1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Linux Socket Filter - Kernel level socket filtering
5 * Based on the design of the Berkeley Packet Filter. The new
6 * internal format has been designed by PLUMgrid:
8 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
12 * Jay Schulist <jschlst@samba.org>
13 * Alexei Starovoitov <ast@plumgrid.com>
14 * Daniel Borkmann <dborkman@redhat.com>
16 * Andi Kleen - Fix a few bad bugs and races.
17 * Kris Katterjohn - Added many additional checks in bpf_check_classic()
20 #include <linux/atomic.h>
21 #include <linux/bpf_verifier.h>
22 #include <linux/module.h>
23 #include <linux/types.h>
25 #include <linux/fcntl.h>
26 #include <linux/socket.h>
27 #include <linux/sock_diag.h>
29 #include <linux/inet.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_packet.h>
32 #include <linux/if_arp.h>
33 #include <linux/gfp.h>
34 #include <net/inet_common.h>
36 #include <net/protocol.h>
37 #include <net/netlink.h>
38 #include <linux/skbuff.h>
39 #include <linux/skmsg.h>
41 #include <net/flow_dissector.h>
42 #include <linux/errno.h>
43 #include <linux/timer.h>
44 #include <linux/uaccess.h>
45 #include <asm/unaligned.h>
46 #include <linux/filter.h>
47 #include <linux/ratelimit.h>
48 #include <linux/seccomp.h>
49 #include <linux/if_vlan.h>
50 #include <linux/bpf.h>
51 #include <linux/btf.h>
52 #include <net/sch_generic.h>
53 #include <net/cls_cgroup.h>
54 #include <net/dst_metadata.h>
56 #include <net/sock_reuseport.h>
57 #include <net/busy_poll.h>
61 #include <linux/bpf_trace.h>
62 #include <net/xdp_sock.h>
63 #include <linux/inetdevice.h>
64 #include <net/inet_hashtables.h>
65 #include <net/inet6_hashtables.h>
66 #include <net/ip_fib.h>
67 #include <net/nexthop.h>
71 #include <net/net_namespace.h>
72 #include <linux/seg6_local.h>
74 #include <net/seg6_local.h>
75 #include <net/lwtunnel.h>
76 #include <net/ipv6_stubs.h>
77 #include <net/bpf_sk_storage.h>
78 #include <net/transp_v6.h>
79 #include <linux/btf_ids.h>
82 #include <net/mptcp.h>
83 #include <net/netfilter/nf_conntrack_bpf.h>
85 static const struct bpf_func_proto *
86 bpf_sk_base_func_proto(enum bpf_func_id func_id);
88 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len)
90 if (in_compat_syscall()) {
91 struct compat_sock_fprog f32;
93 if (len != sizeof(f32))
95 if (copy_from_sockptr(&f32, src, sizeof(f32)))
97 memset(dst, 0, sizeof(*dst));
99 dst->filter = compat_ptr(f32.filter);
101 if (len != sizeof(*dst))
103 if (copy_from_sockptr(dst, src, sizeof(*dst)))
109 EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user);
112 * sk_filter_trim_cap - run a packet through a socket filter
113 * @sk: sock associated with &sk_buff
114 * @skb: buffer to filter
115 * @cap: limit on how short the eBPF program may trim the packet
117 * Run the eBPF program and then cut skb->data to correct size returned by
118 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
119 * than pkt_len we keep whole skb->data. This is the socket level
120 * wrapper to bpf_prog_run. It returns 0 if the packet should
121 * be accepted or -EPERM if the packet should be tossed.
124 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
127 struct sk_filter *filter;
130 * If the skb was allocated from pfmemalloc reserves, only
131 * allow SOCK_MEMALLOC sockets to use it as this socket is
132 * helping free memory
134 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
135 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
138 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
142 err = security_sock_rcv_skb(sk, skb);
147 filter = rcu_dereference(sk->sk_filter);
149 struct sock *save_sk = skb->sk;
150 unsigned int pkt_len;
153 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
155 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
161 EXPORT_SYMBOL(sk_filter_trim_cap);
163 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
165 return skb_get_poff(skb);
168 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
172 if (skb_is_nonlinear(skb))
175 if (skb->len < sizeof(struct nlattr))
178 if (a > skb->len - sizeof(struct nlattr))
181 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
183 return (void *) nla - (void *) skb->data;
188 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
192 if (skb_is_nonlinear(skb))
195 if (skb->len < sizeof(struct nlattr))
198 if (a > skb->len - sizeof(struct nlattr))
201 nla = (struct nlattr *) &skb->data[a];
202 if (nla->nla_len > skb->len - a)
205 nla = nla_find_nested(nla, x);
207 return (void *) nla - (void *) skb->data;
212 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
213 data, int, headlen, int, offset)
216 const int len = sizeof(tmp);
219 if (headlen - offset >= len)
220 return *(u8 *)(data + offset);
221 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
224 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
232 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
235 return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
239 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
240 data, int, headlen, int, offset)
243 const int len = sizeof(tmp);
246 if (headlen - offset >= len)
247 return get_unaligned_be16(data + offset);
248 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
249 return be16_to_cpu(tmp);
251 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
253 return get_unaligned_be16(ptr);
259 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
262 return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
266 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
267 data, int, headlen, int, offset)
270 const int len = sizeof(tmp);
272 if (likely(offset >= 0)) {
273 if (headlen - offset >= len)
274 return get_unaligned_be32(data + offset);
275 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
276 return be32_to_cpu(tmp);
278 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
280 return get_unaligned_be32(ptr);
286 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
289 return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
293 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
294 struct bpf_insn *insn_buf)
296 struct bpf_insn *insn = insn_buf;
300 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4);
302 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
303 offsetof(struct sk_buff, mark));
307 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET);
308 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
309 #ifdef __BIG_ENDIAN_BITFIELD
310 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
315 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2);
317 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
318 offsetof(struct sk_buff, queue_mapping));
321 case SKF_AD_VLAN_TAG:
322 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2);
324 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
325 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
326 offsetof(struct sk_buff, vlan_tci));
328 case SKF_AD_VLAN_TAG_PRESENT:
329 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_VLAN_PRESENT_OFFSET);
330 if (PKT_VLAN_PRESENT_BIT)
331 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, PKT_VLAN_PRESENT_BIT);
332 if (PKT_VLAN_PRESENT_BIT < 7)
333 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
337 return insn - insn_buf;
340 static bool convert_bpf_extensions(struct sock_filter *fp,
341 struct bpf_insn **insnp)
343 struct bpf_insn *insn = *insnp;
347 case SKF_AD_OFF + SKF_AD_PROTOCOL:
348 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2);
350 /* A = *(u16 *) (CTX + offsetof(protocol)) */
351 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
352 offsetof(struct sk_buff, protocol));
353 /* A = ntohs(A) [emitting a nop or swap16] */
354 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
357 case SKF_AD_OFF + SKF_AD_PKTTYPE:
358 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
362 case SKF_AD_OFF + SKF_AD_IFINDEX:
363 case SKF_AD_OFF + SKF_AD_HATYPE:
364 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4);
365 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2);
367 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
368 BPF_REG_TMP, BPF_REG_CTX,
369 offsetof(struct sk_buff, dev));
370 /* if (tmp != 0) goto pc + 1 */
371 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
372 *insn++ = BPF_EXIT_INSN();
373 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
374 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
375 offsetof(struct net_device, ifindex));
377 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
378 offsetof(struct net_device, type));
381 case SKF_AD_OFF + SKF_AD_MARK:
382 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
386 case SKF_AD_OFF + SKF_AD_RXHASH:
387 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4);
389 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
390 offsetof(struct sk_buff, hash));
393 case SKF_AD_OFF + SKF_AD_QUEUE:
394 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
398 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
399 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
400 BPF_REG_A, BPF_REG_CTX, insn);
404 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
405 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
406 BPF_REG_A, BPF_REG_CTX, insn);
410 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
411 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2);
413 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
414 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
415 offsetof(struct sk_buff, vlan_proto));
416 /* A = ntohs(A) [emitting a nop or swap16] */
417 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
420 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
421 case SKF_AD_OFF + SKF_AD_NLATTR:
422 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
423 case SKF_AD_OFF + SKF_AD_CPU:
424 case SKF_AD_OFF + SKF_AD_RANDOM:
426 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
428 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
430 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
431 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
433 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
434 *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
436 case SKF_AD_OFF + SKF_AD_NLATTR:
437 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
439 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
440 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
442 case SKF_AD_OFF + SKF_AD_CPU:
443 *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
445 case SKF_AD_OFF + SKF_AD_RANDOM:
446 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
447 bpf_user_rnd_init_once();
452 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
454 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
458 /* This is just a dummy call to avoid letting the compiler
459 * evict __bpf_call_base() as an optimization. Placed here
460 * where no-one bothers.
