1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Linux Socket Filter - Kernel level socket filtering
5 * Based on the design of the Berkeley Packet Filter. The new
6 * internal format has been designed by PLUMgrid:
8 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
12 * Jay Schulist <jschlst@samba.org>
13 * Alexei Starovoitov <ast@plumgrid.com>
14 * Daniel Borkmann <dborkman@redhat.com>
16 * Andi Kleen - Fix a few bad bugs and races.
17 * Kris Katterjohn - Added many additional checks in bpf_check_classic()
20 #include <linux/atomic.h>
21 #include <linux/module.h>
22 #include <linux/types.h>
24 #include <linux/fcntl.h>
25 #include <linux/socket.h>
26 #include <linux/sock_diag.h>
28 #include <linux/inet.h>
29 #include <linux/netdevice.h>
30 #include <linux/if_packet.h>
31 #include <linux/if_arp.h>
32 #include <linux/gfp.h>
33 #include <net/inet_common.h>
35 #include <net/protocol.h>
36 #include <net/netlink.h>
37 #include <linux/skbuff.h>
38 #include <linux/skmsg.h>
40 #include <net/flow_dissector.h>
41 #include <linux/errno.h>
42 #include <linux/timer.h>
43 #include <linux/uaccess.h>
44 #include <asm/unaligned.h>
45 #include <linux/filter.h>
46 #include <linux/ratelimit.h>
47 #include <linux/seccomp.h>
48 #include <linux/if_vlan.h>
49 #include <linux/bpf.h>
50 #include <linux/btf.h>
51 #include <net/sch_generic.h>
52 #include <net/cls_cgroup.h>
53 #include <net/dst_metadata.h>
55 #include <net/sock_reuseport.h>
56 #include <net/busy_poll.h>
60 #include <linux/bpf_trace.h>
61 #include <net/xdp_sock.h>
62 #include <linux/inetdevice.h>
63 #include <net/inet_hashtables.h>
64 #include <net/inet6_hashtables.h>
65 #include <net/ip_fib.h>
66 #include <net/nexthop.h>
70 #include <net/net_namespace.h>
71 #include <linux/seg6_local.h>
73 #include <net/seg6_local.h>
74 #include <net/lwtunnel.h>
75 #include <net/ipv6_stubs.h>
76 #include <net/bpf_sk_storage.h>
77 #include <net/transp_v6.h>
78 #include <linux/btf_ids.h>
81 #include <net/mptcp.h>
83 static const struct bpf_func_proto *
84 bpf_sk_base_func_proto(enum bpf_func_id func_id);
86 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len)
88 if (in_compat_syscall()) {
89 struct compat_sock_fprog f32;
91 if (len != sizeof(f32))
93 if (copy_from_sockptr(&f32, src, sizeof(f32)))
95 memset(dst, 0, sizeof(*dst));
97 dst->filter = compat_ptr(f32.filter);
99 if (len != sizeof(*dst))
101 if (copy_from_sockptr(dst, src, sizeof(*dst)))
107 EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user);
110 * sk_filter_trim_cap - run a packet through a socket filter
111 * @sk: sock associated with &sk_buff
112 * @skb: buffer to filter
113 * @cap: limit on how short the eBPF program may trim the packet
115 * Run the eBPF program and then cut skb->data to correct size returned by
116 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
117 * than pkt_len we keep whole skb->data. This is the socket level
118 * wrapper to bpf_prog_run. It returns 0 if the packet should
119 * be accepted or -EPERM if the packet should be tossed.
122 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
125 struct sk_filter *filter;
128 * If the skb was allocated from pfmemalloc reserves, only
129 * allow SOCK_MEMALLOC sockets to use it as this socket is
130 * helping free memory
132 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
133 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
136 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
140 err = security_sock_rcv_skb(sk, skb);
145 filter = rcu_dereference(sk->sk_filter);
147 struct sock *save_sk = skb->sk;
148 unsigned int pkt_len;
151 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
153 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
159 EXPORT_SYMBOL(sk_filter_trim_cap);
161 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
163 return skb_get_poff(skb);
166 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
170 if (skb_is_nonlinear(skb))
173 if (skb->len < sizeof(struct nlattr))
176 if (a > skb->len - sizeof(struct nlattr))
179 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
181 return (void *) nla - (void *) skb->data;
186 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
190 if (skb_is_nonlinear(skb))
193 if (skb->len < sizeof(struct nlattr))
196 if (a > skb->len - sizeof(struct nlattr))
199 nla = (struct nlattr *) &skb->data[a];
200 if (nla->nla_len > skb->len - a)
203 nla = nla_find_nested(nla, x);
205 return (void *) nla - (void *) skb->data;
210 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
211 data, int, headlen, int, offset)
214 const int len = sizeof(tmp);
217 if (headlen - offset >= len)
218 return *(u8 *)(data + offset);
219 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
222 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
230 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
233 return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
237 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
238 data, int, headlen, int, offset)
241 const int len = sizeof(tmp);
244 if (headlen - offset >= len)
245 return get_unaligned_be16(data + offset);
246 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
247 return be16_to_cpu(tmp);
249 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
251 return get_unaligned_be16(ptr);
257 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
260 return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
264 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
265 data, int, headlen, int, offset)
268 const int len = sizeof(tmp);
270 if (likely(offset >= 0)) {
271 if (headlen - offset >= len)
272 return get_unaligned_be32(data + offset);
273 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
274 return be32_to_cpu(tmp);
276 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
278 return get_unaligned_be32(ptr);
284 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
287 return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
291 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
292 struct bpf_insn *insn_buf)
294 struct bpf_insn *insn = insn_buf;
298 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4);
300 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
301 offsetof(struct sk_buff, mark));
305 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET);
306 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
307 #ifdef __BIG_ENDIAN_BITFIELD
308 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
313 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2);
315 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
316 offsetof(struct sk_buff, queue_mapping));
319 case SKF_AD_VLAN_TAG:
320 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2);
322 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
323 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
324 offsetof(struct sk_buff, vlan_tci));
326 case SKF_AD_VLAN_TAG_PRESENT:
327 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_VLAN_PRESENT_OFFSET);
328 if (PKT_VLAN_PRESENT_BIT)
329 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, PKT_VLAN_PRESENT_BIT);
330 if (PKT_VLAN_PRESENT_BIT < 7)
331 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
335 return insn - insn_buf;
338 static bool convert_bpf_extensions(struct sock_filter *fp,
339 struct bpf_insn **insnp)
341 struct bpf_insn *insn = *insnp;
345 case SKF_AD_OFF + SKF_AD_PROTOCOL:
346 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2);
348 /* A = *(u16 *) (CTX + offsetof(protocol)) */
349 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
350 offsetof(struct sk_buff, protocol));
351 /* A = ntohs(A) [emitting a nop or swap16] */
352 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
355 case SKF_AD_OFF + SKF_AD_PKTTYPE:
356 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
360 case SKF_AD_OFF + SKF_AD_IFINDEX:
361 case SKF_AD_OFF + SKF_AD_HATYPE:
362 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4);
363 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2);
365 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
366 BPF_REG_TMP, BPF_REG_CTX,
367 offsetof(struct sk_buff, dev));
368 /* if (tmp != 0) goto pc + 1 */
369 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
370 *insn++ = BPF_EXIT_INSN();
371 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
372 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
373 offsetof(struct net_device, ifindex));
375 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
376 offsetof(struct net_device, type));
379 case SKF_AD_OFF + SKF_AD_MARK:
380 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
384 case SKF_AD_OFF + SKF_AD_RXHASH:
385 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4);
387 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
388 offsetof(struct sk_buff, hash));
391 case SKF_AD_OFF + SKF_AD_QUEUE:
392 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
396 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
397 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
398 BPF_REG_A, BPF_REG_CTX, insn);
402 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
403 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
404 BPF_REG_A, BPF_REG_CTX, insn);
408 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
409 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2);
411 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
412 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
413 offsetof(struct sk_buff, vlan_proto));
414 /* A = ntohs(A) [emitting a nop or swap16] */
415 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
418 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
419 case SKF_AD_OFF + SKF_AD_NLATTR:
420 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
421 case SKF_AD_OFF + SKF_AD_CPU:
422 case SKF_AD_OFF + SKF_AD_RANDOM:
424 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
426 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
428 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
429 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
431 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
432 *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
434 case SKF_AD_OFF + SKF_AD_NLATTR:
435 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
437 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
438 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
440 case SKF_AD_OFF + SKF_AD_CPU:
441 *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
443 case SKF_AD_OFF + SKF_AD_RANDOM:
444 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
445 bpf_user_rnd_init_once();
450 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
452 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
456 /* This is just a dummy call to avoid letting the compiler
457 * evict __bpf_call_base() as an optimization. Placed here
458 * where no-one bothers.
460 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
468 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
470 const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
471 int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
472 bool endian = BPF_SIZE(fp->code) == BPF_H ||
473 BPF_SIZE(fp->code) == BPF_W;
474 bool indirect = BPF_MODE(fp->code) == BPF_IND;
475 const int ip_align = NET_IP_ALIGN;
476 struct bpf_insn *insn = *insnp;
480 ((unaligned_ok && offset >= 0) ||
481 (!unaligned_ok && offset >= 0 &&
482 offset + ip_align >= 0 &&
483 offset + ip_align % size == 0))) {
484 bool ldx_off_ok = offset <= S16_MAX;
486 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
488 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
489 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
490 size, 2 + endian + (!ldx_off_ok * 2));
492 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
495 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
496 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
497 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
501 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
502 *insn++ = BPF_JMP_A(8);
505 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
506 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
507 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
509 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
511 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
513 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
516 switch (BPF_SIZE(fp->code)) {
518 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
521 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
524 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
530 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
531 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
532 *insn = BPF_EXIT_INSN();
539 * bpf_convert_filter - convert filter program
540 * @prog: the user passed filter program
541 * @len: the length of the user passed filter program
542 * @new_prog: allocated 'struct bpf_prog' or NULL
543 * @new_len: pointer to store length of converted program
544 * @seen_ld_abs: bool whether we've seen ld_abs/ind
546 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
547 * style extended BPF (eBPF).
548 * Conversion workflow:
550 * 1) First pass for calculating the new program length:
551 * bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
553 * 2) 2nd pass to remap in two passes: 1st pass finds new
554 * jump offsets, 2nd pass remapping:
555 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
557 static int bpf_convert_filter(struct sock_filter *prog, int len,
558 struct bpf_prog *new_prog, int *new_len,
561 int new_flen = 0, pass = 0, target, i, stack_off;
562 struct bpf_insn *new_insn, *first_insn = NULL;
563 struct sock_filter *fp;
567 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
568 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
570 if (len <= 0 || len > BPF_MAXINSNS)
574 first_insn = new_prog->insnsi;
575 addrs = kcalloc(len, sizeof(*addrs),
576 GFP_KERNEL | __GFP_NOWARN);
582 new_insn = first_insn;
585 /* Classic BPF related prologue emission. */
587 /* Classic BPF expects A and X to be reset first. These need
588 * to be guaranteed to be the first two instructions.
590 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
591 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
593 /* All programs must keep CTX in callee saved BPF_REG_CTX.
594 * In eBPF case it's done by the compiler, here we need to
595 * do this ourself. Initial CTX is present in BPF_REG_ARG1.
597 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
599 /* For packet access in classic BPF, cache skb->data
600 * in callee-saved BPF R8 and skb->len - skb->data_len
601 * (headlen) in BPF R9. Since classic BPF is read-only
602 * on CTX, we only need to cache it once.
604 *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
605 BPF_REG_D, BPF_REG_CTX,
606 offsetof(struct sk_buff, data));
607 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
608 offsetof(struct sk_buff, len));
609 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
610 offsetof(struct sk_buff, data_len));
611 *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
617 for (i = 0; i < len; fp++, i++) {
618 struct bpf_insn tmp_insns[32] = { };
619 struct bpf_insn *insn = tmp_insns;
622 addrs[i] = new_insn - first_insn;
625 /* All arithmetic insns and skb loads map as-is. */
626 case BPF_ALU | BPF_ADD | BPF_X:
627 case BPF_ALU | BPF_ADD | BPF_K:
628 case BPF_ALU | BPF_SUB | BPF_X:
629 case BPF_ALU | BPF_SUB | BPF_K:
630 case BPF_ALU | BPF_AND | BPF_X:
631 case BPF_ALU | BPF_AND | BPF_K:
632 case BPF_ALU | BPF_OR | BPF_X:
633 case BPF_ALU | BPF_OR | BPF_K:
634 case BPF_ALU | BPF_LSH | BPF_X:
635 case BPF_ALU | BPF_LSH | BPF_K:
636 case BPF_ALU | BPF_RSH | BPF_X:
637 case BPF_ALU | BPF_RSH | BPF_K:
638 case BPF_ALU | BPF_XOR | BPF_X:
639 case BPF_ALU | BPF_XOR | BPF_K:
640 case BPF_ALU | BPF_MUL | BPF_X:
641 case BPF_ALU | BPF_MUL | BPF_K:
642 case BPF_ALU | BPF_DIV | BPF_X:
643 case BPF_ALU | BPF_DIV | BPF_K:
644 case BPF_ALU | BPF_MOD | BPF_X:
645 case BPF_ALU | BPF_MOD | BPF_K:
646 case BPF_ALU | BPF_NEG:
647 case BPF_LD | BPF_ABS | BPF_W:
648 case BPF_LD | BPF_ABS | BPF_H:
649 case BPF_LD | BPF_ABS | BPF_B:
650 case BPF_LD | BPF_IND | BPF_W:
651 case BPF_LD | BPF_IND | BPF_H:
652 case BPF_LD | BPF_IND | BPF_B:
653 /* Check for overloaded BPF extension and
654 * directly convert it if found, otherwise
655 * just move on with mapping.
657 if (BPF_CLASS(fp->code) == BPF_LD &&
658 BPF_MODE(fp->code) == BPF_ABS &&
659 convert_bpf_extensions(fp, &insn))
661 if (BPF_CLASS(fp->code) == BPF_LD &&
662 convert_bpf_ld_abs(fp, &insn)) {
667 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
668 fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
669 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
670 /* Error with exception code on div/mod by 0.
671 * For cBPF programs, this was always return 0.
673 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
674 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
675 *insn++ = BPF_EXIT_INSN();
678 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
681 /* Jump transformation cannot use BPF block macros
682 * everywhere as offset calculation and target updates
683 * require a bit more work than the rest, i.e. jump
684 * opcodes map as-is, but offsets need adjustment.
687 #define BPF_EMIT_JMP \
689 const s32 off_min = S16_MIN, off_max = S16_MAX; \
692 if (target >= len || target < 0) \
694 off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
695 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
696 off -= insn - tmp_insns; \
697 /* Reject anything not fitting into insn->off. */ \
698 if (off < off_min || off > off_max) \
703 case BPF_JMP | BPF_JA:
704 target = i + fp->k + 1;
705 insn->code = fp->code;
709 case BPF_JMP | BPF_JEQ | BPF_K:
710 case BPF_JMP | BPF_JEQ | BPF_X:
711 case BPF_JMP | BPF_JSET | BPF_K:
712 case BPF_JMP | BPF_JSET | BPF_X:
713 case BPF_JMP | BPF_JGT | BPF_K:
714 case BPF_JMP | BPF_JGT | BPF_X:
715 case BPF_JMP | BPF_JGE | BPF_K:
716 case BPF_JMP | BPF_JGE | BPF_X:
717 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
718 /* BPF immediates are signed, zero extend
719 * immediate into tmp register and use it
722 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
724 insn->dst_reg = BPF_REG_A;
725 insn->src_reg = BPF_REG_TMP;
728 insn->dst_reg = BPF_REG_A;
730 bpf_src = BPF_SRC(fp->code);
731 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
734 /* Common case where 'jump_false' is next insn. */
736 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
737 target = i + fp->jt + 1;
742 /* Convert some jumps when 'jump_true' is next insn. */
744 switch (BPF_OP(fp->code)) {
746 insn->code = BPF_JMP | BPF_JNE | bpf_src;
749 insn->code = BPF_JMP | BPF_JLE | bpf_src;
752 insn->code = BPF_JMP | BPF_JLT | bpf_src;
758 target = i + fp->jf + 1;
763 /* Other jumps are mapped into two insns: Jxx and JA. */
764 target = i + fp->jt + 1;
765 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
769 insn->code = BPF_JMP | BPF_JA;
770 target = i + fp->jf + 1;
774 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
775 case BPF_LDX | BPF_MSH | BPF_B: {
776 struct sock_filter tmp = {
777 .code = BPF_LD | BPF_ABS | BPF_B,
784 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
785 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
786 convert_bpf_ld_abs(&tmp, &insn);
789 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
791 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
793 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
795 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
797 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
800 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
801 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
803 case BPF_RET | BPF_A:
804 case BPF_RET | BPF_K:
805 if (BPF_RVAL(fp->code) == BPF_K)
806 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
808 *insn = BPF_EXIT_INSN();
811 /* Store to stack. */
814 stack_off = fp->k * 4 + 4;
815 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
816 BPF_ST ? BPF_REG_A : BPF_REG_X,
818 /* check_load_and_stores() verifies that classic BPF can
819 * load from stack only after write, so tracking
820 * stack_depth for ST|STX insns is enough
822 if (new_prog && new_prog->aux->stack_depth < stack_off)
823 new_prog->aux->stack_depth = stack_off;
826 /* Load from stack. */
827 case BPF_LD | BPF_MEM:
828 case BPF_LDX | BPF_MEM:
829 stack_off = fp->k * 4 + 4;
830 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
831 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
836 case BPF_LD | BPF_IMM:
837 case BPF_LDX | BPF_IMM:
838 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
839 BPF_REG_A : BPF_REG_X, fp->k);
843 case BPF_MISC | BPF_TAX:
844 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
848 case BPF_MISC | BPF_TXA:
849 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
852 /* A = skb->len or X = skb->len */
853 case BPF_LD | BPF_W | BPF_LEN:
854 case BPF_LDX | BPF_W | BPF_LEN:
855 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
856 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
857 offsetof(struct sk_buff, len));
860 /* Access seccomp_data fields. */
861 case BPF_LDX | BPF_ABS | BPF_W:
862 /* A = *(u32 *) (ctx + K) */
863 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
866 /* Unknown instruction. */
873 memcpy(new_insn, tmp_insns,
874 sizeof(*insn) * (insn - tmp_insns));
875 new_insn += insn - tmp_insns;
879 /* Only calculating new length. */
880 *new_len = new_insn - first_insn;
882 *new_len += 4; /* Prologue bits. */
887 if (new_flen != new_insn - first_insn) {
888 new_flen = new_insn - first_insn;
895 BUG_ON(*new_len != new_flen);
904 * As we dont want to clear mem[] array for each packet going through
905 * __bpf_prog_run(), we check that filter loaded by user never try to read
906 * a cell if not previously written, and we check all branches to be sure
907 * a malicious user doesn't try to abuse us.
909 static int check_load_and_stores(const struct sock_filter *filter, int flen)
911 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
914 BUILD_BUG_ON(BPF_MEMWORDS > 16);
916 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
920 memset(masks, 0xff, flen * sizeof(*masks));
922 for (pc = 0; pc < flen; pc++) {
923 memvalid &= masks[pc];
925 switch (filter[pc].code) {
928 memvalid |= (1 << filter[pc].k);
930 case BPF_LD | BPF_MEM:
931 case BPF_LDX | BPF_MEM:
932 if (!(memvalid & (1 << filter[pc].k))) {
937 case BPF_JMP | BPF_JA:
938 /* A jump must set masks on target */
939 masks[pc + 1 + filter[pc].k] &= memvalid;
942 case BPF_JMP | BPF_JEQ | BPF_K:
943 case BPF_JMP | BPF_JEQ | BPF_X:
944 case BPF_JMP | BPF_JGE | BPF_K:
945 case BPF_JMP | BPF_JGE | BPF_X:
946 case BPF_JMP | BPF_JGT | BPF_K:
947 case BPF_JMP | BPF_JGT | BPF_X:
948 case BPF_JMP | BPF_JSET | BPF_K:
949 case BPF_JMP | BPF_JSET | BPF_X:
950 /* A jump must set masks on targets */
951 masks[pc + 1 + filter[pc].jt] &= memvalid;
952 masks[pc + 1 + filter[pc].jf] &= memvalid;
962 static bool chk_code_allowed(u16 code_to_probe)
964 static const bool codes[] = {
965 /* 32 bit ALU operations */
966 [BPF_ALU | BPF_ADD | BPF_K] = true,
967 [BPF_ALU | BPF_ADD | BPF_X] = true,
968 [BPF_ALU | BPF_SUB | BPF_K] = true,
969 [BPF_ALU | BPF_SUB | BPF_X] = true,
970 [BPF_ALU | BPF_MUL | BPF_K] = true,
971 [BPF_ALU | BPF_MUL | BPF_X] = true,
972 [BPF_ALU | BPF_DIV | BPF_K] = true,
973 [BPF_ALU | BPF_DIV | BPF_X] = true,
974 [BPF_ALU | BPF_MOD | BPF_K] = true,
975 [BPF_ALU | BPF_MOD | BPF_X] = true,
976 [BPF_ALU | BPF_AND | BPF_K] = true,
977 [BPF_ALU | BPF_AND | BPF_X] = true,
978 [BPF_ALU | BPF_OR | BPF_K] = true,
979 [BPF_ALU | BPF_OR | BPF_X] = true,
980 [BPF_ALU | BPF_XOR | BPF_K] = true,
981 [BPF_ALU | BPF_XOR | BPF_X] = true,
982 [BPF_ALU | BPF_LSH | BPF_K] = true,
983 [BPF_ALU | BPF_LSH | BPF_X] = true,
984 [BPF_ALU | BPF_RSH | BPF_K] = true,
985 [BPF_ALU | BPF_RSH | BPF_X] = true,
986 [BPF_ALU | BPF_NEG] = true,
987 /* Load instructions */
988 [BPF_LD | BPF_W | BPF_ABS] = true,
989 [BPF_LD | BPF_H | BPF_ABS] = true,
990 [BPF_LD | BPF_B | BPF_ABS] = true,
991 [BPF_LD | BPF_W | BPF_LEN] = true,
992 [BPF_LD | BPF_W | BPF_IND] = true,
993 [BPF_LD | BPF_H | BPF_IND] = true,
994 [BPF_LD | BPF_B | BPF_IND] = true,
995 [BPF_LD | BPF_IMM] = true,
996 [BPF_LD | BPF_MEM] = true,
997 [BPF_LDX | BPF_W | BPF_LEN] = true,
998 [BPF_LDX | BPF_B | BPF_MSH] = true,
999 [BPF_LDX | BPF_IMM] = true,
1000 [BPF_LDX | BPF_MEM] = true,
1001 /* Store instructions */
1004 /* Misc instructions */
1005 [BPF_MISC | BPF_TAX] = true,
1006 [BPF_MISC | BPF_TXA] = true,
1007 /* Return instructions */
1008 [BPF_RET | BPF_K] = true,
1009 [BPF_RET | BPF_A] = true,
1010 /* Jump instructions */
1011 [BPF_JMP | BPF_JA] = true,
1012 [BPF_JMP | BPF_JEQ | BPF_K] = true,
1013 [BPF_JMP | BPF_JEQ | BPF_X] = true,
1014 [BPF_JMP | BPF_JGE | BPF_K] = true,
1015 [BPF_JMP | BPF_JGE | BPF_X] = true,
1016 [BPF_JMP | BPF_JGT | BPF_K] = true,
1017 [BPF_JMP | BPF_JGT | BPF_X] = true,
1018 [BPF_JMP | BPF_JSET | BPF_K] = true,
1019 [BPF_JMP | BPF_JSET | BPF_X] = true,
1022 if (code_to_probe >= ARRAY_SIZE(codes))
1025 return codes[code_to_probe];
1028 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1033 if (flen == 0 || flen > BPF_MAXINSNS)
1040 * bpf_check_classic - verify socket filter code
1041 * @filter: filter to verify
1042 * @flen: length of filter
1044 * Check the user's filter code. If we let some ugly
1045 * filter code slip through kaboom! The filter must contain
1046 * no references or jumps that are out of range, no illegal
1047 * instructions, and must end with a RET instruction.
1049 * All jumps are forward as they are not signed.
1051 * Returns 0 if the rule set is legal or -EINVAL if not.
1053 static int bpf_check_classic(const struct sock_filter *filter,
1059 /* Check the filter code now */
1060 for (pc = 0; pc < flen; pc++) {
1061 const struct sock_filter *ftest = &filter[pc];
1063 /* May we actually operate on this code? */
1064 if (!chk_code_allowed(ftest->code))
1067 /* Some instructions need special checks */
1068 switch (ftest->code) {
1069 case BPF_ALU | BPF_DIV | BPF_K:
1070 case BPF_ALU | BPF_MOD | BPF_K:
1071 /* Check for division by zero */
1075 case BPF_ALU | BPF_LSH | BPF_K:
1076 case BPF_ALU | BPF_RSH | BPF_K:
1080 case BPF_LD | BPF_MEM:
1081 case BPF_LDX | BPF_MEM:
1084 /* Check for invalid memory addresses */
1085 if (ftest->k >= BPF_MEMWORDS)
1088 case BPF_JMP | BPF_JA:
1089 /* Note, the large ftest->k might cause loops.
1090 * Compare this with conditional jumps below,
1091 * where offsets are limited. --ANK (981016)
1093 if (ftest->k >= (unsigned int)(flen - pc - 1))
1096 case BPF_JMP | BPF_JEQ | BPF_K:
1097 case BPF_JMP | BPF_JEQ | BPF_X:
1098 case BPF_JMP | BPF_JGE | BPF_K:
1099 case BPF_JMP | BPF_JGE | BPF_X:
1100 case BPF_JMP | BPF_JGT | BPF_K:
1101 case BPF_JMP | BPF_JGT | BPF_X:
1102 case BPF_JMP | BPF_JSET | BPF_K:
1103 case BPF_JMP | BPF_JSET | BPF_X:
1104 /* Both conditionals must be safe */
1105 if (pc + ftest->jt + 1 >= flen ||
1106 pc + ftest->jf + 1 >= flen)
1109 case BPF_LD | BPF_W | BPF_ABS:
1110 case BPF_LD | BPF_H | BPF_ABS:
1111 case BPF_LD | BPF_B | BPF_ABS:
1113 if (bpf_anc_helper(ftest) & BPF_ANC)
1115 /* Ancillary operation unknown or unsupported */
1116 if (anc_found == false && ftest->k >= SKF_AD_OFF)
1121 /* Last instruction must be a RET code */
1122 switch (filter[flen - 1].code) {
1123 case BPF_RET | BPF_K:
1124 case BPF_RET | BPF_A:
1125 return check_load_and_stores(filter, flen);
1131 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1132 const struct sock_fprog *fprog)
1134 unsigned int fsize = bpf_classic_proglen(fprog);
1135 struct sock_fprog_kern *fkprog;
1137 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1141 fkprog = fp->orig_prog;
1142 fkprog->len = fprog->len;
1144 fkprog->filter = kmemdup(fp->insns, fsize,
1145 GFP_KERNEL | __GFP_NOWARN);
1146 if (!fkprog->filter) {
1147 kfree(fp->orig_prog);
1154 static void bpf_release_orig_filter(struct bpf_prog *fp)
1156 struct sock_fprog_kern *fprog = fp->orig_prog;
1159 kfree(fprog->filter);
1164 static void __bpf_prog_release(struct bpf_prog *prog)
1166 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1169 bpf_release_orig_filter(prog);
1170 bpf_prog_free(prog);
1174 static void __sk_filter_release(struct sk_filter *fp)
1176 __bpf_prog_release(fp->prog);
1181 * sk_filter_release_rcu - Release a socket filter by rcu_head
1182 * @rcu: rcu_head that contains the sk_filter to free
1184 static void sk_filter_release_rcu(struct rcu_head *rcu)
1186 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1188 __sk_filter_release(fp);
1192 * sk_filter_release - release a socket filter
1193 * @fp: filter to remove
1195 * Remove a filter from a socket and release its resources.
1197 static void sk_filter_release(struct sk_filter *fp)
1199 if (refcount_dec_and_test(&fp->refcnt))
1200 call_rcu(&fp->rcu, sk_filter_release_rcu);
1203 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1205 u32 filter_size = bpf_prog_size(fp->prog->len);
1207 atomic_sub(filter_size, &sk->sk_omem_alloc);
1208 sk_filter_release(fp);
1211 /* try to charge the socket memory if there is space available
1212 * return true on success
1214 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1216 u32 filter_size = bpf_prog_size(fp->prog->len);
1218 /* same check as in sock_kmalloc() */
1219 if (filter_size <= sysctl_optmem_max &&
1220 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
1221 atomic_add(filter_size, &sk->sk_omem_alloc);
1227 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1229 if (!refcount_inc_not_zero(&fp->refcnt))
1232 if (!__sk_filter_charge(sk, fp)) {
1233 sk_filter_release(fp);
1239 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1241 struct sock_filter *old_prog;
1242 struct bpf_prog *old_fp;
1243 int err, new_len, old_len = fp->len;
1244 bool seen_ld_abs = false;
1246 /* We are free to overwrite insns et al right here as it won't be used at
1247 * this point in time anymore internally after the migration to the eBPF
1248 * instruction representation.
1250 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1251 sizeof(struct bpf_insn));
1253 /* Conversion cannot happen on overlapping memory areas,
1254 * so we need to keep the user BPF around until the 2nd
1255 * pass. At this time, the user BPF is stored in fp->insns.
1257 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1258 GFP_KERNEL | __GFP_NOWARN);
1264 /* 1st pass: calculate the new program length. */
1265 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1270 /* Expand fp for appending the new filter representation. */
1272 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1274 /* The old_fp is still around in case we couldn't
1275 * allocate new memory, so uncharge on that one.
1284 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1285 err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1288 /* 2nd bpf_convert_filter() can fail only if it fails
1289 * to allocate memory, remapping must succeed. Note,
1290 * that at this time old_fp has already been released
1295 fp = bpf_prog_select_runtime(fp, &err);
1305 __bpf_prog_release(fp);
1306 return ERR_PTR(err);
1309 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1310 bpf_aux_classic_check_t trans)
1314 fp->bpf_func = NULL;
1317 err = bpf_check_classic(fp->insns, fp->len);
1319 __bpf_prog_release(fp);
1320 return ERR_PTR(err);
1323 /* There might be additional checks and transformations
1324 * needed on classic filters, f.e. in case of seccomp.
1327 err = trans(fp->insns, fp->len);
1329 __bpf_prog_release(fp);
1330 return ERR_PTR(err);
1334 /* Probe if we can JIT compile the filter and if so, do
1335 * the compilation of the filter.
1337 bpf_jit_compile(fp);
1339 /* JIT compiler couldn't process this filter, so do the eBPF translation
1340 * for the optimized interpreter.
1343 fp = bpf_migrate_filter(fp);
1349 * bpf_prog_create - create an unattached filter
1350 * @pfp: the unattached filter that is created
1351 * @fprog: the filter program
1353 * Create a filter independent of any socket. We first run some
1354 * sanity checks on it to make sure it does not explode on us later.
1355 * If an error occurs or there is insufficient memory for the filter
1356 * a negative errno code is returned. On success the return is zero.
1358 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1360 unsigned int fsize = bpf_classic_proglen(fprog);
1361 struct bpf_prog *fp;
1363 /* Make sure new filter is there and in the right amounts. */
1364 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1367 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1371 memcpy(fp->insns, fprog->filter, fsize);
1373 fp->len = fprog->len;
1374 /* Since unattached filters are not copied back to user
1375 * space through sk_get_filter(), we do not need to hold
1376 * a copy here, and can spare us the work.
1378 fp->orig_prog = NULL;
1380 /* bpf_prepare_filter() already takes care of freeing
1381 * memory in case something goes wrong.
1383 fp = bpf_prepare_filter(fp, NULL);
1390 EXPORT_SYMBOL_GPL(bpf_prog_create);
1393 * bpf_prog_create_from_user - create an unattached filter from user buffer
1394 * @pfp: the unattached filter that is created
1395 * @fprog: the filter program
1396 * @trans: post-classic verifier transformation handler
1397 * @save_orig: save classic BPF program
1399 * This function effectively does the same as bpf_prog_create(), only
1400 * that it builds up its insns buffer from user space provided buffer.
1401 * It also allows for passing a bpf_aux_classic_check_t handler.
1403 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1404 bpf_aux_classic_check_t trans, bool save_orig)
1406 unsigned int fsize = bpf_classic_proglen(fprog);
1407 struct bpf_prog *fp;
1410 /* Make sure new filter is there and in the right amounts. */
1411 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1414 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1418 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1419 __bpf_prog_free(fp);
1423 fp->len = fprog->len;
1424 fp->orig_prog = NULL;
1427 err = bpf_prog_store_orig_filter(fp, fprog);
1429 __bpf_prog_free(fp);
1434 /* bpf_prepare_filter() already takes care of freeing
1435 * memory in case something goes wrong.
1437 fp = bpf_prepare_filter(fp, trans);
1444 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1446 void bpf_prog_destroy(struct bpf_prog *fp)
1448 __bpf_prog_release(fp);
1450 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1452 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1454 struct sk_filter *fp, *old_fp;
1456 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1462 if (!__sk_filter_charge(sk, fp)) {
1466 refcount_set(&fp->refcnt, 1);
1468 old_fp = rcu_dereference_protected(sk->sk_filter,
1469 lockdep_sock_is_held(sk));
1470 rcu_assign_pointer(sk->sk_filter, fp);
1473 sk_filter_uncharge(sk, old_fp);
1479 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1481 unsigned int fsize = bpf_classic_proglen(fprog);
1482 struct bpf_prog *prog;
1485 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1486 return ERR_PTR(-EPERM);
1488 /* Make sure new filter is there and in the right amounts. */
1489 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1490 return ERR_PTR(-EINVAL);
1492 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1494 return ERR_PTR(-ENOMEM);
1496 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1497 __bpf_prog_free(prog);
1498 return ERR_PTR(-EFAULT);
1501 prog->len = fprog->len;
1503 err = bpf_prog_store_orig_filter(prog, fprog);
1505 __bpf_prog_free(prog);
1506 return ERR_PTR(-ENOMEM);
1509 /* bpf_prepare_filter() already takes care of freeing
1510 * memory in case something goes wrong.
1512 return bpf_prepare_filter(prog, NULL);
1516 * sk_attach_filter - attach a socket filter
1517 * @fprog: the filter program
1518 * @sk: the socket to use
1520 * Attach the user's filter code. We first run some sanity checks on
1521 * it to make sure it does not explode on us later. If an error
1522 * occurs or there is insufficient memory for the filter a negative
1523 * errno code is returned. On success the return is zero.
