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>
42 #include <net/flow_dissector.h>
43 #include <linux/errno.h>
44 #include <linux/timer.h>
45 #include <linux/uaccess.h>
46 #include <asm/unaligned.h>
47 #include <asm/cmpxchg.h>
48 #include <linux/filter.h>
49 #include <linux/ratelimit.h>
50 #include <linux/seccomp.h>
51 #include <linux/if_vlan.h>
52 #include <linux/bpf.h>
53 #include <net/sch_generic.h>
54 #include <net/cls_cgroup.h>
55 #include <net/dst_metadata.h>
57 #include <net/sock_reuseport.h>
58 #include <net/busy_poll.h>
61 #include <linux/bpf_trace.h>
62 #include <net/xdp_sock.h>
63 #include <linux/inetdevice.h>
64 #include <net/ip_fib.h>
68 #include <linux/seg6_local.h>
70 #include <net/seg6_local.h>
73 * sk_filter_trim_cap - run a packet through a socket filter
74 * @sk: sock associated with &sk_buff
75 * @skb: buffer to filter
76 * @cap: limit on how short the eBPF program may trim the packet
78 * Run the eBPF program and then cut skb->data to correct size returned by
79 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
80 * than pkt_len we keep whole skb->data. This is the socket level
81 * wrapper to BPF_PROG_RUN. It returns 0 if the packet should
82 * be accepted or -EPERM if the packet should be tossed.
85 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
88 struct sk_filter *filter;
91 * If the skb was allocated from pfmemalloc reserves, only
92 * allow SOCK_MEMALLOC sockets to use it as this socket is
95 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
96 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
99 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
103 err = security_sock_rcv_skb(sk, skb);
108 filter = rcu_dereference(sk->sk_filter);
110 struct sock *save_sk = skb->sk;
111 unsigned int pkt_len;
114 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
116 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
122 EXPORT_SYMBOL(sk_filter_trim_cap);
124 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
126 return skb_get_poff(skb);
129 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
133 if (skb_is_nonlinear(skb))
136 if (skb->len < sizeof(struct nlattr))
139 if (a > skb->len - sizeof(struct nlattr))
142 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
144 return (void *) nla - (void *) skb->data;
149 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
153 if (skb_is_nonlinear(skb))
156 if (skb->len < sizeof(struct nlattr))
159 if (a > skb->len - sizeof(struct nlattr))
162 nla = (struct nlattr *) &skb->data[a];
163 if (nla->nla_len > skb->len - a)
166 nla = nla_find_nested(nla, x);
168 return (void *) nla - (void *) skb->data;
173 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
174 data, int, headlen, int, offset)
177 const int len = sizeof(tmp);
180 if (headlen - offset >= len)
181 return *(u8 *)(data + offset);
182 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
185 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
193 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
196 return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
200 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
201 data, int, headlen, int, offset)
204 const int len = sizeof(tmp);
207 if (headlen - offset >= len)
208 return get_unaligned_be16(data + offset);
209 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
210 return be16_to_cpu(tmp);
212 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
214 return get_unaligned_be16(ptr);
220 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
223 return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
227 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
228 data, int, headlen, int, offset)
231 const int len = sizeof(tmp);
233 if (likely(offset >= 0)) {
234 if (headlen - offset >= len)
235 return get_unaligned_be32(data + offset);
236 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
237 return be32_to_cpu(tmp);
239 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
241 return get_unaligned_be32(ptr);
247 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
250 return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
254 BPF_CALL_0(bpf_get_raw_cpu_id)
256 return raw_smp_processor_id();
259 static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = {
260 .func = bpf_get_raw_cpu_id,
262 .ret_type = RET_INTEGER,
265 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
266 struct bpf_insn *insn_buf)
268 struct bpf_insn *insn = insn_buf;
272 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
274 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
275 offsetof(struct sk_buff, mark));
279 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
280 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
281 #ifdef __BIG_ENDIAN_BITFIELD
282 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
287 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2);
289 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
290 offsetof(struct sk_buff, queue_mapping));
293 case SKF_AD_VLAN_TAG:
294 case SKF_AD_VLAN_TAG_PRESENT:
295 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2);
296 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000);
298 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
299 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
300 offsetof(struct sk_buff, vlan_tci));
301 if (skb_field == SKF_AD_VLAN_TAG) {
302 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg,
306 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 12);
308 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
313 return insn - insn_buf;
316 static bool convert_bpf_extensions(struct sock_filter *fp,
317 struct bpf_insn **insnp)
319 struct bpf_insn *insn = *insnp;
323 case SKF_AD_OFF + SKF_AD_PROTOCOL:
324 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
326 /* A = *(u16 *) (CTX + offsetof(protocol)) */
327 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
328 offsetof(struct sk_buff, protocol));
329 /* A = ntohs(A) [emitting a nop or swap16] */
330 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
333 case SKF_AD_OFF + SKF_AD_PKTTYPE:
334 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
338 case SKF_AD_OFF + SKF_AD_IFINDEX:
339 case SKF_AD_OFF + SKF_AD_HATYPE:
340 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
341 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, type) != 2);
343 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
344 BPF_REG_TMP, BPF_REG_CTX,
345 offsetof(struct sk_buff, dev));
346 /* if (tmp != 0) goto pc + 1 */
347 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
348 *insn++ = BPF_EXIT_INSN();
349 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
350 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
351 offsetof(struct net_device, ifindex));
353 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
354 offsetof(struct net_device, type));
357 case SKF_AD_OFF + SKF_AD_MARK:
358 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
362 case SKF_AD_OFF + SKF_AD_RXHASH:
363 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
365 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
366 offsetof(struct sk_buff, hash));
369 case SKF_AD_OFF + SKF_AD_QUEUE:
370 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
374 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
375 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
376 BPF_REG_A, BPF_REG_CTX, insn);
380 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
381 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
382 BPF_REG_A, BPF_REG_CTX, insn);
386 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
387 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2);
389 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
390 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
391 offsetof(struct sk_buff, vlan_proto));
392 /* A = ntohs(A) [emitting a nop or swap16] */
393 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
396 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
397 case SKF_AD_OFF + SKF_AD_NLATTR:
398 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
399 case SKF_AD_OFF + SKF_AD_CPU:
400 case SKF_AD_OFF + SKF_AD_RANDOM:
402 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
404 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
406 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
407 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
409 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
410 *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
412 case SKF_AD_OFF + SKF_AD_NLATTR:
413 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
415 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
416 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
418 case SKF_AD_OFF + SKF_AD_CPU:
419 *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
421 case SKF_AD_OFF + SKF_AD_RANDOM:
422 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
423 bpf_user_rnd_init_once();
428 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
430 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
434 /* This is just a dummy call to avoid letting the compiler
435 * evict __bpf_call_base() as an optimization. Placed here
436 * where no-one bothers.
438 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
446 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
448 const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
449 int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
450 bool endian = BPF_SIZE(fp->code) == BPF_H ||
451 BPF_SIZE(fp->code) == BPF_W;
452 bool indirect = BPF_MODE(fp->code) == BPF_IND;
453 const int ip_align = NET_IP_ALIGN;
454 struct bpf_insn *insn = *insnp;
458 ((unaligned_ok && offset >= 0) ||
459 (!unaligned_ok && offset >= 0 &&
460 offset + ip_align >= 0 &&
461 offset + ip_align % size == 0))) {
462 bool ldx_off_ok = offset <= S16_MAX;
464 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
465 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
466 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
467 size, 2 + endian + (!ldx_off_ok * 2));
469 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
472 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
473 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
474 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
478 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
479 *insn++ = BPF_JMP_A(8);
482 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
483 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
484 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
486 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
488 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
490 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
493 switch (BPF_SIZE(fp->code)) {
495 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
498 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
501 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
507 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
508 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
509 *insn = BPF_EXIT_INSN();
516 * bpf_convert_filter - convert filter program
517 * @prog: the user passed filter program
518 * @len: the length of the user passed filter program
519 * @new_prog: allocated 'struct bpf_prog' or NULL
520 * @new_len: pointer to store length of converted program
521 * @seen_ld_abs: bool whether we've seen ld_abs/ind
523 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
524 * style extended BPF (eBPF).
525 * Conversion workflow:
527 * 1) First pass for calculating the new program length:
528 * bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
530 * 2) 2nd pass to remap in two passes: 1st pass finds new
531 * jump offsets, 2nd pass remapping:
532 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
534 static int bpf_convert_filter(struct sock_filter *prog, int len,
535 struct bpf_prog *new_prog, int *new_len,
538 int new_flen = 0, pass = 0, target, i, stack_off;
539 struct bpf_insn *new_insn, *first_insn = NULL;
540 struct sock_filter *fp;
544 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
545 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
547 if (len <= 0 || len > BPF_MAXINSNS)
551 first_insn = new_prog->insnsi;
552 addrs = kcalloc(len, sizeof(*addrs),
553 GFP_KERNEL | __GFP_NOWARN);
559 new_insn = first_insn;
562 /* Classic BPF related prologue emission. */
564 /* Classic BPF expects A and X to be reset first. These need
565 * to be guaranteed to be the first two instructions.
567 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
568 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
570 /* All programs must keep CTX in callee saved BPF_REG_CTX.
571 * In eBPF case it's done by the compiler, here we need to
572 * do this ourself. Initial CTX is present in BPF_REG_ARG1.
574 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
576 /* For packet access in classic BPF, cache skb->data
577 * in callee-saved BPF R8 and skb->len - skb->data_len
578 * (headlen) in BPF R9. Since classic BPF is read-only
579 * on CTX, we only need to cache it once.
581 *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
582 BPF_REG_D, BPF_REG_CTX,
583 offsetof(struct sk_buff, data));
584 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
585 offsetof(struct sk_buff, len));
586 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
587 offsetof(struct sk_buff, data_len));
588 *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
594 for (i = 0; i < len; fp++, i++) {
595 struct bpf_insn tmp_insns[32] = { };
596 struct bpf_insn *insn = tmp_insns;
599 addrs[i] = new_insn - first_insn;
602 /* All arithmetic insns and skb loads map as-is. */
603 case BPF_ALU | BPF_ADD | BPF_X:
604 case BPF_ALU | BPF_ADD | BPF_K:
605 case BPF_ALU | BPF_SUB | BPF_X:
606 case BPF_ALU | BPF_SUB | BPF_K:
607 case BPF_ALU | BPF_AND | BPF_X:
608 case BPF_ALU | BPF_AND | BPF_K:
609 case BPF_ALU | BPF_OR | BPF_X:
610 case BPF_ALU | BPF_OR | BPF_K:
611 case BPF_ALU | BPF_LSH | BPF_X:
612 case BPF_ALU | BPF_LSH | BPF_K:
613 case BPF_ALU | BPF_RSH | BPF_X:
614 case BPF_ALU | BPF_RSH | BPF_K:
615 case BPF_ALU | BPF_XOR | BPF_X:
616 case BPF_ALU | BPF_XOR | BPF_K:
617 case BPF_ALU | BPF_MUL | BPF_X:
618 case BPF_ALU | BPF_MUL | BPF_K:
619 case BPF_ALU | BPF_DIV | BPF_X:
620 case BPF_ALU | BPF_DIV | BPF_K:
621 case BPF_ALU | BPF_MOD | BPF_X:
622 case BPF_ALU | BPF_MOD | BPF_K:
623 case BPF_ALU | BPF_NEG:
624 case BPF_LD | BPF_ABS | BPF_W:
625 case BPF_LD | BPF_ABS | BPF_H:
626 case BPF_LD | BPF_ABS | BPF_B:
627 case BPF_LD | BPF_IND | BPF_W:
628 case BPF_LD | BPF_IND | BPF_H:
629 case BPF_LD | BPF_IND | BPF_B:
630 /* Check for overloaded BPF extension and
631 * directly convert it if found, otherwise
632 * just move on with mapping.
634 if (BPF_CLASS(fp->code) == BPF_LD &&
635 BPF_MODE(fp->code) == BPF_ABS &&
636 convert_bpf_extensions(fp, &insn))
638 if (BPF_CLASS(fp->code) == BPF_LD &&
639 convert_bpf_ld_abs(fp, &insn)) {
644 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
645 fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
646 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
647 /* Error with exception code on div/mod by 0.
648 * For cBPF programs, this was always return 0.
650 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
651 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
652 *insn++ = BPF_EXIT_INSN();
655 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
658 /* Jump transformation cannot use BPF block macros
659 * everywhere as offset calculation and target updates
660 * require a bit more work than the rest, i.e. jump
661 * opcodes map as-is, but offsets need adjustment.
664 #define BPF_EMIT_JMP \
666 const s32 off_min = S16_MIN, off_max = S16_MAX; \
669 if (target >= len || target < 0) \
671 off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
672 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
673 off -= insn - tmp_insns; \
674 /* Reject anything not fitting into insn->off. */ \
675 if (off < off_min || off > off_max) \
680 case BPF_JMP | BPF_JA:
681 target = i + fp->k + 1;
682 insn->code = fp->code;
686 case BPF_JMP | BPF_JEQ | BPF_K:
687 case BPF_JMP | BPF_JEQ | BPF_X:
688 case BPF_JMP | BPF_JSET | BPF_K:
689 case BPF_JMP | BPF_JSET | BPF_X:
690 case BPF_JMP | BPF_JGT | BPF_K:
691 case BPF_JMP | BPF_JGT | BPF_X:
692 case BPF_JMP | BPF_JGE | BPF_K:
693 case BPF_JMP | BPF_JGE | BPF_X:
694 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
695 /* BPF immediates are signed, zero extend
696 * immediate into tmp register and use it
699 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
701 insn->dst_reg = BPF_REG_A;
702 insn->src_reg = BPF_REG_TMP;
705 insn->dst_reg = BPF_REG_A;
707 bpf_src = BPF_SRC(fp->code);
708 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
711 /* Common case where 'jump_false' is next insn. */
713 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
714 target = i + fp->jt + 1;
719 /* Convert some jumps when 'jump_true' is next insn. */
721 switch (BPF_OP(fp->code)) {
723 insn->code = BPF_JMP | BPF_JNE | bpf_src;
726 insn->code = BPF_JMP | BPF_JLE | bpf_src;
729 insn->code = BPF_JMP | BPF_JLT | bpf_src;
735 target = i + fp->jf + 1;
740 /* Other jumps are mapped into two insns: Jxx and JA. */
741 target = i + fp->jt + 1;
742 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
746 insn->code = BPF_JMP | BPF_JA;
747 target = i + fp->jf + 1;
751 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
752 case BPF_LDX | BPF_MSH | BPF_B: {
753 struct sock_filter tmp = {
754 .code = BPF_LD | BPF_ABS | BPF_B,
761 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
762 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
763 convert_bpf_ld_abs(&tmp, &insn);
766 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
768 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
770 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
772 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
774 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
777 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
778 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
780 case BPF_RET | BPF_A:
781 case BPF_RET | BPF_K:
782 if (BPF_RVAL(fp->code) == BPF_K)
783 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
785 *insn = BPF_EXIT_INSN();
788 /* Store to stack. */
791 stack_off = fp->k * 4 + 4;
792 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
793 BPF_ST ? BPF_REG_A : BPF_REG_X,
795 /* check_load_and_stores() verifies that classic BPF can
796 * load from stack only after write, so tracking
797 * stack_depth for ST|STX insns is enough
799 if (new_prog && new_prog->aux->stack_depth < stack_off)
800 new_prog->aux->stack_depth = stack_off;
803 /* Load from stack. */
804 case BPF_LD | BPF_MEM:
805 case BPF_LDX | BPF_MEM:
806 stack_off = fp->k * 4 + 4;
807 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
808 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
813 case BPF_LD | BPF_IMM:
814 case BPF_LDX | BPF_IMM:
815 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
816 BPF_REG_A : BPF_REG_X, fp->k);
820 case BPF_MISC | BPF_TAX:
821 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
825 case BPF_MISC | BPF_TXA:
826 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
829 /* A = skb->len or X = skb->len */
830 case BPF_LD | BPF_W | BPF_LEN:
831 case BPF_LDX | BPF_W | BPF_LEN:
832 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
833 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
834 offsetof(struct sk_buff, len));
837 /* Access seccomp_data fields. */
838 case BPF_LDX | BPF_ABS | BPF_W:
839 /* A = *(u32 *) (ctx + K) */
840 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
843 /* Unknown instruction. */
850 memcpy(new_insn, tmp_insns,
851 sizeof(*insn) * (insn - tmp_insns));
852 new_insn += insn - tmp_insns;
856 /* Only calculating new length. */
857 *new_len = new_insn - first_insn;
859 *new_len += 4; /* Prologue bits. */
864 if (new_flen != new_insn - first_insn) {
865 new_flen = new_insn - first_insn;
872 BUG_ON(*new_len != new_flen);
881 * As we dont want to clear mem[] array for each packet going through
882 * __bpf_prog_run(), we check that filter loaded by user never try to read
883 * a cell if not previously written, and we check all branches to be sure
884 * a malicious user doesn't try to abuse us.
886 static int check_load_and_stores(const struct sock_filter *filter, int flen)
888 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
891 BUILD_BUG_ON(BPF_MEMWORDS > 16);
893 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
897 memset(masks, 0xff, flen * sizeof(*masks));
899 for (pc = 0; pc < flen; pc++) {
900 memvalid &= masks[pc];
902 switch (filter[pc].code) {
905 memvalid |= (1 << filter[pc].k);
907 case BPF_LD | BPF_MEM:
908 case BPF_LDX | BPF_MEM:
909 if (!(memvalid & (1 << filter[pc].k))) {
914 case BPF_JMP | BPF_JA:
915 /* A jump must set masks on target */
916 masks[pc + 1 + filter[pc].k] &= memvalid;
919 case BPF_JMP | BPF_JEQ | BPF_K:
920 case BPF_JMP | BPF_JEQ | BPF_X:
921 case BPF_JMP | BPF_JGE | BPF_K:
922 case BPF_JMP | BPF_JGE | BPF_X:
923 case BPF_JMP | BPF_JGT | BPF_K:
924 case BPF_JMP | BPF_JGT | BPF_X:
925 case BPF_JMP | BPF_JSET | BPF_K:
926 case BPF_JMP | BPF_JSET | BPF_X:
927 /* A jump must set masks on targets */
928 masks[pc + 1 + filter[pc].jt] &= memvalid;
929 masks[pc + 1 + filter[pc].jf] &= memvalid;
939 static bool chk_code_allowed(u16 code_to_probe)
941 static const bool codes[] = {
942 /* 32 bit ALU operations */
943 [BPF_ALU | BPF_ADD | BPF_K] = true,
944 [BPF_ALU | BPF_ADD | BPF_X] = true,
945 [BPF_ALU | BPF_SUB | BPF_K] = true,
946 [BPF_ALU | BPF_SUB | BPF_X] = true,
947 [BPF_ALU | BPF_MUL | BPF_K] = true,
948 [BPF_ALU | BPF_MUL | BPF_X] = true,
949 [BPF_ALU | BPF_DIV | BPF_K] = true,
950 [BPF_ALU | BPF_DIV | BPF_X] = true,
951 [BPF_ALU | BPF_MOD | BPF_K] = true,
952 [BPF_ALU | BPF_MOD | BPF_X] = true,
953 [BPF_ALU | BPF_AND | BPF_K] = true,
954 [BPF_ALU | BPF_AND | BPF_X] = true,
955 [BPF_ALU | BPF_OR | BPF_K] = true,
956 [BPF_ALU | BPF_OR | BPF_X] = true,
957 [BPF_ALU | BPF_XOR | BPF_K] = true,
958 [BPF_ALU | BPF_XOR | BPF_X] = true,
959 [BPF_ALU | BPF_LSH | BPF_K] = true,
960 [BPF_ALU | BPF_LSH | BPF_X] = true,
961 [BPF_ALU | BPF_RSH | BPF_K] = true,
962 [BPF_ALU | BPF_RSH | BPF_X] = true,
963 [BPF_ALU | BPF_NEG] = true,
964 /* Load instructions */
965 [BPF_LD | BPF_W | BPF_ABS] = true,
966 [BPF_LD | BPF_H | BPF_ABS] = true,
967 [BPF_LD | BPF_B | BPF_ABS] = true,
968 [BPF_LD | BPF_W | BPF_LEN] = true,
969 [BPF_LD | BPF_W | BPF_IND] = true,
970 [BPF_LD | BPF_H | BPF_IND] = true,
971 [BPF_LD | BPF_B | BPF_IND] = true,
972 [BPF_LD | BPF_IMM] = true,
973 [BPF_LD | BPF_MEM] = true,
974 [BPF_LDX | BPF_W | BPF_LEN] = true,
975 [BPF_LDX | BPF_B | BPF_MSH] = true,
976 [BPF_LDX | BPF_IMM] = true,
977 [BPF_LDX | BPF_MEM] = true,
978 /* Store instructions */
981 /* Misc instructions */
982 [BPF_MISC | BPF_TAX] = true,
983 [BPF_MISC | BPF_TXA] = true,
984 /* Return instructions */
985 [BPF_RET | BPF_K] = true,
986 [BPF_RET | BPF_A] = true,
987 /* Jump instructions */
988 [BPF_JMP | BPF_JA] = true,
989 [BPF_JMP | BPF_JEQ | BPF_K] = true,
990 [BPF_JMP | BPF_JEQ | BPF_X] = true,
991 [BPF_JMP | BPF_JGE | BPF_K] = true,
992 [BPF_JMP | BPF_JGE | BPF_X] = true,
993 [BPF_JMP | BPF_JGT | BPF_K] = true,
994 [BPF_JMP | BPF_JGT | BPF_X] = true,
995 [BPF_JMP | BPF_JSET | BPF_K] = true,
996 [BPF_JMP | BPF_JSET | BPF_X] = true,
999 if (code_to_probe >= ARRAY_SIZE(codes))
1002 return codes[code_to_probe];
1005 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1010 if (flen == 0 || flen > BPF_MAXINSNS)
1017 * bpf_check_classic - verify socket filter code
1018 * @filter: filter to verify
1019 * @flen: length of filter
1021 * Check the user's filter code. If we let some ugly
1022 * filter code slip through kaboom! The filter must contain
1023 * no references or jumps that are out of range, no illegal
1024 * instructions, and must end with a RET instruction.