462 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
470 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
472 const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
473 int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
474 bool endian = BPF_SIZE(fp->code) == BPF_H ||
475 BPF_SIZE(fp->code) == BPF_W;
476 bool indirect = BPF_MODE(fp->code) == BPF_IND;
477 const int ip_align = NET_IP_ALIGN;
478 struct bpf_insn *insn = *insnp;
482 ((unaligned_ok && offset >= 0) ||
483 (!unaligned_ok && offset >= 0 &&
484 offset + ip_align >= 0 &&
485 offset + ip_align % size == 0))) {
486 bool ldx_off_ok = offset <= S16_MAX;
488 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
490 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
491 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
492 size, 2 + endian + (!ldx_off_ok * 2));
494 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
497 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
498 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
499 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
503 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
504 *insn++ = BPF_JMP_A(8);
507 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
508 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
509 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
511 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
513 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
515 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
518 switch (BPF_SIZE(fp->code)) {
520 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
523 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
526 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
532 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
533 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
534 *insn = BPF_EXIT_INSN();
541 * bpf_convert_filter - convert filter program
542 * @prog: the user passed filter program
543 * @len: the length of the user passed filter program
544 * @new_prog: allocated 'struct bpf_prog' or NULL
545 * @new_len: pointer to store length of converted program
546 * @seen_ld_abs: bool whether we've seen ld_abs/ind
548 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
549 * style extended BPF (eBPF).
550 * Conversion workflow:
552 * 1) First pass for calculating the new program length:
553 * bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
555 * 2) 2nd pass to remap in two passes: 1st pass finds new
556 * jump offsets, 2nd pass remapping:
557 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
559 static int bpf_convert_filter(struct sock_filter *prog, int len,
560 struct bpf_prog *new_prog, int *new_len,
563 int new_flen = 0, pass = 0, target, i, stack_off;
564 struct bpf_insn *new_insn, *first_insn = NULL;
565 struct sock_filter *fp;
569 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
570 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
572 if (len <= 0 || len > BPF_MAXINSNS)
576 first_insn = new_prog->insnsi;
577 addrs = kcalloc(len, sizeof(*addrs),
578 GFP_KERNEL | __GFP_NOWARN);
584 new_insn = first_insn;
587 /* Classic BPF related prologue emission. */
589 /* Classic BPF expects A and X to be reset first. These need
590 * to be guaranteed to be the first two instructions.
592 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
593 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
595 /* All programs must keep CTX in callee saved BPF_REG_CTX.
596 * In eBPF case it's done by the compiler, here we need to
597 * do this ourself. Initial CTX is present in BPF_REG_ARG1.
599 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
601 /* For packet access in classic BPF, cache skb->data
602 * in callee-saved BPF R8 and skb->len - skb->data_len
603 * (headlen) in BPF R9. Since classic BPF is read-only
604 * on CTX, we only need to cache it once.
606 *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
607 BPF_REG_D, BPF_REG_CTX,
608 offsetof(struct sk_buff, data));
609 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
610 offsetof(struct sk_buff, len));
611 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
612 offsetof(struct sk_buff, data_len));
613 *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
619 for (i = 0; i < len; fp++, i++) {
620 struct bpf_insn tmp_insns[32] = { };
621 struct bpf_insn *insn = tmp_insns;
624 addrs[i] = new_insn - first_insn;
627 /* All arithmetic insns and skb loads map as-is. */
628 case BPF_ALU | BPF_ADD | BPF_X:
629 case BPF_ALU | BPF_ADD | BPF_K:
630 case BPF_ALU | BPF_SUB | BPF_X:
631 case BPF_ALU | BPF_SUB | BPF_K:
632 case BPF_ALU | BPF_AND | BPF_X:
633 case BPF_ALU | BPF_AND | BPF_K:
634 case BPF_ALU | BPF_OR | BPF_X:
635 case BPF_ALU | BPF_OR | BPF_K:
636 case BPF_ALU | BPF_LSH | BPF_X:
637 case BPF_ALU | BPF_LSH | BPF_K:
638 case BPF_ALU | BPF_RSH | BPF_X:
639 case BPF_ALU | BPF_RSH | BPF_K:
640 case BPF_ALU | BPF_XOR | BPF_X:
641 case BPF_ALU | BPF_XOR | BPF_K:
642 case BPF_ALU | BPF_MUL | BPF_X:
643 case BPF_ALU | BPF_MUL | BPF_K:
644 case BPF_ALU | BPF_DIV | BPF_X:
645 case BPF_ALU | BPF_DIV | BPF_K:
646 case BPF_ALU | BPF_MOD | BPF_X:
647 case BPF_ALU | BPF_MOD | BPF_K:
648 case BPF_ALU | BPF_NEG:
649 case BPF_LD | BPF_ABS | BPF_W:
650 case BPF_LD | BPF_ABS | BPF_H:
651 case BPF_LD | BPF_ABS | BPF_B:
652 case BPF_LD | BPF_IND | BPF_W:
653 case BPF_LD | BPF_IND | BPF_H:
654 case BPF_LD | BPF_IND | BPF_B:
655 /* Check for overloaded BPF extension and
656 * directly convert it if found, otherwise
657 * just move on with mapping.
659 if (BPF_CLASS(fp->code) == BPF_LD &&
660 BPF_MODE(fp->code) == BPF_ABS &&
661 convert_bpf_extensions(fp, &insn))
663 if (BPF_CLASS(fp->code) == BPF_LD &&
664 convert_bpf_ld_abs(fp, &insn)) {
669 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
670 fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
671 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
672 /* Error with exception code on div/mod by 0.
673 * For cBPF programs, this was always return 0.
675 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
676 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
677 *insn++ = BPF_EXIT_INSN();
680 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
683 /* Jump transformation cannot use BPF block macros
684 * everywhere as offset calculation and target updates
685 * require a bit more work than the rest, i.e. jump
686 * opcodes map as-is, but offsets need adjustment.
689 #define BPF_EMIT_JMP \
691 const s32 off_min = S16_MIN, off_max = S16_MAX; \
694 if (target >= len || target < 0) \
696 off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
697 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
698 off -= insn - tmp_insns; \
699 /* Reject anything not fitting into insn->off. */ \
700 if (off < off_min || off > off_max) \
705 case BPF_JMP | BPF_JA:
706 target = i + fp->k + 1;
707 insn->code = fp->code;
711 case BPF_JMP | BPF_JEQ | BPF_K:
712 case BPF_JMP | BPF_JEQ | BPF_X:
713 case BPF_JMP | BPF_JSET | BPF_K:
714 case BPF_JMP | BPF_JSET | BPF_X:
715 case BPF_JMP | BPF_JGT | BPF_K:
716 case BPF_JMP | BPF_JGT | BPF_X:
717 case BPF_JMP | BPF_JGE | BPF_K:
718 case BPF_JMP | BPF_JGE | BPF_X:
719 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
720 /* BPF immediates are signed, zero extend
721 * immediate into tmp register and use it
724 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
726 insn->dst_reg = BPF_REG_A;
727 insn->src_reg = BPF_REG_TMP;
730 insn->dst_reg = BPF_REG_A;
732 bpf_src = BPF_SRC(fp->code);
733 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
736 /* Common case where 'jump_false' is next insn. */
738 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
739 target = i + fp->jt + 1;
744 /* Convert some jumps when 'jump_true' is next insn. */
746 switch (BPF_OP(fp->code)) {
748 insn->code = BPF_JMP | BPF_JNE | bpf_src;
751 insn->code = BPF_JMP | BPF_JLE | bpf_src;
754 insn->code = BPF_JMP | BPF_JLT | bpf_src;
760 target = i + fp->jf + 1;
765 /* Other jumps are mapped into two insns: Jxx and JA. */
766 target = i + fp->jt + 1;
767 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
771 insn->code = BPF_JMP | BPF_JA;
772 target = i + fp->jf + 1;
776 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
777 case BPF_LDX | BPF_MSH | BPF_B: {
778 struct sock_filter tmp = {
779 .code = BPF_LD | BPF_ABS | BPF_B,
786 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
787 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
788 convert_bpf_ld_abs(&tmp, &insn);
791 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
793 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
795 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
797 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
799 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
802 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
803 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
805 case BPF_RET | BPF_A:
806 case BPF_RET | BPF_K:
807 if (BPF_RVAL(fp->code) == BPF_K)
808 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
810 *insn = BPF_EXIT_INSN();
813 /* Store to stack. */
816 stack_off = fp->k * 4 + 4;
817 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
818 BPF_ST ? BPF_REG_A : BPF_REG_X,
820 /* check_load_and_stores() verifies that classic BPF can
821 * load from stack only after write, so tracking
822 * stack_depth for ST|STX insns is enough
824 if (new_prog && new_prog->aux->stack_depth < stack_off)
825 new_prog->aux->stack_depth = stack_off;
828 /* Load from stack. */
829 case BPF_LD | BPF_MEM:
830 case BPF_LDX | BPF_MEM:
831 stack_off = fp->k * 4 + 4;
832 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
833 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
838 case BPF_LD | BPF_IMM:
839 case BPF_LDX | BPF_IMM:
840 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
841 BPF_REG_A : BPF_REG_X, fp->k);
845 case BPF_MISC | BPF_TAX:
846 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
850 case BPF_MISC | BPF_TXA:
851 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
854 /* A = skb->len or X = skb->len */
855 case BPF_LD | BPF_W | BPF_LEN:
856 case BPF_LDX | BPF_W | BPF_LEN:
857 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
858 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
859 offsetof(struct sk_buff, len));
862 /* Access seccomp_data fields. */
863 case BPF_LDX | BPF_ABS | BPF_W:
864 /* A = *(u32 *) (ctx + K) */
865 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
868 /* Unknown instruction. */
875 memcpy(new_insn, tmp_insns,
876 sizeof(*insn) * (insn - tmp_insns));
877 new_insn += insn - tmp_insns;
881 /* Only calculating new length. */
882 *new_len = new_insn - first_insn;
884 *new_len += 4; /* Prologue bits. */
889 if (new_flen != new_insn - first_insn) {
890 new_flen = new_insn - first_insn;
897 BUG_ON(*new_len != new_flen);
906 * As we dont want to clear mem[] array for each packet going through
907 * __bpf_prog_run(), we check that filter loaded by user never try to read
908 * a cell if not previously written, and we check all branches to be sure
909 * a malicious user doesn't try to abuse us.
