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 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
463 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
464 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP, size, 2 + endian);
465 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A, BPF_REG_D,
468 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
469 *insn++ = BPF_JMP_A(8);
472 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
473 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
474 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
476 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
478 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
480 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
483 switch (BPF_SIZE(fp->code)) {
485 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
488 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
491 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
497 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
498 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
499 *insn = BPF_EXIT_INSN();
506 * bpf_convert_filter - convert filter program
507 * @prog: the user passed filter program
508 * @len: the length of the user passed filter program
509 * @new_prog: allocated 'struct bpf_prog' or NULL
510 * @new_len: pointer to store length of converted program
511 * @seen_ld_abs: bool whether we've seen ld_abs/ind
513 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
514 * style extended BPF (eBPF).
515 * Conversion workflow:
517 * 1) First pass for calculating the new program length:
518 * bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
520 * 2) 2nd pass to remap in two passes: 1st pass finds new
521 * jump offsets, 2nd pass remapping:
522 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
524 static int bpf_convert_filter(struct sock_filter *prog, int len,
525 struct bpf_prog *new_prog, int *new_len,
528 int new_flen = 0, pass = 0, target, i, stack_off;
529 struct bpf_insn *new_insn, *first_insn = NULL;
530 struct sock_filter *fp;
534 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
535 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
537 if (len <= 0 || len > BPF_MAXINSNS)
541 first_insn = new_prog->insnsi;
542 addrs = kcalloc(len, sizeof(*addrs),
543 GFP_KERNEL | __GFP_NOWARN);
549 new_insn = first_insn;
552 /* Classic BPF related prologue emission. */
554 /* Classic BPF expects A and X to be reset first. These need
555 * to be guaranteed to be the first two instructions.
557 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
558 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
560 /* All programs must keep CTX in callee saved BPF_REG_CTX.
561 * In eBPF case it's done by the compiler, here we need to
562 * do this ourself. Initial CTX is present in BPF_REG_ARG1.
564 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
566 /* For packet access in classic BPF, cache skb->data
567 * in callee-saved BPF R8 and skb->len - skb->data_len
568 * (headlen) in BPF R9. Since classic BPF is read-only
569 * on CTX, we only need to cache it once.
571 *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
572 BPF_REG_D, BPF_REG_CTX,
573 offsetof(struct sk_buff, data));
574 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
575 offsetof(struct sk_buff, len));
576 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
577 offsetof(struct sk_buff, data_len));
578 *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
584 for (i = 0; i < len; fp++, i++) {
585 struct bpf_insn tmp_insns[32] = { };
586 struct bpf_insn *insn = tmp_insns;
589 addrs[i] = new_insn - first_insn;
592 /* All arithmetic insns and skb loads map as-is. */
593 case BPF_ALU | BPF_ADD | BPF_X:
594 case BPF_ALU | BPF_ADD | BPF_K:
595 case BPF_ALU | BPF_SUB | BPF_X:
596 case BPF_ALU | BPF_SUB | BPF_K:
597 case BPF_ALU | BPF_AND | BPF_X:
598 case BPF_ALU | BPF_AND | BPF_K:
599 case BPF_ALU | BPF_OR | BPF_X:
600 case BPF_ALU | BPF_OR | BPF_K:
601 case BPF_ALU | BPF_LSH | BPF_X:
602 case BPF_ALU | BPF_LSH | BPF_K:
603 case BPF_ALU | BPF_RSH | BPF_X:
604 case BPF_ALU | BPF_RSH | BPF_K:
605 case BPF_ALU | BPF_XOR | BPF_X:
606 case BPF_ALU | BPF_XOR | BPF_K:
607 case BPF_ALU | BPF_MUL | BPF_X:
608 case BPF_ALU | BPF_MUL | BPF_K:
609 case BPF_ALU | BPF_DIV | BPF_X:
610 case BPF_ALU | BPF_DIV | BPF_K:
611 case BPF_ALU | BPF_MOD | BPF_X:
612 case BPF_ALU | BPF_MOD | BPF_K:
613 case BPF_ALU | BPF_NEG:
614 case BPF_LD | BPF_ABS | BPF_W:
615 case BPF_LD | BPF_ABS | BPF_H:
616 case BPF_LD | BPF_ABS | BPF_B:
617 case BPF_LD | BPF_IND | BPF_W:
618 case BPF_LD | BPF_IND | BPF_H:
619 case BPF_LD | BPF_IND | BPF_B:
620 /* Check for overloaded BPF extension and
621 * directly convert it if found, otherwise
622 * just move on with mapping.
624 if (BPF_CLASS(fp->code) == BPF_LD &&
625 BPF_MODE(fp->code) == BPF_ABS &&
626 convert_bpf_extensions(fp, &insn))
628 if (BPF_CLASS(fp->code) == BPF_LD &&
629 convert_bpf_ld_abs(fp, &insn)) {
634 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
635 fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
636 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
637 /* Error with exception code on div/mod by 0.
638 * For cBPF programs, this was always return 0.
640 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
641 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
642 *insn++ = BPF_EXIT_INSN();
645 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
648 /* Jump transformation cannot use BPF block macros
649 * everywhere as offset calculation and target updates
650 * require a bit more work than the rest, i.e. jump
651 * opcodes map as-is, but offsets need adjustment.
654 #define BPF_EMIT_JMP \
656 const s32 off_min = S16_MIN, off_max = S16_MAX; \
659 if (target >= len || target < 0) \
661 off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
662 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
663 off -= insn - tmp_insns; \
664 /* Reject anything not fitting into insn->off. */ \
665 if (off < off_min || off > off_max) \
670 case BPF_JMP | BPF_JA:
671 target = i + fp->k + 1;
672 insn->code = fp->code;
676 case BPF_JMP | BPF_JEQ | BPF_K:
677 case BPF_JMP | BPF_JEQ | BPF_X:
678 case BPF_JMP | BPF_JSET | BPF_K:
679 case BPF_JMP | BPF_JSET | BPF_X:
680 case BPF_JMP | BPF_JGT | BPF_K:
681 case BPF_JMP | BPF_JGT | BPF_X:
682 case BPF_JMP | BPF_JGE | BPF_K:
683 case BPF_JMP | BPF_JGE | BPF_X:
684 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
685 /* BPF immediates are signed, zero extend
686 * immediate into tmp register and use it
689 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
691 insn->dst_reg = BPF_REG_A;
692 insn->src_reg = BPF_REG_TMP;
695 insn->dst_reg = BPF_REG_A;
697 bpf_src = BPF_SRC(fp->code);
698 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
701 /* Common case where 'jump_false' is next insn. */
703 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
704 target = i + fp->jt + 1;
709 /* Convert some jumps when 'jump_true' is next insn. */
711 switch (BPF_OP(fp->code)) {
713 insn->code = BPF_JMP | BPF_JNE | bpf_src;
716 insn->code = BPF_JMP | BPF_JLE | bpf_src;
719 insn->code = BPF_JMP | BPF_JLT | bpf_src;
725 target = i + fp->jf + 1;
730 /* Other jumps are mapped into two insns: Jxx and JA. */
731 target = i + fp->jt + 1;
732 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
736 insn->code = BPF_JMP | BPF_JA;
737 target = i + fp->jf + 1;
741 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
742 case BPF_LDX | BPF_MSH | BPF_B: {
743 struct sock_filter tmp = {
744 .code = BPF_LD | BPF_ABS | BPF_B,
751 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
752 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
753 convert_bpf_ld_abs(&tmp, &insn);
756 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
758 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
760 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
762 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
764 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
767 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
768 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
770 case BPF_RET | BPF_A:
771 case BPF_RET | BPF_K:
772 if (BPF_RVAL(fp->code) == BPF_K)
773 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
775 *insn = BPF_EXIT_INSN();
778 /* Store to stack. */
781 stack_off = fp->k * 4 + 4;
782 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
783 BPF_ST ? BPF_REG_A : BPF_REG_X,
785 /* check_load_and_stores() verifies that classic BPF can
786 * load from stack only after write, so tracking
787 * stack_depth for ST|STX insns is enough
789 if (new_prog && new_prog->aux->stack_depth < stack_off)
790 new_prog->aux->stack_depth = stack_off;
793 /* Load from stack. */
794 case BPF_LD | BPF_MEM:
795 case BPF_LDX | BPF_MEM:
796 stack_off = fp->k * 4 + 4;
797 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
798 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
803 case BPF_LD | BPF_IMM:
804 case BPF_LDX | BPF_IMM:
805 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
806 BPF_REG_A : BPF_REG_X, fp->k);
810 case BPF_MISC | BPF_TAX:
811 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
815 case BPF_MISC | BPF_TXA:
816 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
819 /* A = skb->len or X = skb->len */
820 case BPF_LD | BPF_W | BPF_LEN:
821 case BPF_LDX | BPF_W | BPF_LEN:
822 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
823 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
824 offsetof(struct sk_buff, len));
827 /* Access seccomp_data fields. */
828 case BPF_LDX | BPF_ABS | BPF_W:
829 /* A = *(u32 *) (ctx + K) */
830 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
833 /* Unknown instruction. */
840 memcpy(new_insn, tmp_insns,
841 sizeof(*insn) * (insn - tmp_insns));
842 new_insn += insn - tmp_insns;
846 /* Only calculating new length. */
847 *new_len = new_insn - first_insn;
849 *new_len += 4; /* Prologue bits. */
854 if (new_flen != new_insn - first_insn) {
855 new_flen = new_insn - first_insn;
862 BUG_ON(*new_len != new_flen);
871 * As we dont want to clear mem[] array for each packet going through
872 * __bpf_prog_run(), we check that filter loaded by user never try to read
873 * a cell if not previously written, and we check all branches to be sure
874 * a malicious user doesn't try to abuse us.
876 static int check_load_and_stores(const struct sock_filter *filter, int flen)
878 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
881 BUILD_BUG_ON(BPF_MEMWORDS > 16);
883 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
887 memset(masks, 0xff, flen * sizeof(*masks));
889 for (pc = 0; pc < flen; pc++) {
890 memvalid &= masks[pc];
892 switch (filter[pc].code) {
895 memvalid |= (1 << filter[pc].k);
897 case BPF_LD | BPF_MEM:
898 case BPF_LDX | BPF_MEM:
899 if (!(memvalid & (1 << filter[pc].k))) {
904 case BPF_JMP | BPF_JA:
905 /* A jump must set masks on target */
906 masks[pc + 1 + filter[pc].k] &= memvalid;
909 case BPF_JMP | BPF_JEQ | BPF_K:
910 case BPF_JMP | BPF_JEQ | BPF_X:
911 case BPF_JMP | BPF_JGE | BPF_K:
912 case BPF_JMP | BPF_JGE | BPF_X:
913 case BPF_JMP | BPF_JGT | BPF_K:
914 case BPF_JMP | BPF_JGT | BPF_X:
915 case BPF_JMP | BPF_JSET | BPF_K:
916 case BPF_JMP | BPF_JSET | BPF_X:
917 /* A jump must set masks on targets */
918 masks[pc + 1 + filter[pc].jt] &= memvalid;
919 masks[pc + 1 + filter[pc].jf] &= memvalid;
929 static bool chk_code_allowed(u16 code_to_probe)
931 static const bool codes[] = {
932 /* 32 bit ALU operations */
933 [BPF_ALU | BPF_ADD | BPF_K] = true,
934 [BPF_ALU | BPF_ADD | BPF_X] = true,
935 [BPF_ALU | BPF_SUB | BPF_K] = true,
936 [BPF_ALU | BPF_SUB | BPF_X] = true,
937 [BPF_ALU | BPF_MUL | BPF_K] = true,
938 [BPF_ALU | BPF_MUL | BPF_X] = true,
939 [BPF_ALU | BPF_DIV | BPF_K] = true,
940 [BPF_ALU | BPF_DIV | BPF_X] = true,
941 [BPF_ALU | BPF_MOD | BPF_K] = true,
942 [BPF_ALU | BPF_MOD | BPF_X] = true,
943 [BPF_ALU | BPF_AND | BPF_K] = true,
944 [BPF_ALU | BPF_AND | BPF_X] = true,
945 [BPF_ALU | BPF_OR | BPF_K] = true,
946 [BPF_ALU | BPF_OR | BPF_X] = true,
947 [BPF_ALU | BPF_XOR | BPF_K] = true,
948 [BPF_ALU | BPF_XOR | BPF_X] = true,
949 [BPF_ALU | BPF_LSH | BPF_K] = true,
950 [BPF_ALU | BPF_LSH | BPF_X] = true,
951 [BPF_ALU | BPF_RSH | BPF_K] = true,
952 [BPF_ALU | BPF_RSH | BPF_X] = true,
953 [BPF_ALU | BPF_NEG] = true,
954 /* Load instructions */
955 [BPF_LD | BPF_W | BPF_ABS] = true,
956 [BPF_LD | BPF_H | BPF_ABS] = true,
957 [BPF_LD | BPF_B | BPF_ABS] = true,
958 [BPF_LD | BPF_W | BPF_LEN] = true,
959 [BPF_LD | BPF_W | BPF_IND] = true,
960 [BPF_LD | BPF_H | BPF_IND] = true,
961 [BPF_LD | BPF_B | BPF_IND] = true,
962 [BPF_LD | BPF_IMM] = true,
963 [BPF_LD | BPF_MEM] = true,
964 [BPF_LDX | BPF_W | BPF_LEN] = true,
965 [BPF_LDX | BPF_B | BPF_MSH] = true,
966 [BPF_LDX | BPF_IMM] = true,
967 [BPF_LDX | BPF_MEM] = true,
968 /* Store instructions */
971 /* Misc instructions */
972 [BPF_MISC | BPF_TAX] = true,
973 [BPF_MISC | BPF_TXA] = true,
974 /* Return instructions */
975 [BPF_RET | BPF_K] = true,
976 [BPF_RET | BPF_A] = true,
977 /* Jump instructions */
978 [BPF_JMP | BPF_JA] = true,
979 [BPF_JMP | BPF_JEQ | BPF_K] = true,
980 [BPF_JMP | BPF_JEQ | BPF_X] = true,
981 [BPF_JMP | BPF_JGE | BPF_K] = true,
982 [BPF_JMP | BPF_JGE | BPF_X] = true,
983 [BPF_JMP | BPF_JGT | BPF_K] = true,
984 [BPF_JMP | BPF_JGT | BPF_X] = true,
985 [BPF_JMP | BPF_JSET | BPF_K] = true,
986 [BPF_JMP | BPF_JSET | BPF_X] = true,
989 if (code_to_probe >= ARRAY_SIZE(codes))
992 return codes[code_to_probe];
995 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1000 if (flen == 0 || flen > BPF_MAXINSNS)
1007 * bpf_check_classic - verify socket filter code
1008 * @filter: filter to verify
1009 * @flen: length of filter
1011 * Check the user's filter code. If we let some ugly
1012 * filter code slip through kaboom! The filter must contain
1013 * no references or jumps that are out of range, no illegal
1014 * instructions, and must end with a RET instruction.
1016 * All jumps are forward as they are not signed.
1018 * Returns 0 if the rule set is legal or -EINVAL if not.
1020 static int bpf_check_classic(const struct sock_filter *filter,
1026 /* Check the filter code now */
1027 for (pc = 0; pc < flen; pc++) {
1028 const struct sock_filter *ftest = &filter[pc];
1030 /* May we actually operate on this code? */
1031 if (!chk_code_allowed(ftest->code))
1034 /* Some instructions need special checks */
1035 switch (ftest->code) {
1036 case BPF_ALU | BPF_DIV | BPF_K:
1037 case BPF_ALU | BPF_MOD | BPF_K:
1038 /* Check for division by zero */
1042 case BPF_ALU | BPF_LSH | BPF_K:
1043 case BPF_ALU | BPF_RSH | BPF_K:
1047 case BPF_LD | BPF_MEM:
1048 case BPF_LDX | BPF_MEM:
1051 /* Check for invalid memory addresses */
1052 if (ftest->k >= BPF_MEMWORDS)
1055 case BPF_JMP | BPF_JA:
1056 /* Note, the large ftest->k might cause loops.
1057 * Compare this with conditional jumps below,
1058 * where offsets are limited. --ANK (981016)
1060 if (ftest->k >= (unsigned int)(flen - pc - 1))
1063 case BPF_JMP | BPF_JEQ | BPF_K:
1064 case BPF_JMP | BPF_JEQ | BPF_X:
1065 case BPF_JMP | BPF_JGE | BPF_K:
1066 case BPF_JMP | BPF_JGE | BPF_X:
1067 case BPF_JMP | BPF_JGT | BPF_K:
1068 case BPF_JMP | BPF_JGT | BPF_X:
1069 case BPF_JMP | BPF_JSET | BPF_K:
1070 case BPF_JMP | BPF_JSET | BPF_X:
1071 /* Both conditionals must be safe */
1072 if (pc + ftest->jt + 1 >= flen ||
1073 pc + ftest->jf + 1 >= flen)
1076 case BPF_LD | BPF_W | BPF_ABS:
1077 case BPF_LD | BPF_H | BPF_ABS:
1078 case BPF_LD | BPF_B | BPF_ABS:
1080 if (bpf_anc_helper(ftest) & BPF_ANC)
1082 /* Ancillary operation unknown or unsupported */
1083 if (anc_found == false && ftest->k >= SKF_AD_OFF)
1088 /* Last instruction must be a RET code */
1089 switch (filter[flen - 1].code) {
1090 case BPF_RET | BPF_K:
1091 case BPF_RET | BPF_A:
1092 return check_load_and_stores(filter, flen);
1098 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1099 const struct sock_fprog *fprog)
1101 unsigned int fsize = bpf_classic_proglen(fprog);
1102 struct sock_fprog_kern *fkprog;
1104 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1108 fkprog = fp->orig_prog;
1109 fkprog->len = fprog->len;
1111 fkprog->filter = kmemdup(fp->insns, fsize,
1112 GFP_KERNEL | __GFP_NOWARN);
1113 if (!fkprog->filter) {
1114 kfree(fp->orig_prog);
1121 static void bpf_release_orig_filter(struct bpf_prog *fp)
1123 struct sock_fprog_kern *fprog = fp->orig_prog;
1126 kfree(fprog->filter);
1131 static void __bpf_prog_release(struct bpf_prog *prog)
1133 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1136 bpf_release_orig_filter(prog);
1137 bpf_prog_free(prog);
1141 static void __sk_filter_release(struct sk_filter *fp)
1143 __bpf_prog_release(fp->prog);
1148 * sk_filter_release_rcu - Release a socket filter by rcu_head
1149 * @rcu: rcu_head that contains the sk_filter to free
1151 static void sk_filter_release_rcu(struct rcu_head *rcu)
1153 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1155 __sk_filter_release(fp);
1159 * sk_filter_release - release a socket filter
1160 * @fp: filter to remove
1162 * Remove a filter from a socket and release its resources.
1164 static void sk_filter_release(struct sk_filter *fp)
1166 if (refcount_dec_and_test(&fp->refcnt))
1167 call_rcu(&fp->rcu, sk_filter_release_rcu);
1170 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1172 u32 filter_size = bpf_prog_size(fp->prog->len);
1174 atomic_sub(filter_size, &sk->sk_omem_alloc);
1175 sk_filter_release(fp);
1178 /* try to charge the socket memory if there is space available
1179 * return true on success
1181 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1183 u32 filter_size = bpf_prog_size(fp->prog->len);
1185 /* same check as in sock_kmalloc() */
1186 if (filter_size <= sysctl_optmem_max &&
1187 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
1188 atomic_add(filter_size, &sk->sk_omem_alloc);
1194 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1196 if (!refcount_inc_not_zero(&fp->refcnt))
1199 if (!__sk_filter_charge(sk, fp)) {
1200 sk_filter_release(fp);
1206 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1208 struct sock_filter *old_prog;
1209 struct bpf_prog *old_fp;
1210 int err, new_len, old_len = fp->len;
1211 bool seen_ld_abs = false;
1213 /* We are free to overwrite insns et al right here as it
1214 * won't be used at this point in time anymore internally
1215 * after the migration to the internal BPF instruction
1218 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1219 sizeof(struct bpf_insn));
1221 /* Conversion cannot happen on overlapping memory areas,
1222 * so we need to keep the user BPF around until the 2nd
1223 * pass. At this time, the user BPF is stored in fp->insns.