1525 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1527 struct bpf_prog *prog = __get_filter(fprog, sk);
1531 return PTR_ERR(prog);
1533 err = __sk_attach_prog(prog, sk);
1535 __bpf_prog_release(prog);
1541 EXPORT_SYMBOL_GPL(sk_attach_filter);
1543 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1545 struct bpf_prog *prog = __get_filter(fprog, sk);
1549 return PTR_ERR(prog);
1551 if (bpf_prog_size(prog->len) > sysctl_optmem_max)
1554 err = reuseport_attach_prog(sk, prog);
1557 __bpf_prog_release(prog);
1562 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1564 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1565 return ERR_PTR(-EPERM);
1567 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1570 int sk_attach_bpf(u32 ufd, struct sock *sk)
1572 struct bpf_prog *prog = __get_bpf(ufd, sk);
1576 return PTR_ERR(prog);
1578 err = __sk_attach_prog(prog, sk);
1587 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1589 struct bpf_prog *prog;
1592 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1595 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1596 if (PTR_ERR(prog) == -EINVAL)
1597 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1599 return PTR_ERR(prog);
1601 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1602 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1603 * bpf prog (e.g. sockmap). It depends on the
1604 * limitation imposed by bpf_prog_load().
1605 * Hence, sysctl_optmem_max is not checked.
1607 if ((sk->sk_type != SOCK_STREAM &&
1608 sk->sk_type != SOCK_DGRAM) ||
1609 (sk->sk_protocol != IPPROTO_UDP &&
1610 sk->sk_protocol != IPPROTO_TCP) ||
1611 (sk->sk_family != AF_INET &&
1612 sk->sk_family != AF_INET6)) {
1617 /* BPF_PROG_TYPE_SOCKET_FILTER */
1618 if (bpf_prog_size(prog->len) > sysctl_optmem_max) {
1624 err = reuseport_attach_prog(sk, prog);
1632 void sk_reuseport_prog_free(struct bpf_prog *prog)
1637 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1640 bpf_prog_destroy(prog);
1643 struct bpf_scratchpad {
1645 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1646 u8 buff[MAX_BPF_STACK];
1650 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1652 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1653 unsigned int write_len)
1655 return skb_ensure_writable(skb, write_len);
1658 static inline int bpf_try_make_writable(struct sk_buff *skb,
1659 unsigned int write_len)
1661 int err = __bpf_try_make_writable(skb, write_len);
1663 bpf_compute_data_pointers(skb);
1667 static int bpf_try_make_head_writable(struct sk_buff *skb)
1669 return bpf_try_make_writable(skb, skb_headlen(skb));
1672 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1674 if (skb_at_tc_ingress(skb))
1675 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1678 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1680 if (skb_at_tc_ingress(skb))
1681 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1684 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1685 const void *, from, u32, len, u64, flags)
1689 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1691 if (unlikely(offset > INT_MAX))
1693 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1696 ptr = skb->data + offset;
1697 if (flags & BPF_F_RECOMPUTE_CSUM)
1698 __skb_postpull_rcsum(skb, ptr, len, offset);
1700 memcpy(ptr, from, len);
1702 if (flags & BPF_F_RECOMPUTE_CSUM)
1703 __skb_postpush_rcsum(skb, ptr, len, offset);
1704 if (flags & BPF_F_INVALIDATE_HASH)
1705 skb_clear_hash(skb);
1710 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1711 .func = bpf_skb_store_bytes,
1713 .ret_type = RET_INTEGER,
1714 .arg1_type = ARG_PTR_TO_CTX,
1715 .arg2_type = ARG_ANYTHING,
1716 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1717 .arg4_type = ARG_CONST_SIZE,
1718 .arg5_type = ARG_ANYTHING,
1721 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1722 void *, to, u32, len)
1726 if (unlikely(offset > INT_MAX))
1729 ptr = skb_header_pointer(skb, offset, len, to);
1733 memcpy(to, ptr, len);
1741 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1742 .func = bpf_skb_load_bytes,
1744 .ret_type = RET_INTEGER,
1745 .arg1_type = ARG_PTR_TO_CTX,
1746 .arg2_type = ARG_ANYTHING,
1747 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1748 .arg4_type = ARG_CONST_SIZE,
1751 BPF_CALL_4(bpf_flow_dissector_load_bytes,
1752 const struct bpf_flow_dissector *, ctx, u32, offset,
1753 void *, to, u32, len)
1757 if (unlikely(offset > 0xffff))
1760 if (unlikely(!ctx->skb))
1763 ptr = skb_header_pointer(ctx->skb, offset, len, to);
1767 memcpy(to, ptr, len);
1775 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1776 .func = bpf_flow_dissector_load_bytes,
1778 .ret_type = RET_INTEGER,
1779 .arg1_type = ARG_PTR_TO_CTX,
1780 .arg2_type = ARG_ANYTHING,
1781 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1782 .arg4_type = ARG_CONST_SIZE,
1785 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1786 u32, offset, void *, to, u32, len, u32, start_header)
1788 u8 *end = skb_tail_pointer(skb);
1791 if (unlikely(offset > 0xffff))
1794 switch (start_header) {
1795 case BPF_HDR_START_MAC:
1796 if (unlikely(!skb_mac_header_was_set(skb)))
1798 start = skb_mac_header(skb);
1800 case BPF_HDR_START_NET:
1801 start = skb_network_header(skb);
1807 ptr = start + offset;
1809 if (likely(ptr + len <= end)) {
1810 memcpy(to, ptr, len);
1819 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1820 .func = bpf_skb_load_bytes_relative,
1822 .ret_type = RET_INTEGER,
1823 .arg1_type = ARG_PTR_TO_CTX,
1824 .arg2_type = ARG_ANYTHING,
1825 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1826 .arg4_type = ARG_CONST_SIZE,
1827 .arg5_type = ARG_ANYTHING,
1830 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1832 /* Idea is the following: should the needed direct read/write
1833 * test fail during runtime, we can pull in more data and redo
1834 * again, since implicitly, we invalidate previous checks here.
1836 * Or, since we know how much we need to make read/writeable,
1837 * this can be done once at the program beginning for direct
1838 * access case. By this we overcome limitations of only current
1839 * headroom being accessible.
1841 return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1844 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1845 .func = bpf_skb_pull_data,
1847 .ret_type = RET_INTEGER,
1848 .arg1_type = ARG_PTR_TO_CTX,
1849 .arg2_type = ARG_ANYTHING,
1852 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1854 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1857 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1858 .func = bpf_sk_fullsock,
1860 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
1861 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
1864 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1865 unsigned int write_len)
1867 return __bpf_try_make_writable(skb, write_len);
1870 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1872 /* Idea is the following: should the needed direct read/write
1873 * test fail during runtime, we can pull in more data and redo
1874 * again, since implicitly, we invalidate previous checks here.
1876 * Or, since we know how much we need to make read/writeable,
1877 * this can be done once at the program beginning for direct
1878 * access case. By this we overcome limitations of only current
1879 * headroom being accessible.
1881 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1884 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1885 .func = sk_skb_pull_data,
1887 .ret_type = RET_INTEGER,
1888 .arg1_type = ARG_PTR_TO_CTX,
1889 .arg2_type = ARG_ANYTHING,
1892 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1893 u64, from, u64, to, u64, flags)
1897 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1899 if (unlikely(offset > 0xffff || offset & 1))
1901 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1904 ptr = (__sum16 *)(skb->data + offset);
1905 switch (flags & BPF_F_HDR_FIELD_MASK) {
1907 if (unlikely(from != 0))
1910 csum_replace_by_diff(ptr, to);
1913 csum_replace2(ptr, from, to);
1916 csum_replace4(ptr, from, to);
1925 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1926 .func = bpf_l3_csum_replace,
1928 .ret_type = RET_INTEGER,
1929 .arg1_type = ARG_PTR_TO_CTX,
1930 .arg2_type = ARG_ANYTHING,
1931 .arg3_type = ARG_ANYTHING,
1932 .arg4_type = ARG_ANYTHING,
1933 .arg5_type = ARG_ANYTHING,
1936 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1937 u64, from, u64, to, u64, flags)
1939 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1940 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1941 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1944 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1945 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1947 if (unlikely(offset > 0xffff || offset & 1))
1949 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1952 ptr = (__sum16 *)(skb->data + offset);
1953 if (is_mmzero && !do_mforce && !*ptr)
1956 switch (flags & BPF_F_HDR_FIELD_MASK) {
1958 if (unlikely(from != 0))
1961 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1964 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1967 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1973 if (is_mmzero && !*ptr)
1974 *ptr = CSUM_MANGLED_0;
1978 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1979 .func = bpf_l4_csum_replace,
1981 .ret_type = RET_INTEGER,
1982 .arg1_type = ARG_PTR_TO_CTX,
1983 .arg2_type = ARG_ANYTHING,
1984 .arg3_type = ARG_ANYTHING,
1985 .arg4_type = ARG_ANYTHING,
1986 .arg5_type = ARG_ANYTHING,
1989 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1990 __be32 *, to, u32, to_size, __wsum, seed)
1992 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1993 u32 diff_size = from_size + to_size;
1996 /* This is quite flexible, some examples:
1998 * from_size == 0, to_size > 0, seed := csum --> pushing data
1999 * from_size > 0, to_size == 0, seed := csum --> pulling data
2000 * from_size > 0, to_size > 0, seed := 0 --> diffing data
2002 * Even for diffing, from_size and to_size don't need to be equal.
2004 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
2005 diff_size > sizeof(sp->diff)))
2008 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
2009 sp->diff[j] = ~from[i];
2010 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
2011 sp->diff[j] = to[i];
2013 return csum_partial(sp->diff, diff_size, seed);
2016 static const struct bpf_func_proto bpf_csum_diff_proto = {
2017 .func = bpf_csum_diff,
2020 .ret_type = RET_INTEGER,
2021 .arg1_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2022 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
2023 .arg3_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2024 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
2025 .arg5_type = ARG_ANYTHING,
2028 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2030 /* The interface is to be used in combination with bpf_csum_diff()
2031 * for direct packet writes. csum rotation for alignment as well
2032 * as emulating csum_sub() can be done from the eBPF program.
2034 if (skb->ip_summed == CHECKSUM_COMPLETE)
2035 return (skb->csum = csum_add(skb->csum, csum));
2040 static const struct bpf_func_proto bpf_csum_update_proto = {
2041 .func = bpf_csum_update,
2043 .ret_type = RET_INTEGER,
2044 .arg1_type = ARG_PTR_TO_CTX,
2045 .arg2_type = ARG_ANYTHING,
2048 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
2050 /* The interface is to be used in combination with bpf_skb_adjust_room()
2051 * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET
2052 * is passed as flags, for example.
2055 case BPF_CSUM_LEVEL_INC:
2056 __skb_incr_checksum_unnecessary(skb);
2058 case BPF_CSUM_LEVEL_DEC:
2059 __skb_decr_checksum_unnecessary(skb);
2061 case BPF_CSUM_LEVEL_RESET:
2062 __skb_reset_checksum_unnecessary(skb);
2064 case BPF_CSUM_LEVEL_QUERY:
2065 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2066 skb->csum_level : -EACCES;
2074 static const struct bpf_func_proto bpf_csum_level_proto = {
2075 .func = bpf_csum_level,
2077 .ret_type = RET_INTEGER,
2078 .arg1_type = ARG_PTR_TO_CTX,
2079 .arg2_type = ARG_ANYTHING,
2082 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2084 return dev_forward_skb_nomtu(dev, skb);
2087 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2088 struct sk_buff *skb)
2090 int ret = ____dev_forward_skb(dev, skb, false);
2094 ret = netif_rx(skb);
2100 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2104 if (dev_xmit_recursion()) {
2105 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2111 skb_clear_tstamp(skb);
2113 dev_xmit_recursion_inc();
2114 ret = dev_queue_xmit(skb);
2115 dev_xmit_recursion_dec();
2120 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2123 unsigned int mlen = skb_network_offset(skb);
2126 __skb_pull(skb, mlen);
2128 /* At ingress, the mac header has already been pulled once.
2129 * At egress, skb_pospull_rcsum has to be done in case that
2130 * the skb is originated from ingress (i.e. a forwarded skb)
2131 * to ensure that rcsum starts at net header.
2133 if (!skb_at_tc_ingress(skb))
2134 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2136 skb_pop_mac_header(skb);
2137 skb_reset_mac_len(skb);
2138 return flags & BPF_F_INGRESS ?
2139 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2142 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2145 /* Verify that a link layer header is carried */
2146 if (unlikely(skb->mac_header >= skb->network_header)) {
2151 bpf_push_mac_rcsum(skb);
2152 return flags & BPF_F_INGRESS ?
2153 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2156 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2159 if (dev_is_mac_header_xmit(dev))
2160 return __bpf_redirect_common(skb, dev, flags);
2162 return __bpf_redirect_no_mac(skb, dev, flags);
2165 #if IS_ENABLED(CONFIG_IPV6)
2166 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2167 struct net_device *dev, struct bpf_nh_params *nh)
2169 u32 hh_len = LL_RESERVED_SPACE(dev);
2170 const struct in6_addr *nexthop;
2171 struct dst_entry *dst = NULL;
2172 struct neighbour *neigh;
2174 if (dev_xmit_recursion()) {
2175 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2180 skb_clear_tstamp(skb);
2182 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2183 skb = skb_expand_head(skb, hh_len);
2191 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2192 &ipv6_hdr(skb)->daddr);
2194 nexthop = &nh->ipv6_nh;
2196 neigh = ip_neigh_gw6(dev, nexthop);
2197 if (likely(!IS_ERR(neigh))) {
2200 sock_confirm_neigh(skb, neigh);
2201 dev_xmit_recursion_inc();
2202 ret = neigh_output(neigh, skb, false);
2203 dev_xmit_recursion_dec();
2204 rcu_read_unlock_bh();
2207 rcu_read_unlock_bh();
2209 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2215 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2216 struct bpf_nh_params *nh)
2218 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2219 struct net *net = dev_net(dev);
2220 int err, ret = NET_XMIT_DROP;
2223 struct dst_entry *dst;
2224 struct flowi6 fl6 = {
2225 .flowi6_flags = FLOWI_FLAG_ANYSRC,
2226 .flowi6_mark = skb->mark,
2227 .flowlabel = ip6_flowinfo(ip6h),
2228 .flowi6_oif = dev->ifindex,
2229 .flowi6_proto = ip6h->nexthdr,
2230 .daddr = ip6h->daddr,
2231 .saddr = ip6h->saddr,
2234 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2238 skb_dst_set(skb, dst);
2239 } else if (nh->nh_family != AF_INET6) {
2243 err = bpf_out_neigh_v6(net, skb, dev, nh);
2244 if (unlikely(net_xmit_eval(err)))
2245 dev->stats.tx_errors++;
2247 ret = NET_XMIT_SUCCESS;
2250 dev->stats.tx_errors++;
2256 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2257 struct bpf_nh_params *nh)
2260 return NET_XMIT_DROP;
2262 #endif /* CONFIG_IPV6 */
2264 #if IS_ENABLED(CONFIG_INET)
2265 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2266 struct net_device *dev, struct bpf_nh_params *nh)
2268 u32 hh_len = LL_RESERVED_SPACE(dev);
2269 struct neighbour *neigh;
2270 bool is_v6gw = false;
2272 if (dev_xmit_recursion()) {
2273 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2278 skb_clear_tstamp(skb);
2280 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2281 skb = skb_expand_head(skb, hh_len);
2288 struct dst_entry *dst = skb_dst(skb);
2289 struct rtable *rt = container_of(dst, struct rtable, dst);
2291 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2292 } else if (nh->nh_family == AF_INET6) {
2293 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2295 } else if (nh->nh_family == AF_INET) {
2296 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2298 rcu_read_unlock_bh();
2302 if (likely(!IS_ERR(neigh))) {
2305 sock_confirm_neigh(skb, neigh);
2306 dev_xmit_recursion_inc();
2307 ret = neigh_output(neigh, skb, is_v6gw);
2308 dev_xmit_recursion_dec();
2309 rcu_read_unlock_bh();
2312 rcu_read_unlock_bh();
2318 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2319 struct bpf_nh_params *nh)
2321 const struct iphdr *ip4h = ip_hdr(skb);
2322 struct net *net = dev_net(dev);
2323 int err, ret = NET_XMIT_DROP;
2326 struct flowi4 fl4 = {
2327 .flowi4_flags = FLOWI_FLAG_ANYSRC,
2328 .flowi4_mark = skb->mark,
2329 .flowi4_tos = RT_TOS(ip4h->tos),
2330 .flowi4_oif = dev->ifindex,
2331 .flowi4_proto = ip4h->protocol,
2332 .daddr = ip4h->daddr,
2333 .saddr = ip4h->saddr,
2337 rt = ip_route_output_flow(net, &fl4, NULL);
2340 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2345 skb_dst_set(skb, &rt->dst);
2348 err = bpf_out_neigh_v4(net, skb, dev, nh);
2349 if (unlikely(net_xmit_eval(err)))
2350 dev->stats.tx_errors++;
2352 ret = NET_XMIT_SUCCESS;
2355 dev->stats.tx_errors++;
2361 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2362 struct bpf_nh_params *nh)
2365 return NET_XMIT_DROP;
2367 #endif /* CONFIG_INET */
2369 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2370 struct bpf_nh_params *nh)
2372 struct ethhdr *ethh = eth_hdr(skb);
2374 if (unlikely(skb->mac_header >= skb->network_header))
2376 bpf_push_mac_rcsum(skb);
2377 if (is_multicast_ether_addr(ethh->h_dest))
2380 skb_pull(skb, sizeof(*ethh));
2381 skb_unset_mac_header(skb);
2382 skb_reset_network_header(skb);
2384 if (skb->protocol == htons(ETH_P_IP))
2385 return __bpf_redirect_neigh_v4(skb, dev, nh);
2386 else if (skb->protocol == htons(ETH_P_IPV6))
2387 return __bpf_redirect_neigh_v6(skb, dev, nh);
2393 /* Internal, non-exposed redirect flags. */
2395 BPF_F_NEIGH = (1ULL << 1),
2396 BPF_F_PEER = (1ULL << 2),
2397 BPF_F_NEXTHOP = (1ULL << 3),
2398 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2401 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2403 struct net_device *dev;
2404 struct sk_buff *clone;
2407 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2410 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2414 clone = skb_clone(skb, GFP_ATOMIC);
2415 if (unlikely(!clone))
2418 /* For direct write, we need to keep the invariant that the skbs
2419 * we're dealing with need to be uncloned. Should uncloning fail
2420 * here, we need to free the just generated clone to unclone once
2423 ret = bpf_try_make_head_writable(skb);
2424 if (unlikely(ret)) {
2429 return __bpf_redirect(clone, dev, flags);
2432 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2433 .func = bpf_clone_redirect,
2435 .ret_type = RET_INTEGER,
2436 .arg1_type = ARG_PTR_TO_CTX,
2437 .arg2_type = ARG_ANYTHING,
2438 .arg3_type = ARG_ANYTHING,
2441 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2442 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2444 int skb_do_redirect(struct sk_buff *skb)
2446 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2447 struct net *net = dev_net(skb->dev);
2448 struct net_device *dev;
2449 u32 flags = ri->flags;
2451 dev = dev_get_by_index_rcu(net, ri->tgt_index);
2456 if (flags & BPF_F_PEER) {
2457 const struct net_device_ops *ops = dev->netdev_ops;
2459 if (unlikely(!ops->ndo_get_peer_dev ||
2460 !skb_at_tc_ingress(skb)))
2462 dev = ops->ndo_get_peer_dev(dev);
2463 if (unlikely(!dev ||
2464 !(dev->flags & IFF_UP) ||
2465 net_eq(net, dev_net(dev))))
2470 return flags & BPF_F_NEIGH ?
2471 __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2473 __bpf_redirect(skb, dev, flags);
2479 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2481 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2483 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2487 ri->tgt_index = ifindex;
2489 return TC_ACT_REDIRECT;
2492 static const struct bpf_func_proto bpf_redirect_proto = {
2493 .func = bpf_redirect,
2495 .ret_type = RET_INTEGER,
2496 .arg1_type = ARG_ANYTHING,
2497 .arg2_type = ARG_ANYTHING,
2500 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2502 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2504 if (unlikely(flags))
2507 ri->flags = BPF_F_PEER;
2508 ri->tgt_index = ifindex;
2510 return TC_ACT_REDIRECT;
2513 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2514 .func = bpf_redirect_peer,
2516 .ret_type = RET_INTEGER,
2517 .arg1_type = ARG_ANYTHING,
2518 .arg2_type = ARG_ANYTHING,
2521 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2522 int, plen, u64, flags)
2524 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2526 if (unlikely((plen && plen < sizeof(*params)) || flags))
2529 ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2530 ri->tgt_index = ifindex;
2532 BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2534 memcpy(&ri->nh, params, sizeof(ri->nh));
2536 return TC_ACT_REDIRECT;
2539 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2540 .func = bpf_redirect_neigh,
2542 .ret_type = RET_INTEGER,
2543 .arg1_type = ARG_ANYTHING,
2544 .arg2_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2545 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
2546 .arg4_type = ARG_ANYTHING,
2549 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2551 msg->apply_bytes = bytes;
2555 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2556 .func = bpf_msg_apply_bytes,
2558 .ret_type = RET_INTEGER,
2559 .arg1_type = ARG_PTR_TO_CTX,
2560 .arg2_type = ARG_ANYTHING,
2563 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2565 msg->cork_bytes = bytes;
2569 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2570 .func = bpf_msg_cork_bytes,
2572 .ret_type = RET_INTEGER,
2573 .arg1_type = ARG_PTR_TO_CTX,
2574 .arg2_type = ARG_ANYTHING,
2577 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2578 u32, end, u64, flags)
2580 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2581 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2582 struct scatterlist *sge;
2583 u8 *raw, *to, *from;
2586 if (unlikely(flags || end <= start))
2589 /* First find the starting scatterlist element */
2593 len = sk_msg_elem(msg, i)->length;
2594 if (start < offset + len)
2596 sk_msg_iter_var_next(i);
2597 } while (i != msg->sg.end);
2599 if (unlikely(start >= offset + len))
2603 /* The start may point into the sg element so we need to also
2604 * account for the headroom.
2606 bytes_sg_total = start - offset + bytes;
2607 if (!test_bit(i, msg->sg.copy) && bytes_sg_total <= len)
2610 /* At this point we need to linearize multiple scatterlist
2611 * elements or a single shared page. Either way we need to
2612 * copy into a linear buffer exclusively owned by BPF. Then
2613 * place the buffer in the scatterlist and fixup the original
2614 * entries by removing the entries now in the linear buffer
2615 * and shifting the remaining entries. For now we do not try
2616 * to copy partial entries to avoid complexity of running out
2617 * of sg_entry slots. The downside is reading a single byte
2618 * will copy the entire sg entry.
2621 copy += sk_msg_elem(msg, i)->length;
2622 sk_msg_iter_var_next(i);
2623 if (bytes_sg_total <= copy)
2625 } while (i != msg->sg.end);
2628 if (unlikely(bytes_sg_total > copy))
2631 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2633 if (unlikely(!page))
2636 raw = page_address(page);
2639 sge = sk_msg_elem(msg, i);
2640 from = sg_virt(sge);
2644 memcpy(to, from, len);
2647 put_page(sg_page(sge));
2649 sk_msg_iter_var_next(i);
2650 } while (i != last_sge);
2652 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2654 /* To repair sg ring we need to shift entries. If we only
2655 * had a single entry though we can just replace it and
2656 * be done. Otherwise walk the ring and shift the entries.
2658 WARN_ON_ONCE(last_sge == first_sge);
2659 shift = last_sge > first_sge ?
2660 last_sge - first_sge - 1 :
2661 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2666 sk_msg_iter_var_next(i);
2670 if (i + shift >= NR_MSG_FRAG_IDS)
2671 move_from = i + shift - NR_MSG_FRAG_IDS;
2673 move_from = i + shift;
2674 if (move_from == msg->sg.end)
2677 msg->sg.data[i] = msg->sg.data[move_from];
2678 msg->sg.data[move_from].length = 0;
2679 msg->sg.data[move_from].page_link = 0;
2680 msg->sg.data[move_from].offset = 0;
2681 sk_msg_iter_var_next(i);
2684 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2685 msg->sg.end - shift + NR_MSG_FRAG_IDS :
2686 msg->sg.end - shift;
2688 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2689 msg->data_end = msg->data + bytes;
2693 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2694 .func = bpf_msg_pull_data,
2696 .ret_type = RET_INTEGER,
2697 .arg1_type = ARG_PTR_TO_CTX,
2698 .arg2_type = ARG_ANYTHING,
2699 .arg3_type = ARG_ANYTHING,
2700 .arg4_type = ARG_ANYTHING,
2703 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2704 u32, len, u64, flags)
2706 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2707 u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2708 u8 *raw, *to, *from;
2711 if (unlikely(flags))
2714 if (unlikely(len == 0))
2717 /* First find the starting scatterlist element */
2721 l = sk_msg_elem(msg, i)->length;
2723 if (start < offset + l)
2725 sk_msg_iter_var_next(i);
2726 } while (i != msg->sg.end);
2728 if (start >= offset + l)
2731 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2733 /* If no space available will fallback to copy, we need at
2734 * least one scatterlist elem available to push data into
2735 * when start aligns to the beginning of an element or two
2736 * when it falls inside an element. We handle the start equals
2737 * offset case because its the common case for inserting a
2740 if (!space || (space == 1 && start != offset))
2741 copy = msg->sg.data[i].length;
2743 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2744 get_order(copy + len));
2745 if (unlikely(!page))
2751 raw = page_address(page);
2753 psge = sk_msg_elem(msg, i);
2754 front = start - offset;
2755 back = psge->length - front;
2756 from = sg_virt(psge);
2759 memcpy(raw, from, front);
2763 to = raw + front + len;
2765 memcpy(to, from, back);
2768 put_page(sg_page(psge));
2769 } else if (start - offset) {
2770 psge = sk_msg_elem(msg, i);
2771 rsge = sk_msg_elem_cpy(msg, i);
2773 psge->length = start - offset;
2774 rsge.length -= psge->length;
2775 rsge.offset += start;
2777 sk_msg_iter_var_next(i);
2778 sg_unmark_end(psge);
2779 sg_unmark_end(&rsge);
2780 sk_msg_iter_next(msg, end);
2783 /* Slot(s) to place newly allocated data */
2786 /* Shift one or two slots as needed */
2788 sge = sk_msg_elem_cpy(msg, i);
2790 sk_msg_iter_var_next(i);
2791 sg_unmark_end(&sge);
2792 sk_msg_iter_next(msg, end);
2794 nsge = sk_msg_elem_cpy(msg, i);
2796 sk_msg_iter_var_next(i);
2797 nnsge = sk_msg_elem_cpy(msg, i);
2800 while (i != msg->sg.end) {
2801 msg->sg.data[i] = sge;
2803 sk_msg_iter_var_next(i);
2806 nnsge = sk_msg_elem_cpy(msg, i);
2808 nsge = sk_msg_elem_cpy(msg, i);
2813 /* Place newly allocated data buffer */
2814 sk_mem_charge(msg->sk, len);
2815 msg->sg.size += len;
2816 __clear_bit(new, msg->sg.copy);
2817 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2819 get_page(sg_page(&rsge));
2820 sk_msg_iter_var_next(new);
2821 msg->sg.data[new] = rsge;
2824 sk_msg_compute_data_pointers(msg);
2828 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2829 .func = bpf_msg_push_data,
2831 .ret_type = RET_INTEGER,
2832 .arg1_type = ARG_PTR_TO_CTX,
2833 .arg2_type = ARG_ANYTHING,
2834 .arg3_type = ARG_ANYTHING,
2835 .arg4_type = ARG_ANYTHING,
2838 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2844 sk_msg_iter_var_next(i);
2845 msg->sg.data[prev] = msg->sg.data[i];
2846 } while (i != msg->sg.end);
2848 sk_msg_iter_prev(msg, end);
2851 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2853 struct scatterlist tmp, sge;
2855 sk_msg_iter_next(msg, end);
2856 sge = sk_msg_elem_cpy(msg, i);
2857 sk_msg_iter_var_next(i);
2858 tmp = sk_msg_elem_cpy(msg, i);
2860 while (i != msg->sg.end) {
2861 msg->sg.data[i] = sge;
2862 sk_msg_iter_var_next(i);
2864 tmp = sk_msg_elem_cpy(msg, i);
2868 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2869 u32, len, u64, flags)
2871 u32 i = 0, l = 0, space, offset = 0;
2872 u64 last = start + len;
2875 if (unlikely(flags))
2878 /* First find the starting scatterlist element */
2882 l = sk_msg_elem(msg, i)->length;
2884 if (start < offset + l)
2886 sk_msg_iter_var_next(i);
2887 } while (i != msg->sg.end);
2889 /* Bounds checks: start and pop must be inside message */
2890 if (start >= offset + l || last >= msg->sg.size)
2893 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2896 /* --------------| offset
2897 * -| start |-------- len -------|
2899 * |----- a ----|-------- pop -------|----- b ----|
2900 * |______________________________________________| length
2903 * a: region at front of scatter element to save
2904 * b: region at back of scatter element to save when length > A + pop
2905 * pop: region to pop from element, same as input 'pop' here will be
2906 * decremented below per iteration.
2908 * Two top-level cases to handle when start != offset, first B is non
2909 * zero and second B is zero corresponding to when a pop includes more
2912 * Then if B is non-zero AND there is no space allocate space and
2913 * compact A, B regions into page. If there is space shift ring to
2914 * the rigth free'ing the next element in ring to place B, leaving
2915 * A untouched except to reduce length.
2917 if (start != offset) {
2918 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2920 int b = sge->length - pop - a;
2922 sk_msg_iter_var_next(i);
2924 if (pop < sge->length - a) {
2927 sk_msg_shift_right(msg, i);
2928 nsge = sk_msg_elem(msg, i);
2929 get_page(sg_page(sge));
2932 b, sge->offset + pop + a);
2934 struct page *page, *orig;
2937 page = alloc_pages(__GFP_NOWARN |
2938 __GFP_COMP | GFP_ATOMIC,
2940 if (unlikely(!page))
2944 orig = sg_page(sge);
2945 from = sg_virt(sge);
2946 to = page_address(page);
2947 memcpy(to, from, a);
2948 memcpy(to + a, from + a + pop, b);
2949 sg_set_page(sge, page, a + b, 0);
2953 } else if (pop >= sge->length - a) {
2954 pop -= (sge->length - a);
2959 /* From above the current layout _must_ be as follows,
2964 * |---- pop ---|---------------- b ------------|
2965 * |____________________________________________| length
2967 * Offset and start of the current msg elem are equal because in the
2968 * previous case we handled offset != start and either consumed the
2969 * entire element and advanced to the next element OR pop == 0.
2971 * Two cases to handle here are first pop is less than the length
2972 * leaving some remainder b above. Simply adjust the element's layout
2973 * in this case. Or pop >= length of the element so that b = 0. In this
2974 * case advance to next element decrementing pop.
2977 struct scatterlist *sge = sk_msg_elem(msg, i);
2979 if (pop < sge->length) {
2985 sk_msg_shift_left(msg, i);
2987 sk_msg_iter_var_next(i);
2990 sk_mem_uncharge(msg->sk, len - pop);
2991 msg->sg.size -= (len - pop);
2992 sk_msg_compute_data_pointers(msg);
2996 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
2997 .func = bpf_msg_pop_data,
2999 .ret_type = RET_INTEGER,
3000 .arg1_type = ARG_PTR_TO_CTX,
3001 .arg2_type = ARG_ANYTHING,
3002 .arg3_type = ARG_ANYTHING,
3003 .arg4_type = ARG_ANYTHING,
3006 #ifdef CONFIG_CGROUP_NET_CLASSID
3007 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3009 return __task_get_classid(current);
3012 static const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3013 .func = bpf_get_cgroup_classid_curr,
3015 .ret_type = RET_INTEGER,
3018 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3020 struct sock *sk = skb_to_full_sk(skb);
3022 if (!sk || !sk_fullsock(sk))
3025 return sock_cgroup_classid(&sk->sk_cgrp_data);
3028 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3029 .func = bpf_skb_cgroup_classid,
3031 .ret_type = RET_INTEGER,
3032 .arg1_type = ARG_PTR_TO_CTX,
3036 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3038 return task_get_classid(skb);
3041 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3042 .func = bpf_get_cgroup_classid,
3044 .ret_type = RET_INTEGER,
3045 .arg1_type = ARG_PTR_TO_CTX,
3048 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3050 return dst_tclassid(skb);
3053 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3054 .func = bpf_get_route_realm,
3056 .ret_type = RET_INTEGER,
3057 .arg1_type = ARG_PTR_TO_CTX,
3060 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3062 /* If skb_clear_hash() was called due to mangling, we can
3063 * trigger SW recalculation here. Later access to hash
3064 * can then use the inline skb->hash via context directly
3065 * instead of calling this helper again.
3067 return skb_get_hash(skb);
3070 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3071 .func = bpf_get_hash_recalc,
3073 .ret_type = RET_INTEGER,
3074 .arg1_type = ARG_PTR_TO_CTX,
3077 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3079 /* After all direct packet write, this can be used once for
3080 * triggering a lazy recalc on next skb_get_hash() invocation.
3082 skb_clear_hash(skb);
3086 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3087 .func = bpf_set_hash_invalid,
3089 .ret_type = RET_INTEGER,
3090 .arg1_type = ARG_PTR_TO_CTX,
3093 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3095 /* Set user specified hash as L4(+), so that it gets returned
3096 * on skb_get_hash() call unless BPF prog later on triggers a
3099 __skb_set_sw_hash(skb, hash, true);
3103 static const struct bpf_func_proto bpf_set_hash_proto = {
3104 .func = bpf_set_hash,
3106 .ret_type = RET_INTEGER,
3107 .arg1_type = ARG_PTR_TO_CTX,
3108 .arg2_type = ARG_ANYTHING,
3111 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3116 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3117 vlan_proto != htons(ETH_P_8021AD)))
3118 vlan_proto = htons(ETH_P_8021Q);
3120 bpf_push_mac_rcsum(skb);
3121 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3122 bpf_pull_mac_rcsum(skb);
3124 bpf_compute_data_pointers(skb);
3128 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3129 .func = bpf_skb_vlan_push,
3131 .ret_type = RET_INTEGER,
3132 .arg1_type = ARG_PTR_TO_CTX,
3133 .arg2_type = ARG_ANYTHING,
3134 .arg3_type = ARG_ANYTHING,
3137 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3141 bpf_push_mac_rcsum(skb);
3142 ret = skb_vlan_pop(skb);
3143 bpf_pull_mac_rcsum(skb);
3145 bpf_compute_data_pointers(skb);
3149 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3150 .func = bpf_skb_vlan_pop,
3152 .ret_type = RET_INTEGER,
3153 .arg1_type = ARG_PTR_TO_CTX,
3156 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3158 /* Caller already did skb_cow() with len as headroom,
3159 * so no need to do it here.
3162 memmove(skb->data, skb->data + len, off);
3163 memset(skb->data + off, 0, len);
3165 /* No skb_postpush_rcsum(skb, skb->data + off, len)
3166 * needed here as it does not change the skb->csum
3167 * result for checksum complete when summing over
3173 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3175 /* skb_ensure_writable() is not needed here, as we're
3176 * already working on an uncloned skb.
3178 if (unlikely(!pskb_may_pull(skb, off + len)))
3181 skb_postpull_rcsum(skb, skb->data + off, len);
3182 memmove(skb->data + len, skb->data, off);
3183 __skb_pull(skb, len);
3188 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3190 bool trans_same = skb->transport_header == skb->network_header;
3193 /* There's no need for __skb_push()/__skb_pull() pair to
3194 * get to the start of the mac header as we're guaranteed
3195 * to always start from here under eBPF.
