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
6739 case offsetof(struct __sk_buff, data):
6740 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
6741 si->dst_reg, si->src_reg,
6742 offsetof(struct sk_buff, data));
6745 case offsetof(struct __sk_buff, data_meta):
6747 off -= offsetof(struct __sk_buff, data_meta);
6748 off += offsetof(struct sk_buff, cb);
6749 off += offsetof(struct bpf_skb_data_end, data_meta);
6750 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
6754 case offsetof(struct __sk_buff, data_end):
6756 off -= offsetof(struct __sk_buff, data_end);
6757 off += offsetof(struct sk_buff, cb);
6758 off += offsetof(struct bpf_skb_data_end, data_end);
6759 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
6763 case offsetof(struct __sk_buff, tc_index):
6764 #ifdef CONFIG_NET_SCHED
6765 if (type == BPF_WRITE)
6766 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
6767 bpf_target_off(struct sk_buff, tc_index, 2,
6770 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6771 bpf_target_off(struct sk_buff, tc_index, 2,
6775 if (type == BPF_WRITE)
6776 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
6778 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
6782 case offsetof(struct __sk_buff, napi_id):
6783 #if defined(CONFIG_NET_RX_BUSY_POLL)
6784 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6785 bpf_target_off(struct sk_buff, napi_id, 4,
6787 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
6788 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
6791 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
6794 case offsetof(struct __sk_buff, family):
6795 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
6797 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6798 si->dst_reg, si->src_reg,
6799 offsetof(struct sk_buff, sk));
6800 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6801 bpf_target_off(struct sock_common,
6805 case offsetof(struct __sk_buff, remote_ip4):
6806 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
6808 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6809 si->dst_reg, si->src_reg,
6810 offsetof(struct sk_buff, sk));
6811 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6812 bpf_target_off(struct sock_common,
6816 case offsetof(struct __sk_buff, local_ip4):
6817 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6818 skc_rcv_saddr) != 4);
6820 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6821 si->dst_reg, si->src_reg,
6822 offsetof(struct sk_buff, sk));
6823 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6824 bpf_target_off(struct sock_common,
6828 case offsetof(struct __sk_buff, remote_ip6[0]) ...
6829 offsetof(struct __sk_buff, remote_ip6[3]):
6830 #if IS_ENABLED(CONFIG_IPV6)
6831 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6832 skc_v6_daddr.s6_addr32[0]) != 4);
6835 off -= offsetof(struct __sk_buff, remote_ip6[0]);
6837 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6838 si->dst_reg, si->src_reg,
6839 offsetof(struct sk_buff, sk));
6840 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6841 offsetof(struct sock_common,
6842 skc_v6_daddr.s6_addr32[0]) +
6845 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6848 case offsetof(struct __sk_buff, local_ip6[0]) ...
6849 offsetof(struct __sk_buff, local_ip6[3]):
6850 #if IS_ENABLED(CONFIG_IPV6)
6851 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6852 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
6855 off -= offsetof(struct __sk_buff, local_ip6[0]);
6857 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6858 si->dst_reg, si->src_reg,
6859 offsetof(struct sk_buff, sk));
6860 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6861 offsetof(struct sock_common,
6862 skc_v6_rcv_saddr.s6_addr32[0]) +
6865 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6869 case offsetof(struct __sk_buff, remote_port):
6870 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
6872 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6873 si->dst_reg, si->src_reg,
6874 offsetof(struct sk_buff, sk));
6875 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6876 bpf_target_off(struct sock_common,
6879 #ifndef __BIG_ENDIAN_BITFIELD
6880 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
6884 case offsetof(struct __sk_buff, local_port):
6885 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
6887 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6888 si->dst_reg, si->src_reg,
6889 offsetof(struct sk_buff, sk));
6890 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6891 bpf_target_off(struct sock_common,
6892 skc_num, 2, target_size));
6895 case offsetof(struct __sk_buff, flow_keys):
6897 off -= offsetof(struct __sk_buff, flow_keys);
6898 off += offsetof(struct sk_buff, cb);
6899 off += offsetof(struct qdisc_skb_cb, flow_keys);
6900 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
6904 case offsetof(struct __sk_buff, tstamp):
6905 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, tstamp) != 8);
6907 if (type == BPF_WRITE)
6908 *insn++ = BPF_STX_MEM(BPF_DW,
6909 si->dst_reg, si->src_reg,
6910 bpf_target_off(struct sk_buff,
6914 *insn++ = BPF_LDX_MEM(BPF_DW,
6915 si->dst_reg, si->src_reg,
6916 bpf_target_off(struct sk_buff,
6921 case offsetof(struct __sk_buff, gso_segs):
6922 /* si->dst_reg = skb_shinfo(SKB); */
6923 #ifdef NET_SKBUFF_DATA_USES_OFFSET
6924 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
6925 si->dst_reg, si->src_reg,
6926 offsetof(struct sk_buff, head));
6927 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