911 static int check_load_and_stores(const struct sock_filter *filter, int flen)
913 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
916 BUILD_BUG_ON(BPF_MEMWORDS > 16);
918 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
922 memset(masks, 0xff, flen * sizeof(*masks));
924 for (pc = 0; pc < flen; pc++) {
925 memvalid &= masks[pc];
927 switch (filter[pc].code) {
930 memvalid |= (1 << filter[pc].k);
932 case BPF_LD | BPF_MEM:
933 case BPF_LDX | BPF_MEM:
934 if (!(memvalid & (1 << filter[pc].k))) {
939 case BPF_JMP | BPF_JA:
940 /* A jump must set masks on target */
941 masks[pc + 1 + filter[pc].k] &= memvalid;
944 case BPF_JMP | BPF_JEQ | BPF_K:
945 case BPF_JMP | BPF_JEQ | BPF_X:
946 case BPF_JMP | BPF_JGE | BPF_K:
947 case BPF_JMP | BPF_JGE | BPF_X:
948 case BPF_JMP | BPF_JGT | BPF_K:
949 case BPF_JMP | BPF_JGT | BPF_X:
950 case BPF_JMP | BPF_JSET | BPF_K:
951 case BPF_JMP | BPF_JSET | BPF_X:
952 /* A jump must set masks on targets */
953 masks[pc + 1 + filter[pc].jt] &= memvalid;
954 masks[pc + 1 + filter[pc].jf] &= memvalid;
964 static bool chk_code_allowed(u16 code_to_probe)
966 static const bool codes[] = {
967 /* 32 bit ALU operations */
968 [BPF_ALU | BPF_ADD | BPF_K] = true,
969 [BPF_ALU | BPF_ADD | BPF_X] = true,
970 [BPF_ALU | BPF_SUB | BPF_K] = true,
971 [BPF_ALU | BPF_SUB | BPF_X] = true,
972 [BPF_ALU | BPF_MUL | BPF_K] = true,
973 [BPF_ALU | BPF_MUL | BPF_X] = true,
974 [BPF_ALU | BPF_DIV | BPF_K] = true,
975 [BPF_ALU | BPF_DIV | BPF_X] = true,
976 [BPF_ALU | BPF_MOD | BPF_K] = true,
977 [BPF_ALU | BPF_MOD | BPF_X] = true,
978 [BPF_ALU | BPF_AND | BPF_K] = true,
979 [BPF_ALU | BPF_AND | BPF_X] = true,
980 [BPF_ALU | BPF_OR | BPF_K] = true,
981 [BPF_ALU | BPF_OR | BPF_X] = true,
982 [BPF_ALU | BPF_XOR | BPF_K] = true,
983 [BPF_ALU | BPF_XOR | BPF_X] = true,
984 [BPF_ALU | BPF_LSH | BPF_K] = true,
985 [BPF_ALU | BPF_LSH | BPF_X] = true,
986 [BPF_ALU | BPF_RSH | BPF_K] = true,
987 [BPF_ALU | BPF_RSH | BPF_X] = true,
988 [BPF_ALU | BPF_NEG] = true,
989 /* Load instructions */
990 [BPF_LD | BPF_W | BPF_ABS] = true,
991 [BPF_LD | BPF_H | BPF_ABS] = true,
992 [BPF_LD | BPF_B | BPF_ABS] = true,
993 [BPF_LD | BPF_W | BPF_LEN] = true,
994 [BPF_LD | BPF_W | BPF_IND] = true,
995 [BPF_LD | BPF_H | BPF_IND] = true,
996 [BPF_LD | BPF_B | BPF_IND] = true,
997 [BPF_LD | BPF_IMM] = true,
998 [BPF_LD | BPF_MEM] = true,
999 [BPF_LDX | BPF_W | BPF_LEN] = true,
1000 [BPF_LDX | BPF_B | BPF_MSH] = true,
1001 [BPF_LDX | BPF_IMM] = true,
1002 [BPF_LDX | BPF_MEM] = true,
1003 /* Store instructions */
1006 /* Misc instructions */
1007 [BPF_MISC | BPF_TAX] = true,
1008 [BPF_MISC | BPF_TXA] = true,
1009 /* Return instructions */
1010 [BPF_RET | BPF_K] = true,
1011 [BPF_RET | BPF_A] = true,
1012 /* Jump instructions */
1013 [BPF_JMP | BPF_JA] = true,
1014 [BPF_JMP | BPF_JEQ | BPF_K] = true,
1015 [BPF_JMP | BPF_JEQ | BPF_X] = true,
1016 [BPF_JMP | BPF_JGE | BPF_K] = true,
1017 [BPF_JMP | BPF_JGE | BPF_X] = true,
1018 [BPF_JMP | BPF_JGT | BPF_K] = true,
1019 [BPF_JMP | BPF_JGT | BPF_X] = true,
1020 [BPF_JMP | BPF_JSET | BPF_K] = true,
1021 [BPF_JMP | BPF_JSET | BPF_X] = true,
1024 if (code_to_probe >= ARRAY_SIZE(codes))
1027 return codes[code_to_probe];
1030 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1035 if (flen == 0 || flen > BPF_MAXINSNS)
1042 * bpf_check_classic - verify socket filter code
1043 * @filter: filter to verify
1044 * @flen: length of filter
1046 * Check the user's filter code. If we let some ugly
1047 * filter code slip through kaboom! The filter must contain
1048 * no references or jumps that are out of range, no illegal
1049 * instructions, and must end with a RET instruction.
1051 * All jumps are forward as they are not signed.
1053 * Returns 0 if the rule set is legal or -EINVAL if not.
1055 static int bpf_check_classic(const struct sock_filter *filter,
1061 /* Check the filter code now */
1062 for (pc = 0; pc < flen; pc++) {
1063 const struct sock_filter *ftest = &filter[pc];
1065 /* May we actually operate on this code? */
1066 if (!chk_code_allowed(ftest->code))
1069 /* Some instructions need special checks */
1070 switch (ftest->code) {
1071 case BPF_ALU | BPF_DIV | BPF_K:
1072 case BPF_ALU | BPF_MOD | BPF_K:
1073 /* Check for division by zero */
1077 case BPF_ALU | BPF_LSH | BPF_K:
1078 case BPF_ALU | BPF_RSH | BPF_K:
1082 case BPF_LD | BPF_MEM:
1083 case BPF_LDX | BPF_MEM:
1086 /* Check for invalid memory addresses */
1087 if (ftest->k >= BPF_MEMWORDS)
1090 case BPF_JMP | BPF_JA:
1091 /* Note, the large ftest->k might cause loops.
