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 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_all) != 4);
330 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
331 offsetof(struct sk_buff, vlan_all));
332 *insn++ = BPF_JMP_IMM(BPF_JEQ, dst_reg, 0, 1);
333 *insn++ = BPF_ALU32_IMM(BPF_MOV, 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);
2136 /* At ingress, the mac header has already been pulled once.
2137 * At egress, skb_pospull_rcsum has to be done in case that
2138 * the skb is originated from ingress (i.e. a forwarded skb)
2139 * to ensure that rcsum starts at net header.
2141 if (!skb_at_tc_ingress(skb))
2142 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2144 skb_pop_mac_header(skb);
2145 skb_reset_mac_len(skb);
2146 return flags & BPF_F_INGRESS ?
2147 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2150 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2153 /* Verify that a link layer header is carried */
2154 if (unlikely(skb->mac_header >= skb->network_header || skb->len == 0)) {
2159 bpf_push_mac_rcsum(skb);
2160 return flags & BPF_F_INGRESS ?
2161 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2164 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2167 if (dev_is_mac_header_xmit(dev))
2168 return __bpf_redirect_common(skb, dev, flags);
2170 return __bpf_redirect_no_mac(skb, dev, flags);
2173 #if IS_ENABLED(CONFIG_IPV6)
2174 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2175 struct net_device *dev, struct bpf_nh_params *nh)
2177 u32 hh_len = LL_RESERVED_SPACE(dev);
2178 const struct in6_addr *nexthop;
2179 struct dst_entry *dst = NULL;
2180 struct neighbour *neigh;
2182 if (dev_xmit_recursion()) {
2183 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2188 skb_clear_tstamp(skb);
2190 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2191 skb = skb_expand_head(skb, hh_len);
2199 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2200 &ipv6_hdr(skb)->daddr);
2202 nexthop = &nh->ipv6_nh;
2204 neigh = ip_neigh_gw6(dev, nexthop);
2205 if (likely(!IS_ERR(neigh))) {
2208 sock_confirm_neigh(skb, neigh);
2209 dev_xmit_recursion_inc();
2210 ret = neigh_output(neigh, skb, false);
2211 dev_xmit_recursion_dec();
2212 rcu_read_unlock_bh();
2215 rcu_read_unlock_bh();
2217 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2223 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2224 struct bpf_nh_params *nh)
2226 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2227 struct net *net = dev_net(dev);
2228 int err, ret = NET_XMIT_DROP;
2231 struct dst_entry *dst;
2232 struct flowi6 fl6 = {
2233 .flowi6_flags = FLOWI_FLAG_ANYSRC,
2234 .flowi6_mark = skb->mark,
2235 .flowlabel = ip6_flowinfo(ip6h),
2236 .flowi6_oif = dev->ifindex,
2237 .flowi6_proto = ip6h->nexthdr,
2238 .daddr = ip6h->daddr,
2239 .saddr = ip6h->saddr,
2242 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2246 skb_dst_set(skb, dst);
2247 } else if (nh->nh_family != AF_INET6) {
2251 err = bpf_out_neigh_v6(net, skb, dev, nh);
2252 if (unlikely(net_xmit_eval(err)))
2253 dev->stats.tx_errors++;
2255 ret = NET_XMIT_SUCCESS;
2258 dev->stats.tx_errors++;
2264 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2265 struct bpf_nh_params *nh)
2268 return NET_XMIT_DROP;
2270 #endif /* CONFIG_IPV6 */
2272 #if IS_ENABLED(CONFIG_INET)
2273 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2274 struct net_device *dev, struct bpf_nh_params *nh)
2276 u32 hh_len = LL_RESERVED_SPACE(dev);
2277 struct neighbour *neigh;
2278 bool is_v6gw = false;
2280 if (dev_xmit_recursion()) {
2281 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2286 skb_clear_tstamp(skb);
2288 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2289 skb = skb_expand_head(skb, hh_len);
2296 struct dst_entry *dst = skb_dst(skb);
2297 struct rtable *rt = container_of(dst, struct rtable, dst);
2299 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2300 } else if (nh->nh_family == AF_INET6) {
2301 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2303 } else if (nh->nh_family == AF_INET) {
2304 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2306 rcu_read_unlock_bh();
2310 if (likely(!IS_ERR(neigh))) {
2313 sock_confirm_neigh(skb, neigh);
2314 dev_xmit_recursion_inc();
2315 ret = neigh_output(neigh, skb, is_v6gw);
2316 dev_xmit_recursion_dec();
2317 rcu_read_unlock_bh();
2320 rcu_read_unlock_bh();
2326 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2327 struct bpf_nh_params *nh)
2329 const struct iphdr *ip4h = ip_hdr(skb);
2330 struct net *net = dev_net(dev);
2331 int err, ret = NET_XMIT_DROP;
2334 struct flowi4 fl4 = {
2335 .flowi4_flags = FLOWI_FLAG_ANYSRC,
2336 .flowi4_mark = skb->mark,
2337 .flowi4_tos = RT_TOS(ip4h->tos),
2338 .flowi4_oif = dev->ifindex,
2339 .flowi4_proto = ip4h->protocol,
2340 .daddr = ip4h->daddr,
2341 .saddr = ip4h->saddr,
2345 rt = ip_route_output_flow(net, &fl4, NULL);
2348 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2353 skb_dst_set(skb, &rt->dst);
2356 err = bpf_out_neigh_v4(net, skb, dev, nh);
2357 if (unlikely(net_xmit_eval(err)))
2358 dev->stats.tx_errors++;
2360 ret = NET_XMIT_SUCCESS;
2363 dev->stats.tx_errors++;
2369 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2370 struct bpf_nh_params *nh)
2373 return NET_XMIT_DROP;
2375 #endif /* CONFIG_INET */
2377 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2378 struct bpf_nh_params *nh)
2380 struct ethhdr *ethh = eth_hdr(skb);
2382 if (unlikely(skb->mac_header >= skb->network_header))
2384 bpf_push_mac_rcsum(skb);
2385 if (is_multicast_ether_addr(ethh->h_dest))
2388 skb_pull(skb, sizeof(*ethh));
2389 skb_unset_mac_header(skb);
2390 skb_reset_network_header(skb);
2392 if (skb->protocol == htons(ETH_P_IP))
2393 return __bpf_redirect_neigh_v4(skb, dev, nh);
2394 else if (skb->protocol == htons(ETH_P_IPV6))
2395 return __bpf_redirect_neigh_v6(skb, dev, nh);
2401 /* Internal, non-exposed redirect flags. */
2403 BPF_F_NEIGH = (1ULL << 1),
2404 BPF_F_PEER = (1ULL << 2),
2405 BPF_F_NEXTHOP = (1ULL << 3),
2406 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2409 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2411 struct net_device *dev;
2412 struct sk_buff *clone;
2415 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2418 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2422 clone = skb_clone(skb, GFP_ATOMIC);
2423 if (unlikely(!clone))
2426 /* For direct write, we need to keep the invariant that the skbs
2427 * we're dealing with need to be uncloned. Should uncloning fail
2428 * here, we need to free the just generated clone to unclone once
2431 ret = bpf_try_make_head_writable(skb);
2432 if (unlikely(ret)) {
2437 return __bpf_redirect(clone, dev, flags);
2440 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2441 .func = bpf_clone_redirect,
2443 .ret_type = RET_INTEGER,
2444 .arg1_type = ARG_PTR_TO_CTX,
2445 .arg2_type = ARG_ANYTHING,
2446 .arg3_type = ARG_ANYTHING,
2449 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2450 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2452 int skb_do_redirect(struct sk_buff *skb)
2454 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2455 struct net *net = dev_net(skb->dev);
2456 struct net_device *dev;
2457 u32 flags = ri->flags;
2459 dev = dev_get_by_index_rcu(net, ri->tgt_index);
2464 if (flags & BPF_F_PEER) {
2465 const struct net_device_ops *ops = dev->netdev_ops;
2467 if (unlikely(!ops->ndo_get_peer_dev ||
2468 !skb_at_tc_ingress(skb)))
2470 dev = ops->ndo_get_peer_dev(dev);
2471 if (unlikely(!dev ||
2472 !(dev->flags & IFF_UP) ||
2473 net_eq(net, dev_net(dev))))
2478 return flags & BPF_F_NEIGH ?
2479 __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2481 __bpf_redirect(skb, dev, flags);
2487 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2489 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2491 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2495 ri->tgt_index = ifindex;
2497 return TC_ACT_REDIRECT;
2500 static const struct bpf_func_proto bpf_redirect_proto = {
2501 .func = bpf_redirect,
2503 .ret_type = RET_INTEGER,
2504 .arg1_type = ARG_ANYTHING,
2505 .arg2_type = ARG_ANYTHING,
2508 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2510 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2512 if (unlikely(flags))
2515 ri->flags = BPF_F_PEER;
2516 ri->tgt_index = ifindex;
2518 return TC_ACT_REDIRECT;
2521 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2522 .func = bpf_redirect_peer,
2524 .ret_type = RET_INTEGER,
2525 .arg1_type = ARG_ANYTHING,
2526 .arg2_type = ARG_ANYTHING,
2529 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2530 int, plen, u64, flags)
2532 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2534 if (unlikely((plen && plen < sizeof(*params)) || flags))
2537 ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2538 ri->tgt_index = ifindex;
2540 BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2542 memcpy(&ri->nh, params, sizeof(ri->nh));
2544 return TC_ACT_REDIRECT;
2547 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2548 .func = bpf_redirect_neigh,
2550 .ret_type = RET_INTEGER,
2551 .arg1_type = ARG_ANYTHING,
2552 .arg2_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2553 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
2554 .arg4_type = ARG_ANYTHING,
2557 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2559 msg->apply_bytes = bytes;
2563 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2564 .func = bpf_msg_apply_bytes,
2566 .ret_type = RET_INTEGER,
2567 .arg1_type = ARG_PTR_TO_CTX,
2568 .arg2_type = ARG_ANYTHING,
2571 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2573 msg->cork_bytes = bytes;
2577 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2578 .func = bpf_msg_cork_bytes,
2580 .ret_type = RET_INTEGER,
2581 .arg1_type = ARG_PTR_TO_CTX,
2582 .arg2_type = ARG_ANYTHING,
2585 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2586 u32, end, u64, flags)
2588 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2589 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2590 struct scatterlist *sge;
2591 u8 *raw, *to, *from;
2594 if (unlikely(flags || end <= start))
2597 /* First find the starting scatterlist element */
2601 len = sk_msg_elem(msg, i)->length;
2602 if (start < offset + len)
2604 sk_msg_iter_var_next(i);
2605 } while (i != msg->sg.end);
2607 if (unlikely(start >= offset + len))
2611 /* The start may point into the sg element so we need to also
2612 * account for the headroom.
2614 bytes_sg_total = start - offset + bytes;
2615 if (!test_bit(i, msg->sg.copy) && bytes_sg_total <= len)
2618 /* At this point we need to linearize multiple scatterlist
2619 * elements or a single shared page. Either way we need to
2620 * copy into a linear buffer exclusively owned by BPF. Then
2621 * place the buffer in the scatterlist and fixup the original
2622 * entries by removing the entries now in the linear buffer
2623 * and shifting the remaining entries. For now we do not try
2624 * to copy partial entries to avoid complexity of running out
2625 * of sg_entry slots. The downside is reading a single byte
2626 * will copy the entire sg entry.
2629 copy += sk_msg_elem(msg, i)->length;
2630 sk_msg_iter_var_next(i);
2631 if (bytes_sg_total <= copy)
2633 } while (i != msg->sg.end);
2636 if (unlikely(bytes_sg_total > copy))
2639 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2641 if (unlikely(!page))
2644 raw = page_address(page);
2647 sge = sk_msg_elem(msg, i);
2648 from = sg_virt(sge);
2652 memcpy(to, from, len);
2655 put_page(sg_page(sge));
2657 sk_msg_iter_var_next(i);
2658 } while (i != last_sge);
2660 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2662 /* To repair sg ring we need to shift entries. If we only
2663 * had a single entry though we can just replace it and
2664 * be done. Otherwise walk the ring and shift the entries.
2666 WARN_ON_ONCE(last_sge == first_sge);
2667 shift = last_sge > first_sge ?
2668 last_sge - first_sge - 1 :
2669 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2674 sk_msg_iter_var_next(i);
2678 if (i + shift >= NR_MSG_FRAG_IDS)
2679 move_from = i + shift - NR_MSG_FRAG_IDS;
2681 move_from = i + shift;
2682 if (move_from == msg->sg.end)
2685 msg->sg.data[i] = msg->sg.data[move_from];
2686 msg->sg.data[move_from].length = 0;
2687 msg->sg.data[move_from].page_link = 0;
2688 msg->sg.data[move_from].offset = 0;
2689 sk_msg_iter_var_next(i);
2692 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2693 msg->sg.end - shift + NR_MSG_FRAG_IDS :
2694 msg->sg.end - shift;
2696 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2697 msg->data_end = msg->data + bytes;
2701 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2702 .func = bpf_msg_pull_data,
2704 .ret_type = RET_INTEGER,
2705 .arg1_type = ARG_PTR_TO_CTX,
2706 .arg2_type = ARG_ANYTHING,
2707 .arg3_type = ARG_ANYTHING,
2708 .arg4_type = ARG_ANYTHING,
2711 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2712 u32, len, u64, flags)
2714 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2715 u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2716 u8 *raw, *to, *from;
2719 if (unlikely(flags))
2722 if (unlikely(len == 0))
2725 /* First find the starting scatterlist element */
2729 l = sk_msg_elem(msg, i)->length;
2731 if (start < offset + l)
2733 sk_msg_iter_var_next(i);
2734 } while (i != msg->sg.end);
2736 if (start >= offset + l)
2739 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2741 /* If no space available will fallback to copy, we need at
2742 * least one scatterlist elem available to push data into
2743 * when start aligns to the beginning of an element or two
2744 * when it falls inside an element. We handle the start equals
2745 * offset case because its the common case for inserting a
2748 if (!space || (space == 1 && start != offset))
2749 copy = msg->sg.data[i].length;
2751 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2752 get_order(copy + len));
2753 if (unlikely(!page))
2759 raw = page_address(page);
2761 psge = sk_msg_elem(msg, i);
2762 front = start - offset;
2763 back = psge->length - front;
2764 from = sg_virt(psge);
2767 memcpy(raw, from, front);
2771 to = raw + front + len;
2773 memcpy(to, from, back);
2776 put_page(sg_page(psge));
2777 } else if (start - offset) {
2778 psge = sk_msg_elem(msg, i);
2779 rsge = sk_msg_elem_cpy(msg, i);
2781 psge->length = start - offset;
2782 rsge.length -= psge->length;
2783 rsge.offset += start;
2785 sk_msg_iter_var_next(i);
2786 sg_unmark_end(psge);
2787 sg_unmark_end(&rsge);
2788 sk_msg_iter_next(msg, end);
2791 /* Slot(s) to place newly allocated data */
2794 /* Shift one or two slots as needed */
2796 sge = sk_msg_elem_cpy(msg, i);
2798 sk_msg_iter_var_next(i);
2799 sg_unmark_end(&sge);
2800 sk_msg_iter_next(msg, end);
2802 nsge = sk_msg_elem_cpy(msg, i);
2804 sk_msg_iter_var_next(i);
2805 nnsge = sk_msg_elem_cpy(msg, i);
2808 while (i != msg->sg.end) {
2809 msg->sg.data[i] = sge;
2811 sk_msg_iter_var_next(i);
2814 nnsge = sk_msg_elem_cpy(msg, i);
2816 nsge = sk_msg_elem_cpy(msg, i);
2821 /* Place newly allocated data buffer */
2822 sk_mem_charge(msg->sk, len);
2823 msg->sg.size += len;
2824 __clear_bit(new, msg->sg.copy);
2825 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2827 get_page(sg_page(&rsge));
2828 sk_msg_iter_var_next(new);
2829 msg->sg.data[new] = rsge;
2832 sk_msg_compute_data_pointers(msg);
2836 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2837 .func = bpf_msg_push_data,
2839 .ret_type = RET_INTEGER,
2840 .arg1_type = ARG_PTR_TO_CTX,
2841 .arg2_type = ARG_ANYTHING,
2842 .arg3_type = ARG_ANYTHING,
2843 .arg4_type = ARG_ANYTHING,
2846 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2852 sk_msg_iter_var_next(i);
2853 msg->sg.data[prev] = msg->sg.data[i];
2854 } while (i != msg->sg.end);
2856 sk_msg_iter_prev(msg, end);
2859 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2861 struct scatterlist tmp, sge;
2863 sk_msg_iter_next(msg, end);
2864 sge = sk_msg_elem_cpy(msg, i);
2865 sk_msg_iter_var_next(i);
2866 tmp = sk_msg_elem_cpy(msg, i);
2868 while (i != msg->sg.end) {
2869 msg->sg.data[i] = sge;
2870 sk_msg_iter_var_next(i);
2872 tmp = sk_msg_elem_cpy(msg, i);
2876 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2877 u32, len, u64, flags)
2879 u32 i = 0, l = 0, space, offset = 0;
2880 u64 last = start + len;
2883 if (unlikely(flags))
2886 /* First find the starting scatterlist element */
2890 l = sk_msg_elem(msg, i)->length;
2892 if (start < offset + l)
2894 sk_msg_iter_var_next(i);
2895 } while (i != msg->sg.end);
2897 /* Bounds checks: start and pop must be inside message */
2898 if (start >= offset + l || last >= msg->sg.size)
2901 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2904 /* --------------| offset
2905 * -| start |-------- len -------|
2907 * |----- a ----|-------- pop -------|----- b ----|
2908 * |______________________________________________| length
2911 * a: region at front of scatter element to save
2912 * b: region at back of scatter element to save when length > A + pop
2913 * pop: region to pop from element, same as input 'pop' here will be
2914 * decremented below per iteration.
2916 * Two top-level cases to handle when start != offset, first B is non
2917 * zero and second B is zero corresponding to when a pop includes more
2920 * Then if B is non-zero AND there is no space allocate space and
2921 * compact A, B regions into page. If there is space shift ring to
2922 * the rigth free'ing the next element in ring to place B, leaving
2923 * A untouched except to reduce length.
2925 if (start != offset) {
2926 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2928 int b = sge->length - pop - a;
2930 sk_msg_iter_var_next(i);
2932 if (pop < sge->length - a) {
2935 sk_msg_shift_right(msg, i);
2936 nsge = sk_msg_elem(msg, i);
2937 get_page(sg_page(sge));
2940 b, sge->offset + pop + a);
2942 struct page *page, *orig;
2945 page = alloc_pages(__GFP_NOWARN |
2946 __GFP_COMP | GFP_ATOMIC,
2948 if (unlikely(!page))
2952 orig = sg_page(sge);
2953 from = sg_virt(sge);
2954 to = page_address(page);
2955 memcpy(to, from, a);
2956 memcpy(to + a, from + a + pop, b);
2957 sg_set_page(sge, page, a + b, 0);
2961 } else if (pop >= sge->length - a) {
2962 pop -= (sge->length - a);
2967 /* From above the current layout _must_ be as follows,
2972 * |---- pop ---|---------------- b ------------|
2973 * |____________________________________________| length
2975 * Offset and start of the current msg elem are equal because in the
2976 * previous case we handled offset != start and either consumed the
2977 * entire element and advanced to the next element OR pop == 0.
2979 * Two cases to handle here are first pop is less than the length
2980 * leaving some remainder b above. Simply adjust the element's layout
2981 * in this case. Or pop >= length of the element so that b = 0. In this
2982 * case advance to next element decrementing pop.
2985 struct scatterlist *sge = sk_msg_elem(msg, i);
2987 if (pop < sge->length) {
2993 sk_msg_shift_left(msg, i);
2995 sk_msg_iter_var_next(i);
2998 sk_mem_uncharge(msg->sk, len - pop);
2999 msg->sg.size -= (len - pop);
3000 sk_msg_compute_data_pointers(msg);
3004 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
3005 .func = bpf_msg_pop_data,
3007 .ret_type = RET_INTEGER,
3008 .arg1_type = ARG_PTR_TO_CTX,
3009 .arg2_type = ARG_ANYTHING,
3010 .arg3_type = ARG_ANYTHING,
3011 .arg4_type = ARG_ANYTHING,
3014 #ifdef CONFIG_CGROUP_NET_CLASSID
3015 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3017 return __task_get_classid(current);
3020 const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3021 .func = bpf_get_cgroup_classid_curr,
3023 .ret_type = RET_INTEGER,
3026 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3028 struct sock *sk = skb_to_full_sk(skb);
3030 if (!sk || !sk_fullsock(sk))
3033 return sock_cgroup_classid(&sk->sk_cgrp_data);
3036 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3037 .func = bpf_skb_cgroup_classid,
3039 .ret_type = RET_INTEGER,
3040 .arg1_type = ARG_PTR_TO_CTX,
3044 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3046 return task_get_classid(skb);
3049 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3050 .func = bpf_get_cgroup_classid,
3052 .ret_type = RET_INTEGER,
3053 .arg1_type = ARG_PTR_TO_CTX,
3056 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3058 return dst_tclassid(skb);
3061 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3062 .func = bpf_get_route_realm,
3064 .ret_type = RET_INTEGER,
3065 .arg1_type = ARG_PTR_TO_CTX,
3068 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3070 /* If skb_clear_hash() was called due to mangling, we can
3071 * trigger SW recalculation here. Later access to hash
3072 * can then use the inline skb->hash via context directly
3073 * instead of calling this helper again.
3075 return skb_get_hash(skb);
3078 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3079 .func = bpf_get_hash_recalc,
3081 .ret_type = RET_INTEGER,
3082 .arg1_type = ARG_PTR_TO_CTX,
3085 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3087 /* After all direct packet write, this can be used once for
3088 * triggering a lazy recalc on next skb_get_hash() invocation.
3090 skb_clear_hash(skb);
3094 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3095 .func = bpf_set_hash_invalid,
3097 .ret_type = RET_INTEGER,
3098 .arg1_type = ARG_PTR_TO_CTX,
3101 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3103 /* Set user specified hash as L4(+), so that it gets returned
3104 * on skb_get_hash() call unless BPF prog later on triggers a
3107 __skb_set_sw_hash(skb, hash, true);
3111 static const struct bpf_func_proto bpf_set_hash_proto = {
3112 .func = bpf_set_hash,
3114 .ret_type = RET_INTEGER,
3115 .arg1_type = ARG_PTR_TO_CTX,
3116 .arg2_type = ARG_ANYTHING,
3119 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3124 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3125 vlan_proto != htons(ETH_P_8021AD)))
3126 vlan_proto = htons(ETH_P_8021Q);
3128 bpf_push_mac_rcsum(skb);
3129 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3130 bpf_pull_mac_rcsum(skb);
3132 bpf_compute_data_pointers(skb);
3136 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3137 .func = bpf_skb_vlan_push,
3139 .ret_type = RET_INTEGER,
3140 .arg1_type = ARG_PTR_TO_CTX,
3141 .arg2_type = ARG_ANYTHING,
3142 .arg3_type = ARG_ANYTHING,
3145 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3149 bpf_push_mac_rcsum(skb);
3150 ret = skb_vlan_pop(skb);
3151 bpf_pull_mac_rcsum(skb);
3153 bpf_compute_data_pointers(skb);
3157 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3158 .func = bpf_skb_vlan_pop,
3160 .ret_type = RET_INTEGER,
3161 .arg1_type = ARG_PTR_TO_CTX,
3164 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3166 /* Caller already did skb_cow() with len as headroom,
3167 * so no need to do it here.
3170 memmove(skb->data, skb->data + len, off);
3171 memset(skb->data + off, 0, len);
3173 /* No skb_postpush_rcsum(skb, skb->data + off, len)
3174 * needed here as it does not change the skb->csum
3175 * result for checksum complete when summing over
3181 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3185 /* skb_ensure_writable() is not needed here, as we're
3186 * already working on an uncloned skb.
3188 if (unlikely(!pskb_may_pull(skb, off + len)))
3191 old_data = skb->data;
3192 __skb_pull(skb, len);
3193 skb_postpull_rcsum(skb, old_data + off, len);
3194 memmove(skb->data, old_data, off);
3199 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3201 bool trans_same = skb->transport_header == skb->network_header;
3204 /* There's no need for __skb_push()/__skb_pull() pair to
3205 * get to the start of the mac header as we're guaranteed
3206 * to always start from here under eBPF.
3208 ret = bpf_skb_generic_push(skb, off, len);
3210 skb->mac_header -= len;
3211 skb->network_header -= len;
3213 skb->transport_header = skb->network_header;
3219 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3221 bool trans_same = skb->transport_header == skb->network_header;
3224 /* Same here, __skb_push()/__skb_pull() pair not needed. */
3225 ret = bpf_skb_generic_pop(skb, off, len);
3227 skb->mac_header += len;
3228 skb->network_header += len;
3230 skb->transport_header = skb->network_header;
3236 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3238 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3239 u32 off = skb_mac_header_len(skb);
3242 ret = skb_cow(skb, len_diff);
3243 if (unlikely(ret < 0))
3246 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3247 if (unlikely(ret < 0))
3250 if (skb_is_gso(skb)) {
3251 struct skb_shared_info *shinfo = skb_shinfo(skb);
3253 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */
3254 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3255 shinfo->gso_type &= ~SKB_GSO_TCPV4;
3256 shinfo->gso_type |= SKB_GSO_TCPV6;
3260 skb->protocol = htons(ETH_P_IPV6);
3261 skb_clear_hash(skb);
3266 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3268 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3269 u32 off = skb_mac_header_len(skb);
3272 ret = skb_unclone(skb, GFP_ATOMIC);
3273 if (unlikely(ret < 0))
3276 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3277 if (unlikely(ret < 0))
3280 if (skb_is_gso(skb)) {
3281 struct skb_shared_info *shinfo = skb_shinfo(skb);
3283 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */
3284 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3285 shinfo->gso_type &= ~SKB_GSO_TCPV6;
3286 shinfo->gso_type |= SKB_GSO_TCPV4;
3290 skb->protocol = htons(ETH_P_IP);
3291 skb_clear_hash(skb);
3296 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3298 __be16 from_proto = skb->protocol;
3300 if (from_proto == htons(ETH_P_IP) &&
3301 to_proto == htons(ETH_P_IPV6))
3302 return bpf_skb_proto_4_to_6(skb);
3304 if (from_proto == htons(ETH_P_IPV6) &&
3305 to_proto == htons(ETH_P_IP))
3306 return bpf_skb_proto_6_to_4(skb);
3311 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3316 if (unlikely(flags))
3319 /* General idea is that this helper does the basic groundwork
3320 * needed for changing the protocol, and eBPF program fills the
3321 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3322 * and other helpers, rather than passing a raw buffer here.
3324 * The rationale is to keep this minimal and without a need to
3325 * deal with raw packet data. F.e. even if we would pass buffers
3326 * here, the program still needs to call the bpf_lX_csum_replace()
3327 * helpers anyway. Plus, this way we keep also separation of
3328 * concerns, since f.e. bpf_skb_store_bytes() should only take
3331 * Currently, additional options and extension header space are
3332 * not supported, but flags register is reserved so we can adapt
3333 * that. For offloads, we mark packet as dodgy, so that headers
3334 * need to be verified first.
