1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Linux Socket Filter - Kernel level socket filtering
5 * Based on the design of the Berkeley Packet Filter. The new
6 * internal format has been designed by PLUMgrid:
8 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
12 * Jay Schulist <jschlst@samba.org>
13 * Alexei Starovoitov <ast@plumgrid.com>
14 * Daniel Borkmann <dborkman@redhat.com>
16 * Andi Kleen - Fix a few bad bugs and races.
17 * Kris Katterjohn - Added many additional checks in bpf_check_classic()
20 #include <linux/atomic.h>
21 #include <linux/bpf_verifier.h>
22 #include <linux/module.h>
23 #include <linux/types.h>
25 #include <linux/fcntl.h>
26 #include <linux/socket.h>
27 #include <linux/sock_diag.h>
29 #include <linux/inet.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_packet.h>
32 #include <linux/if_arp.h>
33 #include <linux/gfp.h>
34 #include <net/inet_common.h>
36 #include <net/protocol.h>
37 #include <net/netlink.h>
38 #include <linux/skbuff.h>
39 #include <linux/skmsg.h>
41 #include <net/flow_dissector.h>
42 #include <linux/errno.h>
43 #include <linux/timer.h>
44 #include <linux/uaccess.h>
45 #include <asm/unaligned.h>
46 #include <linux/filter.h>
47 #include <linux/ratelimit.h>
48 #include <linux/seccomp.h>
49 #include <linux/if_vlan.h>
50 #include <linux/bpf.h>
51 #include <linux/btf.h>
52 #include <net/sch_generic.h>
53 #include <net/cls_cgroup.h>
54 #include <net/dst_metadata.h>
56 #include <net/sock_reuseport.h>
57 #include <net/busy_poll.h>
61 #include <linux/bpf_trace.h>
62 #include <net/xdp_sock.h>
63 #include <linux/inetdevice.h>
64 #include <net/inet_hashtables.h>
65 #include <net/inet6_hashtables.h>
66 #include <net/ip_fib.h>
67 #include <net/nexthop.h>
71 #include <net/net_namespace.h>
72 #include <linux/seg6_local.h>
74 #include <net/seg6_local.h>
75 #include <net/lwtunnel.h>
76 #include <net/ipv6_stubs.h>
77 #include <net/bpf_sk_storage.h>
78 #include <net/transp_v6.h>
79 #include <linux/btf_ids.h>
82 #include <net/mptcp.h>
83 #include <net/netfilter/nf_conntrack_bpf.h>
85 static const struct bpf_func_proto *
86 bpf_sk_base_func_proto(enum bpf_func_id func_id);
88 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len)
90 if (in_compat_syscall()) {
91 struct compat_sock_fprog f32;
93 if (len != sizeof(f32))
95 if (copy_from_sockptr(&f32, src, sizeof(f32)))
97 memset(dst, 0, sizeof(*dst));
99 dst->filter = compat_ptr(f32.filter);
101 if (len != sizeof(*dst))
103 if (copy_from_sockptr(dst, src, sizeof(*dst)))
109 EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user);
112 * sk_filter_trim_cap - run a packet through a socket filter
113 * @sk: sock associated with &sk_buff
114 * @skb: buffer to filter
115 * @cap: limit on how short the eBPF program may trim the packet
117 * Run the eBPF program and then cut skb->data to correct size returned by
118 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
119 * than pkt_len we keep whole skb->data. This is the socket level
120 * wrapper to bpf_prog_run. It returns 0 if the packet should
121 * be accepted or -EPERM if the packet should be tossed.
124 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
127 struct sk_filter *filter;
130 * If the skb was allocated from pfmemalloc reserves, only
131 * allow SOCK_MEMALLOC sockets to use it as this socket is
132 * helping free memory
134 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
135 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
138 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
142 err = security_sock_rcv_skb(sk, skb);
147 filter = rcu_dereference(sk->sk_filter);
149 struct sock *save_sk = skb->sk;
150 unsigned int pkt_len;
153 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
155 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
161 EXPORT_SYMBOL(sk_filter_trim_cap);
163 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
165 return skb_get_poff(skb);
168 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
172 if (skb_is_nonlinear(skb))
175 if (skb->len < sizeof(struct nlattr))
178 if (a > skb->len - sizeof(struct nlattr))
181 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
183 return (void *) nla - (void *) skb->data;
188 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
192 if (skb_is_nonlinear(skb))
195 if (skb->len < sizeof(struct nlattr))
198 if (a > skb->len - sizeof(struct nlattr))
201 nla = (struct nlattr *) &skb->data[a];
202 if (nla->nla_len > skb->len - a)
205 nla = nla_find_nested(nla, x);
207 return (void *) nla - (void *) skb->data;
212 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
213 data, int, headlen, int, offset)
216 const int len = sizeof(tmp);
219 if (headlen - offset >= len)
220 return *(u8 *)(data + offset);
221 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
224 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
232 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
235 return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
239 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
240 data, int, headlen, int, offset)
243 const int len = sizeof(tmp);
246 if (headlen - offset >= len)
247 return get_unaligned_be16(data + offset);
248 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
249 return be16_to_cpu(tmp);
251 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
253 return get_unaligned_be16(ptr);
259 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
262 return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
266 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
267 data, int, headlen, int, offset)
270 const int len = sizeof(tmp);
272 if (likely(offset >= 0)) {
273 if (headlen - offset >= len)
274 return get_unaligned_be32(data + offset);
275 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
276 return be32_to_cpu(tmp);
278 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
280 return get_unaligned_be32(ptr);
286 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
289 return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
293 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
294 struct bpf_insn *insn_buf)
296 struct bpf_insn *insn = insn_buf;
300 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4);
302 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
303 offsetof(struct sk_buff, mark));
307 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET);
308 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
309 #ifdef __BIG_ENDIAN_BITFIELD
310 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
315 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2);
317 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
318 offsetof(struct sk_buff, queue_mapping));
321 case SKF_AD_VLAN_TAG:
322 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2);
324 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
325 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
326 offsetof(struct sk_buff, vlan_tci));
328 case SKF_AD_VLAN_TAG_PRESENT:
329 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_VLAN_PRESENT_OFFSET);
330 if (PKT_VLAN_PRESENT_BIT)
331 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, PKT_VLAN_PRESENT_BIT);
332 if (PKT_VLAN_PRESENT_BIT < 7)
333 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
337 return insn - insn_buf;
340 static bool convert_bpf_extensions(struct sock_filter *fp,
341 struct bpf_insn **insnp)
343 struct bpf_insn *insn = *insnp;
347 case SKF_AD_OFF + SKF_AD_PROTOCOL:
348 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2);
350 /* A = *(u16 *) (CTX + offsetof(protocol)) */
351 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
352 offsetof(struct sk_buff, protocol));
353 /* A = ntohs(A) [emitting a nop or swap16] */
354 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
357 case SKF_AD_OFF + SKF_AD_PKTTYPE:
358 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
362 case SKF_AD_OFF + SKF_AD_IFINDEX:
363 case SKF_AD_OFF + SKF_AD_HATYPE:
364 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4);
365 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2);
367 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
368 BPF_REG_TMP, BPF_REG_CTX,
369 offsetof(struct sk_buff, dev));
370 /* if (tmp != 0) goto pc + 1 */
371 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
372 *insn++ = BPF_EXIT_INSN();
373 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
374 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
375 offsetof(struct net_device, ifindex));
377 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
378 offsetof(struct net_device, type));
381 case SKF_AD_OFF + SKF_AD_MARK:
382 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
386 case SKF_AD_OFF + SKF_AD_RXHASH:
387 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4);
389 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
390 offsetof(struct sk_buff, hash));
393 case SKF_AD_OFF + SKF_AD_QUEUE:
394 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
398 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
399 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
400 BPF_REG_A, BPF_REG_CTX, insn);
404 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
405 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
406 BPF_REG_A, BPF_REG_CTX, insn);
410 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
411 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2);
413 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
414 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
415 offsetof(struct sk_buff, vlan_proto));
416 /* A = ntohs(A) [emitting a nop or swap16] */
417 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
420 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
421 case SKF_AD_OFF + SKF_AD_NLATTR:
422 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
423 case SKF_AD_OFF + SKF_AD_CPU:
424 case SKF_AD_OFF + SKF_AD_RANDOM:
426 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
428 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
430 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
431 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
433 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
434 *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
436 case SKF_AD_OFF + SKF_AD_NLATTR:
437 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
439 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
440 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
442 case SKF_AD_OFF + SKF_AD_CPU:
443 *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
445 case SKF_AD_OFF + SKF_AD_RANDOM:
446 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
447 bpf_user_rnd_init_once();
452 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
454 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
458 /* This is just a dummy call to avoid letting the compiler
459 * evict __bpf_call_base() as an optimization. Placed here
460 * where no-one bothers.
462 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
470 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
472 const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
473 int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
474 bool endian = BPF_SIZE(fp->code) == BPF_H ||
475 BPF_SIZE(fp->code) == BPF_W;
476 bool indirect = BPF_MODE(fp->code) == BPF_IND;
477 const int ip_align = NET_IP_ALIGN;
478 struct bpf_insn *insn = *insnp;
482 ((unaligned_ok && offset >= 0) ||
483 (!unaligned_ok && offset >= 0 &&
484 offset + ip_align >= 0 &&
485 offset + ip_align % size == 0))) {
486 bool ldx_off_ok = offset <= S16_MAX;
488 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
490 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
491 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
492 size, 2 + endian + (!ldx_off_ok * 2));
494 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
497 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
498 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
499 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
503 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
504 *insn++ = BPF_JMP_A(8);
507 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
508 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
509 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
511 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
513 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
515 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
518 switch (BPF_SIZE(fp->code)) {
520 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
523 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
526 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
532 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
533 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
534 *insn = BPF_EXIT_INSN();
541 * bpf_convert_filter - convert filter program
542 * @prog: the user passed filter program
543 * @len: the length of the user passed filter program
544 * @new_prog: allocated 'struct bpf_prog' or NULL
545 * @new_len: pointer to store length of converted program
546 * @seen_ld_abs: bool whether we've seen ld_abs/ind
548 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
549 * style extended BPF (eBPF).
550 * Conversion workflow:
552 * 1) First pass for calculating the new program length:
553 * bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
555 * 2) 2nd pass to remap in two passes: 1st pass finds new
556 * jump offsets, 2nd pass remapping:
557 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
559 static int bpf_convert_filter(struct sock_filter *prog, int len,
560 struct bpf_prog *new_prog, int *new_len,
563 int new_flen = 0, pass = 0, target, i, stack_off;
564 struct bpf_insn *new_insn, *first_insn = NULL;
565 struct sock_filter *fp;
569 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
570 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
572 if (len <= 0 || len > BPF_MAXINSNS)
576 first_insn = new_prog->insnsi;
577 addrs = kcalloc(len, sizeof(*addrs),
578 GFP_KERNEL | __GFP_NOWARN);
584 new_insn = first_insn;
587 /* Classic BPF related prologue emission. */
589 /* Classic BPF expects A and X to be reset first. These need
590 * to be guaranteed to be the first two instructions.
592 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
593 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
595 /* All programs must keep CTX in callee saved BPF_REG_CTX.
596 * In eBPF case it's done by the compiler, here we need to
597 * do this ourself. Initial CTX is present in BPF_REG_ARG1.
599 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
601 /* For packet access in classic BPF, cache skb->data
602 * in callee-saved BPF R8 and skb->len - skb->data_len
603 * (headlen) in BPF R9. Since classic BPF is read-only
604 * on CTX, we only need to cache it once.
606 *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
607 BPF_REG_D, BPF_REG_CTX,
608 offsetof(struct sk_buff, data));
609 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
610 offsetof(struct sk_buff, len));
611 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
612 offsetof(struct sk_buff, data_len));
613 *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
619 for (i = 0; i < len; fp++, i++) {
620 struct bpf_insn tmp_insns[32] = { };
621 struct bpf_insn *insn = tmp_insns;
624 addrs[i] = new_insn - first_insn;
627 /* All arithmetic insns and skb loads map as-is. */
628 case BPF_ALU | BPF_ADD | BPF_X:
629 case BPF_ALU | BPF_ADD | BPF_K:
630 case BPF_ALU | BPF_SUB | BPF_X:
631 case BPF_ALU | BPF_SUB | BPF_K:
632 case BPF_ALU | BPF_AND | BPF_X:
633 case BPF_ALU | BPF_AND | BPF_K:
634 case BPF_ALU | BPF_OR | BPF_X:
635 case BPF_ALU | BPF_OR | BPF_K:
636 case BPF_ALU | BPF_LSH | BPF_X:
637 case BPF_ALU | BPF_LSH | BPF_K:
638 case BPF_ALU | BPF_RSH | BPF_X:
639 case BPF_ALU | BPF_RSH | BPF_K:
640 case BPF_ALU | BPF_XOR | BPF_X:
641 case BPF_ALU | BPF_XOR | BPF_K:
642 case BPF_ALU | BPF_MUL | BPF_X:
643 case BPF_ALU | BPF_MUL | BPF_K:
644 case BPF_ALU | BPF_DIV | BPF_X:
645 case BPF_ALU | BPF_DIV | BPF_K:
646 case BPF_ALU | BPF_MOD | BPF_X:
647 case BPF_ALU | BPF_MOD | BPF_K:
648 case BPF_ALU | BPF_NEG:
649 case BPF_LD | BPF_ABS | BPF_W:
650 case BPF_LD | BPF_ABS | BPF_H:
651 case BPF_LD | BPF_ABS | BPF_B:
652 case BPF_LD | BPF_IND | BPF_W:
653 case BPF_LD | BPF_IND | BPF_H:
654 case BPF_LD | BPF_IND | BPF_B:
655 /* Check for overloaded BPF extension and
656 * directly convert it if found, otherwise
657 * just move on with mapping.
659 if (BPF_CLASS(fp->code) == BPF_LD &&
660 BPF_MODE(fp->code) == BPF_ABS &&
661 convert_bpf_extensions(fp, &insn))
663 if (BPF_CLASS(fp->code) == BPF_LD &&
664 convert_bpf_ld_abs(fp, &insn)) {
669 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
670 fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
671 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
672 /* Error with exception code on div/mod by 0.
673 * For cBPF programs, this was always return 0.
675 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
676 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
677 *insn++ = BPF_EXIT_INSN();
680 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
683 /* Jump transformation cannot use BPF block macros
684 * everywhere as offset calculation and target updates
685 * require a bit more work than the rest, i.e. jump
686 * opcodes map as-is, but offsets need adjustment.
689 #define BPF_EMIT_JMP \
691 const s32 off_min = S16_MIN, off_max = S16_MAX; \
694 if (target >= len || target < 0) \
696 off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
697 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
698 off -= insn - tmp_insns; \
699 /* Reject anything not fitting into insn->off. */ \
700 if (off < off_min || off > off_max) \
705 case BPF_JMP | BPF_JA:
706 target = i + fp->k + 1;
707 insn->code = fp->code;
711 case BPF_JMP | BPF_JEQ | BPF_K:
712 case BPF_JMP | BPF_JEQ | BPF_X:
713 case BPF_JMP | BPF_JSET | BPF_K:
714 case BPF_JMP | BPF_JSET | BPF_X:
715 case BPF_JMP | BPF_JGT | BPF_K:
716 case BPF_JMP | BPF_JGT | BPF_X:
717 case BPF_JMP | BPF_JGE | BPF_K:
718 case BPF_JMP | BPF_JGE | BPF_X:
719 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
720 /* BPF immediates are signed, zero extend
721 * immediate into tmp register and use it
724 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
726 insn->dst_reg = BPF_REG_A;
727 insn->src_reg = BPF_REG_TMP;
730 insn->dst_reg = BPF_REG_A;
732 bpf_src = BPF_SRC(fp->code);
733 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
736 /* Common case where 'jump_false' is next insn. */
738 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
739 target = i + fp->jt + 1;
744 /* Convert some jumps when 'jump_true' is next insn. */
746 switch (BPF_OP(fp->code)) {
748 insn->code = BPF_JMP | BPF_JNE | bpf_src;
751 insn->code = BPF_JMP | BPF_JLE | bpf_src;
754 insn->code = BPF_JMP | BPF_JLT | bpf_src;
760 target = i + fp->jf + 1;
765 /* Other jumps are mapped into two insns: Jxx and JA. */
766 target = i + fp->jt + 1;
767 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
771 insn->code = BPF_JMP | BPF_JA;
772 target = i + fp->jf + 1;
776 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
777 case BPF_LDX | BPF_MSH | BPF_B: {
778 struct sock_filter tmp = {
779 .code = BPF_LD | BPF_ABS | BPF_B,
786 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
787 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
788 convert_bpf_ld_abs(&tmp, &insn);
791 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
793 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
795 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
797 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
799 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
802 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
803 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
805 case BPF_RET | BPF_A:
806 case BPF_RET | BPF_K:
807 if (BPF_RVAL(fp->code) == BPF_K)
808 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
810 *insn = BPF_EXIT_INSN();
813 /* Store to stack. */
816 stack_off = fp->k * 4 + 4;
817 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
818 BPF_ST ? BPF_REG_A : BPF_REG_X,
820 /* check_load_and_stores() verifies that classic BPF can
821 * load from stack only after write, so tracking
822 * stack_depth for ST|STX insns is enough
824 if (new_prog && new_prog->aux->stack_depth < stack_off)
825 new_prog->aux->stack_depth = stack_off;
828 /* Load from stack. */
829 case BPF_LD | BPF_MEM:
830 case BPF_LDX | BPF_MEM:
831 stack_off = fp->k * 4 + 4;
832 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
833 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
838 case BPF_LD | BPF_IMM:
839 case BPF_LDX | BPF_IMM:
840 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
841 BPF_REG_A : BPF_REG_X, fp->k);
845 case BPF_MISC | BPF_TAX:
846 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
850 case BPF_MISC | BPF_TXA:
851 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
854 /* A = skb->len or X = skb->len */
855 case BPF_LD | BPF_W | BPF_LEN:
856 case BPF_LDX | BPF_W | BPF_LEN:
857 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
858 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
859 offsetof(struct sk_buff, len));
862 /* Access seccomp_data fields. */
863 case BPF_LDX | BPF_ABS | BPF_W:
864 /* A = *(u32 *) (ctx + K) */
865 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
868 /* Unknown instruction. */
875 memcpy(new_insn, tmp_insns,
876 sizeof(*insn) * (insn - tmp_insns));
877 new_insn += insn - tmp_insns;
881 /* Only calculating new length. */
882 *new_len = new_insn - first_insn;
884 *new_len += 4; /* Prologue bits. */
889 if (new_flen != new_insn - first_insn) {
890 new_flen = new_insn - first_insn;
897 BUG_ON(*new_len != new_flen);
906 * As we dont want to clear mem[] array for each packet going through
907 * __bpf_prog_run(), we check that filter loaded by user never try to read
908 * a cell if not previously written, and we check all branches to be sure
909 * a malicious user doesn't try to abuse us.
911 static int check_load_and_stores(const struct sock_filter *filter, int flen)
913 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
916 BUILD_BUG_ON(BPF_MEMWORDS > 16);
918 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
922 memset(masks, 0xff, flen * sizeof(*masks));
924 for (pc = 0; pc < flen; pc++) {
925 memvalid &= masks[pc];
927 switch (filter[pc].code) {
930 memvalid |= (1 << filter[pc].k);
932 case BPF_LD | BPF_MEM:
933 case BPF_LDX | BPF_MEM:
934 if (!(memvalid & (1 << filter[pc].k))) {
939 case BPF_JMP | BPF_JA:
940 /* A jump must set masks on target */
941 masks[pc + 1 + filter[pc].k] &= memvalid;
944 case BPF_JMP | BPF_JEQ | BPF_K:
945 case BPF_JMP | BPF_JEQ | BPF_X:
946 case BPF_JMP | BPF_JGE | BPF_K:
947 case BPF_JMP | BPF_JGE | BPF_X:
948 case BPF_JMP | BPF_JGT | BPF_K:
949 case BPF_JMP | BPF_JGT | BPF_X:
950 case BPF_JMP | BPF_JSET | BPF_K:
951 case BPF_JMP | BPF_JSET | BPF_X:
952 /* A jump must set masks on targets */
953 masks[pc + 1 + filter[pc].jt] &= memvalid;
954 masks[pc + 1 + filter[pc].jf] &= memvalid;
964 static bool chk_code_allowed(u16 code_to_probe)
966 static const bool codes[] = {
967 /* 32 bit ALU operations */
968 [BPF_ALU | BPF_ADD | BPF_K] = true,
969 [BPF_ALU | BPF_ADD | BPF_X] = true,
970 [BPF_ALU | BPF_SUB | BPF_K] = true,
971 [BPF_ALU | BPF_SUB | BPF_X] = true,
972 [BPF_ALU | BPF_MUL | BPF_K] = true,
973 [BPF_ALU | BPF_MUL | BPF_X] = true,
974 [BPF_ALU | BPF_DIV | BPF_K] = true,
975 [BPF_ALU | BPF_DIV | BPF_X] = true,
976 [BPF_ALU | BPF_MOD | BPF_K] = true,
977 [BPF_ALU | BPF_MOD | BPF_X] = true,
978 [BPF_ALU | BPF_AND | BPF_K] = true,
979 [BPF_ALU | BPF_AND | BPF_X] = true,
980 [BPF_ALU | BPF_OR | BPF_K] = true,
981 [BPF_ALU | BPF_OR | BPF_X] = true,
982 [BPF_ALU | BPF_XOR | BPF_K] = true,
983 [BPF_ALU | BPF_XOR | BPF_X] = true,
984 [BPF_ALU | BPF_LSH | BPF_K] = true,
985 [BPF_ALU | BPF_LSH | BPF_X] = true,
986 [BPF_ALU | BPF_RSH | BPF_K] = true,
987 [BPF_ALU | BPF_RSH | BPF_X] = true,
988 [BPF_ALU | BPF_NEG] = true,
989 /* Load instructions */
990 [BPF_LD | BPF_W | BPF_ABS] = true,
991 [BPF_LD | BPF_H | BPF_ABS] = true,
992 [BPF_LD | BPF_B | BPF_ABS] = true,
993 [BPF_LD | BPF_W | BPF_LEN] = true,
994 [BPF_LD | BPF_W | BPF_IND] = true,
995 [BPF_LD | BPF_H | BPF_IND] = true,
996 [BPF_LD | BPF_B | BPF_IND] = true,
997 [BPF_LD | BPF_IMM] = true,
998 [BPF_LD | BPF_MEM] = true,
999 [BPF_LDX | BPF_W | BPF_LEN] = true,
1000 [BPF_LDX | BPF_B | BPF_MSH] = true,
1001 [BPF_LDX | BPF_IMM] = true,
1002 [BPF_LDX | BPF_MEM] = true,
1003 /* Store instructions */
1006 /* Misc instructions */
1007 [BPF_MISC | BPF_TAX] = true,
1008 [BPF_MISC | BPF_TXA] = true,
1009 /* Return instructions */
1010 [BPF_RET | BPF_K] = true,
1011 [BPF_RET | BPF_A] = true,
1012 /* Jump instructions */
1013 [BPF_JMP | BPF_JA] = true,
1014 [BPF_JMP | BPF_JEQ | BPF_K] = true,
1015 [BPF_JMP | BPF_JEQ | BPF_X] = true,
1016 [BPF_JMP | BPF_JGE | BPF_K] = true,
1017 [BPF_JMP | BPF_JGE | BPF_X] = true,
1018 [BPF_JMP | BPF_JGT | BPF_K] = true,
1019 [BPF_JMP | BPF_JGT | BPF_X] = true,
1020 [BPF_JMP | BPF_JSET | BPF_K] = true,
1021 [BPF_JMP | BPF_JSET | BPF_X] = true,
1024 if (code_to_probe >= ARRAY_SIZE(codes))
1027 return codes[code_to_probe];
1030 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1035 if (flen == 0 || flen > BPF_MAXINSNS)
1042 * bpf_check_classic - verify socket filter code
1043 * @filter: filter to verify
1044 * @flen: length of filter
1046 * Check the user's filter code. If we let some ugly
1047 * filter code slip through kaboom! The filter must contain
1048 * no references or jumps that are out of range, no illegal
1049 * instructions, and must end with a RET instruction.
1051 * All jumps are forward as they are not signed.
1053 * Returns 0 if the rule set is legal or -EINVAL if not.
1055 static int bpf_check_classic(const struct sock_filter *filter,
1061 /* Check the filter code now */
1062 for (pc = 0; pc < flen; pc++) {
1063 const struct sock_filter *ftest = &filter[pc];
1065 /* May we actually operate on this code? */
1066 if (!chk_code_allowed(ftest->code))
1069 /* Some instructions need special checks */
1070 switch (ftest->code) {
1071 case BPF_ALU | BPF_DIV | BPF_K:
1072 case BPF_ALU | BPF_MOD | BPF_K:
1073 /* Check for division by zero */
1077 case BPF_ALU | BPF_LSH | BPF_K:
1078 case BPF_ALU | BPF_RSH | BPF_K:
1082 case BPF_LD | BPF_MEM:
1083 case BPF_LDX | BPF_MEM:
1086 /* Check for invalid memory addresses */
1087 if (ftest->k >= BPF_MEMWORDS)
1090 case BPF_JMP | BPF_JA:
1091 /* Note, the large ftest->k might cause loops.
1092 * Compare this with conditional jumps below,
1093 * where offsets are limited. --ANK (981016)
1095 if (ftest->k >= (unsigned int)(flen - pc - 1))
1098 case BPF_JMP | BPF_JEQ | BPF_K:
1099 case BPF_JMP | BPF_JEQ | BPF_X:
1100 case BPF_JMP | BPF_JGE | BPF_K:
1101 case BPF_JMP | BPF_JGE | BPF_X:
1102 case BPF_JMP | BPF_JGT | BPF_K:
1103 case BPF_JMP | BPF_JGT | BPF_X:
1104 case BPF_JMP | BPF_JSET | BPF_K:
1105 case BPF_JMP | BPF_JSET | BPF_X:
1106 /* Both conditionals must be safe */
1107 if (pc + ftest->jt + 1 >= flen ||
1108 pc + ftest->jf + 1 >= flen)
1111 case BPF_LD | BPF_W | BPF_ABS:
1112 case BPF_LD | BPF_H | BPF_ABS:
1113 case BPF_LD | BPF_B | BPF_ABS:
1115 if (bpf_anc_helper(ftest) & BPF_ANC)
1117 /* Ancillary operation unknown or unsupported */
1118 if (anc_found == false && ftest->k >= SKF_AD_OFF)
1123 /* Last instruction must be a RET code */
1124 switch (filter[flen - 1].code) {
1125 case BPF_RET | BPF_K:
1126 case BPF_RET | BPF_A:
1127 return check_load_and_stores(filter, flen);
1133 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1134 const struct sock_fprog *fprog)
1136 unsigned int fsize = bpf_classic_proglen(fprog);
1137 struct sock_fprog_kern *fkprog;
1139 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1143 fkprog = fp->orig_prog;
1144 fkprog->len = fprog->len;
1146 fkprog->filter = kmemdup(fp->insns, fsize,
1147 GFP_KERNEL | __GFP_NOWARN);
1148 if (!fkprog->filter) {
1149 kfree(fp->orig_prog);
1156 static void bpf_release_orig_filter(struct bpf_prog *fp)
1158 struct sock_fprog_kern *fprog = fp->orig_prog;
1161 kfree(fprog->filter);
1166 static void __bpf_prog_release(struct bpf_prog *prog)
1168 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1171 bpf_release_orig_filter(prog);
1172 bpf_prog_free(prog);
1176 static void __sk_filter_release(struct sk_filter *fp)
1178 __bpf_prog_release(fp->prog);
1183 * sk_filter_release_rcu - Release a socket filter by rcu_head
1184 * @rcu: rcu_head that contains the sk_filter to free
1186 static void sk_filter_release_rcu(struct rcu_head *rcu)
1188 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1190 __sk_filter_release(fp);
1194 * sk_filter_release - release a socket filter
1195 * @fp: filter to remove
1197 * Remove a filter from a socket and release its resources.
1199 static void sk_filter_release(struct sk_filter *fp)
1201 if (refcount_dec_and_test(&fp->refcnt))
1202 call_rcu(&fp->rcu, sk_filter_release_rcu);
1205 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1207 u32 filter_size = bpf_prog_size(fp->prog->len);
1209 atomic_sub(filter_size, &sk->sk_omem_alloc);
1210 sk_filter_release(fp);
1213 /* try to charge the socket memory if there is space available
1214 * return true on success
1216 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1218 u32 filter_size = bpf_prog_size(fp->prog->len);
1219 int optmem_max = READ_ONCE(sysctl_optmem_max);
1221 /* same check as in sock_kmalloc() */
1222 if (filter_size <= optmem_max &&
1223 atomic_read(&sk->sk_omem_alloc) + filter_size < optmem_max) {
1224 atomic_add(filter_size, &sk->sk_omem_alloc);
1230 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1232 if (!refcount_inc_not_zero(&fp->refcnt))
1235 if (!__sk_filter_charge(sk, fp)) {
1236 sk_filter_release(fp);
1242 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1244 struct sock_filter *old_prog;
1245 struct bpf_prog *old_fp;
1246 int err, new_len, old_len = fp->len;
1247 bool seen_ld_abs = false;
1249 /* We are free to overwrite insns et al right here as it won't be used at
1250 * this point in time anymore internally after the migration to the eBPF
1251 * instruction representation.
1253 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1254 sizeof(struct bpf_insn));
1256 /* Conversion cannot happen on overlapping memory areas,
1257 * so we need to keep the user BPF around until the 2nd
1258 * pass. At this time, the user BPF is stored in fp->insns.
1260 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1261 GFP_KERNEL | __GFP_NOWARN);
1267 /* 1st pass: calculate the new program length. */
1268 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1273 /* Expand fp for appending the new filter representation. */
1275 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1277 /* The old_fp is still around in case we couldn't
1278 * allocate new memory, so uncharge on that one.
1287 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1288 err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1291 /* 2nd bpf_convert_filter() can fail only if it fails
1292 * to allocate memory, remapping must succeed. Note,
1293 * that at this time old_fp has already been released
1298 fp = bpf_prog_select_runtime(fp, &err);
1308 __bpf_prog_release(fp);
1309 return ERR_PTR(err);
1312 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1313 bpf_aux_classic_check_t trans)
1317 fp->bpf_func = NULL;
1320 err = bpf_check_classic(fp->insns, fp->len);
1322 __bpf_prog_release(fp);
1323 return ERR_PTR(err);
1326 /* There might be additional checks and transformations
1327 * needed on classic filters, f.e. in case of seccomp.
1330 err = trans(fp->insns, fp->len);
1332 __bpf_prog_release(fp);
1333 return ERR_PTR(err);
1337 /* Probe if we can JIT compile the filter and if so, do
1338 * the compilation of the filter.
1340 bpf_jit_compile(fp);
1342 /* JIT compiler couldn't process this filter, so do the eBPF translation
1343 * for the optimized interpreter.
1346 fp = bpf_migrate_filter(fp);
1352 * bpf_prog_create - create an unattached filter
1353 * @pfp: the unattached filter that is created
1354 * @fprog: the filter program
1356 * Create a filter independent of any socket. We first run some
1357 * sanity checks on it to make sure it does not explode on us later.
1358 * If an error occurs or there is insufficient memory for the filter
1359 * a negative errno code is returned. On success the return is zero.
1361 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1363 unsigned int fsize = bpf_classic_proglen(fprog);
1364 struct bpf_prog *fp;
1366 /* Make sure new filter is there and in the right amounts. */
1367 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1370 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1374 memcpy(fp->insns, fprog->filter, fsize);
1376 fp->len = fprog->len;
1377 /* Since unattached filters are not copied back to user
1378 * space through sk_get_filter(), we do not need to hold
1379 * a copy here, and can spare us the work.
1381 fp->orig_prog = NULL;
1383 /* bpf_prepare_filter() already takes care of freeing
1384 * memory in case something goes wrong.
1386 fp = bpf_prepare_filter(fp, NULL);
1393 EXPORT_SYMBOL_GPL(bpf_prog_create);
1396 * bpf_prog_create_from_user - create an unattached filter from user buffer
1397 * @pfp: the unattached filter that is created
1398 * @fprog: the filter program
1399 * @trans: post-classic verifier transformation handler
1400 * @save_orig: save classic BPF program
1402 * This function effectively does the same as bpf_prog_create(), only
1403 * that it builds up its insns buffer from user space provided buffer.
1404 * It also allows for passing a bpf_aux_classic_check_t handler.
1406 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1407 bpf_aux_classic_check_t trans, bool save_orig)
1409 unsigned int fsize = bpf_classic_proglen(fprog);
1410 struct bpf_prog *fp;
1413 /* Make sure new filter is there and in the right amounts. */
1414 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1417 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1421 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1422 __bpf_prog_free(fp);
1426 fp->len = fprog->len;
1427 fp->orig_prog = NULL;
1430 err = bpf_prog_store_orig_filter(fp, fprog);
1432 __bpf_prog_free(fp);
1437 /* bpf_prepare_filter() already takes care of freeing
1438 * memory in case something goes wrong.
1440 fp = bpf_prepare_filter(fp, trans);
1447 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1449 void bpf_prog_destroy(struct bpf_prog *fp)
1451 __bpf_prog_release(fp);
1453 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1455 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1457 struct sk_filter *fp, *old_fp;
1459 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1465 if (!__sk_filter_charge(sk, fp)) {
1469 refcount_set(&fp->refcnt, 1);
1471 old_fp = rcu_dereference_protected(sk->sk_filter,
1472 lockdep_sock_is_held(sk));
1473 rcu_assign_pointer(sk->sk_filter, fp);
1476 sk_filter_uncharge(sk, old_fp);
1482 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1484 unsigned int fsize = bpf_classic_proglen(fprog);
1485 struct bpf_prog *prog;
1488 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1489 return ERR_PTR(-EPERM);
1491 /* Make sure new filter is there and in the right amounts. */
1492 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1493 return ERR_PTR(-EINVAL);
1495 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1497 return ERR_PTR(-ENOMEM);
1499 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1500 __bpf_prog_free(prog);
1501 return ERR_PTR(-EFAULT);
1504 prog->len = fprog->len;
1506 err = bpf_prog_store_orig_filter(prog, fprog);
1508 __bpf_prog_free(prog);
1509 return ERR_PTR(-ENOMEM);
1512 /* bpf_prepare_filter() already takes care of freeing
1513 * memory in case something goes wrong.
1515 return bpf_prepare_filter(prog, NULL);
1519 * sk_attach_filter - attach a socket filter
1520 * @fprog: the filter program
1521 * @sk: the socket to use
1523 * Attach the user's filter code. We first run some sanity checks on
1524 * it to make sure it does not explode on us later. If an error
1525 * occurs or there is insufficient memory for the filter a negative
1526 * errno code is returned. On success the return is zero.