1092 * Compare this with conditional jumps below,
1093 * where offsets are limited. --ANK (981016)
1095 if (ftest->k >= (unsigned int)(flen - pc - 1))
1098 case BPF_JMP | BPF_JEQ | BPF_K:
1099 case BPF_JMP | BPF_JEQ | BPF_X:
1100 case BPF_JMP | BPF_JGE | BPF_K:
1101 case BPF_JMP | BPF_JGE | BPF_X:
1102 case BPF_JMP | BPF_JGT | BPF_K:
1103 case BPF_JMP | BPF_JGT | BPF_X:
1104 case BPF_JMP | BPF_JSET | BPF_K:
1105 case BPF_JMP | BPF_JSET | BPF_X:
1106 /* Both conditionals must be safe */
1107 if (pc + ftest->jt + 1 >= flen ||
1108 pc + ftest->jf + 1 >= flen)
1111 case BPF_LD | BPF_W | BPF_ABS:
1112 case BPF_LD | BPF_H | BPF_ABS:
1113 case BPF_LD | BPF_B | BPF_ABS:
1115 if (bpf_anc_helper(ftest) & BPF_ANC)
1117 /* Ancillary operation unknown or unsupported */
1118 if (anc_found == false && ftest->k >= SKF_AD_OFF)
1123 /* Last instruction must be a RET code */
1124 switch (filter[flen - 1].code) {
1125 case BPF_RET | BPF_K:
1126 case BPF_RET | BPF_A:
1127 return check_load_and_stores(filter, flen);
1133 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1134 const struct sock_fprog *fprog)
1136 unsigned int fsize = bpf_classic_proglen(fprog);
1137 struct sock_fprog_kern *fkprog;
1139 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1143 fkprog = fp->orig_prog;
1144 fkprog->len = fprog->len;
1146 fkprog->filter = kmemdup(fp->insns, fsize,
1147 GFP_KERNEL | __GFP_NOWARN);
1148 if (!fkprog->filter) {
1149 kfree(fp->orig_prog);
1156 static void bpf_release_orig_filter(struct bpf_prog *fp)
1158 struct sock_fprog_kern *fprog = fp->orig_prog;
1161 kfree(fprog->filter);
1166 static void __bpf_prog_release(struct bpf_prog *prog)
1168 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1171 bpf_release_orig_filter(prog);
1172 bpf_prog_free(prog);
1176 static void __sk_filter_release(struct sk_filter *fp)
1178 __bpf_prog_release(fp->prog);
1183 * sk_filter_release_rcu - Release a socket filter by rcu_head
1184 * @rcu: rcu_head that contains the sk_filter to free
1186 static void sk_filter_release_rcu(struct rcu_head *rcu)
1188 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1190 __sk_filter_release(fp);
1194 * sk_filter_release - release a socket filter
1195 * @fp: filter to remove
1197 * Remove a filter from a socket and release its resources.
1199 static void sk_filter_release(struct sk_filter *fp)
1201 if (refcount_dec_and_test(&fp->refcnt))
1202 call_rcu(&fp->rcu, sk_filter_release_rcu);
1205 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1207 u32 filter_size = bpf_prog_size(fp->prog->len);
1209 atomic_sub(filter_size, &sk->sk_omem_alloc);
1210 sk_filter_release(fp);
1213 /* try to charge the socket memory if there is space available
1214 * return true on success
1216 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1218 u32 filter_size = bpf_prog_size(fp->prog->len);
1219 int optmem_max = READ_ONCE(sysctl_optmem_max);
1221 /* same check as in sock_kmalloc() */
1222 if (filter_size <= optmem_max &&
1223 atomic_read(&sk->sk_omem_alloc) + filter_size < optmem_max) {
1224 atomic_add(filter_size, &sk->sk_omem_alloc);
1230 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1232 if (!refcount_inc_not_zero(&fp->refcnt))
1235 if (!__sk_filter_charge(sk, fp)) {
1236 sk_filter_release(fp);
1242 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1244 struct sock_filter *old_prog;
1245 struct bpf_prog *old_fp;
1246 int err, new_len, old_len = fp->len;
1247 bool seen_ld_abs = false;
1249 /* We are free to overwrite insns et al right here as it won't be used at
1250 * this point in time anymore internally after the migration to the eBPF
1251 * instruction representation.
1253 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1254 sizeof(struct bpf_insn));
1256 /* Conversion cannot happen on overlapping memory areas,
1257 * so we need to keep the user BPF around until the 2nd
1258 * pass. At this time, the user BPF is stored in fp->insns.
1260 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1261 GFP_KERNEL | __GFP_NOWARN);
1267 /* 1st pass: calculate the new program length. */
1268 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1273 /* Expand fp for appending the new filter representation. */
1275 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1277 /* The old_fp is still around in case we couldn't
1278 * allocate new memory, so uncharge on that one.
1287 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1288 err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1291 /* 2nd bpf_convert_filter() can fail only if it fails
1292 * to allocate memory, remapping must succeed. Note,
1293 * that at this time old_fp has already been released
1298 fp = bpf_prog_select_runtime(fp, &err);
1308 __bpf_prog_release(fp);
1309 return ERR_PTR(err);
1312 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1313 bpf_aux_classic_check_t trans)
1317 fp->bpf_func = NULL;
1320 err = bpf_check_classic(fp->insns, fp->len);
1322 __bpf_prog_release(fp);
1323 return ERR_PTR(err);
1326 /* There might be additional checks and transformations
1327 * needed on classic filters, f.e. in case of seccomp.
1330 err = trans(fp->insns, fp->len);
1332 __bpf_prog_release(fp);
1333 return ERR_PTR(err);
1337 /* Probe if we can JIT compile the filter and if so, do
1338 * the compilation of the filter.
1340 bpf_jit_compile(fp);
1342 /* JIT compiler couldn't process this filter, so do the eBPF translation
1343 * for the optimized interpreter.
1346 fp = bpf_migrate_filter(fp);
1352 * bpf_prog_create - create an unattached filter
1353 * @pfp: the unattached filter that is created
1354 * @fprog: the filter program
1356 * Create a filter independent of any socket. We first run some
1357 * sanity checks on it to make sure it does not explode on us later.
1358 * If an error occurs or there is insufficient memory for the filter
1359 * a negative errno code is returned. On success the return is zero.
1361 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1363 unsigned int fsize = bpf_classic_proglen(fprog);
1364 struct bpf_prog *fp;
1366 /* Make sure new filter is there and in the right amounts. */
1367 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1370 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1374 memcpy(fp->insns, fprog->filter, fsize);
1376 fp->len = fprog->len;
1377 /* Since unattached filters are not copied back to user
1378 * space through sk_get_filter(), we do not need to hold
1379 * a copy here, and can spare us the work.
1381 fp->orig_prog = NULL;
1383 /* bpf_prepare_filter() already takes care of freeing
1384 * memory in case something goes wrong.
1386 fp = bpf_prepare_filter(fp, NULL);
1393 EXPORT_SYMBOL_GPL(bpf_prog_create);
1396 * bpf_prog_create_from_user - create an unattached filter from user buffer
1397 * @pfp: the unattached filter that is created
1398 * @fprog: the filter program
1399 * @trans: post-classic verifier transformation handler
1400 * @save_orig: save classic BPF program
1402 * This function effectively does the same as bpf_prog_create(), only
1403 * that it builds up its insns buffer from user space provided buffer.
1404 * It also allows for passing a bpf_aux_classic_check_t handler.
1406 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1407 bpf_aux_classic_check_t trans, bool save_orig)
1409 unsigned int fsize = bpf_classic_proglen(fprog);
1410 struct bpf_prog *fp;
1413 /* Make sure new filter is there and in the right amounts. */
1414 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1417 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1421 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1422 __bpf_prog_free(fp);
1426 fp->len = fprog->len;
1427 fp->orig_prog = NULL;
1430 err = bpf_prog_store_orig_filter(fp, fprog);
1432 __bpf_prog_free(fp);
1437 /* bpf_prepare_filter() already takes care of freeing
1438 * memory in case something goes wrong.
