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);
1456 static int __reuseport_attach_prog(struct bpf_prog *prog, struct sock *sk)
1458 struct bpf_prog *old_prog;
1461 if (bpf_prog_size(prog->len) > sysctl_optmem_max)
1464 if (sk_unhashed(sk) && sk->sk_reuseport) {
1465 err = reuseport_alloc(sk);
1468 } else if (!rcu_access_pointer(sk->sk_reuseport_cb)) {
1469 /* The socket wasn't bound with SO_REUSEPORT */
1473 old_prog = reuseport_attach_prog(sk, prog);
1475 bpf_prog_destroy(old_prog);
1481 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1483 unsigned int fsize = bpf_classic_proglen(fprog);
1484 struct bpf_prog *prog;
1487 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1488 return ERR_PTR(-EPERM);
1490 /* Make sure new filter is there and in the right amounts. */
1491 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1492 return ERR_PTR(-EINVAL);
1494 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1496 return ERR_PTR(-ENOMEM);
1498 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1499 __bpf_prog_free(prog);
1500 return ERR_PTR(-EFAULT);
1503 prog->len = fprog->len;
1505 err = bpf_prog_store_orig_filter(prog, fprog);
1507 __bpf_prog_free(prog);
1508 return ERR_PTR(-ENOMEM);
1511 /* bpf_prepare_filter() already takes care of freeing
1512 * memory in case something goes wrong.
1514 return bpf_prepare_filter(prog, NULL);
1518 * sk_attach_filter - attach a socket filter
1519 * @fprog: the filter program
1520 * @sk: the socket to use
1522 * Attach the user's filter code. We first run some sanity checks on
1523 * it to make sure it does not explode on us later. If an error
1524 * occurs or there is insufficient memory for the filter a negative
1525 * errno code is returned. On success the return is zero.
1527 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1529 struct bpf_prog *prog = __get_filter(fprog, sk);
1533 return PTR_ERR(prog);
1535 err = __sk_attach_prog(prog, sk);
1537 __bpf_prog_release(prog);
1543 EXPORT_SYMBOL_GPL(sk_attach_filter);
1545 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1547 struct bpf_prog *prog = __get_filter(fprog, sk);
1551 return PTR_ERR(prog);
1553 err = __reuseport_attach_prog(prog, sk);
1555 __bpf_prog_release(prog);
1562 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1564 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1565 return ERR_PTR(-EPERM);
1567 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1570 int sk_attach_bpf(u32 ufd, struct sock *sk)
1572 struct bpf_prog *prog = __get_bpf(ufd, sk);
1576 return PTR_ERR(prog);
1578 err = __sk_attach_prog(prog, sk);
1587 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1589 struct bpf_prog *prog = __get_bpf(ufd, sk);
1593 return PTR_ERR(prog);
1595 err = __reuseport_attach_prog(prog, sk);
1604 struct bpf_scratchpad {
1606 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1607 u8 buff[MAX_BPF_STACK];
1611 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1613 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1614 unsigned int write_len)
1616 return skb_ensure_writable(skb, write_len);
1619 static inline int bpf_try_make_writable(struct sk_buff *skb,
1620 unsigned int write_len)
1622 int err = __bpf_try_make_writable(skb, write_len);
1624 bpf_compute_data_pointers(skb);
1628 static int bpf_try_make_head_writable(struct sk_buff *skb)
1630 return bpf_try_make_writable(skb, skb_headlen(skb));
1633 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1635 if (skb_at_tc_ingress(skb))
1636 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1639 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1641 if (skb_at_tc_ingress(skb))
1642 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1645 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1646 const void *, from, u32, len, u64, flags)
1650 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1652 if (unlikely(offset > 0xffff))
1654 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1657 ptr = skb->data + offset;
1658 if (flags & BPF_F_RECOMPUTE_CSUM)
1659 __skb_postpull_rcsum(skb, ptr, len, offset);
1661 memcpy(ptr, from, len);
1663 if (flags & BPF_F_RECOMPUTE_CSUM)
1664 __skb_postpush_rcsum(skb, ptr, len, offset);
1665 if (flags & BPF_F_INVALIDATE_HASH)
1666 skb_clear_hash(skb);
1671 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1672 .func = bpf_skb_store_bytes,
1674 .ret_type = RET_INTEGER,
1675 .arg1_type = ARG_PTR_TO_CTX,
1676 .arg2_type = ARG_ANYTHING,
1677 .arg3_type = ARG_PTR_TO_MEM,
1678 .arg4_type = ARG_CONST_SIZE,
1679 .arg5_type = ARG_ANYTHING,
1682 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1683 void *, to, u32, len)
1687 if (unlikely(offset > 0xffff))
1690 ptr = skb_header_pointer(skb, offset, len, to);
1694 memcpy(to, ptr, len);
1702 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1703 .func = bpf_skb_load_bytes,
1705 .ret_type = RET_INTEGER,
1706 .arg1_type = ARG_PTR_TO_CTX,
1707 .arg2_type = ARG_ANYTHING,
1708 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1709 .arg4_type = ARG_CONST_SIZE,
1712 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1713 u32, offset, void *, to, u32, len, u32, start_header)
1715 u8 *end = skb_tail_pointer(skb);
1716 u8 *net = skb_network_header(skb);
1717 u8 *mac = skb_mac_header(skb);
1720 if (unlikely(offset > 0xffff || len > (end - mac)))
1723 switch (start_header) {
1724 case BPF_HDR_START_MAC:
1727 case BPF_HDR_START_NET:
1734 if (likely(ptr >= mac && ptr + len <= end)) {
1735 memcpy(to, ptr, len);
1744 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1745 .func = bpf_skb_load_bytes_relative,
1747 .ret_type = RET_INTEGER,
1748 .arg1_type = ARG_PTR_TO_CTX,
1749 .arg2_type = ARG_ANYTHING,
1750 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1751 .arg4_type = ARG_CONST_SIZE,
1752 .arg5_type = ARG_ANYTHING,
1755 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1757 /* Idea is the following: should the needed direct read/write
1758 * test fail during runtime, we can pull in more data and redo
1759 * again, since implicitly, we invalidate previous checks here.
1761 * Or, since we know how much we need to make read/writeable,
1762 * this can be done once at the program beginning for direct
1763 * access case. By this we overcome limitations of only current
1764 * headroom being accessible.
1766 return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1769 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1770 .func = bpf_skb_pull_data,
1772 .ret_type = RET_INTEGER,
1773 .arg1_type = ARG_PTR_TO_CTX,
1774 .arg2_type = ARG_ANYTHING,
1777 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1778 unsigned int write_len)
1780 int err = __bpf_try_make_writable(skb, write_len);
1782 bpf_compute_data_end_sk_skb(skb);
1786 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1788 /* Idea is the following: should the needed direct read/write
1789 * test fail during runtime, we can pull in more data and redo
1790 * again, since implicitly, we invalidate previous checks here.
1792 * Or, since we know how much we need to make read/writeable,
1793 * this can be done once at the program beginning for direct
1794 * access case. By this we overcome limitations of only current
1795 * headroom being accessible.
1797 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1800 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1801 .func = sk_skb_pull_data,
1803 .ret_type = RET_INTEGER,
1804 .arg1_type = ARG_PTR_TO_CTX,
1805 .arg2_type = ARG_ANYTHING,
1808 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1809 u64, from, u64, to, u64, flags)
1813 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1815 if (unlikely(offset > 0xffff || offset & 1))
1817 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1820 ptr = (__sum16 *)(skb->data + offset);
1821 switch (flags & BPF_F_HDR_FIELD_MASK) {
1823 if (unlikely(from != 0))
1826 csum_replace_by_diff(ptr, to);
1829 csum_replace2(ptr, from, to);
1832 csum_replace4(ptr, from, to);
1841 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1842 .func = bpf_l3_csum_replace,
1844 .ret_type = RET_INTEGER,
1845 .arg1_type = ARG_PTR_TO_CTX,
1846 .arg2_type = ARG_ANYTHING,
1847 .arg3_type = ARG_ANYTHING,
1848 .arg4_type = ARG_ANYTHING,
1849 .arg5_type = ARG_ANYTHING,
1852 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1853 u64, from, u64, to, u64, flags)
1855 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1856 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1857 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1860 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1861 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1863 if (unlikely(offset > 0xffff || offset & 1))
1865 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1868 ptr = (__sum16 *)(skb->data + offset);
1869 if (is_mmzero && !do_mforce && !*ptr)
1872 switch (flags & BPF_F_HDR_FIELD_MASK) {
1874 if (unlikely(from != 0))
1877 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1880 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1883 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1889 if (is_mmzero && !*ptr)
1890 *ptr = CSUM_MANGLED_0;
1894 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1895 .func = bpf_l4_csum_replace,
1897 .ret_type = RET_INTEGER,
1898 .arg1_type = ARG_PTR_TO_CTX,
1899 .arg2_type = ARG_ANYTHING,
1900 .arg3_type = ARG_ANYTHING,
1901 .arg4_type = ARG_ANYTHING,
1902 .arg5_type = ARG_ANYTHING,
1905 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1906 __be32 *, to, u32, to_size, __wsum, seed)
1908 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1909 u32 diff_size = from_size + to_size;
1912 /* This is quite flexible, some examples:
1914 * from_size == 0, to_size > 0, seed := csum --> pushing data
1915 * from_size > 0, to_size == 0, seed := csum --> pulling data
1916 * from_size > 0, to_size > 0, seed := 0 --> diffing data
1918 * Even for diffing, from_size and to_size don't need to be equal.
1920 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
1921 diff_size > sizeof(sp->diff)))
1924 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
1925 sp->diff[j] = ~from[i];
1926 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
1927 sp->diff[j] = to[i];
1929 return csum_partial(sp->diff, diff_size, seed);
1932 static const struct bpf_func_proto bpf_csum_diff_proto = {
1933 .func = bpf_csum_diff,
1936 .ret_type = RET_INTEGER,
1937 .arg1_type = ARG_PTR_TO_MEM_OR_NULL,
1938 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
1939 .arg3_type = ARG_PTR_TO_MEM_OR_NULL,
1940 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
1941 .arg5_type = ARG_ANYTHING,
1944 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
1946 /* The interface is to be used in combination with bpf_csum_diff()
1947 * for direct packet writes. csum rotation for alignment as well
1948 * as emulating csum_sub() can be done from the eBPF program.
1950 if (skb->ip_summed == CHECKSUM_COMPLETE)
1951 return (skb->csum = csum_add(skb->csum, csum));
1956 static const struct bpf_func_proto bpf_csum_update_proto = {
1957 .func = bpf_csum_update,
1959 .ret_type = RET_INTEGER,
1960 .arg1_type = ARG_PTR_TO_CTX,
1961 .arg2_type = ARG_ANYTHING,
1964 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
1966 return dev_forward_skb(dev, skb);
1969 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
1970 struct sk_buff *skb)
1972 int ret = ____dev_forward_skb(dev, skb);
1976 ret = netif_rx(skb);
1982 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
1986 if (unlikely(__this_cpu_read(xmit_recursion) > XMIT_RECURSION_LIMIT)) {
1987 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
1994 __this_cpu_inc(xmit_recursion);
1995 ret = dev_queue_xmit(skb);
1996 __this_cpu_dec(xmit_recursion);
2001 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2004 /* skb->mac_len is not set on normal egress */
2005 unsigned int mlen = skb->network_header - skb->mac_header;
2007 __skb_pull(skb, mlen);
2009 /* At ingress, the mac header has already been pulled once.
2010 * At egress, skb_pospull_rcsum has to be done in case that
2011 * the skb is originated from ingress (i.e. a forwarded skb)
2012 * to ensure that rcsum starts at net header.
2014 if (!skb_at_tc_ingress(skb))
2015 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2016 skb_pop_mac_header(skb);
2017 skb_reset_mac_len(skb);
2018 return flags & BPF_F_INGRESS ?
2019 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2022 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2025 /* Verify that a link layer header is carried */
2026 if (unlikely(skb->mac_header >= skb->network_header)) {
2031 bpf_push_mac_rcsum(skb);
2032 return flags & BPF_F_INGRESS ?
2033 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2036 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2039 if (dev_is_mac_header_xmit(dev))
2040 return __bpf_redirect_common(skb, dev, flags);
2042 return __bpf_redirect_no_mac(skb, dev, flags);
2045 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2047 struct net_device *dev;
2048 struct sk_buff *clone;
2051 if (unlikely(flags & ~(BPF_F_INGRESS)))
2054 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2058 clone = skb_clone(skb, GFP_ATOMIC);
2059 if (unlikely(!clone))
2062 /* For direct write, we need to keep the invariant that the skbs
2063 * we're dealing with need to be uncloned. Should uncloning fail
2064 * here, we need to free the just generated clone to unclone once
2067 ret = bpf_try_make_head_writable(skb);
2068 if (unlikely(ret)) {
2073 return __bpf_redirect(clone, dev, flags);
2076 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2077 .func = bpf_clone_redirect,
2079 .ret_type = RET_INTEGER,
2080 .arg1_type = ARG_PTR_TO_CTX,
2081 .arg2_type = ARG_ANYTHING,
2082 .arg3_type = ARG_ANYTHING,
2085 struct redirect_info {
2088 struct bpf_map *map;
2089 struct bpf_map *map_to_flush;
2090 unsigned long map_owner;
2093 static DEFINE_PER_CPU(struct redirect_info, redirect_info);
2095 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2097 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2099 if (unlikely(flags & ~(BPF_F_INGRESS)))
2102 ri->ifindex = ifindex;
2105 return TC_ACT_REDIRECT;
2108 int skb_do_redirect(struct sk_buff *skb)
2110 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2111 struct net_device *dev;
2113 dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->ifindex);
2115 if (unlikely(!dev)) {
2120 return __bpf_redirect(skb, dev, ri->flags);
2123 static const struct bpf_func_proto bpf_redirect_proto = {
2124 .func = bpf_redirect,
2126 .ret_type = RET_INTEGER,
2127 .arg1_type = ARG_ANYTHING,
2128 .arg2_type = ARG_ANYTHING,
2131 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
2132 struct bpf_map *, map, void *, key, u64, flags)
2134 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
2136 /* If user passes invalid input drop the packet. */
2137 if (unlikely(flags & ~(BPF_F_INGRESS)))
2140 tcb->bpf.flags = flags;
2141 tcb->bpf.sk_redir = __sock_hash_lookup_elem(map, key);
2142 if (!tcb->bpf.sk_redir)
2148 static const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
2149 .func = bpf_sk_redirect_hash,
2151 .ret_type = RET_INTEGER,
2152 .arg1_type = ARG_PTR_TO_CTX,
2153 .arg2_type = ARG_CONST_MAP_PTR,
2154 .arg3_type = ARG_PTR_TO_MAP_KEY,
2155 .arg4_type = ARG_ANYTHING,
2158 BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
2159 struct bpf_map *, map, u32, key, u64, flags)
2161 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
2163 /* If user passes invalid input drop the packet. */
2164 if (unlikely(flags & ~(BPF_F_INGRESS)))
2167 tcb->bpf.flags = flags;
2168 tcb->bpf.sk_redir = __sock_map_lookup_elem(map, key);
2169 if (!tcb->bpf.sk_redir)
2175 struct sock *do_sk_redirect_map(struct sk_buff *skb)
2177 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
2179 return tcb->bpf.sk_redir;
2182 static const struct bpf_func_proto bpf_sk_redirect_map_proto = {
2183 .func = bpf_sk_redirect_map,
2185 .ret_type = RET_INTEGER,
2186 .arg1_type = ARG_PTR_TO_CTX,
2187 .arg2_type = ARG_CONST_MAP_PTR,
2188 .arg3_type = ARG_ANYTHING,
2189 .arg4_type = ARG_ANYTHING,
2192 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg_buff *, msg,
2193 struct bpf_map *, map, void *, key, u64, flags)
2195 /* If user passes invalid input drop the packet. */
2196 if (unlikely(flags & ~(BPF_F_INGRESS)))
2200 msg->sk_redir = __sock_hash_lookup_elem(map, key);
2207 static const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
2208 .func = bpf_msg_redirect_hash,
2210 .ret_type = RET_INTEGER,
2211 .arg1_type = ARG_PTR_TO_CTX,
2212 .arg2_type = ARG_CONST_MAP_PTR,
2213 .arg3_type = ARG_PTR_TO_MAP_KEY,
2214 .arg4_type = ARG_ANYTHING,
2217 BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg_buff *, msg,
2218 struct bpf_map *, map, u32, key, u64, flags)
2220 /* If user passes invalid input drop the packet. */
2221 if (unlikely(flags & ~(BPF_F_INGRESS)))
2225 msg->sk_redir = __sock_map_lookup_elem(map, key);
2232 struct sock *do_msg_redirect_map(struct sk_msg_buff *msg)
2234 return msg->sk_redir;
2237 static const struct bpf_func_proto bpf_msg_redirect_map_proto = {
2238 .func = bpf_msg_redirect_map,
2240 .ret_type = RET_INTEGER,
2241 .arg1_type = ARG_PTR_TO_CTX,
2242 .arg2_type = ARG_CONST_MAP_PTR,
2243 .arg3_type = ARG_ANYTHING,
2244 .arg4_type = ARG_ANYTHING,
2247 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg_buff *, msg, u32, bytes)
2249 msg->apply_bytes = bytes;
2253 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2254 .func = bpf_msg_apply_bytes,
2256 .ret_type = RET_INTEGER,
2257 .arg1_type = ARG_PTR_TO_CTX,
2258 .arg2_type = ARG_ANYTHING,
2261 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg_buff *, msg, u32, bytes)
2263 msg->cork_bytes = bytes;
2267 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2268 .func = bpf_msg_cork_bytes,
2270 .ret_type = RET_INTEGER,
2271 .arg1_type = ARG_PTR_TO_CTX,
2272 .arg2_type = ARG_ANYTHING,
2275 BPF_CALL_4(bpf_msg_pull_data,
2276 struct sk_msg_buff *, msg, u32, start, u32, end, u64, flags)
2278 unsigned int len = 0, offset = 0, copy = 0;
2279 struct scatterlist *sg = msg->sg_data;
2280 int first_sg, last_sg, i, shift;
2281 unsigned char *p, *to, *from;
2282 int bytes = end - start;
2285 if (unlikely(flags || end <= start))
2288 /* First find the starting scatterlist element */
2293 if (start < offset + len)
2296 if (i == MAX_SKB_FRAGS)
2298 } while (i != msg->sg_end);
2300 if (unlikely(start >= offset + len))
2303 if (!msg->sg_copy[i] && bytes <= len)
2308 /* At this point we need to linearize multiple scatterlist
2309 * elements or a single shared page. Either way we need to
2310 * copy into a linear buffer exclusively owned by BPF. Then
2311 * place the buffer in the scatterlist and fixup the original
2312 * entries by removing the entries now in the linear buffer
2313 * and shifting the remaining entries. For now we do not try
2314 * to copy partial entries to avoid complexity of running out
2315 * of sg_entry slots. The downside is reading a single byte
2316 * will copy the entire sg entry.