1026 * All jumps are forward as they are not signed.
1028 * Returns 0 if the rule set is legal or -EINVAL if not.
1030 static int bpf_check_classic(const struct sock_filter *filter,
1036 /* Check the filter code now */
1037 for (pc = 0; pc < flen; pc++) {
1038 const struct sock_filter *ftest = &filter[pc];
1040 /* May we actually operate on this code? */
1041 if (!chk_code_allowed(ftest->code))
1044 /* Some instructions need special checks */
1045 switch (ftest->code) {
1046 case BPF_ALU | BPF_DIV | BPF_K:
1047 case BPF_ALU | BPF_MOD | BPF_K:
1048 /* Check for division by zero */
1052 case BPF_ALU | BPF_LSH | BPF_K:
1053 case BPF_ALU | BPF_RSH | BPF_K:
1057 case BPF_LD | BPF_MEM:
1058 case BPF_LDX | BPF_MEM:
1061 /* Check for invalid memory addresses */
1062 if (ftest->k >= BPF_MEMWORDS)
1065 case BPF_JMP | BPF_JA:
1066 /* Note, the large ftest->k might cause loops.
1067 * Compare this with conditional jumps below,
1068 * where offsets are limited. --ANK (981016)
1070 if (ftest->k >= (unsigned int)(flen - pc - 1))
1073 case BPF_JMP | BPF_JEQ | BPF_K:
1074 case BPF_JMP | BPF_JEQ | BPF_X:
1075 case BPF_JMP | BPF_JGE | BPF_K:
1076 case BPF_JMP | BPF_JGE | BPF_X:
1077 case BPF_JMP | BPF_JGT | BPF_K:
1078 case BPF_JMP | BPF_JGT | BPF_X:
1079 case BPF_JMP | BPF_JSET | BPF_K:
1080 case BPF_JMP | BPF_JSET | BPF_X:
1081 /* Both conditionals must be safe */
1082 if (pc + ftest->jt + 1 >= flen ||
1083 pc + ftest->jf + 1 >= flen)
1086 case BPF_LD | BPF_W | BPF_ABS:
1087 case BPF_LD | BPF_H | BPF_ABS:
1088 case BPF_LD | BPF_B | BPF_ABS:
1090 if (bpf_anc_helper(ftest) & BPF_ANC)
1092 /* Ancillary operation unknown or unsupported */
1093 if (anc_found == false && ftest->k >= SKF_AD_OFF)
1098 /* Last instruction must be a RET code */
1099 switch (filter[flen - 1].code) {
1100 case BPF_RET | BPF_K:
1101 case BPF_RET | BPF_A:
1102 return check_load_and_stores(filter, flen);
1108 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1109 const struct sock_fprog *fprog)
1111 unsigned int fsize = bpf_classic_proglen(fprog);
1112 struct sock_fprog_kern *fkprog;
1114 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1118 fkprog = fp->orig_prog;
1119 fkprog->len = fprog->len;
1121 fkprog->filter = kmemdup(fp->insns, fsize,
1122 GFP_KERNEL | __GFP_NOWARN);
1123 if (!fkprog->filter) {
1124 kfree(fp->orig_prog);
1131 static void bpf_release_orig_filter(struct bpf_prog *fp)
1133 struct sock_fprog_kern *fprog = fp->orig_prog;
1136 kfree(fprog->filter);
1141 static void __bpf_prog_release(struct bpf_prog *prog)
1143 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1146 bpf_release_orig_filter(prog);
1147 bpf_prog_free(prog);
1151 static void __sk_filter_release(struct sk_filter *fp)
1153 __bpf_prog_release(fp->prog);
1158 * sk_filter_release_rcu - Release a socket filter by rcu_head
1159 * @rcu: rcu_head that contains the sk_filter to free
1161 static void sk_filter_release_rcu(struct rcu_head *rcu)
1163 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1165 __sk_filter_release(fp);
1169 * sk_filter_release - release a socket filter
1170 * @fp: filter to remove
1172 * Remove a filter from a socket and release its resources.
1174 static void sk_filter_release(struct sk_filter *fp)
1176 if (refcount_dec_and_test(&fp->refcnt))
1177 call_rcu(&fp->rcu, sk_filter_release_rcu);
1180 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1182 u32 filter_size = bpf_prog_size(fp->prog->len);
1184 atomic_sub(filter_size, &sk->sk_omem_alloc);
1185 sk_filter_release(fp);
1188 /* try to charge the socket memory if there is space available
1189 * return true on success
1191 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1193 u32 filter_size = bpf_prog_size(fp->prog->len);
1195 /* same check as in sock_kmalloc() */
1196 if (filter_size <= sysctl_optmem_max &&
1197 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
1198 atomic_add(filter_size, &sk->sk_omem_alloc);
1204 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1206 if (!refcount_inc_not_zero(&fp->refcnt))
1209 if (!__sk_filter_charge(sk, fp)) {
1210 sk_filter_release(fp);
1216 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1218 struct sock_filter *old_prog;
1219 struct bpf_prog *old_fp;
1220 int err, new_len, old_len = fp->len;
1221 bool seen_ld_abs = false;
1223 /* We are free to overwrite insns et al right here as it
1224 * won't be used at this point in time anymore internally
1225 * after the migration to the internal BPF instruction
1228 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1229 sizeof(struct bpf_insn));
1231 /* Conversion cannot happen on overlapping memory areas,
1232 * so we need to keep the user BPF around until the 2nd
1233 * pass. At this time, the user BPF is stored in fp->insns.
1235 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1236 GFP_KERNEL | __GFP_NOWARN);
1242 /* 1st pass: calculate the new program length. */
1243 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1248 /* Expand fp for appending the new filter representation. */
1250 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1252 /* The old_fp is still around in case we couldn't
1253 * allocate new memory, so uncharge on that one.
1262 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1263 err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1266 /* 2nd bpf_convert_filter() can fail only if it fails
1267 * to allocate memory, remapping must succeed. Note,
1268 * that at this time old_fp has already been released
1273 fp = bpf_prog_select_runtime(fp, &err);
1283 __bpf_prog_release(fp);
1284 return ERR_PTR(err);
1287 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1288 bpf_aux_classic_check_t trans)
1292 fp->bpf_func = NULL;
1295 err = bpf_check_classic(fp->insns, fp->len);
1297 __bpf_prog_release(fp);
1298 return ERR_PTR(err);
1301 /* There might be additional checks and transformations
1302 * needed on classic filters, f.e. in case of seccomp.
1305 err = trans(fp->insns, fp->len);
1307 __bpf_prog_release(fp);
1308 return ERR_PTR(err);
1312 /* Probe if we can JIT compile the filter and if so, do
1313 * the compilation of the filter.
1315 bpf_jit_compile(fp);
1317 /* JIT compiler couldn't process this filter, so do the
1318 * internal BPF translation for the optimized interpreter.
1321 fp = bpf_migrate_filter(fp);
1327 * bpf_prog_create - create an unattached filter
1328 * @pfp: the unattached filter that is created
1329 * @fprog: the filter program
1331 * Create a filter independent of any socket. We first run some
1332 * sanity checks on it to make sure it does not explode on us later.
1333 * If an error occurs or there is insufficient memory for the filter
1334 * a negative errno code is returned. On success the return is zero.
1336 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1338 unsigned int fsize = bpf_classic_proglen(fprog);
1339 struct bpf_prog *fp;
1341 /* Make sure new filter is there and in the right amounts. */
1342 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1345 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1349 memcpy(fp->insns, fprog->filter, fsize);
1351 fp->len = fprog->len;
1352 /* Since unattached filters are not copied back to user
1353 * space through sk_get_filter(), we do not need to hold
1354 * a copy here, and can spare us the work.
1356 fp->orig_prog = NULL;
1358 /* bpf_prepare_filter() already takes care of freeing
1359 * memory in case something goes wrong.
1361 fp = bpf_prepare_filter(fp, NULL);
1368 EXPORT_SYMBOL_GPL(bpf_prog_create);
1371 * bpf_prog_create_from_user - create an unattached filter from user buffer
1372 * @pfp: the unattached filter that is created
1373 * @fprog: the filter program
1374 * @trans: post-classic verifier transformation handler
1375 * @save_orig: save classic BPF program
1377 * This function effectively does the same as bpf_prog_create(), only
1378 * that it builds up its insns buffer from user space provided buffer.
1379 * It also allows for passing a bpf_aux_classic_check_t handler.
1381 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1382 bpf_aux_classic_check_t trans, bool save_orig)
1384 unsigned int fsize = bpf_classic_proglen(fprog);
1385 struct bpf_prog *fp;
1388 /* Make sure new filter is there and in the right amounts. */
1389 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1392 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1396 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1397 __bpf_prog_free(fp);
1401 fp->len = fprog->len;
1402 fp->orig_prog = NULL;
1405 err = bpf_prog_store_orig_filter(fp, fprog);
1407 __bpf_prog_free(fp);
1412 /* bpf_prepare_filter() already takes care of freeing
1413 * memory in case something goes wrong.
1415 fp = bpf_prepare_filter(fp, trans);
1422 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1424 void bpf_prog_destroy(struct bpf_prog *fp)
1426 __bpf_prog_release(fp);
1428 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1430 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1432 struct sk_filter *fp, *old_fp;
1434 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1440 if (!__sk_filter_charge(sk, fp)) {
1444 refcount_set(&fp->refcnt, 1);
1446 old_fp = rcu_dereference_protected(sk->sk_filter,
1447 lockdep_sock_is_held(sk));
1448 rcu_assign_pointer(sk->sk_filter, fp);
1451 sk_filter_uncharge(sk, old_fp);
1457 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1459 unsigned int fsize = bpf_classic_proglen(fprog);
1460 struct bpf_prog *prog;
1463 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1464 return ERR_PTR(-EPERM);
1466 /* Make sure new filter is there and in the right amounts. */
1467 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1468 return ERR_PTR(-EINVAL);
1470 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1472 return ERR_PTR(-ENOMEM);
1474 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1475 __bpf_prog_free(prog);
1476 return ERR_PTR(-EFAULT);
1479 prog->len = fprog->len;
1481 err = bpf_prog_store_orig_filter(prog, fprog);
1483 __bpf_prog_free(prog);
1484 return ERR_PTR(-ENOMEM);
1487 /* bpf_prepare_filter() already takes care of freeing
1488 * memory in case something goes wrong.
1490 return bpf_prepare_filter(prog, NULL);
1494 * sk_attach_filter - attach a socket filter
1495 * @fprog: the filter program
1496 * @sk: the socket to use
1498 * Attach the user's filter code. We first run some sanity checks on
1499 * it to make sure it does not explode on us later. If an error
1500 * occurs or there is insufficient memory for the filter a negative
1501 * errno code is returned. On success the return is zero.
1503 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1505 struct bpf_prog *prog = __get_filter(fprog, sk);
1509 return PTR_ERR(prog);
1511 err = __sk_attach_prog(prog, sk);
1513 __bpf_prog_release(prog);
1519 EXPORT_SYMBOL_GPL(sk_attach_filter);
1521 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1523 struct bpf_prog *prog = __get_filter(fprog, sk);
1527 return PTR_ERR(prog);
1529 if (bpf_prog_size(prog->len) > sysctl_optmem_max)
1532 err = reuseport_attach_prog(sk, prog);
1535 __bpf_prog_release(prog);
1540 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1542 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1543 return ERR_PTR(-EPERM);
1545 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1548 int sk_attach_bpf(u32 ufd, struct sock *sk)
1550 struct bpf_prog *prog = __get_bpf(ufd, sk);
1554 return PTR_ERR(prog);
1556 err = __sk_attach_prog(prog, sk);
1565 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1567 struct bpf_prog *prog;
1570 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1573 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1574 if (IS_ERR(prog) && PTR_ERR(prog) == -EINVAL)
1575 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1577 return PTR_ERR(prog);
1579 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1580 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1581 * bpf prog (e.g. sockmap). It depends on the
1582 * limitation imposed by bpf_prog_load().
1583 * Hence, sysctl_optmem_max is not checked.
1585 if ((sk->sk_type != SOCK_STREAM &&
1586 sk->sk_type != SOCK_DGRAM) ||
1587 (sk->sk_protocol != IPPROTO_UDP &&
1588 sk->sk_protocol != IPPROTO_TCP) ||
1589 (sk->sk_family != AF_INET &&
1590 sk->sk_family != AF_INET6)) {
1595 /* BPF_PROG_TYPE_SOCKET_FILTER */
1596 if (bpf_prog_size(prog->len) > sysctl_optmem_max) {
1602 err = reuseport_attach_prog(sk, prog);
1610 void sk_reuseport_prog_free(struct bpf_prog *prog)
1615 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1618 bpf_prog_destroy(prog);
1621 struct bpf_scratchpad {
1623 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1624 u8 buff[MAX_BPF_STACK];
1628 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1630 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1631 unsigned int write_len)
1633 return skb_ensure_writable(skb, write_len);
1636 static inline int bpf_try_make_writable(struct sk_buff *skb,
1637 unsigned int write_len)
1639 int err = __bpf_try_make_writable(skb, write_len);
1641 bpf_compute_data_pointers(skb);
1645 static int bpf_try_make_head_writable(struct sk_buff *skb)
1647 return bpf_try_make_writable(skb, skb_headlen(skb));
1650 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1652 if (skb_at_tc_ingress(skb))
1653 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1656 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1658 if (skb_at_tc_ingress(skb))
1659 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1662 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1663 const void *, from, u32, len, u64, flags)
1667 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1669 if (unlikely(offset > 0xffff))
1671 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1674 ptr = skb->data + offset;
1675 if (flags & BPF_F_RECOMPUTE_CSUM)
1676 __skb_postpull_rcsum(skb, ptr, len, offset);
1678 memcpy(ptr, from, len);
1680 if (flags & BPF_F_RECOMPUTE_CSUM)
1681 __skb_postpush_rcsum(skb, ptr, len, offset);
1682 if (flags & BPF_F_INVALIDATE_HASH)
1683 skb_clear_hash(skb);
1688 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1689 .func = bpf_skb_store_bytes,
1691 .ret_type = RET_INTEGER,
1692 .arg1_type = ARG_PTR_TO_CTX,
1693 .arg2_type = ARG_ANYTHING,
1694 .arg3_type = ARG_PTR_TO_MEM,
1695 .arg4_type = ARG_CONST_SIZE,
1696 .arg5_type = ARG_ANYTHING,
1699 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1700 void *, to, u32, len)
1704 if (unlikely(offset > 0xffff))
1707 ptr = skb_header_pointer(skb, offset, len, to);
1711 memcpy(to, ptr, len);
1719 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1720 .func = bpf_skb_load_bytes,
1722 .ret_type = RET_INTEGER,
1723 .arg1_type = ARG_PTR_TO_CTX,
1724 .arg2_type = ARG_ANYTHING,
1725 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1726 .arg4_type = ARG_CONST_SIZE,
1729 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1730 u32, offset, void *, to, u32, len, u32, start_header)
1732 u8 *end = skb_tail_pointer(skb);
1733 u8 *net = skb_network_header(skb);
1734 u8 *mac = skb_mac_header(skb);
1737 if (unlikely(offset > 0xffff || len > (end - mac)))
1740 switch (start_header) {
1741 case BPF_HDR_START_MAC:
1744 case BPF_HDR_START_NET:
1751 if (likely(ptr >= mac && ptr + len <= end)) {
1752 memcpy(to, ptr, len);
1761 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1762 .func = bpf_skb_load_bytes_relative,
1764 .ret_type = RET_INTEGER,
1765 .arg1_type = ARG_PTR_TO_CTX,
1766 .arg2_type = ARG_ANYTHING,
1767 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1768 .arg4_type = ARG_CONST_SIZE,
1769 .arg5_type = ARG_ANYTHING,
1772 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1774 /* Idea is the following: should the needed direct read/write
1775 * test fail during runtime, we can pull in more data and redo
1776 * again, since implicitly, we invalidate previous checks here.
1778 * Or, since we know how much we need to make read/writeable,
1779 * this can be done once at the program beginning for direct
1780 * access case. By this we overcome limitations of only current
1781 * headroom being accessible.
1783 return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1786 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1787 .func = bpf_skb_pull_data,
1789 .ret_type = RET_INTEGER,
1790 .arg1_type = ARG_PTR_TO_CTX,
1791 .arg2_type = ARG_ANYTHING,
1794 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1795 unsigned int write_len)
1797 int err = __bpf_try_make_writable(skb, write_len);
1799 bpf_compute_data_end_sk_skb(skb);
1803 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1805 /* Idea is the following: should the needed direct read/write
1806 * test fail during runtime, we can pull in more data and redo
1807 * again, since implicitly, we invalidate previous checks here.
1809 * Or, since we know how much we need to make read/writeable,
1810 * this can be done once at the program beginning for direct
1811 * access case. By this we overcome limitations of only current
1812 * headroom being accessible.
1814 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1817 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1818 .func = sk_skb_pull_data,
1820 .ret_type = RET_INTEGER,
1821 .arg1_type = ARG_PTR_TO_CTX,
1822 .arg2_type = ARG_ANYTHING,
1825 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1826 u64, from, u64, to, u64, flags)
1830 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1832 if (unlikely(offset > 0xffff || offset & 1))
1834 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1837 ptr = (__sum16 *)(skb->data + offset);
1838 switch (flags & BPF_F_HDR_FIELD_MASK) {
1840 if (unlikely(from != 0))
1843 csum_replace_by_diff(ptr, to);
1846 csum_replace2(ptr, from, to);
1849 csum_replace4(ptr, from, to);
1858 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1859 .func = bpf_l3_csum_replace,
1861 .ret_type = RET_INTEGER,
1862 .arg1_type = ARG_PTR_TO_CTX,
1863 .arg2_type = ARG_ANYTHING,
1864 .arg3_type = ARG_ANYTHING,
1865 .arg4_type = ARG_ANYTHING,
1866 .arg5_type = ARG_ANYTHING,
1869 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1870 u64, from, u64, to, u64, flags)
1872 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1873 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1874 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1877 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1878 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1880 if (unlikely(offset > 0xffff || offset & 1))
1882 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1885 ptr = (__sum16 *)(skb->data + offset);
1886 if (is_mmzero && !do_mforce && !*ptr)
1889 switch (flags & BPF_F_HDR_FIELD_MASK) {
1891 if (unlikely(from != 0))
1894 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1897 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1900 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1906 if (is_mmzero && !*ptr)
1907 *ptr = CSUM_MANGLED_0;
1911 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1912 .func = bpf_l4_csum_replace,
1914 .ret_type = RET_INTEGER,
1915 .arg1_type = ARG_PTR_TO_CTX,
1916 .arg2_type = ARG_ANYTHING,
1917 .arg3_type = ARG_ANYTHING,
1918 .arg4_type = ARG_ANYTHING,
1919 .arg5_type = ARG_ANYTHING,
1922 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1923 __be32 *, to, u32, to_size, __wsum, seed)
1925 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1926 u32 diff_size = from_size + to_size;
1929 /* This is quite flexible, some examples:
1931 * from_size == 0, to_size > 0, seed := csum --> pushing data
1932 * from_size > 0, to_size == 0, seed := csum --> pulling data
1933 * from_size > 0, to_size > 0, seed := 0 --> diffing data
1935 * Even for diffing, from_size and to_size don't need to be equal.
1937 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
1938 diff_size > sizeof(sp->diff)))
1941 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
1942 sp->diff[j] = ~from[i];
1943 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
1944 sp->diff[j] = to[i];
1946 return csum_partial(sp->diff, diff_size, seed);
1949 static const struct bpf_func_proto bpf_csum_diff_proto = {
1950 .func = bpf_csum_diff,
1953 .ret_type = RET_INTEGER,
1954 .arg1_type = ARG_PTR_TO_MEM_OR_NULL,
1955 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
1956 .arg3_type = ARG_PTR_TO_MEM_OR_NULL,
1957 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
1958 .arg5_type = ARG_ANYTHING,
1961 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
1963 /* The interface is to be used in combination with bpf_csum_diff()
1964 * for direct packet writes. csum rotation for alignment as well
1965 * as emulating csum_sub() can be done from the eBPF program.
1967 if (skb->ip_summed == CHECKSUM_COMPLETE)
1968 return (skb->csum = csum_add(skb->csum, csum));
1973 static const struct bpf_func_proto bpf_csum_update_proto = {
1974 .func = bpf_csum_update,
1976 .ret_type = RET_INTEGER,
1977 .arg1_type = ARG_PTR_TO_CTX,
1978 .arg2_type = ARG_ANYTHING,
1981 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
1983 return dev_forward_skb(dev, skb);
1986 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
1987 struct sk_buff *skb)
1989 int ret = ____dev_forward_skb(dev, skb);
1993 ret = netif_rx(skb);
1999 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2003 if (unlikely(__this_cpu_read(xmit_recursion) > XMIT_RECURSION_LIMIT)) {
2004 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2011 __this_cpu_inc(xmit_recursion);
2012 ret = dev_queue_xmit(skb);
2013 __this_cpu_dec(xmit_recursion);
2018 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2021 /* skb->mac_len is not set on normal egress */
2022 unsigned int mlen = skb->network_header - skb->mac_header;
2024 __skb_pull(skb, mlen);
2026 /* At ingress, the mac header has already been pulled once.
2027 * At egress, skb_pospull_rcsum has to be done in case that
2028 * the skb is originated from ingress (i.e. a forwarded skb)
2029 * to ensure that rcsum starts at net header.
2031 if (!skb_at_tc_ingress(skb))
2032 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2033 skb_pop_mac_header(skb);
2034 skb_reset_mac_len(skb);
2035 return flags & BPF_F_INGRESS ?
2036 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2039 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2042 /* Verify that a link layer header is carried */
2043 if (unlikely(skb->mac_header >= skb->network_header)) {
2048 bpf_push_mac_rcsum(skb);
2049 return flags & BPF_F_INGRESS ?