1528 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1530 struct bpf_prog *prog = __get_filter(fprog, sk);
1534 return PTR_ERR(prog);
1536 err = __sk_attach_prog(prog, sk);
1538 __bpf_prog_release(prog);
1544 EXPORT_SYMBOL_GPL(sk_attach_filter);
1546 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1548 struct bpf_prog *prog = __get_filter(fprog, sk);
1552 return PTR_ERR(prog);
1554 if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max))
1557 err = reuseport_attach_prog(sk, prog);
1560 __bpf_prog_release(prog);
1565 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1567 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1568 return ERR_PTR(-EPERM);
1570 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1573 int sk_attach_bpf(u32 ufd, struct sock *sk)
1575 struct bpf_prog *prog = __get_bpf(ufd, sk);
1579 return PTR_ERR(prog);
1581 err = __sk_attach_prog(prog, sk);
1590 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1592 struct bpf_prog *prog;
1595 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1598 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1599 if (PTR_ERR(prog) == -EINVAL)
1600 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1602 return PTR_ERR(prog);
1604 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1605 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1606 * bpf prog (e.g. sockmap). It depends on the
1607 * limitation imposed by bpf_prog_load().
1608 * Hence, sysctl_optmem_max is not checked.
1610 if ((sk->sk_type != SOCK_STREAM &&
1611 sk->sk_type != SOCK_DGRAM) ||
1612 (sk->sk_protocol != IPPROTO_UDP &&
1613 sk->sk_protocol != IPPROTO_TCP) ||
1614 (sk->sk_family != AF_INET &&
1615 sk->sk_family != AF_INET6)) {
1620 /* BPF_PROG_TYPE_SOCKET_FILTER */
1621 if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max)) {
1627 err = reuseport_attach_prog(sk, prog);
1635 void sk_reuseport_prog_free(struct bpf_prog *prog)
1640 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1643 bpf_prog_destroy(prog);
1646 struct bpf_scratchpad {
1648 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1649 u8 buff[MAX_BPF_STACK];
1653 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1655 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1656 unsigned int write_len)
1658 return skb_ensure_writable(skb, write_len);
1661 static inline int bpf_try_make_writable(struct sk_buff *skb,
1662 unsigned int write_len)
1664 int err = __bpf_try_make_writable(skb, write_len);
1666 bpf_compute_data_pointers(skb);
1670 static int bpf_try_make_head_writable(struct sk_buff *skb)
1672 return bpf_try_make_writable(skb, skb_headlen(skb));
1675 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1677 if (skb_at_tc_ingress(skb))
1678 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1681 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1683 if (skb_at_tc_ingress(skb))
1684 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1687 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1688 const void *, from, u32, len, u64, flags)
1692 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1694 if (unlikely(offset > INT_MAX))
1696 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1699 ptr = skb->data + offset;
1700 if (flags & BPF_F_RECOMPUTE_CSUM)
1701 __skb_postpull_rcsum(skb, ptr, len, offset);
1703 memcpy(ptr, from, len);
1705 if (flags & BPF_F_RECOMPUTE_CSUM)
1706 __skb_postpush_rcsum(skb, ptr, len, offset);
1707 if (flags & BPF_F_INVALIDATE_HASH)
1708 skb_clear_hash(skb);
1713 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1714 .func = bpf_skb_store_bytes,
1716 .ret_type = RET_INTEGER,
1717 .arg1_type = ARG_PTR_TO_CTX,
1718 .arg2_type = ARG_ANYTHING,
1719 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1720 .arg4_type = ARG_CONST_SIZE,
1721 .arg5_type = ARG_ANYTHING,
1724 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1725 void *, to, u32, len)
1729 if (unlikely(offset > INT_MAX))
1732 ptr = skb_header_pointer(skb, offset, len, to);
1736 memcpy(to, ptr, len);
1744 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1745 .func = bpf_skb_load_bytes,
1747 .ret_type = RET_INTEGER,
1748 .arg1_type = ARG_PTR_TO_CTX,
1749 .arg2_type = ARG_ANYTHING,
1750 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1751 .arg4_type = ARG_CONST_SIZE,
1754 BPF_CALL_4(bpf_flow_dissector_load_bytes,
1755 const struct bpf_flow_dissector *, ctx, u32, offset,
1756 void *, to, u32, len)
1760 if (unlikely(offset > 0xffff))
1763 if (unlikely(!ctx->skb))
1766 ptr = skb_header_pointer(ctx->skb, offset, len, to);
1770 memcpy(to, ptr, len);
1778 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1779 .func = bpf_flow_dissector_load_bytes,
1781 .ret_type = RET_INTEGER,
1782 .arg1_type = ARG_PTR_TO_CTX,
1783 .arg2_type = ARG_ANYTHING,
1784 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1785 .arg4_type = ARG_CONST_SIZE,
1788 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1789 u32, offset, void *, to, u32, len, u32, start_header)
1791 u8 *end = skb_tail_pointer(skb);
1794 if (unlikely(offset > 0xffff))
1797 switch (start_header) {
1798 case BPF_HDR_START_MAC:
1799 if (unlikely(!skb_mac_header_was_set(skb)))
1801 start = skb_mac_header(skb);
1803 case BPF_HDR_START_NET:
1804 start = skb_network_header(skb);
1810 ptr = start + offset;
1812 if (likely(ptr + len <= end)) {
1813 memcpy(to, ptr, len);
1822 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1823 .func = bpf_skb_load_bytes_relative,
1825 .ret_type = RET_INTEGER,
1826 .arg1_type = ARG_PTR_TO_CTX,
1827 .arg2_type = ARG_ANYTHING,
1828 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1829 .arg4_type = ARG_CONST_SIZE,
1830 .arg5_type = ARG_ANYTHING,
1833 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1835 /* Idea is the following: should the needed direct read/write
1836 * test fail during runtime, we can pull in more data and redo
1837 * again, since implicitly, we invalidate previous checks here.
1839 * Or, since we know how much we need to make read/writeable,
1840 * this can be done once at the program beginning for direct
1841 * access case. By this we overcome limitations of only current
1842 * headroom being accessible.
1844 return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1847 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1848 .func = bpf_skb_pull_data,
1850 .ret_type = RET_INTEGER,
1851 .arg1_type = ARG_PTR_TO_CTX,
1852 .arg2_type = ARG_ANYTHING,
1855 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1857 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1860 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1861 .func = bpf_sk_fullsock,
1863 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
1864 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
1867 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1868 unsigned int write_len)
1870 return __bpf_try_make_writable(skb, write_len);
1873 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1875 /* Idea is the following: should the needed direct read/write
1876 * test fail during runtime, we can pull in more data and redo
1877 * again, since implicitly, we invalidate previous checks here.
1879 * Or, since we know how much we need to make read/writeable,
1880 * this can be done once at the program beginning for direct
1881 * access case. By this we overcome limitations of only current
1882 * headroom being accessible.
1884 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1887 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1888 .func = sk_skb_pull_data,
1890 .ret_type = RET_INTEGER,
1891 .arg1_type = ARG_PTR_TO_CTX,
1892 .arg2_type = ARG_ANYTHING,
1895 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1896 u64, from, u64, to, u64, flags)
1900 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1902 if (unlikely(offset > 0xffff || offset & 1))
1904 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1907 ptr = (__sum16 *)(skb->data + offset);
1908 switch (flags & BPF_F_HDR_FIELD_MASK) {
1910 if (unlikely(from != 0))
1913 csum_replace_by_diff(ptr, to);
1916 csum_replace2(ptr, from, to);
1919 csum_replace4(ptr, from, to);
1928 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1929 .func = bpf_l3_csum_replace,
1931 .ret_type = RET_INTEGER,
1932 .arg1_type = ARG_PTR_TO_CTX,
1933 .arg2_type = ARG_ANYTHING,
1934 .arg3_type = ARG_ANYTHING,
1935 .arg4_type = ARG_ANYTHING,
1936 .arg5_type = ARG_ANYTHING,
1939 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1940 u64, from, u64, to, u64, flags)
1942 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1943 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1944 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1947 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1948 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1950 if (unlikely(offset > 0xffff || offset & 1))
1952 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1955 ptr = (__sum16 *)(skb->data + offset);
1956 if (is_mmzero && !do_mforce && !*ptr)
1959 switch (flags & BPF_F_HDR_FIELD_MASK) {
1961 if (unlikely(from != 0))
1964 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1967 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1970 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1976 if (is_mmzero && !*ptr)
1977 *ptr = CSUM_MANGLED_0;
1981 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1982 .func = bpf_l4_csum_replace,
1984 .ret_type = RET_INTEGER,
1985 .arg1_type = ARG_PTR_TO_CTX,
1986 .arg2_type = ARG_ANYTHING,
1987 .arg3_type = ARG_ANYTHING,
1988 .arg4_type = ARG_ANYTHING,
1989 .arg5_type = ARG_ANYTHING,
1992 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1993 __be32 *, to, u32, to_size, __wsum, seed)
1995 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1996 u32 diff_size = from_size + to_size;
1999 /* This is quite flexible, some examples:
2001 * from_size == 0, to_size > 0, seed := csum --> pushing data
2002 * from_size > 0, to_size == 0, seed := csum --> pulling data
2003 * from_size > 0, to_size > 0, seed := 0 --> diffing data
2005 * Even for diffing, from_size and to_size don't need to be equal.
2007 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
2008 diff_size > sizeof(sp->diff)))
2011 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
2012 sp->diff[j] = ~from[i];
2013 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
2014 sp->diff[j] = to[i];
2016 return csum_partial(sp->diff, diff_size, seed);
2019 static const struct bpf_func_proto bpf_csum_diff_proto = {
2020 .func = bpf_csum_diff,
2023 .ret_type = RET_INTEGER,
2024 .arg1_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2025 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
2026 .arg3_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2027 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
2028 .arg5_type = ARG_ANYTHING,
2031 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2033 /* The interface is to be used in combination with bpf_csum_diff()
2034 * for direct packet writes. csum rotation for alignment as well
2035 * as emulating csum_sub() can be done from the eBPF program.
2037 if (skb->ip_summed == CHECKSUM_COMPLETE)
2038 return (skb->csum = csum_add(skb->csum, csum));
2043 static const struct bpf_func_proto bpf_csum_update_proto = {
2044 .func = bpf_csum_update,
2046 .ret_type = RET_INTEGER,
2047 .arg1_type = ARG_PTR_TO_CTX,
2048 .arg2_type = ARG_ANYTHING,
2051 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
2053 /* The interface is to be used in combination with bpf_skb_adjust_room()
2054 * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET
2055 * is passed as flags, for example.
2058 case BPF_CSUM_LEVEL_INC:
2059 __skb_incr_checksum_unnecessary(skb);
2061 case BPF_CSUM_LEVEL_DEC:
2062 __skb_decr_checksum_unnecessary(skb);
2064 case BPF_CSUM_LEVEL_RESET:
2065 __skb_reset_checksum_unnecessary(skb);
2067 case BPF_CSUM_LEVEL_QUERY:
2068 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2069 skb->csum_level : -EACCES;
2077 static const struct bpf_func_proto bpf_csum_level_proto = {
2078 .func = bpf_csum_level,
2080 .ret_type = RET_INTEGER,
2081 .arg1_type = ARG_PTR_TO_CTX,
2082 .arg2_type = ARG_ANYTHING,
2085 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2087 return dev_forward_skb_nomtu(dev, skb);
2090 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2091 struct sk_buff *skb)
2093 int ret = ____dev_forward_skb(dev, skb, false);
2097 ret = netif_rx(skb);
2103 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2107 if (dev_xmit_recursion()) {
2108 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2114 skb_clear_tstamp(skb);
2116 dev_xmit_recursion_inc();
2117 ret = dev_queue_xmit(skb);
2118 dev_xmit_recursion_dec();
2123 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2126 unsigned int mlen = skb_network_offset(skb);
2128 if (unlikely(skb->len <= mlen)) {
2134 __skb_pull(skb, mlen);
2135 if (unlikely(!skb->len)) {
2140 /* At ingress, the mac header has already been pulled once.
2141 * At egress, skb_pospull_rcsum has to be done in case that
2142 * the skb is originated from ingress (i.e. a forwarded skb)
2143 * to ensure that rcsum starts at net header.
2145 if (!skb_at_tc_ingress(skb))
2146 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2148 skb_pop_mac_header(skb);
2149 skb_reset_mac_len(skb);
2150 return flags & BPF_F_INGRESS ?
2151 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2154 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2157 /* Verify that a link layer header is carried */
2158 if (unlikely(skb->mac_header >= skb->network_header || skb->len == 0)) {
2163 bpf_push_mac_rcsum(skb);
2164 return flags & BPF_F_INGRESS ?
2165 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2168 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2171 if (dev_is_mac_header_xmit(dev))
2172 return __bpf_redirect_common(skb, dev, flags);
2174 return __bpf_redirect_no_mac(skb, dev, flags);
2177 #if IS_ENABLED(CONFIG_IPV6)
2178 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2179 struct net_device *dev, struct bpf_nh_params *nh)
2181 u32 hh_len = LL_RESERVED_SPACE(dev);
2182 const struct in6_addr *nexthop;
2183 struct dst_entry *dst = NULL;
2184 struct neighbour *neigh;
2186 if (dev_xmit_recursion()) {
2187 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2192 skb_clear_tstamp(skb);
2194 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2195 skb = skb_expand_head(skb, hh_len);
2203 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2204 &ipv6_hdr(skb)->daddr);
2206 nexthop = &nh->ipv6_nh;
2208 neigh = ip_neigh_gw6(dev, nexthop);
2209 if (likely(!IS_ERR(neigh))) {
2212 sock_confirm_neigh(skb, neigh);
2213 dev_xmit_recursion_inc();
2214 ret = neigh_output(neigh, skb, false);
2215 dev_xmit_recursion_dec();
2216 rcu_read_unlock_bh();
2219 rcu_read_unlock_bh();
2221 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2227 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2228 struct bpf_nh_params *nh)
2230 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2231 struct net *net = dev_net(dev);
2232 int err, ret = NET_XMIT_DROP;
2235 struct dst_entry *dst;
2236 struct flowi6 fl6 = {
2237 .flowi6_flags = FLOWI_FLAG_ANYSRC,
2238 .flowi6_mark = skb->mark,
2239 .flowlabel = ip6_flowinfo(ip6h),
2240 .flowi6_oif = dev->ifindex,
2241 .flowi6_proto = ip6h->nexthdr,
2242 .daddr = ip6h->daddr,
2243 .saddr = ip6h->saddr,
2246 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2250 skb_dst_set(skb, dst);
2251 } else if (nh->nh_family != AF_INET6) {
2255 err = bpf_out_neigh_v6(net, skb, dev, nh);
2256 if (unlikely(net_xmit_eval(err)))
2257 dev->stats.tx_errors++;
2259 ret = NET_XMIT_SUCCESS;
2262 dev->stats.tx_errors++;
2268 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2269 struct bpf_nh_params *nh)
2272 return NET_XMIT_DROP;
2274 #endif /* CONFIG_IPV6 */
2276 #if IS_ENABLED(CONFIG_INET)
2277 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2278 struct net_device *dev, struct bpf_nh_params *nh)
2280 u32 hh_len = LL_RESERVED_SPACE(dev);
2281 struct neighbour *neigh;
2282 bool is_v6gw = false;
2284 if (dev_xmit_recursion()) {
2285 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2290 skb_clear_tstamp(skb);
2292 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2293 skb = skb_expand_head(skb, hh_len);
2300 struct dst_entry *dst = skb_dst(skb);
2301 struct rtable *rt = container_of(dst, struct rtable, dst);
2303 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2304 } else if (nh->nh_family == AF_INET6) {
2305 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2307 } else if (nh->nh_family == AF_INET) {
2308 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2310 rcu_read_unlock_bh();
2314 if (likely(!IS_ERR(neigh))) {
2317 sock_confirm_neigh(skb, neigh);
2318 dev_xmit_recursion_inc();
2319 ret = neigh_output(neigh, skb, is_v6gw);
2320 dev_xmit_recursion_dec();
2321 rcu_read_unlock_bh();
2324 rcu_read_unlock_bh();
2330 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2331 struct bpf_nh_params *nh)
2333 const struct iphdr *ip4h = ip_hdr(skb);
2334 struct net *net = dev_net(dev);
2335 int err, ret = NET_XMIT_DROP;
2338 struct flowi4 fl4 = {
2339 .flowi4_flags = FLOWI_FLAG_ANYSRC,
2340 .flowi4_mark = skb->mark,
2341 .flowi4_tos = RT_TOS(ip4h->tos),
2342 .flowi4_oif = dev->ifindex,
2343 .flowi4_proto = ip4h->protocol,
2344 .daddr = ip4h->daddr,
2345 .saddr = ip4h->saddr,
2349 rt = ip_route_output_flow(net, &fl4, NULL);
2352 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2357 skb_dst_set(skb, &rt->dst);
2360 err = bpf_out_neigh_v4(net, skb, dev, nh);
2361 if (unlikely(net_xmit_eval(err)))
2362 dev->stats.tx_errors++;
2364 ret = NET_XMIT_SUCCESS;
2367 dev->stats.tx_errors++;
2373 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2374 struct bpf_nh_params *nh)
2377 return NET_XMIT_DROP;
2379 #endif /* CONFIG_INET */
2381 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2382 struct bpf_nh_params *nh)
2384 struct ethhdr *ethh = eth_hdr(skb);
2386 if (unlikely(skb->mac_header >= skb->network_header))
2388 bpf_push_mac_rcsum(skb);
2389 if (is_multicast_ether_addr(ethh->h_dest))
2392 skb_pull(skb, sizeof(*ethh));
2393 skb_unset_mac_header(skb);
2394 skb_reset_network_header(skb);
2396 if (skb->protocol == htons(ETH_P_IP))
2397 return __bpf_redirect_neigh_v4(skb, dev, nh);
2398 else if (skb->protocol == htons(ETH_P_IPV6))
2399 return __bpf_redirect_neigh_v6(skb, dev, nh);
2405 /* Internal, non-exposed redirect flags. */
2407 BPF_F_NEIGH = (1ULL << 1),
2408 BPF_F_PEER = (1ULL << 2),
2409 BPF_F_NEXTHOP = (1ULL << 3),
2410 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2413 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2415 struct net_device *dev;
2416 struct sk_buff *clone;
2419 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2422 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2426 clone = skb_clone(skb, GFP_ATOMIC);
2427 if (unlikely(!clone))
2430 /* For direct write, we need to keep the invariant that the skbs
2431 * we're dealing with need to be uncloned. Should uncloning fail
2432 * here, we need to free the just generated clone to unclone once
2435 ret = bpf_try_make_head_writable(skb);
2436 if (unlikely(ret)) {
2441 return __bpf_redirect(clone, dev, flags);
2444 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2445 .func = bpf_clone_redirect,
2447 .ret_type = RET_INTEGER,
2448 .arg1_type = ARG_PTR_TO_CTX,
2449 .arg2_type = ARG_ANYTHING,
2450 .arg3_type = ARG_ANYTHING,
2453 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2454 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2456 int skb_do_redirect(struct sk_buff *skb)
2458 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2459 struct net *net = dev_net(skb->dev);
2460 struct net_device *dev;
2461 u32 flags = ri->flags;
2463 dev = dev_get_by_index_rcu(net, ri->tgt_index);
2468 if (flags & BPF_F_PEER) {
2469 const struct net_device_ops *ops = dev->netdev_ops;
2471 if (unlikely(!ops->ndo_get_peer_dev ||
2472 !skb_at_tc_ingress(skb)))
2474 dev = ops->ndo_get_peer_dev(dev);
2475 if (unlikely(!dev ||
2476 !(dev->flags & IFF_UP) ||
2477 net_eq(net, dev_net(dev))))
2482 return flags & BPF_F_NEIGH ?
2483 __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2485 __bpf_redirect(skb, dev, flags);
2491 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2493 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2495 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2499 ri->tgt_index = ifindex;
2501 return TC_ACT_REDIRECT;
2504 static const struct bpf_func_proto bpf_redirect_proto = {
2505 .func = bpf_redirect,
2507 .ret_type = RET_INTEGER,
2508 .arg1_type = ARG_ANYTHING,
2509 .arg2_type = ARG_ANYTHING,
2512 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2514 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2516 if (unlikely(flags))
2519 ri->flags = BPF_F_PEER;
2520 ri->tgt_index = ifindex;
2522 return TC_ACT_REDIRECT;
2525 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2526 .func = bpf_redirect_peer,
2528 .ret_type = RET_INTEGER,
2529 .arg1_type = ARG_ANYTHING,
2530 .arg2_type = ARG_ANYTHING,
2533 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2534 int, plen, u64, flags)
2536 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2538 if (unlikely((plen && plen < sizeof(*params)) || flags))
2541 ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2542 ri->tgt_index = ifindex;
2544 BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2546 memcpy(&ri->nh, params, sizeof(ri->nh));
2548 return TC_ACT_REDIRECT;
2551 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2552 .func = bpf_redirect_neigh,
2554 .ret_type = RET_INTEGER,
2555 .arg1_type = ARG_ANYTHING,
2556 .arg2_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2557 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
2558 .arg4_type = ARG_ANYTHING,
2561 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2563 msg->apply_bytes = bytes;
2567 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2568 .func = bpf_msg_apply_bytes,
2570 .ret_type = RET_INTEGER,
2571 .arg1_type = ARG_PTR_TO_CTX,
2572 .arg2_type = ARG_ANYTHING,
2575 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2577 msg->cork_bytes = bytes;
2581 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2582 .func = bpf_msg_cork_bytes,
2584 .ret_type = RET_INTEGER,
2585 .arg1_type = ARG_PTR_TO_CTX,
2586 .arg2_type = ARG_ANYTHING,
2589 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2590 u32, end, u64, flags)
2592 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2593 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2594 struct scatterlist *sge;
2595 u8 *raw, *to, *from;
2598 if (unlikely(flags || end <= start))
2601 /* First find the starting scatterlist element */
2605 len = sk_msg_elem(msg, i)->length;
2606 if (start < offset + len)
2608 sk_msg_iter_var_next(i);
2609 } while (i != msg->sg.end);
2611 if (unlikely(start >= offset + len))
2615 /* The start may point into the sg element so we need to also
2616 * account for the headroom.
2618 bytes_sg_total = start - offset + bytes;
2619 if (!test_bit(i, msg->sg.copy) && bytes_sg_total <= len)
2622 /* At this point we need to linearize multiple scatterlist
2623 * elements or a single shared page. Either way we need to
2624 * copy into a linear buffer exclusively owned by BPF. Then
2625 * place the buffer in the scatterlist and fixup the original
2626 * entries by removing the entries now in the linear buffer
2627 * and shifting the remaining entries. For now we do not try
2628 * to copy partial entries to avoid complexity of running out
2629 * of sg_entry slots. The downside is reading a single byte
2630 * will copy the entire sg entry.
2633 copy += sk_msg_elem(msg, i)->length;
2634 sk_msg_iter_var_next(i);
2635 if (bytes_sg_total <= copy)
2637 } while (i != msg->sg.end);
2640 if (unlikely(bytes_sg_total > copy))
2643 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2645 if (unlikely(!page))
2648 raw = page_address(page);
2651 sge = sk_msg_elem(msg, i);
2652 from = sg_virt(sge);
2656 memcpy(to, from, len);
2659 put_page(sg_page(sge));
2661 sk_msg_iter_var_next(i);
2662 } while (i != last_sge);
2664 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2666 /* To repair sg ring we need to shift entries. If we only
2667 * had a single entry though we can just replace it and
2668 * be done. Otherwise walk the ring and shift the entries.
2670 WARN_ON_ONCE(last_sge == first_sge);
2671 shift = last_sge > first_sge ?
2672 last_sge - first_sge - 1 :
2673 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2678 sk_msg_iter_var_next(i);
2682 if (i + shift >= NR_MSG_FRAG_IDS)
2683 move_from = i + shift - NR_MSG_FRAG_IDS;
2685 move_from = i + shift;
2686 if (move_from == msg->sg.end)
2689 msg->sg.data[i] = msg->sg.data[move_from];
2690 msg->sg.data[move_from].length = 0;
2691 msg->sg.data[move_from].page_link = 0;
2692 msg->sg.data[move_from].offset = 0;
2693 sk_msg_iter_var_next(i);
2696 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2697 msg->sg.end - shift + NR_MSG_FRAG_IDS :
2698 msg->sg.end - shift;
2700 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2701 msg->data_end = msg->data + bytes;
2705 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2706 .func = bpf_msg_pull_data,
2708 .ret_type = RET_INTEGER,
2709 .arg1_type = ARG_PTR_TO_CTX,
2710 .arg2_type = ARG_ANYTHING,
2711 .arg3_type = ARG_ANYTHING,
2712 .arg4_type = ARG_ANYTHING,
2715 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2716 u32, len, u64, flags)
2718 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2719 u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2720 u8 *raw, *to, *from;
2723 if (unlikely(flags))
2726 if (unlikely(len == 0))
2729 /* First find the starting scatterlist element */
2733 l = sk_msg_elem(msg, i)->length;
2735 if (start < offset + l)
2737 sk_msg_iter_var_next(i);
2738 } while (i != msg->sg.end);
2740 if (start >= offset + l)
2743 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2745 /* If no space available will fallback to copy, we need at
2746 * least one scatterlist elem available to push data into
2747 * when start aligns to the beginning of an element or two
2748 * when it falls inside an element. We handle the start equals
2749 * offset case because its the common case for inserting a
2752 if (!space || (space == 1 && start != offset))
2753 copy = msg->sg.data[i].length;
2755 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2756 get_order(copy + len));
2757 if (unlikely(!page))
2763 raw = page_address(page);
2765 psge = sk_msg_elem(msg, i);
2766 front = start - offset;
2767 back = psge->length - front;
2768 from = sg_virt(psge);
2771 memcpy(raw, from, front);
2775 to = raw + front + len;
2777 memcpy(to, from, back);
2780 put_page(sg_page(psge));
2781 } else if (start - offset) {
2782 psge = sk_msg_elem(msg, i);
2783 rsge = sk_msg_elem_cpy(msg, i);
2785 psge->length = start - offset;
2786 rsge.length -= psge->length;
2787 rsge.offset += start;
2789 sk_msg_iter_var_next(i);
2790 sg_unmark_end(psge);
2791 sg_unmark_end(&rsge);
2792 sk_msg_iter_next(msg, end);
2795 /* Slot(s) to place newly allocated data */
2798 /* Shift one or two slots as needed */
2800 sge = sk_msg_elem_cpy(msg, i);
2802 sk_msg_iter_var_next(i);
2803 sg_unmark_end(&sge);
2804 sk_msg_iter_next(msg, end);
2806 nsge = sk_msg_elem_cpy(msg, i);
2808 sk_msg_iter_var_next(i);
2809 nnsge = sk_msg_elem_cpy(msg, i);
2812 while (i != msg->sg.end) {
2813 msg->sg.data[i] = sge;
2815 sk_msg_iter_var_next(i);
2818 nnsge = sk_msg_elem_cpy(msg, i);
2820 nsge = sk_msg_elem_cpy(msg, i);
2825 /* Place newly allocated data buffer */
2826 sk_mem_charge(msg->sk, len);
2827 msg->sg.size += len;
2828 __clear_bit(new, msg->sg.copy);
2829 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2831 get_page(sg_page(&rsge));
2832 sk_msg_iter_var_next(new);
2833 msg->sg.data[new] = rsge;
2836 sk_msg_compute_data_pointers(msg);
2840 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2841 .func = bpf_msg_push_data,
2843 .ret_type = RET_INTEGER,
2844 .arg1_type = ARG_PTR_TO_CTX,
2845 .arg2_type = ARG_ANYTHING,
2846 .arg3_type = ARG_ANYTHING,
2847 .arg4_type = ARG_ANYTHING,
2850 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2856 sk_msg_iter_var_next(i);
2857 msg->sg.data[prev] = msg->sg.data[i];
2858 } while (i != msg->sg.end);
2860 sk_msg_iter_prev(msg, end);
2863 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2865 struct scatterlist tmp, sge;
2867 sk_msg_iter_next(msg, end);
2868 sge = sk_msg_elem_cpy(msg, i);
2869 sk_msg_iter_var_next(i);
2870 tmp = sk_msg_elem_cpy(msg, i);
2872 while (i != msg->sg.end) {
2873 msg->sg.data[i] = sge;
2874 sk_msg_iter_var_next(i);
2876 tmp = sk_msg_elem_cpy(msg, i);
2880 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2881 u32, len, u64, flags)
2883 u32 i = 0, l = 0, space, offset = 0;
2884 u64 last = start + len;
2887 if (unlikely(flags))
2890 /* First find the starting scatterlist element */
2894 l = sk_msg_elem(msg, i)->length;
2896 if (start < offset + l)
2898 sk_msg_iter_var_next(i);
2899 } while (i != msg->sg.end);
2901 /* Bounds checks: start and pop must be inside message */
2902 if (start >= offset + l || last >= msg->sg.size)
2905 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2908 /* --------------| offset
2909 * -| start |-------- len -------|
2911 * |----- a ----|-------- pop -------|----- b ----|
2912 * |______________________________________________| length
2915 * a: region at front of scatter element to save
2916 * b: region at back of scatter element to save when length > A + pop
2917 * pop: region to pop from element, same as input 'pop' here will be
2918 * decremented below per iteration.
2920 * Two top-level cases to handle when start != offset, first B is non
2921 * zero and second B is zero corresponding to when a pop includes more
2924 * Then if B is non-zero AND there is no space allocate space and
2925 * compact A, B regions into page. If there is space shift ring to
2926 * the rigth free'ing the next element in ring to place B, leaving
2927 * A untouched except to reduce length.
2929 if (start != offset) {
2930 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2932 int b = sge->length - pop - a;
2934 sk_msg_iter_var_next(i);
2936 if (pop < sge->length - a) {
2939 sk_msg_shift_right(msg, i);
2940 nsge = sk_msg_elem(msg, i);
2941 get_page(sg_page(sge));
2944 b, sge->offset + pop + a);
2946 struct page *page, *orig;
2949 page = alloc_pages(__GFP_NOWARN |
2950 __GFP_COMP | GFP_ATOMIC,
2952 if (unlikely(!page))
2956 orig = sg_page(sge);
2957 from = sg_virt(sge);
2958 to = page_address(page);
2959 memcpy(to, from, a);
2960 memcpy(to + a, from + a + pop, b);
2961 sg_set_page(sge, page, a + b, 0);
2965 } else if (pop >= sge->length - a) {
2966 pop -= (sge->length - a);
2971 /* From above the current layout _must_ be as follows,
2976 * |---- pop ---|---------------- b ------------|
2977 * |____________________________________________| length
2979 * Offset and start of the current msg elem are equal because in the
2980 * previous case we handled offset != start and either consumed the
2981 * entire element and advanced to the next element OR pop == 0.
2983 * Two cases to handle here are first pop is less than the length
2984 * leaving some remainder b above. Simply adjust the element's layout
2985 * in this case. Or pop >= length of the element so that b = 0. In this
2986 * case advance to next element decrementing pop.
2989 struct scatterlist *sge = sk_msg_elem(msg, i);
2991 if (pop < sge->length) {
2997 sk_msg_shift_left(msg, i);
2999 sk_msg_iter_var_next(i);
3002 sk_mem_uncharge(msg->sk, len - pop);
3003 msg->sg.size -= (len - pop);
3004 sk_msg_compute_data_pointers(msg);
3008 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
3009 .func = bpf_msg_pop_data,
3011 .ret_type = RET_INTEGER,
3012 .arg1_type = ARG_PTR_TO_CTX,
3013 .arg2_type = ARG_ANYTHING,
3014 .arg3_type = ARG_ANYTHING,
3015 .arg4_type = ARG_ANYTHING,
3018 #ifdef CONFIG_CGROUP_NET_CLASSID
3019 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3021 return __task_get_classid(current);
3024 const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3025 .func = bpf_get_cgroup_classid_curr,
3027 .ret_type = RET_INTEGER,
3030 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3032 struct sock *sk = skb_to_full_sk(skb);
3034 if (!sk || !sk_fullsock(sk))
3037 return sock_cgroup_classid(&sk->sk_cgrp_data);
3040 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3041 .func = bpf_skb_cgroup_classid,
3043 .ret_type = RET_INTEGER,
3044 .arg1_type = ARG_PTR_TO_CTX,
3048 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3050 return task_get_classid(skb);
3053 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3054 .func = bpf_get_cgroup_classid,
3056 .ret_type = RET_INTEGER,
3057 .arg1_type = ARG_PTR_TO_CTX,
3060 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3062 return dst_tclassid(skb);
3065 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3066 .func = bpf_get_route_realm,
3068 .ret_type = RET_INTEGER,
3069 .arg1_type = ARG_PTR_TO_CTX,
3072 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3074 /* If skb_clear_hash() was called due to mangling, we can
3075 * trigger SW recalculation here. Later access to hash
3076 * can then use the inline skb->hash via context directly
3077 * instead of calling this helper again.
3079 return skb_get_hash(skb);
3082 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3083 .func = bpf_get_hash_recalc,
3085 .ret_type = RET_INTEGER,
3086 .arg1_type = ARG_PTR_TO_CTX,
3089 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3091 /* After all direct packet write, this can be used once for
3092 * triggering a lazy recalc on next skb_get_hash() invocation.
3094 skb_clear_hash(skb);
3098 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3099 .func = bpf_set_hash_invalid,
3101 .ret_type = RET_INTEGER,
3102 .arg1_type = ARG_PTR_TO_CTX,
3105 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3107 /* Set user specified hash as L4(+), so that it gets returned
3108 * on skb_get_hash() call unless BPF prog later on triggers a
3111 __skb_set_sw_hash(skb, hash, true);
3115 static const struct bpf_func_proto bpf_set_hash_proto = {
3116 .func = bpf_set_hash,
3118 .ret_type = RET_INTEGER,
3119 .arg1_type = ARG_PTR_TO_CTX,
3120 .arg2_type = ARG_ANYTHING,
3123 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3128 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3129 vlan_proto != htons(ETH_P_8021AD)))
3130 vlan_proto = htons(ETH_P_8021Q);
3132 bpf_push_mac_rcsum(skb);
3133 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3134 bpf_pull_mac_rcsum(skb);
3136 bpf_compute_data_pointers(skb);
3140 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3141 .func = bpf_skb_vlan_push,
3143 .ret_type = RET_INTEGER,
3144 .arg1_type = ARG_PTR_TO_CTX,
3145 .arg2_type = ARG_ANYTHING,
3146 .arg3_type = ARG_ANYTHING,
3149 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3153 bpf_push_mac_rcsum(skb);
3154 ret = skb_vlan_pop(skb);
3155 bpf_pull_mac_rcsum(skb);
3157 bpf_compute_data_pointers(skb);
3161 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3162 .func = bpf_skb_vlan_pop,
3164 .ret_type = RET_INTEGER,
3165 .arg1_type = ARG_PTR_TO_CTX,
3168 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3170 /* Caller already did skb_cow() with len as headroom,
3171 * so no need to do it here.
3174 memmove(skb->data, skb->data + len, off);
3175 memset(skb->data + off, 0, len);
3177 /* No skb_postpush_rcsum(skb, skb->data + off, len)
3178 * needed here as it does not change the skb->csum
3179 * result for checksum complete when summing over
3185 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3189 /* skb_ensure_writable() is not needed here, as we're
3190 * already working on an uncloned skb.
3192 if (unlikely(!pskb_may_pull(skb, off + len)))
3195 old_data = skb->data;
3196 __skb_pull(skb, len);
3197 skb_postpull_rcsum(skb, old_data + off, len);
3198 memmove(skb->data, old_data, off);
3203 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3205 bool trans_same = skb->transport_header == skb->network_header;
3208 /* There's no need for __skb_push()/__skb_pull() pair to
3209 * get to the start of the mac header as we're guaranteed
3210 * to always start from here under eBPF.