1440 fp = bpf_prepare_filter(fp, trans);
1447 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1449 void bpf_prog_destroy(struct bpf_prog *fp)
1451 __bpf_prog_release(fp);
1453 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1455 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1457 struct sk_filter *fp, *old_fp;
1459 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1465 if (!__sk_filter_charge(sk, fp)) {
1469 refcount_set(&fp->refcnt, 1);
1471 old_fp = rcu_dereference_protected(sk->sk_filter,
1472 lockdep_sock_is_held(sk));
1473 rcu_assign_pointer(sk->sk_filter, fp);
1476 sk_filter_uncharge(sk, old_fp);
1482 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1484 unsigned int fsize = bpf_classic_proglen(fprog);
1485 struct bpf_prog *prog;
1488 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1489 return ERR_PTR(-EPERM);
1491 /* Make sure new filter is there and in the right amounts. */
1492 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1493 return ERR_PTR(-EINVAL);
1495 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1497 return ERR_PTR(-ENOMEM);
1499 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1500 __bpf_prog_free(prog);
1501 return ERR_PTR(-EFAULT);
1504 prog->len = fprog->len;
1506 err = bpf_prog_store_orig_filter(prog, fprog);
1508 __bpf_prog_free(prog);
1509 return ERR_PTR(-ENOMEM);
1512 /* bpf_prepare_filter() already takes care of freeing
1513 * memory in case something goes wrong.
1515 return bpf_prepare_filter(prog, NULL);
1519 * sk_attach_filter - attach a socket filter
1520 * @fprog: the filter program
1521 * @sk: the socket to use
1523 * Attach the user's filter code. We first run some sanity checks on
1524 * it to make sure it does not explode on us later. If an error
1525 * occurs or there is insufficient memory for the filter a negative
1526 * errno code is returned. On success the return is zero.
1528 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1530 struct bpf_prog *prog = __get_filter(fprog, sk);
1534 return PTR_ERR(prog);
1536 err = __sk_attach_prog(prog, sk);
1538 __bpf_prog_release(prog);
1544 EXPORT_SYMBOL_GPL(sk_attach_filter);
1546 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1548 struct bpf_prog *prog = __get_filter(fprog, sk);
1552 return PTR_ERR(prog);
1554 if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max))
1557 err = reuseport_attach_prog(sk, prog);
1560 __bpf_prog_release(prog);
1565 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1567 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1568 return ERR_PTR(-EPERM);
1570 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1573 int sk_attach_bpf(u32 ufd, struct sock *sk)
1575 struct bpf_prog *prog = __get_bpf(ufd, sk);
1579 return PTR_ERR(prog);
1581 err = __sk_attach_prog(prog, sk);
1590 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1592 struct bpf_prog *prog;
1595 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1598 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1599 if (PTR_ERR(prog) == -EINVAL)
1600 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1602 return PTR_ERR(prog);
1604 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1605 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1606 * bpf prog (e.g. sockmap). It depends on the
1607 * limitation imposed by bpf_prog_load().
1608 * Hence, sysctl_optmem_max is not checked.
1610 if ((sk->sk_type != SOCK_STREAM &&
1611 sk->sk_type != SOCK_DGRAM) ||
1612 (sk->sk_protocol != IPPROTO_UDP &&
1613 sk->sk_protocol != IPPROTO_TCP) ||
1614 (sk->sk_family != AF_INET &&
1615 sk->sk_family != AF_INET6)) {
1620 /* BPF_PROG_TYPE_SOCKET_FILTER */
1621 if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max)) {
1627 err = reuseport_attach_prog(sk, prog);
1635 void sk_reuseport_prog_free(struct bpf_prog *prog)
1640 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1643 bpf_prog_destroy(prog);
1646 struct bpf_scratchpad {
1648 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1649 u8 buff[MAX_BPF_STACK];
1653 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1655 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1656 unsigned int write_len)
1658 return skb_ensure_writable(skb, write_len);
1661 static inline int bpf_try_make_writable(struct sk_buff *skb,
1662 unsigned int write_len)
1664 int err = __bpf_try_make_writable(skb, write_len);
1666 bpf_compute_data_pointers(skb);
1670 static int bpf_try_make_head_writable(struct sk_buff *skb)
1672 return bpf_try_make_writable(skb, skb_headlen(skb));
1675 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1677 if (skb_at_tc_ingress(skb))
1678 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1681 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1683 if (skb_at_tc_ingress(skb))
1684 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1687 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1688 const void *, from, u32, len, u64, flags)
1692 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1694 if (unlikely(offset > INT_MAX))
1696 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1699 ptr = skb->data + offset;
1700 if (flags & BPF_F_RECOMPUTE_CSUM)
1701 __skb_postpull_rcsum(skb, ptr, len, offset);
1703 memcpy(ptr, from, len);
1705 if (flags & BPF_F_RECOMPUTE_CSUM)
1706 __skb_postpush_rcsum(skb, ptr, len, offset);
1707 if (flags & BPF_F_INVALIDATE_HASH)
1708 skb_clear_hash(skb);
1713 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1714 .func = bpf_skb_store_bytes,
1716 .ret_type = RET_INTEGER,
1717 .arg1_type = ARG_PTR_TO_CTX,
1718 .arg2_type = ARG_ANYTHING,
1719 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1720 .arg4_type = ARG_CONST_SIZE,
1721 .arg5_type = ARG_ANYTHING,
1724 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1725 void *, to, u32, len)
1729 if (unlikely(offset > INT_MAX))
1732 ptr = skb_header_pointer(skb, offset, len, to);
1736 memcpy(to, ptr, len);
1744 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1745 .func = bpf_skb_load_bytes,
1747 .ret_type = RET_INTEGER,
1748 .arg1_type = ARG_PTR_TO_CTX,
1749 .arg2_type = ARG_ANYTHING,
1750 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1751 .arg4_type = ARG_CONST_SIZE,
1754 BPF_CALL_4(bpf_flow_dissector_load_bytes,
1755 const struct bpf_flow_dissector *, ctx, u32, offset,
1756 void *, to, u32, len)
1760 if (unlikely(offset > 0xffff))
1763 if (unlikely(!ctx->skb))
1766 ptr = skb_header_pointer(ctx->skb, offset, len, to);
1770 memcpy(to, ptr, len);
1778 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1779 .func = bpf_flow_dissector_load_bytes,
1781 .ret_type = RET_INTEGER,
1782 .arg1_type = ARG_PTR_TO_CTX,
1783 .arg2_type = ARG_ANYTHING,
1784 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1785 .arg4_type = ARG_CONST_SIZE,
1788 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1789 u32, offset, void *, to, u32, len, u32, start_header)
1791 u8 *end = skb_tail_pointer(skb);
1794 if (unlikely(offset > 0xffff))
1797 switch (start_header) {
1798 case BPF_HDR_START_MAC:
1799 if (unlikely(!skb_mac_header_was_set(skb)))
1801 start = skb_mac_header(skb);
1803 case BPF_HDR_START_NET:
1804 start = skb_network_header(skb);
1810 ptr = start + offset;
1812 if (likely(ptr + len <= end)) {
1813 memcpy(to, ptr, len);
1822 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1823 .func = bpf_skb_load_bytes_relative,
1825 .ret_type = RET_INTEGER,
1826 .arg1_type = ARG_PTR_TO_CTX,
1827 .arg2_type = ARG_ANYTHING,
1828 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1829 .arg4_type = ARG_CONST_SIZE,
1830 .arg5_type = ARG_ANYTHING,
1833 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1835 /* Idea is the following: should the needed direct read/write
1836 * test fail during runtime, we can pull in more data and redo
1837 * again, since implicitly, we invalidate previous checks here.
1839 * Or, since we know how much we need to make read/writeable,
1840 * this can be done once at the program beginning for direct
1841 * access case. By this we overcome limitations of only current
1842 * headroom being accessible.
1844 return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1847 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1848 .func = bpf_skb_pull_data,
1850 .ret_type = RET_INTEGER,
1851 .arg1_type = ARG_PTR_TO_CTX,
1852 .arg2_type = ARG_ANYTHING,
1855 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1857 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1860 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1861 .func = bpf_sk_fullsock,
1863 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
1864 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
1867 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1868 unsigned int write_len)
1870 return __bpf_try_make_writable(skb, write_len);
1873 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1875 /* Idea is the following: should the needed direct read/write
1876 * test fail during runtime, we can pull in more data and redo
1877 * again, since implicitly, we invalidate previous checks here.
1879 * Or, since we know how much we need to make read/writeable,
1880 * this can be done once at the program beginning for direct
1881 * access case. By this we overcome limitations of only current
1882 * headroom being accessible.