3336 ret = bpf_skb_proto_xlat(skb, proto);
3337 bpf_compute_data_pointers(skb);
3341 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3342 .func = bpf_skb_change_proto,
3344 .ret_type = RET_INTEGER,
3345 .arg1_type = ARG_PTR_TO_CTX,
3346 .arg2_type = ARG_ANYTHING,
3347 .arg3_type = ARG_ANYTHING,
3350 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3352 /* We only allow a restricted subset to be changed for now. */
3353 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3354 !skb_pkt_type_ok(pkt_type)))
3357 skb->pkt_type = pkt_type;
3361 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3362 .func = bpf_skb_change_type,
3364 .ret_type = RET_INTEGER,
3365 .arg1_type = ARG_PTR_TO_CTX,
3366 .arg2_type = ARG_ANYTHING,
3369 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3371 switch (skb->protocol) {
3372 case htons(ETH_P_IP):
3373 return sizeof(struct iphdr);
3374 case htons(ETH_P_IPV6):
3375 return sizeof(struct ipv6hdr);
3381 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3382 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3384 #define BPF_F_ADJ_ROOM_DECAP_L3_MASK (BPF_F_ADJ_ROOM_DECAP_L3_IPV4 | \
3385 BPF_F_ADJ_ROOM_DECAP_L3_IPV6)
3387 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \
3388 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3389 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3390 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3391 BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \
3392 BPF_F_ADJ_ROOM_ENCAP_L2( \
3393 BPF_ADJ_ROOM_ENCAP_L2_MASK) | \
3394 BPF_F_ADJ_ROOM_DECAP_L3_MASK)
3396 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3399 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3400 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3401 u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3402 unsigned int gso_type = SKB_GSO_DODGY;
3405 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3406 /* udp gso_size delineates datagrams, only allow if fixed */
3407 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3408 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3412 ret = skb_cow_head(skb, len_diff);
3413 if (unlikely(ret < 0))
3417 if (skb->protocol != htons(ETH_P_IP) &&
3418 skb->protocol != htons(ETH_P_IPV6))
3421 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3422 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3425 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3426 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3429 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH &&
3430 inner_mac_len < ETH_HLEN)
3433 if (skb->encapsulation)
3436 mac_len = skb->network_header - skb->mac_header;
3437 inner_net = skb->network_header;
3438 if (inner_mac_len > len_diff)
3440 inner_trans = skb->transport_header;
3443 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3444 if (unlikely(ret < 0))
3448 skb->inner_mac_header = inner_net - inner_mac_len;
3449 skb->inner_network_header = inner_net;
3450 skb->inner_transport_header = inner_trans;
3452 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH)
3453 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
3455 skb_set_inner_protocol(skb, skb->protocol);
3457 skb->encapsulation = 1;
3458 skb_set_network_header(skb, mac_len);
3460 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3461 gso_type |= SKB_GSO_UDP_TUNNEL;
3462 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3463 gso_type |= SKB_GSO_GRE;
3464 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3465 gso_type |= SKB_GSO_IPXIP6;
3466 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3467 gso_type |= SKB_GSO_IPXIP4;
3469 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3470 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3471 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3472 sizeof(struct ipv6hdr) :
3473 sizeof(struct iphdr);
3475 skb_set_transport_header(skb, mac_len + nh_len);
3478 /* Match skb->protocol to new outer l3 protocol */
3479 if (skb->protocol == htons(ETH_P_IP) &&
3480 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3481 skb->protocol = htons(ETH_P_IPV6);
3482 else if (skb->protocol == htons(ETH_P_IPV6) &&
3483 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3484 skb->protocol = htons(ETH_P_IP);
3487 if (skb_is_gso(skb)) {
3488 struct skb_shared_info *shinfo = skb_shinfo(skb);
3490 /* Due to header grow, MSS needs to be downgraded. */
3491 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3492 skb_decrease_gso_size(shinfo, len_diff);
3494 /* Header must be checked, and gso_segs recomputed. */
3495 shinfo->gso_type |= gso_type;
3496 shinfo->gso_segs = 0;
3502 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3507 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3508 BPF_F_ADJ_ROOM_DECAP_L3_MASK |
3509 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3512 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3513 /* udp gso_size delineates datagrams, only allow if fixed */
3514 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3515 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3519 ret = skb_unclone(skb, GFP_ATOMIC);
3520 if (unlikely(ret < 0))
3523 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3524 if (unlikely(ret < 0))
3527 /* Match skb->protocol to new outer l3 protocol */
3528 if (skb->protocol == htons(ETH_P_IP) &&
3529 flags & BPF_F_ADJ_ROOM_DECAP_L3_IPV6)
3530 skb->protocol = htons(ETH_P_IPV6);
3531 else if (skb->protocol == htons(ETH_P_IPV6) &&
3532 flags & BPF_F_ADJ_ROOM_DECAP_L3_IPV4)
3533 skb->protocol = htons(ETH_P_IP);
3535 if (skb_is_gso(skb)) {
3536 struct skb_shared_info *shinfo = skb_shinfo(skb);
3538 /* Due to header shrink, MSS can be upgraded. */
3539 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3540 skb_increase_gso_size(shinfo, len_diff);
3542 /* Header must be checked, and gso_segs recomputed. */
3543 shinfo->gso_type |= SKB_GSO_DODGY;
3544 shinfo->gso_segs = 0;
3550 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3552 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3553 u32, mode, u64, flags)
3555 u32 len_diff_abs = abs(len_diff);
3556 bool shrink = len_diff < 0;
3559 if (unlikely(flags || mode))
3561 if (unlikely(len_diff_abs > 0xfffU))
3565 ret = skb_cow(skb, len_diff);
3566 if (unlikely(ret < 0))
3568 __skb_push(skb, len_diff_abs);
3569 memset(skb->data, 0, len_diff_abs);
3571 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3573 __skb_pull(skb, len_diff_abs);
3575 if (tls_sw_has_ctx_rx(skb->sk)) {
3576 struct strp_msg *rxm = strp_msg(skb);
3578 rxm->full_len += len_diff;
3583 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3584 .func = sk_skb_adjust_room,
3586 .ret_type = RET_INTEGER,
3587 .arg1_type = ARG_PTR_TO_CTX,
3588 .arg2_type = ARG_ANYTHING,
3589 .arg3_type = ARG_ANYTHING,
3590 .arg4_type = ARG_ANYTHING,
3593 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3594 u32, mode, u64, flags)
3596 u32 len_cur, len_diff_abs = abs(len_diff);
3597 u32 len_min = bpf_skb_net_base_len(skb);
3598 u32 len_max = BPF_SKB_MAX_LEN;
3599 __be16 proto = skb->protocol;
3600 bool shrink = len_diff < 0;
3604 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3605 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3607 if (unlikely(len_diff_abs > 0xfffU))
3609 if (unlikely(proto != htons(ETH_P_IP) &&
3610 proto != htons(ETH_P_IPV6)))
3613 off = skb_mac_header_len(skb);
3615 case BPF_ADJ_ROOM_NET:
3616 off += bpf_skb_net_base_len(skb);
3618 case BPF_ADJ_ROOM_MAC:
3624 if (flags & BPF_F_ADJ_ROOM_DECAP_L3_MASK) {
3628 switch (flags & BPF_F_ADJ_ROOM_DECAP_L3_MASK) {
3629 case BPF_F_ADJ_ROOM_DECAP_L3_IPV4:
3630 len_min = sizeof(struct iphdr);
3632 case BPF_F_ADJ_ROOM_DECAP_L3_IPV6:
3633 len_min = sizeof(struct ipv6hdr);
3640 len_cur = skb->len - skb_network_offset(skb);
3641 if ((shrink && (len_diff_abs >= len_cur ||
3642 len_cur - len_diff_abs < len_min)) ||
3643 (!shrink && (skb->len + len_diff_abs > len_max &&
3647 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3648 bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3649 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3650 __skb_reset_checksum_unnecessary(skb);
3652 bpf_compute_data_pointers(skb);
3656 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3657 .func = bpf_skb_adjust_room,
3659 .ret_type = RET_INTEGER,
3660 .arg1_type = ARG_PTR_TO_CTX,
3661 .arg2_type = ARG_ANYTHING,
3662 .arg3_type = ARG_ANYTHING,
3663 .arg4_type = ARG_ANYTHING,
3666 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3668 u32 min_len = skb_network_offset(skb);
3670 if (skb_transport_header_was_set(skb))
3671 min_len = skb_transport_offset(skb);
3672 if (skb->ip_summed == CHECKSUM_PARTIAL)
3673 min_len = skb_checksum_start_offset(skb) +
3674 skb->csum_offset + sizeof(__sum16);
3678 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3680 unsigned int old_len = skb->len;
3683 ret = __skb_grow_rcsum(skb, new_len);
3685 memset(skb->data + old_len, 0, new_len - old_len);
3689 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3691 return __skb_trim_rcsum(skb, new_len);
3694 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3697 u32 max_len = BPF_SKB_MAX_LEN;
3698 u32 min_len = __bpf_skb_min_len(skb);
3701 if (unlikely(flags || new_len > max_len || new_len < min_len))
3703 if (skb->encapsulation)
3706 /* The basic idea of this helper is that it's performing the
3707 * needed work to either grow or trim an skb, and eBPF program
3708 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3709 * bpf_lX_csum_replace() and others rather than passing a raw
3710 * buffer here. This one is a slow path helper and intended
3711 * for replies with control messages.
3713 * Like in bpf_skb_change_proto(), we want to keep this rather
3714 * minimal and without protocol specifics so that we are able
3715 * to separate concerns as in bpf_skb_store_bytes() should only
3716 * be the one responsible for writing buffers.
3718 * It's really expected to be a slow path operation here for
3719 * control message replies, so we're implicitly linearizing,
3720 * uncloning and drop offloads from the skb by this.
3722 ret = __bpf_try_make_writable(skb, skb->len);
3724 if (new_len > skb->len)
3725 ret = bpf_skb_grow_rcsum(skb, new_len);
3726 else if (new_len < skb->len)
3727 ret = bpf_skb_trim_rcsum(skb, new_len);
3728 if (!ret && skb_is_gso(skb))
3734 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3737 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3739 bpf_compute_data_pointers(skb);
3743 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3744 .func = bpf_skb_change_tail,
3746 .ret_type = RET_INTEGER,
3747 .arg1_type = ARG_PTR_TO_CTX,
3748 .arg2_type = ARG_ANYTHING,
3749 .arg3_type = ARG_ANYTHING,
3752 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3755 return __bpf_skb_change_tail(skb, new_len, flags);
3758 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3759 .func = sk_skb_change_tail,
3761 .ret_type = RET_INTEGER,
3762 .arg1_type = ARG_PTR_TO_CTX,
3763 .arg2_type = ARG_ANYTHING,
3764 .arg3_type = ARG_ANYTHING,
3767 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3770 u32 max_len = BPF_SKB_MAX_LEN;
3771 u32 new_len = skb->len + head_room;
3774 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3775 new_len < skb->len))
3778 ret = skb_cow(skb, head_room);
3780 /* Idea for this helper is that we currently only
3781 * allow to expand on mac header. This means that
3782 * skb->protocol network header, etc, stay as is.
3783 * Compared to bpf_skb_change_tail(), we're more
3784 * flexible due to not needing to linearize or
3785 * reset GSO. Intention for this helper is to be
3786 * used by an L3 skb that needs to push mac header
3787 * for redirection into L2 device.
3789 __skb_push(skb, head_room);
3790 memset(skb->data, 0, head_room);
3791 skb_reset_mac_header(skb);
3792 skb_reset_mac_len(skb);
3798 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3801 int ret = __bpf_skb_change_head(skb, head_room, flags);
3803 bpf_compute_data_pointers(skb);
3807 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3808 .func = bpf_skb_change_head,
3810 .ret_type = RET_INTEGER,
3811 .arg1_type = ARG_PTR_TO_CTX,
3812 .arg2_type = ARG_ANYTHING,
3813 .arg3_type = ARG_ANYTHING,
3816 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3819 return __bpf_skb_change_head(skb, head_room, flags);
3822 static const struct bpf_func_proto sk_skb_change_head_proto = {
3823 .func = sk_skb_change_head,
3825 .ret_type = RET_INTEGER,
3826 .arg1_type = ARG_PTR_TO_CTX,
3827 .arg2_type = ARG_ANYTHING,
3828 .arg3_type = ARG_ANYTHING,
3831 BPF_CALL_1(bpf_xdp_get_buff_len, struct xdp_buff*, xdp)
3833 return xdp_get_buff_len(xdp);
3836 static const struct bpf_func_proto bpf_xdp_get_buff_len_proto = {
3837 .func = bpf_xdp_get_buff_len,
3839 .ret_type = RET_INTEGER,
3840 .arg1_type = ARG_PTR_TO_CTX,
3843 BTF_ID_LIST_SINGLE(bpf_xdp_get_buff_len_bpf_ids, struct, xdp_buff)
3845 const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto = {
3846 .func = bpf_xdp_get_buff_len,
3848 .arg1_type = ARG_PTR_TO_BTF_ID,
3849 .arg1_btf_id = &bpf_xdp_get_buff_len_bpf_ids[0],
3852 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3854 return xdp_data_meta_unsupported(xdp) ? 0 :
3855 xdp->data - xdp->data_meta;
3858 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3860 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3861 unsigned long metalen = xdp_get_metalen(xdp);
3862 void *data_start = xdp_frame_end + metalen;
3863 void *data = xdp->data + offset;
3865 if (unlikely(data < data_start ||
3866 data > xdp->data_end - ETH_HLEN))
3870 memmove(xdp->data_meta + offset,
3871 xdp->data_meta, metalen);
3872 xdp->data_meta += offset;
3878 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3879 .func = bpf_xdp_adjust_head,
3881 .ret_type = RET_INTEGER,
3882 .arg1_type = ARG_PTR_TO_CTX,
3883 .arg2_type = ARG_ANYTHING,
3886 static void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off,
3887 void *buf, unsigned long len, bool flush)
3889 unsigned long ptr_len, ptr_off = 0;
3890 skb_frag_t *next_frag, *end_frag;
3891 struct skb_shared_info *sinfo;
3895 if (likely(xdp->data_end - xdp->data >= off + len)) {
3896 src = flush ? buf : xdp->data + off;
3897 dst = flush ? xdp->data + off : buf;
3898 memcpy(dst, src, len);
3902 sinfo = xdp_get_shared_info_from_buff(xdp);
3903 end_frag = &sinfo->frags[sinfo->nr_frags];
3904 next_frag = &sinfo->frags[0];
3906 ptr_len = xdp->data_end - xdp->data;
3907 ptr_buf = xdp->data;
3910 if (off < ptr_off + ptr_len) {
3911 unsigned long copy_off = off - ptr_off;
3912 unsigned long copy_len = min(len, ptr_len - copy_off);
3914 src = flush ? buf : ptr_buf + copy_off;
3915 dst = flush ? ptr_buf + copy_off : buf;
3916 memcpy(dst, src, copy_len);
3923 if (!len || next_frag == end_frag)
3927 ptr_buf = skb_frag_address(next_frag);
3928 ptr_len = skb_frag_size(next_frag);
3933 static void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len)
3935 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
3936 u32 size = xdp->data_end - xdp->data;
3937 void *addr = xdp->data;
3940 if (unlikely(offset > 0xffff || len > 0xffff))
3941 return ERR_PTR(-EFAULT);
3943 if (offset + len > xdp_get_buff_len(xdp))
3944 return ERR_PTR(-EINVAL);
3946 if (offset < size) /* linear area */
3950 for (i = 0; i < sinfo->nr_frags; i++) { /* paged area */
3951 u32 frag_size = skb_frag_size(&sinfo->frags[i]);
3953 if (offset < frag_size) {
3954 addr = skb_frag_address(&sinfo->frags[i]);
3958 offset -= frag_size;
3961 return offset + len <= size ? addr + offset : NULL;
3964 BPF_CALL_4(bpf_xdp_load_bytes, struct xdp_buff *, xdp, u32, offset,
3965 void *, buf, u32, len)
3969 ptr = bpf_xdp_pointer(xdp, offset, len);
3971 return PTR_ERR(ptr);
3974 bpf_xdp_copy_buf(xdp, offset, buf, len, false);
3976 memcpy(buf, ptr, len);
3981 static const struct bpf_func_proto bpf_xdp_load_bytes_proto = {
3982 .func = bpf_xdp_load_bytes,
3984 .ret_type = RET_INTEGER,
3985 .arg1_type = ARG_PTR_TO_CTX,
3986 .arg2_type = ARG_ANYTHING,
3987 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
3988 .arg4_type = ARG_CONST_SIZE,
3991 BPF_CALL_4(bpf_xdp_store_bytes, struct xdp_buff *, xdp, u32, offset,
3992 void *, buf, u32, len)
3996 ptr = bpf_xdp_pointer(xdp, offset, len);
3998 return PTR_ERR(ptr);
4001 bpf_xdp_copy_buf(xdp, offset, buf, len, true);
4003 memcpy(ptr, buf, len);
4008 static const struct bpf_func_proto bpf_xdp_store_bytes_proto = {
4009 .func = bpf_xdp_store_bytes,
4011 .ret_type = RET_INTEGER,
4012 .arg1_type = ARG_PTR_TO_CTX,
4013 .arg2_type = ARG_ANYTHING,
4014 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
4015 .arg4_type = ARG_CONST_SIZE,
4018 static int bpf_xdp_frags_increase_tail(struct xdp_buff *xdp, int offset)
4020 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
4021 skb_frag_t *frag = &sinfo->frags[sinfo->nr_frags - 1];
4022 struct xdp_rxq_info *rxq = xdp->rxq;
4023 unsigned int tailroom;
4025 if (!rxq->frag_size || rxq->frag_size > xdp->frame_sz)
4028 tailroom = rxq->frag_size - skb_frag_size(frag) - skb_frag_off(frag);
4029 if (unlikely(offset > tailroom))
4032 memset(skb_frag_address(frag) + skb_frag_size(frag), 0, offset);
4033 skb_frag_size_add(frag, offset);
4034 sinfo->xdp_frags_size += offset;
4039 static int bpf_xdp_frags_shrink_tail(struct xdp_buff *xdp, int offset)
4041 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
4042 int i, n_frags_free = 0, len_free = 0;
4044 if (unlikely(offset > (int)xdp_get_buff_len(xdp) - ETH_HLEN))
4047 for (i = sinfo->nr_frags - 1; i >= 0 && offset > 0; i--) {
4048 skb_frag_t *frag = &sinfo->frags[i];
4049 int shrink = min_t(int, offset, skb_frag_size(frag));
4054 if (skb_frag_size(frag) == shrink) {
4055 struct page *page = skb_frag_page(frag);
4057 __xdp_return(page_address(page), &xdp->rxq->mem,
4061 skb_frag_size_sub(frag, shrink);
4065 sinfo->nr_frags -= n_frags_free;
4066 sinfo->xdp_frags_size -= len_free;
4068 if (unlikely(!sinfo->nr_frags)) {
4069 xdp_buff_clear_frags_flag(xdp);
4070 xdp->data_end -= offset;
4076 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
4078 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
4079 void *data_end = xdp->data_end + offset;
4081 if (unlikely(xdp_buff_has_frags(xdp))) { /* non-linear xdp buff */
4083 return bpf_xdp_frags_shrink_tail(xdp, -offset);
4085 return bpf_xdp_frags_increase_tail(xdp, offset);
4088 /* Notice that xdp_data_hard_end have reserved some tailroom */
4089 if (unlikely(data_end > data_hard_end))
4092 /* ALL drivers MUST init xdp->frame_sz, chicken check below */
4093 if (unlikely(xdp->frame_sz > PAGE_SIZE)) {
4094 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz);
4098 if (unlikely(data_end < xdp->data + ETH_HLEN))
4101 /* Clear memory area on grow, can contain uninit kernel memory */
4103 memset(xdp->data_end, 0, offset);
4105 xdp->data_end = data_end;
4110 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
4111 .func = bpf_xdp_adjust_tail,
4113 .ret_type = RET_INTEGER,
4114 .arg1_type = ARG_PTR_TO_CTX,
4115 .arg2_type = ARG_ANYTHING,
4118 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
4120 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
4121 void *meta = xdp->data_meta + offset;
4122 unsigned long metalen = xdp->data - meta;
4124 if (xdp_data_meta_unsupported(xdp))
4126 if (unlikely(meta < xdp_frame_end ||
4129 if (unlikely(xdp_metalen_invalid(metalen)))
4132 xdp->data_meta = meta;
4137 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
4138 .func = bpf_xdp_adjust_meta,
4140 .ret_type = RET_INTEGER,
4141 .arg1_type = ARG_PTR_TO_CTX,
4142 .arg2_type = ARG_ANYTHING,
4148 * XDP_REDIRECT works by a three-step process, implemented in the functions
4151 * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target
4152 * of the redirect and store it (along with some other metadata) in a per-CPU
4153 * struct bpf_redirect_info.
4155 * 2. When the program returns the XDP_REDIRECT return code, the driver will
4156 * call xdp_do_redirect() which will use the information in struct
4157 * bpf_redirect_info to actually enqueue the frame into a map type-specific
4158 * bulk queue structure.
4160 * 3. Before exiting its NAPI poll loop, the driver will call
4161 * xdp_do_flush(), which will flush all the different bulk queues,
4162 * thus completing the redirect. Note that xdp_do_flush() must be
4163 * called before napi_complete_done() in the driver, as the
4164 * XDP_REDIRECT logic relies on being inside a single NAPI instance
4165 * through to the xdp_do_flush() call for RCU protection of all
4166 * in-kernel data structures.
4169 * Pointers to the map entries will be kept around for this whole sequence of
4170 * steps, protected by RCU. However, there is no top-level rcu_read_lock() in
4171 * the core code; instead, the RCU protection relies on everything happening
4172 * inside a single NAPI poll sequence, which means it's between a pair of calls
4173 * to local_bh_disable()/local_bh_enable().
4175 * The map entries are marked as __rcu and the map code makes sure to
4176 * dereference those pointers with rcu_dereference_check() in a way that works
4177 * for both sections that to hold an rcu_read_lock() and sections that are
4178 * called from NAPI without a separate rcu_read_lock(). The code below does not
4179 * use RCU annotations, but relies on those in the map code.
4181 void xdp_do_flush(void)
4187 EXPORT_SYMBOL_GPL(xdp_do_flush);
4189 void bpf_clear_redirect_map(struct bpf_map *map)
4191 struct bpf_redirect_info *ri;
4194 for_each_possible_cpu(cpu) {
4195 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
4196 /* Avoid polluting remote cacheline due to writes if
4197 * not needed. Once we pass this test, we need the
4198 * cmpxchg() to make sure it hasn't been changed in
4199 * the meantime by remote CPU.
4201 if (unlikely(READ_ONCE(ri->map) == map))
4202 cmpxchg(&ri->map, map, NULL);
4206 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
4207 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key);
4209 u32 xdp_master_redirect(struct xdp_buff *xdp)
4211 struct net_device *master, *slave;
4212 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4214 master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev);
4215 slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp);
4216 if (slave && slave != xdp->rxq->dev) {
4217 /* The target device is different from the receiving device, so
4218 * redirect it to the new device.
4219 * Using XDP_REDIRECT gets the correct behaviour from XDP enabled
4220 * drivers to unmap the packet from their rx ring.
4222 ri->tgt_index = slave->ifindex;
4223 ri->map_id = INT_MAX;
4224 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4225 return XDP_REDIRECT;
4229 EXPORT_SYMBOL_GPL(xdp_master_redirect);
4231 static inline int __xdp_do_redirect_xsk(struct bpf_redirect_info *ri,
4232 struct net_device *dev,
4233 struct xdp_buff *xdp,
4234 struct bpf_prog *xdp_prog)
4236 enum bpf_map_type map_type = ri->map_type;
4237 void *fwd = ri->tgt_value;
4238 u32 map_id = ri->map_id;
4241 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4242 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4244 err = __xsk_map_redirect(fwd, xdp);
4248 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4251 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4255 static __always_inline int __xdp_do_redirect_frame(struct bpf_redirect_info *ri,
4256 struct net_device *dev,
4257 struct xdp_frame *xdpf,
4258 struct bpf_prog *xdp_prog)
4260 enum bpf_map_type map_type = ri->map_type;
4261 void *fwd = ri->tgt_value;
4262 u32 map_id = ri->map_id;
4263 struct bpf_map *map;
4266 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4267 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4269 if (unlikely(!xdpf)) {
4275 case BPF_MAP_TYPE_DEVMAP:
4277 case BPF_MAP_TYPE_DEVMAP_HASH:
4278 map = READ_ONCE(ri->map);
4279 if (unlikely(map)) {
4280 WRITE_ONCE(ri->map, NULL);
4281 err = dev_map_enqueue_multi(xdpf, dev, map,
4282 ri->flags & BPF_F_EXCLUDE_INGRESS);
4284 err = dev_map_enqueue(fwd, xdpf, dev);
4287 case BPF_MAP_TYPE_CPUMAP:
4288 err = cpu_map_enqueue(fwd, xdpf, dev);
4290 case BPF_MAP_TYPE_UNSPEC:
4291 if (map_id == INT_MAX) {
4292 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4293 if (unlikely(!fwd)) {
4297 err = dev_xdp_enqueue(fwd, xdpf, dev);
4308 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4311 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4315 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
4316 struct bpf_prog *xdp_prog)
4318 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4319 enum bpf_map_type map_type = ri->map_type;
4321 if (map_type == BPF_MAP_TYPE_XSKMAP) {
4322 /* XDP_REDIRECT is not supported AF_XDP yet. */
4323 if (unlikely(xdp_buff_has_frags(xdp)))
4326 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4329 return __xdp_do_redirect_frame(ri, dev, xdp_convert_buff_to_frame(xdp),
4332 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4334 int xdp_do_redirect_frame(struct net_device *dev, struct xdp_buff *xdp,
4335 struct xdp_frame *xdpf, struct bpf_prog *xdp_prog)
4337 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4338 enum bpf_map_type map_type = ri->map_type;
4340 if (map_type == BPF_MAP_TYPE_XSKMAP)
4341 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4343 return __xdp_do_redirect_frame(ri, dev, xdpf, xdp_prog);
4345 EXPORT_SYMBOL_GPL(xdp_do_redirect_frame);
4347 static int xdp_do_generic_redirect_map(struct net_device *dev,
4348 struct sk_buff *skb,
4349 struct xdp_buff *xdp,
4350 struct bpf_prog *xdp_prog,
4352 enum bpf_map_type map_type, u32 map_id)
4354 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4355 struct bpf_map *map;
4359 case BPF_MAP_TYPE_DEVMAP:
4361 case BPF_MAP_TYPE_DEVMAP_HASH:
4362 map = READ_ONCE(ri->map);
4363 if (unlikely(map)) {
4364 WRITE_ONCE(ri->map, NULL);
4365 err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4366 ri->flags & BPF_F_EXCLUDE_INGRESS);
4368 err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4373 case BPF_MAP_TYPE_XSKMAP:
4374 err = xsk_generic_rcv(fwd, xdp);
4379 case BPF_MAP_TYPE_CPUMAP:
4380 err = cpu_map_generic_redirect(fwd, skb);
4389 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4392 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4396 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4397 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4399 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4400 enum bpf_map_type map_type = ri->map_type;
4401 void *fwd = ri->tgt_value;
4402 u32 map_id = ri->map_id;
4405 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4406 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4408 if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4409 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4410 if (unlikely(!fwd)) {
4415 err = xdp_ok_fwd_dev(fwd, skb->len);
4420 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4421 generic_xdp_tx(skb, xdp_prog);
4425 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id);
4427 _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4431 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4433 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4435 if (unlikely(flags))
4438 /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4439 * by map_idr) is used for ifindex based XDP redirect.
4441 ri->tgt_index = ifindex;
4442 ri->map_id = INT_MAX;
4443 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4445 return XDP_REDIRECT;
4448 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4449 .func = bpf_xdp_redirect,
4451 .ret_type = RET_INTEGER,
4452 .arg1_type = ARG_ANYTHING,
4453 .arg2_type = ARG_ANYTHING,
4456 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u64, key,
4459 return map->ops->map_redirect(map, key, flags);
4462 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4463 .func = bpf_xdp_redirect_map,
4465 .ret_type = RET_INTEGER,
4466 .arg1_type = ARG_CONST_MAP_PTR,
4467 .arg2_type = ARG_ANYTHING,
4468 .arg3_type = ARG_ANYTHING,
4471 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4472 unsigned long off, unsigned long len)
4474 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4478 if (ptr != dst_buff)
4479 memcpy(dst_buff, ptr, len);
4484 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4485 u64, flags, void *, meta, u64, meta_size)
4487 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4489 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4491 if (unlikely(!skb || skb_size > skb->len))
4494 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4498 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4499 .func = bpf_skb_event_output,
4501 .ret_type = RET_INTEGER,
4502 .arg1_type = ARG_PTR_TO_CTX,
4503 .arg2_type = ARG_CONST_MAP_PTR,
4504 .arg3_type = ARG_ANYTHING,
4505 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4506 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4509 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4511 const struct bpf_func_proto bpf_skb_output_proto = {
4512 .func = bpf_skb_event_output,
4514 .ret_type = RET_INTEGER,
4515 .arg1_type = ARG_PTR_TO_BTF_ID,
4516 .arg1_btf_id = &bpf_skb_output_btf_ids[0],
4517 .arg2_type = ARG_CONST_MAP_PTR,
4518 .arg3_type = ARG_ANYTHING,
4519 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4520 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4523 static unsigned short bpf_tunnel_key_af(u64 flags)
4525 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4528 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4529 u32, size, u64, flags)
4531 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4532 u8 compat[sizeof(struct bpf_tunnel_key)];
4536 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6 |
4537 BPF_F_TUNINFO_FLAGS)))) {
4541 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4545 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4548 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4549 case offsetof(struct bpf_tunnel_key, tunnel_label):
4550 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4552 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4553 /* Fixup deprecated structure layouts here, so we have
4554 * a common path later on.
4556 if (ip_tunnel_info_af(info) != AF_INET)
4559 to = (struct bpf_tunnel_key *)compat;
4566 to->tunnel_id = be64_to_cpu(info->key.tun_id);
4567 to->tunnel_tos = info->key.tos;
4568 to->tunnel_ttl = info->key.ttl;
4569 if (flags & BPF_F_TUNINFO_FLAGS)
4570 to->tunnel_flags = info->key.tun_flags;
4574 if (flags & BPF_F_TUNINFO_IPV6) {
4575 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4576 sizeof(to->remote_ipv6));
4577 memcpy(to->local_ipv6, &info->key.u.ipv6.dst,
4578 sizeof(to->local_ipv6));
4579 to->tunnel_label = be32_to_cpu(info->key.label);
4581 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4582 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4583 to->local_ipv4 = be32_to_cpu(info->key.u.ipv4.dst);
4584 memset(&to->local_ipv6[1], 0, sizeof(__u32) * 3);
4585 to->tunnel_label = 0;
4588 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4589 memcpy(to_orig, to, size);
4593 memset(to_orig, 0, size);
4597 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4598 .func = bpf_skb_get_tunnel_key,
4600 .ret_type = RET_INTEGER,
4601 .arg1_type = ARG_PTR_TO_CTX,
4602 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4603 .arg3_type = ARG_CONST_SIZE,
4604 .arg4_type = ARG_ANYTHING,
4607 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4609 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4612 if (unlikely(!info ||
4613 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4617 if (unlikely(size < info->options_len)) {
4622 ip_tunnel_info_opts_get(to, info);
4623 if (size > info->options_len)
4624 memset(to + info->options_len, 0, size - info->options_len);
4626 return info->options_len;
4628 memset(to, 0, size);
4632 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4633 .func = bpf_skb_get_tunnel_opt,
4635 .ret_type = RET_INTEGER,
4636 .arg1_type = ARG_PTR_TO_CTX,
4637 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4638 .arg3_type = ARG_CONST_SIZE,
4641 static struct metadata_dst __percpu *md_dst;
4643 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4644 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4646 struct metadata_dst *md = this_cpu_ptr(md_dst);
4647 u8 compat[sizeof(struct bpf_tunnel_key)];
4648 struct ip_tunnel_info *info;
4650 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4651 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER |
4652 BPF_F_NO_TUNNEL_KEY)))
4654 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4656 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4657 case offsetof(struct bpf_tunnel_key, tunnel_label):
4658 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4659 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4660 /* Fixup deprecated structure layouts here, so we have
4661 * a common path later on.