1225 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1226 GFP_KERNEL | __GFP_NOWARN);
1232 /* 1st pass: calculate the new program length. */
1233 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1238 /* Expand fp for appending the new filter representation. */
1240 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1242 /* The old_fp is still around in case we couldn't
1243 * allocate new memory, so uncharge on that one.
1252 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1253 err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1256 /* 2nd bpf_convert_filter() can fail only if it fails
1257 * to allocate memory, remapping must succeed. Note,
1258 * that at this time old_fp has already been released
1263 fp = bpf_prog_select_runtime(fp, &err);
1273 __bpf_prog_release(fp);
1274 return ERR_PTR(err);
1277 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1278 bpf_aux_classic_check_t trans)
1282 fp->bpf_func = NULL;
1285 err = bpf_check_classic(fp->insns, fp->len);
1287 __bpf_prog_release(fp);
1288 return ERR_PTR(err);
1291 /* There might be additional checks and transformations
1292 * needed on classic filters, f.e. in case of seccomp.
1295 err = trans(fp->insns, fp->len);
1297 __bpf_prog_release(fp);
1298 return ERR_PTR(err);
1302 /* Probe if we can JIT compile the filter and if so, do
1303 * the compilation of the filter.
1305 bpf_jit_compile(fp);
1307 /* JIT compiler couldn't process this filter, so do the
1308 * internal BPF translation for the optimized interpreter.
1311 fp = bpf_migrate_filter(fp);
1317 * bpf_prog_create - create an unattached filter
1318 * @pfp: the unattached filter that is created
1319 * @fprog: the filter program
1321 * Create a filter independent of any socket. We first run some
1322 * sanity checks on it to make sure it does not explode on us later.
1323 * If an error occurs or there is insufficient memory for the filter
1324 * a negative errno code is returned. On success the return is zero.
1326 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1328 unsigned int fsize = bpf_classic_proglen(fprog);
1329 struct bpf_prog *fp;
1331 /* Make sure new filter is there and in the right amounts. */
1332 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1335 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1339 memcpy(fp->insns, fprog->filter, fsize);
1341 fp->len = fprog->len;
1342 /* Since unattached filters are not copied back to user
1343 * space through sk_get_filter(), we do not need to hold
1344 * a copy here, and can spare us the work.
1346 fp->orig_prog = NULL;
1348 /* bpf_prepare_filter() already takes care of freeing
1349 * memory in case something goes wrong.
1351 fp = bpf_prepare_filter(fp, NULL);
1358 EXPORT_SYMBOL_GPL(bpf_prog_create);
1361 * bpf_prog_create_from_user - create an unattached filter from user buffer
1362 * @pfp: the unattached filter that is created
1363 * @fprog: the filter program
1364 * @trans: post-classic verifier transformation handler
1365 * @save_orig: save classic BPF program
1367 * This function effectively does the same as bpf_prog_create(), only
1368 * that it builds up its insns buffer from user space provided buffer.
1369 * It also allows for passing a bpf_aux_classic_check_t handler.
1371 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1372 bpf_aux_classic_check_t trans, bool save_orig)
1374 unsigned int fsize = bpf_classic_proglen(fprog);
1375 struct bpf_prog *fp;
1378 /* Make sure new filter is there and in the right amounts. */
1379 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1382 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1386 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1387 __bpf_prog_free(fp);
1391 fp->len = fprog->len;
1392 fp->orig_prog = NULL;
1395 err = bpf_prog_store_orig_filter(fp, fprog);
1397 __bpf_prog_free(fp);
1402 /* bpf_prepare_filter() already takes care of freeing
1403 * memory in case something goes wrong.
1405 fp = bpf_prepare_filter(fp, trans);
1412 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1414 void bpf_prog_destroy(struct bpf_prog *fp)
1416 __bpf_prog_release(fp);
1418 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1420 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1422 struct sk_filter *fp, *old_fp;
1424 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1430 if (!__sk_filter_charge(sk, fp)) {
1434 refcount_set(&fp->refcnt, 1);
1436 old_fp = rcu_dereference_protected(sk->sk_filter,
1437 lockdep_sock_is_held(sk));
1438 rcu_assign_pointer(sk->sk_filter, fp);
1441 sk_filter_uncharge(sk, old_fp);
1446 static int __reuseport_attach_prog(struct bpf_prog *prog, struct sock *sk)
1448 struct bpf_prog *old_prog;
1451 if (bpf_prog_size(prog->len) > sysctl_optmem_max)
1454 if (sk_unhashed(sk) && sk->sk_reuseport) {
1455 err = reuseport_alloc(sk);
1458 } else if (!rcu_access_pointer(sk->sk_reuseport_cb)) {
1459 /* The socket wasn't bound with SO_REUSEPORT */
1463 old_prog = reuseport_attach_prog(sk, prog);
1465 bpf_prog_destroy(old_prog);
1471 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1473 unsigned int fsize = bpf_classic_proglen(fprog);
1474 struct bpf_prog *prog;
1477 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1478 return ERR_PTR(-EPERM);
1480 /* Make sure new filter is there and in the right amounts. */
1481 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1482 return ERR_PTR(-EINVAL);
1484 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1486 return ERR_PTR(-ENOMEM);
1488 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1489 __bpf_prog_free(prog);
1490 return ERR_PTR(-EFAULT);
1493 prog->len = fprog->len;
1495 err = bpf_prog_store_orig_filter(prog, fprog);
1497 __bpf_prog_free(prog);
1498 return ERR_PTR(-ENOMEM);
1501 /* bpf_prepare_filter() already takes care of freeing
1502 * memory in case something goes wrong.
1504 return bpf_prepare_filter(prog, NULL);
1508 * sk_attach_filter - attach a socket filter
1509 * @fprog: the filter program
1510 * @sk: the socket to use
1512 * Attach the user's filter code. We first run some sanity checks on
1513 * it to make sure it does not explode on us later. If an error
1514 * occurs or there is insufficient memory for the filter a negative
1515 * errno code is returned. On success the return is zero.
1517 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1519 struct bpf_prog *prog = __get_filter(fprog, sk);
1523 return PTR_ERR(prog);
1525 err = __sk_attach_prog(prog, sk);
1527 __bpf_prog_release(prog);
1533 EXPORT_SYMBOL_GPL(sk_attach_filter);
1535 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1537 struct bpf_prog *prog = __get_filter(fprog, sk);
1541 return PTR_ERR(prog);
1543 err = __reuseport_attach_prog(prog, sk);
1545 __bpf_prog_release(prog);
1552 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1554 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1555 return ERR_PTR(-EPERM);
1557 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1560 int sk_attach_bpf(u32 ufd, struct sock *sk)
1562 struct bpf_prog *prog = __get_bpf(ufd, sk);
1566 return PTR_ERR(prog);
1568 err = __sk_attach_prog(prog, sk);
1577 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1579 struct bpf_prog *prog = __get_bpf(ufd, sk);
1583 return PTR_ERR(prog);
1585 err = __reuseport_attach_prog(prog, sk);
1594 struct bpf_scratchpad {
1596 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1597 u8 buff[MAX_BPF_STACK];
1601 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1603 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1604 unsigned int write_len)
1606 return skb_ensure_writable(skb, write_len);
1609 static inline int bpf_try_make_writable(struct sk_buff *skb,
1610 unsigned int write_len)
1612 int err = __bpf_try_make_writable(skb, write_len);
1614 bpf_compute_data_pointers(skb);
1618 static int bpf_try_make_head_writable(struct sk_buff *skb)
1620 return bpf_try_make_writable(skb, skb_headlen(skb));
1623 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1625 if (skb_at_tc_ingress(skb))
1626 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1629 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1631 if (skb_at_tc_ingress(skb))
1632 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1635 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1636 const void *, from, u32, len, u64, flags)
1640 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1642 if (unlikely(offset > 0xffff))
1644 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1647 ptr = skb->data + offset;
1648 if (flags & BPF_F_RECOMPUTE_CSUM)
1649 __skb_postpull_rcsum(skb, ptr, len, offset);
1651 memcpy(ptr, from, len);
1653 if (flags & BPF_F_RECOMPUTE_CSUM)
1654 __skb_postpush_rcsum(skb, ptr, len, offset);
1655 if (flags & BPF_F_INVALIDATE_HASH)
1656 skb_clear_hash(skb);
1661 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1662 .func = bpf_skb_store_bytes,
1664 .ret_type = RET_INTEGER,
1665 .arg1_type = ARG_PTR_TO_CTX,
1666 .arg2_type = ARG_ANYTHING,
1667 .arg3_type = ARG_PTR_TO_MEM,
1668 .arg4_type = ARG_CONST_SIZE,
1669 .arg5_type = ARG_ANYTHING,
1672 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1673 void *, to, u32, len)
1677 if (unlikely(offset > 0xffff))
1680 ptr = skb_header_pointer(skb, offset, len, to);
1684 memcpy(to, ptr, len);
1692 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1693 .func = bpf_skb_load_bytes,
1695 .ret_type = RET_INTEGER,
1696 .arg1_type = ARG_PTR_TO_CTX,
1697 .arg2_type = ARG_ANYTHING,
1698 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1699 .arg4_type = ARG_CONST_SIZE,
1702 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1703 u32, offset, void *, to, u32, len, u32, start_header)
1707 if (unlikely(offset > 0xffff || len > skb_headlen(skb)))
1710 switch (start_header) {
1711 case BPF_HDR_START_MAC:
1712 ptr = skb_mac_header(skb) + offset;
1714 case BPF_HDR_START_NET:
1715 ptr = skb_network_header(skb) + offset;
1721 if (likely(ptr >= skb_mac_header(skb) &&
1722 ptr + len <= skb_tail_pointer(skb))) {
1723 memcpy(to, ptr, len);
1732 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1733 .func = bpf_skb_load_bytes_relative,
1735 .ret_type = RET_INTEGER,
1736 .arg1_type = ARG_PTR_TO_CTX,
1737 .arg2_type = ARG_ANYTHING,
1738 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1739 .arg4_type = ARG_CONST_SIZE,
1740 .arg5_type = ARG_ANYTHING,
1743 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1745 /* Idea is the following: should the needed direct read/write
1746 * test fail during runtime, we can pull in more data and redo
1747 * again, since implicitly, we invalidate previous checks here.
1749 * Or, since we know how much we need to make read/writeable,
1750 * this can be done once at the program beginning for direct
1751 * access case. By this we overcome limitations of only current
1752 * headroom being accessible.
1754 return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1757 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1758 .func = bpf_skb_pull_data,
1760 .ret_type = RET_INTEGER,
1761 .arg1_type = ARG_PTR_TO_CTX,
1762 .arg2_type = ARG_ANYTHING,
1765 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1766 unsigned int write_len)
1768 int err = __bpf_try_make_writable(skb, write_len);
1770 bpf_compute_data_end_sk_skb(skb);
1774 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1776 /* Idea is the following: should the needed direct read/write
1777 * test fail during runtime, we can pull in more data and redo
1778 * again, since implicitly, we invalidate previous checks here.
1780 * Or, since we know how much we need to make read/writeable,
1781 * this can be done once at the program beginning for direct
1782 * access case. By this we overcome limitations of only current
1783 * headroom being accessible.
1785 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1788 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1789 .func = sk_skb_pull_data,
1791 .ret_type = RET_INTEGER,
1792 .arg1_type = ARG_PTR_TO_CTX,
1793 .arg2_type = ARG_ANYTHING,
1796 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1797 u64, from, u64, to, u64, flags)
1801 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1803 if (unlikely(offset > 0xffff || offset & 1))
1805 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1808 ptr = (__sum16 *)(skb->data + offset);
1809 switch (flags & BPF_F_HDR_FIELD_MASK) {
1811 if (unlikely(from != 0))
1814 csum_replace_by_diff(ptr, to);
1817 csum_replace2(ptr, from, to);
1820 csum_replace4(ptr, from, to);
1829 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1830 .func = bpf_l3_csum_replace,
1832 .ret_type = RET_INTEGER,
1833 .arg1_type = ARG_PTR_TO_CTX,
1834 .arg2_type = ARG_ANYTHING,
1835 .arg3_type = ARG_ANYTHING,
1836 .arg4_type = ARG_ANYTHING,
1837 .arg5_type = ARG_ANYTHING,
1840 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1841 u64, from, u64, to, u64, flags)
1843 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1844 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1845 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1848 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1849 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1851 if (unlikely(offset > 0xffff || offset & 1))
1853 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1856 ptr = (__sum16 *)(skb->data + offset);
1857 if (is_mmzero && !do_mforce && !*ptr)
1860 switch (flags & BPF_F_HDR_FIELD_MASK) {
1862 if (unlikely(from != 0))
1865 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1868 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1871 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1877 if (is_mmzero && !*ptr)
1878 *ptr = CSUM_MANGLED_0;
1882 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1883 .func = bpf_l4_csum_replace,
1885 .ret_type = RET_INTEGER,
1886 .arg1_type = ARG_PTR_TO_CTX,
1887 .arg2_type = ARG_ANYTHING,
1888 .arg3_type = ARG_ANYTHING,
1889 .arg4_type = ARG_ANYTHING,
1890 .arg5_type = ARG_ANYTHING,
1893 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1894 __be32 *, to, u32, to_size, __wsum, seed)
1896 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1897 u32 diff_size = from_size + to_size;
1900 /* This is quite flexible, some examples:
1902 * from_size == 0, to_size > 0, seed := csum --> pushing data
1903 * from_size > 0, to_size == 0, seed := csum --> pulling data
1904 * from_size > 0, to_size > 0, seed := 0 --> diffing data
1906 * Even for diffing, from_size and to_size don't need to be equal.
1908 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
1909 diff_size > sizeof(sp->diff)))
1912 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
1913 sp->diff[j] = ~from[i];
1914 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
1915 sp->diff[j] = to[i];
1917 return csum_partial(sp->diff, diff_size, seed);
1920 static const struct bpf_func_proto bpf_csum_diff_proto = {
1921 .func = bpf_csum_diff,
1924 .ret_type = RET_INTEGER,
1925 .arg1_type = ARG_PTR_TO_MEM_OR_NULL,
1926 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
1927 .arg3_type = ARG_PTR_TO_MEM_OR_NULL,
1928 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
1929 .arg5_type = ARG_ANYTHING,
1932 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
1934 /* The interface is to be used in combination with bpf_csum_diff()
1935 * for direct packet writes. csum rotation for alignment as well
1936 * as emulating csum_sub() can be done from the eBPF program.
1938 if (skb->ip_summed == CHECKSUM_COMPLETE)
1939 return (skb->csum = csum_add(skb->csum, csum));
1944 static const struct bpf_func_proto bpf_csum_update_proto = {
1945 .func = bpf_csum_update,
1947 .ret_type = RET_INTEGER,
1948 .arg1_type = ARG_PTR_TO_CTX,
1949 .arg2_type = ARG_ANYTHING,
1952 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
1954 return dev_forward_skb(dev, skb);
1957 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
1958 struct sk_buff *skb)
1960 int ret = ____dev_forward_skb(dev, skb);
1964 ret = netif_rx(skb);
1970 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
1974 if (unlikely(__this_cpu_read(xmit_recursion) > XMIT_RECURSION_LIMIT)) {
1975 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
1982 __this_cpu_inc(xmit_recursion);
1983 ret = dev_queue_xmit(skb);
1984 __this_cpu_dec(xmit_recursion);
1989 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
1992 /* skb->mac_len is not set on normal egress */
1993 unsigned int mlen = skb->network_header - skb->mac_header;
1995 __skb_pull(skb, mlen);
1997 /* At ingress, the mac header has already been pulled once.
1998 * At egress, skb_pospull_rcsum has to be done in case that
1999 * the skb is originated from ingress (i.e. a forwarded skb)
2000 * to ensure that rcsum starts at net header.
2002 if (!skb_at_tc_ingress(skb))
2003 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2004 skb_pop_mac_header(skb);
2005 skb_reset_mac_len(skb);
2006 return flags & BPF_F_INGRESS ?
2007 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2010 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2013 /* Verify that a link layer header is carried */
2014 if (unlikely(skb->mac_header >= skb->network_header)) {
2019 bpf_push_mac_rcsum(skb);
2020 return flags & BPF_F_INGRESS ?