1884 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1887 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1888 .func = sk_skb_pull_data,
1890 .ret_type = RET_INTEGER,
1891 .arg1_type = ARG_PTR_TO_CTX,
1892 .arg2_type = ARG_ANYTHING,
1895 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1896 u64, from, u64, to, u64, flags)
1900 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1902 if (unlikely(offset > 0xffff || offset & 1))
1904 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1907 ptr = (__sum16 *)(skb->data + offset);
1908 switch (flags & BPF_F_HDR_FIELD_MASK) {
1910 if (unlikely(from != 0))
1913 csum_replace_by_diff(ptr, to);
1916 csum_replace2(ptr, from, to);
1919 csum_replace4(ptr, from, to);
1928 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1929 .func = bpf_l3_csum_replace,
1931 .ret_type = RET_INTEGER,
1932 .arg1_type = ARG_PTR_TO_CTX,
1933 .arg2_type = ARG_ANYTHING,
1934 .arg3_type = ARG_ANYTHING,
1935 .arg4_type = ARG_ANYTHING,
1936 .arg5_type = ARG_ANYTHING,
1939 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1940 u64, from, u64, to, u64, flags)
1942 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1943 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1944 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1947 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1948 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1950 if (unlikely(offset > 0xffff || offset & 1))
1952 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1955 ptr = (__sum16 *)(skb->data + offset);
1956 if (is_mmzero && !do_mforce && !*ptr)
1959 switch (flags & BPF_F_HDR_FIELD_MASK) {
1961 if (unlikely(from != 0))
1964 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1967 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1970 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1976 if (is_mmzero && !*ptr)
1977 *ptr = CSUM_MANGLED_0;
1981 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1982 .func = bpf_l4_csum_replace,
1984 .ret_type = RET_INTEGER,
1985 .arg1_type = ARG_PTR_TO_CTX,
1986 .arg2_type = ARG_ANYTHING,
1987 .arg3_type = ARG_ANYTHING,
1988 .arg4_type = ARG_ANYTHING,
1989 .arg5_type = ARG_ANYTHING,
1992 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1993 __be32 *, to, u32, to_size, __wsum, seed)
1995 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1996 u32 diff_size = from_size + to_size;
1999 /* This is quite flexible, some examples:
2001 * from_size == 0, to_size > 0, seed := csum --> pushing data
2002 * from_size > 0, to_size == 0, seed := csum --> pulling data
2003 * from_size > 0, to_size > 0, seed := 0 --> diffing data
2005 * Even for diffing, from_size and to_size don't need to be equal.
2007 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
2008 diff_size > sizeof(sp->diff)))
2011 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
2012 sp->diff[j] = ~from[i];
2013 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
2014 sp->diff[j] = to[i];
2016 return csum_partial(sp->diff, diff_size, seed);
2019 static const struct bpf_func_proto bpf_csum_diff_proto = {
2020 .func = bpf_csum_diff,
2023 .ret_type = RET_INTEGER,
2024 .arg1_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2025 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
2026 .arg3_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2027 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
2028 .arg5_type = ARG_ANYTHING,
2031 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2033 /* The interface is to be used in combination with bpf_csum_diff()
2034 * for direct packet writes. csum rotation for alignment as well
2035 * as emulating csum_sub() can be done from the eBPF program.
2037 if (skb->ip_summed == CHECKSUM_COMPLETE)
2038 return (skb->csum = csum_add(skb->csum, csum));
2043 static const struct bpf_func_proto bpf_csum_update_proto = {
2044 .func = bpf_csum_update,
2046 .ret_type = RET_INTEGER,
2047 .arg1_type = ARG_PTR_TO_CTX,
2048 .arg2_type = ARG_ANYTHING,
2051 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
2053 /* The interface is to be used in combination with bpf_skb_adjust_room()
2054 * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET
2055 * is passed as flags, for example.
2058 case BPF_CSUM_LEVEL_INC:
2059 __skb_incr_checksum_unnecessary(skb);
2061 case BPF_CSUM_LEVEL_DEC:
2062 __skb_decr_checksum_unnecessary(skb);
2064 case BPF_CSUM_LEVEL_RESET:
2065 __skb_reset_checksum_unnecessary(skb);
2067 case BPF_CSUM_LEVEL_QUERY:
2068 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2069 skb->csum_level : -EACCES;
2077 static const struct bpf_func_proto bpf_csum_level_proto = {
2078 .func = bpf_csum_level,
2080 .ret_type = RET_INTEGER,
2081 .arg1_type = ARG_PTR_TO_CTX,
2082 .arg2_type = ARG_ANYTHING,
2085 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2087 return dev_forward_skb_nomtu(dev, skb);
2090 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2091 struct sk_buff *skb)
2093 int ret = ____dev_forward_skb(dev, skb, false);
2097 ret = netif_rx(skb);
2103 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2107 if (dev_xmit_recursion()) {
2108 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2114 skb_clear_tstamp(skb);
2116 dev_xmit_recursion_inc();
2117 ret = dev_queue_xmit(skb);
2118 dev_xmit_recursion_dec();
2123 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2126 unsigned int mlen = skb_network_offset(skb);
2128 if (unlikely(skb->len <= mlen)) {
2134 __skb_pull(skb, mlen);
2135 if (unlikely(!skb->len)) {
2140 /* At ingress, the mac header has already been pulled once.
2141 * At egress, skb_pospull_rcsum has to be done in case that
2142 * the skb is originated from ingress (i.e. a forwarded skb)
2143 * to ensure that rcsum starts at net header.
2145 if (!skb_at_tc_ingress(skb))
2146 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2148 skb_pop_mac_header(skb);
2149 skb_reset_mac_len(skb);
2150 return flags & BPF_F_INGRESS ?
2151 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2154 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2157 /* Verify that a link layer header is carried */
2158 if (unlikely(skb->mac_header >= skb->network_header || skb->len == 0)) {
2163 bpf_push_mac_rcsum(skb);
2164 return flags & BPF_F_INGRESS ?
2165 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2168 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2171 if (dev_is_mac_header_xmit(dev))
2172 return __bpf_redirect_common(skb, dev, flags);
2174 return __bpf_redirect_no_mac(skb, dev, flags);
2177 #if IS_ENABLED(CONFIG_IPV6)
2178 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2179 struct net_device *dev, struct bpf_nh_params *nh)
2181 u32 hh_len = LL_RESERVED_SPACE(dev);
2182 const struct in6_addr *nexthop;
2183 struct dst_entry *dst = NULL;
2184 struct neighbour *neigh;
2186 if (dev_xmit_recursion()) {
2187 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2192 skb_clear_tstamp(skb);
2194 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2195 skb = skb_expand_head(skb, hh_len);
2203 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2204 &ipv6_hdr(skb)->daddr);
2206 nexthop = &nh->ipv6_nh;
2208 neigh = ip_neigh_gw6(dev, nexthop);
2209 if (likely(!IS_ERR(neigh))) {
2212 sock_confirm_neigh(skb, neigh);
2214 dev_xmit_recursion_inc();
2215 ret = neigh_output(neigh, skb, false);
2216 dev_xmit_recursion_dec();
2221 rcu_read_unlock_bh();
2223 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2229 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2230 struct bpf_nh_params *nh)
2232 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2233 struct net *net = dev_net(dev);
2234 int err, ret = NET_XMIT_DROP;
2237 struct dst_entry *dst;
2238 struct flowi6 fl6 = {
2239 .flowi6_flags = FLOWI_FLAG_ANYSRC,
2240 .flowi6_mark = skb->mark,
2241 .flowlabel = ip6_flowinfo(ip6h),
2242 .flowi6_oif = dev->ifindex,
2243 .flowi6_proto = ip6h->nexthdr,
2244 .daddr = ip6h->daddr,
2245 .saddr = ip6h->saddr,
2248 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2252 skb_dst_set(skb, dst);
2253 } else if (nh->nh_family != AF_INET6) {
2257 err = bpf_out_neigh_v6(net, skb, dev, nh);
2258 if (unlikely(net_xmit_eval(err)))
2259 dev->stats.tx_errors++;
2261 ret = NET_XMIT_SUCCESS;
2264 dev->stats.tx_errors++;
2270 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2271 struct bpf_nh_params *nh)
2274 return NET_XMIT_DROP;
2276 #endif /* CONFIG_IPV6 */
2278 #if IS_ENABLED(CONFIG_INET)
2279 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2280 struct net_device *dev, struct bpf_nh_params *nh)
2282 u32 hh_len = LL_RESERVED_SPACE(dev);
2283 struct neighbour *neigh;