2319 copy += sg[i].length;
2321 if (i == MAX_SKB_FRAGS)
2325 } while (i != msg->sg_end);
2328 if (unlikely(copy < end - start))
2331 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC, get_order(copy));
2332 if (unlikely(!page))
2334 p = page_address(page);
2339 from = sg_virt(&sg[i]);
2343 memcpy(to, from, len);
2346 put_page(sg_page(&sg[i]));
2349 if (i == MAX_SKB_FRAGS)
2351 } while (i != last_sg);
2353 sg[first_sg].length = copy;
2354 sg_set_page(&sg[first_sg], page, copy, 0);
2356 /* To repair sg ring we need to shift entries. If we only
2357 * had a single entry though we can just replace it and
2358 * be done. Otherwise walk the ring and shift the entries.
2360 shift = last_sg - first_sg - 1;
2368 if (i + shift >= MAX_SKB_FRAGS)
2369 move_from = i + shift - MAX_SKB_FRAGS;
2371 move_from = i + shift;
2373 if (move_from == msg->sg_end)
2376 sg[i] = sg[move_from];
2377 sg[move_from].length = 0;
2378 sg[move_from].page_link = 0;
2379 sg[move_from].offset = 0;
2382 if (i == MAX_SKB_FRAGS)
2385 msg->sg_end -= shift;
2386 if (msg->sg_end < 0)
2387 msg->sg_end += MAX_SKB_FRAGS;
2389 msg->data = sg_virt(&sg[i]) + start - offset;
2390 msg->data_end = msg->data + bytes;
2395 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2396 .func = bpf_msg_pull_data,
2398 .ret_type = RET_INTEGER,
2399 .arg1_type = ARG_PTR_TO_CTX,
2400 .arg2_type = ARG_ANYTHING,
2401 .arg3_type = ARG_ANYTHING,
2402 .arg4_type = ARG_ANYTHING,
2405 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
2407 return task_get_classid(skb);
2410 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
2411 .func = bpf_get_cgroup_classid,
2413 .ret_type = RET_INTEGER,
2414 .arg1_type = ARG_PTR_TO_CTX,
2417 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
2419 return dst_tclassid(skb);
2422 static const struct bpf_func_proto bpf_get_route_realm_proto = {
2423 .func = bpf_get_route_realm,
2425 .ret_type = RET_INTEGER,
2426 .arg1_type = ARG_PTR_TO_CTX,
2429 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
2431 /* If skb_clear_hash() was called due to mangling, we can
2432 * trigger SW recalculation here. Later access to hash
2433 * can then use the inline skb->hash via context directly
2434 * instead of calling this helper again.
2436 return skb_get_hash(skb);
2439 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
2440 .func = bpf_get_hash_recalc,
2442 .ret_type = RET_INTEGER,
2443 .arg1_type = ARG_PTR_TO_CTX,
2446 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
2448 /* After all direct packet write, this can be used once for
2449 * triggering a lazy recalc on next skb_get_hash() invocation.
2451 skb_clear_hash(skb);
2455 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
2456 .func = bpf_set_hash_invalid,
2458 .ret_type = RET_INTEGER,
2459 .arg1_type = ARG_PTR_TO_CTX,
2462 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
2464 /* Set user specified hash as L4(+), so that it gets returned
2465 * on skb_get_hash() call unless BPF prog later on triggers a
2468 __skb_set_sw_hash(skb, hash, true);
2472 static const struct bpf_func_proto bpf_set_hash_proto = {
2473 .func = bpf_set_hash,
2475 .ret_type = RET_INTEGER,
2476 .arg1_type = ARG_PTR_TO_CTX,
2477 .arg2_type = ARG_ANYTHING,
2480 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
2485 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
2486 vlan_proto != htons(ETH_P_8021AD)))
2487 vlan_proto = htons(ETH_P_8021Q);
2489 bpf_push_mac_rcsum(skb);
2490 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
2491 bpf_pull_mac_rcsum(skb);
2493 bpf_compute_data_pointers(skb);
2497 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
2498 .func = bpf_skb_vlan_push,
2500 .ret_type = RET_INTEGER,
2501 .arg1_type = ARG_PTR_TO_CTX,
2502 .arg2_type = ARG_ANYTHING,
2503 .arg3_type = ARG_ANYTHING,
2506 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
2510 bpf_push_mac_rcsum(skb);
2511 ret = skb_vlan_pop(skb);
2512 bpf_pull_mac_rcsum(skb);
2514 bpf_compute_data_pointers(skb);
2518 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
2519 .func = bpf_skb_vlan_pop,
2521 .ret_type = RET_INTEGER,
2522 .arg1_type = ARG_PTR_TO_CTX,
2525 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
2527 /* Caller already did skb_cow() with len as headroom,
2528 * so no need to do it here.
2531 memmove(skb->data, skb->data + len, off);
2532 memset(skb->data + off, 0, len);
2534 /* No skb_postpush_rcsum(skb, skb->data + off, len)
2535 * needed here as it does not change the skb->csum
2536 * result for checksum complete when summing over
2542 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
2544 /* skb_ensure_writable() is not needed here, as we're
2545 * already working on an uncloned skb.
2547 if (unlikely(!pskb_may_pull(skb, off + len)))
2550 skb_postpull_rcsum(skb, skb->data + off, len);
2551 memmove(skb->data + len, skb->data, off);
2552 __skb_pull(skb, len);
2557 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
2559 bool trans_same = skb->transport_header == skb->network_header;
2562 /* There's no need for __skb_push()/__skb_pull() pair to
2563 * get to the start of the mac header as we're guaranteed
2564 * to always start from here under eBPF.
2566 ret = bpf_skb_generic_push(skb, off, len);
2568 skb->mac_header -= len;
2569 skb->network_header -= len;
2571 skb->transport_header = skb->network_header;
2577 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
2579 bool trans_same = skb->transport_header == skb->network_header;
2582 /* Same here, __skb_push()/__skb_pull() pair not needed. */
2583 ret = bpf_skb_generic_pop(skb, off, len);
2585 skb->mac_header += len;
2586 skb->network_header += len;
2588 skb->transport_header = skb->network_header;
2594 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
2596 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2597 u32 off = skb_mac_header_len(skb);
2600 /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2601 if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2604 ret = skb_cow(skb, len_diff);
2605 if (unlikely(ret < 0))
2608 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2609 if (unlikely(ret < 0))
2612 if (skb_is_gso(skb)) {
2613 struct skb_shared_info *shinfo = skb_shinfo(skb);
2615 /* SKB_GSO_TCPV4 needs to be changed into
2618 if (shinfo->gso_type & SKB_GSO_TCPV4) {
2619 shinfo->gso_type &= ~SKB_GSO_TCPV4;
2620 shinfo->gso_type |= SKB_GSO_TCPV6;
2623 /* Due to IPv6 header, MSS needs to be downgraded. */
2624 skb_decrease_gso_size(shinfo, len_diff);
2625 /* Header must be checked, and gso_segs recomputed. */
2626 shinfo->gso_type |= SKB_GSO_DODGY;
2627 shinfo->gso_segs = 0;
2630 skb->protocol = htons(ETH_P_IPV6);
2631 skb_clear_hash(skb);
2636 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
2638 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2639 u32 off = skb_mac_header_len(skb);
2642 /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2643 if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2646 ret = skb_unclone(skb, GFP_ATOMIC);
2647 if (unlikely(ret < 0))
2650 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2651 if (unlikely(ret < 0))
2654 if (skb_is_gso(skb)) {
2655 struct skb_shared_info *shinfo = skb_shinfo(skb);
2657 /* SKB_GSO_TCPV6 needs to be changed into
2660 if (shinfo->gso_type & SKB_GSO_TCPV6) {
2661 shinfo->gso_type &= ~SKB_GSO_TCPV6;
2662 shinfo->gso_type |= SKB_GSO_TCPV4;
2665 /* Due to IPv4 header, MSS can be upgraded. */
2666 skb_increase_gso_size(shinfo, len_diff);
2667 /* Header must be checked, and gso_segs recomputed. */
2668 shinfo->gso_type |= SKB_GSO_DODGY;
2669 shinfo->gso_segs = 0;
2672 skb->protocol = htons(ETH_P_IP);
2673 skb_clear_hash(skb);
2678 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
2680 __be16 from_proto = skb->protocol;
2682 if (from_proto == htons(ETH_P_IP) &&
2683 to_proto == htons(ETH_P_IPV6))
2684 return bpf_skb_proto_4_to_6(skb);
2686 if (from_proto == htons(ETH_P_IPV6) &&
2687 to_proto == htons(ETH_P_IP))
2688 return bpf_skb_proto_6_to_4(skb);
2693 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
2698 if (unlikely(flags))
2701 /* General idea is that this helper does the basic groundwork
2702 * needed for changing the protocol, and eBPF program fills the
2703 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
2704 * and other helpers, rather than passing a raw buffer here.
2706 * The rationale is to keep this minimal and without a need to
2707 * deal with raw packet data. F.e. even if we would pass buffers
2708 * here, the program still needs to call the bpf_lX_csum_replace()
2709 * helpers anyway. Plus, this way we keep also separation of
2710 * concerns, since f.e. bpf_skb_store_bytes() should only take
2713 * Currently, additional options and extension header space are
2714 * not supported, but flags register is reserved so we can adapt
2715 * that. For offloads, we mark packet as dodgy, so that headers
2716 * need to be verified first.
2718 ret = bpf_skb_proto_xlat(skb, proto);
2719 bpf_compute_data_pointers(skb);
2723 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
2724 .func = bpf_skb_change_proto,
2726 .ret_type = RET_INTEGER,
2727 .arg1_type = ARG_PTR_TO_CTX,
2728 .arg2_type = ARG_ANYTHING,
2729 .arg3_type = ARG_ANYTHING,
2732 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
2734 /* We only allow a restricted subset to be changed for now. */
2735 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
2736 !skb_pkt_type_ok(pkt_type)))
2739 skb->pkt_type = pkt_type;
2743 static const struct bpf_func_proto bpf_skb_change_type_proto = {
2744 .func = bpf_skb_change_type,
2746 .ret_type = RET_INTEGER,
2747 .arg1_type = ARG_PTR_TO_CTX,
2748 .arg2_type = ARG_ANYTHING,
2751 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
2753 switch (skb->protocol) {
2754 case htons(ETH_P_IP):
2755 return sizeof(struct iphdr);
2756 case htons(ETH_P_IPV6):
2757 return sizeof(struct ipv6hdr);
2763 static int bpf_skb_net_grow(struct sk_buff *skb, u32 len_diff)
2765 u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
2768 /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2769 if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2772 ret = skb_cow(skb, len_diff);
2773 if (unlikely(ret < 0))
2776 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2777 if (unlikely(ret < 0))
2780 if (skb_is_gso(skb)) {
2781 struct skb_shared_info *shinfo = skb_shinfo(skb);
2783 /* Due to header grow, MSS needs to be downgraded. */
2784 skb_decrease_gso_size(shinfo, len_diff);
2785 /* Header must be checked, and gso_segs recomputed. */
2786 shinfo->gso_type |= SKB_GSO_DODGY;
2787 shinfo->gso_segs = 0;
2793 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 len_diff)
2795 u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
2798 /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2799 if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2802 ret = skb_unclone(skb, GFP_ATOMIC);
2803 if (unlikely(ret < 0))
2806 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2807 if (unlikely(ret < 0))
2810 if (skb_is_gso(skb)) {
2811 struct skb_shared_info *shinfo = skb_shinfo(skb);
2813 /* Due to header shrink, MSS can be upgraded. */
2814 skb_increase_gso_size(shinfo, len_diff);
2815 /* Header must be checked, and gso_segs recomputed. */
2816 shinfo->gso_type |= SKB_GSO_DODGY;
2817 shinfo->gso_segs = 0;
2823 static u32 __bpf_skb_max_len(const struct sk_buff *skb)
2825 return skb->dev ? skb->dev->mtu + skb->dev->hard_header_len :
2829 static int bpf_skb_adjust_net(struct sk_buff *skb, s32 len_diff)
2831 bool trans_same = skb->transport_header == skb->network_header;
2832 u32 len_cur, len_diff_abs = abs(len_diff);
2833 u32 len_min = bpf_skb_net_base_len(skb);
2834 u32 len_max = __bpf_skb_max_len(skb);
2835 __be16 proto = skb->protocol;
2836 bool shrink = len_diff < 0;
2839 if (unlikely(len_diff_abs > 0xfffU))
2841 if (unlikely(proto != htons(ETH_P_IP) &&
2842 proto != htons(ETH_P_IPV6)))
2845 len_cur = skb->len - skb_network_offset(skb);
2846 if (skb_transport_header_was_set(skb) && !trans_same)
2847 len_cur = skb_network_header_len(skb);
2848 if ((shrink && (len_diff_abs >= len_cur ||
2849 len_cur - len_diff_abs < len_min)) ||
2850 (!shrink && (skb->len + len_diff_abs > len_max &&
2854 ret = shrink ? bpf_skb_net_shrink(skb, len_diff_abs) :
2855 bpf_skb_net_grow(skb, len_diff_abs);
2857 bpf_compute_data_pointers(skb);
2861 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
2862 u32, mode, u64, flags)
2864 if (unlikely(flags))
2866 if (likely(mode == BPF_ADJ_ROOM_NET))
2867 return bpf_skb_adjust_net(skb, len_diff);
2872 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
2873 .func = bpf_skb_adjust_room,
2875 .ret_type = RET_INTEGER,
2876 .arg1_type = ARG_PTR_TO_CTX,
2877 .arg2_type = ARG_ANYTHING,
2878 .arg3_type = ARG_ANYTHING,
2879 .arg4_type = ARG_ANYTHING,
2882 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
2884 u32 min_len = skb_network_offset(skb);
2886 if (skb_transport_header_was_set(skb))
2887 min_len = skb_transport_offset(skb);
2888 if (skb->ip_summed == CHECKSUM_PARTIAL)
2889 min_len = skb_checksum_start_offset(skb) +
2890 skb->csum_offset + sizeof(__sum16);
2894 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
2896 unsigned int old_len = skb->len;
2899 ret = __skb_grow_rcsum(skb, new_len);
2901 memset(skb->data + old_len, 0, new_len - old_len);
2905 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
2907 return __skb_trim_rcsum(skb, new_len);
2910 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
2913 u32 max_len = __bpf_skb_max_len(skb);
2914 u32 min_len = __bpf_skb_min_len(skb);
2917 if (unlikely(flags || new_len > max_len || new_len < min_len))
2919 if (skb->encapsulation)
2922 /* The basic idea of this helper is that it's performing the
2923 * needed work to either grow or trim an skb, and eBPF program
2924 * rewrites the rest via helpers like bpf_skb_store_bytes(),
2925 * bpf_lX_csum_replace() and others rather than passing a raw
2926 * buffer here. This one is a slow path helper and intended
2927 * for replies with control messages.
