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/bpf_verifier.h>
22 #include <linux/module.h>
23 #include <linux/types.h>
25 #include <linux/fcntl.h>
26 #include <linux/socket.h>
27 #include <linux/sock_diag.h>
29 #include <linux/inet.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_packet.h>
32 #include <linux/if_arp.h>
33 #include <linux/gfp.h>
34 #include <net/inet_common.h>
36 #include <net/protocol.h>
37 #include <net/netlink.h>
38 #include <linux/skbuff.h>
39 #include <linux/skmsg.h>
41 #include <net/flow_dissector.h>
42 #include <linux/errno.h>
43 #include <linux/timer.h>
44 #include <linux/uaccess.h>
45 #include <asm/unaligned.h>
46 #include <linux/filter.h>
47 #include <linux/ratelimit.h>
48 #include <linux/seccomp.h>
49 #include <linux/if_vlan.h>
50 #include <linux/bpf.h>
51 #include <linux/btf.h>
52 #include <net/sch_generic.h>
53 #include <net/cls_cgroup.h>
54 #include <net/dst_metadata.h>
56 #include <net/sock_reuseport.h>
57 #include <net/busy_poll.h>
61 #include <linux/bpf_trace.h>
62 #include <net/xdp_sock.h>
63 #include <linux/inetdevice.h>
64 #include <net/inet_hashtables.h>
65 #include <net/inet6_hashtables.h>
66 #include <net/ip_fib.h>
67 #include <net/nexthop.h>
71 #include <net/net_namespace.h>
72 #include <linux/seg6_local.h>
74 #include <net/seg6_local.h>
75 #include <net/lwtunnel.h>
76 #include <net/ipv6_stubs.h>
77 #include <net/bpf_sk_storage.h>
78 #include <net/transp_v6.h>
79 #include <linux/btf_ids.h>
82 #include <net/mptcp.h>
83 #include <net/netfilter/nf_conntrack_bpf.h>
85 static const struct bpf_func_proto *
86 bpf_sk_base_func_proto(enum bpf_func_id func_id);
88 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len)
90 if (in_compat_syscall()) {
91 struct compat_sock_fprog f32;
93 if (len != sizeof(f32))
95 if (copy_from_sockptr(&f32, src, sizeof(f32)))
97 memset(dst, 0, sizeof(*dst));
99 dst->filter = compat_ptr(f32.filter);
101 if (len != sizeof(*dst))
103 if (copy_from_sockptr(dst, src, sizeof(*dst)))
109 EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user);
112 * sk_filter_trim_cap - run a packet through a socket filter
113 * @sk: sock associated with &sk_buff
114 * @skb: buffer to filter
115 * @cap: limit on how short the eBPF program may trim the packet
117 * Run the eBPF program and then cut skb->data to correct size returned by
118 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
119 * than pkt_len we keep whole skb->data. This is the socket level
120 * wrapper to bpf_prog_run. It returns 0 if the packet should
121 * be accepted or -EPERM if the packet should be tossed.
124 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
127 struct sk_filter *filter;
130 * If the skb was allocated from pfmemalloc reserves, only
131 * allow SOCK_MEMALLOC sockets to use it as this socket is
132 * helping free memory
134 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
135 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
138 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
142 err = security_sock_rcv_skb(sk, skb);
147 filter = rcu_dereference(sk->sk_filter);
149 struct sock *save_sk = skb->sk;
150 unsigned int pkt_len;
153 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
155 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
161 EXPORT_SYMBOL(sk_filter_trim_cap);
163 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
165 return skb_get_poff(skb);
168 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
172 if (skb_is_nonlinear(skb))
175 if (skb->len < sizeof(struct nlattr))
178 if (a > skb->len - sizeof(struct nlattr))
181 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
183 return (void *) nla - (void *) skb->data;
188 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
192 if (skb_is_nonlinear(skb))
195 if (skb->len < sizeof(struct nlattr))
198 if (a > skb->len - sizeof(struct nlattr))
201 nla = (struct nlattr *) &skb->data[a];
202 if (nla->nla_len > skb->len - a)
205 nla = nla_find_nested(nla, x);
207 return (void *) nla - (void *) skb->data;
212 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
213 data, int, headlen, int, offset)
216 const int len = sizeof(tmp);
219 if (headlen - offset >= len)
220 return *(u8 *)(data + offset);
221 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
224 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
232 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
235 return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
239 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
240 data, int, headlen, int, offset)
243 const int len = sizeof(tmp);
246 if (headlen - offset >= len)
247 return get_unaligned_be16(data + offset);
248 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
249 return be16_to_cpu(tmp);
251 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
253 return get_unaligned_be16(ptr);
259 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
262 return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
266 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
267 data, int, headlen, int, offset)
270 const int len = sizeof(tmp);
272 if (likely(offset >= 0)) {
273 if (headlen - offset >= len)
274 return get_unaligned_be32(data + offset);
275 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
276 return be32_to_cpu(tmp);
278 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
280 return get_unaligned_be32(ptr);
286 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
289 return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
293 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
294 struct bpf_insn *insn_buf)
296 struct bpf_insn *insn = insn_buf;
300 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4);
302 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
303 offsetof(struct sk_buff, mark));
307 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET);
308 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
309 #ifdef __BIG_ENDIAN_BITFIELD
310 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
315 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2);
317 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
318 offsetof(struct sk_buff, queue_mapping));
321 case SKF_AD_VLAN_TAG:
322 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2);
324 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
325 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
326 offsetof(struct sk_buff, vlan_tci));
328 case SKF_AD_VLAN_TAG_PRESENT:
329 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_VLAN_PRESENT_OFFSET);
330 if (PKT_VLAN_PRESENT_BIT)
331 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, PKT_VLAN_PRESENT_BIT);
332 if (PKT_VLAN_PRESENT_BIT < 7)
333 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
337 return insn - insn_buf;
340 static bool convert_bpf_extensions(struct sock_filter *fp,
341 struct bpf_insn **insnp)
343 struct bpf_insn *insn = *insnp;
347 case SKF_AD_OFF + SKF_AD_PROTOCOL:
348 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2);
350 /* A = *(u16 *) (CTX + offsetof(protocol)) */
351 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
352 offsetof(struct sk_buff, protocol));
353 /* A = ntohs(A) [emitting a nop or swap16] */
354 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
357 case SKF_AD_OFF + SKF_AD_PKTTYPE:
358 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
362 case SKF_AD_OFF + SKF_AD_IFINDEX:
363 case SKF_AD_OFF + SKF_AD_HATYPE:
364 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4);
365 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2);
367 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
368 BPF_REG_TMP, BPF_REG_CTX,
369 offsetof(struct sk_buff, dev));
370 /* if (tmp != 0) goto pc + 1 */
371 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
372 *insn++ = BPF_EXIT_INSN();
373 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
374 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
375 offsetof(struct net_device, ifindex));
377 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
378 offsetof(struct net_device, type));
381 case SKF_AD_OFF + SKF_AD_MARK:
382 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
386 case SKF_AD_OFF + SKF_AD_RXHASH:
387 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4);
389 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
390 offsetof(struct sk_buff, hash));
393 case SKF_AD_OFF + SKF_AD_QUEUE:
394 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
398 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
399 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
400 BPF_REG_A, BPF_REG_CTX, insn);
404 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
405 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
406 BPF_REG_A, BPF_REG_CTX, insn);
410 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
411 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2);
413 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
414 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
415 offsetof(struct sk_buff, vlan_proto));
416 /* A = ntohs(A) [emitting a nop or swap16] */
417 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
420 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
421 case SKF_AD_OFF + SKF_AD_NLATTR:
422 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
423 case SKF_AD_OFF + SKF_AD_CPU:
424 case SKF_AD_OFF + SKF_AD_RANDOM:
426 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
428 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
430 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
431 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
433 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
434 *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
436 case SKF_AD_OFF + SKF_AD_NLATTR:
437 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
439 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
440 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
442 case SKF_AD_OFF + SKF_AD_CPU:
443 *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
445 case SKF_AD_OFF + SKF_AD_RANDOM:
446 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
447 bpf_user_rnd_init_once();
452 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
454 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
458 /* This is just a dummy call to avoid letting the compiler
459 * evict __bpf_call_base() as an optimization. Placed here
460 * where no-one bothers.
462 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
470 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
472 const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
473 int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
474 bool endian = BPF_SIZE(fp->code) == BPF_H ||
475 BPF_SIZE(fp->code) == BPF_W;
476 bool indirect = BPF_MODE(fp->code) == BPF_IND;
477 const int ip_align = NET_IP_ALIGN;
478 struct bpf_insn *insn = *insnp;
482 ((unaligned_ok && offset >= 0) ||
483 (!unaligned_ok && offset >= 0 &&
484 offset + ip_align >= 0 &&
485 offset + ip_align % size == 0))) {
486 bool ldx_off_ok = offset <= S16_MAX;
488 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
490 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
491 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
492 size, 2 + endian + (!ldx_off_ok * 2));
494 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
497 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
498 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
499 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
503 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
504 *insn++ = BPF_JMP_A(8);
507 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
508 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
509 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
511 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
513 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
515 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
518 switch (BPF_SIZE(fp->code)) {
520 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
523 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
526 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
532 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
533 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
534 *insn = BPF_EXIT_INSN();
541 * bpf_convert_filter - convert filter program
542 * @prog: the user passed filter program
543 * @len: the length of the user passed filter program
544 * @new_prog: allocated 'struct bpf_prog' or NULL
545 * @new_len: pointer to store length of converted program
546 * @seen_ld_abs: bool whether we've seen ld_abs/ind
548 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
549 * style extended BPF (eBPF).
550 * Conversion workflow:
552 * 1) First pass for calculating the new program length:
553 * bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
555 * 2) 2nd pass to remap in two passes: 1st pass finds new
556 * jump offsets, 2nd pass remapping:
557 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
559 static int bpf_convert_filter(struct sock_filter *prog, int len,
560 struct bpf_prog *new_prog, int *new_len,
563 int new_flen = 0, pass = 0, target, i, stack_off;
564 struct bpf_insn *new_insn, *first_insn = NULL;
565 struct sock_filter *fp;
569 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
570 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
572 if (len <= 0 || len > BPF_MAXINSNS)
576 first_insn = new_prog->insnsi;
577 addrs = kcalloc(len, sizeof(*addrs),
578 GFP_KERNEL | __GFP_NOWARN);
584 new_insn = first_insn;
587 /* Classic BPF related prologue emission. */
589 /* Classic BPF expects A and X to be reset first. These need
590 * to be guaranteed to be the first two instructions.
592 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
593 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
595 /* All programs must keep CTX in callee saved BPF_REG_CTX.
596 * In eBPF case it's done by the compiler, here we need to
597 * do this ourself. Initial CTX is present in BPF_REG_ARG1.
599 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
601 /* For packet access in classic BPF, cache skb->data
602 * in callee-saved BPF R8 and skb->len - skb->data_len
603 * (headlen) in BPF R9. Since classic BPF is read-only
604 * on CTX, we only need to cache it once.
606 *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
607 BPF_REG_D, BPF_REG_CTX,
608 offsetof(struct sk_buff, data));
609 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
610 offsetof(struct sk_buff, len));
611 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
612 offsetof(struct sk_buff, data_len));
613 *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
619 for (i = 0; i < len; fp++, i++) {
620 struct bpf_insn tmp_insns[32] = { };
621 struct bpf_insn *insn = tmp_insns;
624 addrs[i] = new_insn - first_insn;
627 /* All arithmetic insns and skb loads map as-is. */
628 case BPF_ALU | BPF_ADD | BPF_X:
629 case BPF_ALU | BPF_ADD | BPF_K:
630 case BPF_ALU | BPF_SUB | BPF_X:
631 case BPF_ALU | BPF_SUB | BPF_K:
632 case BPF_ALU | BPF_AND | BPF_X:
633 case BPF_ALU | BPF_AND | BPF_K:
634 case BPF_ALU | BPF_OR | BPF_X:
635 case BPF_ALU | BPF_OR | BPF_K:
636 case BPF_ALU | BPF_LSH | BPF_X:
637 case BPF_ALU | BPF_LSH | BPF_K:
638 case BPF_ALU | BPF_RSH | BPF_X:
639 case BPF_ALU | BPF_RSH | BPF_K:
640 case BPF_ALU | BPF_XOR | BPF_X:
641 case BPF_ALU | BPF_XOR | BPF_K:
642 case BPF_ALU | BPF_MUL | BPF_X:
643 case BPF_ALU | BPF_MUL | BPF_K:
644 case BPF_ALU | BPF_DIV | BPF_X:
645 case BPF_ALU | BPF_DIV | BPF_K:
646 case BPF_ALU | BPF_MOD | BPF_X:
647 case BPF_ALU | BPF_MOD | BPF_K:
648 case BPF_ALU | BPF_NEG:
649 case BPF_LD | BPF_ABS | BPF_W:
650 case BPF_LD | BPF_ABS | BPF_H:
651 case BPF_LD | BPF_ABS | BPF_B:
652 case BPF_LD | BPF_IND | BPF_W:
653 case BPF_LD | BPF_IND | BPF_H:
654 case BPF_LD | BPF_IND | BPF_B:
655 /* Check for overloaded BPF extension and
656 * directly convert it if found, otherwise
657 * just move on with mapping.
659 if (BPF_CLASS(fp->code) == BPF_LD &&
660 BPF_MODE(fp->code) == BPF_ABS &&
661 convert_bpf_extensions(fp, &insn))
663 if (BPF_CLASS(fp->code) == BPF_LD &&
664 convert_bpf_ld_abs(fp, &insn)) {
669 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
670 fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
671 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
672 /* Error with exception code on div/mod by 0.
673 * For cBPF programs, this was always return 0.
675 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
676 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
677 *insn++ = BPF_EXIT_INSN();
680 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
683 /* Jump transformation cannot use BPF block macros
684 * everywhere as offset calculation and target updates
685 * require a bit more work than the rest, i.e. jump
686 * opcodes map as-is, but offsets need adjustment.
689 #define BPF_EMIT_JMP \
691 const s32 off_min = S16_MIN, off_max = S16_MAX; \
694 if (target >= len || target < 0) \
696 off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
697 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
698 off -= insn - tmp_insns; \
699 /* Reject anything not fitting into insn->off. */ \
700 if (off < off_min || off > off_max) \
705 case BPF_JMP | BPF_JA:
706 target = i + fp->k + 1;
707 insn->code = fp->code;
711 case BPF_JMP | BPF_JEQ | BPF_K:
712 case BPF_JMP | BPF_JEQ | BPF_X:
713 case BPF_JMP | BPF_JSET | BPF_K:
714 case BPF_JMP | BPF_JSET | BPF_X:
715 case BPF_JMP | BPF_JGT | BPF_K:
716 case BPF_JMP | BPF_JGT | BPF_X:
717 case BPF_JMP | BPF_JGE | BPF_K:
718 case BPF_JMP | BPF_JGE | BPF_X:
719 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
720 /* BPF immediates are signed, zero extend
721 * immediate into tmp register and use it
724 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
726 insn->dst_reg = BPF_REG_A;
727 insn->src_reg = BPF_REG_TMP;
730 insn->dst_reg = BPF_REG_A;
732 bpf_src = BPF_SRC(fp->code);
733 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
736 /* Common case where 'jump_false' is next insn. */
738 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
739 target = i + fp->jt + 1;
744 /* Convert some jumps when 'jump_true' is next insn. */
746 switch (BPF_OP(fp->code)) {
748 insn->code = BPF_JMP | BPF_JNE | bpf_src;
751 insn->code = BPF_JMP | BPF_JLE | bpf_src;
754 insn->code = BPF_JMP | BPF_JLT | bpf_src;
760 target = i + fp->jf + 1;
765 /* Other jumps are mapped into two insns: Jxx and JA. */
766 target = i + fp->jt + 1;
767 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
771 insn->code = BPF_JMP | BPF_JA;
772 target = i + fp->jf + 1;
776 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
777 case BPF_LDX | BPF_MSH | BPF_B: {
778 struct sock_filter tmp = {
779 .code = BPF_LD | BPF_ABS | BPF_B,
786 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
787 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
788 convert_bpf_ld_abs(&tmp, &insn);
791 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
793 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
795 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
797 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
799 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
802 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
803 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
805 case BPF_RET | BPF_A:
806 case BPF_RET | BPF_K:
807 if (BPF_RVAL(fp->code) == BPF_K)
808 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
810 *insn = BPF_EXIT_INSN();
813 /* Store to stack. */
816 stack_off = fp->k * 4 + 4;
817 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
818 BPF_ST ? BPF_REG_A : BPF_REG_X,
820 /* check_load_and_stores() verifies that classic BPF can
821 * load from stack only after write, so tracking
822 * stack_depth for ST|STX insns is enough
824 if (new_prog && new_prog->aux->stack_depth < stack_off)
825 new_prog->aux->stack_depth = stack_off;
828 /* Load from stack. */
829 case BPF_LD | BPF_MEM:
830 case BPF_LDX | BPF_MEM:
831 stack_off = fp->k * 4 + 4;
832 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
833 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
838 case BPF_LD | BPF_IMM:
839 case BPF_LDX | BPF_IMM:
840 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
841 BPF_REG_A : BPF_REG_X, fp->k);
845 case BPF_MISC | BPF_TAX:
846 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
850 case BPF_MISC | BPF_TXA:
851 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
854 /* A = skb->len or X = skb->len */
855 case BPF_LD | BPF_W | BPF_LEN:
856 case BPF_LDX | BPF_W | BPF_LEN:
857 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
858 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
859 offsetof(struct sk_buff, len));
862 /* Access seccomp_data fields. */
863 case BPF_LDX | BPF_ABS | BPF_W:
864 /* A = *(u32 *) (ctx + K) */
865 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
868 /* Unknown instruction. */
875 memcpy(new_insn, tmp_insns,
876 sizeof(*insn) * (insn - tmp_insns));
877 new_insn += insn - tmp_insns;
881 /* Only calculating new length. */
882 *new_len = new_insn - first_insn;
884 *new_len += 4; /* Prologue bits. */
889 if (new_flen != new_insn - first_insn) {
890 new_flen = new_insn - first_insn;
897 BUG_ON(*new_len != new_flen);
906 * As we dont want to clear mem[] array for each packet going through
907 * __bpf_prog_run(), we check that filter loaded by user never try to read
908 * a cell if not previously written, and we check all branches to be sure
909 * a malicious user doesn't try to abuse us.
911 static int check_load_and_stores(const struct sock_filter *filter, int flen)
913 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
916 BUILD_BUG_ON(BPF_MEMWORDS > 16);
918 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
922 memset(masks, 0xff, flen * sizeof(*masks));
924 for (pc = 0; pc < flen; pc++) {
925 memvalid &= masks[pc];
927 switch (filter[pc].code) {
930 memvalid |= (1 << filter[pc].k);
932 case BPF_LD | BPF_MEM:
933 case BPF_LDX | BPF_MEM:
934 if (!(memvalid & (1 << filter[pc].k))) {
939 case BPF_JMP | BPF_JA:
940 /* A jump must set masks on target */
941 masks[pc + 1 + filter[pc].k] &= memvalid;
944 case BPF_JMP | BPF_JEQ | BPF_K:
945 case BPF_JMP | BPF_JEQ | BPF_X:
946 case BPF_JMP | BPF_JGE | BPF_K:
947 case BPF_JMP | BPF_JGE | BPF_X:
948 case BPF_JMP | BPF_JGT | BPF_K:
949 case BPF_JMP | BPF_JGT | BPF_X:
950 case BPF_JMP | BPF_JSET | BPF_K:
951 case BPF_JMP | BPF_JSET | BPF_X:
952 /* A jump must set masks on targets */
953 masks[pc + 1 + filter[pc].jt] &= memvalid;
954 masks[pc + 1 + filter[pc].jf] &= memvalid;
964 static bool chk_code_allowed(u16 code_to_probe)
966 static const bool codes[] = {
967 /* 32 bit ALU operations */
968 [BPF_ALU | BPF_ADD | BPF_K] = true,
969 [BPF_ALU | BPF_ADD | BPF_X] = true,
970 [BPF_ALU | BPF_SUB | BPF_K] = true,
971 [BPF_ALU | BPF_SUB | BPF_X] = true,
972 [BPF_ALU | BPF_MUL | BPF_K] = true,
973 [BPF_ALU | BPF_MUL | BPF_X] = true,
974 [BPF_ALU | BPF_DIV | BPF_K] = true,
975 [BPF_ALU | BPF_DIV | BPF_X] = true,
976 [BPF_ALU | BPF_MOD | BPF_K] = true,
977 [BPF_ALU | BPF_MOD | BPF_X] = true,
978 [BPF_ALU | BPF_AND | BPF_K] = true,
979 [BPF_ALU | BPF_AND | BPF_X] = true,
980 [BPF_ALU | BPF_OR | BPF_K] = true,
981 [BPF_ALU | BPF_OR | BPF_X] = true,
982 [BPF_ALU | BPF_XOR | BPF_K] = true,
983 [BPF_ALU | BPF_XOR | BPF_X] = true,
984 [BPF_ALU | BPF_LSH | BPF_K] = true,
985 [BPF_ALU | BPF_LSH | BPF_X] = true,
986 [BPF_ALU | BPF_RSH | BPF_K] = true,
987 [BPF_ALU | BPF_RSH | BPF_X] = true,
988 [BPF_ALU | BPF_NEG] = true,
989 /* Load instructions */
990 [BPF_LD | BPF_W | BPF_ABS] = true,
991 [BPF_LD | BPF_H | BPF_ABS] = true,
992 [BPF_LD | BPF_B | BPF_ABS] = true,
993 [BPF_LD | BPF_W | BPF_LEN] = true,
994 [BPF_LD | BPF_W | BPF_IND] = true,
995 [BPF_LD | BPF_H | BPF_IND] = true,
996 [BPF_LD | BPF_B | BPF_IND] = true,
997 [BPF_LD | BPF_IMM] = true,
998 [BPF_LD | BPF_MEM] = true,
999 [BPF_LDX | BPF_W | BPF_LEN] = true,
1000 [BPF_LDX | BPF_B | BPF_MSH] = true,
1001 [BPF_LDX | BPF_IMM] = true,
1002 [BPF_LDX | BPF_MEM] = true,
1003 /* Store instructions */
1006 /* Misc instructions */
1007 [BPF_MISC | BPF_TAX] = true,
1008 [BPF_MISC | BPF_TXA] = true,
1009 /* Return instructions */
1010 [BPF_RET | BPF_K] = true,
1011 [BPF_RET | BPF_A] = true,
1012 /* Jump instructions */
1013 [BPF_JMP | BPF_JA] = true,
1014 [BPF_JMP | BPF_JEQ | BPF_K] = true,
1015 [BPF_JMP | BPF_JEQ | BPF_X] = true,
1016 [BPF_JMP | BPF_JGE | BPF_K] = true,
1017 [BPF_JMP | BPF_JGE | BPF_X] = true,
1018 [BPF_JMP | BPF_JGT | BPF_K] = true,
1019 [BPF_JMP | BPF_JGT | BPF_X] = true,
1020 [BPF_JMP | BPF_JSET | BPF_K] = true,
1021 [BPF_JMP | BPF_JSET | BPF_X] = true,
1024 if (code_to_probe >= ARRAY_SIZE(codes))
1027 return codes[code_to_probe];
1030 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1035 if (flen == 0 || flen > BPF_MAXINSNS)
1042 * bpf_check_classic - verify socket filter code
1043 * @filter: filter to verify
1044 * @flen: length of filter
1046 * Check the user's filter code. If we let some ugly
1047 * filter code slip through kaboom! The filter must contain
1048 * no references or jumps that are out of range, no illegal
1049 * instructions, and must end with a RET instruction.
1051 * All jumps are forward as they are not signed.
1053 * Returns 0 if the rule set is legal or -EINVAL if not.
1055 static int bpf_check_classic(const struct sock_filter *filter,
1061 /* Check the filter code now */
1062 for (pc = 0; pc < flen; pc++) {
1063 const struct sock_filter *ftest = &filter[pc];
1065 /* May we actually operate on this code? */
1066 if (!chk_code_allowed(ftest->code))
1069 /* Some instructions need special checks */
1070 switch (ftest->code) {
1071 case BPF_ALU | BPF_DIV | BPF_K:
1072 case BPF_ALU | BPF_MOD | BPF_K:
1073 /* Check for division by zero */
1077 case BPF_ALU | BPF_LSH | BPF_K:
1078 case BPF_ALU | BPF_RSH | BPF_K:
1082 case BPF_LD | BPF_MEM:
1083 case BPF_LDX | BPF_MEM:
1086 /* Check for invalid memory addresses */
1087 if (ftest->k >= BPF_MEMWORDS)
1090 case BPF_JMP | BPF_JA:
1091 /* Note, the large ftest->k might cause loops.
1092 * Compare this with conditional jumps below,
1093 * where offsets are limited. --ANK (981016)
1095 if (ftest->k >= (unsigned int)(flen - pc - 1))
1098 case BPF_JMP | BPF_JEQ | BPF_K:
1099 case BPF_JMP | BPF_JEQ | BPF_X:
1100 case BPF_JMP | BPF_JGE | BPF_K:
1101 case BPF_JMP | BPF_JGE | BPF_X:
1102 case BPF_JMP | BPF_JGT | BPF_K:
1103 case BPF_JMP | BPF_JGT | BPF_X:
1104 case BPF_JMP | BPF_JSET | BPF_K:
1105 case BPF_JMP | BPF_JSET | BPF_X:
1106 /* Both conditionals must be safe */
1107 if (pc + ftest->jt + 1 >= flen ||
1108 pc + ftest->jf + 1 >= flen)
1111 case BPF_LD | BPF_W | BPF_ABS:
1112 case BPF_LD | BPF_H | BPF_ABS:
1113 case BPF_LD | BPF_B | BPF_ABS:
1115 if (bpf_anc_helper(ftest) & BPF_ANC)
1117 /* Ancillary operation unknown or unsupported */
1118 if (anc_found == false && ftest->k >= SKF_AD_OFF)
1123 /* Last instruction must be a RET code */
1124 switch (filter[flen - 1].code) {
1125 case BPF_RET | BPF_K:
1126 case BPF_RET | BPF_A:
1127 return check_load_and_stores(filter, flen);
1133 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1134 const struct sock_fprog *fprog)
1136 unsigned int fsize = bpf_classic_proglen(fprog);
1137 struct sock_fprog_kern *fkprog;
1139 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1143 fkprog = fp->orig_prog;
1144 fkprog->len = fprog->len;
1146 fkprog->filter = kmemdup(fp->insns, fsize,
1147 GFP_KERNEL | __GFP_NOWARN);
1148 if (!fkprog->filter) {
1149 kfree(fp->orig_prog);
1156 static void bpf_release_orig_filter(struct bpf_prog *fp)
1158 struct sock_fprog_kern *fprog = fp->orig_prog;
1161 kfree(fprog->filter);
1166 static void __bpf_prog_release(struct bpf_prog *prog)
1168 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1171 bpf_release_orig_filter(prog);
1172 bpf_prog_free(prog);
1176 static void __sk_filter_release(struct sk_filter *fp)
1178 __bpf_prog_release(fp->prog);
1183 * sk_filter_release_rcu - Release a socket filter by rcu_head
1184 * @rcu: rcu_head that contains the sk_filter to free
1186 static void sk_filter_release_rcu(struct rcu_head *rcu)
1188 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1190 __sk_filter_release(fp);
1194 * sk_filter_release - release a socket filter
1195 * @fp: filter to remove
1197 * Remove a filter from a socket and release its resources.
1199 static void sk_filter_release(struct sk_filter *fp)
1201 if (refcount_dec_and_test(&fp->refcnt))
1202 call_rcu(&fp->rcu, sk_filter_release_rcu);
1205 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1207 u32 filter_size = bpf_prog_size(fp->prog->len);
1209 atomic_sub(filter_size, &sk->sk_omem_alloc);
1210 sk_filter_release(fp);
1213 /* try to charge the socket memory if there is space available
1214 * return true on success
1216 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1218 u32 filter_size = bpf_prog_size(fp->prog->len);
1219 int optmem_max = READ_ONCE(sysctl_optmem_max);
1221 /* same check as in sock_kmalloc() */
1222 if (filter_size <= optmem_max &&
1223 atomic_read(&sk->sk_omem_alloc) + filter_size < optmem_max) {
1224 atomic_add(filter_size, &sk->sk_omem_alloc);
1230 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1232 if (!refcount_inc_not_zero(&fp->refcnt))
1235 if (!__sk_filter_charge(sk, fp)) {
1236 sk_filter_release(fp);
1242 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1244 struct sock_filter *old_prog;
1245 struct bpf_prog *old_fp;
1246 int err, new_len, old_len = fp->len;
1247 bool seen_ld_abs = false;
1249 /* We are free to overwrite insns et al right here as it won't be used at
1250 * this point in time anymore internally after the migration to the eBPF
1251 * instruction representation.
1253 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1254 sizeof(struct bpf_insn));
1256 /* Conversion cannot happen on overlapping memory areas,
1257 * so we need to keep the user BPF around until the 2nd
1258 * pass. At this time, the user BPF is stored in fp->insns.
1260 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1261 GFP_KERNEL | __GFP_NOWARN);
1267 /* 1st pass: calculate the new program length. */
1268 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1273 /* Expand fp for appending the new filter representation. */
1275 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1277 /* The old_fp is still around in case we couldn't
1278 * allocate new memory, so uncharge on that one.
1287 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1288 err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1291 /* 2nd bpf_convert_filter() can fail only if it fails
1292 * to allocate memory, remapping must succeed. Note,
1293 * that at this time old_fp has already been released
1298 fp = bpf_prog_select_runtime(fp, &err);
1308 __bpf_prog_release(fp);
1309 return ERR_PTR(err);
1312 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1313 bpf_aux_classic_check_t trans)
1317 fp->bpf_func = NULL;
1320 err = bpf_check_classic(fp->insns, fp->len);
1322 __bpf_prog_release(fp);
1323 return ERR_PTR(err);
1326 /* There might be additional checks and transformations
1327 * needed on classic filters, f.e. in case of seccomp.
1330 err = trans(fp->insns, fp->len);
1332 __bpf_prog_release(fp);
1333 return ERR_PTR(err);
1337 /* Probe if we can JIT compile the filter and if so, do
1338 * the compilation of the filter.
1340 bpf_jit_compile(fp);
1342 /* JIT compiler couldn't process this filter, so do the eBPF translation
1343 * for the optimized interpreter.
1346 fp = bpf_migrate_filter(fp);
1352 * bpf_prog_create - create an unattached filter
1353 * @pfp: the unattached filter that is created
1354 * @fprog: the filter program
1356 * Create a filter independent of any socket. We first run some
1357 * sanity checks on it to make sure it does not explode on us later.
1358 * If an error occurs or there is insufficient memory for the filter
1359 * a negative errno code is returned. On success the return is zero.
1361 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1363 unsigned int fsize = bpf_classic_proglen(fprog);
1364 struct bpf_prog *fp;
1366 /* Make sure new filter is there and in the right amounts. */
1367 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1370 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1374 memcpy(fp->insns, fprog->filter, fsize);
1376 fp->len = fprog->len;
1377 /* Since unattached filters are not copied back to user
1378 * space through sk_get_filter(), we do not need to hold
1379 * a copy here, and can spare us the work.
1381 fp->orig_prog = NULL;
1383 /* bpf_prepare_filter() already takes care of freeing
1384 * memory in case something goes wrong.
1386 fp = bpf_prepare_filter(fp, NULL);
1393 EXPORT_SYMBOL_GPL(bpf_prog_create);
1396 * bpf_prog_create_from_user - create an unattached filter from user buffer
1397 * @pfp: the unattached filter that is created
1398 * @fprog: the filter program
1399 * @trans: post-classic verifier transformation handler
1400 * @save_orig: save classic BPF program
1402 * This function effectively does the same as bpf_prog_create(), only
1403 * that it builds up its insns buffer from user space provided buffer.
1404 * It also allows for passing a bpf_aux_classic_check_t handler.
1406 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1407 bpf_aux_classic_check_t trans, bool save_orig)
1409 unsigned int fsize = bpf_classic_proglen(fprog);
1410 struct bpf_prog *fp;
1413 /* Make sure new filter is there and in the right amounts. */
1414 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1417 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1421 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1422 __bpf_prog_free(fp);
1426 fp->len = fprog->len;
1427 fp->orig_prog = NULL;
1430 err = bpf_prog_store_orig_filter(fp, fprog);
1432 __bpf_prog_free(fp);
1437 /* bpf_prepare_filter() already takes care of freeing
1438 * memory in case something goes wrong.
1440 fp = bpf_prepare_filter(fp, trans);
1447 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1449 void bpf_prog_destroy(struct bpf_prog *fp)
1451 __bpf_prog_release(fp);
1453 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1455 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1457 struct sk_filter *fp, *old_fp;
1459 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1465 if (!__sk_filter_charge(sk, fp)) {
1469 refcount_set(&fp->refcnt, 1);
1471 old_fp = rcu_dereference_protected(sk->sk_filter,
1472 lockdep_sock_is_held(sk));
1473 rcu_assign_pointer(sk->sk_filter, fp);
1476 sk_filter_uncharge(sk, old_fp);
1482 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1484 unsigned int fsize = bpf_classic_proglen(fprog);
1485 struct bpf_prog *prog;
1488 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1489 return ERR_PTR(-EPERM);
1491 /* Make sure new filter is there and in the right amounts. */
1492 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1493 return ERR_PTR(-EINVAL);
1495 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1497 return ERR_PTR(-ENOMEM);
1499 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1500 __bpf_prog_free(prog);
1501 return ERR_PTR(-EFAULT);
1504 prog->len = fprog->len;
1506 err = bpf_prog_store_orig_filter(prog, fprog);
1508 __bpf_prog_free(prog);
1509 return ERR_PTR(-ENOMEM);
1512 /* bpf_prepare_filter() already takes care of freeing
1513 * memory in case something goes wrong.
1515 return bpf_prepare_filter(prog, NULL);
1519 * sk_attach_filter - attach a socket filter
1520 * @fprog: the filter program
1521 * @sk: the socket to use
1523 * Attach the user's filter code. We first run some sanity checks on
1524 * it to make sure it does not explode on us later. If an error
1525 * occurs or there is insufficient memory for the filter a negative
1526 * errno code is returned. On success the return is zero.