3197 ret = bpf_skb_generic_push(skb, off, len);
3199 skb->mac_header -= len;
3200 skb->network_header -= len;
3202 skb->transport_header = skb->network_header;
3208 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3210 bool trans_same = skb->transport_header == skb->network_header;
3213 /* Same here, __skb_push()/__skb_pull() pair not needed. */
3214 ret = bpf_skb_generic_pop(skb, off, len);
3216 skb->mac_header += len;
3217 skb->network_header += len;
3219 skb->transport_header = skb->network_header;
3225 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3227 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3228 u32 off = skb_mac_header_len(skb);
3231 ret = skb_cow(skb, len_diff);
3232 if (unlikely(ret < 0))
3235 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3236 if (unlikely(ret < 0))
3239 if (skb_is_gso(skb)) {
3240 struct skb_shared_info *shinfo = skb_shinfo(skb);
3242 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */
3243 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3244 shinfo->gso_type &= ~SKB_GSO_TCPV4;
3245 shinfo->gso_type |= SKB_GSO_TCPV6;
3249 skb->protocol = htons(ETH_P_IPV6);
3250 skb_clear_hash(skb);
3255 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3257 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3258 u32 off = skb_mac_header_len(skb);
3261 ret = skb_unclone(skb, GFP_ATOMIC);
3262 if (unlikely(ret < 0))
3265 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3266 if (unlikely(ret < 0))
3269 if (skb_is_gso(skb)) {
3270 struct skb_shared_info *shinfo = skb_shinfo(skb);
3272 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */
3273 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3274 shinfo->gso_type &= ~SKB_GSO_TCPV6;
3275 shinfo->gso_type |= SKB_GSO_TCPV4;
3279 skb->protocol = htons(ETH_P_IP);
3280 skb_clear_hash(skb);
3285 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3287 __be16 from_proto = skb->protocol;
3289 if (from_proto == htons(ETH_P_IP) &&
3290 to_proto == htons(ETH_P_IPV6))
3291 return bpf_skb_proto_4_to_6(skb);
3293 if (from_proto == htons(ETH_P_IPV6) &&
3294 to_proto == htons(ETH_P_IP))
3295 return bpf_skb_proto_6_to_4(skb);
3300 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3305 if (unlikely(flags))
3308 /* General idea is that this helper does the basic groundwork
3309 * needed for changing the protocol, and eBPF program fills the
3310 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3311 * and other helpers, rather than passing a raw buffer here.
3313 * The rationale is to keep this minimal and without a need to
3314 * deal with raw packet data. F.e. even if we would pass buffers
3315 * here, the program still needs to call the bpf_lX_csum_replace()
3316 * helpers anyway. Plus, this way we keep also separation of
3317 * concerns, since f.e. bpf_skb_store_bytes() should only take
3320 * Currently, additional options and extension header space are
3321 * not supported, but flags register is reserved so we can adapt
3322 * that. For offloads, we mark packet as dodgy, so that headers
3323 * need to be verified first.
3325 ret = bpf_skb_proto_xlat(skb, proto);
3326 bpf_compute_data_pointers(skb);
3330 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3331 .func = bpf_skb_change_proto,
3333 .ret_type = RET_INTEGER,
3334 .arg1_type = ARG_PTR_TO_CTX,
3335 .arg2_type = ARG_ANYTHING,
3336 .arg3_type = ARG_ANYTHING,
3339 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3341 /* We only allow a restricted subset to be changed for now. */
3342 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3343 !skb_pkt_type_ok(pkt_type)))
3346 skb->pkt_type = pkt_type;
3350 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3351 .func = bpf_skb_change_type,
3353 .ret_type = RET_INTEGER,
3354 .arg1_type = ARG_PTR_TO_CTX,
3355 .arg2_type = ARG_ANYTHING,
3358 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3360 switch (skb->protocol) {
3361 case htons(ETH_P_IP):
3362 return sizeof(struct iphdr);
3363 case htons(ETH_P_IPV6):
3364 return sizeof(struct ipv6hdr);
3370 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3371 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3373 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \
3374 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3375 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3376 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3377 BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \
3378 BPF_F_ADJ_ROOM_ENCAP_L2( \
3379 BPF_ADJ_ROOM_ENCAP_L2_MASK))
3381 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3384 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3385 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3386 u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3387 unsigned int gso_type = SKB_GSO_DODGY;
3390 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3391 /* udp gso_size delineates datagrams, only allow if fixed */
3392 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3393 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3397 ret = skb_cow_head(skb, len_diff);
3398 if (unlikely(ret < 0))
3402 if (skb->protocol != htons(ETH_P_IP) &&
3403 skb->protocol != htons(ETH_P_IPV6))
3406 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3407 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3410 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3411 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3414 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH &&
3415 inner_mac_len < ETH_HLEN)
3418 if (skb->encapsulation)
3421 mac_len = skb->network_header - skb->mac_header;
3422 inner_net = skb->network_header;
3423 if (inner_mac_len > len_diff)
3425 inner_trans = skb->transport_header;
3428 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3429 if (unlikely(ret < 0))
3433 skb->inner_mac_header = inner_net - inner_mac_len;
3434 skb->inner_network_header = inner_net;
3435 skb->inner_transport_header = inner_trans;
3437 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH)
3438 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
3440 skb_set_inner_protocol(skb, skb->protocol);
3442 skb->encapsulation = 1;
3443 skb_set_network_header(skb, mac_len);
3445 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3446 gso_type |= SKB_GSO_UDP_TUNNEL;
3447 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3448 gso_type |= SKB_GSO_GRE;
3449 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3450 gso_type |= SKB_GSO_IPXIP6;
3451 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3452 gso_type |= SKB_GSO_IPXIP4;
3454 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3455 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3456 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3457 sizeof(struct ipv6hdr) :
3458 sizeof(struct iphdr);
3460 skb_set_transport_header(skb, mac_len + nh_len);
3463 /* Match skb->protocol to new outer l3 protocol */
3464 if (skb->protocol == htons(ETH_P_IP) &&
3465 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3466 skb->protocol = htons(ETH_P_IPV6);
3467 else if (skb->protocol == htons(ETH_P_IPV6) &&
3468 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3469 skb->protocol = htons(ETH_P_IP);
3472 if (skb_is_gso(skb)) {
3473 struct skb_shared_info *shinfo = skb_shinfo(skb);
3475 /* Due to header grow, MSS needs to be downgraded. */
3476 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3477 skb_decrease_gso_size(shinfo, len_diff);
3479 /* Header must be checked, and gso_segs recomputed. */
3480 shinfo->gso_type |= gso_type;
3481 shinfo->gso_segs = 0;
3487 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3492 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3493 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3496 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3497 /* udp gso_size delineates datagrams, only allow if fixed */
3498 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3499 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3503 ret = skb_unclone(skb, GFP_ATOMIC);
3504 if (unlikely(ret < 0))
3507 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3508 if (unlikely(ret < 0))
3511 if (skb_is_gso(skb)) {
3512 struct skb_shared_info *shinfo = skb_shinfo(skb);
3514 /* Due to header shrink, MSS can be upgraded. */
3515 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3516 skb_increase_gso_size(shinfo, len_diff);
3518 /* Header must be checked, and gso_segs recomputed. */
3519 shinfo->gso_type |= SKB_GSO_DODGY;
3520 shinfo->gso_segs = 0;
3526 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3528 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3529 u32, mode, u64, flags)
3531 u32 len_diff_abs = abs(len_diff);
3532 bool shrink = len_diff < 0;
3535 if (unlikely(flags || mode))
3537 if (unlikely(len_diff_abs > 0xfffU))
3541 ret = skb_cow(skb, len_diff);
3542 if (unlikely(ret < 0))
3544 __skb_push(skb, len_diff_abs);
3545 memset(skb->data, 0, len_diff_abs);
3547 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3549 __skb_pull(skb, len_diff_abs);
3551 if (tls_sw_has_ctx_rx(skb->sk)) {
3552 struct strp_msg *rxm = strp_msg(skb);
3554 rxm->full_len += len_diff;
3559 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3560 .func = sk_skb_adjust_room,
3562 .ret_type = RET_INTEGER,
3563 .arg1_type = ARG_PTR_TO_CTX,
3564 .arg2_type = ARG_ANYTHING,
3565 .arg3_type = ARG_ANYTHING,
3566 .arg4_type = ARG_ANYTHING,
3569 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3570 u32, mode, u64, flags)
3572 u32 len_cur, len_diff_abs = abs(len_diff);
3573 u32 len_min = bpf_skb_net_base_len(skb);
3574 u32 len_max = BPF_SKB_MAX_LEN;
3575 __be16 proto = skb->protocol;
3576 bool shrink = len_diff < 0;
3580 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3581 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3583 if (unlikely(len_diff_abs > 0xfffU))
3585 if (unlikely(proto != htons(ETH_P_IP) &&
3586 proto != htons(ETH_P_IPV6)))
3589 off = skb_mac_header_len(skb);
3591 case BPF_ADJ_ROOM_NET:
3592 off += bpf_skb_net_base_len(skb);
3594 case BPF_ADJ_ROOM_MAC:
3600 len_cur = skb->len - skb_network_offset(skb);
3601 if ((shrink && (len_diff_abs >= len_cur ||
3602 len_cur - len_diff_abs < len_min)) ||
3603 (!shrink && (skb->len + len_diff_abs > len_max &&
3607 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3608 bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3609 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3610 __skb_reset_checksum_unnecessary(skb);
3612 bpf_compute_data_pointers(skb);
3616 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3617 .func = bpf_skb_adjust_room,
3619 .ret_type = RET_INTEGER,
3620 .arg1_type = ARG_PTR_TO_CTX,
3621 .arg2_type = ARG_ANYTHING,
3622 .arg3_type = ARG_ANYTHING,
3623 .arg4_type = ARG_ANYTHING,
3626 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3628 u32 min_len = skb_network_offset(skb);
3630 if (skb_transport_header_was_set(skb))
3631 min_len = skb_transport_offset(skb);
3632 if (skb->ip_summed == CHECKSUM_PARTIAL)
3633 min_len = skb_checksum_start_offset(skb) +
3634 skb->csum_offset + sizeof(__sum16);
3638 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3640 unsigned int old_len = skb->len;
3643 ret = __skb_grow_rcsum(skb, new_len);
3645 memset(skb->data + old_len, 0, new_len - old_len);
3649 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3651 return __skb_trim_rcsum(skb, new_len);
3654 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3657 u32 max_len = BPF_SKB_MAX_LEN;
3658 u32 min_len = __bpf_skb_min_len(skb);
3661 if (unlikely(flags || new_len > max_len || new_len < min_len))
3663 if (skb->encapsulation)
3666 /* The basic idea of this helper is that it's performing the
3667 * needed work to either grow or trim an skb, and eBPF program
3668 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3669 * bpf_lX_csum_replace() and others rather than passing a raw
3670 * buffer here. This one is a slow path helper and intended
3671 * for replies with control messages.
3673 * Like in bpf_skb_change_proto(), we want to keep this rather
3674 * minimal and without protocol specifics so that we are able
3675 * to separate concerns as in bpf_skb_store_bytes() should only
3676 * be the one responsible for writing buffers.
3678 * It's really expected to be a slow path operation here for
3679 * control message replies, so we're implicitly linearizing,
3680 * uncloning and drop offloads from the skb by this.
3682 ret = __bpf_try_make_writable(skb, skb->len);
3684 if (new_len > skb->len)
3685 ret = bpf_skb_grow_rcsum(skb, new_len);
3686 else if (new_len < skb->len)
3687 ret = bpf_skb_trim_rcsum(skb, new_len);
3688 if (!ret && skb_is_gso(skb))
3694 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3697 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3699 bpf_compute_data_pointers(skb);
3703 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3704 .func = bpf_skb_change_tail,
3706 .ret_type = RET_INTEGER,
3707 .arg1_type = ARG_PTR_TO_CTX,
3708 .arg2_type = ARG_ANYTHING,
3709 .arg3_type = ARG_ANYTHING,
3712 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3715 return __bpf_skb_change_tail(skb, new_len, flags);
3718 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3719 .func = sk_skb_change_tail,
3721 .ret_type = RET_INTEGER,
3722 .arg1_type = ARG_PTR_TO_CTX,
3723 .arg2_type = ARG_ANYTHING,
3724 .arg3_type = ARG_ANYTHING,
3727 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3730 u32 max_len = BPF_SKB_MAX_LEN;
3731 u32 new_len = skb->len + head_room;
3734 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3735 new_len < skb->len))
3738 ret = skb_cow(skb, head_room);
3740 /* Idea for this helper is that we currently only
3741 * allow to expand on mac header. This means that
3742 * skb->protocol network header, etc, stay as is.
3743 * Compared to bpf_skb_change_tail(), we're more
3744 * flexible due to not needing to linearize or
3745 * reset GSO. Intention for this helper is to be
3746 * used by an L3 skb that needs to push mac header
3747 * for redirection into L2 device.
3749 __skb_push(skb, head_room);
3750 memset(skb->data, 0, head_room);
3751 skb_reset_mac_header(skb);
3752 skb_reset_mac_len(skb);
3758 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3761 int ret = __bpf_skb_change_head(skb, head_room, flags);
3763 bpf_compute_data_pointers(skb);
3767 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3768 .func = bpf_skb_change_head,
3770 .ret_type = RET_INTEGER,
3771 .arg1_type = ARG_PTR_TO_CTX,
3772 .arg2_type = ARG_ANYTHING,
3773 .arg3_type = ARG_ANYTHING,
3776 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3779 return __bpf_skb_change_head(skb, head_room, flags);
3782 static const struct bpf_func_proto sk_skb_change_head_proto = {
3783 .func = sk_skb_change_head,
3785 .ret_type = RET_INTEGER,
3786 .arg1_type = ARG_PTR_TO_CTX,
3787 .arg2_type = ARG_ANYTHING,
3788 .arg3_type = ARG_ANYTHING,
3791 BPF_CALL_1(bpf_xdp_get_buff_len, struct xdp_buff*, xdp)
3793 return xdp_get_buff_len(xdp);
3796 static const struct bpf_func_proto bpf_xdp_get_buff_len_proto = {
3797 .func = bpf_xdp_get_buff_len,
3799 .ret_type = RET_INTEGER,
3800 .arg1_type = ARG_PTR_TO_CTX,
3803 BTF_ID_LIST_SINGLE(bpf_xdp_get_buff_len_bpf_ids, struct, xdp_buff)
3805 const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto = {
3806 .func = bpf_xdp_get_buff_len,
3808 .arg1_type = ARG_PTR_TO_BTF_ID,
3809 .arg1_btf_id = &bpf_xdp_get_buff_len_bpf_ids[0],
3812 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3814 return xdp_data_meta_unsupported(xdp) ? 0 :
3815 xdp->data - xdp->data_meta;
3818 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3820 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3821 unsigned long metalen = xdp_get_metalen(xdp);
3822 void *data_start = xdp_frame_end + metalen;
3823 void *data = xdp->data + offset;
3825 if (unlikely(data < data_start ||
3826 data > xdp->data_end - ETH_HLEN))
3830 memmove(xdp->data_meta + offset,
3831 xdp->data_meta, metalen);
3832 xdp->data_meta += offset;
3838 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3839 .func = bpf_xdp_adjust_head,
3841 .ret_type = RET_INTEGER,
3842 .arg1_type = ARG_PTR_TO_CTX,
3843 .arg2_type = ARG_ANYTHING,
3846 static void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off,
3847 void *buf, unsigned long len, bool flush)
3849 unsigned long ptr_len, ptr_off = 0;
3850 skb_frag_t *next_frag, *end_frag;
3851 struct skb_shared_info *sinfo;
3855 if (likely(xdp->data_end - xdp->data >= off + len)) {
3856 src = flush ? buf : xdp->data + off;
3857 dst = flush ? xdp->data + off : buf;
3858 memcpy(dst, src, len);
3862 sinfo = xdp_get_shared_info_from_buff(xdp);
3863 end_frag = &sinfo->frags[sinfo->nr_frags];
3864 next_frag = &sinfo->frags[0];
3866 ptr_len = xdp->data_end - xdp->data;
3867 ptr_buf = xdp->data;
3870 if (off < ptr_off + ptr_len) {
3871 unsigned long copy_off = off - ptr_off;
3872 unsigned long copy_len = min(len, ptr_len - copy_off);
3874 src = flush ? buf : ptr_buf + copy_off;
3875 dst = flush ? ptr_buf + copy_off : buf;
3876 memcpy(dst, src, copy_len);
3883 if (!len || next_frag == end_frag)
3887 ptr_buf = skb_frag_address(next_frag);
3888 ptr_len = skb_frag_size(next_frag);
3893 static void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len)
3895 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
3896 u32 size = xdp->data_end - xdp->data;
3897 void *addr = xdp->data;
3900 if (unlikely(offset > 0xffff || len > 0xffff))
3901 return ERR_PTR(-EFAULT);
3903 if (offset + len > xdp_get_buff_len(xdp))
3904 return ERR_PTR(-EINVAL);
3906 if (offset < size) /* linear area */
3910 for (i = 0; i < sinfo->nr_frags; i++) { /* paged area */
3911 u32 frag_size = skb_frag_size(&sinfo->frags[i]);
3913 if (offset < frag_size) {
3914 addr = skb_frag_address(&sinfo->frags[i]);
3918 offset -= frag_size;
3921 return offset + len < size ? addr + offset : NULL;
3924 BPF_CALL_4(bpf_xdp_load_bytes, struct xdp_buff *, xdp, u32, offset,
3925 void *, buf, u32, len)
3929 ptr = bpf_xdp_pointer(xdp, offset, len);
3931 return PTR_ERR(ptr);
3934 bpf_xdp_copy_buf(xdp, offset, buf, len, false);
3936 memcpy(buf, ptr, len);
3941 static const struct bpf_func_proto bpf_xdp_load_bytes_proto = {
3942 .func = bpf_xdp_load_bytes,
3944 .ret_type = RET_INTEGER,
3945 .arg1_type = ARG_PTR_TO_CTX,
3946 .arg2_type = ARG_ANYTHING,
3947 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
3948 .arg4_type = ARG_CONST_SIZE,
3951 BPF_CALL_4(bpf_xdp_store_bytes, struct xdp_buff *, xdp, u32, offset,
3952 void *, buf, u32, len)
3956 ptr = bpf_xdp_pointer(xdp, offset, len);
3958 return PTR_ERR(ptr);
3961 bpf_xdp_copy_buf(xdp, offset, buf, len, true);
3963 memcpy(ptr, buf, len);
3968 static const struct bpf_func_proto bpf_xdp_store_bytes_proto = {
3969 .func = bpf_xdp_store_bytes,
3971 .ret_type = RET_INTEGER,
3972 .arg1_type = ARG_PTR_TO_CTX,
3973 .arg2_type = ARG_ANYTHING,
3974 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
3975 .arg4_type = ARG_CONST_SIZE,
3978 static int bpf_xdp_frags_increase_tail(struct xdp_buff *xdp, int offset)
3980 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
3981 skb_frag_t *frag = &sinfo->frags[sinfo->nr_frags - 1];
3982 struct xdp_rxq_info *rxq = xdp->rxq;
3983 unsigned int tailroom;
3985 if (!rxq->frag_size || rxq->frag_size > xdp->frame_sz)
3988 tailroom = rxq->frag_size - skb_frag_size(frag) - skb_frag_off(frag);
3989 if (unlikely(offset > tailroom))
3992 memset(skb_frag_address(frag) + skb_frag_size(frag), 0, offset);
3993 skb_frag_size_add(frag, offset);
3994 sinfo->xdp_frags_size += offset;
3999 static int bpf_xdp_frags_shrink_tail(struct xdp_buff *xdp, int offset)
4001 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
4002 int i, n_frags_free = 0, len_free = 0;
4004 if (unlikely(offset > (int)xdp_get_buff_len(xdp) - ETH_HLEN))
4007 for (i = sinfo->nr_frags - 1; i >= 0 && offset > 0; i--) {
4008 skb_frag_t *frag = &sinfo->frags[i];
4009 int shrink = min_t(int, offset, skb_frag_size(frag));
4014 if (skb_frag_size(frag) == shrink) {
4015 struct page *page = skb_frag_page(frag);
4017 __xdp_return(page_address(page), &xdp->rxq->mem,
4021 skb_frag_size_sub(frag, shrink);
4025 sinfo->nr_frags -= n_frags_free;
4026 sinfo->xdp_frags_size -= len_free;
4028 if (unlikely(!sinfo->nr_frags)) {
4029 xdp_buff_clear_frags_flag(xdp);
4030 xdp->data_end -= offset;
4036 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
4038 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
4039 void *data_end = xdp->data_end + offset;
4041 if (unlikely(xdp_buff_has_frags(xdp))) { /* non-linear xdp buff */
4043 return bpf_xdp_frags_shrink_tail(xdp, -offset);
4045 return bpf_xdp_frags_increase_tail(xdp, offset);
4048 /* Notice that xdp_data_hard_end have reserved some tailroom */
4049 if (unlikely(data_end > data_hard_end))
4052 /* ALL drivers MUST init xdp->frame_sz, chicken check below */
4053 if (unlikely(xdp->frame_sz > PAGE_SIZE)) {
4054 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz);
4058 if (unlikely(data_end < xdp->data + ETH_HLEN))
4061 /* Clear memory area on grow, can contain uninit kernel memory */
4063 memset(xdp->data_end, 0, offset);
4065 xdp->data_end = data_end;
4070 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
4071 .func = bpf_xdp_adjust_tail,
4073 .ret_type = RET_INTEGER,
4074 .arg1_type = ARG_PTR_TO_CTX,
4075 .arg2_type = ARG_ANYTHING,
4078 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
4080 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
4081 void *meta = xdp->data_meta + offset;
4082 unsigned long metalen = xdp->data - meta;
4084 if (xdp_data_meta_unsupported(xdp))
4086 if (unlikely(meta < xdp_frame_end ||
4089 if (unlikely(xdp_metalen_invalid(metalen)))
4092 xdp->data_meta = meta;
4097 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
4098 .func = bpf_xdp_adjust_meta,
4100 .ret_type = RET_INTEGER,
4101 .arg1_type = ARG_PTR_TO_CTX,
4102 .arg2_type = ARG_ANYTHING,
4105 /* XDP_REDIRECT works by a three-step process, implemented in the functions
4108 * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target
4109 * of the redirect and store it (along with some other metadata) in a per-CPU
4110 * struct bpf_redirect_info.
4112 * 2. When the program returns the XDP_REDIRECT return code, the driver will
4113 * call xdp_do_redirect() which will use the information in struct
4114 * bpf_redirect_info to actually enqueue the frame into a map type-specific
4115 * bulk queue structure.
4117 * 3. Before exiting its NAPI poll loop, the driver will call xdp_do_flush(),
4118 * which will flush all the different bulk queues, thus completing the
4121 * Pointers to the map entries will be kept around for this whole sequence of
4122 * steps, protected by RCU. However, there is no top-level rcu_read_lock() in
4123 * the core code; instead, the RCU protection relies on everything happening
4124 * inside a single NAPI poll sequence, which means it's between a pair of calls
4125 * to local_bh_disable()/local_bh_enable().
4127 * The map entries are marked as __rcu and the map code makes sure to
4128 * dereference those pointers with rcu_dereference_check() in a way that works
4129 * for both sections that to hold an rcu_read_lock() and sections that are
4130 * called from NAPI without a separate rcu_read_lock(). The code below does not
4131 * use RCU annotations, but relies on those in the map code.
4133 void xdp_do_flush(void)
4139 EXPORT_SYMBOL_GPL(xdp_do_flush);
4141 void bpf_clear_redirect_map(struct bpf_map *map)
4143 struct bpf_redirect_info *ri;
4146 for_each_possible_cpu(cpu) {
4147 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
4148 /* Avoid polluting remote cacheline due to writes if
4149 * not needed. Once we pass this test, we need the
4150 * cmpxchg() to make sure it hasn't been changed in
4151 * the meantime by remote CPU.
4153 if (unlikely(READ_ONCE(ri->map) == map))
4154 cmpxchg(&ri->map, map, NULL);
4158 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
4159 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key);
4161 u32 xdp_master_redirect(struct xdp_buff *xdp)
4163 struct net_device *master, *slave;
4164 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4166 master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev);
4167 slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp);
4168 if (slave && slave != xdp->rxq->dev) {
4169 /* The target device is different from the receiving device, so
4170 * redirect it to the new device.
4171 * Using XDP_REDIRECT gets the correct behaviour from XDP enabled
4172 * drivers to unmap the packet from their rx ring.
4174 ri->tgt_index = slave->ifindex;
4175 ri->map_id = INT_MAX;
4176 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4177 return XDP_REDIRECT;
4181 EXPORT_SYMBOL_GPL(xdp_master_redirect);
4183 static inline int __xdp_do_redirect_xsk(struct bpf_redirect_info *ri,
4184 struct net_device *dev,
4185 struct xdp_buff *xdp,
4186 struct bpf_prog *xdp_prog)
4188 enum bpf_map_type map_type = ri->map_type;
4189 void *fwd = ri->tgt_value;
4190 u32 map_id = ri->map_id;
4193 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4194 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4196 err = __xsk_map_redirect(fwd, xdp);
4200 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4203 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4207 static __always_inline int __xdp_do_redirect_frame(struct bpf_redirect_info *ri,
4208 struct net_device *dev,
4209 struct xdp_frame *xdpf,
4210 struct bpf_prog *xdp_prog)
4212 enum bpf_map_type map_type = ri->map_type;
4213 void *fwd = ri->tgt_value;
4214 u32 map_id = ri->map_id;
4215 struct bpf_map *map;
4218 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4219 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4221 if (unlikely(!xdpf)) {
4227 case BPF_MAP_TYPE_DEVMAP:
4229 case BPF_MAP_TYPE_DEVMAP_HASH:
4230 map = READ_ONCE(ri->map);
4231 if (unlikely(map)) {
4232 WRITE_ONCE(ri->map, NULL);
4233 err = dev_map_enqueue_multi(xdpf, dev, map,
4234 ri->flags & BPF_F_EXCLUDE_INGRESS);
4236 err = dev_map_enqueue(fwd, xdpf, dev);
4239 case BPF_MAP_TYPE_CPUMAP:
4240 err = cpu_map_enqueue(fwd, xdpf, dev);
4242 case BPF_MAP_TYPE_UNSPEC:
4243 if (map_id == INT_MAX) {
4244 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4245 if (unlikely(!fwd)) {
4249 err = dev_xdp_enqueue(fwd, xdpf, dev);
4260 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4263 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4267 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
4268 struct bpf_prog *xdp_prog)
4270 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4271 enum bpf_map_type map_type = ri->map_type;
4273 /* XDP_REDIRECT is not fully supported yet for xdp frags since
4274 * not all XDP capable drivers can map non-linear xdp_frame in
4277 if (unlikely(xdp_buff_has_frags(xdp) &&
4278 map_type != BPF_MAP_TYPE_CPUMAP))
4281 if (map_type == BPF_MAP_TYPE_XSKMAP)
4282 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4284 return __xdp_do_redirect_frame(ri, dev, xdp_convert_buff_to_frame(xdp),
4287 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4289 int xdp_do_redirect_frame(struct net_device *dev, struct xdp_buff *xdp,
4290 struct xdp_frame *xdpf, struct bpf_prog *xdp_prog)
4292 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4293 enum bpf_map_type map_type = ri->map_type;
4295 if (map_type == BPF_MAP_TYPE_XSKMAP)
4296 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4298 return __xdp_do_redirect_frame(ri, dev, xdpf, xdp_prog);
4300 EXPORT_SYMBOL_GPL(xdp_do_redirect_frame);
4302 static int xdp_do_generic_redirect_map(struct net_device *dev,
4303 struct sk_buff *skb,
4304 struct xdp_buff *xdp,
4305 struct bpf_prog *xdp_prog,
4307 enum bpf_map_type map_type, u32 map_id)
4309 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4310 struct bpf_map *map;
4314 case BPF_MAP_TYPE_DEVMAP:
4316 case BPF_MAP_TYPE_DEVMAP_HASH:
4317 map = READ_ONCE(ri->map);
4318 if (unlikely(map)) {
4319 WRITE_ONCE(ri->map, NULL);
4320 err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4321 ri->flags & BPF_F_EXCLUDE_INGRESS);
4323 err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4328 case BPF_MAP_TYPE_XSKMAP:
4329 err = xsk_generic_rcv(fwd, xdp);
4334 case BPF_MAP_TYPE_CPUMAP:
4335 err = cpu_map_generic_redirect(fwd, skb);
4344 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4347 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4351 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4352 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4354 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4355 enum bpf_map_type map_type = ri->map_type;
4356 void *fwd = ri->tgt_value;
4357 u32 map_id = ri->map_id;
4360 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4361 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4363 if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4364 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4365 if (unlikely(!fwd)) {
4370 err = xdp_ok_fwd_dev(fwd, skb->len);
4375 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4376 generic_xdp_tx(skb, xdp_prog);
4380 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id);
4382 _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4386 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4388 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4390 if (unlikely(flags))
4393 /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4394 * by map_idr) is used for ifindex based XDP redirect.
4396 ri->tgt_index = ifindex;
4397 ri->map_id = INT_MAX;
4398 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4400 return XDP_REDIRECT;
4403 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4404 .func = bpf_xdp_redirect,
4406 .ret_type = RET_INTEGER,
4407 .arg1_type = ARG_ANYTHING,
4408 .arg2_type = ARG_ANYTHING,
4411 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
4414 return map->ops->map_redirect(map, ifindex, flags);
4417 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4418 .func = bpf_xdp_redirect_map,
4420 .ret_type = RET_INTEGER,
4421 .arg1_type = ARG_CONST_MAP_PTR,
4422 .arg2_type = ARG_ANYTHING,
4423 .arg3_type = ARG_ANYTHING,
4426 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4427 unsigned long off, unsigned long len)
4429 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4433 if (ptr != dst_buff)
4434 memcpy(dst_buff, ptr, len);
4439 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4440 u64, flags, void *, meta, u64, meta_size)
4442 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4444 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4446 if (unlikely(!skb || skb_size > skb->len))
4449 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4453 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4454 .func = bpf_skb_event_output,
4456 .ret_type = RET_INTEGER,
4457 .arg1_type = ARG_PTR_TO_CTX,
4458 .arg2_type = ARG_CONST_MAP_PTR,
4459 .arg3_type = ARG_ANYTHING,
4460 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4461 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4464 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4466 const struct bpf_func_proto bpf_skb_output_proto = {
4467 .func = bpf_skb_event_output,
4469 .ret_type = RET_INTEGER,
4470 .arg1_type = ARG_PTR_TO_BTF_ID,
4471 .arg1_btf_id = &bpf_skb_output_btf_ids[0],
4472 .arg2_type = ARG_CONST_MAP_PTR,
4473 .arg3_type = ARG_ANYTHING,
4474 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4475 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4478 static unsigned short bpf_tunnel_key_af(u64 flags)
4480 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4483 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4484 u32, size, u64, flags)
4486 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4487 u8 compat[sizeof(struct bpf_tunnel_key)];
4491 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
4495 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4499 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4502 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4503 case offsetof(struct bpf_tunnel_key, tunnel_label):
4504 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4506 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4507 /* Fixup deprecated structure layouts here, so we have
4508 * a common path later on.
4510 if (ip_tunnel_info_af(info) != AF_INET)
4513 to = (struct bpf_tunnel_key *)compat;
4520 to->tunnel_id = be64_to_cpu(info->key.tun_id);
4521 to->tunnel_tos = info->key.tos;
4522 to->tunnel_ttl = info->key.ttl;
4525 if (flags & BPF_F_TUNINFO_IPV6) {
4526 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4527 sizeof(to->remote_ipv6));
4528 memcpy(to->local_ipv6, &info->key.u.ipv6.dst,
4529 sizeof(to->local_ipv6));
4530 to->tunnel_label = be32_to_cpu(info->key.label);
4532 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4533 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4534 to->local_ipv4 = be32_to_cpu(info->key.u.ipv4.dst);
4535 memset(&to->local_ipv6[1], 0, sizeof(__u32) * 3);
4536 to->tunnel_label = 0;
4539 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4540 memcpy(to_orig, to, size);
4544 memset(to_orig, 0, size);
4548 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4549 .func = bpf_skb_get_tunnel_key,
4551 .ret_type = RET_INTEGER,
4552 .arg1_type = ARG_PTR_TO_CTX,
4553 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4554 .arg3_type = ARG_CONST_SIZE,
4555 .arg4_type = ARG_ANYTHING,
4558 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4560 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4563 if (unlikely(!info ||
4564 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4568 if (unlikely(size < info->options_len)) {
4573 ip_tunnel_info_opts_get(to, info);
4574 if (size > info->options_len)
4575 memset(to + info->options_len, 0, size - info->options_len);
4577 return info->options_len;
4579 memset(to, 0, size);
4583 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4584 .func = bpf_skb_get_tunnel_opt,
4586 .ret_type = RET_INTEGER,
4587 .arg1_type = ARG_PTR_TO_CTX,
4588 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4589 .arg3_type = ARG_CONST_SIZE,
4592 static struct metadata_dst __percpu *md_dst;
4594 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4595 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4597 struct metadata_dst *md = this_cpu_ptr(md_dst);
4598 u8 compat[sizeof(struct bpf_tunnel_key)];
4599 struct ip_tunnel_info *info;
4601 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4602 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
4604 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4606 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4607 case offsetof(struct bpf_tunnel_key, tunnel_label):
4608 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4609 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4610 /* Fixup deprecated structure layouts here, so we have
4611 * a common path later on.