2021 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2024 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2027 if (dev_is_mac_header_xmit(dev))
2028 return __bpf_redirect_common(skb, dev, flags);
2030 return __bpf_redirect_no_mac(skb, dev, flags);
2033 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2035 struct net_device *dev;
2036 struct sk_buff *clone;
2039 if (unlikely(flags & ~(BPF_F_INGRESS)))
2042 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2046 clone = skb_clone(skb, GFP_ATOMIC);
2047 if (unlikely(!clone))
2050 /* For direct write, we need to keep the invariant that the skbs
2051 * we're dealing with need to be uncloned. Should uncloning fail
2052 * here, we need to free the just generated clone to unclone once
2055 ret = bpf_try_make_head_writable(skb);
2056 if (unlikely(ret)) {
2061 return __bpf_redirect(clone, dev, flags);
2064 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2065 .func = bpf_clone_redirect,
2067 .ret_type = RET_INTEGER,
2068 .arg1_type = ARG_PTR_TO_CTX,
2069 .arg2_type = ARG_ANYTHING,
2070 .arg3_type = ARG_ANYTHING,
2073 struct redirect_info {
2076 struct bpf_map *map;
2077 struct bpf_map *map_to_flush;
2078 unsigned long map_owner;
2081 static DEFINE_PER_CPU(struct redirect_info, redirect_info);
2083 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2085 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2087 if (unlikely(flags & ~(BPF_F_INGRESS)))
2090 ri->ifindex = ifindex;
2093 return TC_ACT_REDIRECT;
2096 int skb_do_redirect(struct sk_buff *skb)
2098 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2099 struct net_device *dev;
2101 dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->ifindex);
2103 if (unlikely(!dev)) {
2108 return __bpf_redirect(skb, dev, ri->flags);
2111 static const struct bpf_func_proto bpf_redirect_proto = {
2112 .func = bpf_redirect,
2114 .ret_type = RET_INTEGER,
2115 .arg1_type = ARG_ANYTHING,
2116 .arg2_type = ARG_ANYTHING,
2119 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
2120 struct bpf_map *, map, void *, key, u64, flags)
2122 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
2124 /* If user passes invalid input drop the packet. */
2125 if (unlikely(flags & ~(BPF_F_INGRESS)))
2128 tcb->bpf.flags = flags;
2129 tcb->bpf.sk_redir = __sock_hash_lookup_elem(map, key);
2130 if (!tcb->bpf.sk_redir)
2136 static const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
2137 .func = bpf_sk_redirect_hash,
2139 .ret_type = RET_INTEGER,
2140 .arg1_type = ARG_PTR_TO_CTX,
2141 .arg2_type = ARG_CONST_MAP_PTR,
2142 .arg3_type = ARG_PTR_TO_MAP_KEY,
2143 .arg4_type = ARG_ANYTHING,
2146 BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
2147 struct bpf_map *, map, u32, key, u64, flags)
2149 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
2151 /* If user passes invalid input drop the packet. */
2152 if (unlikely(flags & ~(BPF_F_INGRESS)))
2155 tcb->bpf.flags = flags;
2156 tcb->bpf.sk_redir = __sock_map_lookup_elem(map, key);
2157 if (!tcb->bpf.sk_redir)
2163 struct sock *do_sk_redirect_map(struct sk_buff *skb)
2165 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
2167 return tcb->bpf.sk_redir;
2170 static const struct bpf_func_proto bpf_sk_redirect_map_proto = {
2171 .func = bpf_sk_redirect_map,
2173 .ret_type = RET_INTEGER,
2174 .arg1_type = ARG_PTR_TO_CTX,
2175 .arg2_type = ARG_CONST_MAP_PTR,
2176 .arg3_type = ARG_ANYTHING,
2177 .arg4_type = ARG_ANYTHING,
2180 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg_buff *, msg,
2181 struct bpf_map *, map, void *, key, u64, flags)
2183 /* If user passes invalid input drop the packet. */
2184 if (unlikely(flags & ~(BPF_F_INGRESS)))
2188 msg->sk_redir = __sock_hash_lookup_elem(map, key);
2195 static const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
2196 .func = bpf_msg_redirect_hash,
2198 .ret_type = RET_INTEGER,
2199 .arg1_type = ARG_PTR_TO_CTX,
2200 .arg2_type = ARG_CONST_MAP_PTR,
2201 .arg3_type = ARG_PTR_TO_MAP_KEY,
2202 .arg4_type = ARG_ANYTHING,
2205 BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg_buff *, msg,
2206 struct bpf_map *, map, u32, key, u64, flags)
2208 /* If user passes invalid input drop the packet. */
2209 if (unlikely(flags & ~(BPF_F_INGRESS)))
2213 msg->sk_redir = __sock_map_lookup_elem(map, key);
2220 struct sock *do_msg_redirect_map(struct sk_msg_buff *msg)
2222 return msg->sk_redir;
2225 static const struct bpf_func_proto bpf_msg_redirect_map_proto = {
2226 .func = bpf_msg_redirect_map,
2228 .ret_type = RET_INTEGER,
2229 .arg1_type = ARG_PTR_TO_CTX,
2230 .arg2_type = ARG_CONST_MAP_PTR,
2231 .arg3_type = ARG_ANYTHING,
2232 .arg4_type = ARG_ANYTHING,
2235 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg_buff *, msg, u32, bytes)
2237 msg->apply_bytes = bytes;
2241 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2242 .func = bpf_msg_apply_bytes,
2244 .ret_type = RET_INTEGER,
2245 .arg1_type = ARG_PTR_TO_CTX,
2246 .arg2_type = ARG_ANYTHING,
2249 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg_buff *, msg, u32, bytes)
2251 msg->cork_bytes = bytes;
2255 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2256 .func = bpf_msg_cork_bytes,
2258 .ret_type = RET_INTEGER,
2259 .arg1_type = ARG_PTR_TO_CTX,
2260 .arg2_type = ARG_ANYTHING,
2263 BPF_CALL_4(bpf_msg_pull_data,
2264 struct sk_msg_buff *, msg, u32, start, u32, end, u64, flags)
2266 unsigned int len = 0, offset = 0, copy = 0;
2267 struct scatterlist *sg = msg->sg_data;
2268 int first_sg, last_sg, i, shift;
2269 unsigned char *p, *to, *from;
2270 int bytes = end - start;
2273 if (unlikely(flags || end <= start))
2276 /* First find the starting scatterlist element */
2281 if (start < offset + len)
2284 if (i == MAX_SKB_FRAGS)
2286 } while (i != msg->sg_end);
2288 if (unlikely(start >= offset + len))
2291 if (!msg->sg_copy[i] && bytes <= len)
2296 /* At this point we need to linearize multiple scatterlist
2297 * elements or a single shared page. Either way we need to
2298 * copy into a linear buffer exclusively owned by BPF. Then
2299 * place the buffer in the scatterlist and fixup the original
2300 * entries by removing the entries now in the linear buffer
2301 * and shifting the remaining entries. For now we do not try
2302 * to copy partial entries to avoid complexity of running out
2303 * of sg_entry slots. The downside is reading a single byte
2304 * will copy the entire sg entry.
2307 copy += sg[i].length;
2309 if (i == MAX_SKB_FRAGS)
2313 } while (i != msg->sg_end);
2316 if (unlikely(copy < end - start))
2319 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC, get_order(copy));
2320 if (unlikely(!page))
2322 p = page_address(page);
2327 from = sg_virt(&sg[i]);
2331 memcpy(to, from, len);
2334 put_page(sg_page(&sg[i]));
2337 if (i == MAX_SKB_FRAGS)
2339 } while (i != last_sg);
2341 sg[first_sg].length = copy;
2342 sg_set_page(&sg[first_sg], page, copy, 0);
2344 /* To repair sg ring we need to shift entries. If we only
2345 * had a single entry though we can just replace it and
2346 * be done. Otherwise walk the ring and shift the entries.
2348 shift = last_sg - first_sg - 1;
2356 if (i + shift >= MAX_SKB_FRAGS)
2357 move_from = i + shift - MAX_SKB_FRAGS;
2359 move_from = i + shift;
2361 if (move_from == msg->sg_end)
2364 sg[i] = sg[move_from];
2365 sg[move_from].length = 0;
2366 sg[move_from].page_link = 0;
2367 sg[move_from].offset = 0;
2370 if (i == MAX_SKB_FRAGS)
2373 msg->sg_end -= shift;
2374 if (msg->sg_end < 0)
2375 msg->sg_end += MAX_SKB_FRAGS;
2377 msg->data = sg_virt(&sg[i]) + start - offset;
2378 msg->data_end = msg->data + bytes;
2383 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2384 .func = bpf_msg_pull_data,
2386 .ret_type = RET_INTEGER,
2387 .arg1_type = ARG_PTR_TO_CTX,
2388 .arg2_type = ARG_ANYTHING,
2389 .arg3_type = ARG_ANYTHING,
2390 .arg4_type = ARG_ANYTHING,
2393 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
2395 return task_get_classid(skb);
2398 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
2399 .func = bpf_get_cgroup_classid,
2401 .ret_type = RET_INTEGER,
2402 .arg1_type = ARG_PTR_TO_CTX,
2405 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
2407 return dst_tclassid(skb);
2410 static const struct bpf_func_proto bpf_get_route_realm_proto = {
2411 .func = bpf_get_route_realm,
2413 .ret_type = RET_INTEGER,
2414 .arg1_type = ARG_PTR_TO_CTX,
2417 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
2419 /* If skb_clear_hash() was called due to mangling, we can
2420 * trigger SW recalculation here. Later access to hash
2421 * can then use the inline skb->hash via context directly
2422 * instead of calling this helper again.
2424 return skb_get_hash(skb);
2427 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
2428 .func = bpf_get_hash_recalc,
2430 .ret_type = RET_INTEGER,
2431 .arg1_type = ARG_PTR_TO_CTX,
2434 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
2436 /* After all direct packet write, this can be used once for
2437 * triggering a lazy recalc on next skb_get_hash() invocation.
2439 skb_clear_hash(skb);
2443 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
2444 .func = bpf_set_hash_invalid,
2446 .ret_type = RET_INTEGER,
2447 .arg1_type = ARG_PTR_TO_CTX,
2450 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
2452 /* Set user specified hash as L4(+), so that it gets returned
2453 * on skb_get_hash() call unless BPF prog later on triggers a
2456 __skb_set_sw_hash(skb, hash, true);
2460 static const struct bpf_func_proto bpf_set_hash_proto = {
2461 .func = bpf_set_hash,
2463 .ret_type = RET_INTEGER,
2464 .arg1_type = ARG_PTR_TO_CTX,
2465 .arg2_type = ARG_ANYTHING,
2468 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
2473 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
2474 vlan_proto != htons(ETH_P_8021AD)))
2475 vlan_proto = htons(ETH_P_8021Q);
2477 bpf_push_mac_rcsum(skb);
2478 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
2479 bpf_pull_mac_rcsum(skb);
2481 bpf_compute_data_pointers(skb);
2485 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
2486 .func = bpf_skb_vlan_push,
2488 .ret_type = RET_INTEGER,
2489 .arg1_type = ARG_PTR_TO_CTX,
2490 .arg2_type = ARG_ANYTHING,
2491 .arg3_type = ARG_ANYTHING,
2494 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
2498 bpf_push_mac_rcsum(skb);
2499 ret = skb_vlan_pop(skb);
2500 bpf_pull_mac_rcsum(skb);
2502 bpf_compute_data_pointers(skb);
2506 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
2507 .func = bpf_skb_vlan_pop,
2509 .ret_type = RET_INTEGER,
2510 .arg1_type = ARG_PTR_TO_CTX,
2513 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
2515 /* Caller already did skb_cow() with len as headroom,
2516 * so no need to do it here.
2519 memmove(skb->data, skb->data + len, off);
2520 memset(skb->data + off, 0, len);
2522 /* No skb_postpush_rcsum(skb, skb->data + off, len)
2523 * needed here as it does not change the skb->csum
2524 * result for checksum complete when summing over
2530 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
2532 /* skb_ensure_writable() is not needed here, as we're
2533 * already working on an uncloned skb.
2535 if (unlikely(!pskb_may_pull(skb, off + len)))
2538 skb_postpull_rcsum(skb, skb->data + off, len);
2539 memmove(skb->data + len, skb->data, off);
2540 __skb_pull(skb, len);
2545 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
2547 bool trans_same = skb->transport_header == skb->network_header;
2550 /* There's no need for __skb_push()/__skb_pull() pair to
2551 * get to the start of the mac header as we're guaranteed
2552 * to always start from here under eBPF.
2554 ret = bpf_skb_generic_push(skb, off, len);
2556 skb->mac_header -= len;
2557 skb->network_header -= len;
2559 skb->transport_header = skb->network_header;
2565 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
2567 bool trans_same = skb->transport_header == skb->network_header;
2570 /* Same here, __skb_push()/__skb_pull() pair not needed. */
2571 ret = bpf_skb_generic_pop(skb, off, len);
2573 skb->mac_header += len;
2574 skb->network_header += len;
2576 skb->transport_header = skb->network_header;
2582 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
2584 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2585 u32 off = skb_mac_header_len(skb);
2588 /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2589 if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2592 ret = skb_cow(skb, len_diff);
2593 if (unlikely(ret < 0))
2596 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2597 if (unlikely(ret < 0))
2600 if (skb_is_gso(skb)) {
2601 struct skb_shared_info *shinfo = skb_shinfo(skb);
2603 /* SKB_GSO_TCPV4 needs to be changed into
2606 if (shinfo->gso_type & SKB_GSO_TCPV4) {
2607 shinfo->gso_type &= ~SKB_GSO_TCPV4;
2608 shinfo->gso_type |= SKB_GSO_TCPV6;
2611 /* Due to IPv6 header, MSS needs to be downgraded. */
2612 skb_decrease_gso_size(shinfo, len_diff);
2613 /* Header must be checked, and gso_segs recomputed. */
2614 shinfo->gso_type |= SKB_GSO_DODGY;
2615 shinfo->gso_segs = 0;
2618 skb->protocol = htons(ETH_P_IPV6);
2619 skb_clear_hash(skb);
2624 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
2626 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2627 u32 off = skb_mac_header_len(skb);
2630 /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2631 if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2634 ret = skb_unclone(skb, GFP_ATOMIC);
2635 if (unlikely(ret < 0))
2638 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2639 if (unlikely(ret < 0))
2642 if (skb_is_gso(skb)) {
2643 struct skb_shared_info *shinfo = skb_shinfo(skb);
2645 /* SKB_GSO_TCPV6 needs to be changed into
2648 if (shinfo->gso_type & SKB_GSO_TCPV6) {
2649 shinfo->gso_type &= ~SKB_GSO_TCPV6;
2650 shinfo->gso_type |= SKB_GSO_TCPV4;
2653 /* Due to IPv4 header, MSS can be upgraded. */
2654 skb_increase_gso_size(shinfo, len_diff);
2655 /* Header must be checked, and gso_segs recomputed. */
2656 shinfo->gso_type |= SKB_GSO_DODGY;
2657 shinfo->gso_segs = 0;
2660 skb->protocol = htons(ETH_P_IP);
2661 skb_clear_hash(skb);
2666 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
2668 __be16 from_proto = skb->protocol;
2670 if (from_proto == htons(ETH_P_IP) &&
2671 to_proto == htons(ETH_P_IPV6))
2672 return bpf_skb_proto_4_to_6(skb);
2674 if (from_proto == htons(ETH_P_IPV6) &&
2675 to_proto == htons(ETH_P_IP))
2676 return bpf_skb_proto_6_to_4(skb);
2681 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
2686 if (unlikely(flags))
2689 /* General idea is that this helper does the basic groundwork
2690 * needed for changing the protocol, and eBPF program fills the
2691 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
2692 * and other helpers, rather than passing a raw buffer here.
2694 * The rationale is to keep this minimal and without a need to
2695 * deal with raw packet data. F.e. even if we would pass buffers
2696 * here, the program still needs to call the bpf_lX_csum_replace()
2697 * helpers anyway. Plus, this way we keep also separation of
2698 * concerns, since f.e. bpf_skb_store_bytes() should only take
2701 * Currently, additional options and extension header space are
2702 * not supported, but flags register is reserved so we can adapt
2703 * that. For offloads, we mark packet as dodgy, so that headers
2704 * need to be verified first.
2706 ret = bpf_skb_proto_xlat(skb, proto);
2707 bpf_compute_data_pointers(skb);
2711 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
2712 .func = bpf_skb_change_proto,
2714 .ret_type = RET_INTEGER,
2715 .arg1_type = ARG_PTR_TO_CTX,
2716 .arg2_type = ARG_ANYTHING,
2717 .arg3_type = ARG_ANYTHING,
2720 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
2722 /* We only allow a restricted subset to be changed for now. */
2723 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
2724 !skb_pkt_type_ok(pkt_type)))
2727 skb->pkt_type = pkt_type;
2731 static const struct bpf_func_proto bpf_skb_change_type_proto = {
2732 .func = bpf_skb_change_type,
2734 .ret_type = RET_INTEGER,
2735 .arg1_type = ARG_PTR_TO_CTX,
2736 .arg2_type = ARG_ANYTHING,
2739 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
2741 switch (skb->protocol) {
2742 case htons(ETH_P_IP):
2743 return sizeof(struct iphdr);
2744 case htons(ETH_P_IPV6):
2745 return sizeof(struct ipv6hdr);
2751 static int bpf_skb_net_grow(struct sk_buff *skb, u32 len_diff)
2753 u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
2756 /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2757 if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2760 ret = skb_cow(skb, len_diff);
2761 if (unlikely(ret < 0))
2764 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2765 if (unlikely(ret < 0))
2768 if (skb_is_gso(skb)) {
2769 struct skb_shared_info *shinfo = skb_shinfo(skb);
2771 /* Due to header grow, MSS needs to be downgraded. */
2772 skb_decrease_gso_size(shinfo, len_diff);
2773 /* Header must be checked, and gso_segs recomputed. */
2774 shinfo->gso_type |= SKB_GSO_DODGY;
2775 shinfo->gso_segs = 0;
2781 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 len_diff)
2783 u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
2786 /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2787 if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2790 ret = skb_unclone(skb, GFP_ATOMIC);
2791 if (unlikely(ret < 0))
2794 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2795 if (unlikely(ret < 0))
2798 if (skb_is_gso(skb)) {
2799 struct skb_shared_info *shinfo = skb_shinfo(skb);
2801 /* Due to header shrink, MSS can be upgraded. */
2802 skb_increase_gso_size(shinfo, len_diff);
2803 /* Header must be checked, and gso_segs recomputed. */
2804 shinfo->gso_type |= SKB_GSO_DODGY;
2805 shinfo->gso_segs = 0;
2811 static u32 __bpf_skb_max_len(const struct sk_buff *skb)
2813 return skb->dev ? skb->dev->mtu + skb->dev->hard_header_len :
2817 static int bpf_skb_adjust_net(struct sk_buff *skb, s32 len_diff)
2819 bool trans_same = skb->transport_header == skb->network_header;
2820 u32 len_cur, len_diff_abs = abs(len_diff);
2821 u32 len_min = bpf_skb_net_base_len(skb);
2822 u32 len_max = __bpf_skb_max_len(skb);
2823 __be16 proto = skb->protocol;
2824 bool shrink = len_diff < 0;
2827 if (unlikely(len_diff_abs > 0xfffU))
2829 if (unlikely(proto != htons(ETH_P_IP) &&
2830 proto != htons(ETH_P_IPV6)))
2833 len_cur = skb->len - skb_network_offset(skb);
2834 if (skb_transport_header_was_set(skb) && !trans_same)
2835 len_cur = skb_network_header_len(skb);
2836 if ((shrink && (len_diff_abs >= len_cur ||
2837 len_cur - len_diff_abs < len_min)) ||
2838 (!shrink && (skb->len + len_diff_abs > len_max &&
2842 ret = shrink ? bpf_skb_net_shrink(skb, len_diff_abs) :
2843 bpf_skb_net_grow(skb, len_diff_abs);
2845 bpf_compute_data_pointers(skb);
2849 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
2850 u32, mode, u64, flags)
2852 if (unlikely(flags))
2854 if (likely(mode == BPF_ADJ_ROOM_NET))
2855 return bpf_skb_adjust_net(skb, len_diff);
2860 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
2861 .func = bpf_skb_adjust_room,
2863 .ret_type = RET_INTEGER,
2864 .arg1_type = ARG_PTR_TO_CTX,
2865 .arg2_type = ARG_ANYTHING,
2866 .arg3_type = ARG_ANYTHING,
2867 .arg4_type = ARG_ANYTHING,
2870 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
2872 u32 min_len = skb_network_offset(skb);
2874 if (skb_transport_header_was_set(skb))
2875 min_len = skb_transport_offset(skb);
2876 if (skb->ip_summed == CHECKSUM_PARTIAL)
2877 min_len = skb_checksum_start_offset(skb) +
2878 skb->csum_offset + sizeof(__sum16);
2882 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
2884 unsigned int old_len = skb->len;
2887 ret = __skb_grow_rcsum(skb, new_len);
2889 memset(skb->data + old_len, 0, new_len - old_len);
2893 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
2895 return __skb_trim_rcsum(skb, new_len);
2898 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
2901 u32 max_len = __bpf_skb_max_len(skb);
2902 u32 min_len = __bpf_skb_min_len(skb);
2905 if (unlikely(flags || new_len > max_len || new_len < min_len))
2907 if (skb->encapsulation)
2910 /* The basic idea of this helper is that it's performing the
2911 * needed work to either grow or trim an skb, and eBPF program
2912 * rewrites the rest via helpers like bpf_skb_store_bytes(),
2913 * bpf_lX_csum_replace() and others rather than passing a raw
2914 * buffer here. This one is a slow path helper and intended
2915 * for replies with control messages.
2917 * Like in bpf_skb_change_proto(), we want to keep this rather
2918 * minimal and without protocol specifics so that we are able
2919 * to separate concerns as in bpf_skb_store_bytes() should only
2920 * be the one responsible for writing buffers.
2922 * It's really expected to be a slow path operation here for
2923 * control message replies, so we're implicitly linearizing,
2924 * uncloning and drop offloads from the skb by this.
2926 ret = __bpf_try_make_writable(skb, skb->len);
2928 if (new_len > skb->len)
2929 ret = bpf_skb_grow_rcsum(skb, new_len);
2930 else if (new_len < skb->len)
2931 ret = bpf_skb_trim_rcsum(skb, new_len);
2932 if (!ret && skb_is_gso(skb))
2938 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
2941 int ret = __bpf_skb_change_tail(skb, new_len, flags);
2943 bpf_compute_data_pointers(skb);
2947 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
2948 .func = bpf_skb_change_tail,
2950 .ret_type = RET_INTEGER,
2951 .arg1_type = ARG_PTR_TO_CTX,
2952 .arg2_type = ARG_ANYTHING,
2953 .arg3_type = ARG_ANYTHING,
2956 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
2959 int ret = __bpf_skb_change_tail(skb, new_len, flags);
2961 bpf_compute_data_end_sk_skb(skb);
2965 static const struct bpf_func_proto sk_skb_change_tail_proto = {
2966 .func = sk_skb_change_tail,
2968 .ret_type = RET_INTEGER,
2969 .arg1_type = ARG_PTR_TO_CTX,
2970 .arg2_type = ARG_ANYTHING,
2971 .arg3_type = ARG_ANYTHING,
2974 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
2977 u32 max_len = __bpf_skb_max_len(skb);
2978 u32 new_len = skb->len + head_room;
2981 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
2982 new_len < skb->len))
2985 ret = skb_cow(skb, head_room);
2987 /* Idea for this helper is that we currently only
2988 * allow to expand on mac header. This means that
2989 * skb->protocol network header, etc, stay as is.