2050 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2053 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2056 if (dev_is_mac_header_xmit(dev))
2057 return __bpf_redirect_common(skb, dev, flags);
2059 return __bpf_redirect_no_mac(skb, dev, flags);
2062 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2064 struct net_device *dev;
2065 struct sk_buff *clone;
2068 if (unlikely(flags & ~(BPF_F_INGRESS)))
2071 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2075 clone = skb_clone(skb, GFP_ATOMIC);
2076 if (unlikely(!clone))
2079 /* For direct write, we need to keep the invariant that the skbs
2080 * we're dealing with need to be uncloned. Should uncloning fail
2081 * here, we need to free the just generated clone to unclone once
2084 ret = bpf_try_make_head_writable(skb);
2085 if (unlikely(ret)) {
2090 return __bpf_redirect(clone, dev, flags);
2093 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2094 .func = bpf_clone_redirect,
2096 .ret_type = RET_INTEGER,
2097 .arg1_type = ARG_PTR_TO_CTX,
2098 .arg2_type = ARG_ANYTHING,
2099 .arg3_type = ARG_ANYTHING,
2102 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2103 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2105 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2107 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2109 if (unlikely(flags & ~(BPF_F_INGRESS)))
2112 ri->ifindex = ifindex;
2115 return TC_ACT_REDIRECT;
2118 int skb_do_redirect(struct sk_buff *skb)
2120 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2121 struct net_device *dev;
2123 dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->ifindex);
2125 if (unlikely(!dev)) {
2130 return __bpf_redirect(skb, dev, ri->flags);
2133 static const struct bpf_func_proto bpf_redirect_proto = {
2134 .func = bpf_redirect,
2136 .ret_type = RET_INTEGER,
2137 .arg1_type = ARG_ANYTHING,
2138 .arg2_type = ARG_ANYTHING,
2141 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
2142 struct bpf_map *, map, void *, key, u64, flags)
2144 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
2146 /* If user passes invalid input drop the packet. */
2147 if (unlikely(flags & ~(BPF_F_INGRESS)))
2150 tcb->bpf.flags = flags;
2151 tcb->bpf.sk_redir = __sock_hash_lookup_elem(map, key);
2152 if (!tcb->bpf.sk_redir)
2158 static const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
2159 .func = bpf_sk_redirect_hash,
2161 .ret_type = RET_INTEGER,
2162 .arg1_type = ARG_PTR_TO_CTX,
2163 .arg2_type = ARG_CONST_MAP_PTR,
2164 .arg3_type = ARG_PTR_TO_MAP_KEY,
2165 .arg4_type = ARG_ANYTHING,
2168 BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
2169 struct bpf_map *, map, u32, key, u64, flags)
2171 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
2173 /* If user passes invalid input drop the packet. */
2174 if (unlikely(flags & ~(BPF_F_INGRESS)))
2177 tcb->bpf.flags = flags;
2178 tcb->bpf.sk_redir = __sock_map_lookup_elem(map, key);
2179 if (!tcb->bpf.sk_redir)
2185 struct sock *do_sk_redirect_map(struct sk_buff *skb)
2187 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
2189 return tcb->bpf.sk_redir;
2192 static const struct bpf_func_proto bpf_sk_redirect_map_proto = {
2193 .func = bpf_sk_redirect_map,
2195 .ret_type = RET_INTEGER,
2196 .arg1_type = ARG_PTR_TO_CTX,
2197 .arg2_type = ARG_CONST_MAP_PTR,
2198 .arg3_type = ARG_ANYTHING,
2199 .arg4_type = ARG_ANYTHING,
2202 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg_buff *, msg,
2203 struct bpf_map *, map, void *, key, u64, flags)
2205 /* If user passes invalid input drop the packet. */
2206 if (unlikely(flags & ~(BPF_F_INGRESS)))
2210 msg->sk_redir = __sock_hash_lookup_elem(map, key);
2217 static const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
2218 .func = bpf_msg_redirect_hash,
2220 .ret_type = RET_INTEGER,
2221 .arg1_type = ARG_PTR_TO_CTX,
2222 .arg2_type = ARG_CONST_MAP_PTR,
2223 .arg3_type = ARG_PTR_TO_MAP_KEY,
2224 .arg4_type = ARG_ANYTHING,
2227 BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg_buff *, msg,
2228 struct bpf_map *, map, u32, key, u64, flags)
2230 /* If user passes invalid input drop the packet. */
2231 if (unlikely(flags & ~(BPF_F_INGRESS)))
2235 msg->sk_redir = __sock_map_lookup_elem(map, key);
2242 struct sock *do_msg_redirect_map(struct sk_msg_buff *msg)
2244 return msg->sk_redir;
2247 static const struct bpf_func_proto bpf_msg_redirect_map_proto = {
2248 .func = bpf_msg_redirect_map,
2250 .ret_type = RET_INTEGER,
2251 .arg1_type = ARG_PTR_TO_CTX,
2252 .arg2_type = ARG_CONST_MAP_PTR,
2253 .arg3_type = ARG_ANYTHING,
2254 .arg4_type = ARG_ANYTHING,
2257 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg_buff *, msg, u32, bytes)
2259 msg->apply_bytes = bytes;
2263 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2264 .func = bpf_msg_apply_bytes,
2266 .ret_type = RET_INTEGER,
2267 .arg1_type = ARG_PTR_TO_CTX,
2268 .arg2_type = ARG_ANYTHING,
2271 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg_buff *, msg, u32, bytes)
2273 msg->cork_bytes = bytes;
2277 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2278 .func = bpf_msg_cork_bytes,
2280 .ret_type = RET_INTEGER,
2281 .arg1_type = ARG_PTR_TO_CTX,
2282 .arg2_type = ARG_ANYTHING,
2285 BPF_CALL_4(bpf_msg_pull_data,
2286 struct sk_msg_buff *, msg, u32, start, u32, end, u64, flags)
2288 unsigned int len = 0, offset = 0, copy = 0;
2289 struct scatterlist *sg = msg->sg_data;
2290 int first_sg, last_sg, i, shift;
2291 unsigned char *p, *to, *from;
2292 int bytes = end - start;
2295 if (unlikely(flags || end <= start))
2298 /* First find the starting scatterlist element */
2303 if (start < offset + len)
2306 if (i == MAX_SKB_FRAGS)
2308 } while (i != msg->sg_end);
2310 if (unlikely(start >= offset + len))
2313 if (!msg->sg_copy[i] && bytes <= len)
2318 /* At this point we need to linearize multiple scatterlist
2319 * elements or a single shared page. Either way we need to
2320 * copy into a linear buffer exclusively owned by BPF. Then
2321 * place the buffer in the scatterlist and fixup the original
2322 * entries by removing the entries now in the linear buffer
2323 * and shifting the remaining entries. For now we do not try
2324 * to copy partial entries to avoid complexity of running out
2325 * of sg_entry slots. The downside is reading a single byte
2326 * will copy the entire sg entry.
2329 copy += sg[i].length;
2331 if (i == MAX_SKB_FRAGS)
2335 } while (i != msg->sg_end);
2338 if (unlikely(copy < end - start))
2341 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC, get_order(copy));
2342 if (unlikely(!page))
2344 p = page_address(page);
2349 from = sg_virt(&sg[i]);
2353 memcpy(to, from, len);
2356 put_page(sg_page(&sg[i]));
2359 if (i == MAX_SKB_FRAGS)
2361 } while (i != last_sg);
2363 sg[first_sg].length = copy;
2364 sg_set_page(&sg[first_sg], page, copy, 0);
2366 /* To repair sg ring we need to shift entries. If we only
2367 * had a single entry though we can just replace it and
2368 * be done. Otherwise walk the ring and shift the entries.
2370 shift = last_sg - first_sg - 1;
2378 if (i + shift >= MAX_SKB_FRAGS)
2379 move_from = i + shift - MAX_SKB_FRAGS;
2381 move_from = i + shift;
2383 if (move_from == msg->sg_end)
2386 sg[i] = sg[move_from];
2387 sg[move_from].length = 0;
2388 sg[move_from].page_link = 0;
2389 sg[move_from].offset = 0;
2392 if (i == MAX_SKB_FRAGS)
2395 msg->sg_end -= shift;
2396 if (msg->sg_end < 0)
2397 msg->sg_end += MAX_SKB_FRAGS;
2399 msg->data = sg_virt(&sg[i]) + start - offset;
2400 msg->data_end = msg->data + bytes;
2405 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2406 .func = bpf_msg_pull_data,
2408 .ret_type = RET_INTEGER,
2409 .arg1_type = ARG_PTR_TO_CTX,
2410 .arg2_type = ARG_ANYTHING,
2411 .arg3_type = ARG_ANYTHING,
2412 .arg4_type = ARG_ANYTHING,
2415 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
2417 return task_get_classid(skb);
2420 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
2421 .func = bpf_get_cgroup_classid,
2423 .ret_type = RET_INTEGER,
2424 .arg1_type = ARG_PTR_TO_CTX,
2427 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
2429 return dst_tclassid(skb);
2432 static const struct bpf_func_proto bpf_get_route_realm_proto = {
2433 .func = bpf_get_route_realm,
2435 .ret_type = RET_INTEGER,
2436 .arg1_type = ARG_PTR_TO_CTX,
2439 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
2441 /* If skb_clear_hash() was called due to mangling, we can
2442 * trigger SW recalculation here. Later access to hash
2443 * can then use the inline skb->hash via context directly
2444 * instead of calling this helper again.
2446 return skb_get_hash(skb);
2449 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
2450 .func = bpf_get_hash_recalc,
2452 .ret_type = RET_INTEGER,
2453 .arg1_type = ARG_PTR_TO_CTX,
2456 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
2458 /* After all direct packet write, this can be used once for
2459 * triggering a lazy recalc on next skb_get_hash() invocation.
2461 skb_clear_hash(skb);
2465 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
2466 .func = bpf_set_hash_invalid,
2468 .ret_type = RET_INTEGER,
2469 .arg1_type = ARG_PTR_TO_CTX,
2472 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
2474 /* Set user specified hash as L4(+), so that it gets returned
2475 * on skb_get_hash() call unless BPF prog later on triggers a
2478 __skb_set_sw_hash(skb, hash, true);
2482 static const struct bpf_func_proto bpf_set_hash_proto = {
2483 .func = bpf_set_hash,
2485 .ret_type = RET_INTEGER,
2486 .arg1_type = ARG_PTR_TO_CTX,
2487 .arg2_type = ARG_ANYTHING,
2490 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
2495 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
2496 vlan_proto != htons(ETH_P_8021AD)))
2497 vlan_proto = htons(ETH_P_8021Q);
2499 bpf_push_mac_rcsum(skb);
2500 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
2501 bpf_pull_mac_rcsum(skb);
2503 bpf_compute_data_pointers(skb);
2507 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
2508 .func = bpf_skb_vlan_push,
2510 .ret_type = RET_INTEGER,
2511 .arg1_type = ARG_PTR_TO_CTX,
2512 .arg2_type = ARG_ANYTHING,
2513 .arg3_type = ARG_ANYTHING,
2516 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
2520 bpf_push_mac_rcsum(skb);
2521 ret = skb_vlan_pop(skb);
2522 bpf_pull_mac_rcsum(skb);
2524 bpf_compute_data_pointers(skb);
2528 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
2529 .func = bpf_skb_vlan_pop,
2531 .ret_type = RET_INTEGER,
2532 .arg1_type = ARG_PTR_TO_CTX,
2535 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
2537 /* Caller already did skb_cow() with len as headroom,
2538 * so no need to do it here.
2541 memmove(skb->data, skb->data + len, off);
2542 memset(skb->data + off, 0, len);
2544 /* No skb_postpush_rcsum(skb, skb->data + off, len)
2545 * needed here as it does not change the skb->csum
2546 * result for checksum complete when summing over
2552 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
2554 /* skb_ensure_writable() is not needed here, as we're
2555 * already working on an uncloned skb.
2557 if (unlikely(!pskb_may_pull(skb, off + len)))
2560 skb_postpull_rcsum(skb, skb->data + off, len);
2561 memmove(skb->data + len, skb->data, off);
2562 __skb_pull(skb, len);
2567 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
2569 bool trans_same = skb->transport_header == skb->network_header;
2572 /* There's no need for __skb_push()/__skb_pull() pair to
2573 * get to the start of the mac header as we're guaranteed
2574 * to always start from here under eBPF.
2576 ret = bpf_skb_generic_push(skb, off, len);
2578 skb->mac_header -= len;
2579 skb->network_header -= len;
2581 skb->transport_header = skb->network_header;
2587 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
2589 bool trans_same = skb->transport_header == skb->network_header;
2592 /* Same here, __skb_push()/__skb_pull() pair not needed. */
2593 ret = bpf_skb_generic_pop(skb, off, len);
2595 skb->mac_header += len;
2596 skb->network_header += len;
2598 skb->transport_header = skb->network_header;
2604 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
2606 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2607 u32 off = skb_mac_header_len(skb);
2610 /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2611 if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2614 ret = skb_cow(skb, len_diff);
2615 if (unlikely(ret < 0))
2618 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2619 if (unlikely(ret < 0))
2622 if (skb_is_gso(skb)) {
2623 struct skb_shared_info *shinfo = skb_shinfo(skb);
2625 /* SKB_GSO_TCPV4 needs to be changed into
2628 if (shinfo->gso_type & SKB_GSO_TCPV4) {
2629 shinfo->gso_type &= ~SKB_GSO_TCPV4;
2630 shinfo->gso_type |= SKB_GSO_TCPV6;
2633 /* Due to IPv6 header, MSS needs to be downgraded. */
2634 skb_decrease_gso_size(shinfo, len_diff);
2635 /* Header must be checked, and gso_segs recomputed. */
2636 shinfo->gso_type |= SKB_GSO_DODGY;
2637 shinfo->gso_segs = 0;
2640 skb->protocol = htons(ETH_P_IPV6);
2641 skb_clear_hash(skb);
2646 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
2648 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2649 u32 off = skb_mac_header_len(skb);
2652 /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2653 if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2656 ret = skb_unclone(skb, GFP_ATOMIC);
2657 if (unlikely(ret < 0))
2660 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2661 if (unlikely(ret < 0))
2664 if (skb_is_gso(skb)) {
2665 struct skb_shared_info *shinfo = skb_shinfo(skb);
2667 /* SKB_GSO_TCPV6 needs to be changed into
2670 if (shinfo->gso_type & SKB_GSO_TCPV6) {
2671 shinfo->gso_type &= ~SKB_GSO_TCPV6;
2672 shinfo->gso_type |= SKB_GSO_TCPV4;
2675 /* Due to IPv4 header, MSS can be upgraded. */
2676 skb_increase_gso_size(shinfo, len_diff);
2677 /* Header must be checked, and gso_segs recomputed. */
2678 shinfo->gso_type |= SKB_GSO_DODGY;
2679 shinfo->gso_segs = 0;
2682 skb->protocol = htons(ETH_P_IP);
2683 skb_clear_hash(skb);
2688 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
2690 __be16 from_proto = skb->protocol;
2692 if (from_proto == htons(ETH_P_IP) &&
2693 to_proto == htons(ETH_P_IPV6))
2694 return bpf_skb_proto_4_to_6(skb);
2696 if (from_proto == htons(ETH_P_IPV6) &&
2697 to_proto == htons(ETH_P_IP))
2698 return bpf_skb_proto_6_to_4(skb);
2703 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
2708 if (unlikely(flags))
2711 /* General idea is that this helper does the basic groundwork
2712 * needed for changing the protocol, and eBPF program fills the
2713 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
2714 * and other helpers, rather than passing a raw buffer here.
2716 * The rationale is to keep this minimal and without a need to
2717 * deal with raw packet data. F.e. even if we would pass buffers
2718 * here, the program still needs to call the bpf_lX_csum_replace()
2719 * helpers anyway. Plus, this way we keep also separation of
2720 * concerns, since f.e. bpf_skb_store_bytes() should only take
2723 * Currently, additional options and extension header space are
2724 * not supported, but flags register is reserved so we can adapt
2725 * that. For offloads, we mark packet as dodgy, so that headers
2726 * need to be verified first.
2728 ret = bpf_skb_proto_xlat(skb, proto);
2729 bpf_compute_data_pointers(skb);
2733 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
2734 .func = bpf_skb_change_proto,
2736 .ret_type = RET_INTEGER,
2737 .arg1_type = ARG_PTR_TO_CTX,
2738 .arg2_type = ARG_ANYTHING,
2739 .arg3_type = ARG_ANYTHING,
2742 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
2744 /* We only allow a restricted subset to be changed for now. */
2745 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
2746 !skb_pkt_type_ok(pkt_type)))
2749 skb->pkt_type = pkt_type;
2753 static const struct bpf_func_proto bpf_skb_change_type_proto = {
2754 .func = bpf_skb_change_type,
2756 .ret_type = RET_INTEGER,
2757 .arg1_type = ARG_PTR_TO_CTX,
2758 .arg2_type = ARG_ANYTHING,
2761 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
2763 switch (skb->protocol) {
2764 case htons(ETH_P_IP):
2765 return sizeof(struct iphdr);
2766 case htons(ETH_P_IPV6):
2767 return sizeof(struct ipv6hdr);
2773 static int bpf_skb_net_grow(struct sk_buff *skb, u32 len_diff)
2775 u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
2778 /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2779 if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2782 ret = skb_cow(skb, len_diff);
2783 if (unlikely(ret < 0))
2786 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2787 if (unlikely(ret < 0))
2790 if (skb_is_gso(skb)) {
2791 struct skb_shared_info *shinfo = skb_shinfo(skb);
2793 /* Due to header grow, MSS needs to be downgraded. */
2794 skb_decrease_gso_size(shinfo, len_diff);
2795 /* Header must be checked, and gso_segs recomputed. */
2796 shinfo->gso_type |= SKB_GSO_DODGY;
2797 shinfo->gso_segs = 0;
2803 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 len_diff)
2805 u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
2808 /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2809 if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2812 ret = skb_unclone(skb, GFP_ATOMIC);
2813 if (unlikely(ret < 0))
2816 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2817 if (unlikely(ret < 0))
2820 if (skb_is_gso(skb)) {
2821 struct skb_shared_info *shinfo = skb_shinfo(skb);
2823 /* Due to header shrink, MSS can be upgraded. */
2824 skb_increase_gso_size(shinfo, len_diff);
2825 /* Header must be checked, and gso_segs recomputed. */
2826 shinfo->gso_type |= SKB_GSO_DODGY;
2827 shinfo->gso_segs = 0;
2833 static u32 __bpf_skb_max_len(const struct sk_buff *skb)
2835 return skb->dev ? skb->dev->mtu + skb->dev->hard_header_len :
2839 static int bpf_skb_adjust_net(struct sk_buff *skb, s32 len_diff)
2841 bool trans_same = skb->transport_header == skb->network_header;
2842 u32 len_cur, len_diff_abs = abs(len_diff);
2843 u32 len_min = bpf_skb_net_base_len(skb);
2844 u32 len_max = __bpf_skb_max_len(skb);
2845 __be16 proto = skb->protocol;
2846 bool shrink = len_diff < 0;
2849 if (unlikely(len_diff_abs > 0xfffU))
2851 if (unlikely(proto != htons(ETH_P_IP) &&
2852 proto != htons(ETH_P_IPV6)))
2855 len_cur = skb->len - skb_network_offset(skb);
2856 if (skb_transport_header_was_set(skb) && !trans_same)
2857 len_cur = skb_network_header_len(skb);
2858 if ((shrink && (len_diff_abs >= len_cur ||
2859 len_cur - len_diff_abs < len_min)) ||
2860 (!shrink && (skb->len + len_diff_abs > len_max &&
2864 ret = shrink ? bpf_skb_net_shrink(skb, len_diff_abs) :
2865 bpf_skb_net_grow(skb, len_diff_abs);
2867 bpf_compute_data_pointers(skb);
2871 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
2872 u32, mode, u64, flags)
2874 if (unlikely(flags))
2876 if (likely(mode == BPF_ADJ_ROOM_NET))
2877 return bpf_skb_adjust_net(skb, len_diff);
2882 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
2883 .func = bpf_skb_adjust_room,
2885 .ret_type = RET_INTEGER,
2886 .arg1_type = ARG_PTR_TO_CTX,
2887 .arg2_type = ARG_ANYTHING,
2888 .arg3_type = ARG_ANYTHING,
2889 .arg4_type = ARG_ANYTHING,
2892 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
2894 u32 min_len = skb_network_offset(skb);
2896 if (skb_transport_header_was_set(skb))
2897 min_len = skb_transport_offset(skb);
2898 if (skb->ip_summed == CHECKSUM_PARTIAL)
2899 min_len = skb_checksum_start_offset(skb) +
2900 skb->csum_offset + sizeof(__sum16);
2904 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
2906 unsigned int old_len = skb->len;
2909 ret = __skb_grow_rcsum(skb, new_len);
2911 memset(skb->data + old_len, 0, new_len - old_len);
2915 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
2917 return __skb_trim_rcsum(skb, new_len);
2920 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
2923 u32 max_len = __bpf_skb_max_len(skb);
2924 u32 min_len = __bpf_skb_min_len(skb);
2927 if (unlikely(flags || new_len > max_len || new_len < min_len))
2929 if (skb->encapsulation)
2932 /* The basic idea of this helper is that it's performing the
2933 * needed work to either grow or trim an skb, and eBPF program
2934 * rewrites the rest via helpers like bpf_skb_store_bytes(),
2935 * bpf_lX_csum_replace() and others rather than passing a raw
2936 * buffer here. This one is a slow path helper and intended
2937 * for replies with control messages.
2939 * Like in bpf_skb_change_proto(), we want to keep this rather
2940 * minimal and without protocol specifics so that we are able
2941 * to separate concerns as in bpf_skb_store_bytes() should only
2942 * be the one responsible for writing buffers.
2944 * It's really expected to be a slow path operation here for
2945 * control message replies, so we're implicitly linearizing,
2946 * uncloning and drop offloads from the skb by this.