4663 memcpy(compat, from, size);
4664 memset(compat + size, 0, sizeof(compat) - size);
4665 from = (const struct bpf_tunnel_key *) compat;
4671 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4676 dst_hold((struct dst_entry *) md);
4677 skb_dst_set(skb, (struct dst_entry *) md);
4679 info = &md->u.tun_info;
4680 memset(info, 0, sizeof(*info));
4681 info->mode = IP_TUNNEL_INFO_TX;
4683 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4684 if (flags & BPF_F_DONT_FRAGMENT)
4685 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4686 if (flags & BPF_F_ZERO_CSUM_TX)
4687 info->key.tun_flags &= ~TUNNEL_CSUM;
4688 if (flags & BPF_F_SEQ_NUMBER)
4689 info->key.tun_flags |= TUNNEL_SEQ;
4690 if (flags & BPF_F_NO_TUNNEL_KEY)
4691 info->key.tun_flags &= ~TUNNEL_KEY;
4693 info->key.tun_id = cpu_to_be64(from->tunnel_id);
4694 info->key.tos = from->tunnel_tos;
4695 info->key.ttl = from->tunnel_ttl;
4697 if (flags & BPF_F_TUNINFO_IPV6) {
4698 info->mode |= IP_TUNNEL_INFO_IPV6;
4699 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4700 sizeof(from->remote_ipv6));
4701 memcpy(&info->key.u.ipv6.src, from->local_ipv6,
4702 sizeof(from->local_ipv6));
4703 info->key.label = cpu_to_be32(from->tunnel_label) &
4704 IPV6_FLOWLABEL_MASK;
4706 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4707 info->key.u.ipv4.src = cpu_to_be32(from->local_ipv4);
4708 info->key.flow_flags = FLOWI_FLAG_ANYSRC;
4714 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4715 .func = bpf_skb_set_tunnel_key,
4717 .ret_type = RET_INTEGER,
4718 .arg1_type = ARG_PTR_TO_CTX,
4719 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4720 .arg3_type = ARG_CONST_SIZE,
4721 .arg4_type = ARG_ANYTHING,
4724 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4725 const u8 *, from, u32, size)
4727 struct ip_tunnel_info *info = skb_tunnel_info(skb);
4728 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4730 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4732 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4735 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4740 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4741 .func = bpf_skb_set_tunnel_opt,
4743 .ret_type = RET_INTEGER,
4744 .arg1_type = ARG_PTR_TO_CTX,
4745 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4746 .arg3_type = ARG_CONST_SIZE,
4749 static const struct bpf_func_proto *
4750 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4753 struct metadata_dst __percpu *tmp;
4755 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4760 if (cmpxchg(&md_dst, NULL, tmp))
4761 metadata_dst_free_percpu(tmp);
4765 case BPF_FUNC_skb_set_tunnel_key:
4766 return &bpf_skb_set_tunnel_key_proto;
4767 case BPF_FUNC_skb_set_tunnel_opt:
4768 return &bpf_skb_set_tunnel_opt_proto;
4774 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4777 struct bpf_array *array = container_of(map, struct bpf_array, map);
4778 struct cgroup *cgrp;
4781 sk = skb_to_full_sk(skb);
4782 if (!sk || !sk_fullsock(sk))
4784 if (unlikely(idx >= array->map.max_entries))
4787 cgrp = READ_ONCE(array->ptrs[idx]);
4788 if (unlikely(!cgrp))
4791 return sk_under_cgroup_hierarchy(sk, cgrp);
4794 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4795 .func = bpf_skb_under_cgroup,
4797 .ret_type = RET_INTEGER,
4798 .arg1_type = ARG_PTR_TO_CTX,
4799 .arg2_type = ARG_CONST_MAP_PTR,
4800 .arg3_type = ARG_ANYTHING,
4803 #ifdef CONFIG_SOCK_CGROUP_DATA
4804 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4806 struct cgroup *cgrp;
4808 sk = sk_to_full_sk(sk);
4809 if (!sk || !sk_fullsock(sk))
4812 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4813 return cgroup_id(cgrp);
4816 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4818 return __bpf_sk_cgroup_id(skb->sk);
4821 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4822 .func = bpf_skb_cgroup_id,
4824 .ret_type = RET_INTEGER,
4825 .arg1_type = ARG_PTR_TO_CTX,
4828 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4831 struct cgroup *ancestor;
4832 struct cgroup *cgrp;
4834 sk = sk_to_full_sk(sk);
4835 if (!sk || !sk_fullsock(sk))
4838 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4839 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4843 return cgroup_id(ancestor);
4846 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4849 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4852 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4853 .func = bpf_skb_ancestor_cgroup_id,
4855 .ret_type = RET_INTEGER,
4856 .arg1_type = ARG_PTR_TO_CTX,
4857 .arg2_type = ARG_ANYTHING,
4860 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4862 return __bpf_sk_cgroup_id(sk);
4865 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4866 .func = bpf_sk_cgroup_id,
4868 .ret_type = RET_INTEGER,
4869 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4872 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4874 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4877 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4878 .func = bpf_sk_ancestor_cgroup_id,
4880 .ret_type = RET_INTEGER,
4881 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4882 .arg2_type = ARG_ANYTHING,
4886 static unsigned long bpf_xdp_copy(void *dst, const void *ctx,
4887 unsigned long off, unsigned long len)
4889 struct xdp_buff *xdp = (struct xdp_buff *)ctx;
4891 bpf_xdp_copy_buf(xdp, off, dst, len, false);
4895 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4896 u64, flags, void *, meta, u64, meta_size)
4898 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4900 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4903 if (unlikely(!xdp || xdp_size > xdp_get_buff_len(xdp)))
4906 return bpf_event_output(map, flags, meta, meta_size, xdp,
4907 xdp_size, bpf_xdp_copy);
4910 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4911 .func = bpf_xdp_event_output,
4913 .ret_type = RET_INTEGER,
4914 .arg1_type = ARG_PTR_TO_CTX,
4915 .arg2_type = ARG_CONST_MAP_PTR,
4916 .arg3_type = ARG_ANYTHING,
4917 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4918 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4921 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4923 const struct bpf_func_proto bpf_xdp_output_proto = {
4924 .func = bpf_xdp_event_output,
4926 .ret_type = RET_INTEGER,
4927 .arg1_type = ARG_PTR_TO_BTF_ID,
4928 .arg1_btf_id = &bpf_xdp_output_btf_ids[0],
4929 .arg2_type = ARG_CONST_MAP_PTR,
4930 .arg3_type = ARG_ANYTHING,
4931 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4932 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4935 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4937 return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4940 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4941 .func = bpf_get_socket_cookie,
4943 .ret_type = RET_INTEGER,
4944 .arg1_type = ARG_PTR_TO_CTX,
4947 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4949 return __sock_gen_cookie(ctx->sk);
4952 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4953 .func = bpf_get_socket_cookie_sock_addr,
4955 .ret_type = RET_INTEGER,
4956 .arg1_type = ARG_PTR_TO_CTX,
4959 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4961 return __sock_gen_cookie(ctx);
4964 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4965 .func = bpf_get_socket_cookie_sock,
4967 .ret_type = RET_INTEGER,
4968 .arg1_type = ARG_PTR_TO_CTX,
4971 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
4973 return sk ? sock_gen_cookie(sk) : 0;
4976 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
4977 .func = bpf_get_socket_ptr_cookie,
4979 .ret_type = RET_INTEGER,
4980 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4983 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4985 return __sock_gen_cookie(ctx->sk);
4988 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4989 .func = bpf_get_socket_cookie_sock_ops,
4991 .ret_type = RET_INTEGER,
4992 .arg1_type = ARG_PTR_TO_CTX,
4995 static u64 __bpf_get_netns_cookie(struct sock *sk)
4997 const struct net *net = sk ? sock_net(sk) : &init_net;
4999 return net->net_cookie;
5002 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
5004 return __bpf_get_netns_cookie(ctx);
5007 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
5008 .func = bpf_get_netns_cookie_sock,
5010 .ret_type = RET_INTEGER,
5011 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5014 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
5016 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5019 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
5020 .func = bpf_get_netns_cookie_sock_addr,
5022 .ret_type = RET_INTEGER,
5023 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5026 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
5028 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5031 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = {
5032 .func = bpf_get_netns_cookie_sock_ops,
5034 .ret_type = RET_INTEGER,
5035 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5038 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx)
5040 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5043 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = {
5044 .func = bpf_get_netns_cookie_sk_msg,
5046 .ret_type = RET_INTEGER,
5047 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5050 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
5052 struct sock *sk = sk_to_full_sk(skb->sk);
5055 if (!sk || !sk_fullsock(sk))
5057 kuid = sock_net_uid(sock_net(sk), sk);
5058 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
5061 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
5062 .func = bpf_get_socket_uid,
5064 .ret_type = RET_INTEGER,
5065 .arg1_type = ARG_PTR_TO_CTX,
5068 static int sol_socket_sockopt(struct sock *sk, int optname,
5069 char *optval, int *optlen,
5081 case SO_MAX_PACING_RATE:
5082 case SO_BINDTOIFINDEX:
5084 if (*optlen != sizeof(int))
5087 case SO_BINDTODEVICE:
5094 if (optname == SO_BINDTODEVICE)
5096 return sk_getsockopt(sk, SOL_SOCKET, optname,
5097 KERNEL_SOCKPTR(optval),
5098 KERNEL_SOCKPTR(optlen));
5101 return sk_setsockopt(sk, SOL_SOCKET, optname,
5102 KERNEL_SOCKPTR(optval), *optlen);
5105 static int bpf_sol_tcp_setsockopt(struct sock *sk, int optname,
5106 char *optval, int optlen)
5108 struct tcp_sock *tp = tcp_sk(sk);
5109 unsigned long timeout;
5112 if (optlen != sizeof(int))
5115 val = *(int *)optval;
5117 /* Only some options are supported */
5120 if (val <= 0 || tp->data_segs_out > tp->syn_data)
5122 tcp_snd_cwnd_set(tp, val);
5124 case TCP_BPF_SNDCWND_CLAMP:
5127 tp->snd_cwnd_clamp = val;
5128 tp->snd_ssthresh = val;
5130 case TCP_BPF_DELACK_MAX:
5131 timeout = usecs_to_jiffies(val);
5132 if (timeout > TCP_DELACK_MAX ||
5133 timeout < TCP_TIMEOUT_MIN)
5135 inet_csk(sk)->icsk_delack_max = timeout;
5137 case TCP_BPF_RTO_MIN:
5138 timeout = usecs_to_jiffies(val);
5139 if (timeout > TCP_RTO_MIN ||
5140 timeout < TCP_TIMEOUT_MIN)
5142 inet_csk(sk)->icsk_rto_min = timeout;
5151 static int sol_tcp_sockopt_congestion(struct sock *sk, char *optval,
5152 int *optlen, bool getopt)
5154 struct tcp_sock *tp;
5161 if (!inet_csk(sk)->icsk_ca_ops)
5163 /* BPF expects NULL-terminated tcp-cc string */
5164 optval[--(*optlen)] = '\0';
5165 return do_tcp_getsockopt(sk, SOL_TCP, TCP_CONGESTION,
5166 KERNEL_SOCKPTR(optval),
5167 KERNEL_SOCKPTR(optlen));
5170 /* "cdg" is the only cc that alloc a ptr
5171 * in inet_csk_ca area. The bpf-tcp-cc may
5172 * overwrite this ptr after switching to cdg.
5174 if (*optlen >= sizeof("cdg") - 1 && !strncmp("cdg", optval, *optlen))
5177 /* It stops this looping
5179 * .init => bpf_setsockopt(tcp_cc) => .init =>
5180 * bpf_setsockopt(tcp_cc)" => .init => ....
5182 * The second bpf_setsockopt(tcp_cc) is not allowed
5183 * in order to break the loop when both .init
5184 * are the same bpf prog.
5186 * This applies even the second bpf_setsockopt(tcp_cc)
5187 * does not cause a loop. This limits only the first
5188 * '.init' can call bpf_setsockopt(TCP_CONGESTION) to
5189 * pick a fallback cc (eg. peer does not support ECN)
5190 * and the second '.init' cannot fallback to
5194 if (tp->bpf_chg_cc_inprogress)
5197 tp->bpf_chg_cc_inprogress = 1;
5198 ret = do_tcp_setsockopt(sk, SOL_TCP, TCP_CONGESTION,
5199 KERNEL_SOCKPTR(optval), *optlen);
5200 tp->bpf_chg_cc_inprogress = 0;
5204 static int sol_tcp_sockopt(struct sock *sk, int optname,
5205 char *optval, int *optlen,
5208 if (sk->sk_protocol != IPPROTO_TCP)
5218 case TCP_WINDOW_CLAMP:
5219 case TCP_THIN_LINEAR_TIMEOUTS:
5220 case TCP_USER_TIMEOUT:
5221 case TCP_NOTSENT_LOWAT:
5223 if (*optlen != sizeof(int))
5226 case TCP_CONGESTION:
5227 return sol_tcp_sockopt_congestion(sk, optval, optlen, getopt);
5235 return bpf_sol_tcp_setsockopt(sk, optname, optval, *optlen);
5239 if (optname == TCP_SAVED_SYN) {
5240 struct tcp_sock *tp = tcp_sk(sk);
5242 if (!tp->saved_syn ||
5243 *optlen > tcp_saved_syn_len(tp->saved_syn))
5245 memcpy(optval, tp->saved_syn->data, *optlen);
5246 /* It cannot free tp->saved_syn here because it
5247 * does not know if the user space still needs it.
5252 return do_tcp_getsockopt(sk, SOL_TCP, optname,
5253 KERNEL_SOCKPTR(optval),
5254 KERNEL_SOCKPTR(optlen));
5257 return do_tcp_setsockopt(sk, SOL_TCP, optname,
5258 KERNEL_SOCKPTR(optval), *optlen);
5261 static int sol_ip_sockopt(struct sock *sk, int optname,
5262 char *optval, int *optlen,
5265 if (sk->sk_family != AF_INET)
5270 if (*optlen != sizeof(int))
5278 return do_ip_getsockopt(sk, SOL_IP, optname,
5279 KERNEL_SOCKPTR(optval),
5280 KERNEL_SOCKPTR(optlen));
5282 return do_ip_setsockopt(sk, SOL_IP, optname,
5283 KERNEL_SOCKPTR(optval), *optlen);
5286 static int sol_ipv6_sockopt(struct sock *sk, int optname,
5287 char *optval, int *optlen,
5290 if (sk->sk_family != AF_INET6)
5295 case IPV6_AUTOFLOWLABEL:
5296 if (*optlen != sizeof(int))
5304 return ipv6_bpf_stub->ipv6_getsockopt(sk, SOL_IPV6, optname,
5305 KERNEL_SOCKPTR(optval),
5306 KERNEL_SOCKPTR(optlen));
5308 return ipv6_bpf_stub->ipv6_setsockopt(sk, SOL_IPV6, optname,
5309 KERNEL_SOCKPTR(optval), *optlen);
5312 static int __bpf_setsockopt(struct sock *sk, int level, int optname,
5313 char *optval, int optlen)
5315 if (!sk_fullsock(sk))
5318 if (level == SOL_SOCKET)
5319 return sol_socket_sockopt(sk, optname, optval, &optlen, false);
5320 else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5321 return sol_ip_sockopt(sk, optname, optval, &optlen, false);
5322 else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5323 return sol_ipv6_sockopt(sk, optname, optval, &optlen, false);
5324 else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5325 return sol_tcp_sockopt(sk, optname, optval, &optlen, false);
5330 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
5331 char *optval, int optlen)
5333 if (sk_fullsock(sk))
5334 sock_owned_by_me(sk);
5335 return __bpf_setsockopt(sk, level, optname, optval, optlen);
5338 static int __bpf_getsockopt(struct sock *sk, int level, int optname,
5339 char *optval, int optlen)
5341 int err, saved_optlen = optlen;
5343 if (!sk_fullsock(sk)) {
5348 if (level == SOL_SOCKET)
5349 err = sol_socket_sockopt(sk, optname, optval, &optlen, true);
5350 else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5351 err = sol_tcp_sockopt(sk, optname, optval, &optlen, true);
5352 else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5353 err = sol_ip_sockopt(sk, optname, optval, &optlen, true);
5354 else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5355 err = sol_ipv6_sockopt(sk, optname, optval, &optlen, true);
5362 if (optlen < saved_optlen)
5363 memset(optval + optlen, 0, saved_optlen - optlen);
5367 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
5368 char *optval, int optlen)
5370 if (sk_fullsock(sk))
5371 sock_owned_by_me(sk);
5372 return __bpf_getsockopt(sk, level, optname, optval, optlen);
5375 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level,
5376 int, optname, char *, optval, int, optlen)
5378 return _bpf_setsockopt(sk, level, optname, optval, optlen);
5381 const struct bpf_func_proto bpf_sk_setsockopt_proto = {
5382 .func = bpf_sk_setsockopt,
5384 .ret_type = RET_INTEGER,
5385 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5386 .arg2_type = ARG_ANYTHING,
5387 .arg3_type = ARG_ANYTHING,
5388 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5389 .arg5_type = ARG_CONST_SIZE,
5392 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level,
5393 int, optname, char *, optval, int, optlen)
5395 return _bpf_getsockopt(sk, level, optname, optval, optlen);
5398 const struct bpf_func_proto bpf_sk_getsockopt_proto = {
5399 .func = bpf_sk_getsockopt,
5401 .ret_type = RET_INTEGER,
5402 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5403 .arg2_type = ARG_ANYTHING,
5404 .arg3_type = ARG_ANYTHING,
5405 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5406 .arg5_type = ARG_CONST_SIZE,
5409 BPF_CALL_5(bpf_unlocked_sk_setsockopt, struct sock *, sk, int, level,
5410 int, optname, char *, optval, int, optlen)
5412 return __bpf_setsockopt(sk, level, optname, optval, optlen);
5415 const struct bpf_func_proto bpf_unlocked_sk_setsockopt_proto = {
5416 .func = bpf_unlocked_sk_setsockopt,
5418 .ret_type = RET_INTEGER,
5419 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5420 .arg2_type = ARG_ANYTHING,
5421 .arg3_type = ARG_ANYTHING,
5422 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5423 .arg5_type = ARG_CONST_SIZE,
5426 BPF_CALL_5(bpf_unlocked_sk_getsockopt, struct sock *, sk, int, level,
5427 int, optname, char *, optval, int, optlen)
5429 return __bpf_getsockopt(sk, level, optname, optval, optlen);
5432 const struct bpf_func_proto bpf_unlocked_sk_getsockopt_proto = {
5433 .func = bpf_unlocked_sk_getsockopt,
5435 .ret_type = RET_INTEGER,
5436 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5437 .arg2_type = ARG_ANYTHING,
5438 .arg3_type = ARG_ANYTHING,
5439 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5440 .arg5_type = ARG_CONST_SIZE,
5443 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5444 int, level, int, optname, char *, optval, int, optlen)
5446 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5449 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5450 .func = bpf_sock_addr_setsockopt,
5452 .ret_type = RET_INTEGER,
5453 .arg1_type = ARG_PTR_TO_CTX,
5454 .arg2_type = ARG_ANYTHING,
5455 .arg3_type = ARG_ANYTHING,
5456 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5457 .arg5_type = ARG_CONST_SIZE,
5460 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5461 int, level, int, optname, char *, optval, int, optlen)
5463 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5466 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5467 .func = bpf_sock_addr_getsockopt,
5469 .ret_type = RET_INTEGER,
5470 .arg1_type = ARG_PTR_TO_CTX,
5471 .arg2_type = ARG_ANYTHING,
5472 .arg3_type = ARG_ANYTHING,
5473 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5474 .arg5_type = ARG_CONST_SIZE,
5477 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5478 int, level, int, optname, char *, optval, int, optlen)
5480 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5483 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5484 .func = bpf_sock_ops_setsockopt,
5486 .ret_type = RET_INTEGER,
5487 .arg1_type = ARG_PTR_TO_CTX,
5488 .arg2_type = ARG_ANYTHING,
5489 .arg3_type = ARG_ANYTHING,
5490 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5491 .arg5_type = ARG_CONST_SIZE,
5494 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5495 int optname, const u8 **start)
5497 struct sk_buff *syn_skb = bpf_sock->syn_skb;
5498 const u8 *hdr_start;
5502 /* sk is a request_sock here */
5504 if (optname == TCP_BPF_SYN) {
5505 hdr_start = syn_skb->data;
5506 ret = tcp_hdrlen(syn_skb);
5507 } else if (optname == TCP_BPF_SYN_IP) {
5508 hdr_start = skb_network_header(syn_skb);
5509 ret = skb_network_header_len(syn_skb) +
5510 tcp_hdrlen(syn_skb);
5512 /* optname == TCP_BPF_SYN_MAC */
5513 hdr_start = skb_mac_header(syn_skb);
5514 ret = skb_mac_header_len(syn_skb) +
5515 skb_network_header_len(syn_skb) +
5516 tcp_hdrlen(syn_skb);
5519 struct sock *sk = bpf_sock->sk;
5520 struct saved_syn *saved_syn;
5522 if (sk->sk_state == TCP_NEW_SYN_RECV)
5523 /* synack retransmit. bpf_sock->syn_skb will
5524 * not be available. It has to resort to
5525 * saved_syn (if it is saved).
5527 saved_syn = inet_reqsk(sk)->saved_syn;
5529 saved_syn = tcp_sk(sk)->saved_syn;
5534 if (optname == TCP_BPF_SYN) {
5535 hdr_start = saved_syn->data +
5536 saved_syn->mac_hdrlen +
5537 saved_syn->network_hdrlen;
5538 ret = saved_syn->tcp_hdrlen;
5539 } else if (optname == TCP_BPF_SYN_IP) {
5540 hdr_start = saved_syn->data +
5541 saved_syn->mac_hdrlen;
5542 ret = saved_syn->network_hdrlen +
5543 saved_syn->tcp_hdrlen;
5545 /* optname == TCP_BPF_SYN_MAC */
5547 /* TCP_SAVE_SYN may not have saved the mac hdr */
5548 if (!saved_syn->mac_hdrlen)
5551 hdr_start = saved_syn->data;
5552 ret = saved_syn->mac_hdrlen +
5553 saved_syn->network_hdrlen +
5554 saved_syn->tcp_hdrlen;
5562 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5563 int, level, int, optname, char *, optval, int, optlen)
5565 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5566 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5567 int ret, copy_len = 0;
5570 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5573 if (optlen < copy_len) {
5578 memcpy(optval, start, copy_len);
5581 /* Zero out unused buffer at the end */
5582 memset(optval + copy_len, 0, optlen - copy_len);
5587 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5590 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5591 .func = bpf_sock_ops_getsockopt,
5593 .ret_type = RET_INTEGER,
5594 .arg1_type = ARG_PTR_TO_CTX,
5595 .arg2_type = ARG_ANYTHING,
5596 .arg3_type = ARG_ANYTHING,
5597 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5598 .arg5_type = ARG_CONST_SIZE,
5601 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5604 struct sock *sk = bpf_sock->sk;
5605 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5607 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5610 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5612 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5615 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5616 .func = bpf_sock_ops_cb_flags_set,
5618 .ret_type = RET_INTEGER,
5619 .arg1_type = ARG_PTR_TO_CTX,
5620 .arg2_type = ARG_ANYTHING,
5623 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5624 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5626 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5630 struct sock *sk = ctx->sk;
5631 u32 flags = BIND_FROM_BPF;
5635 if (addr_len < offsetofend(struct sockaddr, sa_family))
5637 if (addr->sa_family == AF_INET) {
5638 if (addr_len < sizeof(struct sockaddr_in))
5640 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5641 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5642 return __inet_bind(sk, addr, addr_len, flags);
5643 #if IS_ENABLED(CONFIG_IPV6)
5644 } else if (addr->sa_family == AF_INET6) {
5645 if (addr_len < SIN6_LEN_RFC2133)
5647 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5648 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5649 /* ipv6_bpf_stub cannot be NULL, since it's called from
5650 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5652 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5653 #endif /* CONFIG_IPV6 */
5655 #endif /* CONFIG_INET */
5657 return -EAFNOSUPPORT;
5660 static const struct bpf_func_proto bpf_bind_proto = {
5663 .ret_type = RET_INTEGER,
5664 .arg1_type = ARG_PTR_TO_CTX,
5665 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5666 .arg3_type = ARG_CONST_SIZE,
5671 #if (IS_BUILTIN(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF)) || \
5672 (IS_MODULE(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES))
5674 struct metadata_dst __percpu *xfrm_bpf_md_dst;
5675 EXPORT_SYMBOL_GPL(xfrm_bpf_md_dst);
5679 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5680 struct bpf_xfrm_state *, to, u32, size, u64, flags)
5682 const struct sec_path *sp = skb_sec_path(skb);
5683 const struct xfrm_state *x;
5685 if (!sp || unlikely(index >= sp->len || flags))
5688 x = sp->xvec[index];
5690 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5693 to->reqid = x->props.reqid;
5694 to->spi = x->id.spi;
5695 to->family = x->props.family;
5698 if (to->family == AF_INET6) {
5699 memcpy(to->remote_ipv6, x->props.saddr.a6,
5700 sizeof(to->remote_ipv6));
5702 to->remote_ipv4 = x->props.saddr.a4;
5703 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5708 memset(to, 0, size);
5712 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5713 .func = bpf_skb_get_xfrm_state,
5715 .ret_type = RET_INTEGER,
5716 .arg1_type = ARG_PTR_TO_CTX,
5717 .arg2_type = ARG_ANYTHING,
5718 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
5719 .arg4_type = ARG_CONST_SIZE,
5720 .arg5_type = ARG_ANYTHING,
5724 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5725 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
5726 const struct neighbour *neigh,
5727 const struct net_device *dev, u32 mtu)
5729 memcpy(params->dmac, neigh->ha, ETH_ALEN);
5730 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5731 params->h_vlan_TCI = 0;
5732 params->h_vlan_proto = 0;
5734 params->mtu_result = mtu; /* union with tot_len */
5740 #if IS_ENABLED(CONFIG_INET)
5741 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5742 u32 flags, bool check_mtu)
5744 struct fib_nh_common *nhc;
5745 struct in_device *in_dev;
5746 struct neighbour *neigh;
5747 struct net_device *dev;
5748 struct fib_result res;
5753 dev = dev_get_by_index_rcu(net, params->ifindex);
5757 /* verify forwarding is enabled on this interface */
5758 in_dev = __in_dev_get_rcu(dev);
5759 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5760 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5762 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5764 fl4.flowi4_oif = params->ifindex;
5766 fl4.flowi4_iif = params->ifindex;
5769 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5770 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5771 fl4.flowi4_flags = 0;
5773 fl4.flowi4_proto = params->l4_protocol;
5774 fl4.daddr = params->ipv4_dst;
5775 fl4.saddr = params->ipv4_src;
5776 fl4.fl4_sport = params->sport;
5777 fl4.fl4_dport = params->dport;
5778 fl4.flowi4_multipath_hash = 0;
5780 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5781 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5782 struct fib_table *tb;
5784 tb = fib_get_table(net, tbid);
5786 return BPF_FIB_LKUP_RET_NOT_FWDED;
5788 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5790 fl4.flowi4_mark = 0;
5791 fl4.flowi4_secid = 0;
5792 fl4.flowi4_tun_key.tun_id = 0;
5793 fl4.flowi4_uid = sock_net_uid(net, NULL);
5795 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5799 /* map fib lookup errors to RTN_ type */
5801 return BPF_FIB_LKUP_RET_BLACKHOLE;
5802 if (err == -EHOSTUNREACH)
5803 return BPF_FIB_LKUP_RET_UNREACHABLE;
5805 return BPF_FIB_LKUP_RET_PROHIBIT;
5807 return BPF_FIB_LKUP_RET_NOT_FWDED;
5810 if (res.type != RTN_UNICAST)
5811 return BPF_FIB_LKUP_RET_NOT_FWDED;
5813 if (fib_info_num_path(res.fi) > 1)
5814 fib_select_path(net, &res, &fl4, NULL);
5817 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5818 if (params->tot_len > mtu) {
5819 params->mtu_result = mtu; /* union with tot_len */
5820 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5826 /* do not handle lwt encaps right now */
5827 if (nhc->nhc_lwtstate)
5828 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5832 params->rt_metric = res.fi->fib_priority;
5833 params->ifindex = dev->ifindex;
5835 /* xdp and cls_bpf programs are run in RCU-bh so
5836 * rcu_read_lock_bh is not needed here
5838 if (likely(nhc->nhc_gw_family != AF_INET6)) {
5839 if (nhc->nhc_gw_family)
5840 params->ipv4_dst = nhc->nhc_gw.ipv4;
5842 neigh = __ipv4_neigh_lookup_noref(dev,
5843 (__force u32)params->ipv4_dst);
5845 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5847 params->family = AF_INET6;
5848 *dst = nhc->nhc_gw.ipv6;
5849 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5853 return BPF_FIB_LKUP_RET_NO_NEIGH;
5855 return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5859 #if IS_ENABLED(CONFIG_IPV6)
5860 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5861 u32 flags, bool check_mtu)
5863 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5864 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5865 struct fib6_result res = {};
5866 struct neighbour *neigh;
5867 struct net_device *dev;
5868 struct inet6_dev *idev;
5874 /* link local addresses are never forwarded */
5875 if (rt6_need_strict(dst) || rt6_need_strict(src))
5876 return BPF_FIB_LKUP_RET_NOT_FWDED;
5878 dev = dev_get_by_index_rcu(net, params->ifindex);
5882 idev = __in6_dev_get_safely(dev);
5883 if (unlikely(!idev || !idev->cnf.forwarding))
5884 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5886 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5888 oif = fl6.flowi6_oif = params->ifindex;
5890 oif = fl6.flowi6_iif = params->ifindex;
5892 strict = RT6_LOOKUP_F_HAS_SADDR;
5894 fl6.flowlabel = params->flowinfo;
5895 fl6.flowi6_scope = 0;
5896 fl6.flowi6_flags = 0;
5899 fl6.flowi6_proto = params->l4_protocol;
5902 fl6.fl6_sport = params->sport;
5903 fl6.fl6_dport = params->dport;
5905 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5906 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5907 struct fib6_table *tb;
5909 tb = ipv6_stub->fib6_get_table(net, tbid);
5911 return BPF_FIB_LKUP_RET_NOT_FWDED;
5913 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5916 fl6.flowi6_mark = 0;
5917 fl6.flowi6_secid = 0;
5918 fl6.flowi6_tun_key.tun_id = 0;
5919 fl6.flowi6_uid = sock_net_uid(net, NULL);
5921 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5924 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5925 res.f6i == net->ipv6.fib6_null_entry))
5926 return BPF_FIB_LKUP_RET_NOT_FWDED;
5928 switch (res.fib6_type) {
5929 /* only unicast is forwarded */
5933 return BPF_FIB_LKUP_RET_BLACKHOLE;
5934 case RTN_UNREACHABLE:
5935 return BPF_FIB_LKUP_RET_UNREACHABLE;
5937 return BPF_FIB_LKUP_RET_PROHIBIT;
5939 return BPF_FIB_LKUP_RET_NOT_FWDED;
5942 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5943 fl6.flowi6_oif != 0, NULL, strict);
5946 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5947 if (params->tot_len > mtu) {
5948 params->mtu_result = mtu; /* union with tot_len */
5949 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5953 if (res.nh->fib_nh_lws)
5954 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5956 if (res.nh->fib_nh_gw_family)
5957 *dst = res.nh->fib_nh_gw6;
5959 dev = res.nh->fib_nh_dev;
5960 params->rt_metric = res.f6i->fib6_metric;
5961 params->ifindex = dev->ifindex;
5963 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
5966 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5968 return BPF_FIB_LKUP_RET_NO_NEIGH;
5970 return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5974 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
5975 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5977 if (plen < sizeof(*params))
5980 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5983 switch (params->family) {
5984 #if IS_ENABLED(CONFIG_INET)
5986 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
5989 #if IS_ENABLED(CONFIG_IPV6)
5991 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
5995 return -EAFNOSUPPORT;
5998 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
5999 .func = bpf_xdp_fib_lookup,
6001 .ret_type = RET_INTEGER,
6002 .arg1_type = ARG_PTR_TO_CTX,
6003 .arg2_type = ARG_PTR_TO_MEM,
6004 .arg3_type = ARG_CONST_SIZE,
6005 .arg4_type = ARG_ANYTHING,
6008 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
6009 struct bpf_fib_lookup *, params, int, plen, u32, flags)
6011 struct net *net = dev_net(skb->dev);
6012 int rc = -EAFNOSUPPORT;
6013 bool check_mtu = false;
6015 if (plen < sizeof(*params))
6018 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
6021 if (params->tot_len)
6024 switch (params->family) {
6025 #if IS_ENABLED(CONFIG_INET)
6027 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
6030 #if IS_ENABLED(CONFIG_IPV6)
6032 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
6037 if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
6038 struct net_device *dev;
6040 /* When tot_len isn't provided by user, check skb
6041 * against MTU of FIB lookup resulting net_device
6043 dev = dev_get_by_index_rcu(net, params->ifindex);
6044 if (!is_skb_forwardable(dev, skb))
6045 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
6047 params->mtu_result = dev->mtu; /* union with tot_len */
6053 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
6054 .func = bpf_skb_fib_lookup,
6056 .ret_type = RET_INTEGER,
6057 .arg1_type = ARG_PTR_TO_CTX,
6058 .arg2_type = ARG_PTR_TO_MEM,
6059 .arg3_type = ARG_CONST_SIZE,
6060 .arg4_type = ARG_ANYTHING,
6063 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
6066 struct net *netns = dev_net(dev_curr);
6068 /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
6072 return dev_get_by_index_rcu(netns, ifindex);
6075 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
6076 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6078 int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6079 struct net_device *dev = skb->dev;
6080 int skb_len, dev_len;
6083 if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
6086 if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
6089 dev = __dev_via_ifindex(dev, ifindex);
6093 mtu = READ_ONCE(dev->mtu);
6095 dev_len = mtu + dev->hard_header_len;
6097 /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6098 skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
6100 skb_len += len_diff; /* minus result pass check */
6101 if (skb_len <= dev_len) {
6102 ret = BPF_MTU_CHK_RET_SUCCESS;
6105 /* At this point, skb->len exceed MTU, but as it include length of all
6106 * segments, it can still be below MTU. The SKB can possibly get
6107 * re-segmented in transmit path (see validate_xmit_skb). Thus, user
6108 * must choose if segs are to be MTU checked.