1528 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1530 struct bpf_prog *prog = __get_filter(fprog, sk);
1534 return PTR_ERR(prog);
1536 err = __sk_attach_prog(prog, sk);
1538 __bpf_prog_release(prog);
1544 EXPORT_SYMBOL_GPL(sk_attach_filter);
1546 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1548 struct bpf_prog *prog = __get_filter(fprog, sk);
1552 return PTR_ERR(prog);
1554 if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max))
1557 err = reuseport_attach_prog(sk, prog);
1560 __bpf_prog_release(prog);
1565 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1567 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1568 return ERR_PTR(-EPERM);
1570 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1573 int sk_attach_bpf(u32 ufd, struct sock *sk)
1575 struct bpf_prog *prog = __get_bpf(ufd, sk);
1579 return PTR_ERR(prog);
1581 err = __sk_attach_prog(prog, sk);
1590 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1592 struct bpf_prog *prog;
1595 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1598 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1599 if (PTR_ERR(prog) == -EINVAL)
1600 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1602 return PTR_ERR(prog);
1604 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1605 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1606 * bpf prog (e.g. sockmap). It depends on the
1607 * limitation imposed by bpf_prog_load().
1608 * Hence, sysctl_optmem_max is not checked.
1610 if ((sk->sk_type != SOCK_STREAM &&
1611 sk->sk_type != SOCK_DGRAM) ||
1612 (sk->sk_protocol != IPPROTO_UDP &&
1613 sk->sk_protocol != IPPROTO_TCP) ||
1614 (sk->sk_family != AF_INET &&
1615 sk->sk_family != AF_INET6)) {
1620 /* BPF_PROG_TYPE_SOCKET_FILTER */
1621 if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max)) {
1627 err = reuseport_attach_prog(sk, prog);
1635 void sk_reuseport_prog_free(struct bpf_prog *prog)
1640 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1643 bpf_prog_destroy(prog);
1646 struct bpf_scratchpad {
1648 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1649 u8 buff[MAX_BPF_STACK];
1653 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1655 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1656 unsigned int write_len)
1658 return skb_ensure_writable(skb, write_len);
1661 static inline int bpf_try_make_writable(struct sk_buff *skb,
1662 unsigned int write_len)
1664 int err = __bpf_try_make_writable(skb, write_len);
1666 bpf_compute_data_pointers(skb);
1670 static int bpf_try_make_head_writable(struct sk_buff *skb)
1672 return bpf_try_make_writable(skb, skb_headlen(skb));
1675 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1677 if (skb_at_tc_ingress(skb))
1678 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1681 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1683 if (skb_at_tc_ingress(skb))
1684 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1687 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1688 const void *, from, u32, len, u64, flags)
1692 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1694 if (unlikely(offset > INT_MAX))
1696 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1699 ptr = skb->data + offset;
1700 if (flags & BPF_F_RECOMPUTE_CSUM)
1701 __skb_postpull_rcsum(skb, ptr, len, offset);
1703 memcpy(ptr, from, len);
1705 if (flags & BPF_F_RECOMPUTE_CSUM)
1706 __skb_postpush_rcsum(skb, ptr, len, offset);
1707 if (flags & BPF_F_INVALIDATE_HASH)
1708 skb_clear_hash(skb);
1713 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1714 .func = bpf_skb_store_bytes,
1716 .ret_type = RET_INTEGER,
1717 .arg1_type = ARG_PTR_TO_CTX,
1718 .arg2_type = ARG_ANYTHING,
1719 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1720 .arg4_type = ARG_CONST_SIZE,
1721 .arg5_type = ARG_ANYTHING,
1724 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1725 void *, to, u32, len)
1729 if (unlikely(offset > INT_MAX))
1732 ptr = skb_header_pointer(skb, offset, len, to);
1736 memcpy(to, ptr, len);
1744 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1745 .func = bpf_skb_load_bytes,
1747 .ret_type = RET_INTEGER,
1748 .arg1_type = ARG_PTR_TO_CTX,
1749 .arg2_type = ARG_ANYTHING,
1750 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1751 .arg4_type = ARG_CONST_SIZE,
1754 BPF_CALL_4(bpf_flow_dissector_load_bytes,
1755 const struct bpf_flow_dissector *, ctx, u32, offset,
1756 void *, to, u32, len)
1760 if (unlikely(offset > 0xffff))
1763 if (unlikely(!ctx->skb))
1766 ptr = skb_header_pointer(ctx->skb, offset, len, to);
1770 memcpy(to, ptr, len);
1778 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1779 .func = bpf_flow_dissector_load_bytes,
1781 .ret_type = RET_INTEGER,
1782 .arg1_type = ARG_PTR_TO_CTX,
1783 .arg2_type = ARG_ANYTHING,
1784 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1785 .arg4_type = ARG_CONST_SIZE,
1788 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1789 u32, offset, void *, to, u32, len, u32, start_header)
1791 u8 *end = skb_tail_pointer(skb);
1794 if (unlikely(offset > 0xffff))
1797 switch (start_header) {
1798 case BPF_HDR_START_MAC:
1799 if (unlikely(!skb_mac_header_was_set(skb)))
1801 start = skb_mac_header(skb);
1803 case BPF_HDR_START_NET:
1804 start = skb_network_header(skb);
1810 ptr = start + offset;
1812 if (likely(ptr + len <= end)) {
1813 memcpy(to, ptr, len);
1822 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1823 .func = bpf_skb_load_bytes_relative,
1825 .ret_type = RET_INTEGER,
1826 .arg1_type = ARG_PTR_TO_CTX,
1827 .arg2_type = ARG_ANYTHING,
1828 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1829 .arg4_type = ARG_CONST_SIZE,
1830 .arg5_type = ARG_ANYTHING,
1833 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1835 /* Idea is the following: should the needed direct read/write
1836 * test fail during runtime, we can pull in more data and redo
1837 * again, since implicitly, we invalidate previous checks here.
1839 * Or, since we know how much we need to make read/writeable,
1840 * this can be done once at the program beginning for direct
1841 * access case. By this we overcome limitations of only current
1842 * headroom being accessible.
1844 return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1847 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1848 .func = bpf_skb_pull_data,
1850 .ret_type = RET_INTEGER,
1851 .arg1_type = ARG_PTR_TO_CTX,
1852 .arg2_type = ARG_ANYTHING,
1855 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1857 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1860 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1861 .func = bpf_sk_fullsock,
1863 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
1864 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
1867 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1868 unsigned int write_len)
1870 return __bpf_try_make_writable(skb, write_len);
1873 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1875 /* Idea is the following: should the needed direct read/write
1876 * test fail during runtime, we can pull in more data and redo
1877 * again, since implicitly, we invalidate previous checks here.
1879 * Or, since we know how much we need to make read/writeable,
1880 * this can be done once at the program beginning for direct
1881 * access case. By this we overcome limitations of only current
1882 * headroom being accessible.
1884 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1887 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1888 .func = sk_skb_pull_data,
1890 .ret_type = RET_INTEGER,
1891 .arg1_type = ARG_PTR_TO_CTX,
1892 .arg2_type = ARG_ANYTHING,
1895 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1896 u64, from, u64, to, u64, flags)
1900 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1902 if (unlikely(offset > 0xffff || offset & 1))
1904 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1907 ptr = (__sum16 *)(skb->data + offset);
1908 switch (flags & BPF_F_HDR_FIELD_MASK) {
1910 if (unlikely(from != 0))
1913 csum_replace_by_diff(ptr, to);
1916 csum_replace2(ptr, from, to);
1919 csum_replace4(ptr, from, to);
1928 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1929 .func = bpf_l3_csum_replace,
1931 .ret_type = RET_INTEGER,
1932 .arg1_type = ARG_PTR_TO_CTX,
1933 .arg2_type = ARG_ANYTHING,
1934 .arg3_type = ARG_ANYTHING,
1935 .arg4_type = ARG_ANYTHING,
1936 .arg5_type = ARG_ANYTHING,
1939 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1940 u64, from, u64, to, u64, flags)
1942 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1943 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1944 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1947 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1948 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1950 if (unlikely(offset > 0xffff || offset & 1))
1952 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1955 ptr = (__sum16 *)(skb->data + offset);
1956 if (is_mmzero && !do_mforce && !*ptr)
1959 switch (flags & BPF_F_HDR_FIELD_MASK) {
1961 if (unlikely(from != 0))
1964 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1967 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1970 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1976 if (is_mmzero && !*ptr)
1977 *ptr = CSUM_MANGLED_0;
1981 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1982 .func = bpf_l4_csum_replace,
1984 .ret_type = RET_INTEGER,
1985 .arg1_type = ARG_PTR_TO_CTX,
1986 .arg2_type = ARG_ANYTHING,
1987 .arg3_type = ARG_ANYTHING,
1988 .arg4_type = ARG_ANYTHING,
1989 .arg5_type = ARG_ANYTHING,
1992 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1993 __be32 *, to, u32, to_size, __wsum, seed)
1995 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1996 u32 diff_size = from_size + to_size;
1999 /* This is quite flexible, some examples:
2001 * from_size == 0, to_size > 0, seed := csum --> pushing data
2002 * from_size > 0, to_size == 0, seed := csum --> pulling data
2003 * from_size > 0, to_size > 0, seed := 0 --> diffing data
2005 * Even for diffing, from_size and to_size don't need to be equal.
2007 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
2008 diff_size > sizeof(sp->diff)))
2011 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
2012 sp->diff[j] = ~from[i];
2013 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
2014 sp->diff[j] = to[i];
2016 return csum_partial(sp->diff, diff_size, seed);
2019 static const struct bpf_func_proto bpf_csum_diff_proto = {
2020 .func = bpf_csum_diff,
2023 .ret_type = RET_INTEGER,
2024 .arg1_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2025 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
2026 .arg3_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2027 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
2028 .arg5_type = ARG_ANYTHING,
2031 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2033 /* The interface is to be used in combination with bpf_csum_diff()
2034 * for direct packet writes. csum rotation for alignment as well
2035 * as emulating csum_sub() can be done from the eBPF program.
2037 if (skb->ip_summed == CHECKSUM_COMPLETE)
2038 return (skb->csum = csum_add(skb->csum, csum));
2043 static const struct bpf_func_proto bpf_csum_update_proto = {
2044 .func = bpf_csum_update,
2046 .ret_type = RET_INTEGER,
2047 .arg1_type = ARG_PTR_TO_CTX,
2048 .arg2_type = ARG_ANYTHING,
2051 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
2053 /* The interface is to be used in combination with bpf_skb_adjust_room()
2054 * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET
2055 * is passed as flags, for example.
2058 case BPF_CSUM_LEVEL_INC:
2059 __skb_incr_checksum_unnecessary(skb);
2061 case BPF_CSUM_LEVEL_DEC:
2062 __skb_decr_checksum_unnecessary(skb);
2064 case BPF_CSUM_LEVEL_RESET:
2065 __skb_reset_checksum_unnecessary(skb);
2067 case BPF_CSUM_LEVEL_QUERY:
2068 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2069 skb->csum_level : -EACCES;
2077 static const struct bpf_func_proto bpf_csum_level_proto = {
2078 .func = bpf_csum_level,
2080 .ret_type = RET_INTEGER,
2081 .arg1_type = ARG_PTR_TO_CTX,
2082 .arg2_type = ARG_ANYTHING,
2085 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2087 return dev_forward_skb_nomtu(dev, skb);
2090 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2091 struct sk_buff *skb)
2093 int ret = ____dev_forward_skb(dev, skb, false);
2097 ret = netif_rx(skb);
2103 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2107 if (dev_xmit_recursion()) {
2108 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2114 skb_clear_tstamp(skb);
2116 dev_xmit_recursion_inc();
2117 ret = dev_queue_xmit(skb);
2118 dev_xmit_recursion_dec();
2123 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2126 unsigned int mlen = skb_network_offset(skb);
2128 if (unlikely(skb->len <= mlen)) {
2134 __skb_pull(skb, mlen);
2135 if (unlikely(!skb->len)) {
2140 /* At ingress, the mac header has already been pulled once.
2141 * At egress, skb_pospull_rcsum has to be done in case that
2142 * the skb is originated from ingress (i.e. a forwarded skb)
2143 * to ensure that rcsum starts at net header.
2145 if (!skb_at_tc_ingress(skb))
2146 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2148 skb_pop_mac_header(skb);
2149 skb_reset_mac_len(skb);
2150 return flags & BPF_F_INGRESS ?
2151 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2154 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2157 /* Verify that a link layer header is carried */
2158 if (unlikely(skb->mac_header >= skb->network_header || skb->len == 0)) {
2163 bpf_push_mac_rcsum(skb);
2164 return flags & BPF_F_INGRESS ?
2165 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2168 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2171 if (dev_is_mac_header_xmit(dev))
2172 return __bpf_redirect_common(skb, dev, flags);
2174 return __bpf_redirect_no_mac(skb, dev, flags);
2177 #if IS_ENABLED(CONFIG_IPV6)
2178 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2179 struct net_device *dev, struct bpf_nh_params *nh)
2181 u32 hh_len = LL_RESERVED_SPACE(dev);
2182 const struct in6_addr *nexthop;
2183 struct dst_entry *dst = NULL;
2184 struct neighbour *neigh;
2186 if (dev_xmit_recursion()) {
2187 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2192 skb_clear_tstamp(skb);
2194 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2195 skb = skb_expand_head(skb, hh_len);
2203 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2204 &ipv6_hdr(skb)->daddr);
2206 nexthop = &nh->ipv6_nh;
2208 neigh = ip_neigh_gw6(dev, nexthop);
2209 if (likely(!IS_ERR(neigh))) {
2212 sock_confirm_neigh(skb, neigh);
2214 dev_xmit_recursion_inc();
2215 ret = neigh_output(neigh, skb, false);
2216 dev_xmit_recursion_dec();
2221 rcu_read_unlock_bh();
2223 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2229 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2230 struct bpf_nh_params *nh)
2232 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2233 struct net *net = dev_net(dev);
2234 int err, ret = NET_XMIT_DROP;
2237 struct dst_entry *dst;
2238 struct flowi6 fl6 = {
2239 .flowi6_flags = FLOWI_FLAG_ANYSRC,
2240 .flowi6_mark = skb->mark,
2241 .flowlabel = ip6_flowinfo(ip6h),
2242 .flowi6_oif = dev->ifindex,
2243 .flowi6_proto = ip6h->nexthdr,
2244 .daddr = ip6h->daddr,
2245 .saddr = ip6h->saddr,
2248 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2252 skb_dst_set(skb, dst);
2253 } else if (nh->nh_family != AF_INET6) {
2257 err = bpf_out_neigh_v6(net, skb, dev, nh);
2258 if (unlikely(net_xmit_eval(err)))
2259 dev->stats.tx_errors++;
2261 ret = NET_XMIT_SUCCESS;
2264 dev->stats.tx_errors++;
2270 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2271 struct bpf_nh_params *nh)
2274 return NET_XMIT_DROP;
2276 #endif /* CONFIG_IPV6 */
2278 #if IS_ENABLED(CONFIG_INET)
2279 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2280 struct net_device *dev, struct bpf_nh_params *nh)
2282 u32 hh_len = LL_RESERVED_SPACE(dev);
2283 struct neighbour *neigh;
2284 bool is_v6gw = false;
2286 if (dev_xmit_recursion()) {
2287 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2292 skb_clear_tstamp(skb);
2294 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2295 skb = skb_expand_head(skb, hh_len);
2302 struct dst_entry *dst = skb_dst(skb);
2303 struct rtable *rt = container_of(dst, struct rtable, dst);
2305 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2306 } else if (nh->nh_family == AF_INET6) {
2307 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2309 } else if (nh->nh_family == AF_INET) {
2310 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2316 if (likely(!IS_ERR(neigh))) {
2319 sock_confirm_neigh(skb, neigh);
2321 dev_xmit_recursion_inc();
2322 ret = neigh_output(neigh, skb, is_v6gw);
2323 dev_xmit_recursion_dec();
2334 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2335 struct bpf_nh_params *nh)
2337 const struct iphdr *ip4h = ip_hdr(skb);
2338 struct net *net = dev_net(dev);
2339 int err, ret = NET_XMIT_DROP;
2342 struct flowi4 fl4 = {
2343 .flowi4_flags = FLOWI_FLAG_ANYSRC,
2344 .flowi4_mark = skb->mark,
2345 .flowi4_tos = RT_TOS(ip4h->tos),
2346 .flowi4_oif = dev->ifindex,
2347 .flowi4_proto = ip4h->protocol,
2348 .daddr = ip4h->daddr,
2349 .saddr = ip4h->saddr,
2353 rt = ip_route_output_flow(net, &fl4, NULL);
2356 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2361 skb_dst_set(skb, &rt->dst);
2364 err = bpf_out_neigh_v4(net, skb, dev, nh);
2365 if (unlikely(net_xmit_eval(err)))
2366 dev->stats.tx_errors++;
2368 ret = NET_XMIT_SUCCESS;
2371 dev->stats.tx_errors++;
2377 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2378 struct bpf_nh_params *nh)
2381 return NET_XMIT_DROP;
2383 #endif /* CONFIG_INET */
2385 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2386 struct bpf_nh_params *nh)
2388 struct ethhdr *ethh = eth_hdr(skb);
2390 if (unlikely(skb->mac_header >= skb->network_header))
2392 bpf_push_mac_rcsum(skb);
2393 if (is_multicast_ether_addr(ethh->h_dest))
2396 skb_pull(skb, sizeof(*ethh));
2397 skb_unset_mac_header(skb);
2398 skb_reset_network_header(skb);
2400 if (skb->protocol == htons(ETH_P_IP))
2401 return __bpf_redirect_neigh_v4(skb, dev, nh);
2402 else if (skb->protocol == htons(ETH_P_IPV6))
2403 return __bpf_redirect_neigh_v6(skb, dev, nh);
2409 /* Internal, non-exposed redirect flags. */
2411 BPF_F_NEIGH = (1ULL << 1),
2412 BPF_F_PEER = (1ULL << 2),
2413 BPF_F_NEXTHOP = (1ULL << 3),
2414 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2417 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2419 struct net_device *dev;
2420 struct sk_buff *clone;
2423 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2426 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2430 clone = skb_clone(skb, GFP_ATOMIC);
2431 if (unlikely(!clone))
2434 /* For direct write, we need to keep the invariant that the skbs
2435 * we're dealing with need to be uncloned. Should uncloning fail
2436 * here, we need to free the just generated clone to unclone once
2439 ret = bpf_try_make_head_writable(skb);
2440 if (unlikely(ret)) {
2445 return __bpf_redirect(clone, dev, flags);
2448 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2449 .func = bpf_clone_redirect,
2451 .ret_type = RET_INTEGER,
2452 .arg1_type = ARG_PTR_TO_CTX,
2453 .arg2_type = ARG_ANYTHING,
2454 .arg3_type = ARG_ANYTHING,
2457 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2458 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2460 int skb_do_redirect(struct sk_buff *skb)
2462 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2463 struct net *net = dev_net(skb->dev);
2464 struct net_device *dev;
2465 u32 flags = ri->flags;
2467 dev = dev_get_by_index_rcu(net, ri->tgt_index);
2472 if (flags & BPF_F_PEER) {
2473 const struct net_device_ops *ops = dev->netdev_ops;
2475 if (unlikely(!ops->ndo_get_peer_dev ||
2476 !skb_at_tc_ingress(skb)))
2478 dev = ops->ndo_get_peer_dev(dev);
2479 if (unlikely(!dev ||
2480 !(dev->flags & IFF_UP) ||
2481 net_eq(net, dev_net(dev))))
2486 return flags & BPF_F_NEIGH ?
2487 __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2489 __bpf_redirect(skb, dev, flags);
2495 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2497 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2499 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2503 ri->tgt_index = ifindex;
2505 return TC_ACT_REDIRECT;
2508 static const struct bpf_func_proto bpf_redirect_proto = {
2509 .func = bpf_redirect,
2511 .ret_type = RET_INTEGER,
2512 .arg1_type = ARG_ANYTHING,
2513 .arg2_type = ARG_ANYTHING,
2516 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2518 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2520 if (unlikely(flags))
2523 ri->flags = BPF_F_PEER;
2524 ri->tgt_index = ifindex;
2526 return TC_ACT_REDIRECT;
2529 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2530 .func = bpf_redirect_peer,
2532 .ret_type = RET_INTEGER,
2533 .arg1_type = ARG_ANYTHING,
2534 .arg2_type = ARG_ANYTHING,
2537 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2538 int, plen, u64, flags)
2540 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2542 if (unlikely((plen && plen < sizeof(*params)) || flags))
2545 ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2546 ri->tgt_index = ifindex;
2548 BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2550 memcpy(&ri->nh, params, sizeof(ri->nh));
2552 return TC_ACT_REDIRECT;
2555 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2556 .func = bpf_redirect_neigh,
2558 .ret_type = RET_INTEGER,
2559 .arg1_type = ARG_ANYTHING,
2560 .arg2_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2561 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
2562 .arg4_type = ARG_ANYTHING,
2565 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2567 msg->apply_bytes = bytes;
2571 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2572 .func = bpf_msg_apply_bytes,
2574 .ret_type = RET_INTEGER,
2575 .arg1_type = ARG_PTR_TO_CTX,
2576 .arg2_type = ARG_ANYTHING,
2579 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2581 msg->cork_bytes = bytes;
2585 static void sk_msg_reset_curr(struct sk_msg *msg)
2587 u32 i = msg->sg.start;
2591 len += sk_msg_elem(msg, i)->length;
2592 sk_msg_iter_var_next(i);
2593 if (len >= msg->sg.size)
2595 } while (i != msg->sg.end);
2598 msg->sg.copybreak = 0;
2601 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2602 .func = bpf_msg_cork_bytes,
2604 .ret_type = RET_INTEGER,
2605 .arg1_type = ARG_PTR_TO_CTX,
2606 .arg2_type = ARG_ANYTHING,
2609 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2610 u32, end, u64, flags)
2612 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2613 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2614 struct scatterlist *sge;
2615 u8 *raw, *to, *from;
2618 if (unlikely(flags || end <= start))
2621 /* First find the starting scatterlist element */
2625 len = sk_msg_elem(msg, i)->length;
2626 if (start < offset + len)
2628 sk_msg_iter_var_next(i);
2629 } while (i != msg->sg.end);
2631 if (unlikely(start >= offset + len))
2635 /* The start may point into the sg element so we need to also
2636 * account for the headroom.
2638 bytes_sg_total = start - offset + bytes;
2639 if (!test_bit(i, msg->sg.copy) && bytes_sg_total <= len)
2642 /* At this point we need to linearize multiple scatterlist
2643 * elements or a single shared page. Either way we need to
2644 * copy into a linear buffer exclusively owned by BPF. Then
2645 * place the buffer in the scatterlist and fixup the original
2646 * entries by removing the entries now in the linear buffer
2647 * and shifting the remaining entries. For now we do not try
2648 * to copy partial entries to avoid complexity of running out
2649 * of sg_entry slots. The downside is reading a single byte
2650 * will copy the entire sg entry.
2653 copy += sk_msg_elem(msg, i)->length;
2654 sk_msg_iter_var_next(i);
2655 if (bytes_sg_total <= copy)
2657 } while (i != msg->sg.end);
2660 if (unlikely(bytes_sg_total > copy))
2663 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2665 if (unlikely(!page))
2668 raw = page_address(page);
2671 sge = sk_msg_elem(msg, i);
2672 from = sg_virt(sge);
2676 memcpy(to, from, len);
2679 put_page(sg_page(sge));
2681 sk_msg_iter_var_next(i);
2682 } while (i != last_sge);
2684 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2686 /* To repair sg ring we need to shift entries. If we only
2687 * had a single entry though we can just replace it and
2688 * be done. Otherwise walk the ring and shift the entries.
2690 WARN_ON_ONCE(last_sge == first_sge);
2691 shift = last_sge > first_sge ?
2692 last_sge - first_sge - 1 :
2693 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2698 sk_msg_iter_var_next(i);
2702 if (i + shift >= NR_MSG_FRAG_IDS)
2703 move_from = i + shift - NR_MSG_FRAG_IDS;
2705 move_from = i + shift;
2706 if (move_from == msg->sg.end)
2709 msg->sg.data[i] = msg->sg.data[move_from];
2710 msg->sg.data[move_from].length = 0;
2711 msg->sg.data[move_from].page_link = 0;
2712 msg->sg.data[move_from].offset = 0;
2713 sk_msg_iter_var_next(i);
2716 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2717 msg->sg.end - shift + NR_MSG_FRAG_IDS :
2718 msg->sg.end - shift;
2720 sk_msg_reset_curr(msg);
2721 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2722 msg->data_end = msg->data + bytes;
2726 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2727 .func = bpf_msg_pull_data,
2729 .ret_type = RET_INTEGER,
2730 .arg1_type = ARG_PTR_TO_CTX,
2731 .arg2_type = ARG_ANYTHING,
2732 .arg3_type = ARG_ANYTHING,
2733 .arg4_type = ARG_ANYTHING,
2736 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2737 u32, len, u64, flags)
2739 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2740 u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2741 u8 *raw, *to, *from;
2744 if (unlikely(flags))
2747 if (unlikely(len == 0))
2750 /* First find the starting scatterlist element */
2754 l = sk_msg_elem(msg, i)->length;
2756 if (start < offset + l)
2758 sk_msg_iter_var_next(i);
2759 } while (i != msg->sg.end);
2761 if (start >= offset + l)
2764 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2766 /* If no space available will fallback to copy, we need at
2767 * least one scatterlist elem available to push data into
2768 * when start aligns to the beginning of an element or two
2769 * when it falls inside an element. We handle the start equals
2770 * offset case because its the common case for inserting a
2773 if (!space || (space == 1 && start != offset))
2774 copy = msg->sg.data[i].length;
2776 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2777 get_order(copy + len));
2778 if (unlikely(!page))
2784 raw = page_address(page);
2786 psge = sk_msg_elem(msg, i);
2787 front = start - offset;
2788 back = psge->length - front;
2789 from = sg_virt(psge);
2792 memcpy(raw, from, front);
2796 to = raw + front + len;
2798 memcpy(to, from, back);
2801 put_page(sg_page(psge));
2802 } else if (start - offset) {
2803 psge = sk_msg_elem(msg, i);
2804 rsge = sk_msg_elem_cpy(msg, i);
2806 psge->length = start - offset;
2807 rsge.length -= psge->length;
2808 rsge.offset += start;
2810 sk_msg_iter_var_next(i);
2811 sg_unmark_end(psge);
2812 sg_unmark_end(&rsge);
2813 sk_msg_iter_next(msg, end);
2816 /* Slot(s) to place newly allocated data */
2819 /* Shift one or two slots as needed */
2821 sge = sk_msg_elem_cpy(msg, i);
2823 sk_msg_iter_var_next(i);
2824 sg_unmark_end(&sge);
2825 sk_msg_iter_next(msg, end);
2827 nsge = sk_msg_elem_cpy(msg, i);
2829 sk_msg_iter_var_next(i);
2830 nnsge = sk_msg_elem_cpy(msg, i);
2833 while (i != msg->sg.end) {
2834 msg->sg.data[i] = sge;
2836 sk_msg_iter_var_next(i);
2839 nnsge = sk_msg_elem_cpy(msg, i);
2841 nsge = sk_msg_elem_cpy(msg, i);
2846 /* Place newly allocated data buffer */
2847 sk_mem_charge(msg->sk, len);
2848 msg->sg.size += len;
2849 __clear_bit(new, msg->sg.copy);
2850 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2852 get_page(sg_page(&rsge));
2853 sk_msg_iter_var_next(new);
2854 msg->sg.data[new] = rsge;
2857 sk_msg_reset_curr(msg);
2858 sk_msg_compute_data_pointers(msg);
2862 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2863 .func = bpf_msg_push_data,
2865 .ret_type = RET_INTEGER,
2866 .arg1_type = ARG_PTR_TO_CTX,
2867 .arg2_type = ARG_ANYTHING,
2868 .arg3_type = ARG_ANYTHING,
2869 .arg4_type = ARG_ANYTHING,
2872 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2878 sk_msg_iter_var_next(i);
2879 msg->sg.data[prev] = msg->sg.data[i];
2880 } while (i != msg->sg.end);
2882 sk_msg_iter_prev(msg, end);
2885 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2887 struct scatterlist tmp, sge;
2889 sk_msg_iter_next(msg, end);
2890 sge = sk_msg_elem_cpy(msg, i);
2891 sk_msg_iter_var_next(i);
2892 tmp = sk_msg_elem_cpy(msg, i);
2894 while (i != msg->sg.end) {
2895 msg->sg.data[i] = sge;
2896 sk_msg_iter_var_next(i);
2898 tmp = sk_msg_elem_cpy(msg, i);
2902 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2903 u32, len, u64, flags)
2905 u32 i = 0, l = 0, space, offset = 0;
2906 u64 last = start + len;
2909 if (unlikely(flags))
2912 /* First find the starting scatterlist element */
2916 l = sk_msg_elem(msg, i)->length;
2918 if (start < offset + l)
2920 sk_msg_iter_var_next(i);
2921 } while (i != msg->sg.end);
2923 /* Bounds checks: start and pop must be inside message */
2924 if (start >= offset + l || last >= msg->sg.size)
2927 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2930 /* --------------| offset
2931 * -| start |-------- len -------|
2933 * |----- a ----|-------- pop -------|----- b ----|
2934 * |______________________________________________| length
2937 * a: region at front of scatter element to save
2938 * b: region at back of scatter element to save when length > A + pop
2939 * pop: region to pop from element, same as input 'pop' here will be
2940 * decremented below per iteration.
2942 * Two top-level cases to handle when start != offset, first B is non
2943 * zero and second B is zero corresponding to when a pop includes more
2946 * Then if B is non-zero AND there is no space allocate space and
2947 * compact A, B regions into page. If there is space shift ring to
2948 * the rigth free'ing the next element in ring to place B, leaving
2949 * A untouched except to reduce length.
2951 if (start != offset) {
2952 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2954 int b = sge->length - pop - a;
2956 sk_msg_iter_var_next(i);
2958 if (pop < sge->length - a) {
2961 sk_msg_shift_right(msg, i);
2962 nsge = sk_msg_elem(msg, i);
2963 get_page(sg_page(sge));
2966 b, sge->offset + pop + a);
2968 struct page *page, *orig;
2971 page = alloc_pages(__GFP_NOWARN |
2972 __GFP_COMP | GFP_ATOMIC,
2974 if (unlikely(!page))
2978 orig = sg_page(sge);
2979 from = sg_virt(sge);
2980 to = page_address(page);
2981 memcpy(to, from, a);
2982 memcpy(to + a, from + a + pop, b);
2983 sg_set_page(sge, page, a + b, 0);
2987 } else if (pop >= sge->length - a) {
2988 pop -= (sge->length - a);
2993 /* From above the current layout _must_ be as follows,
2998 * |---- pop ---|---------------- b ------------|
2999 * |____________________________________________| length
3001 * Offset and start of the current msg elem are equal because in the
3002 * previous case we handled offset != start and either consumed the
3003 * entire element and advanced to the next element OR pop == 0.
3005 * Two cases to handle here are first pop is less than the length
3006 * leaving some remainder b above. Simply adjust the element's layout
3007 * in this case. Or pop >= length of the element so that b = 0. In this
3008 * case advance to next element decrementing pop.
3011 struct scatterlist *sge = sk_msg_elem(msg, i);
3013 if (pop < sge->length) {
3019 sk_msg_shift_left(msg, i);
3021 sk_msg_iter_var_next(i);
3024 sk_mem_uncharge(msg->sk, len - pop);
3025 msg->sg.size -= (len - pop);
3026 sk_msg_reset_curr(msg);
3027 sk_msg_compute_data_pointers(msg);
3031 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
3032 .func = bpf_msg_pop_data,
3034 .ret_type = RET_INTEGER,
3035 .arg1_type = ARG_PTR_TO_CTX,
3036 .arg2_type = ARG_ANYTHING,
3037 .arg3_type = ARG_ANYTHING,
3038 .arg4_type = ARG_ANYTHING,
3041 #ifdef CONFIG_CGROUP_NET_CLASSID
3042 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3044 return __task_get_classid(current);
3047 const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3048 .func = bpf_get_cgroup_classid_curr,
3050 .ret_type = RET_INTEGER,
3053 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3055 struct sock *sk = skb_to_full_sk(skb);
3057 if (!sk || !sk_fullsock(sk))
3060 return sock_cgroup_classid(&sk->sk_cgrp_data);
3063 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3064 .func = bpf_skb_cgroup_classid,
3066 .ret_type = RET_INTEGER,
3067 .arg1_type = ARG_PTR_TO_CTX,
3071 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3073 return task_get_classid(skb);
3076 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3077 .func = bpf_get_cgroup_classid,
3079 .ret_type = RET_INTEGER,
3080 .arg1_type = ARG_PTR_TO_CTX,
3083 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3085 return dst_tclassid(skb);
3088 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3089 .func = bpf_get_route_realm,
3091 .ret_type = RET_INTEGER,
3092 .arg1_type = ARG_PTR_TO_CTX,
3095 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3097 /* If skb_clear_hash() was called due to mangling, we can
3098 * trigger SW recalculation here. Later access to hash
3099 * can then use the inline skb->hash via context directly
3100 * instead of calling this helper again.
3102 return skb_get_hash(skb);
3105 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3106 .func = bpf_get_hash_recalc,
3108 .ret_type = RET_INTEGER,
3109 .arg1_type = ARG_PTR_TO_CTX,
3112 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3114 /* After all direct packet write, this can be used once for
3115 * triggering a lazy recalc on next skb_get_hash() invocation.
3117 skb_clear_hash(skb);
3121 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3122 .func = bpf_set_hash_invalid,
3124 .ret_type = RET_INTEGER,
3125 .arg1_type = ARG_PTR_TO_CTX,
3128 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3130 /* Set user specified hash as L4(+), so that it gets returned
3131 * on skb_get_hash() call unless BPF prog later on triggers a
3134 __skb_set_sw_hash(skb, hash, true);
3138 static const struct bpf_func_proto bpf_set_hash_proto = {
3139 .func = bpf_set_hash,
3141 .ret_type = RET_INTEGER,
3142 .arg1_type = ARG_PTR_TO_CTX,
3143 .arg2_type = ARG_ANYTHING,
3146 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3151 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3152 vlan_proto != htons(ETH_P_8021AD)))
3153 vlan_proto = htons(ETH_P_8021Q);
3155 bpf_push_mac_rcsum(skb);
3156 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3157 bpf_pull_mac_rcsum(skb);
3159 bpf_compute_data_pointers(skb);
3163 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3164 .func = bpf_skb_vlan_push,
3166 .ret_type = RET_INTEGER,
3167 .arg1_type = ARG_PTR_TO_CTX,
3168 .arg2_type = ARG_ANYTHING,
3169 .arg3_type = ARG_ANYTHING,
3172 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3176 bpf_push_mac_rcsum(skb);
3177 ret = skb_vlan_pop(skb);
3178 bpf_pull_mac_rcsum(skb);
3180 bpf_compute_data_pointers(skb);
3184 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3185 .func = bpf_skb_vlan_pop,
3187 .ret_type = RET_INTEGER,
3188 .arg1_type = ARG_PTR_TO_CTX,
3191 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3193 /* Caller already did skb_cow() with len as headroom,
3194 * so no need to do it here.