6928 BPF_REG_AX, si->src_reg,
6929 offsetof(struct sk_buff, end));
6930 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
6932 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
6933 si->dst_reg, si->src_reg,
6934 offsetof(struct sk_buff, end));
6936 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
6937 si->dst_reg, si->dst_reg,
6938 bpf_target_off(struct skb_shared_info,
6942 case offsetof(struct __sk_buff, wire_len):
6943 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, pkt_len) != 4);
6946 off -= offsetof(struct __sk_buff, wire_len);
6947 off += offsetof(struct sk_buff, cb);
6948 off += offsetof(struct qdisc_skb_cb, pkt_len);
6950 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
6953 case offsetof(struct __sk_buff, sk):
6954 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6955 si->dst_reg, si->src_reg,
6956 offsetof(struct sk_buff, sk));
6960 return insn - insn_buf;
6963 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
6964 const struct bpf_insn *si,
6965 struct bpf_insn *insn_buf,
6966 struct bpf_prog *prog, u32 *target_size)
6968 struct bpf_insn *insn = insn_buf;
6972 case offsetof(struct bpf_sock, bound_dev_if):
6973 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_bound_dev_if) != 4);
6975 if (type == BPF_WRITE)
6976 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6977 offsetof(struct sock, sk_bound_dev_if));
6979 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6980 offsetof(struct sock, sk_bound_dev_if));
6983 case offsetof(struct bpf_sock, mark):
6984 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_mark) != 4);
6986 if (type == BPF_WRITE)
6987 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6988 offsetof(struct sock, sk_mark));
6990 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6991 offsetof(struct sock, sk_mark));
6994 case offsetof(struct bpf_sock, priority):
6995 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_priority) != 4);
6997 if (type == BPF_WRITE)
6998 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6999 offsetof(struct sock, sk_priority));
7001 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7002 offsetof(struct sock, sk_priority));
7005 case offsetof(struct bpf_sock, family):
7006 *insn++ = BPF_LDX_MEM(
7007 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
7008 si->dst_reg, si->src_reg,
7009 bpf_target_off(struct sock_common,
7011 FIELD_SIZEOF(struct sock_common,
7016 case offsetof(struct bpf_sock, type):
7017 BUILD_BUG_ON(HWEIGHT32(SK_FL_TYPE_MASK) != BITS_PER_BYTE * 2);
7018 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7019 offsetof(struct sock, __sk_flags_offset));
7020 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
7021 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
7025 case offsetof(struct bpf_sock, protocol):
7026 BUILD_BUG_ON(HWEIGHT32(SK_FL_PROTO_MASK) != BITS_PER_BYTE);
7027 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7028 offsetof(struct sock, __sk_flags_offset));
7029 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
7030 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_PROTO_SHIFT);
7034 case offsetof(struct bpf_sock, src_ip4):
7035 *insn++ = BPF_LDX_MEM(
7036 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7037 bpf_target_off(struct sock_common, skc_rcv_saddr,
7038 FIELD_SIZEOF(struct sock_common,
7043 case offsetof(struct bpf_sock, dst_ip4):
7044 *insn++ = BPF_LDX_MEM(
7045 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7046 bpf_target_off(struct sock_common, skc_daddr,
7047 FIELD_SIZEOF(struct sock_common,
7052 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7053 #if IS_ENABLED(CONFIG_IPV6)
7055 off -= offsetof(struct bpf_sock, src_ip6[0]);
7056 *insn++ = BPF_LDX_MEM(
7057 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7060 skc_v6_rcv_saddr.s6_addr32[0],
7061 FIELD_SIZEOF(struct sock_common,
7062 skc_v6_rcv_saddr.s6_addr32[0]),
7063 target_size) + off);
7066 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7070 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
7071 #if IS_ENABLED(CONFIG_IPV6)
7073 off -= offsetof(struct bpf_sock, dst_ip6[0]);
7074 *insn++ = BPF_LDX_MEM(
7075 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7076 bpf_target_off(struct sock_common,
7077 skc_v6_daddr.s6_addr32[0],
7078 FIELD_SIZEOF(struct sock_common,
7079 skc_v6_daddr.s6_addr32[0]),
7080 target_size) + off);
7082 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7087 case offsetof(struct bpf_sock, src_port):
7088 *insn++ = BPF_LDX_MEM(
7089 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
7090 si->dst_reg, si->src_reg,
7091 bpf_target_off(struct sock_common, skc_num,
7092 FIELD_SIZEOF(struct sock_common,
7097 case offsetof(struct bpf_sock, dst_port):
7098 *insn++ = BPF_LDX_MEM(
7099 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
7100 si->dst_reg, si->src_reg,
7101 bpf_target_off(struct sock_common, skc_dport,
7102 FIELD_SIZEOF(struct sock_common,
7107 case offsetof(struct bpf_sock, state):
7108 *insn++ = BPF_LDX_MEM(
7109 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
7110 si->dst_reg, si->src_reg,
7111 bpf_target_off(struct sock_common, skc_state,
7112 FIELD_SIZEOF(struct sock_common,
7118 return insn - insn_buf;
7121 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
7122 const struct bpf_insn *si,
7123 struct bpf_insn *insn_buf,
7124 struct bpf_prog *prog, u32 *target_size)
7126 struct bpf_insn *insn = insn_buf;
7129 case offsetof(struct __sk_buff, ifindex):
