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);
2136 /* At ingress, the mac header has already been pulled once.
2137 * At egress, skb_pospull_rcsum has to be done in case that
2138 * the skb is originated from ingress (i.e. a forwarded skb)
2139 * to ensure that rcsum starts at net header.
2141 if (!skb_at_tc_ingress(skb))
2142 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2144 skb_pop_mac_header(skb);
2145 skb_reset_mac_len(skb);
2146 return flags & BPF_F_INGRESS ?
2147 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2150 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2153 /* Verify that a link layer header is carried */
2154 if (unlikely(skb->mac_header >= skb->network_header || skb->len == 0)) {
2159 bpf_push_mac_rcsum(skb);
2160 return flags & BPF_F_INGRESS ?
2161 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2164 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2167 if (dev_is_mac_header_xmit(dev))
2168 return __bpf_redirect_common(skb, dev, flags);
2170 return __bpf_redirect_no_mac(skb, dev, flags);
2173 #if IS_ENABLED(CONFIG_IPV6)
2174 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2175 struct net_device *dev, struct bpf_nh_params *nh)
2177 u32 hh_len = LL_RESERVED_SPACE(dev);
2178 const struct in6_addr *nexthop;
2179 struct dst_entry *dst = NULL;
2180 struct neighbour *neigh;
2182 if (dev_xmit_recursion()) {
2183 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2188 skb_clear_tstamp(skb);
2190 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2191 skb = skb_expand_head(skb, hh_len);
2199 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2200 &ipv6_hdr(skb)->daddr);
2202 nexthop = &nh->ipv6_nh;
2204 neigh = ip_neigh_gw6(dev, nexthop);
2205 if (likely(!IS_ERR(neigh))) {
2208 sock_confirm_neigh(skb, neigh);
2209 dev_xmit_recursion_inc();
2210 ret = neigh_output(neigh, skb, false);
2211 dev_xmit_recursion_dec();
2212 rcu_read_unlock_bh();
2215 rcu_read_unlock_bh();
2217 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2223 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2224 struct bpf_nh_params *nh)
2226 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2227 struct net *net = dev_net(dev);
2228 int err, ret = NET_XMIT_DROP;
2231 struct dst_entry *dst;
2232 struct flowi6 fl6 = {
2233 .flowi6_flags = FLOWI_FLAG_ANYSRC,
2234 .flowi6_mark = skb->mark,
2235 .flowlabel = ip6_flowinfo(ip6h),
2236 .flowi6_oif = dev->ifindex,
2237 .flowi6_proto = ip6h->nexthdr,
2238 .daddr = ip6h->daddr,
2239 .saddr = ip6h->saddr,
2242 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2246 skb_dst_set(skb, dst);
2247 } else if (nh->nh_family != AF_INET6) {
2251 err = bpf_out_neigh_v6(net, skb, dev, nh);
2252 if (unlikely(net_xmit_eval(err)))
2253 dev->stats.tx_errors++;
2255 ret = NET_XMIT_SUCCESS;
2258 dev->stats.tx_errors++;
2264 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2265 struct bpf_nh_params *nh)
2268 return NET_XMIT_DROP;
2270 #endif /* CONFIG_IPV6 */
2272 #if IS_ENABLED(CONFIG_INET)
2273 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2274 struct net_device *dev, struct bpf_nh_params *nh)
2276 u32 hh_len = LL_RESERVED_SPACE(dev);
2277 struct neighbour *neigh;
2278 bool is_v6gw = false;
2280 if (dev_xmit_recursion()) {
2281 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2286 skb_clear_tstamp(skb);
2288 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2289 skb = skb_expand_head(skb, hh_len);
2296 struct dst_entry *dst = skb_dst(skb);
2297 struct rtable *rt = container_of(dst, struct rtable, dst);
2299 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2300 } else if (nh->nh_family == AF_INET6) {
2301 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2303 } else if (nh->nh_family == AF_INET) {
2304 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2306 rcu_read_unlock_bh();
2310 if (likely(!IS_ERR(neigh))) {
2313 sock_confirm_neigh(skb, neigh);
2314 dev_xmit_recursion_inc();
2315 ret = neigh_output(neigh, skb, is_v6gw);
2316 dev_xmit_recursion_dec();
2317 rcu_read_unlock_bh();
2320 rcu_read_unlock_bh();
2326 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2327 struct bpf_nh_params *nh)
2329 const struct iphdr *ip4h = ip_hdr(skb);
2330 struct net *net = dev_net(dev);
2331 int err, ret = NET_XMIT_DROP;
2334 struct flowi4 fl4 = {
2335 .flowi4_flags = FLOWI_FLAG_ANYSRC,
2336 .flowi4_mark = skb->mark,
2337 .flowi4_tos = RT_TOS(ip4h->tos),
2338 .flowi4_oif = dev->ifindex,
2339 .flowi4_proto = ip4h->protocol,
2340 .daddr = ip4h->daddr,
2341 .saddr = ip4h->saddr,
2345 rt = ip_route_output_flow(net, &fl4, NULL);
2348 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2353 skb_dst_set(skb, &rt->dst);
2356 err = bpf_out_neigh_v4(net, skb, dev, nh);
2357 if (unlikely(net_xmit_eval(err)))
2358 dev->stats.tx_errors++;
2360 ret = NET_XMIT_SUCCESS;
2363 dev->stats.tx_errors++;
2369 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2370 struct bpf_nh_params *nh)
2373 return NET_XMIT_DROP;
2375 #endif /* CONFIG_INET */
2377 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2378 struct bpf_nh_params *nh)
2380 struct ethhdr *ethh = eth_hdr(skb);
2382 if (unlikely(skb->mac_header >= skb->network_header))
2384 bpf_push_mac_rcsum(skb);
2385 if (is_multicast_ether_addr(ethh->h_dest))
2388 skb_pull(skb, sizeof(*ethh));
2389 skb_unset_mac_header(skb);
2390 skb_reset_network_header(skb);
2392 if (skb->protocol == htons(ETH_P_IP))
2393 return __bpf_redirect_neigh_v4(skb, dev, nh);
2394 else if (skb->protocol == htons(ETH_P_IPV6))
2395 return __bpf_redirect_neigh_v6(skb, dev, nh);
2401 /* Internal, non-exposed redirect flags. */
2403 BPF_F_NEIGH = (1ULL << 1),
2404 BPF_F_PEER = (1ULL << 2),
2405 BPF_F_NEXTHOP = (1ULL << 3),
2406 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2409 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2411 struct net_device *dev;
2412 struct sk_buff *clone;
2415 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2418 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2422 clone = skb_clone(skb, GFP_ATOMIC);
2423 if (unlikely(!clone))
2426 /* For direct write, we need to keep the invariant that the skbs
2427 * we're dealing with need to be uncloned. Should uncloning fail
2428 * here, we need to free the just generated clone to unclone once
2431 ret = bpf_try_make_head_writable(skb);
2432 if (unlikely(ret)) {
2437 return __bpf_redirect(clone, dev, flags);
2440 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2441 .func = bpf_clone_redirect,
2443 .ret_type = RET_INTEGER,
2444 .arg1_type = ARG_PTR_TO_CTX,
2445 .arg2_type = ARG_ANYTHING,
2446 .arg3_type = ARG_ANYTHING,
2449 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2450 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2452 int skb_do_redirect(struct sk_buff *skb)
2454 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2455 struct net *net = dev_net(skb->dev);
2456 struct net_device *dev;
2457 u32 flags = ri->flags;
2459 dev = dev_get_by_index_rcu(net, ri->tgt_index);
2464 if (flags & BPF_F_PEER) {
2465 const struct net_device_ops *ops = dev->netdev_ops;
2467 if (unlikely(!ops->ndo_get_peer_dev ||
2468 !skb_at_tc_ingress(skb)))
2470 dev = ops->ndo_get_peer_dev(dev);
2471 if (unlikely(!dev ||
2472 !(dev->flags & IFF_UP) ||
2473 net_eq(net, dev_net(dev))))
2478 return flags & BPF_F_NEIGH ?
2479 __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2481 __bpf_redirect(skb, dev, flags);
2487 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2489 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2491 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2495 ri->tgt_index = ifindex;
2497 return TC_ACT_REDIRECT;
2500 static const struct bpf_func_proto bpf_redirect_proto = {
2501 .func = bpf_redirect,
2503 .ret_type = RET_INTEGER,
2504 .arg1_type = ARG_ANYTHING,
2505 .arg2_type = ARG_ANYTHING,
2508 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2510 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2512 if (unlikely(flags))
2515 ri->flags = BPF_F_PEER;
2516 ri->tgt_index = ifindex;
2518 return TC_ACT_REDIRECT;
2521 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2522 .func = bpf_redirect_peer,
2524 .ret_type = RET_INTEGER,
2525 .arg1_type = ARG_ANYTHING,
2526 .arg2_type = ARG_ANYTHING,
2529 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2530 int, plen, u64, flags)
2532 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2534 if (unlikely((plen && plen < sizeof(*params)) || flags))
2537 ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2538 ri->tgt_index = ifindex;
2540 BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2542 memcpy(&ri->nh, params, sizeof(ri->nh));
2544 return TC_ACT_REDIRECT;
2547 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2548 .func = bpf_redirect_neigh,
2550 .ret_type = RET_INTEGER,
2551 .arg1_type = ARG_ANYTHING,
2552 .arg2_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2553 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
2554 .arg4_type = ARG_ANYTHING,
2557 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2559 msg->apply_bytes = bytes;
2563 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2564 .func = bpf_msg_apply_bytes,
2566 .ret_type = RET_INTEGER,
2567 .arg1_type = ARG_PTR_TO_CTX,
2568 .arg2_type = ARG_ANYTHING,
2571 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2573 msg->cork_bytes = bytes;
2577 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2578 .func = bpf_msg_cork_bytes,
2580 .ret_type = RET_INTEGER,
2581 .arg1_type = ARG_PTR_TO_CTX,
2582 .arg2_type = ARG_ANYTHING,
2585 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2586 u32, end, u64, flags)
2588 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2589 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2590 struct scatterlist *sge;
2591 u8 *raw, *to, *from;
2594 if (unlikely(flags || end <= start))
2597 /* First find the starting scatterlist element */
2601 len = sk_msg_elem(msg, i)->length;
2602 if (start < offset + len)
2604 sk_msg_iter_var_next(i);
2605 } while (i != msg->sg.end);
2607 if (unlikely(start >= offset + len))
2611 /* The start may point into the sg element so we need to also
2612 * account for the headroom.
2614 bytes_sg_total = start - offset + bytes;
2615 if (!test_bit(i, msg->sg.copy) && bytes_sg_total <= len)
2618 /* At this point we need to linearize multiple scatterlist
2619 * elements or a single shared page. Either way we need to
2620 * copy into a linear buffer exclusively owned by BPF. Then
2621 * place the buffer in the scatterlist and fixup the original
2622 * entries by removing the entries now in the linear buffer
2623 * and shifting the remaining entries. For now we do not try
2624 * to copy partial entries to avoid complexity of running out
2625 * of sg_entry slots. The downside is reading a single byte
2626 * will copy the entire sg entry.
2629 copy += sk_msg_elem(msg, i)->length;
2630 sk_msg_iter_var_next(i);
2631 if (bytes_sg_total <= copy)
2633 } while (i != msg->sg.end);
2636 if (unlikely(bytes_sg_total > copy))
2639 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2641 if (unlikely(!page))
2644 raw = page_address(page);
2647 sge = sk_msg_elem(msg, i);
2648 from = sg_virt(sge);
2652 memcpy(to, from, len);
2655 put_page(sg_page(sge));
2657 sk_msg_iter_var_next(i);
2658 } while (i != last_sge);
2660 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2662 /* To repair sg ring we need to shift entries. If we only
2663 * had a single entry though we can just replace it and
2664 * be done. Otherwise walk the ring and shift the entries.
2666 WARN_ON_ONCE(last_sge == first_sge);
2667 shift = last_sge > first_sge ?
2668 last_sge - first_sge - 1 :
2669 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2674 sk_msg_iter_var_next(i);
2678 if (i + shift >= NR_MSG_FRAG_IDS)
2679 move_from = i + shift - NR_MSG_FRAG_IDS;
2681 move_from = i + shift;
2682 if (move_from == msg->sg.end)
2685 msg->sg.data[i] = msg->sg.data[move_from];
2686 msg->sg.data[move_from].length = 0;
2687 msg->sg.data[move_from].page_link = 0;
2688 msg->sg.data[move_from].offset = 0;
2689 sk_msg_iter_var_next(i);
2692 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2693 msg->sg.end - shift + NR_MSG_FRAG_IDS :
2694 msg->sg.end - shift;
2696 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2697 msg->data_end = msg->data + bytes;
2701 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2702 .func = bpf_msg_pull_data,
2704 .ret_type = RET_INTEGER,
2705 .arg1_type = ARG_PTR_TO_CTX,
2706 .arg2_type = ARG_ANYTHING,
2707 .arg3_type = ARG_ANYTHING,
2708 .arg4_type = ARG_ANYTHING,
2711 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2712 u32, len, u64, flags)
2714 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2715 u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2716 u8 *raw, *to, *from;
2719 if (unlikely(flags))
2722 if (unlikely(len == 0))
2725 /* First find the starting scatterlist element */
2729 l = sk_msg_elem(msg, i)->length;
2731 if (start < offset + l)
2733 sk_msg_iter_var_next(i);
2734 } while (i != msg->sg.end);
2736 if (start >= offset + l)
2739 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2741 /* If no space available will fallback to copy, we need at
2742 * least one scatterlist elem available to push data into
2743 * when start aligns to the beginning of an element or two
2744 * when it falls inside an element. We handle the start equals
2745 * offset case because its the common case for inserting a
2748 if (!space || (space == 1 && start != offset))
2749 copy = msg->sg.data[i].length;
2751 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2752 get_order(copy + len));
2753 if (unlikely(!page))
2759 raw = page_address(page);
2761 psge = sk_msg_elem(msg, i);
2762 front = start - offset;
2763 back = psge->length - front;
2764 from = sg_virt(psge);
2767 memcpy(raw, from, front);
2771 to = raw + front + len;
2773 memcpy(to, from, back);
2776 put_page(sg_page(psge));
2777 } else if (start - offset) {
2778 psge = sk_msg_elem(msg, i);
2779 rsge = sk_msg_elem_cpy(msg, i);
2781 psge->length = start - offset;
2782 rsge.length -= psge->length;
2783 rsge.offset += start;
2785 sk_msg_iter_var_next(i);
2786 sg_unmark_end(psge);
2787 sg_unmark_end(&rsge);
2788 sk_msg_iter_next(msg, end);
2791 /* Slot(s) to place newly allocated data */
2794 /* Shift one or two slots as needed */
2796 sge = sk_msg_elem_cpy(msg, i);
2798 sk_msg_iter_var_next(i);
2799 sg_unmark_end(&sge);
2800 sk_msg_iter_next(msg, end);
2802 nsge = sk_msg_elem_cpy(msg, i);
2804 sk_msg_iter_var_next(i);
2805 nnsge = sk_msg_elem_cpy(msg, i);
2808 while (i != msg->sg.end) {
2809 msg->sg.data[i] = sge;
2811 sk_msg_iter_var_next(i);
2814 nnsge = sk_msg_elem_cpy(msg, i);
2816 nsge = sk_msg_elem_cpy(msg, i);
2821 /* Place newly allocated data buffer */
2822 sk_mem_charge(msg->sk, len);
2823 msg->sg.size += len;
2824 __clear_bit(new, msg->sg.copy);
2825 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2827 get_page(sg_page(&rsge));
2828 sk_msg_iter_var_next(new);
2829 msg->sg.data[new] = rsge;
2832 sk_msg_compute_data_pointers(msg);
2836 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2837 .func = bpf_msg_push_data,
2839 .ret_type = RET_INTEGER,
2840 .arg1_type = ARG_PTR_TO_CTX,
2841 .arg2_type = ARG_ANYTHING,
2842 .arg3_type = ARG_ANYTHING,
2843 .arg4_type = ARG_ANYTHING,
2846 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2852 sk_msg_iter_var_next(i);
2853 msg->sg.data[prev] = msg->sg.data[i];
2854 } while (i != msg->sg.end);
2856 sk_msg_iter_prev(msg, end);
2859 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2861 struct scatterlist tmp, sge;
2863 sk_msg_iter_next(msg, end);
2864 sge = sk_msg_elem_cpy(msg, i);
2865 sk_msg_iter_var_next(i);
2866 tmp = sk_msg_elem_cpy(msg, i);
2868 while (i != msg->sg.end) {
2869 msg->sg.data[i] = sge;
2870 sk_msg_iter_var_next(i);
2872 tmp = sk_msg_elem_cpy(msg, i);
2876 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2877 u32, len, u64, flags)
2879 u32 i = 0, l = 0, space, offset = 0;
2880 u64 last = start + len;
2883 if (unlikely(flags))
2886 /* First find the starting scatterlist element */
2890 l = sk_msg_elem(msg, i)->length;
2892 if (start < offset + l)
2894 sk_msg_iter_var_next(i);
2895 } while (i != msg->sg.end);
2897 /* Bounds checks: start and pop must be inside message */
2898 if (start >= offset + l || last >= msg->sg.size)
2901 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2904 /* --------------| offset
2905 * -| start |-------- len -------|
2907 * |----- a ----|-------- pop -------|----- b ----|
2908 * |______________________________________________| length
2911 * a: region at front of scatter element to save
2912 * b: region at back of scatter element to save when length > A + pop
2913 * pop: region to pop from element, same as input 'pop' here will be
2914 * decremented below per iteration.
2916 * Two top-level cases to handle when start != offset, first B is non
2917 * zero and second B is zero corresponding to when a pop includes more
2920 * Then if B is non-zero AND there is no space allocate space and
2921 * compact A, B regions into page. If there is space shift ring to
2922 * the rigth free'ing the next element in ring to place B, leaving
2923 * A untouched except to reduce length.
2925 if (start != offset) {
2926 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2928 int b = sge->length - pop - a;
2930 sk_msg_iter_var_next(i);
2932 if (pop < sge->length - a) {
2935 sk_msg_shift_right(msg, i);
2936 nsge = sk_msg_elem(msg, i);
2937 get_page(sg_page(sge));
2940 b, sge->offset + pop + a);
2942 struct page *page, *orig;
2945 page = alloc_pages(__GFP_NOWARN |
2946 __GFP_COMP | GFP_ATOMIC,
2948 if (unlikely(!page))
2952 orig = sg_page(sge);
2953 from = sg_virt(sge);
2954 to = page_address(page);
2955 memcpy(to, from, a);
2956 memcpy(to + a, from + a + pop, b);
2957 sg_set_page(sge, page, a + b, 0);
2961 } else if (pop >= sge->length - a) {
2962 pop -= (sge->length - a);
2967 /* From above the current layout _must_ be as follows,
2972 * |---- pop ---|---------------- b ------------|
2973 * |____________________________________________| length
2975 * Offset and start of the current msg elem are equal because in the
2976 * previous case we handled offset != start and either consumed the
2977 * entire element and advanced to the next element OR pop == 0.
2979 * Two cases to handle here are first pop is less than the length
2980 * leaving some remainder b above. Simply adjust the element's layout
2981 * in this case. Or pop >= length of the element so that b = 0. In this
2982 * case advance to next element decrementing pop.
2985 struct scatterlist *sge = sk_msg_elem(msg, i);
2987 if (pop < sge->length) {
2993 sk_msg_shift_left(msg, i);
2995 sk_msg_iter_var_next(i);
2998 sk_mem_uncharge(msg->sk, len - pop);
2999 msg->sg.size -= (len - pop);
3000 sk_msg_compute_data_pointers(msg);
3004 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
3005 .func = bpf_msg_pop_data,
3007 .ret_type = RET_INTEGER,
3008 .arg1_type = ARG_PTR_TO_CTX,
3009 .arg2_type = ARG_ANYTHING,
3010 .arg3_type = ARG_ANYTHING,
3011 .arg4_type = ARG_ANYTHING,
3014 #ifdef CONFIG_CGROUP_NET_CLASSID
3015 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3017 return __task_get_classid(current);
3020 const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3021 .func = bpf_get_cgroup_classid_curr,
3023 .ret_type = RET_INTEGER,
3026 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3028 struct sock *sk = skb_to_full_sk(skb);
3030 if (!sk || !sk_fullsock(sk))
3033 return sock_cgroup_classid(&sk->sk_cgrp_data);
3036 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3037 .func = bpf_skb_cgroup_classid,
3039 .ret_type = RET_INTEGER,
3040 .arg1_type = ARG_PTR_TO_CTX,
3044 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3046 return task_get_classid(skb);
3049 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3050 .func = bpf_get_cgroup_classid,
3052 .ret_type = RET_INTEGER,
3053 .arg1_type = ARG_PTR_TO_CTX,
3056 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3058 return dst_tclassid(skb);
3061 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3062 .func = bpf_get_route_realm,
3064 .ret_type = RET_INTEGER,
3065 .arg1_type = ARG_PTR_TO_CTX,
3068 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3070 /* If skb_clear_hash() was called due to mangling, we can
3071 * trigger SW recalculation here. Later access to hash
3072 * can then use the inline skb->hash via context directly
3073 * instead of calling this helper again.
3075 return skb_get_hash(skb);
3078 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3079 .func = bpf_get_hash_recalc,
3081 .ret_type = RET_INTEGER,
3082 .arg1_type = ARG_PTR_TO_CTX,
3085 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3087 /* After all direct packet write, this can be used once for
3088 * triggering a lazy recalc on next skb_get_hash() invocation.
3090 skb_clear_hash(skb);
3094 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3095 .func = bpf_set_hash_invalid,
3097 .ret_type = RET_INTEGER,
3098 .arg1_type = ARG_PTR_TO_CTX,
3101 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3103 /* Set user specified hash as L4(+), so that it gets returned
3104 * on skb_get_hash() call unless BPF prog later on triggers a
3107 __skb_set_sw_hash(skb, hash, true);
3111 static const struct bpf_func_proto bpf_set_hash_proto = {
3112 .func = bpf_set_hash,
3114 .ret_type = RET_INTEGER,
3115 .arg1_type = ARG_PTR_TO_CTX,
3116 .arg2_type = ARG_ANYTHING,
3119 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3124 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3125 vlan_proto != htons(ETH_P_8021AD)))
3126 vlan_proto = htons(ETH_P_8021Q);
3128 bpf_push_mac_rcsum(skb);
3129 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3130 bpf_pull_mac_rcsum(skb);
3132 bpf_compute_data_pointers(skb);
3136 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3137 .func = bpf_skb_vlan_push,
3139 .ret_type = RET_INTEGER,
3140 .arg1_type = ARG_PTR_TO_CTX,
3141 .arg2_type = ARG_ANYTHING,
3142 .arg3_type = ARG_ANYTHING,
3145 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3149 bpf_push_mac_rcsum(skb);
3150 ret = skb_vlan_pop(skb);
3151 bpf_pull_mac_rcsum(skb);
3153 bpf_compute_data_pointers(skb);
3157 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3158 .func = bpf_skb_vlan_pop,
3160 .ret_type = RET_INTEGER,
3161 .arg1_type = ARG_PTR_TO_CTX,
3164 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3166 /* Caller already did skb_cow() with len as headroom,
3167 * so no need to do it here.
3170 memmove(skb->data, skb->data + len, off);
3171 memset(skb->data + off, 0, len);
3173 /* No skb_postpush_rcsum(skb, skb->data + off, len)
3174 * needed here as it does not change the skb->csum
3175 * result for checksum complete when summing over
3181 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3183 /* skb_ensure_writable() is not needed here, as we're
3184 * already working on an uncloned skb.
3186 if (unlikely(!pskb_may_pull(skb, off + len)))
3189 skb_postpull_rcsum(skb, skb->data + off, len);
3190 memmove(skb->data + len, skb->data, off);
3191 __skb_pull(skb, len);
3196 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3198 bool trans_same = skb->transport_header == skb->network_header;
3201 /* There's no need for __skb_push()/__skb_pull() pair to
3202 * get to the start of the mac header as we're guaranteed
3203 * to always start from here under eBPF.
3205 ret = bpf_skb_generic_push(skb, off, len);
3207 skb->mac_header -= len;
3208 skb->network_header -= len;
3210 skb->transport_header = skb->network_header;
3216 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3218 bool trans_same = skb->transport_header == skb->network_header;
3221 /* Same here, __skb_push()/__skb_pull() pair not needed. */
3222 ret = bpf_skb_generic_pop(skb, off, len);
3224 skb->mac_header += len;
3225 skb->network_header += len;
3227 skb->transport_header = skb->network_header;
3233 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3235 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3236 u32 off = skb_mac_header_len(skb);
3239 ret = skb_cow(skb, len_diff);
3240 if (unlikely(ret < 0))
3243 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3244 if (unlikely(ret < 0))
3247 if (skb_is_gso(skb)) {
3248 struct skb_shared_info *shinfo = skb_shinfo(skb);
3250 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */
3251 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3252 shinfo->gso_type &= ~SKB_GSO_TCPV4;
3253 shinfo->gso_type |= SKB_GSO_TCPV6;
3257 skb->protocol = htons(ETH_P_IPV6);
3258 skb_clear_hash(skb);
3263 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3265 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3266 u32 off = skb_mac_header_len(skb);
3269 ret = skb_unclone(skb, GFP_ATOMIC);
3270 if (unlikely(ret < 0))
3273 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3274 if (unlikely(ret < 0))
3277 if (skb_is_gso(skb)) {
3278 struct skb_shared_info *shinfo = skb_shinfo(skb);
3280 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */
3281 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3282 shinfo->gso_type &= ~SKB_GSO_TCPV6;
3283 shinfo->gso_type |= SKB_GSO_TCPV4;
3287 skb->protocol = htons(ETH_P_IP);
3288 skb_clear_hash(skb);
3293 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3295 __be16 from_proto = skb->protocol;
3297 if (from_proto == htons(ETH_P_IP) &&
3298 to_proto == htons(ETH_P_IPV6))
3299 return bpf_skb_proto_4_to_6(skb);
3301 if (from_proto == htons(ETH_P_IPV6) &&
3302 to_proto == htons(ETH_P_IP))
3303 return bpf_skb_proto_6_to_4(skb);
3308 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3313 if (unlikely(flags))
3316 /* General idea is that this helper does the basic groundwork
3317 * needed for changing the protocol, and eBPF program fills the
3318 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3319 * and other helpers, rather than passing a raw buffer here.
3321 * The rationale is to keep this minimal and without a need to
3322 * deal with raw packet data. F.e. even if we would pass buffers
3323 * here, the program still needs to call the bpf_lX_csum_replace()
3324 * helpers anyway. Plus, this way we keep also separation of
3325 * concerns, since f.e. bpf_skb_store_bytes() should only take
3328 * Currently, additional options and extension header space are
3329 * not supported, but flags register is reserved so we can adapt
3330 * that. For offloads, we mark packet as dodgy, so that headers
3331 * need to be verified first.