2929 * Like in bpf_skb_change_proto(), we want to keep this rather
2930 * minimal and without protocol specifics so that we are able
2931 * to separate concerns as in bpf_skb_store_bytes() should only
2932 * be the one responsible for writing buffers.
2934 * It's really expected to be a slow path operation here for
2935 * control message replies, so we're implicitly linearizing,
2936 * uncloning and drop offloads from the skb by this.
2938 ret = __bpf_try_make_writable(skb, skb->len);
2940 if (new_len > skb->len)
2941 ret = bpf_skb_grow_rcsum(skb, new_len);
2942 else if (new_len < skb->len)
2943 ret = bpf_skb_trim_rcsum(skb, new_len);
2944 if (!ret && skb_is_gso(skb))
2950 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
2953 int ret = __bpf_skb_change_tail(skb, new_len, flags);
2955 bpf_compute_data_pointers(skb);
2959 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
2960 .func = bpf_skb_change_tail,
2962 .ret_type = RET_INTEGER,
2963 .arg1_type = ARG_PTR_TO_CTX,
2964 .arg2_type = ARG_ANYTHING,
2965 .arg3_type = ARG_ANYTHING,
2968 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
2971 int ret = __bpf_skb_change_tail(skb, new_len, flags);
2973 bpf_compute_data_end_sk_skb(skb);
2977 static const struct bpf_func_proto sk_skb_change_tail_proto = {
2978 .func = sk_skb_change_tail,
2980 .ret_type = RET_INTEGER,
2981 .arg1_type = ARG_PTR_TO_CTX,
2982 .arg2_type = ARG_ANYTHING,
2983 .arg3_type = ARG_ANYTHING,
2986 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
2989 u32 max_len = __bpf_skb_max_len(skb);
2990 u32 new_len = skb->len + head_room;
2993 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
2994 new_len < skb->len))
2997 ret = skb_cow(skb, head_room);
2999 /* Idea for this helper is that we currently only
3000 * allow to expand on mac header. This means that
3001 * skb->protocol network header, etc, stay as is.
3002 * Compared to bpf_skb_change_tail(), we're more
3003 * flexible due to not needing to linearize or
3004 * reset GSO. Intention for this helper is to be
3005 * used by an L3 skb that needs to push mac header
3006 * for redirection into L2 device.
3008 __skb_push(skb, head_room);
3009 memset(skb->data, 0, head_room);
3010 skb_reset_mac_header(skb);
3016 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3019 int ret = __bpf_skb_change_head(skb, head_room, flags);
3021 bpf_compute_data_pointers(skb);
3025 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3026 .func = bpf_skb_change_head,
3028 .ret_type = RET_INTEGER,
3029 .arg1_type = ARG_PTR_TO_CTX,
3030 .arg2_type = ARG_ANYTHING,
3031 .arg3_type = ARG_ANYTHING,
3034 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3037 int ret = __bpf_skb_change_head(skb, head_room, flags);
3039 bpf_compute_data_end_sk_skb(skb);
3043 static const struct bpf_func_proto sk_skb_change_head_proto = {
3044 .func = sk_skb_change_head,
3046 .ret_type = RET_INTEGER,
3047 .arg1_type = ARG_PTR_TO_CTX,
3048 .arg2_type = ARG_ANYTHING,
3049 .arg3_type = ARG_ANYTHING,
3051 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3053 return xdp_data_meta_unsupported(xdp) ? 0 :
3054 xdp->data - xdp->data_meta;
3057 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3059 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3060 unsigned long metalen = xdp_get_metalen(xdp);
3061 void *data_start = xdp_frame_end + metalen;
3062 void *data = xdp->data + offset;
3064 if (unlikely(data < data_start ||
3065 data > xdp->data_end - ETH_HLEN))
3069 memmove(xdp->data_meta + offset,
3070 xdp->data_meta, metalen);
3071 xdp->data_meta += offset;
3077 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3078 .func = bpf_xdp_adjust_head,
3080 .ret_type = RET_INTEGER,
3081 .arg1_type = ARG_PTR_TO_CTX,
3082 .arg2_type = ARG_ANYTHING,
3085 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3087 void *data_end = xdp->data_end + offset;
3089 /* only shrinking is allowed for now. */
3090 if (unlikely(offset >= 0))
3093 if (unlikely(data_end < xdp->data + ETH_HLEN))
3096 xdp->data_end = data_end;
3101 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3102 .func = bpf_xdp_adjust_tail,
3104 .ret_type = RET_INTEGER,
3105 .arg1_type = ARG_PTR_TO_CTX,
3106 .arg2_type = ARG_ANYTHING,
3109 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3111 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3112 void *meta = xdp->data_meta + offset;
3113 unsigned long metalen = xdp->data - meta;
3115 if (xdp_data_meta_unsupported(xdp))
3117 if (unlikely(meta < xdp_frame_end ||
3120 if (unlikely((metalen & (sizeof(__u32) - 1)) ||
3124 xdp->data_meta = meta;
3129 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3130 .func = bpf_xdp_adjust_meta,
3132 .ret_type = RET_INTEGER,
3133 .arg1_type = ARG_PTR_TO_CTX,
3134 .arg2_type = ARG_ANYTHING,
3137 static int __bpf_tx_xdp(struct net_device *dev,
3138 struct bpf_map *map,
3139 struct xdp_buff *xdp,
3142 struct xdp_frame *xdpf;
3145 if (!dev->netdev_ops->ndo_xdp_xmit) {
3149 err = xdp_ok_fwd_dev(dev, xdp->data_end - xdp->data);
3153 xdpf = convert_to_xdp_frame(xdp);
3154 if (unlikely(!xdpf))
3157 sent = dev->netdev_ops->ndo_xdp_xmit(dev, 1, &xdpf, XDP_XMIT_FLUSH);
3163 static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd,
3164 struct bpf_map *map,
3165 struct xdp_buff *xdp,
3170 switch (map->map_type) {
3171 case BPF_MAP_TYPE_DEVMAP: {
3172 struct bpf_dtab_netdev *dst = fwd;
3174 err = dev_map_enqueue(dst, xdp, dev_rx);
3177 __dev_map_insert_ctx(map, index);
3180 case BPF_MAP_TYPE_CPUMAP: {
3181 struct bpf_cpu_map_entry *rcpu = fwd;
3183 err = cpu_map_enqueue(rcpu, xdp, dev_rx);
3186 __cpu_map_insert_ctx(map, index);
3189 case BPF_MAP_TYPE_XSKMAP: {
3190 struct xdp_sock *xs = fwd;
3192 err = __xsk_map_redirect(map, xdp, xs);
3201 void xdp_do_flush_map(void)
3203 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
3204 struct bpf_map *map = ri->map_to_flush;
3206 ri->map_to_flush = NULL;
3208 switch (map->map_type) {
3209 case BPF_MAP_TYPE_DEVMAP:
3210 __dev_map_flush(map);
3212 case BPF_MAP_TYPE_CPUMAP:
3213 __cpu_map_flush(map);
3215 case BPF_MAP_TYPE_XSKMAP:
3216 __xsk_map_flush(map);
3223 EXPORT_SYMBOL_GPL(xdp_do_flush_map);
3225 static void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index)
3227 switch (map->map_type) {
3228 case BPF_MAP_TYPE_DEVMAP:
3229 return __dev_map_lookup_elem(map, index);
3230 case BPF_MAP_TYPE_CPUMAP:
3231 return __cpu_map_lookup_elem(map, index);
3232 case BPF_MAP_TYPE_XSKMAP:
3233 return __xsk_map_lookup_elem(map, index);
3239 static inline bool xdp_map_invalid(const struct bpf_prog *xdp_prog,
3242 return (unsigned long)xdp_prog->aux != aux;
3245 static int xdp_do_redirect_map(struct net_device *dev, struct xdp_buff *xdp,
3246 struct bpf_prog *xdp_prog)
3248 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
3249 unsigned long map_owner = ri->map_owner;
3250 struct bpf_map *map = ri->map;
3251 u32 index = ri->ifindex;
3259 if (unlikely(xdp_map_invalid(xdp_prog, map_owner))) {
3265 fwd = __xdp_map_lookup_elem(map, index);
3270 if (ri->map_to_flush && ri->map_to_flush != map)
3273 err = __bpf_tx_xdp_map(dev, fwd, map, xdp, index);
3277 ri->map_to_flush = map;
3278 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3281 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3285 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3286 struct bpf_prog *xdp_prog)
3288 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
3289 struct net_device *fwd;
3290 u32 index = ri->ifindex;
3294 return xdp_do_redirect_map(dev, xdp, xdp_prog);
3296 fwd = dev_get_by_index_rcu(dev_net(dev), index);
3298 if (unlikely(!fwd)) {
3303 err = __bpf_tx_xdp(fwd, NULL, xdp, 0);
3307 _trace_xdp_redirect(dev, xdp_prog, index);
3310 _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3313 EXPORT_SYMBOL_GPL(xdp_do_redirect);
3315 static int xdp_do_generic_redirect_map(struct net_device *dev,
3316 struct sk_buff *skb,
3317 struct xdp_buff *xdp,
3318 struct bpf_prog *xdp_prog)
3320 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
3321 unsigned long map_owner = ri->map_owner;
3322 struct bpf_map *map = ri->map;
3323 u32 index = ri->ifindex;
3331 if (unlikely(xdp_map_invalid(xdp_prog, map_owner))) {
3336 fwd = __xdp_map_lookup_elem(map, index);
3337 if (unlikely(!fwd)) {
3342 if (map->map_type == BPF_MAP_TYPE_DEVMAP) {
3343 struct bpf_dtab_netdev *dst = fwd;
3345 err = dev_map_generic_redirect(dst, skb, xdp_prog);
3348 } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) {
3349 struct xdp_sock *xs = fwd;
3351 err = xsk_generic_rcv(xs, xdp);
3356 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
3361 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3364 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3368 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
3369 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
3371 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
3372 u32 index = ri->ifindex;
3373 struct net_device *fwd;
3377 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog);
3380 fwd = dev_get_by_index_rcu(dev_net(dev), index);
3381 if (unlikely(!fwd)) {
3386 err = xdp_ok_fwd_dev(fwd, skb->len);
3391 _trace_xdp_redirect(dev, xdp_prog, index);
3392 generic_xdp_tx(skb, xdp_prog);
3395 _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3398 EXPORT_SYMBOL_GPL(xdp_do_generic_redirect);
3400 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
3402 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
3404 if (unlikely(flags))
3407 ri->ifindex = ifindex;
3412 return XDP_REDIRECT;
3415 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
3416 .func = bpf_xdp_redirect,
3418 .ret_type = RET_INTEGER,
3419 .arg1_type = ARG_ANYTHING,
3420 .arg2_type = ARG_ANYTHING,
3423 BPF_CALL_4(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex, u64, flags,
3424 unsigned long, map_owner)
3426 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
3428 if (unlikely(flags))
3431 ri->ifindex = ifindex;
3434 ri->map_owner = map_owner;
3436 return XDP_REDIRECT;
3439 /* Note, arg4 is hidden from users and populated by the verifier
3440 * with the right pointer.
3442 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
3443 .func = bpf_xdp_redirect_map,
3445 .ret_type = RET_INTEGER,
3446 .arg1_type = ARG_CONST_MAP_PTR,
3447 .arg2_type = ARG_ANYTHING,
3448 .arg3_type = ARG_ANYTHING,
3451 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
3452 unsigned long off, unsigned long len)
3454 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
3458 if (ptr != dst_buff)
3459 memcpy(dst_buff, ptr, len);
3464 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
3465 u64, flags, void *, meta, u64, meta_size)
3467 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
3469 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
3471 if (unlikely(skb_size > skb->len))
3474 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
3478 static const struct bpf_func_proto bpf_skb_event_output_proto = {
3479 .func = bpf_skb_event_output,
3481 .ret_type = RET_INTEGER,
3482 .arg1_type = ARG_PTR_TO_CTX,
3483 .arg2_type = ARG_CONST_MAP_PTR,
3484 .arg3_type = ARG_ANYTHING,
3485 .arg4_type = ARG_PTR_TO_MEM,
3486 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
3489 static unsigned short bpf_tunnel_key_af(u64 flags)
3491 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
3494 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
3495 u32, size, u64, flags)
3497 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3498 u8 compat[sizeof(struct bpf_tunnel_key)];
3502 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
3506 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
3510 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3513 case offsetof(struct bpf_tunnel_key, tunnel_label):
3514 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3516 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3517 /* Fixup deprecated structure layouts here, so we have
3518 * a common path later on.
3520 if (ip_tunnel_info_af(info) != AF_INET)
3523 to = (struct bpf_tunnel_key *)compat;
3530 to->tunnel_id = be64_to_cpu(info->key.tun_id);
3531 to->tunnel_tos = info->key.tos;
3532 to->tunnel_ttl = info->key.ttl;
3535 if (flags & BPF_F_TUNINFO_IPV6) {
3536 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
3537 sizeof(to->remote_ipv6));
3538 to->tunnel_label = be32_to_cpu(info->key.label);
3540 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
3541 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
3542 to->tunnel_label = 0;
3545 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
3546 memcpy(to_orig, to, size);
3550 memset(to_orig, 0, size);
3554 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
3555 .func = bpf_skb_get_tunnel_key,
3557 .ret_type = RET_INTEGER,
3558 .arg1_type = ARG_PTR_TO_CTX,
3559 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
3560 .arg3_type = ARG_CONST_SIZE,
3561 .arg4_type = ARG_ANYTHING,
3564 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
3566 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3569 if (unlikely(!info ||
3570 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
3574 if (unlikely(size < info->options_len)) {
3579 ip_tunnel_info_opts_get(to, info);
3580 if (size > info->options_len)
3581 memset(to + info->options_len, 0, size - info->options_len);
3583 return info->options_len;
3585 memset(to, 0, size);
3589 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
3590 .func = bpf_skb_get_tunnel_opt,
3592 .ret_type = RET_INTEGER,
3593 .arg1_type = ARG_PTR_TO_CTX,
3594 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
3595 .arg3_type = ARG_CONST_SIZE,
3598 static struct metadata_dst __percpu *md_dst;
3600 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
3601 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
3603 struct metadata_dst *md = this_cpu_ptr(md_dst);
3604 u8 compat[sizeof(struct bpf_tunnel_key)];
3605 struct ip_tunnel_info *info;
3607 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
3608 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
3610 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3612 case offsetof(struct bpf_tunnel_key, tunnel_label):
3613 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3614 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3615 /* Fixup deprecated structure layouts here, so we have
3616 * a common path later on.