2990 * Compared to bpf_skb_change_tail(), we're more
2991 * flexible due to not needing to linearize or
2992 * reset GSO. Intention for this helper is to be
2993 * used by an L3 skb that needs to push mac header
2994 * for redirection into L2 device.
2996 __skb_push(skb, head_room);
2997 memset(skb->data, 0, head_room);
2998 skb_reset_mac_header(skb);
3004 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3007 int ret = __bpf_skb_change_head(skb, head_room, flags);
3009 bpf_compute_data_pointers(skb);
3013 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3014 .func = bpf_skb_change_head,
3016 .ret_type = RET_INTEGER,
3017 .arg1_type = ARG_PTR_TO_CTX,
3018 .arg2_type = ARG_ANYTHING,
3019 .arg3_type = ARG_ANYTHING,
3022 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3025 int ret = __bpf_skb_change_head(skb, head_room, flags);
3027 bpf_compute_data_end_sk_skb(skb);
3031 static const struct bpf_func_proto sk_skb_change_head_proto = {
3032 .func = sk_skb_change_head,
3034 .ret_type = RET_INTEGER,
3035 .arg1_type = ARG_PTR_TO_CTX,
3036 .arg2_type = ARG_ANYTHING,
3037 .arg3_type = ARG_ANYTHING,
3039 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3041 return xdp_data_meta_unsupported(xdp) ? 0 :
3042 xdp->data - xdp->data_meta;
3045 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3047 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3048 unsigned long metalen = xdp_get_metalen(xdp);
3049 void *data_start = xdp_frame_end + metalen;
3050 void *data = xdp->data + offset;
3052 if (unlikely(data < data_start ||
3053 data > xdp->data_end - ETH_HLEN))
3057 memmove(xdp->data_meta + offset,
3058 xdp->data_meta, metalen);
3059 xdp->data_meta += offset;
3065 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3066 .func = bpf_xdp_adjust_head,
3068 .ret_type = RET_INTEGER,
3069 .arg1_type = ARG_PTR_TO_CTX,
3070 .arg2_type = ARG_ANYTHING,
3073 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3075 void *data_end = xdp->data_end + offset;
3077 /* only shrinking is allowed for now. */
3078 if (unlikely(offset >= 0))
3081 if (unlikely(data_end < xdp->data + ETH_HLEN))
3084 xdp->data_end = data_end;
3089 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3090 .func = bpf_xdp_adjust_tail,
3092 .ret_type = RET_INTEGER,
3093 .arg1_type = ARG_PTR_TO_CTX,
3094 .arg2_type = ARG_ANYTHING,
3097 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3099 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3100 void *meta = xdp->data_meta + offset;
3101 unsigned long metalen = xdp->data - meta;
3103 if (xdp_data_meta_unsupported(xdp))
3105 if (unlikely(meta < xdp_frame_end ||
3108 if (unlikely((metalen & (sizeof(__u32) - 1)) ||
3112 xdp->data_meta = meta;
3117 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3118 .func = bpf_xdp_adjust_meta,
3120 .ret_type = RET_INTEGER,
3121 .arg1_type = ARG_PTR_TO_CTX,
3122 .arg2_type = ARG_ANYTHING,
3125 static int __bpf_tx_xdp(struct net_device *dev,
3126 struct bpf_map *map,
3127 struct xdp_buff *xdp,
3130 struct xdp_frame *xdpf;
3133 if (!dev->netdev_ops->ndo_xdp_xmit) {
3137 xdpf = convert_to_xdp_frame(xdp);
3138 if (unlikely(!xdpf))
3141 sent = dev->netdev_ops->ndo_xdp_xmit(dev, 1, &xdpf, XDP_XMIT_FLUSH);
3147 static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd,
3148 struct bpf_map *map,
3149 struct xdp_buff *xdp,
3154 switch (map->map_type) {
3155 case BPF_MAP_TYPE_DEVMAP: {
3156 struct bpf_dtab_netdev *dst = fwd;
3158 err = dev_map_enqueue(dst, xdp, dev_rx);
3161 __dev_map_insert_ctx(map, index);
3164 case BPF_MAP_TYPE_CPUMAP: {
3165 struct bpf_cpu_map_entry *rcpu = fwd;
3167 err = cpu_map_enqueue(rcpu, xdp, dev_rx);
3170 __cpu_map_insert_ctx(map, index);
3173 case BPF_MAP_TYPE_XSKMAP: {
3174 struct xdp_sock *xs = fwd;
3176 err = __xsk_map_redirect(map, xdp, xs);
3185 void xdp_do_flush_map(void)
3187 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
3188 struct bpf_map *map = ri->map_to_flush;
3190 ri->map_to_flush = NULL;
3192 switch (map->map_type) {
3193 case BPF_MAP_TYPE_DEVMAP:
3194 __dev_map_flush(map);
3196 case BPF_MAP_TYPE_CPUMAP:
3197 __cpu_map_flush(map);
3199 case BPF_MAP_TYPE_XSKMAP:
3200 __xsk_map_flush(map);
3207 EXPORT_SYMBOL_GPL(xdp_do_flush_map);
3209 static void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index)
3211 switch (map->map_type) {
3212 case BPF_MAP_TYPE_DEVMAP:
3213 return __dev_map_lookup_elem(map, index);
3214 case BPF_MAP_TYPE_CPUMAP:
3215 return __cpu_map_lookup_elem(map, index);
3216 case BPF_MAP_TYPE_XSKMAP:
3217 return __xsk_map_lookup_elem(map, index);
3223 static inline bool xdp_map_invalid(const struct bpf_prog *xdp_prog,
3226 return (unsigned long)xdp_prog->aux != aux;
3229 static int xdp_do_redirect_map(struct net_device *dev, struct xdp_buff *xdp,
3230 struct bpf_prog *xdp_prog)
3232 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
3233 unsigned long map_owner = ri->map_owner;
3234 struct bpf_map *map = ri->map;
3235 u32 index = ri->ifindex;
3243 if (unlikely(xdp_map_invalid(xdp_prog, map_owner))) {
3249 fwd = __xdp_map_lookup_elem(map, index);
3254 if (ri->map_to_flush && ri->map_to_flush != map)
3257 err = __bpf_tx_xdp_map(dev, fwd, map, xdp, index);
3261 ri->map_to_flush = map;
3262 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3265 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3269 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3270 struct bpf_prog *xdp_prog)
3272 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
3273 struct net_device *fwd;
3274 u32 index = ri->ifindex;
3278 return xdp_do_redirect_map(dev, xdp, xdp_prog);
3280 fwd = dev_get_by_index_rcu(dev_net(dev), index);
3282 if (unlikely(!fwd)) {
3287 err = __bpf_tx_xdp(fwd, NULL, xdp, 0);
3291 _trace_xdp_redirect(dev, xdp_prog, index);
3294 _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3297 EXPORT_SYMBOL_GPL(xdp_do_redirect);
3299 static int xdp_do_generic_redirect_map(struct net_device *dev,
3300 struct sk_buff *skb,
3301 struct xdp_buff *xdp,
3302 struct bpf_prog *xdp_prog)
3304 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
3305 unsigned long map_owner = ri->map_owner;
3306 struct bpf_map *map = ri->map;
3307 u32 index = ri->ifindex;
3315 if (unlikely(xdp_map_invalid(xdp_prog, map_owner))) {
3320 fwd = __xdp_map_lookup_elem(map, index);
3321 if (unlikely(!fwd)) {
3326 if (map->map_type == BPF_MAP_TYPE_DEVMAP) {
3327 struct bpf_dtab_netdev *dst = fwd;
3329 err = dev_map_generic_redirect(dst, skb, xdp_prog);
3332 } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) {
3333 struct xdp_sock *xs = fwd;
3335 err = xsk_generic_rcv(xs, xdp);
3340 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
3345 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3348 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3352 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
3353 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
3355 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
3356 u32 index = ri->ifindex;
3357 struct net_device *fwd;
3361 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog);
3364 fwd = dev_get_by_index_rcu(dev_net(dev), index);
3365 if (unlikely(!fwd)) {
3370 if (unlikely((err = __xdp_generic_ok_fwd_dev(skb, fwd))))
3374 _trace_xdp_redirect(dev, xdp_prog, index);
3375 generic_xdp_tx(skb, xdp_prog);
3378 _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3381 EXPORT_SYMBOL_GPL(xdp_do_generic_redirect);
3383 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
3385 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
3387 if (unlikely(flags))
3390 ri->ifindex = ifindex;
3395 return XDP_REDIRECT;
3398 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
3399 .func = bpf_xdp_redirect,
3401 .ret_type = RET_INTEGER,
3402 .arg1_type = ARG_ANYTHING,
3403 .arg2_type = ARG_ANYTHING,
3406 BPF_CALL_4(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex, u64, flags,
3407 unsigned long, map_owner)
3409 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
3411 if (unlikely(flags))
3414 ri->ifindex = ifindex;
3417 ri->map_owner = map_owner;
3419 return XDP_REDIRECT;
3422 /* Note, arg4 is hidden from users and populated by the verifier
3423 * with the right pointer.
3425 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
3426 .func = bpf_xdp_redirect_map,
3428 .ret_type = RET_INTEGER,
3429 .arg1_type = ARG_CONST_MAP_PTR,
3430 .arg2_type = ARG_ANYTHING,
3431 .arg3_type = ARG_ANYTHING,
3434 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
3435 unsigned long off, unsigned long len)
3437 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
3441 if (ptr != dst_buff)
3442 memcpy(dst_buff, ptr, len);
3447 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
3448 u64, flags, void *, meta, u64, meta_size)
3450 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
3452 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
3454 if (unlikely(skb_size > skb->len))
3457 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
3461 static const struct bpf_func_proto bpf_skb_event_output_proto = {
3462 .func = bpf_skb_event_output,
3464 .ret_type = RET_INTEGER,
3465 .arg1_type = ARG_PTR_TO_CTX,
3466 .arg2_type = ARG_CONST_MAP_PTR,
3467 .arg3_type = ARG_ANYTHING,
3468 .arg4_type = ARG_PTR_TO_MEM,
3469 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
3472 static unsigned short bpf_tunnel_key_af(u64 flags)
3474 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
3477 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
3478 u32, size, u64, flags)
3480 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3481 u8 compat[sizeof(struct bpf_tunnel_key)];
3485 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
3489 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
3493 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3496 case offsetof(struct bpf_tunnel_key, tunnel_label):
3497 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3499 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3500 /* Fixup deprecated structure layouts here, so we have
3501 * a common path later on.
3503 if (ip_tunnel_info_af(info) != AF_INET)
3506 to = (struct bpf_tunnel_key *)compat;
3513 to->tunnel_id = be64_to_cpu(info->key.tun_id);
3514 to->tunnel_tos = info->key.tos;
3515 to->tunnel_ttl = info->key.ttl;
3518 if (flags & BPF_F_TUNINFO_IPV6) {
3519 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
3520 sizeof(to->remote_ipv6));
3521 to->tunnel_label = be32_to_cpu(info->key.label);
3523 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
3524 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
3525 to->tunnel_label = 0;
3528 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
3529 memcpy(to_orig, to, size);
3533 memset(to_orig, 0, size);
3537 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
3538 .func = bpf_skb_get_tunnel_key,
3540 .ret_type = RET_INTEGER,
3541 .arg1_type = ARG_PTR_TO_CTX,
3542 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
3543 .arg3_type = ARG_CONST_SIZE,
3544 .arg4_type = ARG_ANYTHING,
3547 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
3549 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3552 if (unlikely(!info ||
3553 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
3557 if (unlikely(size < info->options_len)) {
3562 ip_tunnel_info_opts_get(to, info);
3563 if (size > info->options_len)
3564 memset(to + info->options_len, 0, size - info->options_len);
3566 return info->options_len;
3568 memset(to, 0, size);
3572 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
3573 .func = bpf_skb_get_tunnel_opt,
3575 .ret_type = RET_INTEGER,
3576 .arg1_type = ARG_PTR_TO_CTX,
3577 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
3578 .arg3_type = ARG_CONST_SIZE,
3581 static struct metadata_dst __percpu *md_dst;
3583 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
3584 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
3586 struct metadata_dst *md = this_cpu_ptr(md_dst);
3587 u8 compat[sizeof(struct bpf_tunnel_key)];
3588 struct ip_tunnel_info *info;
3590 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
3591 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
3593 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3595 case offsetof(struct bpf_tunnel_key, tunnel_label):
3596 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3597 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3598 /* Fixup deprecated structure layouts here, so we have
3599 * a common path later on.
3601 memcpy(compat, from, size);
3602 memset(compat + size, 0, sizeof(compat) - size);
3603 from = (const struct bpf_tunnel_key *) compat;
3609 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
3614 dst_hold((struct dst_entry *) md);
3615 skb_dst_set(skb, (struct dst_entry *) md);
3617 info = &md->u.tun_info;
3618 memset(info, 0, sizeof(*info));
3619 info->mode = IP_TUNNEL_INFO_TX;
3621 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
3622 if (flags & BPF_F_DONT_FRAGMENT)
3623 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
3624 if (flags & BPF_F_ZERO_CSUM_TX)
3625 info->key.tun_flags &= ~TUNNEL_CSUM;
3626 if (flags & BPF_F_SEQ_NUMBER)
3627 info->key.tun_flags |= TUNNEL_SEQ;
3629 info->key.tun_id = cpu_to_be64(from->tunnel_id);
3630 info->key.tos = from->tunnel_tos;
3631 info->key.ttl = from->tunnel_ttl;
3633 if (flags & BPF_F_TUNINFO_IPV6) {
3634 info->mode |= IP_TUNNEL_INFO_IPV6;
3635 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
3636 sizeof(from->remote_ipv6));
3637 info->key.label = cpu_to_be32(from->tunnel_label) &
3638 IPV6_FLOWLABEL_MASK;
3640 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
3646 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
3647 .func = bpf_skb_set_tunnel_key,
3649 .ret_type = RET_INTEGER,
3650 .arg1_type = ARG_PTR_TO_CTX,
3651 .arg2_type = ARG_PTR_TO_MEM,
3652 .arg3_type = ARG_CONST_SIZE,
3653 .arg4_type = ARG_ANYTHING,
3656 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
3657 const u8 *, from, u32, size)
3659 struct ip_tunnel_info *info = skb_tunnel_info(skb);
3660 const struct metadata_dst *md = this_cpu_ptr(md_dst);
3662 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
3664 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
3667 ip_tunnel_info_opts_set(info, from, size);
3672 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
3673 .func = bpf_skb_set_tunnel_opt,
3675 .ret_type = RET_INTEGER,
3676 .arg1_type = ARG_PTR_TO_CTX,
3677 .arg2_type = ARG_PTR_TO_MEM,
3678 .arg3_type = ARG_CONST_SIZE,
3681 static const struct bpf_func_proto *
3682 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
3685 struct metadata_dst __percpu *tmp;
3687 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
3692 if (cmpxchg(&md_dst, NULL, tmp))
3693 metadata_dst_free_percpu(tmp);
3697 case BPF_FUNC_skb_set_tunnel_key:
3698 return &bpf_skb_set_tunnel_key_proto;
3699 case BPF_FUNC_skb_set_tunnel_opt:
3700 return &bpf_skb_set_tunnel_opt_proto;
3706 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
3709 struct bpf_array *array = container_of(map, struct bpf_array, map);
3710 struct cgroup *cgrp;
3713 sk = skb_to_full_sk(skb);
3714 if (!sk || !sk_fullsock(sk))
3716 if (unlikely(idx >= array->map.max_entries))
3719 cgrp = READ_ONCE(array->ptrs[idx]);
3720 if (unlikely(!cgrp))
3723 return sk_under_cgroup_hierarchy(sk, cgrp);
3726 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
3727 .func = bpf_skb_under_cgroup,
3729 .ret_type = RET_INTEGER,
3730 .arg1_type = ARG_PTR_TO_CTX,
3731 .arg2_type = ARG_CONST_MAP_PTR,
3732 .arg3_type = ARG_ANYTHING,
3735 #ifdef CONFIG_SOCK_CGROUP_DATA
3736 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
3738 struct sock *sk = skb_to_full_sk(skb);
3739 struct cgroup *cgrp;
3741 if (!sk || !sk_fullsock(sk))
3744 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
3745 return cgrp->kn->id.id;
3748 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
3749 .func = bpf_skb_cgroup_id,
3751 .ret_type = RET_INTEGER,
3752 .arg1_type = ARG_PTR_TO_CTX,
3756 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
3757 unsigned long off, unsigned long len)
3759 memcpy(dst_buff, src_buff + off, len);
3763 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
3764 u64, flags, void *, meta, u64, meta_size)
3766 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
3768 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
3770 if (unlikely(xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
3773 return bpf_event_output(map, flags, meta, meta_size, xdp->data,
3774 xdp_size, bpf_xdp_copy);
3777 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
3778 .func = bpf_xdp_event_output,
3780 .ret_type = RET_INTEGER,
3781 .arg1_type = ARG_PTR_TO_CTX,
3782 .arg2_type = ARG_CONST_MAP_PTR,
3783 .arg3_type = ARG_ANYTHING,
3784 .arg4_type = ARG_PTR_TO_MEM,
3785 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
3788 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
3790 return skb->sk ? sock_gen_cookie(skb->sk) : 0;
3793 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
3794 .func = bpf_get_socket_cookie,
3796 .ret_type = RET_INTEGER,
3797 .arg1_type = ARG_PTR_TO_CTX,
3800 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
3802 struct sock *sk = sk_to_full_sk(skb->sk);
3805 if (!sk || !sk_fullsock(sk))
3807 kuid = sock_net_uid(sock_net(sk), sk);
3808 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
3811 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
3812 .func = bpf_get_socket_uid,
3814 .ret_type = RET_INTEGER,
3815 .arg1_type = ARG_PTR_TO_CTX,
3818 BPF_CALL_5(bpf_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
3819 int, level, int, optname, char *, optval, int, optlen)
3821 struct sock *sk = bpf_sock->sk;
3825 if (!sk_fullsock(sk))
3828 if (level == SOL_SOCKET) {
3829 if (optlen != sizeof(int))
3831 val = *((int *)optval);
3833 /* Only some socketops are supported */
3836 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
3837 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
3840 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
3841 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
3843 case SO_MAX_PACING_RATE:
3844 sk->sk_max_pacing_rate = val;
3845 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
3846 sk->sk_max_pacing_rate);
3849 sk->sk_priority = val;
3854 sk->sk_rcvlowat = val ? : 1;
3863 } else if (level == SOL_IP) {
3864 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
3867 val = *((int *)optval);
3868 /* Only some options are supported */
3871 if (val < -1 || val > 0xff) {
3874 struct inet_sock *inet = inet_sk(sk);
3884 #if IS_ENABLED(CONFIG_IPV6)
3885 } else if (level == SOL_IPV6) {
3886 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
3889 val = *((int *)optval);
3890 /* Only some options are supported */
3893 if (val < -1 || val > 0xff) {
3896 struct ipv6_pinfo *np = inet6_sk(sk);
3907 } else if (level == SOL_TCP &&
3908 sk->sk_prot->setsockopt == tcp_setsockopt) {
3909 if (optname == TCP_CONGESTION) {
3910 char name[TCP_CA_NAME_MAX];
3911 bool reinit = bpf_sock->op > BPF_SOCK_OPS_NEEDS_ECN;
3913 strncpy(name, optval, min_t(long, optlen,
3914 TCP_CA_NAME_MAX-1));
3915 name[TCP_CA_NAME_MAX-1] = 0;
3916 ret = tcp_set_congestion_control(sk, name, false,
3919 struct tcp_sock *tp = tcp_sk(sk);
3921 if (optlen != sizeof(int))
3924 val = *((int *)optval);
3925 /* Only some options are supported */
3928 if (val <= 0 || tp->data_segs_out > 0)
3933 case TCP_BPF_SNDCWND_CLAMP:
3937 tp->snd_cwnd_clamp = val;
3938 tp->snd_ssthresh = val;
3952 static const struct bpf_func_proto bpf_setsockopt_proto = {
3953 .func = bpf_setsockopt,
3955 .ret_type = RET_INTEGER,
3956 .arg1_type = ARG_PTR_TO_CTX,
3957 .arg2_type = ARG_ANYTHING,
3958 .arg3_type = ARG_ANYTHING,
3959 .arg4_type = ARG_PTR_TO_MEM,
3960 .arg5_type = ARG_CONST_SIZE,
3963 BPF_CALL_5(bpf_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
3964 int, level, int, optname, char *, optval, int, optlen)
3966 struct sock *sk = bpf_sock->sk;
3968 if (!sk_fullsock(sk))
3972 if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
3973 if (optname == TCP_CONGESTION) {
3974 struct inet_connection_sock *icsk = inet_csk(sk);
3976 if (!icsk->icsk_ca_ops || optlen <= 1)
3978 strncpy(optval, icsk->icsk_ca_ops->name, optlen);
3979 optval[optlen - 1] = 0;
3983 } else if (level == SOL_IP) {
3984 struct inet_sock *inet = inet_sk(sk);
3986 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
3989 /* Only some options are supported */
3992 *((int *)optval) = (int)inet->tos;
3997 #if IS_ENABLED(CONFIG_IPV6)
3998 } else if (level == SOL_IPV6) {
3999 struct ipv6_pinfo *np = inet6_sk(sk);
4001 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4004 /* Only some options are supported */
4007 *((int *)optval) = (int)np->tclass;
4019 memset(optval, 0, optlen);
4023 static const struct bpf_func_proto bpf_getsockopt_proto = {
4024 .func = bpf_getsockopt,
4026 .ret_type = RET_INTEGER,
4027 .arg1_type = ARG_PTR_TO_CTX,
4028 .arg2_type = ARG_ANYTHING,
4029 .arg3_type = ARG_ANYTHING,
4030 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
4031 .arg5_type = ARG_CONST_SIZE,
4034 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
4037 struct sock *sk = bpf_sock->sk;
4038 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
4040 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
4044 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
4046 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
4049 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
4050 .func = bpf_sock_ops_cb_flags_set,
4052 .ret_type = RET_INTEGER,
4053 .arg1_type = ARG_PTR_TO_CTX,
4054 .arg2_type = ARG_ANYTHING,
4057 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
4058 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
4060 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
4064 struct sock *sk = ctx->sk;
4067 /* Binding to port can be expensive so it's prohibited in the helper.