3212 ret = bpf_skb_generic_push(skb, off, len);
3214 skb->mac_header -= len;
3215 skb->network_header -= len;
3217 skb->transport_header = skb->network_header;
3223 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3225 bool trans_same = skb->transport_header == skb->network_header;
3228 /* Same here, __skb_push()/__skb_pull() pair not needed. */
3229 ret = bpf_skb_generic_pop(skb, off, len);
3231 skb->mac_header += len;
3232 skb->network_header += len;
3234 skb->transport_header = skb->network_header;
3240 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3242 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3243 u32 off = skb_mac_header_len(skb);
3246 ret = skb_cow(skb, len_diff);
3247 if (unlikely(ret < 0))
3250 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3251 if (unlikely(ret < 0))
3254 if (skb_is_gso(skb)) {
3255 struct skb_shared_info *shinfo = skb_shinfo(skb);
3257 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */
3258 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3259 shinfo->gso_type &= ~SKB_GSO_TCPV4;
3260 shinfo->gso_type |= SKB_GSO_TCPV6;
3264 skb->protocol = htons(ETH_P_IPV6);
3265 skb_clear_hash(skb);
3270 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3272 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3273 u32 off = skb_mac_header_len(skb);
3276 ret = skb_unclone(skb, GFP_ATOMIC);
3277 if (unlikely(ret < 0))
3280 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3281 if (unlikely(ret < 0))
3284 if (skb_is_gso(skb)) {
3285 struct skb_shared_info *shinfo = skb_shinfo(skb);
3287 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */
3288 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3289 shinfo->gso_type &= ~SKB_GSO_TCPV6;
3290 shinfo->gso_type |= SKB_GSO_TCPV4;
3294 skb->protocol = htons(ETH_P_IP);
3295 skb_clear_hash(skb);
3300 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3302 __be16 from_proto = skb->protocol;
3304 if (from_proto == htons(ETH_P_IP) &&
3305 to_proto == htons(ETH_P_IPV6))
3306 return bpf_skb_proto_4_to_6(skb);
3308 if (from_proto == htons(ETH_P_IPV6) &&
3309 to_proto == htons(ETH_P_IP))
3310 return bpf_skb_proto_6_to_4(skb);
3315 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3320 if (unlikely(flags))
3323 /* General idea is that this helper does the basic groundwork
3324 * needed for changing the protocol, and eBPF program fills the
3325 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3326 * and other helpers, rather than passing a raw buffer here.
3328 * The rationale is to keep this minimal and without a need to
3329 * deal with raw packet data. F.e. even if we would pass buffers
3330 * here, the program still needs to call the bpf_lX_csum_replace()
3331 * helpers anyway. Plus, this way we keep also separation of
3332 * concerns, since f.e. bpf_skb_store_bytes() should only take
3335 * Currently, additional options and extension header space are
3336 * not supported, but flags register is reserved so we can adapt
3337 * that. For offloads, we mark packet as dodgy, so that headers
3338 * need to be verified first.
3340 ret = bpf_skb_proto_xlat(skb, proto);
3341 bpf_compute_data_pointers(skb);
3345 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3346 .func = bpf_skb_change_proto,
3348 .ret_type = RET_INTEGER,
3349 .arg1_type = ARG_PTR_TO_CTX,
3350 .arg2_type = ARG_ANYTHING,
3351 .arg3_type = ARG_ANYTHING,
3354 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3356 /* We only allow a restricted subset to be changed for now. */
3357 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3358 !skb_pkt_type_ok(pkt_type)))
3361 skb->pkt_type = pkt_type;
3365 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3366 .func = bpf_skb_change_type,
3368 .ret_type = RET_INTEGER,
3369 .arg1_type = ARG_PTR_TO_CTX,
3370 .arg2_type = ARG_ANYTHING,
3373 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3375 switch (skb->protocol) {
3376 case htons(ETH_P_IP):
3377 return sizeof(struct iphdr);
3378 case htons(ETH_P_IPV6):
3379 return sizeof(struct ipv6hdr);
3385 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3386 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3388 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \
3389 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3390 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3391 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3392 BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \
3393 BPF_F_ADJ_ROOM_ENCAP_L2( \
3394 BPF_ADJ_ROOM_ENCAP_L2_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_NO_CSUM_RESET)))
3511 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3512 /* udp gso_size delineates datagrams, only allow if fixed */
3513 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3514 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3518 ret = skb_unclone(skb, GFP_ATOMIC);
3519 if (unlikely(ret < 0))
3522 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3523 if (unlikely(ret < 0))
3526 if (skb_is_gso(skb)) {
3527 struct skb_shared_info *shinfo = skb_shinfo(skb);
3529 /* Due to header shrink, MSS can be upgraded. */
3530 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3531 skb_increase_gso_size(shinfo, len_diff);
3533 /* Header must be checked, and gso_segs recomputed. */
3534 shinfo->gso_type |= SKB_GSO_DODGY;
3535 shinfo->gso_segs = 0;
3541 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3543 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3544 u32, mode, u64, flags)
3546 u32 len_diff_abs = abs(len_diff);
3547 bool shrink = len_diff < 0;
3550 if (unlikely(flags || mode))
3552 if (unlikely(len_diff_abs > 0xfffU))
3556 ret = skb_cow(skb, len_diff);
3557 if (unlikely(ret < 0))
3559 __skb_push(skb, len_diff_abs);
3560 memset(skb->data, 0, len_diff_abs);
3562 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3564 __skb_pull(skb, len_diff_abs);
3566 if (tls_sw_has_ctx_rx(skb->sk)) {
3567 struct strp_msg *rxm = strp_msg(skb);
3569 rxm->full_len += len_diff;
3574 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3575 .func = sk_skb_adjust_room,
3577 .ret_type = RET_INTEGER,
3578 .arg1_type = ARG_PTR_TO_CTX,
3579 .arg2_type = ARG_ANYTHING,
3580 .arg3_type = ARG_ANYTHING,
3581 .arg4_type = ARG_ANYTHING,
3584 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3585 u32, mode, u64, flags)
3587 u32 len_cur, len_diff_abs = abs(len_diff);
3588 u32 len_min = bpf_skb_net_base_len(skb);
3589 u32 len_max = BPF_SKB_MAX_LEN;
3590 __be16 proto = skb->protocol;
3591 bool shrink = len_diff < 0;
3595 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3596 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3598 if (unlikely(len_diff_abs > 0xfffU))
3600 if (unlikely(proto != htons(ETH_P_IP) &&
3601 proto != htons(ETH_P_IPV6)))
3604 off = skb_mac_header_len(skb);
3606 case BPF_ADJ_ROOM_NET:
3607 off += bpf_skb_net_base_len(skb);
3609 case BPF_ADJ_ROOM_MAC:
3615 len_cur = skb->len - skb_network_offset(skb);
3616 if ((shrink && (len_diff_abs >= len_cur ||
3617 len_cur - len_diff_abs < len_min)) ||
3618 (!shrink && (skb->len + len_diff_abs > len_max &&
3622 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3623 bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3624 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3625 __skb_reset_checksum_unnecessary(skb);
3627 bpf_compute_data_pointers(skb);
3631 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3632 .func = bpf_skb_adjust_room,
3634 .ret_type = RET_INTEGER,
3635 .arg1_type = ARG_PTR_TO_CTX,
3636 .arg2_type = ARG_ANYTHING,
3637 .arg3_type = ARG_ANYTHING,
3638 .arg4_type = ARG_ANYTHING,
3641 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3643 u32 min_len = skb_network_offset(skb);
3645 if (skb_transport_header_was_set(skb))
3646 min_len = skb_transport_offset(skb);
3647 if (skb->ip_summed == CHECKSUM_PARTIAL)
3648 min_len = skb_checksum_start_offset(skb) +
3649 skb->csum_offset + sizeof(__sum16);
3653 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3655 unsigned int old_len = skb->len;
3658 ret = __skb_grow_rcsum(skb, new_len);
3660 memset(skb->data + old_len, 0, new_len - old_len);
3664 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3666 return __skb_trim_rcsum(skb, new_len);
3669 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3672 u32 max_len = BPF_SKB_MAX_LEN;
3673 u32 min_len = __bpf_skb_min_len(skb);
3676 if (unlikely(flags || new_len > max_len || new_len < min_len))
3678 if (skb->encapsulation)
3681 /* The basic idea of this helper is that it's performing the
3682 * needed work to either grow or trim an skb, and eBPF program
3683 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3684 * bpf_lX_csum_replace() and others rather than passing a raw
3685 * buffer here. This one is a slow path helper and intended
3686 * for replies with control messages.
3688 * Like in bpf_skb_change_proto(), we want to keep this rather
3689 * minimal and without protocol specifics so that we are able
3690 * to separate concerns as in bpf_skb_store_bytes() should only
3691 * be the one responsible for writing buffers.
3693 * It's really expected to be a slow path operation here for
3694 * control message replies, so we're implicitly linearizing,
3695 * uncloning and drop offloads from the skb by this.
3697 ret = __bpf_try_make_writable(skb, skb->len);
3699 if (new_len > skb->len)
3700 ret = bpf_skb_grow_rcsum(skb, new_len);
3701 else if (new_len < skb->len)
3702 ret = bpf_skb_trim_rcsum(skb, new_len);
3703 if (!ret && skb_is_gso(skb))
3709 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3712 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3714 bpf_compute_data_pointers(skb);
3718 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3719 .func = bpf_skb_change_tail,
3721 .ret_type = RET_INTEGER,
3722 .arg1_type = ARG_PTR_TO_CTX,
3723 .arg2_type = ARG_ANYTHING,
3724 .arg3_type = ARG_ANYTHING,
3727 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3730 return __bpf_skb_change_tail(skb, new_len, flags);
3733 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3734 .func = sk_skb_change_tail,
3736 .ret_type = RET_INTEGER,
3737 .arg1_type = ARG_PTR_TO_CTX,
3738 .arg2_type = ARG_ANYTHING,
3739 .arg3_type = ARG_ANYTHING,
3742 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3745 u32 max_len = BPF_SKB_MAX_LEN;
3746 u32 new_len = skb->len + head_room;
3749 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3750 new_len < skb->len))
3753 ret = skb_cow(skb, head_room);
3755 /* Idea for this helper is that we currently only
3756 * allow to expand on mac header. This means that
3757 * skb->protocol network header, etc, stay as is.
3758 * Compared to bpf_skb_change_tail(), we're more
3759 * flexible due to not needing to linearize or
3760 * reset GSO. Intention for this helper is to be
3761 * used by an L3 skb that needs to push mac header
3762 * for redirection into L2 device.
3764 __skb_push(skb, head_room);
3765 memset(skb->data, 0, head_room);
3766 skb_reset_mac_header(skb);
3767 skb_reset_mac_len(skb);
3773 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3776 int ret = __bpf_skb_change_head(skb, head_room, flags);
3778 bpf_compute_data_pointers(skb);
3782 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3783 .func = bpf_skb_change_head,
3785 .ret_type = RET_INTEGER,
3786 .arg1_type = ARG_PTR_TO_CTX,
3787 .arg2_type = ARG_ANYTHING,
3788 .arg3_type = ARG_ANYTHING,
3791 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3794 return __bpf_skb_change_head(skb, head_room, flags);
3797 static const struct bpf_func_proto sk_skb_change_head_proto = {
3798 .func = sk_skb_change_head,
3800 .ret_type = RET_INTEGER,
3801 .arg1_type = ARG_PTR_TO_CTX,
3802 .arg2_type = ARG_ANYTHING,
3803 .arg3_type = ARG_ANYTHING,
3806 BPF_CALL_1(bpf_xdp_get_buff_len, struct xdp_buff*, xdp)
3808 return xdp_get_buff_len(xdp);
3811 static const struct bpf_func_proto bpf_xdp_get_buff_len_proto = {
3812 .func = bpf_xdp_get_buff_len,
3814 .ret_type = RET_INTEGER,
3815 .arg1_type = ARG_PTR_TO_CTX,
3818 BTF_ID_LIST_SINGLE(bpf_xdp_get_buff_len_bpf_ids, struct, xdp_buff)
3820 const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto = {
3821 .func = bpf_xdp_get_buff_len,
3823 .arg1_type = ARG_PTR_TO_BTF_ID,
3824 .arg1_btf_id = &bpf_xdp_get_buff_len_bpf_ids[0],
3827 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3829 return xdp_data_meta_unsupported(xdp) ? 0 :
3830 xdp->data - xdp->data_meta;
3833 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3835 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3836 unsigned long metalen = xdp_get_metalen(xdp);
3837 void *data_start = xdp_frame_end + metalen;
3838 void *data = xdp->data + offset;
3840 if (unlikely(data < data_start ||
3841 data > xdp->data_end - ETH_HLEN))
3845 memmove(xdp->data_meta + offset,
3846 xdp->data_meta, metalen);
3847 xdp->data_meta += offset;
3853 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3854 .func = bpf_xdp_adjust_head,
3856 .ret_type = RET_INTEGER,
3857 .arg1_type = ARG_PTR_TO_CTX,
3858 .arg2_type = ARG_ANYTHING,
3861 static void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off,
3862 void *buf, unsigned long len, bool flush)
3864 unsigned long ptr_len, ptr_off = 0;
3865 skb_frag_t *next_frag, *end_frag;
3866 struct skb_shared_info *sinfo;
3870 if (likely(xdp->data_end - xdp->data >= off + len)) {
3871 src = flush ? buf : xdp->data + off;
3872 dst = flush ? xdp->data + off : buf;
3873 memcpy(dst, src, len);
3877 sinfo = xdp_get_shared_info_from_buff(xdp);
3878 end_frag = &sinfo->frags[sinfo->nr_frags];
3879 next_frag = &sinfo->frags[0];
3881 ptr_len = xdp->data_end - xdp->data;
3882 ptr_buf = xdp->data;
3885 if (off < ptr_off + ptr_len) {
3886 unsigned long copy_off = off - ptr_off;
3887 unsigned long copy_len = min(len, ptr_len - copy_off);
3889 src = flush ? buf : ptr_buf + copy_off;
3890 dst = flush ? ptr_buf + copy_off : buf;
3891 memcpy(dst, src, copy_len);
3898 if (!len || next_frag == end_frag)
3902 ptr_buf = skb_frag_address(next_frag);
3903 ptr_len = skb_frag_size(next_frag);
3908 static void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len)
3910 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
3911 u32 size = xdp->data_end - xdp->data;
3912 void *addr = xdp->data;
3915 if (unlikely(offset > 0xffff || len > 0xffff))
3916 return ERR_PTR(-EFAULT);
3918 if (offset + len > xdp_get_buff_len(xdp))
3919 return ERR_PTR(-EINVAL);
3921 if (offset < size) /* linear area */
3925 for (i = 0; i < sinfo->nr_frags; i++) { /* paged area */
3926 u32 frag_size = skb_frag_size(&sinfo->frags[i]);
3928 if (offset < frag_size) {
3929 addr = skb_frag_address(&sinfo->frags[i]);
3933 offset -= frag_size;
3936 return offset + len <= size ? addr + offset : NULL;
3939 BPF_CALL_4(bpf_xdp_load_bytes, struct xdp_buff *, xdp, u32, offset,
3940 void *, buf, u32, len)
3944 ptr = bpf_xdp_pointer(xdp, offset, len);
3946 return PTR_ERR(ptr);
3949 bpf_xdp_copy_buf(xdp, offset, buf, len, false);
3951 memcpy(buf, ptr, len);
3956 static const struct bpf_func_proto bpf_xdp_load_bytes_proto = {
3957 .func = bpf_xdp_load_bytes,
3959 .ret_type = RET_INTEGER,
3960 .arg1_type = ARG_PTR_TO_CTX,
3961 .arg2_type = ARG_ANYTHING,
3962 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
3963 .arg4_type = ARG_CONST_SIZE,
3966 BPF_CALL_4(bpf_xdp_store_bytes, struct xdp_buff *, xdp, u32, offset,
3967 void *, buf, u32, len)
3971 ptr = bpf_xdp_pointer(xdp, offset, len);
3973 return PTR_ERR(ptr);
3976 bpf_xdp_copy_buf(xdp, offset, buf, len, true);
3978 memcpy(ptr, buf, len);
3983 static const struct bpf_func_proto bpf_xdp_store_bytes_proto = {
3984 .func = bpf_xdp_store_bytes,
3986 .ret_type = RET_INTEGER,
3987 .arg1_type = ARG_PTR_TO_CTX,
3988 .arg2_type = ARG_ANYTHING,
3989 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
3990 .arg4_type = ARG_CONST_SIZE,
3993 static int bpf_xdp_frags_increase_tail(struct xdp_buff *xdp, int offset)
3995 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
3996 skb_frag_t *frag = &sinfo->frags[sinfo->nr_frags - 1];
3997 struct xdp_rxq_info *rxq = xdp->rxq;
3998 unsigned int tailroom;
4000 if (!rxq->frag_size || rxq->frag_size > xdp->frame_sz)
4003 tailroom = rxq->frag_size - skb_frag_size(frag) - skb_frag_off(frag);
4004 if (unlikely(offset > tailroom))
4007 memset(skb_frag_address(frag) + skb_frag_size(frag), 0, offset);
4008 skb_frag_size_add(frag, offset);
4009 sinfo->xdp_frags_size += offset;
4014 static int bpf_xdp_frags_shrink_tail(struct xdp_buff *xdp, int offset)
4016 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
4017 int i, n_frags_free = 0, len_free = 0;
4019 if (unlikely(offset > (int)xdp_get_buff_len(xdp) - ETH_HLEN))
4022 for (i = sinfo->nr_frags - 1; i >= 0 && offset > 0; i--) {
4023 skb_frag_t *frag = &sinfo->frags[i];
4024 int shrink = min_t(int, offset, skb_frag_size(frag));
4029 if (skb_frag_size(frag) == shrink) {
4030 struct page *page = skb_frag_page(frag);
4032 __xdp_return(page_address(page), &xdp->rxq->mem,
4036 skb_frag_size_sub(frag, shrink);
4040 sinfo->nr_frags -= n_frags_free;
4041 sinfo->xdp_frags_size -= len_free;
4043 if (unlikely(!sinfo->nr_frags)) {
4044 xdp_buff_clear_frags_flag(xdp);
4045 xdp->data_end -= offset;
4051 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
4053 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
4054 void *data_end = xdp->data_end + offset;
4056 if (unlikely(xdp_buff_has_frags(xdp))) { /* non-linear xdp buff */
4058 return bpf_xdp_frags_shrink_tail(xdp, -offset);
4060 return bpf_xdp_frags_increase_tail(xdp, offset);
4063 /* Notice that xdp_data_hard_end have reserved some tailroom */
4064 if (unlikely(data_end > data_hard_end))
4067 /* ALL drivers MUST init xdp->frame_sz, chicken check below */
4068 if (unlikely(xdp->frame_sz > PAGE_SIZE)) {
4069 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz);
4073 if (unlikely(data_end < xdp->data + ETH_HLEN))
4076 /* Clear memory area on grow, can contain uninit kernel memory */
4078 memset(xdp->data_end, 0, offset);
4080 xdp->data_end = data_end;
4085 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
4086 .func = bpf_xdp_adjust_tail,
4088 .ret_type = RET_INTEGER,
4089 .arg1_type = ARG_PTR_TO_CTX,
4090 .arg2_type = ARG_ANYTHING,
4093 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
4095 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
4096 void *meta = xdp->data_meta + offset;
4097 unsigned long metalen = xdp->data - meta;
4099 if (xdp_data_meta_unsupported(xdp))
4101 if (unlikely(meta < xdp_frame_end ||
4104 if (unlikely(xdp_metalen_invalid(metalen)))
4107 xdp->data_meta = meta;
4112 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
4113 .func = bpf_xdp_adjust_meta,
4115 .ret_type = RET_INTEGER,
4116 .arg1_type = ARG_PTR_TO_CTX,
4117 .arg2_type = ARG_ANYTHING,
4120 /* XDP_REDIRECT works by a three-step process, implemented in the functions
4123 * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target
4124 * of the redirect and store it (along with some other metadata) in a per-CPU
4125 * struct bpf_redirect_info.
4127 * 2. When the program returns the XDP_REDIRECT return code, the driver will
4128 * call xdp_do_redirect() which will use the information in struct
4129 * bpf_redirect_info to actually enqueue the frame into a map type-specific
4130 * bulk queue structure.
4132 * 3. Before exiting its NAPI poll loop, the driver will call xdp_do_flush(),
4133 * which will flush all the different bulk queues, thus completing the
4136 * Pointers to the map entries will be kept around for this whole sequence of
4137 * steps, protected by RCU. However, there is no top-level rcu_read_lock() in
4138 * the core code; instead, the RCU protection relies on everything happening
4139 * inside a single NAPI poll sequence, which means it's between a pair of calls
4140 * to local_bh_disable()/local_bh_enable().
4142 * The map entries are marked as __rcu and the map code makes sure to
4143 * dereference those pointers with rcu_dereference_check() in a way that works
4144 * for both sections that to hold an rcu_read_lock() and sections that are
4145 * called from NAPI without a separate rcu_read_lock(). The code below does not
4146 * use RCU annotations, but relies on those in the map code.
4148 void xdp_do_flush(void)
4154 EXPORT_SYMBOL_GPL(xdp_do_flush);
4156 void bpf_clear_redirect_map(struct bpf_map *map)
4158 struct bpf_redirect_info *ri;
4161 for_each_possible_cpu(cpu) {
4162 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
4163 /* Avoid polluting remote cacheline due to writes if
4164 * not needed. Once we pass this test, we need the
4165 * cmpxchg() to make sure it hasn't been changed in
4166 * the meantime by remote CPU.
4168 if (unlikely(READ_ONCE(ri->map) == map))
4169 cmpxchg(&ri->map, map, NULL);
4173 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
4174 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key);
4176 u32 xdp_master_redirect(struct xdp_buff *xdp)
4178 struct net_device *master, *slave;
4179 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4181 master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev);
4182 slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp);
4183 if (slave && slave != xdp->rxq->dev) {
4184 /* The target device is different from the receiving device, so
4185 * redirect it to the new device.
4186 * Using XDP_REDIRECT gets the correct behaviour from XDP enabled
4187 * drivers to unmap the packet from their rx ring.
4189 ri->tgt_index = slave->ifindex;
4190 ri->map_id = INT_MAX;
4191 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4192 return XDP_REDIRECT;
4196 EXPORT_SYMBOL_GPL(xdp_master_redirect);
4198 static inline int __xdp_do_redirect_xsk(struct bpf_redirect_info *ri,
4199 struct net_device *dev,
4200 struct xdp_buff *xdp,
4201 struct bpf_prog *xdp_prog)
4203 enum bpf_map_type map_type = ri->map_type;
4204 void *fwd = ri->tgt_value;
4205 u32 map_id = ri->map_id;
4208 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4209 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4211 err = __xsk_map_redirect(fwd, xdp);
4215 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4218 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4222 static __always_inline int __xdp_do_redirect_frame(struct bpf_redirect_info *ri,
4223 struct net_device *dev,
4224 struct xdp_frame *xdpf,
4225 struct bpf_prog *xdp_prog)
4227 enum bpf_map_type map_type = ri->map_type;
4228 void *fwd = ri->tgt_value;
4229 u32 map_id = ri->map_id;
4230 struct bpf_map *map;
4233 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4234 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4236 if (unlikely(!xdpf)) {
4242 case BPF_MAP_TYPE_DEVMAP:
4244 case BPF_MAP_TYPE_DEVMAP_HASH:
4245 map = READ_ONCE(ri->map);
4246 if (unlikely(map)) {
4247 WRITE_ONCE(ri->map, NULL);
4248 err = dev_map_enqueue_multi(xdpf, dev, map,
4249 ri->flags & BPF_F_EXCLUDE_INGRESS);
4251 err = dev_map_enqueue(fwd, xdpf, dev);
4254 case BPF_MAP_TYPE_CPUMAP:
4255 err = cpu_map_enqueue(fwd, xdpf, dev);
4257 case BPF_MAP_TYPE_UNSPEC:
4258 if (map_id == INT_MAX) {
4259 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4260 if (unlikely(!fwd)) {
4264 err = dev_xdp_enqueue(fwd, xdpf, dev);
4275 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4278 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4282 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
4283 struct bpf_prog *xdp_prog)
4285 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4286 enum bpf_map_type map_type = ri->map_type;
4288 /* XDP_REDIRECT is not fully supported yet for xdp frags since
4289 * not all XDP capable drivers can map non-linear xdp_frame in
4292 if (unlikely(xdp_buff_has_frags(xdp) &&
4293 map_type != BPF_MAP_TYPE_CPUMAP))
4296 if (map_type == BPF_MAP_TYPE_XSKMAP)
4297 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4299 return __xdp_do_redirect_frame(ri, dev, xdp_convert_buff_to_frame(xdp),
4302 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4304 int xdp_do_redirect_frame(struct net_device *dev, struct xdp_buff *xdp,
4305 struct xdp_frame *xdpf, struct bpf_prog *xdp_prog)
4307 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4308 enum bpf_map_type map_type = ri->map_type;
4310 if (map_type == BPF_MAP_TYPE_XSKMAP)
4311 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4313 return __xdp_do_redirect_frame(ri, dev, xdpf, xdp_prog);
4315 EXPORT_SYMBOL_GPL(xdp_do_redirect_frame);
4317 static int xdp_do_generic_redirect_map(struct net_device *dev,
4318 struct sk_buff *skb,
4319 struct xdp_buff *xdp,
4320 struct bpf_prog *xdp_prog,
4322 enum bpf_map_type map_type, u32 map_id)
4324 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4325 struct bpf_map *map;
4329 case BPF_MAP_TYPE_DEVMAP:
4331 case BPF_MAP_TYPE_DEVMAP_HASH:
4332 map = READ_ONCE(ri->map);
4333 if (unlikely(map)) {
4334 WRITE_ONCE(ri->map, NULL);
4335 err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4336 ri->flags & BPF_F_EXCLUDE_INGRESS);
4338 err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4343 case BPF_MAP_TYPE_XSKMAP:
4344 err = xsk_generic_rcv(fwd, xdp);
4349 case BPF_MAP_TYPE_CPUMAP:
4350 err = cpu_map_generic_redirect(fwd, skb);
4359 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4362 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4366 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4367 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4369 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4370 enum bpf_map_type map_type = ri->map_type;
4371 void *fwd = ri->tgt_value;
4372 u32 map_id = ri->map_id;
4375 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4376 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4378 if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4379 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4380 if (unlikely(!fwd)) {
4385 err = xdp_ok_fwd_dev(fwd, skb->len);
4390 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4391 generic_xdp_tx(skb, xdp_prog);
4395 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id);
4397 _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4401 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4403 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4405 if (unlikely(flags))
4408 /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4409 * by map_idr) is used for ifindex based XDP redirect.
4411 ri->tgt_index = ifindex;
4412 ri->map_id = INT_MAX;
4413 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4415 return XDP_REDIRECT;
4418 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4419 .func = bpf_xdp_redirect,
4421 .ret_type = RET_INTEGER,
4422 .arg1_type = ARG_ANYTHING,
4423 .arg2_type = ARG_ANYTHING,
4426 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
4429 return map->ops->map_redirect(map, ifindex, flags);
4432 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4433 .func = bpf_xdp_redirect_map,
4435 .ret_type = RET_INTEGER,
4436 .arg1_type = ARG_CONST_MAP_PTR,
4437 .arg2_type = ARG_ANYTHING,
4438 .arg3_type = ARG_ANYTHING,
4441 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4442 unsigned long off, unsigned long len)
4444 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4448 if (ptr != dst_buff)
4449 memcpy(dst_buff, ptr, len);
4454 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4455 u64, flags, void *, meta, u64, meta_size)
4457 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4459 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4461 if (unlikely(!skb || skb_size > skb->len))
4464 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4468 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4469 .func = bpf_skb_event_output,
4471 .ret_type = RET_INTEGER,
4472 .arg1_type = ARG_PTR_TO_CTX,
4473 .arg2_type = ARG_CONST_MAP_PTR,
4474 .arg3_type = ARG_ANYTHING,
4475 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4476 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4479 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4481 const struct bpf_func_proto bpf_skb_output_proto = {
4482 .func = bpf_skb_event_output,
4484 .ret_type = RET_INTEGER,
4485 .arg1_type = ARG_PTR_TO_BTF_ID,
4486 .arg1_btf_id = &bpf_skb_output_btf_ids[0],
4487 .arg2_type = ARG_CONST_MAP_PTR,
4488 .arg3_type = ARG_ANYTHING,
4489 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4490 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4493 static unsigned short bpf_tunnel_key_af(u64 flags)
4495 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4498 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4499 u32, size, u64, flags)
4501 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4502 u8 compat[sizeof(struct bpf_tunnel_key)];
4506 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6 |
4507 BPF_F_TUNINFO_FLAGS)))) {
4511 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4515 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4518 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4519 case offsetof(struct bpf_tunnel_key, tunnel_label):
4520 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4522 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4523 /* Fixup deprecated structure layouts here, so we have
4524 * a common path later on.
4526 if (ip_tunnel_info_af(info) != AF_INET)
4529 to = (struct bpf_tunnel_key *)compat;
4536 to->tunnel_id = be64_to_cpu(info->key.tun_id);
4537 to->tunnel_tos = info->key.tos;
4538 to->tunnel_ttl = info->key.ttl;
4539 if (flags & BPF_F_TUNINFO_FLAGS)
4540 to->tunnel_flags = info->key.tun_flags;
4544 if (flags & BPF_F_TUNINFO_IPV6) {
4545 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4546 sizeof(to->remote_ipv6));
4547 memcpy(to->local_ipv6, &info->key.u.ipv6.dst,
4548 sizeof(to->local_ipv6));
4549 to->tunnel_label = be32_to_cpu(info->key.label);
4551 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4552 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4553 to->local_ipv4 = be32_to_cpu(info->key.u.ipv4.dst);
4554 memset(&to->local_ipv6[1], 0, sizeof(__u32) * 3);
4555 to->tunnel_label = 0;
4558 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4559 memcpy(to_orig, to, size);
4563 memset(to_orig, 0, size);
4567 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4568 .func = bpf_skb_get_tunnel_key,
4570 .ret_type = RET_INTEGER,
4571 .arg1_type = ARG_PTR_TO_CTX,
4572 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4573 .arg3_type = ARG_CONST_SIZE,
4574 .arg4_type = ARG_ANYTHING,
4577 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4579 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4582 if (unlikely(!info ||
4583 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4587 if (unlikely(size < info->options_len)) {
4592 ip_tunnel_info_opts_get(to, info);
4593 if (size > info->options_len)
4594 memset(to + info->options_len, 0, size - info->options_len);
4596 return info->options_len;
4598 memset(to, 0, size);
4602 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4603 .func = bpf_skb_get_tunnel_opt,
4605 .ret_type = RET_INTEGER,
4606 .arg1_type = ARG_PTR_TO_CTX,
4607 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4608 .arg3_type = ARG_CONST_SIZE,
4611 static struct metadata_dst __percpu *md_dst;
4613 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4614 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4616 struct metadata_dst *md = this_cpu_ptr(md_dst);
4617 u8 compat[sizeof(struct bpf_tunnel_key)];
4618 struct ip_tunnel_info *info;
4620 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4621 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
4623 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4625 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4626 case offsetof(struct bpf_tunnel_key, tunnel_label):
4627 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4628 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4629 /* Fixup deprecated structure layouts here, so we have
4630 * a common path later on.