3618 memcpy(compat, from, size);
3619 memset(compat + size, 0, sizeof(compat) - size);
3620 from = (const struct bpf_tunnel_key *) compat;
3626 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
3631 dst_hold((struct dst_entry *) md);
3632 skb_dst_set(skb, (struct dst_entry *) md);
3634 info = &md->u.tun_info;
3635 memset(info, 0, sizeof(*info));
3636 info->mode = IP_TUNNEL_INFO_TX;
3638 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
3639 if (flags & BPF_F_DONT_FRAGMENT)
3640 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
3641 if (flags & BPF_F_ZERO_CSUM_TX)
3642 info->key.tun_flags &= ~TUNNEL_CSUM;
3643 if (flags & BPF_F_SEQ_NUMBER)
3644 info->key.tun_flags |= TUNNEL_SEQ;
3646 info->key.tun_id = cpu_to_be64(from->tunnel_id);
3647 info->key.tos = from->tunnel_tos;
3648 info->key.ttl = from->tunnel_ttl;
3650 if (flags & BPF_F_TUNINFO_IPV6) {
3651 info->mode |= IP_TUNNEL_INFO_IPV6;
3652 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
3653 sizeof(from->remote_ipv6));
3654 info->key.label = cpu_to_be32(from->tunnel_label) &
3655 IPV6_FLOWLABEL_MASK;
3657 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
3663 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
3664 .func = bpf_skb_set_tunnel_key,
3666 .ret_type = RET_INTEGER,
3667 .arg1_type = ARG_PTR_TO_CTX,
3668 .arg2_type = ARG_PTR_TO_MEM,
3669 .arg3_type = ARG_CONST_SIZE,
3670 .arg4_type = ARG_ANYTHING,
3673 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
3674 const u8 *, from, u32, size)
3676 struct ip_tunnel_info *info = skb_tunnel_info(skb);
3677 const struct metadata_dst *md = this_cpu_ptr(md_dst);
3679 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
3681 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
3684 ip_tunnel_info_opts_set(info, from, size);
3689 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
3690 .func = bpf_skb_set_tunnel_opt,
3692 .ret_type = RET_INTEGER,
3693 .arg1_type = ARG_PTR_TO_CTX,
3694 .arg2_type = ARG_PTR_TO_MEM,
3695 .arg3_type = ARG_CONST_SIZE,
3698 static const struct bpf_func_proto *
3699 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
3702 struct metadata_dst __percpu *tmp;
3704 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
3709 if (cmpxchg(&md_dst, NULL, tmp))
3710 metadata_dst_free_percpu(tmp);
3714 case BPF_FUNC_skb_set_tunnel_key:
3715 return &bpf_skb_set_tunnel_key_proto;
3716 case BPF_FUNC_skb_set_tunnel_opt:
3717 return &bpf_skb_set_tunnel_opt_proto;
3723 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
3726 struct bpf_array *array = container_of(map, struct bpf_array, map);
3727 struct cgroup *cgrp;
3730 sk = skb_to_full_sk(skb);
3731 if (!sk || !sk_fullsock(sk))
3733 if (unlikely(idx >= array->map.max_entries))
3736 cgrp = READ_ONCE(array->ptrs[idx]);
3737 if (unlikely(!cgrp))
3740 return sk_under_cgroup_hierarchy(sk, cgrp);
3743 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
3744 .func = bpf_skb_under_cgroup,
3746 .ret_type = RET_INTEGER,
3747 .arg1_type = ARG_PTR_TO_CTX,
3748 .arg2_type = ARG_CONST_MAP_PTR,
3749 .arg3_type = ARG_ANYTHING,
3752 #ifdef CONFIG_SOCK_CGROUP_DATA
3753 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
3755 struct sock *sk = skb_to_full_sk(skb);
3756 struct cgroup *cgrp;
3758 if (!sk || !sk_fullsock(sk))
3761 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
3762 return cgrp->kn->id.id;
3765 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
3766 .func = bpf_skb_cgroup_id,
3768 .ret_type = RET_INTEGER,
3769 .arg1_type = ARG_PTR_TO_CTX,
3773 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
3774 unsigned long off, unsigned long len)
3776 memcpy(dst_buff, src_buff + off, len);
3780 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
3781 u64, flags, void *, meta, u64, meta_size)
3783 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
3785 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
3787 if (unlikely(xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
3790 return bpf_event_output(map, flags, meta, meta_size, xdp->data,
3791 xdp_size, bpf_xdp_copy);
3794 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
3795 .func = bpf_xdp_event_output,
3797 .ret_type = RET_INTEGER,
3798 .arg1_type = ARG_PTR_TO_CTX,
3799 .arg2_type = ARG_CONST_MAP_PTR,
3800 .arg3_type = ARG_ANYTHING,
3801 .arg4_type = ARG_PTR_TO_MEM,
3802 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
3805 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
3807 return skb->sk ? sock_gen_cookie(skb->sk) : 0;
3810 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
3811 .func = bpf_get_socket_cookie,
3813 .ret_type = RET_INTEGER,
3814 .arg1_type = ARG_PTR_TO_CTX,
3817 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
3819 struct sock *sk = sk_to_full_sk(skb->sk);
3822 if (!sk || !sk_fullsock(sk))
3824 kuid = sock_net_uid(sock_net(sk), sk);
3825 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
3828 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
3829 .func = bpf_get_socket_uid,
3831 .ret_type = RET_INTEGER,
3832 .arg1_type = ARG_PTR_TO_CTX,
3835 BPF_CALL_5(bpf_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
3836 int, level, int, optname, char *, optval, int, optlen)
3838 struct sock *sk = bpf_sock->sk;
3842 if (!sk_fullsock(sk))
3845 if (level == SOL_SOCKET) {
3846 if (optlen != sizeof(int))
3848 val = *((int *)optval);
3850 /* Only some socketops are supported */
3853 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
3854 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
3857 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
3858 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
3860 case SO_MAX_PACING_RATE:
3861 sk->sk_max_pacing_rate = val;
3862 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
3863 sk->sk_max_pacing_rate);
3866 sk->sk_priority = val;
3871 sk->sk_rcvlowat = val ? : 1;
3880 } else if (level == SOL_IP) {
3881 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
3884 val = *((int *)optval);
3885 /* Only some options are supported */
3888 if (val < -1 || val > 0xff) {
3891 struct inet_sock *inet = inet_sk(sk);
3901 #if IS_ENABLED(CONFIG_IPV6)
3902 } else if (level == SOL_IPV6) {
3903 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
3906 val = *((int *)optval);
3907 /* Only some options are supported */
3910 if (val < -1 || val > 0xff) {
3913 struct ipv6_pinfo *np = inet6_sk(sk);
3924 } else if (level == SOL_TCP &&
3925 sk->sk_prot->setsockopt == tcp_setsockopt) {
3926 if (optname == TCP_CONGESTION) {
3927 char name[TCP_CA_NAME_MAX];
3928 bool reinit = bpf_sock->op > BPF_SOCK_OPS_NEEDS_ECN;
3930 strncpy(name, optval, min_t(long, optlen,
3931 TCP_CA_NAME_MAX-1));
3932 name[TCP_CA_NAME_MAX-1] = 0;
3933 ret = tcp_set_congestion_control(sk, name, false,
3936 struct tcp_sock *tp = tcp_sk(sk);
3938 if (optlen != sizeof(int))
3941 val = *((int *)optval);
3942 /* Only some options are supported */
3945 if (val <= 0 || tp->data_segs_out > 0)
3950 case TCP_BPF_SNDCWND_CLAMP:
3954 tp->snd_cwnd_clamp = val;
3955 tp->snd_ssthresh = val;
3969 static const struct bpf_func_proto bpf_setsockopt_proto = {
3970 .func = bpf_setsockopt,
3972 .ret_type = RET_INTEGER,
3973 .arg1_type = ARG_PTR_TO_CTX,
3974 .arg2_type = ARG_ANYTHING,
3975 .arg3_type = ARG_ANYTHING,
3976 .arg4_type = ARG_PTR_TO_MEM,
3977 .arg5_type = ARG_CONST_SIZE,
3980 BPF_CALL_5(bpf_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
3981 int, level, int, optname, char *, optval, int, optlen)
3983 struct sock *sk = bpf_sock->sk;
3985 if (!sk_fullsock(sk))
3989 if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
3990 if (optname == TCP_CONGESTION) {
3991 struct inet_connection_sock *icsk = inet_csk(sk);
3993 if (!icsk->icsk_ca_ops || optlen <= 1)
3995 strncpy(optval, icsk->icsk_ca_ops->name, optlen);
3996 optval[optlen - 1] = 0;
4000 } else if (level == SOL_IP) {
4001 struct inet_sock *inet = inet_sk(sk);
4003 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4006 /* Only some options are supported */
4009 *((int *)optval) = (int)inet->tos;
4014 #if IS_ENABLED(CONFIG_IPV6)
4015 } else if (level == SOL_IPV6) {
4016 struct ipv6_pinfo *np = inet6_sk(sk);
4018 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4021 /* Only some options are supported */
4024 *((int *)optval) = (int)np->tclass;
4036 memset(optval, 0, optlen);
4040 static const struct bpf_func_proto bpf_getsockopt_proto = {
4041 .func = bpf_getsockopt,
4043 .ret_type = RET_INTEGER,
4044 .arg1_type = ARG_PTR_TO_CTX,
4045 .arg2_type = ARG_ANYTHING,
4046 .arg3_type = ARG_ANYTHING,
4047 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
4048 .arg5_type = ARG_CONST_SIZE,
4051 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
4054 struct sock *sk = bpf_sock->sk;
4055 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
4057 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
4061 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
4063 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
4066 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
4067 .func = bpf_sock_ops_cb_flags_set,
4069 .ret_type = RET_INTEGER,
4070 .arg1_type = ARG_PTR_TO_CTX,
4071 .arg2_type = ARG_ANYTHING,
4074 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
4075 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
4077 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
4081 struct sock *sk = ctx->sk;
4084 /* Binding to port can be expensive so it's prohibited in the helper.
4085 * Only binding to IP is supported.
4088 if (addr->sa_family == AF_INET) {
4089 if (addr_len < sizeof(struct sockaddr_in))
4091 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
4093 return __inet_bind(sk, addr, addr_len, true, false);
4094 #if IS_ENABLED(CONFIG_IPV6)
4095 } else if (addr->sa_family == AF_INET6) {
4096 if (addr_len < SIN6_LEN_RFC2133)
4098 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
4100 /* ipv6_bpf_stub cannot be NULL, since it's called from
4101 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
4103 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, true, false);
4104 #endif /* CONFIG_IPV6 */
4106 #endif /* CONFIG_INET */
4108 return -EAFNOSUPPORT;
4111 static const struct bpf_func_proto bpf_bind_proto = {
4114 .ret_type = RET_INTEGER,
4115 .arg1_type = ARG_PTR_TO_CTX,
4116 .arg2_type = ARG_PTR_TO_MEM,
4117 .arg3_type = ARG_CONST_SIZE,
4121 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
4122 struct bpf_xfrm_state *, to, u32, size, u64, flags)
4124 const struct sec_path *sp = skb_sec_path(skb);
4125 const struct xfrm_state *x;
4127 if (!sp || unlikely(index >= sp->len || flags))
4130 x = sp->xvec[index];
4132 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
4135 to->reqid = x->props.reqid;
4136 to->spi = x->id.spi;
4137 to->family = x->props.family;
4140 if (to->family == AF_INET6) {
4141 memcpy(to->remote_ipv6, x->props.saddr.a6,
4142 sizeof(to->remote_ipv6));
4144 to->remote_ipv4 = x->props.saddr.a4;
4145 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4150 memset(to, 0, size);
4154 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
4155 .func = bpf_skb_get_xfrm_state,
4157 .ret_type = RET_INTEGER,
4158 .arg1_type = ARG_PTR_TO_CTX,
4159 .arg2_type = ARG_ANYTHING,
4160 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
4161 .arg4_type = ARG_CONST_SIZE,
4162 .arg5_type = ARG_ANYTHING,
4166 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
4167 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
4168 const struct neighbour *neigh,
4169 const struct net_device *dev)
4171 memcpy(params->dmac, neigh->ha, ETH_ALEN);
4172 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
4173 params->h_vlan_TCI = 0;
4174 params->h_vlan_proto = 0;
4175 params->ifindex = dev->ifindex;
4181 #if IS_ENABLED(CONFIG_INET)
4182 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4183 u32 flags, bool check_mtu)
4185 struct in_device *in_dev;
4186 struct neighbour *neigh;
4187 struct net_device *dev;
4188 struct fib_result res;
4194 dev = dev_get_by_index_rcu(net, params->ifindex);
4198 /* verify forwarding is enabled on this interface */
4199 in_dev = __in_dev_get_rcu(dev);
4200 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
4201 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4203 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4205 fl4.flowi4_oif = params->ifindex;
4207 fl4.flowi4_iif = params->ifindex;
4210 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
4211 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
4212 fl4.flowi4_flags = 0;
4214 fl4.flowi4_proto = params->l4_protocol;
4215 fl4.daddr = params->ipv4_dst;
4216 fl4.saddr = params->ipv4_src;
4217 fl4.fl4_sport = params->sport;
4218 fl4.fl4_dport = params->dport;
4220 if (flags & BPF_FIB_LOOKUP_DIRECT) {
4221 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4222 struct fib_table *tb;
4224 tb = fib_get_table(net, tbid);
4226 return BPF_FIB_LKUP_RET_NOT_FWDED;
4228 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
4230 fl4.flowi4_mark = 0;
4231 fl4.flowi4_secid = 0;
4232 fl4.flowi4_tun_key.tun_id = 0;
4233 fl4.flowi4_uid = sock_net_uid(net, NULL);
4235 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
4239 /* map fib lookup errors to RTN_ type */
4241 return BPF_FIB_LKUP_RET_BLACKHOLE;
4242 if (err == -EHOSTUNREACH)
4243 return BPF_FIB_LKUP_RET_UNREACHABLE;
4245 return BPF_FIB_LKUP_RET_PROHIBIT;
4247 return BPF_FIB_LKUP_RET_NOT_FWDED;
4250 if (res.type != RTN_UNICAST)
4251 return BPF_FIB_LKUP_RET_NOT_FWDED;
4253 if (res.fi->fib_nhs > 1)
4254 fib_select_path(net, &res, &fl4, NULL);
4257 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
4258 if (params->tot_len > mtu)
4259 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4262 nh = &res.fi->fib_nh[res.nh_sel];
4264 /* do not handle lwt encaps right now */
4265 if (nh->nh_lwtstate)
4266 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4270 params->ipv4_dst = nh->nh_gw;
4272 params->rt_metric = res.fi->fib_priority;
4274 /* xdp and cls_bpf programs are run in RCU-bh so
4275 * rcu_read_lock_bh is not needed here
4277 neigh = __ipv4_neigh_lookup_noref(dev, (__force u32)params->ipv4_dst);
4279 return BPF_FIB_LKUP_RET_NO_NEIGH;
4281 return bpf_fib_set_fwd_params(params, neigh, dev);
4285 #if IS_ENABLED(CONFIG_IPV6)
4286 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4287 u32 flags, bool check_mtu)
4289 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
4290 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
4291 struct neighbour *neigh;
4292 struct net_device *dev;
4293 struct inet6_dev *idev;
4294 struct fib6_info *f6i;
4300 /* link local addresses are never forwarded */
4301 if (rt6_need_strict(dst) || rt6_need_strict(src))
4302 return BPF_FIB_LKUP_RET_NOT_FWDED;
4304 dev = dev_get_by_index_rcu(net, params->ifindex);
4308 idev = __in6_dev_get_safely(dev);
4309 if (unlikely(!idev || !net->ipv6.devconf_all->forwarding))
4310 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4312 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4314 oif = fl6.flowi6_oif = params->ifindex;
4316 oif = fl6.flowi6_iif = params->ifindex;
4318 strict = RT6_LOOKUP_F_HAS_SADDR;
4320 fl6.flowlabel = params->flowinfo;
4321 fl6.flowi6_scope = 0;
4322 fl6.flowi6_flags = 0;
4325 fl6.flowi6_proto = params->l4_protocol;
4328 fl6.fl6_sport = params->sport;
4329 fl6.fl6_dport = params->dport;
4331 if (flags & BPF_FIB_LOOKUP_DIRECT) {
4332 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4333 struct fib6_table *tb;
4335 tb = ipv6_stub->fib6_get_table(net, tbid);
4337 return BPF_FIB_LKUP_RET_NOT_FWDED;
4339 f6i = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, strict);
4341 fl6.flowi6_mark = 0;
4342 fl6.flowi6_secid = 0;
4343 fl6.flowi6_tun_key.tun_id = 0;
4344 fl6.flowi6_uid = sock_net_uid(net, NULL);
4346 f6i = ipv6_stub->fib6_lookup(net, oif, &fl6, strict);
4349 if (unlikely(IS_ERR_OR_NULL(f6i) || f6i == net->ipv6.fib6_null_entry))
4350 return BPF_FIB_LKUP_RET_NOT_FWDED;
4352 if (unlikely(f6i->fib6_flags & RTF_REJECT)) {
4353 switch (f6i->fib6_type) {
4355 return BPF_FIB_LKUP_RET_BLACKHOLE;
4356 case RTN_UNREACHABLE:
4357 return BPF_FIB_LKUP_RET_UNREACHABLE;
4359 return BPF_FIB_LKUP_RET_PROHIBIT;
4361 return BPF_FIB_LKUP_RET_NOT_FWDED;
4365 if (f6i->fib6_type != RTN_UNICAST)
4366 return BPF_FIB_LKUP_RET_NOT_FWDED;
4368 if (f6i->fib6_nsiblings && fl6.flowi6_oif == 0)
4369 f6i = ipv6_stub->fib6_multipath_select(net, f6i, &fl6,
4370 fl6.flowi6_oif, NULL,
4374 mtu = ipv6_stub->ip6_mtu_from_fib6(f6i, dst, src);
4375 if (params->tot_len > mtu)