6110 if (skb_is_gso(skb)) {
6111 ret = BPF_MTU_CHK_RET_SUCCESS;
6113 if (flags & BPF_MTU_CHK_SEGS &&
6114 !skb_gso_validate_network_len(skb, mtu))
6115 ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
6118 /* BPF verifier guarantees valid pointer */
6124 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
6125 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6127 struct net_device *dev = xdp->rxq->dev;
6128 int xdp_len = xdp->data_end - xdp->data;
6129 int ret = BPF_MTU_CHK_RET_SUCCESS;
6132 /* XDP variant doesn't support multi-buffer segment check (yet) */
6133 if (unlikely(flags))
6136 dev = __dev_via_ifindex(dev, ifindex);
6140 mtu = READ_ONCE(dev->mtu);
6142 /* Add L2-header as dev MTU is L3 size */
6143 dev_len = mtu + dev->hard_header_len;
6145 /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6147 xdp_len = *mtu_len + dev->hard_header_len;
6149 xdp_len += len_diff; /* minus result pass check */
6150 if (xdp_len > dev_len)
6151 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6153 /* BPF verifier guarantees valid pointer */
6159 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
6160 .func = bpf_skb_check_mtu,
6162 .ret_type = RET_INTEGER,
6163 .arg1_type = ARG_PTR_TO_CTX,
6164 .arg2_type = ARG_ANYTHING,
6165 .arg3_type = ARG_PTR_TO_INT,
6166 .arg4_type = ARG_ANYTHING,
6167 .arg5_type = ARG_ANYTHING,
6170 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
6171 .func = bpf_xdp_check_mtu,
6173 .ret_type = RET_INTEGER,
6174 .arg1_type = ARG_PTR_TO_CTX,
6175 .arg2_type = ARG_ANYTHING,
6176 .arg3_type = ARG_PTR_TO_INT,
6177 .arg4_type = ARG_ANYTHING,
6178 .arg5_type = ARG_ANYTHING,
6181 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6182 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
6185 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
6187 if (!seg6_validate_srh(srh, len, false))
6191 case BPF_LWT_ENCAP_SEG6_INLINE:
6192 if (skb->protocol != htons(ETH_P_IPV6))
6195 err = seg6_do_srh_inline(skb, srh);
6197 case BPF_LWT_ENCAP_SEG6:
6198 skb_reset_inner_headers(skb);
6199 skb->encapsulation = 1;
6200 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
6206 bpf_compute_data_pointers(skb);
6210 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
6212 return seg6_lookup_nexthop(skb, NULL, 0);
6214 #endif /* CONFIG_IPV6_SEG6_BPF */
6216 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6217 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
6220 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
6224 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
6228 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6229 case BPF_LWT_ENCAP_SEG6:
6230 case BPF_LWT_ENCAP_SEG6_INLINE:
6231 return bpf_push_seg6_encap(skb, type, hdr, len);
6233 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6234 case BPF_LWT_ENCAP_IP:
6235 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
6242 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
6243 void *, hdr, u32, len)
6246 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6247 case BPF_LWT_ENCAP_IP:
6248 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
6255 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
6256 .func = bpf_lwt_in_push_encap,
6258 .ret_type = RET_INTEGER,
6259 .arg1_type = ARG_PTR_TO_CTX,
6260 .arg2_type = ARG_ANYTHING,
6261 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6262 .arg4_type = ARG_CONST_SIZE
6265 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
6266 .func = bpf_lwt_xmit_push_encap,
6268 .ret_type = RET_INTEGER,
6269 .arg1_type = ARG_PTR_TO_CTX,
6270 .arg2_type = ARG_ANYTHING,
6271 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6272 .arg4_type = ARG_CONST_SIZE
6275 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6276 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
6277 const void *, from, u32, len)
6279 struct seg6_bpf_srh_state *srh_state =
6280 this_cpu_ptr(&seg6_bpf_srh_states);
6281 struct ipv6_sr_hdr *srh = srh_state->srh;
6282 void *srh_tlvs, *srh_end, *ptr;
6288 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
6289 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
6291 ptr = skb->data + offset;
6292 if (ptr >= srh_tlvs && ptr + len <= srh_end)
6293 srh_state->valid = false;
6294 else if (ptr < (void *)&srh->flags ||
6295 ptr + len > (void *)&srh->segments)
6298 if (unlikely(bpf_try_make_writable(skb, offset + len)))
6300 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6302 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6304 memcpy(skb->data + offset, from, len);
6308 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
6309 .func = bpf_lwt_seg6_store_bytes,
6311 .ret_type = RET_INTEGER,
6312 .arg1_type = ARG_PTR_TO_CTX,
6313 .arg2_type = ARG_ANYTHING,
6314 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6315 .arg4_type = ARG_CONST_SIZE
6318 static void bpf_update_srh_state(struct sk_buff *skb)
6320 struct seg6_bpf_srh_state *srh_state =
6321 this_cpu_ptr(&seg6_bpf_srh_states);
6324 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
6325 srh_state->srh = NULL;
6327 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6328 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
6329 srh_state->valid = true;
6333 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
6334 u32, action, void *, param, u32, param_len)
6336 struct seg6_bpf_srh_state *srh_state =
6337 this_cpu_ptr(&seg6_bpf_srh_states);
6342 case SEG6_LOCAL_ACTION_END_X:
6343 if (!seg6_bpf_has_valid_srh(skb))
6345 if (param_len != sizeof(struct in6_addr))
6347 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
6348 case SEG6_LOCAL_ACTION_END_T:
6349 if (!seg6_bpf_has_valid_srh(skb))
6351 if (param_len != sizeof(int))
6353 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6354 case SEG6_LOCAL_ACTION_END_DT6:
6355 if (!seg6_bpf_has_valid_srh(skb))
6357 if (param_len != sizeof(int))
6360 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
6362 if (!pskb_pull(skb, hdroff))
6365 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
6366 skb_reset_network_header(skb);
6367 skb_reset_transport_header(skb);
6368 skb->encapsulation = 0;
6370 bpf_compute_data_pointers(skb);
6371 bpf_update_srh_state(skb);
6372 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6373 case SEG6_LOCAL_ACTION_END_B6:
6374 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6376 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
6379 bpf_update_srh_state(skb);
6382 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
6383 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6385 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
6388 bpf_update_srh_state(skb);
6396 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
6397 .func = bpf_lwt_seg6_action,
6399 .ret_type = RET_INTEGER,
6400 .arg1_type = ARG_PTR_TO_CTX,
6401 .arg2_type = ARG_ANYTHING,
6402 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6403 .arg4_type = ARG_CONST_SIZE
6406 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
6409 struct seg6_bpf_srh_state *srh_state =
6410 this_cpu_ptr(&seg6_bpf_srh_states);
6411 struct ipv6_sr_hdr *srh = srh_state->srh;
6412 void *srh_end, *srh_tlvs, *ptr;
6413 struct ipv6hdr *hdr;
6417 if (unlikely(srh == NULL))
6420 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
6421 ((srh->first_segment + 1) << 4));
6422 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
6424 ptr = skb->data + offset;
6426 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
6428 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
6432 ret = skb_cow_head(skb, len);
6433 if (unlikely(ret < 0))
6436 ret = bpf_skb_net_hdr_push(skb, offset, len);
6438 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6441 bpf_compute_data_pointers(skb);
6442 if (unlikely(ret < 0))
6445 hdr = (struct ipv6hdr *)skb->data;
6446 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6448 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6450 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6451 srh_state->hdrlen += len;
6452 srh_state->valid = false;
6456 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6457 .func = bpf_lwt_seg6_adjust_srh,
6459 .ret_type = RET_INTEGER,
6460 .arg1_type = ARG_PTR_TO_CTX,
6461 .arg2_type = ARG_ANYTHING,
6462 .arg3_type = ARG_ANYTHING,
6464 #endif /* CONFIG_IPV6_SEG6_BPF */
6467 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6468 int dif, int sdif, u8 family, u8 proto)
6470 struct inet_hashinfo *hinfo = net->ipv4.tcp_death_row.hashinfo;
6471 bool refcounted = false;
6472 struct sock *sk = NULL;
6474 if (family == AF_INET) {
6475 __be32 src4 = tuple->ipv4.saddr;
6476 __be32 dst4 = tuple->ipv4.daddr;
6478 if (proto == IPPROTO_TCP)
6479 sk = __inet_lookup(net, hinfo, NULL, 0,
6480 src4, tuple->ipv4.sport,
6481 dst4, tuple->ipv4.dport,
6482 dif, sdif, &refcounted);
6484 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6485 dst4, tuple->ipv4.dport,
6486 dif, sdif, net->ipv4.udp_table, NULL);
6487 #if IS_ENABLED(CONFIG_IPV6)
6489 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6490 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6492 if (proto == IPPROTO_TCP)
6493 sk = __inet6_lookup(net, hinfo, NULL, 0,
6494 src6, tuple->ipv6.sport,
6495 dst6, ntohs(tuple->ipv6.dport),
6496 dif, sdif, &refcounted);
6497 else if (likely(ipv6_bpf_stub))
6498 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6499 src6, tuple->ipv6.sport,
6500 dst6, tuple->ipv6.dport,
6502 net->ipv4.udp_table, NULL);
6506 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6507 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6513 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6514 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6516 static struct sock *
6517 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6518 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6521 struct sock *sk = NULL;
6526 if (len == sizeof(tuple->ipv4))
6528 else if (len == sizeof(tuple->ipv6))
6533 if (unlikely(flags || !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6536 if (family == AF_INET)
6537 sdif = inet_sdif(skb);
6539 sdif = inet6_sdif(skb);
6541 if ((s32)netns_id < 0) {
6543 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6545 net = get_net_ns_by_id(caller_net, netns_id);
6548 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6556 static struct sock *
6557 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6558 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6561 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6562 ifindex, proto, netns_id, flags);
6565 struct sock *sk2 = sk_to_full_sk(sk);
6567 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6568 * sock refcnt is decremented to prevent a request_sock leak.
6570 if (!sk_fullsock(sk2))
6574 /* Ensure there is no need to bump sk2 refcnt */
6575 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6576 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6586 static struct sock *
6587 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6588 u8 proto, u64 netns_id, u64 flags)
6590 struct net *caller_net;
6594 caller_net = dev_net(skb->dev);
6595 ifindex = skb->dev->ifindex;
6597 caller_net = sock_net(skb->sk);
6601 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6605 static struct sock *
6606 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6607 u8 proto, u64 netns_id, u64 flags)
6609 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6613 struct sock *sk2 = sk_to_full_sk(sk);
6615 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6616 * sock refcnt is decremented to prevent a request_sock leak.
6618 if (!sk_fullsock(sk2))
6622 /* Ensure there is no need to bump sk2 refcnt */
6623 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6624 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6634 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6635 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6637 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6641 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6642 .func = bpf_skc_lookup_tcp,
6645 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6646 .arg1_type = ARG_PTR_TO_CTX,
6647 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6648 .arg3_type = ARG_CONST_SIZE,
6649 .arg4_type = ARG_ANYTHING,
6650 .arg5_type = ARG_ANYTHING,
6653 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6654 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6656 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6660 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6661 .func = bpf_sk_lookup_tcp,
6664 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6665 .arg1_type = ARG_PTR_TO_CTX,
6666 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6667 .arg3_type = ARG_CONST_SIZE,
6668 .arg4_type = ARG_ANYTHING,
6669 .arg5_type = ARG_ANYTHING,
6672 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6673 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6675 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6679 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6680 .func = bpf_sk_lookup_udp,
6683 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6684 .arg1_type = ARG_PTR_TO_CTX,
6685 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6686 .arg3_type = ARG_CONST_SIZE,
6687 .arg4_type = ARG_ANYTHING,
6688 .arg5_type = ARG_ANYTHING,
6691 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6693 if (sk && sk_is_refcounted(sk))
6698 static const struct bpf_func_proto bpf_sk_release_proto = {
6699 .func = bpf_sk_release,
6701 .ret_type = RET_INTEGER,
6702 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON | OBJ_RELEASE,
6705 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6706 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6708 struct net *caller_net = dev_net(ctx->rxq->dev);
6709 int ifindex = ctx->rxq->dev->ifindex;
6711 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6712 ifindex, IPPROTO_UDP, netns_id,
6716 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6717 .func = bpf_xdp_sk_lookup_udp,
6720 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6721 .arg1_type = ARG_PTR_TO_CTX,
6722 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6723 .arg3_type = ARG_CONST_SIZE,
6724 .arg4_type = ARG_ANYTHING,
6725 .arg5_type = ARG_ANYTHING,
6728 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6729 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6731 struct net *caller_net = dev_net(ctx->rxq->dev);
6732 int ifindex = ctx->rxq->dev->ifindex;
6734 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6735 ifindex, IPPROTO_TCP, netns_id,
6739 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6740 .func = bpf_xdp_skc_lookup_tcp,
6743 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6744 .arg1_type = ARG_PTR_TO_CTX,
6745 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6746 .arg3_type = ARG_CONST_SIZE,
6747 .arg4_type = ARG_ANYTHING,
6748 .arg5_type = ARG_ANYTHING,
6751 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6752 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6754 struct net *caller_net = dev_net(ctx->rxq->dev);
6755 int ifindex = ctx->rxq->dev->ifindex;
6757 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6758 ifindex, IPPROTO_TCP, netns_id,
6762 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6763 .func = bpf_xdp_sk_lookup_tcp,
6766 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6767 .arg1_type = ARG_PTR_TO_CTX,
6768 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6769 .arg3_type = ARG_CONST_SIZE,
6770 .arg4_type = ARG_ANYTHING,
6771 .arg5_type = ARG_ANYTHING,
6774 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6775 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6777 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6778 sock_net(ctx->sk), 0,
6779 IPPROTO_TCP, netns_id, flags);
6782 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6783 .func = bpf_sock_addr_skc_lookup_tcp,
6785 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6786 .arg1_type = ARG_PTR_TO_CTX,
6787 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6788 .arg3_type = ARG_CONST_SIZE,
6789 .arg4_type = ARG_ANYTHING,
6790 .arg5_type = ARG_ANYTHING,
6793 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6794 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6796 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6797 sock_net(ctx->sk), 0, IPPROTO_TCP,
6801 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6802 .func = bpf_sock_addr_sk_lookup_tcp,
6804 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6805 .arg1_type = ARG_PTR_TO_CTX,
6806 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6807 .arg3_type = ARG_CONST_SIZE,
6808 .arg4_type = ARG_ANYTHING,
6809 .arg5_type = ARG_ANYTHING,
6812 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6813 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6815 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6816 sock_net(ctx->sk), 0, IPPROTO_UDP,
6820 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6821 .func = bpf_sock_addr_sk_lookup_udp,
6823 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6824 .arg1_type = ARG_PTR_TO_CTX,
6825 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6826 .arg3_type = ARG_CONST_SIZE,
6827 .arg4_type = ARG_ANYTHING,
6828 .arg5_type = ARG_ANYTHING,
6831 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6832 struct bpf_insn_access_aux *info)
6834 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6838 if (off % size != 0)
6842 case offsetof(struct bpf_tcp_sock, bytes_received):
6843 case offsetof(struct bpf_tcp_sock, bytes_acked):
6844 return size == sizeof(__u64);
6846 return size == sizeof(__u32);
6850 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6851 const struct bpf_insn *si,
6852 struct bpf_insn *insn_buf,
6853 struct bpf_prog *prog, u32 *target_size)
6855 struct bpf_insn *insn = insn_buf;
6857 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \
6859 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \
6860 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6861 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6862 si->dst_reg, si->src_reg, \
6863 offsetof(struct tcp_sock, FIELD)); \
6866 #define BPF_INET_SOCK_GET_COMMON(FIELD) \
6868 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \
6870 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6871 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6872 struct inet_connection_sock, \
6874 si->dst_reg, si->src_reg, \
6876 struct inet_connection_sock, \
6881 case offsetof(struct bpf_tcp_sock, rtt_min):
6882 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
6883 sizeof(struct minmax));
6884 BUILD_BUG_ON(sizeof(struct minmax) <
6885 sizeof(struct minmax_sample));
6887 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6888 offsetof(struct tcp_sock, rtt_min) +
6889 offsetof(struct minmax_sample, v));
6891 case offsetof(struct bpf_tcp_sock, snd_cwnd):
6892 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
6894 case offsetof(struct bpf_tcp_sock, srtt_us):
6895 BPF_TCP_SOCK_GET_COMMON(srtt_us);
6897 case offsetof(struct bpf_tcp_sock, snd_ssthresh):
6898 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
6900 case offsetof(struct bpf_tcp_sock, rcv_nxt):
6901 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
6903 case offsetof(struct bpf_tcp_sock, snd_nxt):
6904 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
6906 case offsetof(struct bpf_tcp_sock, snd_una):
6907 BPF_TCP_SOCK_GET_COMMON(snd_una);
6909 case offsetof(struct bpf_tcp_sock, mss_cache):
6910 BPF_TCP_SOCK_GET_COMMON(mss_cache);
6912 case offsetof(struct bpf_tcp_sock, ecn_flags):
6913 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
6915 case offsetof(struct bpf_tcp_sock, rate_delivered):
6916 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
6918 case offsetof(struct bpf_tcp_sock, rate_interval_us):
6919 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
6921 case offsetof(struct bpf_tcp_sock, packets_out):
6922 BPF_TCP_SOCK_GET_COMMON(packets_out);
6924 case offsetof(struct bpf_tcp_sock, retrans_out):
6925 BPF_TCP_SOCK_GET_COMMON(retrans_out);
6927 case offsetof(struct bpf_tcp_sock, total_retrans):
6928 BPF_TCP_SOCK_GET_COMMON(total_retrans);
6930 case offsetof(struct bpf_tcp_sock, segs_in):
6931 BPF_TCP_SOCK_GET_COMMON(segs_in);
6933 case offsetof(struct bpf_tcp_sock, data_segs_in):
6934 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
6936 case offsetof(struct bpf_tcp_sock, segs_out):
6937 BPF_TCP_SOCK_GET_COMMON(segs_out);
6939 case offsetof(struct bpf_tcp_sock, data_segs_out):
6940 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
6942 case offsetof(struct bpf_tcp_sock, lost_out):
6943 BPF_TCP_SOCK_GET_COMMON(lost_out);
6945 case offsetof(struct bpf_tcp_sock, sacked_out):
6946 BPF_TCP_SOCK_GET_COMMON(sacked_out);
6948 case offsetof(struct bpf_tcp_sock, bytes_received):
6949 BPF_TCP_SOCK_GET_COMMON(bytes_received);
6951 case offsetof(struct bpf_tcp_sock, bytes_acked):
6952 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
6954 case offsetof(struct bpf_tcp_sock, dsack_dups):
6955 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
6957 case offsetof(struct bpf_tcp_sock, delivered):
6958 BPF_TCP_SOCK_GET_COMMON(delivered);
6960 case offsetof(struct bpf_tcp_sock, delivered_ce):
6961 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
6963 case offsetof(struct bpf_tcp_sock, icsk_retransmits):
6964 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
6968 return insn - insn_buf;
6971 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
6973 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
6974 return (unsigned long)sk;
6976 return (unsigned long)NULL;
6979 const struct bpf_func_proto bpf_tcp_sock_proto = {
6980 .func = bpf_tcp_sock,
6982 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
6983 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6986 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
6988 sk = sk_to_full_sk(sk);
6990 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
6991 return (unsigned long)sk;
6993 return (unsigned long)NULL;
6996 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
6997 .func = bpf_get_listener_sock,
6999 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
7000 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
7003 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
7005 unsigned int iphdr_len;
7007 switch (skb_protocol(skb, true)) {
7008 case cpu_to_be16(ETH_P_IP):
7009 iphdr_len = sizeof(struct iphdr);
7011 case cpu_to_be16(ETH_P_IPV6):
7012 iphdr_len = sizeof(struct ipv6hdr);
7018 if (skb_headlen(skb) < iphdr_len)
7021 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
7024 return INET_ECN_set_ce(skb);
7027 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7028 struct bpf_insn_access_aux *info)
7030 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
7033 if (off % size != 0)
7038 return size == sizeof(__u32);
7042 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
7043 const struct bpf_insn *si,
7044 struct bpf_insn *insn_buf,
7045 struct bpf_prog *prog, u32 *target_size)
7047 struct bpf_insn *insn = insn_buf;
7049 #define BPF_XDP_SOCK_GET(FIELD) \
7051 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \
7052 sizeof_field(struct bpf_xdp_sock, FIELD)); \
7053 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
7054 si->dst_reg, si->src_reg, \
7055 offsetof(struct xdp_sock, FIELD)); \
7059 case offsetof(struct bpf_xdp_sock, queue_id):
7060 BPF_XDP_SOCK_GET(queue_id);
7064 return insn - insn_buf;
7067 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
7068 .func = bpf_skb_ecn_set_ce,
7070 .ret_type = RET_INTEGER,
7071 .arg1_type = ARG_PTR_TO_CTX,
7074 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7075 struct tcphdr *, th, u32, th_len)
7077 #ifdef CONFIG_SYN_COOKIES
7081 if (unlikely(!sk || th_len < sizeof(*th)))
7084 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
7085 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7088 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7091 if (!th->ack || th->rst || th->syn)
7094 if (unlikely(iph_len < sizeof(struct iphdr)))
7097 if (tcp_synq_no_recent_overflow(sk))
7100 cookie = ntohl(th->ack_seq) - 1;
7102 /* Both struct iphdr and struct ipv6hdr have the version field at the
7103 * same offset so we can cast to the shorter header (struct iphdr).
7105 switch (((struct iphdr *)iph)->version) {
7107 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7110 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
7113 #if IS_BUILTIN(CONFIG_IPV6)
7115 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7118 if (sk->sk_family != AF_INET6)
7121 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
7123 #endif /* CONFIG_IPV6 */
7126 return -EPROTONOSUPPORT;
7138 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
7139 .func = bpf_tcp_check_syncookie,
7142 .ret_type = RET_INTEGER,
7143 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7144 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7145 .arg3_type = ARG_CONST_SIZE,
7146 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7147 .arg5_type = ARG_CONST_SIZE,
7150 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7151 struct tcphdr *, th, u32, th_len)
7153 #ifdef CONFIG_SYN_COOKIES
7157 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
7160 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7163 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7166 if (!th->syn || th->ack || th->fin || th->rst)
7169 if (unlikely(iph_len < sizeof(struct iphdr)))
7172 /* Both struct iphdr and struct ipv6hdr have the version field at the
7173 * same offset so we can cast to the shorter header (struct iphdr).
7175 switch (((struct iphdr *)iph)->version) {
7177 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7180 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
7183 #if IS_BUILTIN(CONFIG_IPV6)
7185 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7188 if (sk->sk_family != AF_INET6)
7191 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
7193 #endif /* CONFIG_IPV6 */
7196 return -EPROTONOSUPPORT;
7201 return cookie | ((u64)mss << 32);
7204 #endif /* CONFIG_SYN_COOKIES */
7207 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
7208 .func = bpf_tcp_gen_syncookie,
7209 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */
7211 .ret_type = RET_INTEGER,
7212 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7213 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7214 .arg3_type = ARG_CONST_SIZE,
7215 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7216 .arg5_type = ARG_CONST_SIZE,
7219 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
7221 if (!sk || flags != 0)
7223 if (!skb_at_tc_ingress(skb))
7225 if (unlikely(dev_net(skb->dev) != sock_net(sk)))
7226 return -ENETUNREACH;
7227 if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
7228 return -ESOCKTNOSUPPORT;
7229 if (sk_is_refcounted(sk) &&
7230 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
7235 skb->destructor = sock_pfree;
7240 static const struct bpf_func_proto bpf_sk_assign_proto = {
7241 .func = bpf_sk_assign,
7243 .ret_type = RET_INTEGER,
7244 .arg1_type = ARG_PTR_TO_CTX,
7245 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7246 .arg3_type = ARG_ANYTHING,
7249 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
7250 u8 search_kind, const u8 *magic,
7251 u8 magic_len, bool *eol)
7257 while (op < opend) {
7260 if (kind == TCPOPT_EOL) {
7262 return ERR_PTR(-ENOMSG);
7263 } else if (kind == TCPOPT_NOP) {
7268 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
7269 /* Something is wrong in the received header.
7270 * Follow the TCP stack's tcp_parse_options()
7271 * and just bail here.
7273 return ERR_PTR(-EFAULT);
7276 if (search_kind == kind) {
7280 if (magic_len > kind_len - 2)
7281 return ERR_PTR(-ENOMSG);
7283 if (!memcmp(&op[2], magic, magic_len))
7290 return ERR_PTR(-ENOMSG);
7293 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7294 void *, search_res, u32, len, u64, flags)
7296 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
7297 const u8 *op, *opend, *magic, *search = search_res;
7298 u8 search_kind, search_len, copy_len, magic_len;
7301 /* 2 byte is the minimal option len except TCPOPT_NOP and
7302 * TCPOPT_EOL which are useless for the bpf prog to learn
7303 * and this helper disallow loading them also.