3197 memmove(skb->data, skb->data + len, off);
3198 memset(skb->data + off, 0, len);
3200 /* No skb_postpush_rcsum(skb, skb->data + off, len)
3201 * needed here as it does not change the skb->csum
3202 * result for checksum complete when summing over
3208 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3212 /* skb_ensure_writable() is not needed here, as we're
3213 * already working on an uncloned skb.
3215 if (unlikely(!pskb_may_pull(skb, off + len)))
3218 old_data = skb->data;
3219 __skb_pull(skb, len);
3220 skb_postpull_rcsum(skb, old_data + off, len);
3221 memmove(skb->data, old_data, off);
3226 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3228 bool trans_same = skb->transport_header == skb->network_header;
3231 /* There's no need for __skb_push()/__skb_pull() pair to
3232 * get to the start of the mac header as we're guaranteed
3233 * to always start from here under eBPF.
3235 ret = bpf_skb_generic_push(skb, off, len);
3237 skb->mac_header -= len;
3238 skb->network_header -= len;
3240 skb->transport_header = skb->network_header;
3246 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3248 bool trans_same = skb->transport_header == skb->network_header;
3251 /* Same here, __skb_push()/__skb_pull() pair not needed. */
3252 ret = bpf_skb_generic_pop(skb, off, len);
3254 skb->mac_header += len;
3255 skb->network_header += len;
3257 skb->transport_header = skb->network_header;
3263 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3265 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3266 u32 off = skb_mac_header_len(skb);
3269 ret = skb_cow(skb, len_diff);
3270 if (unlikely(ret < 0))
3273 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3274 if (unlikely(ret < 0))
3277 if (skb_is_gso(skb)) {
3278 struct skb_shared_info *shinfo = skb_shinfo(skb);
3280 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */
3281 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3282 shinfo->gso_type &= ~SKB_GSO_TCPV4;
3283 shinfo->gso_type |= SKB_GSO_TCPV6;
3287 skb->protocol = htons(ETH_P_IPV6);
3288 skb_clear_hash(skb);
3293 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3295 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3296 u32 off = skb_mac_header_len(skb);
3299 ret = skb_unclone(skb, GFP_ATOMIC);
3300 if (unlikely(ret < 0))
3303 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3304 if (unlikely(ret < 0))
3307 if (skb_is_gso(skb)) {
3308 struct skb_shared_info *shinfo = skb_shinfo(skb);
3310 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */
3311 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3312 shinfo->gso_type &= ~SKB_GSO_TCPV6;
3313 shinfo->gso_type |= SKB_GSO_TCPV4;
3317 skb->protocol = htons(ETH_P_IP);
3318 skb_clear_hash(skb);
3323 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3325 __be16 from_proto = skb->protocol;
3327 if (from_proto == htons(ETH_P_IP) &&
3328 to_proto == htons(ETH_P_IPV6))
3329 return bpf_skb_proto_4_to_6(skb);
3331 if (from_proto == htons(ETH_P_IPV6) &&
3332 to_proto == htons(ETH_P_IP))
3333 return bpf_skb_proto_6_to_4(skb);
3338 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3343 if (unlikely(flags))
3346 /* General idea is that this helper does the basic groundwork
3347 * needed for changing the protocol, and eBPF program fills the
3348 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3349 * and other helpers, rather than passing a raw buffer here.
3351 * The rationale is to keep this minimal and without a need to
3352 * deal with raw packet data. F.e. even if we would pass buffers
3353 * here, the program still needs to call the bpf_lX_csum_replace()
3354 * helpers anyway. Plus, this way we keep also separation of
3355 * concerns, since f.e. bpf_skb_store_bytes() should only take
3358 * Currently, additional options and extension header space are
3359 * not supported, but flags register is reserved so we can adapt
3360 * that. For offloads, we mark packet as dodgy, so that headers
3361 * need to be verified first.
3363 ret = bpf_skb_proto_xlat(skb, proto);
3364 bpf_compute_data_pointers(skb);
3368 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3369 .func = bpf_skb_change_proto,
3371 .ret_type = RET_INTEGER,
3372 .arg1_type = ARG_PTR_TO_CTX,
3373 .arg2_type = ARG_ANYTHING,
3374 .arg3_type = ARG_ANYTHING,
3377 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3379 /* We only allow a restricted subset to be changed for now. */
3380 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3381 !skb_pkt_type_ok(pkt_type)))
3384 skb->pkt_type = pkt_type;
3388 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3389 .func = bpf_skb_change_type,
3391 .ret_type = RET_INTEGER,
3392 .arg1_type = ARG_PTR_TO_CTX,
3393 .arg2_type = ARG_ANYTHING,
3396 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3398 switch (skb->protocol) {
3399 case htons(ETH_P_IP):
3400 return sizeof(struct iphdr);
3401 case htons(ETH_P_IPV6):
3402 return sizeof(struct ipv6hdr);
3408 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3409 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3411 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \
3412 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3413 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3414 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3415 BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \
3416 BPF_F_ADJ_ROOM_ENCAP_L2( \
3417 BPF_ADJ_ROOM_ENCAP_L2_MASK))
3419 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3422 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3423 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3424 u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3425 unsigned int gso_type = SKB_GSO_DODGY;
3428 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3429 /* udp gso_size delineates datagrams, only allow if fixed */
3430 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3431 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3435 ret = skb_cow_head(skb, len_diff);
3436 if (unlikely(ret < 0))
3440 if (skb->protocol != htons(ETH_P_IP) &&
3441 skb->protocol != htons(ETH_P_IPV6))
3444 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3445 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3448 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3449 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3452 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH &&
3453 inner_mac_len < ETH_HLEN)
3456 if (skb->encapsulation)
3459 mac_len = skb->network_header - skb->mac_header;
3460 inner_net = skb->network_header;
3461 if (inner_mac_len > len_diff)
3463 inner_trans = skb->transport_header;
3466 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3467 if (unlikely(ret < 0))
3471 skb->inner_mac_header = inner_net - inner_mac_len;
3472 skb->inner_network_header = inner_net;
3473 skb->inner_transport_header = inner_trans;
3475 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH)
3476 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
3478 skb_set_inner_protocol(skb, skb->protocol);
3480 skb->encapsulation = 1;
3481 skb_set_network_header(skb, mac_len);
3483 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3484 gso_type |= SKB_GSO_UDP_TUNNEL;
3485 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3486 gso_type |= SKB_GSO_GRE;
3487 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3488 gso_type |= SKB_GSO_IPXIP6;
3489 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3490 gso_type |= SKB_GSO_IPXIP4;
3492 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3493 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3494 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3495 sizeof(struct ipv6hdr) :
3496 sizeof(struct iphdr);
3498 skb_set_transport_header(skb, mac_len + nh_len);
3501 /* Match skb->protocol to new outer l3 protocol */
3502 if (skb->protocol == htons(ETH_P_IP) &&
3503 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3504 skb->protocol = htons(ETH_P_IPV6);
3505 else if (skb->protocol == htons(ETH_P_IPV6) &&
3506 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3507 skb->protocol = htons(ETH_P_IP);
3510 if (skb_is_gso(skb)) {
3511 struct skb_shared_info *shinfo = skb_shinfo(skb);
3513 /* Due to header grow, MSS needs to be downgraded. */
3514 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3515 skb_decrease_gso_size(shinfo, len_diff);
3517 /* Header must be checked, and gso_segs recomputed. */
3518 shinfo->gso_type |= gso_type;
3519 shinfo->gso_segs = 0;
3525 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3530 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3531 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3534 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3535 /* udp gso_size delineates datagrams, only allow if fixed */
3536 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3537 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3541 ret = skb_unclone(skb, GFP_ATOMIC);
3542 if (unlikely(ret < 0))
3545 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3546 if (unlikely(ret < 0))
3549 if (skb_is_gso(skb)) {
3550 struct skb_shared_info *shinfo = skb_shinfo(skb);
3552 /* Due to header shrink, MSS can be upgraded. */
3553 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3554 skb_increase_gso_size(shinfo, len_diff);
3556 /* Header must be checked, and gso_segs recomputed. */
3557 shinfo->gso_type |= SKB_GSO_DODGY;
3558 shinfo->gso_segs = 0;
3564 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3566 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3567 u32, mode, u64, flags)
3569 u32 len_diff_abs = abs(len_diff);
3570 bool shrink = len_diff < 0;
3573 if (unlikely(flags || mode))
3575 if (unlikely(len_diff_abs > 0xfffU))
3579 ret = skb_cow(skb, len_diff);
3580 if (unlikely(ret < 0))
3582 __skb_push(skb, len_diff_abs);
3583 memset(skb->data, 0, len_diff_abs);
3585 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3587 __skb_pull(skb, len_diff_abs);
3589 if (tls_sw_has_ctx_rx(skb->sk)) {
3590 struct strp_msg *rxm = strp_msg(skb);
3592 rxm->full_len += len_diff;
3597 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3598 .func = sk_skb_adjust_room,
3600 .ret_type = RET_INTEGER,
3601 .arg1_type = ARG_PTR_TO_CTX,
3602 .arg2_type = ARG_ANYTHING,
3603 .arg3_type = ARG_ANYTHING,
3604 .arg4_type = ARG_ANYTHING,
3607 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3608 u32, mode, u64, flags)
3610 u32 len_cur, len_diff_abs = abs(len_diff);
3611 u32 len_min = bpf_skb_net_base_len(skb);
3612 u32 len_max = BPF_SKB_MAX_LEN;
3613 __be16 proto = skb->protocol;
3614 bool shrink = len_diff < 0;
3618 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3619 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3621 if (unlikely(len_diff_abs > 0xfffU))
3623 if (unlikely(proto != htons(ETH_P_IP) &&
3624 proto != htons(ETH_P_IPV6)))
3627 off = skb_mac_header_len(skb);
3629 case BPF_ADJ_ROOM_NET:
3630 off += bpf_skb_net_base_len(skb);
3632 case BPF_ADJ_ROOM_MAC:
3638 len_cur = skb->len - skb_network_offset(skb);
3639 if ((shrink && (len_diff_abs >= len_cur ||
3640 len_cur - len_diff_abs < len_min)) ||
3641 (!shrink && (skb->len + len_diff_abs > len_max &&
3645 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3646 bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3647 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3648 __skb_reset_checksum_unnecessary(skb);
3650 bpf_compute_data_pointers(skb);
3654 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3655 .func = bpf_skb_adjust_room,
3657 .ret_type = RET_INTEGER,
3658 .arg1_type = ARG_PTR_TO_CTX,
3659 .arg2_type = ARG_ANYTHING,
3660 .arg3_type = ARG_ANYTHING,
3661 .arg4_type = ARG_ANYTHING,
3664 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3666 u32 min_len = skb_network_offset(skb);
3668 if (skb_transport_header_was_set(skb))
3669 min_len = skb_transport_offset(skb);
3670 if (skb->ip_summed == CHECKSUM_PARTIAL)
3671 min_len = skb_checksum_start_offset(skb) +
3672 skb->csum_offset + sizeof(__sum16);
3676 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3678 unsigned int old_len = skb->len;
3681 ret = __skb_grow_rcsum(skb, new_len);
3683 memset(skb->data + old_len, 0, new_len - old_len);
3687 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3689 return __skb_trim_rcsum(skb, new_len);
3692 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3695 u32 max_len = BPF_SKB_MAX_LEN;
3696 u32 min_len = __bpf_skb_min_len(skb);
3699 if (unlikely(flags || new_len > max_len || new_len < min_len))
3701 if (skb->encapsulation)
3704 /* The basic idea of this helper is that it's performing the
3705 * needed work to either grow or trim an skb, and eBPF program
3706 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3707 * bpf_lX_csum_replace() and others rather than passing a raw
3708 * buffer here. This one is a slow path helper and intended
3709 * for replies with control messages.
3711 * Like in bpf_skb_change_proto(), we want to keep this rather
3712 * minimal and without protocol specifics so that we are able
3713 * to separate concerns as in bpf_skb_store_bytes() should only
3714 * be the one responsible for writing buffers.
3716 * It's really expected to be a slow path operation here for
3717 * control message replies, so we're implicitly linearizing,
3718 * uncloning and drop offloads from the skb by this.
3720 ret = __bpf_try_make_writable(skb, skb->len);
3722 if (new_len > skb->len)
3723 ret = bpf_skb_grow_rcsum(skb, new_len);
3724 else if (new_len < skb->len)
3725 ret = bpf_skb_trim_rcsum(skb, new_len);
3726 if (!ret && skb_is_gso(skb))
3732 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3735 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3737 bpf_compute_data_pointers(skb);
3741 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3742 .func = bpf_skb_change_tail,
3744 .ret_type = RET_INTEGER,
3745 .arg1_type = ARG_PTR_TO_CTX,
3746 .arg2_type = ARG_ANYTHING,
3747 .arg3_type = ARG_ANYTHING,
3750 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3753 return __bpf_skb_change_tail(skb, new_len, flags);
3756 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3757 .func = sk_skb_change_tail,
3759 .ret_type = RET_INTEGER,
3760 .arg1_type = ARG_PTR_TO_CTX,
3761 .arg2_type = ARG_ANYTHING,
3762 .arg3_type = ARG_ANYTHING,
3765 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3768 u32 max_len = BPF_SKB_MAX_LEN;
3769 u32 new_len = skb->len + head_room;
3772 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3773 new_len < skb->len))
3776 ret = skb_cow(skb, head_room);
3778 /* Idea for this helper is that we currently only
3779 * allow to expand on mac header. This means that
3780 * skb->protocol network header, etc, stay as is.
3781 * Compared to bpf_skb_change_tail(), we're more
3782 * flexible due to not needing to linearize or
3783 * reset GSO. Intention for this helper is to be
3784 * used by an L3 skb that needs to push mac header
3785 * for redirection into L2 device.
3787 __skb_push(skb, head_room);
3788 memset(skb->data, 0, head_room);
3789 skb_reset_mac_header(skb);
3790 skb_reset_mac_len(skb);
3796 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3799 int ret = __bpf_skb_change_head(skb, head_room, flags);
3801 bpf_compute_data_pointers(skb);
3805 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3806 .func = bpf_skb_change_head,
3808 .ret_type = RET_INTEGER,
3809 .arg1_type = ARG_PTR_TO_CTX,
3810 .arg2_type = ARG_ANYTHING,
3811 .arg3_type = ARG_ANYTHING,
3814 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3817 return __bpf_skb_change_head(skb, head_room, flags);
3820 static const struct bpf_func_proto sk_skb_change_head_proto = {
3821 .func = sk_skb_change_head,
3823 .ret_type = RET_INTEGER,
3824 .arg1_type = ARG_PTR_TO_CTX,
3825 .arg2_type = ARG_ANYTHING,
3826 .arg3_type = ARG_ANYTHING,
3829 BPF_CALL_1(bpf_xdp_get_buff_len, struct xdp_buff*, xdp)
3831 return xdp_get_buff_len(xdp);
3834 static const struct bpf_func_proto bpf_xdp_get_buff_len_proto = {
3835 .func = bpf_xdp_get_buff_len,
3837 .ret_type = RET_INTEGER,
3838 .arg1_type = ARG_PTR_TO_CTX,
3841 BTF_ID_LIST_SINGLE(bpf_xdp_get_buff_len_bpf_ids, struct, xdp_buff)
3843 const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto = {
3844 .func = bpf_xdp_get_buff_len,
3846 .arg1_type = ARG_PTR_TO_BTF_ID,
3847 .arg1_btf_id = &bpf_xdp_get_buff_len_bpf_ids[0],
3850 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3852 return xdp_data_meta_unsupported(xdp) ? 0 :
3853 xdp->data - xdp->data_meta;
3856 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3858 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3859 unsigned long metalen = xdp_get_metalen(xdp);
3860 void *data_start = xdp_frame_end + metalen;
3861 void *data = xdp->data + offset;
3863 if (unlikely(data < data_start ||
3864 data > xdp->data_end - ETH_HLEN))
3868 memmove(xdp->data_meta + offset,
3869 xdp->data_meta, metalen);
3870 xdp->data_meta += offset;
3876 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3877 .func = bpf_xdp_adjust_head,
3879 .ret_type = RET_INTEGER,
3880 .arg1_type = ARG_PTR_TO_CTX,
3881 .arg2_type = ARG_ANYTHING,
3884 static void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off,
3885 void *buf, unsigned long len, bool flush)
3887 unsigned long ptr_len, ptr_off = 0;
3888 skb_frag_t *next_frag, *end_frag;
3889 struct skb_shared_info *sinfo;
3893 if (likely(xdp->data_end - xdp->data >= off + len)) {
3894 src = flush ? buf : xdp->data + off;
3895 dst = flush ? xdp->data + off : buf;
3896 memcpy(dst, src, len);
3900 sinfo = xdp_get_shared_info_from_buff(xdp);
3901 end_frag = &sinfo->frags[sinfo->nr_frags];
3902 next_frag = &sinfo->frags[0];
3904 ptr_len = xdp->data_end - xdp->data;
3905 ptr_buf = xdp->data;
3908 if (off < ptr_off + ptr_len) {
3909 unsigned long copy_off = off - ptr_off;
3910 unsigned long copy_len = min(len, ptr_len - copy_off);
3912 src = flush ? buf : ptr_buf + copy_off;
3913 dst = flush ? ptr_buf + copy_off : buf;
3914 memcpy(dst, src, copy_len);
3921 if (!len || next_frag == end_frag)
3925 ptr_buf = skb_frag_address(next_frag);
3926 ptr_len = skb_frag_size(next_frag);
3931 static void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len)
3933 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
3934 u32 size = xdp->data_end - xdp->data;
3935 void *addr = xdp->data;
3938 if (unlikely(offset > 0xffff || len > 0xffff))
3939 return ERR_PTR(-EFAULT);
3941 if (offset + len > xdp_get_buff_len(xdp))
3942 return ERR_PTR(-EINVAL);
3944 if (offset < size) /* linear area */
3948 for (i = 0; i < sinfo->nr_frags; i++) { /* paged area */
3949 u32 frag_size = skb_frag_size(&sinfo->frags[i]);
3951 if (offset < frag_size) {
3952 addr = skb_frag_address(&sinfo->frags[i]);
3956 offset -= frag_size;
3959 return offset + len <= size ? addr + offset : NULL;
3962 BPF_CALL_4(bpf_xdp_load_bytes, struct xdp_buff *, xdp, u32, offset,
3963 void *, buf, u32, len)
3967 ptr = bpf_xdp_pointer(xdp, offset, len);
3969 return PTR_ERR(ptr);
3972 bpf_xdp_copy_buf(xdp, offset, buf, len, false);
3974 memcpy(buf, ptr, len);
3979 static const struct bpf_func_proto bpf_xdp_load_bytes_proto = {
3980 .func = bpf_xdp_load_bytes,
3982 .ret_type = RET_INTEGER,
3983 .arg1_type = ARG_PTR_TO_CTX,
3984 .arg2_type = ARG_ANYTHING,
3985 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
3986 .arg4_type = ARG_CONST_SIZE,
3989 BPF_CALL_4(bpf_xdp_store_bytes, struct xdp_buff *, xdp, u32, offset,
3990 void *, buf, u32, len)
3994 ptr = bpf_xdp_pointer(xdp, offset, len);
3996 return PTR_ERR(ptr);
3999 bpf_xdp_copy_buf(xdp, offset, buf, len, true);
4001 memcpy(ptr, buf, len);
4006 static const struct bpf_func_proto bpf_xdp_store_bytes_proto = {
4007 .func = bpf_xdp_store_bytes,
4009 .ret_type = RET_INTEGER,
4010 .arg1_type = ARG_PTR_TO_CTX,
4011 .arg2_type = ARG_ANYTHING,
4012 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
4013 .arg4_type = ARG_CONST_SIZE,
4016 static int bpf_xdp_frags_increase_tail(struct xdp_buff *xdp, int offset)
4018 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
4019 skb_frag_t *frag = &sinfo->frags[sinfo->nr_frags - 1];
4020 struct xdp_rxq_info *rxq = xdp->rxq;
4021 unsigned int tailroom;
4023 if (!rxq->frag_size || rxq->frag_size > xdp->frame_sz)
4026 tailroom = rxq->frag_size - skb_frag_size(frag) - skb_frag_off(frag);
4027 if (unlikely(offset > tailroom))
4030 memset(skb_frag_address(frag) + skb_frag_size(frag), 0, offset);
4031 skb_frag_size_add(frag, offset);
4032 sinfo->xdp_frags_size += offset;
4037 static int bpf_xdp_frags_shrink_tail(struct xdp_buff *xdp, int offset)
4039 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
4040 int i, n_frags_free = 0, len_free = 0;
4042 if (unlikely(offset > (int)xdp_get_buff_len(xdp) - ETH_HLEN))
4045 for (i = sinfo->nr_frags - 1; i >= 0 && offset > 0; i--) {
4046 skb_frag_t *frag = &sinfo->frags[i];
4047 int shrink = min_t(int, offset, skb_frag_size(frag));
4052 if (skb_frag_size(frag) == shrink) {
4053 struct page *page = skb_frag_page(frag);
4055 __xdp_return(page_address(page), &xdp->rxq->mem,
4059 skb_frag_size_sub(frag, shrink);
4063 sinfo->nr_frags -= n_frags_free;
4064 sinfo->xdp_frags_size -= len_free;
4066 if (unlikely(!sinfo->nr_frags)) {
4067 xdp_buff_clear_frags_flag(xdp);
4068 xdp->data_end -= offset;
4074 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
4076 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
4077 void *data_end = xdp->data_end + offset;
4079 if (unlikely(xdp_buff_has_frags(xdp))) { /* non-linear xdp buff */
4081 return bpf_xdp_frags_shrink_tail(xdp, -offset);
4083 return bpf_xdp_frags_increase_tail(xdp, offset);
4086 /* Notice that xdp_data_hard_end have reserved some tailroom */
4087 if (unlikely(data_end > data_hard_end))
4090 if (unlikely(data_end < xdp->data + ETH_HLEN))
4093 /* Clear memory area on grow, can contain uninit kernel memory */
4095 memset(xdp->data_end, 0, offset);
4097 xdp->data_end = data_end;
4102 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
4103 .func = bpf_xdp_adjust_tail,
4105 .ret_type = RET_INTEGER,
4106 .arg1_type = ARG_PTR_TO_CTX,
4107 .arg2_type = ARG_ANYTHING,
4110 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
4112 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
4113 void *meta = xdp->data_meta + offset;
4114 unsigned long metalen = xdp->data - meta;
4116 if (xdp_data_meta_unsupported(xdp))
4118 if (unlikely(meta < xdp_frame_end ||
4121 if (unlikely(xdp_metalen_invalid(metalen)))
4124 xdp->data_meta = meta;
4129 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
4130 .func = bpf_xdp_adjust_meta,
4132 .ret_type = RET_INTEGER,
4133 .arg1_type = ARG_PTR_TO_CTX,
4134 .arg2_type = ARG_ANYTHING,
4137 /* XDP_REDIRECT works by a three-step process, implemented in the functions
4140 * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target
4141 * of the redirect and store it (along with some other metadata) in a per-CPU
4142 * struct bpf_redirect_info.
4144 * 2. When the program returns the XDP_REDIRECT return code, the driver will
4145 * call xdp_do_redirect() which will use the information in struct
4146 * bpf_redirect_info to actually enqueue the frame into a map type-specific
4147 * bulk queue structure.
4149 * 3. Before exiting its NAPI poll loop, the driver will call xdp_do_flush(),
4150 * which will flush all the different bulk queues, thus completing the
4153 * Pointers to the map entries will be kept around for this whole sequence of
4154 * steps, protected by RCU. However, there is no top-level rcu_read_lock() in
4155 * the core code; instead, the RCU protection relies on everything happening
4156 * inside a single NAPI poll sequence, which means it's between a pair of calls
4157 * to local_bh_disable()/local_bh_enable().
4159 * The map entries are marked as __rcu and the map code makes sure to
4160 * dereference those pointers with rcu_dereference_check() in a way that works
4161 * for both sections that to hold an rcu_read_lock() and sections that are
4162 * called from NAPI without a separate rcu_read_lock(). The code below does not
4163 * use RCU annotations, but relies on those in the map code.
4165 void xdp_do_flush(void)
4171 EXPORT_SYMBOL_GPL(xdp_do_flush);
4173 void bpf_clear_redirect_map(struct bpf_map *map)
4175 struct bpf_redirect_info *ri;
4178 for_each_possible_cpu(cpu) {
4179 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
4180 /* Avoid polluting remote cacheline due to writes if
4181 * not needed. Once we pass this test, we need the
4182 * cmpxchg() to make sure it hasn't been changed in
4183 * the meantime by remote CPU.
4185 if (unlikely(READ_ONCE(ri->map) == map))
4186 cmpxchg(&ri->map, map, NULL);
4190 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
4191 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key);
4193 u32 xdp_master_redirect(struct xdp_buff *xdp)
4195 struct net_device *master, *slave;
4196 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4198 master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev);
4199 slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp);
4200 if (slave && slave != xdp->rxq->dev) {
4201 /* The target device is different from the receiving device, so
4202 * redirect it to the new device.
4203 * Using XDP_REDIRECT gets the correct behaviour from XDP enabled
4204 * drivers to unmap the packet from their rx ring.
4206 ri->tgt_index = slave->ifindex;
4207 ri->map_id = INT_MAX;
4208 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4209 return XDP_REDIRECT;
4213 EXPORT_SYMBOL_GPL(xdp_master_redirect);
4215 static inline int __xdp_do_redirect_xsk(struct bpf_redirect_info *ri,
4216 struct net_device *dev,
4217 struct xdp_buff *xdp,
4218 struct bpf_prog *xdp_prog)
4220 enum bpf_map_type map_type = ri->map_type;
4221 void *fwd = ri->tgt_value;
4222 u32 map_id = ri->map_id;
4225 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4226 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4228 err = __xsk_map_redirect(fwd, xdp);
4232 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4235 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4239 static __always_inline int __xdp_do_redirect_frame(struct bpf_redirect_info *ri,
4240 struct net_device *dev,
4241 struct xdp_frame *xdpf,
4242 struct bpf_prog *xdp_prog)
4244 enum bpf_map_type map_type = ri->map_type;
4245 void *fwd = ri->tgt_value;
4246 u32 map_id = ri->map_id;
4247 u32 flags = ri->flags;
4248 struct bpf_map *map;
4251 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4253 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4255 if (unlikely(!xdpf)) {
4261 case BPF_MAP_TYPE_DEVMAP:
4263 case BPF_MAP_TYPE_DEVMAP_HASH:
4264 if (unlikely(flags & BPF_F_BROADCAST)) {
4265 map = READ_ONCE(ri->map);
4267 /* The map pointer is cleared when the map is being torn
4268 * down by bpf_clear_redirect_map()
4270 if (unlikely(!map)) {
4275 WRITE_ONCE(ri->map, NULL);
4276 err = dev_map_enqueue_multi(xdpf, dev, map,
4277 flags & BPF_F_EXCLUDE_INGRESS);
4279 err = dev_map_enqueue(fwd, xdpf, dev);
4282 case BPF_MAP_TYPE_CPUMAP:
4283 err = cpu_map_enqueue(fwd, xdpf, dev);
4285 case BPF_MAP_TYPE_UNSPEC:
4286 if (map_id == INT_MAX) {
4287 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4288 if (unlikely(!fwd)) {
4292 err = dev_xdp_enqueue(fwd, xdpf, dev);
4303 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4306 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4310 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
4311 struct bpf_prog *xdp_prog)
4313 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4314 enum bpf_map_type map_type = ri->map_type;
4316 /* XDP_REDIRECT is not fully supported yet for xdp frags since
4317 * not all XDP capable drivers can map non-linear xdp_frame in
4320 if (unlikely(xdp_buff_has_frags(xdp) &&
4321 map_type != BPF_MAP_TYPE_CPUMAP))
4324 if (map_type == BPF_MAP_TYPE_XSKMAP)
4325 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4327 return __xdp_do_redirect_frame(ri, dev, xdp_convert_buff_to_frame(xdp),
4330 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4332 int xdp_do_redirect_frame(struct net_device *dev, struct xdp_buff *xdp,
4333 struct xdp_frame *xdpf, struct bpf_prog *xdp_prog)
4335 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4336 enum bpf_map_type map_type = ri->map_type;
4338 if (map_type == BPF_MAP_TYPE_XSKMAP)
4339 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4341 return __xdp_do_redirect_frame(ri, dev, xdpf, xdp_prog);
4343 EXPORT_SYMBOL_GPL(xdp_do_redirect_frame);
4345 static int xdp_do_generic_redirect_map(struct net_device *dev,
4346 struct sk_buff *skb,
4347 struct xdp_buff *xdp,
4348 struct bpf_prog *xdp_prog, void *fwd,
4349 enum bpf_map_type map_type, u32 map_id,
4352 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4353 struct bpf_map *map;
4357 case BPF_MAP_TYPE_DEVMAP:
4359 case BPF_MAP_TYPE_DEVMAP_HASH:
4360 if (unlikely(flags & BPF_F_BROADCAST)) {
4361 map = READ_ONCE(ri->map);
4363 /* The map pointer is cleared when the map is being torn
4364 * down by bpf_clear_redirect_map()
4366 if (unlikely(!map)) {
4371 WRITE_ONCE(ri->map, NULL);
4372 err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4373 flags & BPF_F_EXCLUDE_INGRESS);
4375 err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4380 case BPF_MAP_TYPE_XSKMAP:
4381 err = xsk_generic_rcv(fwd, xdp);
4386 case BPF_MAP_TYPE_CPUMAP:
4387 err = cpu_map_generic_redirect(fwd, skb);
4396 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4399 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4403 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4404 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4406 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4407 enum bpf_map_type map_type = ri->map_type;
4408 void *fwd = ri->tgt_value;
4409 u32 map_id = ri->map_id;
4410 u32 flags = ri->flags;
4413 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4415 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4417 if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4418 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4419 if (unlikely(!fwd)) {
4424 err = xdp_ok_fwd_dev(fwd, skb->len);
4429 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4430 generic_xdp_tx(skb, xdp_prog);
4434 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id, flags);
4436 _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4440 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4442 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4444 if (unlikely(flags))
4447 /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4448 * by map_idr) is used for ifindex based XDP redirect.
4450 ri->tgt_index = ifindex;
4451 ri->map_id = INT_MAX;
4452 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4454 return XDP_REDIRECT;
4457 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4458 .func = bpf_xdp_redirect,
4460 .ret_type = RET_INTEGER,
4461 .arg1_type = ARG_ANYTHING,
4462 .arg2_type = ARG_ANYTHING,
4465 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
4468 return map->ops->map_redirect(map, ifindex, flags);
4471 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4472 .func = bpf_xdp_redirect_map,
4474 .ret_type = RET_INTEGER,
4475 .arg1_type = ARG_CONST_MAP_PTR,
4476 .arg2_type = ARG_ANYTHING,
4477 .arg3_type = ARG_ANYTHING,
4480 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4481 unsigned long off, unsigned long len)
4483 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4487 if (ptr != dst_buff)
4488 memcpy(dst_buff, ptr, len);
4493 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4494 u64, flags, void *, meta, u64, meta_size)
4496 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4498 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4500 if (unlikely(!skb || skb_size > skb->len))
4503 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4507 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4508 .func = bpf_skb_event_output,
4510 .ret_type = RET_INTEGER,
4511 .arg1_type = ARG_PTR_TO_CTX,
4512 .arg2_type = ARG_CONST_MAP_PTR,
4513 .arg3_type = ARG_ANYTHING,
4514 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4515 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4518 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4520 const struct bpf_func_proto bpf_skb_output_proto = {
4521 .func = bpf_skb_event_output,
4523 .ret_type = RET_INTEGER,
4524 .arg1_type = ARG_PTR_TO_BTF_ID,
4525 .arg1_btf_id = &bpf_skb_output_btf_ids[0],
4526 .arg2_type = ARG_CONST_MAP_PTR,
4527 .arg3_type = ARG_ANYTHING,
4528 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4529 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4532 static unsigned short bpf_tunnel_key_af(u64 flags)
4534 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4537 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4538 u32, size, u64, flags)
4540 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4541 u8 compat[sizeof(struct bpf_tunnel_key)];
4545 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6 |
4546 BPF_F_TUNINFO_FLAGS)))) {
4550 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4554 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4557 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4558 case offsetof(struct bpf_tunnel_key, tunnel_label):
4559 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4561 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4562 /* Fixup deprecated structure layouts here, so we have
4563 * a common path later on.
4565 if (ip_tunnel_info_af(info) != AF_INET)
4568 to = (struct bpf_tunnel_key *)compat;
4575 to->tunnel_id = be64_to_cpu(info->key.tun_id);
4576 to->tunnel_tos = info->key.tos;
4577 to->tunnel_ttl = info->key.ttl;
4578 if (flags & BPF_F_TUNINFO_FLAGS)
4579 to->tunnel_flags = info->key.tun_flags;
4583 if (flags & BPF_F_TUNINFO_IPV6) {
4584 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4585 sizeof(to->remote_ipv6));
4586 memcpy(to->local_ipv6, &info->key.u.ipv6.dst,
4587 sizeof(to->local_ipv6));
4588 to->tunnel_label = be32_to_cpu(info->key.label);
4590 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4591 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4592 to->local_ipv4 = be32_to_cpu(info->key.u.ipv4.dst);
4593 memset(&to->local_ipv6[1], 0, sizeof(__u32) * 3);
4594 to->tunnel_label = 0;
4597 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4598 memcpy(to_orig, to, size);
4602 memset(to_orig, 0, size);
4606 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4607 .func = bpf_skb_get_tunnel_key,
4609 .ret_type = RET_INTEGER,
4610 .arg1_type = ARG_PTR_TO_CTX,
4611 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4612 .arg3_type = ARG_CONST_SIZE,
4613 .arg4_type = ARG_ANYTHING,
4616 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4618 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4621 if (unlikely(!info ||
4622 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4626 if (unlikely(size < info->options_len)) {
4631 ip_tunnel_info_opts_get(to, info);
4632 if (size > info->options_len)
4633 memset(to + info->options_len, 0, size - info->options_len);
4635 return info->options_len;
4637 memset(to, 0, size);
4641 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4642 .func = bpf_skb_get_tunnel_opt,
4644 .ret_type = RET_INTEGER,
4645 .arg1_type = ARG_PTR_TO_CTX,
4646 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4647 .arg3_type = ARG_CONST_SIZE,
4650 static struct metadata_dst __percpu *md_dst;
4652 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4653 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4655 struct metadata_dst *md = this_cpu_ptr(md_dst);
4656 u8 compat[sizeof(struct bpf_tunnel_key)];
4657 struct ip_tunnel_info *info;
4659 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4660 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
4662 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4664 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4665 case offsetof(struct bpf_tunnel_key, tunnel_label):
4666 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4667 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4668 /* Fixup deprecated structure layouts here, so we have
4669 * a common path later on.