3333 ret = bpf_skb_proto_xlat(skb, proto);
3334 bpf_compute_data_pointers(skb);
3338 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3339 .func = bpf_skb_change_proto,
3341 .ret_type = RET_INTEGER,
3342 .arg1_type = ARG_PTR_TO_CTX,
3343 .arg2_type = ARG_ANYTHING,
3344 .arg3_type = ARG_ANYTHING,
3347 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3349 /* We only allow a restricted subset to be changed for now. */
3350 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3351 !skb_pkt_type_ok(pkt_type)))
3354 skb->pkt_type = pkt_type;
3358 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3359 .func = bpf_skb_change_type,
3361 .ret_type = RET_INTEGER,
3362 .arg1_type = ARG_PTR_TO_CTX,
3363 .arg2_type = ARG_ANYTHING,
3366 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3368 switch (skb->protocol) {
3369 case htons(ETH_P_IP):
3370 return sizeof(struct iphdr);
3371 case htons(ETH_P_IPV6):
3372 return sizeof(struct ipv6hdr);
3378 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3379 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3381 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \
3382 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3383 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3384 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3385 BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \
3386 BPF_F_ADJ_ROOM_ENCAP_L2( \
3387 BPF_ADJ_ROOM_ENCAP_L2_MASK))
3389 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3392 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3393 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3394 u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3395 unsigned int gso_type = SKB_GSO_DODGY;
3398 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3399 /* udp gso_size delineates datagrams, only allow if fixed */
3400 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3401 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3405 ret = skb_cow_head(skb, len_diff);
3406 if (unlikely(ret < 0))
3410 if (skb->protocol != htons(ETH_P_IP) &&
3411 skb->protocol != htons(ETH_P_IPV6))
3414 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3415 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3418 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3419 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3422 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH &&
3423 inner_mac_len < ETH_HLEN)
3426 if (skb->encapsulation)
3429 mac_len = skb->network_header - skb->mac_header;
3430 inner_net = skb->network_header;
3431 if (inner_mac_len > len_diff)
3433 inner_trans = skb->transport_header;
3436 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3437 if (unlikely(ret < 0))
3441 skb->inner_mac_header = inner_net - inner_mac_len;
3442 skb->inner_network_header = inner_net;
3443 skb->inner_transport_header = inner_trans;
3445 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH)
3446 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
3448 skb_set_inner_protocol(skb, skb->protocol);
3450 skb->encapsulation = 1;
3451 skb_set_network_header(skb, mac_len);
3453 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3454 gso_type |= SKB_GSO_UDP_TUNNEL;
3455 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3456 gso_type |= SKB_GSO_GRE;
3457 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3458 gso_type |= SKB_GSO_IPXIP6;
3459 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3460 gso_type |= SKB_GSO_IPXIP4;
3462 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3463 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3464 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3465 sizeof(struct ipv6hdr) :
3466 sizeof(struct iphdr);
3468 skb_set_transport_header(skb, mac_len + nh_len);
3471 /* Match skb->protocol to new outer l3 protocol */
3472 if (skb->protocol == htons(ETH_P_IP) &&
3473 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3474 skb->protocol = htons(ETH_P_IPV6);
3475 else if (skb->protocol == htons(ETH_P_IPV6) &&
3476 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3477 skb->protocol = htons(ETH_P_IP);
3480 if (skb_is_gso(skb)) {
3481 struct skb_shared_info *shinfo = skb_shinfo(skb);
3483 /* Due to header grow, MSS needs to be downgraded. */
3484 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3485 skb_decrease_gso_size(shinfo, len_diff);
3487 /* Header must be checked, and gso_segs recomputed. */
3488 shinfo->gso_type |= gso_type;
3489 shinfo->gso_segs = 0;
3495 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3500 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3501 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3504 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3505 /* udp gso_size delineates datagrams, only allow if fixed */
3506 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3507 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3511 ret = skb_unclone(skb, GFP_ATOMIC);
3512 if (unlikely(ret < 0))
3515 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3516 if (unlikely(ret < 0))
3519 if (skb_is_gso(skb)) {
3520 struct skb_shared_info *shinfo = skb_shinfo(skb);
3522 /* Due to header shrink, MSS can be upgraded. */
3523 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3524 skb_increase_gso_size(shinfo, len_diff);
3526 /* Header must be checked, and gso_segs recomputed. */
3527 shinfo->gso_type |= SKB_GSO_DODGY;
3528 shinfo->gso_segs = 0;
3534 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3536 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3537 u32, mode, u64, flags)
3539 u32 len_diff_abs = abs(len_diff);
3540 bool shrink = len_diff < 0;
3543 if (unlikely(flags || mode))
3545 if (unlikely(len_diff_abs > 0xfffU))
3549 ret = skb_cow(skb, len_diff);
3550 if (unlikely(ret < 0))
3552 __skb_push(skb, len_diff_abs);
3553 memset(skb->data, 0, len_diff_abs);
3555 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3557 __skb_pull(skb, len_diff_abs);
3559 if (tls_sw_has_ctx_rx(skb->sk)) {
3560 struct strp_msg *rxm = strp_msg(skb);
3562 rxm->full_len += len_diff;
3567 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3568 .func = sk_skb_adjust_room,
3570 .ret_type = RET_INTEGER,
3571 .arg1_type = ARG_PTR_TO_CTX,
3572 .arg2_type = ARG_ANYTHING,
3573 .arg3_type = ARG_ANYTHING,
3574 .arg4_type = ARG_ANYTHING,
3577 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3578 u32, mode, u64, flags)
3580 u32 len_cur, len_diff_abs = abs(len_diff);
3581 u32 len_min = bpf_skb_net_base_len(skb);
3582 u32 len_max = BPF_SKB_MAX_LEN;
3583 __be16 proto = skb->protocol;
3584 bool shrink = len_diff < 0;
3588 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3589 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3591 if (unlikely(len_diff_abs > 0xfffU))
3593 if (unlikely(proto != htons(ETH_P_IP) &&
3594 proto != htons(ETH_P_IPV6)))
3597 off = skb_mac_header_len(skb);
3599 case BPF_ADJ_ROOM_NET:
3600 off += bpf_skb_net_base_len(skb);
3602 case BPF_ADJ_ROOM_MAC:
3608 len_cur = skb->len - skb_network_offset(skb);
3609 if ((shrink && (len_diff_abs >= len_cur ||
3610 len_cur - len_diff_abs < len_min)) ||
3611 (!shrink && (skb->len + len_diff_abs > len_max &&
3615 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3616 bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3617 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3618 __skb_reset_checksum_unnecessary(skb);
3620 bpf_compute_data_pointers(skb);
3624 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3625 .func = bpf_skb_adjust_room,
3627 .ret_type = RET_INTEGER,
3628 .arg1_type = ARG_PTR_TO_CTX,
3629 .arg2_type = ARG_ANYTHING,
3630 .arg3_type = ARG_ANYTHING,
3631 .arg4_type = ARG_ANYTHING,
3634 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3636 u32 min_len = skb_network_offset(skb);
3638 if (skb_transport_header_was_set(skb))
3639 min_len = skb_transport_offset(skb);
3640 if (skb->ip_summed == CHECKSUM_PARTIAL)
3641 min_len = skb_checksum_start_offset(skb) +
3642 skb->csum_offset + sizeof(__sum16);
3646 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3648 unsigned int old_len = skb->len;
3651 ret = __skb_grow_rcsum(skb, new_len);
3653 memset(skb->data + old_len, 0, new_len - old_len);
3657 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3659 return __skb_trim_rcsum(skb, new_len);
3662 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3665 u32 max_len = BPF_SKB_MAX_LEN;
3666 u32 min_len = __bpf_skb_min_len(skb);
3669 if (unlikely(flags || new_len > max_len || new_len < min_len))
3671 if (skb->encapsulation)
3674 /* The basic idea of this helper is that it's performing the
3675 * needed work to either grow or trim an skb, and eBPF program
3676 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3677 * bpf_lX_csum_replace() and others rather than passing a raw
3678 * buffer here. This one is a slow path helper and intended
3679 * for replies with control messages.
3681 * Like in bpf_skb_change_proto(), we want to keep this rather
3682 * minimal and without protocol specifics so that we are able
3683 * to separate concerns as in bpf_skb_store_bytes() should only
3684 * be the one responsible for writing buffers.
3686 * It's really expected to be a slow path operation here for
3687 * control message replies, so we're implicitly linearizing,
3688 * uncloning and drop offloads from the skb by this.
3690 ret = __bpf_try_make_writable(skb, skb->len);
3692 if (new_len > skb->len)
3693 ret = bpf_skb_grow_rcsum(skb, new_len);
3694 else if (new_len < skb->len)
3695 ret = bpf_skb_trim_rcsum(skb, new_len);
3696 if (!ret && skb_is_gso(skb))
3702 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3705 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3707 bpf_compute_data_pointers(skb);
3711 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3712 .func = bpf_skb_change_tail,
3714 .ret_type = RET_INTEGER,
3715 .arg1_type = ARG_PTR_TO_CTX,
3716 .arg2_type = ARG_ANYTHING,
3717 .arg3_type = ARG_ANYTHING,
3720 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3723 return __bpf_skb_change_tail(skb, new_len, flags);
3726 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3727 .func = sk_skb_change_tail,
3729 .ret_type = RET_INTEGER,
3730 .arg1_type = ARG_PTR_TO_CTX,
3731 .arg2_type = ARG_ANYTHING,
3732 .arg3_type = ARG_ANYTHING,
3735 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3738 u32 max_len = BPF_SKB_MAX_LEN;
3739 u32 new_len = skb->len + head_room;
3742 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3743 new_len < skb->len))
3746 ret = skb_cow(skb, head_room);
3748 /* Idea for this helper is that we currently only
3749 * allow to expand on mac header. This means that
3750 * skb->protocol network header, etc, stay as is.
3751 * Compared to bpf_skb_change_tail(), we're more
3752 * flexible due to not needing to linearize or
3753 * reset GSO. Intention for this helper is to be
3754 * used by an L3 skb that needs to push mac header
3755 * for redirection into L2 device.
3757 __skb_push(skb, head_room);
3758 memset(skb->data, 0, head_room);
3759 skb_reset_mac_header(skb);
3760 skb_reset_mac_len(skb);
3766 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3769 int ret = __bpf_skb_change_head(skb, head_room, flags);
3771 bpf_compute_data_pointers(skb);
3775 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3776 .func = bpf_skb_change_head,
3778 .ret_type = RET_INTEGER,
3779 .arg1_type = ARG_PTR_TO_CTX,
3780 .arg2_type = ARG_ANYTHING,
3781 .arg3_type = ARG_ANYTHING,
3784 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3787 return __bpf_skb_change_head(skb, head_room, flags);
3790 static const struct bpf_func_proto sk_skb_change_head_proto = {
3791 .func = sk_skb_change_head,
3793 .ret_type = RET_INTEGER,
3794 .arg1_type = ARG_PTR_TO_CTX,
3795 .arg2_type = ARG_ANYTHING,
3796 .arg3_type = ARG_ANYTHING,
3799 BPF_CALL_1(bpf_xdp_get_buff_len, struct xdp_buff*, xdp)
3801 return xdp_get_buff_len(xdp);
3804 static const struct bpf_func_proto bpf_xdp_get_buff_len_proto = {
3805 .func = bpf_xdp_get_buff_len,
3807 .ret_type = RET_INTEGER,
3808 .arg1_type = ARG_PTR_TO_CTX,
3811 BTF_ID_LIST_SINGLE(bpf_xdp_get_buff_len_bpf_ids, struct, xdp_buff)
3813 const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto = {
3814 .func = bpf_xdp_get_buff_len,
3816 .arg1_type = ARG_PTR_TO_BTF_ID,
3817 .arg1_btf_id = &bpf_xdp_get_buff_len_bpf_ids[0],
3820 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3822 return xdp_data_meta_unsupported(xdp) ? 0 :
3823 xdp->data - xdp->data_meta;
3826 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3828 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3829 unsigned long metalen = xdp_get_metalen(xdp);
3830 void *data_start = xdp_frame_end + metalen;
3831 void *data = xdp->data + offset;
3833 if (unlikely(data < data_start ||
3834 data > xdp->data_end - ETH_HLEN))
3838 memmove(xdp->data_meta + offset,
3839 xdp->data_meta, metalen);
3840 xdp->data_meta += offset;
3846 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3847 .func = bpf_xdp_adjust_head,
3849 .ret_type = RET_INTEGER,
3850 .arg1_type = ARG_PTR_TO_CTX,
3851 .arg2_type = ARG_ANYTHING,
3854 static void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off,
3855 void *buf, unsigned long len, bool flush)
3857 unsigned long ptr_len, ptr_off = 0;
3858 skb_frag_t *next_frag, *end_frag;
3859 struct skb_shared_info *sinfo;
3863 if (likely(xdp->data_end - xdp->data >= off + len)) {
3864 src = flush ? buf : xdp->data + off;
3865 dst = flush ? xdp->data + off : buf;
3866 memcpy(dst, src, len);
3870 sinfo = xdp_get_shared_info_from_buff(xdp);
3871 end_frag = &sinfo->frags[sinfo->nr_frags];
3872 next_frag = &sinfo->frags[0];
3874 ptr_len = xdp->data_end - xdp->data;
3875 ptr_buf = xdp->data;
3878 if (off < ptr_off + ptr_len) {
3879 unsigned long copy_off = off - ptr_off;
3880 unsigned long copy_len = min(len, ptr_len - copy_off);
3882 src = flush ? buf : ptr_buf + copy_off;
3883 dst = flush ? ptr_buf + copy_off : buf;
3884 memcpy(dst, src, copy_len);
3891 if (!len || next_frag == end_frag)
3895 ptr_buf = skb_frag_address(next_frag);
3896 ptr_len = skb_frag_size(next_frag);
3901 static void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len)
3903 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
3904 u32 size = xdp->data_end - xdp->data;
3905 void *addr = xdp->data;
3908 if (unlikely(offset > 0xffff || len > 0xffff))
3909 return ERR_PTR(-EFAULT);
3911 if (offset + len > xdp_get_buff_len(xdp))
3912 return ERR_PTR(-EINVAL);
3914 if (offset < size) /* linear area */
3918 for (i = 0; i < sinfo->nr_frags; i++) { /* paged area */
3919 u32 frag_size = skb_frag_size(&sinfo->frags[i]);
3921 if (offset < frag_size) {
3922 addr = skb_frag_address(&sinfo->frags[i]);
3926 offset -= frag_size;
3929 return offset + len <= size ? addr + offset : NULL;
3932 BPF_CALL_4(bpf_xdp_load_bytes, struct xdp_buff *, xdp, u32, offset,
3933 void *, buf, u32, len)
3937 ptr = bpf_xdp_pointer(xdp, offset, len);
3939 return PTR_ERR(ptr);
3942 bpf_xdp_copy_buf(xdp, offset, buf, len, false);
3944 memcpy(buf, ptr, len);
3949 static const struct bpf_func_proto bpf_xdp_load_bytes_proto = {
3950 .func = bpf_xdp_load_bytes,
3952 .ret_type = RET_INTEGER,
3953 .arg1_type = ARG_PTR_TO_CTX,
3954 .arg2_type = ARG_ANYTHING,
3955 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
3956 .arg4_type = ARG_CONST_SIZE,
3959 BPF_CALL_4(bpf_xdp_store_bytes, struct xdp_buff *, xdp, u32, offset,
3960 void *, buf, u32, len)
3964 ptr = bpf_xdp_pointer(xdp, offset, len);
3966 return PTR_ERR(ptr);
3969 bpf_xdp_copy_buf(xdp, offset, buf, len, true);
3971 memcpy(ptr, buf, len);
3976 static const struct bpf_func_proto bpf_xdp_store_bytes_proto = {
3977 .func = bpf_xdp_store_bytes,
3979 .ret_type = RET_INTEGER,
3980 .arg1_type = ARG_PTR_TO_CTX,
3981 .arg2_type = ARG_ANYTHING,
3982 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
3983 .arg4_type = ARG_CONST_SIZE,
3986 static int bpf_xdp_frags_increase_tail(struct xdp_buff *xdp, int offset)
3988 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
3989 skb_frag_t *frag = &sinfo->frags[sinfo->nr_frags - 1];
3990 struct xdp_rxq_info *rxq = xdp->rxq;
3991 unsigned int tailroom;
3993 if (!rxq->frag_size || rxq->frag_size > xdp->frame_sz)
3996 tailroom = rxq->frag_size - skb_frag_size(frag) - skb_frag_off(frag);
3997 if (unlikely(offset > tailroom))
4000 memset(skb_frag_address(frag) + skb_frag_size(frag), 0, offset);
4001 skb_frag_size_add(frag, offset);
4002 sinfo->xdp_frags_size += offset;
4007 static int bpf_xdp_frags_shrink_tail(struct xdp_buff *xdp, int offset)
4009 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
4010 int i, n_frags_free = 0, len_free = 0;
4012 if (unlikely(offset > (int)xdp_get_buff_len(xdp) - ETH_HLEN))
4015 for (i = sinfo->nr_frags - 1; i >= 0 && offset > 0; i--) {
4016 skb_frag_t *frag = &sinfo->frags[i];
4017 int shrink = min_t(int, offset, skb_frag_size(frag));
4022 if (skb_frag_size(frag) == shrink) {
4023 struct page *page = skb_frag_page(frag);
4025 __xdp_return(page_address(page), &xdp->rxq->mem,
4029 skb_frag_size_sub(frag, shrink);
4033 sinfo->nr_frags -= n_frags_free;
4034 sinfo->xdp_frags_size -= len_free;
4036 if (unlikely(!sinfo->nr_frags)) {
4037 xdp_buff_clear_frags_flag(xdp);
4038 xdp->data_end -= offset;
4044 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
4046 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
4047 void *data_end = xdp->data_end + offset;
4049 if (unlikely(xdp_buff_has_frags(xdp))) { /* non-linear xdp buff */
4051 return bpf_xdp_frags_shrink_tail(xdp, -offset);
4053 return bpf_xdp_frags_increase_tail(xdp, offset);
4056 /* Notice that xdp_data_hard_end have reserved some tailroom */
4057 if (unlikely(data_end > data_hard_end))
4060 /* ALL drivers MUST init xdp->frame_sz, chicken check below */
4061 if (unlikely(xdp->frame_sz > PAGE_SIZE)) {
4062 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz);
4066 if (unlikely(data_end < xdp->data + ETH_HLEN))
4069 /* Clear memory area on grow, can contain uninit kernel memory */
4071 memset(xdp->data_end, 0, offset);
4073 xdp->data_end = data_end;
4078 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
4079 .func = bpf_xdp_adjust_tail,
4081 .ret_type = RET_INTEGER,
4082 .arg1_type = ARG_PTR_TO_CTX,
4083 .arg2_type = ARG_ANYTHING,
4086 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
4088 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
4089 void *meta = xdp->data_meta + offset;
4090 unsigned long metalen = xdp->data - meta;
4092 if (xdp_data_meta_unsupported(xdp))
4094 if (unlikely(meta < xdp_frame_end ||
4097 if (unlikely(xdp_metalen_invalid(metalen)))
4100 xdp->data_meta = meta;
4105 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
4106 .func = bpf_xdp_adjust_meta,
4108 .ret_type = RET_INTEGER,
4109 .arg1_type = ARG_PTR_TO_CTX,
4110 .arg2_type = ARG_ANYTHING,
4113 /* XDP_REDIRECT works by a three-step process, implemented in the functions
4116 * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target
4117 * of the redirect and store it (along with some other metadata) in a per-CPU
4118 * struct bpf_redirect_info.
4120 * 2. When the program returns the XDP_REDIRECT return code, the driver will
4121 * call xdp_do_redirect() which will use the information in struct
4122 * bpf_redirect_info to actually enqueue the frame into a map type-specific
4123 * bulk queue structure.
4125 * 3. Before exiting its NAPI poll loop, the driver will call xdp_do_flush(),
4126 * which will flush all the different bulk queues, thus completing the
4129 * Pointers to the map entries will be kept around for this whole sequence of
4130 * steps, protected by RCU. However, there is no top-level rcu_read_lock() in
4131 * the core code; instead, the RCU protection relies on everything happening
4132 * inside a single NAPI poll sequence, which means it's between a pair of calls
4133 * to local_bh_disable()/local_bh_enable().
4135 * The map entries are marked as __rcu and the map code makes sure to
4136 * dereference those pointers with rcu_dereference_check() in a way that works
4137 * for both sections that to hold an rcu_read_lock() and sections that are
4138 * called from NAPI without a separate rcu_read_lock(). The code below does not
4139 * use RCU annotations, but relies on those in the map code.
4141 void xdp_do_flush(void)
4147 EXPORT_SYMBOL_GPL(xdp_do_flush);
4149 void bpf_clear_redirect_map(struct bpf_map *map)
4151 struct bpf_redirect_info *ri;
4154 for_each_possible_cpu(cpu) {
4155 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
4156 /* Avoid polluting remote cacheline due to writes if
4157 * not needed. Once we pass this test, we need the
4158 * cmpxchg() to make sure it hasn't been changed in
4159 * the meantime by remote CPU.
4161 if (unlikely(READ_ONCE(ri->map) == map))
4162 cmpxchg(&ri->map, map, NULL);
4166 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
4167 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key);
4169 u32 xdp_master_redirect(struct xdp_buff *xdp)
4171 struct net_device *master, *slave;
4172 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4174 master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev);
4175 slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp);
4176 if (slave && slave != xdp->rxq->dev) {
4177 /* The target device is different from the receiving device, so
4178 * redirect it to the new device.
4179 * Using XDP_REDIRECT gets the correct behaviour from XDP enabled
4180 * drivers to unmap the packet from their rx ring.
4182 ri->tgt_index = slave->ifindex;
4183 ri->map_id = INT_MAX;
4184 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4185 return XDP_REDIRECT;
4189 EXPORT_SYMBOL_GPL(xdp_master_redirect);
4191 static inline int __xdp_do_redirect_xsk(struct bpf_redirect_info *ri,
4192 struct net_device *dev,
4193 struct xdp_buff *xdp,
4194 struct bpf_prog *xdp_prog)
4196 enum bpf_map_type map_type = ri->map_type;
4197 void *fwd = ri->tgt_value;
4198 u32 map_id = ri->map_id;
4201 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4202 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4204 err = __xsk_map_redirect(fwd, xdp);
4208 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4211 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4215 static __always_inline int __xdp_do_redirect_frame(struct bpf_redirect_info *ri,
4216 struct net_device *dev,
4217 struct xdp_frame *xdpf,
4218 struct bpf_prog *xdp_prog)
4220 enum bpf_map_type map_type = ri->map_type;
4221 void *fwd = ri->tgt_value;
4222 u32 map_id = ri->map_id;
4223 struct bpf_map *map;
4226 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4227 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4229 if (unlikely(!xdpf)) {
4235 case BPF_MAP_TYPE_DEVMAP:
4237 case BPF_MAP_TYPE_DEVMAP_HASH:
4238 map = READ_ONCE(ri->map);
4239 if (unlikely(map)) {
4240 WRITE_ONCE(ri->map, NULL);
4241 err = dev_map_enqueue_multi(xdpf, dev, map,
4242 ri->flags & BPF_F_EXCLUDE_INGRESS);
4244 err = dev_map_enqueue(fwd, xdpf, dev);
4247 case BPF_MAP_TYPE_CPUMAP:
4248 err = cpu_map_enqueue(fwd, xdpf, dev);
4250 case BPF_MAP_TYPE_UNSPEC:
4251 if (map_id == INT_MAX) {
4252 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4253 if (unlikely(!fwd)) {
4257 err = dev_xdp_enqueue(fwd, xdpf, dev);
4268 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4271 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4275 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
4276 struct bpf_prog *xdp_prog)
4278 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4279 enum bpf_map_type map_type = ri->map_type;
4281 /* XDP_REDIRECT is not fully supported yet for xdp frags since
4282 * not all XDP capable drivers can map non-linear xdp_frame in
4285 if (unlikely(xdp_buff_has_frags(xdp) &&
4286 map_type != BPF_MAP_TYPE_CPUMAP))
4289 if (map_type == BPF_MAP_TYPE_XSKMAP)
4290 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4292 return __xdp_do_redirect_frame(ri, dev, xdp_convert_buff_to_frame(xdp),
4295 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4297 int xdp_do_redirect_frame(struct net_device *dev, struct xdp_buff *xdp,
4298 struct xdp_frame *xdpf, struct bpf_prog *xdp_prog)
4300 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4301 enum bpf_map_type map_type = ri->map_type;
4303 if (map_type == BPF_MAP_TYPE_XSKMAP)
4304 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4306 return __xdp_do_redirect_frame(ri, dev, xdpf, xdp_prog);
4308 EXPORT_SYMBOL_GPL(xdp_do_redirect_frame);
4310 static int xdp_do_generic_redirect_map(struct net_device *dev,
4311 struct sk_buff *skb,
4312 struct xdp_buff *xdp,
4313 struct bpf_prog *xdp_prog,
4315 enum bpf_map_type map_type, u32 map_id)
4317 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4318 struct bpf_map *map;
4322 case BPF_MAP_TYPE_DEVMAP:
4324 case BPF_MAP_TYPE_DEVMAP_HASH:
4325 map = READ_ONCE(ri->map);
4326 if (unlikely(map)) {
4327 WRITE_ONCE(ri->map, NULL);
4328 err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4329 ri->flags & BPF_F_EXCLUDE_INGRESS);
4331 err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4336 case BPF_MAP_TYPE_XSKMAP:
4337 err = xsk_generic_rcv(fwd, xdp);
4342 case BPF_MAP_TYPE_CPUMAP:
4343 err = cpu_map_generic_redirect(fwd, skb);
4352 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4355 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4359 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4360 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4362 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4363 enum bpf_map_type map_type = ri->map_type;
4364 void *fwd = ri->tgt_value;
4365 u32 map_id = ri->map_id;
4368 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4369 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4371 if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4372 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4373 if (unlikely(!fwd)) {
4378 err = xdp_ok_fwd_dev(fwd, skb->len);
4383 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4384 generic_xdp_tx(skb, xdp_prog);
4388 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id);
4390 _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4394 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4396 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4398 if (unlikely(flags))
4401 /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4402 * by map_idr) is used for ifindex based XDP redirect.
4404 ri->tgt_index = ifindex;
4405 ri->map_id = INT_MAX;
4406 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4408 return XDP_REDIRECT;
4411 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4412 .func = bpf_xdp_redirect,
4414 .ret_type = RET_INTEGER,
4415 .arg1_type = ARG_ANYTHING,
4416 .arg2_type = ARG_ANYTHING,
4419 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
4422 return map->ops->map_redirect(map, ifindex, flags);
4425 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4426 .func = bpf_xdp_redirect_map,
4428 .ret_type = RET_INTEGER,
4429 .arg1_type = ARG_CONST_MAP_PTR,
4430 .arg2_type = ARG_ANYTHING,
4431 .arg3_type = ARG_ANYTHING,
4434 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4435 unsigned long off, unsigned long len)
4437 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4441 if (ptr != dst_buff)
4442 memcpy(dst_buff, ptr, len);
4447 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4448 u64, flags, void *, meta, u64, meta_size)
4450 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4452 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4454 if (unlikely(!skb || skb_size > skb->len))
4457 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4461 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4462 .func = bpf_skb_event_output,
4464 .ret_type = RET_INTEGER,
4465 .arg1_type = ARG_PTR_TO_CTX,
4466 .arg2_type = ARG_CONST_MAP_PTR,
4467 .arg3_type = ARG_ANYTHING,
4468 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4469 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4472 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4474 const struct bpf_func_proto bpf_skb_output_proto = {
4475 .func = bpf_skb_event_output,
4477 .ret_type = RET_INTEGER,
4478 .arg1_type = ARG_PTR_TO_BTF_ID,
4479 .arg1_btf_id = &bpf_skb_output_btf_ids[0],
4480 .arg2_type = ARG_CONST_MAP_PTR,
4481 .arg3_type = ARG_ANYTHING,
4482 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4483 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4486 static unsigned short bpf_tunnel_key_af(u64 flags)
4488 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4491 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4492 u32, size, u64, flags)
4494 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4495 u8 compat[sizeof(struct bpf_tunnel_key)];
4499 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6 |
4500 BPF_F_TUNINFO_FLAGS)))) {
4504 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4508 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4511 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4512 case offsetof(struct bpf_tunnel_key, tunnel_label):
4513 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4515 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4516 /* Fixup deprecated structure layouts here, so we have
4517 * a common path later on.
4519 if (ip_tunnel_info_af(info) != AF_INET)
4522 to = (struct bpf_tunnel_key *)compat;
4529 to->tunnel_id = be64_to_cpu(info->key.tun_id);
4530 to->tunnel_tos = info->key.tos;
4531 to->tunnel_ttl = info->key.ttl;
4532 if (flags & BPF_F_TUNINFO_FLAGS)
4533 to->tunnel_flags = info->key.tun_flags;
4537 if (flags & BPF_F_TUNINFO_IPV6) {
4538 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4539 sizeof(to->remote_ipv6));
4540 memcpy(to->local_ipv6, &info->key.u.ipv6.dst,
4541 sizeof(to->local_ipv6));
4542 to->tunnel_label = be32_to_cpu(info->key.label);
4544 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4545 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4546 to->local_ipv4 = be32_to_cpu(info->key.u.ipv4.dst);
4547 memset(&to->local_ipv6[1], 0, sizeof(__u32) * 3);
4548 to->tunnel_label = 0;
4551 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4552 memcpy(to_orig, to, size);
4556 memset(to_orig, 0, size);
4560 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4561 .func = bpf_skb_get_tunnel_key,
4563 .ret_type = RET_INTEGER,
4564 .arg1_type = ARG_PTR_TO_CTX,
4565 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4566 .arg3_type = ARG_CONST_SIZE,
4567 .arg4_type = ARG_ANYTHING,
4570 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4572 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4575 if (unlikely(!info ||
4576 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4580 if (unlikely(size < info->options_len)) {
4585 ip_tunnel_info_opts_get(to, info);
4586 if (size > info->options_len)
4587 memset(to + info->options_len, 0, size - info->options_len);
4589 return info->options_len;
4591 memset(to, 0, size);
4595 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4596 .func = bpf_skb_get_tunnel_opt,
4598 .ret_type = RET_INTEGER,
4599 .arg1_type = ARG_PTR_TO_CTX,
4600 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4601 .arg3_type = ARG_CONST_SIZE,
4604 static struct metadata_dst __percpu *md_dst;
4606 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4607 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4609 struct metadata_dst *md = this_cpu_ptr(md_dst);
4610 u8 compat[sizeof(struct bpf_tunnel_key)];
4611 struct ip_tunnel_info *info;
4613 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4614 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
4616 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4618 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4619 case offsetof(struct bpf_tunnel_key, tunnel_label):
4620 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4621 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4622 /* Fixup deprecated structure layouts here, so we have
4623 * a common path later on.