4068 * Only binding to IP is supported.
4071 if (addr->sa_family == AF_INET) {
4072 if (addr_len < sizeof(struct sockaddr_in))
4074 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
4076 return __inet_bind(sk, addr, addr_len, true, false);
4077 #if IS_ENABLED(CONFIG_IPV6)
4078 } else if (addr->sa_family == AF_INET6) {
4079 if (addr_len < SIN6_LEN_RFC2133)
4081 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
4083 /* ipv6_bpf_stub cannot be NULL, since it's called from
4084 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
4086 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, true, false);
4087 #endif /* CONFIG_IPV6 */
4089 #endif /* CONFIG_INET */
4091 return -EAFNOSUPPORT;
4094 static const struct bpf_func_proto bpf_bind_proto = {
4097 .ret_type = RET_INTEGER,
4098 .arg1_type = ARG_PTR_TO_CTX,
4099 .arg2_type = ARG_PTR_TO_MEM,
4100 .arg3_type = ARG_CONST_SIZE,
4104 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
4105 struct bpf_xfrm_state *, to, u32, size, u64, flags)
4107 const struct sec_path *sp = skb_sec_path(skb);
4108 const struct xfrm_state *x;
4110 if (!sp || unlikely(index >= sp->len || flags))
4113 x = sp->xvec[index];
4115 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
4118 to->reqid = x->props.reqid;
4119 to->spi = x->id.spi;
4120 to->family = x->props.family;
4123 if (to->family == AF_INET6) {
4124 memcpy(to->remote_ipv6, x->props.saddr.a6,
4125 sizeof(to->remote_ipv6));
4127 to->remote_ipv4 = x->props.saddr.a4;
4128 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4133 memset(to, 0, size);
4137 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
4138 .func = bpf_skb_get_xfrm_state,
4140 .ret_type = RET_INTEGER,
4141 .arg1_type = ARG_PTR_TO_CTX,
4142 .arg2_type = ARG_ANYTHING,
4143 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
4144 .arg4_type = ARG_CONST_SIZE,
4145 .arg5_type = ARG_ANYTHING,
4149 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
4150 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
4151 const struct neighbour *neigh,
4152 const struct net_device *dev)
4154 memcpy(params->dmac, neigh->ha, ETH_ALEN);
4155 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
4156 params->h_vlan_TCI = 0;
4157 params->h_vlan_proto = 0;
4158 params->ifindex = dev->ifindex;
4164 #if IS_ENABLED(CONFIG_INET)
4165 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4166 u32 flags, bool check_mtu)
4168 struct in_device *in_dev;
4169 struct neighbour *neigh;
4170 struct net_device *dev;
4171 struct fib_result res;
4177 dev = dev_get_by_index_rcu(net, params->ifindex);
4181 /* verify forwarding is enabled on this interface */
4182 in_dev = __in_dev_get_rcu(dev);
4183 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
4184 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4186 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4188 fl4.flowi4_oif = params->ifindex;
4190 fl4.flowi4_iif = params->ifindex;
4193 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
4194 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
4195 fl4.flowi4_flags = 0;
4197 fl4.flowi4_proto = params->l4_protocol;
4198 fl4.daddr = params->ipv4_dst;
4199 fl4.saddr = params->ipv4_src;
4200 fl4.fl4_sport = params->sport;
4201 fl4.fl4_dport = params->dport;
4203 if (flags & BPF_FIB_LOOKUP_DIRECT) {
4204 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4205 struct fib_table *tb;
4207 tb = fib_get_table(net, tbid);
4209 return BPF_FIB_LKUP_RET_NOT_FWDED;
4211 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
4213 fl4.flowi4_mark = 0;
4214 fl4.flowi4_secid = 0;
4215 fl4.flowi4_tun_key.tun_id = 0;
4216 fl4.flowi4_uid = sock_net_uid(net, NULL);
4218 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
4222 /* map fib lookup errors to RTN_ type */
4224 return BPF_FIB_LKUP_RET_BLACKHOLE;
4225 if (err == -EHOSTUNREACH)
4226 return BPF_FIB_LKUP_RET_UNREACHABLE;
4228 return BPF_FIB_LKUP_RET_PROHIBIT;
4230 return BPF_FIB_LKUP_RET_NOT_FWDED;
4233 if (res.type != RTN_UNICAST)
4234 return BPF_FIB_LKUP_RET_NOT_FWDED;
4236 if (res.fi->fib_nhs > 1)
4237 fib_select_path(net, &res, &fl4, NULL);
4240 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
4241 if (params->tot_len > mtu)
4242 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4245 nh = &res.fi->fib_nh[res.nh_sel];
4247 /* do not handle lwt encaps right now */
4248 if (nh->nh_lwtstate)
4249 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4253 params->ipv4_dst = nh->nh_gw;
4255 params->rt_metric = res.fi->fib_priority;
4257 /* xdp and cls_bpf programs are run in RCU-bh so
4258 * rcu_read_lock_bh is not needed here
4260 neigh = __ipv4_neigh_lookup_noref(dev, (__force u32)params->ipv4_dst);
4262 return BPF_FIB_LKUP_RET_NO_NEIGH;
4264 return bpf_fib_set_fwd_params(params, neigh, dev);
4268 #if IS_ENABLED(CONFIG_IPV6)
4269 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4270 u32 flags, bool check_mtu)
4272 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
4273 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
4274 struct neighbour *neigh;
4275 struct net_device *dev;
4276 struct inet6_dev *idev;
4277 struct fib6_info *f6i;
4283 /* link local addresses are never forwarded */
4284 if (rt6_need_strict(dst) || rt6_need_strict(src))
4285 return BPF_FIB_LKUP_RET_NOT_FWDED;
4287 dev = dev_get_by_index_rcu(net, params->ifindex);
4291 idev = __in6_dev_get_safely(dev);
4292 if (unlikely(!idev || !net->ipv6.devconf_all->forwarding))
4293 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4295 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4297 oif = fl6.flowi6_oif = params->ifindex;
4299 oif = fl6.flowi6_iif = params->ifindex;
4301 strict = RT6_LOOKUP_F_HAS_SADDR;
4303 fl6.flowlabel = params->flowinfo;
4304 fl6.flowi6_scope = 0;
4305 fl6.flowi6_flags = 0;
4308 fl6.flowi6_proto = params->l4_protocol;
4311 fl6.fl6_sport = params->sport;
4312 fl6.fl6_dport = params->dport;
4314 if (flags & BPF_FIB_LOOKUP_DIRECT) {
4315 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4316 struct fib6_table *tb;
4318 tb = ipv6_stub->fib6_get_table(net, tbid);
4320 return BPF_FIB_LKUP_RET_NOT_FWDED;
4322 f6i = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, strict);
4324 fl6.flowi6_mark = 0;
4325 fl6.flowi6_secid = 0;
4326 fl6.flowi6_tun_key.tun_id = 0;
4327 fl6.flowi6_uid = sock_net_uid(net, NULL);
4329 f6i = ipv6_stub->fib6_lookup(net, oif, &fl6, strict);
4332 if (unlikely(IS_ERR_OR_NULL(f6i) || f6i == net->ipv6.fib6_null_entry))
4333 return BPF_FIB_LKUP_RET_NOT_FWDED;
4335 if (unlikely(f6i->fib6_flags & RTF_REJECT)) {
4336 switch (f6i->fib6_type) {
4338 return BPF_FIB_LKUP_RET_BLACKHOLE;
4339 case RTN_UNREACHABLE:
4340 return BPF_FIB_LKUP_RET_UNREACHABLE;
4342 return BPF_FIB_LKUP_RET_PROHIBIT;
4344 return BPF_FIB_LKUP_RET_NOT_FWDED;
4348 if (f6i->fib6_type != RTN_UNICAST)
4349 return BPF_FIB_LKUP_RET_NOT_FWDED;
4351 if (f6i->fib6_nsiblings && fl6.flowi6_oif == 0)
4352 f6i = ipv6_stub->fib6_multipath_select(net, f6i, &fl6,
4353 fl6.flowi6_oif, NULL,
4357 mtu = ipv6_stub->ip6_mtu_from_fib6(f6i, dst, src);
4358 if (params->tot_len > mtu)
4359 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4362 if (f6i->fib6_nh.nh_lwtstate)
4363 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4365 if (f6i->fib6_flags & RTF_GATEWAY)
4366 *dst = f6i->fib6_nh.nh_gw;
4368 dev = f6i->fib6_nh.nh_dev;
4369 params->rt_metric = f6i->fib6_metric;
4371 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
4372 * not needed here. Can not use __ipv6_neigh_lookup_noref here
4373 * because we need to get nd_tbl via the stub
4375 neigh = ___neigh_lookup_noref(ipv6_stub->nd_tbl, neigh_key_eq128,
4376 ndisc_hashfn, dst, dev);
4378 return BPF_FIB_LKUP_RET_NO_NEIGH;
4380 return bpf_fib_set_fwd_params(params, neigh, dev);
4384 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
4385 struct bpf_fib_lookup *, params, int, plen, u32, flags)
4387 if (plen < sizeof(*params))
4390 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4393 switch (params->family) {
4394 #if IS_ENABLED(CONFIG_INET)
4396 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
4399 #if IS_ENABLED(CONFIG_IPV6)
4401 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
4405 return -EAFNOSUPPORT;
4408 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
4409 .func = bpf_xdp_fib_lookup,
4411 .ret_type = RET_INTEGER,
4412 .arg1_type = ARG_PTR_TO_CTX,
4413 .arg2_type = ARG_PTR_TO_MEM,
4414 .arg3_type = ARG_CONST_SIZE,
4415 .arg4_type = ARG_ANYTHING,
4418 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
4419 struct bpf_fib_lookup *, params, int, plen, u32, flags)
4421 struct net *net = dev_net(skb->dev);
4422 int rc = -EAFNOSUPPORT;
4424 if (plen < sizeof(*params))
4427 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4430 switch (params->family) {
4431 #if IS_ENABLED(CONFIG_INET)
4433 rc = bpf_ipv4_fib_lookup(net, params, flags, false);
4436 #if IS_ENABLED(CONFIG_IPV6)
4438 rc = bpf_ipv6_fib_lookup(net, params, flags, false);
4444 struct net_device *dev;
4446 dev = dev_get_by_index_rcu(net, params->ifindex);
4447 if (!is_skb_forwardable(dev, skb))
4448 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
4454 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
4455 .func = bpf_skb_fib_lookup,
4457 .ret_type = RET_INTEGER,
4458 .arg1_type = ARG_PTR_TO_CTX,
4459 .arg2_type = ARG_PTR_TO_MEM,
4460 .arg3_type = ARG_CONST_SIZE,
4461 .arg4_type = ARG_ANYTHING,
4464 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4465 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
4468 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
4470 if (!seg6_validate_srh(srh, len))
4474 case BPF_LWT_ENCAP_SEG6_INLINE:
4475 if (skb->protocol != htons(ETH_P_IPV6))
4478 err = seg6_do_srh_inline(skb, srh);
4480 case BPF_LWT_ENCAP_SEG6:
4481 skb_reset_inner_headers(skb);
4482 skb->encapsulation = 1;
4483 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
4489 bpf_compute_data_pointers(skb);
4493 ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
4494 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
4496 return seg6_lookup_nexthop(skb, NULL, 0);
4498 #endif /* CONFIG_IPV6_SEG6_BPF */
4500 BPF_CALL_4(bpf_lwt_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
4504 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4505 case BPF_LWT_ENCAP_SEG6:
4506 case BPF_LWT_ENCAP_SEG6_INLINE:
4507 return bpf_push_seg6_encap(skb, type, hdr, len);
4514 static const struct bpf_func_proto bpf_lwt_push_encap_proto = {
4515 .func = bpf_lwt_push_encap,
4517 .ret_type = RET_INTEGER,
4518 .arg1_type = ARG_PTR_TO_CTX,
4519 .arg2_type = ARG_ANYTHING,
4520 .arg3_type = ARG_PTR_TO_MEM,
4521 .arg4_type = ARG_CONST_SIZE
4524 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
4525 const void *, from, u32, len)
4527 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4528 struct seg6_bpf_srh_state *srh_state =
4529 this_cpu_ptr(&seg6_bpf_srh_states);
4530 void *srh_tlvs, *srh_end, *ptr;
4531 struct ipv6_sr_hdr *srh;
4534 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
4537 srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4538 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
4539 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
4541 ptr = skb->data + offset;
4542 if (ptr >= srh_tlvs && ptr + len <= srh_end)
4543 srh_state->valid = 0;
4544 else if (ptr < (void *)&srh->flags ||
4545 ptr + len > (void *)&srh->segments)
4548 if (unlikely(bpf_try_make_writable(skb, offset + len)))
4551 memcpy(skb->data + offset, from, len);
4553 #else /* CONFIG_IPV6_SEG6_BPF */
4558 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
4559 .func = bpf_lwt_seg6_store_bytes,
4561 .ret_type = RET_INTEGER,
4562 .arg1_type = ARG_PTR_TO_CTX,
4563 .arg2_type = ARG_ANYTHING,
4564 .arg3_type = ARG_PTR_TO_MEM,
4565 .arg4_type = ARG_CONST_SIZE
4568 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
4569 u32, action, void *, param, u32, param_len)
4571 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4572 struct seg6_bpf_srh_state *srh_state =
4573 this_cpu_ptr(&seg6_bpf_srh_states);
4574 struct ipv6_sr_hdr *srh;
4578 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
4580 srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4582 if (!srh_state->valid) {
4583 if (unlikely((srh_state->hdrlen & 7) != 0))
4586 srh->hdrlen = (u8)(srh_state->hdrlen >> 3);
4587 if (unlikely(!seg6_validate_srh(srh, (srh->hdrlen + 1) << 3)))
4590 srh_state->valid = 1;
4594 case SEG6_LOCAL_ACTION_END_X:
4595 if (param_len != sizeof(struct in6_addr))
4597 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
4598 case SEG6_LOCAL_ACTION_END_T:
4599 if (param_len != sizeof(int))
4601 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
4602 case SEG6_LOCAL_ACTION_END_B6:
4603 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
4607 ((struct ipv6_sr_hdr *)param)->hdrlen << 3;
4609 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
4610 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
4614 ((struct ipv6_sr_hdr *)param)->hdrlen << 3;
4619 #else /* CONFIG_IPV6_SEG6_BPF */
4624 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
4625 .func = bpf_lwt_seg6_action,
4627 .ret_type = RET_INTEGER,
4628 .arg1_type = ARG_PTR_TO_CTX,
4629 .arg2_type = ARG_ANYTHING,
4630 .arg3_type = ARG_PTR_TO_MEM,
4631 .arg4_type = ARG_CONST_SIZE
4634 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
4637 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4638 struct seg6_bpf_srh_state *srh_state =
4639 this_cpu_ptr(&seg6_bpf_srh_states);
4640 void *srh_end, *srh_tlvs, *ptr;
4641 struct ipv6_sr_hdr *srh;
4642 struct ipv6hdr *hdr;
4646 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
4648 srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4650 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
4651 ((srh->first_segment + 1) << 4));
4652 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
4654 ptr = skb->data + offset;
4656 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
4658 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
4662 ret = skb_cow_head(skb, len);
4663 if (unlikely(ret < 0))
4666 ret = bpf_skb_net_hdr_push(skb, offset, len);
4668 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
4671 bpf_compute_data_pointers(skb);
4672 if (unlikely(ret < 0))
4675 hdr = (struct ipv6hdr *)skb->data;
4676 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
4678 srh_state->hdrlen += len;
4679 srh_state->valid = 0;
4681 #else /* CONFIG_IPV6_SEG6_BPF */
4686 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
4687 .func = bpf_lwt_seg6_adjust_srh,
4689 .ret_type = RET_INTEGER,
4690 .arg1_type = ARG_PTR_TO_CTX,
4691 .arg2_type = ARG_ANYTHING,
4692 .arg3_type = ARG_ANYTHING,
4695 bool bpf_helper_changes_pkt_data(void *func)
4697 if (func == bpf_skb_vlan_push ||
4698 func == bpf_skb_vlan_pop ||
4699 func == bpf_skb_store_bytes ||
4700 func == bpf_skb_change_proto ||
4701 func == bpf_skb_change_head ||
4702 func == sk_skb_change_head ||