7130 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
7131 si->dst_reg, si->src_reg,
7132 offsetof(struct sk_buff, dev));
7133 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7134 bpf_target_off(struct net_device, ifindex, 4,
7138 return bpf_convert_ctx_access(type, si, insn_buf, prog,
7142 return insn - insn_buf;
7145 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
7146 const struct bpf_insn *si,
7147 struct bpf_insn *insn_buf,
7148 struct bpf_prog *prog, u32 *target_size)
7150 struct bpf_insn *insn = insn_buf;
7153 case offsetof(struct xdp_md, data):
7154 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
7155 si->dst_reg, si->src_reg,
7156 offsetof(struct xdp_buff, data));
7158 case offsetof(struct xdp_md, data_meta):
7159 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
7160 si->dst_reg, si->src_reg,
7161 offsetof(struct xdp_buff, data_meta));
7163 case offsetof(struct xdp_md, data_end):
7164 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
7165 si->dst_reg, si->src_reg,
7166 offsetof(struct xdp_buff, data_end));
7168 case offsetof(struct xdp_md, ingress_ifindex):
7169 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
7170 si->dst_reg, si->src_reg,
7171 offsetof(struct xdp_buff, rxq));
7172 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
7173 si->dst_reg, si->dst_reg,
7174 offsetof(struct xdp_rxq_info, dev));
7175 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7176 offsetof(struct net_device, ifindex));
7178 case offsetof(struct xdp_md, rx_queue_index):
7179 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
7180 si->dst_reg, si->src_reg,
7181 offsetof(struct xdp_buff, rxq));
7182 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7183 offsetof(struct xdp_rxq_info,
7188 return insn - insn_buf;
7191 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
7192 * context Structure, F is Field in context structure that contains a pointer
7193 * to Nested Structure of type NS that has the field NF.
7195 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
7196 * sure that SIZE is not greater than actual size of S.F.NF.
7198 * If offset OFF is provided, the load happens from that offset relative to
7201 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
7203 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
7204 si->src_reg, offsetof(S, F)); \
7205 *insn++ = BPF_LDX_MEM( \
7206 SIZE, si->dst_reg, si->dst_reg, \
7207 bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF), \
7212 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
7213 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
7214 BPF_FIELD_SIZEOF(NS, NF), 0)
7216 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
7217 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
7219 * It doesn't support SIZE argument though since narrow stores are not
7220 * supported for now.
7222 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
7223 * "register" since two registers available in convert_ctx_access are not
7224 * enough: we can't override neither SRC, since it contains value to store, nor
7225 * DST since it contains pointer to context that may be used by later
7226 * instructions. But we need a temporary place to save pointer to nested
7227 * structure whose field we want to store to.
7229 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, OFF, TF) \
7231 int tmp_reg = BPF_REG_9; \
7232 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
7234 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
7236 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
7238 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
7239 si->dst_reg, offsetof(S, F)); \
7240 *insn++ = BPF_STX_MEM( \
7241 BPF_FIELD_SIZEOF(NS, NF), tmp_reg, si->src_reg, \
7242 bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF), \
7245 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
7249 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
7252 if (type == BPF_WRITE) { \
7253 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, OFF, \
7256 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
7257 S, NS, F, NF, SIZE, OFF); \
7261 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
7262 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
7263 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
7265 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
7266 const struct bpf_insn *si,
7267 struct bpf_insn *insn_buf,
7268 struct bpf_prog *prog, u32 *target_size)
7270 struct bpf_insn *insn = insn_buf;
7274 case offsetof(struct bpf_sock_addr, user_family):
7275 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
7276 struct sockaddr, uaddr, sa_family);
7279 case offsetof(struct bpf_sock_addr, user_ip4):
7280 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7281 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
7282 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
7285 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
7287 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
7288 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7289 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
7290 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
7294 case offsetof(struct bpf_sock_addr, user_port):
7295 /* To get port we need to know sa_family first and then treat
7296 * sockaddr as either sockaddr_in or sockaddr_in6.
7297 * Though we can simplify since port field has same offset and
7298 * size in both structures.
7299 * Here we check this invariant and use just one of the
7300 * structures if it's true.