4632 memcpy(compat, from, size);
4633 memset(compat + size, 0, sizeof(compat) - size);
4634 from = (const struct bpf_tunnel_key *) compat;
4640 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4645 dst_hold((struct dst_entry *) md);
4646 skb_dst_set(skb, (struct dst_entry *) md);
4648 info = &md->u.tun_info;
4649 memset(info, 0, sizeof(*info));
4650 info->mode = IP_TUNNEL_INFO_TX;
4652 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4653 if (flags & BPF_F_DONT_FRAGMENT)
4654 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4655 if (flags & BPF_F_ZERO_CSUM_TX)
4656 info->key.tun_flags &= ~TUNNEL_CSUM;
4657 if (flags & BPF_F_SEQ_NUMBER)
4658 info->key.tun_flags |= TUNNEL_SEQ;
4660 info->key.tun_id = cpu_to_be64(from->tunnel_id);
4661 info->key.tos = from->tunnel_tos;
4662 info->key.ttl = from->tunnel_ttl;
4664 if (flags & BPF_F_TUNINFO_IPV6) {
4665 info->mode |= IP_TUNNEL_INFO_IPV6;
4666 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4667 sizeof(from->remote_ipv6));
4668 memcpy(&info->key.u.ipv6.src, from->local_ipv6,
4669 sizeof(from->local_ipv6));
4670 info->key.label = cpu_to_be32(from->tunnel_label) &
4671 IPV6_FLOWLABEL_MASK;
4673 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4674 info->key.u.ipv4.src = cpu_to_be32(from->local_ipv4);
4675 info->key.flow_flags = FLOWI_FLAG_ANYSRC;
4681 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4682 .func = bpf_skb_set_tunnel_key,
4684 .ret_type = RET_INTEGER,
4685 .arg1_type = ARG_PTR_TO_CTX,
4686 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4687 .arg3_type = ARG_CONST_SIZE,
4688 .arg4_type = ARG_ANYTHING,
4691 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4692 const u8 *, from, u32, size)
4694 struct ip_tunnel_info *info = skb_tunnel_info(skb);
4695 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4697 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4699 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4702 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4707 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4708 .func = bpf_skb_set_tunnel_opt,
4710 .ret_type = RET_INTEGER,
4711 .arg1_type = ARG_PTR_TO_CTX,
4712 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4713 .arg3_type = ARG_CONST_SIZE,
4716 static const struct bpf_func_proto *
4717 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4720 struct metadata_dst __percpu *tmp;
4722 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4727 if (cmpxchg(&md_dst, NULL, tmp))
4728 metadata_dst_free_percpu(tmp);
4732 case BPF_FUNC_skb_set_tunnel_key:
4733 return &bpf_skb_set_tunnel_key_proto;
4734 case BPF_FUNC_skb_set_tunnel_opt:
4735 return &bpf_skb_set_tunnel_opt_proto;
4741 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4744 struct bpf_array *array = container_of(map, struct bpf_array, map);
4745 struct cgroup *cgrp;
4748 sk = skb_to_full_sk(skb);
4749 if (!sk || !sk_fullsock(sk))
4751 if (unlikely(idx >= array->map.max_entries))
4754 cgrp = READ_ONCE(array->ptrs[idx]);
4755 if (unlikely(!cgrp))
4758 return sk_under_cgroup_hierarchy(sk, cgrp);
4761 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4762 .func = bpf_skb_under_cgroup,
4764 .ret_type = RET_INTEGER,
4765 .arg1_type = ARG_PTR_TO_CTX,
4766 .arg2_type = ARG_CONST_MAP_PTR,
4767 .arg3_type = ARG_ANYTHING,
4770 #ifdef CONFIG_SOCK_CGROUP_DATA
4771 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4773 struct cgroup *cgrp;
4775 sk = sk_to_full_sk(sk);
4776 if (!sk || !sk_fullsock(sk))
4779 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4780 return cgroup_id(cgrp);
4783 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4785 return __bpf_sk_cgroup_id(skb->sk);
4788 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4789 .func = bpf_skb_cgroup_id,
4791 .ret_type = RET_INTEGER,
4792 .arg1_type = ARG_PTR_TO_CTX,
4795 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4798 struct cgroup *ancestor;
4799 struct cgroup *cgrp;
4801 sk = sk_to_full_sk(sk);
4802 if (!sk || !sk_fullsock(sk))
4805 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4806 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4810 return cgroup_id(ancestor);
4813 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4816 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4819 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4820 .func = bpf_skb_ancestor_cgroup_id,
4822 .ret_type = RET_INTEGER,
4823 .arg1_type = ARG_PTR_TO_CTX,
4824 .arg2_type = ARG_ANYTHING,
4827 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4829 return __bpf_sk_cgroup_id(sk);
4832 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4833 .func = bpf_sk_cgroup_id,
4835 .ret_type = RET_INTEGER,
4836 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4839 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4841 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4844 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4845 .func = bpf_sk_ancestor_cgroup_id,
4847 .ret_type = RET_INTEGER,
4848 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4849 .arg2_type = ARG_ANYTHING,
4853 static unsigned long bpf_xdp_copy(void *dst, const void *ctx,
4854 unsigned long off, unsigned long len)
4856 struct xdp_buff *xdp = (struct xdp_buff *)ctx;
4858 bpf_xdp_copy_buf(xdp, off, dst, len, false);
4862 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4863 u64, flags, void *, meta, u64, meta_size)
4865 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4867 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4870 if (unlikely(!xdp || xdp_size > xdp_get_buff_len(xdp)))
4873 return bpf_event_output(map, flags, meta, meta_size, xdp,
4874 xdp_size, bpf_xdp_copy);
4877 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4878 .func = bpf_xdp_event_output,
4880 .ret_type = RET_INTEGER,
4881 .arg1_type = ARG_PTR_TO_CTX,
4882 .arg2_type = ARG_CONST_MAP_PTR,
4883 .arg3_type = ARG_ANYTHING,
4884 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4885 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4888 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4890 const struct bpf_func_proto bpf_xdp_output_proto = {
4891 .func = bpf_xdp_event_output,
4893 .ret_type = RET_INTEGER,
4894 .arg1_type = ARG_PTR_TO_BTF_ID,
4895 .arg1_btf_id = &bpf_xdp_output_btf_ids[0],
4896 .arg2_type = ARG_CONST_MAP_PTR,
4897 .arg3_type = ARG_ANYTHING,
4898 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4899 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4902 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4904 return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4907 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4908 .func = bpf_get_socket_cookie,
4910 .ret_type = RET_INTEGER,
4911 .arg1_type = ARG_PTR_TO_CTX,
4914 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4916 return __sock_gen_cookie(ctx->sk);
4919 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4920 .func = bpf_get_socket_cookie_sock_addr,
4922 .ret_type = RET_INTEGER,
4923 .arg1_type = ARG_PTR_TO_CTX,
4926 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4928 return __sock_gen_cookie(ctx);
4931 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4932 .func = bpf_get_socket_cookie_sock,
4934 .ret_type = RET_INTEGER,
4935 .arg1_type = ARG_PTR_TO_CTX,
4938 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
4940 return sk ? sock_gen_cookie(sk) : 0;
4943 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
4944 .func = bpf_get_socket_ptr_cookie,
4946 .ret_type = RET_INTEGER,
4947 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4950 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4952 return __sock_gen_cookie(ctx->sk);
4955 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4956 .func = bpf_get_socket_cookie_sock_ops,
4958 .ret_type = RET_INTEGER,
4959 .arg1_type = ARG_PTR_TO_CTX,
4962 static u64 __bpf_get_netns_cookie(struct sock *sk)
4964 const struct net *net = sk ? sock_net(sk) : &init_net;
4966 return net->net_cookie;
4969 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
4971 return __bpf_get_netns_cookie(ctx);
4974 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
4975 .func = bpf_get_netns_cookie_sock,
4977 .ret_type = RET_INTEGER,
4978 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4981 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4983 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4986 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
4987 .func = bpf_get_netns_cookie_sock_addr,
4989 .ret_type = RET_INTEGER,
4990 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4993 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4995 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4998 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = {
4999 .func = bpf_get_netns_cookie_sock_ops,
5001 .ret_type = RET_INTEGER,
5002 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5005 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx)
5007 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5010 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = {
5011 .func = bpf_get_netns_cookie_sk_msg,
5013 .ret_type = RET_INTEGER,
5014 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5017 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
5019 struct sock *sk = sk_to_full_sk(skb->sk);
5022 if (!sk || !sk_fullsock(sk))
5024 kuid = sock_net_uid(sock_net(sk), sk);
5025 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
5028 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
5029 .func = bpf_get_socket_uid,
5031 .ret_type = RET_INTEGER,
5032 .arg1_type = ARG_PTR_TO_CTX,
5035 static int sol_socket_sockopt(struct sock *sk, int optname,
5036 char *optval, int *optlen,
5048 case SO_MAX_PACING_RATE:
5049 case SO_BINDTOIFINDEX:
5051 if (*optlen != sizeof(int))
5054 case SO_BINDTODEVICE:
5061 if (optname == SO_BINDTODEVICE)
5063 return sk_getsockopt(sk, SOL_SOCKET, optname,
5064 KERNEL_SOCKPTR(optval),
5065 KERNEL_SOCKPTR(optlen));
5068 return sk_setsockopt(sk, SOL_SOCKET, optname,
5069 KERNEL_SOCKPTR(optval), *optlen);
5072 static int bpf_sol_tcp_setsockopt(struct sock *sk, int optname,
5073 char *optval, int optlen)
5075 struct tcp_sock *tp = tcp_sk(sk);
5076 unsigned long timeout;
5079 if (optlen != sizeof(int))
5082 val = *(int *)optval;
5084 /* Only some options are supported */
5087 if (val <= 0 || tp->data_segs_out > tp->syn_data)
5089 tcp_snd_cwnd_set(tp, val);
5091 case TCP_BPF_SNDCWND_CLAMP:
5094 tp->snd_cwnd_clamp = val;
5095 tp->snd_ssthresh = val;
5097 case TCP_BPF_DELACK_MAX:
5098 timeout = usecs_to_jiffies(val);
5099 if (timeout > TCP_DELACK_MAX ||
5100 timeout < TCP_TIMEOUT_MIN)
5102 inet_csk(sk)->icsk_delack_max = timeout;
5104 case TCP_BPF_RTO_MIN:
5105 timeout = usecs_to_jiffies(val);
5106 if (timeout > TCP_RTO_MIN ||
5107 timeout < TCP_TIMEOUT_MIN)
5109 inet_csk(sk)->icsk_rto_min = timeout;
5118 static int sol_tcp_sockopt_congestion(struct sock *sk, char *optval,
5119 int *optlen, bool getopt)
5121 struct tcp_sock *tp;
5128 if (!inet_csk(sk)->icsk_ca_ops)
5130 /* BPF expects NULL-terminated tcp-cc string */
5131 optval[--(*optlen)] = '\0';
5132 return do_tcp_getsockopt(sk, SOL_TCP, TCP_CONGESTION,
5133 KERNEL_SOCKPTR(optval),
5134 KERNEL_SOCKPTR(optlen));
5137 /* "cdg" is the only cc that alloc a ptr
5138 * in inet_csk_ca area. The bpf-tcp-cc may
5139 * overwrite this ptr after switching to cdg.
5141 if (*optlen >= sizeof("cdg") - 1 && !strncmp("cdg", optval, *optlen))
5144 /* It stops this looping
5146 * .init => bpf_setsockopt(tcp_cc) => .init =>
5147 * bpf_setsockopt(tcp_cc)" => .init => ....
5149 * The second bpf_setsockopt(tcp_cc) is not allowed
5150 * in order to break the loop when both .init
5151 * are the same bpf prog.
5153 * This applies even the second bpf_setsockopt(tcp_cc)
5154 * does not cause a loop. This limits only the first
5155 * '.init' can call bpf_setsockopt(TCP_CONGESTION) to
5156 * pick a fallback cc (eg. peer does not support ECN)
5157 * and the second '.init' cannot fallback to
5161 if (tp->bpf_chg_cc_inprogress)
5164 tp->bpf_chg_cc_inprogress = 1;
5165 ret = do_tcp_setsockopt(sk, SOL_TCP, TCP_CONGESTION,
5166 KERNEL_SOCKPTR(optval), *optlen);
5167 tp->bpf_chg_cc_inprogress = 0;
5171 static int sol_tcp_sockopt(struct sock *sk, int optname,
5172 char *optval, int *optlen,
5175 if (sk->sk_prot->setsockopt != tcp_setsockopt)
5185 case TCP_WINDOW_CLAMP:
5186 case TCP_THIN_LINEAR_TIMEOUTS:
5187 case TCP_USER_TIMEOUT:
5188 case TCP_NOTSENT_LOWAT:
5190 if (*optlen != sizeof(int))
5193 case TCP_CONGESTION:
5194 return sol_tcp_sockopt_congestion(sk, optval, optlen, getopt);
5202 return bpf_sol_tcp_setsockopt(sk, optname, optval, *optlen);
5206 if (optname == TCP_SAVED_SYN) {
5207 struct tcp_sock *tp = tcp_sk(sk);
5209 if (!tp->saved_syn ||
5210 *optlen > tcp_saved_syn_len(tp->saved_syn))
5212 memcpy(optval, tp->saved_syn->data, *optlen);
5213 /* It cannot free tp->saved_syn here because it
5214 * does not know if the user space still needs it.
5219 return do_tcp_getsockopt(sk, SOL_TCP, optname,
5220 KERNEL_SOCKPTR(optval),
5221 KERNEL_SOCKPTR(optlen));
5224 return do_tcp_setsockopt(sk, SOL_TCP, optname,
5225 KERNEL_SOCKPTR(optval), *optlen);
5228 static int sol_ip_sockopt(struct sock *sk, int optname,
5229 char *optval, int *optlen,
5232 if (sk->sk_family != AF_INET)
5237 if (*optlen != sizeof(int))
5245 return do_ip_getsockopt(sk, SOL_IP, optname,
5246 KERNEL_SOCKPTR(optval),
5247 KERNEL_SOCKPTR(optlen));
5249 return do_ip_setsockopt(sk, SOL_IP, optname,
5250 KERNEL_SOCKPTR(optval), *optlen);
5253 static int sol_ipv6_sockopt(struct sock *sk, int optname,
5254 char *optval, int *optlen,
5257 if (sk->sk_family != AF_INET6)
5262 case IPV6_AUTOFLOWLABEL:
5263 if (*optlen != sizeof(int))
5271 return ipv6_bpf_stub->ipv6_getsockopt(sk, SOL_IPV6, optname,
5272 KERNEL_SOCKPTR(optval),
5273 KERNEL_SOCKPTR(optlen));
5275 return ipv6_bpf_stub->ipv6_setsockopt(sk, SOL_IPV6, optname,
5276 KERNEL_SOCKPTR(optval), *optlen);
5279 static int __bpf_setsockopt(struct sock *sk, int level, int optname,
5280 char *optval, int optlen)
5282 if (!sk_fullsock(sk))
5285 if (level == SOL_SOCKET)
5286 return sol_socket_sockopt(sk, optname, optval, &optlen, false);
5287 else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5288 return sol_ip_sockopt(sk, optname, optval, &optlen, false);
5289 else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5290 return sol_ipv6_sockopt(sk, optname, optval, &optlen, false);
5291 else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5292 return sol_tcp_sockopt(sk, optname, optval, &optlen, false);
5297 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
5298 char *optval, int optlen)
5300 if (sk_fullsock(sk))
5301 sock_owned_by_me(sk);
5302 return __bpf_setsockopt(sk, level, optname, optval, optlen);
5305 static int __bpf_getsockopt(struct sock *sk, int level, int optname,
5306 char *optval, int optlen)
5308 int err, saved_optlen = optlen;
5310 if (!sk_fullsock(sk)) {
5315 if (level == SOL_SOCKET)
5316 err = sol_socket_sockopt(sk, optname, optval, &optlen, true);
5317 else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5318 err = sol_tcp_sockopt(sk, optname, optval, &optlen, true);
5319 else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5320 err = sol_ip_sockopt(sk, optname, optval, &optlen, true);
5321 else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5322 err = sol_ipv6_sockopt(sk, optname, optval, &optlen, true);
5329 if (optlen < saved_optlen)
5330 memset(optval + optlen, 0, saved_optlen - optlen);
5334 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
5335 char *optval, int optlen)
5337 if (sk_fullsock(sk))
5338 sock_owned_by_me(sk);
5339 return __bpf_getsockopt(sk, level, optname, optval, optlen);
5342 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level,
5343 int, optname, char *, optval, int, optlen)
5345 return _bpf_setsockopt(sk, level, optname, optval, optlen);
5348 const struct bpf_func_proto bpf_sk_setsockopt_proto = {
5349 .func = bpf_sk_setsockopt,
5351 .ret_type = RET_INTEGER,
5352 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5353 .arg2_type = ARG_ANYTHING,
5354 .arg3_type = ARG_ANYTHING,
5355 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5356 .arg5_type = ARG_CONST_SIZE,
5359 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level,
5360 int, optname, char *, optval, int, optlen)
5362 return _bpf_getsockopt(sk, level, optname, optval, optlen);
5365 const struct bpf_func_proto bpf_sk_getsockopt_proto = {
5366 .func = bpf_sk_getsockopt,
5368 .ret_type = RET_INTEGER,
5369 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5370 .arg2_type = ARG_ANYTHING,
5371 .arg3_type = ARG_ANYTHING,
5372 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5373 .arg5_type = ARG_CONST_SIZE,
5376 BPF_CALL_5(bpf_unlocked_sk_setsockopt, struct sock *, sk, int, level,
5377 int, optname, char *, optval, int, optlen)
5379 return __bpf_setsockopt(sk, level, optname, optval, optlen);
5382 const struct bpf_func_proto bpf_unlocked_sk_setsockopt_proto = {
5383 .func = bpf_unlocked_sk_setsockopt,
5385 .ret_type = RET_INTEGER,
5386 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5387 .arg2_type = ARG_ANYTHING,
5388 .arg3_type = ARG_ANYTHING,
5389 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5390 .arg5_type = ARG_CONST_SIZE,
5393 BPF_CALL_5(bpf_unlocked_sk_getsockopt, struct sock *, sk, int, level,
5394 int, optname, char *, optval, int, optlen)
5396 return __bpf_getsockopt(sk, level, optname, optval, optlen);
5399 const struct bpf_func_proto bpf_unlocked_sk_getsockopt_proto = {
5400 .func = bpf_unlocked_sk_getsockopt,
5402 .ret_type = RET_INTEGER,
5403 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5404 .arg2_type = ARG_ANYTHING,
5405 .arg3_type = ARG_ANYTHING,
5406 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5407 .arg5_type = ARG_CONST_SIZE,
5410 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5411 int, level, int, optname, char *, optval, int, optlen)
5413 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5416 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5417 .func = bpf_sock_addr_setsockopt,
5419 .ret_type = RET_INTEGER,
5420 .arg1_type = ARG_PTR_TO_CTX,
5421 .arg2_type = ARG_ANYTHING,
5422 .arg3_type = ARG_ANYTHING,
5423 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5424 .arg5_type = ARG_CONST_SIZE,
5427 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5428 int, level, int, optname, char *, optval, int, optlen)
5430 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5433 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5434 .func = bpf_sock_addr_getsockopt,
5436 .ret_type = RET_INTEGER,
5437 .arg1_type = ARG_PTR_TO_CTX,
5438 .arg2_type = ARG_ANYTHING,
5439 .arg3_type = ARG_ANYTHING,
5440 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5441 .arg5_type = ARG_CONST_SIZE,
5444 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5445 int, level, int, optname, char *, optval, int, optlen)
5447 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5450 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5451 .func = bpf_sock_ops_setsockopt,
5453 .ret_type = RET_INTEGER,
5454 .arg1_type = ARG_PTR_TO_CTX,
5455 .arg2_type = ARG_ANYTHING,
5456 .arg3_type = ARG_ANYTHING,
5457 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5458 .arg5_type = ARG_CONST_SIZE,
5461 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5462 int optname, const u8 **start)
5464 struct sk_buff *syn_skb = bpf_sock->syn_skb;
5465 const u8 *hdr_start;
5469 /* sk is a request_sock here */
5471 if (optname == TCP_BPF_SYN) {
5472 hdr_start = syn_skb->data;
5473 ret = tcp_hdrlen(syn_skb);
5474 } else if (optname == TCP_BPF_SYN_IP) {
5475 hdr_start = skb_network_header(syn_skb);
5476 ret = skb_network_header_len(syn_skb) +
5477 tcp_hdrlen(syn_skb);
5479 /* optname == TCP_BPF_SYN_MAC */
5480 hdr_start = skb_mac_header(syn_skb);
5481 ret = skb_mac_header_len(syn_skb) +
5482 skb_network_header_len(syn_skb) +
5483 tcp_hdrlen(syn_skb);
5486 struct sock *sk = bpf_sock->sk;
5487 struct saved_syn *saved_syn;
5489 if (sk->sk_state == TCP_NEW_SYN_RECV)
5490 /* synack retransmit. bpf_sock->syn_skb will
5491 * not be available. It has to resort to
5492 * saved_syn (if it is saved).
5494 saved_syn = inet_reqsk(sk)->saved_syn;
5496 saved_syn = tcp_sk(sk)->saved_syn;
5501 if (optname == TCP_BPF_SYN) {
5502 hdr_start = saved_syn->data +
5503 saved_syn->mac_hdrlen +
5504 saved_syn->network_hdrlen;
5505 ret = saved_syn->tcp_hdrlen;
5506 } else if (optname == TCP_BPF_SYN_IP) {
5507 hdr_start = saved_syn->data +
5508 saved_syn->mac_hdrlen;
5509 ret = saved_syn->network_hdrlen +
5510 saved_syn->tcp_hdrlen;
5512 /* optname == TCP_BPF_SYN_MAC */
5514 /* TCP_SAVE_SYN may not have saved the mac hdr */
5515 if (!saved_syn->mac_hdrlen)
5518 hdr_start = saved_syn->data;
5519 ret = saved_syn->mac_hdrlen +
5520 saved_syn->network_hdrlen +
5521 saved_syn->tcp_hdrlen;
5529 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5530 int, level, int, optname, char *, optval, int, optlen)
5532 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5533 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5534 int ret, copy_len = 0;
5537 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5540 if (optlen < copy_len) {
5545 memcpy(optval, start, copy_len);
5548 /* Zero out unused buffer at the end */
5549 memset(optval + copy_len, 0, optlen - copy_len);
5554 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5557 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5558 .func = bpf_sock_ops_getsockopt,
5560 .ret_type = RET_INTEGER,
5561 .arg1_type = ARG_PTR_TO_CTX,
5562 .arg2_type = ARG_ANYTHING,
5563 .arg3_type = ARG_ANYTHING,
5564 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5565 .arg5_type = ARG_CONST_SIZE,
5568 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5571 struct sock *sk = bpf_sock->sk;
5572 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5574 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5577 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5579 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5582 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5583 .func = bpf_sock_ops_cb_flags_set,
5585 .ret_type = RET_INTEGER,
5586 .arg1_type = ARG_PTR_TO_CTX,
5587 .arg2_type = ARG_ANYTHING,
5590 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5591 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5593 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5597 struct sock *sk = ctx->sk;
5598 u32 flags = BIND_FROM_BPF;
5602 if (addr_len < offsetofend(struct sockaddr, sa_family))
5604 if (addr->sa_family == AF_INET) {
5605 if (addr_len < sizeof(struct sockaddr_in))
5607 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5608 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5609 return __inet_bind(sk, addr, addr_len, flags);
5610 #if IS_ENABLED(CONFIG_IPV6)
5611 } else if (addr->sa_family == AF_INET6) {
5612 if (addr_len < SIN6_LEN_RFC2133)
5614 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5615 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5616 /* ipv6_bpf_stub cannot be NULL, since it's called from
5617 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5619 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5620 #endif /* CONFIG_IPV6 */
5622 #endif /* CONFIG_INET */
5624 return -EAFNOSUPPORT;
5627 static const struct bpf_func_proto bpf_bind_proto = {
5630 .ret_type = RET_INTEGER,
5631 .arg1_type = ARG_PTR_TO_CTX,
5632 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5633 .arg3_type = ARG_CONST_SIZE,
5637 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5638 struct bpf_xfrm_state *, to, u32, size, u64, flags)
5640 const struct sec_path *sp = skb_sec_path(skb);
5641 const struct xfrm_state *x;
5643 if (!sp || unlikely(index >= sp->len || flags))
5646 x = sp->xvec[index];
5648 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5651 to->reqid = x->props.reqid;
5652 to->spi = x->id.spi;
5653 to->family = x->props.family;
5656 if (to->family == AF_INET6) {
5657 memcpy(to->remote_ipv6, x->props.saddr.a6,
5658 sizeof(to->remote_ipv6));
5660 to->remote_ipv4 = x->props.saddr.a4;
5661 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5666 memset(to, 0, size);
5670 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5671 .func = bpf_skb_get_xfrm_state,
5673 .ret_type = RET_INTEGER,
5674 .arg1_type = ARG_PTR_TO_CTX,
5675 .arg2_type = ARG_ANYTHING,
5676 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
5677 .arg4_type = ARG_CONST_SIZE,
5678 .arg5_type = ARG_ANYTHING,
5682 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5683 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
5684 const struct neighbour *neigh,
5685 const struct net_device *dev, u32 mtu)
5687 memcpy(params->dmac, neigh->ha, ETH_ALEN);
5688 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5689 params->h_vlan_TCI = 0;
5690 params->h_vlan_proto = 0;
5692 params->mtu_result = mtu; /* union with tot_len */
5698 #if IS_ENABLED(CONFIG_INET)
5699 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5700 u32 flags, bool check_mtu)
5702 struct fib_nh_common *nhc;
5703 struct in_device *in_dev;
5704 struct neighbour *neigh;
5705 struct net_device *dev;
5706 struct fib_result res;
5711 dev = dev_get_by_index_rcu(net, params->ifindex);
5715 /* verify forwarding is enabled on this interface */
5716 in_dev = __in_dev_get_rcu(dev);
5717 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5718 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5720 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5722 fl4.flowi4_oif = params->ifindex;
5724 fl4.flowi4_iif = params->ifindex;
5727 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5728 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5729 fl4.flowi4_flags = 0;
5731 fl4.flowi4_proto = params->l4_protocol;
5732 fl4.daddr = params->ipv4_dst;
5733 fl4.saddr = params->ipv4_src;
5734 fl4.fl4_sport = params->sport;
5735 fl4.fl4_dport = params->dport;
5736 fl4.flowi4_multipath_hash = 0;
5738 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5739 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5740 struct fib_table *tb;
5742 tb = fib_get_table(net, tbid);
5744 return BPF_FIB_LKUP_RET_NOT_FWDED;
5746 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5748 fl4.flowi4_mark = 0;
5749 fl4.flowi4_secid = 0;
5750 fl4.flowi4_tun_key.tun_id = 0;
5751 fl4.flowi4_uid = sock_net_uid(net, NULL);
5753 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5757 /* map fib lookup errors to RTN_ type */
5759 return BPF_FIB_LKUP_RET_BLACKHOLE;
5760 if (err == -EHOSTUNREACH)
5761 return BPF_FIB_LKUP_RET_UNREACHABLE;
5763 return BPF_FIB_LKUP_RET_PROHIBIT;
5765 return BPF_FIB_LKUP_RET_NOT_FWDED;
5768 if (res.type != RTN_UNICAST)
5769 return BPF_FIB_LKUP_RET_NOT_FWDED;
5771 if (fib_info_num_path(res.fi) > 1)
5772 fib_select_path(net, &res, &fl4, NULL);
5775 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5776 if (params->tot_len > mtu) {
5777 params->mtu_result = mtu; /* union with tot_len */
5778 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5784 /* do not handle lwt encaps right now */
5785 if (nhc->nhc_lwtstate)
5786 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5790 params->rt_metric = res.fi->fib_priority;
5791 params->ifindex = dev->ifindex;
5793 /* xdp and cls_bpf programs are run in RCU-bh so
5794 * rcu_read_lock_bh is not needed here
5796 if (likely(nhc->nhc_gw_family != AF_INET6)) {
5797 if (nhc->nhc_gw_family)
5798 params->ipv4_dst = nhc->nhc_gw.ipv4;
5800 neigh = __ipv4_neigh_lookup_noref(dev,
5801 (__force u32)params->ipv4_dst);
5803 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5805 params->family = AF_INET6;
5806 *dst = nhc->nhc_gw.ipv6;
5807 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5810 if (!neigh || !(neigh->nud_state & NUD_VALID))
5811 return BPF_FIB_LKUP_RET_NO_NEIGH;
5813 return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5817 #if IS_ENABLED(CONFIG_IPV6)
5818 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5819 u32 flags, bool check_mtu)
5821 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5822 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5823 struct fib6_result res = {};
5824 struct neighbour *neigh;
5825 struct net_device *dev;
5826 struct inet6_dev *idev;
5832 /* link local addresses are never forwarded */
5833 if (rt6_need_strict(dst) || rt6_need_strict(src))
5834 return BPF_FIB_LKUP_RET_NOT_FWDED;
5836 dev = dev_get_by_index_rcu(net, params->ifindex);
5840 idev = __in6_dev_get_safely(dev);
5841 if (unlikely(!idev || !idev->cnf.forwarding))
5842 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5844 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5846 oif = fl6.flowi6_oif = params->ifindex;
5848 oif = fl6.flowi6_iif = params->ifindex;
5850 strict = RT6_LOOKUP_F_HAS_SADDR;
5852 fl6.flowlabel = params->flowinfo;
5853 fl6.flowi6_scope = 0;
5854 fl6.flowi6_flags = 0;
5857 fl6.flowi6_proto = params->l4_protocol;
5860 fl6.fl6_sport = params->sport;
5861 fl6.fl6_dport = params->dport;
5863 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5864 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5865 struct fib6_table *tb;
5867 tb = ipv6_stub->fib6_get_table(net, tbid);
5869 return BPF_FIB_LKUP_RET_NOT_FWDED;
5871 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5874 fl6.flowi6_mark = 0;
5875 fl6.flowi6_secid = 0;
5876 fl6.flowi6_tun_key.tun_id = 0;
5877 fl6.flowi6_uid = sock_net_uid(net, NULL);
5879 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5882 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5883 res.f6i == net->ipv6.fib6_null_entry))
5884 return BPF_FIB_LKUP_RET_NOT_FWDED;
5886 switch (res.fib6_type) {
5887 /* only unicast is forwarded */
5891 return BPF_FIB_LKUP_RET_BLACKHOLE;
5892 case RTN_UNREACHABLE:
5893 return BPF_FIB_LKUP_RET_UNREACHABLE;
5895 return BPF_FIB_LKUP_RET_PROHIBIT;
5897 return BPF_FIB_LKUP_RET_NOT_FWDED;
5900 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5901 fl6.flowi6_oif != 0, NULL, strict);
5904 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5905 if (params->tot_len > mtu) {
5906 params->mtu_result = mtu; /* union with tot_len */
5907 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5911 if (res.nh->fib_nh_lws)
5912 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5914 if (res.nh->fib_nh_gw_family)
5915 *dst = res.nh->fib_nh_gw6;
5917 dev = res.nh->fib_nh_dev;
5918 params->rt_metric = res.f6i->fib6_metric;
5919 params->ifindex = dev->ifindex;
5921 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
5924 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5925 if (!neigh || !(neigh->nud_state & NUD_VALID))
5926 return BPF_FIB_LKUP_RET_NO_NEIGH;
5928 return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5932 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
5933 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5935 if (plen < sizeof(*params))
5938 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5941 switch (params->family) {
5942 #if IS_ENABLED(CONFIG_INET)
5944 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
5947 #if IS_ENABLED(CONFIG_IPV6)
5949 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
5953 return -EAFNOSUPPORT;
5956 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
5957 .func = bpf_xdp_fib_lookup,
5959 .ret_type = RET_INTEGER,
5960 .arg1_type = ARG_PTR_TO_CTX,
5961 .arg2_type = ARG_PTR_TO_MEM,
5962 .arg3_type = ARG_CONST_SIZE,
5963 .arg4_type = ARG_ANYTHING,
5966 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
5967 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5969 struct net *net = dev_net(skb->dev);
5970 int rc = -EAFNOSUPPORT;
5971 bool check_mtu = false;
5973 if (plen < sizeof(*params))
5976 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5979 if (params->tot_len)
5982 switch (params->family) {
5983 #if IS_ENABLED(CONFIG_INET)
5985 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
5988 #if IS_ENABLED(CONFIG_IPV6)
5990 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
5995 if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
5996 struct net_device *dev;
5998 /* When tot_len isn't provided by user, check skb
5999 * against MTU of FIB lookup resulting net_device
6001 dev = dev_get_by_index_rcu(net, params->ifindex);
6002 if (!is_skb_forwardable(dev, skb))
6003 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
6005 params->mtu_result = dev->mtu; /* union with tot_len */
6011 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
6012 .func = bpf_skb_fib_lookup,
6014 .ret_type = RET_INTEGER,
6015 .arg1_type = ARG_PTR_TO_CTX,
6016 .arg2_type = ARG_PTR_TO_MEM,
6017 .arg3_type = ARG_CONST_SIZE,
6018 .arg4_type = ARG_ANYTHING,
6021 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
6024 struct net *netns = dev_net(dev_curr);
6026 /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
6030 return dev_get_by_index_rcu(netns, ifindex);
6033 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
6034 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6036 int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6037 struct net_device *dev = skb->dev;
6038 int skb_len, dev_len;
6041 if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
6044 if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
6047 dev = __dev_via_ifindex(dev, ifindex);
6051 mtu = READ_ONCE(dev->mtu);
6053 dev_len = mtu + dev->hard_header_len;
6055 /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6056 skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
6058 skb_len += len_diff; /* minus result pass check */
6059 if (skb_len <= dev_len) {
6060 ret = BPF_MTU_CHK_RET_SUCCESS;
6063 /* At this point, skb->len exceed MTU, but as it include length of all
6064 * segments, it can still be below MTU. The SKB can possibly get
6065 * re-segmented in transmit path (see validate_xmit_skb). Thus, user
6066 * must choose if segs are to be MTU checked.
6068 if (skb_is_gso(skb)) {
6069 ret = BPF_MTU_CHK_RET_SUCCESS;
6071 if (flags & BPF_MTU_CHK_SEGS &&
6072 !skb_gso_validate_network_len(skb, mtu))
6073 ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
6076 /* BPF verifier guarantees valid pointer */
6082 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
6083 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6085 struct net_device *dev = xdp->rxq->dev;
6086 int xdp_len = xdp->data_end - xdp->data;
6087 int ret = BPF_MTU_CHK_RET_SUCCESS;
6090 /* XDP variant doesn't support multi-buffer segment check (yet) */
6091 if (unlikely(flags))
6094 dev = __dev_via_ifindex(dev, ifindex);
6098 mtu = READ_ONCE(dev->mtu);
6100 /* Add L2-header as dev MTU is L3 size */
6101 dev_len = mtu + dev->hard_header_len;
6103 /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6105 xdp_len = *mtu_len + dev->hard_header_len;
6107 xdp_len += len_diff; /* minus result pass check */
6108 if (xdp_len > dev_len)
6109 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6111 /* BPF verifier guarantees valid pointer */
6117 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
6118 .func = bpf_skb_check_mtu,
6120 .ret_type = RET_INTEGER,
6121 .arg1_type = ARG_PTR_TO_CTX,
6122 .arg2_type = ARG_ANYTHING,
6123 .arg3_type = ARG_PTR_TO_INT,
6124 .arg4_type = ARG_ANYTHING,
6125 .arg5_type = ARG_ANYTHING,
6128 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
6129 .func = bpf_xdp_check_mtu,
6131 .ret_type = RET_INTEGER,
6132 .arg1_type = ARG_PTR_TO_CTX,
6133 .arg2_type = ARG_ANYTHING,
6134 .arg3_type = ARG_PTR_TO_INT,
6135 .arg4_type = ARG_ANYTHING,
6136 .arg5_type = ARG_ANYTHING,
6139 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6140 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
6143 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
6145 if (!seg6_validate_srh(srh, len, false))
6149 case BPF_LWT_ENCAP_SEG6_INLINE:
6150 if (skb->protocol != htons(ETH_P_IPV6))
6153 err = seg6_do_srh_inline(skb, srh);
6155 case BPF_LWT_ENCAP_SEG6:
6156 skb_reset_inner_headers(skb);
6157 skb->encapsulation = 1;
6158 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
6164 bpf_compute_data_pointers(skb);
6168 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
6170 return seg6_lookup_nexthop(skb, NULL, 0);
6172 #endif /* CONFIG_IPV6_SEG6_BPF */
6174 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6175 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
6178 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
6182 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
6186 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6187 case BPF_LWT_ENCAP_SEG6:
6188 case BPF_LWT_ENCAP_SEG6_INLINE:
6189 return bpf_push_seg6_encap(skb, type, hdr, len);
6191 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6192 case BPF_LWT_ENCAP_IP:
6193 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
6200 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
6201 void *, hdr, u32, len)
6204 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6205 case BPF_LWT_ENCAP_IP:
6206 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
6213 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
6214 .func = bpf_lwt_in_push_encap,
6216 .ret_type = RET_INTEGER,
6217 .arg1_type = ARG_PTR_TO_CTX,
6218 .arg2_type = ARG_ANYTHING,
6219 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6220 .arg4_type = ARG_CONST_SIZE
6223 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
6224 .func = bpf_lwt_xmit_push_encap,
6226 .ret_type = RET_INTEGER,
6227 .arg1_type = ARG_PTR_TO_CTX,
6228 .arg2_type = ARG_ANYTHING,
6229 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6230 .arg4_type = ARG_CONST_SIZE
6233 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6234 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
6235 const void *, from, u32, len)
6237 struct seg6_bpf_srh_state *srh_state =
6238 this_cpu_ptr(&seg6_bpf_srh_states);
6239 struct ipv6_sr_hdr *srh = srh_state->srh;
6240 void *srh_tlvs, *srh_end, *ptr;
6246 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
6247 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
6249 ptr = skb->data + offset;
6250 if (ptr >= srh_tlvs && ptr + len <= srh_end)
6251 srh_state->valid = false;
6252 else if (ptr < (void *)&srh->flags ||
6253 ptr + len > (void *)&srh->segments)
6256 if (unlikely(bpf_try_make_writable(skb, offset + len)))
6258 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6260 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6262 memcpy(skb->data + offset, from, len);
6266 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
6267 .func = bpf_lwt_seg6_store_bytes,
6269 .ret_type = RET_INTEGER,
6270 .arg1_type = ARG_PTR_TO_CTX,
6271 .arg2_type = ARG_ANYTHING,
6272 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6273 .arg4_type = ARG_CONST_SIZE
6276 static void bpf_update_srh_state(struct sk_buff *skb)
6278 struct seg6_bpf_srh_state *srh_state =
6279 this_cpu_ptr(&seg6_bpf_srh_states);
6282 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
6283 srh_state->srh = NULL;
6285 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6286 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
6287 srh_state->valid = true;
6291 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
6292 u32, action, void *, param, u32, param_len)
6294 struct seg6_bpf_srh_state *srh_state =
6295 this_cpu_ptr(&seg6_bpf_srh_states);
6300 case SEG6_LOCAL_ACTION_END_X:
6301 if (!seg6_bpf_has_valid_srh(skb))
6303 if (param_len != sizeof(struct in6_addr))
6305 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
6306 case SEG6_LOCAL_ACTION_END_T:
6307 if (!seg6_bpf_has_valid_srh(skb))
6309 if (param_len != sizeof(int))
6311 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6312 case SEG6_LOCAL_ACTION_END_DT6:
6313 if (!seg6_bpf_has_valid_srh(skb))
6315 if (param_len != sizeof(int))
6318 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
6320 if (!pskb_pull(skb, hdroff))
6323 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
6324 skb_reset_network_header(skb);
6325 skb_reset_transport_header(skb);
6326 skb->encapsulation = 0;
6328 bpf_compute_data_pointers(skb);
6329 bpf_update_srh_state(skb);
6330 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6331 case SEG6_LOCAL_ACTION_END_B6:
6332 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6334 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
6337 bpf_update_srh_state(skb);
6340 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
6341 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6343 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
6346 bpf_update_srh_state(skb);
6354 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
6355 .func = bpf_lwt_seg6_action,
6357 .ret_type = RET_INTEGER,
6358 .arg1_type = ARG_PTR_TO_CTX,
6359 .arg2_type = ARG_ANYTHING,
6360 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6361 .arg4_type = ARG_CONST_SIZE
6364 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
6367 struct seg6_bpf_srh_state *srh_state =
6368 this_cpu_ptr(&seg6_bpf_srh_states);
6369 struct ipv6_sr_hdr *srh = srh_state->srh;
6370 void *srh_end, *srh_tlvs, *ptr;
6371 struct ipv6hdr *hdr;
6375 if (unlikely(srh == NULL))
6378 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
6379 ((srh->first_segment + 1) << 4));
6380 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
6382 ptr = skb->data + offset;
6384 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
6386 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
6390 ret = skb_cow_head(skb, len);
6391 if (unlikely(ret < 0))
6394 ret = bpf_skb_net_hdr_push(skb, offset, len);
6396 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6399 bpf_compute_data_pointers(skb);
6400 if (unlikely(ret < 0))
6403 hdr = (struct ipv6hdr *)skb->data;
6404 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6406 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6408 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6409 srh_state->hdrlen += len;
6410 srh_state->valid = false;
6414 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6415 .func = bpf_lwt_seg6_adjust_srh,
6417 .ret_type = RET_INTEGER,
6418 .arg1_type = ARG_PTR_TO_CTX,
6419 .arg2_type = ARG_ANYTHING,
6420 .arg3_type = ARG_ANYTHING,
6422 #endif /* CONFIG_IPV6_SEG6_BPF */
6425 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6426 int dif, int sdif, u8 family, u8 proto)
6428 struct inet_hashinfo *hinfo = net->ipv4.tcp_death_row.hashinfo;
6429 bool refcounted = false;
6430 struct sock *sk = NULL;
6432 if (family == AF_INET) {
6433 __be32 src4 = tuple->ipv4.saddr;
6434 __be32 dst4 = tuple->ipv4.daddr;
6436 if (proto == IPPROTO_TCP)
6437 sk = __inet_lookup(net, hinfo, NULL, 0,
6438 src4, tuple->ipv4.sport,
6439 dst4, tuple->ipv4.dport,
6440 dif, sdif, &refcounted);
6442 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6443 dst4, tuple->ipv4.dport,
6444 dif, sdif, &udp_table, NULL);
6445 #if IS_ENABLED(CONFIG_IPV6)
6447 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6448 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6450 if (proto == IPPROTO_TCP)
6451 sk = __inet6_lookup(net, hinfo, NULL, 0,
6452 src6, tuple->ipv6.sport,
6453 dst6, ntohs(tuple->ipv6.dport),
6454 dif, sdif, &refcounted);
6455 else if (likely(ipv6_bpf_stub))
6456 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6457 src6, tuple->ipv6.sport,
6458 dst6, tuple->ipv6.dport,
6464 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6465 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6471 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6472 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6474 static struct sock *
6475 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6476 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6477 u64 flags, int sdif)
6479 struct sock *sk = NULL;
6483 if (len == sizeof(tuple->ipv4))
6485 else if (len == sizeof(tuple->ipv6))
6490 if (unlikely(flags || !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6494 if (family == AF_INET)
6495 sdif = inet_sdif(skb);
6497 sdif = inet6_sdif(skb);
6500 if ((s32)netns_id < 0) {
6502 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6504 net = get_net_ns_by_id(caller_net, netns_id);
6507 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6515 static struct sock *
6516 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6517 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6518 u64 flags, int sdif)
6520 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6521 ifindex, proto, netns_id, flags,
6525 struct sock *sk2 = sk_to_full_sk(sk);
6527 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6528 * sock refcnt is decremented to prevent a request_sock leak.