4703 func == bpf_skb_change_tail ||
4704 func == sk_skb_change_tail ||
4705 func == bpf_skb_adjust_room ||
4706 func == bpf_skb_pull_data ||
4707 func == sk_skb_pull_data ||
4708 func == bpf_clone_redirect ||
4709 func == bpf_l3_csum_replace ||
4710 func == bpf_l4_csum_replace ||
4711 func == bpf_xdp_adjust_head ||
4712 func == bpf_xdp_adjust_meta ||
4713 func == bpf_msg_pull_data ||
4714 func == bpf_xdp_adjust_tail ||
4715 func == bpf_lwt_push_encap ||
4716 func == bpf_lwt_seg6_store_bytes ||
4717 func == bpf_lwt_seg6_adjust_srh ||
4718 func == bpf_lwt_seg6_action
4725 static const struct bpf_func_proto *
4726 bpf_base_func_proto(enum bpf_func_id func_id)
4729 case BPF_FUNC_map_lookup_elem:
4730 return &bpf_map_lookup_elem_proto;
4731 case BPF_FUNC_map_update_elem:
4732 return &bpf_map_update_elem_proto;
4733 case BPF_FUNC_map_delete_elem:
4734 return &bpf_map_delete_elem_proto;
4735 case BPF_FUNC_get_prandom_u32:
4736 return &bpf_get_prandom_u32_proto;
4737 case BPF_FUNC_get_smp_processor_id:
4738 return &bpf_get_raw_smp_processor_id_proto;
4739 case BPF_FUNC_get_numa_node_id:
4740 return &bpf_get_numa_node_id_proto;
4741 case BPF_FUNC_tail_call:
4742 return &bpf_tail_call_proto;
4743 case BPF_FUNC_ktime_get_ns:
4744 return &bpf_ktime_get_ns_proto;
4745 case BPF_FUNC_trace_printk:
4746 if (capable(CAP_SYS_ADMIN))
4747 return bpf_get_trace_printk_proto();
4753 static const struct bpf_func_proto *
4754 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4757 /* inet and inet6 sockets are created in a process
4758 * context so there is always a valid uid/gid
4760 case BPF_FUNC_get_current_uid_gid:
4761 return &bpf_get_current_uid_gid_proto;
4763 return bpf_base_func_proto(func_id);
4767 static const struct bpf_func_proto *
4768 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4771 /* inet and inet6 sockets are created in a process
4772 * context so there is always a valid uid/gid
4774 case BPF_FUNC_get_current_uid_gid:
4775 return &bpf_get_current_uid_gid_proto;
4777 switch (prog->expected_attach_type) {
4778 case BPF_CGROUP_INET4_CONNECT:
4779 case BPF_CGROUP_INET6_CONNECT:
4780 return &bpf_bind_proto;
4785 return bpf_base_func_proto(func_id);
4789 static const struct bpf_func_proto *
4790 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4793 case BPF_FUNC_skb_load_bytes:
4794 return &bpf_skb_load_bytes_proto;
4795 case BPF_FUNC_skb_load_bytes_relative:
4796 return &bpf_skb_load_bytes_relative_proto;
4797 case BPF_FUNC_get_socket_cookie:
4798 return &bpf_get_socket_cookie_proto;
4799 case BPF_FUNC_get_socket_uid:
4800 return &bpf_get_socket_uid_proto;
4802 return bpf_base_func_proto(func_id);
4806 static const struct bpf_func_proto *
4807 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4810 case BPF_FUNC_skb_store_bytes:
4811 return &bpf_skb_store_bytes_proto;
4812 case BPF_FUNC_skb_load_bytes:
4813 return &bpf_skb_load_bytes_proto;
4814 case BPF_FUNC_skb_load_bytes_relative:
4815 return &bpf_skb_load_bytes_relative_proto;
4816 case BPF_FUNC_skb_pull_data:
4817 return &bpf_skb_pull_data_proto;
4818 case BPF_FUNC_csum_diff:
4819 return &bpf_csum_diff_proto;
4820 case BPF_FUNC_csum_update:
4821 return &bpf_csum_update_proto;
4822 case BPF_FUNC_l3_csum_replace:
4823 return &bpf_l3_csum_replace_proto;
4824 case BPF_FUNC_l4_csum_replace:
4825 return &bpf_l4_csum_replace_proto;
4826 case BPF_FUNC_clone_redirect:
4827 return &bpf_clone_redirect_proto;
4828 case BPF_FUNC_get_cgroup_classid:
4829 return &bpf_get_cgroup_classid_proto;
4830 case BPF_FUNC_skb_vlan_push:
4831 return &bpf_skb_vlan_push_proto;
4832 case BPF_FUNC_skb_vlan_pop:
4833 return &bpf_skb_vlan_pop_proto;
4834 case BPF_FUNC_skb_change_proto:
4835 return &bpf_skb_change_proto_proto;
4836 case BPF_FUNC_skb_change_type:
4837 return &bpf_skb_change_type_proto;
4838 case BPF_FUNC_skb_adjust_room:
4839 return &bpf_skb_adjust_room_proto;
4840 case BPF_FUNC_skb_change_tail:
4841 return &bpf_skb_change_tail_proto;
4842 case BPF_FUNC_skb_get_tunnel_key:
4843 return &bpf_skb_get_tunnel_key_proto;
4844 case BPF_FUNC_skb_set_tunnel_key:
4845 return bpf_get_skb_set_tunnel_proto(func_id);
4846 case BPF_FUNC_skb_get_tunnel_opt:
4847 return &bpf_skb_get_tunnel_opt_proto;
4848 case BPF_FUNC_skb_set_tunnel_opt:
4849 return bpf_get_skb_set_tunnel_proto(func_id);
4850 case BPF_FUNC_redirect:
4851 return &bpf_redirect_proto;
4852 case BPF_FUNC_get_route_realm:
4853 return &bpf_get_route_realm_proto;
4854 case BPF_FUNC_get_hash_recalc:
4855 return &bpf_get_hash_recalc_proto;
4856 case BPF_FUNC_set_hash_invalid:
4857 return &bpf_set_hash_invalid_proto;
4858 case BPF_FUNC_set_hash:
4859 return &bpf_set_hash_proto;
4860 case BPF_FUNC_perf_event_output:
4861 return &bpf_skb_event_output_proto;
4862 case BPF_FUNC_get_smp_processor_id:
4863 return &bpf_get_smp_processor_id_proto;
4864 case BPF_FUNC_skb_under_cgroup:
4865 return &bpf_skb_under_cgroup_proto;
4866 case BPF_FUNC_get_socket_cookie:
4867 return &bpf_get_socket_cookie_proto;
4868 case BPF_FUNC_get_socket_uid:
4869 return &bpf_get_socket_uid_proto;
4870 case BPF_FUNC_fib_lookup:
4871 return &bpf_skb_fib_lookup_proto;
4873 case BPF_FUNC_skb_get_xfrm_state:
4874 return &bpf_skb_get_xfrm_state_proto;
4876 #ifdef CONFIG_SOCK_CGROUP_DATA
4877 case BPF_FUNC_skb_cgroup_id:
4878 return &bpf_skb_cgroup_id_proto;
4881 return bpf_base_func_proto(func_id);
4885 static const struct bpf_func_proto *
4886 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4889 case BPF_FUNC_perf_event_output:
4890 return &bpf_xdp_event_output_proto;
4891 case BPF_FUNC_get_smp_processor_id:
4892 return &bpf_get_smp_processor_id_proto;
4893 case BPF_FUNC_csum_diff:
4894 return &bpf_csum_diff_proto;
4895 case BPF_FUNC_xdp_adjust_head:
4896 return &bpf_xdp_adjust_head_proto;
4897 case BPF_FUNC_xdp_adjust_meta:
4898 return &bpf_xdp_adjust_meta_proto;
4899 case BPF_FUNC_redirect:
4900 return &bpf_xdp_redirect_proto;
4901 case BPF_FUNC_redirect_map:
4902 return &bpf_xdp_redirect_map_proto;
4903 case BPF_FUNC_xdp_adjust_tail:
4904 return &bpf_xdp_adjust_tail_proto;
4905 case BPF_FUNC_fib_lookup:
4906 return &bpf_xdp_fib_lookup_proto;
4908 return bpf_base_func_proto(func_id);
4912 static const struct bpf_func_proto *
4913 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4916 case BPF_FUNC_setsockopt:
4917 return &bpf_setsockopt_proto;
4918 case BPF_FUNC_getsockopt:
4919 return &bpf_getsockopt_proto;
4920 case BPF_FUNC_sock_ops_cb_flags_set:
4921 return &bpf_sock_ops_cb_flags_set_proto;
4922 case BPF_FUNC_sock_map_update:
4923 return &bpf_sock_map_update_proto;
4924 case BPF_FUNC_sock_hash_update:
4925 return &bpf_sock_hash_update_proto;
4927 return bpf_base_func_proto(func_id);
4931 static const struct bpf_func_proto *
4932 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4935 case BPF_FUNC_msg_redirect_map:
4936 return &bpf_msg_redirect_map_proto;
4937 case BPF_FUNC_msg_redirect_hash:
4938 return &bpf_msg_redirect_hash_proto;
4939 case BPF_FUNC_msg_apply_bytes:
4940 return &bpf_msg_apply_bytes_proto;
4941 case BPF_FUNC_msg_cork_bytes:
4942 return &bpf_msg_cork_bytes_proto;
4943 case BPF_FUNC_msg_pull_data:
4944 return &bpf_msg_pull_data_proto;
4946 return bpf_base_func_proto(func_id);
4950 static const struct bpf_func_proto *
4951 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4954 case BPF_FUNC_skb_store_bytes:
4955 return &bpf_skb_store_bytes_proto;
4956 case BPF_FUNC_skb_load_bytes:
4957 return &bpf_skb_load_bytes_proto;
4958 case BPF_FUNC_skb_pull_data:
4959 return &sk_skb_pull_data_proto;
4960 case BPF_FUNC_skb_change_tail:
4961 return &sk_skb_change_tail_proto;
4962 case BPF_FUNC_skb_change_head:
4963 return &sk_skb_change_head_proto;
4964 case BPF_FUNC_get_socket_cookie:
4965 return &bpf_get_socket_cookie_proto;
4966 case BPF_FUNC_get_socket_uid:
4967 return &bpf_get_socket_uid_proto;
4968 case BPF_FUNC_sk_redirect_map:
4969 return &bpf_sk_redirect_map_proto;
4970 case BPF_FUNC_sk_redirect_hash:
4971 return &bpf_sk_redirect_hash_proto;
4973 return bpf_base_func_proto(func_id);
4977 static const struct bpf_func_proto *
4978 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4981 case BPF_FUNC_skb_load_bytes:
4982 return &bpf_skb_load_bytes_proto;
4983 case BPF_FUNC_skb_pull_data:
4984 return &bpf_skb_pull_data_proto;
4985 case BPF_FUNC_csum_diff:
4986 return &bpf_csum_diff_proto;
4987 case BPF_FUNC_get_cgroup_classid:
4988 return &bpf_get_cgroup_classid_proto;
4989 case BPF_FUNC_get_route_realm:
4990 return &bpf_get_route_realm_proto;
4991 case BPF_FUNC_get_hash_recalc:
4992 return &bpf_get_hash_recalc_proto;
4993 case BPF_FUNC_perf_event_output:
4994 return &bpf_skb_event_output_proto;
4995 case BPF_FUNC_get_smp_processor_id:
4996 return &bpf_get_smp_processor_id_proto;
4997 case BPF_FUNC_skb_under_cgroup:
4998 return &bpf_skb_under_cgroup_proto;
5000 return bpf_base_func_proto(func_id);
5004 static const struct bpf_func_proto *
5005 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5008 case BPF_FUNC_lwt_push_encap:
5009 return &bpf_lwt_push_encap_proto;
5011 return lwt_out_func_proto(func_id, prog);
5015 static const struct bpf_func_proto *
5016 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5019 case BPF_FUNC_skb_get_tunnel_key:
5020 return &bpf_skb_get_tunnel_key_proto;
5021 case BPF_FUNC_skb_set_tunnel_key:
5022 return bpf_get_skb_set_tunnel_proto(func_id);
5023 case BPF_FUNC_skb_get_tunnel_opt:
5024 return &bpf_skb_get_tunnel_opt_proto;
5025 case BPF_FUNC_skb_set_tunnel_opt:
5026 return bpf_get_skb_set_tunnel_proto(func_id);
5027 case BPF_FUNC_redirect:
5028 return &bpf_redirect_proto;
5029 case BPF_FUNC_clone_redirect:
5030 return &bpf_clone_redirect_proto;
5031 case BPF_FUNC_skb_change_tail:
5032 return &bpf_skb_change_tail_proto;
5033 case BPF_FUNC_skb_change_head:
5034 return &bpf_skb_change_head_proto;
5035 case BPF_FUNC_skb_store_bytes:
5036 return &bpf_skb_store_bytes_proto;
5037 case BPF_FUNC_csum_update:
5038 return &bpf_csum_update_proto;
5039 case BPF_FUNC_l3_csum_replace:
5040 return &bpf_l3_csum_replace_proto;
5041 case BPF_FUNC_l4_csum_replace:
5042 return &bpf_l4_csum_replace_proto;
5043 case BPF_FUNC_set_hash_invalid:
5044 return &bpf_set_hash_invalid_proto;
5046 return lwt_out_func_proto(func_id, prog);
5050 static const struct bpf_func_proto *
5051 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5054 case BPF_FUNC_lwt_seg6_store_bytes:
5055 return &bpf_lwt_seg6_store_bytes_proto;
5056 case BPF_FUNC_lwt_seg6_action:
5057 return &bpf_lwt_seg6_action_proto;
5058 case BPF_FUNC_lwt_seg6_adjust_srh:
5059 return &bpf_lwt_seg6_adjust_srh_proto;
5061 return lwt_out_func_proto(func_id, prog);
5065 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
5066 const struct bpf_prog *prog,
5067 struct bpf_insn_access_aux *info)
5069 const int size_default = sizeof(__u32);
5071 if (off < 0 || off >= sizeof(struct __sk_buff))
5074 /* The verifier guarantees that size > 0. */
5075 if (off % size != 0)
5079 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5080 if (off + size > offsetofend(struct __sk_buff, cb[4]))
5083 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
5084 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
5085 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
5086 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
5087 case bpf_ctx_range(struct __sk_buff, data):
5088 case bpf_ctx_range(struct __sk_buff, data_meta):
5089 case bpf_ctx_range(struct __sk_buff, data_end):
5090 if (size != size_default)
5094 /* Only narrow read access allowed for now. */
5095 if (type == BPF_WRITE) {
5096 if (size != size_default)
5099 bpf_ctx_record_field_size(info, size_default);
5100 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
5108 static bool sk_filter_is_valid_access(int off, int size,
5109 enum bpf_access_type type,
5110 const struct bpf_prog *prog,
5111 struct bpf_insn_access_aux *info)
5114 case bpf_ctx_range(struct __sk_buff, tc_classid):
5115 case bpf_ctx_range(struct __sk_buff, data):
5116 case bpf_ctx_range(struct __sk_buff, data_meta):
5117 case bpf_ctx_range(struct __sk_buff, data_end):
5118 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
5122 if (type == BPF_WRITE) {
5124 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5131 return bpf_skb_is_valid_access(off, size, type, prog, info);
5134 static bool lwt_is_valid_access(int off, int size,
5135 enum bpf_access_type type,
5136 const struct bpf_prog *prog,
5137 struct bpf_insn_access_aux *info)
5140 case bpf_ctx_range(struct __sk_buff, tc_classid):
5141 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
5142 case bpf_ctx_range(struct __sk_buff, data_meta):
5146 if (type == BPF_WRITE) {
5148 case bpf_ctx_range(struct __sk_buff, mark):
5149 case bpf_ctx_range(struct __sk_buff, priority):
5150 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5158 case bpf_ctx_range(struct __sk_buff, data):
5159 info->reg_type = PTR_TO_PACKET;
5161 case bpf_ctx_range(struct __sk_buff, data_end):
5162 info->reg_type = PTR_TO_PACKET_END;
5166 return bpf_skb_is_valid_access(off, size, type, prog, info);
5169 /* Attach type specific accesses */
5170 static bool __sock_filter_check_attach_type(int off,
5171 enum bpf_access_type access_type,
5172 enum bpf_attach_type attach_type)
5175 case offsetof(struct bpf_sock, bound_dev_if):
5176 case offsetof(struct bpf_sock, mark):
5177 case offsetof(struct bpf_sock, priority):
5178 switch (attach_type) {
5179 case BPF_CGROUP_INET_SOCK_CREATE:
5184 case bpf_ctx_range(struct bpf_sock, src_ip4):
5185 switch (attach_type) {
5186 case BPF_CGROUP_INET4_POST_BIND:
5191 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
5192 switch (attach_type) {
5193 case BPF_CGROUP_INET6_POST_BIND:
5198 case bpf_ctx_range(struct bpf_sock, src_port):
5199 switch (attach_type) {
5200 case BPF_CGROUP_INET4_POST_BIND:
5201 case BPF_CGROUP_INET6_POST_BIND:
5208 return access_type == BPF_READ;
5213 static bool __sock_filter_check_size(int off, int size,
5214 struct bpf_insn_access_aux *info)
5216 const int size_default = sizeof(__u32);
5219 case bpf_ctx_range(struct bpf_sock, src_ip4):
5220 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
5221 bpf_ctx_record_field_size(info, size_default);
5222 return bpf_ctx_narrow_access_ok(off, size, size_default);
5225 return size == size_default;
5228 static bool sock_filter_is_valid_access(int off, int size,
5229 enum bpf_access_type type,
5230 const struct bpf_prog *prog,
5231 struct bpf_insn_access_aux *info)
5233 if (off < 0 || off >= sizeof(struct bpf_sock))
5235 if (off % size != 0)
5237 if (!__sock_filter_check_attach_type(off, type,
5238 prog->expected_attach_type))
5240 if (!__sock_filter_check_size(off, size, info))
5245 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
5246 const struct bpf_prog *prog, int drop_verdict)
5248 struct bpf_insn *insn = insn_buf;
5253 /* if (!skb->cloned)
5256 * (Fast-path, otherwise approximation that we might be
5257 * a clone, do the rest in helper.)