7302 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
7303 offsetof(struct sockaddr_in6, sin6_port));
7304 BUILD_BUG_ON(FIELD_SIZEOF(struct sockaddr_in, sin_port) !=
7305 FIELD_SIZEOF(struct sockaddr_in6, sin6_port));
7306 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(struct bpf_sock_addr_kern,
7307 struct sockaddr_in6, uaddr,
7308 sin6_port, tmp_reg);
7311 case offsetof(struct bpf_sock_addr, family):
7312 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
7313 struct sock, sk, sk_family);
7316 case offsetof(struct bpf_sock_addr, type):
7317 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(
7318 struct bpf_sock_addr_kern, struct sock, sk,
7319 __sk_flags_offset, BPF_W, 0);
7320 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
7321 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
7324 case offsetof(struct bpf_sock_addr, protocol):
7325 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(
7326 struct bpf_sock_addr_kern, struct sock, sk,
7327 __sk_flags_offset, BPF_W, 0);
7328 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
7329 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg,
7333 case offsetof(struct bpf_sock_addr, msg_src_ip4):
7334 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
7335 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7336 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
7337 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
7340 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
7343 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
7344 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
7345 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7346 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
7347 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
7351 return insn - insn_buf;
7354 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
7355 const struct bpf_insn *si,
7356 struct bpf_insn *insn_buf,
7357 struct bpf_prog *prog,
7360 struct bpf_insn *insn = insn_buf;
7363 /* Helper macro for adding read access to tcp_sock or sock fields. */
7364 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
7366 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) > \
7367 FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD)); \
7368 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
7369 struct bpf_sock_ops_kern, \
7371 si->dst_reg, si->src_reg, \
7372 offsetof(struct bpf_sock_ops_kern, \
7374 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 2); \
7375 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
7376 struct bpf_sock_ops_kern, sk),\
7377 si->dst_reg, si->src_reg, \
7378 offsetof(struct bpf_sock_ops_kern, sk));\
7379 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
7381 si->dst_reg, si->dst_reg, \
7382 offsetof(OBJ, OBJ_FIELD)); \
7385 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
7386 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
7388 /* Helper macro for adding write access to tcp_sock or sock fields.
7389 * The macro is called with two registers, dst_reg which contains a pointer
7390 * to ctx (context) and src_reg which contains the value that should be
7391 * stored. However, we need an additional register since we cannot overwrite
7392 * dst_reg because it may be used later in the program.
7393 * Instead we "borrow" one of the other register. We first save its value
7394 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
7395 * it at the end of the macro.
7397 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
7399 int reg = BPF_REG_9; \
7400 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) > \
7401 FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD)); \
7402 if (si->dst_reg == reg || si->src_reg == reg) \
7404 if (si->dst_reg == reg || si->src_reg == reg) \
7406 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
7407 offsetof(struct bpf_sock_ops_kern, \
7409 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
7410 struct bpf_sock_ops_kern, \
7413 offsetof(struct bpf_sock_ops_kern, \
7415 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
7416 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
7417 struct bpf_sock_ops_kern, sk),\
7419 offsetof(struct bpf_sock_ops_kern, sk));\
7420 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \
7422 offsetof(OBJ, OBJ_FIELD)); \
7423 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
7424 offsetof(struct bpf_sock_ops_kern, \
7428 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
7430 if (TYPE == BPF_WRITE) \
7431 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
7433 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
7436 CONVERT_COMMON_TCP_SOCK_FIELDS(struct bpf_sock_ops,
7437 SOCK_OPS_GET_TCP_SOCK_FIELD);
7439 if (insn > insn_buf)
7440 return insn - insn_buf;
7443 case offsetof(struct bpf_sock_ops, op) ...
7444 offsetof(struct bpf_sock_ops, replylong[3]):
7445 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, op) !=
7446 FIELD_SIZEOF(struct bpf_sock_ops_kern, op));
7447 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, reply) !=
7448 FIELD_SIZEOF(struct bpf_sock_ops_kern, reply));
7449 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, replylong) !=
7450 FIELD_SIZEOF(struct bpf_sock_ops_kern, replylong));
7452 off -= offsetof(struct bpf_sock_ops, op);
7453 off += offsetof(struct bpf_sock_ops_kern, op);
7454 if (type == BPF_WRITE)
7455 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7458 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7462 case offsetof(struct bpf_sock_ops, family):
7463 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
7465 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7466 struct bpf_sock_ops_kern, sk),
7467 si->dst_reg, si->src_reg,
7468 offsetof(struct bpf_sock_ops_kern, sk));
7469 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7470 offsetof(struct sock_common, skc_family));
7473 case offsetof(struct bpf_sock_ops, remote_ip4):
7474 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
7476 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7477 struct bpf_sock_ops_kern, sk),
7478 si->dst_reg, si->src_reg,
7479 offsetof(struct bpf_sock_ops_kern, sk));
7480 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7481 offsetof(struct sock_common, skc_daddr));
7484 case offsetof(struct bpf_sock_ops, local_ip4):
7485 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7486 skc_rcv_saddr) != 4);
7488 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7489 struct bpf_sock_ops_kern, sk),
7490 si->dst_reg, si->src_reg,
7491 offsetof(struct bpf_sock_ops_kern, sk));
7492 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7493 offsetof(struct sock_common,