6530 if (!sk_fullsock(sk2))
6534 /* Ensure there is no need to bump sk2 refcnt */
6535 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6536 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6546 static struct sock *
6547 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6548 u8 proto, u64 netns_id, u64 flags)
6550 struct net *caller_net;
6554 caller_net = dev_net(skb->dev);
6555 ifindex = skb->dev->ifindex;
6557 caller_net = sock_net(skb->sk);
6561 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6562 netns_id, flags, -1);
6565 static struct sock *
6566 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6567 u8 proto, u64 netns_id, u64 flags)
6569 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6573 struct sock *sk2 = sk_to_full_sk(sk);
6575 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6576 * sock refcnt is decremented to prevent a request_sock leak.
6578 if (!sk_fullsock(sk2))
6582 /* Ensure there is no need to bump sk2 refcnt */
6583 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6584 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6594 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6595 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6597 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6601 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6602 .func = bpf_skc_lookup_tcp,
6605 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6606 .arg1_type = ARG_PTR_TO_CTX,
6607 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6608 .arg3_type = ARG_CONST_SIZE,
6609 .arg4_type = ARG_ANYTHING,
6610 .arg5_type = ARG_ANYTHING,
6613 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6614 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6616 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6620 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6621 .func = bpf_sk_lookup_tcp,
6624 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6625 .arg1_type = ARG_PTR_TO_CTX,
6626 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6627 .arg3_type = ARG_CONST_SIZE,
6628 .arg4_type = ARG_ANYTHING,
6629 .arg5_type = ARG_ANYTHING,
6632 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6633 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6635 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6639 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6640 .func = bpf_sk_lookup_udp,
6643 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6644 .arg1_type = ARG_PTR_TO_CTX,
6645 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6646 .arg3_type = ARG_CONST_SIZE,
6647 .arg4_type = ARG_ANYTHING,
6648 .arg5_type = ARG_ANYTHING,
6651 BPF_CALL_5(bpf_tc_skc_lookup_tcp, struct sk_buff *, skb,
6652 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6654 struct net_device *dev = skb->dev;
6655 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6656 struct net *caller_net = dev_net(dev);
6658 return (unsigned long)__bpf_skc_lookup(skb, tuple, len, caller_net,
6659 ifindex, IPPROTO_TCP, netns_id,
6663 static const struct bpf_func_proto bpf_tc_skc_lookup_tcp_proto = {
6664 .func = bpf_tc_skc_lookup_tcp,
6667 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6668 .arg1_type = ARG_PTR_TO_CTX,
6669 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6670 .arg3_type = ARG_CONST_SIZE,
6671 .arg4_type = ARG_ANYTHING,
6672 .arg5_type = ARG_ANYTHING,
6675 BPF_CALL_5(bpf_tc_sk_lookup_tcp, struct sk_buff *, skb,
6676 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6678 struct net_device *dev = skb->dev;
6679 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6680 struct net *caller_net = dev_net(dev);
6682 return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6683 ifindex, IPPROTO_TCP, netns_id,
6687 static const struct bpf_func_proto bpf_tc_sk_lookup_tcp_proto = {
6688 .func = bpf_tc_sk_lookup_tcp,
6691 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6692 .arg1_type = ARG_PTR_TO_CTX,
6693 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6694 .arg3_type = ARG_CONST_SIZE,
6695 .arg4_type = ARG_ANYTHING,
6696 .arg5_type = ARG_ANYTHING,
6699 BPF_CALL_5(bpf_tc_sk_lookup_udp, struct sk_buff *, skb,
6700 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6702 struct net_device *dev = skb->dev;
6703 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6704 struct net *caller_net = dev_net(dev);
6706 return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6707 ifindex, IPPROTO_UDP, netns_id,
6711 static const struct bpf_func_proto bpf_tc_sk_lookup_udp_proto = {
6712 .func = bpf_tc_sk_lookup_udp,
6715 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6716 .arg1_type = ARG_PTR_TO_CTX,
6717 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6718 .arg3_type = ARG_CONST_SIZE,
6719 .arg4_type = ARG_ANYTHING,
6720 .arg5_type = ARG_ANYTHING,
6723 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6725 if (sk && sk_is_refcounted(sk))
6730 static const struct bpf_func_proto bpf_sk_release_proto = {
6731 .func = bpf_sk_release,
6733 .ret_type = RET_INTEGER,
6734 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON | OBJ_RELEASE,
6737 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6738 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6740 struct net_device *dev = ctx->rxq->dev;
6741 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6742 struct net *caller_net = dev_net(dev);
6744 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6745 ifindex, IPPROTO_UDP, netns_id,
6749 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6750 .func = bpf_xdp_sk_lookup_udp,
6753 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6754 .arg1_type = ARG_PTR_TO_CTX,
6755 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6756 .arg3_type = ARG_CONST_SIZE,
6757 .arg4_type = ARG_ANYTHING,
6758 .arg5_type = ARG_ANYTHING,
6761 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6762 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6764 struct net_device *dev = ctx->rxq->dev;
6765 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6766 struct net *caller_net = dev_net(dev);
6768 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6769 ifindex, IPPROTO_TCP, netns_id,
6773 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6774 .func = bpf_xdp_skc_lookup_tcp,
6777 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6778 .arg1_type = ARG_PTR_TO_CTX,
6779 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6780 .arg3_type = ARG_CONST_SIZE,
6781 .arg4_type = ARG_ANYTHING,
6782 .arg5_type = ARG_ANYTHING,
6785 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6786 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6788 struct net_device *dev = ctx->rxq->dev;
6789 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6790 struct net *caller_net = dev_net(dev);
6792 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6793 ifindex, IPPROTO_TCP, netns_id,
6797 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6798 .func = bpf_xdp_sk_lookup_tcp,
6801 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6802 .arg1_type = ARG_PTR_TO_CTX,
6803 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6804 .arg3_type = ARG_CONST_SIZE,
6805 .arg4_type = ARG_ANYTHING,
6806 .arg5_type = ARG_ANYTHING,
6809 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6810 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6812 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6813 sock_net(ctx->sk), 0,
6814 IPPROTO_TCP, netns_id, flags,
6818 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6819 .func = bpf_sock_addr_skc_lookup_tcp,
6821 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6822 .arg1_type = ARG_PTR_TO_CTX,
6823 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6824 .arg3_type = ARG_CONST_SIZE,
6825 .arg4_type = ARG_ANYTHING,
6826 .arg5_type = ARG_ANYTHING,
6829 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6830 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6832 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6833 sock_net(ctx->sk), 0, IPPROTO_TCP,
6834 netns_id, flags, -1);
6837 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6838 .func = bpf_sock_addr_sk_lookup_tcp,
6840 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6841 .arg1_type = ARG_PTR_TO_CTX,
6842 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6843 .arg3_type = ARG_CONST_SIZE,
6844 .arg4_type = ARG_ANYTHING,
6845 .arg5_type = ARG_ANYTHING,
6848 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6849 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6851 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6852 sock_net(ctx->sk), 0, IPPROTO_UDP,
6853 netns_id, flags, -1);
6856 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6857 .func = bpf_sock_addr_sk_lookup_udp,
6859 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6860 .arg1_type = ARG_PTR_TO_CTX,
6861 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6862 .arg3_type = ARG_CONST_SIZE,
6863 .arg4_type = ARG_ANYTHING,
6864 .arg5_type = ARG_ANYTHING,
6867 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6868 struct bpf_insn_access_aux *info)
6870 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6874 if (off % size != 0)
6878 case offsetof(struct bpf_tcp_sock, bytes_received):
6879 case offsetof(struct bpf_tcp_sock, bytes_acked):
6880 return size == sizeof(__u64);
6882 return size == sizeof(__u32);
6886 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6887 const struct bpf_insn *si,
6888 struct bpf_insn *insn_buf,
6889 struct bpf_prog *prog, u32 *target_size)
6891 struct bpf_insn *insn = insn_buf;
6893 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \
6895 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \
6896 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6897 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6898 si->dst_reg, si->src_reg, \
6899 offsetof(struct tcp_sock, FIELD)); \
6902 #define BPF_INET_SOCK_GET_COMMON(FIELD) \
6904 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \
6906 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6907 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6908 struct inet_connection_sock, \
6910 si->dst_reg, si->src_reg, \
6912 struct inet_connection_sock, \
6916 if (insn > insn_buf)
6917 return insn - insn_buf;
6920 case offsetof(struct bpf_tcp_sock, rtt_min):
6921 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
6922 sizeof(struct minmax));
6923 BUILD_BUG_ON(sizeof(struct minmax) <
6924 sizeof(struct minmax_sample));
6926 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6927 offsetof(struct tcp_sock, rtt_min) +
6928 offsetof(struct minmax_sample, v));
6930 case offsetof(struct bpf_tcp_sock, snd_cwnd):
6931 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
6933 case offsetof(struct bpf_tcp_sock, srtt_us):
6934 BPF_TCP_SOCK_GET_COMMON(srtt_us);
6936 case offsetof(struct bpf_tcp_sock, snd_ssthresh):
6937 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
6939 case offsetof(struct bpf_tcp_sock, rcv_nxt):
6940 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
6942 case offsetof(struct bpf_tcp_sock, snd_nxt):
6943 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
6945 case offsetof(struct bpf_tcp_sock, snd_una):
6946 BPF_TCP_SOCK_GET_COMMON(snd_una);
6948 case offsetof(struct bpf_tcp_sock, mss_cache):
6949 BPF_TCP_SOCK_GET_COMMON(mss_cache);
6951 case offsetof(struct bpf_tcp_sock, ecn_flags):
6952 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
6954 case offsetof(struct bpf_tcp_sock, rate_delivered):
6955 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
6957 case offsetof(struct bpf_tcp_sock, rate_interval_us):
6958 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
6960 case offsetof(struct bpf_tcp_sock, packets_out):
6961 BPF_TCP_SOCK_GET_COMMON(packets_out);
6963 case offsetof(struct bpf_tcp_sock, retrans_out):
6964 BPF_TCP_SOCK_GET_COMMON(retrans_out);
6966 case offsetof(struct bpf_tcp_sock, total_retrans):
6967 BPF_TCP_SOCK_GET_COMMON(total_retrans);
6969 case offsetof(struct bpf_tcp_sock, segs_in):
6970 BPF_TCP_SOCK_GET_COMMON(segs_in);
6972 case offsetof(struct bpf_tcp_sock, data_segs_in):
6973 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
6975 case offsetof(struct bpf_tcp_sock, segs_out):
6976 BPF_TCP_SOCK_GET_COMMON(segs_out);
6978 case offsetof(struct bpf_tcp_sock, data_segs_out):
6979 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
6981 case offsetof(struct bpf_tcp_sock, lost_out):
6982 BPF_TCP_SOCK_GET_COMMON(lost_out);
6984 case offsetof(struct bpf_tcp_sock, sacked_out):
6985 BPF_TCP_SOCK_GET_COMMON(sacked_out);
6987 case offsetof(struct bpf_tcp_sock, bytes_received):
6988 BPF_TCP_SOCK_GET_COMMON(bytes_received);
6990 case offsetof(struct bpf_tcp_sock, bytes_acked):
6991 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
6993 case offsetof(struct bpf_tcp_sock, dsack_dups):
6994 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
6996 case offsetof(struct bpf_tcp_sock, delivered):
6997 BPF_TCP_SOCK_GET_COMMON(delivered);
6999 case offsetof(struct bpf_tcp_sock, delivered_ce):
7000 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
7002 case offsetof(struct bpf_tcp_sock, icsk_retransmits):
7003 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
7007 return insn - insn_buf;
7010 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
7012 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
7013 return (unsigned long)sk;
7015 return (unsigned long)NULL;
7018 const struct bpf_func_proto bpf_tcp_sock_proto = {
7019 .func = bpf_tcp_sock,
7021 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
7022 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
7025 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
7027 sk = sk_to_full_sk(sk);
7029 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
7030 return (unsigned long)sk;
7032 return (unsigned long)NULL;
7035 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
7036 .func = bpf_get_listener_sock,
7038 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
7039 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
7042 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
7044 unsigned int iphdr_len;
7046 switch (skb_protocol(skb, true)) {
7047 case cpu_to_be16(ETH_P_IP):
7048 iphdr_len = sizeof(struct iphdr);
7050 case cpu_to_be16(ETH_P_IPV6):
7051 iphdr_len = sizeof(struct ipv6hdr);
7057 if (skb_headlen(skb) < iphdr_len)
7060 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
7063 return INET_ECN_set_ce(skb);
7066 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7067 struct bpf_insn_access_aux *info)
7069 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
7072 if (off % size != 0)
7077 return size == sizeof(__u32);
7081 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
7082 const struct bpf_insn *si,
7083 struct bpf_insn *insn_buf,
7084 struct bpf_prog *prog, u32 *target_size)
7086 struct bpf_insn *insn = insn_buf;
7088 #define BPF_XDP_SOCK_GET(FIELD) \
7090 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \
7091 sizeof_field(struct bpf_xdp_sock, FIELD)); \
7092 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
7093 si->dst_reg, si->src_reg, \
7094 offsetof(struct xdp_sock, FIELD)); \
7098 case offsetof(struct bpf_xdp_sock, queue_id):
7099 BPF_XDP_SOCK_GET(queue_id);
7103 return insn - insn_buf;
7106 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
7107 .func = bpf_skb_ecn_set_ce,
7109 .ret_type = RET_INTEGER,
7110 .arg1_type = ARG_PTR_TO_CTX,
7113 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7114 struct tcphdr *, th, u32, th_len)
7116 #ifdef CONFIG_SYN_COOKIES
7120 if (unlikely(!sk || th_len < sizeof(*th)))
7123 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
7124 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7127 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7130 if (!th->ack || th->rst || th->syn)
7133 if (unlikely(iph_len < sizeof(struct iphdr)))
7136 if (tcp_synq_no_recent_overflow(sk))
7139 cookie = ntohl(th->ack_seq) - 1;
7141 /* Both struct iphdr and struct ipv6hdr have the version field at the
7142 * same offset so we can cast to the shorter header (struct iphdr).
7144 switch (((struct iphdr *)iph)->version) {
7146 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7149 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
7152 #if IS_BUILTIN(CONFIG_IPV6)
7154 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7157 if (sk->sk_family != AF_INET6)
7160 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
7162 #endif /* CONFIG_IPV6 */
7165 return -EPROTONOSUPPORT;
7177 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
7178 .func = bpf_tcp_check_syncookie,
7181 .ret_type = RET_INTEGER,
7182 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7183 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7184 .arg3_type = ARG_CONST_SIZE,
7185 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7186 .arg5_type = ARG_CONST_SIZE,
7189 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7190 struct tcphdr *, th, u32, th_len)
7192 #ifdef CONFIG_SYN_COOKIES
7196 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
7199 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7202 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7205 if (!th->syn || th->ack || th->fin || th->rst)
7208 if (unlikely(iph_len < sizeof(struct iphdr)))
7211 /* Both struct iphdr and struct ipv6hdr have the version field at the
7212 * same offset so we can cast to the shorter header (struct iphdr).
7214 switch (((struct iphdr *)iph)->version) {
7216 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7219 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
7222 #if IS_BUILTIN(CONFIG_IPV6)
7224 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7227 if (sk->sk_family != AF_INET6)
7230 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
7232 #endif /* CONFIG_IPV6 */
7235 return -EPROTONOSUPPORT;
7240 return cookie | ((u64)mss << 32);
7243 #endif /* CONFIG_SYN_COOKIES */
7246 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
7247 .func = bpf_tcp_gen_syncookie,
7248 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */
7250 .ret_type = RET_INTEGER,
7251 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7252 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7253 .arg3_type = ARG_CONST_SIZE,
7254 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7255 .arg5_type = ARG_CONST_SIZE,
7258 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
7260 if (!sk || flags != 0)
7262 if (!skb_at_tc_ingress(skb))
7264 if (unlikely(dev_net(skb->dev) != sock_net(sk)))
7265 return -ENETUNREACH;
7266 if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
7267 return -ESOCKTNOSUPPORT;
7268 if (sk_is_refcounted(sk) &&
7269 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
7274 skb->destructor = sock_pfree;
7279 static const struct bpf_func_proto bpf_sk_assign_proto = {
7280 .func = bpf_sk_assign,
7282 .ret_type = RET_INTEGER,
7283 .arg1_type = ARG_PTR_TO_CTX,
7284 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7285 .arg3_type = ARG_ANYTHING,
7288 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
7289 u8 search_kind, const u8 *magic,
7290 u8 magic_len, bool *eol)
7296 while (op < opend) {
7299 if (kind == TCPOPT_EOL) {
7301 return ERR_PTR(-ENOMSG);
7302 } else if (kind == TCPOPT_NOP) {
7307 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
7308 /* Something is wrong in the received header.
7309 * Follow the TCP stack's tcp_parse_options()
7310 * and just bail here.
7312 return ERR_PTR(-EFAULT);
7315 if (search_kind == kind) {
7319 if (magic_len > kind_len - 2)
7320 return ERR_PTR(-ENOMSG);
7322 if (!memcmp(&op[2], magic, magic_len))
7329 return ERR_PTR(-ENOMSG);
7332 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7333 void *, search_res, u32, len, u64, flags)
7335 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
7336 const u8 *op, *opend, *magic, *search = search_res;
7337 u8 search_kind, search_len, copy_len, magic_len;
7340 /* 2 byte is the minimal option len except TCPOPT_NOP and
7341 * TCPOPT_EOL which are useless for the bpf prog to learn
7342 * and this helper disallow loading them also.
7344 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
7347 search_kind = search[0];
7348 search_len = search[1];
7350 if (search_len > len || search_kind == TCPOPT_NOP ||
7351 search_kind == TCPOPT_EOL)
7354 if (search_kind == TCPOPT_EXP || search_kind == 253) {
7355 /* 16 or 32 bit magic. +2 for kind and kind length */
7356 if (search_len != 4 && search_len != 6)
7359 magic_len = search_len - 2;
7368 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
7373 op += sizeof(struct tcphdr);
7375 if (!bpf_sock->skb ||
7376 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7377 /* This bpf_sock->op cannot call this helper */
7380 opend = bpf_sock->skb_data_end;
7381 op = bpf_sock->skb->data + sizeof(struct tcphdr);
7384 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
7391 if (copy_len > len) {
7396 memcpy(search_res, op, copy_len);
7400 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
7401 .func = bpf_sock_ops_load_hdr_opt,
7403 .ret_type = RET_INTEGER,
7404 .arg1_type = ARG_PTR_TO_CTX,
7405 .arg2_type = ARG_PTR_TO_MEM,
7406 .arg3_type = ARG_CONST_SIZE,
7407 .arg4_type = ARG_ANYTHING,
7410 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7411 const void *, from, u32, len, u64, flags)
7413 u8 new_kind, new_kind_len, magic_len = 0, *opend;
7414 const u8 *op, *new_op, *magic = NULL;
7415 struct sk_buff *skb;
7418 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
7421 if (len < 2 || flags)
7425 new_kind = new_op[0];
7426 new_kind_len = new_op[1];
7428 if (new_kind_len > len || new_kind == TCPOPT_NOP ||
7429 new_kind == TCPOPT_EOL)
7432 if (new_kind_len > bpf_sock->remaining_opt_len)
7435 /* 253 is another experimental kind */
7436 if (new_kind == TCPOPT_EXP || new_kind == 253) {
7437 if (new_kind_len < 4)
7439 /* Match for the 2 byte magic also.
7440 * RFC 6994: the magic could be 2 or 4 bytes.
7441 * Hence, matching by 2 byte only is on the
7442 * conservative side but it is the right
7443 * thing to do for the 'search-for-duplication'
7450 /* Check for duplication */
7451 skb = bpf_sock->skb;
7452 op = skb->data + sizeof(struct tcphdr);
7453 opend = bpf_sock->skb_data_end;
7455 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
7460 if (PTR_ERR(op) != -ENOMSG)
7464 /* The option has been ended. Treat it as no more
7465 * header option can be written.
7469 /* No duplication found. Store the header option. */
7470 memcpy(opend, from, new_kind_len);
7472 bpf_sock->remaining_opt_len -= new_kind_len;
7473 bpf_sock->skb_data_end += new_kind_len;
7478 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
7479 .func = bpf_sock_ops_store_hdr_opt,
7481 .ret_type = RET_INTEGER,
7482 .arg1_type = ARG_PTR_TO_CTX,
7483 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7484 .arg3_type = ARG_CONST_SIZE,
7485 .arg4_type = ARG_ANYTHING,
7488 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7489 u32, len, u64, flags)
7491 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7494 if (flags || len < 2)
7497 if (len > bpf_sock->remaining_opt_len)
7500 bpf_sock->remaining_opt_len -= len;
7505 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7506 .func = bpf_sock_ops_reserve_hdr_opt,
7508 .ret_type = RET_INTEGER,
7509 .arg1_type = ARG_PTR_TO_CTX,
7510 .arg2_type = ARG_ANYTHING,
7511 .arg3_type = ARG_ANYTHING,
7514 BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb,
7515 u64, tstamp, u32, tstamp_type)
7517 /* skb_clear_delivery_time() is done for inet protocol */
7518 if (skb->protocol != htons(ETH_P_IP) &&
7519 skb->protocol != htons(ETH_P_IPV6))
7522 switch (tstamp_type) {
7523 case BPF_SKB_TSTAMP_DELIVERY_MONO:
7526 skb->tstamp = tstamp;
7527 skb->mono_delivery_time = 1;
7529 case BPF_SKB_TSTAMP_UNSPEC:
7533 skb->mono_delivery_time = 0;
7542 static const struct bpf_func_proto bpf_skb_set_tstamp_proto = {
7543 .func = bpf_skb_set_tstamp,
7545 .ret_type = RET_INTEGER,
7546 .arg1_type = ARG_PTR_TO_CTX,
7547 .arg2_type = ARG_ANYTHING,
7548 .arg3_type = ARG_ANYTHING,
7551 #ifdef CONFIG_SYN_COOKIES
7552 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv4, struct iphdr *, iph,
7553 struct tcphdr *, th, u32, th_len)
7558 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7561 mss = tcp_parse_mss_option(th, 0) ?: TCP_MSS_DEFAULT;
7562 cookie = __cookie_v4_init_sequence(iph, th, &mss);
7564 return cookie | ((u64)mss << 32);
7567 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv4_proto = {
7568 .func = bpf_tcp_raw_gen_syncookie_ipv4,
7569 .gpl_only = true, /* __cookie_v4_init_sequence() is GPL */
7571 .ret_type = RET_INTEGER,
7572 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7573 .arg1_size = sizeof(struct iphdr),
7574 .arg2_type = ARG_PTR_TO_MEM,
7575 .arg3_type = ARG_CONST_SIZE,
7578 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv6, struct ipv6hdr *, iph,
7579 struct tcphdr *, th, u32, th_len)
7581 #if IS_BUILTIN(CONFIG_IPV6)
7582 const u16 mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
7583 sizeof(struct ipv6hdr);
7587 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7590 mss = tcp_parse_mss_option(th, 0) ?: mss_clamp;
7591 cookie = __cookie_v6_init_sequence(iph, th, &mss);
7593 return cookie | ((u64)mss << 32);
7595 return -EPROTONOSUPPORT;
7599 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv6_proto = {
7600 .func = bpf_tcp_raw_gen_syncookie_ipv6,
7601 .gpl_only = true, /* __cookie_v6_init_sequence() is GPL */
7603 .ret_type = RET_INTEGER,
7604 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7605 .arg1_size = sizeof(struct ipv6hdr),
7606 .arg2_type = ARG_PTR_TO_MEM,
7607 .arg3_type = ARG_CONST_SIZE,
7610 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv4, struct iphdr *, iph,
7611 struct tcphdr *, th)
7613 u32 cookie = ntohl(th->ack_seq) - 1;
7615 if (__cookie_v4_check(iph, th, cookie) > 0)
7621 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv4_proto = {
7622 .func = bpf_tcp_raw_check_syncookie_ipv4,
7623 .gpl_only = true, /* __cookie_v4_check is GPL */
7625 .ret_type = RET_INTEGER,
7626 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7627 .arg1_size = sizeof(struct iphdr),
7628 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7629 .arg2_size = sizeof(struct tcphdr),
7632 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv6, struct ipv6hdr *, iph,
7633 struct tcphdr *, th)
7635 #if IS_BUILTIN(CONFIG_IPV6)
7636 u32 cookie = ntohl(th->ack_seq) - 1;
7638 if (__cookie_v6_check(iph, th, cookie) > 0)
7643 return -EPROTONOSUPPORT;
7647 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv6_proto = {
7648 .func = bpf_tcp_raw_check_syncookie_ipv6,
7649 .gpl_only = true, /* __cookie_v6_check is GPL */
7651 .ret_type = RET_INTEGER,
7652 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7653 .arg1_size = sizeof(struct ipv6hdr),
7654 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7655 .arg2_size = sizeof(struct tcphdr),
7657 #endif /* CONFIG_SYN_COOKIES */
7659 #endif /* CONFIG_INET */
7661 bool bpf_helper_changes_pkt_data(void *func)
7663 if (func == bpf_skb_vlan_push ||
7664 func == bpf_skb_vlan_pop ||
7665 func == bpf_skb_store_bytes ||
7666 func == bpf_skb_change_proto ||
7667 func == bpf_skb_change_head ||
7668 func == sk_skb_change_head ||
7669 func == bpf_skb_change_tail ||
7670 func == sk_skb_change_tail ||
7671 func == bpf_skb_adjust_room ||
7672 func == sk_skb_adjust_room ||
7673 func == bpf_skb_pull_data ||
7674 func == sk_skb_pull_data ||
7675 func == bpf_clone_redirect ||
7676 func == bpf_l3_csum_replace ||
7677 func == bpf_l4_csum_replace ||
7678 func == bpf_xdp_adjust_head ||
7679 func == bpf_xdp_adjust_meta ||
7680 func == bpf_msg_pull_data ||
7681 func == bpf_msg_push_data ||
7682 func == bpf_msg_pop_data ||
7683 func == bpf_xdp_adjust_tail ||
7684 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7685 func == bpf_lwt_seg6_store_bytes ||
7686 func == bpf_lwt_seg6_adjust_srh ||
7687 func == bpf_lwt_seg6_action ||
7690 func == bpf_sock_ops_store_hdr_opt ||
7692 func == bpf_lwt_in_push_encap ||
7693 func == bpf_lwt_xmit_push_encap)
7699 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7700 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7702 static const struct bpf_func_proto *
7703 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7705 const struct bpf_func_proto *func_proto;
7707 func_proto = cgroup_common_func_proto(func_id, prog);
7711 func_proto = cgroup_current_func_proto(func_id, prog);
7716 case BPF_FUNC_get_socket_cookie:
7717 return &bpf_get_socket_cookie_sock_proto;
7718 case BPF_FUNC_get_netns_cookie:
7719 return &bpf_get_netns_cookie_sock_proto;
7720 case BPF_FUNC_perf_event_output:
7721 return &bpf_event_output_data_proto;
7722 case BPF_FUNC_sk_storage_get:
7723 return &bpf_sk_storage_get_cg_sock_proto;
7724 case BPF_FUNC_ktime_get_coarse_ns:
7725 return &bpf_ktime_get_coarse_ns_proto;
7727 return bpf_base_func_proto(func_id);
7731 static const struct bpf_func_proto *
7732 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7734 const struct bpf_func_proto *func_proto;
7736 func_proto = cgroup_common_func_proto(func_id, prog);
7740 func_proto = cgroup_current_func_proto(func_id, prog);
7746 switch (prog->expected_attach_type) {
7747 case BPF_CGROUP_INET4_CONNECT:
7748 case BPF_CGROUP_INET6_CONNECT:
7749 return &bpf_bind_proto;
7753 case BPF_FUNC_get_socket_cookie:
7754 return &bpf_get_socket_cookie_sock_addr_proto;