4625 memcpy(compat, from, size);
4626 memset(compat + size, 0, sizeof(compat) - size);
4627 from = (const struct bpf_tunnel_key *) compat;
4633 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4638 dst_hold((struct dst_entry *) md);
4639 skb_dst_set(skb, (struct dst_entry *) md);
4641 info = &md->u.tun_info;
4642 memset(info, 0, sizeof(*info));
4643 info->mode = IP_TUNNEL_INFO_TX;
4645 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4646 if (flags & BPF_F_DONT_FRAGMENT)
4647 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4648 if (flags & BPF_F_ZERO_CSUM_TX)
4649 info->key.tun_flags &= ~TUNNEL_CSUM;
4650 if (flags & BPF_F_SEQ_NUMBER)
4651 info->key.tun_flags |= TUNNEL_SEQ;
4653 info->key.tun_id = cpu_to_be64(from->tunnel_id);
4654 info->key.tos = from->tunnel_tos;
4655 info->key.ttl = from->tunnel_ttl;
4657 if (flags & BPF_F_TUNINFO_IPV6) {
4658 info->mode |= IP_TUNNEL_INFO_IPV6;
4659 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4660 sizeof(from->remote_ipv6));
4661 memcpy(&info->key.u.ipv6.src, from->local_ipv6,
4662 sizeof(from->local_ipv6));
4663 info->key.label = cpu_to_be32(from->tunnel_label) &
4664 IPV6_FLOWLABEL_MASK;
4666 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4667 info->key.u.ipv4.src = cpu_to_be32(from->local_ipv4);
4668 info->key.flow_flags = FLOWI_FLAG_ANYSRC;
4674 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4675 .func = bpf_skb_set_tunnel_key,
4677 .ret_type = RET_INTEGER,
4678 .arg1_type = ARG_PTR_TO_CTX,
4679 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4680 .arg3_type = ARG_CONST_SIZE,
4681 .arg4_type = ARG_ANYTHING,
4684 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4685 const u8 *, from, u32, size)
4687 struct ip_tunnel_info *info = skb_tunnel_info(skb);
4688 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4690 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4692 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4695 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4700 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4701 .func = bpf_skb_set_tunnel_opt,
4703 .ret_type = RET_INTEGER,
4704 .arg1_type = ARG_PTR_TO_CTX,
4705 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4706 .arg3_type = ARG_CONST_SIZE,
4709 static const struct bpf_func_proto *
4710 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4713 struct metadata_dst __percpu *tmp;
4715 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4720 if (cmpxchg(&md_dst, NULL, tmp))
4721 metadata_dst_free_percpu(tmp);
4725 case BPF_FUNC_skb_set_tunnel_key:
4726 return &bpf_skb_set_tunnel_key_proto;
4727 case BPF_FUNC_skb_set_tunnel_opt:
4728 return &bpf_skb_set_tunnel_opt_proto;
4734 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4737 struct bpf_array *array = container_of(map, struct bpf_array, map);
4738 struct cgroup *cgrp;
4741 sk = skb_to_full_sk(skb);
4742 if (!sk || !sk_fullsock(sk))
4744 if (unlikely(idx >= array->map.max_entries))
4747 cgrp = READ_ONCE(array->ptrs[idx]);
4748 if (unlikely(!cgrp))
4751 return sk_under_cgroup_hierarchy(sk, cgrp);
4754 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4755 .func = bpf_skb_under_cgroup,
4757 .ret_type = RET_INTEGER,
4758 .arg1_type = ARG_PTR_TO_CTX,
4759 .arg2_type = ARG_CONST_MAP_PTR,
4760 .arg3_type = ARG_ANYTHING,
4763 #ifdef CONFIG_SOCK_CGROUP_DATA
4764 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4766 struct cgroup *cgrp;
4768 sk = sk_to_full_sk(sk);
4769 if (!sk || !sk_fullsock(sk))
4772 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4773 return cgroup_id(cgrp);
4776 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4778 return __bpf_sk_cgroup_id(skb->sk);
4781 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4782 .func = bpf_skb_cgroup_id,
4784 .ret_type = RET_INTEGER,
4785 .arg1_type = ARG_PTR_TO_CTX,
4788 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4791 struct cgroup *ancestor;
4792 struct cgroup *cgrp;
4794 sk = sk_to_full_sk(sk);
4795 if (!sk || !sk_fullsock(sk))
4798 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4799 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4803 return cgroup_id(ancestor);
4806 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4809 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4812 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4813 .func = bpf_skb_ancestor_cgroup_id,
4815 .ret_type = RET_INTEGER,
4816 .arg1_type = ARG_PTR_TO_CTX,
4817 .arg2_type = ARG_ANYTHING,
4820 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4822 return __bpf_sk_cgroup_id(sk);
4825 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4826 .func = bpf_sk_cgroup_id,
4828 .ret_type = RET_INTEGER,
4829 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4832 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4834 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4837 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4838 .func = bpf_sk_ancestor_cgroup_id,
4840 .ret_type = RET_INTEGER,
4841 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4842 .arg2_type = ARG_ANYTHING,
4846 static unsigned long bpf_xdp_copy(void *dst, const void *ctx,
4847 unsigned long off, unsigned long len)
4849 struct xdp_buff *xdp = (struct xdp_buff *)ctx;
4851 bpf_xdp_copy_buf(xdp, off, dst, len, false);
4855 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4856 u64, flags, void *, meta, u64, meta_size)
4858 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4860 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4863 if (unlikely(!xdp || xdp_size > xdp_get_buff_len(xdp)))
4866 return bpf_event_output(map, flags, meta, meta_size, xdp,
4867 xdp_size, bpf_xdp_copy);
4870 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4871 .func = bpf_xdp_event_output,
4873 .ret_type = RET_INTEGER,
4874 .arg1_type = ARG_PTR_TO_CTX,
4875 .arg2_type = ARG_CONST_MAP_PTR,
4876 .arg3_type = ARG_ANYTHING,
4877 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4878 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4881 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4883 const struct bpf_func_proto bpf_xdp_output_proto = {
4884 .func = bpf_xdp_event_output,
4886 .ret_type = RET_INTEGER,
4887 .arg1_type = ARG_PTR_TO_BTF_ID,
4888 .arg1_btf_id = &bpf_xdp_output_btf_ids[0],
4889 .arg2_type = ARG_CONST_MAP_PTR,
4890 .arg3_type = ARG_ANYTHING,
4891 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4892 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4895 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4897 return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4900 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4901 .func = bpf_get_socket_cookie,
4903 .ret_type = RET_INTEGER,
4904 .arg1_type = ARG_PTR_TO_CTX,
4907 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4909 return __sock_gen_cookie(ctx->sk);
4912 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4913 .func = bpf_get_socket_cookie_sock_addr,
4915 .ret_type = RET_INTEGER,
4916 .arg1_type = ARG_PTR_TO_CTX,
4919 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4921 return __sock_gen_cookie(ctx);
4924 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4925 .func = bpf_get_socket_cookie_sock,
4927 .ret_type = RET_INTEGER,
4928 .arg1_type = ARG_PTR_TO_CTX,
4931 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
4933 return sk ? sock_gen_cookie(sk) : 0;
4936 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
4937 .func = bpf_get_socket_ptr_cookie,
4939 .ret_type = RET_INTEGER,
4940 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4943 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4945 return __sock_gen_cookie(ctx->sk);
4948 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4949 .func = bpf_get_socket_cookie_sock_ops,
4951 .ret_type = RET_INTEGER,
4952 .arg1_type = ARG_PTR_TO_CTX,
4955 static u64 __bpf_get_netns_cookie(struct sock *sk)
4957 const struct net *net = sk ? sock_net(sk) : &init_net;
4959 return net->net_cookie;
4962 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
4964 return __bpf_get_netns_cookie(ctx);
4967 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
4968 .func = bpf_get_netns_cookie_sock,
4970 .ret_type = RET_INTEGER,
4971 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4974 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4976 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4979 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
4980 .func = bpf_get_netns_cookie_sock_addr,
4982 .ret_type = RET_INTEGER,
4983 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4986 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4988 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4991 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = {
4992 .func = bpf_get_netns_cookie_sock_ops,
4994 .ret_type = RET_INTEGER,
4995 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4998 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx)
5000 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5003 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = {
5004 .func = bpf_get_netns_cookie_sk_msg,
5006 .ret_type = RET_INTEGER,
5007 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5010 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
5012 struct sock *sk = sk_to_full_sk(skb->sk);
5015 if (!sk || !sk_fullsock(sk))
5017 kuid = sock_net_uid(sock_net(sk), sk);
5018 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
5021 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
5022 .func = bpf_get_socket_uid,
5024 .ret_type = RET_INTEGER,
5025 .arg1_type = ARG_PTR_TO_CTX,
5028 static int sol_socket_sockopt(struct sock *sk, int optname,
5029 char *optval, int *optlen,
5041 case SO_MAX_PACING_RATE:
5042 case SO_BINDTOIFINDEX:
5044 if (*optlen != sizeof(int))
5047 case SO_BINDTODEVICE:
5054 if (optname == SO_BINDTODEVICE)
5056 return sk_getsockopt(sk, SOL_SOCKET, optname,
5057 KERNEL_SOCKPTR(optval),
5058 KERNEL_SOCKPTR(optlen));
5061 return sk_setsockopt(sk, SOL_SOCKET, optname,
5062 KERNEL_SOCKPTR(optval), *optlen);
5065 static int bpf_sol_tcp_setsockopt(struct sock *sk, int optname,
5066 char *optval, int optlen)
5068 struct tcp_sock *tp = tcp_sk(sk);
5069 unsigned long timeout;
5072 if (optlen != sizeof(int))
5075 val = *(int *)optval;
5077 /* Only some options are supported */
5080 if (val <= 0 || tp->data_segs_out > tp->syn_data)
5082 tcp_snd_cwnd_set(tp, val);
5084 case TCP_BPF_SNDCWND_CLAMP:
5087 tp->snd_cwnd_clamp = val;
5088 tp->snd_ssthresh = val;
5090 case TCP_BPF_DELACK_MAX:
5091 timeout = usecs_to_jiffies(val);
5092 if (timeout > TCP_DELACK_MAX ||
5093 timeout < TCP_TIMEOUT_MIN)
5095 inet_csk(sk)->icsk_delack_max = timeout;
5097 case TCP_BPF_RTO_MIN:
5098 timeout = usecs_to_jiffies(val);
5099 if (timeout > TCP_RTO_MIN ||
5100 timeout < TCP_TIMEOUT_MIN)
5102 inet_csk(sk)->icsk_rto_min = timeout;
5111 static int sol_tcp_sockopt_congestion(struct sock *sk, char *optval,
5112 int *optlen, bool getopt)
5114 struct tcp_sock *tp;
5121 if (!inet_csk(sk)->icsk_ca_ops)
5123 /* BPF expects NULL-terminated tcp-cc string */
5124 optval[--(*optlen)] = '\0';
5125 return do_tcp_getsockopt(sk, SOL_TCP, TCP_CONGESTION,
5126 KERNEL_SOCKPTR(optval),
5127 KERNEL_SOCKPTR(optlen));
5130 /* "cdg" is the only cc that alloc a ptr
5131 * in inet_csk_ca area. The bpf-tcp-cc may
5132 * overwrite this ptr after switching to cdg.
5134 if (*optlen >= sizeof("cdg") - 1 && !strncmp("cdg", optval, *optlen))
5137 /* It stops this looping
5139 * .init => bpf_setsockopt(tcp_cc) => .init =>
5140 * bpf_setsockopt(tcp_cc)" => .init => ....
5142 * The second bpf_setsockopt(tcp_cc) is not allowed
5143 * in order to break the loop when both .init
5144 * are the same bpf prog.
5146 * This applies even the second bpf_setsockopt(tcp_cc)
5147 * does not cause a loop. This limits only the first
5148 * '.init' can call bpf_setsockopt(TCP_CONGESTION) to
5149 * pick a fallback cc (eg. peer does not support ECN)
5150 * and the second '.init' cannot fallback to
5154 if (tp->bpf_chg_cc_inprogress)
5157 tp->bpf_chg_cc_inprogress = 1;
5158 ret = do_tcp_setsockopt(sk, SOL_TCP, TCP_CONGESTION,
5159 KERNEL_SOCKPTR(optval), *optlen);
5160 tp->bpf_chg_cc_inprogress = 0;
5164 static int sol_tcp_sockopt(struct sock *sk, int optname,
5165 char *optval, int *optlen,
5168 if (sk->sk_prot->setsockopt != tcp_setsockopt)
5178 case TCP_WINDOW_CLAMP:
5179 case TCP_THIN_LINEAR_TIMEOUTS:
5180 case TCP_USER_TIMEOUT:
5181 case TCP_NOTSENT_LOWAT:
5183 if (*optlen != sizeof(int))
5186 case TCP_CONGESTION:
5187 return sol_tcp_sockopt_congestion(sk, optval, optlen, getopt);
5195 return bpf_sol_tcp_setsockopt(sk, optname, optval, *optlen);
5199 if (optname == TCP_SAVED_SYN) {
5200 struct tcp_sock *tp = tcp_sk(sk);
5202 if (!tp->saved_syn ||
5203 *optlen > tcp_saved_syn_len(tp->saved_syn))
5205 memcpy(optval, tp->saved_syn->data, *optlen);
5206 /* It cannot free tp->saved_syn here because it
5207 * does not know if the user space still needs it.
5212 return do_tcp_getsockopt(sk, SOL_TCP, optname,
5213 KERNEL_SOCKPTR(optval),
5214 KERNEL_SOCKPTR(optlen));
5217 return do_tcp_setsockopt(sk, SOL_TCP, optname,
5218 KERNEL_SOCKPTR(optval), *optlen);
5221 static int sol_ip_sockopt(struct sock *sk, int optname,
5222 char *optval, int *optlen,
5225 if (sk->sk_family != AF_INET)
5230 if (*optlen != sizeof(int))
5238 return do_ip_getsockopt(sk, SOL_IP, optname,
5239 KERNEL_SOCKPTR(optval),
5240 KERNEL_SOCKPTR(optlen));
5242 return do_ip_setsockopt(sk, SOL_IP, optname,
5243 KERNEL_SOCKPTR(optval), *optlen);
5246 static int sol_ipv6_sockopt(struct sock *sk, int optname,
5247 char *optval, int *optlen,
5250 if (sk->sk_family != AF_INET6)
5255 case IPV6_AUTOFLOWLABEL:
5256 if (*optlen != sizeof(int))
5264 return ipv6_bpf_stub->ipv6_getsockopt(sk, SOL_IPV6, optname,
5265 KERNEL_SOCKPTR(optval),
5266 KERNEL_SOCKPTR(optlen));
5268 return ipv6_bpf_stub->ipv6_setsockopt(sk, SOL_IPV6, optname,
5269 KERNEL_SOCKPTR(optval), *optlen);
5272 static int __bpf_setsockopt(struct sock *sk, int level, int optname,
5273 char *optval, int optlen)
5275 if (!sk_fullsock(sk))
5278 if (level == SOL_SOCKET)
5279 return sol_socket_sockopt(sk, optname, optval, &optlen, false);
5280 else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5281 return sol_ip_sockopt(sk, optname, optval, &optlen, false);
5282 else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5283 return sol_ipv6_sockopt(sk, optname, optval, &optlen, false);
5284 else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5285 return sol_tcp_sockopt(sk, optname, optval, &optlen, false);
5290 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
5291 char *optval, int optlen)
5293 if (sk_fullsock(sk))
5294 sock_owned_by_me(sk);
5295 return __bpf_setsockopt(sk, level, optname, optval, optlen);
5298 static int __bpf_getsockopt(struct sock *sk, int level, int optname,
5299 char *optval, int optlen)
5301 int err, saved_optlen = optlen;
5303 if (!sk_fullsock(sk)) {
5308 if (level == SOL_SOCKET)
5309 err = sol_socket_sockopt(sk, optname, optval, &optlen, true);
5310 else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5311 err = sol_tcp_sockopt(sk, optname, optval, &optlen, true);
5312 else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5313 err = sol_ip_sockopt(sk, optname, optval, &optlen, true);
5314 else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5315 err = sol_ipv6_sockopt(sk, optname, optval, &optlen, true);
5322 if (optlen < saved_optlen)
5323 memset(optval + optlen, 0, saved_optlen - optlen);
5327 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
5328 char *optval, int optlen)
5330 if (sk_fullsock(sk))
5331 sock_owned_by_me(sk);
5332 return __bpf_getsockopt(sk, level, optname, optval, optlen);
5335 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level,
5336 int, optname, char *, optval, int, optlen)
5338 return _bpf_setsockopt(sk, level, optname, optval, optlen);
5341 const struct bpf_func_proto bpf_sk_setsockopt_proto = {
5342 .func = bpf_sk_setsockopt,
5344 .ret_type = RET_INTEGER,
5345 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5346 .arg2_type = ARG_ANYTHING,
5347 .arg3_type = ARG_ANYTHING,
5348 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5349 .arg5_type = ARG_CONST_SIZE,
5352 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level,
5353 int, optname, char *, optval, int, optlen)
5355 return _bpf_getsockopt(sk, level, optname, optval, optlen);
5358 const struct bpf_func_proto bpf_sk_getsockopt_proto = {
5359 .func = bpf_sk_getsockopt,
5361 .ret_type = RET_INTEGER,
5362 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5363 .arg2_type = ARG_ANYTHING,
5364 .arg3_type = ARG_ANYTHING,
5365 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5366 .arg5_type = ARG_CONST_SIZE,
5369 BPF_CALL_5(bpf_unlocked_sk_setsockopt, struct sock *, sk, int, level,
5370 int, optname, char *, optval, int, optlen)
5372 return __bpf_setsockopt(sk, level, optname, optval, optlen);
5375 const struct bpf_func_proto bpf_unlocked_sk_setsockopt_proto = {
5376 .func = bpf_unlocked_sk_setsockopt,
5378 .ret_type = RET_INTEGER,
5379 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5380 .arg2_type = ARG_ANYTHING,
5381 .arg3_type = ARG_ANYTHING,
5382 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5383 .arg5_type = ARG_CONST_SIZE,
5386 BPF_CALL_5(bpf_unlocked_sk_getsockopt, struct sock *, sk, int, level,
5387 int, optname, char *, optval, int, optlen)
5389 return __bpf_getsockopt(sk, level, optname, optval, optlen);
5392 const struct bpf_func_proto bpf_unlocked_sk_getsockopt_proto = {
5393 .func = bpf_unlocked_sk_getsockopt,
5395 .ret_type = RET_INTEGER,
5396 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5397 .arg2_type = ARG_ANYTHING,
5398 .arg3_type = ARG_ANYTHING,
5399 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5400 .arg5_type = ARG_CONST_SIZE,
5403 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5404 int, level, int, optname, char *, optval, int, optlen)
5406 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5409 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5410 .func = bpf_sock_addr_setsockopt,
5412 .ret_type = RET_INTEGER,
5413 .arg1_type = ARG_PTR_TO_CTX,
5414 .arg2_type = ARG_ANYTHING,
5415 .arg3_type = ARG_ANYTHING,
5416 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5417 .arg5_type = ARG_CONST_SIZE,
5420 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5421 int, level, int, optname, char *, optval, int, optlen)
5423 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5426 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5427 .func = bpf_sock_addr_getsockopt,
5429 .ret_type = RET_INTEGER,
5430 .arg1_type = ARG_PTR_TO_CTX,
5431 .arg2_type = ARG_ANYTHING,
5432 .arg3_type = ARG_ANYTHING,
5433 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5434 .arg5_type = ARG_CONST_SIZE,
5437 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5438 int, level, int, optname, char *, optval, int, optlen)
5440 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5443 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5444 .func = bpf_sock_ops_setsockopt,
5446 .ret_type = RET_INTEGER,
5447 .arg1_type = ARG_PTR_TO_CTX,
5448 .arg2_type = ARG_ANYTHING,
5449 .arg3_type = ARG_ANYTHING,
5450 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5451 .arg5_type = ARG_CONST_SIZE,
5454 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5455 int optname, const u8 **start)
5457 struct sk_buff *syn_skb = bpf_sock->syn_skb;
5458 const u8 *hdr_start;
5462 /* sk is a request_sock here */
5464 if (optname == TCP_BPF_SYN) {
5465 hdr_start = syn_skb->data;
5466 ret = tcp_hdrlen(syn_skb);
5467 } else if (optname == TCP_BPF_SYN_IP) {
5468 hdr_start = skb_network_header(syn_skb);
5469 ret = skb_network_header_len(syn_skb) +
5470 tcp_hdrlen(syn_skb);
5472 /* optname == TCP_BPF_SYN_MAC */
5473 hdr_start = skb_mac_header(syn_skb);
5474 ret = skb_mac_header_len(syn_skb) +
5475 skb_network_header_len(syn_skb) +
5476 tcp_hdrlen(syn_skb);
5479 struct sock *sk = bpf_sock->sk;
5480 struct saved_syn *saved_syn;
5482 if (sk->sk_state == TCP_NEW_SYN_RECV)
5483 /* synack retransmit. bpf_sock->syn_skb will
5484 * not be available. It has to resort to
5485 * saved_syn (if it is saved).
5487 saved_syn = inet_reqsk(sk)->saved_syn;
5489 saved_syn = tcp_sk(sk)->saved_syn;
5494 if (optname == TCP_BPF_SYN) {
5495 hdr_start = saved_syn->data +
5496 saved_syn->mac_hdrlen +
5497 saved_syn->network_hdrlen;
5498 ret = saved_syn->tcp_hdrlen;
5499 } else if (optname == TCP_BPF_SYN_IP) {
5500 hdr_start = saved_syn->data +
5501 saved_syn->mac_hdrlen;
5502 ret = saved_syn->network_hdrlen +
5503 saved_syn->tcp_hdrlen;
5505 /* optname == TCP_BPF_SYN_MAC */
5507 /* TCP_SAVE_SYN may not have saved the mac hdr */
5508 if (!saved_syn->mac_hdrlen)
5511 hdr_start = saved_syn->data;
5512 ret = saved_syn->mac_hdrlen +
5513 saved_syn->network_hdrlen +
5514 saved_syn->tcp_hdrlen;
5522 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5523 int, level, int, optname, char *, optval, int, optlen)
5525 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5526 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5527 int ret, copy_len = 0;
5530 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5533 if (optlen < copy_len) {
5538 memcpy(optval, start, copy_len);
5541 /* Zero out unused buffer at the end */
5542 memset(optval + copy_len, 0, optlen - copy_len);
5547 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5550 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5551 .func = bpf_sock_ops_getsockopt,
5553 .ret_type = RET_INTEGER,
5554 .arg1_type = ARG_PTR_TO_CTX,
5555 .arg2_type = ARG_ANYTHING,
5556 .arg3_type = ARG_ANYTHING,
5557 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5558 .arg5_type = ARG_CONST_SIZE,
5561 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5564 struct sock *sk = bpf_sock->sk;
5565 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5567 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5570 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5572 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5575 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5576 .func = bpf_sock_ops_cb_flags_set,
5578 .ret_type = RET_INTEGER,
5579 .arg1_type = ARG_PTR_TO_CTX,
5580 .arg2_type = ARG_ANYTHING,
5583 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5584 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5586 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5590 struct sock *sk = ctx->sk;
5591 u32 flags = BIND_FROM_BPF;
5595 if (addr_len < offsetofend(struct sockaddr, sa_family))
5597 if (addr->sa_family == AF_INET) {
5598 if (addr_len < sizeof(struct sockaddr_in))
5600 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5601 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5602 return __inet_bind(sk, addr, addr_len, flags);
5603 #if IS_ENABLED(CONFIG_IPV6)
5604 } else if (addr->sa_family == AF_INET6) {
5605 if (addr_len < SIN6_LEN_RFC2133)
5607 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5608 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5609 /* ipv6_bpf_stub cannot be NULL, since it's called from
5610 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5612 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5613 #endif /* CONFIG_IPV6 */
5615 #endif /* CONFIG_INET */
5617 return -EAFNOSUPPORT;
5620 static const struct bpf_func_proto bpf_bind_proto = {
5623 .ret_type = RET_INTEGER,
5624 .arg1_type = ARG_PTR_TO_CTX,
5625 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5626 .arg3_type = ARG_CONST_SIZE,
5630 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5631 struct bpf_xfrm_state *, to, u32, size, u64, flags)
5633 const struct sec_path *sp = skb_sec_path(skb);
5634 const struct xfrm_state *x;
5636 if (!sp || unlikely(index >= sp->len || flags))
5639 x = sp->xvec[index];
5641 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5644 to->reqid = x->props.reqid;
5645 to->spi = x->id.spi;
5646 to->family = x->props.family;
5649 if (to->family == AF_INET6) {
5650 memcpy(to->remote_ipv6, x->props.saddr.a6,
5651 sizeof(to->remote_ipv6));
5653 to->remote_ipv4 = x->props.saddr.a4;
5654 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5659 memset(to, 0, size);
5663 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5664 .func = bpf_skb_get_xfrm_state,
5666 .ret_type = RET_INTEGER,
5667 .arg1_type = ARG_PTR_TO_CTX,
5668 .arg2_type = ARG_ANYTHING,
5669 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
5670 .arg4_type = ARG_CONST_SIZE,
5671 .arg5_type = ARG_ANYTHING,
5675 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5676 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
5677 const struct neighbour *neigh,
5678 const struct net_device *dev, u32 mtu)
5680 memcpy(params->dmac, neigh->ha, ETH_ALEN);
5681 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5682 params->h_vlan_TCI = 0;
5683 params->h_vlan_proto = 0;
5685 params->mtu_result = mtu; /* union with tot_len */
5691 #if IS_ENABLED(CONFIG_INET)
5692 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5693 u32 flags, bool check_mtu)
5695 struct fib_nh_common *nhc;
5696 struct in_device *in_dev;
5697 struct neighbour *neigh;
5698 struct net_device *dev;
5699 struct fib_result res;
5704 dev = dev_get_by_index_rcu(net, params->ifindex);
5708 /* verify forwarding is enabled on this interface */
5709 in_dev = __in_dev_get_rcu(dev);
5710 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5711 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5713 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5715 fl4.flowi4_oif = params->ifindex;
5717 fl4.flowi4_iif = params->ifindex;
5720 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5721 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5722 fl4.flowi4_flags = 0;
5724 fl4.flowi4_proto = params->l4_protocol;
5725 fl4.daddr = params->ipv4_dst;
5726 fl4.saddr = params->ipv4_src;
5727 fl4.fl4_sport = params->sport;
5728 fl4.fl4_dport = params->dport;
5729 fl4.flowi4_multipath_hash = 0;
5731 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5732 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5733 struct fib_table *tb;
5735 tb = fib_get_table(net, tbid);
5737 return BPF_FIB_LKUP_RET_NOT_FWDED;
5739 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5741 fl4.flowi4_mark = 0;
5742 fl4.flowi4_secid = 0;
5743 fl4.flowi4_tun_key.tun_id = 0;
5744 fl4.flowi4_uid = sock_net_uid(net, NULL);
5746 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5750 /* map fib lookup errors to RTN_ type */
5752 return BPF_FIB_LKUP_RET_BLACKHOLE;
5753 if (err == -EHOSTUNREACH)
5754 return BPF_FIB_LKUP_RET_UNREACHABLE;
5756 return BPF_FIB_LKUP_RET_PROHIBIT;
5758 return BPF_FIB_LKUP_RET_NOT_FWDED;
5761 if (res.type != RTN_UNICAST)
5762 return BPF_FIB_LKUP_RET_NOT_FWDED;
5764 if (fib_info_num_path(res.fi) > 1)
5765 fib_select_path(net, &res, &fl4, NULL);
5768 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5769 if (params->tot_len > mtu) {
5770 params->mtu_result = mtu; /* union with tot_len */
5771 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5777 /* do not handle lwt encaps right now */
5778 if (nhc->nhc_lwtstate)
5779 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5783 params->rt_metric = res.fi->fib_priority;
5784 params->ifindex = dev->ifindex;
5786 /* xdp and cls_bpf programs are run in RCU-bh so
5787 * rcu_read_lock_bh is not needed here
5789 if (likely(nhc->nhc_gw_family != AF_INET6)) {
5790 if (nhc->nhc_gw_family)
5791 params->ipv4_dst = nhc->nhc_gw.ipv4;
5793 neigh = __ipv4_neigh_lookup_noref(dev,
5794 (__force u32)params->ipv4_dst);
5796 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5798 params->family = AF_INET6;
5799 *dst = nhc->nhc_gw.ipv6;
5800 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5804 return BPF_FIB_LKUP_RET_NO_NEIGH;
5806 return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5810 #if IS_ENABLED(CONFIG_IPV6)
5811 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5812 u32 flags, bool check_mtu)
5814 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5815 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5816 struct fib6_result res = {};
5817 struct neighbour *neigh;
5818 struct net_device *dev;
5819 struct inet6_dev *idev;
5825 /* link local addresses are never forwarded */
5826 if (rt6_need_strict(dst) || rt6_need_strict(src))
5827 return BPF_FIB_LKUP_RET_NOT_FWDED;
5829 dev = dev_get_by_index_rcu(net, params->ifindex);
5833 idev = __in6_dev_get_safely(dev);
5834 if (unlikely(!idev || !idev->cnf.forwarding))
5835 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5837 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5839 oif = fl6.flowi6_oif = params->ifindex;
5841 oif = fl6.flowi6_iif = params->ifindex;
5843 strict = RT6_LOOKUP_F_HAS_SADDR;
5845 fl6.flowlabel = params->flowinfo;
5846 fl6.flowi6_scope = 0;
5847 fl6.flowi6_flags = 0;
5850 fl6.flowi6_proto = params->l4_protocol;
5853 fl6.fl6_sport = params->sport;
5854 fl6.fl6_dport = params->dport;
5856 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5857 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5858 struct fib6_table *tb;
5860 tb = ipv6_stub->fib6_get_table(net, tbid);
5862 return BPF_FIB_LKUP_RET_NOT_FWDED;
5864 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5867 fl6.flowi6_mark = 0;
5868 fl6.flowi6_secid = 0;
5869 fl6.flowi6_tun_key.tun_id = 0;
5870 fl6.flowi6_uid = sock_net_uid(net, NULL);
5872 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5875 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5876 res.f6i == net->ipv6.fib6_null_entry))
5877 return BPF_FIB_LKUP_RET_NOT_FWDED;
5879 switch (res.fib6_type) {
5880 /* only unicast is forwarded */
5884 return BPF_FIB_LKUP_RET_BLACKHOLE;
5885 case RTN_UNREACHABLE:
5886 return BPF_FIB_LKUP_RET_UNREACHABLE;
5888 return BPF_FIB_LKUP_RET_PROHIBIT;
5890 return BPF_FIB_LKUP_RET_NOT_FWDED;
5893 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5894 fl6.flowi6_oif != 0, NULL, strict);
5897 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5898 if (params->tot_len > mtu) {
5899 params->mtu_result = mtu; /* union with tot_len */
5900 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5904 if (res.nh->fib_nh_lws)
5905 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5907 if (res.nh->fib_nh_gw_family)
5908 *dst = res.nh->fib_nh_gw6;
5910 dev = res.nh->fib_nh_dev;
5911 params->rt_metric = res.f6i->fib6_metric;
5912 params->ifindex = dev->ifindex;
5914 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
5917 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5919 return BPF_FIB_LKUP_RET_NO_NEIGH;
5921 return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5925 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
5926 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5928 if (plen < sizeof(*params))
5931 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5934 switch (params->family) {
5935 #if IS_ENABLED(CONFIG_INET)
5937 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
5940 #if IS_ENABLED(CONFIG_IPV6)
5942 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
5946 return -EAFNOSUPPORT;
5949 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
5950 .func = bpf_xdp_fib_lookup,
5952 .ret_type = RET_INTEGER,
5953 .arg1_type = ARG_PTR_TO_CTX,
5954 .arg2_type = ARG_PTR_TO_MEM,
5955 .arg3_type = ARG_CONST_SIZE,
5956 .arg4_type = ARG_ANYTHING,
5959 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
5960 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5962 struct net *net = dev_net(skb->dev);
5963 int rc = -EAFNOSUPPORT;
5964 bool check_mtu = false;
5966 if (plen < sizeof(*params))
5969 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5972 if (params->tot_len)
5975 switch (params->family) {
5976 #if IS_ENABLED(CONFIG_INET)
5978 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
5981 #if IS_ENABLED(CONFIG_IPV6)
5983 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
5988 if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
5989 struct net_device *dev;
5991 /* When tot_len isn't provided by user, check skb
5992 * against MTU of FIB lookup resulting net_device
5994 dev = dev_get_by_index_rcu(net, params->ifindex);
5995 if (!is_skb_forwardable(dev, skb))
5996 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
5998 params->mtu_result = dev->mtu; /* union with tot_len */
6004 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
6005 .func = bpf_skb_fib_lookup,
6007 .ret_type = RET_INTEGER,
6008 .arg1_type = ARG_PTR_TO_CTX,
6009 .arg2_type = ARG_PTR_TO_MEM,
6010 .arg3_type = ARG_CONST_SIZE,
6011 .arg4_type = ARG_ANYTHING,
6014 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
6017 struct net *netns = dev_net(dev_curr);
6019 /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
6023 return dev_get_by_index_rcu(netns, ifindex);
6026 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
6027 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6029 int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6030 struct net_device *dev = skb->dev;
6031 int skb_len, dev_len;
6034 if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
6037 if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
6040 dev = __dev_via_ifindex(dev, ifindex);
6044 mtu = READ_ONCE(dev->mtu);
6046 dev_len = mtu + dev->hard_header_len;
6048 /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6049 skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
6051 skb_len += len_diff; /* minus result pass check */
6052 if (skb_len <= dev_len) {
6053 ret = BPF_MTU_CHK_RET_SUCCESS;
6056 /* At this point, skb->len exceed MTU, but as it include length of all
6057 * segments, it can still be below MTU. The SKB can possibly get
6058 * re-segmented in transmit path (see validate_xmit_skb). Thus, user
6059 * must choose if segs are to be MTU checked.