7305 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
7308 search_kind = search[0];
7309 search_len = search[1];
7311 if (search_len > len || search_kind == TCPOPT_NOP ||
7312 search_kind == TCPOPT_EOL)
7315 if (search_kind == TCPOPT_EXP || search_kind == 253) {
7316 /* 16 or 32 bit magic. +2 for kind and kind length */
7317 if (search_len != 4 && search_len != 6)
7320 magic_len = search_len - 2;
7329 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
7334 op += sizeof(struct tcphdr);
7336 if (!bpf_sock->skb ||
7337 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7338 /* This bpf_sock->op cannot call this helper */
7341 opend = bpf_sock->skb_data_end;
7342 op = bpf_sock->skb->data + sizeof(struct tcphdr);
7345 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
7352 if (copy_len > len) {
7357 memcpy(search_res, op, copy_len);
7361 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
7362 .func = bpf_sock_ops_load_hdr_opt,
7364 .ret_type = RET_INTEGER,
7365 .arg1_type = ARG_PTR_TO_CTX,
7366 .arg2_type = ARG_PTR_TO_MEM,
7367 .arg3_type = ARG_CONST_SIZE,
7368 .arg4_type = ARG_ANYTHING,
7371 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7372 const void *, from, u32, len, u64, flags)
7374 u8 new_kind, new_kind_len, magic_len = 0, *opend;
7375 const u8 *op, *new_op, *magic = NULL;
7376 struct sk_buff *skb;
7379 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
7382 if (len < 2 || flags)
7386 new_kind = new_op[0];
7387 new_kind_len = new_op[1];
7389 if (new_kind_len > len || new_kind == TCPOPT_NOP ||
7390 new_kind == TCPOPT_EOL)
7393 if (new_kind_len > bpf_sock->remaining_opt_len)
7396 /* 253 is another experimental kind */
7397 if (new_kind == TCPOPT_EXP || new_kind == 253) {
7398 if (new_kind_len < 4)
7400 /* Match for the 2 byte magic also.
7401 * RFC 6994: the magic could be 2 or 4 bytes.
7402 * Hence, matching by 2 byte only is on the
7403 * conservative side but it is the right
7404 * thing to do for the 'search-for-duplication'
7411 /* Check for duplication */
7412 skb = bpf_sock->skb;
7413 op = skb->data + sizeof(struct tcphdr);
7414 opend = bpf_sock->skb_data_end;
7416 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
7421 if (PTR_ERR(op) != -ENOMSG)
7425 /* The option has been ended. Treat it as no more
7426 * header option can be written.
7430 /* No duplication found. Store the header option. */
7431 memcpy(opend, from, new_kind_len);
7433 bpf_sock->remaining_opt_len -= new_kind_len;
7434 bpf_sock->skb_data_end += new_kind_len;
7439 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
7440 .func = bpf_sock_ops_store_hdr_opt,
7442 .ret_type = RET_INTEGER,
7443 .arg1_type = ARG_PTR_TO_CTX,
7444 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7445 .arg3_type = ARG_CONST_SIZE,
7446 .arg4_type = ARG_ANYTHING,
7449 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7450 u32, len, u64, flags)
7452 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7455 if (flags || len < 2)
7458 if (len > bpf_sock->remaining_opt_len)
7461 bpf_sock->remaining_opt_len -= len;
7466 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7467 .func = bpf_sock_ops_reserve_hdr_opt,
7469 .ret_type = RET_INTEGER,
7470 .arg1_type = ARG_PTR_TO_CTX,
7471 .arg2_type = ARG_ANYTHING,
7472 .arg3_type = ARG_ANYTHING,
7475 BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb,
7476 u64, tstamp, u32, tstamp_type)
7478 /* skb_clear_delivery_time() is done for inet protocol */
7479 if (skb->protocol != htons(ETH_P_IP) &&
7480 skb->protocol != htons(ETH_P_IPV6))
7483 switch (tstamp_type) {
7484 case BPF_SKB_TSTAMP_DELIVERY_MONO:
7487 skb->tstamp = tstamp;
7488 skb->mono_delivery_time = 1;
7490 case BPF_SKB_TSTAMP_UNSPEC:
7494 skb->mono_delivery_time = 0;
7503 static const struct bpf_func_proto bpf_skb_set_tstamp_proto = {
7504 .func = bpf_skb_set_tstamp,
7506 .ret_type = RET_INTEGER,
7507 .arg1_type = ARG_PTR_TO_CTX,
7508 .arg2_type = ARG_ANYTHING,
7509 .arg3_type = ARG_ANYTHING,
7512 #ifdef CONFIG_SYN_COOKIES
7513 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv4, struct iphdr *, iph,
7514 struct tcphdr *, th, u32, th_len)
7519 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7522 mss = tcp_parse_mss_option(th, 0) ?: TCP_MSS_DEFAULT;
7523 cookie = __cookie_v4_init_sequence(iph, th, &mss);
7525 return cookie | ((u64)mss << 32);
7528 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv4_proto = {
7529 .func = bpf_tcp_raw_gen_syncookie_ipv4,
7530 .gpl_only = true, /* __cookie_v4_init_sequence() is GPL */
7532 .ret_type = RET_INTEGER,
7533 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7534 .arg1_size = sizeof(struct iphdr),
7535 .arg2_type = ARG_PTR_TO_MEM,
7536 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7539 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv6, struct ipv6hdr *, iph,
7540 struct tcphdr *, th, u32, th_len)
7542 #if IS_BUILTIN(CONFIG_IPV6)
7543 const u16 mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
7544 sizeof(struct ipv6hdr);
7548 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7551 mss = tcp_parse_mss_option(th, 0) ?: mss_clamp;
7552 cookie = __cookie_v6_init_sequence(iph, th, &mss);
7554 return cookie | ((u64)mss << 32);
7556 return -EPROTONOSUPPORT;
7560 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv6_proto = {
7561 .func = bpf_tcp_raw_gen_syncookie_ipv6,
7562 .gpl_only = true, /* __cookie_v6_init_sequence() is GPL */
7564 .ret_type = RET_INTEGER,
7565 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7566 .arg1_size = sizeof(struct ipv6hdr),
7567 .arg2_type = ARG_PTR_TO_MEM,
7568 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7571 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv4, struct iphdr *, iph,
7572 struct tcphdr *, th)
7574 u32 cookie = ntohl(th->ack_seq) - 1;
7576 if (__cookie_v4_check(iph, th, cookie) > 0)
7582 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv4_proto = {
7583 .func = bpf_tcp_raw_check_syncookie_ipv4,
7584 .gpl_only = true, /* __cookie_v4_check is GPL */
7586 .ret_type = RET_INTEGER,
7587 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7588 .arg1_size = sizeof(struct iphdr),
7589 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7590 .arg2_size = sizeof(struct tcphdr),
7593 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv6, struct ipv6hdr *, iph,
7594 struct tcphdr *, th)
7596 #if IS_BUILTIN(CONFIG_IPV6)
7597 u32 cookie = ntohl(th->ack_seq) - 1;
7599 if (__cookie_v6_check(iph, th, cookie) > 0)
7604 return -EPROTONOSUPPORT;
7608 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv6_proto = {
7609 .func = bpf_tcp_raw_check_syncookie_ipv6,
7610 .gpl_only = true, /* __cookie_v6_check is GPL */
7612 .ret_type = RET_INTEGER,
7613 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7614 .arg1_size = sizeof(struct ipv6hdr),
7615 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7616 .arg2_size = sizeof(struct tcphdr),
7618 #endif /* CONFIG_SYN_COOKIES */
7620 #endif /* CONFIG_INET */
7622 bool bpf_helper_changes_pkt_data(void *func)
7624 if (func == bpf_skb_vlan_push ||
7625 func == bpf_skb_vlan_pop ||
7626 func == bpf_skb_store_bytes ||
7627 func == bpf_skb_change_proto ||
7628 func == bpf_skb_change_head ||
7629 func == sk_skb_change_head ||
7630 func == bpf_skb_change_tail ||
7631 func == sk_skb_change_tail ||
7632 func == bpf_skb_adjust_room ||
7633 func == sk_skb_adjust_room ||
7634 func == bpf_skb_pull_data ||
7635 func == sk_skb_pull_data ||
7636 func == bpf_clone_redirect ||
7637 func == bpf_l3_csum_replace ||
7638 func == bpf_l4_csum_replace ||
7639 func == bpf_xdp_adjust_head ||
7640 func == bpf_xdp_adjust_meta ||
7641 func == bpf_msg_pull_data ||
7642 func == bpf_msg_push_data ||
7643 func == bpf_msg_pop_data ||
7644 func == bpf_xdp_adjust_tail ||
7645 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7646 func == bpf_lwt_seg6_store_bytes ||
7647 func == bpf_lwt_seg6_adjust_srh ||
7648 func == bpf_lwt_seg6_action ||
7651 func == bpf_sock_ops_store_hdr_opt ||
7653 func == bpf_lwt_in_push_encap ||
7654 func == bpf_lwt_xmit_push_encap)
7660 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7661 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7663 static const struct bpf_func_proto *
7664 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7666 const struct bpf_func_proto *func_proto;
7668 func_proto = cgroup_common_func_proto(func_id, prog);
7672 func_proto = cgroup_current_func_proto(func_id, prog);
7677 case BPF_FUNC_get_socket_cookie:
7678 return &bpf_get_socket_cookie_sock_proto;
7679 case BPF_FUNC_get_netns_cookie:
7680 return &bpf_get_netns_cookie_sock_proto;
7681 case BPF_FUNC_perf_event_output:
7682 return &bpf_event_output_data_proto;
7683 case BPF_FUNC_sk_storage_get:
7684 return &bpf_sk_storage_get_cg_sock_proto;
7685 case BPF_FUNC_ktime_get_coarse_ns:
7686 return &bpf_ktime_get_coarse_ns_proto;
7688 return bpf_base_func_proto(func_id);
7692 static const struct bpf_func_proto *
7693 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7695 const struct bpf_func_proto *func_proto;
7697 func_proto = cgroup_common_func_proto(func_id, prog);
7701 func_proto = cgroup_current_func_proto(func_id, prog);
7707 switch (prog->expected_attach_type) {
7708 case BPF_CGROUP_INET4_CONNECT:
7709 case BPF_CGROUP_INET6_CONNECT:
7710 return &bpf_bind_proto;
7714 case BPF_FUNC_get_socket_cookie:
7715 return &bpf_get_socket_cookie_sock_addr_proto;
7716 case BPF_FUNC_get_netns_cookie:
7717 return &bpf_get_netns_cookie_sock_addr_proto;
7718 case BPF_FUNC_perf_event_output:
7719 return &bpf_event_output_data_proto;
7721 case BPF_FUNC_sk_lookup_tcp:
7722 return &bpf_sock_addr_sk_lookup_tcp_proto;
7723 case BPF_FUNC_sk_lookup_udp:
7724 return &bpf_sock_addr_sk_lookup_udp_proto;
7725 case BPF_FUNC_sk_release:
7726 return &bpf_sk_release_proto;
7727 case BPF_FUNC_skc_lookup_tcp:
7728 return &bpf_sock_addr_skc_lookup_tcp_proto;
7729 #endif /* CONFIG_INET */
7730 case BPF_FUNC_sk_storage_get:
7731 return &bpf_sk_storage_get_proto;
7732 case BPF_FUNC_sk_storage_delete:
7733 return &bpf_sk_storage_delete_proto;
7734 case BPF_FUNC_setsockopt:
7735 switch (prog->expected_attach_type) {
7736 case BPF_CGROUP_INET4_BIND:
7737 case BPF_CGROUP_INET6_BIND:
7738 case BPF_CGROUP_INET4_CONNECT:
7739 case BPF_CGROUP_INET6_CONNECT:
7740 case BPF_CGROUP_UDP4_RECVMSG:
7741 case BPF_CGROUP_UDP6_RECVMSG:
7742 case BPF_CGROUP_UDP4_SENDMSG:
7743 case BPF_CGROUP_UDP6_SENDMSG:
7744 case BPF_CGROUP_INET4_GETPEERNAME:
7745 case BPF_CGROUP_INET6_GETPEERNAME:
7746 case BPF_CGROUP_INET4_GETSOCKNAME:
7747 case BPF_CGROUP_INET6_GETSOCKNAME:
7748 return &bpf_sock_addr_setsockopt_proto;
7752 case BPF_FUNC_getsockopt:
7753 switch (prog->expected_attach_type) {
7754 case BPF_CGROUP_INET4_BIND:
7755 case BPF_CGROUP_INET6_BIND:
7756 case BPF_CGROUP_INET4_CONNECT:
7757 case BPF_CGROUP_INET6_CONNECT:
7758 case BPF_CGROUP_UDP4_RECVMSG:
7759 case BPF_CGROUP_UDP6_RECVMSG:
7760 case BPF_CGROUP_UDP4_SENDMSG:
7761 case BPF_CGROUP_UDP6_SENDMSG:
7762 case BPF_CGROUP_INET4_GETPEERNAME:
7763 case BPF_CGROUP_INET6_GETPEERNAME:
7764 case BPF_CGROUP_INET4_GETSOCKNAME:
7765 case BPF_CGROUP_INET6_GETSOCKNAME:
7766 return &bpf_sock_addr_getsockopt_proto;
7771 return bpf_sk_base_func_proto(func_id);
7775 static const struct bpf_func_proto *
7776 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7779 case BPF_FUNC_skb_load_bytes:
7780 return &bpf_skb_load_bytes_proto;
7781 case BPF_FUNC_skb_load_bytes_relative:
7782 return &bpf_skb_load_bytes_relative_proto;
7783 case BPF_FUNC_get_socket_cookie:
7784 return &bpf_get_socket_cookie_proto;
7785 case BPF_FUNC_get_socket_uid:
7786 return &bpf_get_socket_uid_proto;
7787 case BPF_FUNC_perf_event_output:
7788 return &bpf_skb_event_output_proto;
7790 return bpf_sk_base_func_proto(func_id);
7794 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7795 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7797 static const struct bpf_func_proto *
7798 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7800 const struct bpf_func_proto *func_proto;
7802 func_proto = cgroup_common_func_proto(func_id, prog);
7807 case BPF_FUNC_sk_fullsock:
7808 return &bpf_sk_fullsock_proto;
7809 case BPF_FUNC_sk_storage_get:
7810 return &bpf_sk_storage_get_proto;
7811 case BPF_FUNC_sk_storage_delete:
7812 return &bpf_sk_storage_delete_proto;
7813 case BPF_FUNC_perf_event_output:
7814 return &bpf_skb_event_output_proto;
7815 #ifdef CONFIG_SOCK_CGROUP_DATA
7816 case BPF_FUNC_skb_cgroup_id:
7817 return &bpf_skb_cgroup_id_proto;
7818 case BPF_FUNC_skb_ancestor_cgroup_id:
7819 return &bpf_skb_ancestor_cgroup_id_proto;
7820 case BPF_FUNC_sk_cgroup_id:
7821 return &bpf_sk_cgroup_id_proto;
7822 case BPF_FUNC_sk_ancestor_cgroup_id:
7823 return &bpf_sk_ancestor_cgroup_id_proto;
7826 case BPF_FUNC_sk_lookup_tcp:
7827 return &bpf_sk_lookup_tcp_proto;
7828 case BPF_FUNC_sk_lookup_udp:
7829 return &bpf_sk_lookup_udp_proto;
7830 case BPF_FUNC_sk_release:
7831 return &bpf_sk_release_proto;
7832 case BPF_FUNC_skc_lookup_tcp:
7833 return &bpf_skc_lookup_tcp_proto;
7834 case BPF_FUNC_tcp_sock:
7835 return &bpf_tcp_sock_proto;
7836 case BPF_FUNC_get_listener_sock:
7837 return &bpf_get_listener_sock_proto;
7838 case BPF_FUNC_skb_ecn_set_ce:
7839 return &bpf_skb_ecn_set_ce_proto;
7842 return sk_filter_func_proto(func_id, prog);
7846 static const struct bpf_func_proto *
7847 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7850 case BPF_FUNC_skb_store_bytes:
7851 return &bpf_skb_store_bytes_proto;
7852 case BPF_FUNC_skb_load_bytes:
7853 return &bpf_skb_load_bytes_proto;
7854 case BPF_FUNC_skb_load_bytes_relative:
7855 return &bpf_skb_load_bytes_relative_proto;
7856 case BPF_FUNC_skb_pull_data:
7857 return &bpf_skb_pull_data_proto;
7858 case BPF_FUNC_csum_diff:
7859 return &bpf_csum_diff_proto;
7860 case BPF_FUNC_csum_update:
7861 return &bpf_csum_update_proto;
7862 case BPF_FUNC_csum_level:
7863 return &bpf_csum_level_proto;
7864 case BPF_FUNC_l3_csum_replace:
7865 return &bpf_l3_csum_replace_proto;
7866 case BPF_FUNC_l4_csum_replace:
7867 return &bpf_l4_csum_replace_proto;
7868 case BPF_FUNC_clone_redirect:
7869 return &bpf_clone_redirect_proto;
7870 case BPF_FUNC_get_cgroup_classid:
7871 return &bpf_get_cgroup_classid_proto;
7872 case BPF_FUNC_skb_vlan_push:
7873 return &bpf_skb_vlan_push_proto;
7874 case BPF_FUNC_skb_vlan_pop:
7875 return &bpf_skb_vlan_pop_proto;
7876 case BPF_FUNC_skb_change_proto:
7877 return &bpf_skb_change_proto_proto;
7878 case BPF_FUNC_skb_change_type:
7879 return &bpf_skb_change_type_proto;
7880 case BPF_FUNC_skb_adjust_room:
7881 return &bpf_skb_adjust_room_proto;
7882 case BPF_FUNC_skb_change_tail:
7883 return &bpf_skb_change_tail_proto;
7884 case BPF_FUNC_skb_change_head:
7885 return &bpf_skb_change_head_proto;
7886 case BPF_FUNC_skb_get_tunnel_key:
7887 return &bpf_skb_get_tunnel_key_proto;
7888 case BPF_FUNC_skb_set_tunnel_key:
7889 return bpf_get_skb_set_tunnel_proto(func_id);
7890 case BPF_FUNC_skb_get_tunnel_opt:
7891 return &bpf_skb_get_tunnel_opt_proto;
7892 case BPF_FUNC_skb_set_tunnel_opt:
7893 return bpf_get_skb_set_tunnel_proto(func_id);
7894 case BPF_FUNC_redirect:
7895 return &bpf_redirect_proto;
7896 case BPF_FUNC_redirect_neigh:
7897 return &bpf_redirect_neigh_proto;
7898 case BPF_FUNC_redirect_peer:
7899 return &bpf_redirect_peer_proto;
7900 case BPF_FUNC_get_route_realm:
7901 return &bpf_get_route_realm_proto;
7902 case BPF_FUNC_get_hash_recalc:
7903 return &bpf_get_hash_recalc_proto;
7904 case BPF_FUNC_set_hash_invalid:
7905 return &bpf_set_hash_invalid_proto;
7906 case BPF_FUNC_set_hash:
7907 return &bpf_set_hash_proto;
7908 case BPF_FUNC_perf_event_output:
7909 return &bpf_skb_event_output_proto;
7910 case BPF_FUNC_get_smp_processor_id:
7911 return &bpf_get_smp_processor_id_proto;
7912 case BPF_FUNC_skb_under_cgroup:
7913 return &bpf_skb_under_cgroup_proto;
7914 case BPF_FUNC_get_socket_cookie:
7915 return &bpf_get_socket_cookie_proto;
7916 case BPF_FUNC_get_socket_uid:
7917 return &bpf_get_socket_uid_proto;
7918 case BPF_FUNC_fib_lookup:
7919 return &bpf_skb_fib_lookup_proto;
7920 case BPF_FUNC_check_mtu:
7921 return &bpf_skb_check_mtu_proto;
7922 case BPF_FUNC_sk_fullsock:
7923 return &bpf_sk_fullsock_proto;
7924 case BPF_FUNC_sk_storage_get:
7925 return &bpf_sk_storage_get_proto;
7926 case BPF_FUNC_sk_storage_delete:
7927 return &bpf_sk_storage_delete_proto;
7929 case BPF_FUNC_skb_get_xfrm_state:
7930 return &bpf_skb_get_xfrm_state_proto;
7932 #ifdef CONFIG_CGROUP_NET_CLASSID
7933 case BPF_FUNC_skb_cgroup_classid:
7934 return &bpf_skb_cgroup_classid_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;
7943 case BPF_FUNC_sk_lookup_tcp:
7944 return &bpf_sk_lookup_tcp_proto;
7945 case BPF_FUNC_sk_lookup_udp:
7946 return &bpf_sk_lookup_udp_proto;
7947 case BPF_FUNC_sk_release:
7948 return &bpf_sk_release_proto;
7949 case BPF_FUNC_tcp_sock:
7950 return &bpf_tcp_sock_proto;
7951 case BPF_FUNC_get_listener_sock:
7952 return &bpf_get_listener_sock_proto;
7953 case BPF_FUNC_skc_lookup_tcp:
7954 return &bpf_skc_lookup_tcp_proto;
7955 case BPF_FUNC_tcp_check_syncookie:
7956 return &bpf_tcp_check_syncookie_proto;
7957 case BPF_FUNC_skb_ecn_set_ce:
7958 return &bpf_skb_ecn_set_ce_proto;
7959 case BPF_FUNC_tcp_gen_syncookie:
7960 return &bpf_tcp_gen_syncookie_proto;
7961 case BPF_FUNC_sk_assign:
7962 return &bpf_sk_assign_proto;
7963 case BPF_FUNC_skb_set_tstamp:
7964 return &bpf_skb_set_tstamp_proto;
7965 #ifdef CONFIG_SYN_COOKIES
7966 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
7967 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
7968 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
7969 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
7970 case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
7971 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
7972 case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
7973 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
7977 return bpf_sk_base_func_proto(func_id);
7981 static const struct bpf_func_proto *
7982 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7985 case BPF_FUNC_perf_event_output:
7986 return &bpf_xdp_event_output_proto;
7987 case BPF_FUNC_get_smp_processor_id:
7988 return &bpf_get_smp_processor_id_proto;
7989 case BPF_FUNC_csum_diff:
7990 return &bpf_csum_diff_proto;
7991 case BPF_FUNC_xdp_adjust_head:
7992 return &bpf_xdp_adjust_head_proto;
7993 case BPF_FUNC_xdp_adjust_meta:
7994 return &bpf_xdp_adjust_meta_proto;
7995 case BPF_FUNC_redirect:
7996 return &bpf_xdp_redirect_proto;
7997 case BPF_FUNC_redirect_map:
7998 return &bpf_xdp_redirect_map_proto;
7999 case BPF_FUNC_xdp_adjust_tail:
8000 return &bpf_xdp_adjust_tail_proto;
8001 case BPF_FUNC_xdp_get_buff_len:
8002 return &bpf_xdp_get_buff_len_proto;
8003 case BPF_FUNC_xdp_load_bytes:
8004 return &bpf_xdp_load_bytes_proto;
8005 case BPF_FUNC_xdp_store_bytes:
8006 return &bpf_xdp_store_bytes_proto;
8007 case BPF_FUNC_fib_lookup:
8008 return &bpf_xdp_fib_lookup_proto;
8009 case BPF_FUNC_check_mtu:
8010 return &bpf_xdp_check_mtu_proto;
8012 case BPF_FUNC_sk_lookup_udp:
8013 return &bpf_xdp_sk_lookup_udp_proto;
8014 case BPF_FUNC_sk_lookup_tcp:
8015 return &bpf_xdp_sk_lookup_tcp_proto;
8016 case BPF_FUNC_sk_release:
8017 return &bpf_sk_release_proto;
8018 case BPF_FUNC_skc_lookup_tcp:
8019 return &bpf_xdp_skc_lookup_tcp_proto;
8020 case BPF_FUNC_tcp_check_syncookie:
8021 return &bpf_tcp_check_syncookie_proto;
8022 case BPF_FUNC_tcp_gen_syncookie:
8023 return &bpf_tcp_gen_syncookie_proto;
8024 #ifdef CONFIG_SYN_COOKIES
8025 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8026 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8027 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8028 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8029 case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8030 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8031 case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8032 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8036 return bpf_sk_base_func_proto(func_id);
8039 #if IS_MODULE(CONFIG_NF_CONNTRACK) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES)
8040 /* The nf_conn___init type is used in the NF_CONNTRACK kfuncs. The
8041 * kfuncs are defined in two different modules, and we want to be able
8042 * to use them interchangably with the same BTF type ID. Because modules
8043 * can't de-duplicate BTF IDs between each other, we need the type to be
8044 * referenced in the vmlinux BTF or the verifier will get confused about
8045 * the different types. So we add this dummy type reference which will
8046 * be included in vmlinux BTF, allowing both modules to refer to the
8049 BTF_TYPE_EMIT(struct nf_conn___init);
8053 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
8054 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
8056 static const struct bpf_func_proto *
8057 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8059 const struct bpf_func_proto *func_proto;
8061 func_proto = cgroup_common_func_proto(func_id, prog);
8066 case BPF_FUNC_setsockopt:
8067 return &bpf_sock_ops_setsockopt_proto;
8068 case BPF_FUNC_getsockopt:
8069 return &bpf_sock_ops_getsockopt_proto;
8070 case BPF_FUNC_sock_ops_cb_flags_set:
8071 return &bpf_sock_ops_cb_flags_set_proto;
8072 case BPF_FUNC_sock_map_update:
8073 return &bpf_sock_map_update_proto;
8074 case BPF_FUNC_sock_hash_update:
8075 return &bpf_sock_hash_update_proto;
8076 case BPF_FUNC_get_socket_cookie:
8077 return &bpf_get_socket_cookie_sock_ops_proto;
8078 case BPF_FUNC_perf_event_output:
8079 return &bpf_event_output_data_proto;
8080 case BPF_FUNC_sk_storage_get:
8081 return &bpf_sk_storage_get_proto;
8082 case BPF_FUNC_sk_storage_delete:
8083 return &bpf_sk_storage_delete_proto;
8084 case BPF_FUNC_get_netns_cookie:
8085 return &bpf_get_netns_cookie_sock_ops_proto;
8087 case BPF_FUNC_load_hdr_opt:
8088 return &bpf_sock_ops_load_hdr_opt_proto;
8089 case BPF_FUNC_store_hdr_opt:
8090 return &bpf_sock_ops_store_hdr_opt_proto;
8091 case BPF_FUNC_reserve_hdr_opt:
8092 return &bpf_sock_ops_reserve_hdr_opt_proto;
8093 case BPF_FUNC_tcp_sock:
8094 return &bpf_tcp_sock_proto;
8095 #endif /* CONFIG_INET */
8097 return bpf_sk_base_func_proto(func_id);
8101 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
8102 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
8104 static const struct bpf_func_proto *
8105 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8108 case BPF_FUNC_msg_redirect_map:
8109 return &bpf_msg_redirect_map_proto;
8110 case BPF_FUNC_msg_redirect_hash:
8111 return &bpf_msg_redirect_hash_proto;
8112 case BPF_FUNC_msg_apply_bytes:
8113 return &bpf_msg_apply_bytes_proto;
8114 case BPF_FUNC_msg_cork_bytes:
8115 return &bpf_msg_cork_bytes_proto;
8116 case BPF_FUNC_msg_pull_data:
8117 return &bpf_msg_pull_data_proto;
8118 case BPF_FUNC_msg_push_data:
8119 return &bpf_msg_push_data_proto;
8120 case BPF_FUNC_msg_pop_data:
8121 return &bpf_msg_pop_data_proto;
8122 case BPF_FUNC_perf_event_output:
8123 return &bpf_event_output_data_proto;
8124 case BPF_FUNC_get_current_uid_gid:
8125 return &bpf_get_current_uid_gid_proto;
8126 case BPF_FUNC_get_current_pid_tgid:
8127 return &bpf_get_current_pid_tgid_proto;
8128 case BPF_FUNC_sk_storage_get:
8129 return &bpf_sk_storage_get_proto;
8130 case BPF_FUNC_sk_storage_delete:
8131 return &bpf_sk_storage_delete_proto;
8132 case BPF_FUNC_get_netns_cookie:
8133 return &bpf_get_netns_cookie_sk_msg_proto;
8134 #ifdef CONFIG_CGROUPS
8135 case BPF_FUNC_get_current_cgroup_id:
8136 return &bpf_get_current_cgroup_id_proto;
8137 case BPF_FUNC_get_current_ancestor_cgroup_id:
8138 return &bpf_get_current_ancestor_cgroup_id_proto;
8140 #ifdef CONFIG_CGROUP_NET_CLASSID
8141 case BPF_FUNC_get_cgroup_classid:
8142 return &bpf_get_cgroup_classid_curr_proto;
8145 return bpf_sk_base_func_proto(func_id);
8149 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
8150 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
8152 static const struct bpf_func_proto *
8153 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8156 case BPF_FUNC_skb_store_bytes:
8157 return &bpf_skb_store_bytes_proto;
8158 case BPF_FUNC_skb_load_bytes:
8159 return &bpf_skb_load_bytes_proto;
8160 case BPF_FUNC_skb_pull_data:
8161 return &sk_skb_pull_data_proto;
8162 case BPF_FUNC_skb_change_tail:
8163 return &sk_skb_change_tail_proto;
8164 case BPF_FUNC_skb_change_head:
8165 return &sk_skb_change_head_proto;
8166 case BPF_FUNC_skb_adjust_room:
8167 return &sk_skb_adjust_room_proto;
8168 case BPF_FUNC_get_socket_cookie:
8169 return &bpf_get_socket_cookie_proto;
8170 case BPF_FUNC_get_socket_uid:
8171 return &bpf_get_socket_uid_proto;
8172 case BPF_FUNC_sk_redirect_map:
8173 return &bpf_sk_redirect_map_proto;
8174 case BPF_FUNC_sk_redirect_hash:
8175 return &bpf_sk_redirect_hash_proto;
8176 case BPF_FUNC_perf_event_output:
8177 return &bpf_skb_event_output_proto;
8179 case BPF_FUNC_sk_lookup_tcp:
8180 return &bpf_sk_lookup_tcp_proto;
8181 case BPF_FUNC_sk_lookup_udp:
8182 return &bpf_sk_lookup_udp_proto;
8183 case BPF_FUNC_sk_release:
8184 return &bpf_sk_release_proto;
8185 case BPF_FUNC_skc_lookup_tcp:
8186 return &bpf_skc_lookup_tcp_proto;
8189 return bpf_sk_base_func_proto(func_id);
8193 static const struct bpf_func_proto *
8194 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8197 case BPF_FUNC_skb_load_bytes:
8198 return &bpf_flow_dissector_load_bytes_proto;
8200 return bpf_sk_base_func_proto(func_id);
8204 static const struct bpf_func_proto *
8205 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8208 case BPF_FUNC_skb_load_bytes:
8209 return &bpf_skb_load_bytes_proto;
8210 case BPF_FUNC_skb_pull_data:
8211 return &bpf_skb_pull_data_proto;
8212 case BPF_FUNC_csum_diff:
8213 return &bpf_csum_diff_proto;
8214 case BPF_FUNC_get_cgroup_classid:
8215 return &bpf_get_cgroup_classid_proto;
8216 case BPF_FUNC_get_route_realm:
8217 return &bpf_get_route_realm_proto;
8218 case BPF_FUNC_get_hash_recalc:
8219 return &bpf_get_hash_recalc_proto;
8220 case BPF_FUNC_perf_event_output:
8221 return &bpf_skb_event_output_proto;
8222 case BPF_FUNC_get_smp_processor_id:
8223 return &bpf_get_smp_processor_id_proto;
8224 case BPF_FUNC_skb_under_cgroup:
8225 return &bpf_skb_under_cgroup_proto;
8227 return bpf_sk_base_func_proto(func_id);
8231 static const struct bpf_func_proto *
8232 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8235 case BPF_FUNC_lwt_push_encap:
8236 return &bpf_lwt_in_push_encap_proto;
8238 return lwt_out_func_proto(func_id, prog);
8242 static const struct bpf_func_proto *
8243 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8246 case BPF_FUNC_skb_get_tunnel_key:
8247 return &bpf_skb_get_tunnel_key_proto;
8248 case BPF_FUNC_skb_set_tunnel_key:
8249 return bpf_get_skb_set_tunnel_proto(func_id);
8250 case BPF_FUNC_skb_get_tunnel_opt:
8251 return &bpf_skb_get_tunnel_opt_proto;
8252 case BPF_FUNC_skb_set_tunnel_opt:
8253 return bpf_get_skb_set_tunnel_proto(func_id);
8254 case BPF_FUNC_redirect:
8255 return &bpf_redirect_proto;
8256 case BPF_FUNC_clone_redirect:
8257 return &bpf_clone_redirect_proto;
8258 case BPF_FUNC_skb_change_tail:
8259 return &bpf_skb_change_tail_proto;
8260 case BPF_FUNC_skb_change_head:
8261 return &bpf_skb_change_head_proto;
8262 case BPF_FUNC_skb_store_bytes:
8263 return &bpf_skb_store_bytes_proto;
8264 case BPF_FUNC_csum_update:
8265 return &bpf_csum_update_proto;
8266 case BPF_FUNC_csum_level:
8267 return &bpf_csum_level_proto;