7755 case BPF_FUNC_get_netns_cookie:
7756 return &bpf_get_netns_cookie_sock_addr_proto;
7757 case BPF_FUNC_perf_event_output:
7758 return &bpf_event_output_data_proto;
7760 case BPF_FUNC_sk_lookup_tcp:
7761 return &bpf_sock_addr_sk_lookup_tcp_proto;
7762 case BPF_FUNC_sk_lookup_udp:
7763 return &bpf_sock_addr_sk_lookup_udp_proto;
7764 case BPF_FUNC_sk_release:
7765 return &bpf_sk_release_proto;
7766 case BPF_FUNC_skc_lookup_tcp:
7767 return &bpf_sock_addr_skc_lookup_tcp_proto;
7768 #endif /* CONFIG_INET */
7769 case BPF_FUNC_sk_storage_get:
7770 return &bpf_sk_storage_get_proto;
7771 case BPF_FUNC_sk_storage_delete:
7772 return &bpf_sk_storage_delete_proto;
7773 case BPF_FUNC_setsockopt:
7774 switch (prog->expected_attach_type) {
7775 case BPF_CGROUP_INET4_BIND:
7776 case BPF_CGROUP_INET6_BIND:
7777 case BPF_CGROUP_INET4_CONNECT:
7778 case BPF_CGROUP_INET6_CONNECT:
7779 case BPF_CGROUP_UDP4_RECVMSG:
7780 case BPF_CGROUP_UDP6_RECVMSG:
7781 case BPF_CGROUP_UDP4_SENDMSG:
7782 case BPF_CGROUP_UDP6_SENDMSG:
7783 case BPF_CGROUP_INET4_GETPEERNAME:
7784 case BPF_CGROUP_INET6_GETPEERNAME:
7785 case BPF_CGROUP_INET4_GETSOCKNAME:
7786 case BPF_CGROUP_INET6_GETSOCKNAME:
7787 return &bpf_sock_addr_setsockopt_proto;
7791 case BPF_FUNC_getsockopt:
7792 switch (prog->expected_attach_type) {
7793 case BPF_CGROUP_INET4_BIND:
7794 case BPF_CGROUP_INET6_BIND:
7795 case BPF_CGROUP_INET4_CONNECT:
7796 case BPF_CGROUP_INET6_CONNECT:
7797 case BPF_CGROUP_UDP4_RECVMSG:
7798 case BPF_CGROUP_UDP6_RECVMSG:
7799 case BPF_CGROUP_UDP4_SENDMSG:
7800 case BPF_CGROUP_UDP6_SENDMSG:
7801 case BPF_CGROUP_INET4_GETPEERNAME:
7802 case BPF_CGROUP_INET6_GETPEERNAME:
7803 case BPF_CGROUP_INET4_GETSOCKNAME:
7804 case BPF_CGROUP_INET6_GETSOCKNAME:
7805 return &bpf_sock_addr_getsockopt_proto;
7810 return bpf_sk_base_func_proto(func_id);
7814 static const struct bpf_func_proto *
7815 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7818 case BPF_FUNC_skb_load_bytes:
7819 return &bpf_skb_load_bytes_proto;
7820 case BPF_FUNC_skb_load_bytes_relative:
7821 return &bpf_skb_load_bytes_relative_proto;
7822 case BPF_FUNC_get_socket_cookie:
7823 return &bpf_get_socket_cookie_proto;
7824 case BPF_FUNC_get_socket_uid:
7825 return &bpf_get_socket_uid_proto;
7826 case BPF_FUNC_perf_event_output:
7827 return &bpf_skb_event_output_proto;
7829 return bpf_sk_base_func_proto(func_id);
7833 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7834 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7836 static const struct bpf_func_proto *
7837 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7839 const struct bpf_func_proto *func_proto;
7841 func_proto = cgroup_common_func_proto(func_id, prog);
7846 case BPF_FUNC_sk_fullsock:
7847 return &bpf_sk_fullsock_proto;
7848 case BPF_FUNC_sk_storage_get:
7849 return &bpf_sk_storage_get_proto;
7850 case BPF_FUNC_sk_storage_delete:
7851 return &bpf_sk_storage_delete_proto;
7852 case BPF_FUNC_perf_event_output:
7853 return &bpf_skb_event_output_proto;
7854 #ifdef CONFIG_SOCK_CGROUP_DATA
7855 case BPF_FUNC_skb_cgroup_id:
7856 return &bpf_skb_cgroup_id_proto;
7857 case BPF_FUNC_skb_ancestor_cgroup_id:
7858 return &bpf_skb_ancestor_cgroup_id_proto;
7859 case BPF_FUNC_sk_cgroup_id:
7860 return &bpf_sk_cgroup_id_proto;
7861 case BPF_FUNC_sk_ancestor_cgroup_id:
7862 return &bpf_sk_ancestor_cgroup_id_proto;
7865 case BPF_FUNC_sk_lookup_tcp:
7866 return &bpf_sk_lookup_tcp_proto;
7867 case BPF_FUNC_sk_lookup_udp:
7868 return &bpf_sk_lookup_udp_proto;
7869 case BPF_FUNC_sk_release:
7870 return &bpf_sk_release_proto;
7871 case BPF_FUNC_skc_lookup_tcp:
7872 return &bpf_skc_lookup_tcp_proto;
7873 case BPF_FUNC_tcp_sock:
7874 return &bpf_tcp_sock_proto;
7875 case BPF_FUNC_get_listener_sock:
7876 return &bpf_get_listener_sock_proto;
7877 case BPF_FUNC_skb_ecn_set_ce:
7878 return &bpf_skb_ecn_set_ce_proto;
7881 return sk_filter_func_proto(func_id, prog);
7885 static const struct bpf_func_proto *
7886 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7889 case BPF_FUNC_skb_store_bytes:
7890 return &bpf_skb_store_bytes_proto;
7891 case BPF_FUNC_skb_load_bytes:
7892 return &bpf_skb_load_bytes_proto;
7893 case BPF_FUNC_skb_load_bytes_relative:
7894 return &bpf_skb_load_bytes_relative_proto;
7895 case BPF_FUNC_skb_pull_data:
7896 return &bpf_skb_pull_data_proto;
7897 case BPF_FUNC_csum_diff:
7898 return &bpf_csum_diff_proto;
7899 case BPF_FUNC_csum_update:
7900 return &bpf_csum_update_proto;
7901 case BPF_FUNC_csum_level:
7902 return &bpf_csum_level_proto;
7903 case BPF_FUNC_l3_csum_replace:
7904 return &bpf_l3_csum_replace_proto;
7905 case BPF_FUNC_l4_csum_replace:
7906 return &bpf_l4_csum_replace_proto;
7907 case BPF_FUNC_clone_redirect:
7908 return &bpf_clone_redirect_proto;
7909 case BPF_FUNC_get_cgroup_classid:
7910 return &bpf_get_cgroup_classid_proto;
7911 case BPF_FUNC_skb_vlan_push:
7912 return &bpf_skb_vlan_push_proto;
7913 case BPF_FUNC_skb_vlan_pop:
7914 return &bpf_skb_vlan_pop_proto;
7915 case BPF_FUNC_skb_change_proto:
7916 return &bpf_skb_change_proto_proto;
7917 case BPF_FUNC_skb_change_type:
7918 return &bpf_skb_change_type_proto;
7919 case BPF_FUNC_skb_adjust_room:
7920 return &bpf_skb_adjust_room_proto;
7921 case BPF_FUNC_skb_change_tail:
7922 return &bpf_skb_change_tail_proto;
7923 case BPF_FUNC_skb_change_head:
7924 return &bpf_skb_change_head_proto;
7925 case BPF_FUNC_skb_get_tunnel_key:
7926 return &bpf_skb_get_tunnel_key_proto;
7927 case BPF_FUNC_skb_set_tunnel_key:
7928 return bpf_get_skb_set_tunnel_proto(func_id);
7929 case BPF_FUNC_skb_get_tunnel_opt:
7930 return &bpf_skb_get_tunnel_opt_proto;
7931 case BPF_FUNC_skb_set_tunnel_opt:
7932 return bpf_get_skb_set_tunnel_proto(func_id);
7933 case BPF_FUNC_redirect:
7934 return &bpf_redirect_proto;
7935 case BPF_FUNC_redirect_neigh:
7936 return &bpf_redirect_neigh_proto;
7937 case BPF_FUNC_redirect_peer:
7938 return &bpf_redirect_peer_proto;
7939 case BPF_FUNC_get_route_realm:
7940 return &bpf_get_route_realm_proto;
7941 case BPF_FUNC_get_hash_recalc:
7942 return &bpf_get_hash_recalc_proto;
7943 case BPF_FUNC_set_hash_invalid:
7944 return &bpf_set_hash_invalid_proto;
7945 case BPF_FUNC_set_hash:
7946 return &bpf_set_hash_proto;
7947 case BPF_FUNC_perf_event_output:
7948 return &bpf_skb_event_output_proto;
7949 case BPF_FUNC_get_smp_processor_id:
7950 return &bpf_get_smp_processor_id_proto;
7951 case BPF_FUNC_skb_under_cgroup:
7952 return &bpf_skb_under_cgroup_proto;
7953 case BPF_FUNC_get_socket_cookie:
7954 return &bpf_get_socket_cookie_proto;
7955 case BPF_FUNC_get_socket_uid:
7956 return &bpf_get_socket_uid_proto;
7957 case BPF_FUNC_fib_lookup:
7958 return &bpf_skb_fib_lookup_proto;
7959 case BPF_FUNC_check_mtu:
7960 return &bpf_skb_check_mtu_proto;
7961 case BPF_FUNC_sk_fullsock:
7962 return &bpf_sk_fullsock_proto;
7963 case BPF_FUNC_sk_storage_get:
7964 return &bpf_sk_storage_get_proto;
7965 case BPF_FUNC_sk_storage_delete:
7966 return &bpf_sk_storage_delete_proto;
7968 case BPF_FUNC_skb_get_xfrm_state:
7969 return &bpf_skb_get_xfrm_state_proto;
7971 #ifdef CONFIG_CGROUP_NET_CLASSID
7972 case BPF_FUNC_skb_cgroup_classid:
7973 return &bpf_skb_cgroup_classid_proto;
7975 #ifdef CONFIG_SOCK_CGROUP_DATA
7976 case BPF_FUNC_skb_cgroup_id:
7977 return &bpf_skb_cgroup_id_proto;
7978 case BPF_FUNC_skb_ancestor_cgroup_id:
7979 return &bpf_skb_ancestor_cgroup_id_proto;
7982 case BPF_FUNC_sk_lookup_tcp:
7983 return &bpf_tc_sk_lookup_tcp_proto;
7984 case BPF_FUNC_sk_lookup_udp:
7985 return &bpf_tc_sk_lookup_udp_proto;
7986 case BPF_FUNC_sk_release:
7987 return &bpf_sk_release_proto;
7988 case BPF_FUNC_tcp_sock:
7989 return &bpf_tcp_sock_proto;
7990 case BPF_FUNC_get_listener_sock:
7991 return &bpf_get_listener_sock_proto;
7992 case BPF_FUNC_skc_lookup_tcp:
7993 return &bpf_tc_skc_lookup_tcp_proto;
7994 case BPF_FUNC_tcp_check_syncookie:
7995 return &bpf_tcp_check_syncookie_proto;
7996 case BPF_FUNC_skb_ecn_set_ce:
7997 return &bpf_skb_ecn_set_ce_proto;
7998 case BPF_FUNC_tcp_gen_syncookie:
7999 return &bpf_tcp_gen_syncookie_proto;
8000 case BPF_FUNC_sk_assign:
8001 return &bpf_sk_assign_proto;
8002 case BPF_FUNC_skb_set_tstamp:
8003 return &bpf_skb_set_tstamp_proto;
8004 #ifdef CONFIG_SYN_COOKIES
8005 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8006 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8007 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8008 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8009 case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8010 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8011 case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8012 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8016 return bpf_sk_base_func_proto(func_id);
8020 static const struct bpf_func_proto *
8021 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8024 case BPF_FUNC_perf_event_output:
8025 return &bpf_xdp_event_output_proto;
8026 case BPF_FUNC_get_smp_processor_id:
8027 return &bpf_get_smp_processor_id_proto;
8028 case BPF_FUNC_csum_diff:
8029 return &bpf_csum_diff_proto;
8030 case BPF_FUNC_xdp_adjust_head:
8031 return &bpf_xdp_adjust_head_proto;
8032 case BPF_FUNC_xdp_adjust_meta:
8033 return &bpf_xdp_adjust_meta_proto;
8034 case BPF_FUNC_redirect:
8035 return &bpf_xdp_redirect_proto;
8036 case BPF_FUNC_redirect_map:
8037 return &bpf_xdp_redirect_map_proto;
8038 case BPF_FUNC_xdp_adjust_tail:
8039 return &bpf_xdp_adjust_tail_proto;
8040 case BPF_FUNC_xdp_get_buff_len:
8041 return &bpf_xdp_get_buff_len_proto;
8042 case BPF_FUNC_xdp_load_bytes:
8043 return &bpf_xdp_load_bytes_proto;
8044 case BPF_FUNC_xdp_store_bytes:
8045 return &bpf_xdp_store_bytes_proto;
8046 case BPF_FUNC_fib_lookup:
8047 return &bpf_xdp_fib_lookup_proto;
8048 case BPF_FUNC_check_mtu:
8049 return &bpf_xdp_check_mtu_proto;
8051 case BPF_FUNC_sk_lookup_udp:
8052 return &bpf_xdp_sk_lookup_udp_proto;
8053 case BPF_FUNC_sk_lookup_tcp:
8054 return &bpf_xdp_sk_lookup_tcp_proto;
8055 case BPF_FUNC_sk_release:
8056 return &bpf_sk_release_proto;
8057 case BPF_FUNC_skc_lookup_tcp:
8058 return &bpf_xdp_skc_lookup_tcp_proto;
8059 case BPF_FUNC_tcp_check_syncookie:
8060 return &bpf_tcp_check_syncookie_proto;
8061 case BPF_FUNC_tcp_gen_syncookie:
8062 return &bpf_tcp_gen_syncookie_proto;
8063 #ifdef CONFIG_SYN_COOKIES
8064 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8065 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8066 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8067 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8068 case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8069 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8070 case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8071 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8075 return bpf_sk_base_func_proto(func_id);
8078 #if IS_MODULE(CONFIG_NF_CONNTRACK) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES)
8079 /* The nf_conn___init type is used in the NF_CONNTRACK kfuncs. The
8080 * kfuncs are defined in two different modules, and we want to be able
8081 * to use them interchangably with the same BTF type ID. Because modules
8082 * can't de-duplicate BTF IDs between each other, we need the type to be
8083 * referenced in the vmlinux BTF or the verifier will get confused about
8084 * the different types. So we add this dummy type reference which will
8085 * be included in vmlinux BTF, allowing both modules to refer to the
8088 BTF_TYPE_EMIT(struct nf_conn___init);
8092 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
8093 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
8095 static const struct bpf_func_proto *
8096 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8098 const struct bpf_func_proto *func_proto;
8100 func_proto = cgroup_common_func_proto(func_id, prog);
8105 case BPF_FUNC_setsockopt:
8106 return &bpf_sock_ops_setsockopt_proto;
8107 case BPF_FUNC_getsockopt:
8108 return &bpf_sock_ops_getsockopt_proto;
8109 case BPF_FUNC_sock_ops_cb_flags_set:
8110 return &bpf_sock_ops_cb_flags_set_proto;
8111 case BPF_FUNC_sock_map_update:
8112 return &bpf_sock_map_update_proto;
8113 case BPF_FUNC_sock_hash_update:
8114 return &bpf_sock_hash_update_proto;
8115 case BPF_FUNC_get_socket_cookie:
8116 return &bpf_get_socket_cookie_sock_ops_proto;
8117 case BPF_FUNC_perf_event_output:
8118 return &bpf_event_output_data_proto;
8119 case BPF_FUNC_sk_storage_get:
8120 return &bpf_sk_storage_get_proto;
8121 case BPF_FUNC_sk_storage_delete:
8122 return &bpf_sk_storage_delete_proto;
8123 case BPF_FUNC_get_netns_cookie:
8124 return &bpf_get_netns_cookie_sock_ops_proto;
8126 case BPF_FUNC_load_hdr_opt:
8127 return &bpf_sock_ops_load_hdr_opt_proto;
8128 case BPF_FUNC_store_hdr_opt:
8129 return &bpf_sock_ops_store_hdr_opt_proto;
8130 case BPF_FUNC_reserve_hdr_opt:
8131 return &bpf_sock_ops_reserve_hdr_opt_proto;
8132 case BPF_FUNC_tcp_sock:
8133 return &bpf_tcp_sock_proto;
8134 #endif /* CONFIG_INET */
8136 return bpf_sk_base_func_proto(func_id);
8140 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
8141 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
8143 static const struct bpf_func_proto *
8144 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8147 case BPF_FUNC_msg_redirect_map:
8148 return &bpf_msg_redirect_map_proto;
8149 case BPF_FUNC_msg_redirect_hash:
8150 return &bpf_msg_redirect_hash_proto;
8151 case BPF_FUNC_msg_apply_bytes:
8152 return &bpf_msg_apply_bytes_proto;
8153 case BPF_FUNC_msg_cork_bytes:
8154 return &bpf_msg_cork_bytes_proto;
8155 case BPF_FUNC_msg_pull_data:
8156 return &bpf_msg_pull_data_proto;
8157 case BPF_FUNC_msg_push_data:
8158 return &bpf_msg_push_data_proto;
8159 case BPF_FUNC_msg_pop_data:
8160 return &bpf_msg_pop_data_proto;
8161 case BPF_FUNC_perf_event_output:
8162 return &bpf_event_output_data_proto;
8163 case BPF_FUNC_get_current_uid_gid:
8164 return &bpf_get_current_uid_gid_proto;
8165 case BPF_FUNC_get_current_pid_tgid:
8166 return &bpf_get_current_pid_tgid_proto;
8167 case BPF_FUNC_sk_storage_get:
8168 return &bpf_sk_storage_get_proto;
8169 case BPF_FUNC_sk_storage_delete:
8170 return &bpf_sk_storage_delete_proto;
8171 case BPF_FUNC_get_netns_cookie:
8172 return &bpf_get_netns_cookie_sk_msg_proto;
8173 #ifdef CONFIG_CGROUPS
8174 case BPF_FUNC_get_current_cgroup_id:
8175 return &bpf_get_current_cgroup_id_proto;
8176 case BPF_FUNC_get_current_ancestor_cgroup_id:
8177 return &bpf_get_current_ancestor_cgroup_id_proto;
8179 #ifdef CONFIG_CGROUP_NET_CLASSID
8180 case BPF_FUNC_get_cgroup_classid:
8181 return &bpf_get_cgroup_classid_curr_proto;
8184 return bpf_sk_base_func_proto(func_id);
8188 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
8189 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
8191 static const struct bpf_func_proto *
8192 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8195 case BPF_FUNC_skb_store_bytes:
8196 return &bpf_skb_store_bytes_proto;
8197 case BPF_FUNC_skb_load_bytes:
8198 return &bpf_skb_load_bytes_proto;
8199 case BPF_FUNC_skb_pull_data:
8200 return &sk_skb_pull_data_proto;
8201 case BPF_FUNC_skb_change_tail:
8202 return &sk_skb_change_tail_proto;
8203 case BPF_FUNC_skb_change_head:
8204 return &sk_skb_change_head_proto;
8205 case BPF_FUNC_skb_adjust_room:
8206 return &sk_skb_adjust_room_proto;
8207 case BPF_FUNC_get_socket_cookie:
8208 return &bpf_get_socket_cookie_proto;
8209 case BPF_FUNC_get_socket_uid:
8210 return &bpf_get_socket_uid_proto;
8211 case BPF_FUNC_sk_redirect_map:
8212 return &bpf_sk_redirect_map_proto;
8213 case BPF_FUNC_sk_redirect_hash:
8214 return &bpf_sk_redirect_hash_proto;
8215 case BPF_FUNC_perf_event_output:
8216 return &bpf_skb_event_output_proto;
8218 case BPF_FUNC_sk_lookup_tcp:
8219 return &bpf_sk_lookup_tcp_proto;
8220 case BPF_FUNC_sk_lookup_udp:
8221 return &bpf_sk_lookup_udp_proto;
8222 case BPF_FUNC_sk_release:
8223 return &bpf_sk_release_proto;
8224 case BPF_FUNC_skc_lookup_tcp:
8225 return &bpf_skc_lookup_tcp_proto;
8228 return bpf_sk_base_func_proto(func_id);
8232 static const struct bpf_func_proto *
8233 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8236 case BPF_FUNC_skb_load_bytes:
8237 return &bpf_flow_dissector_load_bytes_proto;
8239 return bpf_sk_base_func_proto(func_id);
8243 static const struct bpf_func_proto *
8244 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8247 case BPF_FUNC_skb_load_bytes:
8248 return &bpf_skb_load_bytes_proto;
8249 case BPF_FUNC_skb_pull_data:
8250 return &bpf_skb_pull_data_proto;
8251 case BPF_FUNC_csum_diff:
8252 return &bpf_csum_diff_proto;
8253 case BPF_FUNC_get_cgroup_classid:
8254 return &bpf_get_cgroup_classid_proto;
8255 case BPF_FUNC_get_route_realm:
8256 return &bpf_get_route_realm_proto;
8257 case BPF_FUNC_get_hash_recalc:
8258 return &bpf_get_hash_recalc_proto;
8259 case BPF_FUNC_perf_event_output:
8260 return &bpf_skb_event_output_proto;
8261 case BPF_FUNC_get_smp_processor_id:
8262 return &bpf_get_smp_processor_id_proto;
8263 case BPF_FUNC_skb_under_cgroup:
8264 return &bpf_skb_under_cgroup_proto;
8266 return bpf_sk_base_func_proto(func_id);
8270 static const struct bpf_func_proto *
8271 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8274 case BPF_FUNC_lwt_push_encap:
8275 return &bpf_lwt_in_push_encap_proto;
8277 return lwt_out_func_proto(func_id, prog);
8281 static const struct bpf_func_proto *
8282 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8285 case BPF_FUNC_skb_get_tunnel_key:
8286 return &bpf_skb_get_tunnel_key_proto;
8287 case BPF_FUNC_skb_set_tunnel_key:
8288 return bpf_get_skb_set_tunnel_proto(func_id);
8289 case BPF_FUNC_skb_get_tunnel_opt:
8290 return &bpf_skb_get_tunnel_opt_proto;
8291 case BPF_FUNC_skb_set_tunnel_opt:
8292 return bpf_get_skb_set_tunnel_proto(func_id);
8293 case BPF_FUNC_redirect:
8294 return &bpf_redirect_proto;
8295 case BPF_FUNC_clone_redirect:
8296 return &bpf_clone_redirect_proto;
8297 case BPF_FUNC_skb_change_tail:
8298 return &bpf_skb_change_tail_proto;
8299 case BPF_FUNC_skb_change_head:
8300 return &bpf_skb_change_head_proto;
8301 case BPF_FUNC_skb_store_bytes:
8302 return &bpf_skb_store_bytes_proto;
8303 case BPF_FUNC_csum_update:
8304 return &bpf_csum_update_proto;
8305 case BPF_FUNC_csum_level:
8306 return &bpf_csum_level_proto;
8307 case BPF_FUNC_l3_csum_replace:
8308 return &bpf_l3_csum_replace_proto;
8309 case BPF_FUNC_l4_csum_replace:
8310 return &bpf_l4_csum_replace_proto;
8311 case BPF_FUNC_set_hash_invalid:
8312 return &bpf_set_hash_invalid_proto;
8313 case BPF_FUNC_lwt_push_encap:
8314 return &bpf_lwt_xmit_push_encap_proto;
8316 return lwt_out_func_proto(func_id, prog);
8320 static const struct bpf_func_proto *
8321 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8324 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
8325 case BPF_FUNC_lwt_seg6_store_bytes:
8326 return &bpf_lwt_seg6_store_bytes_proto;
8327 case BPF_FUNC_lwt_seg6_action:
8328 return &bpf_lwt_seg6_action_proto;
8329 case BPF_FUNC_lwt_seg6_adjust_srh:
8330 return &bpf_lwt_seg6_adjust_srh_proto;
8333 return lwt_out_func_proto(func_id, prog);
8337 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
8338 const struct bpf_prog *prog,
8339 struct bpf_insn_access_aux *info)
8341 const int size_default = sizeof(__u32);
8343 if (off < 0 || off >= sizeof(struct __sk_buff))
8346 /* The verifier guarantees that size > 0. */
8347 if (off % size != 0)
8351 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8352 if (off + size > offsetofend(struct __sk_buff, cb[4]))
8355 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
8356 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
8357 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
8358 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
8359 case bpf_ctx_range(struct __sk_buff, data):
8360 case bpf_ctx_range(struct __sk_buff, data_meta):
8361 case bpf_ctx_range(struct __sk_buff, data_end):
8362 if (size != size_default)
8365 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8367 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8368 if (type == BPF_WRITE || size != sizeof(__u64))
8371 case bpf_ctx_range(struct __sk_buff, tstamp):
8372 if (size != sizeof(__u64))
8375 case offsetof(struct __sk_buff, sk):
8376 if (type == BPF_WRITE || size != sizeof(__u64))
8378 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
8380 case offsetof(struct __sk_buff, tstamp_type):
8382 case offsetofend(struct __sk_buff, tstamp_type) ... offsetof(struct __sk_buff, hwtstamp) - 1:
8383 /* Explicitly prohibit access to padding in __sk_buff. */
8386 /* Only narrow read access allowed for now. */
8387 if (type == BPF_WRITE) {
8388 if (size != size_default)
8391 bpf_ctx_record_field_size(info, size_default);
8392 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8400 static bool sk_filter_is_valid_access(int off, int size,
8401 enum bpf_access_type type,
8402 const struct bpf_prog *prog,
8403 struct bpf_insn_access_aux *info)
8406 case bpf_ctx_range(struct __sk_buff, tc_classid):
8407 case bpf_ctx_range(struct __sk_buff, data):
8408 case bpf_ctx_range(struct __sk_buff, data_meta):
8409 case bpf_ctx_range(struct __sk_buff, data_end):
8410 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8411 case bpf_ctx_range(struct __sk_buff, tstamp):
8412 case bpf_ctx_range(struct __sk_buff, wire_len):
8413 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8417 if (type == BPF_WRITE) {
8419 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8426 return bpf_skb_is_valid_access(off, size, type, prog, info);
8429 static bool cg_skb_is_valid_access(int off, int size,
8430 enum bpf_access_type type,
8431 const struct bpf_prog *prog,
8432 struct bpf_insn_access_aux *info)
8435 case bpf_ctx_range(struct __sk_buff, tc_classid):
8436 case bpf_ctx_range(struct __sk_buff, data_meta):
8437 case bpf_ctx_range(struct __sk_buff, wire_len):
8439 case bpf_ctx_range(struct __sk_buff, data):
8440 case bpf_ctx_range(struct __sk_buff, data_end):
8446 if (type == BPF_WRITE) {
8448 case bpf_ctx_range(struct __sk_buff, mark):
8449 case bpf_ctx_range(struct __sk_buff, priority):
8450 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8452 case bpf_ctx_range(struct __sk_buff, tstamp):
8462 case bpf_ctx_range(struct __sk_buff, data):
8463 info->reg_type = PTR_TO_PACKET;
8465 case bpf_ctx_range(struct __sk_buff, data_end):
8466 info->reg_type = PTR_TO_PACKET_END;
8470 return bpf_skb_is_valid_access(off, size, type, prog, info);
8473 static bool lwt_is_valid_access(int off, int size,
8474 enum bpf_access_type type,
8475 const struct bpf_prog *prog,
8476 struct bpf_insn_access_aux *info)
8479 case bpf_ctx_range(struct __sk_buff, tc_classid):
8480 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8481 case bpf_ctx_range(struct __sk_buff, data_meta):
8482 case bpf_ctx_range(struct __sk_buff, tstamp):
8483 case bpf_ctx_range(struct __sk_buff, wire_len):
8484 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8488 if (type == BPF_WRITE) {
8490 case bpf_ctx_range(struct __sk_buff, mark):
8491 case bpf_ctx_range(struct __sk_buff, priority):
8492 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8500 case bpf_ctx_range(struct __sk_buff, data):
8501 info->reg_type = PTR_TO_PACKET;
8503 case bpf_ctx_range(struct __sk_buff, data_end):
8504 info->reg_type = PTR_TO_PACKET_END;
8508 return bpf_skb_is_valid_access(off, size, type, prog, info);
8511 /* Attach type specific accesses */
8512 static bool __sock_filter_check_attach_type(int off,
8513 enum bpf_access_type access_type,
8514 enum bpf_attach_type attach_type)
8517 case offsetof(struct bpf_sock, bound_dev_if):
8518 case offsetof(struct bpf_sock, mark):
8519 case offsetof(struct bpf_sock, priority):
8520 switch (attach_type) {
8521 case BPF_CGROUP_INET_SOCK_CREATE:
8522 case BPF_CGROUP_INET_SOCK_RELEASE:
8527 case bpf_ctx_range(struct bpf_sock, src_ip4):
8528 switch (attach_type) {
8529 case BPF_CGROUP_INET4_POST_BIND:
8534 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8535 switch (attach_type) {
8536 case BPF_CGROUP_INET6_POST_BIND:
8541 case bpf_ctx_range(struct bpf_sock, src_port):
8542 switch (attach_type) {
8543 case BPF_CGROUP_INET4_POST_BIND:
8544 case BPF_CGROUP_INET6_POST_BIND:
8551 return access_type == BPF_READ;
8556 bool bpf_sock_common_is_valid_access(int off, int size,
8557 enum bpf_access_type type,
8558 struct bpf_insn_access_aux *info)
8561 case bpf_ctx_range_till(struct bpf_sock, type, priority):
8564 return bpf_sock_is_valid_access(off, size, type, info);
8568 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
8569 struct bpf_insn_access_aux *info)
8571 const int size_default = sizeof(__u32);
8574 if (off < 0 || off >= sizeof(struct bpf_sock))
8576 if (off % size != 0)
8580 case offsetof(struct bpf_sock, state):
8581 case offsetof(struct bpf_sock, family):
8582 case offsetof(struct bpf_sock, type):
8583 case offsetof(struct bpf_sock, protocol):
8584 case offsetof(struct bpf_sock, src_port):
8585 case offsetof(struct bpf_sock, rx_queue_mapping):
8586 case bpf_ctx_range(struct bpf_sock, src_ip4):
8587 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8588 case bpf_ctx_range(struct bpf_sock, dst_ip4):
8589 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8590 bpf_ctx_record_field_size(info, size_default);
8591 return bpf_ctx_narrow_access_ok(off, size, size_default);
8592 case bpf_ctx_range(struct bpf_sock, dst_port):
8593 field_size = size == size_default ?
8594 size_default : sizeof_field(struct bpf_sock, dst_port);
8595 bpf_ctx_record_field_size(info, field_size);
8596 return bpf_ctx_narrow_access_ok(off, size, field_size);
8597 case offsetofend(struct bpf_sock, dst_port) ...
8598 offsetof(struct bpf_sock, dst_ip4) - 1:
8602 return size == size_default;
8605 static bool sock_filter_is_valid_access(int off, int size,
8606 enum bpf_access_type type,
8607 const struct bpf_prog *prog,
8608 struct bpf_insn_access_aux *info)
8610 if (!bpf_sock_is_valid_access(off, size, type, info))
8612 return __sock_filter_check_attach_type(off, type,
8613 prog->expected_attach_type);
8616 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
8617 const struct bpf_prog *prog)
8619 /* Neither direct read nor direct write requires any preliminary
8625 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
8626 const struct bpf_prog *prog, int drop_verdict)
8628 struct bpf_insn *insn = insn_buf;
8633 /* if (!skb->cloned)
8636 * (Fast-path, otherwise approximation that we might be
8637 * a clone, do the rest in helper.)
8639 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET);
8640 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
8641 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
8643 /* ret = bpf_skb_pull_data(skb, 0); */
8644 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
8645 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
8646 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
8647 BPF_FUNC_skb_pull_data);
8650 * return TC_ACT_SHOT;
8652 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
8653 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
8654 *insn++ = BPF_EXIT_INSN();
8657 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
8659 *insn++ = prog->insnsi[0];
8661 return insn - insn_buf;
8664 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
8665 struct bpf_insn *insn_buf)
8667 bool indirect = BPF_MODE(orig->code) == BPF_IND;
8668 struct bpf_insn *insn = insn_buf;
8671 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
8673 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
8675 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
8677 /* We're guaranteed here that CTX is in R6. */
8678 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
8680 switch (BPF_SIZE(orig->code)) {
8682 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
8685 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
8688 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
8692 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
8693 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
8694 *insn++ = BPF_EXIT_INSN();
8696 return insn - insn_buf;
8699 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8700 const struct bpf_prog *prog)
8702 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8705 static bool tc_cls_act_is_valid_access(int off, int size,
8706 enum bpf_access_type type,
8707 const struct bpf_prog *prog,
8708 struct bpf_insn_access_aux *info)
8710 if (type == BPF_WRITE) {
8712 case bpf_ctx_range(struct __sk_buff, mark):
8713 case bpf_ctx_range(struct __sk_buff, tc_index):
8714 case bpf_ctx_range(struct __sk_buff, priority):
8715 case bpf_ctx_range(struct __sk_buff, tc_classid):
8716 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8717 case bpf_ctx_range(struct __sk_buff, tstamp):
8718 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8726 case bpf_ctx_range(struct __sk_buff, data):
8727 info->reg_type = PTR_TO_PACKET;
8729 case bpf_ctx_range(struct __sk_buff, data_meta):
8730 info->reg_type = PTR_TO_PACKET_META;
8732 case bpf_ctx_range(struct __sk_buff, data_end):
8733 info->reg_type = PTR_TO_PACKET_END;
8735 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8737 case offsetof(struct __sk_buff, tstamp_type):
8738 /* The convert_ctx_access() on reading and writing
8739 * __sk_buff->tstamp depends on whether the bpf prog
8740 * has used __sk_buff->tstamp_type or not.
8741 * Thus, we need to set prog->tstamp_type_access
8742 * earlier during is_valid_access() here.