2948 ret = __bpf_try_make_writable(skb, skb->len);
2950 if (new_len > skb->len)
2951 ret = bpf_skb_grow_rcsum(skb, new_len);
2952 else if (new_len < skb->len)
2953 ret = bpf_skb_trim_rcsum(skb, new_len);
2954 if (!ret && skb_is_gso(skb))
2960 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
2963 int ret = __bpf_skb_change_tail(skb, new_len, flags);
2965 bpf_compute_data_pointers(skb);
2969 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
2970 .func = bpf_skb_change_tail,
2972 .ret_type = RET_INTEGER,
2973 .arg1_type = ARG_PTR_TO_CTX,
2974 .arg2_type = ARG_ANYTHING,
2975 .arg3_type = ARG_ANYTHING,
2978 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
2981 int ret = __bpf_skb_change_tail(skb, new_len, flags);
2983 bpf_compute_data_end_sk_skb(skb);
2987 static const struct bpf_func_proto sk_skb_change_tail_proto = {
2988 .func = sk_skb_change_tail,
2990 .ret_type = RET_INTEGER,
2991 .arg1_type = ARG_PTR_TO_CTX,
2992 .arg2_type = ARG_ANYTHING,
2993 .arg3_type = ARG_ANYTHING,
2996 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
2999 u32 max_len = __bpf_skb_max_len(skb);
3000 u32 new_len = skb->len + head_room;
3003 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3004 new_len < skb->len))
3007 ret = skb_cow(skb, head_room);
3009 /* Idea for this helper is that we currently only
3010 * allow to expand on mac header. This means that
3011 * skb->protocol network header, etc, stay as is.
3012 * Compared to bpf_skb_change_tail(), we're more
3013 * flexible due to not needing to linearize or
3014 * reset GSO. Intention for this helper is to be
3015 * used by an L3 skb that needs to push mac header
3016 * for redirection into L2 device.
3018 __skb_push(skb, head_room);
3019 memset(skb->data, 0, head_room);
3020 skb_reset_mac_header(skb);
3026 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3029 int ret = __bpf_skb_change_head(skb, head_room, flags);
3031 bpf_compute_data_pointers(skb);
3035 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3036 .func = bpf_skb_change_head,
3038 .ret_type = RET_INTEGER,
3039 .arg1_type = ARG_PTR_TO_CTX,
3040 .arg2_type = ARG_ANYTHING,
3041 .arg3_type = ARG_ANYTHING,
3044 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3047 int ret = __bpf_skb_change_head(skb, head_room, flags);
3049 bpf_compute_data_end_sk_skb(skb);
3053 static const struct bpf_func_proto sk_skb_change_head_proto = {
3054 .func = sk_skb_change_head,
3056 .ret_type = RET_INTEGER,
3057 .arg1_type = ARG_PTR_TO_CTX,
3058 .arg2_type = ARG_ANYTHING,
3059 .arg3_type = ARG_ANYTHING,
3061 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3063 return xdp_data_meta_unsupported(xdp) ? 0 :
3064 xdp->data - xdp->data_meta;
3067 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3069 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3070 unsigned long metalen = xdp_get_metalen(xdp);
3071 void *data_start = xdp_frame_end + metalen;
3072 void *data = xdp->data + offset;
3074 if (unlikely(data < data_start ||
3075 data > xdp->data_end - ETH_HLEN))
3079 memmove(xdp->data_meta + offset,
3080 xdp->data_meta, metalen);
3081 xdp->data_meta += offset;
3087 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3088 .func = bpf_xdp_adjust_head,
3090 .ret_type = RET_INTEGER,
3091 .arg1_type = ARG_PTR_TO_CTX,
3092 .arg2_type = ARG_ANYTHING,
3095 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3097 void *data_end = xdp->data_end + offset;
3099 /* only shrinking is allowed for now. */
3100 if (unlikely(offset >= 0))
3103 if (unlikely(data_end < xdp->data + ETH_HLEN))
3106 xdp->data_end = data_end;
3111 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3112 .func = bpf_xdp_adjust_tail,
3114 .ret_type = RET_INTEGER,
3115 .arg1_type = ARG_PTR_TO_CTX,
3116 .arg2_type = ARG_ANYTHING,
3119 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3121 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3122 void *meta = xdp->data_meta + offset;
3123 unsigned long metalen = xdp->data - meta;
3125 if (xdp_data_meta_unsupported(xdp))
3127 if (unlikely(meta < xdp_frame_end ||
3130 if (unlikely((metalen & (sizeof(__u32) - 1)) ||
3134 xdp->data_meta = meta;
3139 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3140 .func = bpf_xdp_adjust_meta,
3142 .ret_type = RET_INTEGER,
3143 .arg1_type = ARG_PTR_TO_CTX,
3144 .arg2_type = ARG_ANYTHING,
3147 static int __bpf_tx_xdp(struct net_device *dev,
3148 struct bpf_map *map,
3149 struct xdp_buff *xdp,
3152 struct xdp_frame *xdpf;
3155 if (!dev->netdev_ops->ndo_xdp_xmit) {
3159 err = xdp_ok_fwd_dev(dev, xdp->data_end - xdp->data);
3163 xdpf = convert_to_xdp_frame(xdp);
3164 if (unlikely(!xdpf))
3167 sent = dev->netdev_ops->ndo_xdp_xmit(dev, 1, &xdpf, XDP_XMIT_FLUSH);
3173 static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd,
3174 struct bpf_map *map,
3175 struct xdp_buff *xdp,
3180 switch (map->map_type) {
3181 case BPF_MAP_TYPE_DEVMAP: {
3182 struct bpf_dtab_netdev *dst = fwd;
3184 err = dev_map_enqueue(dst, xdp, dev_rx);
3187 __dev_map_insert_ctx(map, index);
3190 case BPF_MAP_TYPE_CPUMAP: {
3191 struct bpf_cpu_map_entry *rcpu = fwd;
3193 err = cpu_map_enqueue(rcpu, xdp, dev_rx);
3196 __cpu_map_insert_ctx(map, index);
3199 case BPF_MAP_TYPE_XSKMAP: {
3200 struct xdp_sock *xs = fwd;
3202 err = __xsk_map_redirect(map, xdp, xs);
3211 void xdp_do_flush_map(void)
3213 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3214 struct bpf_map *map = ri->map_to_flush;
3216 ri->map_to_flush = NULL;
3218 switch (map->map_type) {
3219 case BPF_MAP_TYPE_DEVMAP:
3220 __dev_map_flush(map);
3222 case BPF_MAP_TYPE_CPUMAP:
3223 __cpu_map_flush(map);
3225 case BPF_MAP_TYPE_XSKMAP:
3226 __xsk_map_flush(map);
3233 EXPORT_SYMBOL_GPL(xdp_do_flush_map);
3235 static void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index)
3237 switch (map->map_type) {
3238 case BPF_MAP_TYPE_DEVMAP:
3239 return __dev_map_lookup_elem(map, index);
3240 case BPF_MAP_TYPE_CPUMAP:
3241 return __cpu_map_lookup_elem(map, index);
3242 case BPF_MAP_TYPE_XSKMAP:
3243 return __xsk_map_lookup_elem(map, index);
3249 void bpf_clear_redirect_map(struct bpf_map *map)
3251 struct bpf_redirect_info *ri;
3254 for_each_possible_cpu(cpu) {
3255 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3256 /* Avoid polluting remote cacheline due to writes if
3257 * not needed. Once we pass this test, we need the
3258 * cmpxchg() to make sure it hasn't been changed in
3259 * the meantime by remote CPU.
3261 if (unlikely(READ_ONCE(ri->map) == map))
3262 cmpxchg(&ri->map, map, NULL);
3266 static int xdp_do_redirect_map(struct net_device *dev, struct xdp_buff *xdp,
3267 struct bpf_prog *xdp_prog, struct bpf_map *map)
3269 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3270 u32 index = ri->ifindex;
3275 WRITE_ONCE(ri->map, NULL);
3277 fwd = __xdp_map_lookup_elem(map, index);
3282 if (ri->map_to_flush && ri->map_to_flush != map)
3285 err = __bpf_tx_xdp_map(dev, fwd, map, xdp, index);
3289 ri->map_to_flush = map;
3290 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3293 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3297 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3298 struct bpf_prog *xdp_prog)
3300 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3301 struct bpf_map *map = READ_ONCE(ri->map);
3302 struct net_device *fwd;
3303 u32 index = ri->ifindex;
3307 return xdp_do_redirect_map(dev, xdp, xdp_prog, map);
3309 fwd = dev_get_by_index_rcu(dev_net(dev), index);
3311 if (unlikely(!fwd)) {
3316 err = __bpf_tx_xdp(fwd, NULL, xdp, 0);
3320 _trace_xdp_redirect(dev, xdp_prog, index);
3323 _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3326 EXPORT_SYMBOL_GPL(xdp_do_redirect);
3328 static int xdp_do_generic_redirect_map(struct net_device *dev,
3329 struct sk_buff *skb,
3330 struct xdp_buff *xdp,
3331 struct bpf_prog *xdp_prog,
3332 struct bpf_map *map)
3334 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3335 u32 index = ri->ifindex;
3340 WRITE_ONCE(ri->map, NULL);
3342 fwd = __xdp_map_lookup_elem(map, index);
3343 if (unlikely(!fwd)) {
3348 if (map->map_type == BPF_MAP_TYPE_DEVMAP) {
3349 struct bpf_dtab_netdev *dst = fwd;
3351 err = dev_map_generic_redirect(dst, skb, xdp_prog);
3354 } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) {
3355 struct xdp_sock *xs = fwd;
3357 err = xsk_generic_rcv(xs, xdp);
3362 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
3367 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3370 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3374 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
3375 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
3377 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3378 struct bpf_map *map = READ_ONCE(ri->map);
3379 u32 index = ri->ifindex;
3380 struct net_device *fwd;
3384 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog,
3387 fwd = dev_get_by_index_rcu(dev_net(dev), index);
3388 if (unlikely(!fwd)) {
3393 err = xdp_ok_fwd_dev(fwd, skb->len);
3398 _trace_xdp_redirect(dev, xdp_prog, index);
3399 generic_xdp_tx(skb, xdp_prog);
3402 _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3405 EXPORT_SYMBOL_GPL(xdp_do_generic_redirect);
3407 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
3409 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3411 if (unlikely(flags))
3414 ri->ifindex = ifindex;
3416 WRITE_ONCE(ri->map, NULL);
3418 return XDP_REDIRECT;
3421 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
3422 .func = bpf_xdp_redirect,
3424 .ret_type = RET_INTEGER,
3425 .arg1_type = ARG_ANYTHING,
3426 .arg2_type = ARG_ANYTHING,
3429 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
3432 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3434 if (unlikely(flags))
3437 ri->ifindex = ifindex;
3439 WRITE_ONCE(ri->map, map);
3441 return XDP_REDIRECT;
3444 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
3445 .func = bpf_xdp_redirect_map,
3447 .ret_type = RET_INTEGER,
3448 .arg1_type = ARG_CONST_MAP_PTR,
3449 .arg2_type = ARG_ANYTHING,
3450 .arg3_type = ARG_ANYTHING,
3453 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
3454 unsigned long off, unsigned long len)
3456 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
3460 if (ptr != dst_buff)
3461 memcpy(dst_buff, ptr, len);
3466 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
3467 u64, flags, void *, meta, u64, meta_size)
3469 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
3471 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
3473 if (unlikely(skb_size > skb->len))
3476 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
3480 static const struct bpf_func_proto bpf_skb_event_output_proto = {
3481 .func = bpf_skb_event_output,
3483 .ret_type = RET_INTEGER,
3484 .arg1_type = ARG_PTR_TO_CTX,
3485 .arg2_type = ARG_CONST_MAP_PTR,
3486 .arg3_type = ARG_ANYTHING,
3487 .arg4_type = ARG_PTR_TO_MEM,
3488 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
3491 static unsigned short bpf_tunnel_key_af(u64 flags)
3493 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
3496 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
3497 u32, size, u64, flags)
3499 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3500 u8 compat[sizeof(struct bpf_tunnel_key)];
3504 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
3508 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
3512 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3515 case offsetof(struct bpf_tunnel_key, tunnel_label):
3516 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3518 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3519 /* Fixup deprecated structure layouts here, so we have
3520 * a common path later on.
3522 if (ip_tunnel_info_af(info) != AF_INET)
3525 to = (struct bpf_tunnel_key *)compat;
3532 to->tunnel_id = be64_to_cpu(info->key.tun_id);
3533 to->tunnel_tos = info->key.tos;
3534 to->tunnel_ttl = info->key.ttl;
3537 if (flags & BPF_F_TUNINFO_IPV6) {
3538 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
3539 sizeof(to->remote_ipv6));
3540 to->tunnel_label = be32_to_cpu(info->key.label);
3542 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
3543 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
3544 to->tunnel_label = 0;
3547 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
3548 memcpy(to_orig, to, size);
3552 memset(to_orig, 0, size);
3556 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
3557 .func = bpf_skb_get_tunnel_key,
3559 .ret_type = RET_INTEGER,
3560 .arg1_type = ARG_PTR_TO_CTX,
3561 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
3562 .arg3_type = ARG_CONST_SIZE,
3563 .arg4_type = ARG_ANYTHING,
3566 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
3568 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3571 if (unlikely(!info ||
3572 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
3576 if (unlikely(size < info->options_len)) {
3581 ip_tunnel_info_opts_get(to, info);
3582 if (size > info->options_len)
3583 memset(to + info->options_len, 0, size - info->options_len);
3585 return info->options_len;
3587 memset(to, 0, size);
3591 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
3592 .func = bpf_skb_get_tunnel_opt,
3594 .ret_type = RET_INTEGER,
3595 .arg1_type = ARG_PTR_TO_CTX,
3596 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
3597 .arg3_type = ARG_CONST_SIZE,
3600 static struct metadata_dst __percpu *md_dst;
3602 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
3603 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
3605 struct metadata_dst *md = this_cpu_ptr(md_dst);
3606 u8 compat[sizeof(struct bpf_tunnel_key)];
3607 struct ip_tunnel_info *info;
3609 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
3610 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
3612 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3614 case offsetof(struct bpf_tunnel_key, tunnel_label):
3615 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3616 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3617 /* Fixup deprecated structure layouts here, so we have
3618 * a common path later on.
3620 memcpy(compat, from, size);
3621 memset(compat + size, 0, sizeof(compat) - size);
3622 from = (const struct bpf_tunnel_key *) compat;
3628 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
3633 dst_hold((struct dst_entry *) md);
3634 skb_dst_set(skb, (struct dst_entry *) md);
3636 info = &md->u.tun_info;
3637 memset(info, 0, sizeof(*info));
3638 info->mode = IP_TUNNEL_INFO_TX;
3640 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
3641 if (flags & BPF_F_DONT_FRAGMENT)
3642 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
3643 if (flags & BPF_F_ZERO_CSUM_TX)
3644 info->key.tun_flags &= ~TUNNEL_CSUM;
3645 if (flags & BPF_F_SEQ_NUMBER)
3646 info->key.tun_flags |= TUNNEL_SEQ;
3648 info->key.tun_id = cpu_to_be64(from->tunnel_id);
3649 info->key.tos = from->tunnel_tos;
3650 info->key.ttl = from->tunnel_ttl;
3652 if (flags & BPF_F_TUNINFO_IPV6) {
3653 info->mode |= IP_TUNNEL_INFO_IPV6;
3654 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
3655 sizeof(from->remote_ipv6));
3656 info->key.label = cpu_to_be32(from->tunnel_label) &
3657 IPV6_FLOWLABEL_MASK;
3659 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
3665 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
3666 .func = bpf_skb_set_tunnel_key,
3668 .ret_type = RET_INTEGER,
3669 .arg1_type = ARG_PTR_TO_CTX,
3670 .arg2_type = ARG_PTR_TO_MEM,
3671 .arg3_type = ARG_CONST_SIZE,
3672 .arg4_type = ARG_ANYTHING,
3675 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
3676 const u8 *, from, u32, size)
3678 struct ip_tunnel_info *info = skb_tunnel_info(skb);
3679 const struct metadata_dst *md = this_cpu_ptr(md_dst);
3681 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
3683 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
3686 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
3691 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
3692 .func = bpf_skb_set_tunnel_opt,
3694 .ret_type = RET_INTEGER,
3695 .arg1_type = ARG_PTR_TO_CTX,
3696 .arg2_type = ARG_PTR_TO_MEM,
3697 .arg3_type = ARG_CONST_SIZE,
3700 static const struct bpf_func_proto *
3701 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
3704 struct metadata_dst __percpu *tmp;
3706 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
3711 if (cmpxchg(&md_dst, NULL, tmp))
3712 metadata_dst_free_percpu(tmp);
3716 case BPF_FUNC_skb_set_tunnel_key:
3717 return &bpf_skb_set_tunnel_key_proto;
3718 case BPF_FUNC_skb_set_tunnel_opt:
3719 return &bpf_skb_set_tunnel_opt_proto;
3725 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
3728 struct bpf_array *array = container_of(map, struct bpf_array, map);
3729 struct cgroup *cgrp;
3732 sk = skb_to_full_sk(skb);
3733 if (!sk || !sk_fullsock(sk))
3735 if (unlikely(idx >= array->map.max_entries))
3738 cgrp = READ_ONCE(array->ptrs[idx]);
3739 if (unlikely(!cgrp))
3742 return sk_under_cgroup_hierarchy(sk, cgrp);
3745 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
3746 .func = bpf_skb_under_cgroup,
3748 .ret_type = RET_INTEGER,
3749 .arg1_type = ARG_PTR_TO_CTX,
3750 .arg2_type = ARG_CONST_MAP_PTR,
3751 .arg3_type = ARG_ANYTHING,
3754 #ifdef CONFIG_SOCK_CGROUP_DATA
3755 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
3757 struct sock *sk = skb_to_full_sk(skb);
3758 struct cgroup *cgrp;
3760 if (!sk || !sk_fullsock(sk))
3763 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
3764 return cgrp->kn->id.id;
3767 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
3768 .func = bpf_skb_cgroup_id,
3770 .ret_type = RET_INTEGER,
3771 .arg1_type = ARG_PTR_TO_CTX,
3774 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
3777 struct sock *sk = skb_to_full_sk(skb);
3778 struct cgroup *ancestor;
3779 struct cgroup *cgrp;
3781 if (!sk || !sk_fullsock(sk))
3784 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
3785 ancestor = cgroup_ancestor(cgrp, ancestor_level);
3789 return ancestor->kn->id.id;
3792 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
3793 .func = bpf_skb_ancestor_cgroup_id,
3795 .ret_type = RET_INTEGER,
3796 .arg1_type = ARG_PTR_TO_CTX,
3797 .arg2_type = ARG_ANYTHING,
3801 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
3802 unsigned long off, unsigned long len)
3804 memcpy(dst_buff, src_buff + off, len);
3808 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
3809 u64, flags, void *, meta, u64, meta_size)
3811 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
3813 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
3815 if (unlikely(xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
3818 return bpf_event_output(map, flags, meta, meta_size, xdp->data,
3819 xdp_size, bpf_xdp_copy);
3822 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
3823 .func = bpf_xdp_event_output,
3825 .ret_type = RET_INTEGER,
3826 .arg1_type = ARG_PTR_TO_CTX,
3827 .arg2_type = ARG_CONST_MAP_PTR,
3828 .arg3_type = ARG_ANYTHING,
3829 .arg4_type = ARG_PTR_TO_MEM,
3830 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
3833 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
3835 return skb->sk ? sock_gen_cookie(skb->sk) : 0;
3838 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
3839 .func = bpf_get_socket_cookie,
3841 .ret_type = RET_INTEGER,
3842 .arg1_type = ARG_PTR_TO_CTX,
3845 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
3847 return sock_gen_cookie(ctx->sk);
3850 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
3851 .func = bpf_get_socket_cookie_sock_addr,
3853 .ret_type = RET_INTEGER,
3854 .arg1_type = ARG_PTR_TO_CTX,
3857 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
3859 return sock_gen_cookie(ctx->sk);
3862 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
3863 .func = bpf_get_socket_cookie_sock_ops,
3865 .ret_type = RET_INTEGER,
3866 .arg1_type = ARG_PTR_TO_CTX,
3869 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
3871 struct sock *sk = sk_to_full_sk(skb->sk);
3874 if (!sk || !sk_fullsock(sk))
3876 kuid = sock_net_uid(sock_net(sk), sk);
3877 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
3880 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
3881 .func = bpf_get_socket_uid,
3883 .ret_type = RET_INTEGER,
3884 .arg1_type = ARG_PTR_TO_CTX,
3887 BPF_CALL_5(bpf_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
3888 int, level, int, optname, char *, optval, int, optlen)
3890 struct sock *sk = bpf_sock->sk;
3894 if (!sk_fullsock(sk))
3897 if (level == SOL_SOCKET) {
3898 if (optlen != sizeof(int))
3900 val = *((int *)optval);
3902 /* Only some socketops are supported */
3905 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
3906 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
3909 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
3910 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
3912 case SO_MAX_PACING_RATE:
3913 sk->sk_max_pacing_rate = val;
3914 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
3915 sk->sk_max_pacing_rate);
3918 sk->sk_priority = val;
3923 sk->sk_rcvlowat = val ? : 1;
3932 } else if (level == SOL_IP) {
3933 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
3936 val = *((int *)optval);
3937 /* Only some options are supported */
3940 if (val < -1 || val > 0xff) {
3943 struct inet_sock *inet = inet_sk(sk);
3953 #if IS_ENABLED(CONFIG_IPV6)
3954 } else if (level == SOL_IPV6) {
3955 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
3958 val = *((int *)optval);
3959 /* Only some options are supported */
3962 if (val < -1 || val > 0xff) {
3965 struct ipv6_pinfo *np = inet6_sk(sk);
3976 } else if (level == SOL_TCP &&
3977 sk->sk_prot->setsockopt == tcp_setsockopt) {
3978 if (optname == TCP_CONGESTION) {
3979 char name[TCP_CA_NAME_MAX];
3980 bool reinit = bpf_sock->op > BPF_SOCK_OPS_NEEDS_ECN;
3982 strncpy(name, optval, min_t(long, optlen,
3983 TCP_CA_NAME_MAX-1));
3984 name[TCP_CA_NAME_MAX-1] = 0;
3985 ret = tcp_set_congestion_control(sk, name, false,
3988 struct tcp_sock *tp = tcp_sk(sk);
3990 if (optlen != sizeof(int))
3993 val = *((int *)optval);
3994 /* Only some options are supported */
3997 if (val <= 0 || tp->data_segs_out > 0)
4002 case TCP_BPF_SNDCWND_CLAMP:
4006 tp->snd_cwnd_clamp = val;
4007 tp->snd_ssthresh = val;
4021 static const struct bpf_func_proto bpf_setsockopt_proto = {
4022 .func = bpf_setsockopt,
4024 .ret_type = RET_INTEGER,
4025 .arg1_type = ARG_PTR_TO_CTX,
4026 .arg2_type = ARG_ANYTHING,
4027 .arg3_type = ARG_ANYTHING,
4028 .arg4_type = ARG_PTR_TO_MEM,
4029 .arg5_type = ARG_CONST_SIZE,
4032 BPF_CALL_5(bpf_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
4033 int, level, int, optname, char *, optval, int, optlen)
4035 struct sock *sk = bpf_sock->sk;
4037 if (!sk_fullsock(sk))
4041 if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4042 if (optname == TCP_CONGESTION) {
4043 struct inet_connection_sock *icsk = inet_csk(sk);
4045 if (!icsk->icsk_ca_ops || optlen <= 1)
4047 strncpy(optval, icsk->icsk_ca_ops->name, optlen);
4048 optval[optlen - 1] = 0;
4052 } else if (level == SOL_IP) {
4053 struct inet_sock *inet = inet_sk(sk);
4055 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4058 /* Only some options are supported */
4061 *((int *)optval) = (int)inet->tos;
4066 #if IS_ENABLED(CONFIG_IPV6)
4067 } else if (level == SOL_IPV6) {
4068 struct ipv6_pinfo *np = inet6_sk(sk);
4070 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4073 /* Only some options are supported */
4076 *((int *)optval) = (int)np->tclass;
4088 memset(optval, 0, optlen);
4092 static const struct bpf_func_proto bpf_getsockopt_proto = {
4093 .func = bpf_getsockopt,
4095 .ret_type = RET_INTEGER,
4096 .arg1_type = ARG_PTR_TO_CTX,
4097 .arg2_type = ARG_ANYTHING,
4098 .arg3_type = ARG_ANYTHING,
4099 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
4100 .arg5_type = ARG_CONST_SIZE,
4103 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
4106 struct sock *sk = bpf_sock->sk;
4107 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
4109 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
4113 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
4115 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
4118 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
4119 .func = bpf_sock_ops_cb_flags_set,
4121 .ret_type = RET_INTEGER,
4122 .arg1_type = ARG_PTR_TO_CTX,
4123 .arg2_type = ARG_ANYTHING,
4126 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
4127 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
4129 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
4133 struct sock *sk = ctx->sk;
4136 /* Binding to port can be expensive so it's prohibited in the helper.