8268 case BPF_FUNC_l3_csum_replace:
8269 return &bpf_l3_csum_replace_proto;
8270 case BPF_FUNC_l4_csum_replace:
8271 return &bpf_l4_csum_replace_proto;
8272 case BPF_FUNC_set_hash_invalid:
8273 return &bpf_set_hash_invalid_proto;
8274 case BPF_FUNC_lwt_push_encap:
8275 return &bpf_lwt_xmit_push_encap_proto;
8277 return lwt_out_func_proto(func_id, prog);
8281 static const struct bpf_func_proto *
8282 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8285 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
8286 case BPF_FUNC_lwt_seg6_store_bytes:
8287 return &bpf_lwt_seg6_store_bytes_proto;
8288 case BPF_FUNC_lwt_seg6_action:
8289 return &bpf_lwt_seg6_action_proto;
8290 case BPF_FUNC_lwt_seg6_adjust_srh:
8291 return &bpf_lwt_seg6_adjust_srh_proto;
8294 return lwt_out_func_proto(func_id, prog);
8298 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
8299 const struct bpf_prog *prog,
8300 struct bpf_insn_access_aux *info)
8302 const int size_default = sizeof(__u32);
8304 if (off < 0 || off >= sizeof(struct __sk_buff))
8307 /* The verifier guarantees that size > 0. */
8308 if (off % size != 0)
8312 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8313 if (off + size > offsetofend(struct __sk_buff, cb[4]))
8316 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
8317 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
8318 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
8319 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
8320 case bpf_ctx_range(struct __sk_buff, data):
8321 case bpf_ctx_range(struct __sk_buff, data_meta):
8322 case bpf_ctx_range(struct __sk_buff, data_end):
8323 if (size != size_default)
8326 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8328 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8329 if (type == BPF_WRITE || size != sizeof(__u64))
8332 case bpf_ctx_range(struct __sk_buff, tstamp):
8333 if (size != sizeof(__u64))
8336 case offsetof(struct __sk_buff, sk):
8337 if (type == BPF_WRITE || size != sizeof(__u64))
8339 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
8341 case offsetof(struct __sk_buff, tstamp_type):
8343 case offsetofend(struct __sk_buff, tstamp_type) ... offsetof(struct __sk_buff, hwtstamp) - 1:
8344 /* Explicitly prohibit access to padding in __sk_buff. */
8347 /* Only narrow read access allowed for now. */
8348 if (type == BPF_WRITE) {
8349 if (size != size_default)
8352 bpf_ctx_record_field_size(info, size_default);
8353 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8361 static bool sk_filter_is_valid_access(int off, int size,
8362 enum bpf_access_type type,
8363 const struct bpf_prog *prog,
8364 struct bpf_insn_access_aux *info)
8367 case bpf_ctx_range(struct __sk_buff, tc_classid):
8368 case bpf_ctx_range(struct __sk_buff, data):
8369 case bpf_ctx_range(struct __sk_buff, data_meta):
8370 case bpf_ctx_range(struct __sk_buff, data_end):
8371 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8372 case bpf_ctx_range(struct __sk_buff, tstamp):
8373 case bpf_ctx_range(struct __sk_buff, wire_len):
8374 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8378 if (type == BPF_WRITE) {
8380 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8387 return bpf_skb_is_valid_access(off, size, type, prog, info);
8390 static bool cg_skb_is_valid_access(int off, int size,
8391 enum bpf_access_type type,
8392 const struct bpf_prog *prog,
8393 struct bpf_insn_access_aux *info)
8396 case bpf_ctx_range(struct __sk_buff, tc_classid):
8397 case bpf_ctx_range(struct __sk_buff, data_meta):
8398 case bpf_ctx_range(struct __sk_buff, wire_len):
8400 case bpf_ctx_range(struct __sk_buff, data):
8401 case bpf_ctx_range(struct __sk_buff, data_end):
8407 if (type == BPF_WRITE) {
8409 case bpf_ctx_range(struct __sk_buff, mark):
8410 case bpf_ctx_range(struct __sk_buff, priority):
8411 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8413 case bpf_ctx_range(struct __sk_buff, tstamp):
8423 case bpf_ctx_range(struct __sk_buff, data):
8424 info->reg_type = PTR_TO_PACKET;
8426 case bpf_ctx_range(struct __sk_buff, data_end):
8427 info->reg_type = PTR_TO_PACKET_END;
8431 return bpf_skb_is_valid_access(off, size, type, prog, info);
8434 static bool lwt_is_valid_access(int off, int size,
8435 enum bpf_access_type type,
8436 const struct bpf_prog *prog,
8437 struct bpf_insn_access_aux *info)
8440 case bpf_ctx_range(struct __sk_buff, tc_classid):
8441 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8442 case bpf_ctx_range(struct __sk_buff, data_meta):
8443 case bpf_ctx_range(struct __sk_buff, tstamp):
8444 case bpf_ctx_range(struct __sk_buff, wire_len):
8445 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8449 if (type == BPF_WRITE) {
8451 case bpf_ctx_range(struct __sk_buff, mark):
8452 case bpf_ctx_range(struct __sk_buff, priority):
8453 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8461 case bpf_ctx_range(struct __sk_buff, data):
8462 info->reg_type = PTR_TO_PACKET;
8464 case bpf_ctx_range(struct __sk_buff, data_end):
8465 info->reg_type = PTR_TO_PACKET_END;
8469 return bpf_skb_is_valid_access(off, size, type, prog, info);
8472 /* Attach type specific accesses */
8473 static bool __sock_filter_check_attach_type(int off,
8474 enum bpf_access_type access_type,
8475 enum bpf_attach_type attach_type)
8478 case offsetof(struct bpf_sock, bound_dev_if):
8479 case offsetof(struct bpf_sock, mark):
8480 case offsetof(struct bpf_sock, priority):
8481 switch (attach_type) {
8482 case BPF_CGROUP_INET_SOCK_CREATE:
8483 case BPF_CGROUP_INET_SOCK_RELEASE:
8488 case bpf_ctx_range(struct bpf_sock, src_ip4):
8489 switch (attach_type) {
8490 case BPF_CGROUP_INET4_POST_BIND:
8495 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8496 switch (attach_type) {
8497 case BPF_CGROUP_INET6_POST_BIND:
8502 case bpf_ctx_range(struct bpf_sock, src_port):
8503 switch (attach_type) {
8504 case BPF_CGROUP_INET4_POST_BIND:
8505 case BPF_CGROUP_INET6_POST_BIND:
8512 return access_type == BPF_READ;
8517 bool bpf_sock_common_is_valid_access(int off, int size,
8518 enum bpf_access_type type,
8519 struct bpf_insn_access_aux *info)
8522 case bpf_ctx_range_till(struct bpf_sock, type, priority):
8525 return bpf_sock_is_valid_access(off, size, type, info);
8529 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
8530 struct bpf_insn_access_aux *info)
8532 const int size_default = sizeof(__u32);
8535 if (off < 0 || off >= sizeof(struct bpf_sock))
8537 if (off % size != 0)
8541 case offsetof(struct bpf_sock, state):
8542 case offsetof(struct bpf_sock, family):
8543 case offsetof(struct bpf_sock, type):
8544 case offsetof(struct bpf_sock, protocol):
8545 case offsetof(struct bpf_sock, src_port):
8546 case offsetof(struct bpf_sock, rx_queue_mapping):
8547 case bpf_ctx_range(struct bpf_sock, src_ip4):
8548 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8549 case bpf_ctx_range(struct bpf_sock, dst_ip4):
8550 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8551 bpf_ctx_record_field_size(info, size_default);
8552 return bpf_ctx_narrow_access_ok(off, size, size_default);
8553 case bpf_ctx_range(struct bpf_sock, dst_port):
8554 field_size = size == size_default ?
8555 size_default : sizeof_field(struct bpf_sock, dst_port);
8556 bpf_ctx_record_field_size(info, field_size);
8557 return bpf_ctx_narrow_access_ok(off, size, field_size);
8558 case offsetofend(struct bpf_sock, dst_port) ...
8559 offsetof(struct bpf_sock, dst_ip4) - 1:
8563 return size == size_default;
8566 static bool sock_filter_is_valid_access(int off, int size,
8567 enum bpf_access_type type,
8568 const struct bpf_prog *prog,
8569 struct bpf_insn_access_aux *info)
8571 if (!bpf_sock_is_valid_access(off, size, type, info))
8573 return __sock_filter_check_attach_type(off, type,
8574 prog->expected_attach_type);
8577 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
8578 const struct bpf_prog *prog)
8580 /* Neither direct read nor direct write requires any preliminary
8586 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
8587 const struct bpf_prog *prog, int drop_verdict)
8589 struct bpf_insn *insn = insn_buf;
8594 /* if (!skb->cloned)
8597 * (Fast-path, otherwise approximation that we might be
8598 * a clone, do the rest in helper.)
8600 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET);
8601 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
8602 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
8604 /* ret = bpf_skb_pull_data(skb, 0); */
8605 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
8606 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
8607 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
8608 BPF_FUNC_skb_pull_data);
8611 * return TC_ACT_SHOT;
8613 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
8614 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
8615 *insn++ = BPF_EXIT_INSN();
8618 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
8620 *insn++ = prog->insnsi[0];
8622 return insn - insn_buf;
8625 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
8626 struct bpf_insn *insn_buf)
8628 bool indirect = BPF_MODE(orig->code) == BPF_IND;
8629 struct bpf_insn *insn = insn_buf;
8632 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
8634 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
8636 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
8638 /* We're guaranteed here that CTX is in R6. */
8639 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
8641 switch (BPF_SIZE(orig->code)) {
8643 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
8646 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
8649 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
8653 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
8654 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
8655 *insn++ = BPF_EXIT_INSN();
8657 return insn - insn_buf;
8660 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8661 const struct bpf_prog *prog)
8663 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8666 static bool tc_cls_act_is_valid_access(int off, int size,
8667 enum bpf_access_type type,
8668 const struct bpf_prog *prog,
8669 struct bpf_insn_access_aux *info)
8671 if (type == BPF_WRITE) {
8673 case bpf_ctx_range(struct __sk_buff, mark):
8674 case bpf_ctx_range(struct __sk_buff, tc_index):
8675 case bpf_ctx_range(struct __sk_buff, priority):
8676 case bpf_ctx_range(struct __sk_buff, tc_classid):
8677 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8678 case bpf_ctx_range(struct __sk_buff, tstamp):
8679 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8687 case bpf_ctx_range(struct __sk_buff, data):
8688 info->reg_type = PTR_TO_PACKET;
8690 case bpf_ctx_range(struct __sk_buff, data_meta):
8691 info->reg_type = PTR_TO_PACKET_META;
8693 case bpf_ctx_range(struct __sk_buff, data_end):
8694 info->reg_type = PTR_TO_PACKET_END;
8696 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8698 case offsetof(struct __sk_buff, tstamp_type):
8699 /* The convert_ctx_access() on reading and writing
8700 * __sk_buff->tstamp depends on whether the bpf prog
8701 * has used __sk_buff->tstamp_type or not.
8702 * Thus, we need to set prog->tstamp_type_access
8703 * earlier during is_valid_access() here.
8705 ((struct bpf_prog *)prog)->tstamp_type_access = 1;
8706 return size == sizeof(__u8);
8709 return bpf_skb_is_valid_access(off, size, type, prog, info);
8712 DEFINE_MUTEX(nf_conn_btf_access_lock);
8713 EXPORT_SYMBOL_GPL(nf_conn_btf_access_lock);
8715 int (*nfct_btf_struct_access)(struct bpf_verifier_log *log,
8716 const struct bpf_reg_state *reg,
8717 int off, int size, enum bpf_access_type atype,
8718 u32 *next_btf_id, enum bpf_type_flag *flag);
8719 EXPORT_SYMBOL_GPL(nfct_btf_struct_access);
8721 static int tc_cls_act_btf_struct_access(struct bpf_verifier_log *log,
8722 const struct bpf_reg_state *reg,
8723 int off, int size, enum bpf_access_type atype,
8724 u32 *next_btf_id, enum bpf_type_flag *flag)
8728 if (atype == BPF_READ)
8729 return btf_struct_access(log, reg, off, size, atype, next_btf_id, flag);
8731 mutex_lock(&nf_conn_btf_access_lock);
8732 if (nfct_btf_struct_access)
8733 ret = nfct_btf_struct_access(log, reg, off, size, atype, next_btf_id, flag);
8734 mutex_unlock(&nf_conn_btf_access_lock);
8739 static bool __is_valid_xdp_access(int off, int size)
8741 if (off < 0 || off >= sizeof(struct xdp_md))
8743 if (off % size != 0)
8745 if (size != sizeof(__u32))
8751 static bool xdp_is_valid_access(int off, int size,
8752 enum bpf_access_type type,
8753 const struct bpf_prog *prog,
8754 struct bpf_insn_access_aux *info)
8756 if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8758 case offsetof(struct xdp_md, egress_ifindex):
8763 if (type == BPF_WRITE) {
8764 if (bpf_prog_is_offloaded(prog->aux)) {
8766 case offsetof(struct xdp_md, rx_queue_index):
8767 return __is_valid_xdp_access(off, size);
8774 case offsetof(struct xdp_md, data):
8775 info->reg_type = PTR_TO_PACKET;
8777 case offsetof(struct xdp_md, data_meta):
8778 info->reg_type = PTR_TO_PACKET_META;
8780 case offsetof(struct xdp_md, data_end):
8781 info->reg_type = PTR_TO_PACKET_END;
8785 return __is_valid_xdp_access(off, size);
8788 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act)
8790 const u32 act_max = XDP_REDIRECT;
8792 pr_warn_once("%s XDP return value %u on prog %s (id %d) dev %s, expect packet loss!\n",
8793 act > act_max ? "Illegal" : "Driver unsupported",
8794 act, prog->aux->name, prog->aux->id, dev ? dev->name : "N/A");
8796 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
8798 static int xdp_btf_struct_access(struct bpf_verifier_log *log,
8799 const struct bpf_reg_state *reg,
8800 int off, int size, enum bpf_access_type atype,
8801 u32 *next_btf_id, enum bpf_type_flag *flag)
8805 if (atype == BPF_READ)
8806 return btf_struct_access(log, reg, off, size, atype, next_btf_id, flag);
8808 mutex_lock(&nf_conn_btf_access_lock);
8809 if (nfct_btf_struct_access)
8810 ret = nfct_btf_struct_access(log, reg, off, size, atype, next_btf_id, flag);
8811 mutex_unlock(&nf_conn_btf_access_lock);
8816 static bool sock_addr_is_valid_access(int off, int size,
8817 enum bpf_access_type type,
8818 const struct bpf_prog *prog,
8819 struct bpf_insn_access_aux *info)
8821 const int size_default = sizeof(__u32);
8823 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
8825 if (off % size != 0)
8828 /* Disallow access to IPv6 fields from IPv4 contex and vise
8832 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8833 switch (prog->expected_attach_type) {
8834 case BPF_CGROUP_INET4_BIND:
8835 case BPF_CGROUP_INET4_CONNECT:
8836 case BPF_CGROUP_INET4_GETPEERNAME:
8837 case BPF_CGROUP_INET4_GETSOCKNAME:
8838 case BPF_CGROUP_UDP4_SENDMSG:
8839 case BPF_CGROUP_UDP4_RECVMSG:
8845 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8846 switch (prog->expected_attach_type) {
8847 case BPF_CGROUP_INET6_BIND:
8848 case BPF_CGROUP_INET6_CONNECT:
8849 case BPF_CGROUP_INET6_GETPEERNAME:
8850 case BPF_CGROUP_INET6_GETSOCKNAME:
8851 case BPF_CGROUP_UDP6_SENDMSG:
8852 case BPF_CGROUP_UDP6_RECVMSG:
8858 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8859 switch (prog->expected_attach_type) {
8860 case BPF_CGROUP_UDP4_SENDMSG:
8866 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8868 switch (prog->expected_attach_type) {
8869 case BPF_CGROUP_UDP6_SENDMSG:
8878 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8879 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8880 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8881 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8883 case bpf_ctx_range(struct bpf_sock_addr, user_port):
8884 if (type == BPF_READ) {
8885 bpf_ctx_record_field_size(info, size_default);
8887 if (bpf_ctx_wide_access_ok(off, size,
8888 struct bpf_sock_addr,
8892 if (bpf_ctx_wide_access_ok(off, size,
8893 struct bpf_sock_addr,
8897 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8900 if (bpf_ctx_wide_access_ok(off, size,
8901 struct bpf_sock_addr,
8905 if (bpf_ctx_wide_access_ok(off, size,
8906 struct bpf_sock_addr,
8910 if (size != size_default)
8914 case offsetof(struct bpf_sock_addr, sk):
8915 if (type != BPF_READ)
8917 if (size != sizeof(__u64))
8919 info->reg_type = PTR_TO_SOCKET;
8922 if (type == BPF_READ) {
8923 if (size != size_default)
8933 static bool sock_ops_is_valid_access(int off, int size,
8934 enum bpf_access_type type,
8935 const struct bpf_prog *prog,
8936 struct bpf_insn_access_aux *info)
8938 const int size_default = sizeof(__u32);
8940 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
8943 /* The verifier guarantees that size > 0. */
8944 if (off % size != 0)
8947 if (type == BPF_WRITE) {
8949 case offsetof(struct bpf_sock_ops, reply):
8950 case offsetof(struct bpf_sock_ops, sk_txhash):
8951 if (size != size_default)
8959 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
8961 if (size != sizeof(__u64))
8964 case offsetof(struct bpf_sock_ops, sk):
8965 if (size != sizeof(__u64))
8967 info->reg_type = PTR_TO_SOCKET_OR_NULL;
8969 case offsetof(struct bpf_sock_ops, skb_data):
8970 if (size != sizeof(__u64))
8972 info->reg_type = PTR_TO_PACKET;
8974 case offsetof(struct bpf_sock_ops, skb_data_end):
8975 if (size != sizeof(__u64))
8977 info->reg_type = PTR_TO_PACKET_END;
8979 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
8980 bpf_ctx_record_field_size(info, size_default);
8981 return bpf_ctx_narrow_access_ok(off, size,
8983 case offsetof(struct bpf_sock_ops, skb_hwtstamp):
8984 if (size != sizeof(__u64))
8988 if (size != size_default)
8997 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
8998 const struct bpf_prog *prog)
9000 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
9003 static bool sk_skb_is_valid_access(int off, int size,
9004 enum bpf_access_type type,
9005 const struct bpf_prog *prog,
9006 struct bpf_insn_access_aux *info)
9009 case bpf_ctx_range(struct __sk_buff, tc_classid):
9010 case bpf_ctx_range(struct __sk_buff, data_meta):
9011 case bpf_ctx_range(struct __sk_buff, tstamp):
9012 case bpf_ctx_range(struct __sk_buff, wire_len):
9013 case bpf_ctx_range(struct __sk_buff, hwtstamp):
9017 if (type == BPF_WRITE) {
9019 case bpf_ctx_range(struct __sk_buff, tc_index):
9020 case bpf_ctx_range(struct __sk_buff, priority):
9028 case bpf_ctx_range(struct __sk_buff, mark):
9030 case bpf_ctx_range(struct __sk_buff, data):
9031 info->reg_type = PTR_TO_PACKET;
9033 case bpf_ctx_range(struct __sk_buff, data_end):
9034 info->reg_type = PTR_TO_PACKET_END;
9038 return bpf_skb_is_valid_access(off, size, type, prog, info);
9041 static bool sk_msg_is_valid_access(int off, int size,
9042 enum bpf_access_type type,
9043 const struct bpf_prog *prog,
9044 struct bpf_insn_access_aux *info)
9046 if (type == BPF_WRITE)
9049 if (off % size != 0)
9053 case offsetof(struct sk_msg_md, data):
9054 info->reg_type = PTR_TO_PACKET;
9055 if (size != sizeof(__u64))
9058 case offsetof(struct sk_msg_md, data_end):
9059 info->reg_type = PTR_TO_PACKET_END;
9060 if (size != sizeof(__u64))
9063 case offsetof(struct sk_msg_md, sk):
9064 if (size != sizeof(__u64))
9066 info->reg_type = PTR_TO_SOCKET;
9068 case bpf_ctx_range(struct sk_msg_md, family):
9069 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
9070 case bpf_ctx_range(struct sk_msg_md, local_ip4):
9071 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
9072 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
9073 case bpf_ctx_range(struct sk_msg_md, remote_port):
9074 case bpf_ctx_range(struct sk_msg_md, local_port):
9075 case bpf_ctx_range(struct sk_msg_md, size):
9076 if (size != sizeof(__u32))
9085 static bool flow_dissector_is_valid_access(int off, int size,
9086 enum bpf_access_type type,
9087 const struct bpf_prog *prog,
9088 struct bpf_insn_access_aux *info)
9090 const int size_default = sizeof(__u32);
9092 if (off < 0 || off >= sizeof(struct __sk_buff))
9095 if (type == BPF_WRITE)
9099 case bpf_ctx_range(struct __sk_buff, data):
9100 if (size != size_default)
9102 info->reg_type = PTR_TO_PACKET;
9104 case bpf_ctx_range(struct __sk_buff, data_end):
9105 if (size != size_default)
9107 info->reg_type = PTR_TO_PACKET_END;
9109 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
9110 if (size != sizeof(__u64))
9112 info->reg_type = PTR_TO_FLOW_KEYS;
9119 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
9120 const struct bpf_insn *si,
9121 struct bpf_insn *insn_buf,
9122 struct bpf_prog *prog,
9126 struct bpf_insn *insn = insn_buf;
9129 case offsetof(struct __sk_buff, data):
9130 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
9131 si->dst_reg, si->src_reg,
9132 offsetof(struct bpf_flow_dissector, data));
9135 case offsetof(struct __sk_buff, data_end):
9136 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
9137 si->dst_reg, si->src_reg,
9138 offsetof(struct bpf_flow_dissector, data_end));
9141 case offsetof(struct __sk_buff, flow_keys):
9142 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
9143 si->dst_reg, si->src_reg,
9144 offsetof(struct bpf_flow_dissector, flow_keys));
9148 return insn - insn_buf;
9151 static struct bpf_insn *bpf_convert_tstamp_type_read(const struct bpf_insn *si,
9152 struct bpf_insn *insn)
9154 __u8 value_reg = si->dst_reg;
9155 __u8 skb_reg = si->src_reg;
9156 /* AX is needed because src_reg and dst_reg could be the same */
9157 __u8 tmp_reg = BPF_REG_AX;
9159 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg,
9160 PKT_VLAN_PRESENT_OFFSET);
9161 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg,
9162 SKB_MONO_DELIVERY_TIME_MASK, 2);
9163 *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_UNSPEC);
9164 *insn++ = BPF_JMP_A(1);
9165 *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_DELIVERY_MONO);
9170 static struct bpf_insn *bpf_convert_shinfo_access(__u8 dst_reg, __u8 skb_reg,
9171 struct bpf_insn *insn)
9173 /* si->dst_reg = skb_shinfo(SKB); */
9174 #ifdef NET_SKBUFF_DATA_USES_OFFSET
9175 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9176 BPF_REG_AX, skb_reg,
9177 offsetof(struct sk_buff, end));
9178 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
9180 offsetof(struct sk_buff, head));
9181 *insn++ = BPF_ALU64_REG(BPF_ADD, dst_reg, BPF_REG_AX);
9183 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9185 offsetof(struct sk_buff, end));
9191 static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog,
9192 const struct bpf_insn *si,
9193 struct bpf_insn *insn)
9195 __u8 value_reg = si->dst_reg;
9196 __u8 skb_reg = si->src_reg;
9198 #ifdef CONFIG_NET_CLS_ACT
9199 /* If the tstamp_type is read,
9200 * the bpf prog is aware the tstamp could have delivery time.
9201 * Thus, read skb->tstamp as is if tstamp_type_access is true.
9203 if (!prog->tstamp_type_access) {
9204 /* AX is needed because src_reg and dst_reg could be the same */
9205 __u8 tmp_reg = BPF_REG_AX;
9207 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, PKT_VLAN_PRESENT_OFFSET);
9208 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg,
9209 TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK);
9210 *insn++ = BPF_JMP32_IMM(BPF_JNE, tmp_reg,
9211 TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK, 2);
9212 /* skb->tc_at_ingress && skb->mono_delivery_time,
9213 * read 0 as the (rcv) timestamp.
9215 *insn++ = BPF_MOV64_IMM(value_reg, 0);
9216 *insn++ = BPF_JMP_A(1);
9220 *insn++ = BPF_LDX_MEM(BPF_DW, value_reg, skb_reg,
9221 offsetof(struct sk_buff, tstamp));
9225 static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog,
9226 const struct bpf_insn *si,
9227 struct bpf_insn *insn)
9229 __u8 value_reg = si->src_reg;
9230 __u8 skb_reg = si->dst_reg;
9232 #ifdef CONFIG_NET_CLS_ACT
9233 /* If the tstamp_type is read,
9234 * the bpf prog is aware the tstamp could have delivery time.
9235 * Thus, write skb->tstamp as is if tstamp_type_access is true.
9236 * Otherwise, writing at ingress will have to clear the
9237 * mono_delivery_time bit also.
9239 if (!prog->tstamp_type_access) {
9240 __u8 tmp_reg = BPF_REG_AX;
9242 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, PKT_VLAN_PRESENT_OFFSET);
9243 /* Writing __sk_buff->tstamp as ingress, goto <clear> */
9244 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1);
9246 *insn++ = BPF_JMP_A(2);
9247 /* <clear>: mono_delivery_time */
9248 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_MONO_DELIVERY_TIME_MASK);
9249 *insn++ = BPF_STX_MEM(BPF_B, skb_reg, tmp_reg, PKT_VLAN_PRESENT_OFFSET);
9253 /* <store>: skb->tstamp = tstamp */
9254 *insn++ = BPF_STX_MEM(BPF_DW, skb_reg, value_reg,
9255 offsetof(struct sk_buff, tstamp));
9259 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
9260 const struct bpf_insn *si,
9261 struct bpf_insn *insn_buf,
9262 struct bpf_prog *prog, u32 *target_size)
9264 struct bpf_insn *insn = insn_buf;
9268 case offsetof(struct __sk_buff, len):
9269 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9270 bpf_target_off(struct sk_buff, len, 4,
9274 case offsetof(struct __sk_buff, protocol):
9275 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9276 bpf_target_off(struct sk_buff, protocol, 2,
9280 case offsetof(struct __sk_buff, vlan_proto):
9281 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9282 bpf_target_off(struct sk_buff, vlan_proto, 2,
9286 case offsetof(struct __sk_buff, priority):
9287 if (type == BPF_WRITE)
9288 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9289 bpf_target_off(struct sk_buff, priority, 4,
9292 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9293 bpf_target_off(struct sk_buff, priority, 4,
9297 case offsetof(struct __sk_buff, ingress_ifindex):
9298 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9299 bpf_target_off(struct sk_buff, skb_iif, 4,
9303 case offsetof(struct __sk_buff, ifindex):
9304 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9305 si->dst_reg, si->src_reg,
9306 offsetof(struct sk_buff, dev));
9307 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9308 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9309 bpf_target_off(struct net_device, ifindex, 4,
9313 case offsetof(struct __sk_buff, hash):
9314 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9315 bpf_target_off(struct sk_buff, hash, 4,
9319 case offsetof(struct __sk_buff, mark):
9320 if (type == BPF_WRITE)
9321 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9322 bpf_target_off(struct sk_buff, mark, 4,
9325 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9326 bpf_target_off(struct sk_buff, mark, 4,
9330 case offsetof(struct __sk_buff, pkt_type):
9332 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
9334 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
9335 #ifdef __BIG_ENDIAN_BITFIELD
9336 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
9340 case offsetof(struct __sk_buff, queue_mapping):
9341 if (type == BPF_WRITE) {
9342 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
9343 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
9344 bpf_target_off(struct sk_buff,
9348 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9349 bpf_target_off(struct sk_buff,
9355 case offsetof(struct __sk_buff, vlan_present):
9356 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9357 bpf_target_off(struct sk_buff,
9358 vlan_all, 4, target_size));
9359 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9360 *insn++ = BPF_ALU32_IMM(BPF_MOV, si->dst_reg, 1);
9363 case offsetof(struct __sk_buff, vlan_tci):
9364 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9365 bpf_target_off(struct sk_buff, vlan_tci, 2,
9369 case offsetof(struct __sk_buff, cb[0]) ...