4613 memcpy(compat, from, size);
4614 memset(compat + size, 0, sizeof(compat) - size);
4615 from = (const struct bpf_tunnel_key *) compat;
4621 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4626 dst_hold((struct dst_entry *) md);
4627 skb_dst_set(skb, (struct dst_entry *) md);
4629 info = &md->u.tun_info;
4630 memset(info, 0, sizeof(*info));
4631 info->mode = IP_TUNNEL_INFO_TX;
4633 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4634 if (flags & BPF_F_DONT_FRAGMENT)
4635 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4636 if (flags & BPF_F_ZERO_CSUM_TX)
4637 info->key.tun_flags &= ~TUNNEL_CSUM;
4638 if (flags & BPF_F_SEQ_NUMBER)
4639 info->key.tun_flags |= TUNNEL_SEQ;
4641 info->key.tun_id = cpu_to_be64(from->tunnel_id);
4642 info->key.tos = from->tunnel_tos;
4643 info->key.ttl = from->tunnel_ttl;
4645 if (flags & BPF_F_TUNINFO_IPV6) {
4646 info->mode |= IP_TUNNEL_INFO_IPV6;
4647 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4648 sizeof(from->remote_ipv6));
4649 memcpy(&info->key.u.ipv6.src, from->local_ipv6,
4650 sizeof(from->local_ipv6));
4651 info->key.label = cpu_to_be32(from->tunnel_label) &
4652 IPV6_FLOWLABEL_MASK;
4654 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4655 info->key.u.ipv4.src = cpu_to_be32(from->local_ipv4);
4661 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4662 .func = bpf_skb_set_tunnel_key,
4664 .ret_type = RET_INTEGER,
4665 .arg1_type = ARG_PTR_TO_CTX,
4666 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4667 .arg3_type = ARG_CONST_SIZE,
4668 .arg4_type = ARG_ANYTHING,
4671 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4672 const u8 *, from, u32, size)
4674 struct ip_tunnel_info *info = skb_tunnel_info(skb);
4675 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4677 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4679 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4682 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4687 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4688 .func = bpf_skb_set_tunnel_opt,
4690 .ret_type = RET_INTEGER,
4691 .arg1_type = ARG_PTR_TO_CTX,
4692 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4693 .arg3_type = ARG_CONST_SIZE,
4696 static const struct bpf_func_proto *
4697 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4700 struct metadata_dst __percpu *tmp;
4702 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4707 if (cmpxchg(&md_dst, NULL, tmp))
4708 metadata_dst_free_percpu(tmp);
4712 case BPF_FUNC_skb_set_tunnel_key:
4713 return &bpf_skb_set_tunnel_key_proto;
4714 case BPF_FUNC_skb_set_tunnel_opt:
4715 return &bpf_skb_set_tunnel_opt_proto;
4721 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4724 struct bpf_array *array = container_of(map, struct bpf_array, map);
4725 struct cgroup *cgrp;
4728 sk = skb_to_full_sk(skb);
4729 if (!sk || !sk_fullsock(sk))
4731 if (unlikely(idx >= array->map.max_entries))
4734 cgrp = READ_ONCE(array->ptrs[idx]);
4735 if (unlikely(!cgrp))
4738 return sk_under_cgroup_hierarchy(sk, cgrp);
4741 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4742 .func = bpf_skb_under_cgroup,
4744 .ret_type = RET_INTEGER,
4745 .arg1_type = ARG_PTR_TO_CTX,
4746 .arg2_type = ARG_CONST_MAP_PTR,
4747 .arg3_type = ARG_ANYTHING,
4750 #ifdef CONFIG_SOCK_CGROUP_DATA
4751 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4753 struct cgroup *cgrp;
4755 sk = sk_to_full_sk(sk);
4756 if (!sk || !sk_fullsock(sk))
4759 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4760 return cgroup_id(cgrp);
4763 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4765 return __bpf_sk_cgroup_id(skb->sk);
4768 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4769 .func = bpf_skb_cgroup_id,
4771 .ret_type = RET_INTEGER,
4772 .arg1_type = ARG_PTR_TO_CTX,
4775 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4778 struct cgroup *ancestor;
4779 struct cgroup *cgrp;
4781 sk = sk_to_full_sk(sk);
4782 if (!sk || !sk_fullsock(sk))
4785 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4786 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4790 return cgroup_id(ancestor);
4793 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4796 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4799 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4800 .func = bpf_skb_ancestor_cgroup_id,
4802 .ret_type = RET_INTEGER,
4803 .arg1_type = ARG_PTR_TO_CTX,
4804 .arg2_type = ARG_ANYTHING,
4807 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4809 return __bpf_sk_cgroup_id(sk);
4812 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4813 .func = bpf_sk_cgroup_id,
4815 .ret_type = RET_INTEGER,
4816 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4819 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4821 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4824 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4825 .func = bpf_sk_ancestor_cgroup_id,
4827 .ret_type = RET_INTEGER,
4828 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4829 .arg2_type = ARG_ANYTHING,
4833 static unsigned long bpf_xdp_copy(void *dst, const void *ctx,
4834 unsigned long off, unsigned long len)
4836 struct xdp_buff *xdp = (struct xdp_buff *)ctx;
4838 bpf_xdp_copy_buf(xdp, off, dst, len, false);
4842 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4843 u64, flags, void *, meta, u64, meta_size)
4845 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4847 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4850 if (unlikely(!xdp || xdp_size > xdp_get_buff_len(xdp)))
4853 return bpf_event_output(map, flags, meta, meta_size, xdp,
4854 xdp_size, bpf_xdp_copy);
4857 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4858 .func = bpf_xdp_event_output,
4860 .ret_type = RET_INTEGER,
4861 .arg1_type = ARG_PTR_TO_CTX,
4862 .arg2_type = ARG_CONST_MAP_PTR,
4863 .arg3_type = ARG_ANYTHING,
4864 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4865 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4868 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4870 const struct bpf_func_proto bpf_xdp_output_proto = {
4871 .func = bpf_xdp_event_output,
4873 .ret_type = RET_INTEGER,
4874 .arg1_type = ARG_PTR_TO_BTF_ID,
4875 .arg1_btf_id = &bpf_xdp_output_btf_ids[0],
4876 .arg2_type = ARG_CONST_MAP_PTR,
4877 .arg3_type = ARG_ANYTHING,
4878 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4879 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4882 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4884 return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4887 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4888 .func = bpf_get_socket_cookie,
4890 .ret_type = RET_INTEGER,
4891 .arg1_type = ARG_PTR_TO_CTX,
4894 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4896 return __sock_gen_cookie(ctx->sk);
4899 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4900 .func = bpf_get_socket_cookie_sock_addr,
4902 .ret_type = RET_INTEGER,
4903 .arg1_type = ARG_PTR_TO_CTX,
4906 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4908 return __sock_gen_cookie(ctx);
4911 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4912 .func = bpf_get_socket_cookie_sock,
4914 .ret_type = RET_INTEGER,
4915 .arg1_type = ARG_PTR_TO_CTX,
4918 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
4920 return sk ? sock_gen_cookie(sk) : 0;
4923 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
4924 .func = bpf_get_socket_ptr_cookie,
4926 .ret_type = RET_INTEGER,
4927 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4930 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4932 return __sock_gen_cookie(ctx->sk);
4935 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4936 .func = bpf_get_socket_cookie_sock_ops,
4938 .ret_type = RET_INTEGER,
4939 .arg1_type = ARG_PTR_TO_CTX,
4942 static u64 __bpf_get_netns_cookie(struct sock *sk)
4944 const struct net *net = sk ? sock_net(sk) : &init_net;
4946 return net->net_cookie;
4949 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
4951 return __bpf_get_netns_cookie(ctx);
4954 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
4955 .func = bpf_get_netns_cookie_sock,
4957 .ret_type = RET_INTEGER,
4958 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4961 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4963 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4966 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
4967 .func = bpf_get_netns_cookie_sock_addr,
4969 .ret_type = RET_INTEGER,
4970 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4973 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4975 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4978 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = {
4979 .func = bpf_get_netns_cookie_sock_ops,
4981 .ret_type = RET_INTEGER,
4982 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4985 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx)
4987 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4990 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = {
4991 .func = bpf_get_netns_cookie_sk_msg,
4993 .ret_type = RET_INTEGER,
4994 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4997 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4999 struct sock *sk = sk_to_full_sk(skb->sk);
5002 if (!sk || !sk_fullsock(sk))
5004 kuid = sock_net_uid(sock_net(sk), sk);
5005 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
5008 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
5009 .func = bpf_get_socket_uid,
5011 .ret_type = RET_INTEGER,
5012 .arg1_type = ARG_PTR_TO_CTX,
5015 static int __bpf_setsockopt(struct sock *sk, int level, int optname,
5016 char *optval, int optlen)
5018 char devname[IFNAMSIZ];
5024 if (!sk_fullsock(sk))
5027 if (level == SOL_SOCKET) {
5028 if (optlen != sizeof(int) && optname != SO_BINDTODEVICE)
5030 val = *((int *)optval);
5031 valbool = val ? 1 : 0;
5033 /* Only some socketops are supported */
5036 val = min_t(u32, val, sysctl_rmem_max);
5037 val = min_t(int, val, INT_MAX / 2);
5038 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
5039 WRITE_ONCE(sk->sk_rcvbuf,
5040 max_t(int, val * 2, SOCK_MIN_RCVBUF));
5043 val = min_t(u32, val, sysctl_wmem_max);
5044 val = min_t(int, val, INT_MAX / 2);
5045 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
5046 WRITE_ONCE(sk->sk_sndbuf,
5047 max_t(int, val * 2, SOCK_MIN_SNDBUF));
5049 case SO_MAX_PACING_RATE: /* 32bit version */
5051 cmpxchg(&sk->sk_pacing_status,
5054 sk->sk_max_pacing_rate = (val == ~0U) ?
5055 ~0UL : (unsigned int)val;
5056 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
5057 sk->sk_max_pacing_rate);
5060 sk->sk_priority = val;
5065 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
5068 if (sk->sk_mark != val) {
5073 case SO_BINDTODEVICE:
5074 optlen = min_t(long, optlen, IFNAMSIZ - 1);
5075 strncpy(devname, optval, optlen);
5076 devname[optlen] = 0;
5079 if (devname[0] != '\0') {
5080 struct net_device *dev;
5085 dev = dev_get_by_name(net, devname);
5088 ifindex = dev->ifindex;
5092 case SO_BINDTOIFINDEX:
5093 if (optname == SO_BINDTOIFINDEX)
5095 ret = sock_bindtoindex(sk, ifindex, false);
5098 if (sk->sk_prot->keepalive)
5099 sk->sk_prot->keepalive(sk, valbool);
5100 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
5103 sk->sk_reuseport = valbool;
5106 if (val < -1 || val > 1) {
5110 sk->sk_txrehash = (u8)val;
5116 } else if (level == SOL_IP) {
5117 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
5120 val = *((int *)optval);
5121 /* Only some options are supported */
5124 if (val < -1 || val > 0xff) {
5127 struct inet_sock *inet = inet_sk(sk);
5137 #if IS_ENABLED(CONFIG_IPV6)
5138 } else if (level == SOL_IPV6) {
5139 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
5142 val = *((int *)optval);
5143 /* Only some options are supported */
5146 if (val < -1 || val > 0xff) {
5149 struct ipv6_pinfo *np = inet6_sk(sk);
5160 } else if (level == SOL_TCP &&
5161 sk->sk_prot->setsockopt == tcp_setsockopt) {
5162 if (optname == TCP_CONGESTION) {
5163 char name[TCP_CA_NAME_MAX];
5165 strncpy(name, optval, min_t(long, optlen,
5166 TCP_CA_NAME_MAX-1));
5167 name[TCP_CA_NAME_MAX-1] = 0;
5168 ret = tcp_set_congestion_control(sk, name, false, true);
5170 struct inet_connection_sock *icsk = inet_csk(sk);
5171 struct tcp_sock *tp = tcp_sk(sk);
5172 unsigned long timeout;
5174 if (optlen != sizeof(int))
5177 val = *((int *)optval);
5178 /* Only some options are supported */
5181 if (val <= 0 || tp->data_segs_out > tp->syn_data)
5184 tcp_snd_cwnd_set(tp, val);
5186 case TCP_BPF_SNDCWND_CLAMP:
5190 tp->snd_cwnd_clamp = val;
5191 tp->snd_ssthresh = val;
5194 case TCP_BPF_DELACK_MAX:
5195 timeout = usecs_to_jiffies(val);
5196 if (timeout > TCP_DELACK_MAX ||
5197 timeout < TCP_TIMEOUT_MIN)
5199 inet_csk(sk)->icsk_delack_max = timeout;
5201 case TCP_BPF_RTO_MIN:
5202 timeout = usecs_to_jiffies(val);
5203 if (timeout > TCP_RTO_MIN ||
5204 timeout < TCP_TIMEOUT_MIN)
5206 inet_csk(sk)->icsk_rto_min = timeout;
5209 if (val < 0 || val > 1)
5215 ret = tcp_sock_set_keepidle_locked(sk, val);
5218 if (val < 1 || val > MAX_TCP_KEEPINTVL)
5221 tp->keepalive_intvl = val * HZ;
5224 if (val < 1 || val > MAX_TCP_KEEPCNT)
5227 tp->keepalive_probes = val;
5230 if (val < 1 || val > MAX_TCP_SYNCNT)
5233 icsk->icsk_syn_retries = val;
5235 case TCP_USER_TIMEOUT:
5239 icsk->icsk_user_timeout = val;
5241 case TCP_NOTSENT_LOWAT:
5242 tp->notsent_lowat = val;
5243 sk->sk_write_space(sk);
5245 case TCP_WINDOW_CLAMP:
5246 ret = tcp_set_window_clamp(sk, val);
5259 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
5260 char *optval, int optlen)
5262 if (sk_fullsock(sk))
5263 sock_owned_by_me(sk);
5264 return __bpf_setsockopt(sk, level, optname, optval, optlen);
5267 static int __bpf_getsockopt(struct sock *sk, int level, int optname,
5268 char *optval, int optlen)
5270 if (!sk_fullsock(sk))
5273 if (level == SOL_SOCKET) {
5274 if (optlen != sizeof(int))
5279 *((int *)optval) = sk->sk_rcvbuf;
5282 *((int *)optval) = sk->sk_sndbuf;
5285 *((int *)optval) = sk->sk_mark;
5288 *((int *)optval) = sk->sk_priority;
5290 case SO_BINDTOIFINDEX:
5291 *((int *)optval) = sk->sk_bound_dev_if;
5294 *((int *)optval) = sk->sk_reuseport;
5297 *((int *)optval) = sk->sk_txrehash;
5303 } else if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
5304 struct inet_connection_sock *icsk;
5305 struct tcp_sock *tp;
5308 case TCP_CONGESTION:
5309 icsk = inet_csk(sk);
5311 if (!icsk->icsk_ca_ops || optlen <= 1)
5313 strncpy(optval, icsk->icsk_ca_ops->name, optlen);
5314 optval[optlen - 1] = 0;
5319 if (optlen <= 0 || !tp->saved_syn ||
5320 optlen > tcp_saved_syn_len(tp->saved_syn))
5322 memcpy(optval, tp->saved_syn->data, optlen);
5327 } else if (level == SOL_IP) {
5328 struct inet_sock *inet = inet_sk(sk);
5330 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
5333 /* Only some options are supported */
5336 *((int *)optval) = (int)inet->tos;
5341 #if IS_ENABLED(CONFIG_IPV6)
5342 } else if (level == SOL_IPV6) {
5343 struct ipv6_pinfo *np = inet6_sk(sk);
5345 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
5348 /* Only some options are supported */
5351 *((int *)optval) = (int)np->tclass;
5363 memset(optval, 0, optlen);
5367 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
5368 char *optval, int optlen)
5370 if (sk_fullsock(sk))
5371 sock_owned_by_me(sk);
5372 return __bpf_getsockopt(sk, level, optname, optval, optlen);
5375 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level,
5376 int, optname, char *, optval, int, optlen)
5378 if (level == SOL_TCP && optname == TCP_CONGESTION) {
5379 if (optlen >= sizeof("cdg") - 1 &&
5380 !strncmp("cdg", optval, optlen))
5384 return _bpf_setsockopt(sk, level, optname, optval, optlen);
5387 const struct bpf_func_proto bpf_sk_setsockopt_proto = {
5388 .func = bpf_sk_setsockopt,
5390 .ret_type = RET_INTEGER,
5391 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5392 .arg2_type = ARG_ANYTHING,
5393 .arg3_type = ARG_ANYTHING,
5394 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5395 .arg5_type = ARG_CONST_SIZE,
5398 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level,
5399 int, optname, char *, optval, int, optlen)
5401 return _bpf_getsockopt(sk, level, optname, optval, optlen);
5404 const struct bpf_func_proto bpf_sk_getsockopt_proto = {
5405 .func = bpf_sk_getsockopt,
5407 .ret_type = RET_INTEGER,
5408 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5409 .arg2_type = ARG_ANYTHING,
5410 .arg3_type = ARG_ANYTHING,
5411 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5412 .arg5_type = ARG_CONST_SIZE,
5415 BPF_CALL_5(bpf_unlocked_sk_setsockopt, struct sock *, sk, int, level,
5416 int, optname, char *, optval, int, optlen)
5418 return __bpf_setsockopt(sk, level, optname, optval, optlen);
5421 const struct bpf_func_proto bpf_unlocked_sk_setsockopt_proto = {
5422 .func = bpf_unlocked_sk_setsockopt,
5424 .ret_type = RET_INTEGER,
5425 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5426 .arg2_type = ARG_ANYTHING,
5427 .arg3_type = ARG_ANYTHING,
5428 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5429 .arg5_type = ARG_CONST_SIZE,
5432 BPF_CALL_5(bpf_unlocked_sk_getsockopt, struct sock *, sk, int, level,
5433 int, optname, char *, optval, int, optlen)
5435 return __bpf_getsockopt(sk, level, optname, optval, optlen);
5438 const struct bpf_func_proto bpf_unlocked_sk_getsockopt_proto = {
5439 .func = bpf_unlocked_sk_getsockopt,
5441 .ret_type = RET_INTEGER,
5442 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5443 .arg2_type = ARG_ANYTHING,
5444 .arg3_type = ARG_ANYTHING,
5445 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5446 .arg5_type = ARG_CONST_SIZE,
5449 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5450 int, level, int, optname, char *, optval, int, optlen)
5452 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5455 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5456 .func = bpf_sock_addr_setsockopt,
5458 .ret_type = RET_INTEGER,
5459 .arg1_type = ARG_PTR_TO_CTX,
5460 .arg2_type = ARG_ANYTHING,
5461 .arg3_type = ARG_ANYTHING,
5462 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5463 .arg5_type = ARG_CONST_SIZE,
5466 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5467 int, level, int, optname, char *, optval, int, optlen)
5469 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5472 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5473 .func = bpf_sock_addr_getsockopt,
5475 .ret_type = RET_INTEGER,
5476 .arg1_type = ARG_PTR_TO_CTX,
5477 .arg2_type = ARG_ANYTHING,
5478 .arg3_type = ARG_ANYTHING,
5479 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5480 .arg5_type = ARG_CONST_SIZE,
5483 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5484 int, level, int, optname, char *, optval, int, optlen)
5486 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5489 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5490 .func = bpf_sock_ops_setsockopt,
5492 .ret_type = RET_INTEGER,
5493 .arg1_type = ARG_PTR_TO_CTX,
5494 .arg2_type = ARG_ANYTHING,
5495 .arg3_type = ARG_ANYTHING,
5496 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5497 .arg5_type = ARG_CONST_SIZE,
5500 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5501 int optname, const u8 **start)
5503 struct sk_buff *syn_skb = bpf_sock->syn_skb;
5504 const u8 *hdr_start;
5508 /* sk is a request_sock here */
5510 if (optname == TCP_BPF_SYN) {
5511 hdr_start = syn_skb->data;
5512 ret = tcp_hdrlen(syn_skb);
5513 } else if (optname == TCP_BPF_SYN_IP) {
5514 hdr_start = skb_network_header(syn_skb);
5515 ret = skb_network_header_len(syn_skb) +
5516 tcp_hdrlen(syn_skb);
5518 /* optname == TCP_BPF_SYN_MAC */
5519 hdr_start = skb_mac_header(syn_skb);
5520 ret = skb_mac_header_len(syn_skb) +
5521 skb_network_header_len(syn_skb) +
5522 tcp_hdrlen(syn_skb);
5525 struct sock *sk = bpf_sock->sk;
5526 struct saved_syn *saved_syn;
5528 if (sk->sk_state == TCP_NEW_SYN_RECV)
5529 /* synack retransmit. bpf_sock->syn_skb will
5530 * not be available. It has to resort to
5531 * saved_syn (if it is saved).
5533 saved_syn = inet_reqsk(sk)->saved_syn;
5535 saved_syn = tcp_sk(sk)->saved_syn;
5540 if (optname == TCP_BPF_SYN) {
5541 hdr_start = saved_syn->data +
5542 saved_syn->mac_hdrlen +
5543 saved_syn->network_hdrlen;
5544 ret = saved_syn->tcp_hdrlen;
5545 } else if (optname == TCP_BPF_SYN_IP) {
5546 hdr_start = saved_syn->data +
5547 saved_syn->mac_hdrlen;
5548 ret = saved_syn->network_hdrlen +
5549 saved_syn->tcp_hdrlen;
5551 /* optname == TCP_BPF_SYN_MAC */
5553 /* TCP_SAVE_SYN may not have saved the mac hdr */
5554 if (!saved_syn->mac_hdrlen)
5557 hdr_start = saved_syn->data;
5558 ret = saved_syn->mac_hdrlen +
5559 saved_syn->network_hdrlen +
5560 saved_syn->tcp_hdrlen;
5568 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5569 int, level, int, optname, char *, optval, int, optlen)
5571 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5572 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5573 int ret, copy_len = 0;
5576 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5579 if (optlen < copy_len) {
5584 memcpy(optval, start, copy_len);
5587 /* Zero out unused buffer at the end */
5588 memset(optval + copy_len, 0, optlen - copy_len);
5593 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5596 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5597 .func = bpf_sock_ops_getsockopt,
5599 .ret_type = RET_INTEGER,
5600 .arg1_type = ARG_PTR_TO_CTX,
5601 .arg2_type = ARG_ANYTHING,
5602 .arg3_type = ARG_ANYTHING,
5603 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5604 .arg5_type = ARG_CONST_SIZE,
5607 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5610 struct sock *sk = bpf_sock->sk;
5611 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5613 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5616 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5618 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5621 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5622 .func = bpf_sock_ops_cb_flags_set,
5624 .ret_type = RET_INTEGER,
5625 .arg1_type = ARG_PTR_TO_CTX,
5626 .arg2_type = ARG_ANYTHING,
5629 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5630 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5632 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5636 struct sock *sk = ctx->sk;
5637 u32 flags = BIND_FROM_BPF;
5641 if (addr_len < offsetofend(struct sockaddr, sa_family))
5643 if (addr->sa_family == AF_INET) {
5644 if (addr_len < sizeof(struct sockaddr_in))
5646 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5647 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5648 return __inet_bind(sk, addr, addr_len, flags);
5649 #if IS_ENABLED(CONFIG_IPV6)
5650 } else if (addr->sa_family == AF_INET6) {
5651 if (addr_len < SIN6_LEN_RFC2133)
5653 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5654 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5655 /* ipv6_bpf_stub cannot be NULL, since it's called from
5656 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5658 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5659 #endif /* CONFIG_IPV6 */
5661 #endif /* CONFIG_INET */
5663 return -EAFNOSUPPORT;
5666 static const struct bpf_func_proto bpf_bind_proto = {
5669 .ret_type = RET_INTEGER,
5670 .arg1_type = ARG_PTR_TO_CTX,
5671 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5672 .arg3_type = ARG_CONST_SIZE,
5676 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5677 struct bpf_xfrm_state *, to, u32, size, u64, flags)
5679 const struct sec_path *sp = skb_sec_path(skb);
5680 const struct xfrm_state *x;
5682 if (!sp || unlikely(index >= sp->len || flags))
5685 x = sp->xvec[index];
5687 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5690 to->reqid = x->props.reqid;
5691 to->spi = x->id.spi;
5692 to->family = x->props.family;
5695 if (to->family == AF_INET6) {
5696 memcpy(to->remote_ipv6, x->props.saddr.a6,
5697 sizeof(to->remote_ipv6));
5699 to->remote_ipv4 = x->props.saddr.a4;
5700 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5705 memset(to, 0, size);
5709 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5710 .func = bpf_skb_get_xfrm_state,
5712 .ret_type = RET_INTEGER,
5713 .arg1_type = ARG_PTR_TO_CTX,
5714 .arg2_type = ARG_ANYTHING,
5715 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
5716 .arg4_type = ARG_CONST_SIZE,
5717 .arg5_type = ARG_ANYTHING,
5721 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5722 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
5723 const struct neighbour *neigh,
5724 const struct net_device *dev, u32 mtu)
5726 memcpy(params->dmac, neigh->ha, ETH_ALEN);
5727 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5728 params->h_vlan_TCI = 0;
5729 params->h_vlan_proto = 0;
5731 params->mtu_result = mtu; /* union with tot_len */
5737 #if IS_ENABLED(CONFIG_INET)
5738 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5739 u32 flags, bool check_mtu)
5741 struct fib_nh_common *nhc;
5742 struct in_device *in_dev;
5743 struct neighbour *neigh;
5744 struct net_device *dev;
5745 struct fib_result res;
5750 dev = dev_get_by_index_rcu(net, params->ifindex);
5754 /* verify forwarding is enabled on this interface */
5755 in_dev = __in_dev_get_rcu(dev);
5756 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5757 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5759 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5761 fl4.flowi4_oif = params->ifindex;
5763 fl4.flowi4_iif = params->ifindex;
5766 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5767 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5768 fl4.flowi4_flags = 0;
5770 fl4.flowi4_proto = params->l4_protocol;
5771 fl4.daddr = params->ipv4_dst;
5772 fl4.saddr = params->ipv4_src;
5773 fl4.fl4_sport = params->sport;
5774 fl4.fl4_dport = params->dport;
5775 fl4.flowi4_multipath_hash = 0;
5777 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5778 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5779 struct fib_table *tb;
5781 tb = fib_get_table(net, tbid);
5783 return BPF_FIB_LKUP_RET_NOT_FWDED;
5785 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5787 fl4.flowi4_mark = 0;
5788 fl4.flowi4_secid = 0;
5789 fl4.flowi4_tun_key.tun_id = 0;
5790 fl4.flowi4_uid = sock_net_uid(net, NULL);
5792 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5796 /* map fib lookup errors to RTN_ type */
5798 return BPF_FIB_LKUP_RET_BLACKHOLE;
5799 if (err == -EHOSTUNREACH)
5800 return BPF_FIB_LKUP_RET_UNREACHABLE;
5802 return BPF_FIB_LKUP_RET_PROHIBIT;
5804 return BPF_FIB_LKUP_RET_NOT_FWDED;
5807 if (res.type != RTN_UNICAST)
5808 return BPF_FIB_LKUP_RET_NOT_FWDED;
5810 if (fib_info_num_path(res.fi) > 1)
5811 fib_select_path(net, &res, &fl4, NULL);
5814 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5815 if (params->tot_len > mtu) {
5816 params->mtu_result = mtu; /* union with tot_len */
5817 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5823 /* do not handle lwt encaps right now */
5824 if (nhc->nhc_lwtstate)
5825 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5829 params->rt_metric = res.fi->fib_priority;
5830 params->ifindex = dev->ifindex;
5832 /* xdp and cls_bpf programs are run in RCU-bh so
5833 * rcu_read_lock_bh is not needed here
5835 if (likely(nhc->nhc_gw_family != AF_INET6)) {
5836 if (nhc->nhc_gw_family)
5837 params->ipv4_dst = nhc->nhc_gw.ipv4;
5839 neigh = __ipv4_neigh_lookup_noref(dev,
5840 (__force u32)params->ipv4_dst);
5842 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5844 params->family = AF_INET6;
5845 *dst = nhc->nhc_gw.ipv6;
5846 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5850 return BPF_FIB_LKUP_RET_NO_NEIGH;
5852 return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5856 #if IS_ENABLED(CONFIG_IPV6)
5857 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5858 u32 flags, bool check_mtu)
5860 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5861 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5862 struct fib6_result res = {};
5863 struct neighbour *neigh;
5864 struct net_device *dev;
5865 struct inet6_dev *idev;
5871 /* link local addresses are never forwarded */
5872 if (rt6_need_strict(dst) || rt6_need_strict(src))
5873 return BPF_FIB_LKUP_RET_NOT_FWDED;
5875 dev = dev_get_by_index_rcu(net, params->ifindex);
5879 idev = __in6_dev_get_safely(dev);
5880 if (unlikely(!idev || !idev->cnf.forwarding))
5881 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5883 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5885 oif = fl6.flowi6_oif = params->ifindex;
5887 oif = fl6.flowi6_iif = params->ifindex;
5889 strict = RT6_LOOKUP_F_HAS_SADDR;
5891 fl6.flowlabel = params->flowinfo;
5892 fl6.flowi6_scope = 0;
5893 fl6.flowi6_flags = 0;
5896 fl6.flowi6_proto = params->l4_protocol;
5899 fl6.fl6_sport = params->sport;
5900 fl6.fl6_dport = params->dport;
5902 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5903 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5904 struct fib6_table *tb;
5906 tb = ipv6_stub->fib6_get_table(net, tbid);
5908 return BPF_FIB_LKUP_RET_NOT_FWDED;
5910 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5913 fl6.flowi6_mark = 0;
5914 fl6.flowi6_secid = 0;
5915 fl6.flowi6_tun_key.tun_id = 0;
5916 fl6.flowi6_uid = sock_net_uid(net, NULL);
5918 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5921 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5922 res.f6i == net->ipv6.fib6_null_entry))
5923 return BPF_FIB_LKUP_RET_NOT_FWDED;
5925 switch (res.fib6_type) {
5926 /* only unicast is forwarded */
5930 return BPF_FIB_LKUP_RET_BLACKHOLE;
5931 case RTN_UNREACHABLE:
5932 return BPF_FIB_LKUP_RET_UNREACHABLE;
5934 return BPF_FIB_LKUP_RET_PROHIBIT;
5936 return BPF_FIB_LKUP_RET_NOT_FWDED;
5939 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5940 fl6.flowi6_oif != 0, NULL, strict);
5943 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5944 if (params->tot_len > mtu) {
5945 params->mtu_result = mtu; /* union with tot_len */
5946 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5950 if (res.nh->fib_nh_lws)
5951 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5953 if (res.nh->fib_nh_gw_family)
5954 *dst = res.nh->fib_nh_gw6;
5956 dev = res.nh->fib_nh_dev;
5957 params->rt_metric = res.f6i->fib6_metric;
5958 params->ifindex = dev->ifindex;
5960 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
5963 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5965 return BPF_FIB_LKUP_RET_NO_NEIGH;
5967 return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5971 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
5972 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5974 if (plen < sizeof(*params))
5977 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5980 switch (params->family) {
5981 #if IS_ENABLED(CONFIG_INET)
5983 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
5986 #if IS_ENABLED(CONFIG_IPV6)
5988 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
5992 return -EAFNOSUPPORT;
5995 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
5996 .func = bpf_xdp_fib_lookup,
5998 .ret_type = RET_INTEGER,
5999 .arg1_type = ARG_PTR_TO_CTX,
6000 .arg2_type = ARG_PTR_TO_MEM,
6001 .arg3_type = ARG_CONST_SIZE,
6002 .arg4_type = ARG_ANYTHING,
6005 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
6006 struct bpf_fib_lookup *, params, int, plen, u32, flags)
6008 struct net *net = dev_net(skb->dev);
6009 int rc = -EAFNOSUPPORT;
6010 bool check_mtu = false;
6012 if (plen < sizeof(*params))
6015 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
6018 if (params->tot_len)
6021 switch (params->family) {
6022 #if IS_ENABLED(CONFIG_INET)
6024 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
6027 #if IS_ENABLED(CONFIG_IPV6)
6029 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
6034 if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
6035 struct net_device *dev;
6037 /* When tot_len isn't provided by user, check skb
6038 * against MTU of FIB lookup resulting net_device
6040 dev = dev_get_by_index_rcu(net, params->ifindex);
6041 if (!is_skb_forwardable(dev, skb))
6042 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
6044 params->mtu_result = dev->mtu; /* union with tot_len */
6050 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
6051 .func = bpf_skb_fib_lookup,
6053 .ret_type = RET_INTEGER,
6054 .arg1_type = ARG_PTR_TO_CTX,
6055 .arg2_type = ARG_PTR_TO_MEM,
6056 .arg3_type = ARG_CONST_SIZE,
6057 .arg4_type = ARG_ANYTHING,
6060 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
6063 struct net *netns = dev_net(dev_curr);
6065 /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
6069 return dev_get_by_index_rcu(netns, ifindex);
6072 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
6073 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6075 int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6076 struct net_device *dev = skb->dev;
6077 int skb_len, dev_len;
6080 if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
6083 if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
6086 dev = __dev_via_ifindex(dev, ifindex);
6090 mtu = READ_ONCE(dev->mtu);
6092 dev_len = mtu + dev->hard_header_len;
6094 /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6095 skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
6097 skb_len += len_diff; /* minus result pass check */
6098 if (skb_len <= dev_len) {
6099 ret = BPF_MTU_CHK_RET_SUCCESS;
6102 /* At this point, skb->len exceed MTU, but as it include length of all
6103 * segments, it can still be below MTU. The SKB can possibly get
6104 * re-segmented in transmit path (see validate_xmit_skb). Thus, user
6105 * must choose if segs are to be MTU checked.