4137 * Only binding to IP is supported.
4140 if (addr->sa_family == AF_INET) {
4141 if (addr_len < sizeof(struct sockaddr_in))
4143 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
4145 return __inet_bind(sk, addr, addr_len, true, false);
4146 #if IS_ENABLED(CONFIG_IPV6)
4147 } else if (addr->sa_family == AF_INET6) {
4148 if (addr_len < SIN6_LEN_RFC2133)
4150 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
4152 /* ipv6_bpf_stub cannot be NULL, since it's called from
4153 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
4155 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, true, false);
4156 #endif /* CONFIG_IPV6 */
4158 #endif /* CONFIG_INET */
4160 return -EAFNOSUPPORT;
4163 static const struct bpf_func_proto bpf_bind_proto = {
4166 .ret_type = RET_INTEGER,
4167 .arg1_type = ARG_PTR_TO_CTX,
4168 .arg2_type = ARG_PTR_TO_MEM,
4169 .arg3_type = ARG_CONST_SIZE,
4173 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
4174 struct bpf_xfrm_state *, to, u32, size, u64, flags)
4176 const struct sec_path *sp = skb_sec_path(skb);
4177 const struct xfrm_state *x;
4179 if (!sp || unlikely(index >= sp->len || flags))
4182 x = sp->xvec[index];
4184 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
4187 to->reqid = x->props.reqid;
4188 to->spi = x->id.spi;
4189 to->family = x->props.family;
4192 if (to->family == AF_INET6) {
4193 memcpy(to->remote_ipv6, x->props.saddr.a6,
4194 sizeof(to->remote_ipv6));
4196 to->remote_ipv4 = x->props.saddr.a4;
4197 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4202 memset(to, 0, size);
4206 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
4207 .func = bpf_skb_get_xfrm_state,
4209 .ret_type = RET_INTEGER,
4210 .arg1_type = ARG_PTR_TO_CTX,
4211 .arg2_type = ARG_ANYTHING,
4212 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
4213 .arg4_type = ARG_CONST_SIZE,
4214 .arg5_type = ARG_ANYTHING,
4218 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
4219 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
4220 const struct neighbour *neigh,
4221 const struct net_device *dev)
4223 memcpy(params->dmac, neigh->ha, ETH_ALEN);
4224 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
4225 params->h_vlan_TCI = 0;
4226 params->h_vlan_proto = 0;
4227 params->ifindex = dev->ifindex;
4233 #if IS_ENABLED(CONFIG_INET)
4234 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4235 u32 flags, bool check_mtu)
4237 struct in_device *in_dev;
4238 struct neighbour *neigh;
4239 struct net_device *dev;
4240 struct fib_result res;
4246 dev = dev_get_by_index_rcu(net, params->ifindex);
4250 /* verify forwarding is enabled on this interface */
4251 in_dev = __in_dev_get_rcu(dev);
4252 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
4253 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4255 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4257 fl4.flowi4_oif = params->ifindex;
4259 fl4.flowi4_iif = params->ifindex;
4262 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
4263 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
4264 fl4.flowi4_flags = 0;
4266 fl4.flowi4_proto = params->l4_protocol;
4267 fl4.daddr = params->ipv4_dst;
4268 fl4.saddr = params->ipv4_src;
4269 fl4.fl4_sport = params->sport;
4270 fl4.fl4_dport = params->dport;
4272 if (flags & BPF_FIB_LOOKUP_DIRECT) {
4273 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4274 struct fib_table *tb;
4276 tb = fib_get_table(net, tbid);
4278 return BPF_FIB_LKUP_RET_NOT_FWDED;
4280 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
4282 fl4.flowi4_mark = 0;
4283 fl4.flowi4_secid = 0;
4284 fl4.flowi4_tun_key.tun_id = 0;
4285 fl4.flowi4_uid = sock_net_uid(net, NULL);
4287 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
4291 /* map fib lookup errors to RTN_ type */
4293 return BPF_FIB_LKUP_RET_BLACKHOLE;
4294 if (err == -EHOSTUNREACH)
4295 return BPF_FIB_LKUP_RET_UNREACHABLE;
4297 return BPF_FIB_LKUP_RET_PROHIBIT;
4299 return BPF_FIB_LKUP_RET_NOT_FWDED;
4302 if (res.type != RTN_UNICAST)
4303 return BPF_FIB_LKUP_RET_NOT_FWDED;
4305 if (res.fi->fib_nhs > 1)
4306 fib_select_path(net, &res, &fl4, NULL);
4309 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
4310 if (params->tot_len > mtu)
4311 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4314 nh = &res.fi->fib_nh[res.nh_sel];
4316 /* do not handle lwt encaps right now */
4317 if (nh->nh_lwtstate)
4318 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4322 params->ipv4_dst = nh->nh_gw;
4324 params->rt_metric = res.fi->fib_priority;
4326 /* xdp and cls_bpf programs are run in RCU-bh so
4327 * rcu_read_lock_bh is not needed here
4329 neigh = __ipv4_neigh_lookup_noref(dev, (__force u32)params->ipv4_dst);
4331 return BPF_FIB_LKUP_RET_NO_NEIGH;
4333 return bpf_fib_set_fwd_params(params, neigh, dev);
4337 #if IS_ENABLED(CONFIG_IPV6)
4338 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4339 u32 flags, bool check_mtu)
4341 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
4342 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
4343 struct neighbour *neigh;
4344 struct net_device *dev;
4345 struct inet6_dev *idev;
4346 struct fib6_info *f6i;
4352 /* link local addresses are never forwarded */
4353 if (rt6_need_strict(dst) || rt6_need_strict(src))
4354 return BPF_FIB_LKUP_RET_NOT_FWDED;
4356 dev = dev_get_by_index_rcu(net, params->ifindex);
4360 idev = __in6_dev_get_safely(dev);
4361 if (unlikely(!idev || !net->ipv6.devconf_all->forwarding))
4362 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4364 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4366 oif = fl6.flowi6_oif = params->ifindex;
4368 oif = fl6.flowi6_iif = params->ifindex;
4370 strict = RT6_LOOKUP_F_HAS_SADDR;
4372 fl6.flowlabel = params->flowinfo;
4373 fl6.flowi6_scope = 0;
4374 fl6.flowi6_flags = 0;
4377 fl6.flowi6_proto = params->l4_protocol;
4380 fl6.fl6_sport = params->sport;
4381 fl6.fl6_dport = params->dport;
4383 if (flags & BPF_FIB_LOOKUP_DIRECT) {
4384 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4385 struct fib6_table *tb;
4387 tb = ipv6_stub->fib6_get_table(net, tbid);
4389 return BPF_FIB_LKUP_RET_NOT_FWDED;
4391 f6i = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, strict);
4393 fl6.flowi6_mark = 0;
4394 fl6.flowi6_secid = 0;
4395 fl6.flowi6_tun_key.tun_id = 0;
4396 fl6.flowi6_uid = sock_net_uid(net, NULL);
4398 f6i = ipv6_stub->fib6_lookup(net, oif, &fl6, strict);
4401 if (unlikely(IS_ERR_OR_NULL(f6i) || f6i == net->ipv6.fib6_null_entry))
4402 return BPF_FIB_LKUP_RET_NOT_FWDED;
4404 if (unlikely(f6i->fib6_flags & RTF_REJECT)) {
4405 switch (f6i->fib6_type) {
4407 return BPF_FIB_LKUP_RET_BLACKHOLE;
4408 case RTN_UNREACHABLE:
4409 return BPF_FIB_LKUP_RET_UNREACHABLE;
4411 return BPF_FIB_LKUP_RET_PROHIBIT;
4413 return BPF_FIB_LKUP_RET_NOT_FWDED;
4417 if (f6i->fib6_type != RTN_UNICAST)
4418 return BPF_FIB_LKUP_RET_NOT_FWDED;
4420 if (f6i->fib6_nsiblings && fl6.flowi6_oif == 0)
4421 f6i = ipv6_stub->fib6_multipath_select(net, f6i, &fl6,
4422 fl6.flowi6_oif, NULL,
4426 mtu = ipv6_stub->ip6_mtu_from_fib6(f6i, dst, src);
4427 if (params->tot_len > mtu)
4428 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4431 if (f6i->fib6_nh.nh_lwtstate)
4432 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4434 if (f6i->fib6_flags & RTF_GATEWAY)
4435 *dst = f6i->fib6_nh.nh_gw;
4437 dev = f6i->fib6_nh.nh_dev;
4438 params->rt_metric = f6i->fib6_metric;
4440 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
4441 * not needed here. Can not use __ipv6_neigh_lookup_noref here
4442 * because we need to get nd_tbl via the stub
4444 neigh = ___neigh_lookup_noref(ipv6_stub->nd_tbl, neigh_key_eq128,
4445 ndisc_hashfn, dst, dev);
4447 return BPF_FIB_LKUP_RET_NO_NEIGH;
4449 return bpf_fib_set_fwd_params(params, neigh, dev);
4453 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
4454 struct bpf_fib_lookup *, params, int, plen, u32, flags)
4456 if (plen < sizeof(*params))
4459 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4462 switch (params->family) {
4463 #if IS_ENABLED(CONFIG_INET)
4465 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
4468 #if IS_ENABLED(CONFIG_IPV6)
4470 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
4474 return -EAFNOSUPPORT;
4477 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
4478 .func = bpf_xdp_fib_lookup,
4480 .ret_type = RET_INTEGER,
4481 .arg1_type = ARG_PTR_TO_CTX,
4482 .arg2_type = ARG_PTR_TO_MEM,
4483 .arg3_type = ARG_CONST_SIZE,
4484 .arg4_type = ARG_ANYTHING,
4487 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
4488 struct bpf_fib_lookup *, params, int, plen, u32, flags)
4490 struct net *net = dev_net(skb->dev);
4491 int rc = -EAFNOSUPPORT;
4493 if (plen < sizeof(*params))
4496 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4499 switch (params->family) {
4500 #if IS_ENABLED(CONFIG_INET)
4502 rc = bpf_ipv4_fib_lookup(net, params, flags, false);
4505 #if IS_ENABLED(CONFIG_IPV6)
4507 rc = bpf_ipv6_fib_lookup(net, params, flags, false);
4513 struct net_device *dev;
4515 dev = dev_get_by_index_rcu(net, params->ifindex);
4516 if (!is_skb_forwardable(dev, skb))
4517 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
4523 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
4524 .func = bpf_skb_fib_lookup,
4526 .ret_type = RET_INTEGER,
4527 .arg1_type = ARG_PTR_TO_CTX,
4528 .arg2_type = ARG_PTR_TO_MEM,
4529 .arg3_type = ARG_CONST_SIZE,
4530 .arg4_type = ARG_ANYTHING,
4533 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4534 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
4537 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
4539 if (!seg6_validate_srh(srh, len))
4543 case BPF_LWT_ENCAP_SEG6_INLINE:
4544 if (skb->protocol != htons(ETH_P_IPV6))
4547 err = seg6_do_srh_inline(skb, srh);
4549 case BPF_LWT_ENCAP_SEG6:
4550 skb_reset_inner_headers(skb);
4551 skb->encapsulation = 1;
4552 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
4558 bpf_compute_data_pointers(skb);
4562 ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
4563 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
4565 return seg6_lookup_nexthop(skb, NULL, 0);
4567 #endif /* CONFIG_IPV6_SEG6_BPF */
4569 BPF_CALL_4(bpf_lwt_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
4573 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4574 case BPF_LWT_ENCAP_SEG6:
4575 case BPF_LWT_ENCAP_SEG6_INLINE:
4576 return bpf_push_seg6_encap(skb, type, hdr, len);
4583 static const struct bpf_func_proto bpf_lwt_push_encap_proto = {
4584 .func = bpf_lwt_push_encap,
4586 .ret_type = RET_INTEGER,
4587 .arg1_type = ARG_PTR_TO_CTX,
4588 .arg2_type = ARG_ANYTHING,
4589 .arg3_type = ARG_PTR_TO_MEM,
4590 .arg4_type = ARG_CONST_SIZE
4593 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4594 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
4595 const void *, from, u32, len)
4597 struct seg6_bpf_srh_state *srh_state =
4598 this_cpu_ptr(&seg6_bpf_srh_states);
4599 struct ipv6_sr_hdr *srh = srh_state->srh;
4600 void *srh_tlvs, *srh_end, *ptr;
4606 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
4607 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
4609 ptr = skb->data + offset;
4610 if (ptr >= srh_tlvs && ptr + len <= srh_end)
4611 srh_state->valid = false;
4612 else if (ptr < (void *)&srh->flags ||
4613 ptr + len > (void *)&srh->segments)
4616 if (unlikely(bpf_try_make_writable(skb, offset + len)))
4618 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
4620 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4622 memcpy(skb->data + offset, from, len);
4626 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
4627 .func = bpf_lwt_seg6_store_bytes,
4629 .ret_type = RET_INTEGER,
4630 .arg1_type = ARG_PTR_TO_CTX,
4631 .arg2_type = ARG_ANYTHING,
4632 .arg3_type = ARG_PTR_TO_MEM,
4633 .arg4_type = ARG_CONST_SIZE
4636 static void bpf_update_srh_state(struct sk_buff *skb)
4638 struct seg6_bpf_srh_state *srh_state =
4639 this_cpu_ptr(&seg6_bpf_srh_states);
4642 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
4643 srh_state->srh = NULL;
4645 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4646 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
4647 srh_state->valid = true;
4651 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
4652 u32, action, void *, param, u32, param_len)
4654 struct seg6_bpf_srh_state *srh_state =
4655 this_cpu_ptr(&seg6_bpf_srh_states);
4660 case SEG6_LOCAL_ACTION_END_X:
4661 if (!seg6_bpf_has_valid_srh(skb))
4663 if (param_len != sizeof(struct in6_addr))
4665 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
4666 case SEG6_LOCAL_ACTION_END_T:
4667 if (!seg6_bpf_has_valid_srh(skb))
4669 if (param_len != sizeof(int))
4671 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
4672 case SEG6_LOCAL_ACTION_END_DT6:
4673 if (!seg6_bpf_has_valid_srh(skb))
4675 if (param_len != sizeof(int))
4678 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
4680 if (!pskb_pull(skb, hdroff))
4683 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
4684 skb_reset_network_header(skb);
4685 skb_reset_transport_header(skb);
4686 skb->encapsulation = 0;
4688 bpf_compute_data_pointers(skb);
4689 bpf_update_srh_state(skb);
4690 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
4691 case SEG6_LOCAL_ACTION_END_B6:
4692 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
4694 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
4697 bpf_update_srh_state(skb);
4700 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
4701 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
4703 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
4706 bpf_update_srh_state(skb);
4714 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
4715 .func = bpf_lwt_seg6_action,
4717 .ret_type = RET_INTEGER,
4718 .arg1_type = ARG_PTR_TO_CTX,
4719 .arg2_type = ARG_ANYTHING,
4720 .arg3_type = ARG_PTR_TO_MEM,
4721 .arg4_type = ARG_CONST_SIZE
4724 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
4727 struct seg6_bpf_srh_state *srh_state =
4728 this_cpu_ptr(&seg6_bpf_srh_states);
4729 struct ipv6_sr_hdr *srh = srh_state->srh;
4730 void *srh_end, *srh_tlvs, *ptr;
4731 struct ipv6hdr *hdr;
4735 if (unlikely(srh == NULL))
4738 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
4739 ((srh->first_segment + 1) << 4));
4740 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
4742 ptr = skb->data + offset;
4744 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
4746 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
4750 ret = skb_cow_head(skb, len);
4751 if (unlikely(ret < 0))
4754 ret = bpf_skb_net_hdr_push(skb, offset, len);
4756 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
4759 bpf_compute_data_pointers(skb);
4760 if (unlikely(ret < 0))
4763 hdr = (struct ipv6hdr *)skb->data;
4764 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
4766 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
4768 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4769 srh_state->hdrlen += len;
4770 srh_state->valid = false;
4774 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
4775 .func = bpf_lwt_seg6_adjust_srh,
4777 .ret_type = RET_INTEGER,
4778 .arg1_type = ARG_PTR_TO_CTX,
4779 .arg2_type = ARG_ANYTHING,
4780 .arg3_type = ARG_ANYTHING,
4782 #endif /* CONFIG_IPV6_SEG6_BPF */
4784 bool bpf_helper_changes_pkt_data(void *func)
4786 if (func == bpf_skb_vlan_push ||
4787 func == bpf_skb_vlan_pop ||
4788 func == bpf_skb_store_bytes ||
4789 func == bpf_skb_change_proto ||
4790 func == bpf_skb_change_head ||
4791 func == sk_skb_change_head ||
4792 func == bpf_skb_change_tail ||
4793 func == sk_skb_change_tail ||
4794 func == bpf_skb_adjust_room ||
4795 func == bpf_skb_pull_data ||
4796 func == sk_skb_pull_data ||
4797 func == bpf_clone_redirect ||
4798 func == bpf_l3_csum_replace ||
4799 func == bpf_l4_csum_replace ||
4800 func == bpf_xdp_adjust_head ||
4801 func == bpf_xdp_adjust_meta ||
4802 func == bpf_msg_pull_data ||
4803 func == bpf_xdp_adjust_tail ||
4804 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4805 func == bpf_lwt_seg6_store_bytes ||
4806 func == bpf_lwt_seg6_adjust_srh ||
4807 func == bpf_lwt_seg6_action ||
4809 func == bpf_lwt_push_encap)
4815 static const struct bpf_func_proto *
4816 bpf_base_func_proto(enum bpf_func_id func_id)
4819 case BPF_FUNC_map_lookup_elem:
4820 return &bpf_map_lookup_elem_proto;
4821 case BPF_FUNC_map_update_elem:
4822 return &bpf_map_update_elem_proto;
4823 case BPF_FUNC_map_delete_elem:
4824 return &bpf_map_delete_elem_proto;
4825 case BPF_FUNC_get_prandom_u32:
4826 return &bpf_get_prandom_u32_proto;
4827 case BPF_FUNC_get_smp_processor_id:
4828 return &bpf_get_raw_smp_processor_id_proto;
4829 case BPF_FUNC_get_numa_node_id:
4830 return &bpf_get_numa_node_id_proto;
4831 case BPF_FUNC_tail_call:
4832 return &bpf_tail_call_proto;
4833 case BPF_FUNC_ktime_get_ns:
4834 return &bpf_ktime_get_ns_proto;
4835 case BPF_FUNC_trace_printk:
4836 if (capable(CAP_SYS_ADMIN))
4837 return bpf_get_trace_printk_proto();
4838 /* else: fall through */
4844 static const struct bpf_func_proto *
4845 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4848 /* inet and inet6 sockets are created in a process
4849 * context so there is always a valid uid/gid
4851 case BPF_FUNC_get_current_uid_gid:
4852 return &bpf_get_current_uid_gid_proto;
4853 case BPF_FUNC_get_local_storage:
4854 return &bpf_get_local_storage_proto;
4856 return bpf_base_func_proto(func_id);
4860 static const struct bpf_func_proto *
4861 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4864 /* inet and inet6 sockets are created in a process
4865 * context so there is always a valid uid/gid
4867 case BPF_FUNC_get_current_uid_gid:
4868 return &bpf_get_current_uid_gid_proto;
4870 switch (prog->expected_attach_type) {
4871 case BPF_CGROUP_INET4_CONNECT:
4872 case BPF_CGROUP_INET6_CONNECT:
4873 return &bpf_bind_proto;
4877 case BPF_FUNC_get_socket_cookie:
4878 return &bpf_get_socket_cookie_sock_addr_proto;
4879 case BPF_FUNC_get_local_storage:
4880 return &bpf_get_local_storage_proto;
4882 return bpf_base_func_proto(func_id);
4886 static const struct bpf_func_proto *
4887 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4890 case BPF_FUNC_skb_load_bytes:
4891 return &bpf_skb_load_bytes_proto;
4892 case BPF_FUNC_skb_load_bytes_relative:
4893 return &bpf_skb_load_bytes_relative_proto;
4894 case BPF_FUNC_get_socket_cookie:
4895 return &bpf_get_socket_cookie_proto;
4896 case BPF_FUNC_get_socket_uid:
4897 return &bpf_get_socket_uid_proto;
4899 return bpf_base_func_proto(func_id);
4903 static const struct bpf_func_proto *
4904 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4907 case BPF_FUNC_get_local_storage:
4908 return &bpf_get_local_storage_proto;
4910 return sk_filter_func_proto(func_id, prog);
4914 static const struct bpf_func_proto *
4915 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4918 case BPF_FUNC_skb_store_bytes:
4919 return &bpf_skb_store_bytes_proto;
4920 case BPF_FUNC_skb_load_bytes:
4921 return &bpf_skb_load_bytes_proto;
4922 case BPF_FUNC_skb_load_bytes_relative:
4923 return &bpf_skb_load_bytes_relative_proto;
4924 case BPF_FUNC_skb_pull_data:
4925 return &bpf_skb_pull_data_proto;
4926 case BPF_FUNC_csum_diff:
4927 return &bpf_csum_diff_proto;
4928 case BPF_FUNC_csum_update:
4929 return &bpf_csum_update_proto;
4930 case BPF_FUNC_l3_csum_replace:
4931 return &bpf_l3_csum_replace_proto;
4932 case BPF_FUNC_l4_csum_replace:
4933 return &bpf_l4_csum_replace_proto;
4934 case BPF_FUNC_clone_redirect:
4935 return &bpf_clone_redirect_proto;
4936 case BPF_FUNC_get_cgroup_classid:
4937 return &bpf_get_cgroup_classid_proto;
4938 case BPF_FUNC_skb_vlan_push:
4939 return &bpf_skb_vlan_push_proto;
4940 case BPF_FUNC_skb_vlan_pop:
4941 return &bpf_skb_vlan_pop_proto;
4942 case BPF_FUNC_skb_change_proto:
4943 return &bpf_skb_change_proto_proto;
4944 case BPF_FUNC_skb_change_type:
4945 return &bpf_skb_change_type_proto;
4946 case BPF_FUNC_skb_adjust_room:
4947 return &bpf_skb_adjust_room_proto;
4948 case BPF_FUNC_skb_change_tail:
4949 return &bpf_skb_change_tail_proto;
4950 case BPF_FUNC_skb_get_tunnel_key:
4951 return &bpf_skb_get_tunnel_key_proto;
4952 case BPF_FUNC_skb_set_tunnel_key:
4953 return bpf_get_skb_set_tunnel_proto(func_id);
4954 case BPF_FUNC_skb_get_tunnel_opt:
4955 return &bpf_skb_get_tunnel_opt_proto;
4956 case BPF_FUNC_skb_set_tunnel_opt:
4957 return bpf_get_skb_set_tunnel_proto(func_id);
4958 case BPF_FUNC_redirect:
4959 return &bpf_redirect_proto;
4960 case BPF_FUNC_get_route_realm:
4961 return &bpf_get_route_realm_proto;
4962 case BPF_FUNC_get_hash_recalc:
4963 return &bpf_get_hash_recalc_proto;
4964 case BPF_FUNC_set_hash_invalid:
4965 return &bpf_set_hash_invalid_proto;
4966 case BPF_FUNC_set_hash:
4967 return &bpf_set_hash_proto;
4968 case BPF_FUNC_perf_event_output:
4969 return &bpf_skb_event_output_proto;
4970 case BPF_FUNC_get_smp_processor_id:
4971 return &bpf_get_smp_processor_id_proto;