9370 offsetofend(struct __sk_buff, cb[4]) - 1:
9371 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
9372 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
9373 offsetof(struct qdisc_skb_cb, data)) %
9376 prog->cb_access = 1;
9378 off -= offsetof(struct __sk_buff, cb[0]);
9379 off += offsetof(struct sk_buff, cb);
9380 off += offsetof(struct qdisc_skb_cb, data);
9381 if (type == BPF_WRITE)
9382 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
9385 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
9389 case offsetof(struct __sk_buff, tc_classid):
9390 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
9393 off -= offsetof(struct __sk_buff, tc_classid);
9394 off += offsetof(struct sk_buff, cb);
9395 off += offsetof(struct qdisc_skb_cb, tc_classid);
9397 if (type == BPF_WRITE)
9398 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
9401 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
9405 case offsetof(struct __sk_buff, data):
9406 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9407 si->dst_reg, si->src_reg,
9408 offsetof(struct sk_buff, data));
9411 case offsetof(struct __sk_buff, data_meta):
9413 off -= offsetof(struct __sk_buff, data_meta);
9414 off += offsetof(struct sk_buff, cb);
9415 off += offsetof(struct bpf_skb_data_end, data_meta);
9416 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9420 case offsetof(struct __sk_buff, data_end):
9422 off -= offsetof(struct __sk_buff, data_end);
9423 off += offsetof(struct sk_buff, cb);
9424 off += offsetof(struct bpf_skb_data_end, data_end);
9425 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9429 case offsetof(struct __sk_buff, tc_index):
9430 #ifdef CONFIG_NET_SCHED
9431 if (type == BPF_WRITE)
9432 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
9433 bpf_target_off(struct sk_buff, tc_index, 2,
9436 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9437 bpf_target_off(struct sk_buff, tc_index, 2,
9441 if (type == BPF_WRITE)
9442 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
9444 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9448 case offsetof(struct __sk_buff, napi_id):
9449 #if defined(CONFIG_NET_RX_BUSY_POLL)
9450 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9451 bpf_target_off(struct sk_buff, napi_id, 4,
9453 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
9454 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9457 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9460 case offsetof(struct __sk_buff, family):
9461 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9463 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9464 si->dst_reg, si->src_reg,
9465 offsetof(struct sk_buff, sk));
9466 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9467 bpf_target_off(struct sock_common,
9471 case offsetof(struct __sk_buff, remote_ip4):
9472 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9474 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9475 si->dst_reg, si->src_reg,
9476 offsetof(struct sk_buff, sk));
9477 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9478 bpf_target_off(struct sock_common,
9482 case offsetof(struct __sk_buff, local_ip4):
9483 BUILD_BUG_ON(sizeof_field(struct sock_common,
9484 skc_rcv_saddr) != 4);
9486 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9487 si->dst_reg, si->src_reg,
9488 offsetof(struct sk_buff, sk));
9489 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9490 bpf_target_off(struct sock_common,
9494 case offsetof(struct __sk_buff, remote_ip6[0]) ...
9495 offsetof(struct __sk_buff, remote_ip6[3]):
9496 #if IS_ENABLED(CONFIG_IPV6)
9497 BUILD_BUG_ON(sizeof_field(struct sock_common,
9498 skc_v6_daddr.s6_addr32[0]) != 4);
9501 off -= offsetof(struct __sk_buff, remote_ip6[0]);
9503 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9504 si->dst_reg, si->src_reg,
9505 offsetof(struct sk_buff, sk));
9506 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9507 offsetof(struct sock_common,
9508 skc_v6_daddr.s6_addr32[0]) +
9511 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9514 case offsetof(struct __sk_buff, local_ip6[0]) ...
9515 offsetof(struct __sk_buff, local_ip6[3]):
9516 #if IS_ENABLED(CONFIG_IPV6)
9517 BUILD_BUG_ON(sizeof_field(struct sock_common,
9518 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9521 off -= offsetof(struct __sk_buff, local_ip6[0]);
9523 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9524 si->dst_reg, si->src_reg,
9525 offsetof(struct sk_buff, sk));
9526 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9527 offsetof(struct sock_common,
9528 skc_v6_rcv_saddr.s6_addr32[0]) +
9531 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9535 case offsetof(struct __sk_buff, remote_port):
9536 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9538 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9539 si->dst_reg, si->src_reg,
9540 offsetof(struct sk_buff, sk));
9541 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9542 bpf_target_off(struct sock_common,
9545 #ifndef __BIG_ENDIAN_BITFIELD
9546 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9550 case offsetof(struct __sk_buff, local_port):
9551 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9553 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9554 si->dst_reg, si->src_reg,
9555 offsetof(struct sk_buff, sk));
9556 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9557 bpf_target_off(struct sock_common,
9558 skc_num, 2, target_size));
9561 case offsetof(struct __sk_buff, tstamp):
9562 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
9564 if (type == BPF_WRITE)
9565 insn = bpf_convert_tstamp_write(prog, si, insn);
9567 insn = bpf_convert_tstamp_read(prog, si, insn);
9570 case offsetof(struct __sk_buff, tstamp_type):
9571 insn = bpf_convert_tstamp_type_read(si, insn);
9574 case offsetof(struct __sk_buff, gso_segs):
9575 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9576 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
9577 si->dst_reg, si->dst_reg,
9578 bpf_target_off(struct skb_shared_info,
9582 case offsetof(struct __sk_buff, gso_size):
9583 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9584 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
9585 si->dst_reg, si->dst_reg,
9586 bpf_target_off(struct skb_shared_info,
9590 case offsetof(struct __sk_buff, wire_len):
9591 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
9594 off -= offsetof(struct __sk_buff, wire_len);
9595 off += offsetof(struct sk_buff, cb);
9596 off += offsetof(struct qdisc_skb_cb, pkt_len);
9598 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
9601 case offsetof(struct __sk_buff, sk):
9602 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9603 si->dst_reg, si->src_reg,
9604 offsetof(struct sk_buff, sk));
9606 case offsetof(struct __sk_buff, hwtstamp):
9607 BUILD_BUG_ON(sizeof_field(struct skb_shared_hwtstamps, hwtstamp) != 8);
9608 BUILD_BUG_ON(offsetof(struct skb_shared_hwtstamps, hwtstamp) != 0);
9610 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9611 *insn++ = BPF_LDX_MEM(BPF_DW,
9612 si->dst_reg, si->dst_reg,
9613 bpf_target_off(struct skb_shared_info,
9619 return insn - insn_buf;
9622 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
9623 const struct bpf_insn *si,
9624 struct bpf_insn *insn_buf,
9625 struct bpf_prog *prog, u32 *target_size)
9627 struct bpf_insn *insn = insn_buf;
9631 case offsetof(struct bpf_sock, bound_dev_if):
9632 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
9634 if (type == BPF_WRITE)
9635 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9636 offsetof(struct sock, sk_bound_dev_if));
9638 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9639 offsetof(struct sock, sk_bound_dev_if));
9642 case offsetof(struct bpf_sock, mark):
9643 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
9645 if (type == BPF_WRITE)
9646 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9647 offsetof(struct sock, sk_mark));
9649 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9650 offsetof(struct sock, sk_mark));
9653 case offsetof(struct bpf_sock, priority):
9654 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
9656 if (type == BPF_WRITE)
9657 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9658 offsetof(struct sock, sk_priority));
9660 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9661 offsetof(struct sock, sk_priority));
9664 case offsetof(struct bpf_sock, family):
9665 *insn++ = BPF_LDX_MEM(
9666 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
9667 si->dst_reg, si->src_reg,
9668 bpf_target_off(struct sock_common,
9670 sizeof_field(struct sock_common,
9675 case offsetof(struct bpf_sock, type):
9676 *insn++ = BPF_LDX_MEM(
9677 BPF_FIELD_SIZEOF(struct sock, sk_type),
9678 si->dst_reg, si->src_reg,
9679 bpf_target_off(struct sock, sk_type,
9680 sizeof_field(struct sock, sk_type),
9684 case offsetof(struct bpf_sock, protocol):
9685 *insn++ = BPF_LDX_MEM(
9686 BPF_FIELD_SIZEOF(struct sock, sk_protocol),
9687 si->dst_reg, si->src_reg,
9688 bpf_target_off(struct sock, sk_protocol,
9689 sizeof_field(struct sock, sk_protocol),
9693 case offsetof(struct bpf_sock, src_ip4):
9694 *insn++ = BPF_LDX_MEM(
9695 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9696 bpf_target_off(struct sock_common, skc_rcv_saddr,
9697 sizeof_field(struct sock_common,
9702 case offsetof(struct bpf_sock, dst_ip4):
9703 *insn++ = BPF_LDX_MEM(
9704 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9705 bpf_target_off(struct sock_common, skc_daddr,
9706 sizeof_field(struct sock_common,
9711 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
9712 #if IS_ENABLED(CONFIG_IPV6)
9714 off -= offsetof(struct bpf_sock, src_ip6[0]);
9715 *insn++ = BPF_LDX_MEM(
9716 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9719 skc_v6_rcv_saddr.s6_addr32[0],
9720 sizeof_field(struct sock_common,
9721 skc_v6_rcv_saddr.s6_addr32[0]),
9722 target_size) + off);
9725 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9729 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
9730 #if IS_ENABLED(CONFIG_IPV6)
9732 off -= offsetof(struct bpf_sock, dst_ip6[0]);
9733 *insn++ = BPF_LDX_MEM(
9734 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9735 bpf_target_off(struct sock_common,
9736 skc_v6_daddr.s6_addr32[0],
9737 sizeof_field(struct sock_common,
9738 skc_v6_daddr.s6_addr32[0]),
9739 target_size) + off);
9741 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9746 case offsetof(struct bpf_sock, src_port):
9747 *insn++ = BPF_LDX_MEM(
9748 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
9749 si->dst_reg, si->src_reg,
9750 bpf_target_off(struct sock_common, skc_num,
9751 sizeof_field(struct sock_common,
9756 case offsetof(struct bpf_sock, dst_port):
9757 *insn++ = BPF_LDX_MEM(
9758 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
9759 si->dst_reg, si->src_reg,
9760 bpf_target_off(struct sock_common, skc_dport,
9761 sizeof_field(struct sock_common,
9766 case offsetof(struct bpf_sock, state):
9767 *insn++ = BPF_LDX_MEM(
9768 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
9769 si->dst_reg, si->src_reg,
9770 bpf_target_off(struct sock_common, skc_state,
9771 sizeof_field(struct sock_common,
9775 case offsetof(struct bpf_sock, rx_queue_mapping):
9776 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
9777 *insn++ = BPF_LDX_MEM(
9778 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
9779 si->dst_reg, si->src_reg,
9780 bpf_target_off(struct sock, sk_rx_queue_mapping,
9781 sizeof_field(struct sock,
9782 sk_rx_queue_mapping),
9784 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
9786 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9788 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9794 return insn - insn_buf;
9797 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
9798 const struct bpf_insn *si,
9799 struct bpf_insn *insn_buf,
9800 struct bpf_prog *prog, u32 *target_size)
9802 struct bpf_insn *insn = insn_buf;
9805 case offsetof(struct __sk_buff, ifindex):
9806 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9807 si->dst_reg, si->src_reg,
9808 offsetof(struct sk_buff, dev));
9809 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9810 bpf_target_off(struct net_device, ifindex, 4,
9814 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9818 return insn - insn_buf;
9821 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
9822 const struct bpf_insn *si,
9823 struct bpf_insn *insn_buf,
9824 struct bpf_prog *prog, u32 *target_size)
9826 struct bpf_insn *insn = insn_buf;
9829 case offsetof(struct xdp_md, data):
9830 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
9831 si->dst_reg, si->src_reg,
9832 offsetof(struct xdp_buff, data));
9834 case offsetof(struct xdp_md, data_meta):
9835 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
9836 si->dst_reg, si->src_reg,
9837 offsetof(struct xdp_buff, data_meta));
9839 case offsetof(struct xdp_md, data_end):
9840 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
9841 si->dst_reg, si->src_reg,
9842 offsetof(struct xdp_buff, data_end));
9844 case offsetof(struct xdp_md, ingress_ifindex):
9845 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9846 si->dst_reg, si->src_reg,
9847 offsetof(struct xdp_buff, rxq));
9848 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
9849 si->dst_reg, si->dst_reg,
9850 offsetof(struct xdp_rxq_info, dev));
9851 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9852 offsetof(struct net_device, ifindex));
9854 case offsetof(struct xdp_md, rx_queue_index):
9855 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9856 si->dst_reg, si->src_reg,
9857 offsetof(struct xdp_buff, rxq));
9858 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9859 offsetof(struct xdp_rxq_info,
9862 case offsetof(struct xdp_md, egress_ifindex):
9863 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
9864 si->dst_reg, si->src_reg,
9865 offsetof(struct xdp_buff, txq));
9866 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
9867 si->dst_reg, si->dst_reg,
9868 offsetof(struct xdp_txq_info, dev));
9869 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9870 offsetof(struct net_device, ifindex));
9874 return insn - insn_buf;
9877 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
9878 * context Structure, F is Field in context structure that contains a pointer
9879 * to Nested Structure of type NS that has the field NF.
9881 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
9882 * sure that SIZE is not greater than actual size of S.F.NF.
9884 * If offset OFF is provided, the load happens from that offset relative to
9887 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
9889 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
9890 si->src_reg, offsetof(S, F)); \
9891 *insn++ = BPF_LDX_MEM( \
9892 SIZE, si->dst_reg, si->dst_reg, \
9893 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
9898 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
9899 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
9900 BPF_FIELD_SIZEOF(NS, NF), 0)
9902 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
9903 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
9905 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
9906 * "register" since two registers available in convert_ctx_access are not
9907 * enough: we can't override neither SRC, since it contains value to store, nor
9908 * DST since it contains pointer to context that may be used by later
9909 * instructions. But we need a temporary place to save pointer to nested
9910 * structure whose field we want to store to.
9912 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
9914 int tmp_reg = BPF_REG_9; \
9915 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
9917 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
9919 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
9921 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
9922 si->dst_reg, offsetof(S, F)); \
9923 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \
9924 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
9927 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
9931 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
9934 if (type == BPF_WRITE) { \
9935 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
9938 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
9939 S, NS, F, NF, SIZE, OFF); \
9943 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
9944 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
9945 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
9947 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
9948 const struct bpf_insn *si,
9949 struct bpf_insn *insn_buf,
9950 struct bpf_prog *prog, u32 *target_size)
9952 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
9953 struct bpf_insn *insn = insn_buf;
9956 case offsetof(struct bpf_sock_addr, user_family):
9957 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9958 struct sockaddr, uaddr, sa_family);
9961 case offsetof(struct bpf_sock_addr, user_ip4):
9962 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9963 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
9964 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
9967 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9969 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
9970 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9971 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9972 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
9976 case offsetof(struct bpf_sock_addr, user_port):
9977 /* To get port we need to know sa_family first and then treat
9978 * sockaddr as either sockaddr_in or sockaddr_in6.
9979 * Though we can simplify since port field has same offset and
9980 * size in both structures.
9981 * Here we check this invariant and use just one of the
9982 * structures if it's true.
9984 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
9985 offsetof(struct sockaddr_in6, sin6_port));
9986 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
9987 sizeof_field(struct sockaddr_in6, sin6_port));
9988 /* Account for sin6_port being smaller than user_port. */
9989 port_size = min(port_size, BPF_LDST_BYTES(si));
9990 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9991 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9992 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
9995 case offsetof(struct bpf_sock_addr, family):
9996 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9997 struct sock, sk, sk_family);
10000 case offsetof(struct bpf_sock_addr, type):
10001 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10002 struct sock, sk, sk_type);
10005 case offsetof(struct bpf_sock_addr, protocol):
10006 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10007 struct sock, sk, sk_protocol);
10010 case offsetof(struct bpf_sock_addr, msg_src_ip4):
10011 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
10012 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10013 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
10014 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
10017 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
10020 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
10021 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
10022 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10023 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
10024 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
10026 case offsetof(struct bpf_sock_addr, sk):
10027 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
10028 si->dst_reg, si->src_reg,
10029 offsetof(struct bpf_sock_addr_kern, sk));
10033 return insn - insn_buf;
10036 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
10037 const struct bpf_insn *si,
10038 struct bpf_insn *insn_buf,
10039 struct bpf_prog *prog,
10042 struct bpf_insn *insn = insn_buf;
10045 /* Helper macro for adding read access to tcp_sock or sock fields. */
10046 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
10048 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \
10049 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
10050 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
10051 if (si->dst_reg == reg || si->src_reg == reg) \
10053 if (si->dst_reg == reg || si->src_reg == reg) \
10055 if (si->dst_reg == si->src_reg) { \
10056 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
10057 offsetof(struct bpf_sock_ops_kern, \
10059 fullsock_reg = reg; \
10062 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10063 struct bpf_sock_ops_kern, \
10065 fullsock_reg, si->src_reg, \
10066 offsetof(struct bpf_sock_ops_kern, \
10068 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
10069 if (si->dst_reg == si->src_reg) \
10070 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10071 offsetof(struct bpf_sock_ops_kern, \
10073 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10074 struct bpf_sock_ops_kern, sk),\
10075 si->dst_reg, si->src_reg, \
10076 offsetof(struct bpf_sock_ops_kern, sk));\
10077 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
10079 si->dst_reg, si->dst_reg, \
10080 offsetof(OBJ, OBJ_FIELD)); \
10081 if (si->dst_reg == si->src_reg) { \
10082 *insn++ = BPF_JMP_A(1); \
10083 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10084 offsetof(struct bpf_sock_ops_kern, \
10089 #define SOCK_OPS_GET_SK() \
10091 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \
10092 if (si->dst_reg == reg || si->src_reg == reg) \
10094 if (si->dst_reg == reg || si->src_reg == reg) \
10096 if (si->dst_reg == si->src_reg) { \
10097 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
10098 offsetof(struct bpf_sock_ops_kern, \
10100 fullsock_reg = reg; \
10103 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10104 struct bpf_sock_ops_kern, \
10106 fullsock_reg, si->src_reg, \
10107 offsetof(struct bpf_sock_ops_kern, \
10109 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
10110 if (si->dst_reg == si->src_reg) \
10111 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10112 offsetof(struct bpf_sock_ops_kern, \
10114 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10115 struct bpf_sock_ops_kern, sk),\
10116 si->dst_reg, si->src_reg, \
10117 offsetof(struct bpf_sock_ops_kern, sk));\
10118 if (si->dst_reg == si->src_reg) { \
10119 *insn++ = BPF_JMP_A(1); \
10120 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10121 offsetof(struct bpf_sock_ops_kern, \
10126 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
10127 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
10129 /* Helper macro for adding write access to tcp_sock or sock fields.
10130 * The macro is called with two registers, dst_reg which contains a pointer
10131 * to ctx (context) and src_reg which contains the value that should be
10132 * stored. However, we need an additional register since we cannot overwrite
10133 * dst_reg because it may be used later in the program.
10134 * Instead we "borrow" one of the other register. We first save its value
10135 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
10136 * it at the end of the macro.
10138 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
10140 int reg = BPF_REG_9; \
10141 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
10142 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
10143 if (si->dst_reg == reg || si->src_reg == reg) \
10145 if (si->dst_reg == reg || si->src_reg == reg) \
10147 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
10148 offsetof(struct bpf_sock_ops_kern, \
10150 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10151 struct bpf_sock_ops_kern, \
10153 reg, si->dst_reg, \
10154 offsetof(struct bpf_sock_ops_kern, \
10156 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
10157 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10158 struct bpf_sock_ops_kern, sk),\
10159 reg, si->dst_reg, \
10160 offsetof(struct bpf_sock_ops_kern, sk));\
10161 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \
10162 reg, si->src_reg, \
10163 offsetof(OBJ, OBJ_FIELD)); \
10164 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
10165 offsetof(struct bpf_sock_ops_kern, \
10169 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
10171 if (TYPE == BPF_WRITE) \
10172 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
10174 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
10178 case offsetof(struct bpf_sock_ops, op):
10179 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10181 si->dst_reg, si->src_reg,
10182 offsetof(struct bpf_sock_ops_kern, op));
10185 case offsetof(struct bpf_sock_ops, replylong[0]) ...
10186 offsetof(struct bpf_sock_ops, replylong[3]):
10187 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
10188 sizeof_field(struct bpf_sock_ops_kern, reply));
10189 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
10190 sizeof_field(struct bpf_sock_ops_kern, replylong));
10192 off -= offsetof(struct bpf_sock_ops, replylong[0]);
10193 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
10194 if (type == BPF_WRITE)
10195 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
10198 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10202 case offsetof(struct bpf_sock_ops, family):
10203 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10205 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10206 struct bpf_sock_ops_kern, sk),
10207 si->dst_reg, si->src_reg,
10208 offsetof(struct bpf_sock_ops_kern, sk));
10209 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10210 offsetof(struct sock_common, skc_family));
10213 case offsetof(struct bpf_sock_ops, remote_ip4):
10214 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10216 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10217 struct bpf_sock_ops_kern, sk),
10218 si->dst_reg, si->src_reg,
10219 offsetof(struct bpf_sock_ops_kern, sk));
10220 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10221 offsetof(struct sock_common, skc_daddr));
10224 case offsetof(struct bpf_sock_ops, local_ip4):
10225 BUILD_BUG_ON(sizeof_field(struct sock_common,
10226 skc_rcv_saddr) != 4);
10228 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10229 struct bpf_sock_ops_kern, sk),
10230 si->dst_reg, si->src_reg,
10231 offsetof(struct bpf_sock_ops_kern, sk));
10232 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10233 offsetof(struct sock_common,
10237 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
10238 offsetof(struct bpf_sock_ops, remote_ip6[3]):
10239 #if IS_ENABLED(CONFIG_IPV6)
10240 BUILD_BUG_ON(sizeof_field(struct sock_common,
10241 skc_v6_daddr.s6_addr32[0]) != 4);
10244 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
10245 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10246 struct bpf_sock_ops_kern, sk),
10247 si->dst_reg, si->src_reg,
10248 offsetof(struct bpf_sock_ops_kern, sk));
10249 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10250 offsetof(struct sock_common,
10251 skc_v6_daddr.s6_addr32[0]) +
10254 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10258 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
10259 offsetof(struct bpf_sock_ops, local_ip6[3]):
10260 #if IS_ENABLED(CONFIG_IPV6)
10261 BUILD_BUG_ON(sizeof_field(struct sock_common,
10262 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10265 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
10266 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10267 struct bpf_sock_ops_kern, sk),
10268 si->dst_reg, si->src_reg,
10269 offsetof(struct bpf_sock_ops_kern, sk));
10270 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10271 offsetof(struct sock_common,
10272 skc_v6_rcv_saddr.s6_addr32[0]) +
10275 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10279 case offsetof(struct bpf_sock_ops, remote_port):
10280 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10282 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10283 struct bpf_sock_ops_kern, sk),
10284 si->dst_reg, si->src_reg,
10285 offsetof(struct bpf_sock_ops_kern, sk));
10286 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10287 offsetof(struct sock_common, skc_dport));
10288 #ifndef __BIG_ENDIAN_BITFIELD
10289 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10293 case offsetof(struct bpf_sock_ops, local_port):
10294 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10296 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10297 struct bpf_sock_ops_kern, sk),
10298 si->dst_reg, si->src_reg,
10299 offsetof(struct bpf_sock_ops_kern, sk));
10300 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10301 offsetof(struct sock_common, skc_num));
10304 case offsetof(struct bpf_sock_ops, is_fullsock):
10305 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10306 struct bpf_sock_ops_kern,
10308 si->dst_reg, si->src_reg,
10309 offsetof(struct bpf_sock_ops_kern,
10313 case offsetof(struct bpf_sock_ops, state):
10314 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
10316 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10317 struct bpf_sock_ops_kern, sk),
10318 si->dst_reg, si->src_reg,
10319 offsetof(struct bpf_sock_ops_kern, sk));
10320 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
10321 offsetof(struct sock_common, skc_state));
10324 case offsetof(struct bpf_sock_ops, rtt_min):
10325 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
10326 sizeof(struct minmax));
10327 BUILD_BUG_ON(sizeof(struct minmax) <
10328 sizeof(struct minmax_sample));
10330 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10331 struct bpf_sock_ops_kern, sk),
10332 si->dst_reg, si->src_reg,
10333 offsetof(struct bpf_sock_ops_kern, sk));
10334 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10335 offsetof(struct tcp_sock, rtt_min) +
10336 sizeof_field(struct minmax_sample, t));
10339 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
10340 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
10344 case offsetof(struct bpf_sock_ops, sk_txhash):
10345 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
10346 struct sock, type);
10348 case offsetof(struct bpf_sock_ops, snd_cwnd):
10349 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
10351 case offsetof(struct bpf_sock_ops, srtt_us):
10352 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
10354 case offsetof(struct bpf_sock_ops, snd_ssthresh):
10355 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
10357 case offsetof(struct bpf_sock_ops, rcv_nxt):
10358 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
10360 case offsetof(struct bpf_sock_ops, snd_nxt):
10361 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
10363 case offsetof(struct bpf_sock_ops, snd_una):
10364 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
10366 case offsetof(struct bpf_sock_ops, mss_cache):
10367 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
10369 case offsetof(struct bpf_sock_ops, ecn_flags):
10370 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
10372 case offsetof(struct bpf_sock_ops, rate_delivered):
10373 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
10375 case offsetof(struct bpf_sock_ops, rate_interval_us):
10376 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
10378 case offsetof(struct bpf_sock_ops, packets_out):
10379 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
10381 case offsetof(struct bpf_sock_ops, retrans_out):
10382 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
10384 case offsetof(struct bpf_sock_ops, total_retrans):
10385 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
10387 case offsetof(struct bpf_sock_ops, segs_in):
10388 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
10390 case offsetof(struct bpf_sock_ops, data_segs_in):
10391 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
10393 case offsetof(struct bpf_sock_ops, segs_out):
10394 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
10396 case offsetof(struct bpf_sock_ops, data_segs_out):
10397 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
10399 case offsetof(struct bpf_sock_ops, lost_out):
10400 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
10402 case offsetof(struct bpf_sock_ops, sacked_out):
10403 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
10405 case offsetof(struct bpf_sock_ops, bytes_received):
10406 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
10408 case offsetof(struct bpf_sock_ops, bytes_acked):
10409 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
10411 case offsetof(struct bpf_sock_ops, sk):
10414 case offsetof(struct bpf_sock_ops, skb_data_end):
10415 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10417 si->dst_reg, si->src_reg,
10418 offsetof(struct bpf_sock_ops_kern,
10421 case offsetof(struct bpf_sock_ops, skb_data):
10422 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10424 si->dst_reg, si->src_reg,
10425 offsetof(struct bpf_sock_ops_kern,
10427 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10428 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10429 si->dst_reg, si->dst_reg,
10430 offsetof(struct sk_buff, data));
10432 case offsetof(struct bpf_sock_ops, skb_len):
10433 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10435 si->dst_reg, si->src_reg,
10436 offsetof(struct bpf_sock_ops_kern,
10438 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10439 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10440 si->dst_reg, si->dst_reg,
10441 offsetof(struct sk_buff, len));
10443 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
10444 off = offsetof(struct sk_buff, cb);
10445 off += offsetof(struct tcp_skb_cb, tcp_flags);
10446 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
10447 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10449 si->dst_reg, si->src_reg,
10450 offsetof(struct bpf_sock_ops_kern,
10452 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10453 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
10455 si->dst_reg, si->dst_reg, off);
10457 case offsetof(struct bpf_sock_ops, skb_hwtstamp): {
10458 struct bpf_insn *jmp_on_null_skb;
10460 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10462 si->dst_reg, si->src_reg,
10463 offsetof(struct bpf_sock_ops_kern,
10465 /* Reserve one insn to test skb == NULL */
10466 jmp_on_null_skb = insn++;
10467 insn = bpf_convert_shinfo_access(si->dst_reg, si->dst_reg, insn);
10468 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
10469 bpf_target_off(struct skb_shared_info,
10472 *jmp_on_null_skb = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0,
10473 insn - jmp_on_null_skb - 1);
10477 return insn - insn_buf;
10480 /* data_end = skb->data + skb_headlen() */
10481 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
10482 struct bpf_insn *insn)
10485 int temp_reg_off = offsetof(struct sk_buff, cb) +
10486 offsetof(struct sk_skb_cb, temp_reg);
10488 if (si->src_reg == si->dst_reg) {
10489 /* We need an extra register, choose and save a register. */
10491 if (si->src_reg == reg || si->dst_reg == reg)
10493 if (si->src_reg == reg || si->dst_reg == reg)
10495 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off);
10500 /* reg = skb->data */
10501 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10503 offsetof(struct sk_buff, data));
10504 /* AX = skb->len */
10505 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10506 BPF_REG_AX, si->src_reg,
10507 offsetof(struct sk_buff, len));
10508 /* reg = skb->data + skb->len */
10509 *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX);
10510 /* AX = skb->data_len */
10511 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
10512 BPF_REG_AX, si->src_reg,
10513 offsetof(struct sk_buff, data_len));
10515 /* reg = skb->data + skb->len - skb->data_len */
10516 *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX);
10518 if (si->src_reg == si->dst_reg) {
10519 /* Restore the saved register */
10520 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg);
10521 *insn++ = BPF_MOV64_REG(si->dst_reg, reg);
10522 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off);
10528 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
10529 const struct bpf_insn *si,
10530 struct bpf_insn *insn_buf,
10531 struct bpf_prog *prog, u32 *target_size)
10533 struct bpf_insn *insn = insn_buf;
10537 case offsetof(struct __sk_buff, data_end):
10538 insn = bpf_convert_data_end_access(si, insn);
10540 case offsetof(struct __sk_buff, cb[0]) ...