7497 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
7498 offsetof(struct bpf_sock_ops, remote_ip6[3]):
7499 #if IS_ENABLED(CONFIG_IPV6)
7500 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7501 skc_v6_daddr.s6_addr32[0]) != 4);
7504 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
7505 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7506 struct bpf_sock_ops_kern, sk),
7507 si->dst_reg, si->src_reg,
7508 offsetof(struct bpf_sock_ops_kern, sk));
7509 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7510 offsetof(struct sock_common,
7511 skc_v6_daddr.s6_addr32[0]) +
7514 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7518 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
7519 offsetof(struct bpf_sock_ops, local_ip6[3]):
7520 #if IS_ENABLED(CONFIG_IPV6)
7521 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7522 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
7525 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
7526 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7527 struct bpf_sock_ops_kern, sk),
7528 si->dst_reg, si->src_reg,
7529 offsetof(struct bpf_sock_ops_kern, sk));
7530 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7531 offsetof(struct sock_common,
7532 skc_v6_rcv_saddr.s6_addr32[0]) +
7535 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7539 case offsetof(struct bpf_sock_ops, remote_port):
7540 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
7542 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7543 struct bpf_sock_ops_kern, sk),
7544 si->dst_reg, si->src_reg,
7545 offsetof(struct bpf_sock_ops_kern, sk));
7546 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7547 offsetof(struct sock_common, skc_dport));
7548 #ifndef __BIG_ENDIAN_BITFIELD
7549 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
7553 case offsetof(struct bpf_sock_ops, local_port):
7554 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
7556 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7557 struct bpf_sock_ops_kern, sk),
7558 si->dst_reg, si->src_reg,
7559 offsetof(struct bpf_sock_ops_kern, sk));
7560 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7561 offsetof(struct sock_common, skc_num));
7564 case offsetof(struct bpf_sock_ops, is_fullsock):
7565 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7566 struct bpf_sock_ops_kern,
7568 si->dst_reg, si->src_reg,
7569 offsetof(struct bpf_sock_ops_kern,
7573 case offsetof(struct bpf_sock_ops, state):
7574 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_state) != 1);
7576 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7577 struct bpf_sock_ops_kern, sk),
7578 si->dst_reg, si->src_reg,
7579 offsetof(struct bpf_sock_ops_kern, sk));
7580 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
7581 offsetof(struct sock_common, skc_state));
7584 case offsetof(struct bpf_sock_ops, rtt_min):
7585 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, rtt_min) !=
7586 sizeof(struct minmax));
7587 BUILD_BUG_ON(sizeof(struct minmax) <
7588 sizeof(struct minmax_sample));
7590 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7591 struct bpf_sock_ops_kern, sk),
7592 si->dst_reg, si->src_reg,
7593 offsetof(struct bpf_sock_ops_kern, sk));
7594 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7595 offsetof(struct tcp_sock, rtt_min) +
7596 FIELD_SIZEOF(struct minmax_sample, t));
7599 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
7600 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
7604 case offsetof(struct bpf_sock_ops, sk_txhash):
7605 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
7609 return insn - insn_buf;
7612 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
7613 const struct bpf_insn *si,
7614 struct bpf_insn *insn_buf,
7615 struct bpf_prog *prog, u32 *target_size)
7617 struct bpf_insn *insn = insn_buf;
7621 case offsetof(struct __sk_buff, data_end):
7623 off -= offsetof(struct __sk_buff, data_end);
7624 off += offsetof(struct sk_buff, cb);
7625 off += offsetof(struct tcp_skb_cb, bpf.data_end);
7626 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
7630 return bpf_convert_ctx_access(type, si, insn_buf, prog,
7634 return insn - insn_buf;
7637 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
7638 const struct bpf_insn *si,
7639 struct bpf_insn *insn_buf,
7640 struct bpf_prog *prog, u32 *target_size)
7642 struct bpf_insn *insn = insn_buf;
7643 #if IS_ENABLED(CONFIG_IPV6)
7647 /* convert ctx uses the fact sg element is first in struct */
7648 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
7651 case offsetof(struct sk_msg_md, data):
7652 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
7653 si->dst_reg, si->src_reg,
7654 offsetof(struct sk_msg, data));
7656 case offsetof(struct sk_msg_md, data_end):
7657 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
7658 si->dst_reg, si->src_reg,
7659 offsetof(struct sk_msg, data_end));
7661 case offsetof(struct sk_msg_md, family):
7662 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
7664 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7666 si->dst_reg, si->src_reg,
7667 offsetof(struct sk_msg, sk));
7668 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7669 offsetof(struct sock_common, skc_family));
7672 case offsetof(struct sk_msg_md, remote_ip4):
7673 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
7675 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7677 si->dst_reg, si->src_reg,
7678 offsetof(struct sk_msg, sk));
7679 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7680 offsetof(struct sock_common, skc_daddr));
7683 case offsetof(struct sk_msg_md, local_ip4):
7684 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7685 skc_rcv_saddr) != 4);
7687 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7689 si->dst_reg, si->src_reg,
7690 offsetof(struct sk_msg, sk));
7691 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7692 offsetof(struct sock_common,
7696 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
7697 offsetof(struct sk_msg_md, remote_ip6[3]):
7698 #if IS_ENABLED(CONFIG_IPV6)
7699 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7700 skc_v6_daddr.s6_addr32[0]) != 4);
7703 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
7704 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7706 si->dst_reg, si->src_reg,
7707 offsetof(struct sk_msg, sk));
7708 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7709 offsetof(struct sock_common,
7710 skc_v6_daddr.s6_addr32[0]) +
7713 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7717 case offsetof(struct sk_msg_md, local_ip6[0]) ...