4972 case BPF_FUNC_skb_under_cgroup:
4973 return &bpf_skb_under_cgroup_proto;
4974 case BPF_FUNC_get_socket_cookie:
4975 return &bpf_get_socket_cookie_proto;
4976 case BPF_FUNC_get_socket_uid:
4977 return &bpf_get_socket_uid_proto;
4978 case BPF_FUNC_fib_lookup:
4979 return &bpf_skb_fib_lookup_proto;
4981 case BPF_FUNC_skb_get_xfrm_state:
4982 return &bpf_skb_get_xfrm_state_proto;
4984 #ifdef CONFIG_SOCK_CGROUP_DATA
4985 case BPF_FUNC_skb_cgroup_id:
4986 return &bpf_skb_cgroup_id_proto;
4987 case BPF_FUNC_skb_ancestor_cgroup_id:
4988 return &bpf_skb_ancestor_cgroup_id_proto;
4991 return bpf_base_func_proto(func_id);
4995 static const struct bpf_func_proto *
4996 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4999 case BPF_FUNC_perf_event_output:
5000 return &bpf_xdp_event_output_proto;
5001 case BPF_FUNC_get_smp_processor_id:
5002 return &bpf_get_smp_processor_id_proto;
5003 case BPF_FUNC_csum_diff:
5004 return &bpf_csum_diff_proto;
5005 case BPF_FUNC_xdp_adjust_head:
5006 return &bpf_xdp_adjust_head_proto;
5007 case BPF_FUNC_xdp_adjust_meta:
5008 return &bpf_xdp_adjust_meta_proto;
5009 case BPF_FUNC_redirect:
5010 return &bpf_xdp_redirect_proto;
5011 case BPF_FUNC_redirect_map:
5012 return &bpf_xdp_redirect_map_proto;
5013 case BPF_FUNC_xdp_adjust_tail:
5014 return &bpf_xdp_adjust_tail_proto;
5015 case BPF_FUNC_fib_lookup:
5016 return &bpf_xdp_fib_lookup_proto;
5018 return bpf_base_func_proto(func_id);
5022 static const struct bpf_func_proto *
5023 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5026 case BPF_FUNC_setsockopt:
5027 return &bpf_setsockopt_proto;
5028 case BPF_FUNC_getsockopt:
5029 return &bpf_getsockopt_proto;
5030 case BPF_FUNC_sock_ops_cb_flags_set:
5031 return &bpf_sock_ops_cb_flags_set_proto;
5032 case BPF_FUNC_sock_map_update:
5033 return &bpf_sock_map_update_proto;
5034 case BPF_FUNC_sock_hash_update:
5035 return &bpf_sock_hash_update_proto;
5036 case BPF_FUNC_get_socket_cookie:
5037 return &bpf_get_socket_cookie_sock_ops_proto;
5038 case BPF_FUNC_get_local_storage:
5039 return &bpf_get_local_storage_proto;
5041 return bpf_base_func_proto(func_id);
5045 static const struct bpf_func_proto *
5046 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5049 case BPF_FUNC_msg_redirect_map:
5050 return &bpf_msg_redirect_map_proto;
5051 case BPF_FUNC_msg_redirect_hash:
5052 return &bpf_msg_redirect_hash_proto;
5053 case BPF_FUNC_msg_apply_bytes:
5054 return &bpf_msg_apply_bytes_proto;
5055 case BPF_FUNC_msg_cork_bytes:
5056 return &bpf_msg_cork_bytes_proto;
5057 case BPF_FUNC_msg_pull_data:
5058 return &bpf_msg_pull_data_proto;
5059 case BPF_FUNC_get_local_storage:
5060 return &bpf_get_local_storage_proto;
5062 return bpf_base_func_proto(func_id);
5066 static const struct bpf_func_proto *
5067 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5070 case BPF_FUNC_skb_store_bytes:
5071 return &bpf_skb_store_bytes_proto;
5072 case BPF_FUNC_skb_load_bytes:
5073 return &bpf_skb_load_bytes_proto;
5074 case BPF_FUNC_skb_pull_data:
5075 return &sk_skb_pull_data_proto;
5076 case BPF_FUNC_skb_change_tail:
5077 return &sk_skb_change_tail_proto;
5078 case BPF_FUNC_skb_change_head:
5079 return &sk_skb_change_head_proto;
5080 case BPF_FUNC_get_socket_cookie:
5081 return &bpf_get_socket_cookie_proto;
5082 case BPF_FUNC_get_socket_uid:
5083 return &bpf_get_socket_uid_proto;
5084 case BPF_FUNC_sk_redirect_map:
5085 return &bpf_sk_redirect_map_proto;
5086 case BPF_FUNC_sk_redirect_hash:
5087 return &bpf_sk_redirect_hash_proto;
5088 case BPF_FUNC_get_local_storage:
5089 return &bpf_get_local_storage_proto;
5091 return bpf_base_func_proto(func_id);
5095 static const struct bpf_func_proto *
5096 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5099 case BPF_FUNC_skb_load_bytes:
5100 return &bpf_skb_load_bytes_proto;
5101 case BPF_FUNC_skb_pull_data:
5102 return &bpf_skb_pull_data_proto;
5103 case BPF_FUNC_csum_diff:
5104 return &bpf_csum_diff_proto;
5105 case BPF_FUNC_get_cgroup_classid:
5106 return &bpf_get_cgroup_classid_proto;
5107 case BPF_FUNC_get_route_realm:
5108 return &bpf_get_route_realm_proto;
5109 case BPF_FUNC_get_hash_recalc:
5110 return &bpf_get_hash_recalc_proto;
5111 case BPF_FUNC_perf_event_output:
5112 return &bpf_skb_event_output_proto;
5113 case BPF_FUNC_get_smp_processor_id:
5114 return &bpf_get_smp_processor_id_proto;
5115 case BPF_FUNC_skb_under_cgroup:
5116 return &bpf_skb_under_cgroup_proto;
5118 return bpf_base_func_proto(func_id);
5122 static const struct bpf_func_proto *
5123 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5126 case BPF_FUNC_lwt_push_encap:
5127 return &bpf_lwt_push_encap_proto;
5129 return lwt_out_func_proto(func_id, prog);
5133 static const struct bpf_func_proto *
5134 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5137 case BPF_FUNC_skb_get_tunnel_key:
5138 return &bpf_skb_get_tunnel_key_proto;
5139 case BPF_FUNC_skb_set_tunnel_key:
5140 return bpf_get_skb_set_tunnel_proto(func_id);
5141 case BPF_FUNC_skb_get_tunnel_opt:
5142 return &bpf_skb_get_tunnel_opt_proto;
5143 case BPF_FUNC_skb_set_tunnel_opt:
5144 return bpf_get_skb_set_tunnel_proto(func_id);
5145 case BPF_FUNC_redirect:
5146 return &bpf_redirect_proto;
5147 case BPF_FUNC_clone_redirect:
5148 return &bpf_clone_redirect_proto;
5149 case BPF_FUNC_skb_change_tail:
5150 return &bpf_skb_change_tail_proto;
5151 case BPF_FUNC_skb_change_head:
5152 return &bpf_skb_change_head_proto;
5153 case BPF_FUNC_skb_store_bytes:
5154 return &bpf_skb_store_bytes_proto;
5155 case BPF_FUNC_csum_update:
5156 return &bpf_csum_update_proto;
5157 case BPF_FUNC_l3_csum_replace:
5158 return &bpf_l3_csum_replace_proto;
5159 case BPF_FUNC_l4_csum_replace:
5160 return &bpf_l4_csum_replace_proto;
5161 case BPF_FUNC_set_hash_invalid:
5162 return &bpf_set_hash_invalid_proto;
5164 return lwt_out_func_proto(func_id, prog);
5168 static const struct bpf_func_proto *
5169 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5172 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5173 case BPF_FUNC_lwt_seg6_store_bytes:
5174 return &bpf_lwt_seg6_store_bytes_proto;
5175 case BPF_FUNC_lwt_seg6_action:
5176 return &bpf_lwt_seg6_action_proto;
5177 case BPF_FUNC_lwt_seg6_adjust_srh:
5178 return &bpf_lwt_seg6_adjust_srh_proto;
5181 return lwt_out_func_proto(func_id, prog);
5185 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
5186 const struct bpf_prog *prog,
5187 struct bpf_insn_access_aux *info)
5189 const int size_default = sizeof(__u32);
5191 if (off < 0 || off >= sizeof(struct __sk_buff))
5194 /* The verifier guarantees that size > 0. */
5195 if (off % size != 0)
5199 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5200 if (off + size > offsetofend(struct __sk_buff, cb[4]))
5203 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
5204 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
5205 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
5206 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
5207 case bpf_ctx_range(struct __sk_buff, data):
5208 case bpf_ctx_range(struct __sk_buff, data_meta):
5209 case bpf_ctx_range(struct __sk_buff, data_end):
5210 if (size != size_default)
5214 /* Only narrow read access allowed for now. */
5215 if (type == BPF_WRITE) {
5216 if (size != size_default)
5219 bpf_ctx_record_field_size(info, size_default);
5220 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
5228 static bool sk_filter_is_valid_access(int off, int size,
5229 enum bpf_access_type type,
5230 const struct bpf_prog *prog,
5231 struct bpf_insn_access_aux *info)
5234 case bpf_ctx_range(struct __sk_buff, tc_classid):
5235 case bpf_ctx_range(struct __sk_buff, data):
5236 case bpf_ctx_range(struct __sk_buff, data_meta):
5237 case bpf_ctx_range(struct __sk_buff, data_end):
5238 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
5242 if (type == BPF_WRITE) {
5244 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5251 return bpf_skb_is_valid_access(off, size, type, prog, info);
5254 static bool lwt_is_valid_access(int off, int size,
5255 enum bpf_access_type type,
5256 const struct bpf_prog *prog,
5257 struct bpf_insn_access_aux *info)
5260 case bpf_ctx_range(struct __sk_buff, tc_classid):
5261 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
5262 case bpf_ctx_range(struct __sk_buff, data_meta):
5266 if (type == BPF_WRITE) {
5268 case bpf_ctx_range(struct __sk_buff, mark):
5269 case bpf_ctx_range(struct __sk_buff, priority):
5270 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5278 case bpf_ctx_range(struct __sk_buff, data):
5279 info->reg_type = PTR_TO_PACKET;
5281 case bpf_ctx_range(struct __sk_buff, data_end):
5282 info->reg_type = PTR_TO_PACKET_END;
5286 return bpf_skb_is_valid_access(off, size, type, prog, info);
5289 /* Attach type specific accesses */
5290 static bool __sock_filter_check_attach_type(int off,
5291 enum bpf_access_type access_type,
5292 enum bpf_attach_type attach_type)
5295 case offsetof(struct bpf_sock, bound_dev_if):
5296 case offsetof(struct bpf_sock, mark):
5297 case offsetof(struct bpf_sock, priority):
5298 switch (attach_type) {
5299 case BPF_CGROUP_INET_SOCK_CREATE:
5304 case bpf_ctx_range(struct bpf_sock, src_ip4):
5305 switch (attach_type) {
5306 case BPF_CGROUP_INET4_POST_BIND:
5311 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
5312 switch (attach_type) {
5313 case BPF_CGROUP_INET6_POST_BIND:
5318 case bpf_ctx_range(struct bpf_sock, src_port):
5319 switch (attach_type) {
5320 case BPF_CGROUP_INET4_POST_BIND:
5321 case BPF_CGROUP_INET6_POST_BIND:
5328 return access_type == BPF_READ;
5333 static bool __sock_filter_check_size(int off, int size,
5334 struct bpf_insn_access_aux *info)
5336 const int size_default = sizeof(__u32);
5339 case bpf_ctx_range(struct bpf_sock, src_ip4):
5340 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
5341 bpf_ctx_record_field_size(info, size_default);
5342 return bpf_ctx_narrow_access_ok(off, size, size_default);
5345 return size == size_default;
5348 static bool sock_filter_is_valid_access(int off, int size,
5349 enum bpf_access_type type,
5350 const struct bpf_prog *prog,
5351 struct bpf_insn_access_aux *info)
5353 if (off < 0 || off >= sizeof(struct bpf_sock))
5355 if (off % size != 0)
5357 if (!__sock_filter_check_attach_type(off, type,
5358 prog->expected_attach_type))
5360 if (!__sock_filter_check_size(off, size, info))
5365 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
5366 const struct bpf_prog *prog, int drop_verdict)
5368 struct bpf_insn *insn = insn_buf;
5373 /* if (!skb->cloned)
5376 * (Fast-path, otherwise approximation that we might be
5377 * a clone, do the rest in helper.)
5379 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
5380 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
5381 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
5383 /* ret = bpf_skb_pull_data(skb, 0); */
5384 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
5385 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
5386 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
5387 BPF_FUNC_skb_pull_data);
5390 * return TC_ACT_SHOT;
5392 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
5393 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
5394 *insn++ = BPF_EXIT_INSN();
5397 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
5399 *insn++ = prog->insnsi[0];
5401 return insn - insn_buf;
5404 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
5405 struct bpf_insn *insn_buf)
5407 bool indirect = BPF_MODE(orig->code) == BPF_IND;
5408 struct bpf_insn *insn = insn_buf;
5410 /* We're guaranteed here that CTX is in R6. */
5411 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
5413 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
5415 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
5417 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
5420 switch (BPF_SIZE(orig->code)) {
5422 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
5425 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
5428 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
5432 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
5433 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
5434 *insn++ = BPF_EXIT_INSN();
5436 return insn - insn_buf;
5439 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
5440 const struct bpf_prog *prog)
5442 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
5445 static bool tc_cls_act_is_valid_access(int off, int size,
5446 enum bpf_access_type type,
5447 const struct bpf_prog *prog,
5448 struct bpf_insn_access_aux *info)
5450 if (type == BPF_WRITE) {
5452 case bpf_ctx_range(struct __sk_buff, mark):
5453 case bpf_ctx_range(struct __sk_buff, tc_index):
5454 case bpf_ctx_range(struct __sk_buff, priority):
5455 case bpf_ctx_range(struct __sk_buff, tc_classid):
5456 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5464 case bpf_ctx_range(struct __sk_buff, data):
5465 info->reg_type = PTR_TO_PACKET;
5467 case bpf_ctx_range(struct __sk_buff, data_meta):
5468 info->reg_type = PTR_TO_PACKET_META;
5470 case bpf_ctx_range(struct __sk_buff, data_end):
5471 info->reg_type = PTR_TO_PACKET_END;
5473 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
5477 return bpf_skb_is_valid_access(off, size, type, prog, info);
5480 static bool __is_valid_xdp_access(int off, int size)
5482 if (off < 0 || off >= sizeof(struct xdp_md))
5484 if (off % size != 0)
5486 if (size != sizeof(__u32))
5492 static bool xdp_is_valid_access(int off, int size,
5493 enum bpf_access_type type,
5494 const struct bpf_prog *prog,
5495 struct bpf_insn_access_aux *info)
5497 if (type == BPF_WRITE) {
5498 if (bpf_prog_is_dev_bound(prog->aux)) {
5500 case offsetof(struct xdp_md, rx_queue_index):
5501 return __is_valid_xdp_access(off, size);
5508 case offsetof(struct xdp_md, data):
5509 info->reg_type = PTR_TO_PACKET;
5511 case offsetof(struct xdp_md, data_meta):
5512 info->reg_type = PTR_TO_PACKET_META;
5514 case offsetof(struct xdp_md, data_end):
5515 info->reg_type = PTR_TO_PACKET_END;
5519 return __is_valid_xdp_access(off, size);
5522 void bpf_warn_invalid_xdp_action(u32 act)
5524 const u32 act_max = XDP_REDIRECT;
5526 WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
5527 act > act_max ? "Illegal" : "Driver unsupported",
5530 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
5532 static bool sock_addr_is_valid_access(int off, int size,
5533 enum bpf_access_type type,
5534 const struct bpf_prog *prog,
5535 struct bpf_insn_access_aux *info)
5537 const int size_default = sizeof(__u32);
5539 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
5541 if (off % size != 0)
5544 /* Disallow access to IPv6 fields from IPv4 contex and vise
5548 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
5549 switch (prog->expected_attach_type) {
5550 case BPF_CGROUP_INET4_BIND:
5551 case BPF_CGROUP_INET4_CONNECT:
5552 case BPF_CGROUP_UDP4_SENDMSG:
5558 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
5559 switch (prog->expected_attach_type) {
5560 case BPF_CGROUP_INET6_BIND:
5561 case BPF_CGROUP_INET6_CONNECT:
5562 case BPF_CGROUP_UDP6_SENDMSG:
5568 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
5569 switch (prog->expected_attach_type) {
5570 case BPF_CGROUP_UDP4_SENDMSG:
5576 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
5578 switch (prog->expected_attach_type) {
5579 case BPF_CGROUP_UDP6_SENDMSG:
5588 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
5589 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
5590 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
5591 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
5593 /* Only narrow read access allowed for now. */
5594 if (type == BPF_READ) {
5595 bpf_ctx_record_field_size(info, size_default);
5596 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
5599 if (size != size_default)
5603 case bpf_ctx_range(struct bpf_sock_addr, user_port):
5604 if (size != size_default)
5608 if (type == BPF_READ) {
5609 if (size != size_default)
5619 static bool sock_ops_is_valid_access(int off, int size,
5620 enum bpf_access_type type,
5621 const struct bpf_prog *prog,
5622 struct bpf_insn_access_aux *info)
5624 const int size_default = sizeof(__u32);
5626 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
5629 /* The verifier guarantees that size > 0. */
5630 if (off % size != 0)
5633 if (type == BPF_WRITE) {
5635 case offsetof(struct bpf_sock_ops, reply):
5636 case offsetof(struct bpf_sock_ops, sk_txhash):
5637 if (size != size_default)
5645 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
5647 if (size != sizeof(__u64))
5651 if (size != size_default)
5660 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
5661 const struct bpf_prog *prog)
5663 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
5666 static bool sk_skb_is_valid_access(int off, int size,
5667 enum bpf_access_type type,
5668 const struct bpf_prog *prog,
5669 struct bpf_insn_access_aux *info)
5672 case bpf_ctx_range(struct __sk_buff, tc_classid):
5673 case bpf_ctx_range(struct __sk_buff, data_meta):
5677 if (type == BPF_WRITE) {
5679 case bpf_ctx_range(struct __sk_buff, tc_index):
5680 case bpf_ctx_range(struct __sk_buff, priority):
5688 case bpf_ctx_range(struct __sk_buff, mark):
5690 case bpf_ctx_range(struct __sk_buff, data):
5691 info->reg_type = PTR_TO_PACKET;
5693 case bpf_ctx_range(struct __sk_buff, data_end):
5694 info->reg_type = PTR_TO_PACKET_END;
5698 return bpf_skb_is_valid_access(off, size, type, prog, info);
5701 static bool sk_msg_is_valid_access(int off, int size,
5702 enum bpf_access_type type,
5703 const struct bpf_prog *prog,
5704 struct bpf_insn_access_aux *info)
5706 if (type == BPF_WRITE)
5710 case offsetof(struct sk_msg_md, data):
5711 info->reg_type = PTR_TO_PACKET;
5712 if (size != sizeof(__u64))
5715 case offsetof(struct sk_msg_md, data_end):
5716 info->reg_type = PTR_TO_PACKET_END;
5717 if (size != sizeof(__u64))
5721 if (size != sizeof(__u32))
5725 if (off < 0 || off >= sizeof(struct sk_msg_md))
5727 if (off % size != 0)
5733 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
5734 const struct bpf_insn *si,
5735 struct bpf_insn *insn_buf,
5736 struct bpf_prog *prog, u32 *target_size)
5738 struct bpf_insn *insn = insn_buf;
5742 case offsetof(struct __sk_buff, len):
5743 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5744 bpf_target_off(struct sk_buff, len, 4,
5748 case offsetof(struct __sk_buff, protocol):
5749 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5750 bpf_target_off(struct sk_buff, protocol, 2,
5754 case offsetof(struct __sk_buff, vlan_proto):
5755 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5756 bpf_target_off(struct sk_buff, vlan_proto, 2,
5760 case offsetof(struct __sk_buff, priority):
5761 if (type == BPF_WRITE)
5762 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
5763 bpf_target_off(struct sk_buff, priority, 4,
5766 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5767 bpf_target_off(struct sk_buff, priority, 4,
5771 case offsetof(struct __sk_buff, ingress_ifindex):
5772 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5773 bpf_target_off(struct sk_buff, skb_iif, 4,
5777 case offsetof(struct __sk_buff, ifindex):
5778 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
5779 si->dst_reg, si->src_reg,
5780 offsetof(struct sk_buff, dev));
5781 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
5782 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
5783 bpf_target_off(struct net_device, ifindex, 4,
5787 case offsetof(struct __sk_buff, hash):
5788 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5789 bpf_target_off(struct sk_buff, hash, 4,
5793 case offsetof(struct __sk_buff, mark):
5794 if (type == BPF_WRITE)
5795 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
5796 bpf_target_off(struct sk_buff, mark, 4,
5799 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5800 bpf_target_off(struct sk_buff, mark, 4,
5804 case offsetof(struct __sk_buff, pkt_type):
5806 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
5808 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
5809 #ifdef __BIG_ENDIAN_BITFIELD
5810 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
5814 case offsetof(struct __sk_buff, queue_mapping):
5815 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5816 bpf_target_off(struct sk_buff, queue_mapping, 2,
5820 case offsetof(struct __sk_buff, vlan_present):
5821 case offsetof(struct __sk_buff, vlan_tci):
5822 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000);
5824 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5825 bpf_target_off(struct sk_buff, vlan_tci, 2,
5827 if (si->off == offsetof(struct __sk_buff, vlan_tci)) {
5828 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg,
5831 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 12);
5832 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
5836 case offsetof(struct __sk_buff, cb[0]) ...