4376 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4379 if (f6i->fib6_nh.nh_lwtstate)
4380 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4382 if (f6i->fib6_flags & RTF_GATEWAY)
4383 *dst = f6i->fib6_nh.nh_gw;
4385 dev = f6i->fib6_nh.nh_dev;
4386 params->rt_metric = f6i->fib6_metric;
4388 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
4389 * not needed here. Can not use __ipv6_neigh_lookup_noref here
4390 * because we need to get nd_tbl via the stub
4392 neigh = ___neigh_lookup_noref(ipv6_stub->nd_tbl, neigh_key_eq128,
4393 ndisc_hashfn, dst, dev);
4395 return BPF_FIB_LKUP_RET_NO_NEIGH;
4397 return bpf_fib_set_fwd_params(params, neigh, dev);
4401 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
4402 struct bpf_fib_lookup *, params, int, plen, u32, flags)
4404 if (plen < sizeof(*params))
4407 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4410 switch (params->family) {
4411 #if IS_ENABLED(CONFIG_INET)
4413 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
4416 #if IS_ENABLED(CONFIG_IPV6)
4418 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
4422 return -EAFNOSUPPORT;
4425 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
4426 .func = bpf_xdp_fib_lookup,
4428 .ret_type = RET_INTEGER,
4429 .arg1_type = ARG_PTR_TO_CTX,
4430 .arg2_type = ARG_PTR_TO_MEM,
4431 .arg3_type = ARG_CONST_SIZE,
4432 .arg4_type = ARG_ANYTHING,
4435 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
4436 struct bpf_fib_lookup *, params, int, plen, u32, flags)
4438 struct net *net = dev_net(skb->dev);
4439 int rc = -EAFNOSUPPORT;
4441 if (plen < sizeof(*params))
4444 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4447 switch (params->family) {
4448 #if IS_ENABLED(CONFIG_INET)
4450 rc = bpf_ipv4_fib_lookup(net, params, flags, false);
4453 #if IS_ENABLED(CONFIG_IPV6)
4455 rc = bpf_ipv6_fib_lookup(net, params, flags, false);
4461 struct net_device *dev;
4463 dev = dev_get_by_index_rcu(net, params->ifindex);
4464 if (!is_skb_forwardable(dev, skb))
4465 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
4471 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
4472 .func = bpf_skb_fib_lookup,
4474 .ret_type = RET_INTEGER,
4475 .arg1_type = ARG_PTR_TO_CTX,
4476 .arg2_type = ARG_PTR_TO_MEM,
4477 .arg3_type = ARG_CONST_SIZE,
4478 .arg4_type = ARG_ANYTHING,
4481 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4482 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
4485 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
4487 if (!seg6_validate_srh(srh, len))
4491 case BPF_LWT_ENCAP_SEG6_INLINE:
4492 if (skb->protocol != htons(ETH_P_IPV6))
4495 err = seg6_do_srh_inline(skb, srh);
4497 case BPF_LWT_ENCAP_SEG6:
4498 skb_reset_inner_headers(skb);
4499 skb->encapsulation = 1;
4500 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
4506 bpf_compute_data_pointers(skb);
4510 ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
4511 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
4513 return seg6_lookup_nexthop(skb, NULL, 0);
4515 #endif /* CONFIG_IPV6_SEG6_BPF */
4517 BPF_CALL_4(bpf_lwt_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
4521 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4522 case BPF_LWT_ENCAP_SEG6:
4523 case BPF_LWT_ENCAP_SEG6_INLINE:
4524 return bpf_push_seg6_encap(skb, type, hdr, len);
4531 static const struct bpf_func_proto bpf_lwt_push_encap_proto = {
4532 .func = bpf_lwt_push_encap,
4534 .ret_type = RET_INTEGER,
4535 .arg1_type = ARG_PTR_TO_CTX,
4536 .arg2_type = ARG_ANYTHING,
4537 .arg3_type = ARG_PTR_TO_MEM,
4538 .arg4_type = ARG_CONST_SIZE
4541 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4542 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
4543 const void *, from, u32, len)
4545 struct seg6_bpf_srh_state *srh_state =
4546 this_cpu_ptr(&seg6_bpf_srh_states);
4547 void *srh_tlvs, *srh_end, *ptr;
4548 struct ipv6_sr_hdr *srh;
4551 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
4554 srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4555 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
4556 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
4558 ptr = skb->data + offset;
4559 if (ptr >= srh_tlvs && ptr + len <= srh_end)
4560 srh_state->valid = 0;
4561 else if (ptr < (void *)&srh->flags ||
4562 ptr + len > (void *)&srh->segments)
4565 if (unlikely(bpf_try_make_writable(skb, offset + len)))
4568 memcpy(skb->data + offset, from, len);
4572 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
4573 .func = bpf_lwt_seg6_store_bytes,
4575 .ret_type = RET_INTEGER,
4576 .arg1_type = ARG_PTR_TO_CTX,
4577 .arg2_type = ARG_ANYTHING,
4578 .arg3_type = ARG_PTR_TO_MEM,
4579 .arg4_type = ARG_CONST_SIZE
4582 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
4583 u32, action, void *, param, u32, param_len)
4585 struct seg6_bpf_srh_state *srh_state =
4586 this_cpu_ptr(&seg6_bpf_srh_states);
4587 struct ipv6_sr_hdr *srh;
4591 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
4593 srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4595 if (!srh_state->valid) {
4596 if (unlikely((srh_state->hdrlen & 7) != 0))
4599 srh->hdrlen = (u8)(srh_state->hdrlen >> 3);
4600 if (unlikely(!seg6_validate_srh(srh, (srh->hdrlen + 1) << 3)))
4603 srh_state->valid = 1;
4607 case SEG6_LOCAL_ACTION_END_X:
4608 if (param_len != sizeof(struct in6_addr))
4610 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
4611 case SEG6_LOCAL_ACTION_END_T:
4612 if (param_len != sizeof(int))
4614 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
4615 case SEG6_LOCAL_ACTION_END_B6:
4616 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
4620 ((struct ipv6_sr_hdr *)param)->hdrlen << 3;
4622 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
4623 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
4627 ((struct ipv6_sr_hdr *)param)->hdrlen << 3;
4634 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
4635 .func = bpf_lwt_seg6_action,
4637 .ret_type = RET_INTEGER,
4638 .arg1_type = ARG_PTR_TO_CTX,
4639 .arg2_type = ARG_ANYTHING,
4640 .arg3_type = ARG_PTR_TO_MEM,
4641 .arg4_type = ARG_CONST_SIZE
4644 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
4647 struct seg6_bpf_srh_state *srh_state =
4648 this_cpu_ptr(&seg6_bpf_srh_states);
4649 void *srh_end, *srh_tlvs, *ptr;
4650 struct ipv6_sr_hdr *srh;
4651 struct ipv6hdr *hdr;
4655 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
4657 srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4659 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
4660 ((srh->first_segment + 1) << 4));
4661 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
4663 ptr = skb->data + offset;
4665 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
4667 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
4671 ret = skb_cow_head(skb, len);
4672 if (unlikely(ret < 0))
4675 ret = bpf_skb_net_hdr_push(skb, offset, len);
4677 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
4680 bpf_compute_data_pointers(skb);
4681 if (unlikely(ret < 0))
4684 hdr = (struct ipv6hdr *)skb->data;
4685 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
4687 srh_state->hdrlen += len;
4688 srh_state->valid = 0;
4692 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
4693 .func = bpf_lwt_seg6_adjust_srh,
4695 .ret_type = RET_INTEGER,
4696 .arg1_type = ARG_PTR_TO_CTX,
4697 .arg2_type = ARG_ANYTHING,
4698 .arg3_type = ARG_ANYTHING,
4700 #endif /* CONFIG_IPV6_SEG6_BPF */
4702 bool bpf_helper_changes_pkt_data(void *func)
4704 if (func == bpf_skb_vlan_push ||
4705 func == bpf_skb_vlan_pop ||
4706 func == bpf_skb_store_bytes ||
4707 func == bpf_skb_change_proto ||
4708 func == bpf_skb_change_head ||
4709 func == sk_skb_change_head ||
4710 func == bpf_skb_change_tail ||
4711 func == sk_skb_change_tail ||
4712 func == bpf_skb_adjust_room ||
4713 func == bpf_skb_pull_data ||
4714 func == sk_skb_pull_data ||
4715 func == bpf_clone_redirect ||
4716 func == bpf_l3_csum_replace ||
4717 func == bpf_l4_csum_replace ||
4718 func == bpf_xdp_adjust_head ||
4719 func == bpf_xdp_adjust_meta ||
4720 func == bpf_msg_pull_data ||
4721 func == bpf_xdp_adjust_tail ||
4722 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4723 func == bpf_lwt_seg6_store_bytes ||
4724 func == bpf_lwt_seg6_adjust_srh ||
4725 func == bpf_lwt_seg6_action ||
4727 func == bpf_lwt_push_encap)
4733 static const struct bpf_func_proto *
4734 bpf_base_func_proto(enum bpf_func_id func_id)
4737 case BPF_FUNC_map_lookup_elem:
4738 return &bpf_map_lookup_elem_proto;
4739 case BPF_FUNC_map_update_elem:
4740 return &bpf_map_update_elem_proto;
4741 case BPF_FUNC_map_delete_elem:
4742 return &bpf_map_delete_elem_proto;
4743 case BPF_FUNC_get_prandom_u32:
4744 return &bpf_get_prandom_u32_proto;
4745 case BPF_FUNC_get_smp_processor_id:
4746 return &bpf_get_raw_smp_processor_id_proto;
4747 case BPF_FUNC_get_numa_node_id:
4748 return &bpf_get_numa_node_id_proto;
4749 case BPF_FUNC_tail_call:
4750 return &bpf_tail_call_proto;
4751 case BPF_FUNC_ktime_get_ns:
4752 return &bpf_ktime_get_ns_proto;
4753 case BPF_FUNC_trace_printk:
4754 if (capable(CAP_SYS_ADMIN))
4755 return bpf_get_trace_printk_proto();
4761 static const struct bpf_func_proto *
4762 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4765 /* inet and inet6 sockets are created in a process
4766 * context so there is always a valid uid/gid
4768 case BPF_FUNC_get_current_uid_gid:
4769 return &bpf_get_current_uid_gid_proto;
4771 return bpf_base_func_proto(func_id);
4775 static const struct bpf_func_proto *
4776 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4779 /* inet and inet6 sockets are created in a process
4780 * context so there is always a valid uid/gid
4782 case BPF_FUNC_get_current_uid_gid:
4783 return &bpf_get_current_uid_gid_proto;
4785 switch (prog->expected_attach_type) {
4786 case BPF_CGROUP_INET4_CONNECT:
4787 case BPF_CGROUP_INET6_CONNECT:
4788 return &bpf_bind_proto;
4793 return bpf_base_func_proto(func_id);
4797 static const struct bpf_func_proto *
4798 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4801 case BPF_FUNC_skb_load_bytes:
4802 return &bpf_skb_load_bytes_proto;
4803 case BPF_FUNC_skb_load_bytes_relative:
4804 return &bpf_skb_load_bytes_relative_proto;
4805 case BPF_FUNC_get_socket_cookie:
4806 return &bpf_get_socket_cookie_proto;
4807 case BPF_FUNC_get_socket_uid:
4808 return &bpf_get_socket_uid_proto;
4810 return bpf_base_func_proto(func_id);
4814 static const struct bpf_func_proto *
4815 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4818 case BPF_FUNC_skb_store_bytes:
4819 return &bpf_skb_store_bytes_proto;
4820 case BPF_FUNC_skb_load_bytes:
4821 return &bpf_skb_load_bytes_proto;
4822 case BPF_FUNC_skb_load_bytes_relative:
4823 return &bpf_skb_load_bytes_relative_proto;
4824 case BPF_FUNC_skb_pull_data:
4825 return &bpf_skb_pull_data_proto;
4826 case BPF_FUNC_csum_diff:
4827 return &bpf_csum_diff_proto;
4828 case BPF_FUNC_csum_update:
4829 return &bpf_csum_update_proto;
4830 case BPF_FUNC_l3_csum_replace:
4831 return &bpf_l3_csum_replace_proto;
4832 case BPF_FUNC_l4_csum_replace:
4833 return &bpf_l4_csum_replace_proto;
4834 case BPF_FUNC_clone_redirect:
4835 return &bpf_clone_redirect_proto;
4836 case BPF_FUNC_get_cgroup_classid:
4837 return &bpf_get_cgroup_classid_proto;
4838 case BPF_FUNC_skb_vlan_push:
4839 return &bpf_skb_vlan_push_proto;
4840 case BPF_FUNC_skb_vlan_pop:
4841 return &bpf_skb_vlan_pop_proto;
4842 case BPF_FUNC_skb_change_proto:
4843 return &bpf_skb_change_proto_proto;
4844 case BPF_FUNC_skb_change_type:
4845 return &bpf_skb_change_type_proto;
4846 case BPF_FUNC_skb_adjust_room:
4847 return &bpf_skb_adjust_room_proto;
4848 case BPF_FUNC_skb_change_tail:
4849 return &bpf_skb_change_tail_proto;
4850 case BPF_FUNC_skb_get_tunnel_key:
4851 return &bpf_skb_get_tunnel_key_proto;
4852 case BPF_FUNC_skb_set_tunnel_key:
4853 return bpf_get_skb_set_tunnel_proto(func_id);
4854 case BPF_FUNC_skb_get_tunnel_opt:
4855 return &bpf_skb_get_tunnel_opt_proto;
4856 case BPF_FUNC_skb_set_tunnel_opt:
4857 return bpf_get_skb_set_tunnel_proto(func_id);
4858 case BPF_FUNC_redirect:
4859 return &bpf_redirect_proto;
4860 case BPF_FUNC_get_route_realm:
4861 return &bpf_get_route_realm_proto;
4862 case BPF_FUNC_get_hash_recalc:
4863 return &bpf_get_hash_recalc_proto;
4864 case BPF_FUNC_set_hash_invalid:
4865 return &bpf_set_hash_invalid_proto;
4866 case BPF_FUNC_set_hash:
4867 return &bpf_set_hash_proto;
4868 case BPF_FUNC_perf_event_output:
4869 return &bpf_skb_event_output_proto;
4870 case BPF_FUNC_get_smp_processor_id:
4871 return &bpf_get_smp_processor_id_proto;
4872 case BPF_FUNC_skb_under_cgroup:
4873 return &bpf_skb_under_cgroup_proto;
4874 case BPF_FUNC_get_socket_cookie:
4875 return &bpf_get_socket_cookie_proto;
4876 case BPF_FUNC_get_socket_uid:
4877 return &bpf_get_socket_uid_proto;
4878 case BPF_FUNC_fib_lookup:
4879 return &bpf_skb_fib_lookup_proto;
4881 case BPF_FUNC_skb_get_xfrm_state:
4882 return &bpf_skb_get_xfrm_state_proto;
4884 #ifdef CONFIG_SOCK_CGROUP_DATA
4885 case BPF_FUNC_skb_cgroup_id:
4886 return &bpf_skb_cgroup_id_proto;
4889 return bpf_base_func_proto(func_id);
4893 static const struct bpf_func_proto *
4894 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4897 case BPF_FUNC_perf_event_output:
4898 return &bpf_xdp_event_output_proto;
4899 case BPF_FUNC_get_smp_processor_id:
4900 return &bpf_get_smp_processor_id_proto;
4901 case BPF_FUNC_csum_diff:
4902 return &bpf_csum_diff_proto;
4903 case BPF_FUNC_xdp_adjust_head:
4904 return &bpf_xdp_adjust_head_proto;
4905 case BPF_FUNC_xdp_adjust_meta:
4906 return &bpf_xdp_adjust_meta_proto;
4907 case BPF_FUNC_redirect:
4908 return &bpf_xdp_redirect_proto;
4909 case BPF_FUNC_redirect_map:
4910 return &bpf_xdp_redirect_map_proto;
4911 case BPF_FUNC_xdp_adjust_tail:
4912 return &bpf_xdp_adjust_tail_proto;
4913 case BPF_FUNC_fib_lookup:
4914 return &bpf_xdp_fib_lookup_proto;
4916 return bpf_base_func_proto(func_id);
4920 static const struct bpf_func_proto *
4921 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4924 case BPF_FUNC_setsockopt:
4925 return &bpf_setsockopt_proto;
4926 case BPF_FUNC_getsockopt:
4927 return &bpf_getsockopt_proto;
4928 case BPF_FUNC_sock_ops_cb_flags_set:
4929 return &bpf_sock_ops_cb_flags_set_proto;
4930 case BPF_FUNC_sock_map_update:
4931 return &bpf_sock_map_update_proto;
4932 case BPF_FUNC_sock_hash_update:
4933 return &bpf_sock_hash_update_proto;
4935 return bpf_base_func_proto(func_id);
4939 static const struct bpf_func_proto *
4940 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4943 case BPF_FUNC_msg_redirect_map:
4944 return &bpf_msg_redirect_map_proto;
4945 case BPF_FUNC_msg_redirect_hash:
4946 return &bpf_msg_redirect_hash_proto;
4947 case BPF_FUNC_msg_apply_bytes:
4948 return &bpf_msg_apply_bytes_proto;
4949 case BPF_FUNC_msg_cork_bytes:
4950 return &bpf_msg_cork_bytes_proto;
4951 case BPF_FUNC_msg_pull_data:
4952 return &bpf_msg_pull_data_proto;
4954 return bpf_base_func_proto(func_id);
4958 static const struct bpf_func_proto *
4959 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4962 case BPF_FUNC_skb_store_bytes:
4963 return &bpf_skb_store_bytes_proto;
4964 case BPF_FUNC_skb_load_bytes:
4965 return &bpf_skb_load_bytes_proto;
4966 case BPF_FUNC_skb_pull_data:
4967 return &sk_skb_pull_data_proto;
4968 case BPF_FUNC_skb_change_tail:
4969 return &sk_skb_change_tail_proto;
4970 case BPF_FUNC_skb_change_head:
4971 return &sk_skb_change_head_proto;
4972 case BPF_FUNC_get_socket_cookie:
4973 return &bpf_get_socket_cookie_proto;
4974 case BPF_FUNC_get_socket_uid:
4975 return &bpf_get_socket_uid_proto;
4976 case BPF_FUNC_sk_redirect_map:
4977 return &bpf_sk_redirect_map_proto;
4978 case BPF_FUNC_sk_redirect_hash:
4979 return &bpf_sk_redirect_hash_proto;
4981 return bpf_base_func_proto(func_id);
4985 static const struct bpf_func_proto *
4986 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4989 case BPF_FUNC_skb_load_bytes:
4990 return &bpf_skb_load_bytes_proto;
4991 case BPF_FUNC_skb_pull_data:
4992 return &bpf_skb_pull_data_proto;
4993 case BPF_FUNC_csum_diff:
4994 return &bpf_csum_diff_proto;
4995 case BPF_FUNC_get_cgroup_classid:
4996 return &bpf_get_cgroup_classid_proto;
4997 case BPF_FUNC_get_route_realm:
4998 return &bpf_get_route_realm_proto;
4999 case BPF_FUNC_get_hash_recalc:
5000 return &bpf_get_hash_recalc_proto;
5001 case BPF_FUNC_perf_event_output:
5002 return &bpf_skb_event_output_proto;
5003 case BPF_FUNC_get_smp_processor_id:
5004 return &bpf_get_smp_processor_id_proto;