7718 offsetof(struct sk_msg_md, local_ip6[3]):
7719 #if IS_ENABLED(CONFIG_IPV6)
7720 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7721 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
7724 off -= offsetof(struct sk_msg_md, local_ip6[0]);
7725 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7727 si->dst_reg, si->src_reg,
7728 offsetof(struct sk_msg, sk));
7729 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7730 offsetof(struct sock_common,
7731 skc_v6_rcv_saddr.s6_addr32[0]) +
7734 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7738 case offsetof(struct sk_msg_md, remote_port):
7739 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
7741 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7743 si->dst_reg, si->src_reg,
7744 offsetof(struct sk_msg, sk));
7745 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7746 offsetof(struct sock_common, skc_dport));
7747 #ifndef __BIG_ENDIAN_BITFIELD
7748 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
7752 case offsetof(struct sk_msg_md, local_port):
7753 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
7755 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7757 si->dst_reg, si->src_reg,
7758 offsetof(struct sk_msg, sk));
7759 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7760 offsetof(struct sock_common, skc_num));
7763 case offsetof(struct sk_msg_md, size):
7764 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
7765 si->dst_reg, si->src_reg,
7766 offsetof(struct sk_msg_sg, size));
7770 return insn - insn_buf;
7773 const struct bpf_verifier_ops sk_filter_verifier_ops = {
7774 .get_func_proto = sk_filter_func_proto,
7775 .is_valid_access = sk_filter_is_valid_access,
7776 .convert_ctx_access = bpf_convert_ctx_access,
7777 .gen_ld_abs = bpf_gen_ld_abs,
7780 const struct bpf_prog_ops sk_filter_prog_ops = {
7781 .test_run = bpf_prog_test_run_skb,
7784 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
7785 .get_func_proto = tc_cls_act_func_proto,
7786 .is_valid_access = tc_cls_act_is_valid_access,
7787 .convert_ctx_access = tc_cls_act_convert_ctx_access,
7788 .gen_prologue = tc_cls_act_prologue,
7789 .gen_ld_abs = bpf_gen_ld_abs,
7792 const struct bpf_prog_ops tc_cls_act_prog_ops = {
7793 .test_run = bpf_prog_test_run_skb,
7796 const struct bpf_verifier_ops xdp_verifier_ops = {
7797 .get_func_proto = xdp_func_proto,
7798 .is_valid_access = xdp_is_valid_access,
7799 .convert_ctx_access = xdp_convert_ctx_access,
7800 .gen_prologue = bpf_noop_prologue,
7803 const struct bpf_prog_ops xdp_prog_ops = {
7804 .test_run = bpf_prog_test_run_xdp,
7807 const struct bpf_verifier_ops cg_skb_verifier_ops = {
7808 .get_func_proto = cg_skb_func_proto,
7809 .is_valid_access = cg_skb_is_valid_access,
7810 .convert_ctx_access = bpf_convert_ctx_access,
7813 const struct bpf_prog_ops cg_skb_prog_ops = {
7814 .test_run = bpf_prog_test_run_skb,
7817 const struct bpf_verifier_ops lwt_in_verifier_ops = {
7818 .get_func_proto = lwt_in_func_proto,
7819 .is_valid_access = lwt_is_valid_access,
7820 .convert_ctx_access = bpf_convert_ctx_access,
7823 const struct bpf_prog_ops lwt_in_prog_ops = {
7824 .test_run = bpf_prog_test_run_skb,
7827 const struct bpf_verifier_ops lwt_out_verifier_ops = {
7828 .get_func_proto = lwt_out_func_proto,
7829 .is_valid_access = lwt_is_valid_access,
7830 .convert_ctx_access = bpf_convert_ctx_access,
7833 const struct bpf_prog_ops lwt_out_prog_ops = {
7834 .test_run = bpf_prog_test_run_skb,
7837 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
7838 .get_func_proto = lwt_xmit_func_proto,
7839 .is_valid_access = lwt_is_valid_access,
7840 .convert_ctx_access = bpf_convert_ctx_access,
7841 .gen_prologue = tc_cls_act_prologue,
7844 const struct bpf_prog_ops lwt_xmit_prog_ops = {
7845 .test_run = bpf_prog_test_run_skb,
7848 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
7849 .get_func_proto = lwt_seg6local_func_proto,
7850 .is_valid_access = lwt_is_valid_access,
7851 .convert_ctx_access = bpf_convert_ctx_access,
7854 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
7855 .test_run = bpf_prog_test_run_skb,
7858 const struct bpf_verifier_ops cg_sock_verifier_ops = {
7859 .get_func_proto = sock_filter_func_proto,