6107 if (skb_is_gso(skb)) {
6108 ret = BPF_MTU_CHK_RET_SUCCESS;
6110 if (flags & BPF_MTU_CHK_SEGS &&
6111 !skb_gso_validate_network_len(skb, mtu))
6112 ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
6115 /* BPF verifier guarantees valid pointer */
6121 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
6122 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6124 struct net_device *dev = xdp->rxq->dev;
6125 int xdp_len = xdp->data_end - xdp->data;
6126 int ret = BPF_MTU_CHK_RET_SUCCESS;
6129 /* XDP variant doesn't support multi-buffer segment check (yet) */
6130 if (unlikely(flags))
6133 dev = __dev_via_ifindex(dev, ifindex);
6137 mtu = READ_ONCE(dev->mtu);
6139 /* Add L2-header as dev MTU is L3 size */
6140 dev_len = mtu + dev->hard_header_len;
6142 /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6144 xdp_len = *mtu_len + dev->hard_header_len;
6146 xdp_len += len_diff; /* minus result pass check */
6147 if (xdp_len > dev_len)
6148 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6150 /* BPF verifier guarantees valid pointer */
6156 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
6157 .func = bpf_skb_check_mtu,
6159 .ret_type = RET_INTEGER,
6160 .arg1_type = ARG_PTR_TO_CTX,
6161 .arg2_type = ARG_ANYTHING,
6162 .arg3_type = ARG_PTR_TO_INT,
6163 .arg4_type = ARG_ANYTHING,
6164 .arg5_type = ARG_ANYTHING,
6167 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
6168 .func = bpf_xdp_check_mtu,
6170 .ret_type = RET_INTEGER,
6171 .arg1_type = ARG_PTR_TO_CTX,
6172 .arg2_type = ARG_ANYTHING,
6173 .arg3_type = ARG_PTR_TO_INT,
6174 .arg4_type = ARG_ANYTHING,
6175 .arg5_type = ARG_ANYTHING,
6178 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6179 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
6182 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
6184 if (!seg6_validate_srh(srh, len, false))
6188 case BPF_LWT_ENCAP_SEG6_INLINE:
6189 if (skb->protocol != htons(ETH_P_IPV6))
6192 err = seg6_do_srh_inline(skb, srh);
6194 case BPF_LWT_ENCAP_SEG6:
6195 skb_reset_inner_headers(skb);
6196 skb->encapsulation = 1;
6197 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
6203 bpf_compute_data_pointers(skb);
6207 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
6209 return seg6_lookup_nexthop(skb, NULL, 0);
6211 #endif /* CONFIG_IPV6_SEG6_BPF */
6213 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6214 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
6217 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
6221 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
6225 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6226 case BPF_LWT_ENCAP_SEG6:
6227 case BPF_LWT_ENCAP_SEG6_INLINE:
6228 return bpf_push_seg6_encap(skb, type, hdr, len);
6230 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6231 case BPF_LWT_ENCAP_IP:
6232 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
6239 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
6240 void *, hdr, u32, len)
6243 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6244 case BPF_LWT_ENCAP_IP:
6245 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
6252 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
6253 .func = bpf_lwt_in_push_encap,
6255 .ret_type = RET_INTEGER,
6256 .arg1_type = ARG_PTR_TO_CTX,
6257 .arg2_type = ARG_ANYTHING,
6258 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6259 .arg4_type = ARG_CONST_SIZE
6262 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
6263 .func = bpf_lwt_xmit_push_encap,
6265 .ret_type = RET_INTEGER,
6266 .arg1_type = ARG_PTR_TO_CTX,
6267 .arg2_type = ARG_ANYTHING,
6268 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6269 .arg4_type = ARG_CONST_SIZE
6272 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6273 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
6274 const void *, from, u32, len)
6276 struct seg6_bpf_srh_state *srh_state =
6277 this_cpu_ptr(&seg6_bpf_srh_states);
6278 struct ipv6_sr_hdr *srh = srh_state->srh;
6279 void *srh_tlvs, *srh_end, *ptr;
6285 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
6286 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
6288 ptr = skb->data + offset;
6289 if (ptr >= srh_tlvs && ptr + len <= srh_end)
6290 srh_state->valid = false;
6291 else if (ptr < (void *)&srh->flags ||
6292 ptr + len > (void *)&srh->segments)
6295 if (unlikely(bpf_try_make_writable(skb, offset + len)))
6297 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6299 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6301 memcpy(skb->data + offset, from, len);
6305 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
6306 .func = bpf_lwt_seg6_store_bytes,
6308 .ret_type = RET_INTEGER,
6309 .arg1_type = ARG_PTR_TO_CTX,
6310 .arg2_type = ARG_ANYTHING,
6311 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6312 .arg4_type = ARG_CONST_SIZE
6315 static void bpf_update_srh_state(struct sk_buff *skb)
6317 struct seg6_bpf_srh_state *srh_state =
6318 this_cpu_ptr(&seg6_bpf_srh_states);
6321 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
6322 srh_state->srh = NULL;
6324 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6325 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
6326 srh_state->valid = true;
6330 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
6331 u32, action, void *, param, u32, param_len)
6333 struct seg6_bpf_srh_state *srh_state =
6334 this_cpu_ptr(&seg6_bpf_srh_states);
6339 case SEG6_LOCAL_ACTION_END_X:
6340 if (!seg6_bpf_has_valid_srh(skb))
6342 if (param_len != sizeof(struct in6_addr))
6344 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
6345 case SEG6_LOCAL_ACTION_END_T:
6346 if (!seg6_bpf_has_valid_srh(skb))
6348 if (param_len != sizeof(int))
6350 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6351 case SEG6_LOCAL_ACTION_END_DT6:
6352 if (!seg6_bpf_has_valid_srh(skb))
6354 if (param_len != sizeof(int))
6357 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
6359 if (!pskb_pull(skb, hdroff))
6362 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
6363 skb_reset_network_header(skb);
6364 skb_reset_transport_header(skb);
6365 skb->encapsulation = 0;
6367 bpf_compute_data_pointers(skb);
6368 bpf_update_srh_state(skb);
6369 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6370 case SEG6_LOCAL_ACTION_END_B6:
6371 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6373 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
6376 bpf_update_srh_state(skb);
6379 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
6380 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6382 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
6385 bpf_update_srh_state(skb);
6393 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
6394 .func = bpf_lwt_seg6_action,
6396 .ret_type = RET_INTEGER,
6397 .arg1_type = ARG_PTR_TO_CTX,
6398 .arg2_type = ARG_ANYTHING,
6399 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6400 .arg4_type = ARG_CONST_SIZE
6403 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
6406 struct seg6_bpf_srh_state *srh_state =
6407 this_cpu_ptr(&seg6_bpf_srh_states);
6408 struct ipv6_sr_hdr *srh = srh_state->srh;
6409 void *srh_end, *srh_tlvs, *ptr;
6410 struct ipv6hdr *hdr;
6414 if (unlikely(srh == NULL))
6417 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
6418 ((srh->first_segment + 1) << 4));
6419 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
6421 ptr = skb->data + offset;
6423 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
6425 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
6429 ret = skb_cow_head(skb, len);
6430 if (unlikely(ret < 0))
6433 ret = bpf_skb_net_hdr_push(skb, offset, len);
6435 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6438 bpf_compute_data_pointers(skb);
6439 if (unlikely(ret < 0))
6442 hdr = (struct ipv6hdr *)skb->data;
6443 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6445 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6447 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6448 srh_state->hdrlen += len;
6449 srh_state->valid = false;
6453 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6454 .func = bpf_lwt_seg6_adjust_srh,
6456 .ret_type = RET_INTEGER,
6457 .arg1_type = ARG_PTR_TO_CTX,
6458 .arg2_type = ARG_ANYTHING,
6459 .arg3_type = ARG_ANYTHING,
6461 #endif /* CONFIG_IPV6_SEG6_BPF */
6464 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6465 int dif, int sdif, u8 family, u8 proto)
6467 bool refcounted = false;
6468 struct sock *sk = NULL;
6470 if (family == AF_INET) {
6471 __be32 src4 = tuple->ipv4.saddr;
6472 __be32 dst4 = tuple->ipv4.daddr;
6474 if (proto == IPPROTO_TCP)
6475 sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
6476 src4, tuple->ipv4.sport,
6477 dst4, tuple->ipv4.dport,
6478 dif, sdif, &refcounted);
6480 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6481 dst4, tuple->ipv4.dport,
6482 dif, sdif, &udp_table, NULL);
6483 #if IS_ENABLED(CONFIG_IPV6)
6485 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6486 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6488 if (proto == IPPROTO_TCP)
6489 sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
6490 src6, tuple->ipv6.sport,
6491 dst6, ntohs(tuple->ipv6.dport),
6492 dif, sdif, &refcounted);
6493 else if (likely(ipv6_bpf_stub))
6494 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6495 src6, tuple->ipv6.sport,
6496 dst6, tuple->ipv6.dport,
6502 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6503 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6509 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6510 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6512 static struct sock *
6513 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6514 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6517 struct sock *sk = NULL;
6522 if (len == sizeof(tuple->ipv4))
6524 else if (len == sizeof(tuple->ipv6))
6529 if (unlikely(flags || !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6532 if (family == AF_INET)
6533 sdif = inet_sdif(skb);
6535 sdif = inet6_sdif(skb);
6537 if ((s32)netns_id < 0) {
6539 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6541 net = get_net_ns_by_id(caller_net, netns_id);
6544 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6552 static struct sock *
6553 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6554 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6557 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6558 ifindex, proto, netns_id, flags);
6561 struct sock *sk2 = sk_to_full_sk(sk);
6563 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6564 * sock refcnt is decremented to prevent a request_sock leak.
6566 if (!sk_fullsock(sk2))
6570 /* Ensure there is no need to bump sk2 refcnt */
6571 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6572 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6582 static struct sock *
6583 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6584 u8 proto, u64 netns_id, u64 flags)
6586 struct net *caller_net;
6590 caller_net = dev_net(skb->dev);
6591 ifindex = skb->dev->ifindex;
6593 caller_net = sock_net(skb->sk);
6597 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6601 static struct sock *
6602 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6603 u8 proto, u64 netns_id, u64 flags)
6605 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6609 struct sock *sk2 = sk_to_full_sk(sk);
6611 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6612 * sock refcnt is decremented to prevent a request_sock leak.
6614 if (!sk_fullsock(sk2))
6618 /* Ensure there is no need to bump sk2 refcnt */
6619 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6620 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6630 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6631 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6633 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6637 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6638 .func = bpf_skc_lookup_tcp,
6641 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6642 .arg1_type = ARG_PTR_TO_CTX,
6643 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6644 .arg3_type = ARG_CONST_SIZE,
6645 .arg4_type = ARG_ANYTHING,
6646 .arg5_type = ARG_ANYTHING,
6649 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6650 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6652 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6656 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6657 .func = bpf_sk_lookup_tcp,
6660 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6661 .arg1_type = ARG_PTR_TO_CTX,
6662 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6663 .arg3_type = ARG_CONST_SIZE,
6664 .arg4_type = ARG_ANYTHING,
6665 .arg5_type = ARG_ANYTHING,
6668 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6669 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6671 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6675 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6676 .func = bpf_sk_lookup_udp,
6679 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6680 .arg1_type = ARG_PTR_TO_CTX,
6681 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6682 .arg3_type = ARG_CONST_SIZE,
6683 .arg4_type = ARG_ANYTHING,
6684 .arg5_type = ARG_ANYTHING,
6687 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6689 if (sk && sk_is_refcounted(sk))
6694 static const struct bpf_func_proto bpf_sk_release_proto = {
6695 .func = bpf_sk_release,
6697 .ret_type = RET_INTEGER,
6698 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON | OBJ_RELEASE,
6701 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6702 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6704 struct net *caller_net = dev_net(ctx->rxq->dev);
6705 int ifindex = ctx->rxq->dev->ifindex;
6707 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6708 ifindex, IPPROTO_UDP, netns_id,
6712 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6713 .func = bpf_xdp_sk_lookup_udp,
6716 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6717 .arg1_type = ARG_PTR_TO_CTX,
6718 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6719 .arg3_type = ARG_CONST_SIZE,
6720 .arg4_type = ARG_ANYTHING,
6721 .arg5_type = ARG_ANYTHING,
6724 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6725 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6727 struct net *caller_net = dev_net(ctx->rxq->dev);
6728 int ifindex = ctx->rxq->dev->ifindex;
6730 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6731 ifindex, IPPROTO_TCP, netns_id,
6735 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6736 .func = bpf_xdp_skc_lookup_tcp,
6739 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6740 .arg1_type = ARG_PTR_TO_CTX,
6741 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6742 .arg3_type = ARG_CONST_SIZE,
6743 .arg4_type = ARG_ANYTHING,
6744 .arg5_type = ARG_ANYTHING,
6747 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6748 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6750 struct net *caller_net = dev_net(ctx->rxq->dev);
6751 int ifindex = ctx->rxq->dev->ifindex;
6753 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6754 ifindex, IPPROTO_TCP, netns_id,
6758 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6759 .func = bpf_xdp_sk_lookup_tcp,
6762 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6763 .arg1_type = ARG_PTR_TO_CTX,
6764 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6765 .arg3_type = ARG_CONST_SIZE,
6766 .arg4_type = ARG_ANYTHING,
6767 .arg5_type = ARG_ANYTHING,
6770 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6771 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6773 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6774 sock_net(ctx->sk), 0,
6775 IPPROTO_TCP, netns_id, flags);
6778 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6779 .func = bpf_sock_addr_skc_lookup_tcp,
6781 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6782 .arg1_type = ARG_PTR_TO_CTX,
6783 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6784 .arg3_type = ARG_CONST_SIZE,
6785 .arg4_type = ARG_ANYTHING,
6786 .arg5_type = ARG_ANYTHING,
6789 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6790 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6792 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6793 sock_net(ctx->sk), 0, IPPROTO_TCP,
6797 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6798 .func = bpf_sock_addr_sk_lookup_tcp,
6800 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6801 .arg1_type = ARG_PTR_TO_CTX,
6802 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6803 .arg3_type = ARG_CONST_SIZE,
6804 .arg4_type = ARG_ANYTHING,
6805 .arg5_type = ARG_ANYTHING,
6808 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6809 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6811 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6812 sock_net(ctx->sk), 0, IPPROTO_UDP,
6816 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6817 .func = bpf_sock_addr_sk_lookup_udp,
6819 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6820 .arg1_type = ARG_PTR_TO_CTX,
6821 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6822 .arg3_type = ARG_CONST_SIZE,
6823 .arg4_type = ARG_ANYTHING,
6824 .arg5_type = ARG_ANYTHING,
6827 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6828 struct bpf_insn_access_aux *info)
6830 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6834 if (off % size != 0)
6838 case offsetof(struct bpf_tcp_sock, bytes_received):
6839 case offsetof(struct bpf_tcp_sock, bytes_acked):
6840 return size == sizeof(__u64);
6842 return size == sizeof(__u32);
6846 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6847 const struct bpf_insn *si,
6848 struct bpf_insn *insn_buf,
6849 struct bpf_prog *prog, u32 *target_size)
6851 struct bpf_insn *insn = insn_buf;
6853 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \
6855 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \
6856 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6857 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6858 si->dst_reg, si->src_reg, \
6859 offsetof(struct tcp_sock, FIELD)); \
6862 #define BPF_INET_SOCK_GET_COMMON(FIELD) \
6864 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \
6866 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6867 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6868 struct inet_connection_sock, \
6870 si->dst_reg, si->src_reg, \
6872 struct inet_connection_sock, \
6876 if (insn > insn_buf)
6877 return insn - insn_buf;
6880 case offsetof(struct bpf_tcp_sock, rtt_min):
6881 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
6882 sizeof(struct minmax));
6883 BUILD_BUG_ON(sizeof(struct minmax) <
6884 sizeof(struct minmax_sample));
6886 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6887 offsetof(struct tcp_sock, rtt_min) +
6888 offsetof(struct minmax_sample, v));
6890 case offsetof(struct bpf_tcp_sock, snd_cwnd):
6891 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
6893 case offsetof(struct bpf_tcp_sock, srtt_us):
6894 BPF_TCP_SOCK_GET_COMMON(srtt_us);
6896 case offsetof(struct bpf_tcp_sock, snd_ssthresh):
6897 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
6899 case offsetof(struct bpf_tcp_sock, rcv_nxt):
6900 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
6902 case offsetof(struct bpf_tcp_sock, snd_nxt):
6903 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
6905 case offsetof(struct bpf_tcp_sock, snd_una):
6906 BPF_TCP_SOCK_GET_COMMON(snd_una);
6908 case offsetof(struct bpf_tcp_sock, mss_cache):
6909 BPF_TCP_SOCK_GET_COMMON(mss_cache);
6911 case offsetof(struct bpf_tcp_sock, ecn_flags):
6912 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
6914 case offsetof(struct bpf_tcp_sock, rate_delivered):
6915 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
6917 case offsetof(struct bpf_tcp_sock, rate_interval_us):
6918 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
6920 case offsetof(struct bpf_tcp_sock, packets_out):
6921 BPF_TCP_SOCK_GET_COMMON(packets_out);
6923 case offsetof(struct bpf_tcp_sock, retrans_out):
6924 BPF_TCP_SOCK_GET_COMMON(retrans_out);
6926 case offsetof(struct bpf_tcp_sock, total_retrans):
6927 BPF_TCP_SOCK_GET_COMMON(total_retrans);
6929 case offsetof(struct bpf_tcp_sock, segs_in):
6930 BPF_TCP_SOCK_GET_COMMON(segs_in);
6932 case offsetof(struct bpf_tcp_sock, data_segs_in):
6933 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
6935 case offsetof(struct bpf_tcp_sock, segs_out):
6936 BPF_TCP_SOCK_GET_COMMON(segs_out);
6938 case offsetof(struct bpf_tcp_sock, data_segs_out):
6939 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
6941 case offsetof(struct bpf_tcp_sock, lost_out):
6942 BPF_TCP_SOCK_GET_COMMON(lost_out);
6944 case offsetof(struct bpf_tcp_sock, sacked_out):
6945 BPF_TCP_SOCK_GET_COMMON(sacked_out);
6947 case offsetof(struct bpf_tcp_sock, bytes_received):
6948 BPF_TCP_SOCK_GET_COMMON(bytes_received);
6950 case offsetof(struct bpf_tcp_sock, bytes_acked):
6951 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
6953 case offsetof(struct bpf_tcp_sock, dsack_dups):
6954 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
6956 case offsetof(struct bpf_tcp_sock, delivered):
6957 BPF_TCP_SOCK_GET_COMMON(delivered);
6959 case offsetof(struct bpf_tcp_sock, delivered_ce):
6960 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
6962 case offsetof(struct bpf_tcp_sock, icsk_retransmits):
6963 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
6967 return insn - insn_buf;
6970 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
6972 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
6973 return (unsigned long)sk;
6975 return (unsigned long)NULL;
6978 const struct bpf_func_proto bpf_tcp_sock_proto = {
6979 .func = bpf_tcp_sock,
6981 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
6982 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6985 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
6987 sk = sk_to_full_sk(sk);
6989 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
6990 return (unsigned long)sk;
6992 return (unsigned long)NULL;
6995 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
6996 .func = bpf_get_listener_sock,
6998 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6999 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
7002 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
7004 unsigned int iphdr_len;
7006 switch (skb_protocol(skb, true)) {
7007 case cpu_to_be16(ETH_P_IP):
7008 iphdr_len = sizeof(struct iphdr);
7010 case cpu_to_be16(ETH_P_IPV6):
7011 iphdr_len = sizeof(struct ipv6hdr);
7017 if (skb_headlen(skb) < iphdr_len)
7020 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
7023 return INET_ECN_set_ce(skb);
7026 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7027 struct bpf_insn_access_aux *info)
7029 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
7032 if (off % size != 0)
7037 return size == sizeof(__u32);
7041 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
7042 const struct bpf_insn *si,
7043 struct bpf_insn *insn_buf,
7044 struct bpf_prog *prog, u32 *target_size)
7046 struct bpf_insn *insn = insn_buf;
7048 #define BPF_XDP_SOCK_GET(FIELD) \
7050 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \
7051 sizeof_field(struct bpf_xdp_sock, FIELD)); \
7052 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
7053 si->dst_reg, si->src_reg, \
7054 offsetof(struct xdp_sock, FIELD)); \
7058 case offsetof(struct bpf_xdp_sock, queue_id):
7059 BPF_XDP_SOCK_GET(queue_id);
7063 return insn - insn_buf;
7066 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
7067 .func = bpf_skb_ecn_set_ce,
7069 .ret_type = RET_INTEGER,
7070 .arg1_type = ARG_PTR_TO_CTX,
7073 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7074 struct tcphdr *, th, u32, th_len)
7076 #ifdef CONFIG_SYN_COOKIES
7080 if (unlikely(!sk || th_len < sizeof(*th)))
7083 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
7084 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7087 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
7090 if (!th->ack || th->rst || th->syn)
7093 if (unlikely(iph_len < sizeof(struct iphdr)))
7096 if (tcp_synq_no_recent_overflow(sk))
7099 cookie = ntohl(th->ack_seq) - 1;
7101 /* Both struct iphdr and struct ipv6hdr have the version field at the
7102 * same offset so we can cast to the shorter header (struct iphdr).
7104 switch (((struct iphdr *)iph)->version) {
7106 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7109 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
7112 #if IS_BUILTIN(CONFIG_IPV6)
7114 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7117 if (sk->sk_family != AF_INET6)
7120 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
7122 #endif /* CONFIG_IPV6 */
7125 return -EPROTONOSUPPORT;
7137 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
7138 .func = bpf_tcp_check_syncookie,
7141 .ret_type = RET_INTEGER,
7142 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7143 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7144 .arg3_type = ARG_CONST_SIZE,
7145 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7146 .arg5_type = ARG_CONST_SIZE,
7149 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7150 struct tcphdr *, th, u32, th_len)
7152 #ifdef CONFIG_SYN_COOKIES
7156 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
7159 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7162 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
7165 if (!th->syn || th->ack || th->fin || th->rst)
7168 if (unlikely(iph_len < sizeof(struct iphdr)))
7171 /* Both struct iphdr and struct ipv6hdr have the version field at the
7172 * same offset so we can cast to the shorter header (struct iphdr).
7174 switch (((struct iphdr *)iph)->version) {
7176 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7179 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
7182 #if IS_BUILTIN(CONFIG_IPV6)
7184 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7187 if (sk->sk_family != AF_INET6)
7190 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
7192 #endif /* CONFIG_IPV6 */
7195 return -EPROTONOSUPPORT;
7200 return cookie | ((u64)mss << 32);
7203 #endif /* CONFIG_SYN_COOKIES */
7206 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
7207 .func = bpf_tcp_gen_syncookie,
7208 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */
7210 .ret_type = RET_INTEGER,
7211 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7212 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7213 .arg3_type = ARG_CONST_SIZE,
7214 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7215 .arg5_type = ARG_CONST_SIZE,
7218 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
7220 if (!sk || flags != 0)
7222 if (!skb_at_tc_ingress(skb))
7224 if (unlikely(dev_net(skb->dev) != sock_net(sk)))
7225 return -ENETUNREACH;
7226 if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
7227 return -ESOCKTNOSUPPORT;
7228 if (sk_is_refcounted(sk) &&
7229 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
7234 skb->destructor = sock_pfree;
7239 static const struct bpf_func_proto bpf_sk_assign_proto = {
7240 .func = bpf_sk_assign,
7242 .ret_type = RET_INTEGER,
7243 .arg1_type = ARG_PTR_TO_CTX,
7244 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7245 .arg3_type = ARG_ANYTHING,
7248 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
7249 u8 search_kind, const u8 *magic,
7250 u8 magic_len, bool *eol)
7256 while (op < opend) {
7259 if (kind == TCPOPT_EOL) {
7261 return ERR_PTR(-ENOMSG);
7262 } else if (kind == TCPOPT_NOP) {
7267 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
7268 /* Something is wrong in the received header.
7269 * Follow the TCP stack's tcp_parse_options()
7270 * and just bail here.
7272 return ERR_PTR(-EFAULT);
7275 if (search_kind == kind) {
7279 if (magic_len > kind_len - 2)
7280 return ERR_PTR(-ENOMSG);
7282 if (!memcmp(&op[2], magic, magic_len))
7289 return ERR_PTR(-ENOMSG);
7292 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7293 void *, search_res, u32, len, u64, flags)
7295 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
7296 const u8 *op, *opend, *magic, *search = search_res;
7297 u8 search_kind, search_len, copy_len, magic_len;
7300 /* 2 byte is the minimal option len except TCPOPT_NOP and
7301 * TCPOPT_EOL which are useless for the bpf prog to learn
7302 * and this helper disallow loading them also.
7304 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
7307 search_kind = search[0];
7308 search_len = search[1];
7310 if (search_len > len || search_kind == TCPOPT_NOP ||
7311 search_kind == TCPOPT_EOL)
7314 if (search_kind == TCPOPT_EXP || search_kind == 253) {
7315 /* 16 or 32 bit magic. +2 for kind and kind length */
7316 if (search_len != 4 && search_len != 6)
7319 magic_len = search_len - 2;
7328 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
7333 op += sizeof(struct tcphdr);
7335 if (!bpf_sock->skb ||
7336 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7337 /* This bpf_sock->op cannot call this helper */
7340 opend = bpf_sock->skb_data_end;
7341 op = bpf_sock->skb->data + sizeof(struct tcphdr);
7344 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
7351 if (copy_len > len) {
7356 memcpy(search_res, op, copy_len);
7360 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
7361 .func = bpf_sock_ops_load_hdr_opt,
7363 .ret_type = RET_INTEGER,
7364 .arg1_type = ARG_PTR_TO_CTX,
7365 .arg2_type = ARG_PTR_TO_MEM,
7366 .arg3_type = ARG_CONST_SIZE,
7367 .arg4_type = ARG_ANYTHING,
7370 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7371 const void *, from, u32, len, u64, flags)
7373 u8 new_kind, new_kind_len, magic_len = 0, *opend;
7374 const u8 *op, *new_op, *magic = NULL;
7375 struct sk_buff *skb;
7378 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
7381 if (len < 2 || flags)
7385 new_kind = new_op[0];
7386 new_kind_len = new_op[1];
7388 if (new_kind_len > len || new_kind == TCPOPT_NOP ||
7389 new_kind == TCPOPT_EOL)
7392 if (new_kind_len > bpf_sock->remaining_opt_len)
7395 /* 253 is another experimental kind */
7396 if (new_kind == TCPOPT_EXP || new_kind == 253) {
7397 if (new_kind_len < 4)
7399 /* Match for the 2 byte magic also.
7400 * RFC 6994: the magic could be 2 or 4 bytes.
7401 * Hence, matching by 2 byte only is on the
7402 * conservative side but it is the right
7403 * thing to do for the 'search-for-duplication'
7410 /* Check for duplication */
7411 skb = bpf_sock->skb;
7412 op = skb->data + sizeof(struct tcphdr);
7413 opend = bpf_sock->skb_data_end;
7415 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
7420 if (PTR_ERR(op) != -ENOMSG)
7424 /* The option has been ended. Treat it as no more
7425 * header option can be written.
7429 /* No duplication found. Store the header option. */
7430 memcpy(opend, from, new_kind_len);
7432 bpf_sock->remaining_opt_len -= new_kind_len;
7433 bpf_sock->skb_data_end += new_kind_len;
7438 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
7439 .func = bpf_sock_ops_store_hdr_opt,
7441 .ret_type = RET_INTEGER,
7442 .arg1_type = ARG_PTR_TO_CTX,
7443 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7444 .arg3_type = ARG_CONST_SIZE,
7445 .arg4_type = ARG_ANYTHING,
7448 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7449 u32, len, u64, flags)
7451 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7454 if (flags || len < 2)
7457 if (len > bpf_sock->remaining_opt_len)
7460 bpf_sock->remaining_opt_len -= len;
7465 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7466 .func = bpf_sock_ops_reserve_hdr_opt,
7468 .ret_type = RET_INTEGER,
7469 .arg1_type = ARG_PTR_TO_CTX,
7470 .arg2_type = ARG_ANYTHING,
7471 .arg3_type = ARG_ANYTHING,
7474 BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb,
7475 u64, tstamp, u32, tstamp_type)
7477 /* skb_clear_delivery_time() is done for inet protocol */
7478 if (skb->protocol != htons(ETH_P_IP) &&
7479 skb->protocol != htons(ETH_P_IPV6))
7482 switch (tstamp_type) {
7483 case BPF_SKB_TSTAMP_DELIVERY_MONO:
7486 skb->tstamp = tstamp;
7487 skb->mono_delivery_time = 1;
7489 case BPF_SKB_TSTAMP_UNSPEC:
7493 skb->mono_delivery_time = 0;
7502 static const struct bpf_func_proto bpf_skb_set_tstamp_proto = {
7503 .func = bpf_skb_set_tstamp,
7505 .ret_type = RET_INTEGER,
7506 .arg1_type = ARG_PTR_TO_CTX,
7507 .arg2_type = ARG_ANYTHING,
7508 .arg3_type = ARG_ANYTHING,
7511 #ifdef CONFIG_SYN_COOKIES
7512 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv4, struct iphdr *, iph,
7513 struct tcphdr *, th, u32, th_len)
7518 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7521 mss = tcp_parse_mss_option(th, 0) ?: TCP_MSS_DEFAULT;
7522 cookie = __cookie_v4_init_sequence(iph, th, &mss);
7524 return cookie | ((u64)mss << 32);
7527 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv4_proto = {
7528 .func = bpf_tcp_raw_gen_syncookie_ipv4,
7529 .gpl_only = true, /* __cookie_v4_init_sequence() is GPL */
7531 .ret_type = RET_INTEGER,
7532 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7533 .arg1_size = sizeof(struct iphdr),
7534 .arg2_type = ARG_PTR_TO_MEM,
7535 .arg3_type = ARG_CONST_SIZE,
7538 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv6, struct ipv6hdr *, iph,
7539 struct tcphdr *, th, u32, th_len)
7541 #if IS_BUILTIN(CONFIG_IPV6)
7542 const u16 mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
7543 sizeof(struct ipv6hdr);
7547 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7550 mss = tcp_parse_mss_option(th, 0) ?: mss_clamp;
7551 cookie = __cookie_v6_init_sequence(iph, th, &mss);
7553 return cookie | ((u64)mss << 32);
7555 return -EPROTONOSUPPORT;
7559 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv6_proto = {
7560 .func = bpf_tcp_raw_gen_syncookie_ipv6,
7561 .gpl_only = true, /* __cookie_v6_init_sequence() is GPL */
7563 .ret_type = RET_INTEGER,
7564 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7565 .arg1_size = sizeof(struct ipv6hdr),