5259 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
5260 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
5261 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
5263 /* ret = bpf_skb_pull_data(skb, 0); */
5264 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
5265 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
5266 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
5267 BPF_FUNC_skb_pull_data);
5270 * return TC_ACT_SHOT;
5272 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
5273 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
5274 *insn++ = BPF_EXIT_INSN();
5277 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
5279 *insn++ = prog->insnsi[0];
5281 return insn - insn_buf;
5284 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
5285 struct bpf_insn *insn_buf)
5287 bool indirect = BPF_MODE(orig->code) == BPF_IND;
5288 struct bpf_insn *insn = insn_buf;
5290 /* We're guaranteed here that CTX is in R6. */
5291 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
5293 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
5295 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
5297 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
5300 switch (BPF_SIZE(orig->code)) {
5302 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
5305 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
5308 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
5312 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
5313 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
5314 *insn++ = BPF_EXIT_INSN();
5316 return insn - insn_buf;
5319 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
5320 const struct bpf_prog *prog)
5322 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
5325 static bool tc_cls_act_is_valid_access(int off, int size,
5326 enum bpf_access_type type,
5327 const struct bpf_prog *prog,
5328 struct bpf_insn_access_aux *info)
5330 if (type == BPF_WRITE) {
5332 case bpf_ctx_range(struct __sk_buff, mark):
5333 case bpf_ctx_range(struct __sk_buff, tc_index):
5334 case bpf_ctx_range(struct __sk_buff, priority):
5335 case bpf_ctx_range(struct __sk_buff, tc_classid):
5336 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5344 case bpf_ctx_range(struct __sk_buff, data):
5345 info->reg_type = PTR_TO_PACKET;
5347 case bpf_ctx_range(struct __sk_buff, data_meta):
5348 info->reg_type = PTR_TO_PACKET_META;
5350 case bpf_ctx_range(struct __sk_buff, data_end):
5351 info->reg_type = PTR_TO_PACKET_END;
5353 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
5357 return bpf_skb_is_valid_access(off, size, type, prog, info);
5360 static bool __is_valid_xdp_access(int off, int size)
5362 if (off < 0 || off >= sizeof(struct xdp_md))
5364 if (off % size != 0)
5366 if (size != sizeof(__u32))
5372 static bool xdp_is_valid_access(int off, int size,
5373 enum bpf_access_type type,
5374 const struct bpf_prog *prog,
5375 struct bpf_insn_access_aux *info)
5377 if (type == BPF_WRITE) {
5378 if (bpf_prog_is_dev_bound(prog->aux)) {
5380 case offsetof(struct xdp_md, rx_queue_index):
5381 return __is_valid_xdp_access(off, size);
5388 case offsetof(struct xdp_md, data):
5389 info->reg_type = PTR_TO_PACKET;
5391 case offsetof(struct xdp_md, data_meta):
5392 info->reg_type = PTR_TO_PACKET_META;
5394 case offsetof(struct xdp_md, data_end):
5395 info->reg_type = PTR_TO_PACKET_END;
5399 return __is_valid_xdp_access(off, size);
5402 void bpf_warn_invalid_xdp_action(u32 act)
5404 const u32 act_max = XDP_REDIRECT;
5406 WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
5407 act > act_max ? "Illegal" : "Driver unsupported",
5410 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
5412 static bool sock_addr_is_valid_access(int off, int size,
5413 enum bpf_access_type type,
5414 const struct bpf_prog *prog,
5415 struct bpf_insn_access_aux *info)
5417 const int size_default = sizeof(__u32);
5419 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
5421 if (off % size != 0)
5424 /* Disallow access to IPv6 fields from IPv4 contex and vise
5428 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
5429 switch (prog->expected_attach_type) {
5430 case BPF_CGROUP_INET4_BIND:
5431 case BPF_CGROUP_INET4_CONNECT:
5432 case BPF_CGROUP_UDP4_SENDMSG:
5438 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
5439 switch (prog->expected_attach_type) {
5440 case BPF_CGROUP_INET6_BIND:
5441 case BPF_CGROUP_INET6_CONNECT:
5442 case BPF_CGROUP_UDP6_SENDMSG:
5448 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
5449 switch (prog->expected_attach_type) {
5450 case BPF_CGROUP_UDP4_SENDMSG:
5456 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
5458 switch (prog->expected_attach_type) {
5459 case BPF_CGROUP_UDP6_SENDMSG:
5468 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
5469 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
5470 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
5471 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
5473 /* Only narrow read access allowed for now. */
5474 if (type == BPF_READ) {
5475 bpf_ctx_record_field_size(info, size_default);
5476 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
5479 if (size != size_default)
5483 case bpf_ctx_range(struct bpf_sock_addr, user_port):
5484 if (size != size_default)
5488 if (type == BPF_READ) {
5489 if (size != size_default)
5499 static bool sock_ops_is_valid_access(int off, int size,
5500 enum bpf_access_type type,
5501 const struct bpf_prog *prog,
5502 struct bpf_insn_access_aux *info)
5504 const int size_default = sizeof(__u32);
5506 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
5509 /* The verifier guarantees that size > 0. */
5510 if (off % size != 0)
5513 if (type == BPF_WRITE) {
5515 case offsetof(struct bpf_sock_ops, reply):
5516 case offsetof(struct bpf_sock_ops, sk_txhash):
5517 if (size != size_default)
5525 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
5527 if (size != sizeof(__u64))
5531 if (size != size_default)
5540 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
5541 const struct bpf_prog *prog)
5543 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
5546 static bool sk_skb_is_valid_access(int off, int size,
5547 enum bpf_access_type type,
5548 const struct bpf_prog *prog,
5549 struct bpf_insn_access_aux *info)
5552 case bpf_ctx_range(struct __sk_buff, tc_classid):
5553 case bpf_ctx_range(struct __sk_buff, data_meta):
5557 if (type == BPF_WRITE) {
5559 case bpf_ctx_range(struct __sk_buff, tc_index):
5560 case bpf_ctx_range(struct __sk_buff, priority):
5568 case bpf_ctx_range(struct __sk_buff, mark):
5570 case bpf_ctx_range(struct __sk_buff, data):
5571 info->reg_type = PTR_TO_PACKET;
5573 case bpf_ctx_range(struct __sk_buff, data_end):
5574 info->reg_type = PTR_TO_PACKET_END;
5578 return bpf_skb_is_valid_access(off, size, type, prog, info);
5581 static bool sk_msg_is_valid_access(int off, int size,
5582 enum bpf_access_type type,
5583 const struct bpf_prog *prog,
5584 struct bpf_insn_access_aux *info)
5586 if (type == BPF_WRITE)
5590 case offsetof(struct sk_msg_md, data):
5591 info->reg_type = PTR_TO_PACKET;
5592 if (size != sizeof(__u64))
5595 case offsetof(struct sk_msg_md, data_end):
5596 info->reg_type = PTR_TO_PACKET_END;
5597 if (size != sizeof(__u64))
5601 if (size != sizeof(__u32))
5605 if (off < 0 || off >= sizeof(struct sk_msg_md))
5607 if (off % size != 0)
5613 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
5614 const struct bpf_insn *si,
5615 struct bpf_insn *insn_buf,
5616 struct bpf_prog *prog, u32 *target_size)
5618 struct bpf_insn *insn = insn_buf;
5622 case offsetof(struct __sk_buff, len):
5623 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5624 bpf_target_off(struct sk_buff, len, 4,
5628 case offsetof(struct __sk_buff, protocol):
5629 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5630 bpf_target_off(struct sk_buff, protocol, 2,
5634 case offsetof(struct __sk_buff, vlan_proto):
5635 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5636 bpf_target_off(struct sk_buff, vlan_proto, 2,
5640 case offsetof(struct __sk_buff, priority):
5641 if (type == BPF_WRITE)
5642 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
5643 bpf_target_off(struct sk_buff, priority, 4,
5646 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5647 bpf_target_off(struct sk_buff, priority, 4,
5651 case offsetof(struct __sk_buff, ingress_ifindex):
5652 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5653 bpf_target_off(struct sk_buff, skb_iif, 4,
5657 case offsetof(struct __sk_buff, ifindex):
5658 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
5659 si->dst_reg, si->src_reg,
5660 offsetof(struct sk_buff, dev));
5661 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
5662 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
5663 bpf_target_off(struct net_device, ifindex, 4,
5667 case offsetof(struct __sk_buff, hash):
5668 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5669 bpf_target_off(struct sk_buff, hash, 4,
5673 case offsetof(struct __sk_buff, mark):
5674 if (type == BPF_WRITE)
5675 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
5676 bpf_target_off(struct sk_buff, mark, 4,
5679 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5680 bpf_target_off(struct sk_buff, mark, 4,
5684 case offsetof(struct __sk_buff, pkt_type):
5686 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
5688 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
5689 #ifdef __BIG_ENDIAN_BITFIELD
5690 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
5694 case offsetof(struct __sk_buff, queue_mapping):
5695 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5696 bpf_target_off(struct sk_buff, queue_mapping, 2,
5700 case offsetof(struct __sk_buff, vlan_present):
5701 case offsetof(struct __sk_buff, vlan_tci):
5702 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000);
5704 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5705 bpf_target_off(struct sk_buff, vlan_tci, 2,
5707 if (si->off == offsetof(struct __sk_buff, vlan_tci)) {
5708 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg,
5711 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 12);
5712 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
5716 case offsetof(struct __sk_buff, cb[0]) ...
5717 offsetofend(struct __sk_buff, cb[4]) - 1:
5718 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20);
5719 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
5720 offsetof(struct qdisc_skb_cb, data)) %
5723 prog->cb_access = 1;
5725 off -= offsetof(struct __sk_buff, cb[0]);
5726 off += offsetof(struct sk_buff, cb);
5727 off += offsetof(struct qdisc_skb_cb, data);
5728 if (type == BPF_WRITE)
5729 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
5732 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
5736 case offsetof(struct __sk_buff, tc_classid):
5737 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, tc_classid) != 2);
5740 off -= offsetof(struct __sk_buff, tc_classid);
5741 off += offsetof(struct sk_buff, cb);
5742 off += offsetof(struct qdisc_skb_cb, tc_classid);
5744 if (type == BPF_WRITE)
5745 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
5748 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
5752 case offsetof(struct __sk_buff, data):
5753 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
5754 si->dst_reg, si->src_reg,
5755 offsetof(struct sk_buff, data));
5758 case offsetof(struct __sk_buff, data_meta):
5760 off -= offsetof(struct __sk_buff, data_meta);
5761 off += offsetof(struct sk_buff, cb);
5762 off += offsetof(struct bpf_skb_data_end, data_meta);
5763 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
5767 case offsetof(struct __sk_buff, data_end):
5769 off -= offsetof(struct __sk_buff, data_end);
5770 off += offsetof(struct sk_buff, cb);
5771 off += offsetof(struct bpf_skb_data_end, data_end);
5772 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
5776 case offsetof(struct __sk_buff, tc_index):
5777 #ifdef CONFIG_NET_SCHED
5778 if (type == BPF_WRITE)
5779 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
5780 bpf_target_off(struct sk_buff, tc_index, 2,
5783 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5784 bpf_target_off(struct sk_buff, tc_index, 2,
5788 if (type == BPF_WRITE)
5789 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
5791 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
5795 case offsetof(struct __sk_buff, napi_id):
5796 #if defined(CONFIG_NET_RX_BUSY_POLL)
5797 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5798 bpf_target_off(struct sk_buff, napi_id, 4,
5800 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
5801 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
5804 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
5807 case offsetof(struct __sk_buff, family):
5808 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
5810 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
5811 si->dst_reg, si->src_reg,
5812 offsetof(struct sk_buff, sk));
5813 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
5814 bpf_target_off(struct sock_common,
5818 case offsetof(struct __sk_buff, remote_ip4):
5819 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
5821 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
5822 si->dst_reg, si->src_reg,
5823 offsetof(struct sk_buff, sk));
5824 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
5825 bpf_target_off(struct sock_common,
5829 case offsetof(struct __sk_buff, local_ip4):
5830 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
5831 skc_rcv_saddr) != 4);
5833 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
5834 si->dst_reg, si->src_reg,
5835 offsetof(struct sk_buff, sk));
5836 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
5837 bpf_target_off(struct sock_common,
5841 case offsetof(struct __sk_buff, remote_ip6[0]) ...
5842 offsetof(struct __sk_buff, remote_ip6[3]):
5843 #if IS_ENABLED(CONFIG_IPV6)
5844 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
5845 skc_v6_daddr.s6_addr32[0]) != 4);
5848 off -= offsetof(struct __sk_buff, remote_ip6[0]);
5850 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
5851 si->dst_reg, si->src_reg,
5852 offsetof(struct sk_buff, sk));
5853 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
5854 offsetof(struct sock_common,
5855 skc_v6_daddr.s6_addr32[0]) +
5858 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
5861 case offsetof(struct __sk_buff, local_ip6[0]) ...
5862 offsetof(struct __sk_buff, local_ip6[3]):
5863 #if IS_ENABLED(CONFIG_IPV6)
5864 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
5865 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
5868 off -= offsetof(struct __sk_buff, local_ip6[0]);
5870 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
5871 si->dst_reg, si->src_reg,
5872 offsetof(struct sk_buff, sk));
5873 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
5874 offsetof(struct sock_common,
5875 skc_v6_rcv_saddr.s6_addr32[0]) +
5878 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
5882 case offsetof(struct __sk_buff, remote_port):
5883 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
5885 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
5886 si->dst_reg, si->src_reg,
5887 offsetof(struct sk_buff, sk));
5888 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
5889 bpf_target_off(struct sock_common,
5892 #ifndef __BIG_ENDIAN_BITFIELD
5893 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
5897 case offsetof(struct __sk_buff, local_port):
5898 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
5900 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
5901 si->dst_reg, si->src_reg,
5902 offsetof(struct sk_buff, sk));
5903 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
5904 bpf_target_off(struct sock_common,
5905 skc_num, 2, target_size));
5909 return insn - insn_buf;
5912 static u32 sock_filter_convert_ctx_access(enum bpf_access_type type,
5913 const struct bpf_insn *si,
5914 struct bpf_insn *insn_buf,
5915 struct bpf_prog *prog, u32 *target_size)
5917 struct bpf_insn *insn = insn_buf;
5921 case offsetof(struct bpf_sock, bound_dev_if):
5922 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_bound_dev_if) != 4);
5924 if (type == BPF_WRITE)
5925 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
5926 offsetof(struct sock, sk_bound_dev_if));
5928 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5929 offsetof(struct sock, sk_bound_dev_if));
5932 case offsetof(struct bpf_sock, mark):
5933 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_mark) != 4);
5935 if (type == BPF_WRITE)
5936 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
5937 offsetof(struct sock, sk_mark));
5939 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5940 offsetof(struct sock, sk_mark));
5943 case offsetof(struct bpf_sock, priority):
5944 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_priority) != 4);
5946 if (type == BPF_WRITE)
5947 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
5948 offsetof(struct sock, sk_priority));
5950 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5951 offsetof(struct sock, sk_priority));
5954 case offsetof(struct bpf_sock, family):
5955 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_family) != 2);
5957 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5958 offsetof(struct sock, sk_family));
5961 case offsetof(struct bpf_sock, type):
5962 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5963 offsetof(struct sock, __sk_flags_offset));
5964 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
5965 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
5968 case offsetof(struct bpf_sock, protocol):
5969 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5970 offsetof(struct sock, __sk_flags_offset));
5971 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
5972 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_PROTO_SHIFT);
5975 case offsetof(struct bpf_sock, src_ip4):
5976 *insn++ = BPF_LDX_MEM(
5977 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
5978 bpf_target_off(struct sock_common, skc_rcv_saddr,
5979 FIELD_SIZEOF(struct sock_common,
5984 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
5985 #if IS_ENABLED(CONFIG_IPV6)
5987 off -= offsetof(struct bpf_sock, src_ip6[0]);
5988 *insn++ = BPF_LDX_MEM(
5989 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
5992 skc_v6_rcv_saddr.s6_addr32[0],
5993 FIELD_SIZEOF(struct sock_common,
5994 skc_v6_rcv_saddr.s6_addr32[0]),
5995 target_size) + off);
5998 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6002 case offsetof(struct bpf_sock, src_port):
6003 *insn++ = BPF_LDX_MEM(
6004 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
6005 si->dst_reg, si->src_reg,
6006 bpf_target_off(struct sock_common, skc_num,
6007 FIELD_SIZEOF(struct sock_common,
6013 return insn - insn_buf;
6016 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
6017 const struct bpf_insn *si,
6018 struct bpf_insn *insn_buf,
6019 struct bpf_prog *prog, u32 *target_size)
6021 struct bpf_insn *insn = insn_buf;
6024 case offsetof(struct __sk_buff, ifindex):
6025 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
6026 si->dst_reg, si->src_reg,
6027 offsetof(struct sk_buff, dev));
6028 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6029 bpf_target_off(struct net_device, ifindex, 4,
6033 return bpf_convert_ctx_access(type, si, insn_buf, prog,
6037 return insn - insn_buf;
6040 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
6041 const struct bpf_insn *si,
6042 struct bpf_insn *insn_buf,
6043 struct bpf_prog *prog, u32 *target_size)
6045 struct bpf_insn *insn = insn_buf;
6048 case offsetof(struct xdp_md, data):
6049 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
6050 si->dst_reg, si->src_reg,
6051 offsetof(struct xdp_buff, data));
6053 case offsetof(struct xdp_md, data_meta):
6054 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
6055 si->dst_reg, si->src_reg,
6056 offsetof(struct xdp_buff, data_meta));
6058 case offsetof(struct xdp_md, data_end):
6059 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
6060 si->dst_reg, si->src_reg,
6061 offsetof(struct xdp_buff, data_end));
6063 case offsetof(struct xdp_md, ingress_ifindex):
6064 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
6065 si->dst_reg, si->src_reg,
6066 offsetof(struct xdp_buff, rxq));
6067 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
6068 si->dst_reg, si->dst_reg,
6069 offsetof(struct xdp_rxq_info, dev));
6070 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6071 offsetof(struct net_device, ifindex));
6073 case offsetof(struct xdp_md, rx_queue_index):
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_W, si->dst_reg, si->dst_reg,
6078 offsetof(struct xdp_rxq_info,
6083 return insn - insn_buf;
6086 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
6087 * context Structure, F is Field in context structure that contains a pointer
6088 * to Nested Structure of type NS that has the field NF.
6090 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
6091 * sure that SIZE is not greater than actual size of S.F.NF.
6093 * If offset OFF is provided, the load happens from that offset relative to
6096 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
6098 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
6099 si->src_reg, offsetof(S, F)); \
6100 *insn++ = BPF_LDX_MEM( \
6101 SIZE, si->dst_reg, si->dst_reg, \
6102 bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF), \
6107 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
6108 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
6109 BPF_FIELD_SIZEOF(NS, NF), 0)
6111 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
6112 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
6114 * It doesn't support SIZE argument though since narrow stores are not
6115 * supported for now.
6117 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
6118 * "register" since two registers available in convert_ctx_access are not
6119 * enough: we can't override neither SRC, since it contains value to store, nor
6120 * DST since it contains pointer to context that may be used by later
6121 * instructions. But we need a temporary place to save pointer to nested
6122 * structure whose field we want to store to.
6124 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, OFF, TF) \
6126 int tmp_reg = BPF_REG_9; \
6127 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
6129 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
6131 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
6133 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
6134 si->dst_reg, offsetof(S, F)); \
6135 *insn++ = BPF_STX_MEM( \
6136 BPF_FIELD_SIZEOF(NS, NF), tmp_reg, si->src_reg, \
6137 bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF), \
6140 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
6144 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
6147 if (type == BPF_WRITE) { \
6148 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, OFF, \
6151 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
6152 S, NS, F, NF, SIZE, OFF); \
6156 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
6157 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
6158 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
6160 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
6161 const struct bpf_insn *si,
6162 struct bpf_insn *insn_buf,
6163 struct bpf_prog *prog, u32 *target_size)
6165 struct bpf_insn *insn = insn_buf;
6169 case offsetof(struct bpf_sock_addr, user_family):
6170 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
6171 struct sockaddr, uaddr, sa_family);
6174 case offsetof(struct bpf_sock_addr, user_ip4):
6175 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
6176 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
6177 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
6180 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
6182 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
6183 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
6184 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
6185 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
6189 case offsetof(struct bpf_sock_addr, user_port):
6190 /* To get port we need to know sa_family first and then treat
6191 * sockaddr as either sockaddr_in or sockaddr_in6.
6192 * Though we can simplify since port field has same offset and
6193 * size in both structures.
6194 * Here we check this invariant and use just one of the
6195 * structures if it's true.
6197 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
6198 offsetof(struct sockaddr_in6, sin6_port));
6199 BUILD_BUG_ON(FIELD_SIZEOF(struct sockaddr_in, sin_port) !=
6200 FIELD_SIZEOF(struct sockaddr_in6, sin6_port));
6201 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(struct bpf_sock_addr_kern,
6202 struct sockaddr_in6, uaddr,
6203 sin6_port, tmp_reg);
6206 case offsetof(struct bpf_sock_addr, family):
6207 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
6208 struct sock, sk, sk_family);
6211 case offsetof(struct bpf_sock_addr, type):
6212 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(
6213 struct bpf_sock_addr_kern, struct sock, sk,
6214 __sk_flags_offset, BPF_W, 0);
6215 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
6216 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
6219 case offsetof(struct bpf_sock_addr, protocol):
6220 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(
6221 struct bpf_sock_addr_kern, struct sock, sk,
6222 __sk_flags_offset, BPF_W, 0);
6223 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
6224 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg,
6228 case offsetof(struct bpf_sock_addr, msg_src_ip4):
6229 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
6230 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
6231 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
6232 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
6235 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
6238 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
6239 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
6240 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
6241 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
6242 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
6246 return insn - insn_buf;
6249 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
6250 const struct bpf_insn *si,
6251 struct bpf_insn *insn_buf,
6252 struct bpf_prog *prog,
6255 struct bpf_insn *insn = insn_buf;
6259 case offsetof(struct bpf_sock_ops, op) ...