8744 ((struct bpf_prog *)prog)->tstamp_type_access = 1;
8745 return size == sizeof(__u8);
8748 return bpf_skb_is_valid_access(off, size, type, prog, info);
8751 DEFINE_MUTEX(nf_conn_btf_access_lock);
8752 EXPORT_SYMBOL_GPL(nf_conn_btf_access_lock);
8754 int (*nfct_btf_struct_access)(struct bpf_verifier_log *log, const struct btf *btf,
8755 const struct btf_type *t, int off, int size,
8756 enum bpf_access_type atype, u32 *next_btf_id,
8757 enum bpf_type_flag *flag);
8758 EXPORT_SYMBOL_GPL(nfct_btf_struct_access);
8760 static int tc_cls_act_btf_struct_access(struct bpf_verifier_log *log,
8761 const struct btf *btf,
8762 const struct btf_type *t, int off,
8763 int size, enum bpf_access_type atype,
8765 enum bpf_type_flag *flag)
8769 if (atype == BPF_READ)
8770 return btf_struct_access(log, btf, t, off, size, atype, next_btf_id,
8773 mutex_lock(&nf_conn_btf_access_lock);
8774 if (nfct_btf_struct_access)
8775 ret = nfct_btf_struct_access(log, btf, t, off, size, atype, next_btf_id, flag);
8776 mutex_unlock(&nf_conn_btf_access_lock);
8781 static bool __is_valid_xdp_access(int off, int size)
8783 if (off < 0 || off >= sizeof(struct xdp_md))
8785 if (off % size != 0)
8787 if (size != sizeof(__u32))
8793 static bool xdp_is_valid_access(int off, int size,
8794 enum bpf_access_type type,
8795 const struct bpf_prog *prog,
8796 struct bpf_insn_access_aux *info)
8798 if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8800 case offsetof(struct xdp_md, egress_ifindex):
8805 if (type == BPF_WRITE) {
8806 if (bpf_prog_is_dev_bound(prog->aux)) {
8808 case offsetof(struct xdp_md, rx_queue_index):
8809 return __is_valid_xdp_access(off, size);
8816 case offsetof(struct xdp_md, data):
8817 info->reg_type = PTR_TO_PACKET;
8819 case offsetof(struct xdp_md, data_meta):
8820 info->reg_type = PTR_TO_PACKET_META;
8822 case offsetof(struct xdp_md, data_end):
8823 info->reg_type = PTR_TO_PACKET_END;
8827 return __is_valid_xdp_access(off, size);
8830 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act)
8832 const u32 act_max = XDP_REDIRECT;
8834 pr_warn_once("%s XDP return value %u on prog %s (id %d) dev %s, expect packet loss!\n",
8835 act > act_max ? "Illegal" : "Driver unsupported",
8836 act, prog->aux->name, prog->aux->id, dev ? dev->name : "N/A");
8838 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
8840 static int xdp_btf_struct_access(struct bpf_verifier_log *log,
8841 const struct btf *btf,
8842 const struct btf_type *t, int off,
8843 int size, enum bpf_access_type atype,
8845 enum bpf_type_flag *flag)
8849 if (atype == BPF_READ)
8850 return btf_struct_access(log, btf, t, off, size, atype, next_btf_id,
8853 mutex_lock(&nf_conn_btf_access_lock);
8854 if (nfct_btf_struct_access)
8855 ret = nfct_btf_struct_access(log, btf, t, off, size, atype, next_btf_id, flag);
8856 mutex_unlock(&nf_conn_btf_access_lock);
8861 static bool sock_addr_is_valid_access(int off, int size,
8862 enum bpf_access_type type,
8863 const struct bpf_prog *prog,
8864 struct bpf_insn_access_aux *info)
8866 const int size_default = sizeof(__u32);
8868 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
8870 if (off % size != 0)
8873 /* Disallow access to IPv6 fields from IPv4 contex and vise
8877 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8878 switch (prog->expected_attach_type) {
8879 case BPF_CGROUP_INET4_BIND:
8880 case BPF_CGROUP_INET4_CONNECT:
8881 case BPF_CGROUP_INET4_GETPEERNAME:
8882 case BPF_CGROUP_INET4_GETSOCKNAME:
8883 case BPF_CGROUP_UDP4_SENDMSG:
8884 case BPF_CGROUP_UDP4_RECVMSG:
8890 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8891 switch (prog->expected_attach_type) {
8892 case BPF_CGROUP_INET6_BIND:
8893 case BPF_CGROUP_INET6_CONNECT:
8894 case BPF_CGROUP_INET6_GETPEERNAME:
8895 case BPF_CGROUP_INET6_GETSOCKNAME:
8896 case BPF_CGROUP_UDP6_SENDMSG:
8897 case BPF_CGROUP_UDP6_RECVMSG:
8903 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8904 switch (prog->expected_attach_type) {
8905 case BPF_CGROUP_UDP4_SENDMSG:
8911 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8913 switch (prog->expected_attach_type) {
8914 case BPF_CGROUP_UDP6_SENDMSG:
8923 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8924 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8925 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8926 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8928 case bpf_ctx_range(struct bpf_sock_addr, user_port):
8929 if (type == BPF_READ) {
8930 bpf_ctx_record_field_size(info, size_default);
8932 if (bpf_ctx_wide_access_ok(off, size,
8933 struct bpf_sock_addr,
8937 if (bpf_ctx_wide_access_ok(off, size,
8938 struct bpf_sock_addr,
8942 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8945 if (bpf_ctx_wide_access_ok(off, size,
8946 struct bpf_sock_addr,
8950 if (bpf_ctx_wide_access_ok(off, size,
8951 struct bpf_sock_addr,
8955 if (size != size_default)
8959 case offsetof(struct bpf_sock_addr, sk):
8960 if (type != BPF_READ)
8962 if (size != sizeof(__u64))
8964 info->reg_type = PTR_TO_SOCKET;
8967 if (type == BPF_READ) {
8968 if (size != size_default)
8978 static bool sock_ops_is_valid_access(int off, int size,
8979 enum bpf_access_type type,
8980 const struct bpf_prog *prog,
8981 struct bpf_insn_access_aux *info)
8983 const int size_default = sizeof(__u32);
8985 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
8988 /* The verifier guarantees that size > 0. */
8989 if (off % size != 0)
8992 if (type == BPF_WRITE) {
8994 case offsetof(struct bpf_sock_ops, reply):
8995 case offsetof(struct bpf_sock_ops, sk_txhash):
8996 if (size != size_default)
9004 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
9006 if (size != sizeof(__u64))
9009 case offsetof(struct bpf_sock_ops, sk):
9010 if (size != sizeof(__u64))
9012 info->reg_type = PTR_TO_SOCKET_OR_NULL;
9014 case offsetof(struct bpf_sock_ops, skb_data):
9015 if (size != sizeof(__u64))
9017 info->reg_type = PTR_TO_PACKET;
9019 case offsetof(struct bpf_sock_ops, skb_data_end):
9020 if (size != sizeof(__u64))
9022 info->reg_type = PTR_TO_PACKET_END;
9024 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9025 bpf_ctx_record_field_size(info, size_default);
9026 return bpf_ctx_narrow_access_ok(off, size,
9029 if (size != size_default)
9038 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
9039 const struct bpf_prog *prog)
9041 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
9044 static bool sk_skb_is_valid_access(int off, int size,
9045 enum bpf_access_type type,
9046 const struct bpf_prog *prog,
9047 struct bpf_insn_access_aux *info)
9050 case bpf_ctx_range(struct __sk_buff, tc_classid):
9051 case bpf_ctx_range(struct __sk_buff, data_meta):
9052 case bpf_ctx_range(struct __sk_buff, tstamp):
9053 case bpf_ctx_range(struct __sk_buff, wire_len):
9054 case bpf_ctx_range(struct __sk_buff, hwtstamp):
9058 if (type == BPF_WRITE) {
9060 case bpf_ctx_range(struct __sk_buff, tc_index):
9061 case bpf_ctx_range(struct __sk_buff, priority):
9069 case bpf_ctx_range(struct __sk_buff, mark):
9071 case bpf_ctx_range(struct __sk_buff, data):
9072 info->reg_type = PTR_TO_PACKET;
9074 case bpf_ctx_range(struct __sk_buff, data_end):
9075 info->reg_type = PTR_TO_PACKET_END;
9079 return bpf_skb_is_valid_access(off, size, type, prog, info);
9082 static bool sk_msg_is_valid_access(int off, int size,
9083 enum bpf_access_type type,
9084 const struct bpf_prog *prog,
9085 struct bpf_insn_access_aux *info)
9087 if (type == BPF_WRITE)
9090 if (off % size != 0)
9094 case offsetof(struct sk_msg_md, data):
9095 info->reg_type = PTR_TO_PACKET;
9096 if (size != sizeof(__u64))
9099 case offsetof(struct sk_msg_md, data_end):
9100 info->reg_type = PTR_TO_PACKET_END;
9101 if (size != sizeof(__u64))
9104 case offsetof(struct sk_msg_md, sk):
9105 if (size != sizeof(__u64))
9107 info->reg_type = PTR_TO_SOCKET;
9109 case bpf_ctx_range(struct sk_msg_md, family):
9110 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
9111 case bpf_ctx_range(struct sk_msg_md, local_ip4):
9112 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
9113 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
9114 case bpf_ctx_range(struct sk_msg_md, remote_port):
9115 case bpf_ctx_range(struct sk_msg_md, local_port):
9116 case bpf_ctx_range(struct sk_msg_md, size):
9117 if (size != sizeof(__u32))
9126 static bool flow_dissector_is_valid_access(int off, int size,
9127 enum bpf_access_type type,
9128 const struct bpf_prog *prog,
9129 struct bpf_insn_access_aux *info)
9131 const int size_default = sizeof(__u32);
9133 if (off < 0 || off >= sizeof(struct __sk_buff))
9136 if (type == BPF_WRITE)
9140 case bpf_ctx_range(struct __sk_buff, data):
9141 if (size != size_default)
9143 info->reg_type = PTR_TO_PACKET;
9145 case bpf_ctx_range(struct __sk_buff, data_end):
9146 if (size != size_default)
9148 info->reg_type = PTR_TO_PACKET_END;
9150 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
9151 if (size != sizeof(__u64))
9153 info->reg_type = PTR_TO_FLOW_KEYS;
9160 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
9161 const struct bpf_insn *si,
9162 struct bpf_insn *insn_buf,
9163 struct bpf_prog *prog,
9167 struct bpf_insn *insn = insn_buf;
9170 case offsetof(struct __sk_buff, data):
9171 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
9172 si->dst_reg, si->src_reg,
9173 offsetof(struct bpf_flow_dissector, data));
9176 case offsetof(struct __sk_buff, data_end):
9177 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
9178 si->dst_reg, si->src_reg,
9179 offsetof(struct bpf_flow_dissector, data_end));
9182 case offsetof(struct __sk_buff, flow_keys):
9183 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
9184 si->dst_reg, si->src_reg,
9185 offsetof(struct bpf_flow_dissector, flow_keys));
9189 return insn - insn_buf;
9192 static struct bpf_insn *bpf_convert_tstamp_type_read(const struct bpf_insn *si,
9193 struct bpf_insn *insn)
9195 __u8 value_reg = si->dst_reg;
9196 __u8 skb_reg = si->src_reg;
9197 /* AX is needed because src_reg and dst_reg could be the same */
9198 __u8 tmp_reg = BPF_REG_AX;
9200 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg,
9201 PKT_VLAN_PRESENT_OFFSET);
9202 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg,
9203 SKB_MONO_DELIVERY_TIME_MASK, 2);
9204 *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_UNSPEC);
9205 *insn++ = BPF_JMP_A(1);
9206 *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_DELIVERY_MONO);
9211 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si,
9212 struct bpf_insn *insn)
9214 /* si->dst_reg = skb_shinfo(SKB); */
9215 #ifdef NET_SKBUFF_DATA_USES_OFFSET
9216 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9217 BPF_REG_AX, si->src_reg,
9218 offsetof(struct sk_buff, end));
9219 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
9220 si->dst_reg, si->src_reg,
9221 offsetof(struct sk_buff, head));
9222 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
9224 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9225 si->dst_reg, si->src_reg,
9226 offsetof(struct sk_buff, end));
9232 static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog,
9233 const struct bpf_insn *si,
9234 struct bpf_insn *insn)
9236 __u8 value_reg = si->dst_reg;
9237 __u8 skb_reg = si->src_reg;
9239 #ifdef CONFIG_NET_CLS_ACT
9240 /* If the tstamp_type is read,
9241 * the bpf prog is aware the tstamp could have delivery time.
9242 * Thus, read skb->tstamp as is if tstamp_type_access is true.
9244 if (!prog->tstamp_type_access) {
9245 /* AX is needed because src_reg and dst_reg could be the same */
9246 __u8 tmp_reg = BPF_REG_AX;
9248 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, PKT_VLAN_PRESENT_OFFSET);
9249 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg,
9250 TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK);
9251 *insn++ = BPF_JMP32_IMM(BPF_JNE, tmp_reg,
9252 TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK, 2);
9253 /* skb->tc_at_ingress && skb->mono_delivery_time,
9254 * read 0 as the (rcv) timestamp.
9256 *insn++ = BPF_MOV64_IMM(value_reg, 0);
9257 *insn++ = BPF_JMP_A(1);
9261 *insn++ = BPF_LDX_MEM(BPF_DW, value_reg, skb_reg,
9262 offsetof(struct sk_buff, tstamp));
9266 static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog,
9267 const struct bpf_insn *si,
9268 struct bpf_insn *insn)
9270 __u8 value_reg = si->src_reg;
9271 __u8 skb_reg = si->dst_reg;
9273 #ifdef CONFIG_NET_CLS_ACT
9274 /* If the tstamp_type is read,
9275 * the bpf prog is aware the tstamp could have delivery time.
9276 * Thus, write skb->tstamp as is if tstamp_type_access is true.
9277 * Otherwise, writing at ingress will have to clear the
9278 * mono_delivery_time bit also.
9280 if (!prog->tstamp_type_access) {
9281 __u8 tmp_reg = BPF_REG_AX;
9283 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, PKT_VLAN_PRESENT_OFFSET);
9284 /* Writing __sk_buff->tstamp as ingress, goto <clear> */
9285 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1);
9287 *insn++ = BPF_JMP_A(2);
9288 /* <clear>: mono_delivery_time */
9289 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_MONO_DELIVERY_TIME_MASK);
9290 *insn++ = BPF_STX_MEM(BPF_B, skb_reg, tmp_reg, PKT_VLAN_PRESENT_OFFSET);
9294 /* <store>: skb->tstamp = tstamp */
9295 *insn++ = BPF_STX_MEM(BPF_DW, skb_reg, value_reg,
9296 offsetof(struct sk_buff, tstamp));
9300 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
9301 const struct bpf_insn *si,
9302 struct bpf_insn *insn_buf,
9303 struct bpf_prog *prog, u32 *target_size)
9305 struct bpf_insn *insn = insn_buf;
9309 case offsetof(struct __sk_buff, len):
9310 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9311 bpf_target_off(struct sk_buff, len, 4,
9315 case offsetof(struct __sk_buff, protocol):
9316 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9317 bpf_target_off(struct sk_buff, protocol, 2,
9321 case offsetof(struct __sk_buff, vlan_proto):
9322 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9323 bpf_target_off(struct sk_buff, vlan_proto, 2,
9327 case offsetof(struct __sk_buff, priority):
9328 if (type == BPF_WRITE)
9329 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9330 bpf_target_off(struct sk_buff, priority, 4,
9333 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9334 bpf_target_off(struct sk_buff, priority, 4,
9338 case offsetof(struct __sk_buff, ingress_ifindex):
9339 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9340 bpf_target_off(struct sk_buff, skb_iif, 4,
9344 case offsetof(struct __sk_buff, ifindex):
9345 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9346 si->dst_reg, si->src_reg,
9347 offsetof(struct sk_buff, dev));
9348 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9349 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9350 bpf_target_off(struct net_device, ifindex, 4,
9354 case offsetof(struct __sk_buff, hash):
9355 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9356 bpf_target_off(struct sk_buff, hash, 4,
9360 case offsetof(struct __sk_buff, mark):
9361 if (type == BPF_WRITE)
9362 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9363 bpf_target_off(struct sk_buff, mark, 4,
9366 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9367 bpf_target_off(struct sk_buff, mark, 4,
9371 case offsetof(struct __sk_buff, pkt_type):
9373 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
9375 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
9376 #ifdef __BIG_ENDIAN_BITFIELD
9377 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
9381 case offsetof(struct __sk_buff, queue_mapping):
9382 if (type == BPF_WRITE) {
9383 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
9384 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
9385 bpf_target_off(struct sk_buff,
9389 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9390 bpf_target_off(struct sk_buff,
9396 case offsetof(struct __sk_buff, vlan_present):
9398 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
9399 PKT_VLAN_PRESENT_OFFSET);
9400 if (PKT_VLAN_PRESENT_BIT)
9401 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
9402 if (PKT_VLAN_PRESENT_BIT < 7)
9403 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
9406 case offsetof(struct __sk_buff, vlan_tci):
9407 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9408 bpf_target_off(struct sk_buff, vlan_tci, 2,
9412 case offsetof(struct __sk_buff, cb[0]) ...
9413 offsetofend(struct __sk_buff, cb[4]) - 1:
9414 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
9415 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
9416 offsetof(struct qdisc_skb_cb, data)) %
9419 prog->cb_access = 1;
9421 off -= offsetof(struct __sk_buff, cb[0]);
9422 off += offsetof(struct sk_buff, cb);
9423 off += offsetof(struct qdisc_skb_cb, data);
9424 if (type == BPF_WRITE)
9425 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
9428 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
9432 case offsetof(struct __sk_buff, tc_classid):
9433 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
9436 off -= offsetof(struct __sk_buff, tc_classid);
9437 off += offsetof(struct sk_buff, cb);
9438 off += offsetof(struct qdisc_skb_cb, tc_classid);
9440 if (type == BPF_WRITE)
9441 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
9444 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
9448 case offsetof(struct __sk_buff, data):
9449 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9450 si->dst_reg, si->src_reg,
9451 offsetof(struct sk_buff, data));
9454 case offsetof(struct __sk_buff, data_meta):
9456 off -= offsetof(struct __sk_buff, data_meta);
9457 off += offsetof(struct sk_buff, cb);
9458 off += offsetof(struct bpf_skb_data_end, data_meta);
9459 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9463 case offsetof(struct __sk_buff, data_end):
9465 off -= offsetof(struct __sk_buff, data_end);
9466 off += offsetof(struct sk_buff, cb);
9467 off += offsetof(struct bpf_skb_data_end, data_end);
9468 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9472 case offsetof(struct __sk_buff, tc_index):
9473 #ifdef CONFIG_NET_SCHED
9474 if (type == BPF_WRITE)
9475 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
9476 bpf_target_off(struct sk_buff, tc_index, 2,
9479 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9480 bpf_target_off(struct sk_buff, tc_index, 2,
9484 if (type == BPF_WRITE)
9485 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
9487 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9491 case offsetof(struct __sk_buff, napi_id):
9492 #if defined(CONFIG_NET_RX_BUSY_POLL)
9493 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9494 bpf_target_off(struct sk_buff, napi_id, 4,
9496 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
9497 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9500 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9503 case offsetof(struct __sk_buff, family):
9504 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9506 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9507 si->dst_reg, si->src_reg,
9508 offsetof(struct sk_buff, sk));
9509 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9510 bpf_target_off(struct sock_common,
9514 case offsetof(struct __sk_buff, remote_ip4):
9515 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9517 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9518 si->dst_reg, si->src_reg,
9519 offsetof(struct sk_buff, sk));
9520 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9521 bpf_target_off(struct sock_common,
9525 case offsetof(struct __sk_buff, local_ip4):
9526 BUILD_BUG_ON(sizeof_field(struct sock_common,
9527 skc_rcv_saddr) != 4);
9529 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9530 si->dst_reg, si->src_reg,
9531 offsetof(struct sk_buff, sk));
9532 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9533 bpf_target_off(struct sock_common,
9537 case offsetof(struct __sk_buff, remote_ip6[0]) ...
9538 offsetof(struct __sk_buff, remote_ip6[3]):
9539 #if IS_ENABLED(CONFIG_IPV6)
9540 BUILD_BUG_ON(sizeof_field(struct sock_common,
9541 skc_v6_daddr.s6_addr32[0]) != 4);
9544 off -= offsetof(struct __sk_buff, remote_ip6[0]);
9546 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9547 si->dst_reg, si->src_reg,
9548 offsetof(struct sk_buff, sk));
9549 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9550 offsetof(struct sock_common,
9551 skc_v6_daddr.s6_addr32[0]) +
9554 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9557 case offsetof(struct __sk_buff, local_ip6[0]) ...
9558 offsetof(struct __sk_buff, local_ip6[3]):
9559 #if IS_ENABLED(CONFIG_IPV6)
9560 BUILD_BUG_ON(sizeof_field(struct sock_common,
9561 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9564 off -= offsetof(struct __sk_buff, local_ip6[0]);
9566 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9567 si->dst_reg, si->src_reg,
9568 offsetof(struct sk_buff, sk));
9569 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9570 offsetof(struct sock_common,
9571 skc_v6_rcv_saddr.s6_addr32[0]) +
9574 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9578 case offsetof(struct __sk_buff, remote_port):
9579 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9581 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9582 si->dst_reg, si->src_reg,
9583 offsetof(struct sk_buff, sk));
9584 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9585 bpf_target_off(struct sock_common,
9588 #ifndef __BIG_ENDIAN_BITFIELD
9589 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9593 case offsetof(struct __sk_buff, local_port):
9594 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9596 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9597 si->dst_reg, si->src_reg,
9598 offsetof(struct sk_buff, sk));
9599 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9600 bpf_target_off(struct sock_common,
9601 skc_num, 2, target_size));
9604 case offsetof(struct __sk_buff, tstamp):
9605 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
9607 if (type == BPF_WRITE)
9608 insn = bpf_convert_tstamp_write(prog, si, insn);
9610 insn = bpf_convert_tstamp_read(prog, si, insn);
9613 case offsetof(struct __sk_buff, tstamp_type):
9614 insn = bpf_convert_tstamp_type_read(si, insn);
9617 case offsetof(struct __sk_buff, gso_segs):
9618 insn = bpf_convert_shinfo_access(si, insn);
9619 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
9620 si->dst_reg, si->dst_reg,
9621 bpf_target_off(struct skb_shared_info,
9625 case offsetof(struct __sk_buff, gso_size):
9626 insn = bpf_convert_shinfo_access(si, insn);
9627 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
9628 si->dst_reg, si->dst_reg,
9629 bpf_target_off(struct skb_shared_info,
9633 case offsetof(struct __sk_buff, wire_len):
9634 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
9637 off -= offsetof(struct __sk_buff, wire_len);
9638 off += offsetof(struct sk_buff, cb);
9639 off += offsetof(struct qdisc_skb_cb, pkt_len);
9641 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
9644 case offsetof(struct __sk_buff, sk):
9645 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9646 si->dst_reg, si->src_reg,
9647 offsetof(struct sk_buff, sk));
9649 case offsetof(struct __sk_buff, hwtstamp):
9650 BUILD_BUG_ON(sizeof_field(struct skb_shared_hwtstamps, hwtstamp) != 8);
9651 BUILD_BUG_ON(offsetof(struct skb_shared_hwtstamps, hwtstamp) != 0);
9653 insn = bpf_convert_shinfo_access(si, insn);
9654 *insn++ = BPF_LDX_MEM(BPF_DW,
9655 si->dst_reg, si->dst_reg,
9656 bpf_target_off(struct skb_shared_info,
9662 return insn - insn_buf;
9665 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
9666 const struct bpf_insn *si,
9667 struct bpf_insn *insn_buf,
9668 struct bpf_prog *prog, u32 *target_size)
9670 struct bpf_insn *insn = insn_buf;
9674 case offsetof(struct bpf_sock, bound_dev_if):
9675 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
9677 if (type == BPF_WRITE)
9678 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9679 offsetof(struct sock, sk_bound_dev_if));
9681 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9682 offsetof(struct sock, sk_bound_dev_if));
9685 case offsetof(struct bpf_sock, mark):
9686 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
9688 if (type == BPF_WRITE)
9689 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9690 offsetof(struct sock, sk_mark));
9692 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9693 offsetof(struct sock, sk_mark));
9696 case offsetof(struct bpf_sock, priority):
9697 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
9699 if (type == BPF_WRITE)
9700 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9701 offsetof(struct sock, sk_priority));
9703 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9704 offsetof(struct sock, sk_priority));
9707 case offsetof(struct bpf_sock, family):
9708 *insn++ = BPF_LDX_MEM(
9709 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
9710 si->dst_reg, si->src_reg,
9711 bpf_target_off(struct sock_common,
9713 sizeof_field(struct sock_common,
9718 case offsetof(struct bpf_sock, type):
9719 *insn++ = BPF_LDX_MEM(
9720 BPF_FIELD_SIZEOF(struct sock, sk_type),
9721 si->dst_reg, si->src_reg,
9722 bpf_target_off(struct sock, sk_type,
9723 sizeof_field(struct sock, sk_type),
9727 case offsetof(struct bpf_sock, protocol):
9728 *insn++ = BPF_LDX_MEM(
9729 BPF_FIELD_SIZEOF(struct sock, sk_protocol),
9730 si->dst_reg, si->src_reg,
9731 bpf_target_off(struct sock, sk_protocol,
9732 sizeof_field(struct sock, sk_protocol),
9736 case offsetof(struct bpf_sock, src_ip4):
9737 *insn++ = BPF_LDX_MEM(
9738 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9739 bpf_target_off(struct sock_common, skc_rcv_saddr,
9740 sizeof_field(struct sock_common,
9745 case offsetof(struct bpf_sock, dst_ip4):
9746 *insn++ = BPF_LDX_MEM(
9747 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9748 bpf_target_off(struct sock_common, skc_daddr,
9749 sizeof_field(struct sock_common,
9754 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
9755 #if IS_ENABLED(CONFIG_IPV6)
9757 off -= offsetof(struct bpf_sock, src_ip6[0]);
9758 *insn++ = BPF_LDX_MEM(
9759 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9762 skc_v6_rcv_saddr.s6_addr32[0],
9763 sizeof_field(struct sock_common,
9764 skc_v6_rcv_saddr.s6_addr32[0]),
9765 target_size) + off);
9768 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9772 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
9773 #if IS_ENABLED(CONFIG_IPV6)
9775 off -= offsetof(struct bpf_sock, dst_ip6[0]);
9776 *insn++ = BPF_LDX_MEM(
9777 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9778 bpf_target_off(struct sock_common,
9779 skc_v6_daddr.s6_addr32[0],
9780 sizeof_field(struct sock_common,
9781 skc_v6_daddr.s6_addr32[0]),
9782 target_size) + off);
9784 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9789 case offsetof(struct bpf_sock, src_port):
9790 *insn++ = BPF_LDX_MEM(
9791 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
9792 si->dst_reg, si->src_reg,
9793 bpf_target_off(struct sock_common, skc_num,
9794 sizeof_field(struct sock_common,
9799 case offsetof(struct bpf_sock, dst_port):
9800 *insn++ = BPF_LDX_MEM(
9801 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
9802 si->dst_reg, si->src_reg,
9803 bpf_target_off(struct sock_common, skc_dport,
9804 sizeof_field(struct sock_common,
9809 case offsetof(struct bpf_sock, state):
9810 *insn++ = BPF_LDX_MEM(
9811 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
9812 si->dst_reg, si->src_reg,
9813 bpf_target_off(struct sock_common, skc_state,
9814 sizeof_field(struct sock_common,
9818 case offsetof(struct bpf_sock, rx_queue_mapping):
9819 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
9820 *insn++ = BPF_LDX_MEM(
9821 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
9822 si->dst_reg, si->src_reg,
9823 bpf_target_off(struct sock, sk_rx_queue_mapping,
9824 sizeof_field(struct sock,
9825 sk_rx_queue_mapping),
9827 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
9829 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9831 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9837 return insn - insn_buf;
9840 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
9841 const struct bpf_insn *si,
9842 struct bpf_insn *insn_buf,
9843 struct bpf_prog *prog, u32 *target_size)
9845 struct bpf_insn *insn = insn_buf;
9848 case offsetof(struct __sk_buff, ifindex):
9849 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9850 si->dst_reg, si->src_reg,
9851 offsetof(struct sk_buff, dev));
9852 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9853 bpf_target_off(struct net_device, ifindex, 4,
9857 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9861 return insn - insn_buf;
9864 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
9865 const struct bpf_insn *si,
9866 struct bpf_insn *insn_buf,
9867 struct bpf_prog *prog, u32 *target_size)
9869 struct bpf_insn *insn = insn_buf;
9872 case offsetof(struct xdp_md, data):
9873 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
9874 si->dst_reg, si->src_reg,
9875 offsetof(struct xdp_buff, data));
9877 case offsetof(struct xdp_md, data_meta):
9878 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
9879 si->dst_reg, si->src_reg,
9880 offsetof(struct xdp_buff, data_meta));
9882 case offsetof(struct xdp_md, data_end):
9883 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
9884 si->dst_reg, si->src_reg,
9885 offsetof(struct xdp_buff, data_end));
9887 case offsetof(struct xdp_md, ingress_ifindex):
9888 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9889 si->dst_reg, si->src_reg,
9890 offsetof(struct xdp_buff, rxq));
9891 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
9892 si->dst_reg, si->dst_reg,
9893 offsetof(struct xdp_rxq_info, dev));
9894 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9895 offsetof(struct net_device, ifindex));
9897 case offsetof(struct xdp_md, rx_queue_index):
9898 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9899 si->dst_reg, si->src_reg,
9900 offsetof(struct xdp_buff, rxq));
9901 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9902 offsetof(struct xdp_rxq_info,
9905 case offsetof(struct xdp_md, egress_ifindex):
9906 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
9907 si->dst_reg, si->src_reg,
9908 offsetof(struct xdp_buff, txq));
9909 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
9910 si->dst_reg, si->dst_reg,
9911 offsetof(struct xdp_txq_info, dev));
9912 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9913 offsetof(struct net_device, ifindex));
9917 return insn - insn_buf;
9920 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
9921 * context Structure, F is Field in context structure that contains a pointer
9922 * to Nested Structure of type NS that has the field NF.
9924 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
9925 * sure that SIZE is not greater than actual size of S.F.NF.
9927 * If offset OFF is provided, the load happens from that offset relative to
9930 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
9932 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
9933 si->src_reg, offsetof(S, F)); \
9934 *insn++ = BPF_LDX_MEM( \
9935 SIZE, si->dst_reg, si->dst_reg, \
9936 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
9941 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
9942 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
9943 BPF_FIELD_SIZEOF(NS, NF), 0)
9945 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
9946 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
9948 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
9949 * "register" since two registers available in convert_ctx_access are not
9950 * enough: we can't override neither SRC, since it contains value to store, nor
9951 * DST since it contains pointer to context that may be used by later
9952 * instructions. But we need a temporary place to save pointer to nested
9953 * structure whose field we want to store to.
9955 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
9957 int tmp_reg = BPF_REG_9; \
9958 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
9960 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
9962 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
9964 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
9965 si->dst_reg, offsetof(S, F)); \
9966 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \
9967 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
9970 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
9974 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
9977 if (type == BPF_WRITE) { \
9978 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
9981 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
9982 S, NS, F, NF, SIZE, OFF); \
9986 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
9987 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
9988 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
9990 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
9991 const struct bpf_insn *si,
9992 struct bpf_insn *insn_buf,
9993 struct bpf_prog *prog, u32 *target_size)
9995 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
9996 struct bpf_insn *insn = insn_buf;
9999 case offsetof(struct bpf_sock_addr, user_family):
10000 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10001 struct sockaddr, uaddr, sa_family);
10004 case offsetof(struct bpf_sock_addr, user_ip4):
10005 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10006 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
10007 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
10010 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
10012 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
10013 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10014 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
10015 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
10019 case offsetof(struct bpf_sock_addr, user_port):
10020 /* To get port we need to know sa_family first and then treat
10021 * sockaddr as either sockaddr_in or sockaddr_in6.
10022 * Though we can simplify since port field has same offset and
10023 * size in both structures.
10024 * Here we check this invariant and use just one of the
10025 * structures if it's true.
10027 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
10028 offsetof(struct sockaddr_in6, sin6_port));
10029 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
10030 sizeof_field(struct sockaddr_in6, sin6_port));
10031 /* Account for sin6_port being smaller than user_port. */
10032 port_size = min(port_size, BPF_LDST_BYTES(si));
10033 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10034 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
10035 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
10038 case offsetof(struct bpf_sock_addr, family):
10039 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10040 struct sock, sk, sk_family);
10043 case offsetof(struct bpf_sock_addr, type):
10044 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10045 struct sock, sk, sk_type);
10048 case offsetof(struct bpf_sock_addr, protocol):
10049 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10050 struct sock, sk, sk_protocol);
10053 case offsetof(struct bpf_sock_addr, msg_src_ip4):
10054 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
10055 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10056 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
10057 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
10060 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
10063 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
10064 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
10065 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10066 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
10067 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
10069 case offsetof(struct bpf_sock_addr, sk):
10070 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
10071 si->dst_reg, si->src_reg,
10072 offsetof(struct bpf_sock_addr_kern, sk));
10076 return insn - insn_buf;
10079 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
10080 const struct bpf_insn *si,
10081 struct bpf_insn *insn_buf,
10082 struct bpf_prog *prog,
10085 struct bpf_insn *insn = insn_buf;
10088 /* Helper macro for adding read access to tcp_sock or sock fields. */
10089 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
10091 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \
10092 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
10093 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
10094 if (si->dst_reg == reg || si->src_reg == reg) \
10096 if (si->dst_reg == reg || si->src_reg == reg) \
10098 if (si->dst_reg == si->src_reg) { \
10099 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
10100 offsetof(struct bpf_sock_ops_kern, \
10102 fullsock_reg = reg; \
10105 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10106 struct bpf_sock_ops_kern, \
10108 fullsock_reg, si->src_reg, \
10109 offsetof(struct bpf_sock_ops_kern, \
10111 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
10112 if (si->dst_reg == si->src_reg) \
10113 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10114 offsetof(struct bpf_sock_ops_kern, \
10116 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10117 struct bpf_sock_ops_kern, sk),\
10118 si->dst_reg, si->src_reg, \
10119 offsetof(struct bpf_sock_ops_kern, sk));\
10120 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
10122 si->dst_reg, si->dst_reg, \
10123 offsetof(OBJ, OBJ_FIELD)); \
10124 if (si->dst_reg == si->src_reg) { \
10125 *insn++ = BPF_JMP_A(1); \
10126 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10127 offsetof(struct bpf_sock_ops_kern, \
10132 #define SOCK_OPS_GET_SK() \
10134 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \
10135 if (si->dst_reg == reg || si->src_reg == reg) \
10137 if (si->dst_reg == reg || si->src_reg == reg) \
10139 if (si->dst_reg == si->src_reg) { \
10140 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
10141 offsetof(struct bpf_sock_ops_kern, \
10143 fullsock_reg = reg; \
10146 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10147 struct bpf_sock_ops_kern, \
10149 fullsock_reg, si->src_reg, \
10150 offsetof(struct bpf_sock_ops_kern, \
10152 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
10153 if (si->dst_reg == si->src_reg) \
10154 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10155 offsetof(struct bpf_sock_ops_kern, \
10157 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10158 struct bpf_sock_ops_kern, sk),\
10159 si->dst_reg, si->src_reg, \
10160 offsetof(struct bpf_sock_ops_kern, sk));\
10161 if (si->dst_reg == si->src_reg) { \
10162 *insn++ = BPF_JMP_A(1); \
10163 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10164 offsetof(struct bpf_sock_ops_kern, \
10169 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
10170 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
10172 /* Helper macro for adding write access to tcp_sock or sock fields.
10173 * The macro is called with two registers, dst_reg which contains a pointer
10174 * to ctx (context) and src_reg which contains the value that should be
10175 * stored. However, we need an additional register since we cannot overwrite
10176 * dst_reg because it may be used later in the program.
10177 * Instead we "borrow" one of the other register. We first save its value
10178 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
10179 * it at the end of the macro.
10181 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
10183 int reg = BPF_REG_9; \
10184 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
10185 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
10186 if (si->dst_reg == reg || si->src_reg == reg) \
10188 if (si->dst_reg == reg || si->src_reg == reg) \
10190 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
10191 offsetof(struct bpf_sock_ops_kern, \
10193 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10194 struct bpf_sock_ops_kern, \
10196 reg, si->dst_reg, \
10197 offsetof(struct bpf_sock_ops_kern, \
10199 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
10200 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10201 struct bpf_sock_ops_kern, sk),\
10202 reg, si->dst_reg, \
10203 offsetof(struct bpf_sock_ops_kern, sk));\
10204 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \
10205 reg, si->src_reg, \
10206 offsetof(OBJ, OBJ_FIELD)); \
10207 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
10208 offsetof(struct bpf_sock_ops_kern, \
10212 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
10214 if (TYPE == BPF_WRITE) \
10215 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
10217 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
10220 if (insn > insn_buf)
10221 return insn - insn_buf;
10224 case offsetof(struct bpf_sock_ops, op):
10225 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10227 si->dst_reg, si->src_reg,
10228 offsetof(struct bpf_sock_ops_kern, op));
10231 case offsetof(struct bpf_sock_ops, replylong[0]) ...