4671 memcpy(compat, from, size);
4672 memset(compat + size, 0, sizeof(compat) - size);
4673 from = (const struct bpf_tunnel_key *) compat;
4679 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4684 dst_hold((struct dst_entry *) md);
4685 skb_dst_set(skb, (struct dst_entry *) md);
4687 info = &md->u.tun_info;
4688 memset(info, 0, sizeof(*info));
4689 info->mode = IP_TUNNEL_INFO_TX;
4691 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4692 if (flags & BPF_F_DONT_FRAGMENT)
4693 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4694 if (flags & BPF_F_ZERO_CSUM_TX)
4695 info->key.tun_flags &= ~TUNNEL_CSUM;
4696 if (flags & BPF_F_SEQ_NUMBER)
4697 info->key.tun_flags |= TUNNEL_SEQ;
4699 info->key.tun_id = cpu_to_be64(from->tunnel_id);
4700 info->key.tos = from->tunnel_tos;
4701 info->key.ttl = from->tunnel_ttl;
4703 if (flags & BPF_F_TUNINFO_IPV6) {
4704 info->mode |= IP_TUNNEL_INFO_IPV6;
4705 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4706 sizeof(from->remote_ipv6));
4707 memcpy(&info->key.u.ipv6.src, from->local_ipv6,
4708 sizeof(from->local_ipv6));
4709 info->key.label = cpu_to_be32(from->tunnel_label) &
4710 IPV6_FLOWLABEL_MASK;
4712 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4713 info->key.u.ipv4.src = cpu_to_be32(from->local_ipv4);
4714 info->key.flow_flags = FLOWI_FLAG_ANYSRC;
4720 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4721 .func = bpf_skb_set_tunnel_key,
4723 .ret_type = RET_INTEGER,
4724 .arg1_type = ARG_PTR_TO_CTX,
4725 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4726 .arg3_type = ARG_CONST_SIZE,
4727 .arg4_type = ARG_ANYTHING,
4730 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4731 const u8 *, from, u32, size)
4733 struct ip_tunnel_info *info = skb_tunnel_info(skb);
4734 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4736 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4738 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4741 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4746 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4747 .func = bpf_skb_set_tunnel_opt,
4749 .ret_type = RET_INTEGER,
4750 .arg1_type = ARG_PTR_TO_CTX,
4751 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4752 .arg3_type = ARG_CONST_SIZE,
4755 static const struct bpf_func_proto *
4756 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4759 struct metadata_dst __percpu *tmp;
4761 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4766 if (cmpxchg(&md_dst, NULL, tmp))
4767 metadata_dst_free_percpu(tmp);
4771 case BPF_FUNC_skb_set_tunnel_key:
4772 return &bpf_skb_set_tunnel_key_proto;
4773 case BPF_FUNC_skb_set_tunnel_opt:
4774 return &bpf_skb_set_tunnel_opt_proto;
4780 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4783 struct bpf_array *array = container_of(map, struct bpf_array, map);
4784 struct cgroup *cgrp;
4787 sk = skb_to_full_sk(skb);
4788 if (!sk || !sk_fullsock(sk))
4790 if (unlikely(idx >= array->map.max_entries))
4793 cgrp = READ_ONCE(array->ptrs[idx]);
4794 if (unlikely(!cgrp))
4797 return sk_under_cgroup_hierarchy(sk, cgrp);
4800 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4801 .func = bpf_skb_under_cgroup,
4803 .ret_type = RET_INTEGER,
4804 .arg1_type = ARG_PTR_TO_CTX,
4805 .arg2_type = ARG_CONST_MAP_PTR,
4806 .arg3_type = ARG_ANYTHING,
4809 #ifdef CONFIG_SOCK_CGROUP_DATA
4810 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4812 struct cgroup *cgrp;
4814 sk = sk_to_full_sk(sk);
4815 if (!sk || !sk_fullsock(sk))
4818 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4819 return cgroup_id(cgrp);
4822 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4824 return __bpf_sk_cgroup_id(skb->sk);
4827 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4828 .func = bpf_skb_cgroup_id,
4830 .ret_type = RET_INTEGER,
4831 .arg1_type = ARG_PTR_TO_CTX,
4834 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4837 struct cgroup *ancestor;
4838 struct cgroup *cgrp;
4840 sk = sk_to_full_sk(sk);
4841 if (!sk || !sk_fullsock(sk))
4844 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4845 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4849 return cgroup_id(ancestor);
4852 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4855 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4858 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4859 .func = bpf_skb_ancestor_cgroup_id,
4861 .ret_type = RET_INTEGER,
4862 .arg1_type = ARG_PTR_TO_CTX,
4863 .arg2_type = ARG_ANYTHING,
4866 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4868 return __bpf_sk_cgroup_id(sk);
4871 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4872 .func = bpf_sk_cgroup_id,
4874 .ret_type = RET_INTEGER,
4875 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4878 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4880 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4883 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4884 .func = bpf_sk_ancestor_cgroup_id,
4886 .ret_type = RET_INTEGER,
4887 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4888 .arg2_type = ARG_ANYTHING,
4892 static unsigned long bpf_xdp_copy(void *dst, const void *ctx,
4893 unsigned long off, unsigned long len)
4895 struct xdp_buff *xdp = (struct xdp_buff *)ctx;
4897 bpf_xdp_copy_buf(xdp, off, dst, len, false);
4901 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4902 u64, flags, void *, meta, u64, meta_size)
4904 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4906 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4909 if (unlikely(!xdp || xdp_size > xdp_get_buff_len(xdp)))
4912 return bpf_event_output(map, flags, meta, meta_size, xdp,
4913 xdp_size, bpf_xdp_copy);
4916 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4917 .func = bpf_xdp_event_output,
4919 .ret_type = RET_INTEGER,
4920 .arg1_type = ARG_PTR_TO_CTX,
4921 .arg2_type = ARG_CONST_MAP_PTR,
4922 .arg3_type = ARG_ANYTHING,
4923 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4924 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4927 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4929 const struct bpf_func_proto bpf_xdp_output_proto = {
4930 .func = bpf_xdp_event_output,
4932 .ret_type = RET_INTEGER,
4933 .arg1_type = ARG_PTR_TO_BTF_ID,
4934 .arg1_btf_id = &bpf_xdp_output_btf_ids[0],
4935 .arg2_type = ARG_CONST_MAP_PTR,
4936 .arg3_type = ARG_ANYTHING,
4937 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4938 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4941 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4943 return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4946 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4947 .func = bpf_get_socket_cookie,
4949 .ret_type = RET_INTEGER,
4950 .arg1_type = ARG_PTR_TO_CTX,
4953 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4955 return __sock_gen_cookie(ctx->sk);
4958 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4959 .func = bpf_get_socket_cookie_sock_addr,
4961 .ret_type = RET_INTEGER,
4962 .arg1_type = ARG_PTR_TO_CTX,
4965 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4967 return __sock_gen_cookie(ctx);
4970 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4971 .func = bpf_get_socket_cookie_sock,
4973 .ret_type = RET_INTEGER,
4974 .arg1_type = ARG_PTR_TO_CTX,
4977 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
4979 return sk ? sock_gen_cookie(sk) : 0;
4982 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
4983 .func = bpf_get_socket_ptr_cookie,
4985 .ret_type = RET_INTEGER,
4986 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4989 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4991 return __sock_gen_cookie(ctx->sk);
4994 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4995 .func = bpf_get_socket_cookie_sock_ops,
4997 .ret_type = RET_INTEGER,
4998 .arg1_type = ARG_PTR_TO_CTX,
5001 static u64 __bpf_get_netns_cookie(struct sock *sk)
5003 const struct net *net = sk ? sock_net(sk) : &init_net;
5005 return net->net_cookie;
5008 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
5010 return __bpf_get_netns_cookie(ctx);
5013 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
5014 .func = bpf_get_netns_cookie_sock,
5016 .ret_type = RET_INTEGER,
5017 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5020 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
5022 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5025 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
5026 .func = bpf_get_netns_cookie_sock_addr,
5028 .ret_type = RET_INTEGER,
5029 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5032 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
5034 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5037 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = {
5038 .func = bpf_get_netns_cookie_sock_ops,
5040 .ret_type = RET_INTEGER,
5041 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5044 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx)
5046 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5049 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = {
5050 .func = bpf_get_netns_cookie_sk_msg,
5052 .ret_type = RET_INTEGER,
5053 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5056 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
5058 struct sock *sk = sk_to_full_sk(skb->sk);
5061 if (!sk || !sk_fullsock(sk))
5063 kuid = sock_net_uid(sock_net(sk), sk);
5064 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
5067 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
5068 .func = bpf_get_socket_uid,
5070 .ret_type = RET_INTEGER,
5071 .arg1_type = ARG_PTR_TO_CTX,
5074 static int sol_socket_sockopt(struct sock *sk, int optname,
5075 char *optval, int *optlen,
5087 case SO_MAX_PACING_RATE:
5088 case SO_BINDTOIFINDEX:
5090 if (*optlen != sizeof(int))
5093 case SO_BINDTODEVICE:
5100 if (optname == SO_BINDTODEVICE)
5102 return sk_getsockopt(sk, SOL_SOCKET, optname,
5103 KERNEL_SOCKPTR(optval),
5104 KERNEL_SOCKPTR(optlen));
5107 return sk_setsockopt(sk, SOL_SOCKET, optname,
5108 KERNEL_SOCKPTR(optval), *optlen);
5111 static int bpf_sol_tcp_setsockopt(struct sock *sk, int optname,
5112 char *optval, int optlen)
5114 struct tcp_sock *tp = tcp_sk(sk);
5115 unsigned long timeout;
5118 if (optlen != sizeof(int))
5121 val = *(int *)optval;
5123 /* Only some options are supported */
5126 if (val <= 0 || tp->data_segs_out > tp->syn_data)
5128 tcp_snd_cwnd_set(tp, val);
5130 case TCP_BPF_SNDCWND_CLAMP:
5133 tp->snd_cwnd_clamp = val;
5134 tp->snd_ssthresh = val;
5136 case TCP_BPF_DELACK_MAX:
5137 timeout = usecs_to_jiffies(val);
5138 if (timeout > TCP_DELACK_MAX ||
5139 timeout < TCP_TIMEOUT_MIN)
5141 inet_csk(sk)->icsk_delack_max = timeout;
5143 case TCP_BPF_RTO_MIN:
5144 timeout = usecs_to_jiffies(val);
5145 if (timeout > TCP_RTO_MIN ||
5146 timeout < TCP_TIMEOUT_MIN)
5148 inet_csk(sk)->icsk_rto_min = timeout;
5157 static int sol_tcp_sockopt_congestion(struct sock *sk, char *optval,
5158 int *optlen, bool getopt)
5160 struct tcp_sock *tp;
5167 if (!inet_csk(sk)->icsk_ca_ops)
5169 /* BPF expects NULL-terminated tcp-cc string */
5170 optval[--(*optlen)] = '\0';
5171 return do_tcp_getsockopt(sk, SOL_TCP, TCP_CONGESTION,
5172 KERNEL_SOCKPTR(optval),
5173 KERNEL_SOCKPTR(optlen));
5176 /* "cdg" is the only cc that alloc a ptr
5177 * in inet_csk_ca area. The bpf-tcp-cc may
5178 * overwrite this ptr after switching to cdg.
5180 if (*optlen >= sizeof("cdg") - 1 && !strncmp("cdg", optval, *optlen))
5183 /* It stops this looping
5185 * .init => bpf_setsockopt(tcp_cc) => .init =>
5186 * bpf_setsockopt(tcp_cc)" => .init => ....
5188 * The second bpf_setsockopt(tcp_cc) is not allowed
5189 * in order to break the loop when both .init
5190 * are the same bpf prog.
5192 * This applies even the second bpf_setsockopt(tcp_cc)
5193 * does not cause a loop. This limits only the first
5194 * '.init' can call bpf_setsockopt(TCP_CONGESTION) to
5195 * pick a fallback cc (eg. peer does not support ECN)
5196 * and the second '.init' cannot fallback to
5200 if (tp->bpf_chg_cc_inprogress)
5203 tp->bpf_chg_cc_inprogress = 1;
5204 ret = do_tcp_setsockopt(sk, SOL_TCP, TCP_CONGESTION,
5205 KERNEL_SOCKPTR(optval), *optlen);
5206 tp->bpf_chg_cc_inprogress = 0;
5210 static int sol_tcp_sockopt(struct sock *sk, int optname,
5211 char *optval, int *optlen,
5214 if (sk->sk_prot->setsockopt != tcp_setsockopt)
5224 case TCP_WINDOW_CLAMP:
5225 case TCP_THIN_LINEAR_TIMEOUTS:
5226 case TCP_USER_TIMEOUT:
5227 case TCP_NOTSENT_LOWAT:
5229 if (*optlen != sizeof(int))
5232 case TCP_CONGESTION:
5233 return sol_tcp_sockopt_congestion(sk, optval, optlen, getopt);
5241 return bpf_sol_tcp_setsockopt(sk, optname, optval, *optlen);
5245 if (optname == TCP_SAVED_SYN) {
5246 struct tcp_sock *tp = tcp_sk(sk);
5248 if (!tp->saved_syn ||
5249 *optlen > tcp_saved_syn_len(tp->saved_syn))
5251 memcpy(optval, tp->saved_syn->data, *optlen);
5252 /* It cannot free tp->saved_syn here because it
5253 * does not know if the user space still needs it.
5258 return do_tcp_getsockopt(sk, SOL_TCP, optname,
5259 KERNEL_SOCKPTR(optval),
5260 KERNEL_SOCKPTR(optlen));
5263 return do_tcp_setsockopt(sk, SOL_TCP, optname,
5264 KERNEL_SOCKPTR(optval), *optlen);
5267 static int sol_ip_sockopt(struct sock *sk, int optname,
5268 char *optval, int *optlen,
5271 if (sk->sk_family != AF_INET)
5276 if (*optlen != sizeof(int))
5284 return do_ip_getsockopt(sk, SOL_IP, optname,
5285 KERNEL_SOCKPTR(optval),
5286 KERNEL_SOCKPTR(optlen));
5288 return do_ip_setsockopt(sk, SOL_IP, optname,
5289 KERNEL_SOCKPTR(optval), *optlen);
5292 static int sol_ipv6_sockopt(struct sock *sk, int optname,
5293 char *optval, int *optlen,
5296 if (sk->sk_family != AF_INET6)
5301 case IPV6_AUTOFLOWLABEL:
5302 if (*optlen != sizeof(int))
5310 return ipv6_bpf_stub->ipv6_getsockopt(sk, SOL_IPV6, optname,
5311 KERNEL_SOCKPTR(optval),
5312 KERNEL_SOCKPTR(optlen));
5314 return ipv6_bpf_stub->ipv6_setsockopt(sk, SOL_IPV6, optname,
5315 KERNEL_SOCKPTR(optval), *optlen);
5318 static int __bpf_setsockopt(struct sock *sk, int level, int optname,
5319 char *optval, int optlen)
5321 if (!sk_fullsock(sk))
5324 if (level == SOL_SOCKET)
5325 return sol_socket_sockopt(sk, optname, optval, &optlen, false);
5326 else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5327 return sol_ip_sockopt(sk, optname, optval, &optlen, false);
5328 else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5329 return sol_ipv6_sockopt(sk, optname, optval, &optlen, false);
5330 else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5331 return sol_tcp_sockopt(sk, optname, optval, &optlen, false);
5336 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
5337 char *optval, int optlen)
5339 if (sk_fullsock(sk))
5340 sock_owned_by_me(sk);
5341 return __bpf_setsockopt(sk, level, optname, optval, optlen);
5344 static int __bpf_getsockopt(struct sock *sk, int level, int optname,
5345 char *optval, int optlen)
5347 int err, saved_optlen = optlen;
5349 if (!sk_fullsock(sk)) {
5354 if (level == SOL_SOCKET)
5355 err = sol_socket_sockopt(sk, optname, optval, &optlen, true);
5356 else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5357 err = sol_tcp_sockopt(sk, optname, optval, &optlen, true);
5358 else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5359 err = sol_ip_sockopt(sk, optname, optval, &optlen, true);
5360 else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5361 err = sol_ipv6_sockopt(sk, optname, optval, &optlen, true);
5368 if (optlen < saved_optlen)
5369 memset(optval + optlen, 0, saved_optlen - optlen);
5373 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
5374 char *optval, int optlen)
5376 if (sk_fullsock(sk))
5377 sock_owned_by_me(sk);
5378 return __bpf_getsockopt(sk, level, optname, optval, optlen);
5381 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level,
5382 int, optname, char *, optval, int, 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, u32 mtu)
5724 params->h_vlan_TCI = 0;
5725 params->h_vlan_proto = 0;
5727 params->mtu_result = mtu; /* union with tot_len */
5733 #if IS_ENABLED(CONFIG_INET)
5734 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5735 u32 flags, bool check_mtu)
5737 struct fib_nh_common *nhc;
5738 struct in_device *in_dev;
5739 struct neighbour *neigh;
5740 struct net_device *dev;
5741 struct fib_result res;
5746 dev = dev_get_by_index_rcu(net, params->ifindex);
5750 /* verify forwarding is enabled on this interface */
5751 in_dev = __in_dev_get_rcu(dev);
5752 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5753 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5755 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5757 fl4.flowi4_oif = params->ifindex;
5759 fl4.flowi4_iif = params->ifindex;
5762 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5763 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5764 fl4.flowi4_flags = 0;
5766 fl4.flowi4_proto = params->l4_protocol;
5767 fl4.daddr = params->ipv4_dst;
5768 fl4.saddr = params->ipv4_src;
5769 fl4.fl4_sport = params->sport;
5770 fl4.fl4_dport = params->dport;
5771 fl4.flowi4_multipath_hash = 0;
5773 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5774 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5775 struct fib_table *tb;
5777 if (flags & BPF_FIB_LOOKUP_TBID) {
5778 tbid = params->tbid;
5779 /* zero out for vlan output */
5783 tb = fib_get_table(net, tbid);
5785 return BPF_FIB_LKUP_RET_NOT_FWDED;
5787 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5789 fl4.flowi4_mark = 0;
5790 fl4.flowi4_secid = 0;
5791 fl4.flowi4_tun_key.tun_id = 0;
5792 fl4.flowi4_uid = sock_net_uid(net, NULL);
5794 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5798 /* map fib lookup errors to RTN_ type */
5800 return BPF_FIB_LKUP_RET_BLACKHOLE;
5801 if (err == -EHOSTUNREACH)
5802 return BPF_FIB_LKUP_RET_UNREACHABLE;
5804 return BPF_FIB_LKUP_RET_PROHIBIT;
5806 return BPF_FIB_LKUP_RET_NOT_FWDED;
5809 if (res.type != RTN_UNICAST)
5810 return BPF_FIB_LKUP_RET_NOT_FWDED;
5812 if (fib_info_num_path(res.fi) > 1)
5813 fib_select_path(net, &res, &fl4, NULL);
5816 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5817 if (params->tot_len > mtu) {
5818 params->mtu_result = mtu; /* union with tot_len */
5819 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5825 /* do not handle lwt encaps right now */
5826 if (nhc->nhc_lwtstate)
5827 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5831 params->rt_metric = res.fi->fib_priority;
5832 params->ifindex = dev->ifindex;
5834 if (flags & BPF_FIB_LOOKUP_SRC)
5835 params->ipv4_src = fib_result_prefsrc(net, &res);
5837 /* xdp and cls_bpf programs are run in RCU-bh so
5838 * rcu_read_lock_bh is not needed here
5840 if (likely(nhc->nhc_gw_family != AF_INET6)) {
5841 if (nhc->nhc_gw_family)
5842 params->ipv4_dst = nhc->nhc_gw.ipv4;
5844 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5846 params->family = AF_INET6;
5847 *dst = nhc->nhc_gw.ipv6;
5850 if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH)
5851 goto set_fwd_params;
5853 if (likely(nhc->nhc_gw_family != AF_INET6))
5854 neigh = __ipv4_neigh_lookup_noref(dev,
5855 (__force u32)params->ipv4_dst);
5857 neigh = __ipv6_neigh_lookup_noref_stub(dev, params->ipv6_dst);
5859 if (!neigh || !(READ_ONCE(neigh->nud_state) & NUD_VALID))
5860 return BPF_FIB_LKUP_RET_NO_NEIGH;
5861 memcpy(params->dmac, neigh->ha, ETH_ALEN);
5862 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5865 return bpf_fib_set_fwd_params(params, mtu);
5869 #if IS_ENABLED(CONFIG_IPV6)
5870 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5871 u32 flags, bool check_mtu)
5873 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5874 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5875 struct fib6_result res = {};
5876 struct neighbour *neigh;
5877 struct net_device *dev;
5878 struct inet6_dev *idev;
5884 /* link local addresses are never forwarded */
5885 if (rt6_need_strict(dst) || rt6_need_strict(src))
5886 return BPF_FIB_LKUP_RET_NOT_FWDED;
5888 dev = dev_get_by_index_rcu(net, params->ifindex);
5892 idev = __in6_dev_get_safely(dev);
5893 if (unlikely(!idev || !idev->cnf.forwarding))
5894 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5896 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5898 oif = fl6.flowi6_oif = params->ifindex;
5900 oif = fl6.flowi6_iif = params->ifindex;
5902 strict = RT6_LOOKUP_F_HAS_SADDR;
5904 fl6.flowlabel = params->flowinfo;
5905 fl6.flowi6_scope = 0;
5906 fl6.flowi6_flags = 0;
5909 fl6.flowi6_proto = params->l4_protocol;
5912 fl6.fl6_sport = params->sport;
5913 fl6.fl6_dport = params->dport;
5915 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5916 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5917 struct fib6_table *tb;
5919 if (flags & BPF_FIB_LOOKUP_TBID) {
5920 tbid = params->tbid;
5921 /* zero out for vlan output */
5925 tb = ipv6_stub->fib6_get_table(net, tbid);
5927 return BPF_FIB_LKUP_RET_NOT_FWDED;
5929 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5932 fl6.flowi6_mark = 0;
5933 fl6.flowi6_secid = 0;
5934 fl6.flowi6_tun_key.tun_id = 0;
5935 fl6.flowi6_uid = sock_net_uid(net, NULL);
5937 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5940 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5941 res.f6i == net->ipv6.fib6_null_entry))
5942 return BPF_FIB_LKUP_RET_NOT_FWDED;
5944 switch (res.fib6_type) {
5945 /* only unicast is forwarded */
5949 return BPF_FIB_LKUP_RET_BLACKHOLE;
5950 case RTN_UNREACHABLE:
5951 return BPF_FIB_LKUP_RET_UNREACHABLE;
5953 return BPF_FIB_LKUP_RET_PROHIBIT;
5955 return BPF_FIB_LKUP_RET_NOT_FWDED;
5958 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5959 fl6.flowi6_oif != 0, NULL, strict);
5962 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5963 if (params->tot_len > mtu) {
5964 params->mtu_result = mtu; /* union with tot_len */
5965 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5969 if (res.nh->fib_nh_lws)
5970 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5972 if (res.nh->fib_nh_gw_family)
5973 *dst = res.nh->fib_nh_gw6;
5975 dev = res.nh->fib_nh_dev;
5976 params->rt_metric = res.f6i->fib6_metric;
5977 params->ifindex = dev->ifindex;
5979 if (flags & BPF_FIB_LOOKUP_SRC) {
5980 if (res.f6i->fib6_prefsrc.plen) {
5981 *src = res.f6i->fib6_prefsrc.addr;
5983 err = ipv6_bpf_stub->ipv6_dev_get_saddr(net, dev,
5987 return BPF_FIB_LKUP_RET_NO_SRC_ADDR;
5991 if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH)
5992 goto set_fwd_params;
5994 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
5997 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5998 if (!neigh || !(READ_ONCE(neigh->nud_state) & NUD_VALID))
5999 return BPF_FIB_LKUP_RET_NO_NEIGH;
6000 memcpy(params->dmac, neigh->ha, ETH_ALEN);
6001 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
6004 return bpf_fib_set_fwd_params(params, mtu);
6008 #define BPF_FIB_LOOKUP_MASK (BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT | \
6009 BPF_FIB_LOOKUP_SKIP_NEIGH | BPF_FIB_LOOKUP_TBID | \
6012 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
6013 struct bpf_fib_lookup *, params, int, plen, u32, flags)
6015 if (plen < sizeof(*params))
6018 if (flags & ~BPF_FIB_LOOKUP_MASK)
6021 switch (params->family) {
6022 #if IS_ENABLED(CONFIG_INET)
6024 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
6027 #if IS_ENABLED(CONFIG_IPV6)
6029 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
6033 return -EAFNOSUPPORT;
6036 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
6037 .func = bpf_xdp_fib_lookup,
6039 .ret_type = RET_INTEGER,
6040 .arg1_type = ARG_PTR_TO_CTX,
6041 .arg2_type = ARG_PTR_TO_MEM,
6042 .arg3_type = ARG_CONST_SIZE,
6043 .arg4_type = ARG_ANYTHING,
6046 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
6047 struct bpf_fib_lookup *, params, int, plen, u32, flags)
6049 struct net *net = dev_net(skb->dev);
6050 int rc = -EAFNOSUPPORT;
6051 bool check_mtu = false;
6053 if (plen < sizeof(*params))
6056 if (flags & ~BPF_FIB_LOOKUP_MASK)
6059 if (params->tot_len)
6062 switch (params->family) {
6063 #if IS_ENABLED(CONFIG_INET)
6065 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
6068 #if IS_ENABLED(CONFIG_IPV6)
6070 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
6075 if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
6076 struct net_device *dev;
6078 /* When tot_len isn't provided by user, check skb
6079 * against MTU of FIB lookup resulting net_device
6081 dev = dev_get_by_index_rcu(net, params->ifindex);
6082 if (!is_skb_forwardable(dev, skb))
6083 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
6085 params->mtu_result = dev->mtu; /* union with tot_len */
6091 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
6092 .func = bpf_skb_fib_lookup,
6094 .ret_type = RET_INTEGER,
6095 .arg1_type = ARG_PTR_TO_CTX,
6096 .arg2_type = ARG_PTR_TO_MEM,
6097 .arg3_type = ARG_CONST_SIZE,
6098 .arg4_type = ARG_ANYTHING,
6101 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
6104 struct net *netns = dev_net(dev_curr);
6106 /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
6110 return dev_get_by_index_rcu(netns, ifindex);
6113 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
6114 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6116 int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6117 struct net_device *dev = skb->dev;
6118 int skb_len, dev_len;
6121 if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
6124 if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
6127 dev = __dev_via_ifindex(dev, ifindex);
6131 mtu = READ_ONCE(dev->mtu);
6133 dev_len = mtu + dev->hard_header_len;
6135 /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6136 skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
6138 skb_len += len_diff; /* minus result pass check */
6139 if (skb_len <= dev_len) {
6140 ret = BPF_MTU_CHK_RET_SUCCESS;
6143 /* At this point, skb->len exceed MTU, but as it include length of all
6144 * segments, it can still be below MTU. The SKB can possibly get
6145 * re-segmented in transmit path (see validate_xmit_skb). Thus, user
6146 * must choose if segs are to be MTU checked.
6148 if (skb_is_gso(skb)) {
6149 ret = BPF_MTU_CHK_RET_SUCCESS;
6151 if (flags & BPF_MTU_CHK_SEGS &&
6152 !skb_gso_validate_network_len(skb, mtu))
6153 ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
6156 /* BPF verifier guarantees valid pointer */
6162 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
6163 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6165 struct net_device *dev = xdp->rxq->dev;
6166 int xdp_len = xdp->data_end - xdp->data;
6167 int ret = BPF_MTU_CHK_RET_SUCCESS;
6170 /* XDP variant doesn't support multi-buffer segment check (yet) */
6171 if (unlikely(flags))
6174 dev = __dev_via_ifindex(dev, ifindex);
6178 mtu = READ_ONCE(dev->mtu);
6180 /* Add L2-header as dev MTU is L3 size */
6181 dev_len = mtu + dev->hard_header_len;
6183 /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6185 xdp_len = *mtu_len + dev->hard_header_len;
6187 xdp_len += len_diff; /* minus result pass check */
6188 if (xdp_len > dev_len)
6189 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6191 /* BPF verifier guarantees valid pointer */
6197 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
6198 .func = bpf_skb_check_mtu,
6200 .ret_type = RET_INTEGER,
6201 .arg1_type = ARG_PTR_TO_CTX,
6202 .arg2_type = ARG_ANYTHING,
6203 .arg3_type = ARG_PTR_TO_INT,
6204 .arg4_type = ARG_ANYTHING,
6205 .arg5_type = ARG_ANYTHING,
6208 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
6209 .func = bpf_xdp_check_mtu,
6211 .ret_type = RET_INTEGER,
6212 .arg1_type = ARG_PTR_TO_CTX,
6213 .arg2_type = ARG_ANYTHING,
6214 .arg3_type = ARG_PTR_TO_INT,
6215 .arg4_type = ARG_ANYTHING,
6216 .arg5_type = ARG_ANYTHING,
6219 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6220 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
6223 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
6225 if (!seg6_validate_srh(srh, len, false))
6229 case BPF_LWT_ENCAP_SEG6_INLINE:
6230 if (skb->protocol != htons(ETH_P_IPV6))
6233 err = seg6_do_srh_inline(skb, srh);
6235 case BPF_LWT_ENCAP_SEG6:
6236 skb_reset_inner_headers(skb);
6237 skb->encapsulation = 1;
6238 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
6244 bpf_compute_data_pointers(skb);
6248 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
6250 return seg6_lookup_nexthop(skb, NULL, 0);
6252 #endif /* CONFIG_IPV6_SEG6_BPF */
6254 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6255 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
6258 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
6262 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
6266 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6267 case BPF_LWT_ENCAP_SEG6:
6268 case BPF_LWT_ENCAP_SEG6_INLINE:
6269 return bpf_push_seg6_encap(skb, type, hdr, len);
6271 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6272 case BPF_LWT_ENCAP_IP:
6273 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
6280 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
6281 void *, hdr, u32, len)
6284 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6285 case BPF_LWT_ENCAP_IP:
6286 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
6293 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
6294 .func = bpf_lwt_in_push_encap,
6296 .ret_type = RET_INTEGER,
6297 .arg1_type = ARG_PTR_TO_CTX,
6298 .arg2_type = ARG_ANYTHING,
6299 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6300 .arg4_type = ARG_CONST_SIZE
6303 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
6304 .func = bpf_lwt_xmit_push_encap,
6306 .ret_type = RET_INTEGER,
6307 .arg1_type = ARG_PTR_TO_CTX,
6308 .arg2_type = ARG_ANYTHING,
6309 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6310 .arg4_type = ARG_CONST_SIZE
6313 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6314 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
6315 const void *, from, u32, len)
6317 struct seg6_bpf_srh_state *srh_state =
6318 this_cpu_ptr(&seg6_bpf_srh_states);
6319 struct ipv6_sr_hdr *srh = srh_state->srh;
6320 void *srh_tlvs, *srh_end, *ptr;
6326 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
6327 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
6329 ptr = skb->data + offset;
6330 if (ptr >= srh_tlvs && ptr + len <= srh_end)
6331 srh_state->valid = false;
6332 else if (ptr < (void *)&srh->flags ||
6333 ptr + len > (void *)&srh->segments)
6336 if (unlikely(bpf_try_make_writable(skb, offset + len)))
6338 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6340 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6342 memcpy(skb->data + offset, from, len);
6346 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
6347 .func = bpf_lwt_seg6_store_bytes,
6349 .ret_type = RET_INTEGER,
6350 .arg1_type = ARG_PTR_TO_CTX,
6351 .arg2_type = ARG_ANYTHING,
6352 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6353 .arg4_type = ARG_CONST_SIZE
6356 static void bpf_update_srh_state(struct sk_buff *skb)
6358 struct seg6_bpf_srh_state *srh_state =
6359 this_cpu_ptr(&seg6_bpf_srh_states);
6362 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
6363 srh_state->srh = NULL;
6365 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6366 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
6367 srh_state->valid = true;
6371 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
6372 u32, action, void *, param, u32, param_len)
6374 struct seg6_bpf_srh_state *srh_state =
6375 this_cpu_ptr(&seg6_bpf_srh_states);
6380 case SEG6_LOCAL_ACTION_END_X:
6381 if (!seg6_bpf_has_valid_srh(skb))
6383 if (param_len != sizeof(struct in6_addr))
6385 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
6386 case SEG6_LOCAL_ACTION_END_T:
6387 if (!seg6_bpf_has_valid_srh(skb))
6389 if (param_len != sizeof(int))
6391 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6392 case SEG6_LOCAL_ACTION_END_DT6:
6393 if (!seg6_bpf_has_valid_srh(skb))
6395 if (param_len != sizeof(int))
6398 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
6400 if (!pskb_pull(skb, hdroff))
6403 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
6404 skb_reset_network_header(skb);
6405 skb_reset_transport_header(skb);
6406 skb->encapsulation = 0;
6408 bpf_compute_data_pointers(skb);
6409 bpf_update_srh_state(skb);
6410 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6411 case SEG6_LOCAL_ACTION_END_B6:
6412 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6414 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
6417 bpf_update_srh_state(skb);
6420 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
6421 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6423 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
6426 bpf_update_srh_state(skb);
6434 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
6435 .func = bpf_lwt_seg6_action,
6437 .ret_type = RET_INTEGER,
6438 .arg1_type = ARG_PTR_TO_CTX,
6439 .arg2_type = ARG_ANYTHING,
6440 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6441 .arg4_type = ARG_CONST_SIZE
6444 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
6447 struct seg6_bpf_srh_state *srh_state =
6448 this_cpu_ptr(&seg6_bpf_srh_states);
6449 struct ipv6_sr_hdr *srh = srh_state->srh;
6450 void *srh_end, *srh_tlvs, *ptr;
6451 struct ipv6hdr *hdr;
6455 if (unlikely(srh == NULL))
6458 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
6459 ((srh->first_segment + 1) << 4));
6460 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
6462 ptr = skb->data + offset;
6464 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
6466 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
6470 ret = skb_cow_head(skb, len);
6471 if (unlikely(ret < 0))
6474 ret = bpf_skb_net_hdr_push(skb, offset, len);
6476 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6479 bpf_compute_data_pointers(skb);
6480 if (unlikely(ret < 0))
6483 hdr = (struct ipv6hdr *)skb->data;
6484 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6486 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6488 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6489 srh_state->hdrlen += len;
6490 srh_state->valid = false;
6494 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6495 .func = bpf_lwt_seg6_adjust_srh,
6497 .ret_type = RET_INTEGER,
6498 .arg1_type = ARG_PTR_TO_CTX,
6499 .arg2_type = ARG_ANYTHING,
6500 .arg3_type = ARG_ANYTHING,
6502 #endif /* CONFIG_IPV6_SEG6_BPF */
6505 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6506 int dif, int sdif, u8 family, u8 proto)
6508 struct inet_hashinfo *hinfo = net->ipv4.tcp_death_row.hashinfo;
6509 bool refcounted = false;
6510 struct sock *sk = NULL;
6512 if (family == AF_INET) {
6513 __be32 src4 = tuple->ipv4.saddr;
6514 __be32 dst4 = tuple->ipv4.daddr;
6516 if (proto == IPPROTO_TCP)
6517 sk = __inet_lookup(net, hinfo, NULL, 0,
6518 src4, tuple->ipv4.sport,
6519 dst4, tuple->ipv4.dport,
6520 dif, sdif, &refcounted);
6522 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6523 dst4, tuple->ipv4.dport,
6524 dif, sdif, &udp_table, NULL);
6525 #if IS_ENABLED(CONFIG_IPV6)
6527 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6528 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6530 if (proto == IPPROTO_TCP)
6531 sk = __inet6_lookup(net, hinfo, NULL, 0,
6532 src6, tuple->ipv6.sport,
6533 dst6, ntohs(tuple->ipv6.dport),
6534 dif, sdif, &refcounted);
6535 else if (likely(ipv6_bpf_stub))
6536 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6537 src6, tuple->ipv6.sport,
6538 dst6, tuple->ipv6.dport,
6544 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6545 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6551 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6552 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6554 static struct sock *
6555 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6556 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6557 u64 flags, int sdif)
6559 struct sock *sk = NULL;
6563 if (len == sizeof(tuple->ipv4))
6565 else if (len == sizeof(tuple->ipv6))
6570 if (unlikely(flags || !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6574 if (family == AF_INET)
6575 sdif = inet_sdif(skb);
6577 sdif = inet6_sdif(skb);
6580 if ((s32)netns_id < 0) {
6582 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6584 net = get_net_ns_by_id(caller_net, netns_id);
6587 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6595 static struct sock *
6596 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6597 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6598 u64 flags, int sdif)
6600 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6601 ifindex, proto, netns_id, flags,
6605 struct sock *sk2 = sk_to_full_sk(sk);
6607 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6608 * sock refcnt is decremented to prevent a request_sock leak.