6061 if (skb_is_gso(skb)) {
6062 ret = BPF_MTU_CHK_RET_SUCCESS;
6064 if (flags & BPF_MTU_CHK_SEGS &&
6065 !skb_gso_validate_network_len(skb, mtu))
6066 ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
6069 /* BPF verifier guarantees valid pointer */
6075 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
6076 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6078 struct net_device *dev = xdp->rxq->dev;
6079 int xdp_len = xdp->data_end - xdp->data;
6080 int ret = BPF_MTU_CHK_RET_SUCCESS;
6083 /* XDP variant doesn't support multi-buffer segment check (yet) */
6084 if (unlikely(flags))
6087 dev = __dev_via_ifindex(dev, ifindex);
6091 mtu = READ_ONCE(dev->mtu);
6093 /* Add L2-header as dev MTU is L3 size */
6094 dev_len = mtu + dev->hard_header_len;
6096 /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6098 xdp_len = *mtu_len + dev->hard_header_len;
6100 xdp_len += len_diff; /* minus result pass check */
6101 if (xdp_len > dev_len)
6102 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6104 /* BPF verifier guarantees valid pointer */
6110 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
6111 .func = bpf_skb_check_mtu,
6113 .ret_type = RET_INTEGER,
6114 .arg1_type = ARG_PTR_TO_CTX,
6115 .arg2_type = ARG_ANYTHING,
6116 .arg3_type = ARG_PTR_TO_INT,
6117 .arg4_type = ARG_ANYTHING,
6118 .arg5_type = ARG_ANYTHING,
6121 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
6122 .func = bpf_xdp_check_mtu,
6124 .ret_type = RET_INTEGER,
6125 .arg1_type = ARG_PTR_TO_CTX,
6126 .arg2_type = ARG_ANYTHING,
6127 .arg3_type = ARG_PTR_TO_INT,
6128 .arg4_type = ARG_ANYTHING,
6129 .arg5_type = ARG_ANYTHING,
6132 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6133 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
6136 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
6138 if (!seg6_validate_srh(srh, len, false))
6142 case BPF_LWT_ENCAP_SEG6_INLINE:
6143 if (skb->protocol != htons(ETH_P_IPV6))
6146 err = seg6_do_srh_inline(skb, srh);
6148 case BPF_LWT_ENCAP_SEG6:
6149 skb_reset_inner_headers(skb);
6150 skb->encapsulation = 1;
6151 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
6157 bpf_compute_data_pointers(skb);
6161 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
6163 return seg6_lookup_nexthop(skb, NULL, 0);
6165 #endif /* CONFIG_IPV6_SEG6_BPF */
6167 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6168 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
6171 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
6175 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
6179 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6180 case BPF_LWT_ENCAP_SEG6:
6181 case BPF_LWT_ENCAP_SEG6_INLINE:
6182 return bpf_push_seg6_encap(skb, type, hdr, len);
6184 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6185 case BPF_LWT_ENCAP_IP:
6186 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
6193 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
6194 void *, hdr, u32, len)
6197 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6198 case BPF_LWT_ENCAP_IP:
6199 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
6206 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
6207 .func = bpf_lwt_in_push_encap,
6209 .ret_type = RET_INTEGER,
6210 .arg1_type = ARG_PTR_TO_CTX,
6211 .arg2_type = ARG_ANYTHING,
6212 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6213 .arg4_type = ARG_CONST_SIZE
6216 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
6217 .func = bpf_lwt_xmit_push_encap,
6219 .ret_type = RET_INTEGER,
6220 .arg1_type = ARG_PTR_TO_CTX,
6221 .arg2_type = ARG_ANYTHING,
6222 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6223 .arg4_type = ARG_CONST_SIZE
6226 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6227 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
6228 const void *, from, u32, len)
6230 struct seg6_bpf_srh_state *srh_state =
6231 this_cpu_ptr(&seg6_bpf_srh_states);
6232 struct ipv6_sr_hdr *srh = srh_state->srh;
6233 void *srh_tlvs, *srh_end, *ptr;
6239 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
6240 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
6242 ptr = skb->data + offset;
6243 if (ptr >= srh_tlvs && ptr + len <= srh_end)
6244 srh_state->valid = false;
6245 else if (ptr < (void *)&srh->flags ||
6246 ptr + len > (void *)&srh->segments)
6249 if (unlikely(bpf_try_make_writable(skb, offset + len)))
6251 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6253 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6255 memcpy(skb->data + offset, from, len);
6259 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
6260 .func = bpf_lwt_seg6_store_bytes,
6262 .ret_type = RET_INTEGER,
6263 .arg1_type = ARG_PTR_TO_CTX,
6264 .arg2_type = ARG_ANYTHING,
6265 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6266 .arg4_type = ARG_CONST_SIZE
6269 static void bpf_update_srh_state(struct sk_buff *skb)
6271 struct seg6_bpf_srh_state *srh_state =
6272 this_cpu_ptr(&seg6_bpf_srh_states);
6275 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
6276 srh_state->srh = NULL;
6278 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6279 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
6280 srh_state->valid = true;
6284 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
6285 u32, action, void *, param, u32, param_len)
6287 struct seg6_bpf_srh_state *srh_state =
6288 this_cpu_ptr(&seg6_bpf_srh_states);
6293 case SEG6_LOCAL_ACTION_END_X:
6294 if (!seg6_bpf_has_valid_srh(skb))
6296 if (param_len != sizeof(struct in6_addr))
6298 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
6299 case SEG6_LOCAL_ACTION_END_T:
6300 if (!seg6_bpf_has_valid_srh(skb))
6302 if (param_len != sizeof(int))
6304 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6305 case SEG6_LOCAL_ACTION_END_DT6:
6306 if (!seg6_bpf_has_valid_srh(skb))
6308 if (param_len != sizeof(int))
6311 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
6313 if (!pskb_pull(skb, hdroff))
6316 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
6317 skb_reset_network_header(skb);
6318 skb_reset_transport_header(skb);
6319 skb->encapsulation = 0;
6321 bpf_compute_data_pointers(skb);
6322 bpf_update_srh_state(skb);
6323 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6324 case SEG6_LOCAL_ACTION_END_B6:
6325 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6327 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
6330 bpf_update_srh_state(skb);
6333 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
6334 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6336 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
6339 bpf_update_srh_state(skb);
6347 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
6348 .func = bpf_lwt_seg6_action,
6350 .ret_type = RET_INTEGER,
6351 .arg1_type = ARG_PTR_TO_CTX,
6352 .arg2_type = ARG_ANYTHING,
6353 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6354 .arg4_type = ARG_CONST_SIZE
6357 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
6360 struct seg6_bpf_srh_state *srh_state =
6361 this_cpu_ptr(&seg6_bpf_srh_states);
6362 struct ipv6_sr_hdr *srh = srh_state->srh;
6363 void *srh_end, *srh_tlvs, *ptr;
6364 struct ipv6hdr *hdr;
6368 if (unlikely(srh == NULL))
6371 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
6372 ((srh->first_segment + 1) << 4));
6373 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
6375 ptr = skb->data + offset;
6377 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
6379 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
6383 ret = skb_cow_head(skb, len);
6384 if (unlikely(ret < 0))
6387 ret = bpf_skb_net_hdr_push(skb, offset, len);
6389 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6392 bpf_compute_data_pointers(skb);
6393 if (unlikely(ret < 0))
6396 hdr = (struct ipv6hdr *)skb->data;
6397 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6399 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6401 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6402 srh_state->hdrlen += len;
6403 srh_state->valid = false;
6407 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6408 .func = bpf_lwt_seg6_adjust_srh,
6410 .ret_type = RET_INTEGER,
6411 .arg1_type = ARG_PTR_TO_CTX,
6412 .arg2_type = ARG_ANYTHING,
6413 .arg3_type = ARG_ANYTHING,
6415 #endif /* CONFIG_IPV6_SEG6_BPF */
6418 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6419 int dif, int sdif, u8 family, u8 proto)
6421 struct inet_hashinfo *hinfo = net->ipv4.tcp_death_row.hashinfo;
6422 bool refcounted = false;
6423 struct sock *sk = NULL;
6425 if (family == AF_INET) {
6426 __be32 src4 = tuple->ipv4.saddr;
6427 __be32 dst4 = tuple->ipv4.daddr;
6429 if (proto == IPPROTO_TCP)
6430 sk = __inet_lookup(net, hinfo, NULL, 0,
6431 src4, tuple->ipv4.sport,
6432 dst4, tuple->ipv4.dport,
6433 dif, sdif, &refcounted);
6435 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6436 dst4, tuple->ipv4.dport,
6437 dif, sdif, &udp_table, NULL);
6438 #if IS_ENABLED(CONFIG_IPV6)
6440 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6441 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6443 if (proto == IPPROTO_TCP)
6444 sk = __inet6_lookup(net, hinfo, NULL, 0,
6445 src6, tuple->ipv6.sport,
6446 dst6, ntohs(tuple->ipv6.dport),
6447 dif, sdif, &refcounted);
6448 else if (likely(ipv6_bpf_stub))
6449 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6450 src6, tuple->ipv6.sport,
6451 dst6, tuple->ipv6.dport,
6457 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6458 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6464 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6465 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6467 static struct sock *
6468 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6469 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6472 struct sock *sk = NULL;
6477 if (len == sizeof(tuple->ipv4))
6479 else if (len == sizeof(tuple->ipv6))
6484 if (unlikely(flags || !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6487 if (family == AF_INET)
6488 sdif = inet_sdif(skb);
6490 sdif = inet6_sdif(skb);
6492 if ((s32)netns_id < 0) {
6494 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6496 net = get_net_ns_by_id(caller_net, netns_id);
6499 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6507 static struct sock *
6508 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6509 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6512 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6513 ifindex, proto, netns_id, flags);
6516 struct sock *sk2 = sk_to_full_sk(sk);
6518 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6519 * sock refcnt is decremented to prevent a request_sock leak.
6521 if (!sk_fullsock(sk2))
6525 /* Ensure there is no need to bump sk2 refcnt */
6526 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6527 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6537 static struct sock *
6538 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6539 u8 proto, u64 netns_id, u64 flags)
6541 struct net *caller_net;
6545 caller_net = dev_net(skb->dev);
6546 ifindex = skb->dev->ifindex;
6548 caller_net = sock_net(skb->sk);
6552 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6556 static struct sock *
6557 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6558 u8 proto, u64 netns_id, u64 flags)
6560 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6564 struct sock *sk2 = sk_to_full_sk(sk);
6566 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6567 * sock refcnt is decremented to prevent a request_sock leak.
6569 if (!sk_fullsock(sk2))
6573 /* Ensure there is no need to bump sk2 refcnt */
6574 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6575 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6585 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6586 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6588 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6592 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6593 .func = bpf_skc_lookup_tcp,
6596 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6597 .arg1_type = ARG_PTR_TO_CTX,
6598 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6599 .arg3_type = ARG_CONST_SIZE,
6600 .arg4_type = ARG_ANYTHING,
6601 .arg5_type = ARG_ANYTHING,
6604 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6605 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6607 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6611 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6612 .func = bpf_sk_lookup_tcp,
6615 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6616 .arg1_type = ARG_PTR_TO_CTX,
6617 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6618 .arg3_type = ARG_CONST_SIZE,
6619 .arg4_type = ARG_ANYTHING,
6620 .arg5_type = ARG_ANYTHING,
6623 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6624 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6626 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6630 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6631 .func = bpf_sk_lookup_udp,
6634 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6635 .arg1_type = ARG_PTR_TO_CTX,
6636 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6637 .arg3_type = ARG_CONST_SIZE,
6638 .arg4_type = ARG_ANYTHING,
6639 .arg5_type = ARG_ANYTHING,
6642 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6644 if (sk && sk_is_refcounted(sk))
6649 static const struct bpf_func_proto bpf_sk_release_proto = {
6650 .func = bpf_sk_release,
6652 .ret_type = RET_INTEGER,
6653 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON | OBJ_RELEASE,
6656 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6657 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6659 struct net *caller_net = dev_net(ctx->rxq->dev);
6660 int ifindex = ctx->rxq->dev->ifindex;
6662 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6663 ifindex, IPPROTO_UDP, netns_id,
6667 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6668 .func = bpf_xdp_sk_lookup_udp,
6671 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6672 .arg1_type = ARG_PTR_TO_CTX,
6673 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6674 .arg3_type = ARG_CONST_SIZE,
6675 .arg4_type = ARG_ANYTHING,
6676 .arg5_type = ARG_ANYTHING,
6679 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6680 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6682 struct net *caller_net = dev_net(ctx->rxq->dev);
6683 int ifindex = ctx->rxq->dev->ifindex;
6685 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6686 ifindex, IPPROTO_TCP, netns_id,
6690 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6691 .func = bpf_xdp_skc_lookup_tcp,
6694 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6695 .arg1_type = ARG_PTR_TO_CTX,
6696 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6697 .arg3_type = ARG_CONST_SIZE,
6698 .arg4_type = ARG_ANYTHING,
6699 .arg5_type = ARG_ANYTHING,
6702 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6703 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6705 struct net *caller_net = dev_net(ctx->rxq->dev);
6706 int ifindex = ctx->rxq->dev->ifindex;
6708 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6709 ifindex, IPPROTO_TCP, netns_id,
6713 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6714 .func = bpf_xdp_sk_lookup_tcp,
6717 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6718 .arg1_type = ARG_PTR_TO_CTX,
6719 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6720 .arg3_type = ARG_CONST_SIZE,
6721 .arg4_type = ARG_ANYTHING,
6722 .arg5_type = ARG_ANYTHING,
6725 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6726 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6728 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6729 sock_net(ctx->sk), 0,
6730 IPPROTO_TCP, netns_id, flags);
6733 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6734 .func = bpf_sock_addr_skc_lookup_tcp,
6736 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6737 .arg1_type = ARG_PTR_TO_CTX,
6738 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6739 .arg3_type = ARG_CONST_SIZE,
6740 .arg4_type = ARG_ANYTHING,
6741 .arg5_type = ARG_ANYTHING,
6744 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6745 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6747 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6748 sock_net(ctx->sk), 0, IPPROTO_TCP,
6752 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6753 .func = bpf_sock_addr_sk_lookup_tcp,
6755 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6756 .arg1_type = ARG_PTR_TO_CTX,
6757 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6758 .arg3_type = ARG_CONST_SIZE,
6759 .arg4_type = ARG_ANYTHING,
6760 .arg5_type = ARG_ANYTHING,
6763 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6764 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6766 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6767 sock_net(ctx->sk), 0, IPPROTO_UDP,
6771 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6772 .func = bpf_sock_addr_sk_lookup_udp,
6774 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6775 .arg1_type = ARG_PTR_TO_CTX,
6776 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6777 .arg3_type = ARG_CONST_SIZE,
6778 .arg4_type = ARG_ANYTHING,
6779 .arg5_type = ARG_ANYTHING,
6782 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6783 struct bpf_insn_access_aux *info)
6785 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6789 if (off % size != 0)
6793 case offsetof(struct bpf_tcp_sock, bytes_received):
6794 case offsetof(struct bpf_tcp_sock, bytes_acked):
6795 return size == sizeof(__u64);
6797 return size == sizeof(__u32);
6801 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6802 const struct bpf_insn *si,
6803 struct bpf_insn *insn_buf,
6804 struct bpf_prog *prog, u32 *target_size)
6806 struct bpf_insn *insn = insn_buf;
6808 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \
6810 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \
6811 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6812 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6813 si->dst_reg, si->src_reg, \
6814 offsetof(struct tcp_sock, FIELD)); \
6817 #define BPF_INET_SOCK_GET_COMMON(FIELD) \
6819 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \
6821 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6822 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6823 struct inet_connection_sock, \
6825 si->dst_reg, si->src_reg, \
6827 struct inet_connection_sock, \
6831 if (insn > insn_buf)
6832 return insn - insn_buf;
6835 case offsetof(struct bpf_tcp_sock, rtt_min):
6836 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
6837 sizeof(struct minmax));
6838 BUILD_BUG_ON(sizeof(struct minmax) <
6839 sizeof(struct minmax_sample));
6841 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6842 offsetof(struct tcp_sock, rtt_min) +
6843 offsetof(struct minmax_sample, v));
6845 case offsetof(struct bpf_tcp_sock, snd_cwnd):
6846 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
6848 case offsetof(struct bpf_tcp_sock, srtt_us):
6849 BPF_TCP_SOCK_GET_COMMON(srtt_us);
6851 case offsetof(struct bpf_tcp_sock, snd_ssthresh):
6852 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
6854 case offsetof(struct bpf_tcp_sock, rcv_nxt):
6855 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
6857 case offsetof(struct bpf_tcp_sock, snd_nxt):
6858 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
6860 case offsetof(struct bpf_tcp_sock, snd_una):
6861 BPF_TCP_SOCK_GET_COMMON(snd_una);
6863 case offsetof(struct bpf_tcp_sock, mss_cache):
6864 BPF_TCP_SOCK_GET_COMMON(mss_cache);
6866 case offsetof(struct bpf_tcp_sock, ecn_flags):
6867 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
6869 case offsetof(struct bpf_tcp_sock, rate_delivered):
6870 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
6872 case offsetof(struct bpf_tcp_sock, rate_interval_us):
6873 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
6875 case offsetof(struct bpf_tcp_sock, packets_out):
6876 BPF_TCP_SOCK_GET_COMMON(packets_out);
6878 case offsetof(struct bpf_tcp_sock, retrans_out):
6879 BPF_TCP_SOCK_GET_COMMON(retrans_out);
6881 case offsetof(struct bpf_tcp_sock, total_retrans):
6882 BPF_TCP_SOCK_GET_COMMON(total_retrans);
6884 case offsetof(struct bpf_tcp_sock, segs_in):
6885 BPF_TCP_SOCK_GET_COMMON(segs_in);
6887 case offsetof(struct bpf_tcp_sock, data_segs_in):
6888 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
6890 case offsetof(struct bpf_tcp_sock, segs_out):
6891 BPF_TCP_SOCK_GET_COMMON(segs_out);
6893 case offsetof(struct bpf_tcp_sock, data_segs_out):
6894 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
6896 case offsetof(struct bpf_tcp_sock, lost_out):
6897 BPF_TCP_SOCK_GET_COMMON(lost_out);
6899 case offsetof(struct bpf_tcp_sock, sacked_out):
6900 BPF_TCP_SOCK_GET_COMMON(sacked_out);
6902 case offsetof(struct bpf_tcp_sock, bytes_received):
6903 BPF_TCP_SOCK_GET_COMMON(bytes_received);
6905 case offsetof(struct bpf_tcp_sock, bytes_acked):
6906 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
6908 case offsetof(struct bpf_tcp_sock, dsack_dups):
6909 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
6911 case offsetof(struct bpf_tcp_sock, delivered):
6912 BPF_TCP_SOCK_GET_COMMON(delivered);
6914 case offsetof(struct bpf_tcp_sock, delivered_ce):
6915 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
6917 case offsetof(struct bpf_tcp_sock, icsk_retransmits):
6918 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
6922 return insn - insn_buf;
6925 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
6927 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
6928 return (unsigned long)sk;
6930 return (unsigned long)NULL;
6933 const struct bpf_func_proto bpf_tcp_sock_proto = {
6934 .func = bpf_tcp_sock,
6936 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
6937 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6940 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
6942 sk = sk_to_full_sk(sk);
6944 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
6945 return (unsigned long)sk;
6947 return (unsigned long)NULL;
6950 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
6951 .func = bpf_get_listener_sock,
6953 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6954 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
6957 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
6959 unsigned int iphdr_len;
6961 switch (skb_protocol(skb, true)) {
6962 case cpu_to_be16(ETH_P_IP):
6963 iphdr_len = sizeof(struct iphdr);
6965 case cpu_to_be16(ETH_P_IPV6):
6966 iphdr_len = sizeof(struct ipv6hdr);
6972 if (skb_headlen(skb) < iphdr_len)
6975 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
6978 return INET_ECN_set_ce(skb);
6981 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6982 struct bpf_insn_access_aux *info)
6984 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
6987 if (off % size != 0)
6992 return size == sizeof(__u32);
6996 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
6997 const struct bpf_insn *si,
6998 struct bpf_insn *insn_buf,
6999 struct bpf_prog *prog, u32 *target_size)
7001 struct bpf_insn *insn = insn_buf;
7003 #define BPF_XDP_SOCK_GET(FIELD) \
7005 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \
7006 sizeof_field(struct bpf_xdp_sock, FIELD)); \
7007 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
7008 si->dst_reg, si->src_reg, \
7009 offsetof(struct xdp_sock, FIELD)); \
7013 case offsetof(struct bpf_xdp_sock, queue_id):
7014 BPF_XDP_SOCK_GET(queue_id);
7018 return insn - insn_buf;
7021 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
7022 .func = bpf_skb_ecn_set_ce,
7024 .ret_type = RET_INTEGER,
7025 .arg1_type = ARG_PTR_TO_CTX,
7028 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7029 struct tcphdr *, th, u32, th_len)
7031 #ifdef CONFIG_SYN_COOKIES
7035 if (unlikely(!sk || th_len < sizeof(*th)))
7038 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
7039 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7042 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7045 if (!th->ack || th->rst || th->syn)
7048 if (unlikely(iph_len < sizeof(struct iphdr)))
7051 if (tcp_synq_no_recent_overflow(sk))
7054 cookie = ntohl(th->ack_seq) - 1;
7056 /* Both struct iphdr and struct ipv6hdr have the version field at the
7057 * same offset so we can cast to the shorter header (struct iphdr).
7059 switch (((struct iphdr *)iph)->version) {
7061 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7064 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
7067 #if IS_BUILTIN(CONFIG_IPV6)
7069 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7072 if (sk->sk_family != AF_INET6)
7075 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
7077 #endif /* CONFIG_IPV6 */
7080 return -EPROTONOSUPPORT;
7092 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
7093 .func = bpf_tcp_check_syncookie,
7096 .ret_type = RET_INTEGER,
7097 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7098 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7099 .arg3_type = ARG_CONST_SIZE,
7100 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7101 .arg5_type = ARG_CONST_SIZE,
7104 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7105 struct tcphdr *, th, u32, th_len)
7107 #ifdef CONFIG_SYN_COOKIES
7111 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
7114 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7117 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7120 if (!th->syn || th->ack || th->fin || th->rst)
7123 if (unlikely(iph_len < sizeof(struct iphdr)))
7126 /* Both struct iphdr and struct ipv6hdr have the version field at the
7127 * same offset so we can cast to the shorter header (struct iphdr).
7129 switch (((struct iphdr *)iph)->version) {
7131 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7134 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
7137 #if IS_BUILTIN(CONFIG_IPV6)
7139 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7142 if (sk->sk_family != AF_INET6)
7145 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
7147 #endif /* CONFIG_IPV6 */
7150 return -EPROTONOSUPPORT;
7155 return cookie | ((u64)mss << 32);
7158 #endif /* CONFIG_SYN_COOKIES */
7161 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
7162 .func = bpf_tcp_gen_syncookie,
7163 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */
7165 .ret_type = RET_INTEGER,
7166 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7167 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7168 .arg3_type = ARG_CONST_SIZE,
7169 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7170 .arg5_type = ARG_CONST_SIZE,
7173 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
7175 if (!sk || flags != 0)
7177 if (!skb_at_tc_ingress(skb))
7179 if (unlikely(dev_net(skb->dev) != sock_net(sk)))
7180 return -ENETUNREACH;
7181 if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
7182 return -ESOCKTNOSUPPORT;
7183 if (sk_is_refcounted(sk) &&
7184 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
7189 skb->destructor = sock_pfree;
7194 static const struct bpf_func_proto bpf_sk_assign_proto = {
7195 .func = bpf_sk_assign,
7197 .ret_type = RET_INTEGER,
7198 .arg1_type = ARG_PTR_TO_CTX,
7199 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7200 .arg3_type = ARG_ANYTHING,
7203 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
7204 u8 search_kind, const u8 *magic,
7205 u8 magic_len, bool *eol)
7211 while (op < opend) {
7214 if (kind == TCPOPT_EOL) {
7216 return ERR_PTR(-ENOMSG);
7217 } else if (kind == TCPOPT_NOP) {
7222 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
7223 /* Something is wrong in the received header.
7224 * Follow the TCP stack's tcp_parse_options()
7225 * and just bail here.
7227 return ERR_PTR(-EFAULT);
7230 if (search_kind == kind) {
7234 if (magic_len > kind_len - 2)
7235 return ERR_PTR(-ENOMSG);
7237 if (!memcmp(&op[2], magic, magic_len))
7244 return ERR_PTR(-ENOMSG);
7247 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7248 void *, search_res, u32, len, u64, flags)
7250 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
7251 const u8 *op, *opend, *magic, *search = search_res;
7252 u8 search_kind, search_len, copy_len, magic_len;
7255 /* 2 byte is the minimal option len except TCPOPT_NOP and
7256 * TCPOPT_EOL which are useless for the bpf prog to learn
7257 * and this helper disallow loading them also.
7259 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
7262 search_kind = search[0];
7263 search_len = search[1];
7265 if (search_len > len || search_kind == TCPOPT_NOP ||
7266 search_kind == TCPOPT_EOL)
7269 if (search_kind == TCPOPT_EXP || search_kind == 253) {
7270 /* 16 or 32 bit magic. +2 for kind and kind length */
7271 if (search_len != 4 && search_len != 6)
7274 magic_len = search_len - 2;
7283 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
7288 op += sizeof(struct tcphdr);
7290 if (!bpf_sock->skb ||
7291 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7292 /* This bpf_sock->op cannot call this helper */
7295 opend = bpf_sock->skb_data_end;
7296 op = bpf_sock->skb->data + sizeof(struct tcphdr);
7299 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
7306 if (copy_len > len) {
7311 memcpy(search_res, op, copy_len);
7315 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
7316 .func = bpf_sock_ops_load_hdr_opt,
7318 .ret_type = RET_INTEGER,
7319 .arg1_type = ARG_PTR_TO_CTX,
7320 .arg2_type = ARG_PTR_TO_MEM,
7321 .arg3_type = ARG_CONST_SIZE,
7322 .arg4_type = ARG_ANYTHING,
7325 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7326 const void *, from, u32, len, u64, flags)
7328 u8 new_kind, new_kind_len, magic_len = 0, *opend;
7329 const u8 *op, *new_op, *magic = NULL;
7330 struct sk_buff *skb;
7333 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
7336 if (len < 2 || flags)
7340 new_kind = new_op[0];
7341 new_kind_len = new_op[1];
7343 if (new_kind_len > len || new_kind == TCPOPT_NOP ||
7344 new_kind == TCPOPT_EOL)
7347 if (new_kind_len > bpf_sock->remaining_opt_len)
7350 /* 253 is another experimental kind */
7351 if (new_kind == TCPOPT_EXP || new_kind == 253) {
7352 if (new_kind_len < 4)
7354 /* Match for the 2 byte magic also.