2284 bool is_v6gw = false;
2286 if (dev_xmit_recursion()) {
2287 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2292 skb_clear_tstamp(skb);
2294 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2295 skb = skb_expand_head(skb, hh_len);
2302 struct dst_entry *dst = skb_dst(skb);
2303 struct rtable *rt = container_of(dst, struct rtable, dst);
2305 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2306 } else if (nh->nh_family == AF_INET6) {
2307 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2309 } else if (nh->nh_family == AF_INET) {
2310 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2316 if (likely(!IS_ERR(neigh))) {
2319 sock_confirm_neigh(skb, neigh);
2321 dev_xmit_recursion_inc();
2322 ret = neigh_output(neigh, skb, is_v6gw);
2323 dev_xmit_recursion_dec();
2334 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2335 struct bpf_nh_params *nh)
2337 const struct iphdr *ip4h = ip_hdr(skb);
2338 struct net *net = dev_net(dev);
2339 int err, ret = NET_XMIT_DROP;
2342 struct flowi4 fl4 = {
2343 .flowi4_flags = FLOWI_FLAG_ANYSRC,
2344 .flowi4_mark = skb->mark,
2345 .flowi4_tos = RT_TOS(ip4h->tos),
2346 .flowi4_oif = dev->ifindex,
2347 .flowi4_proto = ip4h->protocol,
2348 .daddr = ip4h->daddr,
2349 .saddr = ip4h->saddr,
2353 rt = ip_route_output_flow(net, &fl4, NULL);
2356 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2361 skb_dst_set(skb, &rt->dst);
2364 err = bpf_out_neigh_v4(net, skb, dev, nh);
2365 if (unlikely(net_xmit_eval(err)))
2366 dev->stats.tx_errors++;
2368 ret = NET_XMIT_SUCCESS;
2371 dev->stats.tx_errors++;
2377 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2378 struct bpf_nh_params *nh)
2381 return NET_XMIT_DROP;
2383 #endif /* CONFIG_INET */
2385 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2386 struct bpf_nh_params *nh)
2388 struct ethhdr *ethh = eth_hdr(skb);
2390 if (unlikely(skb->mac_header >= skb->network_header))
2392 bpf_push_mac_rcsum(skb);
2393 if (is_multicast_ether_addr(ethh->h_dest))
2396 skb_pull(skb, sizeof(*ethh));
2397 skb_unset_mac_header(skb);
2398 skb_reset_network_header(skb);
2400 if (skb->protocol == htons(ETH_P_IP))
2401 return __bpf_redirect_neigh_v4(skb, dev, nh);
2402 else if (skb->protocol == htons(ETH_P_IPV6))
2403 return __bpf_redirect_neigh_v6(skb, dev, nh);
2409 /* Internal, non-exposed redirect flags. */
2411 BPF_F_NEIGH = (1ULL << 1),
2412 BPF_F_PEER = (1ULL << 2),
2413 BPF_F_NEXTHOP = (1ULL << 3),
2414 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2417 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2419 struct net_device *dev;
2420 struct sk_buff *clone;
2423 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2426 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2430 clone = skb_clone(skb, GFP_ATOMIC);
2431 if (unlikely(!clone))
2434 /* For direct write, we need to keep the invariant that the skbs
2435 * we're dealing with need to be uncloned. Should uncloning fail
2436 * here, we need to free the just generated clone to unclone once
2439 ret = bpf_try_make_head_writable(skb);
2440 if (unlikely(ret)) {
2445 return __bpf_redirect(clone, dev, flags);
2448 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2449 .func = bpf_clone_redirect,
2451 .ret_type = RET_INTEGER,
2452 .arg1_type = ARG_PTR_TO_CTX,
2453 .arg2_type = ARG_ANYTHING,
2454 .arg3_type = ARG_ANYTHING,
2457 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2458 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2460 int skb_do_redirect(struct sk_buff *skb)
2462 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2463 struct net *net = dev_net(skb->dev);
2464 struct net_device *dev;
2465 u32 flags = ri->flags;
2467 dev = dev_get_by_index_rcu(net, ri->tgt_index);
2472 if (flags & BPF_F_PEER) {
2473 const struct net_device_ops *ops = dev->netdev_ops;
2475 if (unlikely(!ops->ndo_get_peer_dev ||
2476 !skb_at_tc_ingress(skb)))
2478 dev = ops->ndo_get_peer_dev(dev);
2479 if (unlikely(!dev ||
2480 !(dev->flags & IFF_UP) ||
2481 net_eq(net, dev_net(dev))))
2486 return flags & BPF_F_NEIGH ?
2487 __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2489 __bpf_redirect(skb, dev, flags);
2495 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2497 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2499 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2503 ri->tgt_index = ifindex;
2505 return TC_ACT_REDIRECT;
2508 static const struct bpf_func_proto bpf_redirect_proto = {
2509 .func = bpf_redirect,
2511 .ret_type = RET_INTEGER,
2512 .arg1_type = ARG_ANYTHING,
2513 .arg2_type = ARG_ANYTHING,
2516 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2518 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2520 if (unlikely(flags))
2523 ri->flags = BPF_F_PEER;
2524 ri->tgt_index = ifindex;
2526 return TC_ACT_REDIRECT;
2529 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2530 .func = bpf_redirect_peer,
2532 .ret_type = RET_INTEGER,
2533 .arg1_type = ARG_ANYTHING,
2534 .arg2_type = ARG_ANYTHING,
2537 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2538 int, plen, u64, flags)
2540 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2542 if (unlikely((plen && plen < sizeof(*params)) || flags))
2545 ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2546 ri->tgt_index = ifindex;
2548 BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2550 memcpy(&ri->nh, params, sizeof(ri->nh));
2552 return TC_ACT_REDIRECT;
2555 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2556 .func = bpf_redirect_neigh,
2558 .ret_type = RET_INTEGER,
2559 .arg1_type = ARG_ANYTHING,
2560 .arg2_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2561 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
2562 .arg4_type = ARG_ANYTHING,
2565 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2567 msg->apply_bytes = bytes;
2571 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2572 .func = bpf_msg_apply_bytes,
2574 .ret_type = RET_INTEGER,
2575 .arg1_type = ARG_PTR_TO_CTX,
2576 .arg2_type = ARG_ANYTHING,
2579 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2581 msg->cork_bytes = bytes;
2585 static void sk_msg_reset_curr(struct sk_msg *msg)
2587 u32 i = msg->sg.start;
2591 len += sk_msg_elem(msg, i)->length;
2592 sk_msg_iter_var_next(i);
2593 if (len >= msg->sg.size)
2595 } while (i != msg->sg.end);
2598 msg->sg.copybreak = 0;
2601 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2602 .func = bpf_msg_cork_bytes,
2604 .ret_type = RET_INTEGER,
2605 .arg1_type = ARG_PTR_TO_CTX,
2606 .arg2_type = ARG_ANYTHING,
2609 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2610 u32, end, u64, flags)
2612 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2613 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2614 struct scatterlist *sge;
2615 u8 *raw, *to, *from;
2618 if (unlikely(flags || end <= start))
2621 /* First find the starting scatterlist element */
2625 len = sk_msg_elem(msg, i)->length;
2626 if (start < offset + len)
2628 sk_msg_iter_var_next(i);
2629 } while (i != msg->sg.end);
2631 if (unlikely(start >= offset + len))
2635 /* The start may point into the sg element so we need to also
2636 * account for the headroom.
2638 bytes_sg_total = start - offset + bytes;
2639 if (!test_bit(i, msg->sg.copy) && bytes_sg_total <= len)
2642 /* At this point we need to linearize multiple scatterlist
2643 * elements or a single shared page. Either way we need to
2644 * copy into a linear buffer exclusively owned by BPF. Then
2645 * place the buffer in the scatterlist and fixup the original
2646 * entries by removing the entries now in the linear buffer
2647 * and shifting the remaining entries. For now we do not try
2648 * to copy partial entries to avoid complexity of running out
2649 * of sg_entry slots. The downside is reading a single byte
2650 * will copy the entire sg entry.
2653 copy += sk_msg_elem(msg, i)->length;
2654 sk_msg_iter_var_next(i);
2655 if (bytes_sg_total <= copy)
2657 } while (i != msg->sg.end);
2660 if (unlikely(bytes_sg_total > copy))
2663 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2665 if (unlikely(!page))
2668 raw = page_address(page);
2671 sge = sk_msg_elem(msg, i);
2672 from = sg_virt(sge);
2676 memcpy(to, from, len);
2679 put_page(sg_page(sge));
2681 sk_msg_iter_var_next(i);
2682 } while (i != last_sge);
2684 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2686 /* To repair sg ring we need to shift entries. If we only
2687 * had a single entry though we can just replace it and
2688 * be done. Otherwise walk the ring and shift the entries.
2690 WARN_ON_ONCE(last_sge == first_sge);
2691 shift = last_sge > first_sge ?