6610 if (!sk_fullsock(sk2))
6614 /* Ensure there is no need to bump sk2 refcnt */
6615 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6616 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6626 static struct sock *
6627 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6628 u8 proto, u64 netns_id, u64 flags)
6630 struct net *caller_net;
6634 caller_net = dev_net(skb->dev);
6635 ifindex = skb->dev->ifindex;
6637 caller_net = sock_net(skb->sk);
6641 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6642 netns_id, flags, -1);
6645 static struct sock *
6646 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6647 u8 proto, u64 netns_id, u64 flags)
6649 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6653 struct sock *sk2 = sk_to_full_sk(sk);
6655 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6656 * sock refcnt is decremented to prevent a request_sock leak.
6658 if (!sk_fullsock(sk2))
6662 /* Ensure there is no need to bump sk2 refcnt */
6663 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6664 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6674 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6675 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6677 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6681 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6682 .func = bpf_skc_lookup_tcp,
6685 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6686 .arg1_type = ARG_PTR_TO_CTX,
6687 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6688 .arg3_type = ARG_CONST_SIZE,
6689 .arg4_type = ARG_ANYTHING,
6690 .arg5_type = ARG_ANYTHING,
6693 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6694 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6696 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6700 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6701 .func = bpf_sk_lookup_tcp,
6704 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6705 .arg1_type = ARG_PTR_TO_CTX,
6706 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6707 .arg3_type = ARG_CONST_SIZE,
6708 .arg4_type = ARG_ANYTHING,
6709 .arg5_type = ARG_ANYTHING,
6712 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6713 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6715 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6719 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6720 .func = bpf_sk_lookup_udp,
6723 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6724 .arg1_type = ARG_PTR_TO_CTX,
6725 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6726 .arg3_type = ARG_CONST_SIZE,
6727 .arg4_type = ARG_ANYTHING,
6728 .arg5_type = ARG_ANYTHING,
6731 BPF_CALL_5(bpf_tc_skc_lookup_tcp, struct sk_buff *, skb,
6732 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6734 struct net_device *dev = skb->dev;
6735 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6736 struct net *caller_net = dev_net(dev);
6738 return (unsigned long)__bpf_skc_lookup(skb, tuple, len, caller_net,
6739 ifindex, IPPROTO_TCP, netns_id,
6743 static const struct bpf_func_proto bpf_tc_skc_lookup_tcp_proto = {
6744 .func = bpf_tc_skc_lookup_tcp,
6747 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6748 .arg1_type = ARG_PTR_TO_CTX,
6749 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6750 .arg3_type = ARG_CONST_SIZE,
6751 .arg4_type = ARG_ANYTHING,
6752 .arg5_type = ARG_ANYTHING,
6755 BPF_CALL_5(bpf_tc_sk_lookup_tcp, struct sk_buff *, skb,
6756 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6758 struct net_device *dev = skb->dev;
6759 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6760 struct net *caller_net = dev_net(dev);
6762 return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6763 ifindex, IPPROTO_TCP, netns_id,
6767 static const struct bpf_func_proto bpf_tc_sk_lookup_tcp_proto = {
6768 .func = bpf_tc_sk_lookup_tcp,
6771 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6772 .arg1_type = ARG_PTR_TO_CTX,
6773 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6774 .arg3_type = ARG_CONST_SIZE,
6775 .arg4_type = ARG_ANYTHING,
6776 .arg5_type = ARG_ANYTHING,
6779 BPF_CALL_5(bpf_tc_sk_lookup_udp, struct sk_buff *, skb,
6780 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6782 struct net_device *dev = skb->dev;
6783 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6784 struct net *caller_net = dev_net(dev);
6786 return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6787 ifindex, IPPROTO_UDP, netns_id,
6791 static const struct bpf_func_proto bpf_tc_sk_lookup_udp_proto = {
6792 .func = bpf_tc_sk_lookup_udp,
6795 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6796 .arg1_type = ARG_PTR_TO_CTX,
6797 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6798 .arg3_type = ARG_CONST_SIZE,
6799 .arg4_type = ARG_ANYTHING,
6800 .arg5_type = ARG_ANYTHING,
6803 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6805 if (sk && sk_is_refcounted(sk))
6810 static const struct bpf_func_proto bpf_sk_release_proto = {
6811 .func = bpf_sk_release,
6813 .ret_type = RET_INTEGER,
6814 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON | OBJ_RELEASE,
6817 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6818 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6820 struct net_device *dev = ctx->rxq->dev;
6821 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6822 struct net *caller_net = dev_net(dev);
6824 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6825 ifindex, IPPROTO_UDP, netns_id,
6829 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6830 .func = bpf_xdp_sk_lookup_udp,
6833 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6834 .arg1_type = ARG_PTR_TO_CTX,
6835 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6836 .arg3_type = ARG_CONST_SIZE,
6837 .arg4_type = ARG_ANYTHING,
6838 .arg5_type = ARG_ANYTHING,
6841 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6842 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6844 struct net_device *dev = ctx->rxq->dev;
6845 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6846 struct net *caller_net = dev_net(dev);
6848 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6849 ifindex, IPPROTO_TCP, netns_id,
6853 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6854 .func = bpf_xdp_skc_lookup_tcp,
6857 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6858 .arg1_type = ARG_PTR_TO_CTX,
6859 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6860 .arg3_type = ARG_CONST_SIZE,
6861 .arg4_type = ARG_ANYTHING,
6862 .arg5_type = ARG_ANYTHING,
6865 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6866 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6868 struct net_device *dev = ctx->rxq->dev;
6869 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6870 struct net *caller_net = dev_net(dev);
6872 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6873 ifindex, IPPROTO_TCP, netns_id,
6877 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6878 .func = bpf_xdp_sk_lookup_tcp,
6881 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6882 .arg1_type = ARG_PTR_TO_CTX,
6883 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6884 .arg3_type = ARG_CONST_SIZE,
6885 .arg4_type = ARG_ANYTHING,
6886 .arg5_type = ARG_ANYTHING,
6889 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6890 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6892 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6893 sock_net(ctx->sk), 0,
6894 IPPROTO_TCP, netns_id, flags,
6898 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6899 .func = bpf_sock_addr_skc_lookup_tcp,
6901 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6902 .arg1_type = ARG_PTR_TO_CTX,
6903 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6904 .arg3_type = ARG_CONST_SIZE,
6905 .arg4_type = ARG_ANYTHING,
6906 .arg5_type = ARG_ANYTHING,
6909 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6910 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6912 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6913 sock_net(ctx->sk), 0, IPPROTO_TCP,
6914 netns_id, flags, -1);
6917 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6918 .func = bpf_sock_addr_sk_lookup_tcp,
6920 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6921 .arg1_type = ARG_PTR_TO_CTX,
6922 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6923 .arg3_type = ARG_CONST_SIZE,
6924 .arg4_type = ARG_ANYTHING,
6925 .arg5_type = ARG_ANYTHING,
6928 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6929 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6931 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6932 sock_net(ctx->sk), 0, IPPROTO_UDP,
6933 netns_id, flags, -1);
6936 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6937 .func = bpf_sock_addr_sk_lookup_udp,
6939 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6940 .arg1_type = ARG_PTR_TO_CTX,
6941 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6942 .arg3_type = ARG_CONST_SIZE,
6943 .arg4_type = ARG_ANYTHING,
6944 .arg5_type = ARG_ANYTHING,
6947 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6948 struct bpf_insn_access_aux *info)
6950 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6954 if (off % size != 0)
6958 case offsetof(struct bpf_tcp_sock, bytes_received):
6959 case offsetof(struct bpf_tcp_sock, bytes_acked):
6960 return size == sizeof(__u64);
6962 return size == sizeof(__u32);
6966 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6967 const struct bpf_insn *si,
6968 struct bpf_insn *insn_buf,
6969 struct bpf_prog *prog, u32 *target_size)
6971 struct bpf_insn *insn = insn_buf;
6973 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \
6975 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \
6976 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6977 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6978 si->dst_reg, si->src_reg, \
6979 offsetof(struct tcp_sock, FIELD)); \
6982 #define BPF_INET_SOCK_GET_COMMON(FIELD) \
6984 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \
6986 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6987 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6988 struct inet_connection_sock, \
6990 si->dst_reg, si->src_reg, \
6992 struct inet_connection_sock, \
6996 if (insn > insn_buf)
6997 return insn - insn_buf;
7000 case offsetof(struct bpf_tcp_sock, rtt_min):
7001 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
7002 sizeof(struct minmax));
7003 BUILD_BUG_ON(sizeof(struct minmax) <
7004 sizeof(struct minmax_sample));
7006 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7007 offsetof(struct tcp_sock, rtt_min) +
7008 offsetof(struct minmax_sample, v));
7010 case offsetof(struct bpf_tcp_sock, snd_cwnd):
7011 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
7013 case offsetof(struct bpf_tcp_sock, srtt_us):
7014 BPF_TCP_SOCK_GET_COMMON(srtt_us);
7016 case offsetof(struct bpf_tcp_sock, snd_ssthresh):
7017 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
7019 case offsetof(struct bpf_tcp_sock, rcv_nxt):
7020 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
7022 case offsetof(struct bpf_tcp_sock, snd_nxt):
7023 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
7025 case offsetof(struct bpf_tcp_sock, snd_una):
7026 BPF_TCP_SOCK_GET_COMMON(snd_una);
7028 case offsetof(struct bpf_tcp_sock, mss_cache):
7029 BPF_TCP_SOCK_GET_COMMON(mss_cache);
7031 case offsetof(struct bpf_tcp_sock, ecn_flags):
7032 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
7034 case offsetof(struct bpf_tcp_sock, rate_delivered):
7035 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
7037 case offsetof(struct bpf_tcp_sock, rate_interval_us):
7038 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
7040 case offsetof(struct bpf_tcp_sock, packets_out):
7041 BPF_TCP_SOCK_GET_COMMON(packets_out);
7043 case offsetof(struct bpf_tcp_sock, retrans_out):
7044 BPF_TCP_SOCK_GET_COMMON(retrans_out);
7046 case offsetof(struct bpf_tcp_sock, total_retrans):
7047 BPF_TCP_SOCK_GET_COMMON(total_retrans);
7049 case offsetof(struct bpf_tcp_sock, segs_in):
7050 BPF_TCP_SOCK_GET_COMMON(segs_in);
7052 case offsetof(struct bpf_tcp_sock, data_segs_in):
7053 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
7055 case offsetof(struct bpf_tcp_sock, segs_out):
7056 BPF_TCP_SOCK_GET_COMMON(segs_out);
7058 case offsetof(struct bpf_tcp_sock, data_segs_out):
7059 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
7061 case offsetof(struct bpf_tcp_sock, lost_out):
7062 BPF_TCP_SOCK_GET_COMMON(lost_out);
7064 case offsetof(struct bpf_tcp_sock, sacked_out):
7065 BPF_TCP_SOCK_GET_COMMON(sacked_out);
7067 case offsetof(struct bpf_tcp_sock, bytes_received):
7068 BPF_TCP_SOCK_GET_COMMON(bytes_received);
7070 case offsetof(struct bpf_tcp_sock, bytes_acked):
7071 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
7073 case offsetof(struct bpf_tcp_sock, dsack_dups):
7074 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
7076 case offsetof(struct bpf_tcp_sock, delivered):
7077 BPF_TCP_SOCK_GET_COMMON(delivered);
7079 case offsetof(struct bpf_tcp_sock, delivered_ce):
7080 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
7082 case offsetof(struct bpf_tcp_sock, icsk_retransmits):
7083 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
7087 return insn - insn_buf;
7090 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
7092 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
7093 return (unsigned long)sk;
7095 return (unsigned long)NULL;
7098 const struct bpf_func_proto bpf_tcp_sock_proto = {
7099 .func = bpf_tcp_sock,
7101 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
7102 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
7105 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
7107 sk = sk_to_full_sk(sk);
7109 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
7110 return (unsigned long)sk;
7112 return (unsigned long)NULL;
7115 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
7116 .func = bpf_get_listener_sock,
7118 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
7119 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
7122 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
7124 unsigned int iphdr_len;
7126 switch (skb_protocol(skb, true)) {
7127 case cpu_to_be16(ETH_P_IP):
7128 iphdr_len = sizeof(struct iphdr);
7130 case cpu_to_be16(ETH_P_IPV6):
7131 iphdr_len = sizeof(struct ipv6hdr);
7137 if (skb_headlen(skb) < iphdr_len)
7140 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
7143 return INET_ECN_set_ce(skb);
7146 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7147 struct bpf_insn_access_aux *info)
7149 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
7152 if (off % size != 0)
7157 return size == sizeof(__u32);
7161 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
7162 const struct bpf_insn *si,
7163 struct bpf_insn *insn_buf,
7164 struct bpf_prog *prog, u32 *target_size)
7166 struct bpf_insn *insn = insn_buf;
7168 #define BPF_XDP_SOCK_GET(FIELD) \
7170 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \
7171 sizeof_field(struct bpf_xdp_sock, FIELD)); \
7172 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
7173 si->dst_reg, si->src_reg, \
7174 offsetof(struct xdp_sock, FIELD)); \
7178 case offsetof(struct bpf_xdp_sock, queue_id):
7179 BPF_XDP_SOCK_GET(queue_id);
7183 return insn - insn_buf;
7186 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
7187 .func = bpf_skb_ecn_set_ce,
7189 .ret_type = RET_INTEGER,
7190 .arg1_type = ARG_PTR_TO_CTX,
7193 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7194 struct tcphdr *, th, u32, th_len)
7196 #ifdef CONFIG_SYN_COOKIES
7200 if (unlikely(!sk || th_len < sizeof(*th)))
7203 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
7204 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7207 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7210 if (!th->ack || th->rst || th->syn)
7213 if (unlikely(iph_len < sizeof(struct iphdr)))
7216 if (tcp_synq_no_recent_overflow(sk))
7219 cookie = ntohl(th->ack_seq) - 1;
7221 /* Both struct iphdr and struct ipv6hdr have the version field at the
7222 * same offset so we can cast to the shorter header (struct iphdr).
7224 switch (((struct iphdr *)iph)->version) {
7226 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7229 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
7232 #if IS_BUILTIN(CONFIG_IPV6)
7234 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7237 if (sk->sk_family != AF_INET6)
7240 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
7242 #endif /* CONFIG_IPV6 */
7245 return -EPROTONOSUPPORT;
7257 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
7258 .func = bpf_tcp_check_syncookie,
7261 .ret_type = RET_INTEGER,
7262 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7263 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7264 .arg3_type = ARG_CONST_SIZE,
7265 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7266 .arg5_type = ARG_CONST_SIZE,
7269 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7270 struct tcphdr *, th, u32, th_len)
7272 #ifdef CONFIG_SYN_COOKIES
7276 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
7279 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7282 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7285 if (!th->syn || th->ack || th->fin || th->rst)
7288 if (unlikely(iph_len < sizeof(struct iphdr)))
7291 /* Both struct iphdr and struct ipv6hdr have the version field at the
7292 * same offset so we can cast to the shorter header (struct iphdr).
7294 switch (((struct iphdr *)iph)->version) {
7296 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7299 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
7302 #if IS_BUILTIN(CONFIG_IPV6)
7304 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7307 if (sk->sk_family != AF_INET6)
7310 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
7312 #endif /* CONFIG_IPV6 */
7315 return -EPROTONOSUPPORT;
7320 return cookie | ((u64)mss << 32);
7323 #endif /* CONFIG_SYN_COOKIES */
7326 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
7327 .func = bpf_tcp_gen_syncookie,
7328 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */
7330 .ret_type = RET_INTEGER,
7331 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7332 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7333 .arg3_type = ARG_CONST_SIZE,
7334 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7335 .arg5_type = ARG_CONST_SIZE,
7338 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
7340 if (!sk || flags != 0)
7342 if (!skb_at_tc_ingress(skb))
7344 if (unlikely(dev_net(skb->dev) != sock_net(sk)))
7345 return -ENETUNREACH;
7346 if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
7347 return -ESOCKTNOSUPPORT;
7348 if (sk_unhashed(sk))
7350 if (sk_is_refcounted(sk) &&
7351 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
7356 skb->destructor = sock_pfree;
7361 static const struct bpf_func_proto bpf_sk_assign_proto = {
7362 .func = bpf_sk_assign,
7364 .ret_type = RET_INTEGER,
7365 .arg1_type = ARG_PTR_TO_CTX,
7366 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7367 .arg3_type = ARG_ANYTHING,
7370 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
7371 u8 search_kind, const u8 *magic,
7372 u8 magic_len, bool *eol)
7378 while (op < opend) {
7381 if (kind == TCPOPT_EOL) {
7383 return ERR_PTR(-ENOMSG);
7384 } else if (kind == TCPOPT_NOP) {
7389 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
7390 /* Something is wrong in the received header.
7391 * Follow the TCP stack's tcp_parse_options()
7392 * and just bail here.
7394 return ERR_PTR(-EFAULT);
7397 if (search_kind == kind) {
7401 if (magic_len > kind_len - 2)
7402 return ERR_PTR(-ENOMSG);
7404 if (!memcmp(&op[2], magic, magic_len))
7411 return ERR_PTR(-ENOMSG);
7414 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7415 void *, search_res, u32, len, u64, flags)
7417 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
7418 const u8 *op, *opend, *magic, *search = search_res;
7419 u8 search_kind, search_len, copy_len, magic_len;
7422 /* 2 byte is the minimal option len except TCPOPT_NOP and
7423 * TCPOPT_EOL which are useless for the bpf prog to learn
7424 * and this helper disallow loading them also.
7426 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
7429 search_kind = search[0];
7430 search_len = search[1];
7432 if (search_len > len || search_kind == TCPOPT_NOP ||
7433 search_kind == TCPOPT_EOL)
7436 if (search_kind == TCPOPT_EXP || search_kind == 253) {
7437 /* 16 or 32 bit magic. +2 for kind and kind length */
7438 if (search_len != 4 && search_len != 6)
7441 magic_len = search_len - 2;
7450 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
7455 op += sizeof(struct tcphdr);
7457 if (!bpf_sock->skb ||
7458 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7459 /* This bpf_sock->op cannot call this helper */
7462 opend = bpf_sock->skb_data_end;
7463 op = bpf_sock->skb->data + sizeof(struct tcphdr);
7466 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
7473 if (copy_len > len) {
7478 memcpy(search_res, op, copy_len);
7482 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
7483 .func = bpf_sock_ops_load_hdr_opt,
7485 .ret_type = RET_INTEGER,
7486 .arg1_type = ARG_PTR_TO_CTX,
7487 .arg2_type = ARG_PTR_TO_MEM,
7488 .arg3_type = ARG_CONST_SIZE,
7489 .arg4_type = ARG_ANYTHING,
7492 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7493 const void *, from, u32, len, u64, flags)
7495 u8 new_kind, new_kind_len, magic_len = 0, *opend;
7496 const u8 *op, *new_op, *magic = NULL;
7497 struct sk_buff *skb;
7500 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
7503 if (len < 2 || flags)
7507 new_kind = new_op[0];
7508 new_kind_len = new_op[1];
7510 if (new_kind_len > len || new_kind == TCPOPT_NOP ||
7511 new_kind == TCPOPT_EOL)
7514 if (new_kind_len > bpf_sock->remaining_opt_len)
7517 /* 253 is another experimental kind */
7518 if (new_kind == TCPOPT_EXP || new_kind == 253) {
7519 if (new_kind_len < 4)
7521 /* Match for the 2 byte magic also.
7522 * RFC 6994: the magic could be 2 or 4 bytes.
7523 * Hence, matching by 2 byte only is on the
7524 * conservative side but it is the right
7525 * thing to do for the 'search-for-duplication'
7532 /* Check for duplication */
7533 skb = bpf_sock->skb;
7534 op = skb->data + sizeof(struct tcphdr);
7535 opend = bpf_sock->skb_data_end;
7537 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
7542 if (PTR_ERR(op) != -ENOMSG)
7546 /* The option has been ended. Treat it as no more
7547 * header option can be written.
7551 /* No duplication found. Store the header option. */
7552 memcpy(opend, from, new_kind_len);
7554 bpf_sock->remaining_opt_len -= new_kind_len;
7555 bpf_sock->skb_data_end += new_kind_len;
7560 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
7561 .func = bpf_sock_ops_store_hdr_opt,
7563 .ret_type = RET_INTEGER,
7564 .arg1_type = ARG_PTR_TO_CTX,
7565 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7566 .arg3_type = ARG_CONST_SIZE,
7567 .arg4_type = ARG_ANYTHING,
7570 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7571 u32, len, u64, flags)
7573 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7576 if (flags || len < 2)
7579 if (len > bpf_sock->remaining_opt_len)
7582 bpf_sock->remaining_opt_len -= len;
7587 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7588 .func = bpf_sock_ops_reserve_hdr_opt,
7590 .ret_type = RET_INTEGER,
7591 .arg1_type = ARG_PTR_TO_CTX,
7592 .arg2_type = ARG_ANYTHING,
7593 .arg3_type = ARG_ANYTHING,
7596 BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb,
7597 u64, tstamp, u32, tstamp_type)
7599 /* skb_clear_delivery_time() is done for inet protocol */
7600 if (skb->protocol != htons(ETH_P_IP) &&
7601 skb->protocol != htons(ETH_P_IPV6))
7604 switch (tstamp_type) {
7605 case BPF_SKB_TSTAMP_DELIVERY_MONO:
7608 skb->tstamp = tstamp;
7609 skb->mono_delivery_time = 1;
7611 case BPF_SKB_TSTAMP_UNSPEC:
7615 skb->mono_delivery_time = 0;
7624 static const struct bpf_func_proto bpf_skb_set_tstamp_proto = {
7625 .func = bpf_skb_set_tstamp,
7627 .ret_type = RET_INTEGER,
7628 .arg1_type = ARG_PTR_TO_CTX,
7629 .arg2_type = ARG_ANYTHING,
7630 .arg3_type = ARG_ANYTHING,
7633 #ifdef CONFIG_SYN_COOKIES
7634 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv4, struct iphdr *, iph,
7635 struct tcphdr *, th, u32, th_len)
7640 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7643 mss = tcp_parse_mss_option(th, 0) ?: TCP_MSS_DEFAULT;
7644 cookie = __cookie_v4_init_sequence(iph, th, &mss);
7646 return cookie | ((u64)mss << 32);
7649 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv4_proto = {
7650 .func = bpf_tcp_raw_gen_syncookie_ipv4,
7651 .gpl_only = true, /* __cookie_v4_init_sequence() is GPL */
7653 .ret_type = RET_INTEGER,
7654 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7655 .arg1_size = sizeof(struct iphdr),
7656 .arg2_type = ARG_PTR_TO_MEM,
7657 .arg3_type = ARG_CONST_SIZE,
7660 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv6, struct ipv6hdr *, iph,
7661 struct tcphdr *, th, u32, th_len)
7663 #if IS_BUILTIN(CONFIG_IPV6)
7664 const u16 mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
7665 sizeof(struct ipv6hdr);
7669 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7672 mss = tcp_parse_mss_option(th, 0) ?: mss_clamp;
7673 cookie = __cookie_v6_init_sequence(iph, th, &mss);
7675 return cookie | ((u64)mss << 32);
7677 return -EPROTONOSUPPORT;
7681 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv6_proto = {
7682 .func = bpf_tcp_raw_gen_syncookie_ipv6,
7683 .gpl_only = true, /* __cookie_v6_init_sequence() is GPL */
7685 .ret_type = RET_INTEGER,
7686 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7687 .arg1_size = sizeof(struct ipv6hdr),
7688 .arg2_type = ARG_PTR_TO_MEM,
7689 .arg3_type = ARG_CONST_SIZE,
7692 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv4, struct iphdr *, iph,
7693 struct tcphdr *, th)
7695 u32 cookie = ntohl(th->ack_seq) - 1;
7697 if (__cookie_v4_check(iph, th, cookie) > 0)
7703 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv4_proto = {
7704 .func = bpf_tcp_raw_check_syncookie_ipv4,
7705 .gpl_only = true, /* __cookie_v4_check is GPL */
7707 .ret_type = RET_INTEGER,
7708 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7709 .arg1_size = sizeof(struct iphdr),
7710 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7711 .arg2_size = sizeof(struct tcphdr),
7714 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv6, struct ipv6hdr *, iph,
7715 struct tcphdr *, th)
7717 #if IS_BUILTIN(CONFIG_IPV6)
7718 u32 cookie = ntohl(th->ack_seq) - 1;
7720 if (__cookie_v6_check(iph, th, cookie) > 0)
7725 return -EPROTONOSUPPORT;
7729 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv6_proto = {
7730 .func = bpf_tcp_raw_check_syncookie_ipv6,
7731 .gpl_only = true, /* __cookie_v6_check is GPL */
7733 .ret_type = RET_INTEGER,
7734 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7735 .arg1_size = sizeof(struct ipv6hdr),
7736 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7737 .arg2_size = sizeof(struct tcphdr),
7739 #endif /* CONFIG_SYN_COOKIES */
7741 #endif /* CONFIG_INET */
7743 bool bpf_helper_changes_pkt_data(void *func)
7745 if (func == bpf_skb_vlan_push ||
7746 func == bpf_skb_vlan_pop ||
7747 func == bpf_skb_store_bytes ||
7748 func == bpf_skb_change_proto ||
7749 func == bpf_skb_change_head ||
7750 func == sk_skb_change_head ||
7751 func == bpf_skb_change_tail ||
7752 func == sk_skb_change_tail ||
7753 func == bpf_skb_adjust_room ||
7754 func == sk_skb_adjust_room ||
7755 func == bpf_skb_pull_data ||
7756 func == sk_skb_pull_data ||
7757 func == bpf_clone_redirect ||
7758 func == bpf_l3_csum_replace ||
7759 func == bpf_l4_csum_replace ||
7760 func == bpf_xdp_adjust_head ||
7761 func == bpf_xdp_adjust_meta ||
7762 func == bpf_msg_pull_data ||
7763 func == bpf_msg_push_data ||
7764 func == bpf_msg_pop_data ||
7765 func == bpf_xdp_adjust_tail ||
7766 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7767 func == bpf_lwt_seg6_store_bytes ||
7768 func == bpf_lwt_seg6_adjust_srh ||
7769 func == bpf_lwt_seg6_action ||
7772 func == bpf_sock_ops_store_hdr_opt ||
7774 func == bpf_lwt_in_push_encap ||
7775 func == bpf_lwt_xmit_push_encap)
7781 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7782 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7784 static const struct bpf_func_proto *
7785 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7787 const struct bpf_func_proto *func_proto;
7789 func_proto = cgroup_common_func_proto(func_id, prog);
7793 func_proto = cgroup_current_func_proto(func_id, prog);
7798 case BPF_FUNC_get_socket_cookie:
7799 return &bpf_get_socket_cookie_sock_proto;
7800 case BPF_FUNC_get_netns_cookie:
7801 return &bpf_get_netns_cookie_sock_proto;
7802 case BPF_FUNC_perf_event_output:
7803 return &bpf_event_output_data_proto;
7804 case BPF_FUNC_sk_storage_get:
7805 return &bpf_sk_storage_get_cg_sock_proto;
7806 case BPF_FUNC_ktime_get_coarse_ns:
7807 return &bpf_ktime_get_coarse_ns_proto;
7809 return bpf_base_func_proto(func_id);
7813 static const struct bpf_func_proto *
7814 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7816 const struct bpf_func_proto *func_proto;
7818 func_proto = cgroup_common_func_proto(func_id, prog);
7822 func_proto = cgroup_current_func_proto(func_id, prog);
7828 switch (prog->expected_attach_type) {
7829 case BPF_CGROUP_INET4_CONNECT:
7830 case BPF_CGROUP_INET6_CONNECT:
7831 return &bpf_bind_proto;
7835 case BPF_FUNC_get_socket_cookie:
7836 return &bpf_get_socket_cookie_sock_addr_proto;
7837 case BPF_FUNC_get_netns_cookie:
7838 return &bpf_get_netns_cookie_sock_addr_proto;
7839 case BPF_FUNC_perf_event_output:
7840 return &bpf_event_output_data_proto;
7842 case BPF_FUNC_sk_lookup_tcp:
7843 return &bpf_sock_addr_sk_lookup_tcp_proto;
7844 case BPF_FUNC_sk_lookup_udp:
7845 return &bpf_sock_addr_sk_lookup_udp_proto;
7846 case BPF_FUNC_sk_release:
7847 return &bpf_sk_release_proto;
7848 case BPF_FUNC_skc_lookup_tcp:
7849 return &bpf_sock_addr_skc_lookup_tcp_proto;
7850 #endif /* CONFIG_INET */
7851 case BPF_FUNC_sk_storage_get:
7852 return &bpf_sk_storage_get_proto;
7853 case BPF_FUNC_sk_storage_delete:
7854 return &bpf_sk_storage_delete_proto;
7855 case BPF_FUNC_setsockopt:
7856 switch (prog->expected_attach_type) {
7857 case BPF_CGROUP_INET4_BIND:
7858 case BPF_CGROUP_INET6_BIND:
7859 case BPF_CGROUP_INET4_CONNECT:
7860 case BPF_CGROUP_INET6_CONNECT:
7861 case BPF_CGROUP_UDP4_RECVMSG:
7862 case BPF_CGROUP_UDP6_RECVMSG:
7863 case BPF_CGROUP_UDP4_SENDMSG:
7864 case BPF_CGROUP_UDP6_SENDMSG:
7865 case BPF_CGROUP_INET4_GETPEERNAME:
7866 case BPF_CGROUP_INET6_GETPEERNAME:
7867 case BPF_CGROUP_INET4_GETSOCKNAME:
7868 case BPF_CGROUP_INET6_GETSOCKNAME:
7869 return &bpf_sock_addr_setsockopt_proto;
7873 case BPF_FUNC_getsockopt:
7874 switch (prog->expected_attach_type) {
7875 case BPF_CGROUP_INET4_BIND:
7876 case BPF_CGROUP_INET6_BIND:
7877 case BPF_CGROUP_INET4_CONNECT:
7878 case BPF_CGROUP_INET6_CONNECT:
7879 case BPF_CGROUP_UDP4_RECVMSG:
7880 case BPF_CGROUP_UDP6_RECVMSG:
7881 case BPF_CGROUP_UDP4_SENDMSG:
7882 case BPF_CGROUP_UDP6_SENDMSG:
7883 case BPF_CGROUP_INET4_GETPEERNAME:
7884 case BPF_CGROUP_INET6_GETPEERNAME:
7885 case BPF_CGROUP_INET4_GETSOCKNAME:
7886 case BPF_CGROUP_INET6_GETSOCKNAME:
7887 return &bpf_sock_addr_getsockopt_proto;
7892 return bpf_sk_base_func_proto(func_id);
7896 static const struct bpf_func_proto *
7897 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7900 case BPF_FUNC_skb_load_bytes:
7901 return &bpf_skb_load_bytes_proto;
7902 case BPF_FUNC_skb_load_bytes_relative:
7903 return &bpf_skb_load_bytes_relative_proto;
7904 case BPF_FUNC_get_socket_cookie:
7905 return &bpf_get_socket_cookie_proto;
7906 case BPF_FUNC_get_socket_uid:
7907 return &bpf_get_socket_uid_proto;
7908 case BPF_FUNC_perf_event_output:
7909 return &bpf_skb_event_output_proto;
7911 return bpf_sk_base_func_proto(func_id);
7915 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7916 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7918 static const struct bpf_func_proto *
7919 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7921 const struct bpf_func_proto *func_proto;
7923 func_proto = cgroup_common_func_proto(func_id, prog);
7928 case BPF_FUNC_sk_fullsock:
7929 return &bpf_sk_fullsock_proto;
7930 case BPF_FUNC_sk_storage_get:
7931 return &bpf_sk_storage_get_proto;
7932 case BPF_FUNC_sk_storage_delete:
7933 return &bpf_sk_storage_delete_proto;
7934 case BPF_FUNC_perf_event_output:
7935 return &bpf_skb_event_output_proto;
7936 #ifdef CONFIG_SOCK_CGROUP_DATA
7937 case BPF_FUNC_skb_cgroup_id:
7938 return &bpf_skb_cgroup_id_proto;
7939 case BPF_FUNC_skb_ancestor_cgroup_id:
7940 return &bpf_skb_ancestor_cgroup_id_proto;
7941 case BPF_FUNC_sk_cgroup_id:
7942 return &bpf_sk_cgroup_id_proto;
7943 case BPF_FUNC_sk_ancestor_cgroup_id:
7944 return &bpf_sk_ancestor_cgroup_id_proto;
7947 case BPF_FUNC_sk_lookup_tcp:
7948 return &bpf_sk_lookup_tcp_proto;
7949 case BPF_FUNC_sk_lookup_udp:
7950 return &bpf_sk_lookup_udp_proto;
7951 case BPF_FUNC_sk_release:
7952 return &bpf_sk_release_proto;
7953 case BPF_FUNC_skc_lookup_tcp:
7954 return &bpf_skc_lookup_tcp_proto;
7955 case BPF_FUNC_tcp_sock:
7956 return &bpf_tcp_sock_proto;
7957 case BPF_FUNC_get_listener_sock:
7958 return &bpf_get_listener_sock_proto;
7959 case BPF_FUNC_skb_ecn_set_ce:
7960 return &bpf_skb_ecn_set_ce_proto;
7963 return sk_filter_func_proto(func_id, prog);
7967 static const struct bpf_func_proto *
7968 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7971 case BPF_FUNC_skb_store_bytes:
7972 return &bpf_skb_store_bytes_proto;
7973 case BPF_FUNC_skb_load_bytes:
7974 return &bpf_skb_load_bytes_proto;
7975 case BPF_FUNC_skb_load_bytes_relative:
7976 return &bpf_skb_load_bytes_relative_proto;
7977 case BPF_FUNC_skb_pull_data:
7978 return &bpf_skb_pull_data_proto;
7979 case BPF_FUNC_csum_diff:
7980 return &bpf_csum_diff_proto;
7981 case BPF_FUNC_csum_update:
7982 return &bpf_csum_update_proto;
7983 case BPF_FUNC_csum_level:
7984 return &bpf_csum_level_proto;
7985 case BPF_FUNC_l3_csum_replace:
7986 return &bpf_l3_csum_replace_proto;
7987 case BPF_FUNC_l4_csum_replace:
7988 return &bpf_l4_csum_replace_proto;
7989 case BPF_FUNC_clone_redirect:
7990 return &bpf_clone_redirect_proto;
7991 case BPF_FUNC_get_cgroup_classid:
7992 return &bpf_get_cgroup_classid_proto;
7993 case BPF_FUNC_skb_vlan_push:
7994 return &bpf_skb_vlan_push_proto;
7995 case BPF_FUNC_skb_vlan_pop:
7996 return &bpf_skb_vlan_pop_proto;
7997 case BPF_FUNC_skb_change_proto:
7998 return &bpf_skb_change_proto_proto;
7999 case BPF_FUNC_skb_change_type:
8000 return &bpf_skb_change_type_proto;
8001 case BPF_FUNC_skb_adjust_room:
8002 return &bpf_skb_adjust_room_proto;
8003 case BPF_FUNC_skb_change_tail:
8004 return &bpf_skb_change_tail_proto;
8005 case BPF_FUNC_skb_change_head:
8006 return &bpf_skb_change_head_proto;
8007 case BPF_FUNC_skb_get_tunnel_key:
8008 return &bpf_skb_get_tunnel_key_proto;
8009 case BPF_FUNC_skb_set_tunnel_key:
8010 return bpf_get_skb_set_tunnel_proto(func_id);
8011 case BPF_FUNC_skb_get_tunnel_opt:
8012 return &bpf_skb_get_tunnel_opt_proto;
8013 case BPF_FUNC_skb_set_tunnel_opt:
8014 return bpf_get_skb_set_tunnel_proto(func_id);
8015 case BPF_FUNC_redirect:
8016 return &bpf_redirect_proto;
8017 case BPF_FUNC_redirect_neigh:
8018 return &bpf_redirect_neigh_proto;
8019 case BPF_FUNC_redirect_peer:
8020 return &bpf_redirect_peer_proto;
8021 case BPF_FUNC_get_route_realm:
8022 return &bpf_get_route_realm_proto;
8023 case BPF_FUNC_get_hash_recalc:
8024 return &bpf_get_hash_recalc_proto;
8025 case BPF_FUNC_set_hash_invalid:
8026 return &bpf_set_hash_invalid_proto;
8027 case BPF_FUNC_set_hash:
8028 return &bpf_set_hash_proto;
8029 case BPF_FUNC_perf_event_output:
8030 return &bpf_skb_event_output_proto;
8031 case BPF_FUNC_get_smp_processor_id:
8032 return &bpf_get_smp_processor_id_proto;
8033 case BPF_FUNC_skb_under_cgroup:
8034 return &bpf_skb_under_cgroup_proto;
8035 case BPF_FUNC_get_socket_cookie:
8036 return &bpf_get_socket_cookie_proto;
8037 case BPF_FUNC_get_socket_uid:
8038 return &bpf_get_socket_uid_proto;
8039 case BPF_FUNC_fib_lookup:
8040 return &bpf_skb_fib_lookup_proto;
8041 case BPF_FUNC_check_mtu:
8042 return &bpf_skb_check_mtu_proto;
8043 case BPF_FUNC_sk_fullsock:
8044 return &bpf_sk_fullsock_proto;
8045 case BPF_FUNC_sk_storage_get:
8046 return &bpf_sk_storage_get_proto;
8047 case BPF_FUNC_sk_storage_delete:
8048 return &bpf_sk_storage_delete_proto;
8050 case BPF_FUNC_skb_get_xfrm_state:
8051 return &bpf_skb_get_xfrm_state_proto;
8053 #ifdef CONFIG_CGROUP_NET_CLASSID
8054 case BPF_FUNC_skb_cgroup_classid:
8055 return &bpf_skb_cgroup_classid_proto;
8057 #ifdef CONFIG_SOCK_CGROUP_DATA
8058 case BPF_FUNC_skb_cgroup_id:
8059 return &bpf_skb_cgroup_id_proto;
8060 case BPF_FUNC_skb_ancestor_cgroup_id:
8061 return &bpf_skb_ancestor_cgroup_id_proto;
8064 case BPF_FUNC_sk_lookup_tcp:
8065 return &bpf_tc_sk_lookup_tcp_proto;
8066 case BPF_FUNC_sk_lookup_udp:
8067 return &bpf_tc_sk_lookup_udp_proto;
8068 case BPF_FUNC_sk_release:
8069 return &bpf_sk_release_proto;
8070 case BPF_FUNC_tcp_sock:
8071 return &bpf_tcp_sock_proto;
8072 case BPF_FUNC_get_listener_sock:
8073 return &bpf_get_listener_sock_proto;
8074 case BPF_FUNC_skc_lookup_tcp:
8075 return &bpf_tc_skc_lookup_tcp_proto;
8076 case BPF_FUNC_tcp_check_syncookie:
8077 return &bpf_tcp_check_syncookie_proto;
8078 case BPF_FUNC_skb_ecn_set_ce:
8079 return &bpf_skb_ecn_set_ce_proto;
8080 case BPF_FUNC_tcp_gen_syncookie:
8081 return &bpf_tcp_gen_syncookie_proto;
8082 case BPF_FUNC_sk_assign:
8083 return &bpf_sk_assign_proto;
8084 case BPF_FUNC_skb_set_tstamp:
8085 return &bpf_skb_set_tstamp_proto;
8086 #ifdef CONFIG_SYN_COOKIES
8087 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8088 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8089 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8090 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8091 case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8092 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8093 case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8094 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8098 return bpf_sk_base_func_proto(func_id);
8102 static const struct bpf_func_proto *
8103 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8106 case BPF_FUNC_perf_event_output:
8107 return &bpf_xdp_event_output_proto;
8108 case BPF_FUNC_get_smp_processor_id:
8109 return &bpf_get_smp_processor_id_proto;
8110 case BPF_FUNC_csum_diff:
8111 return &bpf_csum_diff_proto;
8112 case BPF_FUNC_xdp_adjust_head:
8113 return &bpf_xdp_adjust_head_proto;
8114 case BPF_FUNC_xdp_adjust_meta:
8115 return &bpf_xdp_adjust_meta_proto;
8116 case BPF_FUNC_redirect:
8117 return &bpf_xdp_redirect_proto;
8118 case BPF_FUNC_redirect_map:
8119 return &bpf_xdp_redirect_map_proto;
8120 case BPF_FUNC_xdp_adjust_tail:
8121 return &bpf_xdp_adjust_tail_proto;
8122 case BPF_FUNC_xdp_get_buff_len:
8123 return &bpf_xdp_get_buff_len_proto;
8124 case BPF_FUNC_xdp_load_bytes:
8125 return &bpf_xdp_load_bytes_proto;
8126 case BPF_FUNC_xdp_store_bytes:
8127 return &bpf_xdp_store_bytes_proto;
8128 case BPF_FUNC_fib_lookup:
8129 return &bpf_xdp_fib_lookup_proto;
8130 case BPF_FUNC_check_mtu:
8131 return &bpf_xdp_check_mtu_proto;
8133 case BPF_FUNC_sk_lookup_udp:
8134 return &bpf_xdp_sk_lookup_udp_proto;
8135 case BPF_FUNC_sk_lookup_tcp:
8136 return &bpf_xdp_sk_lookup_tcp_proto;
8137 case BPF_FUNC_sk_release:
8138 return &bpf_sk_release_proto;
8139 case BPF_FUNC_skc_lookup_tcp:
8140 return &bpf_xdp_skc_lookup_tcp_proto;
8141 case BPF_FUNC_tcp_check_syncookie:
8142 return &bpf_tcp_check_syncookie_proto;
8143 case BPF_FUNC_tcp_gen_syncookie:
8144 return &bpf_tcp_gen_syncookie_proto;
8145 #ifdef CONFIG_SYN_COOKIES
8146 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8147 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8148 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8149 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8150 case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8151 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8152 case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8153 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8157 return bpf_sk_base_func_proto(func_id);
8160 #if IS_MODULE(CONFIG_NF_CONNTRACK) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES)
8161 /* The nf_conn___init type is used in the NF_CONNTRACK kfuncs. The
8162 * kfuncs are defined in two different modules, and we want to be able
8163 * to use them interchangably with the same BTF type ID. Because modules
8164 * can't de-duplicate BTF IDs between each other, we need the type to be
8165 * referenced in the vmlinux BTF or the verifier will get confused about
8166 * the different types. So we add this dummy type reference which will
8167 * be included in vmlinux BTF, allowing both modules to refer to the
8170 BTF_TYPE_EMIT(struct nf_conn___init);
8174 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
8175 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
8177 static const struct bpf_func_proto *
8178 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8180 const struct bpf_func_proto *func_proto;
8182 func_proto = cgroup_common_func_proto(func_id, prog);
8187 case BPF_FUNC_setsockopt:
8188 return &bpf_sock_ops_setsockopt_proto;
8189 case BPF_FUNC_getsockopt:
8190 return &bpf_sock_ops_getsockopt_proto;
8191 case BPF_FUNC_sock_ops_cb_flags_set:
8192 return &bpf_sock_ops_cb_flags_set_proto;
8193 case BPF_FUNC_sock_map_update:
8194 return &bpf_sock_map_update_proto;
8195 case BPF_FUNC_sock_hash_update:
8196 return &bpf_sock_hash_update_proto;
8197 case BPF_FUNC_get_socket_cookie:
8198 return &bpf_get_socket_cookie_sock_ops_proto;
8199 case BPF_FUNC_perf_event_output:
8200 return &bpf_event_output_data_proto;
8201 case BPF_FUNC_sk_storage_get:
8202 return &bpf_sk_storage_get_proto;
8203 case BPF_FUNC_sk_storage_delete:
8204 return &bpf_sk_storage_delete_proto;
8205 case BPF_FUNC_get_netns_cookie:
8206 return &bpf_get_netns_cookie_sock_ops_proto;
8208 case BPF_FUNC_load_hdr_opt:
8209 return &bpf_sock_ops_load_hdr_opt_proto;
8210 case BPF_FUNC_store_hdr_opt:
8211 return &bpf_sock_ops_store_hdr_opt_proto;
8212 case BPF_FUNC_reserve_hdr_opt:
8213 return &bpf_sock_ops_reserve_hdr_opt_proto;
8214 case BPF_FUNC_tcp_sock:
8215 return &bpf_tcp_sock_proto;
8216 #endif /* CONFIG_INET */
8218 return bpf_sk_base_func_proto(func_id);
8222 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
8223 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
8225 static const struct bpf_func_proto *
8226 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8229 case BPF_FUNC_msg_redirect_map:
8230 return &bpf_msg_redirect_map_proto;
8231 case BPF_FUNC_msg_redirect_hash:
8232 return &bpf_msg_redirect_hash_proto;
8233 case BPF_FUNC_msg_apply_bytes:
8234 return &bpf_msg_apply_bytes_proto;
8235 case BPF_FUNC_msg_cork_bytes:
8236 return &bpf_msg_cork_bytes_proto;
8237 case BPF_FUNC_msg_pull_data:
8238 return &bpf_msg_pull_data_proto;
8239 case BPF_FUNC_msg_push_data:
8240 return &bpf_msg_push_data_proto;
8241 case BPF_FUNC_msg_pop_data:
8242 return &bpf_msg_pop_data_proto;
8243 case BPF_FUNC_perf_event_output:
8244 return &bpf_event_output_data_proto;
8245 case BPF_FUNC_get_current_uid_gid:
8246 return &bpf_get_current_uid_gid_proto;
8247 case BPF_FUNC_get_current_pid_tgid:
8248 return &bpf_get_current_pid_tgid_proto;
8249 case BPF_FUNC_sk_storage_get:
8250 return &bpf_sk_storage_get_proto;
8251 case BPF_FUNC_sk_storage_delete:
8252 return &bpf_sk_storage_delete_proto;
8253 case BPF_FUNC_get_netns_cookie:
8254 return &bpf_get_netns_cookie_sk_msg_proto;
8255 #ifdef CONFIG_CGROUPS
8256 case BPF_FUNC_get_current_cgroup_id:
8257 return &bpf_get_current_cgroup_id_proto;
8258 case BPF_FUNC_get_current_ancestor_cgroup_id:
8259 return &bpf_get_current_ancestor_cgroup_id_proto;
8261 #ifdef CONFIG_CGROUP_NET_CLASSID
8262 case BPF_FUNC_get_cgroup_classid:
8263 return &bpf_get_cgroup_classid_curr_proto;
8266 return bpf_sk_base_func_proto(func_id);
8270 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
8271 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
8273 static const struct bpf_func_proto *
8274 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8277 case BPF_FUNC_skb_store_bytes:
8278 return &bpf_skb_store_bytes_proto;
8279 case BPF_FUNC_skb_load_bytes:
8280 return &bpf_skb_load_bytes_proto;
8281 case BPF_FUNC_skb_pull_data:
8282 return &sk_skb_pull_data_proto;
8283 case BPF_FUNC_skb_change_tail:
8284 return &sk_skb_change_tail_proto;
8285 case BPF_FUNC_skb_change_head:
8286 return &sk_skb_change_head_proto;
8287 case BPF_FUNC_skb_adjust_room:
8288 return &sk_skb_adjust_room_proto;
8289 case BPF_FUNC_get_socket_cookie:
8290 return &bpf_get_socket_cookie_proto;
8291 case BPF_FUNC_get_socket_uid:
8292 return &bpf_get_socket_uid_proto;
8293 case BPF_FUNC_sk_redirect_map:
8294 return &bpf_sk_redirect_map_proto;
8295 case BPF_FUNC_sk_redirect_hash:
8296 return &bpf_sk_redirect_hash_proto;
8297 case BPF_FUNC_perf_event_output:
8298 return &bpf_skb_event_output_proto;
8300 case BPF_FUNC_sk_lookup_tcp:
8301 return &bpf_sk_lookup_tcp_proto;
8302 case BPF_FUNC_sk_lookup_udp:
8303 return &bpf_sk_lookup_udp_proto;
8304 case BPF_FUNC_sk_release:
8305 return &bpf_sk_release_proto;
8306 case BPF_FUNC_skc_lookup_tcp:
8307 return &bpf_skc_lookup_tcp_proto;
8310 return bpf_sk_base_func_proto(func_id);
8314 static const struct bpf_func_proto *
8315 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8318 case BPF_FUNC_skb_load_bytes:
8319 return &bpf_flow_dissector_load_bytes_proto;
8321 return bpf_sk_base_func_proto(func_id);
8325 static const struct bpf_func_proto *
8326 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8329 case BPF_FUNC_skb_load_bytes:
8330 return &bpf_skb_load_bytes_proto;
8331 case BPF_FUNC_skb_pull_data:
8332 return &bpf_skb_pull_data_proto;
8333 case BPF_FUNC_csum_diff:
8334 return &bpf_csum_diff_proto;
8335 case BPF_FUNC_get_cgroup_classid:
8336 return &bpf_get_cgroup_classid_proto;
8337 case BPF_FUNC_get_route_realm:
8338 return &bpf_get_route_realm_proto;
8339 case BPF_FUNC_get_hash_recalc:
8340 return &bpf_get_hash_recalc_proto;
8341 case BPF_FUNC_perf_event_output:
8342 return &bpf_skb_event_output_proto;
8343 case BPF_FUNC_get_smp_processor_id:
8344 return &bpf_get_smp_processor_id_proto;
8345 case BPF_FUNC_skb_under_cgroup:
8346 return &bpf_skb_under_cgroup_proto;
8348 return bpf_sk_base_func_proto(func_id);
8352 static const struct bpf_func_proto *
8353 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8356 case BPF_FUNC_lwt_push_encap:
8357 return &bpf_lwt_in_push_encap_proto;
8359 return lwt_out_func_proto(func_id, prog);
8363 static const struct bpf_func_proto *
8364 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8367 case BPF_FUNC_skb_get_tunnel_key:
8368 return &bpf_skb_get_tunnel_key_proto;
8369 case BPF_FUNC_skb_set_tunnel_key:
8370 return bpf_get_skb_set_tunnel_proto(func_id);
8371 case BPF_FUNC_skb_get_tunnel_opt:
8372 return &bpf_skb_get_tunnel_opt_proto;
8373 case BPF_FUNC_skb_set_tunnel_opt:
8374 return bpf_get_skb_set_tunnel_proto(func_id);
8375 case BPF_FUNC_redirect:
8376 return &bpf_redirect_proto;
8377 case BPF_FUNC_clone_redirect:
8378 return &bpf_clone_redirect_proto;
8379 case BPF_FUNC_skb_change_tail:
8380 return &bpf_skb_change_tail_proto;
8381 case BPF_FUNC_skb_change_head:
8382 return &bpf_skb_change_head_proto;
8383 case BPF_FUNC_skb_store_bytes:
8384 return &bpf_skb_store_bytes_proto;
8385 case BPF_FUNC_csum_update:
8386 return &bpf_csum_update_proto;
8387 case BPF_FUNC_csum_level:
8388 return &bpf_csum_level_proto;
8389 case BPF_FUNC_l3_csum_replace:
8390 return &bpf_l3_csum_replace_proto;
8391 case BPF_FUNC_l4_csum_replace:
8392 return &bpf_l4_csum_replace_proto;
8393 case BPF_FUNC_set_hash_invalid:
8394 return &bpf_set_hash_invalid_proto;
8395 case BPF_FUNC_lwt_push_encap:
8396 return &bpf_lwt_xmit_push_encap_proto;
8398 return lwt_out_func_proto(func_id, prog);
8402 static const struct bpf_func_proto *
8403 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8406 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
8407 case BPF_FUNC_lwt_seg6_store_bytes:
8408 return &bpf_lwt_seg6_store_bytes_proto;
8409 case BPF_FUNC_lwt_seg6_action:
8410 return &bpf_lwt_seg6_action_proto;
8411 case BPF_FUNC_lwt_seg6_adjust_srh:
8412 return &bpf_lwt_seg6_adjust_srh_proto;
8415 return lwt_out_func_proto(func_id, prog);
8419 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
8420 const struct bpf_prog *prog,
8421 struct bpf_insn_access_aux *info)
8423 const int size_default = sizeof(__u32);
8425 if (off < 0 || off >= sizeof(struct __sk_buff))
8428 /* The verifier guarantees that size > 0. */
8429 if (off % size != 0)
8433 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8434 if (off + size > offsetofend(struct __sk_buff, cb[4]))
8437 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
8438 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
8439 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
8440 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
8441 case bpf_ctx_range(struct __sk_buff, data):
8442 case bpf_ctx_range(struct __sk_buff, data_meta):
8443 case bpf_ctx_range(struct __sk_buff, data_end):
8444 if (size != size_default)
8447 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8449 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8450 if (type == BPF_WRITE || size != sizeof(__u64))
8453 case bpf_ctx_range(struct __sk_buff, tstamp):
8454 if (size != sizeof(__u64))
8457 case offsetof(struct __sk_buff, sk):
8458 if (type == BPF_WRITE || size != sizeof(__u64))
8460 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
8462 case offsetof(struct __sk_buff, tstamp_type):
8464 case offsetofend(struct __sk_buff, tstamp_type) ... offsetof(struct __sk_buff, hwtstamp) - 1:
8465 /* Explicitly prohibit access to padding in __sk_buff. */
8468 /* Only narrow read access allowed for now. */
8469 if (type == BPF_WRITE) {
8470 if (size != size_default)
8473 bpf_ctx_record_field_size(info, size_default);
8474 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8482 static bool sk_filter_is_valid_access(int off, int size,
8483 enum bpf_access_type type,
8484 const struct bpf_prog *prog,
8485 struct bpf_insn_access_aux *info)
8488 case bpf_ctx_range(struct __sk_buff, tc_classid):
8489 case bpf_ctx_range(struct __sk_buff, data):
8490 case bpf_ctx_range(struct __sk_buff, data_meta):
8491 case bpf_ctx_range(struct __sk_buff, data_end):
8492 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8493 case bpf_ctx_range(struct __sk_buff, tstamp):
8494 case bpf_ctx_range(struct __sk_buff, wire_len):
8495 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8499 if (type == BPF_WRITE) {
8501 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8508 return bpf_skb_is_valid_access(off, size, type, prog, info);
8511 static bool cg_skb_is_valid_access(int off, int size,
8512 enum bpf_access_type type,
8513 const struct bpf_prog *prog,
8514 struct bpf_insn_access_aux *info)
8517 case bpf_ctx_range(struct __sk_buff, tc_classid):
8518 case bpf_ctx_range(struct __sk_buff, data_meta):
8519 case bpf_ctx_range(struct __sk_buff, wire_len):
8521 case bpf_ctx_range(struct __sk_buff, data):
8522 case bpf_ctx_range(struct __sk_buff, data_end):
8528 if (type == BPF_WRITE) {
8530 case bpf_ctx_range(struct __sk_buff, mark):
8531 case bpf_ctx_range(struct __sk_buff, priority):
8532 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8534 case bpf_ctx_range(struct __sk_buff, tstamp):
8544 case bpf_ctx_range(struct __sk_buff, data):
8545 info->reg_type = PTR_TO_PACKET;
8547 case bpf_ctx_range(struct __sk_buff, data_end):
8548 info->reg_type = PTR_TO_PACKET_END;
8552 return bpf_skb_is_valid_access(off, size, type, prog, info);
8555 static bool lwt_is_valid_access(int off, int size,
8556 enum bpf_access_type type,
8557 const struct bpf_prog *prog,
8558 struct bpf_insn_access_aux *info)
8561 case bpf_ctx_range(struct __sk_buff, tc_classid):
8562 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8563 case bpf_ctx_range(struct __sk_buff, data_meta):
8564 case bpf_ctx_range(struct __sk_buff, tstamp):
8565 case bpf_ctx_range(struct __sk_buff, wire_len):
8566 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8570 if (type == BPF_WRITE) {
8572 case bpf_ctx_range(struct __sk_buff, mark):
8573 case bpf_ctx_range(struct __sk_buff, priority):
8574 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8582 case bpf_ctx_range(struct __sk_buff, data):
8583 info->reg_type = PTR_TO_PACKET;
8585 case bpf_ctx_range(struct __sk_buff, data_end):
8586 info->reg_type = PTR_TO_PACKET_END;
8590 return bpf_skb_is_valid_access(off, size, type, prog, info);
8593 /* Attach type specific accesses */
8594 static bool __sock_filter_check_attach_type(int off,
8595 enum bpf_access_type access_type,
8596 enum bpf_attach_type attach_type)
8599 case offsetof(struct bpf_sock, bound_dev_if):
8600 case offsetof(struct bpf_sock, mark):
8601 case offsetof(struct bpf_sock, priority):
8602 switch (attach_type) {
8603 case BPF_CGROUP_INET_SOCK_CREATE:
8604 case BPF_CGROUP_INET_SOCK_RELEASE:
8609 case bpf_ctx_range(struct bpf_sock, src_ip4):
8610 switch (attach_type) {
8611 case BPF_CGROUP_INET4_POST_BIND:
8616 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8617 switch (attach_type) {
8618 case BPF_CGROUP_INET6_POST_BIND:
8623 case bpf_ctx_range(struct bpf_sock, src_port):
8624 switch (attach_type) {
8625 case BPF_CGROUP_INET4_POST_BIND:
8626 case BPF_CGROUP_INET6_POST_BIND:
8633 return access_type == BPF_READ;
8638 bool bpf_sock_common_is_valid_access(int off, int size,
8639 enum bpf_access_type type,
8640 struct bpf_insn_access_aux *info)
8643 case bpf_ctx_range_till(struct bpf_sock, type, priority):
8646 return bpf_sock_is_valid_access(off, size, type, info);
8650 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
8651 struct bpf_insn_access_aux *info)
8653 const int size_default = sizeof(__u32);
8656 if (off < 0 || off >= sizeof(struct bpf_sock))
8658 if (off % size != 0)
8662 case offsetof(struct bpf_sock, state):
8663 case offsetof(struct bpf_sock, family):
8664 case offsetof(struct bpf_sock, type):
8665 case offsetof(struct bpf_sock, protocol):
8666 case offsetof(struct bpf_sock, src_port):
8667 case offsetof(struct bpf_sock, rx_queue_mapping):
8668 case bpf_ctx_range(struct bpf_sock, src_ip4):
8669 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8670 case bpf_ctx_range(struct bpf_sock, dst_ip4):
8671 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8672 bpf_ctx_record_field_size(info, size_default);
8673 return bpf_ctx_narrow_access_ok(off, size, size_default);
8674 case bpf_ctx_range(struct bpf_sock, dst_port):
8675 field_size = size == size_default ?
8676 size_default : sizeof_field(struct bpf_sock, dst_port);
8677 bpf_ctx_record_field_size(info, field_size);
8678 return bpf_ctx_narrow_access_ok(off, size, field_size);
8679 case offsetofend(struct bpf_sock, dst_port) ...
8680 offsetof(struct bpf_sock, dst_ip4) - 1:
8684 return size == size_default;
8687 static bool sock_filter_is_valid_access(int off, int size,
8688 enum bpf_access_type type,
8689 const struct bpf_prog *prog,
8690 struct bpf_insn_access_aux *info)
8692 if (!bpf_sock_is_valid_access(off, size, type, info))
8694 return __sock_filter_check_attach_type(off, type,
8695 prog->expected_attach_type);
8698 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
8699 const struct bpf_prog *prog)
8701 /* Neither direct read nor direct write requires any preliminary
8707 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
8708 const struct bpf_prog *prog, int drop_verdict)
8710 struct bpf_insn *insn = insn_buf;
8715 /* if (!skb->cloned)
8718 * (Fast-path, otherwise approximation that we might be
8719 * a clone, do the rest in helper.)
8721 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET);
8722 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
8723 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
8725 /* ret = bpf_skb_pull_data(skb, 0); */
8726 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
8727 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
8728 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
8729 BPF_FUNC_skb_pull_data);
8732 * return TC_ACT_SHOT;
8734 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
8735 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
8736 *insn++ = BPF_EXIT_INSN();
8739 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
8741 *insn++ = prog->insnsi[0];
8743 return insn - insn_buf;
8746 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
8747 struct bpf_insn *insn_buf)
8749 bool indirect = BPF_MODE(orig->code) == BPF_IND;
8750 struct bpf_insn *insn = insn_buf;
8753 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
8755 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
8757 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
8759 /* We're guaranteed here that CTX is in R6. */
8760 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
8762 switch (BPF_SIZE(orig->code)) {
8764 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
8767 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
8770 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
8774 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
8775 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
8776 *insn++ = BPF_EXIT_INSN();
8778 return insn - insn_buf;
8781 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8782 const struct bpf_prog *prog)
8784 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8787 static bool tc_cls_act_is_valid_access(int off, int size,
8788 enum bpf_access_type type,
8789 const struct bpf_prog *prog,
8790 struct bpf_insn_access_aux *info)
8792 if (type == BPF_WRITE) {
8794 case bpf_ctx_range(struct __sk_buff, mark):
8795 case bpf_ctx_range(struct __sk_buff, tc_index):
8796 case bpf_ctx_range(struct __sk_buff, priority):
8797 case bpf_ctx_range(struct __sk_buff, tc_classid):
8798 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8799 case bpf_ctx_range(struct __sk_buff, tstamp):
8800 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8808 case bpf_ctx_range(struct __sk_buff, data):
8809 info->reg_type = PTR_TO_PACKET;
8811 case bpf_ctx_range(struct __sk_buff, data_meta):
8812 info->reg_type = PTR_TO_PACKET_META;
8814 case bpf_ctx_range(struct __sk_buff, data_end):
8815 info->reg_type = PTR_TO_PACKET_END;
8817 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8819 case offsetof(struct __sk_buff, tstamp_type):
8820 /* The convert_ctx_access() on reading and writing
8821 * __sk_buff->tstamp depends on whether the bpf prog
8822 * has used __sk_buff->tstamp_type or not.
8823 * Thus, we need to set prog->tstamp_type_access
8824 * earlier during is_valid_access() here.