7355 * RFC 6994: the magic could be 2 or 4 bytes.
7356 * Hence, matching by 2 byte only is on the
7357 * conservative side but it is the right
7358 * thing to do for the 'search-for-duplication'
7365 /* Check for duplication */
7366 skb = bpf_sock->skb;
7367 op = skb->data + sizeof(struct tcphdr);
7368 opend = bpf_sock->skb_data_end;
7370 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
7375 if (PTR_ERR(op) != -ENOMSG)
7379 /* The option has been ended. Treat it as no more
7380 * header option can be written.
7384 /* No duplication found. Store the header option. */
7385 memcpy(opend, from, new_kind_len);
7387 bpf_sock->remaining_opt_len -= new_kind_len;
7388 bpf_sock->skb_data_end += new_kind_len;
7393 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
7394 .func = bpf_sock_ops_store_hdr_opt,
7396 .ret_type = RET_INTEGER,
7397 .arg1_type = ARG_PTR_TO_CTX,
7398 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7399 .arg3_type = ARG_CONST_SIZE,
7400 .arg4_type = ARG_ANYTHING,
7403 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7404 u32, len, u64, flags)
7406 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7409 if (flags || len < 2)
7412 if (len > bpf_sock->remaining_opt_len)
7415 bpf_sock->remaining_opt_len -= len;
7420 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7421 .func = bpf_sock_ops_reserve_hdr_opt,
7423 .ret_type = RET_INTEGER,
7424 .arg1_type = ARG_PTR_TO_CTX,
7425 .arg2_type = ARG_ANYTHING,
7426 .arg3_type = ARG_ANYTHING,
7429 BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb,
7430 u64, tstamp, u32, tstamp_type)
7432 /* skb_clear_delivery_time() is done for inet protocol */
7433 if (skb->protocol != htons(ETH_P_IP) &&
7434 skb->protocol != htons(ETH_P_IPV6))
7437 switch (tstamp_type) {
7438 case BPF_SKB_TSTAMP_DELIVERY_MONO:
7441 skb->tstamp = tstamp;
7442 skb->mono_delivery_time = 1;
7444 case BPF_SKB_TSTAMP_UNSPEC:
7448 skb->mono_delivery_time = 0;
7457 static const struct bpf_func_proto bpf_skb_set_tstamp_proto = {
7458 .func = bpf_skb_set_tstamp,
7460 .ret_type = RET_INTEGER,
7461 .arg1_type = ARG_PTR_TO_CTX,
7462 .arg2_type = ARG_ANYTHING,
7463 .arg3_type = ARG_ANYTHING,
7466 #ifdef CONFIG_SYN_COOKIES
7467 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv4, struct iphdr *, iph,
7468 struct tcphdr *, th, u32, th_len)
7473 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7476 mss = tcp_parse_mss_option(th, 0) ?: TCP_MSS_DEFAULT;
7477 cookie = __cookie_v4_init_sequence(iph, th, &mss);
7479 return cookie | ((u64)mss << 32);
7482 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv4_proto = {
7483 .func = bpf_tcp_raw_gen_syncookie_ipv4,
7484 .gpl_only = true, /* __cookie_v4_init_sequence() is GPL */
7486 .ret_type = RET_INTEGER,
7487 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7488 .arg1_size = sizeof(struct iphdr),
7489 .arg2_type = ARG_PTR_TO_MEM,
7490 .arg3_type = ARG_CONST_SIZE,
7493 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv6, struct ipv6hdr *, iph,
7494 struct tcphdr *, th, u32, th_len)
7496 #if IS_BUILTIN(CONFIG_IPV6)
7497 const u16 mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
7498 sizeof(struct ipv6hdr);
7502 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7505 mss = tcp_parse_mss_option(th, 0) ?: mss_clamp;
7506 cookie = __cookie_v6_init_sequence(iph, th, &mss);
7508 return cookie | ((u64)mss << 32);
7510 return -EPROTONOSUPPORT;
7514 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv6_proto = {
7515 .func = bpf_tcp_raw_gen_syncookie_ipv6,
7516 .gpl_only = true, /* __cookie_v6_init_sequence() is GPL */
7518 .ret_type = RET_INTEGER,
7519 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7520 .arg1_size = sizeof(struct ipv6hdr),
7521 .arg2_type = ARG_PTR_TO_MEM,
7522 .arg3_type = ARG_CONST_SIZE,
7525 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv4, struct iphdr *, iph,
7526 struct tcphdr *, th)
7528 u32 cookie = ntohl(th->ack_seq) - 1;
7530 if (__cookie_v4_check(iph, th, cookie) > 0)
7536 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv4_proto = {
7537 .func = bpf_tcp_raw_check_syncookie_ipv4,
7538 .gpl_only = true, /* __cookie_v4_check is GPL */
7540 .ret_type = RET_INTEGER,
7541 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7542 .arg1_size = sizeof(struct iphdr),
7543 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7544 .arg2_size = sizeof(struct tcphdr),
7547 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv6, struct ipv6hdr *, iph,
7548 struct tcphdr *, th)
7550 #if IS_BUILTIN(CONFIG_IPV6)
7551 u32 cookie = ntohl(th->ack_seq) - 1;
7553 if (__cookie_v6_check(iph, th, cookie) > 0)
7558 return -EPROTONOSUPPORT;
7562 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv6_proto = {
7563 .func = bpf_tcp_raw_check_syncookie_ipv6,
7564 .gpl_only = true, /* __cookie_v6_check is GPL */
7566 .ret_type = RET_INTEGER,
7567 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7568 .arg1_size = sizeof(struct ipv6hdr),
7569 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7570 .arg2_size = sizeof(struct tcphdr),
7572 #endif /* CONFIG_SYN_COOKIES */
7574 #endif /* CONFIG_INET */
7576 bool bpf_helper_changes_pkt_data(void *func)
7578 if (func == bpf_skb_vlan_push ||
7579 func == bpf_skb_vlan_pop ||
7580 func == bpf_skb_store_bytes ||
7581 func == bpf_skb_change_proto ||
7582 func == bpf_skb_change_head ||
7583 func == sk_skb_change_head ||
7584 func == bpf_skb_change_tail ||
7585 func == sk_skb_change_tail ||
7586 func == bpf_skb_adjust_room ||
7587 func == sk_skb_adjust_room ||
7588 func == bpf_skb_pull_data ||
7589 func == sk_skb_pull_data ||
7590 func == bpf_clone_redirect ||
7591 func == bpf_l3_csum_replace ||
7592 func == bpf_l4_csum_replace ||
7593 func == bpf_xdp_adjust_head ||
7594 func == bpf_xdp_adjust_meta ||
7595 func == bpf_msg_pull_data ||
7596 func == bpf_msg_push_data ||
7597 func == bpf_msg_pop_data ||
7598 func == bpf_xdp_adjust_tail ||
7599 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7600 func == bpf_lwt_seg6_store_bytes ||
7601 func == bpf_lwt_seg6_adjust_srh ||
7602 func == bpf_lwt_seg6_action ||
7605 func == bpf_sock_ops_store_hdr_opt ||
7607 func == bpf_lwt_in_push_encap ||
7608 func == bpf_lwt_xmit_push_encap)
7614 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7615 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7617 static const struct bpf_func_proto *
7618 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7620 const struct bpf_func_proto *func_proto;
7622 func_proto = cgroup_common_func_proto(func_id, prog);
7626 func_proto = cgroup_current_func_proto(func_id, prog);
7631 case BPF_FUNC_get_socket_cookie:
7632 return &bpf_get_socket_cookie_sock_proto;
7633 case BPF_FUNC_get_netns_cookie:
7634 return &bpf_get_netns_cookie_sock_proto;
7635 case BPF_FUNC_perf_event_output:
7636 return &bpf_event_output_data_proto;
7637 case BPF_FUNC_sk_storage_get:
7638 return &bpf_sk_storage_get_cg_sock_proto;
7639 case BPF_FUNC_ktime_get_coarse_ns:
7640 return &bpf_ktime_get_coarse_ns_proto;
7642 return bpf_base_func_proto(func_id);
7646 static const struct bpf_func_proto *
7647 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7649 const struct bpf_func_proto *func_proto;
7651 func_proto = cgroup_common_func_proto(func_id, prog);
7655 func_proto = cgroup_current_func_proto(func_id, prog);
7661 switch (prog->expected_attach_type) {
7662 case BPF_CGROUP_INET4_CONNECT:
7663 case BPF_CGROUP_INET6_CONNECT:
7664 return &bpf_bind_proto;
7668 case BPF_FUNC_get_socket_cookie:
7669 return &bpf_get_socket_cookie_sock_addr_proto;
7670 case BPF_FUNC_get_netns_cookie:
7671 return &bpf_get_netns_cookie_sock_addr_proto;
7672 case BPF_FUNC_perf_event_output:
7673 return &bpf_event_output_data_proto;
7675 case BPF_FUNC_sk_lookup_tcp:
7676 return &bpf_sock_addr_sk_lookup_tcp_proto;
7677 case BPF_FUNC_sk_lookup_udp:
7678 return &bpf_sock_addr_sk_lookup_udp_proto;
7679 case BPF_FUNC_sk_release:
7680 return &bpf_sk_release_proto;
7681 case BPF_FUNC_skc_lookup_tcp:
7682 return &bpf_sock_addr_skc_lookup_tcp_proto;
7683 #endif /* CONFIG_INET */
7684 case BPF_FUNC_sk_storage_get:
7685 return &bpf_sk_storage_get_proto;
7686 case BPF_FUNC_sk_storage_delete:
7687 return &bpf_sk_storage_delete_proto;
7688 case BPF_FUNC_setsockopt:
7689 switch (prog->expected_attach_type) {
7690 case BPF_CGROUP_INET4_BIND:
7691 case BPF_CGROUP_INET6_BIND:
7692 case BPF_CGROUP_INET4_CONNECT:
7693 case BPF_CGROUP_INET6_CONNECT:
7694 case BPF_CGROUP_UDP4_RECVMSG:
7695 case BPF_CGROUP_UDP6_RECVMSG:
7696 case BPF_CGROUP_UDP4_SENDMSG:
7697 case BPF_CGROUP_UDP6_SENDMSG:
7698 case BPF_CGROUP_INET4_GETPEERNAME:
7699 case BPF_CGROUP_INET6_GETPEERNAME:
7700 case BPF_CGROUP_INET4_GETSOCKNAME:
7701 case BPF_CGROUP_INET6_GETSOCKNAME:
7702 return &bpf_sock_addr_setsockopt_proto;
7706 case BPF_FUNC_getsockopt:
7707 switch (prog->expected_attach_type) {
7708 case BPF_CGROUP_INET4_BIND:
7709 case BPF_CGROUP_INET6_BIND:
7710 case BPF_CGROUP_INET4_CONNECT:
7711 case BPF_CGROUP_INET6_CONNECT:
7712 case BPF_CGROUP_UDP4_RECVMSG:
7713 case BPF_CGROUP_UDP6_RECVMSG:
7714 case BPF_CGROUP_UDP4_SENDMSG:
7715 case BPF_CGROUP_UDP6_SENDMSG:
7716 case BPF_CGROUP_INET4_GETPEERNAME:
7717 case BPF_CGROUP_INET6_GETPEERNAME:
7718 case BPF_CGROUP_INET4_GETSOCKNAME:
7719 case BPF_CGROUP_INET6_GETSOCKNAME:
7720 return &bpf_sock_addr_getsockopt_proto;
7725 return bpf_sk_base_func_proto(func_id);
7729 static const struct bpf_func_proto *
7730 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7733 case BPF_FUNC_skb_load_bytes:
7734 return &bpf_skb_load_bytes_proto;
7735 case BPF_FUNC_skb_load_bytes_relative:
7736 return &bpf_skb_load_bytes_relative_proto;
7737 case BPF_FUNC_get_socket_cookie:
7738 return &bpf_get_socket_cookie_proto;
7739 case BPF_FUNC_get_socket_uid:
7740 return &bpf_get_socket_uid_proto;
7741 case BPF_FUNC_perf_event_output:
7742 return &bpf_skb_event_output_proto;
7744 return bpf_sk_base_func_proto(func_id);
7748 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7749 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7751 static const struct bpf_func_proto *
7752 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7754 const struct bpf_func_proto *func_proto;
7756 func_proto = cgroup_common_func_proto(func_id, prog);
7761 case BPF_FUNC_sk_fullsock:
7762 return &bpf_sk_fullsock_proto;
7763 case BPF_FUNC_sk_storage_get:
7764 return &bpf_sk_storage_get_proto;
7765 case BPF_FUNC_sk_storage_delete:
7766 return &bpf_sk_storage_delete_proto;
7767 case BPF_FUNC_perf_event_output:
7768 return &bpf_skb_event_output_proto;
7769 #ifdef CONFIG_SOCK_CGROUP_DATA
7770 case BPF_FUNC_skb_cgroup_id:
7771 return &bpf_skb_cgroup_id_proto;
7772 case BPF_FUNC_skb_ancestor_cgroup_id:
7773 return &bpf_skb_ancestor_cgroup_id_proto;
7774 case BPF_FUNC_sk_cgroup_id:
7775 return &bpf_sk_cgroup_id_proto;
7776 case BPF_FUNC_sk_ancestor_cgroup_id:
7777 return &bpf_sk_ancestor_cgroup_id_proto;
7780 case BPF_FUNC_sk_lookup_tcp:
7781 return &bpf_sk_lookup_tcp_proto;
7782 case BPF_FUNC_sk_lookup_udp:
7783 return &bpf_sk_lookup_udp_proto;
7784 case BPF_FUNC_sk_release:
7785 return &bpf_sk_release_proto;
7786 case BPF_FUNC_skc_lookup_tcp:
7787 return &bpf_skc_lookup_tcp_proto;
7788 case BPF_FUNC_tcp_sock:
7789 return &bpf_tcp_sock_proto;
7790 case BPF_FUNC_get_listener_sock:
7791 return &bpf_get_listener_sock_proto;
7792 case BPF_FUNC_skb_ecn_set_ce:
7793 return &bpf_skb_ecn_set_ce_proto;
7796 return sk_filter_func_proto(func_id, prog);
7800 static const struct bpf_func_proto *
7801 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7804 case BPF_FUNC_skb_store_bytes:
7805 return &bpf_skb_store_bytes_proto;
7806 case BPF_FUNC_skb_load_bytes:
7807 return &bpf_skb_load_bytes_proto;
7808 case BPF_FUNC_skb_load_bytes_relative:
7809 return &bpf_skb_load_bytes_relative_proto;
7810 case BPF_FUNC_skb_pull_data:
7811 return &bpf_skb_pull_data_proto;
7812 case BPF_FUNC_csum_diff:
7813 return &bpf_csum_diff_proto;
7814 case BPF_FUNC_csum_update:
7815 return &bpf_csum_update_proto;
7816 case BPF_FUNC_csum_level:
7817 return &bpf_csum_level_proto;
7818 case BPF_FUNC_l3_csum_replace:
7819 return &bpf_l3_csum_replace_proto;
7820 case BPF_FUNC_l4_csum_replace:
7821 return &bpf_l4_csum_replace_proto;
7822 case BPF_FUNC_clone_redirect:
7823 return &bpf_clone_redirect_proto;
7824 case BPF_FUNC_get_cgroup_classid:
7825 return &bpf_get_cgroup_classid_proto;
7826 case BPF_FUNC_skb_vlan_push:
7827 return &bpf_skb_vlan_push_proto;
7828 case BPF_FUNC_skb_vlan_pop:
7829 return &bpf_skb_vlan_pop_proto;
7830 case BPF_FUNC_skb_change_proto:
7831 return &bpf_skb_change_proto_proto;
7832 case BPF_FUNC_skb_change_type:
7833 return &bpf_skb_change_type_proto;
7834 case BPF_FUNC_skb_adjust_room:
7835 return &bpf_skb_adjust_room_proto;
7836 case BPF_FUNC_skb_change_tail:
7837 return &bpf_skb_change_tail_proto;
7838 case BPF_FUNC_skb_change_head:
7839 return &bpf_skb_change_head_proto;
7840 case BPF_FUNC_skb_get_tunnel_key:
7841 return &bpf_skb_get_tunnel_key_proto;
7842 case BPF_FUNC_skb_set_tunnel_key:
7843 return bpf_get_skb_set_tunnel_proto(func_id);
7844 case BPF_FUNC_skb_get_tunnel_opt:
7845 return &bpf_skb_get_tunnel_opt_proto;
7846 case BPF_FUNC_skb_set_tunnel_opt:
7847 return bpf_get_skb_set_tunnel_proto(func_id);
7848 case BPF_FUNC_redirect:
7849 return &bpf_redirect_proto;
7850 case BPF_FUNC_redirect_neigh:
7851 return &bpf_redirect_neigh_proto;
7852 case BPF_FUNC_redirect_peer:
7853 return &bpf_redirect_peer_proto;
7854 case BPF_FUNC_get_route_realm:
7855 return &bpf_get_route_realm_proto;
7856 case BPF_FUNC_get_hash_recalc:
7857 return &bpf_get_hash_recalc_proto;
7858 case BPF_FUNC_set_hash_invalid:
7859 return &bpf_set_hash_invalid_proto;
7860 case BPF_FUNC_set_hash:
7861 return &bpf_set_hash_proto;
7862 case BPF_FUNC_perf_event_output:
7863 return &bpf_skb_event_output_proto;
7864 case BPF_FUNC_get_smp_processor_id:
7865 return &bpf_get_smp_processor_id_proto;
7866 case BPF_FUNC_skb_under_cgroup:
7867 return &bpf_skb_under_cgroup_proto;
7868 case BPF_FUNC_get_socket_cookie:
7869 return &bpf_get_socket_cookie_proto;
7870 case BPF_FUNC_get_socket_uid:
7871 return &bpf_get_socket_uid_proto;
7872 case BPF_FUNC_fib_lookup:
7873 return &bpf_skb_fib_lookup_proto;
7874 case BPF_FUNC_check_mtu:
7875 return &bpf_skb_check_mtu_proto;
7876 case BPF_FUNC_sk_fullsock:
7877 return &bpf_sk_fullsock_proto;
7878 case BPF_FUNC_sk_storage_get:
7879 return &bpf_sk_storage_get_proto;
7880 case BPF_FUNC_sk_storage_delete:
7881 return &bpf_sk_storage_delete_proto;
7883 case BPF_FUNC_skb_get_xfrm_state:
7884 return &bpf_skb_get_xfrm_state_proto;
7886 #ifdef CONFIG_CGROUP_NET_CLASSID
7887 case BPF_FUNC_skb_cgroup_classid:
7888 return &bpf_skb_cgroup_classid_proto;
7890 #ifdef CONFIG_SOCK_CGROUP_DATA
7891 case BPF_FUNC_skb_cgroup_id:
7892 return &bpf_skb_cgroup_id_proto;
7893 case BPF_FUNC_skb_ancestor_cgroup_id:
7894 return &bpf_skb_ancestor_cgroup_id_proto;
7897 case BPF_FUNC_sk_lookup_tcp:
7898 return &bpf_sk_lookup_tcp_proto;
7899 case BPF_FUNC_sk_lookup_udp:
7900 return &bpf_sk_lookup_udp_proto;
7901 case BPF_FUNC_sk_release:
7902 return &bpf_sk_release_proto;
7903 case BPF_FUNC_tcp_sock:
7904 return &bpf_tcp_sock_proto;
7905 case BPF_FUNC_get_listener_sock:
7906 return &bpf_get_listener_sock_proto;
7907 case BPF_FUNC_skc_lookup_tcp:
7908 return &bpf_skc_lookup_tcp_proto;
7909 case BPF_FUNC_tcp_check_syncookie:
7910 return &bpf_tcp_check_syncookie_proto;
7911 case BPF_FUNC_skb_ecn_set_ce:
7912 return &bpf_skb_ecn_set_ce_proto;
7913 case BPF_FUNC_tcp_gen_syncookie:
7914 return &bpf_tcp_gen_syncookie_proto;
7915 case BPF_FUNC_sk_assign:
7916 return &bpf_sk_assign_proto;
7917 case BPF_FUNC_skb_set_tstamp:
7918 return &bpf_skb_set_tstamp_proto;
7919 #ifdef CONFIG_SYN_COOKIES
7920 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
7921 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
7922 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
7923 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
7924 case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
7925 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
7926 case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
7927 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
7931 return bpf_sk_base_func_proto(func_id);
7935 static const struct bpf_func_proto *
7936 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7939 case BPF_FUNC_perf_event_output:
7940 return &bpf_xdp_event_output_proto;
7941 case BPF_FUNC_get_smp_processor_id:
7942 return &bpf_get_smp_processor_id_proto;
7943 case BPF_FUNC_csum_diff:
7944 return &bpf_csum_diff_proto;
7945 case BPF_FUNC_xdp_adjust_head:
7946 return &bpf_xdp_adjust_head_proto;
7947 case BPF_FUNC_xdp_adjust_meta:
7948 return &bpf_xdp_adjust_meta_proto;
7949 case BPF_FUNC_redirect:
7950 return &bpf_xdp_redirect_proto;
7951 case BPF_FUNC_redirect_map:
7952 return &bpf_xdp_redirect_map_proto;
7953 case BPF_FUNC_xdp_adjust_tail:
7954 return &bpf_xdp_adjust_tail_proto;
7955 case BPF_FUNC_xdp_get_buff_len:
7956 return &bpf_xdp_get_buff_len_proto;
7957 case BPF_FUNC_xdp_load_bytes:
7958 return &bpf_xdp_load_bytes_proto;
7959 case BPF_FUNC_xdp_store_bytes:
7960 return &bpf_xdp_store_bytes_proto;
7961 case BPF_FUNC_fib_lookup:
7962 return &bpf_xdp_fib_lookup_proto;
7963 case BPF_FUNC_check_mtu:
7964 return &bpf_xdp_check_mtu_proto;
7966 case BPF_FUNC_sk_lookup_udp:
7967 return &bpf_xdp_sk_lookup_udp_proto;
7968 case BPF_FUNC_sk_lookup_tcp:
7969 return &bpf_xdp_sk_lookup_tcp_proto;
7970 case BPF_FUNC_sk_release:
7971 return &bpf_sk_release_proto;
7972 case BPF_FUNC_skc_lookup_tcp:
7973 return &bpf_xdp_skc_lookup_tcp_proto;
7974 case BPF_FUNC_tcp_check_syncookie:
7975 return &bpf_tcp_check_syncookie_proto;
7976 case BPF_FUNC_tcp_gen_syncookie:
7977 return &bpf_tcp_gen_syncookie_proto;
7978 #ifdef CONFIG_SYN_COOKIES
7979 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
7980 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
7981 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
7982 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
7983 case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
7984 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
7985 case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
7986 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
7990 return bpf_sk_base_func_proto(func_id);
7993 #if IS_MODULE(CONFIG_NF_CONNTRACK) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES)
7994 /* The nf_conn___init type is used in the NF_CONNTRACK kfuncs. The
7995 * kfuncs are defined in two different modules, and we want to be able
7996 * to use them interchangably with the same BTF type ID. Because modules
7997 * can't de-duplicate BTF IDs between each other, we need the type to be
7998 * referenced in the vmlinux BTF or the verifier will get confused about
7999 * the different types. So we add this dummy type reference which will
8000 * be included in vmlinux BTF, allowing both modules to refer to the
8003 BTF_TYPE_EMIT(struct nf_conn___init);
8007 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
8008 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
8010 static const struct bpf_func_proto *
8011 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8013 const struct bpf_func_proto *func_proto;
8015 func_proto = cgroup_common_func_proto(func_id, prog);
8020 case BPF_FUNC_setsockopt:
8021 return &bpf_sock_ops_setsockopt_proto;
8022 case BPF_FUNC_getsockopt:
8023 return &bpf_sock_ops_getsockopt_proto;
8024 case BPF_FUNC_sock_ops_cb_flags_set:
8025 return &bpf_sock_ops_cb_flags_set_proto;
8026 case BPF_FUNC_sock_map_update:
8027 return &bpf_sock_map_update_proto;
8028 case BPF_FUNC_sock_hash_update:
8029 return &bpf_sock_hash_update_proto;
8030 case BPF_FUNC_get_socket_cookie:
8031 return &bpf_get_socket_cookie_sock_ops_proto;
8032 case BPF_FUNC_perf_event_output:
8033 return &bpf_event_output_data_proto;
8034 case BPF_FUNC_sk_storage_get:
8035 return &bpf_sk_storage_get_proto;
8036 case BPF_FUNC_sk_storage_delete:
8037 return &bpf_sk_storage_delete_proto;
8038 case BPF_FUNC_get_netns_cookie:
8039 return &bpf_get_netns_cookie_sock_ops_proto;
8041 case BPF_FUNC_load_hdr_opt:
8042 return &bpf_sock_ops_load_hdr_opt_proto;
8043 case BPF_FUNC_store_hdr_opt:
8044 return &bpf_sock_ops_store_hdr_opt_proto;
8045 case BPF_FUNC_reserve_hdr_opt:
8046 return &bpf_sock_ops_reserve_hdr_opt_proto;
8047 case BPF_FUNC_tcp_sock:
8048 return &bpf_tcp_sock_proto;
8049 #endif /* CONFIG_INET */
8051 return bpf_sk_base_func_proto(func_id);
8055 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
8056 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
8058 static const struct bpf_func_proto *
8059 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8062 case BPF_FUNC_msg_redirect_map:
8063 return &bpf_msg_redirect_map_proto;
8064 case BPF_FUNC_msg_redirect_hash:
8065 return &bpf_msg_redirect_hash_proto;
8066 case BPF_FUNC_msg_apply_bytes:
8067 return &bpf_msg_apply_bytes_proto;
8068 case BPF_FUNC_msg_cork_bytes:
8069 return &bpf_msg_cork_bytes_proto;
8070 case BPF_FUNC_msg_pull_data:
8071 return &bpf_msg_pull_data_proto;
8072 case BPF_FUNC_msg_push_data:
8073 return &bpf_msg_push_data_proto;
8074 case BPF_FUNC_msg_pop_data:
8075 return &bpf_msg_pop_data_proto;
8076 case BPF_FUNC_perf_event_output:
8077 return &bpf_event_output_data_proto;
8078 case BPF_FUNC_get_current_uid_gid:
8079 return &bpf_get_current_uid_gid_proto;
8080 case BPF_FUNC_get_current_pid_tgid:
8081 return &bpf_get_current_pid_tgid_proto;
8082 case BPF_FUNC_sk_storage_get:
8083 return &bpf_sk_storage_get_proto;
8084 case BPF_FUNC_sk_storage_delete:
8085 return &bpf_sk_storage_delete_proto;
8086 case BPF_FUNC_get_netns_cookie:
8087 return &bpf_get_netns_cookie_sk_msg_proto;
8088 #ifdef CONFIG_CGROUPS
8089 case BPF_FUNC_get_current_cgroup_id:
8090 return &bpf_get_current_cgroup_id_proto;
8091 case BPF_FUNC_get_current_ancestor_cgroup_id:
8092 return &bpf_get_current_ancestor_cgroup_id_proto;
8094 #ifdef CONFIG_CGROUP_NET_CLASSID
8095 case BPF_FUNC_get_cgroup_classid:
8096 return &bpf_get_cgroup_classid_curr_proto;
8099 return bpf_sk_base_func_proto(func_id);
8103 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
8104 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
8106 static const struct bpf_func_proto *
8107 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8110 case BPF_FUNC_skb_store_bytes:
8111 return &bpf_skb_store_bytes_proto;
8112 case BPF_FUNC_skb_load_bytes:
8113 return &bpf_skb_load_bytes_proto;
8114 case BPF_FUNC_skb_pull_data:
8115 return &sk_skb_pull_data_proto;
8116 case BPF_FUNC_skb_change_tail:
8117 return &sk_skb_change_tail_proto;
8118 case BPF_FUNC_skb_change_head:
8119 return &sk_skb_change_head_proto;
8120 case BPF_FUNC_skb_adjust_room:
8121 return &sk_skb_adjust_room_proto;
8122 case BPF_FUNC_get_socket_cookie:
8123 return &bpf_get_socket_cookie_proto;
8124 case BPF_FUNC_get_socket_uid:
8125 return &bpf_get_socket_uid_proto;
8126 case BPF_FUNC_sk_redirect_map:
8127 return &bpf_sk_redirect_map_proto;
8128 case BPF_FUNC_sk_redirect_hash:
8129 return &bpf_sk_redirect_hash_proto;
8130 case BPF_FUNC_perf_event_output:
8131 return &bpf_skb_event_output_proto;
8133 case BPF_FUNC_sk_lookup_tcp:
8134 return &bpf_sk_lookup_tcp_proto;
8135 case BPF_FUNC_sk_lookup_udp:
8136 return &bpf_sk_lookup_udp_proto;
8137 case BPF_FUNC_sk_release:
8138 return &bpf_sk_release_proto;
8139 case BPF_FUNC_skc_lookup_tcp:
8140 return &bpf_skc_lookup_tcp_proto;
8143 return bpf_sk_base_func_proto(func_id);
8147 static const struct bpf_func_proto *
8148 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8151 case BPF_FUNC_skb_load_bytes:
8152 return &bpf_flow_dissector_load_bytes_proto;
8154 return bpf_sk_base_func_proto(func_id);
8158 static const struct bpf_func_proto *
8159 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8162 case BPF_FUNC_skb_load_bytes:
8163 return &bpf_skb_load_bytes_proto;
8164 case BPF_FUNC_skb_pull_data:
8165 return &bpf_skb_pull_data_proto;
8166 case BPF_FUNC_csum_diff:
8167 return &bpf_csum_diff_proto;
8168 case BPF_FUNC_get_cgroup_classid:
8169 return &bpf_get_cgroup_classid_proto;
8170 case BPF_FUNC_get_route_realm:
8171 return &bpf_get_route_realm_proto;
8172 case BPF_FUNC_get_hash_recalc:
8173 return &bpf_get_hash_recalc_proto;
8174 case BPF_FUNC_perf_event_output:
8175 return &bpf_skb_event_output_proto;
8176 case BPF_FUNC_get_smp_processor_id:
8177 return &bpf_get_smp_processor_id_proto;
8178 case BPF_FUNC_skb_under_cgroup:
8179 return &bpf_skb_under_cgroup_proto;
8181 return bpf_sk_base_func_proto(func_id);
8185 static const struct bpf_func_proto *
8186 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8189 case BPF_FUNC_lwt_push_encap:
8190 return &bpf_lwt_in_push_encap_proto;
8192 return lwt_out_func_proto(func_id, prog);
8196 static const struct bpf_func_proto *
8197 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8200 case BPF_FUNC_skb_get_tunnel_key:
8201 return &bpf_skb_get_tunnel_key_proto;
8202 case BPF_FUNC_skb_set_tunnel_key:
8203 return bpf_get_skb_set_tunnel_proto(func_id);
8204 case BPF_FUNC_skb_get_tunnel_opt:
8205 return &bpf_skb_get_tunnel_opt_proto;
8206 case BPF_FUNC_skb_set_tunnel_opt:
8207 return bpf_get_skb_set_tunnel_proto(func_id);
8208 case BPF_FUNC_redirect:
8209 return &bpf_redirect_proto;
8210 case BPF_FUNC_clone_redirect:
8211 return &bpf_clone_redirect_proto;
8212 case BPF_FUNC_skb_change_tail:
8213 return &bpf_skb_change_tail_proto;
8214 case BPF_FUNC_skb_change_head:
8215 return &bpf_skb_change_head_proto;
8216 case BPF_FUNC_skb_store_bytes:
8217 return &bpf_skb_store_bytes_proto;
8218 case BPF_FUNC_csum_update:
8219 return &bpf_csum_update_proto;
8220 case BPF_FUNC_csum_level:
8221 return &bpf_csum_level_proto;
8222 case BPF_FUNC_l3_csum_replace:
8223 return &bpf_l3_csum_replace_proto;
8224 case BPF_FUNC_l4_csum_replace:
8225 return &bpf_l4_csum_replace_proto;
8226 case BPF_FUNC_set_hash_invalid:
8227 return &bpf_set_hash_invalid_proto;
8228 case BPF_FUNC_lwt_push_encap:
8229 return &bpf_lwt_xmit_push_encap_proto;
8231 return lwt_out_func_proto(func_id, prog);
8235 static const struct bpf_func_proto *
8236 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8239 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
8240 case BPF_FUNC_lwt_seg6_store_bytes:
8241 return &bpf_lwt_seg6_store_bytes_proto;
8242 case BPF_FUNC_lwt_seg6_action:
8243 return &bpf_lwt_seg6_action_proto;
8244 case BPF_FUNC_lwt_seg6_adjust_srh:
8245 return &bpf_lwt_seg6_adjust_srh_proto;
8248 return lwt_out_func_proto(func_id, prog);
8252 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