7860 .is_valid_access = sock_filter_is_valid_access,
7861 .convert_ctx_access = bpf_sock_convert_ctx_access,
7864 const struct bpf_prog_ops cg_sock_prog_ops = {
7867 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
7868 .get_func_proto = sock_addr_func_proto,
7869 .is_valid_access = sock_addr_is_valid_access,
7870 .convert_ctx_access = sock_addr_convert_ctx_access,
7873 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
7876 const struct bpf_verifier_ops sock_ops_verifier_ops = {
7877 .get_func_proto = sock_ops_func_proto,
7878 .is_valid_access = sock_ops_is_valid_access,
7879 .convert_ctx_access = sock_ops_convert_ctx_access,
7882 const struct bpf_prog_ops sock_ops_prog_ops = {
7885 const struct bpf_verifier_ops sk_skb_verifier_ops = {
7886 .get_func_proto = sk_skb_func_proto,
7887 .is_valid_access = sk_skb_is_valid_access,
7888 .convert_ctx_access = sk_skb_convert_ctx_access,
7889 .gen_prologue = sk_skb_prologue,
7892 const struct bpf_prog_ops sk_skb_prog_ops = {
7895 const struct bpf_verifier_ops sk_msg_verifier_ops = {
7896 .get_func_proto = sk_msg_func_proto,
7897 .is_valid_access = sk_msg_is_valid_access,
7898 .convert_ctx_access = sk_msg_convert_ctx_access,
7899 .gen_prologue = bpf_noop_prologue,
7902 const struct bpf_prog_ops sk_msg_prog_ops = {
7905 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
7906 .get_func_proto = flow_dissector_func_proto,
7907 .is_valid_access = flow_dissector_is_valid_access,
7908 .convert_ctx_access = bpf_convert_ctx_access,
7911 const struct bpf_prog_ops flow_dissector_prog_ops = {
7912 .test_run = bpf_prog_test_run_flow_dissector,
7915 int sk_detach_filter(struct sock *sk)
7918 struct sk_filter *filter;
7920 if (sock_flag(sk, SOCK_FILTER_LOCKED))
7923 filter = rcu_dereference_protected(sk->sk_filter,
7924 lockdep_sock_is_held(sk));
7926 RCU_INIT_POINTER(sk->sk_filter, NULL);
7927 sk_filter_uncharge(sk, filter);
7933 EXPORT_SYMBOL_GPL(sk_detach_filter);
7935 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
7938 struct sock_fprog_kern *fprog;
7939 struct sk_filter *filter;
7943 filter = rcu_dereference_protected(sk->sk_filter,
7944 lockdep_sock_is_held(sk));
7948 /* We're copying the filter that has been originally attached,
7949 * so no conversion/decode needed anymore. eBPF programs that
7950 * have no original program cannot be dumped through this.
7953 fprog = filter->prog->orig_prog;
7959 /* User space only enquires number of filter blocks. */
7963 if (len < fprog->len)
7967 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
7970 /* Instead of bytes, the API requests to return the number
7980 struct sk_reuseport_kern {
7981 struct sk_buff *skb;
7983 struct sock *selected_sk;
7990 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
7991 struct sock_reuseport *reuse,
7992 struct sock *sk, struct sk_buff *skb,
7995 reuse_kern->skb = skb;
7996 reuse_kern->sk = sk;
7997 reuse_kern->selected_sk = NULL;
7998 reuse_kern->data_end = skb->data + skb_headlen(skb);
7999 reuse_kern->hash = hash;
8000 reuse_kern->reuseport_id = reuse->reuseport_id;
8001 reuse_kern->bind_inany = reuse->bind_inany;
8004 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
8005 struct bpf_prog *prog, struct sk_buff *skb,
8008 struct sk_reuseport_kern reuse_kern;
8009 enum sk_action action;
8011 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, hash);
8012 action = BPF_PROG_RUN(prog, &reuse_kern);
8014 if (action == SK_PASS)
8015 return reuse_kern.selected_sk;
8017 return ERR_PTR(-ECONNREFUSED);
8020 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
8021 struct bpf_map *, map, void *, key, u32, flags)
8023 struct sock_reuseport *reuse;
8024 struct sock *selected_sk;
8026 selected_sk = map->ops->map_lookup_elem(map, key);
8030 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
8032 /* selected_sk is unhashed (e.g. by close()) after the
8033 * above map_lookup_elem(). Treat selected_sk has already
8034 * been removed from the map.
8038 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
8041 if (unlikely(!reuse_kern->reuseport_id))
8042 /* There is a small race between adding the
8043 * sk to the map and setting the
8044 * reuse_kern->reuseport_id.