6260 offsetof(struct bpf_sock_ops, replylong[3]):
6261 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, op) !=
6262 FIELD_SIZEOF(struct bpf_sock_ops_kern, op));
6263 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, reply) !=
6264 FIELD_SIZEOF(struct bpf_sock_ops_kern, reply));
6265 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, replylong) !=
6266 FIELD_SIZEOF(struct bpf_sock_ops_kern, replylong));
6268 off -= offsetof(struct bpf_sock_ops, op);
6269 off += offsetof(struct bpf_sock_ops_kern, op);
6270 if (type == BPF_WRITE)
6271 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6274 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6278 case offsetof(struct bpf_sock_ops, family):
6279 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
6281 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6282 struct bpf_sock_ops_kern, sk),
6283 si->dst_reg, si->src_reg,
6284 offsetof(struct bpf_sock_ops_kern, sk));
6285 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6286 offsetof(struct sock_common, skc_family));
6289 case offsetof(struct bpf_sock_ops, remote_ip4):
6290 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
6292 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6293 struct bpf_sock_ops_kern, sk),
6294 si->dst_reg, si->src_reg,
6295 offsetof(struct bpf_sock_ops_kern, sk));
6296 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6297 offsetof(struct sock_common, skc_daddr));
6300 case offsetof(struct bpf_sock_ops, local_ip4):
6301 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6302 skc_rcv_saddr) != 4);
6304 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6305 struct bpf_sock_ops_kern, sk),
6306 si->dst_reg, si->src_reg,
6307 offsetof(struct bpf_sock_ops_kern, sk));
6308 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6309 offsetof(struct sock_common,
6313 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
6314 offsetof(struct bpf_sock_ops, remote_ip6[3]):
6315 #if IS_ENABLED(CONFIG_IPV6)
6316 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6317 skc_v6_daddr.s6_addr32[0]) != 4);
6320 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
6321 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6322 struct bpf_sock_ops_kern, sk),
6323 si->dst_reg, si->src_reg,
6324 offsetof(struct bpf_sock_ops_kern, sk));
6325 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6326 offsetof(struct sock_common,
6327 skc_v6_daddr.s6_addr32[0]) +
6330 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6334 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
6335 offsetof(struct bpf_sock_ops, local_ip6[3]):
6336 #if IS_ENABLED(CONFIG_IPV6)
6337 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6338 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
6341 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
6342 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6343 struct bpf_sock_ops_kern, sk),
6344 si->dst_reg, si->src_reg,
6345 offsetof(struct bpf_sock_ops_kern, sk));
6346 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6347 offsetof(struct sock_common,
6348 skc_v6_rcv_saddr.s6_addr32[0]) +
6351 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6355 case offsetof(struct bpf_sock_ops, remote_port):
6356 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
6358 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6359 struct bpf_sock_ops_kern, sk),
6360 si->dst_reg, si->src_reg,
6361 offsetof(struct bpf_sock_ops_kern, sk));
6362 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6363 offsetof(struct sock_common, skc_dport));
6364 #ifndef __BIG_ENDIAN_BITFIELD
6365 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
6369 case offsetof(struct bpf_sock_ops, local_port):
6370 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
6372 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6373 struct bpf_sock_ops_kern, sk),
6374 si->dst_reg, si->src_reg,
6375 offsetof(struct bpf_sock_ops_kern, sk));
6376 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6377 offsetof(struct sock_common, skc_num));
6380 case offsetof(struct bpf_sock_ops, is_fullsock):
6381 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6382 struct bpf_sock_ops_kern,
6384 si->dst_reg, si->src_reg,
6385 offsetof(struct bpf_sock_ops_kern,
6389 case offsetof(struct bpf_sock_ops, state):
6390 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_state) != 1);
6392 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6393 struct bpf_sock_ops_kern, sk),
6394 si->dst_reg, si->src_reg,
6395 offsetof(struct bpf_sock_ops_kern, sk));
6396 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
6397 offsetof(struct sock_common, skc_state));
6400 case offsetof(struct bpf_sock_ops, rtt_min):
6401 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, rtt_min) !=
6402 sizeof(struct minmax));
6403 BUILD_BUG_ON(sizeof(struct minmax) <
6404 sizeof(struct minmax_sample));
6406 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6407 struct bpf_sock_ops_kern, sk),
6408 si->dst_reg, si->src_reg,
6409 offsetof(struct bpf_sock_ops_kern, sk));
6410 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6411 offsetof(struct tcp_sock, rtt_min) +
6412 FIELD_SIZEOF(struct minmax_sample, t));
6415 /* Helper macro for adding read access to tcp_sock or sock fields. */
6416 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
6418 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) > \
6419 FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD)); \
6420 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6421 struct bpf_sock_ops_kern, \
6423 si->dst_reg, si->src_reg, \
6424 offsetof(struct bpf_sock_ops_kern, \
6426 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 2); \
6427 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6428 struct bpf_sock_ops_kern, sk),\
6429 si->dst_reg, si->src_reg, \
6430 offsetof(struct bpf_sock_ops_kern, sk));\
6431 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
6433 si->dst_reg, si->dst_reg, \
6434 offsetof(OBJ, OBJ_FIELD)); \
6437 /* Helper macro for adding write access to tcp_sock or sock fields.
6438 * The macro is called with two registers, dst_reg which contains a pointer
6439 * to ctx (context) and src_reg which contains the value that should be
6440 * stored. However, we need an additional register since we cannot overwrite
6441 * dst_reg because it may be used later in the program.
6442 * Instead we "borrow" one of the other register. We first save its value
6443 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
6444 * it at the end of the macro.
6446 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
6448 int reg = BPF_REG_9; \
6449 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) > \
6450 FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD)); \
6451 if (si->dst_reg == reg || si->src_reg == reg) \
6453 if (si->dst_reg == reg || si->src_reg == reg) \
6455 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
6456 offsetof(struct bpf_sock_ops_kern, \
6458 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6459 struct bpf_sock_ops_kern, \
6462 offsetof(struct bpf_sock_ops_kern, \
6464 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
6465 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6466 struct bpf_sock_ops_kern, sk),\
6468 offsetof(struct bpf_sock_ops_kern, sk));\
6469 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \
6471 offsetof(OBJ, OBJ_FIELD)); \
6472 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
6473 offsetof(struct bpf_sock_ops_kern, \
6477 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
6479 if (TYPE == BPF_WRITE) \
6480 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
6482 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
6485 case offsetof(struct bpf_sock_ops, snd_cwnd):
6486 SOCK_OPS_GET_FIELD(snd_cwnd, snd_cwnd, struct tcp_sock);
6489 case offsetof(struct bpf_sock_ops, srtt_us):
6490 SOCK_OPS_GET_FIELD(srtt_us, srtt_us, struct tcp_sock);
6493 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
6494 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
6498 case offsetof(struct bpf_sock_ops, snd_ssthresh):
6499 SOCK_OPS_GET_FIELD(snd_ssthresh, snd_ssthresh, struct tcp_sock);
6502 case offsetof(struct bpf_sock_ops, rcv_nxt):
6503 SOCK_OPS_GET_FIELD(rcv_nxt, rcv_nxt, struct tcp_sock);
6506 case offsetof(struct bpf_sock_ops, snd_nxt):
6507 SOCK_OPS_GET_FIELD(snd_nxt, snd_nxt, struct tcp_sock);
6510 case offsetof(struct bpf_sock_ops, snd_una):
6511 SOCK_OPS_GET_FIELD(snd_una, snd_una, struct tcp_sock);
6514 case offsetof(struct bpf_sock_ops, mss_cache):
6515 SOCK_OPS_GET_FIELD(mss_cache, mss_cache, struct tcp_sock);
6518 case offsetof(struct bpf_sock_ops, ecn_flags):
6519 SOCK_OPS_GET_FIELD(ecn_flags, ecn_flags, struct tcp_sock);
6522 case offsetof(struct bpf_sock_ops, rate_delivered):
6523 SOCK_OPS_GET_FIELD(rate_delivered, rate_delivered,
6527 case offsetof(struct bpf_sock_ops, rate_interval_us):
6528 SOCK_OPS_GET_FIELD(rate_interval_us, rate_interval_us,
6532 case offsetof(struct bpf_sock_ops, packets_out):
6533 SOCK_OPS_GET_FIELD(packets_out, packets_out, struct tcp_sock);
6536 case offsetof(struct bpf_sock_ops, retrans_out):
6537 SOCK_OPS_GET_FIELD(retrans_out, retrans_out, struct tcp_sock);
6540 case offsetof(struct bpf_sock_ops, total_retrans):
6541 SOCK_OPS_GET_FIELD(total_retrans, total_retrans,
6545 case offsetof(struct bpf_sock_ops, segs_in):
6546 SOCK_OPS_GET_FIELD(segs_in, segs_in, struct tcp_sock);
6549 case offsetof(struct bpf_sock_ops, data_segs_in):
6550 SOCK_OPS_GET_FIELD(data_segs_in, data_segs_in, struct tcp_sock);
6553 case offsetof(struct bpf_sock_ops, segs_out):
6554 SOCK_OPS_GET_FIELD(segs_out, segs_out, struct tcp_sock);
6557 case offsetof(struct bpf_sock_ops, data_segs_out):
6558 SOCK_OPS_GET_FIELD(data_segs_out, data_segs_out,
6562 case offsetof(struct bpf_sock_ops, lost_out):
6563 SOCK_OPS_GET_FIELD(lost_out, lost_out, struct tcp_sock);
6566 case offsetof(struct bpf_sock_ops, sacked_out):
6567 SOCK_OPS_GET_FIELD(sacked_out, sacked_out, struct tcp_sock);
6570 case offsetof(struct bpf_sock_ops, sk_txhash):
6571 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
6575 case offsetof(struct bpf_sock_ops, bytes_received):
6576 SOCK_OPS_GET_FIELD(bytes_received, bytes_received,
6580 case offsetof(struct bpf_sock_ops, bytes_acked):
6581 SOCK_OPS_GET_FIELD(bytes_acked, bytes_acked, struct tcp_sock);
6585 return insn - insn_buf;
6588 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
6589 const struct bpf_insn *si,
6590 struct bpf_insn *insn_buf,
6591 struct bpf_prog *prog, u32 *target_size)
6593 struct bpf_insn *insn = insn_buf;
6597 case offsetof(struct __sk_buff, data_end):
6599 off -= offsetof(struct __sk_buff, data_end);
6600 off += offsetof(struct sk_buff, cb);
6601 off += offsetof(struct tcp_skb_cb, bpf.data_end);
6602 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
6606 return bpf_convert_ctx_access(type, si, insn_buf, prog,
6610 return insn - insn_buf;
6613 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
6614 const struct bpf_insn *si,
6615 struct bpf_insn *insn_buf,
6616 struct bpf_prog *prog, u32 *target_size)
6618 struct bpf_insn *insn = insn_buf;
6619 #if IS_ENABLED(CONFIG_IPV6)
6624 case offsetof(struct sk_msg_md, data):
6625 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_buff, data),
6626 si->dst_reg, si->src_reg,
6627 offsetof(struct sk_msg_buff, data));
6629 case offsetof(struct sk_msg_md, data_end):
6630 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_buff, data_end),
6631 si->dst_reg, si->src_reg,
6632 offsetof(struct sk_msg_buff, data_end));
6634 case offsetof(struct sk_msg_md, family):
6635 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
6637 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6638 struct sk_msg_buff, sk),
6639 si->dst_reg, si->src_reg,
6640 offsetof(struct sk_msg_buff, sk));
6641 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6642 offsetof(struct sock_common, skc_family));
6645 case offsetof(struct sk_msg_md, remote_ip4):
6646 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
6648 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6649 struct sk_msg_buff, sk),
6650 si->dst_reg, si->src_reg,
6651 offsetof(struct sk_msg_buff, sk));
6652 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6653 offsetof(struct sock_common, skc_daddr));
6656 case offsetof(struct sk_msg_md, local_ip4):
6657 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6658 skc_rcv_saddr) != 4);
6660 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6661 struct sk_msg_buff, sk),
6662 si->dst_reg, si->src_reg,
6663 offsetof(struct sk_msg_buff, sk));
6664 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6665 offsetof(struct sock_common,
6669 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
6670 offsetof(struct sk_msg_md, remote_ip6[3]):
6671 #if IS_ENABLED(CONFIG_IPV6)
6672 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6673 skc_v6_daddr.s6_addr32[0]) != 4);
6676 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
6677 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6678 struct sk_msg_buff, sk),
6679 si->dst_reg, si->src_reg,
6680 offsetof(struct sk_msg_buff, sk));
6681 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6682 offsetof(struct sock_common,
6683 skc_v6_daddr.s6_addr32[0]) +
6686 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6690 case offsetof(struct sk_msg_md, local_ip6[0]) ...
6691 offsetof(struct sk_msg_md, local_ip6[3]):
6692 #if IS_ENABLED(CONFIG_IPV6)
6693 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6694 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
6697 off -= offsetof(struct sk_msg_md, local_ip6[0]);
6698 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6699 struct sk_msg_buff, sk),
6700 si->dst_reg, si->src_reg,
6701 offsetof(struct sk_msg_buff, sk));
6702 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6703 offsetof(struct sock_common,
6704 skc_v6_rcv_saddr.s6_addr32[0]) +
6707 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6711 case offsetof(struct sk_msg_md, remote_port):
6712 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
6714 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6715 struct sk_msg_buff, sk),
6716 si->dst_reg, si->src_reg,
6717 offsetof(struct sk_msg_buff, sk));
6718 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6719 offsetof(struct sock_common, skc_dport));
6720 #ifndef __BIG_ENDIAN_BITFIELD
6721 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
6725 case offsetof(struct sk_msg_md, local_port):
6726 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
6728 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6729 struct sk_msg_buff, sk),
6730 si->dst_reg, si->src_reg,
6731 offsetof(struct sk_msg_buff, sk));
6732 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6733 offsetof(struct sock_common, skc_num));
6737 return insn - insn_buf;
6740 const struct bpf_verifier_ops sk_filter_verifier_ops = {
6741 .get_func_proto = sk_filter_func_proto,
6742 .is_valid_access = sk_filter_is_valid_access,
6743 .convert_ctx_access = bpf_convert_ctx_access,
6744 .gen_ld_abs = bpf_gen_ld_abs,
6747 const struct bpf_prog_ops sk_filter_prog_ops = {
6748 .test_run = bpf_prog_test_run_skb,
6751 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
6752 .get_func_proto = tc_cls_act_func_proto,
6753 .is_valid_access = tc_cls_act_is_valid_access,
6754 .convert_ctx_access = tc_cls_act_convert_ctx_access,
6755 .gen_prologue = tc_cls_act_prologue,
6756 .gen_ld_abs = bpf_gen_ld_abs,
6759 const struct bpf_prog_ops tc_cls_act_prog_ops = {
6760 .test_run = bpf_prog_test_run_skb,
6763 const struct bpf_verifier_ops xdp_verifier_ops = {
6764 .get_func_proto = xdp_func_proto,
6765 .is_valid_access = xdp_is_valid_access,
6766 .convert_ctx_access = xdp_convert_ctx_access,
6769 const struct bpf_prog_ops xdp_prog_ops = {
6770 .test_run = bpf_prog_test_run_xdp,
6773 const struct bpf_verifier_ops cg_skb_verifier_ops = {
6774 .get_func_proto = sk_filter_func_proto,
6775 .is_valid_access = sk_filter_is_valid_access,
6776 .convert_ctx_access = bpf_convert_ctx_access,
6779 const struct bpf_prog_ops cg_skb_prog_ops = {
6780 .test_run = bpf_prog_test_run_skb,
6783 const struct bpf_verifier_ops lwt_in_verifier_ops = {
6784 .get_func_proto = lwt_in_func_proto,
6785 .is_valid_access = lwt_is_valid_access,
6786 .convert_ctx_access = bpf_convert_ctx_access,
6789 const struct bpf_prog_ops lwt_in_prog_ops = {
6790 .test_run = bpf_prog_test_run_skb,
6793 const struct bpf_verifier_ops lwt_out_verifier_ops = {
6794 .get_func_proto = lwt_out_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_out_prog_ops = {
6800 .test_run = bpf_prog_test_run_skb,
6803 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
6804 .get_func_proto = lwt_xmit_func_proto,
6805 .is_valid_access = lwt_is_valid_access,
6806 .convert_ctx_access = bpf_convert_ctx_access,
6807 .gen_prologue = tc_cls_act_prologue,
6810 const struct bpf_prog_ops lwt_xmit_prog_ops = {
6811 .test_run = bpf_prog_test_run_skb,
6814 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
6815 .get_func_proto = lwt_seg6local_func_proto,
6816 .is_valid_access = lwt_is_valid_access,
6817 .convert_ctx_access = bpf_convert_ctx_access,
6820 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
6821 .test_run = bpf_prog_test_run_skb,
6824 const struct bpf_verifier_ops cg_sock_verifier_ops = {
6825 .get_func_proto = sock_filter_func_proto,
6826 .is_valid_access = sock_filter_is_valid_access,
6827 .convert_ctx_access = sock_filter_convert_ctx_access,
6830 const struct bpf_prog_ops cg_sock_prog_ops = {
6833 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
6834 .get_func_proto = sock_addr_func_proto,
6835 .is_valid_access = sock_addr_is_valid_access,
6836 .convert_ctx_access = sock_addr_convert_ctx_access,
6839 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
6842 const struct bpf_verifier_ops sock_ops_verifier_ops = {
6843 .get_func_proto = sock_ops_func_proto,
6844 .is_valid_access = sock_ops_is_valid_access,
6845 .convert_ctx_access = sock_ops_convert_ctx_access,
6848 const struct bpf_prog_ops sock_ops_prog_ops = {
6851 const struct bpf_verifier_ops sk_skb_verifier_ops = {
6852 .get_func_proto = sk_skb_func_proto,
6853 .is_valid_access = sk_skb_is_valid_access,
6854 .convert_ctx_access = sk_skb_convert_ctx_access,
6855 .gen_prologue = sk_skb_prologue,
6858 const struct bpf_prog_ops sk_skb_prog_ops = {
6861 const struct bpf_verifier_ops sk_msg_verifier_ops = {
6862 .get_func_proto = sk_msg_func_proto,
6863 .is_valid_access = sk_msg_is_valid_access,
6864 .convert_ctx_access = sk_msg_convert_ctx_access,
6867 const struct bpf_prog_ops sk_msg_prog_ops = {
6870 int sk_detach_filter(struct sock *sk)
6873 struct sk_filter *filter;
6875 if (sock_flag(sk, SOCK_FILTER_LOCKED))
6878 filter = rcu_dereference_protected(sk->sk_filter,
6879 lockdep_sock_is_held(sk));
6881 RCU_INIT_POINTER(sk->sk_filter, NULL);
6882 sk_filter_uncharge(sk, filter);
6888 EXPORT_SYMBOL_GPL(sk_detach_filter);
6890 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
6893 struct sock_fprog_kern *fprog;
6894 struct sk_filter *filter;
6898 filter = rcu_dereference_protected(sk->sk_filter,
6899 lockdep_sock_is_held(sk));
6903 /* We're copying the filter that has been originally attached,
6904 * so no conversion/decode needed anymore. eBPF programs that
6905 * have no original program cannot be dumped through this.
6908 fprog = filter->prog->orig_prog;
6914 /* User space only enquires number of filter blocks. */
6918 if (len < fprog->len)
6922 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
6925 /* Instead of bytes, the API requests to return the number