8253 const struct bpf_prog *prog,
8254 struct bpf_insn_access_aux *info)
8256 const int size_default = sizeof(__u32);
8258 if (off < 0 || off >= sizeof(struct __sk_buff))
8261 /* The verifier guarantees that size > 0. */
8262 if (off % size != 0)
8266 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8267 if (off + size > offsetofend(struct __sk_buff, cb[4]))
8270 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
8271 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
8272 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
8273 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
8274 case bpf_ctx_range(struct __sk_buff, data):
8275 case bpf_ctx_range(struct __sk_buff, data_meta):
8276 case bpf_ctx_range(struct __sk_buff, data_end):
8277 if (size != size_default)
8280 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8282 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8283 if (type == BPF_WRITE || size != sizeof(__u64))
8286 case bpf_ctx_range(struct __sk_buff, tstamp):
8287 if (size != sizeof(__u64))
8290 case offsetof(struct __sk_buff, sk):
8291 if (type == BPF_WRITE || size != sizeof(__u64))
8293 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
8295 case offsetof(struct __sk_buff, tstamp_type):
8297 case offsetofend(struct __sk_buff, tstamp_type) ... offsetof(struct __sk_buff, hwtstamp) - 1:
8298 /* Explicitly prohibit access to padding in __sk_buff. */
8301 /* Only narrow read access allowed for now. */
8302 if (type == BPF_WRITE) {
8303 if (size != size_default)
8306 bpf_ctx_record_field_size(info, size_default);
8307 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8315 static bool sk_filter_is_valid_access(int off, int size,
8316 enum bpf_access_type type,
8317 const struct bpf_prog *prog,
8318 struct bpf_insn_access_aux *info)
8321 case bpf_ctx_range(struct __sk_buff, tc_classid):
8322 case bpf_ctx_range(struct __sk_buff, data):
8323 case bpf_ctx_range(struct __sk_buff, data_meta):
8324 case bpf_ctx_range(struct __sk_buff, data_end):
8325 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8326 case bpf_ctx_range(struct __sk_buff, tstamp):
8327 case bpf_ctx_range(struct __sk_buff, wire_len):
8328 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8332 if (type == BPF_WRITE) {
8334 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8341 return bpf_skb_is_valid_access(off, size, type, prog, info);
8344 static bool cg_skb_is_valid_access(int off, int size,
8345 enum bpf_access_type type,
8346 const struct bpf_prog *prog,
8347 struct bpf_insn_access_aux *info)
8350 case bpf_ctx_range(struct __sk_buff, tc_classid):
8351 case bpf_ctx_range(struct __sk_buff, data_meta):
8352 case bpf_ctx_range(struct __sk_buff, wire_len):
8354 case bpf_ctx_range(struct __sk_buff, data):
8355 case bpf_ctx_range(struct __sk_buff, data_end):
8361 if (type == BPF_WRITE) {
8363 case bpf_ctx_range(struct __sk_buff, mark):
8364 case bpf_ctx_range(struct __sk_buff, priority):
8365 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8367 case bpf_ctx_range(struct __sk_buff, tstamp):
8377 case bpf_ctx_range(struct __sk_buff, data):
8378 info->reg_type = PTR_TO_PACKET;
8380 case bpf_ctx_range(struct __sk_buff, data_end):
8381 info->reg_type = PTR_TO_PACKET_END;
8385 return bpf_skb_is_valid_access(off, size, type, prog, info);
8388 static bool lwt_is_valid_access(int off, int size,
8389 enum bpf_access_type type,
8390 const struct bpf_prog *prog,
8391 struct bpf_insn_access_aux *info)
8394 case bpf_ctx_range(struct __sk_buff, tc_classid):
8395 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8396 case bpf_ctx_range(struct __sk_buff, data_meta):
8397 case bpf_ctx_range(struct __sk_buff, tstamp):
8398 case bpf_ctx_range(struct __sk_buff, wire_len):
8399 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8403 if (type == BPF_WRITE) {
8405 case bpf_ctx_range(struct __sk_buff, mark):
8406 case bpf_ctx_range(struct __sk_buff, priority):
8407 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8415 case bpf_ctx_range(struct __sk_buff, data):
8416 info->reg_type = PTR_TO_PACKET;
8418 case bpf_ctx_range(struct __sk_buff, data_end):
8419 info->reg_type = PTR_TO_PACKET_END;
8423 return bpf_skb_is_valid_access(off, size, type, prog, info);
8426 /* Attach type specific accesses */
8427 static bool __sock_filter_check_attach_type(int off,
8428 enum bpf_access_type access_type,
8429 enum bpf_attach_type attach_type)
8432 case offsetof(struct bpf_sock, bound_dev_if):
8433 case offsetof(struct bpf_sock, mark):
8434 case offsetof(struct bpf_sock, priority):
8435 switch (attach_type) {
8436 case BPF_CGROUP_INET_SOCK_CREATE:
8437 case BPF_CGROUP_INET_SOCK_RELEASE:
8442 case bpf_ctx_range(struct bpf_sock, src_ip4):
8443 switch (attach_type) {
8444 case BPF_CGROUP_INET4_POST_BIND:
8449 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8450 switch (attach_type) {
8451 case BPF_CGROUP_INET6_POST_BIND:
8456 case bpf_ctx_range(struct bpf_sock, src_port):
8457 switch (attach_type) {
8458 case BPF_CGROUP_INET4_POST_BIND:
8459 case BPF_CGROUP_INET6_POST_BIND:
8466 return access_type == BPF_READ;
8471 bool bpf_sock_common_is_valid_access(int off, int size,
8472 enum bpf_access_type type,
8473 struct bpf_insn_access_aux *info)
8476 case bpf_ctx_range_till(struct bpf_sock, type, priority):
8479 return bpf_sock_is_valid_access(off, size, type, info);
8483 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
8484 struct bpf_insn_access_aux *info)
8486 const int size_default = sizeof(__u32);
8489 if (off < 0 || off >= sizeof(struct bpf_sock))
8491 if (off % size != 0)
8495 case offsetof(struct bpf_sock, state):
8496 case offsetof(struct bpf_sock, family):
8497 case offsetof(struct bpf_sock, type):
8498 case offsetof(struct bpf_sock, protocol):
8499 case offsetof(struct bpf_sock, src_port):
8500 case offsetof(struct bpf_sock, rx_queue_mapping):
8501 case bpf_ctx_range(struct bpf_sock, src_ip4):
8502 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8503 case bpf_ctx_range(struct bpf_sock, dst_ip4):
8504 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8505 bpf_ctx_record_field_size(info, size_default);
8506 return bpf_ctx_narrow_access_ok(off, size, size_default);
8507 case bpf_ctx_range(struct bpf_sock, dst_port):
8508 field_size = size == size_default ?
8509 size_default : sizeof_field(struct bpf_sock, dst_port);
8510 bpf_ctx_record_field_size(info, field_size);
8511 return bpf_ctx_narrow_access_ok(off, size, field_size);
8512 case offsetofend(struct bpf_sock, dst_port) ...
8513 offsetof(struct bpf_sock, dst_ip4) - 1:
8517 return size == size_default;
8520 static bool sock_filter_is_valid_access(int off, int size,
8521 enum bpf_access_type type,
8522 const struct bpf_prog *prog,
8523 struct bpf_insn_access_aux *info)
8525 if (!bpf_sock_is_valid_access(off, size, type, info))
8527 return __sock_filter_check_attach_type(off, type,
8528 prog->expected_attach_type);
8531 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
8532 const struct bpf_prog *prog)
8534 /* Neither direct read nor direct write requires any preliminary
8540 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
8541 const struct bpf_prog *prog, int drop_verdict)
8543 struct bpf_insn *insn = insn_buf;
8548 /* if (!skb->cloned)
8551 * (Fast-path, otherwise approximation that we might be
8552 * a clone, do the rest in helper.)
8554 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET);
8555 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
8556 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
8558 /* ret = bpf_skb_pull_data(skb, 0); */
8559 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
8560 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
8561 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
8562 BPF_FUNC_skb_pull_data);
8565 * return TC_ACT_SHOT;
8567 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
8568 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
8569 *insn++ = BPF_EXIT_INSN();
8572 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
8574 *insn++ = prog->insnsi[0];
8576 return insn - insn_buf;
8579 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
8580 struct bpf_insn *insn_buf)
8582 bool indirect = BPF_MODE(orig->code) == BPF_IND;
8583 struct bpf_insn *insn = insn_buf;
8586 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
8588 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
8590 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
8592 /* We're guaranteed here that CTX is in R6. */
8593 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
8595 switch (BPF_SIZE(orig->code)) {
8597 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
8600 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
8603 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
8607 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
8608 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
8609 *insn++ = BPF_EXIT_INSN();
8611 return insn - insn_buf;
8614 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8615 const struct bpf_prog *prog)
8617 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8620 static bool tc_cls_act_is_valid_access(int off, int size,
8621 enum bpf_access_type type,
8622 const struct bpf_prog *prog,
8623 struct bpf_insn_access_aux *info)
8625 if (type == BPF_WRITE) {
8627 case bpf_ctx_range(struct __sk_buff, mark):
8628 case bpf_ctx_range(struct __sk_buff, tc_index):
8629 case bpf_ctx_range(struct __sk_buff, priority):
8630 case bpf_ctx_range(struct __sk_buff, tc_classid):
8631 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8632 case bpf_ctx_range(struct __sk_buff, tstamp):
8633 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8641 case bpf_ctx_range(struct __sk_buff, data):
8642 info->reg_type = PTR_TO_PACKET;
8644 case bpf_ctx_range(struct __sk_buff, data_meta):
8645 info->reg_type = PTR_TO_PACKET_META;
8647 case bpf_ctx_range(struct __sk_buff, data_end):
8648 info->reg_type = PTR_TO_PACKET_END;
8650 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8652 case offsetof(struct __sk_buff, tstamp_type):
8653 /* The convert_ctx_access() on reading and writing
8654 * __sk_buff->tstamp depends on whether the bpf prog
8655 * has used __sk_buff->tstamp_type or not.
8656 * Thus, we need to set prog->tstamp_type_access
8657 * earlier during is_valid_access() here.
8659 ((struct bpf_prog *)prog)->tstamp_type_access = 1;
8660 return size == sizeof(__u8);
8663 return bpf_skb_is_valid_access(off, size, type, prog, info);
8666 DEFINE_MUTEX(nf_conn_btf_access_lock);
8667 EXPORT_SYMBOL_GPL(nf_conn_btf_access_lock);
8669 int (*nfct_btf_struct_access)(struct bpf_verifier_log *log, const struct btf *btf,
8670 const struct btf_type *t, int off, int size,
8671 enum bpf_access_type atype, u32 *next_btf_id,
8672 enum bpf_type_flag *flag);
8673 EXPORT_SYMBOL_GPL(nfct_btf_struct_access);
8675 static int tc_cls_act_btf_struct_access(struct bpf_verifier_log *log,
8676 const struct btf *btf,
8677 const struct btf_type *t, int off,
8678 int size, enum bpf_access_type atype,
8680 enum bpf_type_flag *flag)
8684 if (atype == BPF_READ)
8685 return btf_struct_access(log, btf, t, off, size, atype, next_btf_id,
8688 mutex_lock(&nf_conn_btf_access_lock);
8689 if (nfct_btf_struct_access)
8690 ret = nfct_btf_struct_access(log, btf, t, off, size, atype, next_btf_id, flag);
8691 mutex_unlock(&nf_conn_btf_access_lock);
8696 static bool __is_valid_xdp_access(int off, int size)
8698 if (off < 0 || off >= sizeof(struct xdp_md))
8700 if (off % size != 0)
8702 if (size != sizeof(__u32))
8708 static bool xdp_is_valid_access(int off, int size,
8709 enum bpf_access_type type,
8710 const struct bpf_prog *prog,
8711 struct bpf_insn_access_aux *info)
8713 if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8715 case offsetof(struct xdp_md, egress_ifindex):
8720 if (type == BPF_WRITE) {
8721 if (bpf_prog_is_dev_bound(prog->aux)) {
8723 case offsetof(struct xdp_md, rx_queue_index):
8724 return __is_valid_xdp_access(off, size);
8731 case offsetof(struct xdp_md, data):
8732 info->reg_type = PTR_TO_PACKET;
8734 case offsetof(struct xdp_md, data_meta):
8735 info->reg_type = PTR_TO_PACKET_META;
8737 case offsetof(struct xdp_md, data_end):
8738 info->reg_type = PTR_TO_PACKET_END;
8742 return __is_valid_xdp_access(off, size);
8745 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act)
8747 const u32 act_max = XDP_REDIRECT;
8749 pr_warn_once("%s XDP return value %u on prog %s (id %d) dev %s, expect packet loss!\n",
8750 act > act_max ? "Illegal" : "Driver unsupported",
8751 act, prog->aux->name, prog->aux->id, dev ? dev->name : "N/A");
8753 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
8755 static int xdp_btf_struct_access(struct bpf_verifier_log *log,
8756 const struct btf *btf,
8757 const struct btf_type *t, int off,
8758 int size, enum bpf_access_type atype,
8760 enum bpf_type_flag *flag)
8764 if (atype == BPF_READ)
8765 return btf_struct_access(log, btf, t, off, size, atype, next_btf_id,
8768 mutex_lock(&nf_conn_btf_access_lock);
8769 if (nfct_btf_struct_access)
8770 ret = nfct_btf_struct_access(log, btf, t, off, size, atype, next_btf_id, flag);
8771 mutex_unlock(&nf_conn_btf_access_lock);
8776 static bool sock_addr_is_valid_access(int off, int size,
8777 enum bpf_access_type type,
8778 const struct bpf_prog *prog,
8779 struct bpf_insn_access_aux *info)
8781 const int size_default = sizeof(__u32);
8783 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
8785 if (off % size != 0)
8788 /* Disallow access to IPv6 fields from IPv4 contex and vise
8792 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8793 switch (prog->expected_attach_type) {
8794 case BPF_CGROUP_INET4_BIND:
8795 case BPF_CGROUP_INET4_CONNECT:
8796 case BPF_CGROUP_INET4_GETPEERNAME:
8797 case BPF_CGROUP_INET4_GETSOCKNAME:
8798 case BPF_CGROUP_UDP4_SENDMSG:
8799 case BPF_CGROUP_UDP4_RECVMSG:
8805 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8806 switch (prog->expected_attach_type) {
8807 case BPF_CGROUP_INET6_BIND:
8808 case BPF_CGROUP_INET6_CONNECT:
8809 case BPF_CGROUP_INET6_GETPEERNAME:
8810 case BPF_CGROUP_INET6_GETSOCKNAME:
8811 case BPF_CGROUP_UDP6_SENDMSG:
8812 case BPF_CGROUP_UDP6_RECVMSG:
8818 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8819 switch (prog->expected_attach_type) {
8820 case BPF_CGROUP_UDP4_SENDMSG:
8826 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8828 switch (prog->expected_attach_type) {
8829 case BPF_CGROUP_UDP6_SENDMSG:
8838 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8839 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8840 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8841 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8843 case bpf_ctx_range(struct bpf_sock_addr, user_port):
8844 if (type == BPF_READ) {
8845 bpf_ctx_record_field_size(info, size_default);
8847 if (bpf_ctx_wide_access_ok(off, size,
8848 struct bpf_sock_addr,
8852 if (bpf_ctx_wide_access_ok(off, size,
8853 struct bpf_sock_addr,
8857 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8860 if (bpf_ctx_wide_access_ok(off, size,
8861 struct bpf_sock_addr,
8865 if (bpf_ctx_wide_access_ok(off, size,
8866 struct bpf_sock_addr,
8870 if (size != size_default)
8874 case offsetof(struct bpf_sock_addr, sk):
8875 if (type != BPF_READ)
8877 if (size != sizeof(__u64))
8879 info->reg_type = PTR_TO_SOCKET;
8882 if (type == BPF_READ) {
8883 if (size != size_default)
8893 static bool sock_ops_is_valid_access(int off, int size,
8894 enum bpf_access_type type,
8895 const struct bpf_prog *prog,
8896 struct bpf_insn_access_aux *info)
8898 const int size_default = sizeof(__u32);
8900 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
8903 /* The verifier guarantees that size > 0. */
8904 if (off % size != 0)
8907 if (type == BPF_WRITE) {
8909 case offsetof(struct bpf_sock_ops, reply):
8910 case offsetof(struct bpf_sock_ops, sk_txhash):
8911 if (size != size_default)
8919 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
8921 if (size != sizeof(__u64))
8924 case offsetof(struct bpf_sock_ops, sk):
8925 if (size != sizeof(__u64))
8927 info->reg_type = PTR_TO_SOCKET_OR_NULL;
8929 case offsetof(struct bpf_sock_ops, skb_data):
8930 if (size != sizeof(__u64))
8932 info->reg_type = PTR_TO_PACKET;
8934 case offsetof(struct bpf_sock_ops, skb_data_end):
8935 if (size != sizeof(__u64))
8937 info->reg_type = PTR_TO_PACKET_END;
8939 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
8940 bpf_ctx_record_field_size(info, size_default);
8941 return bpf_ctx_narrow_access_ok(off, size,
8944 if (size != size_default)
8953 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
8954 const struct bpf_prog *prog)
8956 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
8959 static bool sk_skb_is_valid_access(int off, int size,
8960 enum bpf_access_type type,
8961 const struct bpf_prog *prog,
8962 struct bpf_insn_access_aux *info)
8965 case bpf_ctx_range(struct __sk_buff, tc_classid):
8966 case bpf_ctx_range(struct __sk_buff, data_meta):
8967 case bpf_ctx_range(struct __sk_buff, tstamp):
8968 case bpf_ctx_range(struct __sk_buff, wire_len):
8969 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8973 if (type == BPF_WRITE) {
8975 case bpf_ctx_range(struct __sk_buff, tc_index):
8976 case bpf_ctx_range(struct __sk_buff, priority):
8984 case bpf_ctx_range(struct __sk_buff, mark):
8986 case bpf_ctx_range(struct __sk_buff, data):
8987 info->reg_type = PTR_TO_PACKET;
8989 case bpf_ctx_range(struct __sk_buff, data_end):
8990 info->reg_type = PTR_TO_PACKET_END;
8994 return bpf_skb_is_valid_access(off, size, type, prog, info);
8997 static bool sk_msg_is_valid_access(int off, int size,
8998 enum bpf_access_type type,
8999 const struct bpf_prog *prog,
9000 struct bpf_insn_access_aux *info)
9002 if (type == BPF_WRITE)
9005 if (off % size != 0)
9009 case offsetof(struct sk_msg_md, data):
9010 info->reg_type = PTR_TO_PACKET;
9011 if (size != sizeof(__u64))
9014 case offsetof(struct sk_msg_md, data_end):
9015 info->reg_type = PTR_TO_PACKET_END;
9016 if (size != sizeof(__u64))
9019 case offsetof(struct sk_msg_md, sk):
9020 if (size != sizeof(__u64))
9022 info->reg_type = PTR_TO_SOCKET;
9024 case bpf_ctx_range(struct sk_msg_md, family):
9025 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
9026 case bpf_ctx_range(struct sk_msg_md, local_ip4):
9027 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
9028 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
9029 case bpf_ctx_range(struct sk_msg_md, remote_port):
9030 case bpf_ctx_range(struct sk_msg_md, local_port):
9031 case bpf_ctx_range(struct sk_msg_md, size):
9032 if (size != sizeof(__u32))
9041 static bool flow_dissector_is_valid_access(int off, int size,
9042 enum bpf_access_type type,
9043 const struct bpf_prog *prog,
9044 struct bpf_insn_access_aux *info)
9046 const int size_default = sizeof(__u32);
9048 if (off < 0 || off >= sizeof(struct __sk_buff))
9051 if (type == BPF_WRITE)
9055 case bpf_ctx_range(struct __sk_buff, data):
9056 if (size != size_default)
9058 info->reg_type = PTR_TO_PACKET;
9060 case bpf_ctx_range(struct __sk_buff, data_end):
9061 if (size != size_default)
9063 info->reg_type = PTR_TO_PACKET_END;
9065 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
9066 if (size != sizeof(__u64))
9068 info->reg_type = PTR_TO_FLOW_KEYS;
9075 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
9076 const struct bpf_insn *si,
9077 struct bpf_insn *insn_buf,
9078 struct bpf_prog *prog,
9082 struct bpf_insn *insn = insn_buf;
9085 case offsetof(struct __sk_buff, data):
9086 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
9087 si->dst_reg, si->src_reg,
9088 offsetof(struct bpf_flow_dissector, data));
9091 case offsetof(struct __sk_buff, data_end):
9092 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
9093 si->dst_reg, si->src_reg,
9094 offsetof(struct bpf_flow_dissector, data_end));
9097 case offsetof(struct __sk_buff, flow_keys):
9098 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
9099 si->dst_reg, si->src_reg,
9100 offsetof(struct bpf_flow_dissector, flow_keys));
9104 return insn - insn_buf;
9107 static struct bpf_insn *bpf_convert_tstamp_type_read(const struct bpf_insn *si,
9108 struct bpf_insn *insn)
9110 __u8 value_reg = si->dst_reg;
9111 __u8 skb_reg = si->src_reg;
9112 /* AX is needed because src_reg and dst_reg could be the same */
9113 __u8 tmp_reg = BPF_REG_AX;
9115 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg,
9116 PKT_VLAN_PRESENT_OFFSET);
9117 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg,
9118 SKB_MONO_DELIVERY_TIME_MASK, 2);
9119 *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_UNSPEC);
9120 *insn++ = BPF_JMP_A(1);
9121 *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_DELIVERY_MONO);
9126 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si,
9127 struct bpf_insn *insn)
9129 /* si->dst_reg = skb_shinfo(SKB); */
9130 #ifdef NET_SKBUFF_DATA_USES_OFFSET
9131 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9132 BPF_REG_AX, si->src_reg,
9133 offsetof(struct sk_buff, end));
9134 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
9135 si->dst_reg, si->src_reg,
9136 offsetof(struct sk_buff, head));
9137 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
9139 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9140 si->dst_reg, si->src_reg,
9141 offsetof(struct sk_buff, end));
9147 static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog,
9148 const struct bpf_insn *si,
9149 struct bpf_insn *insn)
9151 __u8 value_reg = si->dst_reg;
9152 __u8 skb_reg = si->src_reg;
9154 #ifdef CONFIG_NET_CLS_ACT
9155 /* If the tstamp_type is read,
9156 * the bpf prog is aware the tstamp could have delivery time.
9157 * Thus, read skb->tstamp as is if tstamp_type_access is true.
9159 if (!prog->tstamp_type_access) {
9160 /* AX is needed because src_reg and dst_reg could be the same */
9161 __u8 tmp_reg = BPF_REG_AX;
9163 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, PKT_VLAN_PRESENT_OFFSET);
9164 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg,
9165 TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK);
9166 *insn++ = BPF_JMP32_IMM(BPF_JNE, tmp_reg,
9167 TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK, 2);
9168 /* skb->tc_at_ingress && skb->mono_delivery_time,
9169 * read 0 as the (rcv) timestamp.
9171 *insn++ = BPF_MOV64_IMM(value_reg, 0);
9172 *insn++ = BPF_JMP_A(1);
9176 *insn++ = BPF_LDX_MEM(BPF_DW, value_reg, skb_reg,
9177 offsetof(struct sk_buff, tstamp));
9181 static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog,
9182 const struct bpf_insn *si,
9183 struct bpf_insn *insn)
9185 __u8 value_reg = si->src_reg;
9186 __u8 skb_reg = si->dst_reg;
9188 #ifdef CONFIG_NET_CLS_ACT
9189 /* If the tstamp_type is read,
9190 * the bpf prog is aware the tstamp could have delivery time.
9191 * Thus, write skb->tstamp as is if tstamp_type_access is true.
9192 * Otherwise, writing at ingress will have to clear the
9193 * mono_delivery_time bit also.
9195 if (!prog->tstamp_type_access) {
9196 __u8 tmp_reg = BPF_REG_AX;
9198 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, PKT_VLAN_PRESENT_OFFSET);
9199 /* Writing __sk_buff->tstamp as ingress, goto <clear> */
9200 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1);
9202 *insn++ = BPF_JMP_A(2);
9203 /* <clear>: mono_delivery_time */
9204 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_MONO_DELIVERY_TIME_MASK);
9205 *insn++ = BPF_STX_MEM(BPF_B, skb_reg, tmp_reg, PKT_VLAN_PRESENT_OFFSET);
9209 /* <store>: skb->tstamp = tstamp */
9210 *insn++ = BPF_STX_MEM(BPF_DW, skb_reg, value_reg,
9211 offsetof(struct sk_buff, tstamp));
9215 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
9216 const struct bpf_insn *si,
9217 struct bpf_insn *insn_buf,
9218 struct bpf_prog *prog, u32 *target_size)
9220 struct bpf_insn *insn = insn_buf;
9224 case offsetof(struct __sk_buff, len):
9225 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9226 bpf_target_off(struct sk_buff, len, 4,
9230 case offsetof(struct __sk_buff, protocol):
9231 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9232 bpf_target_off(struct sk_buff, protocol, 2,
9236 case offsetof(struct __sk_buff, vlan_proto):
9237 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9238 bpf_target_off(struct sk_buff, vlan_proto, 2,
9242 case offsetof(struct __sk_buff, priority):
9243 if (type == BPF_WRITE)
9244 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9245 bpf_target_off(struct sk_buff, priority, 4,
9248 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9249 bpf_target_off(struct sk_buff, priority, 4,
9253 case offsetof(struct __sk_buff, ingress_ifindex):
9254 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9255 bpf_target_off(struct sk_buff, skb_iif, 4,
9259 case offsetof(struct __sk_buff, ifindex):
9260 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9261 si->dst_reg, si->src_reg,
9262 offsetof(struct sk_buff, dev));
9263 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9264 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9265 bpf_target_off(struct net_device, ifindex, 4,
9269 case offsetof(struct __sk_buff, hash):
9270 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9271 bpf_target_off(struct sk_buff, hash, 4,
9275 case offsetof(struct __sk_buff, mark):
9276 if (type == BPF_WRITE)
9277 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9278 bpf_target_off(struct sk_buff, mark, 4,
9281 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9282 bpf_target_off(struct sk_buff, mark, 4,
9286 case offsetof(struct __sk_buff, pkt_type):
9288 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
9290 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
9291 #ifdef __BIG_ENDIAN_BITFIELD
9292 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
9296 case offsetof(struct __sk_buff, queue_mapping):
9297 if (type == BPF_WRITE) {
9298 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
9299 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
9300 bpf_target_off(struct sk_buff,
9304 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9305 bpf_target_off(struct sk_buff,
9311 case offsetof(struct __sk_buff, vlan_present):
9313 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
9314 PKT_VLAN_PRESENT_OFFSET);
9315 if (PKT_VLAN_PRESENT_BIT)
9316 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
9317 if (PKT_VLAN_PRESENT_BIT < 7)
9318 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
9321 case offsetof(struct __sk_buff, vlan_tci):
9322 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9323 bpf_target_off(struct sk_buff, vlan_tci, 2,
9327 case offsetof(struct __sk_buff, cb[0]) ...