10232 offsetof(struct bpf_sock_ops, replylong[3]):
10233 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
10234 sizeof_field(struct bpf_sock_ops_kern, reply));
10235 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
10236 sizeof_field(struct bpf_sock_ops_kern, replylong));
10238 off -= offsetof(struct bpf_sock_ops, replylong[0]);
10239 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
10240 if (type == BPF_WRITE)
10241 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
10244 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10248 case offsetof(struct bpf_sock_ops, family):
10249 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10251 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10252 struct bpf_sock_ops_kern, sk),
10253 si->dst_reg, si->src_reg,
10254 offsetof(struct bpf_sock_ops_kern, sk));
10255 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10256 offsetof(struct sock_common, skc_family));
10259 case offsetof(struct bpf_sock_ops, remote_ip4):
10260 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10262 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10263 struct bpf_sock_ops_kern, sk),
10264 si->dst_reg, si->src_reg,
10265 offsetof(struct bpf_sock_ops_kern, sk));
10266 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10267 offsetof(struct sock_common, skc_daddr));
10270 case offsetof(struct bpf_sock_ops, local_ip4):
10271 BUILD_BUG_ON(sizeof_field(struct sock_common,
10272 skc_rcv_saddr) != 4);
10274 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10275 struct bpf_sock_ops_kern, sk),
10276 si->dst_reg, si->src_reg,
10277 offsetof(struct bpf_sock_ops_kern, sk));
10278 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10279 offsetof(struct sock_common,
10283 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
10284 offsetof(struct bpf_sock_ops, remote_ip6[3]):
10285 #if IS_ENABLED(CONFIG_IPV6)
10286 BUILD_BUG_ON(sizeof_field(struct sock_common,
10287 skc_v6_daddr.s6_addr32[0]) != 4);
10290 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
10291 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10292 struct bpf_sock_ops_kern, sk),
10293 si->dst_reg, si->src_reg,
10294 offsetof(struct bpf_sock_ops_kern, sk));
10295 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10296 offsetof(struct sock_common,
10297 skc_v6_daddr.s6_addr32[0]) +
10300 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10304 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
10305 offsetof(struct bpf_sock_ops, local_ip6[3]):
10306 #if IS_ENABLED(CONFIG_IPV6)
10307 BUILD_BUG_ON(sizeof_field(struct sock_common,
10308 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10311 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
10312 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10313 struct bpf_sock_ops_kern, sk),
10314 si->dst_reg, si->src_reg,
10315 offsetof(struct bpf_sock_ops_kern, sk));
10316 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10317 offsetof(struct sock_common,
10318 skc_v6_rcv_saddr.s6_addr32[0]) +
10321 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10325 case offsetof(struct bpf_sock_ops, remote_port):
10326 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10328 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10329 struct bpf_sock_ops_kern, sk),
10330 si->dst_reg, si->src_reg,
10331 offsetof(struct bpf_sock_ops_kern, sk));
10332 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10333 offsetof(struct sock_common, skc_dport));
10334 #ifndef __BIG_ENDIAN_BITFIELD
10335 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10339 case offsetof(struct bpf_sock_ops, local_port):
10340 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10342 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10343 struct bpf_sock_ops_kern, sk),
10344 si->dst_reg, si->src_reg,
10345 offsetof(struct bpf_sock_ops_kern, sk));
10346 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10347 offsetof(struct sock_common, skc_num));
10350 case offsetof(struct bpf_sock_ops, is_fullsock):
10351 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10352 struct bpf_sock_ops_kern,
10354 si->dst_reg, si->src_reg,
10355 offsetof(struct bpf_sock_ops_kern,
10359 case offsetof(struct bpf_sock_ops, state):
10360 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
10362 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10363 struct bpf_sock_ops_kern, sk),
10364 si->dst_reg, si->src_reg,
10365 offsetof(struct bpf_sock_ops_kern, sk));
10366 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
10367 offsetof(struct sock_common, skc_state));
10370 case offsetof(struct bpf_sock_ops, rtt_min):
10371 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
10372 sizeof(struct minmax));
10373 BUILD_BUG_ON(sizeof(struct minmax) <
10374 sizeof(struct minmax_sample));
10376 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10377 struct bpf_sock_ops_kern, sk),
10378 si->dst_reg, si->src_reg,
10379 offsetof(struct bpf_sock_ops_kern, sk));
10380 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10381 offsetof(struct tcp_sock, rtt_min) +
10382 sizeof_field(struct minmax_sample, t));
10385 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
10386 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
10390 case offsetof(struct bpf_sock_ops, sk_txhash):
10391 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
10392 struct sock, type);
10394 case offsetof(struct bpf_sock_ops, snd_cwnd):
10395 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
10397 case offsetof(struct bpf_sock_ops, srtt_us):
10398 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
10400 case offsetof(struct bpf_sock_ops, snd_ssthresh):
10401 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
10403 case offsetof(struct bpf_sock_ops, rcv_nxt):
10404 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
10406 case offsetof(struct bpf_sock_ops, snd_nxt):
10407 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
10409 case offsetof(struct bpf_sock_ops, snd_una):
10410 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
10412 case offsetof(struct bpf_sock_ops, mss_cache):
10413 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
10415 case offsetof(struct bpf_sock_ops, ecn_flags):
10416 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
10418 case offsetof(struct bpf_sock_ops, rate_delivered):
10419 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
10421 case offsetof(struct bpf_sock_ops, rate_interval_us):
10422 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
10424 case offsetof(struct bpf_sock_ops, packets_out):
10425 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
10427 case offsetof(struct bpf_sock_ops, retrans_out):
10428 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
10430 case offsetof(struct bpf_sock_ops, total_retrans):
10431 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
10433 case offsetof(struct bpf_sock_ops, segs_in):
10434 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
10436 case offsetof(struct bpf_sock_ops, data_segs_in):
10437 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
10439 case offsetof(struct bpf_sock_ops, segs_out):
10440 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
10442 case offsetof(struct bpf_sock_ops, data_segs_out):
10443 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
10445 case offsetof(struct bpf_sock_ops, lost_out):
10446 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
10448 case offsetof(struct bpf_sock_ops, sacked_out):
10449 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
10451 case offsetof(struct bpf_sock_ops, bytes_received):
10452 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
10454 case offsetof(struct bpf_sock_ops, bytes_acked):
10455 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
10457 case offsetof(struct bpf_sock_ops, sk):
10460 case offsetof(struct bpf_sock_ops, skb_data_end):
10461 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10463 si->dst_reg, si->src_reg,
10464 offsetof(struct bpf_sock_ops_kern,
10467 case offsetof(struct bpf_sock_ops, skb_data):
10468 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10470 si->dst_reg, si->src_reg,
10471 offsetof(struct bpf_sock_ops_kern,
10473 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10474 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10475 si->dst_reg, si->dst_reg,
10476 offsetof(struct sk_buff, data));
10478 case offsetof(struct bpf_sock_ops, skb_len):
10479 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10481 si->dst_reg, si->src_reg,
10482 offsetof(struct bpf_sock_ops_kern,
10484 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10485 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10486 si->dst_reg, si->dst_reg,
10487 offsetof(struct sk_buff, len));
10489 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
10490 off = offsetof(struct sk_buff, cb);
10491 off += offsetof(struct tcp_skb_cb, tcp_flags);
10492 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
10493 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10495 si->dst_reg, si->src_reg,
10496 offsetof(struct bpf_sock_ops_kern,
10498 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10499 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
10501 si->dst_reg, si->dst_reg, off);
10504 return insn - insn_buf;
10507 /* data_end = skb->data + skb_headlen() */
10508 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
10509 struct bpf_insn *insn)
10512 int temp_reg_off = offsetof(struct sk_buff, cb) +
10513 offsetof(struct sk_skb_cb, temp_reg);
10515 if (si->src_reg == si->dst_reg) {
10516 /* We need an extra register, choose and save a register. */
10518 if (si->src_reg == reg || si->dst_reg == reg)
10520 if (si->src_reg == reg || si->dst_reg == reg)
10522 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off);
10527 /* reg = skb->data */
10528 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10530 offsetof(struct sk_buff, data));
10531 /* AX = skb->len */
10532 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10533 BPF_REG_AX, si->src_reg,
10534 offsetof(struct sk_buff, len));
10535 /* reg = skb->data + skb->len */
10536 *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX);
10537 /* AX = skb->data_len */
10538 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
10539 BPF_REG_AX, si->src_reg,
10540 offsetof(struct sk_buff, data_len));
10542 /* reg = skb->data + skb->len - skb->data_len */
10543 *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX);
10545 if (si->src_reg == si->dst_reg) {
10546 /* Restore the saved register */
10547 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg);
10548 *insn++ = BPF_MOV64_REG(si->dst_reg, reg);
10549 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off);
10555 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
10556 const struct bpf_insn *si,
10557 struct bpf_insn *insn_buf,
10558 struct bpf_prog *prog, u32 *target_size)
10560 struct bpf_insn *insn = insn_buf;
10564 case offsetof(struct __sk_buff, data_end):
10565 insn = bpf_convert_data_end_access(si, insn);
10567 case offsetof(struct __sk_buff, cb[0]) ...
10568 offsetofend(struct __sk_buff, cb[4]) - 1:
10569 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20);
10570 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
10571 offsetof(struct sk_skb_cb, data)) %
10574 prog->cb_access = 1;
10576 off -= offsetof(struct __sk_buff, cb[0]);
10577 off += offsetof(struct sk_buff, cb);
10578 off += offsetof(struct sk_skb_cb, data);
10579 if (type == BPF_WRITE)
10580 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
10583 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
10589 return bpf_convert_ctx_access(type, si, insn_buf, prog,
10593 return insn - insn_buf;
10596 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
10597 const struct bpf_insn *si,
10598 struct bpf_insn *insn_buf,
10599 struct bpf_prog *prog, u32 *target_size)
10601 struct bpf_insn *insn = insn_buf;
10602 #if IS_ENABLED(CONFIG_IPV6)
10606 /* convert ctx uses the fact sg element is first in struct */
10607 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
10610 case offsetof(struct sk_msg_md, data):
10611 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
10612 si->dst_reg, si->src_reg,
10613 offsetof(struct sk_msg, data));
10615 case offsetof(struct sk_msg_md, data_end):
10616 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
10617 si->dst_reg, si->src_reg,
10618 offsetof(struct sk_msg, data_end));
10620 case offsetof(struct sk_msg_md, family):
10621 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10623 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10624 struct sk_msg, sk),
10625 si->dst_reg, si->src_reg,
10626 offsetof(struct sk_msg, sk));
10627 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10628 offsetof(struct sock_common, skc_family));
10631 case offsetof(struct sk_msg_md, remote_ip4):
10632 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10634 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10635 struct sk_msg, sk),
10636 si->dst_reg, si->src_reg,
10637 offsetof(struct sk_msg, sk));
10638 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10639 offsetof(struct sock_common, skc_daddr));
10642 case offsetof(struct sk_msg_md, local_ip4):
10643 BUILD_BUG_ON(sizeof_field(struct sock_common,
10644 skc_rcv_saddr) != 4);
10646 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10647 struct sk_msg, sk),
10648 si->dst_reg, si->src_reg,
10649 offsetof(struct sk_msg, sk));
10650 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10651 offsetof(struct sock_common,
10655 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
10656 offsetof(struct sk_msg_md, remote_ip6[3]):
10657 #if IS_ENABLED(CONFIG_IPV6)
10658 BUILD_BUG_ON(sizeof_field(struct sock_common,
10659 skc_v6_daddr.s6_addr32[0]) != 4);
10662 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
10663 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10664 struct sk_msg, sk),
10665 si->dst_reg, si->src_reg,
10666 offsetof(struct sk_msg, sk));
10667 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10668 offsetof(struct sock_common,
10669 skc_v6_daddr.s6_addr32[0]) +
10672 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10676 case offsetof(struct sk_msg_md, local_ip6[0]) ...
10677 offsetof(struct sk_msg_md, local_ip6[3]):
10678 #if IS_ENABLED(CONFIG_IPV6)
10679 BUILD_BUG_ON(sizeof_field(struct sock_common,
10680 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10683 off -= offsetof(struct sk_msg_md, local_ip6[0]);
10684 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10685 struct sk_msg, sk),
10686 si->dst_reg, si->src_reg,
10687 offsetof(struct sk_msg, sk));
10688 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10689 offsetof(struct sock_common,
10690 skc_v6_rcv_saddr.s6_addr32[0]) +
10693 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10697 case offsetof(struct sk_msg_md, remote_port):
10698 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10700 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10701 struct sk_msg, sk),
10702 si->dst_reg, si->src_reg,
10703 offsetof(struct sk_msg, sk));
10704 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10705 offsetof(struct sock_common, skc_dport));
10706 #ifndef __BIG_ENDIAN_BITFIELD
10707 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10711 case offsetof(struct sk_msg_md, local_port):
10712 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10714 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10715 struct sk_msg, sk),
10716 si->dst_reg, si->src_reg,
10717 offsetof(struct sk_msg, sk));
10718 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10719 offsetof(struct sock_common, skc_num));
10722 case offsetof(struct sk_msg_md, size):
10723 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
10724 si->dst_reg, si->src_reg,
10725 offsetof(struct sk_msg_sg, size));
10728 case offsetof(struct sk_msg_md, sk):
10729 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
10730 si->dst_reg, si->src_reg,
10731 offsetof(struct sk_msg, sk));
10735 return insn - insn_buf;
10738 const struct bpf_verifier_ops sk_filter_verifier_ops = {
10739 .get_func_proto = sk_filter_func_proto,
10740 .is_valid_access = sk_filter_is_valid_access,
10741 .convert_ctx_access = bpf_convert_ctx_access,
10742 .gen_ld_abs = bpf_gen_ld_abs,
10745 const struct bpf_prog_ops sk_filter_prog_ops = {
10746 .test_run = bpf_prog_test_run_skb,
10749 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
10750 .get_func_proto = tc_cls_act_func_proto,
10751 .is_valid_access = tc_cls_act_is_valid_access,
10752 .convert_ctx_access = tc_cls_act_convert_ctx_access,
10753 .gen_prologue = tc_cls_act_prologue,
10754 .gen_ld_abs = bpf_gen_ld_abs,
10755 .btf_struct_access = tc_cls_act_btf_struct_access,
10758 const struct bpf_prog_ops tc_cls_act_prog_ops = {
10759 .test_run = bpf_prog_test_run_skb,
10762 const struct bpf_verifier_ops xdp_verifier_ops = {
10763 .get_func_proto = xdp_func_proto,
10764 .is_valid_access = xdp_is_valid_access,
10765 .convert_ctx_access = xdp_convert_ctx_access,
10766 .gen_prologue = bpf_noop_prologue,
10767 .btf_struct_access = xdp_btf_struct_access,
10770 const struct bpf_prog_ops xdp_prog_ops = {
10771 .test_run = bpf_prog_test_run_xdp,
10774 const struct bpf_verifier_ops cg_skb_verifier_ops = {
10775 .get_func_proto = cg_skb_func_proto,
10776 .is_valid_access = cg_skb_is_valid_access,
10777 .convert_ctx_access = bpf_convert_ctx_access,
10780 const struct bpf_prog_ops cg_skb_prog_ops = {
10781 .test_run = bpf_prog_test_run_skb,
10784 const struct bpf_verifier_ops lwt_in_verifier_ops = {
10785 .get_func_proto = lwt_in_func_proto,
10786 .is_valid_access = lwt_is_valid_access,
10787 .convert_ctx_access = bpf_convert_ctx_access,
10790 const struct bpf_prog_ops lwt_in_prog_ops = {
10791 .test_run = bpf_prog_test_run_skb,
10794 const struct bpf_verifier_ops lwt_out_verifier_ops = {
10795 .get_func_proto = lwt_out_func_proto,
10796 .is_valid_access = lwt_is_valid_access,
10797 .convert_ctx_access = bpf_convert_ctx_access,
10800 const struct bpf_prog_ops lwt_out_prog_ops = {
10801 .test_run = bpf_prog_test_run_skb,
10804 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
10805 .get_func_proto = lwt_xmit_func_proto,
10806 .is_valid_access = lwt_is_valid_access,
10807 .convert_ctx_access = bpf_convert_ctx_access,
10808 .gen_prologue = tc_cls_act_prologue,
10811 const struct bpf_prog_ops lwt_xmit_prog_ops = {
10812 .test_run = bpf_prog_test_run_skb,
10815 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
10816 .get_func_proto = lwt_seg6local_func_proto,
10817 .is_valid_access = lwt_is_valid_access,
10818 .convert_ctx_access = bpf_convert_ctx_access,
10821 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
10822 .test_run = bpf_prog_test_run_skb,
10825 const struct bpf_verifier_ops cg_sock_verifier_ops = {
10826 .get_func_proto = sock_filter_func_proto,
10827 .is_valid_access = sock_filter_is_valid_access,
10828 .convert_ctx_access = bpf_sock_convert_ctx_access,
10831 const struct bpf_prog_ops cg_sock_prog_ops = {
10834 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
10835 .get_func_proto = sock_addr_func_proto,
10836 .is_valid_access = sock_addr_is_valid_access,
10837 .convert_ctx_access = sock_addr_convert_ctx_access,
10840 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
10843 const struct bpf_verifier_ops sock_ops_verifier_ops = {
10844 .get_func_proto = sock_ops_func_proto,
10845 .is_valid_access = sock_ops_is_valid_access,
10846 .convert_ctx_access = sock_ops_convert_ctx_access,
10849 const struct bpf_prog_ops sock_ops_prog_ops = {
10852 const struct bpf_verifier_ops sk_skb_verifier_ops = {
10853 .get_func_proto = sk_skb_func_proto,
10854 .is_valid_access = sk_skb_is_valid_access,
10855 .convert_ctx_access = sk_skb_convert_ctx_access,
10856 .gen_prologue = sk_skb_prologue,
10859 const struct bpf_prog_ops sk_skb_prog_ops = {
10862 const struct bpf_verifier_ops sk_msg_verifier_ops = {
10863 .get_func_proto = sk_msg_func_proto,
10864 .is_valid_access = sk_msg_is_valid_access,
10865 .convert_ctx_access = sk_msg_convert_ctx_access,
10866 .gen_prologue = bpf_noop_prologue,
10869 const struct bpf_prog_ops sk_msg_prog_ops = {
10872 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
10873 .get_func_proto = flow_dissector_func_proto,
10874 .is_valid_access = flow_dissector_is_valid_access,
10875 .convert_ctx_access = flow_dissector_convert_ctx_access,
10878 const struct bpf_prog_ops flow_dissector_prog_ops = {
10879 .test_run = bpf_prog_test_run_flow_dissector,
10882 int sk_detach_filter(struct sock *sk)
10885 struct sk_filter *filter;
10887 if (sock_flag(sk, SOCK_FILTER_LOCKED))
10890 filter = rcu_dereference_protected(sk->sk_filter,
10891 lockdep_sock_is_held(sk));
10893 RCU_INIT_POINTER(sk->sk_filter, NULL);
10894 sk_filter_uncharge(sk, filter);
10900 EXPORT_SYMBOL_GPL(sk_detach_filter);
10902 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len)
10904 struct sock_fprog_kern *fprog;
10905 struct sk_filter *filter;
10908 sockopt_lock_sock(sk);
10909 filter = rcu_dereference_protected(sk->sk_filter,
10910 lockdep_sock_is_held(sk));
10914 /* We're copying the filter that has been originally attached,
10915 * so no conversion/decode needed anymore. eBPF programs that
10916 * have no original program cannot be dumped through this.
10919 fprog = filter->prog->orig_prog;
10925 /* User space only enquires number of filter blocks. */
10929 if (len < fprog->len)
10933 if (copy_to_sockptr(optval, fprog->filter, bpf_classic_proglen(fprog)))
10936 /* Instead of bytes, the API requests to return the number
10937 * of filter blocks.
10941 sockopt_release_sock(sk);
10946 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
10947 struct sock_reuseport *reuse,
10948 struct sock *sk, struct sk_buff *skb,
10949 struct sock *migrating_sk,
10952 reuse_kern->skb = skb;
10953 reuse_kern->sk = sk;
10954 reuse_kern->selected_sk = NULL;
10955 reuse_kern->migrating_sk = migrating_sk;
10956 reuse_kern->data_end = skb->data + skb_headlen(skb);
10957 reuse_kern->hash = hash;
10958 reuse_kern->reuseport_id = reuse->reuseport_id;
10959 reuse_kern->bind_inany = reuse->bind_inany;
10962 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
10963 struct bpf_prog *prog, struct sk_buff *skb,
10964 struct sock *migrating_sk,
10967 struct sk_reuseport_kern reuse_kern;
10968 enum sk_action action;
10970 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
10971 action = bpf_prog_run(prog, &reuse_kern);
10973 if (action == SK_PASS)
10974 return reuse_kern.selected_sk;
10976 return ERR_PTR(-ECONNREFUSED);
10979 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
10980 struct bpf_map *, map, void *, key, u32, flags)
10982 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
10983 struct sock_reuseport *reuse;
10984 struct sock *selected_sk;
10986 selected_sk = map->ops->map_lookup_elem(map, key);
10990 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
10992 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
10993 if (sk_is_refcounted(selected_sk))
10994 sock_put(selected_sk);
10996 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
10997 * The only (!reuse) case here is - the sk has already been
10998 * unhashed (e.g. by close()), so treat it as -ENOENT.
11000 * Other maps (e.g. sock_map) do not provide this guarantee and
11001 * the sk may never be in the reuseport group to begin with.
11003 return is_sockarray ? -ENOENT : -EINVAL;
11006 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
11007 struct sock *sk = reuse_kern->sk;
11009 if (sk->sk_protocol != selected_sk->sk_protocol)
11010 return -EPROTOTYPE;
11011 else if (sk->sk_family != selected_sk->sk_family)
11012 return -EAFNOSUPPORT;
11014 /* Catch all. Likely bound to a different sockaddr. */
11018 reuse_kern->selected_sk = selected_sk;
11023 static const struct bpf_func_proto sk_select_reuseport_proto = {
11024 .func = sk_select_reuseport,
11026 .ret_type = RET_INTEGER,
11027 .arg1_type = ARG_PTR_TO_CTX,
11028 .arg2_type = ARG_CONST_MAP_PTR,
11029 .arg3_type = ARG_PTR_TO_MAP_KEY,
11030 .arg4_type = ARG_ANYTHING,
11033 BPF_CALL_4(sk_reuseport_load_bytes,
11034 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11035 void *, to, u32, len)
11037 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
11040 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
11041 .func = sk_reuseport_load_bytes,
11043 .ret_type = RET_INTEGER,
11044 .arg1_type = ARG_PTR_TO_CTX,
11045 .arg2_type = ARG_ANYTHING,
11046 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
11047 .arg4_type = ARG_CONST_SIZE,
11050 BPF_CALL_5(sk_reuseport_load_bytes_relative,
11051 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11052 void *, to, u32, len, u32, start_header)
11054 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
11055 len, start_header);
11058 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
11059 .func = sk_reuseport_load_bytes_relative,
11061 .ret_type = RET_INTEGER,
11062 .arg1_type = ARG_PTR_TO_CTX,
11063 .arg2_type = ARG_ANYTHING,
11064 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
11065 .arg4_type = ARG_CONST_SIZE,
11066 .arg5_type = ARG_ANYTHING,
11069 static const struct bpf_func_proto *
11070 sk_reuseport_func_proto(enum bpf_func_id func_id,
11071 const struct bpf_prog *prog)
11074 case BPF_FUNC_sk_select_reuseport:
11075 return &sk_select_reuseport_proto;
11076 case BPF_FUNC_skb_load_bytes:
11077 return &sk_reuseport_load_bytes_proto;
11078 case BPF_FUNC_skb_load_bytes_relative:
11079 return &sk_reuseport_load_bytes_relative_proto;
11080 case BPF_FUNC_get_socket_cookie:
11081 return &bpf_get_socket_ptr_cookie_proto;
11082 case BPF_FUNC_ktime_get_coarse_ns:
11083 return &bpf_ktime_get_coarse_ns_proto;
11085 return bpf_base_func_proto(func_id);
11090 sk_reuseport_is_valid_access(int off, int size,
11091 enum bpf_access_type type,
11092 const struct bpf_prog *prog,
11093 struct bpf_insn_access_aux *info)
11095 const u32 size_default = sizeof(__u32);
11097 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
11098 off % size || type != BPF_READ)
11102 case offsetof(struct sk_reuseport_md, data):
11103 info->reg_type = PTR_TO_PACKET;
11104 return size == sizeof(__u64);
11106 case offsetof(struct sk_reuseport_md, data_end):
11107 info->reg_type = PTR_TO_PACKET_END;
11108 return size == sizeof(__u64);
11110 case offsetof(struct sk_reuseport_md, hash):
11111 return size == size_default;
11113 case offsetof(struct sk_reuseport_md, sk):
11114 info->reg_type = PTR_TO_SOCKET;
11115 return size == sizeof(__u64);
11117 case offsetof(struct sk_reuseport_md, migrating_sk):
11118 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
11119 return size == sizeof(__u64);
11121 /* Fields that allow narrowing */
11122 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
11123 if (size < sizeof_field(struct sk_buff, protocol))
11126 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
11127 case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
11128 case bpf_ctx_range(struct sk_reuseport_md, len):
11129 bpf_ctx_record_field_size(info, size_default);
11130 return bpf_ctx_narrow_access_ok(off, size, size_default);
11137 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
11138 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
11139 si->dst_reg, si->src_reg, \
11140 bpf_target_off(struct sk_reuseport_kern, F, \
11141 sizeof_field(struct sk_reuseport_kern, F), \
11145 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
11146 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
11151 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \
11152 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
11157 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
11158 const struct bpf_insn *si,
11159 struct bpf_insn *insn_buf,
11160 struct bpf_prog *prog,
11163 struct bpf_insn *insn = insn_buf;
11166 case offsetof(struct sk_reuseport_md, data):
11167 SK_REUSEPORT_LOAD_SKB_FIELD(data);
11170 case offsetof(struct sk_reuseport_md, len):
11171 SK_REUSEPORT_LOAD_SKB_FIELD(len);
11174 case offsetof(struct sk_reuseport_md, eth_protocol):
11175 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
11178 case offsetof(struct sk_reuseport_md, ip_protocol):
11179 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
11182 case offsetof(struct sk_reuseport_md, data_end):
11183 SK_REUSEPORT_LOAD_FIELD(data_end);
11186 case offsetof(struct sk_reuseport_md, hash):
11187 SK_REUSEPORT_LOAD_FIELD(hash);
11190 case offsetof(struct sk_reuseport_md, bind_inany):
11191 SK_REUSEPORT_LOAD_FIELD(bind_inany);
11194 case offsetof(struct sk_reuseport_md, sk):
11195 SK_REUSEPORT_LOAD_FIELD(sk);
11198 case offsetof(struct sk_reuseport_md, migrating_sk):
11199 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
11203 return insn - insn_buf;
11206 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
11207 .get_func_proto = sk_reuseport_func_proto,
11208 .is_valid_access = sk_reuseport_is_valid_access,
11209 .convert_ctx_access = sk_reuseport_convert_ctx_access,
11212 const struct bpf_prog_ops sk_reuseport_prog_ops = {
11215 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
11216 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
11218 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
11219 struct sock *, sk, u64, flags)
11221 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
11222 BPF_SK_LOOKUP_F_NO_REUSEPORT)))
11224 if (unlikely(sk && sk_is_refcounted(sk)))
11225 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
11226 if (unlikely(sk && sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN))
11227 return -ESOCKTNOSUPPORT; /* only accept TCP socket in LISTEN */
11228 if (unlikely(sk && sk_is_udp(sk) && sk->sk_state != TCP_CLOSE))
11229 return -ESOCKTNOSUPPORT; /* only accept UDP socket in CLOSE */
11231 /* Check if socket is suitable for packet L3/L4 protocol */
11232 if (sk && sk->sk_protocol != ctx->protocol)
11233 return -EPROTOTYPE;
11234 if (sk && sk->sk_family != ctx->family &&
11235 (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
11236 return -EAFNOSUPPORT;
11238 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
11241 /* Select socket as lookup result */
11242 ctx->selected_sk = sk;
11243 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
11247 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
11248 .func = bpf_sk_lookup_assign,
11250 .ret_type = RET_INTEGER,
11251 .arg1_type = ARG_PTR_TO_CTX,
11252 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL,
11253 .arg3_type = ARG_ANYTHING,
11256 static const struct bpf_func_proto *
11257 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
11260 case BPF_FUNC_perf_event_output:
11261 return &bpf_event_output_data_proto;
11262 case BPF_FUNC_sk_assign:
11263 return &bpf_sk_lookup_assign_proto;
11264 case BPF_FUNC_sk_release:
11265 return &bpf_sk_release_proto;
11267 return bpf_sk_base_func_proto(func_id);
11271 static bool sk_lookup_is_valid_access(int off, int size,
11272 enum bpf_access_type type,
11273 const struct bpf_prog *prog,
11274 struct bpf_insn_access_aux *info)
11276 if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
11278 if (off % size != 0)
11280 if (type != BPF_READ)
11284 case offsetof(struct bpf_sk_lookup, sk):
11285 info->reg_type = PTR_TO_SOCKET_OR_NULL;
11286 return size == sizeof(__u64);
11288 case bpf_ctx_range(struct bpf_sk_lookup, family):
11289 case bpf_ctx_range(struct bpf_sk_lookup, protocol):
11290 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
11291 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
11292 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
11293 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
11294 case bpf_ctx_range(struct bpf_sk_lookup, local_port):
11295 case bpf_ctx_range(struct bpf_sk_lookup, ingress_ifindex):
11296 bpf_ctx_record_field_size(info, sizeof(__u32));
11297 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
11299 case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
11300 /* Allow 4-byte access to 2-byte field for backward compatibility */
11301 if (size == sizeof(__u32))
11303 bpf_ctx_record_field_size(info, sizeof(__be16));
11304 return bpf_ctx_narrow_access_ok(off, size, sizeof(__be16));
11306 case offsetofend(struct bpf_sk_lookup, remote_port) ...
11307 offsetof(struct bpf_sk_lookup, local_ip4) - 1:
11308 /* Allow access to zero padding for backward compatibility */
11309 bpf_ctx_record_field_size(info, sizeof(__u16));
11310 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u16));
11317 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
11318 const struct bpf_insn *si,
11319 struct bpf_insn *insn_buf,
11320 struct bpf_prog *prog,
11323 struct bpf_insn *insn = insn_buf;
11326 case offsetof(struct bpf_sk_lookup, sk):
11327 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11328 offsetof(struct bpf_sk_lookup_kern, selected_sk));
11331 case offsetof(struct bpf_sk_lookup, family):
11332 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11333 bpf_target_off(struct bpf_sk_lookup_kern,
11334 family, 2, target_size));
11337 case offsetof(struct bpf_sk_lookup, protocol):
11338 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11339 bpf_target_off(struct bpf_sk_lookup_kern,
11340 protocol, 2, target_size));
11343 case offsetof(struct bpf_sk_lookup, remote_ip4):
11344 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11345 bpf_target_off(struct bpf_sk_lookup_kern,
11346 v4.saddr, 4, target_size));
11349 case offsetof(struct bpf_sk_lookup, local_ip4):
11350 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11351 bpf_target_off(struct bpf_sk_lookup_kern,
11352 v4.daddr, 4, target_size));
11355 case bpf_ctx_range_till(struct bpf_sk_lookup,
11356 remote_ip6[0], remote_ip6[3]): {
11357 #if IS_ENABLED(CONFIG_IPV6)
11360 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
11361 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11362 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11363 offsetof(struct bpf_sk_lookup_kern, v6.saddr));
11364 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11365 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11367 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11371 case bpf_ctx_range_till(struct bpf_sk_lookup,
11372 local_ip6[0], local_ip6[3]): {
11373 #if IS_ENABLED(CONFIG_IPV6)
11376 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
11377 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11378 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11379 offsetof(struct bpf_sk_lookup_kern, v6.daddr));
11380 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11381 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11383 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11387 case offsetof(struct bpf_sk_lookup, remote_port):
11388 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11389 bpf_target_off(struct bpf_sk_lookup_kern,
11390 sport, 2, target_size));
11393 case offsetofend(struct bpf_sk_lookup, remote_port):
11395 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11398 case offsetof(struct bpf_sk_lookup, local_port):
11399 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11400 bpf_target_off(struct bpf_sk_lookup_kern,
11401 dport, 2, target_size));
11404 case offsetof(struct bpf_sk_lookup, ingress_ifindex):
11405 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11406 bpf_target_off(struct bpf_sk_lookup_kern,
11407 ingress_ifindex, 4, target_size));
11411 return insn - insn_buf;
11414 const struct bpf_prog_ops sk_lookup_prog_ops = {
11415 .test_run = bpf_prog_test_run_sk_lookup,
11418 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
11419 .get_func_proto = sk_lookup_func_proto,
11420 .is_valid_access = sk_lookup_is_valid_access,
11421 .convert_ctx_access = sk_lookup_convert_ctx_access,
11424 #endif /* CONFIG_INET */
11426 DEFINE_BPF_DISPATCHER(xdp)
11428 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
11430 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
11433 BTF_ID_LIST_GLOBAL(btf_sock_ids, MAX_BTF_SOCK_TYPE)
11434 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
11436 #undef BTF_SOCK_TYPE
11438 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
11440 /* tcp6_sock type is not generated in dwarf and hence btf,
11441 * trigger an explicit type generation here.
11443 BTF_TYPE_EMIT(struct tcp6_sock);
11444 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
11445 sk->sk_family == AF_INET6)
11446 return (unsigned long)sk;
11448 return (unsigned long)NULL;
11451 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
11452 .func = bpf_skc_to_tcp6_sock,
11454 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11455 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11456 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
11459 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
11461 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
11462 return (unsigned long)sk;
11464 return (unsigned long)NULL;
11467 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
11468 .func = bpf_skc_to_tcp_sock,
11470 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11471 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11472 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
11475 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
11477 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
11478 * generated if CONFIG_INET=n. Trigger an explicit generation here.
11480 BTF_TYPE_EMIT(struct inet_timewait_sock);
11481 BTF_TYPE_EMIT(struct tcp_timewait_sock);
11484 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
11485 return (unsigned long)sk;
11488 #if IS_BUILTIN(CONFIG_IPV6)
11489 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
11490 return (unsigned long)sk;
11493 return (unsigned long)NULL;
11496 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
11497 .func = bpf_skc_to_tcp_timewait_sock,
11499 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11500 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11501 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
11504 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
11507 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11508 return (unsigned long)sk;
11511 #if IS_BUILTIN(CONFIG_IPV6)
11512 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11513 return (unsigned long)sk;
11516 return (unsigned long)NULL;
11519 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
11520 .func = bpf_skc_to_tcp_request_sock,
11522 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11523 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11524 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
11527 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
11529 /* udp6_sock type is not generated in dwarf and hence btf,
11530 * trigger an explicit type generation here.
11532 BTF_TYPE_EMIT(struct udp6_sock);
11533 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
11534 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
11535 return (unsigned long)sk;
11537 return (unsigned long)NULL;
11540 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
11541 .func = bpf_skc_to_udp6_sock,
11543 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11544 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11545 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
11548 BPF_CALL_1(bpf_skc_to_unix_sock, struct sock *, sk)
11550 /* unix_sock type is not generated in dwarf and hence btf,
11551 * trigger an explicit type generation here.
11553 BTF_TYPE_EMIT(struct unix_sock);
11554 if (sk && sk_fullsock(sk) && sk->sk_family == AF_UNIX)
11555 return (unsigned long)sk;
11557 return (unsigned long)NULL;
11560 const struct bpf_func_proto bpf_skc_to_unix_sock_proto = {
11561 .func = bpf_skc_to_unix_sock,
11563 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11564 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11565 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UNIX],
11568 BPF_CALL_1(bpf_skc_to_mptcp_sock, struct sock *, sk)
11570 BTF_TYPE_EMIT(struct mptcp_sock);
11571 return (unsigned long)bpf_mptcp_sock_from_subflow(sk);
11574 const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto = {
11575 .func = bpf_skc_to_mptcp_sock,
11577 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11578 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
11579 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_MPTCP],
11582 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
11584 return (unsigned long)sock_from_file(file);
11587 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
11588 BTF_ID(struct, socket)
11589 BTF_ID(struct, file)
11591 const struct bpf_func_proto bpf_sock_from_file_proto = {
11592 .func = bpf_sock_from_file,
11594 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11595 .ret_btf_id = &bpf_sock_from_file_btf_ids[0],
11596 .arg1_type = ARG_PTR_TO_BTF_ID,
11597 .arg1_btf_id = &bpf_sock_from_file_btf_ids[1],
11600 static const struct bpf_func_proto *
11601 bpf_sk_base_func_proto(enum bpf_func_id func_id)
11603 const struct bpf_func_proto *func;
11606 case BPF_FUNC_skc_to_tcp6_sock:
11607 func = &bpf_skc_to_tcp6_sock_proto;
11609 case BPF_FUNC_skc_to_tcp_sock:
11610 func = &bpf_skc_to_tcp_sock_proto;
11612 case BPF_FUNC_skc_to_tcp_timewait_sock:
11613 func = &bpf_skc_to_tcp_timewait_sock_proto;
11615 case BPF_FUNC_skc_to_tcp_request_sock:
11616 func = &bpf_skc_to_tcp_request_sock_proto;
11618 case BPF_FUNC_skc_to_udp6_sock:
11619 func = &bpf_skc_to_udp6_sock_proto;
11621 case BPF_FUNC_skc_to_unix_sock:
11622 func = &bpf_skc_to_unix_sock_proto;
11624 case BPF_FUNC_skc_to_mptcp_sock:
11625 func = &bpf_skc_to_mptcp_sock_proto;
11627 case BPF_FUNC_ktime_get_coarse_ns:
11628 return &bpf_ktime_get_coarse_ns_proto;
11630 return bpf_base_func_proto(func_id);
11633 if (!perfmon_capable())