7566 .arg2_type = ARG_PTR_TO_MEM,
7567 .arg3_type = ARG_CONST_SIZE,
7570 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv4, struct iphdr *, iph,
7571 struct tcphdr *, th)
7573 u32 cookie = ntohl(th->ack_seq) - 1;
7575 if (__cookie_v4_check(iph, th, cookie) > 0)
7581 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv4_proto = {
7582 .func = bpf_tcp_raw_check_syncookie_ipv4,
7583 .gpl_only = true, /* __cookie_v4_check is GPL */
7585 .ret_type = RET_INTEGER,
7586 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7587 .arg1_size = sizeof(struct iphdr),
7588 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7589 .arg2_size = sizeof(struct tcphdr),
7592 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv6, struct ipv6hdr *, iph,
7593 struct tcphdr *, th)
7595 #if IS_BUILTIN(CONFIG_IPV6)
7596 u32 cookie = ntohl(th->ack_seq) - 1;
7598 if (__cookie_v6_check(iph, th, cookie) > 0)
7603 return -EPROTONOSUPPORT;
7607 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv6_proto = {
7608 .func = bpf_tcp_raw_check_syncookie_ipv6,
7609 .gpl_only = true, /* __cookie_v6_check is GPL */
7611 .ret_type = RET_INTEGER,
7612 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7613 .arg1_size = sizeof(struct ipv6hdr),
7614 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7615 .arg2_size = sizeof(struct tcphdr),
7617 #endif /* CONFIG_SYN_COOKIES */
7619 #endif /* CONFIG_INET */
7621 bool bpf_helper_changes_pkt_data(void *func)
7623 if (func == bpf_skb_vlan_push ||
7624 func == bpf_skb_vlan_pop ||
7625 func == bpf_skb_store_bytes ||
7626 func == bpf_skb_change_proto ||
7627 func == bpf_skb_change_head ||
7628 func == sk_skb_change_head ||
7629 func == bpf_skb_change_tail ||
7630 func == sk_skb_change_tail ||
7631 func == bpf_skb_adjust_room ||
7632 func == sk_skb_adjust_room ||
7633 func == bpf_skb_pull_data ||
7634 func == sk_skb_pull_data ||
7635 func == bpf_clone_redirect ||
7636 func == bpf_l3_csum_replace ||
7637 func == bpf_l4_csum_replace ||
7638 func == bpf_xdp_adjust_head ||
7639 func == bpf_xdp_adjust_meta ||
7640 func == bpf_msg_pull_data ||
7641 func == bpf_msg_push_data ||
7642 func == bpf_msg_pop_data ||
7643 func == bpf_xdp_adjust_tail ||
7644 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7645 func == bpf_lwt_seg6_store_bytes ||
7646 func == bpf_lwt_seg6_adjust_srh ||
7647 func == bpf_lwt_seg6_action ||
7650 func == bpf_sock_ops_store_hdr_opt ||
7652 func == bpf_lwt_in_push_encap ||
7653 func == bpf_lwt_xmit_push_encap)
7659 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7660 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7662 static const struct bpf_func_proto *
7663 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7666 /* inet and inet6 sockets are created in a process
7667 * context so there is always a valid uid/gid
7669 case BPF_FUNC_get_current_uid_gid:
7670 return &bpf_get_current_uid_gid_proto;
7671 case BPF_FUNC_get_local_storage:
7672 return &bpf_get_local_storage_proto;
7673 case BPF_FUNC_get_socket_cookie:
7674 return &bpf_get_socket_cookie_sock_proto;
7675 case BPF_FUNC_get_netns_cookie:
7676 return &bpf_get_netns_cookie_sock_proto;
7677 case BPF_FUNC_perf_event_output:
7678 return &bpf_event_output_data_proto;
7679 case BPF_FUNC_get_current_pid_tgid:
7680 return &bpf_get_current_pid_tgid_proto;
7681 case BPF_FUNC_get_current_comm:
7682 return &bpf_get_current_comm_proto;
7683 #ifdef CONFIG_CGROUPS
7684 case BPF_FUNC_get_current_cgroup_id:
7685 return &bpf_get_current_cgroup_id_proto;
7686 case BPF_FUNC_get_current_ancestor_cgroup_id:
7687 return &bpf_get_current_ancestor_cgroup_id_proto;
7689 #ifdef CONFIG_CGROUP_NET_CLASSID
7690 case BPF_FUNC_get_cgroup_classid:
7691 return &bpf_get_cgroup_classid_curr_proto;
7693 case BPF_FUNC_sk_storage_get:
7694 return &bpf_sk_storage_get_cg_sock_proto;
7695 case BPF_FUNC_ktime_get_coarse_ns:
7696 return &bpf_ktime_get_coarse_ns_proto;
7698 return bpf_base_func_proto(func_id);
7702 static const struct bpf_func_proto *
7703 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7706 /* inet and inet6 sockets are created in a process
7707 * context so there is always a valid uid/gid
7709 case BPF_FUNC_get_current_uid_gid:
7710 return &bpf_get_current_uid_gid_proto;
7712 switch (prog->expected_attach_type) {
7713 case BPF_CGROUP_INET4_CONNECT:
7714 case BPF_CGROUP_INET6_CONNECT:
7715 return &bpf_bind_proto;
7719 case BPF_FUNC_get_socket_cookie:
7720 return &bpf_get_socket_cookie_sock_addr_proto;
7721 case BPF_FUNC_get_netns_cookie:
7722 return &bpf_get_netns_cookie_sock_addr_proto;
7723 case BPF_FUNC_get_local_storage:
7724 return &bpf_get_local_storage_proto;
7725 case BPF_FUNC_perf_event_output:
7726 return &bpf_event_output_data_proto;
7727 case BPF_FUNC_get_current_pid_tgid:
7728 return &bpf_get_current_pid_tgid_proto;
7729 case BPF_FUNC_get_current_comm:
7730 return &bpf_get_current_comm_proto;
7731 #ifdef CONFIG_CGROUPS
7732 case BPF_FUNC_get_current_cgroup_id:
7733 return &bpf_get_current_cgroup_id_proto;
7734 case BPF_FUNC_get_current_ancestor_cgroup_id:
7735 return &bpf_get_current_ancestor_cgroup_id_proto;
7737 #ifdef CONFIG_CGROUP_NET_CLASSID
7738 case BPF_FUNC_get_cgroup_classid:
7739 return &bpf_get_cgroup_classid_curr_proto;
7742 case BPF_FUNC_sk_lookup_tcp:
7743 return &bpf_sock_addr_sk_lookup_tcp_proto;
7744 case BPF_FUNC_sk_lookup_udp:
7745 return &bpf_sock_addr_sk_lookup_udp_proto;
7746 case BPF_FUNC_sk_release:
7747 return &bpf_sk_release_proto;
7748 case BPF_FUNC_skc_lookup_tcp:
7749 return &bpf_sock_addr_skc_lookup_tcp_proto;
7750 #endif /* CONFIG_INET */
7751 case BPF_FUNC_sk_storage_get:
7752 return &bpf_sk_storage_get_proto;
7753 case BPF_FUNC_sk_storage_delete:
7754 return &bpf_sk_storage_delete_proto;
7755 case BPF_FUNC_setsockopt:
7756 switch (prog->expected_attach_type) {
7757 case BPF_CGROUP_INET4_BIND:
7758 case BPF_CGROUP_INET6_BIND:
7759 case BPF_CGROUP_INET4_CONNECT:
7760 case BPF_CGROUP_INET6_CONNECT:
7761 case BPF_CGROUP_UDP4_RECVMSG:
7762 case BPF_CGROUP_UDP6_RECVMSG:
7763 case BPF_CGROUP_UDP4_SENDMSG:
7764 case BPF_CGROUP_UDP6_SENDMSG:
7765 case BPF_CGROUP_INET4_GETPEERNAME:
7766 case BPF_CGROUP_INET6_GETPEERNAME:
7767 case BPF_CGROUP_INET4_GETSOCKNAME:
7768 case BPF_CGROUP_INET6_GETSOCKNAME:
7769 return &bpf_sock_addr_setsockopt_proto;
7773 case BPF_FUNC_getsockopt:
7774 switch (prog->expected_attach_type) {
7775 case BPF_CGROUP_INET4_BIND:
7776 case BPF_CGROUP_INET6_BIND:
7777 case BPF_CGROUP_INET4_CONNECT:
7778 case BPF_CGROUP_INET6_CONNECT:
7779 case BPF_CGROUP_UDP4_RECVMSG:
7780 case BPF_CGROUP_UDP6_RECVMSG:
7781 case BPF_CGROUP_UDP4_SENDMSG:
7782 case BPF_CGROUP_UDP6_SENDMSG:
7783 case BPF_CGROUP_INET4_GETPEERNAME:
7784 case BPF_CGROUP_INET6_GETPEERNAME:
7785 case BPF_CGROUP_INET4_GETSOCKNAME:
7786 case BPF_CGROUP_INET6_GETSOCKNAME:
7787 return &bpf_sock_addr_getsockopt_proto;
7792 return bpf_sk_base_func_proto(func_id);
7796 static const struct bpf_func_proto *
7797 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7800 case BPF_FUNC_skb_load_bytes:
7801 return &bpf_skb_load_bytes_proto;
7802 case BPF_FUNC_skb_load_bytes_relative:
7803 return &bpf_skb_load_bytes_relative_proto;
7804 case BPF_FUNC_get_socket_cookie:
7805 return &bpf_get_socket_cookie_proto;
7806 case BPF_FUNC_get_socket_uid:
7807 return &bpf_get_socket_uid_proto;
7808 case BPF_FUNC_perf_event_output:
7809 return &bpf_skb_event_output_proto;
7811 return bpf_sk_base_func_proto(func_id);
7815 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7816 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7818 static const struct bpf_func_proto *
7819 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7822 case BPF_FUNC_get_local_storage:
7823 return &bpf_get_local_storage_proto;
7824 case BPF_FUNC_sk_fullsock:
7825 return &bpf_sk_fullsock_proto;
7826 case BPF_FUNC_sk_storage_get:
7827 return &bpf_sk_storage_get_proto;
7828 case BPF_FUNC_sk_storage_delete:
7829 return &bpf_sk_storage_delete_proto;
7830 case BPF_FUNC_perf_event_output:
7831 return &bpf_skb_event_output_proto;
7832 #ifdef CONFIG_SOCK_CGROUP_DATA
7833 case BPF_FUNC_skb_cgroup_id:
7834 return &bpf_skb_cgroup_id_proto;
7835 case BPF_FUNC_skb_ancestor_cgroup_id:
7836 return &bpf_skb_ancestor_cgroup_id_proto;
7837 case BPF_FUNC_sk_cgroup_id:
7838 return &bpf_sk_cgroup_id_proto;
7839 case BPF_FUNC_sk_ancestor_cgroup_id:
7840 return &bpf_sk_ancestor_cgroup_id_proto;
7843 case BPF_FUNC_sk_lookup_tcp:
7844 return &bpf_sk_lookup_tcp_proto;
7845 case BPF_FUNC_sk_lookup_udp:
7846 return &bpf_sk_lookup_udp_proto;
7847 case BPF_FUNC_sk_release:
7848 return &bpf_sk_release_proto;
7849 case BPF_FUNC_skc_lookup_tcp:
7850 return &bpf_skc_lookup_tcp_proto;
7851 case BPF_FUNC_tcp_sock:
7852 return &bpf_tcp_sock_proto;
7853 case BPF_FUNC_get_listener_sock:
7854 return &bpf_get_listener_sock_proto;
7855 case BPF_FUNC_skb_ecn_set_ce:
7856 return &bpf_skb_ecn_set_ce_proto;
7859 return sk_filter_func_proto(func_id, prog);
7863 static const struct bpf_func_proto *
7864 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7867 case BPF_FUNC_skb_store_bytes:
7868 return &bpf_skb_store_bytes_proto;
7869 case BPF_FUNC_skb_load_bytes:
7870 return &bpf_skb_load_bytes_proto;
7871 case BPF_FUNC_skb_load_bytes_relative:
7872 return &bpf_skb_load_bytes_relative_proto;
7873 case BPF_FUNC_skb_pull_data:
7874 return &bpf_skb_pull_data_proto;
7875 case BPF_FUNC_csum_diff:
7876 return &bpf_csum_diff_proto;
7877 case BPF_FUNC_csum_update:
7878 return &bpf_csum_update_proto;
7879 case BPF_FUNC_csum_level:
7880 return &bpf_csum_level_proto;
7881 case BPF_FUNC_l3_csum_replace:
7882 return &bpf_l3_csum_replace_proto;
7883 case BPF_FUNC_l4_csum_replace:
7884 return &bpf_l4_csum_replace_proto;
7885 case BPF_FUNC_clone_redirect:
7886 return &bpf_clone_redirect_proto;
7887 case BPF_FUNC_get_cgroup_classid:
7888 return &bpf_get_cgroup_classid_proto;
7889 case BPF_FUNC_skb_vlan_push:
7890 return &bpf_skb_vlan_push_proto;
7891 case BPF_FUNC_skb_vlan_pop:
7892 return &bpf_skb_vlan_pop_proto;
7893 case BPF_FUNC_skb_change_proto:
7894 return &bpf_skb_change_proto_proto;
7895 case BPF_FUNC_skb_change_type:
7896 return &bpf_skb_change_type_proto;
7897 case BPF_FUNC_skb_adjust_room:
7898 return &bpf_skb_adjust_room_proto;
7899 case BPF_FUNC_skb_change_tail:
7900 return &bpf_skb_change_tail_proto;
7901 case BPF_FUNC_skb_change_head:
7902 return &bpf_skb_change_head_proto;
7903 case BPF_FUNC_skb_get_tunnel_key:
7904 return &bpf_skb_get_tunnel_key_proto;
7905 case BPF_FUNC_skb_set_tunnel_key:
7906 return bpf_get_skb_set_tunnel_proto(func_id);
7907 case BPF_FUNC_skb_get_tunnel_opt:
7908 return &bpf_skb_get_tunnel_opt_proto;
7909 case BPF_FUNC_skb_set_tunnel_opt:
7910 return bpf_get_skb_set_tunnel_proto(func_id);
7911 case BPF_FUNC_redirect:
7912 return &bpf_redirect_proto;
7913 case BPF_FUNC_redirect_neigh:
7914 return &bpf_redirect_neigh_proto;
7915 case BPF_FUNC_redirect_peer:
7916 return &bpf_redirect_peer_proto;
7917 case BPF_FUNC_get_route_realm:
7918 return &bpf_get_route_realm_proto;
7919 case BPF_FUNC_get_hash_recalc:
7920 return &bpf_get_hash_recalc_proto;
7921 case BPF_FUNC_set_hash_invalid:
7922 return &bpf_set_hash_invalid_proto;
7923 case BPF_FUNC_set_hash:
7924 return &bpf_set_hash_proto;
7925 case BPF_FUNC_perf_event_output:
7926 return &bpf_skb_event_output_proto;
7927 case BPF_FUNC_get_smp_processor_id:
7928 return &bpf_get_smp_processor_id_proto;
7929 case BPF_FUNC_skb_under_cgroup:
7930 return &bpf_skb_under_cgroup_proto;
7931 case BPF_FUNC_get_socket_cookie:
7932 return &bpf_get_socket_cookie_proto;
7933 case BPF_FUNC_get_socket_uid:
7934 return &bpf_get_socket_uid_proto;
7935 case BPF_FUNC_fib_lookup:
7936 return &bpf_skb_fib_lookup_proto;
7937 case BPF_FUNC_check_mtu:
7938 return &bpf_skb_check_mtu_proto;
7939 case BPF_FUNC_sk_fullsock:
7940 return &bpf_sk_fullsock_proto;
7941 case BPF_FUNC_sk_storage_get:
7942 return &bpf_sk_storage_get_proto;
7943 case BPF_FUNC_sk_storage_delete:
7944 return &bpf_sk_storage_delete_proto;
7946 case BPF_FUNC_skb_get_xfrm_state:
7947 return &bpf_skb_get_xfrm_state_proto;
7949 #ifdef CONFIG_CGROUP_NET_CLASSID
7950 case BPF_FUNC_skb_cgroup_classid:
7951 return &bpf_skb_cgroup_classid_proto;
7953 #ifdef CONFIG_SOCK_CGROUP_DATA
7954 case BPF_FUNC_skb_cgroup_id:
7955 return &bpf_skb_cgroup_id_proto;
7956 case BPF_FUNC_skb_ancestor_cgroup_id:
7957 return &bpf_skb_ancestor_cgroup_id_proto;
7960 case BPF_FUNC_sk_lookup_tcp:
7961 return &bpf_sk_lookup_tcp_proto;
7962 case BPF_FUNC_sk_lookup_udp:
7963 return &bpf_sk_lookup_udp_proto;
7964 case BPF_FUNC_sk_release:
7965 return &bpf_sk_release_proto;
7966 case BPF_FUNC_tcp_sock:
7967 return &bpf_tcp_sock_proto;
7968 case BPF_FUNC_get_listener_sock:
7969 return &bpf_get_listener_sock_proto;
7970 case BPF_FUNC_skc_lookup_tcp:
7971 return &bpf_skc_lookup_tcp_proto;
7972 case BPF_FUNC_tcp_check_syncookie:
7973 return &bpf_tcp_check_syncookie_proto;
7974 case BPF_FUNC_skb_ecn_set_ce:
7975 return &bpf_skb_ecn_set_ce_proto;
7976 case BPF_FUNC_tcp_gen_syncookie:
7977 return &bpf_tcp_gen_syncookie_proto;
7978 case BPF_FUNC_sk_assign:
7979 return &bpf_sk_assign_proto;
7980 case BPF_FUNC_skb_set_tstamp:
7981 return &bpf_skb_set_tstamp_proto;
7982 #ifdef CONFIG_SYN_COOKIES
7983 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
7984 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
7985 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
7986 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
7987 case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
7988 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
7989 case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
7990 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
7994 return bpf_sk_base_func_proto(func_id);
7998 static const struct bpf_func_proto *
7999 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8002 case BPF_FUNC_perf_event_output:
8003 return &bpf_xdp_event_output_proto;
8004 case BPF_FUNC_get_smp_processor_id:
8005 return &bpf_get_smp_processor_id_proto;
8006 case BPF_FUNC_csum_diff:
8007 return &bpf_csum_diff_proto;
8008 case BPF_FUNC_xdp_adjust_head:
8009 return &bpf_xdp_adjust_head_proto;
8010 case BPF_FUNC_xdp_adjust_meta:
8011 return &bpf_xdp_adjust_meta_proto;
8012 case BPF_FUNC_redirect:
8013 return &bpf_xdp_redirect_proto;
8014 case BPF_FUNC_redirect_map:
8015 return &bpf_xdp_redirect_map_proto;
8016 case BPF_FUNC_xdp_adjust_tail:
8017 return &bpf_xdp_adjust_tail_proto;
8018 case BPF_FUNC_xdp_get_buff_len:
8019 return &bpf_xdp_get_buff_len_proto;
8020 case BPF_FUNC_xdp_load_bytes:
8021 return &bpf_xdp_load_bytes_proto;
8022 case BPF_FUNC_xdp_store_bytes:
8023 return &bpf_xdp_store_bytes_proto;
8024 case BPF_FUNC_fib_lookup:
8025 return &bpf_xdp_fib_lookup_proto;
8026 case BPF_FUNC_check_mtu:
8027 return &bpf_xdp_check_mtu_proto;
8029 case BPF_FUNC_sk_lookup_udp:
8030 return &bpf_xdp_sk_lookup_udp_proto;
8031 case BPF_FUNC_sk_lookup_tcp:
8032 return &bpf_xdp_sk_lookup_tcp_proto;
8033 case BPF_FUNC_sk_release:
8034 return &bpf_sk_release_proto;
8035 case BPF_FUNC_skc_lookup_tcp:
8036 return &bpf_xdp_skc_lookup_tcp_proto;
8037 case BPF_FUNC_tcp_check_syncookie:
8038 return &bpf_tcp_check_syncookie_proto;
8039 case BPF_FUNC_tcp_gen_syncookie:
8040 return &bpf_tcp_gen_syncookie_proto;
8041 #ifdef CONFIG_SYN_COOKIES
8042 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8043 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8044 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8045 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8046 case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8047 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8048 case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8049 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8053 return bpf_sk_base_func_proto(func_id);
8057 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
8058 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
8060 static const struct bpf_func_proto *
8061 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8064 case BPF_FUNC_setsockopt:
8065 return &bpf_sock_ops_setsockopt_proto;
8066 case BPF_FUNC_getsockopt:
8067 return &bpf_sock_ops_getsockopt_proto;
8068 case BPF_FUNC_sock_ops_cb_flags_set:
8069 return &bpf_sock_ops_cb_flags_set_proto;
8070 case BPF_FUNC_sock_map_update:
8071 return &bpf_sock_map_update_proto;
8072 case BPF_FUNC_sock_hash_update:
8073 return &bpf_sock_hash_update_proto;
8074 case BPF_FUNC_get_socket_cookie:
8075 return &bpf_get_socket_cookie_sock_ops_proto;
8076 case BPF_FUNC_get_local_storage:
8077 return &bpf_get_local_storage_proto;
8078 case BPF_FUNC_perf_event_output:
8079 return &bpf_event_output_data_proto;
8080 case BPF_FUNC_sk_storage_get:
8081 return &bpf_sk_storage_get_proto;
8082 case BPF_FUNC_sk_storage_delete:
8083 return &bpf_sk_storage_delete_proto;
8084 case BPF_FUNC_get_netns_cookie:
8085 return &bpf_get_netns_cookie_sock_ops_proto;
8087 case BPF_FUNC_load_hdr_opt:
8088 return &bpf_sock_ops_load_hdr_opt_proto;
8089 case BPF_FUNC_store_hdr_opt:
8090 return &bpf_sock_ops_store_hdr_opt_proto;
8091 case BPF_FUNC_reserve_hdr_opt:
8092 return &bpf_sock_ops_reserve_hdr_opt_proto;
8093 case BPF_FUNC_tcp_sock:
8094 return &bpf_tcp_sock_proto;
8095 #endif /* CONFIG_INET */
8097 return bpf_sk_base_func_proto(func_id);
8101 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
8102 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
8104 static const struct bpf_func_proto *
8105 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8108 case BPF_FUNC_msg_redirect_map:
8109 return &bpf_msg_redirect_map_proto;
8110 case BPF_FUNC_msg_redirect_hash:
8111 return &bpf_msg_redirect_hash_proto;
8112 case BPF_FUNC_msg_apply_bytes:
8113 return &bpf_msg_apply_bytes_proto;
8114 case BPF_FUNC_msg_cork_bytes:
8115 return &bpf_msg_cork_bytes_proto;
8116 case BPF_FUNC_msg_pull_data:
8117 return &bpf_msg_pull_data_proto;
8118 case BPF_FUNC_msg_push_data:
8119 return &bpf_msg_push_data_proto;
8120 case BPF_FUNC_msg_pop_data:
8121 return &bpf_msg_pop_data_proto;
8122 case BPF_FUNC_perf_event_output:
8123 return &bpf_event_output_data_proto;
8124 case BPF_FUNC_get_current_uid_gid:
8125 return &bpf_get_current_uid_gid_proto;
8126 case BPF_FUNC_get_current_pid_tgid:
8127 return &bpf_get_current_pid_tgid_proto;
8128 case BPF_FUNC_sk_storage_get:
8129 return &bpf_sk_storage_get_proto;
8130 case BPF_FUNC_sk_storage_delete:
8131 return &bpf_sk_storage_delete_proto;
8132 case BPF_FUNC_get_netns_cookie:
8133 return &bpf_get_netns_cookie_sk_msg_proto;
8134 #ifdef CONFIG_CGROUPS
8135 case BPF_FUNC_get_current_cgroup_id:
8136 return &bpf_get_current_cgroup_id_proto;
8137 case BPF_FUNC_get_current_ancestor_cgroup_id:
8138 return &bpf_get_current_ancestor_cgroup_id_proto;
8140 #ifdef CONFIG_CGROUP_NET_CLASSID
8141 case BPF_FUNC_get_cgroup_classid:
8142 return &bpf_get_cgroup_classid_curr_proto;
8145 return bpf_sk_base_func_proto(func_id);
8149 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
8150 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
8152 static const struct bpf_func_proto *
8153 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8156 case BPF_FUNC_skb_store_bytes:
8157 return &bpf_skb_store_bytes_proto;
8158 case BPF_FUNC_skb_load_bytes:
8159 return &bpf_skb_load_bytes_proto;
8160 case BPF_FUNC_skb_pull_data:
8161 return &sk_skb_pull_data_proto;
8162 case BPF_FUNC_skb_change_tail:
8163 return &sk_skb_change_tail_proto;
8164 case BPF_FUNC_skb_change_head:
8165 return &sk_skb_change_head_proto;
8166 case BPF_FUNC_skb_adjust_room:
8167 return &sk_skb_adjust_room_proto;
8168 case BPF_FUNC_get_socket_cookie:
8169 return &bpf_get_socket_cookie_proto;
8170 case BPF_FUNC_get_socket_uid:
8171 return &bpf_get_socket_uid_proto;
8172 case BPF_FUNC_sk_redirect_map:
8173 return &bpf_sk_redirect_map_proto;
8174 case BPF_FUNC_sk_redirect_hash:
8175 return &bpf_sk_redirect_hash_proto;
8176 case BPF_FUNC_perf_event_output:
8177 return &bpf_skb_event_output_proto;
8179 case BPF_FUNC_sk_lookup_tcp:
8180 return &bpf_sk_lookup_tcp_proto;
8181 case BPF_FUNC_sk_lookup_udp:
8182 return &bpf_sk_lookup_udp_proto;
8183 case BPF_FUNC_sk_release:
8184 return &bpf_sk_release_proto;
8185 case BPF_FUNC_skc_lookup_tcp:
8186 return &bpf_skc_lookup_tcp_proto;
8189 return bpf_sk_base_func_proto(func_id);
8193 static const struct bpf_func_proto *
8194 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8197 case BPF_FUNC_skb_load_bytes:
8198 return &bpf_flow_dissector_load_bytes_proto;
8200 return bpf_sk_base_func_proto(func_id);
8204 static const struct bpf_func_proto *
8205 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8208 case BPF_FUNC_skb_load_bytes:
8209 return &bpf_skb_load_bytes_proto;
8210 case BPF_FUNC_skb_pull_data:
8211 return &bpf_skb_pull_data_proto;
8212 case BPF_FUNC_csum_diff:
8213 return &bpf_csum_diff_proto;
8214 case BPF_FUNC_get_cgroup_classid:
8215 return &bpf_get_cgroup_classid_proto;
8216 case BPF_FUNC_get_route_realm:
8217 return &bpf_get_route_realm_proto;
8218 case BPF_FUNC_get_hash_recalc:
8219 return &bpf_get_hash_recalc_proto;
8220 case BPF_FUNC_perf_event_output:
8221 return &bpf_skb_event_output_proto;
8222 case BPF_FUNC_get_smp_processor_id:
8223 return &bpf_get_smp_processor_id_proto;
8224 case BPF_FUNC_skb_under_cgroup:
8225 return &bpf_skb_under_cgroup_proto;
8227 return bpf_sk_base_func_proto(func_id);
8231 static const struct bpf_func_proto *
8232 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8235 case BPF_FUNC_lwt_push_encap:
8236 return &bpf_lwt_in_push_encap_proto;
8238 return lwt_out_func_proto(func_id, prog);
8242 static const struct bpf_func_proto *
8243 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8246 case BPF_FUNC_skb_get_tunnel_key:
8247 return &bpf_skb_get_tunnel_key_proto;
8248 case BPF_FUNC_skb_set_tunnel_key:
8249 return bpf_get_skb_set_tunnel_proto(func_id);
8250 case BPF_FUNC_skb_get_tunnel_opt:
8251 return &bpf_skb_get_tunnel_opt_proto;
8252 case BPF_FUNC_skb_set_tunnel_opt:
8253 return bpf_get_skb_set_tunnel_proto(func_id);
8254 case BPF_FUNC_redirect:
8255 return &bpf_redirect_proto;
8256 case BPF_FUNC_clone_redirect:
8257 return &bpf_clone_redirect_proto;
8258 case BPF_FUNC_skb_change_tail:
8259 return &bpf_skb_change_tail_proto;
8260 case BPF_FUNC_skb_change_head:
8261 return &bpf_skb_change_head_proto;
8262 case BPF_FUNC_skb_store_bytes:
8263 return &bpf_skb_store_bytes_proto;
8264 case BPF_FUNC_csum_update:
8265 return &bpf_csum_update_proto;
8266 case BPF_FUNC_csum_level:
8267 return &bpf_csum_level_proto;
8268 case BPF_FUNC_l3_csum_replace:
8269 return &bpf_l3_csum_replace_proto;
8270 case BPF_FUNC_l4_csum_replace:
8271 return &bpf_l4_csum_replace_proto;
8272 case BPF_FUNC_set_hash_invalid:
8273 return &bpf_set_hash_invalid_proto;
8274 case BPF_FUNC_lwt_push_encap:
8275 return &bpf_lwt_xmit_push_encap_proto;
8277 return lwt_out_func_proto(func_id, prog);
8281 static const struct bpf_func_proto *
8282 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8285 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
8286 case BPF_FUNC_lwt_seg6_store_bytes:
8287 return &bpf_lwt_seg6_store_bytes_proto;
8288 case BPF_FUNC_lwt_seg6_action:
8289 return &bpf_lwt_seg6_action_proto;
8290 case BPF_FUNC_lwt_seg6_adjust_srh:
8291 return &bpf_lwt_seg6_adjust_srh_proto;
8294 return lwt_out_func_proto(func_id, prog);
8298 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
8299 const struct bpf_prog *prog,
8300 struct bpf_insn_access_aux *info)
8302 const int size_default = sizeof(__u32);
8304 if (off < 0 || off >= sizeof(struct __sk_buff))
8307 /* The verifier guarantees that size > 0. */
8308 if (off % size != 0)
8312 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8313 if (off + size > offsetofend(struct __sk_buff, cb[4]))
8316 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
8317 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
8318 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
8319 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
8320 case bpf_ctx_range(struct __sk_buff, data):
8321 case bpf_ctx_range(struct __sk_buff, data_meta):
8322 case bpf_ctx_range(struct __sk_buff, data_end):
8323 if (size != size_default)
8326 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8328 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8329 if (type == BPF_WRITE || size != sizeof(__u64))
8332 case bpf_ctx_range(struct __sk_buff, tstamp):
8333 if (size != sizeof(__u64))
8336 case offsetof(struct __sk_buff, sk):
8337 if (type == BPF_WRITE || size != sizeof(__u64))
8339 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
8341 case offsetof(struct __sk_buff, tstamp_type):
8343 case offsetofend(struct __sk_buff, tstamp_type) ... offsetof(struct __sk_buff, hwtstamp) - 1:
8344 /* Explicitly prohibit access to padding in __sk_buff. */
8347 /* Only narrow read access allowed for now. */
8348 if (type == BPF_WRITE) {
8349 if (size != size_default)
8352 bpf_ctx_record_field_size(info, size_default);
8353 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8361 static bool sk_filter_is_valid_access(int off, int size,
8362 enum bpf_access_type type,
8363 const struct bpf_prog *prog,
8364 struct bpf_insn_access_aux *info)
8367 case bpf_ctx_range(struct __sk_buff, tc_classid):
8368 case bpf_ctx_range(struct __sk_buff, data):
8369 case bpf_ctx_range(struct __sk_buff, data_meta):
8370 case bpf_ctx_range(struct __sk_buff, data_end):
8371 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8372 case bpf_ctx_range(struct __sk_buff, tstamp):
8373 case bpf_ctx_range(struct __sk_buff, wire_len):
8374 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8378 if (type == BPF_WRITE) {
8380 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8387 return bpf_skb_is_valid_access(off, size, type, prog, info);
8390 static bool cg_skb_is_valid_access(int off, int size,
8391 enum bpf_access_type type,
8392 const struct bpf_prog *prog,
8393 struct bpf_insn_access_aux *info)
8396 case bpf_ctx_range(struct __sk_buff, tc_classid):
8397 case bpf_ctx_range(struct __sk_buff, data_meta):
8398 case bpf_ctx_range(struct __sk_buff, wire_len):
8400 case bpf_ctx_range(struct __sk_buff, data):
8401 case bpf_ctx_range(struct __sk_buff, data_end):
8407 if (type == BPF_WRITE) {
8409 case bpf_ctx_range(struct __sk_buff, mark):
8410 case bpf_ctx_range(struct __sk_buff, priority):
8411 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8413 case bpf_ctx_range(struct __sk_buff, tstamp):
8423 case bpf_ctx_range(struct __sk_buff, data):
8424 info->reg_type = PTR_TO_PACKET;
8426 case bpf_ctx_range(struct __sk_buff, data_end):
8427 info->reg_type = PTR_TO_PACKET_END;
8431 return bpf_skb_is_valid_access(off, size, type, prog, info);
8434 static bool lwt_is_valid_access(int off, int size,
8435 enum bpf_access_type type,
8436 const struct bpf_prog *prog,
8437 struct bpf_insn_access_aux *info)
8440 case bpf_ctx_range(struct __sk_buff, tc_classid):
8441 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8442 case bpf_ctx_range(struct __sk_buff, data_meta):
8443 case bpf_ctx_range(struct __sk_buff, tstamp):
8444 case bpf_ctx_range(struct __sk_buff, wire_len):
8445 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8449 if (type == BPF_WRITE) {
8451 case bpf_ctx_range(struct __sk_buff, mark):
8452 case bpf_ctx_range(struct __sk_buff, priority):
8453 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8461 case bpf_ctx_range(struct __sk_buff, data):
8462 info->reg_type = PTR_TO_PACKET;
8464 case bpf_ctx_range(struct __sk_buff, data_end):
8465 info->reg_type = PTR_TO_PACKET_END;
8469 return bpf_skb_is_valid_access(off, size, type, prog, info);
8472 /* Attach type specific accesses */
8473 static bool __sock_filter_check_attach_type(int off,
8474 enum bpf_access_type access_type,
8475 enum bpf_attach_type attach_type)
8478 case offsetof(struct bpf_sock, bound_dev_if):
8479 case offsetof(struct bpf_sock, mark):
8480 case offsetof(struct bpf_sock, priority):
8481 switch (attach_type) {
8482 case BPF_CGROUP_INET_SOCK_CREATE:
8483 case BPF_CGROUP_INET_SOCK_RELEASE:
8488 case bpf_ctx_range(struct bpf_sock, src_ip4):
8489 switch (attach_type) {
8490 case BPF_CGROUP_INET4_POST_BIND:
8495 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8496 switch (attach_type) {
8497 case BPF_CGROUP_INET6_POST_BIND:
8502 case bpf_ctx_range(struct bpf_sock, src_port):
8503 switch (attach_type) {
8504 case BPF_CGROUP_INET4_POST_BIND:
8505 case BPF_CGROUP_INET6_POST_BIND:
8512 return access_type == BPF_READ;
8517 bool bpf_sock_common_is_valid_access(int off, int size,
8518 enum bpf_access_type type,
8519 struct bpf_insn_access_aux *info)
8522 case bpf_ctx_range_till(struct bpf_sock, type, priority):
8525 return bpf_sock_is_valid_access(off, size, type, info);
8529 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
8530 struct bpf_insn_access_aux *info)
8532 const int size_default = sizeof(__u32);
8535 if (off < 0 || off >= sizeof(struct bpf_sock))
8537 if (off % size != 0)
8541 case offsetof(struct bpf_sock, state):
8542 case offsetof(struct bpf_sock, family):
8543 case offsetof(struct bpf_sock, type):
8544 case offsetof(struct bpf_sock, protocol):
8545 case offsetof(struct bpf_sock, src_port):
8546 case offsetof(struct bpf_sock, rx_queue_mapping):
8547 case bpf_ctx_range(struct bpf_sock, src_ip4):
8548 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8549 case bpf_ctx_range(struct bpf_sock, dst_ip4):
8550 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8551 bpf_ctx_record_field_size(info, size_default);
8552 return bpf_ctx_narrow_access_ok(off, size, size_default);
8553 case bpf_ctx_range(struct bpf_sock, dst_port):
8554 field_size = size == size_default ?
8555 size_default : sizeof_field(struct bpf_sock, dst_port);
8556 bpf_ctx_record_field_size(info, field_size);
8557 return bpf_ctx_narrow_access_ok(off, size, field_size);
8558 case offsetofend(struct bpf_sock, dst_port) ...
8559 offsetof(struct bpf_sock, dst_ip4) - 1:
8563 return size == size_default;
8566 static bool sock_filter_is_valid_access(int off, int size,
8567 enum bpf_access_type type,
8568 const struct bpf_prog *prog,
8569 struct bpf_insn_access_aux *info)
8571 if (!bpf_sock_is_valid_access(off, size, type, info))
8573 return __sock_filter_check_attach_type(off, type,
8574 prog->expected_attach_type);
8577 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
8578 const struct bpf_prog *prog)
8580 /* Neither direct read nor direct write requires any preliminary
8586 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
8587 const struct bpf_prog *prog, int drop_verdict)
8589 struct bpf_insn *insn = insn_buf;
8594 /* if (!skb->cloned)
8597 * (Fast-path, otherwise approximation that we might be
8598 * a clone, do the rest in helper.)
8600 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET);
8601 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
8602 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
8604 /* ret = bpf_skb_pull_data(skb, 0); */
8605 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
8606 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
8607 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
8608 BPF_FUNC_skb_pull_data);
8611 * return TC_ACT_SHOT;
8613 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
8614 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
8615 *insn++ = BPF_EXIT_INSN();
8618 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
8620 *insn++ = prog->insnsi[0];
8622 return insn - insn_buf;
8625 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
8626 struct bpf_insn *insn_buf)
8628 bool indirect = BPF_MODE(orig->code) == BPF_IND;
8629 struct bpf_insn *insn = insn_buf;
8632 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
8634 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
8636 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
8638 /* We're guaranteed here that CTX is in R6. */
8639 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
8641 switch (BPF_SIZE(orig->code)) {
8643 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
8646 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
8649 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
8653 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
8654 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
8655 *insn++ = BPF_EXIT_INSN();
8657 return insn - insn_buf;
8660 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8661 const struct bpf_prog *prog)
8663 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8666 static bool tc_cls_act_is_valid_access(int off, int size,
8667 enum bpf_access_type type,
8668 const struct bpf_prog *prog,
8669 struct bpf_insn_access_aux *info)
8671 if (type == BPF_WRITE) {
8673 case bpf_ctx_range(struct __sk_buff, mark):
8674 case bpf_ctx_range(struct __sk_buff, tc_index):
8675 case bpf_ctx_range(struct __sk_buff, priority):
8676 case bpf_ctx_range(struct __sk_buff, tc_classid):
8677 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8678 case bpf_ctx_range(struct __sk_buff, tstamp):
8679 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8687 case bpf_ctx_range(struct __sk_buff, data):
8688 info->reg_type = PTR_TO_PACKET;
8690 case bpf_ctx_range(struct __sk_buff, data_meta):
8691 info->reg_type = PTR_TO_PACKET_META;
8693 case bpf_ctx_range(struct __sk_buff, data_end):
8694 info->reg_type = PTR_TO_PACKET_END;
8696 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8698 case offsetof(struct __sk_buff, tstamp_type):
8699 /* The convert_ctx_access() on reading and writing
8700 * __sk_buff->tstamp depends on whether the bpf prog
8701 * has used __sk_buff->tstamp_type or not.
8702 * Thus, we need to set prog->tstamp_type_access
8703 * earlier during is_valid_access() here.