5837 offsetofend(struct __sk_buff, cb[4]) - 1:
5838 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20);
5839 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
5840 offsetof(struct qdisc_skb_cb, data)) %
5843 prog->cb_access = 1;
5845 off -= offsetof(struct __sk_buff, cb[0]);
5846 off += offsetof(struct sk_buff, cb);
5847 off += offsetof(struct qdisc_skb_cb, data);
5848 if (type == BPF_WRITE)
5849 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
5852 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
5856 case offsetof(struct __sk_buff, tc_classid):
5857 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, tc_classid) != 2);
5860 off -= offsetof(struct __sk_buff, tc_classid);
5861 off += offsetof(struct sk_buff, cb);
5862 off += offsetof(struct qdisc_skb_cb, tc_classid);
5864 if (type == BPF_WRITE)
5865 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
5868 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
5872 case offsetof(struct __sk_buff, data):
5873 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
5874 si->dst_reg, si->src_reg,
5875 offsetof(struct sk_buff, data));
5878 case offsetof(struct __sk_buff, data_meta):
5880 off -= offsetof(struct __sk_buff, data_meta);
5881 off += offsetof(struct sk_buff, cb);
5882 off += offsetof(struct bpf_skb_data_end, data_meta);
5883 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
5887 case offsetof(struct __sk_buff, data_end):
5889 off -= offsetof(struct __sk_buff, data_end);
5890 off += offsetof(struct sk_buff, cb);
5891 off += offsetof(struct bpf_skb_data_end, data_end);
5892 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
5896 case offsetof(struct __sk_buff, tc_index):
5897 #ifdef CONFIG_NET_SCHED
5898 if (type == BPF_WRITE)
5899 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
5900 bpf_target_off(struct sk_buff, tc_index, 2,
5903 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5904 bpf_target_off(struct sk_buff, tc_index, 2,
5908 if (type == BPF_WRITE)
5909 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
5911 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
5915 case offsetof(struct __sk_buff, napi_id):
5916 #if defined(CONFIG_NET_RX_BUSY_POLL)
5917 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5918 bpf_target_off(struct sk_buff, napi_id, 4,
5920 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
5921 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
5924 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
5927 case offsetof(struct __sk_buff, family):
5928 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
5930 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
5931 si->dst_reg, si->src_reg,
5932 offsetof(struct sk_buff, sk));
5933 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
5934 bpf_target_off(struct sock_common,
5938 case offsetof(struct __sk_buff, remote_ip4):
5939 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
5941 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
5942 si->dst_reg, si->src_reg,
5943 offsetof(struct sk_buff, sk));
5944 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
5945 bpf_target_off(struct sock_common,
5949 case offsetof(struct __sk_buff, local_ip4):
5950 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
5951 skc_rcv_saddr) != 4);
5953 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
5954 si->dst_reg, si->src_reg,
5955 offsetof(struct sk_buff, sk));
5956 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
5957 bpf_target_off(struct sock_common,
5961 case offsetof(struct __sk_buff, remote_ip6[0]) ...
5962 offsetof(struct __sk_buff, remote_ip6[3]):
5963 #if IS_ENABLED(CONFIG_IPV6)
5964 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
5965 skc_v6_daddr.s6_addr32[0]) != 4);
5968 off -= offsetof(struct __sk_buff, remote_ip6[0]);
5970 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
5971 si->dst_reg, si->src_reg,
5972 offsetof(struct sk_buff, sk));
5973 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
5974 offsetof(struct sock_common,
5975 skc_v6_daddr.s6_addr32[0]) +
5978 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
5981 case offsetof(struct __sk_buff, local_ip6[0]) ...
5982 offsetof(struct __sk_buff, local_ip6[3]):
5983 #if IS_ENABLED(CONFIG_IPV6)
5984 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
5985 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
5988 off -= offsetof(struct __sk_buff, local_ip6[0]);
5990 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
5991 si->dst_reg, si->src_reg,
5992 offsetof(struct sk_buff, sk));
5993 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
5994 offsetof(struct sock_common,
5995 skc_v6_rcv_saddr.s6_addr32[0]) +
5998 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6002 case offsetof(struct __sk_buff, remote_port):
6003 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
6005 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6006 si->dst_reg, si->src_reg,
6007 offsetof(struct sk_buff, sk));
6008 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6009 bpf_target_off(struct sock_common,
6012 #ifndef __BIG_ENDIAN_BITFIELD
6013 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
6017 case offsetof(struct __sk_buff, local_port):
6018 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
6020 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6021 si->dst_reg, si->src_reg,
6022 offsetof(struct sk_buff, sk));
6023 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6024 bpf_target_off(struct sock_common,
6025 skc_num, 2, target_size));
6029 return insn - insn_buf;
6032 static u32 sock_filter_convert_ctx_access(enum bpf_access_type type,
6033 const struct bpf_insn *si,
6034 struct bpf_insn *insn_buf,
6035 struct bpf_prog *prog, u32 *target_size)
6037 struct bpf_insn *insn = insn_buf;
6041 case offsetof(struct bpf_sock, bound_dev_if):
6042 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_bound_dev_if) != 4);
6044 if (type == BPF_WRITE)
6045 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6046 offsetof(struct sock, sk_bound_dev_if));
6048 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6049 offsetof(struct sock, sk_bound_dev_if));
6052 case offsetof(struct bpf_sock, mark):
6053 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_mark) != 4);
6055 if (type == BPF_WRITE)
6056 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6057 offsetof(struct sock, sk_mark));
6059 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6060 offsetof(struct sock, sk_mark));
6063 case offsetof(struct bpf_sock, priority):
6064 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_priority) != 4);
6066 if (type == BPF_WRITE)
6067 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6068 offsetof(struct sock, sk_priority));
6070 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6071 offsetof(struct sock, sk_priority));
6074 case offsetof(struct bpf_sock, family):
6075 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_family) != 2);
6077 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6078 offsetof(struct sock, sk_family));
6081 case offsetof(struct bpf_sock, type):
6082 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6083 offsetof(struct sock, __sk_flags_offset));
6084 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
6085 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
6088 case offsetof(struct bpf_sock, protocol):
6089 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6090 offsetof(struct sock, __sk_flags_offset));
6091 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
6092 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_PROTO_SHIFT);
6095 case offsetof(struct bpf_sock, src_ip4):
6096 *insn++ = BPF_LDX_MEM(
6097 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
6098 bpf_target_off(struct sock_common, skc_rcv_saddr,
6099 FIELD_SIZEOF(struct sock_common,
6104 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
6105 #if IS_ENABLED(CONFIG_IPV6)
6107 off -= offsetof(struct bpf_sock, src_ip6[0]);
6108 *insn++ = BPF_LDX_MEM(
6109 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
6112 skc_v6_rcv_saddr.s6_addr32[0],
6113 FIELD_SIZEOF(struct sock_common,
6114 skc_v6_rcv_saddr.s6_addr32[0]),
6115 target_size) + off);
6118 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6122 case offsetof(struct bpf_sock, src_port):
6123 *insn++ = BPF_LDX_MEM(
6124 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
6125 si->dst_reg, si->src_reg,
6126 bpf_target_off(struct sock_common, skc_num,
6127 FIELD_SIZEOF(struct sock_common,
6133 return insn - insn_buf;
6136 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
6137 const struct bpf_insn *si,
6138 struct bpf_insn *insn_buf,
6139 struct bpf_prog *prog, u32 *target_size)
6141 struct bpf_insn *insn = insn_buf;
6144 case offsetof(struct __sk_buff, ifindex):
6145 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
6146 si->dst_reg, si->src_reg,
6147 offsetof(struct sk_buff, dev));
6148 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6149 bpf_target_off(struct net_device, ifindex, 4,
6153 return bpf_convert_ctx_access(type, si, insn_buf, prog,
6157 return insn - insn_buf;
6160 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
6161 const struct bpf_insn *si,
6162 struct bpf_insn *insn_buf,
6163 struct bpf_prog *prog, u32 *target_size)
6165 struct bpf_insn *insn = insn_buf;
6168 case offsetof(struct xdp_md, data):
6169 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
6170 si->dst_reg, si->src_reg,
6171 offsetof(struct xdp_buff, data));
6173 case offsetof(struct xdp_md, data_meta):
6174 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
6175 si->dst_reg, si->src_reg,
6176 offsetof(struct xdp_buff, data_meta));
6178 case offsetof(struct xdp_md, data_end):
6179 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
6180 si->dst_reg, si->src_reg,
6181 offsetof(struct xdp_buff, data_end));
6183 case offsetof(struct xdp_md, ingress_ifindex):
6184 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
6185 si->dst_reg, si->src_reg,
6186 offsetof(struct xdp_buff, rxq));
6187 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
6188 si->dst_reg, si->dst_reg,
6189 offsetof(struct xdp_rxq_info, dev));
6190 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6191 offsetof(struct net_device, ifindex));
6193 case offsetof(struct xdp_md, rx_queue_index):
6194 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
6195 si->dst_reg, si->src_reg,
6196 offsetof(struct xdp_buff, rxq));
6197 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6198 offsetof(struct xdp_rxq_info,
6203 return insn - insn_buf;
6206 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
6207 * context Structure, F is Field in context structure that contains a pointer
6208 * to Nested Structure of type NS that has the field NF.
6210 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
6211 * sure that SIZE is not greater than actual size of S.F.NF.
6213 * If offset OFF is provided, the load happens from that offset relative to
6216 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
6218 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
6219 si->src_reg, offsetof(S, F)); \
6220 *insn++ = BPF_LDX_MEM( \
6221 SIZE, si->dst_reg, si->dst_reg, \
6222 bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF), \
6227 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
6228 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
6229 BPF_FIELD_SIZEOF(NS, NF), 0)
6231 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
6232 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
6234 * It doesn't support SIZE argument though since narrow stores are not
6235 * supported for now.
6237 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
6238 * "register" since two registers available in convert_ctx_access are not
6239 * enough: we can't override neither SRC, since it contains value to store, nor
6240 * DST since it contains pointer to context that may be used by later
6241 * instructions. But we need a temporary place to save pointer to nested
6242 * structure whose field we want to store to.
6244 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, OFF, TF) \
6246 int tmp_reg = BPF_REG_9; \
6247 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
6249 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
6251 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
6253 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
6254 si->dst_reg, offsetof(S, F)); \
6255 *insn++ = BPF_STX_MEM( \
6256 BPF_FIELD_SIZEOF(NS, NF), tmp_reg, si->src_reg, \
6257 bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF), \
6260 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
6264 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
6267 if (type == BPF_WRITE) { \
6268 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, OFF, \
6271 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
6272 S, NS, F, NF, SIZE, OFF); \
6276 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
6277 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
6278 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
6280 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
6281 const struct bpf_insn *si,
6282 struct bpf_insn *insn_buf,
6283 struct bpf_prog *prog, u32 *target_size)
6285 struct bpf_insn *insn = insn_buf;
6289 case offsetof(struct bpf_sock_addr, user_family):
6290 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
6291 struct sockaddr, uaddr, sa_family);
6294 case offsetof(struct bpf_sock_addr, user_ip4):
6295 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
6296 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
6297 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
6300 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
6302 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
6303 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
6304 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
6305 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
6309 case offsetof(struct bpf_sock_addr, user_port):
6310 /* To get port we need to know sa_family first and then treat
6311 * sockaddr as either sockaddr_in or sockaddr_in6.
6312 * Though we can simplify since port field has same offset and
6313 * size in both structures.
6314 * Here we check this invariant and use just one of the
6315 * structures if it's true.
6317 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
6318 offsetof(struct sockaddr_in6, sin6_port));
6319 BUILD_BUG_ON(FIELD_SIZEOF(struct sockaddr_in, sin_port) !=
6320 FIELD_SIZEOF(struct sockaddr_in6, sin6_port));
6321 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(struct bpf_sock_addr_kern,
6322 struct sockaddr_in6, uaddr,
6323 sin6_port, tmp_reg);
6326 case offsetof(struct bpf_sock_addr, family):
6327 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
6328 struct sock, sk, sk_family);
6331 case offsetof(struct bpf_sock_addr, type):
6332 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(
6333 struct bpf_sock_addr_kern, struct sock, sk,
6334 __sk_flags_offset, BPF_W, 0);
6335 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
6336 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
6339 case offsetof(struct bpf_sock_addr, protocol):
6340 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(
6341 struct bpf_sock_addr_kern, struct sock, sk,
6342 __sk_flags_offset, BPF_W, 0);
6343 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
6344 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg,
6348 case offsetof(struct bpf_sock_addr, msg_src_ip4):
6349 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
6350 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
6351 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
6352 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
6355 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
6358 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
6359 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
6360 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
6361 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
6362 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
6366 return insn - insn_buf;
6369 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
6370 const struct bpf_insn *si,
6371 struct bpf_insn *insn_buf,
6372 struct bpf_prog *prog,
6375 struct bpf_insn *insn = insn_buf;
6379 case offsetof(struct bpf_sock_ops, op) ...
6380 offsetof(struct bpf_sock_ops, replylong[3]):
6381 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, op) !=
6382 FIELD_SIZEOF(struct bpf_sock_ops_kern, op));
6383 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, reply) !=
6384 FIELD_SIZEOF(struct bpf_sock_ops_kern, reply));
6385 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, replylong) !=
6386 FIELD_SIZEOF(struct bpf_sock_ops_kern, replylong));
6388 off -= offsetof(struct bpf_sock_ops, op);
6389 off += offsetof(struct bpf_sock_ops_kern, op);
6390 if (type == BPF_WRITE)
6391 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6394 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6398 case offsetof(struct bpf_sock_ops, family):
6399 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
6401 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6402 struct bpf_sock_ops_kern, sk),
6403 si->dst_reg, si->src_reg,
6404 offsetof(struct bpf_sock_ops_kern, sk));
6405 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6406 offsetof(struct sock_common, skc_family));
6409 case offsetof(struct bpf_sock_ops, remote_ip4):
6410 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
6412 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6413 struct bpf_sock_ops_kern, sk),
6414 si->dst_reg, si->src_reg,
6415 offsetof(struct bpf_sock_ops_kern, sk));
6416 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6417 offsetof(struct sock_common, skc_daddr));
6420 case offsetof(struct bpf_sock_ops, local_ip4):
6421 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6422 skc_rcv_saddr) != 4);
6424 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6425 struct bpf_sock_ops_kern, sk),
6426 si->dst_reg, si->src_reg,
6427 offsetof(struct bpf_sock_ops_kern, sk));
6428 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6429 offsetof(struct sock_common,
6433 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
6434 offsetof(struct bpf_sock_ops, remote_ip6[3]):
6435 #if IS_ENABLED(CONFIG_IPV6)
6436 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6437 skc_v6_daddr.s6_addr32[0]) != 4);
6440 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
6441 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6442 struct bpf_sock_ops_kern, sk),
6443 si->dst_reg, si->src_reg,
6444 offsetof(struct bpf_sock_ops_kern, sk));
6445 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6446 offsetof(struct sock_common,
6447 skc_v6_daddr.s6_addr32[0]) +
6450 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6454 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
6455 offsetof(struct bpf_sock_ops, local_ip6[3]):
6456 #if IS_ENABLED(CONFIG_IPV6)
6457 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6458 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
6461 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
6462 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6463 struct bpf_sock_ops_kern, sk),
6464 si->dst_reg, si->src_reg,
6465 offsetof(struct bpf_sock_ops_kern, sk));
6466 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6467 offsetof(struct sock_common,
6468 skc_v6_rcv_saddr.s6_addr32[0]) +
6471 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6475 case offsetof(struct bpf_sock_ops, remote_port):
6476 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
6478 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6479 struct bpf_sock_ops_kern, sk),
6480 si->dst_reg, si->src_reg,
6481 offsetof(struct bpf_sock_ops_kern, sk));
6482 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6483 offsetof(struct sock_common, skc_dport));
6484 #ifndef __BIG_ENDIAN_BITFIELD
6485 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
6489 case offsetof(struct bpf_sock_ops, local_port):
6490 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
6492 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6493 struct bpf_sock_ops_kern, sk),
6494 si->dst_reg, si->src_reg,
6495 offsetof(struct bpf_sock_ops_kern, sk));
6496 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6497 offsetof(struct sock_common, skc_num));
6500 case offsetof(struct bpf_sock_ops, is_fullsock):
6501 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6502 struct bpf_sock_ops_kern,
6504 si->dst_reg, si->src_reg,
6505 offsetof(struct bpf_sock_ops_kern,
6509 case offsetof(struct bpf_sock_ops, state):
6510 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_state) != 1);
6512 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6513 struct bpf_sock_ops_kern, sk),
6514 si->dst_reg, si->src_reg,
6515 offsetof(struct bpf_sock_ops_kern, sk));
6516 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
6517 offsetof(struct sock_common, skc_state));
6520 case offsetof(struct bpf_sock_ops, rtt_min):
6521 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, rtt_min) !=
6522 sizeof(struct minmax));
6523 BUILD_BUG_ON(sizeof(struct minmax) <
6524 sizeof(struct minmax_sample));
6526 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6527 struct bpf_sock_ops_kern, sk),
6528 si->dst_reg, si->src_reg,
6529 offsetof(struct bpf_sock_ops_kern, sk));
6530 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6531 offsetof(struct tcp_sock, rtt_min) +
6532 FIELD_SIZEOF(struct minmax_sample, t));
6535 /* Helper macro for adding read access to tcp_sock or sock fields. */
6536 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
6538 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) > \
6539 FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD)); \
6540 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6541 struct bpf_sock_ops_kern, \
6543 si->dst_reg, si->src_reg, \
6544 offsetof(struct bpf_sock_ops_kern, \
6546 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 2); \
6547 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6548 struct bpf_sock_ops_kern, sk),\
6549 si->dst_reg, si->src_reg, \
6550 offsetof(struct bpf_sock_ops_kern, sk));\
6551 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
6553 si->dst_reg, si->dst_reg, \
6554 offsetof(OBJ, OBJ_FIELD)); \
6557 /* Helper macro for adding write access to tcp_sock or sock fields.