9328 offsetofend(struct __sk_buff, cb[4]) - 1:
9329 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
9330 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
9331 offsetof(struct qdisc_skb_cb, data)) %
9334 prog->cb_access = 1;
9336 off -= offsetof(struct __sk_buff, cb[0]);
9337 off += offsetof(struct sk_buff, cb);
9338 off += offsetof(struct qdisc_skb_cb, data);
9339 if (type == BPF_WRITE)
9340 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
9343 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
9347 case offsetof(struct __sk_buff, tc_classid):
9348 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
9351 off -= offsetof(struct __sk_buff, tc_classid);
9352 off += offsetof(struct sk_buff, cb);
9353 off += offsetof(struct qdisc_skb_cb, tc_classid);
9355 if (type == BPF_WRITE)
9356 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
9359 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
9363 case offsetof(struct __sk_buff, data):
9364 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9365 si->dst_reg, si->src_reg,
9366 offsetof(struct sk_buff, data));
9369 case offsetof(struct __sk_buff, data_meta):
9371 off -= offsetof(struct __sk_buff, data_meta);
9372 off += offsetof(struct sk_buff, cb);
9373 off += offsetof(struct bpf_skb_data_end, data_meta);
9374 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9378 case offsetof(struct __sk_buff, data_end):
9380 off -= offsetof(struct __sk_buff, data_end);
9381 off += offsetof(struct sk_buff, cb);
9382 off += offsetof(struct bpf_skb_data_end, data_end);
9383 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9387 case offsetof(struct __sk_buff, tc_index):
9388 #ifdef CONFIG_NET_SCHED
9389 if (type == BPF_WRITE)
9390 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
9391 bpf_target_off(struct sk_buff, tc_index, 2,
9394 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9395 bpf_target_off(struct sk_buff, tc_index, 2,
9399 if (type == BPF_WRITE)
9400 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
9402 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9406 case offsetof(struct __sk_buff, napi_id):
9407 #if defined(CONFIG_NET_RX_BUSY_POLL)
9408 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9409 bpf_target_off(struct sk_buff, napi_id, 4,
9411 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
9412 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9415 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9418 case offsetof(struct __sk_buff, family):
9419 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9421 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9422 si->dst_reg, si->src_reg,
9423 offsetof(struct sk_buff, sk));
9424 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9425 bpf_target_off(struct sock_common,
9429 case offsetof(struct __sk_buff, remote_ip4):
9430 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9432 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9433 si->dst_reg, si->src_reg,
9434 offsetof(struct sk_buff, sk));
9435 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9436 bpf_target_off(struct sock_common,
9440 case offsetof(struct __sk_buff, local_ip4):
9441 BUILD_BUG_ON(sizeof_field(struct sock_common,
9442 skc_rcv_saddr) != 4);
9444 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9445 si->dst_reg, si->src_reg,
9446 offsetof(struct sk_buff, sk));
9447 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9448 bpf_target_off(struct sock_common,
9452 case offsetof(struct __sk_buff, remote_ip6[0]) ...
9453 offsetof(struct __sk_buff, remote_ip6[3]):
9454 #if IS_ENABLED(CONFIG_IPV6)
9455 BUILD_BUG_ON(sizeof_field(struct sock_common,
9456 skc_v6_daddr.s6_addr32[0]) != 4);
9459 off -= offsetof(struct __sk_buff, remote_ip6[0]);
9461 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9462 si->dst_reg, si->src_reg,
9463 offsetof(struct sk_buff, sk));
9464 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9465 offsetof(struct sock_common,
9466 skc_v6_daddr.s6_addr32[0]) +
9469 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9472 case offsetof(struct __sk_buff, local_ip6[0]) ...
9473 offsetof(struct __sk_buff, local_ip6[3]):
9474 #if IS_ENABLED(CONFIG_IPV6)
9475 BUILD_BUG_ON(sizeof_field(struct sock_common,
9476 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9479 off -= offsetof(struct __sk_buff, local_ip6[0]);
9481 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9482 si->dst_reg, si->src_reg,
9483 offsetof(struct sk_buff, sk));
9484 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9485 offsetof(struct sock_common,
9486 skc_v6_rcv_saddr.s6_addr32[0]) +
9489 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9493 case offsetof(struct __sk_buff, remote_port):
9494 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9496 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9497 si->dst_reg, si->src_reg,
9498 offsetof(struct sk_buff, sk));
9499 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9500 bpf_target_off(struct sock_common,
9503 #ifndef __BIG_ENDIAN_BITFIELD
9504 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9508 case offsetof(struct __sk_buff, local_port):
9509 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9511 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9512 si->dst_reg, si->src_reg,
9513 offsetof(struct sk_buff, sk));
9514 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9515 bpf_target_off(struct sock_common,
9516 skc_num, 2, target_size));
9519 case offsetof(struct __sk_buff, tstamp):
9520 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
9522 if (type == BPF_WRITE)
9523 insn = bpf_convert_tstamp_write(prog, si, insn);
9525 insn = bpf_convert_tstamp_read(prog, si, insn);
9528 case offsetof(struct __sk_buff, tstamp_type):
9529 insn = bpf_convert_tstamp_type_read(si, insn);
9532 case offsetof(struct __sk_buff, gso_segs):
9533 insn = bpf_convert_shinfo_access(si, insn);
9534 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
9535 si->dst_reg, si->dst_reg,
9536 bpf_target_off(struct skb_shared_info,
9540 case offsetof(struct __sk_buff, gso_size):
9541 insn = bpf_convert_shinfo_access(si, insn);
9542 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
9543 si->dst_reg, si->dst_reg,
9544 bpf_target_off(struct skb_shared_info,
9548 case offsetof(struct __sk_buff, wire_len):
9549 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
9552 off -= offsetof(struct __sk_buff, wire_len);
9553 off += offsetof(struct sk_buff, cb);
9554 off += offsetof(struct qdisc_skb_cb, pkt_len);
9556 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
9559 case offsetof(struct __sk_buff, sk):
9560 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9561 si->dst_reg, si->src_reg,
9562 offsetof(struct sk_buff, sk));
9564 case offsetof(struct __sk_buff, hwtstamp):
9565 BUILD_BUG_ON(sizeof_field(struct skb_shared_hwtstamps, hwtstamp) != 8);
9566 BUILD_BUG_ON(offsetof(struct skb_shared_hwtstamps, hwtstamp) != 0);
9568 insn = bpf_convert_shinfo_access(si, insn);
9569 *insn++ = BPF_LDX_MEM(BPF_DW,
9570 si->dst_reg, si->dst_reg,
9571 bpf_target_off(struct skb_shared_info,
9577 return insn - insn_buf;
9580 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
9581 const struct bpf_insn *si,
9582 struct bpf_insn *insn_buf,
9583 struct bpf_prog *prog, u32 *target_size)
9585 struct bpf_insn *insn = insn_buf;
9589 case offsetof(struct bpf_sock, bound_dev_if):
9590 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
9592 if (type == BPF_WRITE)
9593 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9594 offsetof(struct sock, sk_bound_dev_if));
9596 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9597 offsetof(struct sock, sk_bound_dev_if));
9600 case offsetof(struct bpf_sock, mark):
9601 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
9603 if (type == BPF_WRITE)
9604 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9605 offsetof(struct sock, sk_mark));
9607 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9608 offsetof(struct sock, sk_mark));
9611 case offsetof(struct bpf_sock, priority):
9612 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
9614 if (type == BPF_WRITE)
9615 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9616 offsetof(struct sock, sk_priority));
9618 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9619 offsetof(struct sock, sk_priority));
9622 case offsetof(struct bpf_sock, family):
9623 *insn++ = BPF_LDX_MEM(
9624 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
9625 si->dst_reg, si->src_reg,
9626 bpf_target_off(struct sock_common,
9628 sizeof_field(struct sock_common,
9633 case offsetof(struct bpf_sock, type):
9634 *insn++ = BPF_LDX_MEM(
9635 BPF_FIELD_SIZEOF(struct sock, sk_type),
9636 si->dst_reg, si->src_reg,
9637 bpf_target_off(struct sock, sk_type,
9638 sizeof_field(struct sock, sk_type),
9642 case offsetof(struct bpf_sock, protocol):
9643 *insn++ = BPF_LDX_MEM(
9644 BPF_FIELD_SIZEOF(struct sock, sk_protocol),
9645 si->dst_reg, si->src_reg,
9646 bpf_target_off(struct sock, sk_protocol,
9647 sizeof_field(struct sock, sk_protocol),
9651 case offsetof(struct bpf_sock, src_ip4):
9652 *insn++ = BPF_LDX_MEM(
9653 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9654 bpf_target_off(struct sock_common, skc_rcv_saddr,
9655 sizeof_field(struct sock_common,
9660 case offsetof(struct bpf_sock, dst_ip4):
9661 *insn++ = BPF_LDX_MEM(
9662 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9663 bpf_target_off(struct sock_common, skc_daddr,
9664 sizeof_field(struct sock_common,
9669 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
9670 #if IS_ENABLED(CONFIG_IPV6)
9672 off -= offsetof(struct bpf_sock, src_ip6[0]);
9673 *insn++ = BPF_LDX_MEM(
9674 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9677 skc_v6_rcv_saddr.s6_addr32[0],
9678 sizeof_field(struct sock_common,
9679 skc_v6_rcv_saddr.s6_addr32[0]),
9680 target_size) + off);
9683 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9687 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
9688 #if IS_ENABLED(CONFIG_IPV6)
9690 off -= offsetof(struct bpf_sock, dst_ip6[0]);
9691 *insn++ = BPF_LDX_MEM(
9692 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9693 bpf_target_off(struct sock_common,
9694 skc_v6_daddr.s6_addr32[0],
9695 sizeof_field(struct sock_common,
9696 skc_v6_daddr.s6_addr32[0]),
9697 target_size) + off);
9699 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9704 case offsetof(struct bpf_sock, src_port):
9705 *insn++ = BPF_LDX_MEM(
9706 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
9707 si->dst_reg, si->src_reg,
9708 bpf_target_off(struct sock_common, skc_num,
9709 sizeof_field(struct sock_common,
9714 case offsetof(struct bpf_sock, dst_port):
9715 *insn++ = BPF_LDX_MEM(
9716 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
9717 si->dst_reg, si->src_reg,
9718 bpf_target_off(struct sock_common, skc_dport,
9719 sizeof_field(struct sock_common,
9724 case offsetof(struct bpf_sock, state):
9725 *insn++ = BPF_LDX_MEM(
9726 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
9727 si->dst_reg, si->src_reg,
9728 bpf_target_off(struct sock_common, skc_state,
9729 sizeof_field(struct sock_common,
9733 case offsetof(struct bpf_sock, rx_queue_mapping):
9734 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
9735 *insn++ = BPF_LDX_MEM(
9736 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
9737 si->dst_reg, si->src_reg,
9738 bpf_target_off(struct sock, sk_rx_queue_mapping,
9739 sizeof_field(struct sock,
9740 sk_rx_queue_mapping),
9742 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
9744 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9746 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9752 return insn - insn_buf;
9755 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
9756 const struct bpf_insn *si,
9757 struct bpf_insn *insn_buf,
9758 struct bpf_prog *prog, u32 *target_size)
9760 struct bpf_insn *insn = insn_buf;
9763 case offsetof(struct __sk_buff, ifindex):
9764 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9765 si->dst_reg, si->src_reg,
9766 offsetof(struct sk_buff, dev));
9767 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9768 bpf_target_off(struct net_device, ifindex, 4,
9772 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9776 return insn - insn_buf;
9779 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
9780 const struct bpf_insn *si,
9781 struct bpf_insn *insn_buf,
9782 struct bpf_prog *prog, u32 *target_size)
9784 struct bpf_insn *insn = insn_buf;
9787 case offsetof(struct xdp_md, data):
9788 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
9789 si->dst_reg, si->src_reg,
9790 offsetof(struct xdp_buff, data));
9792 case offsetof(struct xdp_md, data_meta):
9793 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
9794 si->dst_reg, si->src_reg,
9795 offsetof(struct xdp_buff, data_meta));
9797 case offsetof(struct xdp_md, data_end):
9798 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
9799 si->dst_reg, si->src_reg,
9800 offsetof(struct xdp_buff, data_end));
9802 case offsetof(struct xdp_md, ingress_ifindex):
9803 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9804 si->dst_reg, si->src_reg,
9805 offsetof(struct xdp_buff, rxq));
9806 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
9807 si->dst_reg, si->dst_reg,
9808 offsetof(struct xdp_rxq_info, dev));
9809 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9810 offsetof(struct net_device, ifindex));
9812 case offsetof(struct xdp_md, rx_queue_index):
9813 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9814 si->dst_reg, si->src_reg,
9815 offsetof(struct xdp_buff, rxq));
9816 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9817 offsetof(struct xdp_rxq_info,
9820 case offsetof(struct xdp_md, egress_ifindex):
9821 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
9822 si->dst_reg, si->src_reg,
9823 offsetof(struct xdp_buff, txq));
9824 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
9825 si->dst_reg, si->dst_reg,
9826 offsetof(struct xdp_txq_info, dev));
9827 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9828 offsetof(struct net_device, ifindex));
9832 return insn - insn_buf;
9835 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
9836 * context Structure, F is Field in context structure that contains a pointer
9837 * to Nested Structure of type NS that has the field NF.
9839 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
9840 * sure that SIZE is not greater than actual size of S.F.NF.
9842 * If offset OFF is provided, the load happens from that offset relative to
9845 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
9847 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
9848 si->src_reg, offsetof(S, F)); \
9849 *insn++ = BPF_LDX_MEM( \
9850 SIZE, si->dst_reg, si->dst_reg, \
9851 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
9856 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
9857 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
9858 BPF_FIELD_SIZEOF(NS, NF), 0)
9860 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
9861 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
9863 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
9864 * "register" since two registers available in convert_ctx_access are not
9865 * enough: we can't override neither SRC, since it contains value to store, nor
9866 * DST since it contains pointer to context that may be used by later
9867 * instructions. But we need a temporary place to save pointer to nested
9868 * structure whose field we want to store to.
9870 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
9872 int tmp_reg = BPF_REG_9; \
9873 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
9875 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
9877 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
9879 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
9880 si->dst_reg, offsetof(S, F)); \
9881 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \
9882 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
9885 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
9889 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
9892 if (type == BPF_WRITE) { \
9893 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
9896 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
9897 S, NS, F, NF, SIZE, OFF); \
9901 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
9902 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
9903 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
9905 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
9906 const struct bpf_insn *si,
9907 struct bpf_insn *insn_buf,
9908 struct bpf_prog *prog, u32 *target_size)
9910 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
9911 struct bpf_insn *insn = insn_buf;
9914 case offsetof(struct bpf_sock_addr, user_family):
9915 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9916 struct sockaddr, uaddr, sa_family);
9919 case offsetof(struct bpf_sock_addr, user_ip4):
9920 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9921 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
9922 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
9925 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9927 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
9928 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9929 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9930 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
9934 case offsetof(struct bpf_sock_addr, user_port):
9935 /* To get port we need to know sa_family first and then treat
9936 * sockaddr as either sockaddr_in or sockaddr_in6.
9937 * Though we can simplify since port field has same offset and
9938 * size in both structures.
9939 * Here we check this invariant and use just one of the
9940 * structures if it's true.
9942 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
9943 offsetof(struct sockaddr_in6, sin6_port));
9944 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
9945 sizeof_field(struct sockaddr_in6, sin6_port));
9946 /* Account for sin6_port being smaller than user_port. */
9947 port_size = min(port_size, BPF_LDST_BYTES(si));
9948 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9949 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9950 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
9953 case offsetof(struct bpf_sock_addr, family):
9954 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9955 struct sock, sk, sk_family);
9958 case offsetof(struct bpf_sock_addr, type):
9959 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9960 struct sock, sk, sk_type);
9963 case offsetof(struct bpf_sock_addr, protocol):
9964 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9965 struct sock, sk, sk_protocol);
9968 case offsetof(struct bpf_sock_addr, msg_src_ip4):
9969 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
9970 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9971 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
9972 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
9975 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9978 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
9979 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
9980 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9981 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
9982 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
9984 case offsetof(struct bpf_sock_addr, sk):
9985 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
9986 si->dst_reg, si->src_reg,
9987 offsetof(struct bpf_sock_addr_kern, sk));
9991 return insn - insn_buf;
9994 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
9995 const struct bpf_insn *si,
9996 struct bpf_insn *insn_buf,
9997 struct bpf_prog *prog,
10000 struct bpf_insn *insn = insn_buf;
10003 /* Helper macro for adding read access to tcp_sock or sock fields. */
10004 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
10006 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \
10007 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
10008 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
10009 if (si->dst_reg == reg || si->src_reg == reg) \
10011 if (si->dst_reg == reg || si->src_reg == reg) \
10013 if (si->dst_reg == si->src_reg) { \
10014 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
10015 offsetof(struct bpf_sock_ops_kern, \
10017 fullsock_reg = reg; \
10020 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10021 struct bpf_sock_ops_kern, \
10023 fullsock_reg, si->src_reg, \
10024 offsetof(struct bpf_sock_ops_kern, \
10026 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
10027 if (si->dst_reg == si->src_reg) \
10028 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10029 offsetof(struct bpf_sock_ops_kern, \
10031 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10032 struct bpf_sock_ops_kern, sk),\
10033 si->dst_reg, si->src_reg, \
10034 offsetof(struct bpf_sock_ops_kern, sk));\
10035 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
10037 si->dst_reg, si->dst_reg, \
10038 offsetof(OBJ, OBJ_FIELD)); \
10039 if (si->dst_reg == si->src_reg) { \
10040 *insn++ = BPF_JMP_A(1); \
10041 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10042 offsetof(struct bpf_sock_ops_kern, \
10047 #define SOCK_OPS_GET_SK() \
10049 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \
10050 if (si->dst_reg == reg || si->src_reg == reg) \
10052 if (si->dst_reg == reg || si->src_reg == reg) \
10054 if (si->dst_reg == si->src_reg) { \
10055 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
10056 offsetof(struct bpf_sock_ops_kern, \
10058 fullsock_reg = reg; \
10061 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10062 struct bpf_sock_ops_kern, \
10064 fullsock_reg, si->src_reg, \
10065 offsetof(struct bpf_sock_ops_kern, \
10067 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
10068 if (si->dst_reg == si->src_reg) \
10069 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10070 offsetof(struct bpf_sock_ops_kern, \
10072 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10073 struct bpf_sock_ops_kern, sk),\
10074 si->dst_reg, si->src_reg, \
10075 offsetof(struct bpf_sock_ops_kern, sk));\
10076 if (si->dst_reg == si->src_reg) { \
10077 *insn++ = BPF_JMP_A(1); \
10078 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10079 offsetof(struct bpf_sock_ops_kern, \
10084 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
10085 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
10087 /* Helper macro for adding write access to tcp_sock or sock fields.
10088 * The macro is called with two registers, dst_reg which contains a pointer
10089 * to ctx (context) and src_reg which contains the value that should be
10090 * stored. However, we need an additional register since we cannot overwrite
10091 * dst_reg because it may be used later in the program.
10092 * Instead we "borrow" one of the other register. We first save its value
10093 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
10094 * it at the end of the macro.
10096 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
10098 int reg = BPF_REG_9; \
10099 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
10100 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
10101 if (si->dst_reg == reg || si->src_reg == reg) \
10103 if (si->dst_reg == reg || si->src_reg == reg) \
10105 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
10106 offsetof(struct bpf_sock_ops_kern, \
10108 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10109 struct bpf_sock_ops_kern, \
10111 reg, si->dst_reg, \
10112 offsetof(struct bpf_sock_ops_kern, \
10114 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
10115 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10116 struct bpf_sock_ops_kern, sk),\
10117 reg, si->dst_reg, \
10118 offsetof(struct bpf_sock_ops_kern, sk));\
10119 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \
10120 reg, si->src_reg, \
10121 offsetof(OBJ, OBJ_FIELD)); \
10122 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
10123 offsetof(struct bpf_sock_ops_kern, \
10127 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
10129 if (TYPE == BPF_WRITE) \
10130 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
10132 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
10135 if (insn > insn_buf)
10136 return insn - insn_buf;
10139 case offsetof(struct bpf_sock_ops, op):
10140 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10142 si->dst_reg, si->src_reg,
10143 offsetof(struct bpf_sock_ops_kern, op));
10146 case offsetof(struct bpf_sock_ops, replylong[0]) ...
10147 offsetof(struct bpf_sock_ops, replylong[3]):
10148 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
10149 sizeof_field(struct bpf_sock_ops_kern, reply));
10150 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
10151 sizeof_field(struct bpf_sock_ops_kern, replylong));
10153 off -= offsetof(struct bpf_sock_ops, replylong[0]);
10154 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
10155 if (type == BPF_WRITE)
10156 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
10159 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10163 case offsetof(struct bpf_sock_ops, family):
10164 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10166 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10167 struct bpf_sock_ops_kern, sk),
10168 si->dst_reg, si->src_reg,
10169 offsetof(struct bpf_sock_ops_kern, sk));
10170 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10171 offsetof(struct sock_common, skc_family));
10174 case offsetof(struct bpf_sock_ops, remote_ip4):
10175 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10177 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10178 struct bpf_sock_ops_kern, sk),
10179 si->dst_reg, si->src_reg,
10180 offsetof(struct bpf_sock_ops_kern, sk));
10181 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10182 offsetof(struct sock_common, skc_daddr));
10185 case offsetof(struct bpf_sock_ops, local_ip4):
10186 BUILD_BUG_ON(sizeof_field(struct sock_common,
10187 skc_rcv_saddr) != 4);
10189 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10190 struct bpf_sock_ops_kern, sk),
10191 si->dst_reg, si->src_reg,
10192 offsetof(struct bpf_sock_ops_kern, sk));
10193 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10194 offsetof(struct sock_common,
10198 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
10199 offsetof(struct bpf_sock_ops, remote_ip6[3]):
10200 #if IS_ENABLED(CONFIG_IPV6)
10201 BUILD_BUG_ON(sizeof_field(struct sock_common,
10202 skc_v6_daddr.s6_addr32[0]) != 4);
10205 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
10206 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10207 struct bpf_sock_ops_kern, sk),
10208 si->dst_reg, si->src_reg,
10209 offsetof(struct bpf_sock_ops_kern, sk));
10210 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10211 offsetof(struct sock_common,
10212 skc_v6_daddr.s6_addr32[0]) +
10215 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10219 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
10220 offsetof(struct bpf_sock_ops, local_ip6[3]):
10221 #if IS_ENABLED(CONFIG_IPV6)
10222 BUILD_BUG_ON(sizeof_field(struct sock_common,
10223 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10226 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
10227 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10228 struct bpf_sock_ops_kern, sk),
10229 si->dst_reg, si->src_reg,
10230 offsetof(struct bpf_sock_ops_kern, sk));
10231 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10232 offsetof(struct sock_common,
10233 skc_v6_rcv_saddr.s6_addr32[0]) +
10236 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10240 case offsetof(struct bpf_sock_ops, remote_port):
10241 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10243 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10244 struct bpf_sock_ops_kern, sk),
10245 si->dst_reg, si->src_reg,
10246 offsetof(struct bpf_sock_ops_kern, sk));
10247 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10248 offsetof(struct sock_common, skc_dport));
10249 #ifndef __BIG_ENDIAN_BITFIELD
10250 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10254 case offsetof(struct bpf_sock_ops, local_port):
10255 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10257 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10258 struct bpf_sock_ops_kern, sk),
10259 si->dst_reg, si->src_reg,
10260 offsetof(struct bpf_sock_ops_kern, sk));
10261 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10262 offsetof(struct sock_common, skc_num));
10265 case offsetof(struct bpf_sock_ops, is_fullsock):
10266 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10267 struct bpf_sock_ops_kern,
10269 si->dst_reg, si->src_reg,
10270 offsetof(struct bpf_sock_ops_kern,
10274 case offsetof(struct bpf_sock_ops, state):
10275 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
10277 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10278 struct bpf_sock_ops_kern, sk),
10279 si->dst_reg, si->src_reg,
10280 offsetof(struct bpf_sock_ops_kern, sk));
10281 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
10282 offsetof(struct sock_common, skc_state));
10285 case offsetof(struct bpf_sock_ops, rtt_min):
10286 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
10287 sizeof(struct minmax));
10288 BUILD_BUG_ON(sizeof(struct minmax) <
10289 sizeof(struct minmax_sample));
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 tcp_sock, rtt_min) +
10297 sizeof_field(struct minmax_sample, t));
10300 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
10301 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
10305 case offsetof(struct bpf_sock_ops, sk_txhash):
10306 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
10307 struct sock, type);
10309 case offsetof(struct bpf_sock_ops, snd_cwnd):
10310 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
10312 case offsetof(struct bpf_sock_ops, srtt_us):
10313 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
10315 case offsetof(struct bpf_sock_ops, snd_ssthresh):
10316 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
10318 case offsetof(struct bpf_sock_ops, rcv_nxt):
10319 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
10321 case offsetof(struct bpf_sock_ops, snd_nxt):
10322 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
10324 case offsetof(struct bpf_sock_ops, snd_una):
10325 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
10327 case offsetof(struct bpf_sock_ops, mss_cache):
10328 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
10330 case offsetof(struct bpf_sock_ops, ecn_flags):
10331 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
10333 case offsetof(struct bpf_sock_ops, rate_delivered):
10334 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
10336 case offsetof(struct bpf_sock_ops, rate_interval_us):
10337 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
10339 case offsetof(struct bpf_sock_ops, packets_out):
10340 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
10342 case offsetof(struct bpf_sock_ops, retrans_out):
10343 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
10345 case offsetof(struct bpf_sock_ops, total_retrans):
10346 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
10348 case offsetof(struct bpf_sock_ops, segs_in):
10349 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
10351 case offsetof(struct bpf_sock_ops, data_segs_in):
10352 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
10354 case offsetof(struct bpf_sock_ops, segs_out):
10355 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
10357 case offsetof(struct bpf_sock_ops, data_segs_out):
10358 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
10360 case offsetof(struct bpf_sock_ops, lost_out):
10361 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
10363 case offsetof(struct bpf_sock_ops, sacked_out):
10364 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
10366 case offsetof(struct bpf_sock_ops, bytes_received):
10367 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
10369 case offsetof(struct bpf_sock_ops, bytes_acked):
10370 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
10372 case offsetof(struct bpf_sock_ops, sk):
10375 case offsetof(struct bpf_sock_ops, skb_data_end):
10376 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10378 si->dst_reg, si->src_reg,
10379 offsetof(struct bpf_sock_ops_kern,
10382 case offsetof(struct bpf_sock_ops, skb_data):
10383 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10385 si->dst_reg, si->src_reg,
10386 offsetof(struct bpf_sock_ops_kern,
10388 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10389 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10390 si->dst_reg, si->dst_reg,
10391 offsetof(struct sk_buff, data));
10393 case offsetof(struct bpf_sock_ops, skb_len):
10394 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10396 si->dst_reg, si->src_reg,
10397 offsetof(struct bpf_sock_ops_kern,
10399 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10400 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10401 si->dst_reg, si->dst_reg,
10402 offsetof(struct sk_buff, len));
10404 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
10405 off = offsetof(struct sk_buff, cb);
10406 off += offsetof(struct tcp_skb_cb, tcp_flags);
10407 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
10408 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10410 si->dst_reg, si->src_reg,
10411 offsetof(struct bpf_sock_ops_kern,
10413 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10414 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
10416 si->dst_reg, si->dst_reg, off);
10419 return insn - insn_buf;
10422 /* data_end = skb->data + skb_headlen() */
10423 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
10424 struct bpf_insn *insn)
10427 int temp_reg_off = offsetof(struct sk_buff, cb) +
10428 offsetof(struct sk_skb_cb, temp_reg);
10430 if (si->src_reg == si->dst_reg) {
10431 /* We need an extra register, choose and save a register. */
10433 if (si->src_reg == reg || si->dst_reg == reg)
10435 if (si->src_reg == reg || si->dst_reg == reg)
10437 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off);
10442 /* reg = skb->data */
10443 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10445 offsetof(struct sk_buff, data));
10446 /* AX = skb->len */
10447 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10448 BPF_REG_AX, si->src_reg,
10449 offsetof(struct sk_buff, len));
10450 /* reg = skb->data + skb->len */
10451 *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX);
10452 /* AX = skb->data_len */
10453 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
10454 BPF_REG_AX, si->src_reg,
10455 offsetof(struct sk_buff, data_len));
10457 /* reg = skb->data + skb->len - skb->data_len */
10458 *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX);
10460 if (si->src_reg == si->dst_reg) {
10461 /* Restore the saved register */
10462 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg);
10463 *insn++ = BPF_MOV64_REG(si->dst_reg, reg);
10464 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off);
10470 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
10471 const struct bpf_insn *si,
10472 struct bpf_insn *insn_buf,
10473 struct bpf_prog *prog, u32 *target_size)
10475 struct bpf_insn *insn = insn_buf;
10479 case offsetof(struct __sk_buff, data_end):
10480 insn = bpf_convert_data_end_access(si, insn);
10482 case offsetof(struct __sk_buff, cb[0]) ...