8045 * Treat it as the sk has not been added to
8050 sk = reuse_kern->sk;
8051 if (sk->sk_protocol != selected_sk->sk_protocol)
8053 else if (sk->sk_family != selected_sk->sk_family)
8054 return -EAFNOSUPPORT;
8056 /* Catch all. Likely bound to a different sockaddr. */
8060 reuse_kern->selected_sk = selected_sk;
8065 static const struct bpf_func_proto sk_select_reuseport_proto = {
8066 .func = sk_select_reuseport,
8068 .ret_type = RET_INTEGER,
8069 .arg1_type = ARG_PTR_TO_CTX,
8070 .arg2_type = ARG_CONST_MAP_PTR,
8071 .arg3_type = ARG_PTR_TO_MAP_KEY,
8072 .arg4_type = ARG_ANYTHING,
8075 BPF_CALL_4(sk_reuseport_load_bytes,
8076 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
8077 void *, to, u32, len)
8079 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
8082 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
8083 .func = sk_reuseport_load_bytes,
8085 .ret_type = RET_INTEGER,
8086 .arg1_type = ARG_PTR_TO_CTX,
8087 .arg2_type = ARG_ANYTHING,
8088 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
8089 .arg4_type = ARG_CONST_SIZE,
8092 BPF_CALL_5(sk_reuseport_load_bytes_relative,
8093 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
8094 void *, to, u32, len, u32, start_header)
8096 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
8100 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
8101 .func = sk_reuseport_load_bytes_relative,
8103 .ret_type = RET_INTEGER,
8104 .arg1_type = ARG_PTR_TO_CTX,
8105 .arg2_type = ARG_ANYTHING,
8106 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
8107 .arg4_type = ARG_CONST_SIZE,
8108 .arg5_type = ARG_ANYTHING,
8111 static const struct bpf_func_proto *
8112 sk_reuseport_func_proto(enum bpf_func_id func_id,
8113 const struct bpf_prog *prog)
8116 case BPF_FUNC_sk_select_reuseport:
8117 return &sk_select_reuseport_proto;
8118 case BPF_FUNC_skb_load_bytes:
8119 return &sk_reuseport_load_bytes_proto;
8120 case BPF_FUNC_skb_load_bytes_relative:
8121 return &sk_reuseport_load_bytes_relative_proto;
8123 return bpf_base_func_proto(func_id);
8128 sk_reuseport_is_valid_access(int off, int size,
8129 enum bpf_access_type type,
8130 const struct bpf_prog *prog,
8131 struct bpf_insn_access_aux *info)
8133 const u32 size_default = sizeof(__u32);
8135 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
8136 off % size || type != BPF_READ)
8140 case offsetof(struct sk_reuseport_md, data):
8141 info->reg_type = PTR_TO_PACKET;
8142 return size == sizeof(__u64);
8144 case offsetof(struct sk_reuseport_md, data_end):
8145 info->reg_type = PTR_TO_PACKET_END;
8146 return size == sizeof(__u64);
8148 case offsetof(struct sk_reuseport_md, hash):
8149 return size == size_default;
8151 /* Fields that allow narrowing */
8152 case offsetof(struct sk_reuseport_md, eth_protocol):
8153 if (size < FIELD_SIZEOF(struct sk_buff, protocol))
8156 case offsetof(struct sk_reuseport_md, ip_protocol):
8157 case offsetof(struct sk_reuseport_md, bind_inany):
8158 case offsetof(struct sk_reuseport_md, len):
8159 bpf_ctx_record_field_size(info, size_default);
8160 return bpf_ctx_narrow_access_ok(off, size, size_default);
8167 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
8168 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
8169 si->dst_reg, si->src_reg, \
8170 bpf_target_off(struct sk_reuseport_kern, F, \
8171 FIELD_SIZEOF(struct sk_reuseport_kern, F), \
8175 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
8176 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
8181 #define SK_REUSEPORT_LOAD_SK_FIELD_SIZE_OFF(SK_FIELD, BPF_SIZE, EXTRA_OFF) \
8182 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(struct sk_reuseport_kern, \
8185 SK_FIELD, BPF_SIZE, EXTRA_OFF)
8187 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
8188 const struct bpf_insn *si,
8189 struct bpf_insn *insn_buf,
8190 struct bpf_prog *prog,
8193 struct bpf_insn *insn = insn_buf;
8196 case offsetof(struct sk_reuseport_md, data):
8197 SK_REUSEPORT_LOAD_SKB_FIELD(data);
8200 case offsetof(struct sk_reuseport_md, len):
8201 SK_REUSEPORT_LOAD_SKB_FIELD(len);
8204 case offsetof(struct sk_reuseport_md, eth_protocol):
8205 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
8208 case offsetof(struct sk_reuseport_md, ip_protocol):
8209 BUILD_BUG_ON(HWEIGHT32(SK_FL_PROTO_MASK) != BITS_PER_BYTE);
8210 SK_REUSEPORT_LOAD_SK_FIELD_SIZE_OFF(__sk_flags_offset,
8212 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
8213 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg,
8215 /* SK_FL_PROTO_MASK and SK_FL_PROTO_SHIFT are endian
8216 * aware. No further narrowing or masking is needed.
8221 case offsetof(struct sk_reuseport_md, data_end):
8222 SK_REUSEPORT_LOAD_FIELD(data_end);
8225 case offsetof(struct sk_reuseport_md, hash):
8226 SK_REUSEPORT_LOAD_FIELD(hash);
8229 case offsetof(struct sk_reuseport_md, bind_inany):
8230 SK_REUSEPORT_LOAD_FIELD(bind_inany);
8234 return insn - insn_buf;
8237 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
8238 .get_func_proto = sk_reuseport_func_proto,
8239 .is_valid_access = sk_reuseport_is_valid_access,
8240 .convert_ctx_access = sk_reuseport_convert_ctx_access,
8243 const struct bpf_prog_ops sk_reuseport_prog_ops = {
8245 #endif /* CONFIG_INET */