5005 case BPF_FUNC_skb_under_cgroup:
5006 return &bpf_skb_under_cgroup_proto;
5008 return bpf_base_func_proto(func_id);
5012 static const struct bpf_func_proto *
5013 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5016 case BPF_FUNC_lwt_push_encap:
5017 return &bpf_lwt_push_encap_proto;
5019 return lwt_out_func_proto(func_id, prog);
5023 static const struct bpf_func_proto *
5024 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5027 case BPF_FUNC_skb_get_tunnel_key:
5028 return &bpf_skb_get_tunnel_key_proto;
5029 case BPF_FUNC_skb_set_tunnel_key:
5030 return bpf_get_skb_set_tunnel_proto(func_id);
5031 case BPF_FUNC_skb_get_tunnel_opt:
5032 return &bpf_skb_get_tunnel_opt_proto;
5033 case BPF_FUNC_skb_set_tunnel_opt:
5034 return bpf_get_skb_set_tunnel_proto(func_id);
5035 case BPF_FUNC_redirect:
5036 return &bpf_redirect_proto;
5037 case BPF_FUNC_clone_redirect:
5038 return &bpf_clone_redirect_proto;
5039 case BPF_FUNC_skb_change_tail:
5040 return &bpf_skb_change_tail_proto;
5041 case BPF_FUNC_skb_change_head:
5042 return &bpf_skb_change_head_proto;
5043 case BPF_FUNC_skb_store_bytes:
5044 return &bpf_skb_store_bytes_proto;
5045 case BPF_FUNC_csum_update:
5046 return &bpf_csum_update_proto;
5047 case BPF_FUNC_l3_csum_replace:
5048 return &bpf_l3_csum_replace_proto;
5049 case BPF_FUNC_l4_csum_replace:
5050 return &bpf_l4_csum_replace_proto;
5051 case BPF_FUNC_set_hash_invalid:
5052 return &bpf_set_hash_invalid_proto;
5054 return lwt_out_func_proto(func_id, prog);
5058 static const struct bpf_func_proto *
5059 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5062 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5063 case BPF_FUNC_lwt_seg6_store_bytes:
5064 return &bpf_lwt_seg6_store_bytes_proto;
5065 case BPF_FUNC_lwt_seg6_action:
5066 return &bpf_lwt_seg6_action_proto;
5067 case BPF_FUNC_lwt_seg6_adjust_srh:
5068 return &bpf_lwt_seg6_adjust_srh_proto;
5071 return lwt_out_func_proto(func_id, prog);
5075 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
5076 const struct bpf_prog *prog,
5077 struct bpf_insn_access_aux *info)
5079 const int size_default = sizeof(__u32);
5081 if (off < 0 || off >= sizeof(struct __sk_buff))
5084 /* The verifier guarantees that size > 0. */
5085 if (off % size != 0)
5089 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5090 if (off + size > offsetofend(struct __sk_buff, cb[4]))
5093 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
5094 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
5095 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
5096 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
5097 case bpf_ctx_range(struct __sk_buff, data):
5098 case bpf_ctx_range(struct __sk_buff, data_meta):
5099 case bpf_ctx_range(struct __sk_buff, data_end):
5100 if (size != size_default)
5104 /* Only narrow read access allowed for now. */
5105 if (type == BPF_WRITE) {
5106 if (size != size_default)
5109 bpf_ctx_record_field_size(info, size_default);
5110 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
5118 static bool sk_filter_is_valid_access(int off, int size,
5119 enum bpf_access_type type,
5120 const struct bpf_prog *prog,
5121 struct bpf_insn_access_aux *info)
5124 case bpf_ctx_range(struct __sk_buff, tc_classid):
5125 case bpf_ctx_range(struct __sk_buff, data):
5126 case bpf_ctx_range(struct __sk_buff, data_meta):
5127 case bpf_ctx_range(struct __sk_buff, data_end):
5128 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
5132 if (type == BPF_WRITE) {
5134 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5141 return bpf_skb_is_valid_access(off, size, type, prog, info);
5144 static bool lwt_is_valid_access(int off, int size,
5145 enum bpf_access_type type,
5146 const struct bpf_prog *prog,
5147 struct bpf_insn_access_aux *info)
5150 case bpf_ctx_range(struct __sk_buff, tc_classid):
5151 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
5152 case bpf_ctx_range(struct __sk_buff, data_meta):
5156 if (type == BPF_WRITE) {
5158 case bpf_ctx_range(struct __sk_buff, mark):
5159 case bpf_ctx_range(struct __sk_buff, priority):
5160 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5168 case bpf_ctx_range(struct __sk_buff, data):
5169 info->reg_type = PTR_TO_PACKET;
5171 case bpf_ctx_range(struct __sk_buff, data_end):
5172 info->reg_type = PTR_TO_PACKET_END;
5176 return bpf_skb_is_valid_access(off, size, type, prog, info);
5179 /* Attach type specific accesses */
5180 static bool __sock_filter_check_attach_type(int off,
5181 enum bpf_access_type access_type,
5182 enum bpf_attach_type attach_type)
5185 case offsetof(struct bpf_sock, bound_dev_if):
5186 case offsetof(struct bpf_sock, mark):
5187 case offsetof(struct bpf_sock, priority):
5188 switch (attach_type) {
5189 case BPF_CGROUP_INET_SOCK_CREATE:
5194 case bpf_ctx_range(struct bpf_sock, src_ip4):
5195 switch (attach_type) {
5196 case BPF_CGROUP_INET4_POST_BIND:
5201 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
5202 switch (attach_type) {
5203 case BPF_CGROUP_INET6_POST_BIND:
5208 case bpf_ctx_range(struct bpf_sock, src_port):
5209 switch (attach_type) {
5210 case BPF_CGROUP_INET4_POST_BIND:
5211 case BPF_CGROUP_INET6_POST_BIND:
5218 return access_type == BPF_READ;
5223 static bool __sock_filter_check_size(int off, int size,
5224 struct bpf_insn_access_aux *info)
5226 const int size_default = sizeof(__u32);
5229 case bpf_ctx_range(struct bpf_sock, src_ip4):
5230 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
5231 bpf_ctx_record_field_size(info, size_default);
5232 return bpf_ctx_narrow_access_ok(off, size, size_default);
5235 return size == size_default;
5238 static bool sock_filter_is_valid_access(int off, int size,
5239 enum bpf_access_type type,
5240 const struct bpf_prog *prog,
5241 struct bpf_insn_access_aux *info)
5243 if (off < 0 || off >= sizeof(struct bpf_sock))
5245 if (off % size != 0)
5247 if (!__sock_filter_check_attach_type(off, type,
5248 prog->expected_attach_type))
5250 if (!__sock_filter_check_size(off, size, info))
5255 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
5256 const struct bpf_prog *prog, int drop_verdict)
5258 struct bpf_insn *insn = insn_buf;
5263 /* if (!skb->cloned)
5266 * (Fast-path, otherwise approximation that we might be
5267 * a clone, do the rest in helper.)
5269 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
5270 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
5271 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
5273 /* ret = bpf_skb_pull_data(skb, 0); */
5274 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
5275 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
5276 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
5277 BPF_FUNC_skb_pull_data);
5280 * return TC_ACT_SHOT;
5282 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
5283 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
5284 *insn++ = BPF_EXIT_INSN();
5287 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
5289 *insn++ = prog->insnsi[0];
5291 return insn - insn_buf;
5294 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
5295 struct bpf_insn *insn_buf)
5297 bool indirect = BPF_MODE(orig->code) == BPF_IND;
5298 struct bpf_insn *insn = insn_buf;
5300 /* We're guaranteed here that CTX is in R6. */
5301 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
5303 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
5305 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
5307 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
5310 switch (BPF_SIZE(orig->code)) {
5312 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
5315 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
5318 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
5322 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
5323 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
5324 *insn++ = BPF_EXIT_INSN();
5326 return insn - insn_buf;
5329 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
5330 const struct bpf_prog *prog)
5332 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
5335 static bool tc_cls_act_is_valid_access(int off, int size,
5336 enum bpf_access_type type,
5337 const struct bpf_prog *prog,
5338 struct bpf_insn_access_aux *info)
5340 if (type == BPF_WRITE) {
5342 case bpf_ctx_range(struct __sk_buff, mark):
5343 case bpf_ctx_range(struct __sk_buff, tc_index):
5344 case bpf_ctx_range(struct __sk_buff, priority):
5345 case bpf_ctx_range(struct __sk_buff, tc_classid):
5346 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5354 case bpf_ctx_range(struct __sk_buff, data):
5355 info->reg_type = PTR_TO_PACKET;
5357 case bpf_ctx_range(struct __sk_buff, data_meta):
5358 info->reg_type = PTR_TO_PACKET_META;
5360 case bpf_ctx_range(struct __sk_buff, data_end):
5361 info->reg_type = PTR_TO_PACKET_END;
5363 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
5367 return bpf_skb_is_valid_access(off, size, type, prog, info);
5370 static bool __is_valid_xdp_access(int off, int size)
5372 if (off < 0 || off >= sizeof(struct xdp_md))
5374 if (off % size != 0)
5376 if (size != sizeof(__u32))
5382 static bool xdp_is_valid_access(int off, int size,
5383 enum bpf_access_type type,
5384 const struct bpf_prog *prog,
5385 struct bpf_insn_access_aux *info)
5387 if (type == BPF_WRITE) {
5388 if (bpf_prog_is_dev_bound(prog->aux)) {
5390 case offsetof(struct xdp_md, rx_queue_index):
5391 return __is_valid_xdp_access(off, size);
5398 case offsetof(struct xdp_md, data):
5399 info->reg_type = PTR_TO_PACKET;
5401 case offsetof(struct xdp_md, data_meta):
5402 info->reg_type = PTR_TO_PACKET_META;
5404 case offsetof(struct xdp_md, data_end):
5405 info->reg_type = PTR_TO_PACKET_END;
5409 return __is_valid_xdp_access(off, size);
5412 void bpf_warn_invalid_xdp_action(u32 act)
5414 const u32 act_max = XDP_REDIRECT;
5416 WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
5417 act > act_max ? "Illegal" : "Driver unsupported",
5420 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
5422 static bool sock_addr_is_valid_access(int off, int size,
5423 enum bpf_access_type type,
5424 const struct bpf_prog *prog,
5425 struct bpf_insn_access_aux *info)
5427 const int size_default = sizeof(__u32);
5429 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
5431 if (off % size != 0)
5434 /* Disallow access to IPv6 fields from IPv4 contex and vise
5438 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
5439 switch (prog->expected_attach_type) {
5440 case BPF_CGROUP_INET4_BIND:
5441 case BPF_CGROUP_INET4_CONNECT:
5442 case BPF_CGROUP_UDP4_SENDMSG:
5448 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
5449 switch (prog->expected_attach_type) {
5450 case BPF_CGROUP_INET6_BIND:
5451 case BPF_CGROUP_INET6_CONNECT:
5452 case BPF_CGROUP_UDP6_SENDMSG:
5458 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
5459 switch (prog->expected_attach_type) {
5460 case BPF_CGROUP_UDP4_SENDMSG:
5466 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
5468 switch (prog->expected_attach_type) {
5469 case BPF_CGROUP_UDP6_SENDMSG:
5478 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
5479 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
5480 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
5481 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
5483 /* Only narrow read access allowed for now. */
5484 if (type == BPF_READ) {
5485 bpf_ctx_record_field_size(info, size_default);
5486 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
5489 if (size != size_default)
5493 case bpf_ctx_range(struct bpf_sock_addr, user_port):
5494 if (size != size_default)
5498 if (type == BPF_READ) {
5499 if (size != size_default)
5509 static bool sock_ops_is_valid_access(int off, int size,
5510 enum bpf_access_type type,
5511 const struct bpf_prog *prog,
5512 struct bpf_insn_access_aux *info)
5514 const int size_default = sizeof(__u32);
5516 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
5519 /* The verifier guarantees that size > 0. */
5520 if (off % size != 0)
5523 if (type == BPF_WRITE) {
5525 case offsetof(struct bpf_sock_ops, reply):
5526 case offsetof(struct bpf_sock_ops, sk_txhash):
5527 if (size != size_default)
5535 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
5537 if (size != sizeof(__u64))
5541 if (size != size_default)
5550 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
5551 const struct bpf_prog *prog)
5553 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
5556 static bool sk_skb_is_valid_access(int off, int size,
5557 enum bpf_access_type type,
5558 const struct bpf_prog *prog,
5559 struct bpf_insn_access_aux *info)
5562 case bpf_ctx_range(struct __sk_buff, tc_classid):
5563 case bpf_ctx_range(struct __sk_buff, data_meta):
5567 if (type == BPF_WRITE) {
5569 case bpf_ctx_range(struct __sk_buff, tc_index):
5570 case bpf_ctx_range(struct __sk_buff, priority):
5578 case bpf_ctx_range(struct __sk_buff, mark):
5580 case bpf_ctx_range(struct __sk_buff, data):
5581 info->reg_type = PTR_TO_PACKET;
5583 case bpf_ctx_range(struct __sk_buff, data_end):
5584 info->reg_type = PTR_TO_PACKET_END;
5588 return bpf_skb_is_valid_access(off, size, type, prog, info);
5591 static bool sk_msg_is_valid_access(int off, int size,
5592 enum bpf_access_type type,
5593 const struct bpf_prog *prog,
5594 struct bpf_insn_access_aux *info)
5596 if (type == BPF_WRITE)
5600 case offsetof(struct sk_msg_md, data):
5601 info->reg_type = PTR_TO_PACKET;
5602 if (size != sizeof(__u64))
5605 case offsetof(struct sk_msg_md, data_end):
5606 info->reg_type = PTR_TO_PACKET_END;
5607 if (size != sizeof(__u64))
5611 if (size != sizeof(__u32))
5615 if (off < 0 || off >= sizeof(struct sk_msg_md))
5617 if (off % size != 0)
5623 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
5624 const struct bpf_insn *si,
5625 struct bpf_insn *insn_buf,
5626 struct bpf_prog *prog, u32 *target_size)
5628 struct bpf_insn *insn = insn_buf;
5632 case offsetof(struct __sk_buff, len):
5633 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5634 bpf_target_off(struct sk_buff, len, 4,
5638 case offsetof(struct __sk_buff, protocol):
5639 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5640 bpf_target_off(struct sk_buff, protocol, 2,
5644 case offsetof(struct __sk_buff, vlan_proto):
5645 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5646 bpf_target_off(struct sk_buff, vlan_proto, 2,
5650 case offsetof(struct __sk_buff, priority):
5651 if (type == BPF_WRITE)
5652 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
5653 bpf_target_off(struct sk_buff, priority, 4,
5656 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5657 bpf_target_off(struct sk_buff, priority, 4,
5661 case offsetof(struct __sk_buff, ingress_ifindex):
5662 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5663 bpf_target_off(struct sk_buff, skb_iif, 4,
5667 case offsetof(struct __sk_buff, ifindex):
5668 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
5669 si->dst_reg, si->src_reg,
5670 offsetof(struct sk_buff, dev));
5671 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
5672 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
5673 bpf_target_off(struct net_device, ifindex, 4,
5677 case offsetof(struct __sk_buff, hash):
5678 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5679 bpf_target_off(struct sk_buff, hash, 4,
5683 case offsetof(struct __sk_buff, mark):
5684 if (type == BPF_WRITE)
5685 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
5686 bpf_target_off(struct sk_buff, mark, 4,
5689 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5690 bpf_target_off(struct sk_buff, mark, 4,
5694 case offsetof(struct __sk_buff, pkt_type):
5696 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
5698 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
5699 #ifdef __BIG_ENDIAN_BITFIELD
5700 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
5704 case offsetof(struct __sk_buff, queue_mapping):
5705 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5706 bpf_target_off(struct sk_buff, queue_mapping, 2,
5710 case offsetof(struct __sk_buff, vlan_present):
5711 case offsetof(struct __sk_buff, vlan_tci):
5712 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000);
5714 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5715 bpf_target_off(struct sk_buff, vlan_tci, 2,
5717 if (si->off == offsetof(struct __sk_buff, vlan_tci)) {
5718 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg,
5721 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 12);
5722 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
5726 case offsetof(struct __sk_buff, cb[0]) ...