10483 offsetofend(struct __sk_buff, cb[4]) - 1:
10484 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20);
10485 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
10486 offsetof(struct sk_skb_cb, data)) %
10489 prog->cb_access = 1;
10491 off -= offsetof(struct __sk_buff, cb[0]);
10492 off += offsetof(struct sk_buff, cb);
10493 off += offsetof(struct sk_skb_cb, data);
10494 if (type == BPF_WRITE)
10495 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
10498 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
10504 return bpf_convert_ctx_access(type, si, insn_buf, prog,
10508 return insn - insn_buf;
10511 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
10512 const struct bpf_insn *si,
10513 struct bpf_insn *insn_buf,
10514 struct bpf_prog *prog, u32 *target_size)
10516 struct bpf_insn *insn = insn_buf;
10517 #if IS_ENABLED(CONFIG_IPV6)
10521 /* convert ctx uses the fact sg element is first in struct */
10522 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
10525 case offsetof(struct sk_msg_md, data):
10526 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
10527 si->dst_reg, si->src_reg,
10528 offsetof(struct sk_msg, data));
10530 case offsetof(struct sk_msg_md, data_end):
10531 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
10532 si->dst_reg, si->src_reg,
10533 offsetof(struct sk_msg, data_end));
10535 case offsetof(struct sk_msg_md, family):
10536 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10538 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10539 struct sk_msg, sk),
10540 si->dst_reg, si->src_reg,
10541 offsetof(struct sk_msg, sk));
10542 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10543 offsetof(struct sock_common, skc_family));
10546 case offsetof(struct sk_msg_md, remote_ip4):
10547 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10549 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10550 struct sk_msg, sk),
10551 si->dst_reg, si->src_reg,
10552 offsetof(struct sk_msg, sk));
10553 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10554 offsetof(struct sock_common, skc_daddr));
10557 case offsetof(struct sk_msg_md, local_ip4):
10558 BUILD_BUG_ON(sizeof_field(struct sock_common,
10559 skc_rcv_saddr) != 4);
10561 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10562 struct sk_msg, sk),
10563 si->dst_reg, si->src_reg,
10564 offsetof(struct sk_msg, sk));
10565 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10566 offsetof(struct sock_common,
10570 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
10571 offsetof(struct sk_msg_md, remote_ip6[3]):
10572 #if IS_ENABLED(CONFIG_IPV6)
10573 BUILD_BUG_ON(sizeof_field(struct sock_common,
10574 skc_v6_daddr.s6_addr32[0]) != 4);
10577 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
10578 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10579 struct sk_msg, sk),
10580 si->dst_reg, si->src_reg,
10581 offsetof(struct sk_msg, sk));
10582 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10583 offsetof(struct sock_common,
10584 skc_v6_daddr.s6_addr32[0]) +
10587 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10591 case offsetof(struct sk_msg_md, local_ip6[0]) ...
10592 offsetof(struct sk_msg_md, local_ip6[3]):
10593 #if IS_ENABLED(CONFIG_IPV6)
10594 BUILD_BUG_ON(sizeof_field(struct sock_common,
10595 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10598 off -= offsetof(struct sk_msg_md, local_ip6[0]);
10599 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10600 struct sk_msg, sk),
10601 si->dst_reg, si->src_reg,
10602 offsetof(struct sk_msg, sk));
10603 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10604 offsetof(struct sock_common,
10605 skc_v6_rcv_saddr.s6_addr32[0]) +
10608 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10612 case offsetof(struct sk_msg_md, remote_port):
10613 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10615 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10616 struct sk_msg, sk),
10617 si->dst_reg, si->src_reg,
10618 offsetof(struct sk_msg, sk));
10619 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10620 offsetof(struct sock_common, skc_dport));
10621 #ifndef __BIG_ENDIAN_BITFIELD
10622 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10626 case offsetof(struct sk_msg_md, local_port):
10627 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10629 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10630 struct sk_msg, sk),
10631 si->dst_reg, si->src_reg,
10632 offsetof(struct sk_msg, sk));
10633 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10634 offsetof(struct sock_common, skc_num));
10637 case offsetof(struct sk_msg_md, size):
10638 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
10639 si->dst_reg, si->src_reg,
10640 offsetof(struct sk_msg_sg, size));
10643 case offsetof(struct sk_msg_md, sk):
10644 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
10645 si->dst_reg, si->src_reg,
10646 offsetof(struct sk_msg, sk));
10650 return insn - insn_buf;
10653 const struct bpf_verifier_ops sk_filter_verifier_ops = {
10654 .get_func_proto = sk_filter_func_proto,
10655 .is_valid_access = sk_filter_is_valid_access,
10656 .convert_ctx_access = bpf_convert_ctx_access,
10657 .gen_ld_abs = bpf_gen_ld_abs,
10660 const struct bpf_prog_ops sk_filter_prog_ops = {
10661 .test_run = bpf_prog_test_run_skb,
10664 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
10665 .get_func_proto = tc_cls_act_func_proto,
10666 .is_valid_access = tc_cls_act_is_valid_access,
10667 .convert_ctx_access = tc_cls_act_convert_ctx_access,
10668 .gen_prologue = tc_cls_act_prologue,
10669 .gen_ld_abs = bpf_gen_ld_abs,
10670 .btf_struct_access = tc_cls_act_btf_struct_access,
10673 const struct bpf_prog_ops tc_cls_act_prog_ops = {
10674 .test_run = bpf_prog_test_run_skb,
10677 const struct bpf_verifier_ops xdp_verifier_ops = {
10678 .get_func_proto = xdp_func_proto,
10679 .is_valid_access = xdp_is_valid_access,
10680 .convert_ctx_access = xdp_convert_ctx_access,
10681 .gen_prologue = bpf_noop_prologue,
10682 .btf_struct_access = xdp_btf_struct_access,
10685 const struct bpf_prog_ops xdp_prog_ops = {
10686 .test_run = bpf_prog_test_run_xdp,
10689 const struct bpf_verifier_ops cg_skb_verifier_ops = {
10690 .get_func_proto = cg_skb_func_proto,
10691 .is_valid_access = cg_skb_is_valid_access,
10692 .convert_ctx_access = bpf_convert_ctx_access,
10695 const struct bpf_prog_ops cg_skb_prog_ops = {
10696 .test_run = bpf_prog_test_run_skb,
10699 const struct bpf_verifier_ops lwt_in_verifier_ops = {
10700 .get_func_proto = lwt_in_func_proto,
10701 .is_valid_access = lwt_is_valid_access,
10702 .convert_ctx_access = bpf_convert_ctx_access,
10705 const struct bpf_prog_ops lwt_in_prog_ops = {
10706 .test_run = bpf_prog_test_run_skb,
10709 const struct bpf_verifier_ops lwt_out_verifier_ops = {
10710 .get_func_proto = lwt_out_func_proto,
10711 .is_valid_access = lwt_is_valid_access,
10712 .convert_ctx_access = bpf_convert_ctx_access,
10715 const struct bpf_prog_ops lwt_out_prog_ops = {
10716 .test_run = bpf_prog_test_run_skb,
10719 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
10720 .get_func_proto = lwt_xmit_func_proto,
10721 .is_valid_access = lwt_is_valid_access,
10722 .convert_ctx_access = bpf_convert_ctx_access,
10723 .gen_prologue = tc_cls_act_prologue,
10726 const struct bpf_prog_ops lwt_xmit_prog_ops = {
10727 .test_run = bpf_prog_test_run_skb,
10730 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
10731 .get_func_proto = lwt_seg6local_func_proto,
10732 .is_valid_access = lwt_is_valid_access,
10733 .convert_ctx_access = bpf_convert_ctx_access,
10736 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
10737 .test_run = bpf_prog_test_run_skb,
10740 const struct bpf_verifier_ops cg_sock_verifier_ops = {
10741 .get_func_proto = sock_filter_func_proto,
10742 .is_valid_access = sock_filter_is_valid_access,
10743 .convert_ctx_access = bpf_sock_convert_ctx_access,
10746 const struct bpf_prog_ops cg_sock_prog_ops = {
10749 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
10750 .get_func_proto = sock_addr_func_proto,
10751 .is_valid_access = sock_addr_is_valid_access,
10752 .convert_ctx_access = sock_addr_convert_ctx_access,
10755 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
10758 const struct bpf_verifier_ops sock_ops_verifier_ops = {
10759 .get_func_proto = sock_ops_func_proto,
10760 .is_valid_access = sock_ops_is_valid_access,
10761 .convert_ctx_access = sock_ops_convert_ctx_access,
10764 const struct bpf_prog_ops sock_ops_prog_ops = {
10767 const struct bpf_verifier_ops sk_skb_verifier_ops = {
10768 .get_func_proto = sk_skb_func_proto,
10769 .is_valid_access = sk_skb_is_valid_access,
10770 .convert_ctx_access = sk_skb_convert_ctx_access,
10771 .gen_prologue = sk_skb_prologue,
10774 const struct bpf_prog_ops sk_skb_prog_ops = {
10777 const struct bpf_verifier_ops sk_msg_verifier_ops = {
10778 .get_func_proto = sk_msg_func_proto,
10779 .is_valid_access = sk_msg_is_valid_access,
10780 .convert_ctx_access = sk_msg_convert_ctx_access,
10781 .gen_prologue = bpf_noop_prologue,
10784 const struct bpf_prog_ops sk_msg_prog_ops = {
10787 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
10788 .get_func_proto = flow_dissector_func_proto,
10789 .is_valid_access = flow_dissector_is_valid_access,
10790 .convert_ctx_access = flow_dissector_convert_ctx_access,
10793 const struct bpf_prog_ops flow_dissector_prog_ops = {
10794 .test_run = bpf_prog_test_run_flow_dissector,
10797 int sk_detach_filter(struct sock *sk)
10800 struct sk_filter *filter;
10802 if (sock_flag(sk, SOCK_FILTER_LOCKED))
10805 filter = rcu_dereference_protected(sk->sk_filter,
10806 lockdep_sock_is_held(sk));
10808 RCU_INIT_POINTER(sk->sk_filter, NULL);
10809 sk_filter_uncharge(sk, filter);
10815 EXPORT_SYMBOL_GPL(sk_detach_filter);
10817 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len)
10819 struct sock_fprog_kern *fprog;
10820 struct sk_filter *filter;
10823 sockopt_lock_sock(sk);
10824 filter = rcu_dereference_protected(sk->sk_filter,
10825 lockdep_sock_is_held(sk));
10829 /* We're copying the filter that has been originally attached,
10830 * so no conversion/decode needed anymore. eBPF programs that
10831 * have no original program cannot be dumped through this.
10834 fprog = filter->prog->orig_prog;
10840 /* User space only enquires number of filter blocks. */
10844 if (len < fprog->len)
10848 if (copy_to_sockptr(optval, fprog->filter, bpf_classic_proglen(fprog)))
10851 /* Instead of bytes, the API requests to return the number
10852 * of filter blocks.
10856 sockopt_release_sock(sk);
10861 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
10862 struct sock_reuseport *reuse,
10863 struct sock *sk, struct sk_buff *skb,
10864 struct sock *migrating_sk,
10867 reuse_kern->skb = skb;
10868 reuse_kern->sk = sk;
10869 reuse_kern->selected_sk = NULL;
10870 reuse_kern->migrating_sk = migrating_sk;
10871 reuse_kern->data_end = skb->data + skb_headlen(skb);
10872 reuse_kern->hash = hash;
10873 reuse_kern->reuseport_id = reuse->reuseport_id;
10874 reuse_kern->bind_inany = reuse->bind_inany;
10877 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
10878 struct bpf_prog *prog, struct sk_buff *skb,
10879 struct sock *migrating_sk,
10882 struct sk_reuseport_kern reuse_kern;
10883 enum sk_action action;
10885 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
10886 action = bpf_prog_run(prog, &reuse_kern);
10888 if (action == SK_PASS)
10889 return reuse_kern.selected_sk;
10891 return ERR_PTR(-ECONNREFUSED);
10894 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
10895 struct bpf_map *, map, void *, key, u32, flags)
10897 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
10898 struct sock_reuseport *reuse;
10899 struct sock *selected_sk;
10901 selected_sk = map->ops->map_lookup_elem(map, key);
10905 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
10907 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
10908 if (sk_is_refcounted(selected_sk))
10909 sock_put(selected_sk);
10911 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
10912 * The only (!reuse) case here is - the sk has already been
10913 * unhashed (e.g. by close()), so treat it as -ENOENT.
10915 * Other maps (e.g. sock_map) do not provide this guarantee and
10916 * the sk may never be in the reuseport group to begin with.
10918 return is_sockarray ? -ENOENT : -EINVAL;
10921 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
10922 struct sock *sk = reuse_kern->sk;
10924 if (sk->sk_protocol != selected_sk->sk_protocol)
10925 return -EPROTOTYPE;
10926 else if (sk->sk_family != selected_sk->sk_family)
10927 return -EAFNOSUPPORT;
10929 /* Catch all. Likely bound to a different sockaddr. */
10933 reuse_kern->selected_sk = selected_sk;
10938 static const struct bpf_func_proto sk_select_reuseport_proto = {
10939 .func = sk_select_reuseport,
10941 .ret_type = RET_INTEGER,
10942 .arg1_type = ARG_PTR_TO_CTX,
10943 .arg2_type = ARG_CONST_MAP_PTR,
10944 .arg3_type = ARG_PTR_TO_MAP_KEY,
10945 .arg4_type = ARG_ANYTHING,
10948 BPF_CALL_4(sk_reuseport_load_bytes,
10949 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10950 void *, to, u32, len)
10952 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
10955 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
10956 .func = sk_reuseport_load_bytes,
10958 .ret_type = RET_INTEGER,
10959 .arg1_type = ARG_PTR_TO_CTX,
10960 .arg2_type = ARG_ANYTHING,
10961 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
10962 .arg4_type = ARG_CONST_SIZE,
10965 BPF_CALL_5(sk_reuseport_load_bytes_relative,
10966 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10967 void *, to, u32, len, u32, start_header)
10969 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
10970 len, start_header);
10973 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
10974 .func = sk_reuseport_load_bytes_relative,
10976 .ret_type = RET_INTEGER,
10977 .arg1_type = ARG_PTR_TO_CTX,
10978 .arg2_type = ARG_ANYTHING,
10979 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
10980 .arg4_type = ARG_CONST_SIZE,
10981 .arg5_type = ARG_ANYTHING,
10984 static const struct bpf_func_proto *
10985 sk_reuseport_func_proto(enum bpf_func_id func_id,
10986 const struct bpf_prog *prog)
10989 case BPF_FUNC_sk_select_reuseport:
10990 return &sk_select_reuseport_proto;
10991 case BPF_FUNC_skb_load_bytes:
10992 return &sk_reuseport_load_bytes_proto;
10993 case BPF_FUNC_skb_load_bytes_relative:
10994 return &sk_reuseport_load_bytes_relative_proto;
10995 case BPF_FUNC_get_socket_cookie:
10996 return &bpf_get_socket_ptr_cookie_proto;
10997 case BPF_FUNC_ktime_get_coarse_ns:
10998 return &bpf_ktime_get_coarse_ns_proto;
11000 return bpf_base_func_proto(func_id);
11005 sk_reuseport_is_valid_access(int off, int size,
11006 enum bpf_access_type type,
11007 const struct bpf_prog *prog,
11008 struct bpf_insn_access_aux *info)
11010 const u32 size_default = sizeof(__u32);
11012 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
11013 off % size || type != BPF_READ)
11017 case offsetof(struct sk_reuseport_md, data):
11018 info->reg_type = PTR_TO_PACKET;
11019 return size == sizeof(__u64);
11021 case offsetof(struct sk_reuseport_md, data_end):
11022 info->reg_type = PTR_TO_PACKET_END;
11023 return size == sizeof(__u64);
11025 case offsetof(struct sk_reuseport_md, hash):
11026 return size == size_default;
11028 case offsetof(struct sk_reuseport_md, sk):
11029 info->reg_type = PTR_TO_SOCKET;
11030 return size == sizeof(__u64);
11032 case offsetof(struct sk_reuseport_md, migrating_sk):
11033 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
11034 return size == sizeof(__u64);
11036 /* Fields that allow narrowing */
11037 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
11038 if (size < sizeof_field(struct sk_buff, protocol))
11041 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
11042 case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
11043 case bpf_ctx_range(struct sk_reuseport_md, len):
11044 bpf_ctx_record_field_size(info, size_default);
11045 return bpf_ctx_narrow_access_ok(off, size, size_default);
11052 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
11053 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
11054 si->dst_reg, si->src_reg, \
11055 bpf_target_off(struct sk_reuseport_kern, F, \
11056 sizeof_field(struct sk_reuseport_kern, F), \
11060 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
11061 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
11066 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \
11067 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
11072 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
11073 const struct bpf_insn *si,
11074 struct bpf_insn *insn_buf,
11075 struct bpf_prog *prog,
11078 struct bpf_insn *insn = insn_buf;
11081 case offsetof(struct sk_reuseport_md, data):
11082 SK_REUSEPORT_LOAD_SKB_FIELD(data);
11085 case offsetof(struct sk_reuseport_md, len):
11086 SK_REUSEPORT_LOAD_SKB_FIELD(len);
11089 case offsetof(struct sk_reuseport_md, eth_protocol):
11090 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
11093 case offsetof(struct sk_reuseport_md, ip_protocol):
11094 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
11097 case offsetof(struct sk_reuseport_md, data_end):
11098 SK_REUSEPORT_LOAD_FIELD(data_end);
11101 case offsetof(struct sk_reuseport_md, hash):
11102 SK_REUSEPORT_LOAD_FIELD(hash);
11105 case offsetof(struct sk_reuseport_md, bind_inany):
11106 SK_REUSEPORT_LOAD_FIELD(bind_inany);
11109 case offsetof(struct sk_reuseport_md, sk):
11110 SK_REUSEPORT_LOAD_FIELD(sk);
11113 case offsetof(struct sk_reuseport_md, migrating_sk):
11114 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
11118 return insn - insn_buf;
11121 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
11122 .get_func_proto = sk_reuseport_func_proto,
11123 .is_valid_access = sk_reuseport_is_valid_access,
11124 .convert_ctx_access = sk_reuseport_convert_ctx_access,
11127 const struct bpf_prog_ops sk_reuseport_prog_ops = {
11130 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
11131 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
11133 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
11134 struct sock *, sk, u64, flags)
11136 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
11137 BPF_SK_LOOKUP_F_NO_REUSEPORT)))
11139 if (unlikely(sk && sk_is_refcounted(sk)))
11140 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
11141 if (unlikely(sk && sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN))
11142 return -ESOCKTNOSUPPORT; /* only accept TCP socket in LISTEN */
11143 if (unlikely(sk && sk_is_udp(sk) && sk->sk_state != TCP_CLOSE))
11144 return -ESOCKTNOSUPPORT; /* only accept UDP socket in CLOSE */
11146 /* Check if socket is suitable for packet L3/L4 protocol */
11147 if (sk && sk->sk_protocol != ctx->protocol)
11148 return -EPROTOTYPE;
11149 if (sk && sk->sk_family != ctx->family &&
11150 (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
11151 return -EAFNOSUPPORT;
11153 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
11156 /* Select socket as lookup result */
11157 ctx->selected_sk = sk;
11158 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
11162 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
11163 .func = bpf_sk_lookup_assign,
11165 .ret_type = RET_INTEGER,
11166 .arg1_type = ARG_PTR_TO_CTX,
11167 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL,
11168 .arg3_type = ARG_ANYTHING,
11171 static const struct bpf_func_proto *
11172 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
11175 case BPF_FUNC_perf_event_output:
11176 return &bpf_event_output_data_proto;
11177 case BPF_FUNC_sk_assign:
11178 return &bpf_sk_lookup_assign_proto;
11179 case BPF_FUNC_sk_release:
11180 return &bpf_sk_release_proto;
11182 return bpf_sk_base_func_proto(func_id);
11186 static bool sk_lookup_is_valid_access(int off, int size,
11187 enum bpf_access_type type,
11188 const struct bpf_prog *prog,
11189 struct bpf_insn_access_aux *info)
11191 if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
11193 if (off % size != 0)
11195 if (type != BPF_READ)
11199 case offsetof(struct bpf_sk_lookup, sk):
11200 info->reg_type = PTR_TO_SOCKET_OR_NULL;
11201 return size == sizeof(__u64);
11203 case bpf_ctx_range(struct bpf_sk_lookup, family):
11204 case bpf_ctx_range(struct bpf_sk_lookup, protocol):
11205 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
11206 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
11207 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
11208 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
11209 case bpf_ctx_range(struct bpf_sk_lookup, local_port):
11210 case bpf_ctx_range(struct bpf_sk_lookup, ingress_ifindex):
11211 bpf_ctx_record_field_size(info, sizeof(__u32));
11212 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
11214 case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
11215 /* Allow 4-byte access to 2-byte field for backward compatibility */
11216 if (size == sizeof(__u32))
11218 bpf_ctx_record_field_size(info, sizeof(__be16));
11219 return bpf_ctx_narrow_access_ok(off, size, sizeof(__be16));
11221 case offsetofend(struct bpf_sk_lookup, remote_port) ...
11222 offsetof(struct bpf_sk_lookup, local_ip4) - 1:
11223 /* Allow access to zero padding for backward compatibility */
11224 bpf_ctx_record_field_size(info, sizeof(__u16));
11225 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u16));
11232 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
11233 const struct bpf_insn *si,
11234 struct bpf_insn *insn_buf,
11235 struct bpf_prog *prog,
11238 struct bpf_insn *insn = insn_buf;
11241 case offsetof(struct bpf_sk_lookup, sk):
11242 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11243 offsetof(struct bpf_sk_lookup_kern, selected_sk));
11246 case offsetof(struct bpf_sk_lookup, family):
11247 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11248 bpf_target_off(struct bpf_sk_lookup_kern,
11249 family, 2, target_size));
11252 case offsetof(struct bpf_sk_lookup, protocol):
11253 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11254 bpf_target_off(struct bpf_sk_lookup_kern,
11255 protocol, 2, target_size));
11258 case offsetof(struct bpf_sk_lookup, remote_ip4):
11259 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11260 bpf_target_off(struct bpf_sk_lookup_kern,
11261 v4.saddr, 4, target_size));
11264 case offsetof(struct bpf_sk_lookup, local_ip4):
11265 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11266 bpf_target_off(struct bpf_sk_lookup_kern,
11267 v4.daddr, 4, target_size));
11270 case bpf_ctx_range_till(struct bpf_sk_lookup,
11271 remote_ip6[0], remote_ip6[3]): {
11272 #if IS_ENABLED(CONFIG_IPV6)
11275 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
11276 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11277 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11278 offsetof(struct bpf_sk_lookup_kern, v6.saddr));
11279 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11280 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11282 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11286 case bpf_ctx_range_till(struct bpf_sk_lookup,
11287 local_ip6[0], local_ip6[3]): {
11288 #if IS_ENABLED(CONFIG_IPV6)
11291 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
11292 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11293 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11294 offsetof(struct bpf_sk_lookup_kern, v6.daddr));
11295 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11296 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11298 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11302 case offsetof(struct bpf_sk_lookup, remote_port):
11303 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11304 bpf_target_off(struct bpf_sk_lookup_kern,
11305 sport, 2, target_size));
11308 case offsetofend(struct bpf_sk_lookup, remote_port):
11310 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11313 case offsetof(struct bpf_sk_lookup, local_port):
11314 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11315 bpf_target_off(struct bpf_sk_lookup_kern,
11316 dport, 2, target_size));
11319 case offsetof(struct bpf_sk_lookup, ingress_ifindex):
11320 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11321 bpf_target_off(struct bpf_sk_lookup_kern,
11322 ingress_ifindex, 4, target_size));
11326 return insn - insn_buf;
11329 const struct bpf_prog_ops sk_lookup_prog_ops = {
11330 .test_run = bpf_prog_test_run_sk_lookup,
11333 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
11334 .get_func_proto = sk_lookup_func_proto,
11335 .is_valid_access = sk_lookup_is_valid_access,
11336 .convert_ctx_access = sk_lookup_convert_ctx_access,
11339 #endif /* CONFIG_INET */
11341 DEFINE_BPF_DISPATCHER(xdp)
11343 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
11345 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
11348 BTF_ID_LIST_GLOBAL(btf_sock_ids, MAX_BTF_SOCK_TYPE)
11349 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
11351 #undef BTF_SOCK_TYPE
11353 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
11355 /* tcp6_sock type is not generated in dwarf and hence btf,
11356 * trigger an explicit type generation here.
11358 BTF_TYPE_EMIT(struct tcp6_sock);
11359 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
11360 sk->sk_family == AF_INET6)
11361 return (unsigned long)sk;
11363 return (unsigned long)NULL;
11366 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
11367 .func = bpf_skc_to_tcp6_sock,
11369 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11370 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11371 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
11374 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
11376 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
11377 return (unsigned long)sk;
11379 return (unsigned long)NULL;
11382 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
11383 .func = bpf_skc_to_tcp_sock,
11385 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11386 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11387 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
11390 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
11392 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
11393 * generated if CONFIG_INET=n. Trigger an explicit generation here.
11395 BTF_TYPE_EMIT(struct inet_timewait_sock);
11396 BTF_TYPE_EMIT(struct tcp_timewait_sock);
11399 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
11400 return (unsigned long)sk;
11403 #if IS_BUILTIN(CONFIG_IPV6)
11404 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
11405 return (unsigned long)sk;
11408 return (unsigned long)NULL;
11411 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
11412 .func = bpf_skc_to_tcp_timewait_sock,
11414 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11415 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11416 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
11419 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
11422 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11423 return (unsigned long)sk;
11426 #if IS_BUILTIN(CONFIG_IPV6)
11427 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11428 return (unsigned long)sk;
11431 return (unsigned long)NULL;
11434 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
11435 .func = bpf_skc_to_tcp_request_sock,
11437 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11438 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11439 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
11442 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
11444 /* udp6_sock type is not generated in dwarf and hence btf,
11445 * trigger an explicit type generation here.
11447 BTF_TYPE_EMIT(struct udp6_sock);
11448 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
11449 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
11450 return (unsigned long)sk;
11452 return (unsigned long)NULL;
11455 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
11456 .func = bpf_skc_to_udp6_sock,
11458 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11459 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11460 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
11463 BPF_CALL_1(bpf_skc_to_unix_sock, struct sock *, sk)
11465 /* unix_sock type is not generated in dwarf and hence btf,
11466 * trigger an explicit type generation here.
11468 BTF_TYPE_EMIT(struct unix_sock);
11469 if (sk && sk_fullsock(sk) && sk->sk_family == AF_UNIX)
11470 return (unsigned long)sk;
11472 return (unsigned long)NULL;
11475 const struct bpf_func_proto bpf_skc_to_unix_sock_proto = {
11476 .func = bpf_skc_to_unix_sock,
11478 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11479 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11480 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UNIX],
11483 BPF_CALL_1(bpf_skc_to_mptcp_sock, struct sock *, sk)
11485 BTF_TYPE_EMIT(struct mptcp_sock);
11486 return (unsigned long)bpf_mptcp_sock_from_subflow(sk);
11489 const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto = {
11490 .func = bpf_skc_to_mptcp_sock,
11492 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11493 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
11494 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_MPTCP],
11497 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
11499 return (unsigned long)sock_from_file(file);
11502 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
11503 BTF_ID(struct, socket)
11504 BTF_ID(struct, file)
11506 const struct bpf_func_proto bpf_sock_from_file_proto = {
11507 .func = bpf_sock_from_file,
11509 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11510 .ret_btf_id = &bpf_sock_from_file_btf_ids[0],
11511 .arg1_type = ARG_PTR_TO_BTF_ID,
11512 .arg1_btf_id = &bpf_sock_from_file_btf_ids[1],
11515 static const struct bpf_func_proto *
11516 bpf_sk_base_func_proto(enum bpf_func_id func_id)
11518 const struct bpf_func_proto *func;
11521 case BPF_FUNC_skc_to_tcp6_sock:
11522 func = &bpf_skc_to_tcp6_sock_proto;
11524 case BPF_FUNC_skc_to_tcp_sock:
11525 func = &bpf_skc_to_tcp_sock_proto;
11527 case BPF_FUNC_skc_to_tcp_timewait_sock:
11528 func = &bpf_skc_to_tcp_timewait_sock_proto;
11530 case BPF_FUNC_skc_to_tcp_request_sock:
11531 func = &bpf_skc_to_tcp_request_sock_proto;
11533 case BPF_FUNC_skc_to_udp6_sock:
11534 func = &bpf_skc_to_udp6_sock_proto;
11536 case BPF_FUNC_skc_to_unix_sock:
11537 func = &bpf_skc_to_unix_sock_proto;
11539 case BPF_FUNC_skc_to_mptcp_sock:
11540 func = &bpf_skc_to_mptcp_sock_proto;
11542 case BPF_FUNC_ktime_get_coarse_ns:
11543 return &bpf_ktime_get_coarse_ns_proto;
11545 return bpf_base_func_proto(func_id);
11548 if (!perfmon_capable())