10541 offsetofend(struct __sk_buff, cb[4]) - 1:
10542 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20);
10543 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
10544 offsetof(struct sk_skb_cb, data)) %
10547 prog->cb_access = 1;
10549 off -= offsetof(struct __sk_buff, cb[0]);
10550 off += offsetof(struct sk_buff, cb);
10551 off += offsetof(struct sk_skb_cb, data);
10552 if (type == BPF_WRITE)
10553 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
10556 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
10562 return bpf_convert_ctx_access(type, si, insn_buf, prog,
10566 return insn - insn_buf;
10569 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
10570 const struct bpf_insn *si,
10571 struct bpf_insn *insn_buf,
10572 struct bpf_prog *prog, u32 *target_size)
10574 struct bpf_insn *insn = insn_buf;
10575 #if IS_ENABLED(CONFIG_IPV6)
10579 /* convert ctx uses the fact sg element is first in struct */
10580 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
10583 case offsetof(struct sk_msg_md, data):
10584 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
10585 si->dst_reg, si->src_reg,
10586 offsetof(struct sk_msg, data));
10588 case offsetof(struct sk_msg_md, data_end):
10589 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
10590 si->dst_reg, si->src_reg,
10591 offsetof(struct sk_msg, data_end));
10593 case offsetof(struct sk_msg_md, family):
10594 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10596 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10597 struct sk_msg, sk),
10598 si->dst_reg, si->src_reg,
10599 offsetof(struct sk_msg, sk));
10600 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10601 offsetof(struct sock_common, skc_family));
10604 case offsetof(struct sk_msg_md, remote_ip4):
10605 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10607 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10608 struct sk_msg, sk),
10609 si->dst_reg, si->src_reg,
10610 offsetof(struct sk_msg, sk));
10611 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10612 offsetof(struct sock_common, skc_daddr));
10615 case offsetof(struct sk_msg_md, local_ip4):
10616 BUILD_BUG_ON(sizeof_field(struct sock_common,
10617 skc_rcv_saddr) != 4);
10619 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10620 struct sk_msg, sk),
10621 si->dst_reg, si->src_reg,
10622 offsetof(struct sk_msg, sk));
10623 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10624 offsetof(struct sock_common,
10628 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
10629 offsetof(struct sk_msg_md, remote_ip6[3]):
10630 #if IS_ENABLED(CONFIG_IPV6)
10631 BUILD_BUG_ON(sizeof_field(struct sock_common,
10632 skc_v6_daddr.s6_addr32[0]) != 4);
10635 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
10636 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10637 struct sk_msg, sk),
10638 si->dst_reg, si->src_reg,
10639 offsetof(struct sk_msg, sk));
10640 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10641 offsetof(struct sock_common,
10642 skc_v6_daddr.s6_addr32[0]) +
10645 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10649 case offsetof(struct sk_msg_md, local_ip6[0]) ...
10650 offsetof(struct sk_msg_md, local_ip6[3]):
10651 #if IS_ENABLED(CONFIG_IPV6)
10652 BUILD_BUG_ON(sizeof_field(struct sock_common,
10653 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10656 off -= offsetof(struct sk_msg_md, local_ip6[0]);
10657 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10658 struct sk_msg, sk),
10659 si->dst_reg, si->src_reg,
10660 offsetof(struct sk_msg, sk));
10661 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10662 offsetof(struct sock_common,
10663 skc_v6_rcv_saddr.s6_addr32[0]) +
10666 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10670 case offsetof(struct sk_msg_md, remote_port):
10671 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10673 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10674 struct sk_msg, sk),
10675 si->dst_reg, si->src_reg,
10676 offsetof(struct sk_msg, sk));
10677 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10678 offsetof(struct sock_common, skc_dport));
10679 #ifndef __BIG_ENDIAN_BITFIELD
10680 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10684 case offsetof(struct sk_msg_md, local_port):
10685 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10687 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10688 struct sk_msg, sk),
10689 si->dst_reg, si->src_reg,
10690 offsetof(struct sk_msg, sk));
10691 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10692 offsetof(struct sock_common, skc_num));
10695 case offsetof(struct sk_msg_md, size):
10696 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
10697 si->dst_reg, si->src_reg,
10698 offsetof(struct sk_msg_sg, size));
10701 case offsetof(struct sk_msg_md, sk):
10702 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
10703 si->dst_reg, si->src_reg,
10704 offsetof(struct sk_msg, sk));
10708 return insn - insn_buf;
10711 const struct bpf_verifier_ops sk_filter_verifier_ops = {
10712 .get_func_proto = sk_filter_func_proto,
10713 .is_valid_access = sk_filter_is_valid_access,
10714 .convert_ctx_access = bpf_convert_ctx_access,
10715 .gen_ld_abs = bpf_gen_ld_abs,
10718 const struct bpf_prog_ops sk_filter_prog_ops = {
10719 .test_run = bpf_prog_test_run_skb,
10722 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
10723 .get_func_proto = tc_cls_act_func_proto,
10724 .is_valid_access = tc_cls_act_is_valid_access,
10725 .convert_ctx_access = tc_cls_act_convert_ctx_access,
10726 .gen_prologue = tc_cls_act_prologue,
10727 .gen_ld_abs = bpf_gen_ld_abs,
10728 .btf_struct_access = tc_cls_act_btf_struct_access,
10731 const struct bpf_prog_ops tc_cls_act_prog_ops = {
10732 .test_run = bpf_prog_test_run_skb,
10735 const struct bpf_verifier_ops xdp_verifier_ops = {
10736 .get_func_proto = xdp_func_proto,
10737 .is_valid_access = xdp_is_valid_access,
10738 .convert_ctx_access = xdp_convert_ctx_access,
10739 .gen_prologue = bpf_noop_prologue,
10740 .btf_struct_access = xdp_btf_struct_access,
10743 const struct bpf_prog_ops xdp_prog_ops = {
10744 .test_run = bpf_prog_test_run_xdp,
10747 const struct bpf_verifier_ops cg_skb_verifier_ops = {
10748 .get_func_proto = cg_skb_func_proto,
10749 .is_valid_access = cg_skb_is_valid_access,
10750 .convert_ctx_access = bpf_convert_ctx_access,
10753 const struct bpf_prog_ops cg_skb_prog_ops = {
10754 .test_run = bpf_prog_test_run_skb,
10757 const struct bpf_verifier_ops lwt_in_verifier_ops = {
10758 .get_func_proto = lwt_in_func_proto,
10759 .is_valid_access = lwt_is_valid_access,
10760 .convert_ctx_access = bpf_convert_ctx_access,
10763 const struct bpf_prog_ops lwt_in_prog_ops = {
10764 .test_run = bpf_prog_test_run_skb,
10767 const struct bpf_verifier_ops lwt_out_verifier_ops = {
10768 .get_func_proto = lwt_out_func_proto,
10769 .is_valid_access = lwt_is_valid_access,
10770 .convert_ctx_access = bpf_convert_ctx_access,
10773 const struct bpf_prog_ops lwt_out_prog_ops = {
10774 .test_run = bpf_prog_test_run_skb,
10777 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
10778 .get_func_proto = lwt_xmit_func_proto,
10779 .is_valid_access = lwt_is_valid_access,
10780 .convert_ctx_access = bpf_convert_ctx_access,
10781 .gen_prologue = tc_cls_act_prologue,
10784 const struct bpf_prog_ops lwt_xmit_prog_ops = {
10785 .test_run = bpf_prog_test_run_skb,
10788 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
10789 .get_func_proto = lwt_seg6local_func_proto,
10790 .is_valid_access = lwt_is_valid_access,
10791 .convert_ctx_access = bpf_convert_ctx_access,
10794 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
10795 .test_run = bpf_prog_test_run_skb,
10798 const struct bpf_verifier_ops cg_sock_verifier_ops = {
10799 .get_func_proto = sock_filter_func_proto,
10800 .is_valid_access = sock_filter_is_valid_access,
10801 .convert_ctx_access = bpf_sock_convert_ctx_access,
10804 const struct bpf_prog_ops cg_sock_prog_ops = {
10807 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
10808 .get_func_proto = sock_addr_func_proto,
10809 .is_valid_access = sock_addr_is_valid_access,
10810 .convert_ctx_access = sock_addr_convert_ctx_access,
10813 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
10816 const struct bpf_verifier_ops sock_ops_verifier_ops = {
10817 .get_func_proto = sock_ops_func_proto,
10818 .is_valid_access = sock_ops_is_valid_access,
10819 .convert_ctx_access = sock_ops_convert_ctx_access,
10822 const struct bpf_prog_ops sock_ops_prog_ops = {
10825 const struct bpf_verifier_ops sk_skb_verifier_ops = {
10826 .get_func_proto = sk_skb_func_proto,
10827 .is_valid_access = sk_skb_is_valid_access,
10828 .convert_ctx_access = sk_skb_convert_ctx_access,
10829 .gen_prologue = sk_skb_prologue,
10832 const struct bpf_prog_ops sk_skb_prog_ops = {
10835 const struct bpf_verifier_ops sk_msg_verifier_ops = {
10836 .get_func_proto = sk_msg_func_proto,
10837 .is_valid_access = sk_msg_is_valid_access,
10838 .convert_ctx_access = sk_msg_convert_ctx_access,
10839 .gen_prologue = bpf_noop_prologue,
10842 const struct bpf_prog_ops sk_msg_prog_ops = {
10845 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
10846 .get_func_proto = flow_dissector_func_proto,
10847 .is_valid_access = flow_dissector_is_valid_access,
10848 .convert_ctx_access = flow_dissector_convert_ctx_access,
10851 const struct bpf_prog_ops flow_dissector_prog_ops = {
10852 .test_run = bpf_prog_test_run_flow_dissector,
10855 int sk_detach_filter(struct sock *sk)
10858 struct sk_filter *filter;
10860 if (sock_flag(sk, SOCK_FILTER_LOCKED))
10863 filter = rcu_dereference_protected(sk->sk_filter,
10864 lockdep_sock_is_held(sk));
10866 RCU_INIT_POINTER(sk->sk_filter, NULL);
10867 sk_filter_uncharge(sk, filter);
10873 EXPORT_SYMBOL_GPL(sk_detach_filter);
10875 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len)
10877 struct sock_fprog_kern *fprog;
10878 struct sk_filter *filter;
10881 sockopt_lock_sock(sk);
10882 filter = rcu_dereference_protected(sk->sk_filter,
10883 lockdep_sock_is_held(sk));
10887 /* We're copying the filter that has been originally attached,
10888 * so no conversion/decode needed anymore. eBPF programs that
10889 * have no original program cannot be dumped through this.
10892 fprog = filter->prog->orig_prog;
10898 /* User space only enquires number of filter blocks. */
10902 if (len < fprog->len)
10906 if (copy_to_sockptr(optval, fprog->filter, bpf_classic_proglen(fprog)))
10909 /* Instead of bytes, the API requests to return the number
10910 * of filter blocks.
10914 sockopt_release_sock(sk);
10919 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
10920 struct sock_reuseport *reuse,
10921 struct sock *sk, struct sk_buff *skb,
10922 struct sock *migrating_sk,
10925 reuse_kern->skb = skb;
10926 reuse_kern->sk = sk;
10927 reuse_kern->selected_sk = NULL;
10928 reuse_kern->migrating_sk = migrating_sk;
10929 reuse_kern->data_end = skb->data + skb_headlen(skb);
10930 reuse_kern->hash = hash;
10931 reuse_kern->reuseport_id = reuse->reuseport_id;
10932 reuse_kern->bind_inany = reuse->bind_inany;
10935 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
10936 struct bpf_prog *prog, struct sk_buff *skb,
10937 struct sock *migrating_sk,
10940 struct sk_reuseport_kern reuse_kern;
10941 enum sk_action action;
10943 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
10944 action = bpf_prog_run(prog, &reuse_kern);
10946 if (action == SK_PASS)
10947 return reuse_kern.selected_sk;
10949 return ERR_PTR(-ECONNREFUSED);
10952 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
10953 struct bpf_map *, map, void *, key, u32, flags)
10955 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
10956 struct sock_reuseport *reuse;
10957 struct sock *selected_sk;
10959 selected_sk = map->ops->map_lookup_elem(map, key);
10963 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
10965 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
10966 if (sk_is_refcounted(selected_sk))
10967 sock_put(selected_sk);
10969 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
10970 * The only (!reuse) case here is - the sk has already been
10971 * unhashed (e.g. by close()), so treat it as -ENOENT.
10973 * Other maps (e.g. sock_map) do not provide this guarantee and
10974 * the sk may never be in the reuseport group to begin with.
10976 return is_sockarray ? -ENOENT : -EINVAL;
10979 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
10980 struct sock *sk = reuse_kern->sk;
10982 if (sk->sk_protocol != selected_sk->sk_protocol)
10983 return -EPROTOTYPE;
10984 else if (sk->sk_family != selected_sk->sk_family)
10985 return -EAFNOSUPPORT;
10987 /* Catch all. Likely bound to a different sockaddr. */
10991 reuse_kern->selected_sk = selected_sk;
10996 static const struct bpf_func_proto sk_select_reuseport_proto = {
10997 .func = sk_select_reuseport,
10999 .ret_type = RET_INTEGER,
11000 .arg1_type = ARG_PTR_TO_CTX,
11001 .arg2_type = ARG_CONST_MAP_PTR,
11002 .arg3_type = ARG_PTR_TO_MAP_KEY,
11003 .arg4_type = ARG_ANYTHING,
11006 BPF_CALL_4(sk_reuseport_load_bytes,
11007 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11008 void *, to, u32, len)
11010 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
11013 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
11014 .func = sk_reuseport_load_bytes,
11016 .ret_type = RET_INTEGER,
11017 .arg1_type = ARG_PTR_TO_CTX,
11018 .arg2_type = ARG_ANYTHING,
11019 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
11020 .arg4_type = ARG_CONST_SIZE,
11023 BPF_CALL_5(sk_reuseport_load_bytes_relative,
11024 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11025 void *, to, u32, len, u32, start_header)
11027 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
11028 len, start_header);
11031 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
11032 .func = sk_reuseport_load_bytes_relative,
11034 .ret_type = RET_INTEGER,
11035 .arg1_type = ARG_PTR_TO_CTX,
11036 .arg2_type = ARG_ANYTHING,
11037 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
11038 .arg4_type = ARG_CONST_SIZE,
11039 .arg5_type = ARG_ANYTHING,
11042 static const struct bpf_func_proto *
11043 sk_reuseport_func_proto(enum bpf_func_id func_id,
11044 const struct bpf_prog *prog)
11047 case BPF_FUNC_sk_select_reuseport:
11048 return &sk_select_reuseport_proto;
11049 case BPF_FUNC_skb_load_bytes:
11050 return &sk_reuseport_load_bytes_proto;
11051 case BPF_FUNC_skb_load_bytes_relative:
11052 return &sk_reuseport_load_bytes_relative_proto;
11053 case BPF_FUNC_get_socket_cookie:
11054 return &bpf_get_socket_ptr_cookie_proto;
11055 case BPF_FUNC_ktime_get_coarse_ns:
11056 return &bpf_ktime_get_coarse_ns_proto;
11058 return bpf_base_func_proto(func_id);
11063 sk_reuseport_is_valid_access(int off, int size,
11064 enum bpf_access_type type,
11065 const struct bpf_prog *prog,
11066 struct bpf_insn_access_aux *info)
11068 const u32 size_default = sizeof(__u32);
11070 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
11071 off % size || type != BPF_READ)
11075 case offsetof(struct sk_reuseport_md, data):
11076 info->reg_type = PTR_TO_PACKET;
11077 return size == sizeof(__u64);
11079 case offsetof(struct sk_reuseport_md, data_end):
11080 info->reg_type = PTR_TO_PACKET_END;
11081 return size == sizeof(__u64);
11083 case offsetof(struct sk_reuseport_md, hash):
11084 return size == size_default;
11086 case offsetof(struct sk_reuseport_md, sk):
11087 info->reg_type = PTR_TO_SOCKET;
11088 return size == sizeof(__u64);
11090 case offsetof(struct sk_reuseport_md, migrating_sk):
11091 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
11092 return size == sizeof(__u64);
11094 /* Fields that allow narrowing */
11095 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
11096 if (size < sizeof_field(struct sk_buff, protocol))
11099 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
11100 case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
11101 case bpf_ctx_range(struct sk_reuseport_md, len):
11102 bpf_ctx_record_field_size(info, size_default);
11103 return bpf_ctx_narrow_access_ok(off, size, size_default);
11110 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
11111 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
11112 si->dst_reg, si->src_reg, \
11113 bpf_target_off(struct sk_reuseport_kern, F, \
11114 sizeof_field(struct sk_reuseport_kern, F), \
11118 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
11119 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
11124 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \
11125 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
11130 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
11131 const struct bpf_insn *si,
11132 struct bpf_insn *insn_buf,
11133 struct bpf_prog *prog,
11136 struct bpf_insn *insn = insn_buf;
11139 case offsetof(struct sk_reuseport_md, data):
11140 SK_REUSEPORT_LOAD_SKB_FIELD(data);
11143 case offsetof(struct sk_reuseport_md, len):
11144 SK_REUSEPORT_LOAD_SKB_FIELD(len);
11147 case offsetof(struct sk_reuseport_md, eth_protocol):
11148 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
11151 case offsetof(struct sk_reuseport_md, ip_protocol):
11152 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
11155 case offsetof(struct sk_reuseport_md, data_end):
11156 SK_REUSEPORT_LOAD_FIELD(data_end);
11159 case offsetof(struct sk_reuseport_md, hash):
11160 SK_REUSEPORT_LOAD_FIELD(hash);
11163 case offsetof(struct sk_reuseport_md, bind_inany):
11164 SK_REUSEPORT_LOAD_FIELD(bind_inany);
11167 case offsetof(struct sk_reuseport_md, sk):
11168 SK_REUSEPORT_LOAD_FIELD(sk);
11171 case offsetof(struct sk_reuseport_md, migrating_sk):
11172 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
11176 return insn - insn_buf;
11179 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
11180 .get_func_proto = sk_reuseport_func_proto,
11181 .is_valid_access = sk_reuseport_is_valid_access,
11182 .convert_ctx_access = sk_reuseport_convert_ctx_access,
11185 const struct bpf_prog_ops sk_reuseport_prog_ops = {
11188 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
11189 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
11191 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
11192 struct sock *, sk, u64, flags)
11194 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
11195 BPF_SK_LOOKUP_F_NO_REUSEPORT)))
11197 if (unlikely(sk && sk_is_refcounted(sk)))
11198 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
11199 if (unlikely(sk && sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN))
11200 return -ESOCKTNOSUPPORT; /* only accept TCP socket in LISTEN */
11201 if (unlikely(sk && sk_is_udp(sk) && sk->sk_state != TCP_CLOSE))
11202 return -ESOCKTNOSUPPORT; /* only accept UDP socket in CLOSE */
11204 /* Check if socket is suitable for packet L3/L4 protocol */
11205 if (sk && sk->sk_protocol != ctx->protocol)
11206 return -EPROTOTYPE;
11207 if (sk && sk->sk_family != ctx->family &&
11208 (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
11209 return -EAFNOSUPPORT;
11211 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
11214 /* Select socket as lookup result */
11215 ctx->selected_sk = sk;
11216 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
11220 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
11221 .func = bpf_sk_lookup_assign,
11223 .ret_type = RET_INTEGER,
11224 .arg1_type = ARG_PTR_TO_CTX,
11225 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL,
11226 .arg3_type = ARG_ANYTHING,
11229 static const struct bpf_func_proto *
11230 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
11233 case BPF_FUNC_perf_event_output:
11234 return &bpf_event_output_data_proto;
11235 case BPF_FUNC_sk_assign:
11236 return &bpf_sk_lookup_assign_proto;
11237 case BPF_FUNC_sk_release:
11238 return &bpf_sk_release_proto;
11240 return bpf_sk_base_func_proto(func_id);
11244 static bool sk_lookup_is_valid_access(int off, int size,
11245 enum bpf_access_type type,
11246 const struct bpf_prog *prog,
11247 struct bpf_insn_access_aux *info)
11249 if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
11251 if (off % size != 0)
11253 if (type != BPF_READ)
11257 case offsetof(struct bpf_sk_lookup, sk):
11258 info->reg_type = PTR_TO_SOCKET_OR_NULL;
11259 return size == sizeof(__u64);
11261 case bpf_ctx_range(struct bpf_sk_lookup, family):
11262 case bpf_ctx_range(struct bpf_sk_lookup, protocol):
11263 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
11264 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
11265 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
11266 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
11267 case bpf_ctx_range(struct bpf_sk_lookup, local_port):
11268 case bpf_ctx_range(struct bpf_sk_lookup, ingress_ifindex):
11269 bpf_ctx_record_field_size(info, sizeof(__u32));
11270 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
11272 case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
11273 /* Allow 4-byte access to 2-byte field for backward compatibility */
11274 if (size == sizeof(__u32))
11276 bpf_ctx_record_field_size(info, sizeof(__be16));
11277 return bpf_ctx_narrow_access_ok(off, size, sizeof(__be16));
11279 case offsetofend(struct bpf_sk_lookup, remote_port) ...
11280 offsetof(struct bpf_sk_lookup, local_ip4) - 1:
11281 /* Allow access to zero padding for backward compatibility */
11282 bpf_ctx_record_field_size(info, sizeof(__u16));
11283 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u16));
11290 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
11291 const struct bpf_insn *si,
11292 struct bpf_insn *insn_buf,
11293 struct bpf_prog *prog,
11296 struct bpf_insn *insn = insn_buf;
11299 case offsetof(struct bpf_sk_lookup, sk):
11300 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11301 offsetof(struct bpf_sk_lookup_kern, selected_sk));
11304 case offsetof(struct bpf_sk_lookup, family):
11305 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11306 bpf_target_off(struct bpf_sk_lookup_kern,
11307 family, 2, target_size));
11310 case offsetof(struct bpf_sk_lookup, protocol):
11311 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11312 bpf_target_off(struct bpf_sk_lookup_kern,
11313 protocol, 2, target_size));
11316 case offsetof(struct bpf_sk_lookup, remote_ip4):
11317 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11318 bpf_target_off(struct bpf_sk_lookup_kern,
11319 v4.saddr, 4, target_size));
11322 case offsetof(struct bpf_sk_lookup, local_ip4):
11323 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11324 bpf_target_off(struct bpf_sk_lookup_kern,
11325 v4.daddr, 4, target_size));
11328 case bpf_ctx_range_till(struct bpf_sk_lookup,
11329 remote_ip6[0], remote_ip6[3]): {
11330 #if IS_ENABLED(CONFIG_IPV6)
11333 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
11334 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11335 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11336 offsetof(struct bpf_sk_lookup_kern, v6.saddr));
11337 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11338 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11340 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11344 case bpf_ctx_range_till(struct bpf_sk_lookup,
11345 local_ip6[0], local_ip6[3]): {
11346 #if IS_ENABLED(CONFIG_IPV6)
11349 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
11350 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11351 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11352 offsetof(struct bpf_sk_lookup_kern, v6.daddr));
11353 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11354 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11356 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11360 case offsetof(struct bpf_sk_lookup, remote_port):
11361 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11362 bpf_target_off(struct bpf_sk_lookup_kern,
11363 sport, 2, target_size));
11366 case offsetofend(struct bpf_sk_lookup, remote_port):
11368 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11371 case offsetof(struct bpf_sk_lookup, local_port):
11372 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11373 bpf_target_off(struct bpf_sk_lookup_kern,
11374 dport, 2, target_size));
11377 case offsetof(struct bpf_sk_lookup, ingress_ifindex):
11378 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11379 bpf_target_off(struct bpf_sk_lookup_kern,
11380 ingress_ifindex, 4, target_size));
11384 return insn - insn_buf;
11387 const struct bpf_prog_ops sk_lookup_prog_ops = {
11388 .test_run = bpf_prog_test_run_sk_lookup,
11391 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
11392 .get_func_proto = sk_lookup_func_proto,
11393 .is_valid_access = sk_lookup_is_valid_access,
11394 .convert_ctx_access = sk_lookup_convert_ctx_access,
11397 #endif /* CONFIG_INET */
11399 DEFINE_BPF_DISPATCHER(xdp)
11401 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
11403 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
11406 BTF_ID_LIST_GLOBAL(btf_sock_ids, MAX_BTF_SOCK_TYPE)
11407 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
11409 #undef BTF_SOCK_TYPE
11411 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
11413 /* tcp6_sock type is not generated in dwarf and hence btf,
11414 * trigger an explicit type generation here.
11416 BTF_TYPE_EMIT(struct tcp6_sock);
11417 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
11418 sk->sk_family == AF_INET6)
11419 return (unsigned long)sk;
11421 return (unsigned long)NULL;
11424 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
11425 .func = bpf_skc_to_tcp6_sock,
11427 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11428 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11429 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
11432 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
11434 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
11435 return (unsigned long)sk;
11437 return (unsigned long)NULL;
11440 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
11441 .func = bpf_skc_to_tcp_sock,
11443 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11444 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11445 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
11448 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
11450 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
11451 * generated if CONFIG_INET=n. Trigger an explicit generation here.
11453 BTF_TYPE_EMIT(struct inet_timewait_sock);
11454 BTF_TYPE_EMIT(struct tcp_timewait_sock);
11457 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
11458 return (unsigned long)sk;
11461 #if IS_BUILTIN(CONFIG_IPV6)
11462 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
11463 return (unsigned long)sk;
11466 return (unsigned long)NULL;
11469 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
11470 .func = bpf_skc_to_tcp_timewait_sock,
11472 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11473 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11474 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
11477 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
11480 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11481 return (unsigned long)sk;
11484 #if IS_BUILTIN(CONFIG_IPV6)
11485 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11486 return (unsigned long)sk;
11489 return (unsigned long)NULL;
11492 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
11493 .func = bpf_skc_to_tcp_request_sock,
11495 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11496 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11497 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
11500 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
11502 /* udp6_sock type is not generated in dwarf and hence btf,
11503 * trigger an explicit type generation here.
11505 BTF_TYPE_EMIT(struct udp6_sock);
11506 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
11507 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
11508 return (unsigned long)sk;
11510 return (unsigned long)NULL;
11513 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
11514 .func = bpf_skc_to_udp6_sock,
11516 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11517 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11518 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
11521 BPF_CALL_1(bpf_skc_to_unix_sock, struct sock *, sk)
11523 /* unix_sock type is not generated in dwarf and hence btf,
11524 * trigger an explicit type generation here.
11526 BTF_TYPE_EMIT(struct unix_sock);
11527 if (sk && sk_fullsock(sk) && sk->sk_family == AF_UNIX)
11528 return (unsigned long)sk;
11530 return (unsigned long)NULL;
11533 const struct bpf_func_proto bpf_skc_to_unix_sock_proto = {
11534 .func = bpf_skc_to_unix_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_UNIX],
11541 BPF_CALL_1(bpf_skc_to_mptcp_sock, struct sock *, sk)
11543 BTF_TYPE_EMIT(struct mptcp_sock);
11544 return (unsigned long)bpf_mptcp_sock_from_subflow(sk);
11547 const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto = {
11548 .func = bpf_skc_to_mptcp_sock,
11550 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11551 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
11552 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_MPTCP],
11555 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
11557 return (unsigned long)sock_from_file(file);
11560 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
11561 BTF_ID(struct, socket)
11562 BTF_ID(struct, file)
11564 const struct bpf_func_proto bpf_sock_from_file_proto = {
11565 .func = bpf_sock_from_file,
11567 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11568 .ret_btf_id = &bpf_sock_from_file_btf_ids[0],
11569 .arg1_type = ARG_PTR_TO_BTF_ID,
11570 .arg1_btf_id = &bpf_sock_from_file_btf_ids[1],
11573 static const struct bpf_func_proto *
11574 bpf_sk_base_func_proto(enum bpf_func_id func_id)
11576 const struct bpf_func_proto *func;
11579 case BPF_FUNC_skc_to_tcp6_sock:
11580 func = &bpf_skc_to_tcp6_sock_proto;
11582 case BPF_FUNC_skc_to_tcp_sock:
11583 func = &bpf_skc_to_tcp_sock_proto;
11585 case BPF_FUNC_skc_to_tcp_timewait_sock:
11586 func = &bpf_skc_to_tcp_timewait_sock_proto;
11588 case BPF_FUNC_skc_to_tcp_request_sock:
11589 func = &bpf_skc_to_tcp_request_sock_proto;
11591 case BPF_FUNC_skc_to_udp6_sock:
11592 func = &bpf_skc_to_udp6_sock_proto;
11594 case BPF_FUNC_skc_to_unix_sock:
11595 func = &bpf_skc_to_unix_sock_proto;
11597 case BPF_FUNC_skc_to_mptcp_sock:
11598 func = &bpf_skc_to_mptcp_sock_proto;
11600 case BPF_FUNC_ktime_get_coarse_ns:
11601 return &bpf_ktime_get_coarse_ns_proto;
11603 return bpf_base_func_proto(func_id);
11606 if (!perfmon_capable())