8705 ((struct bpf_prog *)prog)->tstamp_type_access = 1;
8706 return size == sizeof(__u8);
8709 return bpf_skb_is_valid_access(off, size, type, prog, info);
8712 static bool __is_valid_xdp_access(int off, int size)
8714 if (off < 0 || off >= sizeof(struct xdp_md))
8716 if (off % size != 0)
8718 if (size != sizeof(__u32))
8724 static bool xdp_is_valid_access(int off, int size,
8725 enum bpf_access_type type,
8726 const struct bpf_prog *prog,
8727 struct bpf_insn_access_aux *info)
8729 if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8731 case offsetof(struct xdp_md, egress_ifindex):
8736 if (type == BPF_WRITE) {
8737 if (bpf_prog_is_dev_bound(prog->aux)) {
8739 case offsetof(struct xdp_md, rx_queue_index):
8740 return __is_valid_xdp_access(off, size);
8747 case offsetof(struct xdp_md, data):
8748 info->reg_type = PTR_TO_PACKET;
8750 case offsetof(struct xdp_md, data_meta):
8751 info->reg_type = PTR_TO_PACKET_META;
8753 case offsetof(struct xdp_md, data_end):
8754 info->reg_type = PTR_TO_PACKET_END;
8758 return __is_valid_xdp_access(off, size);
8761 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act)
8763 const u32 act_max = XDP_REDIRECT;
8765 pr_warn_once("%s XDP return value %u on prog %s (id %d) dev %s, expect packet loss!\n",
8766 act > act_max ? "Illegal" : "Driver unsupported",
8767 act, prog->aux->name, prog->aux->id, dev ? dev->name : "N/A");
8769 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
8771 static bool sock_addr_is_valid_access(int off, int size,
8772 enum bpf_access_type type,
8773 const struct bpf_prog *prog,
8774 struct bpf_insn_access_aux *info)
8776 const int size_default = sizeof(__u32);
8778 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
8780 if (off % size != 0)
8783 /* Disallow access to IPv6 fields from IPv4 contex and vise
8787 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8788 switch (prog->expected_attach_type) {
8789 case BPF_CGROUP_INET4_BIND:
8790 case BPF_CGROUP_INET4_CONNECT:
8791 case BPF_CGROUP_INET4_GETPEERNAME:
8792 case BPF_CGROUP_INET4_GETSOCKNAME:
8793 case BPF_CGROUP_UDP4_SENDMSG:
8794 case BPF_CGROUP_UDP4_RECVMSG:
8800 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8801 switch (prog->expected_attach_type) {
8802 case BPF_CGROUP_INET6_BIND:
8803 case BPF_CGROUP_INET6_CONNECT:
8804 case BPF_CGROUP_INET6_GETPEERNAME:
8805 case BPF_CGROUP_INET6_GETSOCKNAME:
8806 case BPF_CGROUP_UDP6_SENDMSG:
8807 case BPF_CGROUP_UDP6_RECVMSG:
8813 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8814 switch (prog->expected_attach_type) {
8815 case BPF_CGROUP_UDP4_SENDMSG:
8821 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8823 switch (prog->expected_attach_type) {
8824 case BPF_CGROUP_UDP6_SENDMSG:
8833 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8834 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8835 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8836 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8838 case bpf_ctx_range(struct bpf_sock_addr, user_port):
8839 if (type == BPF_READ) {
8840 bpf_ctx_record_field_size(info, size_default);
8842 if (bpf_ctx_wide_access_ok(off, size,
8843 struct bpf_sock_addr,
8847 if (bpf_ctx_wide_access_ok(off, size,
8848 struct bpf_sock_addr,
8852 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8855 if (bpf_ctx_wide_access_ok(off, size,
8856 struct bpf_sock_addr,
8860 if (bpf_ctx_wide_access_ok(off, size,
8861 struct bpf_sock_addr,
8865 if (size != size_default)
8869 case offsetof(struct bpf_sock_addr, sk):
8870 if (type != BPF_READ)
8872 if (size != sizeof(__u64))
8874 info->reg_type = PTR_TO_SOCKET;
8877 if (type == BPF_READ) {
8878 if (size != size_default)
8888 static bool sock_ops_is_valid_access(int off, int size,
8889 enum bpf_access_type type,
8890 const struct bpf_prog *prog,
8891 struct bpf_insn_access_aux *info)
8893 const int size_default = sizeof(__u32);
8895 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
8898 /* The verifier guarantees that size > 0. */
8899 if (off % size != 0)
8902 if (type == BPF_WRITE) {
8904 case offsetof(struct bpf_sock_ops, reply):
8905 case offsetof(struct bpf_sock_ops, sk_txhash):
8906 if (size != size_default)
8914 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
8916 if (size != sizeof(__u64))
8919 case offsetof(struct bpf_sock_ops, sk):
8920 if (size != sizeof(__u64))
8922 info->reg_type = PTR_TO_SOCKET_OR_NULL;
8924 case offsetof(struct bpf_sock_ops, skb_data):
8925 if (size != sizeof(__u64))
8927 info->reg_type = PTR_TO_PACKET;
8929 case offsetof(struct bpf_sock_ops, skb_data_end):
8930 if (size != sizeof(__u64))
8932 info->reg_type = PTR_TO_PACKET_END;
8934 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
8935 bpf_ctx_record_field_size(info, size_default);
8936 return bpf_ctx_narrow_access_ok(off, size,
8939 if (size != size_default)
8948 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
8949 const struct bpf_prog *prog)
8951 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
8954 static bool sk_skb_is_valid_access(int off, int size,
8955 enum bpf_access_type type,
8956 const struct bpf_prog *prog,
8957 struct bpf_insn_access_aux *info)
8960 case bpf_ctx_range(struct __sk_buff, tc_classid):
8961 case bpf_ctx_range(struct __sk_buff, data_meta):
8962 case bpf_ctx_range(struct __sk_buff, tstamp):
8963 case bpf_ctx_range(struct __sk_buff, wire_len):
8964 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8968 if (type == BPF_WRITE) {
8970 case bpf_ctx_range(struct __sk_buff, tc_index):
8971 case bpf_ctx_range(struct __sk_buff, priority):
8979 case bpf_ctx_range(struct __sk_buff, mark):
8981 case bpf_ctx_range(struct __sk_buff, data):
8982 info->reg_type = PTR_TO_PACKET;
8984 case bpf_ctx_range(struct __sk_buff, data_end):
8985 info->reg_type = PTR_TO_PACKET_END;
8989 return bpf_skb_is_valid_access(off, size, type, prog, info);
8992 static bool sk_msg_is_valid_access(int off, int size,
8993 enum bpf_access_type type,
8994 const struct bpf_prog *prog,
8995 struct bpf_insn_access_aux *info)
8997 if (type == BPF_WRITE)
9000 if (off % size != 0)
9004 case offsetof(struct sk_msg_md, data):
9005 info->reg_type = PTR_TO_PACKET;
9006 if (size != sizeof(__u64))
9009 case offsetof(struct sk_msg_md, data_end):
9010 info->reg_type = PTR_TO_PACKET_END;
9011 if (size != sizeof(__u64))
9014 case offsetof(struct sk_msg_md, sk):
9015 if (size != sizeof(__u64))
9017 info->reg_type = PTR_TO_SOCKET;
9019 case bpf_ctx_range(struct sk_msg_md, family):
9020 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
9021 case bpf_ctx_range(struct sk_msg_md, local_ip4):
9022 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
9023 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
9024 case bpf_ctx_range(struct sk_msg_md, remote_port):
9025 case bpf_ctx_range(struct sk_msg_md, local_port):
9026 case bpf_ctx_range(struct sk_msg_md, size):
9027 if (size != sizeof(__u32))
9036 static bool flow_dissector_is_valid_access(int off, int size,
9037 enum bpf_access_type type,
9038 const struct bpf_prog *prog,
9039 struct bpf_insn_access_aux *info)
9041 const int size_default = sizeof(__u32);
9043 if (off < 0 || off >= sizeof(struct __sk_buff))
9046 if (type == BPF_WRITE)
9050 case bpf_ctx_range(struct __sk_buff, data):
9051 if (size != size_default)
9053 info->reg_type = PTR_TO_PACKET;
9055 case bpf_ctx_range(struct __sk_buff, data_end):
9056 if (size != size_default)
9058 info->reg_type = PTR_TO_PACKET_END;
9060 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
9061 if (size != sizeof(__u64))
9063 info->reg_type = PTR_TO_FLOW_KEYS;
9070 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
9071 const struct bpf_insn *si,
9072 struct bpf_insn *insn_buf,
9073 struct bpf_prog *prog,
9077 struct bpf_insn *insn = insn_buf;
9080 case offsetof(struct __sk_buff, data):
9081 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
9082 si->dst_reg, si->src_reg,
9083 offsetof(struct bpf_flow_dissector, data));
9086 case offsetof(struct __sk_buff, data_end):
9087 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
9088 si->dst_reg, si->src_reg,
9089 offsetof(struct bpf_flow_dissector, data_end));
9092 case offsetof(struct __sk_buff, flow_keys):
9093 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
9094 si->dst_reg, si->src_reg,
9095 offsetof(struct bpf_flow_dissector, flow_keys));
9099 return insn - insn_buf;
9102 static struct bpf_insn *bpf_convert_tstamp_type_read(const struct bpf_insn *si,
9103 struct bpf_insn *insn)
9105 __u8 value_reg = si->dst_reg;
9106 __u8 skb_reg = si->src_reg;
9107 /* AX is needed because src_reg and dst_reg could be the same */
9108 __u8 tmp_reg = BPF_REG_AX;
9110 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg,
9111 PKT_VLAN_PRESENT_OFFSET);
9112 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg,
9113 SKB_MONO_DELIVERY_TIME_MASK, 2);
9114 *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_UNSPEC);
9115 *insn++ = BPF_JMP_A(1);
9116 *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_DELIVERY_MONO);
9121 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si,
9122 struct bpf_insn *insn)
9124 /* si->dst_reg = skb_shinfo(SKB); */
9125 #ifdef NET_SKBUFF_DATA_USES_OFFSET
9126 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9127 BPF_REG_AX, si->src_reg,
9128 offsetof(struct sk_buff, end));
9129 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
9130 si->dst_reg, si->src_reg,
9131 offsetof(struct sk_buff, head));
9132 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
9134 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9135 si->dst_reg, si->src_reg,
9136 offsetof(struct sk_buff, end));
9142 static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog,
9143 const struct bpf_insn *si,
9144 struct bpf_insn *insn)
9146 __u8 value_reg = si->dst_reg;
9147 __u8 skb_reg = si->src_reg;
9149 #ifdef CONFIG_NET_CLS_ACT
9150 /* If the tstamp_type is read,
9151 * the bpf prog is aware the tstamp could have delivery time.
9152 * Thus, read skb->tstamp as is if tstamp_type_access is true.
9154 if (!prog->tstamp_type_access) {
9155 /* AX is needed because src_reg and dst_reg could be the same */
9156 __u8 tmp_reg = BPF_REG_AX;
9158 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, PKT_VLAN_PRESENT_OFFSET);
9159 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg,
9160 TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK);
9161 *insn++ = BPF_JMP32_IMM(BPF_JNE, tmp_reg,
9162 TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK, 2);
9163 /* skb->tc_at_ingress && skb->mono_delivery_time,
9164 * read 0 as the (rcv) timestamp.
9166 *insn++ = BPF_MOV64_IMM(value_reg, 0);
9167 *insn++ = BPF_JMP_A(1);
9171 *insn++ = BPF_LDX_MEM(BPF_DW, value_reg, skb_reg,
9172 offsetof(struct sk_buff, tstamp));
9176 static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog,
9177 const struct bpf_insn *si,
9178 struct bpf_insn *insn)
9180 __u8 value_reg = si->src_reg;
9181 __u8 skb_reg = si->dst_reg;
9183 #ifdef CONFIG_NET_CLS_ACT
9184 /* If the tstamp_type is read,
9185 * the bpf prog is aware the tstamp could have delivery time.
9186 * Thus, write skb->tstamp as is if tstamp_type_access is true.
9187 * Otherwise, writing at ingress will have to clear the
9188 * mono_delivery_time bit also.
9190 if (!prog->tstamp_type_access) {
9191 __u8 tmp_reg = BPF_REG_AX;
9193 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, PKT_VLAN_PRESENT_OFFSET);
9194 /* Writing __sk_buff->tstamp as ingress, goto <clear> */
9195 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1);
9197 *insn++ = BPF_JMP_A(2);
9198 /* <clear>: mono_delivery_time */
9199 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_MONO_DELIVERY_TIME_MASK);
9200 *insn++ = BPF_STX_MEM(BPF_B, skb_reg, tmp_reg, PKT_VLAN_PRESENT_OFFSET);
9204 /* <store>: skb->tstamp = tstamp */
9205 *insn++ = BPF_STX_MEM(BPF_DW, skb_reg, value_reg,
9206 offsetof(struct sk_buff, tstamp));
9210 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
9211 const struct bpf_insn *si,
9212 struct bpf_insn *insn_buf,
9213 struct bpf_prog *prog, u32 *target_size)
9215 struct bpf_insn *insn = insn_buf;
9219 case offsetof(struct __sk_buff, len):
9220 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9221 bpf_target_off(struct sk_buff, len, 4,
9225 case offsetof(struct __sk_buff, protocol):
9226 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9227 bpf_target_off(struct sk_buff, protocol, 2,
9231 case offsetof(struct __sk_buff, vlan_proto):
9232 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9233 bpf_target_off(struct sk_buff, vlan_proto, 2,
9237 case offsetof(struct __sk_buff, priority):
9238 if (type == BPF_WRITE)
9239 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9240 bpf_target_off(struct sk_buff, priority, 4,
9243 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9244 bpf_target_off(struct sk_buff, priority, 4,
9248 case offsetof(struct __sk_buff, ingress_ifindex):
9249 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9250 bpf_target_off(struct sk_buff, skb_iif, 4,
9254 case offsetof(struct __sk_buff, ifindex):
9255 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9256 si->dst_reg, si->src_reg,
9257 offsetof(struct sk_buff, dev));
9258 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9259 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9260 bpf_target_off(struct net_device, ifindex, 4,
9264 case offsetof(struct __sk_buff, hash):
9265 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9266 bpf_target_off(struct sk_buff, hash, 4,
9270 case offsetof(struct __sk_buff, mark):
9271 if (type == BPF_WRITE)
9272 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9273 bpf_target_off(struct sk_buff, mark, 4,
9276 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9277 bpf_target_off(struct sk_buff, mark, 4,
9281 case offsetof(struct __sk_buff, pkt_type):
9283 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
9285 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
9286 #ifdef __BIG_ENDIAN_BITFIELD
9287 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
9291 case offsetof(struct __sk_buff, queue_mapping):
9292 if (type == BPF_WRITE) {
9293 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
9294 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
9295 bpf_target_off(struct sk_buff,
9299 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9300 bpf_target_off(struct sk_buff,
9306 case offsetof(struct __sk_buff, vlan_present):
9308 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
9309 PKT_VLAN_PRESENT_OFFSET);
9310 if (PKT_VLAN_PRESENT_BIT)
9311 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
9312 if (PKT_VLAN_PRESENT_BIT < 7)
9313 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
9316 case offsetof(struct __sk_buff, vlan_tci):
9317 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9318 bpf_target_off(struct sk_buff, vlan_tci, 2,
9322 case offsetof(struct __sk_buff, cb[0]) ...
9323 offsetofend(struct __sk_buff, cb[4]) - 1:
9324 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
9325 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
9326 offsetof(struct qdisc_skb_cb, data)) %
9329 prog->cb_access = 1;
9331 off -= offsetof(struct __sk_buff, cb[0]);
9332 off += offsetof(struct sk_buff, cb);
9333 off += offsetof(struct qdisc_skb_cb, data);
9334 if (type == BPF_WRITE)
9335 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
9338 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
9342 case offsetof(struct __sk_buff, tc_classid):
9343 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
9346 off -= offsetof(struct __sk_buff, tc_classid);
9347 off += offsetof(struct sk_buff, cb);
9348 off += offsetof(struct qdisc_skb_cb, tc_classid);
9350 if (type == BPF_WRITE)
9351 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
9354 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
9358 case offsetof(struct __sk_buff, data):
9359 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9360 si->dst_reg, si->src_reg,
9361 offsetof(struct sk_buff, data));
9364 case offsetof(struct __sk_buff, data_meta):
9366 off -= offsetof(struct __sk_buff, data_meta);
9367 off += offsetof(struct sk_buff, cb);
9368 off += offsetof(struct bpf_skb_data_end, data_meta);
9369 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9373 case offsetof(struct __sk_buff, data_end):
9375 off -= offsetof(struct __sk_buff, data_end);
9376 off += offsetof(struct sk_buff, cb);
9377 off += offsetof(struct bpf_skb_data_end, data_end);
9378 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9382 case offsetof(struct __sk_buff, tc_index):
9383 #ifdef CONFIG_NET_SCHED
9384 if (type == BPF_WRITE)
9385 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
9386 bpf_target_off(struct sk_buff, tc_index, 2,
9389 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9390 bpf_target_off(struct sk_buff, tc_index, 2,
9394 if (type == BPF_WRITE)
9395 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
9397 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9401 case offsetof(struct __sk_buff, napi_id):
9402 #if defined(CONFIG_NET_RX_BUSY_POLL)
9403 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9404 bpf_target_off(struct sk_buff, napi_id, 4,
9406 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
9407 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9410 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9413 case offsetof(struct __sk_buff, family):
9414 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9416 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9417 si->dst_reg, si->src_reg,
9418 offsetof(struct sk_buff, sk));
9419 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9420 bpf_target_off(struct sock_common,
9424 case offsetof(struct __sk_buff, remote_ip4):
9425 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9427 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9428 si->dst_reg, si->src_reg,
9429 offsetof(struct sk_buff, sk));
9430 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9431 bpf_target_off(struct sock_common,
9435 case offsetof(struct __sk_buff, local_ip4):
9436 BUILD_BUG_ON(sizeof_field(struct sock_common,
9437 skc_rcv_saddr) != 4);
9439 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9440 si->dst_reg, si->src_reg,
9441 offsetof(struct sk_buff, sk));
9442 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9443 bpf_target_off(struct sock_common,
9447 case offsetof(struct __sk_buff, remote_ip6[0]) ...
9448 offsetof(struct __sk_buff, remote_ip6[3]):
9449 #if IS_ENABLED(CONFIG_IPV6)
9450 BUILD_BUG_ON(sizeof_field(struct sock_common,
9451 skc_v6_daddr.s6_addr32[0]) != 4);
9454 off -= offsetof(struct __sk_buff, remote_ip6[0]);
9456 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9457 si->dst_reg, si->src_reg,
9458 offsetof(struct sk_buff, sk));
9459 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9460 offsetof(struct sock_common,
9461 skc_v6_daddr.s6_addr32[0]) +
9464 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9467 case offsetof(struct __sk_buff, local_ip6[0]) ...
9468 offsetof(struct __sk_buff, local_ip6[3]):
9469 #if IS_ENABLED(CONFIG_IPV6)
9470 BUILD_BUG_ON(sizeof_field(struct sock_common,
9471 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9474 off -= offsetof(struct __sk_buff, local_ip6[0]);
9476 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9477 si->dst_reg, si->src_reg,
9478 offsetof(struct sk_buff, sk));
9479 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9480 offsetof(struct sock_common,
9481 skc_v6_rcv_saddr.s6_addr32[0]) +
9484 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9488 case offsetof(struct __sk_buff, remote_port):
9489 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9491 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9492 si->dst_reg, si->src_reg,
9493 offsetof(struct sk_buff, sk));
9494 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9495 bpf_target_off(struct sock_common,
9498 #ifndef __BIG_ENDIAN_BITFIELD
9499 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9503 case offsetof(struct __sk_buff, local_port):
9504 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9506 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9507 si->dst_reg, si->src_reg,
9508 offsetof(struct sk_buff, sk));
9509 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9510 bpf_target_off(struct sock_common,
9511 skc_num, 2, target_size));
9514 case offsetof(struct __sk_buff, tstamp):
9515 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
9517 if (type == BPF_WRITE)
9518 insn = bpf_convert_tstamp_write(prog, si, insn);
9520 insn = bpf_convert_tstamp_read(prog, si, insn);
9523 case offsetof(struct __sk_buff, tstamp_type):
9524 insn = bpf_convert_tstamp_type_read(si, insn);
9527 case offsetof(struct __sk_buff, gso_segs):
9528 insn = bpf_convert_shinfo_access(si, insn);
9529 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
9530 si->dst_reg, si->dst_reg,
9531 bpf_target_off(struct skb_shared_info,
9535 case offsetof(struct __sk_buff, gso_size):
9536 insn = bpf_convert_shinfo_access(si, insn);
9537 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
9538 si->dst_reg, si->dst_reg,
9539 bpf_target_off(struct skb_shared_info,
9543 case offsetof(struct __sk_buff, wire_len):
9544 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
9547 off -= offsetof(struct __sk_buff, wire_len);
9548 off += offsetof(struct sk_buff, cb);
9549 off += offsetof(struct qdisc_skb_cb, pkt_len);
9551 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
9554 case offsetof(struct __sk_buff, sk):
9555 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9556 si->dst_reg, si->src_reg,
9557 offsetof(struct sk_buff, sk));
9559 case offsetof(struct __sk_buff, hwtstamp):
9560 BUILD_BUG_ON(sizeof_field(struct skb_shared_hwtstamps, hwtstamp) != 8);
9561 BUILD_BUG_ON(offsetof(struct skb_shared_hwtstamps, hwtstamp) != 0);
9563 insn = bpf_convert_shinfo_access(si, insn);
9564 *insn++ = BPF_LDX_MEM(BPF_DW,
9565 si->dst_reg, si->dst_reg,
9566 bpf_target_off(struct skb_shared_info,
9572 return insn - insn_buf;
9575 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
9576 const struct bpf_insn *si,
9577 struct bpf_insn *insn_buf,
9578 struct bpf_prog *prog, u32 *target_size)
9580 struct bpf_insn *insn = insn_buf;
9584 case offsetof(struct bpf_sock, bound_dev_if):
9585 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
9587 if (type == BPF_WRITE)
9588 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9589 offsetof(struct sock, sk_bound_dev_if));
9591 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9592 offsetof(struct sock, sk_bound_dev_if));
9595 case offsetof(struct bpf_sock, mark):
9596 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
9598 if (type == BPF_WRITE)
9599 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9600 offsetof(struct sock, sk_mark));
9602 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9603 offsetof(struct sock, sk_mark));
9606 case offsetof(struct bpf_sock, priority):
9607 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
9609 if (type == BPF_WRITE)
9610 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9611 offsetof(struct sock, sk_priority));
9613 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9614 offsetof(struct sock, sk_priority));
9617 case offsetof(struct bpf_sock, family):
9618 *insn++ = BPF_LDX_MEM(
9619 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
9620 si->dst_reg, si->src_reg,
9621 bpf_target_off(struct sock_common,
9623 sizeof_field(struct sock_common,
9628 case offsetof(struct bpf_sock, type):
9629 *insn++ = BPF_LDX_MEM(
9630 BPF_FIELD_SIZEOF(struct sock, sk_type),
9631 si->dst_reg, si->src_reg,
9632 bpf_target_off(struct sock, sk_type,
9633 sizeof_field(struct sock, sk_type),
9637 case offsetof(struct bpf_sock, protocol):
9638 *insn++ = BPF_LDX_MEM(
9639 BPF_FIELD_SIZEOF(struct sock, sk_protocol),
9640 si->dst_reg, si->src_reg,
9641 bpf_target_off(struct sock, sk_protocol,
9642 sizeof_field(struct sock, sk_protocol),
9646 case offsetof(struct bpf_sock, src_ip4):
9647 *insn++ = BPF_LDX_MEM(
9648 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9649 bpf_target_off(struct sock_common, skc_rcv_saddr,
9650 sizeof_field(struct sock_common,
9655 case offsetof(struct bpf_sock, dst_ip4):
9656 *insn++ = BPF_LDX_MEM(
9657 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9658 bpf_target_off(struct sock_common, skc_daddr,
9659 sizeof_field(struct sock_common,
9664 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
9665 #if IS_ENABLED(CONFIG_IPV6)
9667 off -= offsetof(struct bpf_sock, src_ip6[0]);
9668 *insn++ = BPF_LDX_MEM(
9669 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9672 skc_v6_rcv_saddr.s6_addr32[0],
9673 sizeof_field(struct sock_common,
9674 skc_v6_rcv_saddr.s6_addr32[0]),
9675 target_size) + off);
9678 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9682 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
9683 #if IS_ENABLED(CONFIG_IPV6)
9685 off -= offsetof(struct bpf_sock, dst_ip6[0]);
9686 *insn++ = BPF_LDX_MEM(
9687 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9688 bpf_target_off(struct sock_common,
9689 skc_v6_daddr.s6_addr32[0],
9690 sizeof_field(struct sock_common,
9691 skc_v6_daddr.s6_addr32[0]),
9692 target_size) + off);
9694 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9699 case offsetof(struct bpf_sock, src_port):
9700 *insn++ = BPF_LDX_MEM(
9701 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
9702 si->dst_reg, si->src_reg,
9703 bpf_target_off(struct sock_common, skc_num,
9704 sizeof_field(struct sock_common,
9709 case offsetof(struct bpf_sock, dst_port):
9710 *insn++ = BPF_LDX_MEM(
9711 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
9712 si->dst_reg, si->src_reg,
9713 bpf_target_off(struct sock_common, skc_dport,
9714 sizeof_field(struct sock_common,
9719 case offsetof(struct bpf_sock, state):
9720 *insn++ = BPF_LDX_MEM(
9721 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
9722 si->dst_reg, si->src_reg,
9723 bpf_target_off(struct sock_common, skc_state,
9724 sizeof_field(struct sock_common,
9728 case offsetof(struct bpf_sock, rx_queue_mapping):
9729 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
9730 *insn++ = BPF_LDX_MEM(
9731 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
9732 si->dst_reg, si->src_reg,
9733 bpf_target_off(struct sock, sk_rx_queue_mapping,
9734 sizeof_field(struct sock,
9735 sk_rx_queue_mapping),
9737 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
9739 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9741 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9747 return insn - insn_buf;
9750 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
9751 const struct bpf_insn *si,
9752 struct bpf_insn *insn_buf,
9753 struct bpf_prog *prog, u32 *target_size)
9755 struct bpf_insn *insn = insn_buf;
9758 case offsetof(struct __sk_buff, ifindex):
9759 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9760 si->dst_reg, si->src_reg,
9761 offsetof(struct sk_buff, dev));
9762 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9763 bpf_target_off(struct net_device, ifindex, 4,
9767 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9771 return insn - insn_buf;
9774 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
9775 const struct bpf_insn *si,
9776 struct bpf_insn *insn_buf,
9777 struct bpf_prog *prog, u32 *target_size)
9779 struct bpf_insn *insn = insn_buf;
9782 case offsetof(struct xdp_md, data):
9783 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
9784 si->dst_reg, si->src_reg,
9785 offsetof(struct xdp_buff, data));
9787 case offsetof(struct xdp_md, data_meta):
9788 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
9789 si->dst_reg, si->src_reg,
9790 offsetof(struct xdp_buff, data_meta));
9792 case offsetof(struct xdp_md, data_end):
9793 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
9794 si->dst_reg, si->src_reg,
9795 offsetof(struct xdp_buff, data_end));
9797 case offsetof(struct xdp_md, ingress_ifindex):
9798 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9799 si->dst_reg, si->src_reg,
9800 offsetof(struct xdp_buff, rxq));
9801 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
9802 si->dst_reg, si->dst_reg,
9803 offsetof(struct xdp_rxq_info, dev));
9804 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9805 offsetof(struct net_device, ifindex));
9807 case offsetof(struct xdp_md, rx_queue_index):
9808 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9809 si->dst_reg, si->src_reg,
9810 offsetof(struct xdp_buff, rxq));
9811 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9812 offsetof(struct xdp_rxq_info,
9815 case offsetof(struct xdp_md, egress_ifindex):
9816 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
9817 si->dst_reg, si->src_reg,
9818 offsetof(struct xdp_buff, txq));
9819 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
9820 si->dst_reg, si->dst_reg,
9821 offsetof(struct xdp_txq_info, dev));
9822 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9823 offsetof(struct net_device, ifindex));
9827 return insn - insn_buf;
9830 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
9831 * context Structure, F is Field in context structure that contains a pointer
9832 * to Nested Structure of type NS that has the field NF.
9834 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
9835 * sure that SIZE is not greater than actual size of S.F.NF.
9837 * If offset OFF is provided, the load happens from that offset relative to
9840 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
9842 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
9843 si->src_reg, offsetof(S, F)); \
9844 *insn++ = BPF_LDX_MEM( \
9845 SIZE, si->dst_reg, si->dst_reg, \
9846 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
9851 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
9852 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
9853 BPF_FIELD_SIZEOF(NS, NF), 0)
9855 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
9856 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
9858 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
9859 * "register" since two registers available in convert_ctx_access are not
9860 * enough: we can't override neither SRC, since it contains value to store, nor
9861 * DST since it contains pointer to context that may be used by later
9862 * instructions. But we need a temporary place to save pointer to nested
9863 * structure whose field we want to store to.
9865 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
9867 int tmp_reg = BPF_REG_9; \
9868 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
9870 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
9872 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
9874 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
9875 si->dst_reg, offsetof(S, F)); \
9876 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \
9877 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
9880 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
9884 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
9887 if (type == BPF_WRITE) { \
9888 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
9891 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
9892 S, NS, F, NF, SIZE, OFF); \
9896 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
9897 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
9898 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
9900 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
9901 const struct bpf_insn *si,
9902 struct bpf_insn *insn_buf,
9903 struct bpf_prog *prog, u32 *target_size)
9905 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
9906 struct bpf_insn *insn = insn_buf;
9909 case offsetof(struct bpf_sock_addr, user_family):
9910 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9911 struct sockaddr, uaddr, sa_family);
9914 case offsetof(struct bpf_sock_addr, user_ip4):
9915 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9916 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
9917 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
9920 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9922 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
9923 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9924 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9925 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
9929 case offsetof(struct bpf_sock_addr, user_port):
9930 /* To get port we need to know sa_family first and then treat
9931 * sockaddr as either sockaddr_in or sockaddr_in6.
9932 * Though we can simplify since port field has same offset and
9933 * size in both structures.
9934 * Here we check this invariant and use just one of the
9935 * structures if it's true.
9937 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
9938 offsetof(struct sockaddr_in6, sin6_port));
9939 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
9940 sizeof_field(struct sockaddr_in6, sin6_port));
9941 /* Account for sin6_port being smaller than user_port. */
9942 port_size = min(port_size, BPF_LDST_BYTES(si));
9943 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9944 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9945 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
9948 case offsetof(struct bpf_sock_addr, family):
9949 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9950 struct sock, sk, sk_family);
9953 case offsetof(struct bpf_sock_addr, type):
9954 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9955 struct sock, sk, sk_type);
9958 case offsetof(struct bpf_sock_addr, protocol):
9959 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9960 struct sock, sk, sk_protocol);
9963 case offsetof(struct bpf_sock_addr, msg_src_ip4):
9964 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
9965 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9966 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
9967 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
9970 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9973 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
9974 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
9975 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9976 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
9977 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
9979 case offsetof(struct bpf_sock_addr, sk):
9980 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
9981 si->dst_reg, si->src_reg,
9982 offsetof(struct bpf_sock_addr_kern, sk));
9986 return insn - insn_buf;
9989 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
9990 const struct bpf_insn *si,
9991 struct bpf_insn *insn_buf,
9992 struct bpf_prog *prog,
9995 struct bpf_insn *insn = insn_buf;
9998 /* Helper macro for adding read access to tcp_sock or sock fields. */
9999 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
10001 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \
10002 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
10003 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
10004 if (si->dst_reg == reg || si->src_reg == reg) \
10006 if (si->dst_reg == reg || si->src_reg == reg) \
10008 if (si->dst_reg == si->src_reg) { \
10009 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
10010 offsetof(struct bpf_sock_ops_kern, \
10012 fullsock_reg = reg; \
10015 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10016 struct bpf_sock_ops_kern, \
10018 fullsock_reg, si->src_reg, \
10019 offsetof(struct bpf_sock_ops_kern, \
10021 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
10022 if (si->dst_reg == si->src_reg) \
10023 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10024 offsetof(struct bpf_sock_ops_kern, \
10026 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10027 struct bpf_sock_ops_kern, sk),\
10028 si->dst_reg, si->src_reg, \
10029 offsetof(struct bpf_sock_ops_kern, sk));\
10030 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
10032 si->dst_reg, si->dst_reg, \
10033 offsetof(OBJ, OBJ_FIELD)); \
10034 if (si->dst_reg == si->src_reg) { \
10035 *insn++ = BPF_JMP_A(1); \
10036 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10037 offsetof(struct bpf_sock_ops_kern, \
10042 #define SOCK_OPS_GET_SK() \
10044 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \
10045 if (si->dst_reg == reg || si->src_reg == reg) \
10047 if (si->dst_reg == reg || si->src_reg == reg) \
10049 if (si->dst_reg == si->src_reg) { \
10050 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
10051 offsetof(struct bpf_sock_ops_kern, \
10053 fullsock_reg = reg; \
10056 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10057 struct bpf_sock_ops_kern, \
10059 fullsock_reg, si->src_reg, \
10060 offsetof(struct bpf_sock_ops_kern, \
10062 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
10063 if (si->dst_reg == si->src_reg) \
10064 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10065 offsetof(struct bpf_sock_ops_kern, \
10067 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10068 struct bpf_sock_ops_kern, sk),\
10069 si->dst_reg, si->src_reg, \
10070 offsetof(struct bpf_sock_ops_kern, sk));\
10071 if (si->dst_reg == si->src_reg) { \
10072 *insn++ = BPF_JMP_A(1); \
10073 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10074 offsetof(struct bpf_sock_ops_kern, \
10079 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
10080 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
10082 /* Helper macro for adding write access to tcp_sock or sock fields.
10083 * The macro is called with two registers, dst_reg which contains a pointer
10084 * to ctx (context) and src_reg which contains the value that should be
10085 * stored. However, we need an additional register since we cannot overwrite
10086 * dst_reg because it may be used later in the program.
10087 * Instead we "borrow" one of the other register. We first save its value
10088 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
10089 * it at the end of the macro.
10091 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
10093 int reg = BPF_REG_9; \
10094 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
10095 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
10096 if (si->dst_reg == reg || si->src_reg == reg) \
10098 if (si->dst_reg == reg || si->src_reg == reg) \
10100 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
10101 offsetof(struct bpf_sock_ops_kern, \
10103 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10104 struct bpf_sock_ops_kern, \
10106 reg, si->dst_reg, \
10107 offsetof(struct bpf_sock_ops_kern, \
10109 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
10110 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10111 struct bpf_sock_ops_kern, sk),\
10112 reg, si->dst_reg, \
10113 offsetof(struct bpf_sock_ops_kern, sk));\
10114 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \
10115 reg, si->src_reg, \
10116 offsetof(OBJ, OBJ_FIELD)); \
10117 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
10118 offsetof(struct bpf_sock_ops_kern, \
10122 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
10124 if (TYPE == BPF_WRITE) \
10125 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
10127 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
10130 if (insn > insn_buf)
10131 return insn - insn_buf;
10134 case offsetof(struct bpf_sock_ops, op):
10135 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10137 si->dst_reg, si->src_reg,
10138 offsetof(struct bpf_sock_ops_kern, op));
10141 case offsetof(struct bpf_sock_ops, replylong[0]) ...