8826 ((struct bpf_prog *)prog)->tstamp_type_access = 1;
8827 return size == sizeof(__u8);
8830 return bpf_skb_is_valid_access(off, size, type, prog, info);
8833 DEFINE_MUTEX(nf_conn_btf_access_lock);
8834 EXPORT_SYMBOL_GPL(nf_conn_btf_access_lock);
8836 int (*nfct_btf_struct_access)(struct bpf_verifier_log *log, const struct btf *btf,
8837 const struct btf_type *t, int off, int size,
8838 enum bpf_access_type atype, u32 *next_btf_id,
8839 enum bpf_type_flag *flag);
8840 EXPORT_SYMBOL_GPL(nfct_btf_struct_access);
8842 static int tc_cls_act_btf_struct_access(struct bpf_verifier_log *log,
8843 const struct btf *btf,
8844 const struct btf_type *t, int off,
8845 int size, enum bpf_access_type atype,
8847 enum bpf_type_flag *flag)
8851 if (atype == BPF_READ)
8852 return btf_struct_access(log, btf, t, off, size, atype, next_btf_id,
8855 mutex_lock(&nf_conn_btf_access_lock);
8856 if (nfct_btf_struct_access)
8857 ret = nfct_btf_struct_access(log, btf, t, off, size, atype, next_btf_id, flag);
8858 mutex_unlock(&nf_conn_btf_access_lock);
8863 static bool __is_valid_xdp_access(int off, int size)
8865 if (off < 0 || off >= sizeof(struct xdp_md))
8867 if (off % size != 0)
8869 if (size != sizeof(__u32))
8875 static bool xdp_is_valid_access(int off, int size,
8876 enum bpf_access_type type,
8877 const struct bpf_prog *prog,
8878 struct bpf_insn_access_aux *info)
8880 if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8882 case offsetof(struct xdp_md, egress_ifindex):
8887 if (type == BPF_WRITE) {
8888 if (bpf_prog_is_dev_bound(prog->aux)) {
8890 case offsetof(struct xdp_md, rx_queue_index):
8891 return __is_valid_xdp_access(off, size);
8898 case offsetof(struct xdp_md, data):
8899 info->reg_type = PTR_TO_PACKET;
8901 case offsetof(struct xdp_md, data_meta):
8902 info->reg_type = PTR_TO_PACKET_META;
8904 case offsetof(struct xdp_md, data_end):
8905 info->reg_type = PTR_TO_PACKET_END;
8909 return __is_valid_xdp_access(off, size);
8912 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act)
8914 const u32 act_max = XDP_REDIRECT;
8916 pr_warn_once("%s XDP return value %u on prog %s (id %d) dev %s, expect packet loss!\n",
8917 act > act_max ? "Illegal" : "Driver unsupported",
8918 act, prog->aux->name, prog->aux->id, dev ? dev->name : "N/A");
8920 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
8922 static int xdp_btf_struct_access(struct bpf_verifier_log *log,
8923 const struct btf *btf,
8924 const struct btf_type *t, int off,
8925 int size, enum bpf_access_type atype,
8927 enum bpf_type_flag *flag)
8931 if (atype == BPF_READ)
8932 return btf_struct_access(log, btf, t, off, size, atype, next_btf_id,
8935 mutex_lock(&nf_conn_btf_access_lock);
8936 if (nfct_btf_struct_access)
8937 ret = nfct_btf_struct_access(log, btf, t, off, size, atype, next_btf_id, flag);
8938 mutex_unlock(&nf_conn_btf_access_lock);
8943 static bool sock_addr_is_valid_access(int off, int size,
8944 enum bpf_access_type type,
8945 const struct bpf_prog *prog,
8946 struct bpf_insn_access_aux *info)
8948 const int size_default = sizeof(__u32);
8950 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
8952 if (off % size != 0)
8955 /* Disallow access to IPv6 fields from IPv4 contex and vise
8959 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8960 switch (prog->expected_attach_type) {
8961 case BPF_CGROUP_INET4_BIND:
8962 case BPF_CGROUP_INET4_CONNECT:
8963 case BPF_CGROUP_INET4_GETPEERNAME:
8964 case BPF_CGROUP_INET4_GETSOCKNAME:
8965 case BPF_CGROUP_UDP4_SENDMSG:
8966 case BPF_CGROUP_UDP4_RECVMSG:
8972 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8973 switch (prog->expected_attach_type) {
8974 case BPF_CGROUP_INET6_BIND:
8975 case BPF_CGROUP_INET6_CONNECT:
8976 case BPF_CGROUP_INET6_GETPEERNAME:
8977 case BPF_CGROUP_INET6_GETSOCKNAME:
8978 case BPF_CGROUP_UDP6_SENDMSG:
8979 case BPF_CGROUP_UDP6_RECVMSG:
8985 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8986 switch (prog->expected_attach_type) {
8987 case BPF_CGROUP_UDP4_SENDMSG:
8993 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8995 switch (prog->expected_attach_type) {
8996 case BPF_CGROUP_UDP6_SENDMSG:
9005 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
9006 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9007 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
9008 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9010 case bpf_ctx_range(struct bpf_sock_addr, user_port):
9011 if (type == BPF_READ) {
9012 bpf_ctx_record_field_size(info, size_default);
9014 if (bpf_ctx_wide_access_ok(off, size,
9015 struct bpf_sock_addr,
9019 if (bpf_ctx_wide_access_ok(off, size,
9020 struct bpf_sock_addr,
9024 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
9027 if (bpf_ctx_wide_access_ok(off, size,
9028 struct bpf_sock_addr,
9032 if (bpf_ctx_wide_access_ok(off, size,
9033 struct bpf_sock_addr,
9037 if (size != size_default)
9041 case offsetof(struct bpf_sock_addr, sk):
9042 if (type != BPF_READ)
9044 if (size != sizeof(__u64))
9046 info->reg_type = PTR_TO_SOCKET;
9049 if (type == BPF_READ) {
9050 if (size != size_default)
9060 static bool sock_ops_is_valid_access(int off, int size,
9061 enum bpf_access_type type,
9062 const struct bpf_prog *prog,
9063 struct bpf_insn_access_aux *info)
9065 const int size_default = sizeof(__u32);
9067 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
9070 /* The verifier guarantees that size > 0. */
9071 if (off % size != 0)
9074 if (type == BPF_WRITE) {
9076 case offsetof(struct bpf_sock_ops, reply):
9077 case offsetof(struct bpf_sock_ops, sk_txhash):
9078 if (size != size_default)
9086 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
9088 if (size != sizeof(__u64))
9091 case offsetof(struct bpf_sock_ops, sk):
9092 if (size != sizeof(__u64))
9094 info->reg_type = PTR_TO_SOCKET_OR_NULL;
9096 case offsetof(struct bpf_sock_ops, skb_data):
9097 if (size != sizeof(__u64))
9099 info->reg_type = PTR_TO_PACKET;
9101 case offsetof(struct bpf_sock_ops, skb_data_end):
9102 if (size != sizeof(__u64))
9104 info->reg_type = PTR_TO_PACKET_END;
9106 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9107 bpf_ctx_record_field_size(info, size_default);
9108 return bpf_ctx_narrow_access_ok(off, size,
9111 if (size != size_default)
9120 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
9121 const struct bpf_prog *prog)
9123 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
9126 static bool sk_skb_is_valid_access(int off, int size,
9127 enum bpf_access_type type,
9128 const struct bpf_prog *prog,
9129 struct bpf_insn_access_aux *info)
9132 case bpf_ctx_range(struct __sk_buff, tc_classid):
9133 case bpf_ctx_range(struct __sk_buff, data_meta):
9134 case bpf_ctx_range(struct __sk_buff, tstamp):
9135 case bpf_ctx_range(struct __sk_buff, wire_len):
9136 case bpf_ctx_range(struct __sk_buff, hwtstamp):
9140 if (type == BPF_WRITE) {
9142 case bpf_ctx_range(struct __sk_buff, tc_index):
9143 case bpf_ctx_range(struct __sk_buff, priority):
9151 case bpf_ctx_range(struct __sk_buff, mark):
9153 case bpf_ctx_range(struct __sk_buff, data):
9154 info->reg_type = PTR_TO_PACKET;
9156 case bpf_ctx_range(struct __sk_buff, data_end):
9157 info->reg_type = PTR_TO_PACKET_END;
9161 return bpf_skb_is_valid_access(off, size, type, prog, info);
9164 static bool sk_msg_is_valid_access(int off, int size,
9165 enum bpf_access_type type,
9166 const struct bpf_prog *prog,
9167 struct bpf_insn_access_aux *info)
9169 if (type == BPF_WRITE)
9172 if (off % size != 0)
9176 case offsetof(struct sk_msg_md, data):
9177 info->reg_type = PTR_TO_PACKET;
9178 if (size != sizeof(__u64))
9181 case offsetof(struct sk_msg_md, data_end):
9182 info->reg_type = PTR_TO_PACKET_END;
9183 if (size != sizeof(__u64))
9186 case offsetof(struct sk_msg_md, sk):
9187 if (size != sizeof(__u64))
9189 info->reg_type = PTR_TO_SOCKET;
9191 case bpf_ctx_range(struct sk_msg_md, family):
9192 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
9193 case bpf_ctx_range(struct sk_msg_md, local_ip4):
9194 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
9195 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
9196 case bpf_ctx_range(struct sk_msg_md, remote_port):
9197 case bpf_ctx_range(struct sk_msg_md, local_port):
9198 case bpf_ctx_range(struct sk_msg_md, size):
9199 if (size != sizeof(__u32))
9208 static bool flow_dissector_is_valid_access(int off, int size,
9209 enum bpf_access_type type,
9210 const struct bpf_prog *prog,
9211 struct bpf_insn_access_aux *info)
9213 const int size_default = sizeof(__u32);
9215 if (off < 0 || off >= sizeof(struct __sk_buff))
9218 if (type == BPF_WRITE)
9222 case bpf_ctx_range(struct __sk_buff, data):
9223 if (size != size_default)
9225 info->reg_type = PTR_TO_PACKET;
9227 case bpf_ctx_range(struct __sk_buff, data_end):
9228 if (size != size_default)
9230 info->reg_type = PTR_TO_PACKET_END;
9232 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
9233 if (size != sizeof(__u64))
9235 info->reg_type = PTR_TO_FLOW_KEYS;
9242 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
9243 const struct bpf_insn *si,
9244 struct bpf_insn *insn_buf,
9245 struct bpf_prog *prog,
9249 struct bpf_insn *insn = insn_buf;
9252 case offsetof(struct __sk_buff, data):
9253 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
9254 si->dst_reg, si->src_reg,
9255 offsetof(struct bpf_flow_dissector, data));
9258 case offsetof(struct __sk_buff, data_end):
9259 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
9260 si->dst_reg, si->src_reg,
9261 offsetof(struct bpf_flow_dissector, data_end));
9264 case offsetof(struct __sk_buff, flow_keys):
9265 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
9266 si->dst_reg, si->src_reg,
9267 offsetof(struct bpf_flow_dissector, flow_keys));
9271 return insn - insn_buf;
9274 static struct bpf_insn *bpf_convert_tstamp_type_read(const struct bpf_insn *si,
9275 struct bpf_insn *insn)
9277 __u8 value_reg = si->dst_reg;
9278 __u8 skb_reg = si->src_reg;
9279 /* AX is needed because src_reg and dst_reg could be the same */
9280 __u8 tmp_reg = BPF_REG_AX;
9282 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg,
9283 PKT_VLAN_PRESENT_OFFSET);
9284 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg,
9285 SKB_MONO_DELIVERY_TIME_MASK, 2);
9286 *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_UNSPEC);
9287 *insn++ = BPF_JMP_A(1);
9288 *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_DELIVERY_MONO);
9293 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si,
9294 struct bpf_insn *insn)
9296 /* si->dst_reg = skb_shinfo(SKB); */
9297 #ifdef NET_SKBUFF_DATA_USES_OFFSET
9298 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9299 BPF_REG_AX, si->src_reg,
9300 offsetof(struct sk_buff, end));
9301 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
9302 si->dst_reg, si->src_reg,
9303 offsetof(struct sk_buff, head));
9304 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
9306 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9307 si->dst_reg, si->src_reg,
9308 offsetof(struct sk_buff, end));
9314 static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog,
9315 const struct bpf_insn *si,
9316 struct bpf_insn *insn)
9318 __u8 value_reg = si->dst_reg;
9319 __u8 skb_reg = si->src_reg;
9321 #ifdef CONFIG_NET_CLS_ACT
9322 /* If the tstamp_type is read,
9323 * the bpf prog is aware the tstamp could have delivery time.
9324 * Thus, read skb->tstamp as is if tstamp_type_access is true.
9326 if (!prog->tstamp_type_access) {
9327 /* AX is needed because src_reg and dst_reg could be the same */
9328 __u8 tmp_reg = BPF_REG_AX;
9330 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, PKT_VLAN_PRESENT_OFFSET);
9331 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg,
9332 TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK);
9333 *insn++ = BPF_JMP32_IMM(BPF_JNE, tmp_reg,
9334 TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK, 2);
9335 /* skb->tc_at_ingress && skb->mono_delivery_time,
9336 * read 0 as the (rcv) timestamp.
9338 *insn++ = BPF_MOV64_IMM(value_reg, 0);
9339 *insn++ = BPF_JMP_A(1);
9343 *insn++ = BPF_LDX_MEM(BPF_DW, value_reg, skb_reg,
9344 offsetof(struct sk_buff, tstamp));
9348 static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog,
9349 const struct bpf_insn *si,
9350 struct bpf_insn *insn)
9352 __u8 value_reg = si->src_reg;
9353 __u8 skb_reg = si->dst_reg;
9355 #ifdef CONFIG_NET_CLS_ACT
9356 /* If the tstamp_type is read,
9357 * the bpf prog is aware the tstamp could have delivery time.
9358 * Thus, write skb->tstamp as is if tstamp_type_access is true.
9359 * Otherwise, writing at ingress will have to clear the
9360 * mono_delivery_time bit also.
9362 if (!prog->tstamp_type_access) {
9363 __u8 tmp_reg = BPF_REG_AX;
9365 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, PKT_VLAN_PRESENT_OFFSET);
9366 /* Writing __sk_buff->tstamp as ingress, goto <clear> */
9367 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1);
9369 *insn++ = BPF_JMP_A(2);
9370 /* <clear>: mono_delivery_time */
9371 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_MONO_DELIVERY_TIME_MASK);
9372 *insn++ = BPF_STX_MEM(BPF_B, skb_reg, tmp_reg, PKT_VLAN_PRESENT_OFFSET);
9376 /* <store>: skb->tstamp = tstamp */
9377 *insn++ = BPF_STX_MEM(BPF_DW, skb_reg, value_reg,
9378 offsetof(struct sk_buff, tstamp));
9382 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
9383 const struct bpf_insn *si,
9384 struct bpf_insn *insn_buf,
9385 struct bpf_prog *prog, u32 *target_size)
9387 struct bpf_insn *insn = insn_buf;
9391 case offsetof(struct __sk_buff, len):
9392 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9393 bpf_target_off(struct sk_buff, len, 4,
9397 case offsetof(struct __sk_buff, protocol):
9398 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9399 bpf_target_off(struct sk_buff, protocol, 2,
9403 case offsetof(struct __sk_buff, vlan_proto):
9404 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9405 bpf_target_off(struct sk_buff, vlan_proto, 2,
9409 case offsetof(struct __sk_buff, priority):
9410 if (type == BPF_WRITE)
9411 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9412 bpf_target_off(struct sk_buff, priority, 4,
9415 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9416 bpf_target_off(struct sk_buff, priority, 4,
9420 case offsetof(struct __sk_buff, ingress_ifindex):
9421 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9422 bpf_target_off(struct sk_buff, skb_iif, 4,
9426 case offsetof(struct __sk_buff, ifindex):
9427 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9428 si->dst_reg, si->src_reg,
9429 offsetof(struct sk_buff, dev));
9430 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9431 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9432 bpf_target_off(struct net_device, ifindex, 4,
9436 case offsetof(struct __sk_buff, hash):
9437 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9438 bpf_target_off(struct sk_buff, hash, 4,
9442 case offsetof(struct __sk_buff, mark):
9443 if (type == BPF_WRITE)
9444 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9445 bpf_target_off(struct sk_buff, mark, 4,
9448 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9449 bpf_target_off(struct sk_buff, mark, 4,
9453 case offsetof(struct __sk_buff, pkt_type):
9455 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
9457 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
9458 #ifdef __BIG_ENDIAN_BITFIELD
9459 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
9463 case offsetof(struct __sk_buff, queue_mapping):
9464 if (type == BPF_WRITE) {
9465 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
9466 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
9467 bpf_target_off(struct sk_buff,
9471 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9472 bpf_target_off(struct sk_buff,
9478 case offsetof(struct __sk_buff, vlan_present):
9480 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
9481 PKT_VLAN_PRESENT_OFFSET);
9482 if (PKT_VLAN_PRESENT_BIT)
9483 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
9484 if (PKT_VLAN_PRESENT_BIT < 7)
9485 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
9488 case offsetof(struct __sk_buff, vlan_tci):
9489 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9490 bpf_target_off(struct sk_buff, vlan_tci, 2,
9494 case offsetof(struct __sk_buff, cb[0]) ...
9495 offsetofend(struct __sk_buff, cb[4]) - 1:
9496 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
9497 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
9498 offsetof(struct qdisc_skb_cb, data)) %
9501 prog->cb_access = 1;
9503 off -= offsetof(struct __sk_buff, cb[0]);
9504 off += offsetof(struct sk_buff, cb);
9505 off += offsetof(struct qdisc_skb_cb, data);
9506 if (type == BPF_WRITE)
9507 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
9510 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
9514 case offsetof(struct __sk_buff, tc_classid):
9515 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
9518 off -= offsetof(struct __sk_buff, tc_classid);
9519 off += offsetof(struct sk_buff, cb);
9520 off += offsetof(struct qdisc_skb_cb, tc_classid);
9522 if (type == BPF_WRITE)
9523 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
9526 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
9530 case offsetof(struct __sk_buff, data):
9531 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9532 si->dst_reg, si->src_reg,
9533 offsetof(struct sk_buff, data));
9536 case offsetof(struct __sk_buff, data_meta):
9538 off -= offsetof(struct __sk_buff, data_meta);
9539 off += offsetof(struct sk_buff, cb);
9540 off += offsetof(struct bpf_skb_data_end, data_meta);
9541 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9545 case offsetof(struct __sk_buff, data_end):
9547 off -= offsetof(struct __sk_buff, data_end);
9548 off += offsetof(struct sk_buff, cb);
9549 off += offsetof(struct bpf_skb_data_end, data_end);
9550 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9554 case offsetof(struct __sk_buff, tc_index):
9555 #ifdef CONFIG_NET_SCHED
9556 if (type == BPF_WRITE)
9557 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
9558 bpf_target_off(struct sk_buff, tc_index, 2,
9561 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9562 bpf_target_off(struct sk_buff, tc_index, 2,
9566 if (type == BPF_WRITE)
9567 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
9569 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9573 case offsetof(struct __sk_buff, napi_id):
9574 #if defined(CONFIG_NET_RX_BUSY_POLL)
9575 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9576 bpf_target_off(struct sk_buff, napi_id, 4,
9578 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
9579 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9582 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9585 case offsetof(struct __sk_buff, family):
9586 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9588 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9589 si->dst_reg, si->src_reg,
9590 offsetof(struct sk_buff, sk));
9591 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9592 bpf_target_off(struct sock_common,
9596 case offsetof(struct __sk_buff, remote_ip4):
9597 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9599 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9600 si->dst_reg, si->src_reg,
9601 offsetof(struct sk_buff, sk));
9602 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9603 bpf_target_off(struct sock_common,
9607 case offsetof(struct __sk_buff, local_ip4):
9608 BUILD_BUG_ON(sizeof_field(struct sock_common,
9609 skc_rcv_saddr) != 4);
9611 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9612 si->dst_reg, si->src_reg,
9613 offsetof(struct sk_buff, sk));
9614 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9615 bpf_target_off(struct sock_common,
9619 case offsetof(struct __sk_buff, remote_ip6[0]) ...
9620 offsetof(struct __sk_buff, remote_ip6[3]):
9621 #if IS_ENABLED(CONFIG_IPV6)
9622 BUILD_BUG_ON(sizeof_field(struct sock_common,
9623 skc_v6_daddr.s6_addr32[0]) != 4);
9626 off -= offsetof(struct __sk_buff, remote_ip6[0]);
9628 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9629 si->dst_reg, si->src_reg,
9630 offsetof(struct sk_buff, sk));
9631 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9632 offsetof(struct sock_common,
9633 skc_v6_daddr.s6_addr32[0]) +
9636 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9639 case offsetof(struct __sk_buff, local_ip6[0]) ...
9640 offsetof(struct __sk_buff, local_ip6[3]):
9641 #if IS_ENABLED(CONFIG_IPV6)
9642 BUILD_BUG_ON(sizeof_field(struct sock_common,
9643 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9646 off -= offsetof(struct __sk_buff, local_ip6[0]);
9648 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9649 si->dst_reg, si->src_reg,
9650 offsetof(struct sk_buff, sk));
9651 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9652 offsetof(struct sock_common,
9653 skc_v6_rcv_saddr.s6_addr32[0]) +
9656 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9660 case offsetof(struct __sk_buff, remote_port):
9661 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9663 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9664 si->dst_reg, si->src_reg,
9665 offsetof(struct sk_buff, sk));
9666 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9667 bpf_target_off(struct sock_common,
9670 #ifndef __BIG_ENDIAN_BITFIELD
9671 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9675 case offsetof(struct __sk_buff, local_port):
9676 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9678 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9679 si->dst_reg, si->src_reg,
9680 offsetof(struct sk_buff, sk));
9681 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9682 bpf_target_off(struct sock_common,
9683 skc_num, 2, target_size));
9686 case offsetof(struct __sk_buff, tstamp):
9687 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
9689 if (type == BPF_WRITE)
9690 insn = bpf_convert_tstamp_write(prog, si, insn);
9692 insn = bpf_convert_tstamp_read(prog, si, insn);
9695 case offsetof(struct __sk_buff, tstamp_type):
9696 insn = bpf_convert_tstamp_type_read(si, insn);
9699 case offsetof(struct __sk_buff, gso_segs):
9700 insn = bpf_convert_shinfo_access(si, insn);
9701 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
9702 si->dst_reg, si->dst_reg,
9703 bpf_target_off(struct skb_shared_info,
9707 case offsetof(struct __sk_buff, gso_size):
9708 insn = bpf_convert_shinfo_access(si, insn);
9709 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
9710 si->dst_reg, si->dst_reg,
9711 bpf_target_off(struct skb_shared_info,
9715 case offsetof(struct __sk_buff, wire_len):
9716 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
9719 off -= offsetof(struct __sk_buff, wire_len);
9720 off += offsetof(struct sk_buff, cb);
9721 off += offsetof(struct qdisc_skb_cb, pkt_len);
9723 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
9726 case offsetof(struct __sk_buff, sk):
9727 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9728 si->dst_reg, si->src_reg,
9729 offsetof(struct sk_buff, sk));
9731 case offsetof(struct __sk_buff, hwtstamp):
9732 BUILD_BUG_ON(sizeof_field(struct skb_shared_hwtstamps, hwtstamp) != 8);
9733 BUILD_BUG_ON(offsetof(struct skb_shared_hwtstamps, hwtstamp) != 0);
9735 insn = bpf_convert_shinfo_access(si, insn);
9736 *insn++ = BPF_LDX_MEM(BPF_DW,
9737 si->dst_reg, si->dst_reg,
9738 bpf_target_off(struct skb_shared_info,
9744 return insn - insn_buf;
9747 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
9748 const struct bpf_insn *si,
9749 struct bpf_insn *insn_buf,
9750 struct bpf_prog *prog, u32 *target_size)
9752 struct bpf_insn *insn = insn_buf;
9756 case offsetof(struct bpf_sock, bound_dev_if):
9757 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
9759 if (type == BPF_WRITE)
9760 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9761 offsetof(struct sock, sk_bound_dev_if));
9763 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9764 offsetof(struct sock, sk_bound_dev_if));
9767 case offsetof(struct bpf_sock, mark):
9768 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
9770 if (type == BPF_WRITE)
9771 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9772 offsetof(struct sock, sk_mark));
9774 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9775 offsetof(struct sock, sk_mark));
9778 case offsetof(struct bpf_sock, priority):
9779 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
9781 if (type == BPF_WRITE)
9782 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9783 offsetof(struct sock, sk_priority));
9785 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9786 offsetof(struct sock, sk_priority));
9789 case offsetof(struct bpf_sock, family):
9790 *insn++ = BPF_LDX_MEM(
9791 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
9792 si->dst_reg, si->src_reg,
9793 bpf_target_off(struct sock_common,
9795 sizeof_field(struct sock_common,
9800 case offsetof(struct bpf_sock, type):
9801 *insn++ = BPF_LDX_MEM(
9802 BPF_FIELD_SIZEOF(struct sock, sk_type),
9803 si->dst_reg, si->src_reg,
9804 bpf_target_off(struct sock, sk_type,
9805 sizeof_field(struct sock, sk_type),
9809 case offsetof(struct bpf_sock, protocol):
9810 *insn++ = BPF_LDX_MEM(
9811 BPF_FIELD_SIZEOF(struct sock, sk_protocol),
9812 si->dst_reg, si->src_reg,
9813 bpf_target_off(struct sock, sk_protocol,
9814 sizeof_field(struct sock, sk_protocol),
9818 case offsetof(struct bpf_sock, src_ip4):
9819 *insn++ = BPF_LDX_MEM(
9820 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9821 bpf_target_off(struct sock_common, skc_rcv_saddr,
9822 sizeof_field(struct sock_common,
9827 case offsetof(struct bpf_sock, dst_ip4):
9828 *insn++ = BPF_LDX_MEM(
9829 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9830 bpf_target_off(struct sock_common, skc_daddr,
9831 sizeof_field(struct sock_common,
9836 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
9837 #if IS_ENABLED(CONFIG_IPV6)
9839 off -= offsetof(struct bpf_sock, src_ip6[0]);
9840 *insn++ = BPF_LDX_MEM(
9841 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9844 skc_v6_rcv_saddr.s6_addr32[0],
9845 sizeof_field(struct sock_common,
9846 skc_v6_rcv_saddr.s6_addr32[0]),
9847 target_size) + off);
9850 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9854 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
9855 #if IS_ENABLED(CONFIG_IPV6)
9857 off -= offsetof(struct bpf_sock, dst_ip6[0]);
9858 *insn++ = BPF_LDX_MEM(
9859 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9860 bpf_target_off(struct sock_common,
9861 skc_v6_daddr.s6_addr32[0],
9862 sizeof_field(struct sock_common,
9863 skc_v6_daddr.s6_addr32[0]),
9864 target_size) + off);
9866 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9871 case offsetof(struct bpf_sock, src_port):
9872 *insn++ = BPF_LDX_MEM(
9873 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
9874 si->dst_reg, si->src_reg,
9875 bpf_target_off(struct sock_common, skc_num,
9876 sizeof_field(struct sock_common,
9881 case offsetof(struct bpf_sock, dst_port):
9882 *insn++ = BPF_LDX_MEM(
9883 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
9884 si->dst_reg, si->src_reg,
9885 bpf_target_off(struct sock_common, skc_dport,
9886 sizeof_field(struct sock_common,
9891 case offsetof(struct bpf_sock, state):
9892 *insn++ = BPF_LDX_MEM(
9893 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
9894 si->dst_reg, si->src_reg,
9895 bpf_target_off(struct sock_common, skc_state,
9896 sizeof_field(struct sock_common,
9900 case offsetof(struct bpf_sock, rx_queue_mapping):
9901 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
9902 *insn++ = BPF_LDX_MEM(
9903 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
9904 si->dst_reg, si->src_reg,
9905 bpf_target_off(struct sock, sk_rx_queue_mapping,
9906 sizeof_field(struct sock,
9907 sk_rx_queue_mapping),
9909 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
9911 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9913 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9919 return insn - insn_buf;
9922 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
9923 const struct bpf_insn *si,
9924 struct bpf_insn *insn_buf,
9925 struct bpf_prog *prog, u32 *target_size)
9927 struct bpf_insn *insn = insn_buf;
9930 case offsetof(struct __sk_buff, ifindex):
9931 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9932 si->dst_reg, si->src_reg,
9933 offsetof(struct sk_buff, dev));
9934 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9935 bpf_target_off(struct net_device, ifindex, 4,
9939 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9943 return insn - insn_buf;
9946 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
9947 const struct bpf_insn *si,
9948 struct bpf_insn *insn_buf,
9949 struct bpf_prog *prog, u32 *target_size)
9951 struct bpf_insn *insn = insn_buf;
9954 case offsetof(struct xdp_md, data):
9955 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
9956 si->dst_reg, si->src_reg,
9957 offsetof(struct xdp_buff, data));
9959 case offsetof(struct xdp_md, data_meta):
9960 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
9961 si->dst_reg, si->src_reg,
9962 offsetof(struct xdp_buff, data_meta));
9964 case offsetof(struct xdp_md, data_end):
9965 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
9966 si->dst_reg, si->src_reg,
9967 offsetof(struct xdp_buff, data_end));
9969 case offsetof(struct xdp_md, ingress_ifindex):
9970 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9971 si->dst_reg, si->src_reg,
9972 offsetof(struct xdp_buff, rxq));
9973 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
9974 si->dst_reg, si->dst_reg,
9975 offsetof(struct xdp_rxq_info, dev));
9976 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9977 offsetof(struct net_device, ifindex));
9979 case offsetof(struct xdp_md, rx_queue_index):
9980 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9981 si->dst_reg, si->src_reg,
9982 offsetof(struct xdp_buff, rxq));
9983 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9984 offsetof(struct xdp_rxq_info,
9987 case offsetof(struct xdp_md, egress_ifindex):
9988 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
9989 si->dst_reg, si->src_reg,
9990 offsetof(struct xdp_buff, txq));
9991 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
9992 si->dst_reg, si->dst_reg,
9993 offsetof(struct xdp_txq_info, dev));
9994 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9995 offsetof(struct net_device, ifindex));
9999 return insn - insn_buf;
10002 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
10003 * context Structure, F is Field in context structure that contains a pointer
10004 * to Nested Structure of type NS that has the field NF.
10006 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
10007 * sure that SIZE is not greater than actual size of S.F.NF.
10009 * If offset OFF is provided, the load happens from that offset relative to
10012 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
10014 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
10015 si->src_reg, offsetof(S, F)); \
10016 *insn++ = BPF_LDX_MEM( \
10017 SIZE, si->dst_reg, si->dst_reg, \
10018 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
10023 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
10024 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
10025 BPF_FIELD_SIZEOF(NS, NF), 0)
10027 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
10028 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
10030 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
10031 * "register" since two registers available in convert_ctx_access are not
10032 * enough: we can't override neither SRC, since it contains value to store, nor
10033 * DST since it contains pointer to context that may be used by later
10034 * instructions. But we need a temporary place to save pointer to nested
10035 * structure whose field we want to store to.
10037 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
10039 int tmp_reg = BPF_REG_9; \
10040 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
10042 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
10044 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
10045 offsetof(S, TF)); \
10046 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
10047 si->dst_reg, offsetof(S, F)); \
10048 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \
10049 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
10052 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
10053 offsetof(S, TF)); \
10056 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
10059 if (type == BPF_WRITE) { \
10060 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
10063 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
10064 S, NS, F, NF, SIZE, OFF); \
10068 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
10069 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
10070 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
10072 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
10073 const struct bpf_insn *si,
10074 struct bpf_insn *insn_buf,
10075 struct bpf_prog *prog, u32 *target_size)
10077 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
10078 struct bpf_insn *insn = insn_buf;
10081 case offsetof(struct bpf_sock_addr, user_family):
10082 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10083 struct sockaddr, uaddr, sa_family);
10086 case offsetof(struct bpf_sock_addr, user_ip4):
10087 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10088 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
10089 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
10092 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
10094 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
10095 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10096 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
10097 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
10101 case offsetof(struct bpf_sock_addr, user_port):
10102 /* To get port we need to know sa_family first and then treat
10103 * sockaddr as either sockaddr_in or sockaddr_in6.
10104 * Though we can simplify since port field has same offset and
10105 * size in both structures.
10106 * Here we check this invariant and use just one of the
10107 * structures if it's true.
10109 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
10110 offsetof(struct sockaddr_in6, sin6_port));
10111 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
10112 sizeof_field(struct sockaddr_in6, sin6_port));
10113 /* Account for sin6_port being smaller than user_port. */
10114 port_size = min(port_size, BPF_LDST_BYTES(si));
10115 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10116 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
10117 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
10120 case offsetof(struct bpf_sock_addr, family):
10121 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10122 struct sock, sk, sk_family);
10125 case offsetof(struct bpf_sock_addr, type):
10126 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10127 struct sock, sk, sk_type);
10130 case offsetof(struct bpf_sock_addr, protocol):
10131 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10132 struct sock, sk, sk_protocol);
10135 case offsetof(struct bpf_sock_addr, msg_src_ip4):
10136 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
10137 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10138 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
10139 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
10142 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
10145 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
10146 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
10147 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10148 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
10149 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
10151 case offsetof(struct bpf_sock_addr, sk):
10152 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
10153 si->dst_reg, si->src_reg,
10154 offsetof(struct bpf_sock_addr_kern, sk));
10158 return insn - insn_buf;
10161 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
10162 const struct bpf_insn *si,
10163 struct bpf_insn *insn_buf,
10164 struct bpf_prog *prog,
10167 struct bpf_insn *insn = insn_buf;
10170 /* Helper macro for adding read access to tcp_sock or sock fields. */
10171 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
10173 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \
10174 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
10175 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
10176 if (si->dst_reg == reg || si->src_reg == reg) \
10178 if (si->dst_reg == reg || si->src_reg == reg) \
10180 if (si->dst_reg == si->src_reg) { \
10181 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
10182 offsetof(struct bpf_sock_ops_kern, \
10184 fullsock_reg = reg; \
10187 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10188 struct bpf_sock_ops_kern, \
10190 fullsock_reg, si->src_reg, \
10191 offsetof(struct bpf_sock_ops_kern, \
10193 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
10194 if (si->dst_reg == si->src_reg) \
10195 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10196 offsetof(struct bpf_sock_ops_kern, \
10198 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10199 struct bpf_sock_ops_kern, sk),\
10200 si->dst_reg, si->src_reg, \
10201 offsetof(struct bpf_sock_ops_kern, sk));\
10202 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
10204 si->dst_reg, si->dst_reg, \
10205 offsetof(OBJ, OBJ_FIELD)); \
10206 if (si->dst_reg == si->src_reg) { \
10207 *insn++ = BPF_JMP_A(1); \
10208 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10209 offsetof(struct bpf_sock_ops_kern, \
10214 #define SOCK_OPS_GET_SK() \
10216 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \
10217 if (si->dst_reg == reg || si->src_reg == reg) \
10219 if (si->dst_reg == reg || si->src_reg == reg) \
10221 if (si->dst_reg == si->src_reg) { \
10222 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
10223 offsetof(struct bpf_sock_ops_kern, \
10225 fullsock_reg = reg; \
10228 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10229 struct bpf_sock_ops_kern, \
10231 fullsock_reg, si->src_reg, \
10232 offsetof(struct bpf_sock_ops_kern, \
10234 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
10235 if (si->dst_reg == si->src_reg) \
10236 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10237 offsetof(struct bpf_sock_ops_kern, \
10239 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10240 struct bpf_sock_ops_kern, sk),\
10241 si->dst_reg, si->src_reg, \
10242 offsetof(struct bpf_sock_ops_kern, sk));\
10243 if (si->dst_reg == si->src_reg) { \
10244 *insn++ = BPF_JMP_A(1); \
10245 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10246 offsetof(struct bpf_sock_ops_kern, \
10251 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
10252 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
10254 /* Helper macro for adding write access to tcp_sock or sock fields.
10255 * The macro is called with two registers, dst_reg which contains a pointer
10256 * to ctx (context) and src_reg which contains the value that should be
10257 * stored. However, we need an additional register since we cannot overwrite
10258 * dst_reg because it may be used later in the program.
10259 * Instead we "borrow" one of the other register. We first save its value
10260 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
10261 * it at the end of the macro.
10263 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
10265 int reg = BPF_REG_9; \
10266 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
10267 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
10268 if (si->dst_reg == reg || si->src_reg == reg) \
10270 if (si->dst_reg == reg || si->src_reg == reg) \
10272 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
10273 offsetof(struct bpf_sock_ops_kern, \
10275 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10276 struct bpf_sock_ops_kern, \
10278 reg, si->dst_reg, \
10279 offsetof(struct bpf_sock_ops_kern, \
10281 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
10282 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10283 struct bpf_sock_ops_kern, sk),\
10284 reg, si->dst_reg, \
10285 offsetof(struct bpf_sock_ops_kern, sk));\
10286 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \
10287 reg, si->src_reg, \
10288 offsetof(OBJ, OBJ_FIELD)); \
10289 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
10290 offsetof(struct bpf_sock_ops_kern, \
10294 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
10296 if (TYPE == BPF_WRITE) \
10297 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
10299 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
10302 if (insn > insn_buf)
10303 return insn - insn_buf;
10306 case offsetof(struct bpf_sock_ops, op):
10307 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10309 si->dst_reg, si->src_reg,
10310 offsetof(struct bpf_sock_ops_kern, op));
10313 case offsetof(struct bpf_sock_ops, replylong[0]) ...