5727 offsetofend(struct __sk_buff, cb[4]) - 1:
5728 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20);
5729 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
5730 offsetof(struct qdisc_skb_cb, data)) %
5733 prog->cb_access = 1;
5735 off -= offsetof(struct __sk_buff, cb[0]);
5736 off += offsetof(struct sk_buff, cb);
5737 off += offsetof(struct qdisc_skb_cb, data);
5738 if (type == BPF_WRITE)
5739 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
5742 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
5746 case offsetof(struct __sk_buff, tc_classid):
5747 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, tc_classid) != 2);
5750 off -= offsetof(struct __sk_buff, tc_classid);
5751 off += offsetof(struct sk_buff, cb);
5752 off += offsetof(struct qdisc_skb_cb, tc_classid);
5754 if (type == BPF_WRITE)
5755 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
5758 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
5762 case offsetof(struct __sk_buff, data):
5763 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
5764 si->dst_reg, si->src_reg,
5765 offsetof(struct sk_buff, data));
5768 case offsetof(struct __sk_buff, data_meta):
5770 off -= offsetof(struct __sk_buff, data_meta);
5771 off += offsetof(struct sk_buff, cb);
5772 off += offsetof(struct bpf_skb_data_end, data_meta);
5773 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
5777 case offsetof(struct __sk_buff, data_end):
5779 off -= offsetof(struct __sk_buff, data_end);
5780 off += offsetof(struct sk_buff, cb);
5781 off += offsetof(struct bpf_skb_data_end, data_end);
5782 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
5786 case offsetof(struct __sk_buff, tc_index):
5787 #ifdef CONFIG_NET_SCHED
5788 if (type == BPF_WRITE)
5789 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
5790 bpf_target_off(struct sk_buff, tc_index, 2,
5793 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5794 bpf_target_off(struct sk_buff, tc_index, 2,
5798 if (type == BPF_WRITE)
5799 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
5801 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
5805 case offsetof(struct __sk_buff, napi_id):
5806 #if defined(CONFIG_NET_RX_BUSY_POLL)
5807 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5808 bpf_target_off(struct sk_buff, napi_id, 4,
5810 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
5811 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
5814 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
5817 case offsetof(struct __sk_buff, family):
5818 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
5820 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
5821 si->dst_reg, si->src_reg,
5822 offsetof(struct sk_buff, sk));
5823 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
5824 bpf_target_off(struct sock_common,
5828 case offsetof(struct __sk_buff, remote_ip4):
5829 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
5831 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
5832 si->dst_reg, si->src_reg,
5833 offsetof(struct sk_buff, sk));
5834 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
5835 bpf_target_off(struct sock_common,
5839 case offsetof(struct __sk_buff, local_ip4):
5840 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
5841 skc_rcv_saddr) != 4);
5843 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
5844 si->dst_reg, si->src_reg,
5845 offsetof(struct sk_buff, sk));
5846 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
5847 bpf_target_off(struct sock_common,
5851 case offsetof(struct __sk_buff, remote_ip6[0]) ...
5852 offsetof(struct __sk_buff, remote_ip6[3]):
5853 #if IS_ENABLED(CONFIG_IPV6)
5854 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
5855 skc_v6_daddr.s6_addr32[0]) != 4);
5858 off -= offsetof(struct __sk_buff, remote_ip6[0]);
5860 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
5861 si->dst_reg, si->src_reg,
5862 offsetof(struct sk_buff, sk));
5863 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
5864 offsetof(struct sock_common,
5865 skc_v6_daddr.s6_addr32[0]) +
5868 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
5871 case offsetof(struct __sk_buff, local_ip6[0]) ...
5872 offsetof(struct __sk_buff, local_ip6[3]):
5873 #if IS_ENABLED(CONFIG_IPV6)
5874 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
5875 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
5878 off -= offsetof(struct __sk_buff, local_ip6[0]);
5880 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
5881 si->dst_reg, si->src_reg,
5882 offsetof(struct sk_buff, sk));
5883 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
5884 offsetof(struct sock_common,
5885 skc_v6_rcv_saddr.s6_addr32[0]) +
5888 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
5892 case offsetof(struct __sk_buff, remote_port):
5893 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
5895 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
5896 si->dst_reg, si->src_reg,
5897 offsetof(struct sk_buff, sk));
5898 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
5899 bpf_target_off(struct sock_common,
5902 #ifndef __BIG_ENDIAN_BITFIELD
5903 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
5907 case offsetof(struct __sk_buff, local_port):
5908 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
5910 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
5911 si->dst_reg, si->src_reg,
5912 offsetof(struct sk_buff, sk));
5913 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
5914 bpf_target_off(struct sock_common,
5915 skc_num, 2, target_size));
5919 return insn - insn_buf;
5922 static u32 sock_filter_convert_ctx_access(enum bpf_access_type type,
5923 const struct bpf_insn *si,
5924 struct bpf_insn *insn_buf,
5925 struct bpf_prog *prog, u32 *target_size)
5927 struct bpf_insn *insn = insn_buf;
5931 case offsetof(struct bpf_sock, bound_dev_if):
5932 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_bound_dev_if) != 4);
5934 if (type == BPF_WRITE)
5935 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
5936 offsetof(struct sock, sk_bound_dev_if));
5938 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5939 offsetof(struct sock, sk_bound_dev_if));
5942 case offsetof(struct bpf_sock, mark):
5943 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_mark) != 4);
5945 if (type == BPF_WRITE)
5946 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
5947 offsetof(struct sock, sk_mark));
5949 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5950 offsetof(struct sock, sk_mark));
5953 case offsetof(struct bpf_sock, priority):
5954 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_priority) != 4);
5956 if (type == BPF_WRITE)
5957 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
5958 offsetof(struct sock, sk_priority));
5960 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5961 offsetof(struct sock, sk_priority));
5964 case offsetof(struct bpf_sock, family):
5965 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_family) != 2);
5967 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5968 offsetof(struct sock, sk_family));
5971 case offsetof(struct bpf_sock, type):
5972 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5973 offsetof(struct sock, __sk_flags_offset));
5974 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
5975 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
5978 case offsetof(struct bpf_sock, protocol):
5979 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5980 offsetof(struct sock, __sk_flags_offset));
5981 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
5982 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_PROTO_SHIFT);
5985 case offsetof(struct bpf_sock, src_ip4):
5986 *insn++ = BPF_LDX_MEM(
5987 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
5988 bpf_target_off(struct sock_common, skc_rcv_saddr,
5989 FIELD_SIZEOF(struct sock_common,
5994 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
5995 #if IS_ENABLED(CONFIG_IPV6)
5997 off -= offsetof(struct bpf_sock, src_ip6[0]);
5998 *insn++ = BPF_LDX_MEM(
5999 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
6002 skc_v6_rcv_saddr.s6_addr32[0],
6003 FIELD_SIZEOF(struct sock_common,
6004 skc_v6_rcv_saddr.s6_addr32[0]),
6005 target_size) + off);
6008 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6012 case offsetof(struct bpf_sock, src_port):
6013 *insn++ = BPF_LDX_MEM(
6014 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
6015 si->dst_reg, si->src_reg,
6016 bpf_target_off(struct sock_common, skc_num,
6017 FIELD_SIZEOF(struct sock_common,
6023 return insn - insn_buf;
6026 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
6027 const struct bpf_insn *si,
6028 struct bpf_insn *insn_buf,
6029 struct bpf_prog *prog, u32 *target_size)
6031 struct bpf_insn *insn = insn_buf;
6034 case offsetof(struct __sk_buff, ifindex):
6035 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
6036 si->dst_reg, si->src_reg,
6037 offsetof(struct sk_buff, dev));
6038 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6039 bpf_target_off(struct net_device, ifindex, 4,
6043 return bpf_convert_ctx_access(type, si, insn_buf, prog,
6047 return insn - insn_buf;
6050 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
6051 const struct bpf_insn *si,
6052 struct bpf_insn *insn_buf,
6053 struct bpf_prog *prog, u32 *target_size)
6055 struct bpf_insn *insn = insn_buf;
6058 case offsetof(struct xdp_md, data):
6059 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
6060 si->dst_reg, si->src_reg,
6061 offsetof(struct xdp_buff, data));
6063 case offsetof(struct xdp_md, data_meta):
6064 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
6065 si->dst_reg, si->src_reg,
6066 offsetof(struct xdp_buff, data_meta));
6068 case offsetof(struct xdp_md, data_end):
6069 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
6070 si->dst_reg, si->src_reg,
6071 offsetof(struct xdp_buff, data_end));
6073 case offsetof(struct xdp_md, ingress_ifindex):
6074 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
6075 si->dst_reg, si->src_reg,
6076 offsetof(struct xdp_buff, rxq));
6077 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
6078 si->dst_reg, si->dst_reg,
6079 offsetof(struct xdp_rxq_info, dev));
6080 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6081 offsetof(struct net_device, ifindex));
6083 case offsetof(struct xdp_md, rx_queue_index):
6084 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
6085 si->dst_reg, si->src_reg,
6086 offsetof(struct xdp_buff, rxq));
6087 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6088 offsetof(struct xdp_rxq_info,
6093 return insn - insn_buf;
6096 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
6097 * context Structure, F is Field in context structure that contains a pointer
6098 * to Nested Structure of type NS that has the field NF.
6100 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
6101 * sure that SIZE is not greater than actual size of S.F.NF.
6103 * If offset OFF is provided, the load happens from that offset relative to
6106 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
6108 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
6109 si->src_reg, offsetof(S, F)); \
6110 *insn++ = BPF_LDX_MEM( \
6111 SIZE, si->dst_reg, si->dst_reg, \
6112 bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF), \
6117 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
6118 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
6119 BPF_FIELD_SIZEOF(NS, NF), 0)
6121 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
6122 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
6124 * It doesn't support SIZE argument though since narrow stores are not
6125 * supported for now.
6127 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
6128 * "register" since two registers available in convert_ctx_access are not
6129 * enough: we can't override neither SRC, since it contains value to store, nor
6130 * DST since it contains pointer to context that may be used by later
6131 * instructions. But we need a temporary place to save pointer to nested
6132 * structure whose field we want to store to.
6134 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, OFF, TF) \
6136 int tmp_reg = BPF_REG_9; \
6137 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
6139 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
6141 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
6143 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
6144 si->dst_reg, offsetof(S, F)); \
6145 *insn++ = BPF_STX_MEM( \
6146 BPF_FIELD_SIZEOF(NS, NF), tmp_reg, si->src_reg, \
6147 bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF), \
6150 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
6154 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
6157 if (type == BPF_WRITE) { \
6158 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, OFF, \
6161 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
6162 S, NS, F, NF, SIZE, OFF); \
6166 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
6167 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
6168 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
6170 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
6171 const struct bpf_insn *si,
6172 struct bpf_insn *insn_buf,
6173 struct bpf_prog *prog, u32 *target_size)
6175 struct bpf_insn *insn = insn_buf;
6179 case offsetof(struct bpf_sock_addr, user_family):
6180 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
6181 struct sockaddr, uaddr, sa_family);
6184 case offsetof(struct bpf_sock_addr, user_ip4):
6185 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
6186 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
6187 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
6190 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
6192 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
6193 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
6194 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
6195 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
6199 case offsetof(struct bpf_sock_addr, user_port):
6200 /* To get port we need to know sa_family first and then treat
6201 * sockaddr as either sockaddr_in or sockaddr_in6.
6202 * Though we can simplify since port field has same offset and
6203 * size in both structures.
6204 * Here we check this invariant and use just one of the
6205 * structures if it's true.
6207 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
6208 offsetof(struct sockaddr_in6, sin6_port));
6209 BUILD_BUG_ON(FIELD_SIZEOF(struct sockaddr_in, sin_port) !=
6210 FIELD_SIZEOF(struct sockaddr_in6, sin6_port));
6211 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(struct bpf_sock_addr_kern,
6212 struct sockaddr_in6, uaddr,
6213 sin6_port, tmp_reg);
6216 case offsetof(struct bpf_sock_addr, family):
6217 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
6218 struct sock, sk, sk_family);
6221 case offsetof(struct bpf_sock_addr, type):
6222 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(
6223 struct bpf_sock_addr_kern, struct sock, sk,
6224 __sk_flags_offset, BPF_W, 0);
6225 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
6226 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
6229 case offsetof(struct bpf_sock_addr, protocol):
6230 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(
6231 struct bpf_sock_addr_kern, struct sock, sk,
6232 __sk_flags_offset, BPF_W, 0);
6233 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
6234 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg,
6238 case offsetof(struct bpf_sock_addr, msg_src_ip4):
6239 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
6240 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
6241 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
6242 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
6245 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
6248 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
6249 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
6250 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
6251 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
6252 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
6256 return insn - insn_buf;
6259 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
6260 const struct bpf_insn *si,
6261 struct bpf_insn *insn_buf,
6262 struct bpf_prog *prog,
6265 struct bpf_insn *insn = insn_buf;
6269 case offsetof(struct bpf_sock_ops, op) ...
6270 offsetof(struct bpf_sock_ops, replylong[3]):
6271 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, op) !=
6272 FIELD_SIZEOF(struct bpf_sock_ops_kern, op));
6273 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, reply) !=
6274 FIELD_SIZEOF(struct bpf_sock_ops_kern, reply));
6275 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, replylong) !=
6276 FIELD_SIZEOF(struct bpf_sock_ops_kern, replylong));
6278 off -= offsetof(struct bpf_sock_ops, op);
6279 off += offsetof(struct bpf_sock_ops_kern, op);
6280 if (type == BPF_WRITE)
6281 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6284 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6288 case offsetof(struct bpf_sock_ops, family):
6289 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
6291 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6292 struct bpf_sock_ops_kern, sk),
6293 si->dst_reg, si->src_reg,
6294 offsetof(struct bpf_sock_ops_kern, sk));
6295 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6296 offsetof(struct sock_common, skc_family));
6299 case offsetof(struct bpf_sock_ops, remote_ip4):
6300 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
6302 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6303 struct bpf_sock_ops_kern, sk),
6304 si->dst_reg, si->src_reg,
6305 offsetof(struct bpf_sock_ops_kern, sk));
6306 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6307 offsetof(struct sock_common, skc_daddr));
6310 case offsetof(struct bpf_sock_ops, local_ip4):
6311 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6312 skc_rcv_saddr) != 4);
6314 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6315 struct bpf_sock_ops_kern, sk),
6316 si->dst_reg, si->src_reg,
6317 offsetof(struct bpf_sock_ops_kern, sk));
6318 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6319 offsetof(struct sock_common,
6323 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
6324 offsetof(struct bpf_sock_ops, remote_ip6[3]):
6325 #if IS_ENABLED(CONFIG_IPV6)
6326 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6327 skc_v6_daddr.s6_addr32[0]) != 4);
6330 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
6331 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6332 struct bpf_sock_ops_kern, sk),
6333 si->dst_reg, si->src_reg,
6334 offsetof(struct bpf_sock_ops_kern, sk));
6335 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6336 offsetof(struct sock_common,
6337 skc_v6_daddr.s6_addr32[0]) +
6340 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6344 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
6345 offsetof(struct bpf_sock_ops, local_ip6[3]):
6346 #if IS_ENABLED(CONFIG_IPV6)
6347 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6348 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
6351 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
6352 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6353 struct bpf_sock_ops_kern, sk),
6354 si->dst_reg, si->src_reg,
6355 offsetof(struct bpf_sock_ops_kern, sk));
6356 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6357 offsetof(struct sock_common,
6358 skc_v6_rcv_saddr.s6_addr32[0]) +
6361 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6365 case offsetof(struct bpf_sock_ops, remote_port):
6366 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
6368 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6369 struct bpf_sock_ops_kern, sk),
6370 si->dst_reg, si->src_reg,
6371 offsetof(struct bpf_sock_ops_kern, sk));
6372 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6373 offsetof(struct sock_common, skc_dport));
6374 #ifndef __BIG_ENDIAN_BITFIELD
6375 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
6379 case offsetof(struct bpf_sock_ops, local_port):
6380 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
6382 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6383 struct bpf_sock_ops_kern, sk),
6384 si->dst_reg, si->src_reg,
6385 offsetof(struct bpf_sock_ops_kern, sk));
6386 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6387 offsetof(struct sock_common, skc_num));
6390 case offsetof(struct bpf_sock_ops, is_fullsock):
6391 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6392 struct bpf_sock_ops_kern,
6394 si->dst_reg, si->src_reg,
6395 offsetof(struct bpf_sock_ops_kern,
6399 case offsetof(struct bpf_sock_ops, state):
6400 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_state) != 1);
6402 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6403 struct bpf_sock_ops_kern, sk),
6404 si->dst_reg, si->src_reg,
6405 offsetof(struct bpf_sock_ops_kern, sk));
6406 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
6407 offsetof(struct sock_common, skc_state));
6410 case offsetof(struct bpf_sock_ops, rtt_min):
6411 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, rtt_min) !=
6412 sizeof(struct minmax));
6413 BUILD_BUG_ON(sizeof(struct minmax) <
6414 sizeof(struct minmax_sample));
6416 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6417 struct bpf_sock_ops_kern, sk),
6418 si->dst_reg, si->src_reg,
6419 offsetof(struct bpf_sock_ops_kern, sk));
6420 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6421 offsetof(struct tcp_sock, rtt_min) +
6422 FIELD_SIZEOF(struct minmax_sample, t));
6425 /* Helper macro for adding read access to tcp_sock or sock fields. */
6426 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
6428 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) > \
6429 FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD)); \
6430 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6431 struct bpf_sock_ops_kern, \
6433 si->dst_reg, si->src_reg, \
6434 offsetof(struct bpf_sock_ops_kern, \
6436 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 2); \
6437 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6438 struct bpf_sock_ops_kern, sk),\
6439 si->dst_reg, si->src_reg, \
6440 offsetof(struct bpf_sock_ops_kern, sk));\
6441 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
6443 si->dst_reg, si->dst_reg, \
6444 offsetof(OBJ, OBJ_FIELD)); \
6447 /* Helper macro for adding write access to tcp_sock or sock fields.