10142 offsetof(struct bpf_sock_ops, replylong[3]):
10143 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
10144 sizeof_field(struct bpf_sock_ops_kern, reply));
10145 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
10146 sizeof_field(struct bpf_sock_ops_kern, replylong));
10148 off -= offsetof(struct bpf_sock_ops, replylong[0]);
10149 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
10150 if (type == BPF_WRITE)
10151 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
10154 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10158 case offsetof(struct bpf_sock_ops, family):
10159 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10161 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10162 struct bpf_sock_ops_kern, sk),
10163 si->dst_reg, si->src_reg,
10164 offsetof(struct bpf_sock_ops_kern, sk));
10165 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10166 offsetof(struct sock_common, skc_family));
10169 case offsetof(struct bpf_sock_ops, remote_ip4):
10170 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10172 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10173 struct bpf_sock_ops_kern, sk),
10174 si->dst_reg, si->src_reg,
10175 offsetof(struct bpf_sock_ops_kern, sk));
10176 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10177 offsetof(struct sock_common, skc_daddr));
10180 case offsetof(struct bpf_sock_ops, local_ip4):
10181 BUILD_BUG_ON(sizeof_field(struct sock_common,
10182 skc_rcv_saddr) != 4);
10184 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10185 struct bpf_sock_ops_kern, sk),
10186 si->dst_reg, si->src_reg,
10187 offsetof(struct bpf_sock_ops_kern, sk));
10188 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10189 offsetof(struct sock_common,
10193 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
10194 offsetof(struct bpf_sock_ops, remote_ip6[3]):
10195 #if IS_ENABLED(CONFIG_IPV6)
10196 BUILD_BUG_ON(sizeof_field(struct sock_common,
10197 skc_v6_daddr.s6_addr32[0]) != 4);
10200 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
10201 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10202 struct bpf_sock_ops_kern, sk),
10203 si->dst_reg, si->src_reg,
10204 offsetof(struct bpf_sock_ops_kern, sk));
10205 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10206 offsetof(struct sock_common,
10207 skc_v6_daddr.s6_addr32[0]) +
10210 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10214 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
10215 offsetof(struct bpf_sock_ops, local_ip6[3]):
10216 #if IS_ENABLED(CONFIG_IPV6)
10217 BUILD_BUG_ON(sizeof_field(struct sock_common,
10218 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10221 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
10222 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10223 struct bpf_sock_ops_kern, sk),
10224 si->dst_reg, si->src_reg,
10225 offsetof(struct bpf_sock_ops_kern, sk));
10226 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10227 offsetof(struct sock_common,
10228 skc_v6_rcv_saddr.s6_addr32[0]) +
10231 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10235 case offsetof(struct bpf_sock_ops, remote_port):
10236 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10238 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10239 struct bpf_sock_ops_kern, sk),
10240 si->dst_reg, si->src_reg,
10241 offsetof(struct bpf_sock_ops_kern, sk));
10242 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10243 offsetof(struct sock_common, skc_dport));
10244 #ifndef __BIG_ENDIAN_BITFIELD
10245 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10249 case offsetof(struct bpf_sock_ops, local_port):
10250 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10252 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10253 struct bpf_sock_ops_kern, sk),
10254 si->dst_reg, si->src_reg,
10255 offsetof(struct bpf_sock_ops_kern, sk));
10256 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10257 offsetof(struct sock_common, skc_num));
10260 case offsetof(struct bpf_sock_ops, is_fullsock):
10261 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10262 struct bpf_sock_ops_kern,
10264 si->dst_reg, si->src_reg,
10265 offsetof(struct bpf_sock_ops_kern,
10269 case offsetof(struct bpf_sock_ops, state):
10270 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
10272 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10273 struct bpf_sock_ops_kern, sk),
10274 si->dst_reg, si->src_reg,
10275 offsetof(struct bpf_sock_ops_kern, sk));
10276 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
10277 offsetof(struct sock_common, skc_state));
10280 case offsetof(struct bpf_sock_ops, rtt_min):
10281 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
10282 sizeof(struct minmax));
10283 BUILD_BUG_ON(sizeof(struct minmax) <
10284 sizeof(struct minmax_sample));
10286 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10287 struct bpf_sock_ops_kern, sk),
10288 si->dst_reg, si->src_reg,
10289 offsetof(struct bpf_sock_ops_kern, sk));
10290 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10291 offsetof(struct tcp_sock, rtt_min) +
10292 sizeof_field(struct minmax_sample, t));
10295 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
10296 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
10300 case offsetof(struct bpf_sock_ops, sk_txhash):
10301 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
10302 struct sock, type);
10304 case offsetof(struct bpf_sock_ops, snd_cwnd):
10305 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
10307 case offsetof(struct bpf_sock_ops, srtt_us):
10308 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
10310 case offsetof(struct bpf_sock_ops, snd_ssthresh):
10311 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
10313 case offsetof(struct bpf_sock_ops, rcv_nxt):
10314 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
10316 case offsetof(struct bpf_sock_ops, snd_nxt):
10317 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
10319 case offsetof(struct bpf_sock_ops, snd_una):
10320 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
10322 case offsetof(struct bpf_sock_ops, mss_cache):
10323 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
10325 case offsetof(struct bpf_sock_ops, ecn_flags):
10326 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
10328 case offsetof(struct bpf_sock_ops, rate_delivered):
10329 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
10331 case offsetof(struct bpf_sock_ops, rate_interval_us):
10332 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
10334 case offsetof(struct bpf_sock_ops, packets_out):
10335 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
10337 case offsetof(struct bpf_sock_ops, retrans_out):
10338 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
10340 case offsetof(struct bpf_sock_ops, total_retrans):
10341 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
10343 case offsetof(struct bpf_sock_ops, segs_in):
10344 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
10346 case offsetof(struct bpf_sock_ops, data_segs_in):
10347 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
10349 case offsetof(struct bpf_sock_ops, segs_out):
10350 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
10352 case offsetof(struct bpf_sock_ops, data_segs_out):
10353 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
10355 case offsetof(struct bpf_sock_ops, lost_out):
10356 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
10358 case offsetof(struct bpf_sock_ops, sacked_out):
10359 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
10361 case offsetof(struct bpf_sock_ops, bytes_received):
10362 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
10364 case offsetof(struct bpf_sock_ops, bytes_acked):
10365 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
10367 case offsetof(struct bpf_sock_ops, sk):
10370 case offsetof(struct bpf_sock_ops, skb_data_end):
10371 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10373 si->dst_reg, si->src_reg,
10374 offsetof(struct bpf_sock_ops_kern,
10377 case offsetof(struct bpf_sock_ops, skb_data):
10378 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10380 si->dst_reg, si->src_reg,
10381 offsetof(struct bpf_sock_ops_kern,
10383 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10384 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10385 si->dst_reg, si->dst_reg,
10386 offsetof(struct sk_buff, data));
10388 case offsetof(struct bpf_sock_ops, skb_len):
10389 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10391 si->dst_reg, si->src_reg,
10392 offsetof(struct bpf_sock_ops_kern,
10394 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10395 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10396 si->dst_reg, si->dst_reg,
10397 offsetof(struct sk_buff, len));
10399 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
10400 off = offsetof(struct sk_buff, cb);
10401 off += offsetof(struct tcp_skb_cb, tcp_flags);
10402 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
10403 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10405 si->dst_reg, si->src_reg,
10406 offsetof(struct bpf_sock_ops_kern,
10408 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10409 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
10411 si->dst_reg, si->dst_reg, off);
10414 return insn - insn_buf;
10417 /* data_end = skb->data + skb_headlen() */
10418 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
10419 struct bpf_insn *insn)
10422 int temp_reg_off = offsetof(struct sk_buff, cb) +
10423 offsetof(struct sk_skb_cb, temp_reg);
10425 if (si->src_reg == si->dst_reg) {
10426 /* We need an extra register, choose and save a register. */
10428 if (si->src_reg == reg || si->dst_reg == reg)
10430 if (si->src_reg == reg || si->dst_reg == reg)
10432 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off);
10437 /* reg = skb->data */
10438 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10440 offsetof(struct sk_buff, data));
10441 /* AX = skb->len */
10442 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10443 BPF_REG_AX, si->src_reg,
10444 offsetof(struct sk_buff, len));
10445 /* reg = skb->data + skb->len */
10446 *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX);
10447 /* AX = skb->data_len */
10448 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
10449 BPF_REG_AX, si->src_reg,
10450 offsetof(struct sk_buff, data_len));
10452 /* reg = skb->data + skb->len - skb->data_len */
10453 *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX);
10455 if (si->src_reg == si->dst_reg) {
10456 /* Restore the saved register */
10457 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg);
10458 *insn++ = BPF_MOV64_REG(si->dst_reg, reg);
10459 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off);
10465 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
10466 const struct bpf_insn *si,
10467 struct bpf_insn *insn_buf,
10468 struct bpf_prog *prog, u32 *target_size)
10470 struct bpf_insn *insn = insn_buf;
10474 case offsetof(struct __sk_buff, data_end):
10475 insn = bpf_convert_data_end_access(si, insn);
10477 case offsetof(struct __sk_buff, cb[0]) ...
10478 offsetofend(struct __sk_buff, cb[4]) - 1:
10479 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20);
10480 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
10481 offsetof(struct sk_skb_cb, data)) %
10484 prog->cb_access = 1;
10486 off -= offsetof(struct __sk_buff, cb[0]);
10487 off += offsetof(struct sk_buff, cb);
10488 off += offsetof(struct sk_skb_cb, data);
10489 if (type == BPF_WRITE)
10490 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
10493 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
10499 return bpf_convert_ctx_access(type, si, insn_buf, prog,
10503 return insn - insn_buf;
10506 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
10507 const struct bpf_insn *si,
10508 struct bpf_insn *insn_buf,
10509 struct bpf_prog *prog, u32 *target_size)
10511 struct bpf_insn *insn = insn_buf;
10512 #if IS_ENABLED(CONFIG_IPV6)
10516 /* convert ctx uses the fact sg element is first in struct */
10517 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
10520 case offsetof(struct sk_msg_md, data):
10521 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
10522 si->dst_reg, si->src_reg,
10523 offsetof(struct sk_msg, data));
10525 case offsetof(struct sk_msg_md, data_end):
10526 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
10527 si->dst_reg, si->src_reg,
10528 offsetof(struct sk_msg, data_end));
10530 case offsetof(struct sk_msg_md, family):
10531 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10533 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10534 struct sk_msg, sk),
10535 si->dst_reg, si->src_reg,
10536 offsetof(struct sk_msg, sk));
10537 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10538 offsetof(struct sock_common, skc_family));
10541 case offsetof(struct sk_msg_md, remote_ip4):
10542 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10544 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10545 struct sk_msg, sk),
10546 si->dst_reg, si->src_reg,
10547 offsetof(struct sk_msg, sk));
10548 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10549 offsetof(struct sock_common, skc_daddr));
10552 case offsetof(struct sk_msg_md, local_ip4):
10553 BUILD_BUG_ON(sizeof_field(struct sock_common,
10554 skc_rcv_saddr) != 4);
10556 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10557 struct sk_msg, sk),
10558 si->dst_reg, si->src_reg,
10559 offsetof(struct sk_msg, sk));
10560 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10561 offsetof(struct sock_common,
10565 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
10566 offsetof(struct sk_msg_md, remote_ip6[3]):
10567 #if IS_ENABLED(CONFIG_IPV6)
10568 BUILD_BUG_ON(sizeof_field(struct sock_common,
10569 skc_v6_daddr.s6_addr32[0]) != 4);
10572 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
10573 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10574 struct sk_msg, sk),
10575 si->dst_reg, si->src_reg,
10576 offsetof(struct sk_msg, sk));
10577 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10578 offsetof(struct sock_common,
10579 skc_v6_daddr.s6_addr32[0]) +
10582 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10586 case offsetof(struct sk_msg_md, local_ip6[0]) ...
10587 offsetof(struct sk_msg_md, local_ip6[3]):
10588 #if IS_ENABLED(CONFIG_IPV6)
10589 BUILD_BUG_ON(sizeof_field(struct sock_common,
10590 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10593 off -= offsetof(struct sk_msg_md, local_ip6[0]);
10594 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10595 struct sk_msg, sk),
10596 si->dst_reg, si->src_reg,
10597 offsetof(struct sk_msg, sk));
10598 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10599 offsetof(struct sock_common,
10600 skc_v6_rcv_saddr.s6_addr32[0]) +
10603 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10607 case offsetof(struct sk_msg_md, remote_port):
10608 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10610 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10611 struct sk_msg, sk),
10612 si->dst_reg, si->src_reg,
10613 offsetof(struct sk_msg, sk));
10614 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10615 offsetof(struct sock_common, skc_dport));
10616 #ifndef __BIG_ENDIAN_BITFIELD
10617 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10621 case offsetof(struct sk_msg_md, local_port):
10622 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10624 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10625 struct sk_msg, sk),
10626 si->dst_reg, si->src_reg,
10627 offsetof(struct sk_msg, sk));
10628 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10629 offsetof(struct sock_common, skc_num));
10632 case offsetof(struct sk_msg_md, size):
10633 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
10634 si->dst_reg, si->src_reg,
10635 offsetof(struct sk_msg_sg, size));
10638 case offsetof(struct sk_msg_md, sk):
10639 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
10640 si->dst_reg, si->src_reg,
10641 offsetof(struct sk_msg, sk));
10645 return insn - insn_buf;
10648 const struct bpf_verifier_ops sk_filter_verifier_ops = {
10649 .get_func_proto = sk_filter_func_proto,
10650 .is_valid_access = sk_filter_is_valid_access,
10651 .convert_ctx_access = bpf_convert_ctx_access,
10652 .gen_ld_abs = bpf_gen_ld_abs,
10655 const struct bpf_prog_ops sk_filter_prog_ops = {
10656 .test_run = bpf_prog_test_run_skb,
10659 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
10660 .get_func_proto = tc_cls_act_func_proto,
10661 .is_valid_access = tc_cls_act_is_valid_access,
10662 .convert_ctx_access = tc_cls_act_convert_ctx_access,
10663 .gen_prologue = tc_cls_act_prologue,
10664 .gen_ld_abs = bpf_gen_ld_abs,
10667 const struct bpf_prog_ops tc_cls_act_prog_ops = {
10668 .test_run = bpf_prog_test_run_skb,
10671 const struct bpf_verifier_ops xdp_verifier_ops = {
10672 .get_func_proto = xdp_func_proto,
10673 .is_valid_access = xdp_is_valid_access,
10674 .convert_ctx_access = xdp_convert_ctx_access,
10675 .gen_prologue = bpf_noop_prologue,
10678 const struct bpf_prog_ops xdp_prog_ops = {
10679 .test_run = bpf_prog_test_run_xdp,
10682 const struct bpf_verifier_ops cg_skb_verifier_ops = {
10683 .get_func_proto = cg_skb_func_proto,
10684 .is_valid_access = cg_skb_is_valid_access,
10685 .convert_ctx_access = bpf_convert_ctx_access,
10688 const struct bpf_prog_ops cg_skb_prog_ops = {
10689 .test_run = bpf_prog_test_run_skb,
10692 const struct bpf_verifier_ops lwt_in_verifier_ops = {
10693 .get_func_proto = lwt_in_func_proto,
10694 .is_valid_access = lwt_is_valid_access,
10695 .convert_ctx_access = bpf_convert_ctx_access,
10698 const struct bpf_prog_ops lwt_in_prog_ops = {
10699 .test_run = bpf_prog_test_run_skb,
10702 const struct bpf_verifier_ops lwt_out_verifier_ops = {
10703 .get_func_proto = lwt_out_func_proto,
10704 .is_valid_access = lwt_is_valid_access,
10705 .convert_ctx_access = bpf_convert_ctx_access,
10708 const struct bpf_prog_ops lwt_out_prog_ops = {
10709 .test_run = bpf_prog_test_run_skb,
10712 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
10713 .get_func_proto = lwt_xmit_func_proto,
10714 .is_valid_access = lwt_is_valid_access,
10715 .convert_ctx_access = bpf_convert_ctx_access,
10716 .gen_prologue = tc_cls_act_prologue,
10719 const struct bpf_prog_ops lwt_xmit_prog_ops = {
10720 .test_run = bpf_prog_test_run_skb,
10723 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
10724 .get_func_proto = lwt_seg6local_func_proto,
10725 .is_valid_access = lwt_is_valid_access,
10726 .convert_ctx_access = bpf_convert_ctx_access,
10729 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
10730 .test_run = bpf_prog_test_run_skb,
10733 const struct bpf_verifier_ops cg_sock_verifier_ops = {
10734 .get_func_proto = sock_filter_func_proto,
10735 .is_valid_access = sock_filter_is_valid_access,
10736 .convert_ctx_access = bpf_sock_convert_ctx_access,
10739 const struct bpf_prog_ops cg_sock_prog_ops = {
10742 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
10743 .get_func_proto = sock_addr_func_proto,
10744 .is_valid_access = sock_addr_is_valid_access,
10745 .convert_ctx_access = sock_addr_convert_ctx_access,
10748 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
10751 const struct bpf_verifier_ops sock_ops_verifier_ops = {
10752 .get_func_proto = sock_ops_func_proto,
10753 .is_valid_access = sock_ops_is_valid_access,
10754 .convert_ctx_access = sock_ops_convert_ctx_access,
10757 const struct bpf_prog_ops sock_ops_prog_ops = {
10760 const struct bpf_verifier_ops sk_skb_verifier_ops = {
10761 .get_func_proto = sk_skb_func_proto,
10762 .is_valid_access = sk_skb_is_valid_access,
10763 .convert_ctx_access = sk_skb_convert_ctx_access,
10764 .gen_prologue = sk_skb_prologue,
10767 const struct bpf_prog_ops sk_skb_prog_ops = {
10770 const struct bpf_verifier_ops sk_msg_verifier_ops = {
10771 .get_func_proto = sk_msg_func_proto,
10772 .is_valid_access = sk_msg_is_valid_access,
10773 .convert_ctx_access = sk_msg_convert_ctx_access,
10774 .gen_prologue = bpf_noop_prologue,
10777 const struct bpf_prog_ops sk_msg_prog_ops = {
10780 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
10781 .get_func_proto = flow_dissector_func_proto,
10782 .is_valid_access = flow_dissector_is_valid_access,
10783 .convert_ctx_access = flow_dissector_convert_ctx_access,
10786 const struct bpf_prog_ops flow_dissector_prog_ops = {
10787 .test_run = bpf_prog_test_run_flow_dissector,
10790 int sk_detach_filter(struct sock *sk)
10793 struct sk_filter *filter;
10795 if (sock_flag(sk, SOCK_FILTER_LOCKED))
10798 filter = rcu_dereference_protected(sk->sk_filter,
10799 lockdep_sock_is_held(sk));
10801 RCU_INIT_POINTER(sk->sk_filter, NULL);
10802 sk_filter_uncharge(sk, filter);
10808 EXPORT_SYMBOL_GPL(sk_detach_filter);
10810 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
10813 struct sock_fprog_kern *fprog;
10814 struct sk_filter *filter;
10818 filter = rcu_dereference_protected(sk->sk_filter,
10819 lockdep_sock_is_held(sk));
10823 /* We're copying the filter that has been originally attached,
10824 * so no conversion/decode needed anymore. eBPF programs that
10825 * have no original program cannot be dumped through this.
10828 fprog = filter->prog->orig_prog;
10834 /* User space only enquires number of filter blocks. */
10838 if (len < fprog->len)
10842 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
10845 /* Instead of bytes, the API requests to return the number
10846 * of filter blocks.
10855 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
10856 struct sock_reuseport *reuse,
10857 struct sock *sk, struct sk_buff *skb,
10858 struct sock *migrating_sk,
10861 reuse_kern->skb = skb;
10862 reuse_kern->sk = sk;
10863 reuse_kern->selected_sk = NULL;
10864 reuse_kern->migrating_sk = migrating_sk;
10865 reuse_kern->data_end = skb->data + skb_headlen(skb);
10866 reuse_kern->hash = hash;
10867 reuse_kern->reuseport_id = reuse->reuseport_id;
10868 reuse_kern->bind_inany = reuse->bind_inany;
10871 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
10872 struct bpf_prog *prog, struct sk_buff *skb,
10873 struct sock *migrating_sk,
10876 struct sk_reuseport_kern reuse_kern;
10877 enum sk_action action;
10879 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
10880 action = bpf_prog_run(prog, &reuse_kern);
10882 if (action == SK_PASS)
10883 return reuse_kern.selected_sk;
10885 return ERR_PTR(-ECONNREFUSED);
10888 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
10889 struct bpf_map *, map, void *, key, u32, flags)
10891 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
10892 struct sock_reuseport *reuse;
10893 struct sock *selected_sk;
10895 selected_sk = map->ops->map_lookup_elem(map, key);
10899 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
10901 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
10902 if (sk_is_refcounted(selected_sk))
10903 sock_put(selected_sk);
10905 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
10906 * The only (!reuse) case here is - the sk has already been
10907 * unhashed (e.g. by close()), so treat it as -ENOENT.
10909 * Other maps (e.g. sock_map) do not provide this guarantee and
10910 * the sk may never be in the reuseport group to begin with.
10912 return is_sockarray ? -ENOENT : -EINVAL;
10915 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
10916 struct sock *sk = reuse_kern->sk;
10918 if (sk->sk_protocol != selected_sk->sk_protocol)
10919 return -EPROTOTYPE;
10920 else if (sk->sk_family != selected_sk->sk_family)
10921 return -EAFNOSUPPORT;
10923 /* Catch all. Likely bound to a different sockaddr. */
10927 reuse_kern->selected_sk = selected_sk;
10932 static const struct bpf_func_proto sk_select_reuseport_proto = {
10933 .func = sk_select_reuseport,
10935 .ret_type = RET_INTEGER,
10936 .arg1_type = ARG_PTR_TO_CTX,
10937 .arg2_type = ARG_CONST_MAP_PTR,
10938 .arg3_type = ARG_PTR_TO_MAP_KEY,
10939 .arg4_type = ARG_ANYTHING,
10942 BPF_CALL_4(sk_reuseport_load_bytes,
10943 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10944 void *, to, u32, len)
10946 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
10949 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
10950 .func = sk_reuseport_load_bytes,
10952 .ret_type = RET_INTEGER,
10953 .arg1_type = ARG_PTR_TO_CTX,
10954 .arg2_type = ARG_ANYTHING,
10955 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
10956 .arg4_type = ARG_CONST_SIZE,
10959 BPF_CALL_5(sk_reuseport_load_bytes_relative,
10960 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10961 void *, to, u32, len, u32, start_header)
10963 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
10964 len, start_header);
10967 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
10968 .func = sk_reuseport_load_bytes_relative,
10970 .ret_type = RET_INTEGER,
10971 .arg1_type = ARG_PTR_TO_CTX,
10972 .arg2_type = ARG_ANYTHING,
10973 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
10974 .arg4_type = ARG_CONST_SIZE,
10975 .arg5_type = ARG_ANYTHING,
10978 static const struct bpf_func_proto *
10979 sk_reuseport_func_proto(enum bpf_func_id func_id,
10980 const struct bpf_prog *prog)
10983 case BPF_FUNC_sk_select_reuseport:
10984 return &sk_select_reuseport_proto;
10985 case BPF_FUNC_skb_load_bytes:
10986 return &sk_reuseport_load_bytes_proto;
10987 case BPF_FUNC_skb_load_bytes_relative:
10988 return &sk_reuseport_load_bytes_relative_proto;
10989 case BPF_FUNC_get_socket_cookie:
10990 return &bpf_get_socket_ptr_cookie_proto;
10991 case BPF_FUNC_ktime_get_coarse_ns:
10992 return &bpf_ktime_get_coarse_ns_proto;
10994 return bpf_base_func_proto(func_id);
10999 sk_reuseport_is_valid_access(int off, int size,
11000 enum bpf_access_type type,
11001 const struct bpf_prog *prog,
11002 struct bpf_insn_access_aux *info)
11004 const u32 size_default = sizeof(__u32);
11006 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
11007 off % size || type != BPF_READ)
11011 case offsetof(struct sk_reuseport_md, data):
11012 info->reg_type = PTR_TO_PACKET;
11013 return size == sizeof(__u64);
11015 case offsetof(struct sk_reuseport_md, data_end):
11016 info->reg_type = PTR_TO_PACKET_END;
11017 return size == sizeof(__u64);
11019 case offsetof(struct sk_reuseport_md, hash):
11020 return size == size_default;
11022 case offsetof(struct sk_reuseport_md, sk):
11023 info->reg_type = PTR_TO_SOCKET;
11024 return size == sizeof(__u64);
11026 case offsetof(struct sk_reuseport_md, migrating_sk):
11027 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
11028 return size == sizeof(__u64);
11030 /* Fields that allow narrowing */
11031 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
11032 if (size < sizeof_field(struct sk_buff, protocol))
11035 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
11036 case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
11037 case bpf_ctx_range(struct sk_reuseport_md, len):
11038 bpf_ctx_record_field_size(info, size_default);
11039 return bpf_ctx_narrow_access_ok(off, size, size_default);
11046 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
11047 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
11048 si->dst_reg, si->src_reg, \
11049 bpf_target_off(struct sk_reuseport_kern, F, \
11050 sizeof_field(struct sk_reuseport_kern, F), \
11054 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
11055 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
11060 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \
11061 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
11066 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
11067 const struct bpf_insn *si,
11068 struct bpf_insn *insn_buf,
11069 struct bpf_prog *prog,
11072 struct bpf_insn *insn = insn_buf;
11075 case offsetof(struct sk_reuseport_md, data):
11076 SK_REUSEPORT_LOAD_SKB_FIELD(data);
11079 case offsetof(struct sk_reuseport_md, len):
11080 SK_REUSEPORT_LOAD_SKB_FIELD(len);
11083 case offsetof(struct sk_reuseport_md, eth_protocol):
11084 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
11087 case offsetof(struct sk_reuseport_md, ip_protocol):
11088 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
11091 case offsetof(struct sk_reuseport_md, data_end):
11092 SK_REUSEPORT_LOAD_FIELD(data_end);
11095 case offsetof(struct sk_reuseport_md, hash):
11096 SK_REUSEPORT_LOAD_FIELD(hash);
11099 case offsetof(struct sk_reuseport_md, bind_inany):
11100 SK_REUSEPORT_LOAD_FIELD(bind_inany);
11103 case offsetof(struct sk_reuseport_md, sk):
11104 SK_REUSEPORT_LOAD_FIELD(sk);
11107 case offsetof(struct sk_reuseport_md, migrating_sk):
11108 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
11112 return insn - insn_buf;
11115 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
11116 .get_func_proto = sk_reuseport_func_proto,
11117 .is_valid_access = sk_reuseport_is_valid_access,
11118 .convert_ctx_access = sk_reuseport_convert_ctx_access,
11121 const struct bpf_prog_ops sk_reuseport_prog_ops = {
11124 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
11125 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
11127 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
11128 struct sock *, sk, u64, flags)
11130 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
11131 BPF_SK_LOOKUP_F_NO_REUSEPORT)))
11133 if (unlikely(sk && sk_is_refcounted(sk)))
11134 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
11135 if (unlikely(sk && sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN))
11136 return -ESOCKTNOSUPPORT; /* only accept TCP socket in LISTEN */
11137 if (unlikely(sk && sk_is_udp(sk) && sk->sk_state != TCP_CLOSE))
11138 return -ESOCKTNOSUPPORT; /* only accept UDP socket in CLOSE */
11140 /* Check if socket is suitable for packet L3/L4 protocol */
11141 if (sk && sk->sk_protocol != ctx->protocol)
11142 return -EPROTOTYPE;
11143 if (sk && sk->sk_family != ctx->family &&
11144 (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
11145 return -EAFNOSUPPORT;
11147 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
11150 /* Select socket as lookup result */
11151 ctx->selected_sk = sk;
11152 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
11156 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
11157 .func = bpf_sk_lookup_assign,
11159 .ret_type = RET_INTEGER,
11160 .arg1_type = ARG_PTR_TO_CTX,
11161 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL,
11162 .arg3_type = ARG_ANYTHING,
11165 static const struct bpf_func_proto *
11166 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
11169 case BPF_FUNC_perf_event_output:
11170 return &bpf_event_output_data_proto;
11171 case BPF_FUNC_sk_assign:
11172 return &bpf_sk_lookup_assign_proto;
11173 case BPF_FUNC_sk_release:
11174 return &bpf_sk_release_proto;
11176 return bpf_sk_base_func_proto(func_id);
11180 static bool sk_lookup_is_valid_access(int off, int size,
11181 enum bpf_access_type type,
11182 const struct bpf_prog *prog,
11183 struct bpf_insn_access_aux *info)
11185 if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
11187 if (off % size != 0)
11189 if (type != BPF_READ)
11193 case offsetof(struct bpf_sk_lookup, sk):
11194 info->reg_type = PTR_TO_SOCKET_OR_NULL;
11195 return size == sizeof(__u64);
11197 case bpf_ctx_range(struct bpf_sk_lookup, family):
11198 case bpf_ctx_range(struct bpf_sk_lookup, protocol):
11199 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
11200 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
11201 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
11202 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
11203 case bpf_ctx_range(struct bpf_sk_lookup, local_port):
11204 case bpf_ctx_range(struct bpf_sk_lookup, ingress_ifindex):
11205 bpf_ctx_record_field_size(info, sizeof(__u32));
11206 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
11208 case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
11209 /* Allow 4-byte access to 2-byte field for backward compatibility */
11210 if (size == sizeof(__u32))
11212 bpf_ctx_record_field_size(info, sizeof(__be16));
11213 return bpf_ctx_narrow_access_ok(off, size, sizeof(__be16));
11215 case offsetofend(struct bpf_sk_lookup, remote_port) ...
11216 offsetof(struct bpf_sk_lookup, local_ip4) - 1:
11217 /* Allow access to zero padding for backward compatibility */
11218 bpf_ctx_record_field_size(info, sizeof(__u16));
11219 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u16));
11226 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
11227 const struct bpf_insn *si,
11228 struct bpf_insn *insn_buf,
11229 struct bpf_prog *prog,
11232 struct bpf_insn *insn = insn_buf;
11235 case offsetof(struct bpf_sk_lookup, sk):
11236 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11237 offsetof(struct bpf_sk_lookup_kern, selected_sk));
11240 case offsetof(struct bpf_sk_lookup, family):
11241 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11242 bpf_target_off(struct bpf_sk_lookup_kern,
11243 family, 2, target_size));
11246 case offsetof(struct bpf_sk_lookup, protocol):
11247 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11248 bpf_target_off(struct bpf_sk_lookup_kern,
11249 protocol, 2, target_size));
11252 case offsetof(struct bpf_sk_lookup, remote_ip4):
11253 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11254 bpf_target_off(struct bpf_sk_lookup_kern,
11255 v4.saddr, 4, target_size));
11258 case offsetof(struct bpf_sk_lookup, local_ip4):
11259 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11260 bpf_target_off(struct bpf_sk_lookup_kern,
11261 v4.daddr, 4, target_size));
11264 case bpf_ctx_range_till(struct bpf_sk_lookup,
11265 remote_ip6[0], remote_ip6[3]): {
11266 #if IS_ENABLED(CONFIG_IPV6)
11269 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
11270 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11271 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11272 offsetof(struct bpf_sk_lookup_kern, v6.saddr));
11273 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11274 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11276 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11280 case bpf_ctx_range_till(struct bpf_sk_lookup,
11281 local_ip6[0], local_ip6[3]): {
11282 #if IS_ENABLED(CONFIG_IPV6)
11285 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
11286 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11287 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11288 offsetof(struct bpf_sk_lookup_kern, v6.daddr));
11289 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11290 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11292 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11296 case offsetof(struct bpf_sk_lookup, remote_port):
11297 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11298 bpf_target_off(struct bpf_sk_lookup_kern,
11299 sport, 2, target_size));
11302 case offsetofend(struct bpf_sk_lookup, remote_port):
11304 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11307 case offsetof(struct bpf_sk_lookup, local_port):
11308 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11309 bpf_target_off(struct bpf_sk_lookup_kern,
11310 dport, 2, target_size));
11313 case offsetof(struct bpf_sk_lookup, ingress_ifindex):
11314 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11315 bpf_target_off(struct bpf_sk_lookup_kern,
11316 ingress_ifindex, 4, target_size));
11320 return insn - insn_buf;
11323 const struct bpf_prog_ops sk_lookup_prog_ops = {
11324 .test_run = bpf_prog_test_run_sk_lookup,
11327 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
11328 .get_func_proto = sk_lookup_func_proto,
11329 .is_valid_access = sk_lookup_is_valid_access,
11330 .convert_ctx_access = sk_lookup_convert_ctx_access,
11333 #endif /* CONFIG_INET */
11335 DEFINE_BPF_DISPATCHER(xdp)
11337 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
11339 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
11342 BTF_ID_LIST_GLOBAL(btf_sock_ids, MAX_BTF_SOCK_TYPE)
11343 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
11345 #undef BTF_SOCK_TYPE
11347 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
11349 /* tcp6_sock type is not generated in dwarf and hence btf,
11350 * trigger an explicit type generation here.
11352 BTF_TYPE_EMIT(struct tcp6_sock);
11353 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
11354 sk->sk_family == AF_INET6)
11355 return (unsigned long)sk;
11357 return (unsigned long)NULL;
11360 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
11361 .func = bpf_skc_to_tcp6_sock,
11363 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11364 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11365 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
11368 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
11370 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
11371 return (unsigned long)sk;
11373 return (unsigned long)NULL;
11376 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
11377 .func = bpf_skc_to_tcp_sock,
11379 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11380 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11381 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
11384 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
11386 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
11387 * generated if CONFIG_INET=n. Trigger an explicit generation here.
11389 BTF_TYPE_EMIT(struct inet_timewait_sock);
11390 BTF_TYPE_EMIT(struct tcp_timewait_sock);
11393 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
11394 return (unsigned long)sk;
11397 #if IS_BUILTIN(CONFIG_IPV6)
11398 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
11399 return (unsigned long)sk;
11402 return (unsigned long)NULL;
11405 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
11406 .func = bpf_skc_to_tcp_timewait_sock,
11408 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11409 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11410 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
11413 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
11416 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11417 return (unsigned long)sk;
11420 #if IS_BUILTIN(CONFIG_IPV6)
11421 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11422 return (unsigned long)sk;
11425 return (unsigned long)NULL;
11428 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
11429 .func = bpf_skc_to_tcp_request_sock,
11431 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11432 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11433 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
11436 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
11438 /* udp6_sock type is not generated in dwarf and hence btf,
11439 * trigger an explicit type generation here.
11441 BTF_TYPE_EMIT(struct udp6_sock);
11442 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
11443 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
11444 return (unsigned long)sk;
11446 return (unsigned long)NULL;
11449 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
11450 .func = bpf_skc_to_udp6_sock,
11452 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11453 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11454 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
11457 BPF_CALL_1(bpf_skc_to_unix_sock, struct sock *, sk)
11459 /* unix_sock type is not generated in dwarf and hence btf,
11460 * trigger an explicit type generation here.
11462 BTF_TYPE_EMIT(struct unix_sock);
11463 if (sk && sk_fullsock(sk) && sk->sk_family == AF_UNIX)
11464 return (unsigned long)sk;
11466 return (unsigned long)NULL;
11469 const struct bpf_func_proto bpf_skc_to_unix_sock_proto = {
11470 .func = bpf_skc_to_unix_sock,
11472 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11473 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11474 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UNIX],
11477 BPF_CALL_1(bpf_skc_to_mptcp_sock, struct sock *, sk)
11479 BTF_TYPE_EMIT(struct mptcp_sock);
11480 return (unsigned long)bpf_mptcp_sock_from_subflow(sk);
11483 const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto = {
11484 .func = bpf_skc_to_mptcp_sock,
11486 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11487 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
11488 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_MPTCP],
11491 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
11493 return (unsigned long)sock_from_file(file);
11496 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
11497 BTF_ID(struct, socket)
11498 BTF_ID(struct, file)
11500 const struct bpf_func_proto bpf_sock_from_file_proto = {
11501 .func = bpf_sock_from_file,
11503 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11504 .ret_btf_id = &bpf_sock_from_file_btf_ids[0],
11505 .arg1_type = ARG_PTR_TO_BTF_ID,
11506 .arg1_btf_id = &bpf_sock_from_file_btf_ids[1],
11509 static const struct bpf_func_proto *
11510 bpf_sk_base_func_proto(enum bpf_func_id func_id)
11512 const struct bpf_func_proto *func;
11515 case BPF_FUNC_skc_to_tcp6_sock:
11516 func = &bpf_skc_to_tcp6_sock_proto;
11518 case BPF_FUNC_skc_to_tcp_sock:
11519 func = &bpf_skc_to_tcp_sock_proto;
11521 case BPF_FUNC_skc_to_tcp_timewait_sock:
11522 func = &bpf_skc_to_tcp_timewait_sock_proto;
11524 case BPF_FUNC_skc_to_tcp_request_sock:
11525 func = &bpf_skc_to_tcp_request_sock_proto;
11527 case BPF_FUNC_skc_to_udp6_sock:
11528 func = &bpf_skc_to_udp6_sock_proto;
11530 case BPF_FUNC_skc_to_unix_sock:
11531 func = &bpf_skc_to_unix_sock_proto;
11533 case BPF_FUNC_skc_to_mptcp_sock:
11534 func = &bpf_skc_to_mptcp_sock_proto;
11536 case BPF_FUNC_ktime_get_coarse_ns:
11537 return &bpf_ktime_get_coarse_ns_proto;
11539 return bpf_base_func_proto(func_id);
11542 if (!perfmon_capable())