6558 * The macro is called with two registers, dst_reg which contains a pointer
6559 * to ctx (context) and src_reg which contains the value that should be
6560 * stored. However, we need an additional register since we cannot overwrite
6561 * dst_reg because it may be used later in the program.
6562 * Instead we "borrow" one of the other register. We first save its value
6563 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
6564 * it at the end of the macro.
6566 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
6568 int reg = BPF_REG_9; \
6569 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) > \
6570 FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD)); \
6571 if (si->dst_reg == reg || si->src_reg == reg) \
6573 if (si->dst_reg == reg || si->src_reg == reg) \
6575 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
6576 offsetof(struct bpf_sock_ops_kern, \
6578 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6579 struct bpf_sock_ops_kern, \
6582 offsetof(struct bpf_sock_ops_kern, \
6584 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
6585 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6586 struct bpf_sock_ops_kern, sk),\
6588 offsetof(struct bpf_sock_ops_kern, sk));\
6589 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \
6591 offsetof(OBJ, OBJ_FIELD)); \
6592 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
6593 offsetof(struct bpf_sock_ops_kern, \
6597 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
6599 if (TYPE == BPF_WRITE) \
6600 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
6602 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
6605 case offsetof(struct bpf_sock_ops, snd_cwnd):
6606 SOCK_OPS_GET_FIELD(snd_cwnd, snd_cwnd, struct tcp_sock);
6609 case offsetof(struct bpf_sock_ops, srtt_us):
6610 SOCK_OPS_GET_FIELD(srtt_us, srtt_us, struct tcp_sock);
6613 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
6614 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
6618 case offsetof(struct bpf_sock_ops, snd_ssthresh):
6619 SOCK_OPS_GET_FIELD(snd_ssthresh, snd_ssthresh, struct tcp_sock);
6622 case offsetof(struct bpf_sock_ops, rcv_nxt):
6623 SOCK_OPS_GET_FIELD(rcv_nxt, rcv_nxt, struct tcp_sock);
6626 case offsetof(struct bpf_sock_ops, snd_nxt):
6627 SOCK_OPS_GET_FIELD(snd_nxt, snd_nxt, struct tcp_sock);
6630 case offsetof(struct bpf_sock_ops, snd_una):
6631 SOCK_OPS_GET_FIELD(snd_una, snd_una, struct tcp_sock);
6634 case offsetof(struct bpf_sock_ops, mss_cache):
6635 SOCK_OPS_GET_FIELD(mss_cache, mss_cache, struct tcp_sock);
6638 case offsetof(struct bpf_sock_ops, ecn_flags):
6639 SOCK_OPS_GET_FIELD(ecn_flags, ecn_flags, struct tcp_sock);
6642 case offsetof(struct bpf_sock_ops, rate_delivered):
6643 SOCK_OPS_GET_FIELD(rate_delivered, rate_delivered,
6647 case offsetof(struct bpf_sock_ops, rate_interval_us):
6648 SOCK_OPS_GET_FIELD(rate_interval_us, rate_interval_us,
6652 case offsetof(struct bpf_sock_ops, packets_out):
6653 SOCK_OPS_GET_FIELD(packets_out, packets_out, struct tcp_sock);
6656 case offsetof(struct bpf_sock_ops, retrans_out):
6657 SOCK_OPS_GET_FIELD(retrans_out, retrans_out, struct tcp_sock);
6660 case offsetof(struct bpf_sock_ops, total_retrans):
6661 SOCK_OPS_GET_FIELD(total_retrans, total_retrans,
6665 case offsetof(struct bpf_sock_ops, segs_in):
6666 SOCK_OPS_GET_FIELD(segs_in, segs_in, struct tcp_sock);
6669 case offsetof(struct bpf_sock_ops, data_segs_in):
6670 SOCK_OPS_GET_FIELD(data_segs_in, data_segs_in, struct tcp_sock);
6673 case offsetof(struct bpf_sock_ops, segs_out):
6674 SOCK_OPS_GET_FIELD(segs_out, segs_out, struct tcp_sock);
6677 case offsetof(struct bpf_sock_ops, data_segs_out):
6678 SOCK_OPS_GET_FIELD(data_segs_out, data_segs_out,
6682 case offsetof(struct bpf_sock_ops, lost_out):
6683 SOCK_OPS_GET_FIELD(lost_out, lost_out, struct tcp_sock);
6686 case offsetof(struct bpf_sock_ops, sacked_out):
6687 SOCK_OPS_GET_FIELD(sacked_out, sacked_out, struct tcp_sock);
6690 case offsetof(struct bpf_sock_ops, sk_txhash):
6691 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
6695 case offsetof(struct bpf_sock_ops, bytes_received):
6696 SOCK_OPS_GET_FIELD(bytes_received, bytes_received,
6700 case offsetof(struct bpf_sock_ops, bytes_acked):
6701 SOCK_OPS_GET_FIELD(bytes_acked, bytes_acked, struct tcp_sock);
6705 return insn - insn_buf;
6708 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
6709 const struct bpf_insn *si,
6710 struct bpf_insn *insn_buf,
6711 struct bpf_prog *prog, u32 *target_size)
6713 struct bpf_insn *insn = insn_buf;
6717 case offsetof(struct __sk_buff, data_end):
6719 off -= offsetof(struct __sk_buff, data_end);
6720 off += offsetof(struct sk_buff, cb);
6721 off += offsetof(struct tcp_skb_cb, bpf.data_end);
6722 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
6726 return bpf_convert_ctx_access(type, si, insn_buf, prog,
6730 return insn - insn_buf;
6733 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
6734 const struct bpf_insn *si,
6735 struct bpf_insn *insn_buf,
6736 struct bpf_prog *prog, u32 *target_size)
6738 struct bpf_insn *insn = insn_buf;
6739 #if IS_ENABLED(CONFIG_IPV6)
6744 case offsetof(struct sk_msg_md, data):
6745 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_buff, data),
6746 si->dst_reg, si->src_reg,
6747 offsetof(struct sk_msg_buff, data));
6749 case offsetof(struct sk_msg_md, data_end):
6750 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_buff, data_end),
6751 si->dst_reg, si->src_reg,
6752 offsetof(struct sk_msg_buff, data_end));
6754 case offsetof(struct sk_msg_md, family):
6755 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
6757 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6758 struct sk_msg_buff, sk),
6759 si->dst_reg, si->src_reg,
6760 offsetof(struct sk_msg_buff, sk));
6761 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6762 offsetof(struct sock_common, skc_family));
6765 case offsetof(struct sk_msg_md, remote_ip4):
6766 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
6768 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6769 struct sk_msg_buff, sk),
6770 si->dst_reg, si->src_reg,
6771 offsetof(struct sk_msg_buff, sk));
6772 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6773 offsetof(struct sock_common, skc_daddr));
6776 case offsetof(struct sk_msg_md, local_ip4):
6777 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6778 skc_rcv_saddr) != 4);
6780 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6781 struct sk_msg_buff, sk),
6782 si->dst_reg, si->src_reg,
6783 offsetof(struct sk_msg_buff, sk));
6784 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6785 offsetof(struct sock_common,
6789 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
6790 offsetof(struct sk_msg_md, remote_ip6[3]):
6791 #if IS_ENABLED(CONFIG_IPV6)
6792 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6793 skc_v6_daddr.s6_addr32[0]) != 4);
6796 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
6797 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6798 struct sk_msg_buff, sk),
6799 si->dst_reg, si->src_reg,
6800 offsetof(struct sk_msg_buff, sk));
6801 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6802 offsetof(struct sock_common,
6803 skc_v6_daddr.s6_addr32[0]) +
6806 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6810 case offsetof(struct sk_msg_md, local_ip6[0]) ...
6811 offsetof(struct sk_msg_md, local_ip6[3]):
6812 #if IS_ENABLED(CONFIG_IPV6)
6813 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6814 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
6817 off -= offsetof(struct sk_msg_md, local_ip6[0]);
6818 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6819 struct sk_msg_buff, sk),
6820 si->dst_reg, si->src_reg,
6821 offsetof(struct sk_msg_buff, sk));
6822 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6823 offsetof(struct sock_common,
6824 skc_v6_rcv_saddr.s6_addr32[0]) +
6827 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6831 case offsetof(struct sk_msg_md, remote_port):
6832 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
6834 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6835 struct sk_msg_buff, sk),
6836 si->dst_reg, si->src_reg,
6837 offsetof(struct sk_msg_buff, sk));
6838 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6839 offsetof(struct sock_common, skc_dport));
6840 #ifndef __BIG_ENDIAN_BITFIELD
6841 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
6845 case offsetof(struct sk_msg_md, local_port):
6846 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
6848 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6849 struct sk_msg_buff, sk),
6850 si->dst_reg, si->src_reg,
6851 offsetof(struct sk_msg_buff, sk));
6852 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6853 offsetof(struct sock_common, skc_num));
6857 return insn - insn_buf;
6860 const struct bpf_verifier_ops sk_filter_verifier_ops = {
6861 .get_func_proto = sk_filter_func_proto,
6862 .is_valid_access = sk_filter_is_valid_access,
6863 .convert_ctx_access = bpf_convert_ctx_access,
6864 .gen_ld_abs = bpf_gen_ld_abs,
6867 const struct bpf_prog_ops sk_filter_prog_ops = {
6868 .test_run = bpf_prog_test_run_skb,
6871 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
6872 .get_func_proto = tc_cls_act_func_proto,
6873 .is_valid_access = tc_cls_act_is_valid_access,
6874 .convert_ctx_access = tc_cls_act_convert_ctx_access,
6875 .gen_prologue = tc_cls_act_prologue,
6876 .gen_ld_abs = bpf_gen_ld_abs,
6879 const struct bpf_prog_ops tc_cls_act_prog_ops = {
6880 .test_run = bpf_prog_test_run_skb,
6883 const struct bpf_verifier_ops xdp_verifier_ops = {
6884 .get_func_proto = xdp_func_proto,
6885 .is_valid_access = xdp_is_valid_access,
6886 .convert_ctx_access = xdp_convert_ctx_access,
6889 const struct bpf_prog_ops xdp_prog_ops = {
6890 .test_run = bpf_prog_test_run_xdp,
6893 const struct bpf_verifier_ops cg_skb_verifier_ops = {
6894 .get_func_proto = cg_skb_func_proto,
6895 .is_valid_access = sk_filter_is_valid_access,
6896 .convert_ctx_access = bpf_convert_ctx_access,
6899 const struct bpf_prog_ops cg_skb_prog_ops = {
6900 .test_run = bpf_prog_test_run_skb,
6903 const struct bpf_verifier_ops lwt_in_verifier_ops = {
6904 .get_func_proto = lwt_in_func_proto,
6905 .is_valid_access = lwt_is_valid_access,
6906 .convert_ctx_access = bpf_convert_ctx_access,
6909 const struct bpf_prog_ops lwt_in_prog_ops = {
6910 .test_run = bpf_prog_test_run_skb,
6913 const struct bpf_verifier_ops lwt_out_verifier_ops = {
6914 .get_func_proto = lwt_out_func_proto,
6915 .is_valid_access = lwt_is_valid_access,
6916 .convert_ctx_access = bpf_convert_ctx_access,
6919 const struct bpf_prog_ops lwt_out_prog_ops = {
6920 .test_run = bpf_prog_test_run_skb,
6923 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
6924 .get_func_proto = lwt_xmit_func_proto,
6925 .is_valid_access = lwt_is_valid_access,
6926 .convert_ctx_access = bpf_convert_ctx_access,
6927 .gen_prologue = tc_cls_act_prologue,
6930 const struct bpf_prog_ops lwt_xmit_prog_ops = {
6931 .test_run = bpf_prog_test_run_skb,
6934 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
6935 .get_func_proto = lwt_seg6local_func_proto,
6936 .is_valid_access = lwt_is_valid_access,
6937 .convert_ctx_access = bpf_convert_ctx_access,
6940 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
6941 .test_run = bpf_prog_test_run_skb,
6944 const struct bpf_verifier_ops cg_sock_verifier_ops = {
6945 .get_func_proto = sock_filter_func_proto,
6946 .is_valid_access = sock_filter_is_valid_access,
6947 .convert_ctx_access = sock_filter_convert_ctx_access,
6950 const struct bpf_prog_ops cg_sock_prog_ops = {
6953 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
6954 .get_func_proto = sock_addr_func_proto,
6955 .is_valid_access = sock_addr_is_valid_access,
6956 .convert_ctx_access = sock_addr_convert_ctx_access,
6959 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
6962 const struct bpf_verifier_ops sock_ops_verifier_ops = {
6963 .get_func_proto = sock_ops_func_proto,
6964 .is_valid_access = sock_ops_is_valid_access,
6965 .convert_ctx_access = sock_ops_convert_ctx_access,
6968 const struct bpf_prog_ops sock_ops_prog_ops = {
6971 const struct bpf_verifier_ops sk_skb_verifier_ops = {
6972 .get_func_proto = sk_skb_func_proto,
6973 .is_valid_access = sk_skb_is_valid_access,
6974 .convert_ctx_access = sk_skb_convert_ctx_access,
6975 .gen_prologue = sk_skb_prologue,
6978 const struct bpf_prog_ops sk_skb_prog_ops = {
6981 const struct bpf_verifier_ops sk_msg_verifier_ops = {
6982 .get_func_proto = sk_msg_func_proto,
6983 .is_valid_access = sk_msg_is_valid_access,
6984 .convert_ctx_access = sk_msg_convert_ctx_access,
6987 const struct bpf_prog_ops sk_msg_prog_ops = {
6990 int sk_detach_filter(struct sock *sk)
6993 struct sk_filter *filter;
6995 if (sock_flag(sk, SOCK_FILTER_LOCKED))
6998 filter = rcu_dereference_protected(sk->sk_filter,
6999 lockdep_sock_is_held(sk));
7001 RCU_INIT_POINTER(sk->sk_filter, NULL);
7002 sk_filter_uncharge(sk, filter);
7008 EXPORT_SYMBOL_GPL(sk_detach_filter);
7010 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
7013 struct sock_fprog_kern *fprog;
7014 struct sk_filter *filter;
7018 filter = rcu_dereference_protected(sk->sk_filter,
7019 lockdep_sock_is_held(sk));
7023 /* We're copying the filter that has been originally attached,
7024 * so no conversion/decode needed anymore. eBPF programs that
7025 * have no original program cannot be dumped through this.
7028 fprog = filter->prog->orig_prog;
7034 /* User space only enquires number of filter blocks. */
7038 if (len < fprog->len)
7042 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
7045 /* Instead of bytes, the API requests to return the number
7055 struct sk_reuseport_kern {
7056 struct sk_buff *skb;
7058 struct sock *selected_sk;
7065 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
7066 struct sock_reuseport *reuse,
7067 struct sock *sk, struct sk_buff *skb,
7070 reuse_kern->skb = skb;
7071 reuse_kern->sk = sk;
7072 reuse_kern->selected_sk = NULL;
7073 reuse_kern->data_end = skb->data + skb_headlen(skb);
7074 reuse_kern->hash = hash;
7075 reuse_kern->reuseport_id = reuse->reuseport_id;
7076 reuse_kern->bind_inany = reuse->bind_inany;
7079 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
7080 struct bpf_prog *prog, struct sk_buff *skb,
7083 struct sk_reuseport_kern reuse_kern;
7084 enum sk_action action;
7086 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, hash);
7087 action = BPF_PROG_RUN(prog, &reuse_kern);
7089 if (action == SK_PASS)
7090 return reuse_kern.selected_sk;
7092 return ERR_PTR(-ECONNREFUSED);
7095 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
7096 struct bpf_map *, map, void *, key, u32, flags)
7098 struct sock_reuseport *reuse;
7099 struct sock *selected_sk;
7101 selected_sk = map->ops->map_lookup_elem(map, key);
7105 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
7107 /* selected_sk is unhashed (e.g. by close()) after the
7108 * above map_lookup_elem(). Treat selected_sk has already
7109 * been removed from the map.
7113 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
7116 if (unlikely(!reuse_kern->reuseport_id))
7117 /* There is a small race between adding the
7118 * sk to the map and setting the
7119 * reuse_kern->reuseport_id.
7120 * Treat it as the sk has not been added to
7125 sk = reuse_kern->sk;
7126 if (sk->sk_protocol != selected_sk->sk_protocol)
7128 else if (sk->sk_family != selected_sk->sk_family)
7129 return -EAFNOSUPPORT;
7131 /* Catch all. Likely bound to a different sockaddr. */
7135 reuse_kern->selected_sk = selected_sk;
7140 static const struct bpf_func_proto sk_select_reuseport_proto = {
7141 .func = sk_select_reuseport,
7143 .ret_type = RET_INTEGER,
7144 .arg1_type = ARG_PTR_TO_CTX,
7145 .arg2_type = ARG_CONST_MAP_PTR,
7146 .arg3_type = ARG_PTR_TO_MAP_KEY,
7147 .arg4_type = ARG_ANYTHING,
7150 BPF_CALL_4(sk_reuseport_load_bytes,
7151 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
7152 void *, to, u32, len)
7154 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
7157 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
7158 .func = sk_reuseport_load_bytes,
7160 .ret_type = RET_INTEGER,
7161 .arg1_type = ARG_PTR_TO_CTX,
7162 .arg2_type = ARG_ANYTHING,
7163 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
7164 .arg4_type = ARG_CONST_SIZE,
7167 BPF_CALL_5(sk_reuseport_load_bytes_relative,
7168 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
7169 void *, to, u32, len, u32, start_header)
7171 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
7175 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
7176 .func = sk_reuseport_load_bytes_relative,
7178 .ret_type = RET_INTEGER,
7179 .arg1_type = ARG_PTR_TO_CTX,
7180 .arg2_type = ARG_ANYTHING,
7181 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
7182 .arg4_type = ARG_CONST_SIZE,
7183 .arg5_type = ARG_ANYTHING,
7186 static const struct bpf_func_proto *
7187 sk_reuseport_func_proto(enum bpf_func_id func_id,
7188 const struct bpf_prog *prog)
7191 case BPF_FUNC_sk_select_reuseport:
7192 return &sk_select_reuseport_proto;
7193 case BPF_FUNC_skb_load_bytes:
7194 return &sk_reuseport_load_bytes_proto;
7195 case BPF_FUNC_skb_load_bytes_relative:
7196 return &sk_reuseport_load_bytes_relative_proto;
7198 return bpf_base_func_proto(func_id);
7203 sk_reuseport_is_valid_access(int off, int size,
7204 enum bpf_access_type type,
7205 const struct bpf_prog *prog,
7206 struct bpf_insn_access_aux *info)
7208 const u32 size_default = sizeof(__u32);
7210 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
7211 off % size || type != BPF_READ)
7215 case offsetof(struct sk_reuseport_md, data):
7216 info->reg_type = PTR_TO_PACKET;
7217 return size == sizeof(__u64);
7219 case offsetof(struct sk_reuseport_md, data_end):
7220 info->reg_type = PTR_TO_PACKET_END;
7221 return size == sizeof(__u64);
7223 case offsetof(struct sk_reuseport_md, hash):
7224 return size == size_default;
7226 /* Fields that allow narrowing */
7227 case offsetof(struct sk_reuseport_md, eth_protocol):
7228 if (size < FIELD_SIZEOF(struct sk_buff, protocol))
7231 case offsetof(struct sk_reuseport_md, ip_protocol):
7232 case offsetof(struct sk_reuseport_md, bind_inany):
7233 case offsetof(struct sk_reuseport_md, len):
7234 bpf_ctx_record_field_size(info, size_default);
7235 return bpf_ctx_narrow_access_ok(off, size, size_default);
7242 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
7243 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
7244 si->dst_reg, si->src_reg, \
7245 bpf_target_off(struct sk_reuseport_kern, F, \
7246 FIELD_SIZEOF(struct sk_reuseport_kern, F), \
7250 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
7251 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
7256 #define SK_REUSEPORT_LOAD_SK_FIELD_SIZE_OFF(SK_FIELD, BPF_SIZE, EXTRA_OFF) \
7257 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(struct sk_reuseport_kern, \
7260 SK_FIELD, BPF_SIZE, EXTRA_OFF)
7262 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
7263 const struct bpf_insn *si,
7264 struct bpf_insn *insn_buf,
7265 struct bpf_prog *prog,
7268 struct bpf_insn *insn = insn_buf;
7271 case offsetof(struct sk_reuseport_md, data):
7272 SK_REUSEPORT_LOAD_SKB_FIELD(data);
7275 case offsetof(struct sk_reuseport_md, len):
7276 SK_REUSEPORT_LOAD_SKB_FIELD(len);
7279 case offsetof(struct sk_reuseport_md, eth_protocol):
7280 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
7283 case offsetof(struct sk_reuseport_md, ip_protocol):
7284 BUILD_BUG_ON(hweight_long(SK_FL_PROTO_MASK) != BITS_PER_BYTE);
7285 SK_REUSEPORT_LOAD_SK_FIELD_SIZE_OFF(__sk_flags_offset,
7287 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
7288 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg,
7290 /* SK_FL_PROTO_MASK and SK_FL_PROTO_SHIFT are endian
7291 * aware. No further narrowing or masking is needed.
7296 case offsetof(struct sk_reuseport_md, data_end):
7297 SK_REUSEPORT_LOAD_FIELD(data_end);
7300 case offsetof(struct sk_reuseport_md, hash):
7301 SK_REUSEPORT_LOAD_FIELD(hash);
7304 case offsetof(struct sk_reuseport_md, bind_inany):
7305 SK_REUSEPORT_LOAD_FIELD(bind_inany);
7309 return insn - insn_buf;
7312 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
7313 .get_func_proto = sk_reuseport_func_proto,
7314 .is_valid_access = sk_reuseport_is_valid_access,
7315 .convert_ctx_access = sk_reuseport_convert_ctx_access,
7318 const struct bpf_prog_ops sk_reuseport_prog_ops = {
7320 #endif /* CONFIG_INET */