10314 offsetof(struct bpf_sock_ops, replylong[3]):
10315 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
10316 sizeof_field(struct bpf_sock_ops_kern, reply));
10317 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
10318 sizeof_field(struct bpf_sock_ops_kern, replylong));
10320 off -= offsetof(struct bpf_sock_ops, replylong[0]);
10321 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
10322 if (type == BPF_WRITE)
10323 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
10326 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10330 case offsetof(struct bpf_sock_ops, family):
10331 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10333 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10334 struct bpf_sock_ops_kern, sk),
10335 si->dst_reg, si->src_reg,
10336 offsetof(struct bpf_sock_ops_kern, sk));
10337 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10338 offsetof(struct sock_common, skc_family));
10341 case offsetof(struct bpf_sock_ops, remote_ip4):
10342 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10344 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10345 struct bpf_sock_ops_kern, sk),
10346 si->dst_reg, si->src_reg,
10347 offsetof(struct bpf_sock_ops_kern, sk));
10348 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10349 offsetof(struct sock_common, skc_daddr));
10352 case offsetof(struct bpf_sock_ops, local_ip4):
10353 BUILD_BUG_ON(sizeof_field(struct sock_common,
10354 skc_rcv_saddr) != 4);
10356 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10357 struct bpf_sock_ops_kern, sk),
10358 si->dst_reg, si->src_reg,
10359 offsetof(struct bpf_sock_ops_kern, sk));
10360 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10361 offsetof(struct sock_common,
10365 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
10366 offsetof(struct bpf_sock_ops, remote_ip6[3]):
10367 #if IS_ENABLED(CONFIG_IPV6)
10368 BUILD_BUG_ON(sizeof_field(struct sock_common,
10369 skc_v6_daddr.s6_addr32[0]) != 4);
10372 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
10373 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10374 struct bpf_sock_ops_kern, sk),
10375 si->dst_reg, si->src_reg,
10376 offsetof(struct bpf_sock_ops_kern, sk));
10377 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10378 offsetof(struct sock_common,
10379 skc_v6_daddr.s6_addr32[0]) +
10382 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10386 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
10387 offsetof(struct bpf_sock_ops, local_ip6[3]):
10388 #if IS_ENABLED(CONFIG_IPV6)
10389 BUILD_BUG_ON(sizeof_field(struct sock_common,
10390 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10393 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
10394 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10395 struct bpf_sock_ops_kern, sk),
10396 si->dst_reg, si->src_reg,
10397 offsetof(struct bpf_sock_ops_kern, sk));
10398 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10399 offsetof(struct sock_common,
10400 skc_v6_rcv_saddr.s6_addr32[0]) +
10403 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10407 case offsetof(struct bpf_sock_ops, remote_port):
10408 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10410 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10411 struct bpf_sock_ops_kern, sk),
10412 si->dst_reg, si->src_reg,
10413 offsetof(struct bpf_sock_ops_kern, sk));
10414 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10415 offsetof(struct sock_common, skc_dport));
10416 #ifndef __BIG_ENDIAN_BITFIELD
10417 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10421 case offsetof(struct bpf_sock_ops, local_port):
10422 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10424 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10425 struct bpf_sock_ops_kern, sk),
10426 si->dst_reg, si->src_reg,
10427 offsetof(struct bpf_sock_ops_kern, sk));
10428 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10429 offsetof(struct sock_common, skc_num));
10432 case offsetof(struct bpf_sock_ops, is_fullsock):
10433 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10434 struct bpf_sock_ops_kern,
10436 si->dst_reg, si->src_reg,
10437 offsetof(struct bpf_sock_ops_kern,
10441 case offsetof(struct bpf_sock_ops, state):
10442 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
10444 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10445 struct bpf_sock_ops_kern, sk),
10446 si->dst_reg, si->src_reg,
10447 offsetof(struct bpf_sock_ops_kern, sk));
10448 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
10449 offsetof(struct sock_common, skc_state));
10452 case offsetof(struct bpf_sock_ops, rtt_min):
10453 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
10454 sizeof(struct minmax));
10455 BUILD_BUG_ON(sizeof(struct minmax) <
10456 sizeof(struct minmax_sample));
10458 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10459 struct bpf_sock_ops_kern, sk),
10460 si->dst_reg, si->src_reg,
10461 offsetof(struct bpf_sock_ops_kern, sk));
10462 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10463 offsetof(struct tcp_sock, rtt_min) +
10464 sizeof_field(struct minmax_sample, t));
10467 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
10468 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
10472 case offsetof(struct bpf_sock_ops, sk_txhash):
10473 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
10474 struct sock, type);
10476 case offsetof(struct bpf_sock_ops, snd_cwnd):
10477 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
10479 case offsetof(struct bpf_sock_ops, srtt_us):
10480 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
10482 case offsetof(struct bpf_sock_ops, snd_ssthresh):
10483 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
10485 case offsetof(struct bpf_sock_ops, rcv_nxt):
10486 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
10488 case offsetof(struct bpf_sock_ops, snd_nxt):
10489 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
10491 case offsetof(struct bpf_sock_ops, snd_una):
10492 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
10494 case offsetof(struct bpf_sock_ops, mss_cache):
10495 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
10497 case offsetof(struct bpf_sock_ops, ecn_flags):
10498 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
10500 case offsetof(struct bpf_sock_ops, rate_delivered):
10501 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
10503 case offsetof(struct bpf_sock_ops, rate_interval_us):
10504 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
10506 case offsetof(struct bpf_sock_ops, packets_out):
10507 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
10509 case offsetof(struct bpf_sock_ops, retrans_out):
10510 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
10512 case offsetof(struct bpf_sock_ops, total_retrans):
10513 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
10515 case offsetof(struct bpf_sock_ops, segs_in):
10516 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
10518 case offsetof(struct bpf_sock_ops, data_segs_in):
10519 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
10521 case offsetof(struct bpf_sock_ops, segs_out):
10522 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
10524 case offsetof(struct bpf_sock_ops, data_segs_out):
10525 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
10527 case offsetof(struct bpf_sock_ops, lost_out):
10528 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
10530 case offsetof(struct bpf_sock_ops, sacked_out):
10531 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
10533 case offsetof(struct bpf_sock_ops, bytes_received):
10534 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
10536 case offsetof(struct bpf_sock_ops, bytes_acked):
10537 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
10539 case offsetof(struct bpf_sock_ops, sk):
10542 case offsetof(struct bpf_sock_ops, skb_data_end):
10543 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10545 si->dst_reg, si->src_reg,
10546 offsetof(struct bpf_sock_ops_kern,
10549 case offsetof(struct bpf_sock_ops, skb_data):
10550 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10552 si->dst_reg, si->src_reg,
10553 offsetof(struct bpf_sock_ops_kern,
10555 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10556 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10557 si->dst_reg, si->dst_reg,
10558 offsetof(struct sk_buff, data));
10560 case offsetof(struct bpf_sock_ops, skb_len):
10561 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10563 si->dst_reg, si->src_reg,
10564 offsetof(struct bpf_sock_ops_kern,
10566 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10567 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10568 si->dst_reg, si->dst_reg,
10569 offsetof(struct sk_buff, len));
10571 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
10572 off = offsetof(struct sk_buff, cb);
10573 off += offsetof(struct tcp_skb_cb, tcp_flags);
10574 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
10575 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10577 si->dst_reg, si->src_reg,
10578 offsetof(struct bpf_sock_ops_kern,
10580 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10581 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
10583 si->dst_reg, si->dst_reg, off);
10586 return insn - insn_buf;
10589 /* data_end = skb->data + skb_headlen() */
10590 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
10591 struct bpf_insn *insn)
10594 int temp_reg_off = offsetof(struct sk_buff, cb) +
10595 offsetof(struct sk_skb_cb, temp_reg);
10597 if (si->src_reg == si->dst_reg) {
10598 /* We need an extra register, choose and save a register. */
10600 if (si->src_reg == reg || si->dst_reg == reg)
10602 if (si->src_reg == reg || si->dst_reg == reg)
10604 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off);
10609 /* reg = skb->data */
10610 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10612 offsetof(struct sk_buff, data));
10613 /* AX = skb->len */
10614 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10615 BPF_REG_AX, si->src_reg,
10616 offsetof(struct sk_buff, len));
10617 /* reg = skb->data + skb->len */
10618 *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX);
10619 /* AX = skb->data_len */
10620 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
10621 BPF_REG_AX, si->src_reg,
10622 offsetof(struct sk_buff, data_len));
10624 /* reg = skb->data + skb->len - skb->data_len */
10625 *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX);
10627 if (si->src_reg == si->dst_reg) {
10628 /* Restore the saved register */
10629 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg);
10630 *insn++ = BPF_MOV64_REG(si->dst_reg, reg);
10631 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off);
10637 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
10638 const struct bpf_insn *si,
10639 struct bpf_insn *insn_buf,
10640 struct bpf_prog *prog, u32 *target_size)
10642 struct bpf_insn *insn = insn_buf;
10646 case offsetof(struct __sk_buff, data_end):
10647 insn = bpf_convert_data_end_access(si, insn);
10649 case offsetof(struct __sk_buff, cb[0]) ...
10650 offsetofend(struct __sk_buff, cb[4]) - 1:
10651 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20);
10652 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
10653 offsetof(struct sk_skb_cb, data)) %
10656 prog->cb_access = 1;
10658 off -= offsetof(struct __sk_buff, cb[0]);
10659 off += offsetof(struct sk_buff, cb);
10660 off += offsetof(struct sk_skb_cb, data);
10661 if (type == BPF_WRITE)
10662 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
10665 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
10671 return bpf_convert_ctx_access(type, si, insn_buf, prog,
10675 return insn - insn_buf;
10678 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
10679 const struct bpf_insn *si,
10680 struct bpf_insn *insn_buf,
10681 struct bpf_prog *prog, u32 *target_size)
10683 struct bpf_insn *insn = insn_buf;
10684 #if IS_ENABLED(CONFIG_IPV6)
10688 /* convert ctx uses the fact sg element is first in struct */
10689 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
10692 case offsetof(struct sk_msg_md, data):
10693 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
10694 si->dst_reg, si->src_reg,
10695 offsetof(struct sk_msg, data));
10697 case offsetof(struct sk_msg_md, data_end):
10698 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
10699 si->dst_reg, si->src_reg,
10700 offsetof(struct sk_msg, data_end));
10702 case offsetof(struct sk_msg_md, family):
10703 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10705 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10706 struct sk_msg, sk),
10707 si->dst_reg, si->src_reg,
10708 offsetof(struct sk_msg, sk));
10709 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10710 offsetof(struct sock_common, skc_family));
10713 case offsetof(struct sk_msg_md, remote_ip4):
10714 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10716 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10717 struct sk_msg, sk),
10718 si->dst_reg, si->src_reg,
10719 offsetof(struct sk_msg, sk));
10720 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10721 offsetof(struct sock_common, skc_daddr));
10724 case offsetof(struct sk_msg_md, local_ip4):
10725 BUILD_BUG_ON(sizeof_field(struct sock_common,
10726 skc_rcv_saddr) != 4);
10728 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10729 struct sk_msg, sk),
10730 si->dst_reg, si->src_reg,
10731 offsetof(struct sk_msg, sk));
10732 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10733 offsetof(struct sock_common,
10737 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
10738 offsetof(struct sk_msg_md, remote_ip6[3]):
10739 #if IS_ENABLED(CONFIG_IPV6)
10740 BUILD_BUG_ON(sizeof_field(struct sock_common,
10741 skc_v6_daddr.s6_addr32[0]) != 4);
10744 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
10745 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10746 struct sk_msg, sk),
10747 si->dst_reg, si->src_reg,
10748 offsetof(struct sk_msg, sk));
10749 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10750 offsetof(struct sock_common,
10751 skc_v6_daddr.s6_addr32[0]) +
10754 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10758 case offsetof(struct sk_msg_md, local_ip6[0]) ...
10759 offsetof(struct sk_msg_md, local_ip6[3]):
10760 #if IS_ENABLED(CONFIG_IPV6)
10761 BUILD_BUG_ON(sizeof_field(struct sock_common,
10762 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10765 off -= offsetof(struct sk_msg_md, local_ip6[0]);
10766 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10767 struct sk_msg, sk),
10768 si->dst_reg, si->src_reg,
10769 offsetof(struct sk_msg, sk));
10770 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10771 offsetof(struct sock_common,
10772 skc_v6_rcv_saddr.s6_addr32[0]) +
10775 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10779 case offsetof(struct sk_msg_md, remote_port):
10780 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10782 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10783 struct sk_msg, sk),
10784 si->dst_reg, si->src_reg,
10785 offsetof(struct sk_msg, sk));
10786 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10787 offsetof(struct sock_common, skc_dport));
10788 #ifndef __BIG_ENDIAN_BITFIELD
10789 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10793 case offsetof(struct sk_msg_md, local_port):
10794 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10796 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10797 struct sk_msg, sk),
10798 si->dst_reg, si->src_reg,
10799 offsetof(struct sk_msg, sk));
10800 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10801 offsetof(struct sock_common, skc_num));
10804 case offsetof(struct sk_msg_md, size):
10805 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
10806 si->dst_reg, si->src_reg,
10807 offsetof(struct sk_msg_sg, size));
10810 case offsetof(struct sk_msg_md, sk):
10811 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
10812 si->dst_reg, si->src_reg,
10813 offsetof(struct sk_msg, sk));
10817 return insn - insn_buf;
10820 const struct bpf_verifier_ops sk_filter_verifier_ops = {
10821 .get_func_proto = sk_filter_func_proto,
10822 .is_valid_access = sk_filter_is_valid_access,
10823 .convert_ctx_access = bpf_convert_ctx_access,
10824 .gen_ld_abs = bpf_gen_ld_abs,
10827 const struct bpf_prog_ops sk_filter_prog_ops = {
10828 .test_run = bpf_prog_test_run_skb,
10831 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
10832 .get_func_proto = tc_cls_act_func_proto,
10833 .is_valid_access = tc_cls_act_is_valid_access,
10834 .convert_ctx_access = tc_cls_act_convert_ctx_access,
10835 .gen_prologue = tc_cls_act_prologue,
10836 .gen_ld_abs = bpf_gen_ld_abs,
10837 .btf_struct_access = tc_cls_act_btf_struct_access,
10840 const struct bpf_prog_ops tc_cls_act_prog_ops = {
10841 .test_run = bpf_prog_test_run_skb,
10844 const struct bpf_verifier_ops xdp_verifier_ops = {
10845 .get_func_proto = xdp_func_proto,
10846 .is_valid_access = xdp_is_valid_access,
10847 .convert_ctx_access = xdp_convert_ctx_access,
10848 .gen_prologue = bpf_noop_prologue,
10849 .btf_struct_access = xdp_btf_struct_access,
10852 const struct bpf_prog_ops xdp_prog_ops = {
10853 .test_run = bpf_prog_test_run_xdp,
10856 const struct bpf_verifier_ops cg_skb_verifier_ops = {
10857 .get_func_proto = cg_skb_func_proto,
10858 .is_valid_access = cg_skb_is_valid_access,
10859 .convert_ctx_access = bpf_convert_ctx_access,
10862 const struct bpf_prog_ops cg_skb_prog_ops = {
10863 .test_run = bpf_prog_test_run_skb,
10866 const struct bpf_verifier_ops lwt_in_verifier_ops = {
10867 .get_func_proto = lwt_in_func_proto,
10868 .is_valid_access = lwt_is_valid_access,
10869 .convert_ctx_access = bpf_convert_ctx_access,
10872 const struct bpf_prog_ops lwt_in_prog_ops = {
10873 .test_run = bpf_prog_test_run_skb,
10876 const struct bpf_verifier_ops lwt_out_verifier_ops = {
10877 .get_func_proto = lwt_out_func_proto,
10878 .is_valid_access = lwt_is_valid_access,
10879 .convert_ctx_access = bpf_convert_ctx_access,
10882 const struct bpf_prog_ops lwt_out_prog_ops = {
10883 .test_run = bpf_prog_test_run_skb,
10886 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
10887 .get_func_proto = lwt_xmit_func_proto,
10888 .is_valid_access = lwt_is_valid_access,
10889 .convert_ctx_access = bpf_convert_ctx_access,
10890 .gen_prologue = tc_cls_act_prologue,
10893 const struct bpf_prog_ops lwt_xmit_prog_ops = {
10894 .test_run = bpf_prog_test_run_skb,
10897 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
10898 .get_func_proto = lwt_seg6local_func_proto,
10899 .is_valid_access = lwt_is_valid_access,
10900 .convert_ctx_access = bpf_convert_ctx_access,
10903 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
10904 .test_run = bpf_prog_test_run_skb,
10907 const struct bpf_verifier_ops cg_sock_verifier_ops = {
10908 .get_func_proto = sock_filter_func_proto,
10909 .is_valid_access = sock_filter_is_valid_access,
10910 .convert_ctx_access = bpf_sock_convert_ctx_access,
10913 const struct bpf_prog_ops cg_sock_prog_ops = {
10916 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
10917 .get_func_proto = sock_addr_func_proto,
10918 .is_valid_access = sock_addr_is_valid_access,
10919 .convert_ctx_access = sock_addr_convert_ctx_access,
10922 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
10925 const struct bpf_verifier_ops sock_ops_verifier_ops = {
10926 .get_func_proto = sock_ops_func_proto,
10927 .is_valid_access = sock_ops_is_valid_access,
10928 .convert_ctx_access = sock_ops_convert_ctx_access,
10931 const struct bpf_prog_ops sock_ops_prog_ops = {
10934 const struct bpf_verifier_ops sk_skb_verifier_ops = {
10935 .get_func_proto = sk_skb_func_proto,
10936 .is_valid_access = sk_skb_is_valid_access,
10937 .convert_ctx_access = sk_skb_convert_ctx_access,
10938 .gen_prologue = sk_skb_prologue,
10941 const struct bpf_prog_ops sk_skb_prog_ops = {
10944 const struct bpf_verifier_ops sk_msg_verifier_ops = {
10945 .get_func_proto = sk_msg_func_proto,
10946 .is_valid_access = sk_msg_is_valid_access,
10947 .convert_ctx_access = sk_msg_convert_ctx_access,
10948 .gen_prologue = bpf_noop_prologue,
10951 const struct bpf_prog_ops sk_msg_prog_ops = {
10954 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
10955 .get_func_proto = flow_dissector_func_proto,
10956 .is_valid_access = flow_dissector_is_valid_access,
10957 .convert_ctx_access = flow_dissector_convert_ctx_access,
10960 const struct bpf_prog_ops flow_dissector_prog_ops = {
10961 .test_run = bpf_prog_test_run_flow_dissector,
10964 int sk_detach_filter(struct sock *sk)
10967 struct sk_filter *filter;
10969 if (sock_flag(sk, SOCK_FILTER_LOCKED))
10972 filter = rcu_dereference_protected(sk->sk_filter,
10973 lockdep_sock_is_held(sk));
10975 RCU_INIT_POINTER(sk->sk_filter, NULL);
10976 sk_filter_uncharge(sk, filter);
10982 EXPORT_SYMBOL_GPL(sk_detach_filter);
10984 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len)
10986 struct sock_fprog_kern *fprog;
10987 struct sk_filter *filter;
10990 sockopt_lock_sock(sk);
10991 filter = rcu_dereference_protected(sk->sk_filter,
10992 lockdep_sock_is_held(sk));
10996 /* We're copying the filter that has been originally attached,
10997 * so no conversion/decode needed anymore. eBPF programs that
10998 * have no original program cannot be dumped through this.
11001 fprog = filter->prog->orig_prog;
11007 /* User space only enquires number of filter blocks. */
11011 if (len < fprog->len)
11015 if (copy_to_sockptr(optval, fprog->filter, bpf_classic_proglen(fprog)))
11018 /* Instead of bytes, the API requests to return the number
11019 * of filter blocks.
11023 sockopt_release_sock(sk);
11028 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
11029 struct sock_reuseport *reuse,
11030 struct sock *sk, struct sk_buff *skb,
11031 struct sock *migrating_sk,
11034 reuse_kern->skb = skb;
11035 reuse_kern->sk = sk;
11036 reuse_kern->selected_sk = NULL;
11037 reuse_kern->migrating_sk = migrating_sk;
11038 reuse_kern->data_end = skb->data + skb_headlen(skb);
11039 reuse_kern->hash = hash;
11040 reuse_kern->reuseport_id = reuse->reuseport_id;
11041 reuse_kern->bind_inany = reuse->bind_inany;
11044 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
11045 struct bpf_prog *prog, struct sk_buff *skb,
11046 struct sock *migrating_sk,
11049 struct sk_reuseport_kern reuse_kern;
11050 enum sk_action action;
11052 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
11053 action = bpf_prog_run(prog, &reuse_kern);
11055 if (action == SK_PASS)
11056 return reuse_kern.selected_sk;
11058 return ERR_PTR(-ECONNREFUSED);
11061 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
11062 struct bpf_map *, map, void *, key, u32, flags)
11064 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
11065 struct sock_reuseport *reuse;
11066 struct sock *selected_sk;
11068 selected_sk = map->ops->map_lookup_elem(map, key);
11072 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
11074 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
11075 if (sk_is_refcounted(selected_sk))
11076 sock_put(selected_sk);
11078 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
11079 * The only (!reuse) case here is - the sk has already been
11080 * unhashed (e.g. by close()), so treat it as -ENOENT.
11082 * Other maps (e.g. sock_map) do not provide this guarantee and
11083 * the sk may never be in the reuseport group to begin with.
11085 return is_sockarray ? -ENOENT : -EINVAL;
11088 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
11089 struct sock *sk = reuse_kern->sk;
11091 if (sk->sk_protocol != selected_sk->sk_protocol)
11092 return -EPROTOTYPE;
11093 else if (sk->sk_family != selected_sk->sk_family)
11094 return -EAFNOSUPPORT;
11096 /* Catch all. Likely bound to a different sockaddr. */
11100 reuse_kern->selected_sk = selected_sk;
11105 static const struct bpf_func_proto sk_select_reuseport_proto = {
11106 .func = sk_select_reuseport,
11108 .ret_type = RET_INTEGER,
11109 .arg1_type = ARG_PTR_TO_CTX,
11110 .arg2_type = ARG_CONST_MAP_PTR,
11111 .arg3_type = ARG_PTR_TO_MAP_KEY,
11112 .arg4_type = ARG_ANYTHING,
11115 BPF_CALL_4(sk_reuseport_load_bytes,
11116 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11117 void *, to, u32, len)
11119 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
11122 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
11123 .func = sk_reuseport_load_bytes,
11125 .ret_type = RET_INTEGER,
11126 .arg1_type = ARG_PTR_TO_CTX,
11127 .arg2_type = ARG_ANYTHING,
11128 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
11129 .arg4_type = ARG_CONST_SIZE,
11132 BPF_CALL_5(sk_reuseport_load_bytes_relative,
11133 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11134 void *, to, u32, len, u32, start_header)
11136 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
11137 len, start_header);
11140 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
11141 .func = sk_reuseport_load_bytes_relative,
11143 .ret_type = RET_INTEGER,
11144 .arg1_type = ARG_PTR_TO_CTX,
11145 .arg2_type = ARG_ANYTHING,
11146 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
11147 .arg4_type = ARG_CONST_SIZE,
11148 .arg5_type = ARG_ANYTHING,
11151 static const struct bpf_func_proto *
11152 sk_reuseport_func_proto(enum bpf_func_id func_id,
11153 const struct bpf_prog *prog)
11156 case BPF_FUNC_sk_select_reuseport:
11157 return &sk_select_reuseport_proto;
11158 case BPF_FUNC_skb_load_bytes:
11159 return &sk_reuseport_load_bytes_proto;
11160 case BPF_FUNC_skb_load_bytes_relative:
11161 return &sk_reuseport_load_bytes_relative_proto;
11162 case BPF_FUNC_get_socket_cookie:
11163 return &bpf_get_socket_ptr_cookie_proto;
11164 case BPF_FUNC_ktime_get_coarse_ns:
11165 return &bpf_ktime_get_coarse_ns_proto;
11167 return bpf_base_func_proto(func_id);
11172 sk_reuseport_is_valid_access(int off, int size,
11173 enum bpf_access_type type,
11174 const struct bpf_prog *prog,
11175 struct bpf_insn_access_aux *info)
11177 const u32 size_default = sizeof(__u32);
11179 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
11180 off % size || type != BPF_READ)
11184 case offsetof(struct sk_reuseport_md, data):
11185 info->reg_type = PTR_TO_PACKET;
11186 return size == sizeof(__u64);
11188 case offsetof(struct sk_reuseport_md, data_end):
11189 info->reg_type = PTR_TO_PACKET_END;
11190 return size == sizeof(__u64);
11192 case offsetof(struct sk_reuseport_md, hash):
11193 return size == size_default;
11195 case offsetof(struct sk_reuseport_md, sk):
11196 info->reg_type = PTR_TO_SOCKET;
11197 return size == sizeof(__u64);
11199 case offsetof(struct sk_reuseport_md, migrating_sk):
11200 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
11201 return size == sizeof(__u64);
11203 /* Fields that allow narrowing */
11204 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
11205 if (size < sizeof_field(struct sk_buff, protocol))
11208 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
11209 case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
11210 case bpf_ctx_range(struct sk_reuseport_md, len):
11211 bpf_ctx_record_field_size(info, size_default);
11212 return bpf_ctx_narrow_access_ok(off, size, size_default);
11219 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
11220 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
11221 si->dst_reg, si->src_reg, \
11222 bpf_target_off(struct sk_reuseport_kern, F, \
11223 sizeof_field(struct sk_reuseport_kern, F), \
11227 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
11228 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
11233 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \
11234 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
11239 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
11240 const struct bpf_insn *si,
11241 struct bpf_insn *insn_buf,
11242 struct bpf_prog *prog,
11245 struct bpf_insn *insn = insn_buf;
11248 case offsetof(struct sk_reuseport_md, data):
11249 SK_REUSEPORT_LOAD_SKB_FIELD(data);
11252 case offsetof(struct sk_reuseport_md, len):
11253 SK_REUSEPORT_LOAD_SKB_FIELD(len);
11256 case offsetof(struct sk_reuseport_md, eth_protocol):
11257 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
11260 case offsetof(struct sk_reuseport_md, ip_protocol):
11261 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
11264 case offsetof(struct sk_reuseport_md, data_end):
11265 SK_REUSEPORT_LOAD_FIELD(data_end);
11268 case offsetof(struct sk_reuseport_md, hash):
11269 SK_REUSEPORT_LOAD_FIELD(hash);
11272 case offsetof(struct sk_reuseport_md, bind_inany):
11273 SK_REUSEPORT_LOAD_FIELD(bind_inany);
11276 case offsetof(struct sk_reuseport_md, sk):
11277 SK_REUSEPORT_LOAD_FIELD(sk);
11280 case offsetof(struct sk_reuseport_md, migrating_sk):
11281 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
11285 return insn - insn_buf;
11288 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
11289 .get_func_proto = sk_reuseport_func_proto,
11290 .is_valid_access = sk_reuseport_is_valid_access,
11291 .convert_ctx_access = sk_reuseport_convert_ctx_access,
11294 const struct bpf_prog_ops sk_reuseport_prog_ops = {
11297 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
11298 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
11300 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
11301 struct sock *, sk, u64, flags)
11303 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
11304 BPF_SK_LOOKUP_F_NO_REUSEPORT)))
11306 if (unlikely(sk && sk_is_refcounted(sk)))
11307 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
11308 if (unlikely(sk && sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN))
11309 return -ESOCKTNOSUPPORT; /* only accept TCP socket in LISTEN */
11310 if (unlikely(sk && sk_is_udp(sk) && sk->sk_state != TCP_CLOSE))
11311 return -ESOCKTNOSUPPORT; /* only accept UDP socket in CLOSE */
11313 /* Check if socket is suitable for packet L3/L4 protocol */
11314 if (sk && sk->sk_protocol != ctx->protocol)
11315 return -EPROTOTYPE;
11316 if (sk && sk->sk_family != ctx->family &&
11317 (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
11318 return -EAFNOSUPPORT;
11320 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
11323 /* Select socket as lookup result */
11324 ctx->selected_sk = sk;
11325 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
11329 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
11330 .func = bpf_sk_lookup_assign,
11332 .ret_type = RET_INTEGER,
11333 .arg1_type = ARG_PTR_TO_CTX,
11334 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL,
11335 .arg3_type = ARG_ANYTHING,
11338 static const struct bpf_func_proto *
11339 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
11342 case BPF_FUNC_perf_event_output:
11343 return &bpf_event_output_data_proto;
11344 case BPF_FUNC_sk_assign:
11345 return &bpf_sk_lookup_assign_proto;
11346 case BPF_FUNC_sk_release:
11347 return &bpf_sk_release_proto;
11349 return bpf_sk_base_func_proto(func_id);
11353 static bool sk_lookup_is_valid_access(int off, int size,
11354 enum bpf_access_type type,
11355 const struct bpf_prog *prog,
11356 struct bpf_insn_access_aux *info)
11358 if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
11360 if (off % size != 0)
11362 if (type != BPF_READ)
11366 case offsetof(struct bpf_sk_lookup, sk):
11367 info->reg_type = PTR_TO_SOCKET_OR_NULL;
11368 return size == sizeof(__u64);
11370 case bpf_ctx_range(struct bpf_sk_lookup, family):
11371 case bpf_ctx_range(struct bpf_sk_lookup, protocol):
11372 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
11373 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
11374 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
11375 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
11376 case bpf_ctx_range(struct bpf_sk_lookup, local_port):
11377 case bpf_ctx_range(struct bpf_sk_lookup, ingress_ifindex):
11378 bpf_ctx_record_field_size(info, sizeof(__u32));
11379 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
11381 case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
11382 /* Allow 4-byte access to 2-byte field for backward compatibility */
11383 if (size == sizeof(__u32))
11385 bpf_ctx_record_field_size(info, sizeof(__be16));
11386 return bpf_ctx_narrow_access_ok(off, size, sizeof(__be16));
11388 case offsetofend(struct bpf_sk_lookup, remote_port) ...
11389 offsetof(struct bpf_sk_lookup, local_ip4) - 1:
11390 /* Allow access to zero padding for backward compatibility */
11391 bpf_ctx_record_field_size(info, sizeof(__u16));
11392 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u16));
11399 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
11400 const struct bpf_insn *si,
11401 struct bpf_insn *insn_buf,
11402 struct bpf_prog *prog,
11405 struct bpf_insn *insn = insn_buf;
11408 case offsetof(struct bpf_sk_lookup, sk):
11409 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11410 offsetof(struct bpf_sk_lookup_kern, selected_sk));
11413 case offsetof(struct bpf_sk_lookup, family):
11414 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11415 bpf_target_off(struct bpf_sk_lookup_kern,
11416 family, 2, target_size));
11419 case offsetof(struct bpf_sk_lookup, protocol):
11420 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11421 bpf_target_off(struct bpf_sk_lookup_kern,
11422 protocol, 2, target_size));
11425 case offsetof(struct bpf_sk_lookup, remote_ip4):
11426 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11427 bpf_target_off(struct bpf_sk_lookup_kern,
11428 v4.saddr, 4, target_size));
11431 case offsetof(struct bpf_sk_lookup, local_ip4):
11432 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11433 bpf_target_off(struct bpf_sk_lookup_kern,
11434 v4.daddr, 4, target_size));
11437 case bpf_ctx_range_till(struct bpf_sk_lookup,
11438 remote_ip6[0], remote_ip6[3]): {
11439 #if IS_ENABLED(CONFIG_IPV6)
11442 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
11443 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11444 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11445 offsetof(struct bpf_sk_lookup_kern, v6.saddr));
11446 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11447 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11449 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11453 case bpf_ctx_range_till(struct bpf_sk_lookup,
11454 local_ip6[0], local_ip6[3]): {
11455 #if IS_ENABLED(CONFIG_IPV6)
11458 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
11459 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11460 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11461 offsetof(struct bpf_sk_lookup_kern, v6.daddr));
11462 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11463 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11465 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11469 case offsetof(struct bpf_sk_lookup, remote_port):
11470 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11471 bpf_target_off(struct bpf_sk_lookup_kern,
11472 sport, 2, target_size));
11475 case offsetofend(struct bpf_sk_lookup, remote_port):
11477 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11480 case offsetof(struct bpf_sk_lookup, local_port):
11481 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11482 bpf_target_off(struct bpf_sk_lookup_kern,
11483 dport, 2, target_size));
11486 case offsetof(struct bpf_sk_lookup, ingress_ifindex):
11487 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11488 bpf_target_off(struct bpf_sk_lookup_kern,
11489 ingress_ifindex, 4, target_size));
11493 return insn - insn_buf;
11496 const struct bpf_prog_ops sk_lookup_prog_ops = {
11497 .test_run = bpf_prog_test_run_sk_lookup,
11500 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
11501 .get_func_proto = sk_lookup_func_proto,
11502 .is_valid_access = sk_lookup_is_valid_access,
11503 .convert_ctx_access = sk_lookup_convert_ctx_access,
11506 #endif /* CONFIG_INET */
11508 DEFINE_BPF_DISPATCHER(xdp)
11510 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
11512 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
11515 BTF_ID_LIST_GLOBAL(btf_sock_ids, MAX_BTF_SOCK_TYPE)
11516 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
11518 #undef BTF_SOCK_TYPE
11520 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
11522 /* tcp6_sock type is not generated in dwarf and hence btf,
11523 * trigger an explicit type generation here.
11525 BTF_TYPE_EMIT(struct tcp6_sock);
11526 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
11527 sk->sk_family == AF_INET6)
11528 return (unsigned long)sk;
11530 return (unsigned long)NULL;
11533 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
11534 .func = bpf_skc_to_tcp6_sock,
11536 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11537 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11538 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
11541 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
11543 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
11544 return (unsigned long)sk;
11546 return (unsigned long)NULL;
11549 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
11550 .func = bpf_skc_to_tcp_sock,
11552 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11553 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11554 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
11557 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
11559 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
11560 * generated if CONFIG_INET=n. Trigger an explicit generation here.
11562 BTF_TYPE_EMIT(struct inet_timewait_sock);
11563 BTF_TYPE_EMIT(struct tcp_timewait_sock);
11566 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
11567 return (unsigned long)sk;
11570 #if IS_BUILTIN(CONFIG_IPV6)
11571 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
11572 return (unsigned long)sk;
11575 return (unsigned long)NULL;
11578 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
11579 .func = bpf_skc_to_tcp_timewait_sock,
11581 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11582 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11583 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
11586 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
11589 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11590 return (unsigned long)sk;
11593 #if IS_BUILTIN(CONFIG_IPV6)
11594 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11595 return (unsigned long)sk;
11598 return (unsigned long)NULL;
11601 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
11602 .func = bpf_skc_to_tcp_request_sock,
11604 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11605 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11606 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
11609 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
11611 /* udp6_sock type is not generated in dwarf and hence btf,
11612 * trigger an explicit type generation here.
11614 BTF_TYPE_EMIT(struct udp6_sock);
11615 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
11616 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
11617 return (unsigned long)sk;
11619 return (unsigned long)NULL;
11622 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
11623 .func = bpf_skc_to_udp6_sock,
11625 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11626 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11627 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
11630 BPF_CALL_1(bpf_skc_to_unix_sock, struct sock *, sk)
11632 /* unix_sock type is not generated in dwarf and hence btf,
11633 * trigger an explicit type generation here.
11635 BTF_TYPE_EMIT(struct unix_sock);
11636 if (sk && sk_fullsock(sk) && sk->sk_family == AF_UNIX)
11637 return (unsigned long)sk;
11639 return (unsigned long)NULL;
11642 const struct bpf_func_proto bpf_skc_to_unix_sock_proto = {
11643 .func = bpf_skc_to_unix_sock,
11645 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11646 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11647 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UNIX],
11650 BPF_CALL_1(bpf_skc_to_mptcp_sock, struct sock *, sk)
11652 BTF_TYPE_EMIT(struct mptcp_sock);
11653 return (unsigned long)bpf_mptcp_sock_from_subflow(sk);
11656 const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto = {
11657 .func = bpf_skc_to_mptcp_sock,
11659 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11660 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
11661 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_MPTCP],
11664 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
11666 return (unsigned long)sock_from_file(file);
11669 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
11670 BTF_ID(struct, socket)
11671 BTF_ID(struct, file)
11673 const struct bpf_func_proto bpf_sock_from_file_proto = {
11674 .func = bpf_sock_from_file,
11676 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11677 .ret_btf_id = &bpf_sock_from_file_btf_ids[0],
11678 .arg1_type = ARG_PTR_TO_BTF_ID,
11679 .arg1_btf_id = &bpf_sock_from_file_btf_ids[1],
11682 static const struct bpf_func_proto *
11683 bpf_sk_base_func_proto(enum bpf_func_id func_id)
11685 const struct bpf_func_proto *func;
11688 case BPF_FUNC_skc_to_tcp6_sock:
11689 func = &bpf_skc_to_tcp6_sock_proto;
11691 case BPF_FUNC_skc_to_tcp_sock:
11692 func = &bpf_skc_to_tcp_sock_proto;
11694 case BPF_FUNC_skc_to_tcp_timewait_sock:
11695 func = &bpf_skc_to_tcp_timewait_sock_proto;
11697 case BPF_FUNC_skc_to_tcp_request_sock:
11698 func = &bpf_skc_to_tcp_request_sock_proto;
11700 case BPF_FUNC_skc_to_udp6_sock:
11701 func = &bpf_skc_to_udp6_sock_proto;
11703 case BPF_FUNC_skc_to_unix_sock:
11704 func = &bpf_skc_to_unix_sock_proto;
11706 case BPF_FUNC_skc_to_mptcp_sock:
11707 func = &bpf_skc_to_mptcp_sock_proto;
11709 case BPF_FUNC_ktime_get_coarse_ns:
11710 return &bpf_ktime_get_coarse_ns_proto;
11712 return bpf_base_func_proto(func_id);
11715 if (!perfmon_capable())