6448 * The macro is called with two registers, dst_reg which contains a pointer
6449 * to ctx (context) and src_reg which contains the value that should be
6450 * stored. However, we need an additional register since we cannot overwrite
6451 * dst_reg because it may be used later in the program.
6452 * Instead we "borrow" one of the other register. We first save its value
6453 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
6454 * it at the end of the macro.
6456 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
6458 int reg = BPF_REG_9; \
6459 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) > \
6460 FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD)); \
6461 if (si->dst_reg == reg || si->src_reg == reg) \
6463 if (si->dst_reg == reg || si->src_reg == reg) \
6465 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
6466 offsetof(struct bpf_sock_ops_kern, \
6468 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6469 struct bpf_sock_ops_kern, \
6472 offsetof(struct bpf_sock_ops_kern, \
6474 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
6475 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6476 struct bpf_sock_ops_kern, sk),\
6478 offsetof(struct bpf_sock_ops_kern, sk));\
6479 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \
6481 offsetof(OBJ, OBJ_FIELD)); \
6482 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
6483 offsetof(struct bpf_sock_ops_kern, \
6487 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
6489 if (TYPE == BPF_WRITE) \
6490 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
6492 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
6495 case offsetof(struct bpf_sock_ops, snd_cwnd):
6496 SOCK_OPS_GET_FIELD(snd_cwnd, snd_cwnd, struct tcp_sock);
6499 case offsetof(struct bpf_sock_ops, srtt_us):
6500 SOCK_OPS_GET_FIELD(srtt_us, srtt_us, struct tcp_sock);
6503 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
6504 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
6508 case offsetof(struct bpf_sock_ops, snd_ssthresh):
6509 SOCK_OPS_GET_FIELD(snd_ssthresh, snd_ssthresh, struct tcp_sock);
6512 case offsetof(struct bpf_sock_ops, rcv_nxt):
6513 SOCK_OPS_GET_FIELD(rcv_nxt, rcv_nxt, struct tcp_sock);
6516 case offsetof(struct bpf_sock_ops, snd_nxt):
6517 SOCK_OPS_GET_FIELD(snd_nxt, snd_nxt, struct tcp_sock);
6520 case offsetof(struct bpf_sock_ops, snd_una):
6521 SOCK_OPS_GET_FIELD(snd_una, snd_una, struct tcp_sock);
6524 case offsetof(struct bpf_sock_ops, mss_cache):
6525 SOCK_OPS_GET_FIELD(mss_cache, mss_cache, struct tcp_sock);
6528 case offsetof(struct bpf_sock_ops, ecn_flags):
6529 SOCK_OPS_GET_FIELD(ecn_flags, ecn_flags, struct tcp_sock);
6532 case offsetof(struct bpf_sock_ops, rate_delivered):
6533 SOCK_OPS_GET_FIELD(rate_delivered, rate_delivered,
6537 case offsetof(struct bpf_sock_ops, rate_interval_us):
6538 SOCK_OPS_GET_FIELD(rate_interval_us, rate_interval_us,
6542 case offsetof(struct bpf_sock_ops, packets_out):
6543 SOCK_OPS_GET_FIELD(packets_out, packets_out, struct tcp_sock);
6546 case offsetof(struct bpf_sock_ops, retrans_out):
6547 SOCK_OPS_GET_FIELD(retrans_out, retrans_out, struct tcp_sock);
6550 case offsetof(struct bpf_sock_ops, total_retrans):
6551 SOCK_OPS_GET_FIELD(total_retrans, total_retrans,
6555 case offsetof(struct bpf_sock_ops, segs_in):
6556 SOCK_OPS_GET_FIELD(segs_in, segs_in, struct tcp_sock);
6559 case offsetof(struct bpf_sock_ops, data_segs_in):
6560 SOCK_OPS_GET_FIELD(data_segs_in, data_segs_in, struct tcp_sock);
6563 case offsetof(struct bpf_sock_ops, segs_out):
6564 SOCK_OPS_GET_FIELD(segs_out, segs_out, struct tcp_sock);
6567 case offsetof(struct bpf_sock_ops, data_segs_out):
6568 SOCK_OPS_GET_FIELD(data_segs_out, data_segs_out,
6572 case offsetof(struct bpf_sock_ops, lost_out):
6573 SOCK_OPS_GET_FIELD(lost_out, lost_out, struct tcp_sock);
6576 case offsetof(struct bpf_sock_ops, sacked_out):
6577 SOCK_OPS_GET_FIELD(sacked_out, sacked_out, struct tcp_sock);
6580 case offsetof(struct bpf_sock_ops, sk_txhash):
6581 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
6585 case offsetof(struct bpf_sock_ops, bytes_received):
6586 SOCK_OPS_GET_FIELD(bytes_received, bytes_received,
6590 case offsetof(struct bpf_sock_ops, bytes_acked):
6591 SOCK_OPS_GET_FIELD(bytes_acked, bytes_acked, struct tcp_sock);
6595 return insn - insn_buf;
6598 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
6599 const struct bpf_insn *si,
6600 struct bpf_insn *insn_buf,
6601 struct bpf_prog *prog, u32 *target_size)
6603 struct bpf_insn *insn = insn_buf;
6607 case offsetof(struct __sk_buff, data_end):
6609 off -= offsetof(struct __sk_buff, data_end);
6610 off += offsetof(struct sk_buff, cb);
6611 off += offsetof(struct tcp_skb_cb, bpf.data_end);
6612 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
6616 return bpf_convert_ctx_access(type, si, insn_buf, prog,
6620 return insn - insn_buf;
6623 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
6624 const struct bpf_insn *si,
6625 struct bpf_insn *insn_buf,
6626 struct bpf_prog *prog, u32 *target_size)
6628 struct bpf_insn *insn = insn_buf;
6629 #if IS_ENABLED(CONFIG_IPV6)
6634 case offsetof(struct sk_msg_md, data):
6635 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_buff, data),
6636 si->dst_reg, si->src_reg,
6637 offsetof(struct sk_msg_buff, data));
6639 case offsetof(struct sk_msg_md, data_end):
6640 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_buff, data_end),
6641 si->dst_reg, si->src_reg,
6642 offsetof(struct sk_msg_buff, data_end));
6644 case offsetof(struct sk_msg_md, family):
6645 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
6647 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6648 struct sk_msg_buff, sk),
6649 si->dst_reg, si->src_reg,
6650 offsetof(struct sk_msg_buff, sk));
6651 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6652 offsetof(struct sock_common, skc_family));
6655 case offsetof(struct sk_msg_md, remote_ip4):
6656 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
6658 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6659 struct sk_msg_buff, sk),
6660 si->dst_reg, si->src_reg,
6661 offsetof(struct sk_msg_buff, sk));
6662 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6663 offsetof(struct sock_common, skc_daddr));
6666 case offsetof(struct sk_msg_md, local_ip4):
6667 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6668 skc_rcv_saddr) != 4);
6670 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6671 struct sk_msg_buff, sk),
6672 si->dst_reg, si->src_reg,
6673 offsetof(struct sk_msg_buff, sk));
6674 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6675 offsetof(struct sock_common,
6679 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
6680 offsetof(struct sk_msg_md, remote_ip6[3]):
6681 #if IS_ENABLED(CONFIG_IPV6)
6682 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6683 skc_v6_daddr.s6_addr32[0]) != 4);
6686 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
6687 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6688 struct sk_msg_buff, sk),
6689 si->dst_reg, si->src_reg,
6690 offsetof(struct sk_msg_buff, sk));
6691 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6692 offsetof(struct sock_common,
6693 skc_v6_daddr.s6_addr32[0]) +
6696 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6700 case offsetof(struct sk_msg_md, local_ip6[0]) ...
6701 offsetof(struct sk_msg_md, local_ip6[3]):
6702 #if IS_ENABLED(CONFIG_IPV6)
6703 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6704 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
6707 off -= offsetof(struct sk_msg_md, local_ip6[0]);
6708 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6709 struct sk_msg_buff, sk),
6710 si->dst_reg, si->src_reg,
6711 offsetof(struct sk_msg_buff, sk));
6712 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6713 offsetof(struct sock_common,
6714 skc_v6_rcv_saddr.s6_addr32[0]) +
6717 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6721 case offsetof(struct sk_msg_md, remote_port):
6722 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
6724 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6725 struct sk_msg_buff, sk),
6726 si->dst_reg, si->src_reg,
6727 offsetof(struct sk_msg_buff, sk));
6728 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6729 offsetof(struct sock_common, skc_dport));
6730 #ifndef __BIG_ENDIAN_BITFIELD
6731 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
6735 case offsetof(struct sk_msg_md, local_port):
6736 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
6738 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6739 struct sk_msg_buff, sk),
6740 si->dst_reg, si->src_reg,
6741 offsetof(struct sk_msg_buff, sk));
6742 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6743 offsetof(struct sock_common, skc_num));
6747 return insn - insn_buf;
6750 const struct bpf_verifier_ops sk_filter_verifier_ops = {
6751 .get_func_proto = sk_filter_func_proto,
6752 .is_valid_access = sk_filter_is_valid_access,
6753 .convert_ctx_access = bpf_convert_ctx_access,
6754 .gen_ld_abs = bpf_gen_ld_abs,
6757 const struct bpf_prog_ops sk_filter_prog_ops = {
6758 .test_run = bpf_prog_test_run_skb,
6761 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
6762 .get_func_proto = tc_cls_act_func_proto,
6763 .is_valid_access = tc_cls_act_is_valid_access,
6764 .convert_ctx_access = tc_cls_act_convert_ctx_access,
6765 .gen_prologue = tc_cls_act_prologue,
6766 .gen_ld_abs = bpf_gen_ld_abs,
6769 const struct bpf_prog_ops tc_cls_act_prog_ops = {
6770 .test_run = bpf_prog_test_run_skb,
6773 const struct bpf_verifier_ops xdp_verifier_ops = {
6774 .get_func_proto = xdp_func_proto,
6775 .is_valid_access = xdp_is_valid_access,
6776 .convert_ctx_access = xdp_convert_ctx_access,
6779 const struct bpf_prog_ops xdp_prog_ops = {
6780 .test_run = bpf_prog_test_run_xdp,
6783 const struct bpf_verifier_ops cg_skb_verifier_ops = {
6784 .get_func_proto = sk_filter_func_proto,
6785 .is_valid_access = sk_filter_is_valid_access,
6786 .convert_ctx_access = bpf_convert_ctx_access,
6789 const struct bpf_prog_ops cg_skb_prog_ops = {
6790 .test_run = bpf_prog_test_run_skb,
6793 const struct bpf_verifier_ops lwt_in_verifier_ops = {
6794 .get_func_proto = lwt_in_func_proto,
6795 .is_valid_access = lwt_is_valid_access,
6796 .convert_ctx_access = bpf_convert_ctx_access,
6799 const struct bpf_prog_ops lwt_in_prog_ops = {
6800 .test_run = bpf_prog_test_run_skb,
6803 const struct bpf_verifier_ops lwt_out_verifier_ops = {
6804 .get_func_proto = lwt_out_func_proto,
6805 .is_valid_access = lwt_is_valid_access,
6806 .convert_ctx_access = bpf_convert_ctx_access,
6809 const struct bpf_prog_ops lwt_out_prog_ops = {
6810 .test_run = bpf_prog_test_run_skb,
6813 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
6814 .get_func_proto = lwt_xmit_func_proto,
6815 .is_valid_access = lwt_is_valid_access,
6816 .convert_ctx_access = bpf_convert_ctx_access,
6817 .gen_prologue = tc_cls_act_prologue,
6820 const struct bpf_prog_ops lwt_xmit_prog_ops = {
6821 .test_run = bpf_prog_test_run_skb,
6824 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
6825 .get_func_proto = lwt_seg6local_func_proto,
6826 .is_valid_access = lwt_is_valid_access,
6827 .convert_ctx_access = bpf_convert_ctx_access,
6830 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
6831 .test_run = bpf_prog_test_run_skb,
6834 const struct bpf_verifier_ops cg_sock_verifier_ops = {
6835 .get_func_proto = sock_filter_func_proto,
6836 .is_valid_access = sock_filter_is_valid_access,
6837 .convert_ctx_access = sock_filter_convert_ctx_access,
6840 const struct bpf_prog_ops cg_sock_prog_ops = {
6843 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
6844 .get_func_proto = sock_addr_func_proto,
6845 .is_valid_access = sock_addr_is_valid_access,
6846 .convert_ctx_access = sock_addr_convert_ctx_access,
6849 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
6852 const struct bpf_verifier_ops sock_ops_verifier_ops = {
6853 .get_func_proto = sock_ops_func_proto,
6854 .is_valid_access = sock_ops_is_valid_access,
6855 .convert_ctx_access = sock_ops_convert_ctx_access,
6858 const struct bpf_prog_ops sock_ops_prog_ops = {
6861 const struct bpf_verifier_ops sk_skb_verifier_ops = {
6862 .get_func_proto = sk_skb_func_proto,
6863 .is_valid_access = sk_skb_is_valid_access,
6864 .convert_ctx_access = sk_skb_convert_ctx_access,
6865 .gen_prologue = sk_skb_prologue,
6868 const struct bpf_prog_ops sk_skb_prog_ops = {
6871 const struct bpf_verifier_ops sk_msg_verifier_ops = {
6872 .get_func_proto = sk_msg_func_proto,
6873 .is_valid_access = sk_msg_is_valid_access,
6874 .convert_ctx_access = sk_msg_convert_ctx_access,
6877 const struct bpf_prog_ops sk_msg_prog_ops = {
6880 int sk_detach_filter(struct sock *sk)
6883 struct sk_filter *filter;
6885 if (sock_flag(sk, SOCK_FILTER_LOCKED))
6888 filter = rcu_dereference_protected(sk->sk_filter,
6889 lockdep_sock_is_held(sk));
6891 RCU_INIT_POINTER(sk->sk_filter, NULL);
6892 sk_filter_uncharge(sk, filter);
6898 EXPORT_SYMBOL_GPL(sk_detach_filter);
6900 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
6903 struct sock_fprog_kern *fprog;
6904 struct sk_filter *filter;
6908 filter = rcu_dereference_protected(sk->sk_filter,
6909 lockdep_sock_is_held(sk));
6913 /* We're copying the filter that has been originally attached,
6914 * so no conversion/decode needed anymore. eBPF programs that
6915 * have no original program cannot be dumped through this.
6918 fprog = filter->prog->orig_prog;
6924 /* User space only enquires number of filter blocks. */
6928 if (len < fprog->len)
6932 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
6935 /* Instead of bytes, the API requests to return the number