2692 last_sge - first_sge - 1 :
2693 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2698 sk_msg_iter_var_next(i);
2702 if (i + shift >= NR_MSG_FRAG_IDS)
2703 move_from = i + shift - NR_MSG_FRAG_IDS;
2705 move_from = i + shift;
2706 if (move_from == msg->sg.end)
2709 msg->sg.data[i] = msg->sg.data[move_from];
2710 msg->sg.data[move_from].length = 0;
2711 msg->sg.data[move_from].page_link = 0;
2712 msg->sg.data[move_from].offset = 0;
2713 sk_msg_iter_var_next(i);
2716 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2717 msg->sg.end - shift + NR_MSG_FRAG_IDS :
2718 msg->sg.end - shift;
2720 sk_msg_reset_curr(msg);
2721 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2722 msg->data_end = msg->data + bytes;
2726 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2727 .func = bpf_msg_pull_data,
2729 .ret_type = RET_INTEGER,
2730 .arg1_type = ARG_PTR_TO_CTX,
2731 .arg2_type = ARG_ANYTHING,
2732 .arg3_type = ARG_ANYTHING,
2733 .arg4_type = ARG_ANYTHING,
2736 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2737 u32, len, u64, flags)
2739 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2740 u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2741 u8 *raw, *to, *from;
2744 if (unlikely(flags))
2747 if (unlikely(len == 0))
2750 /* First find the starting scatterlist element */
2754 l = sk_msg_elem(msg, i)->length;
2756 if (start < offset + l)
2758 sk_msg_iter_var_next(i);
2759 } while (i != msg->sg.end);
2761 if (start >= offset + l)
2764 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2766 /* If no space available will fallback to copy, we need at
2767 * least one scatterlist elem available to push data into
2768 * when start aligns to the beginning of an element or two
2769 * when it falls inside an element. We handle the start equals
2770 * offset case because its the common case for inserting a
2773 if (!space || (space == 1 && start != offset))
2774 copy = msg->sg.data[i].length;
2776 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2777 get_order(copy + len));
2778 if (unlikely(!page))
2784 raw = page_address(page);
2786 psge = sk_msg_elem(msg, i);
2787 front = start - offset;
2788 back = psge->length - front;
2789 from = sg_virt(psge);
2792 memcpy(raw, from, front);
2796 to = raw + front + len;
2798 memcpy(to, from, back);
2801 put_page(sg_page(psge));
2802 } else if (start - offset) {
2803 psge = sk_msg_elem(msg, i);
2804 rsge = sk_msg_elem_cpy(msg, i);
2806 psge->length = start - offset;
2807 rsge.length -= psge->length;
2808 rsge.offset += start;
2810 sk_msg_iter_var_next(i);
2811 sg_unmark_end(psge);
2812 sg_unmark_end(&rsge);
2813 sk_msg_iter_next(msg, end);
2816 /* Slot(s) to place newly allocated data */
2819 /* Shift one or two slots as needed */
2821 sge = sk_msg_elem_cpy(msg, i);
2823 sk_msg_iter_var_next(i);
2824 sg_unmark_end(&sge);
2825 sk_msg_iter_next(msg, end);
2827 nsge = sk_msg_elem_cpy(msg, i);
2829 sk_msg_iter_var_next(i);
2830 nnsge = sk_msg_elem_cpy(msg, i);
2833 while (i != msg->sg.end) {
2834 msg->sg.data[i] = sge;
2836 sk_msg_iter_var_next(i);
2839 nnsge = sk_msg_elem_cpy(msg, i);
2841 nsge = sk_msg_elem_cpy(msg, i);
2846 /* Place newly allocated data buffer */
2847 sk_mem_charge(msg->sk, len);
2848 msg->sg.size += len;
2849 __clear_bit(new, msg->sg.copy);
2850 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2852 get_page(sg_page(&rsge));
2853 sk_msg_iter_var_next(new);
2854 msg->sg.data[new] = rsge;
2857 sk_msg_reset_curr(msg);
2858 sk_msg_compute_data_pointers(msg);
2862 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2863 .func = bpf_msg_push_data,
2865 .ret_type = RET_INTEGER,
2866 .arg1_type = ARG_PTR_TO_CTX,
2867 .arg2_type = ARG_ANYTHING,
2868 .arg3_type = ARG_ANYTHING,
2869 .arg4_type = ARG_ANYTHING,
2872 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2878 sk_msg_iter_var_next(i);
2879 msg->sg.data[prev] = msg->sg.data[i];
2880 } while (i != msg->sg.end);
2882 sk_msg_iter_prev(msg, end);
2885 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2887 struct scatterlist tmp, sge;
2889 sk_msg_iter_next(msg, end);
2890 sge = sk_msg_elem_cpy(msg, i);
2891 sk_msg_iter_var_next(i);
2892 tmp = sk_msg_elem_cpy(msg, i);
2894 while (i != msg->sg.end) {
2895 msg->sg.data[i] = sge;
2896 sk_msg_iter_var_next(i);
2898 tmp = sk_msg_elem_cpy(msg, i);
2902 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2903 u32, len, u64, flags)
2905 u32 i = 0, l = 0, space, offset = 0;
2906 u64 last = start + len;
2909 if (unlikely(flags))
2912 /* First find the starting scatterlist element */
2916 l = sk_msg_elem(msg, i)->length;
2918 if (start < offset + l)
2920 sk_msg_iter_var_next(i);
2921 } while (i != msg->sg.end);
2923 /* Bounds checks: start and pop must be inside message */
2924 if (start >= offset + l || last >= msg->sg.size)
2927 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2930 /* --------------| offset
2931 * -| start |-------- len -------|
2933 * |----- a ----|-------- pop -------|----- b ----|
2934 * |______________________________________________| length
2937 * a: region at front of scatter element to save
2938 * b: region at back of scatter element to save when length > A + pop
2939 * pop: region to pop from element, same as input 'pop' here will be
2940 * decremented below per iteration.
2942 * Two top-level cases to handle when start != offset, first B is non
2943 * zero and second B is zero corresponding to when a pop includes more
2946 * Then if B is non-zero AND there is no space allocate space and
2947 * compact A, B regions into page. If there is space shift ring to
2948 * the rigth free'ing the next element in ring to place B, leaving
2949 * A untouched except to reduce length.
2951 if (start != offset) {
2952 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2954 int b = sge->length - pop - a;
2956 sk_msg_iter_var_next(i);
2958 if (pop < sge->length - a) {
2961 sk_msg_shift_right(msg, i);
2962 nsge = sk_msg_elem(msg, i);
2963 get_page(sg_page(sge));
2966 b, sge->offset + pop + a);
2968 struct page *page, *orig;
2971 page = alloc_pages(__GFP_NOWARN |
2972 __GFP_COMP | GFP_ATOMIC,
2974 if (unlikely(!page))
2978 orig = sg_page(sge);
2979 from = sg_virt(sge);
2980 to = page_address(page);
2981 memcpy(to, from, a);
2982 memcpy(to + a, from + a + pop, b);
2983 sg_set_page(sge, page, a + b, 0);
2987 } else if (pop >= sge->length - a) {
2988 pop -= (sge->length - a);
2993 /* From above the current layout _must_ be as follows,
2998 * |---- pop ---|---------------- b ------------|
2999 * |____________________________________________| length
3001 * Offset and start of the current msg elem are equal because in the
3002 * previous case we handled offset != start and either consumed the
3003 * entire element and advanced to the next element OR pop == 0.
3005 * Two cases to handle here are first pop is less than the length
3006 * leaving some remainder b above. Simply adjust the element's layout
3007 * in this case. Or pop >= length of the element so that b = 0. In this
3008 * case advance to next element decrementing pop.
3011 struct scatterlist *sge = sk_msg_elem(msg, i);
3013 if (pop < sge->length) {
3019 sk_msg_shift_left(msg, i);
3021 sk_msg_iter_var_next(i);
3024 sk_mem_uncharge(msg->sk, len - pop);
3025 msg->sg.size -= (len - pop);
3026 sk_msg_reset_curr(msg);
3027 sk_msg_compute_data_pointers(msg);
3031 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
3032 .func = bpf_msg_pop_data,
3034 .ret_type = RET_INTEGER,
3035 .arg1_type = ARG_PTR_TO_CTX,
3036 .arg2_type = ARG_ANYTHING,
3037 .arg3_type = ARG_ANYTHING,
3038 .arg4_type = ARG_ANYTHING,
3041 #ifdef CONFIG_CGROUP_NET_CLASSID
3042 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3044 return __task_get_classid(current);
3047 const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3048 .func = bpf_get_cgroup_classid_curr,
3050 .ret_type = RET_INTEGER,
3053 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3055 struct sock *sk = skb_to_full_sk(skb);
3057 if (!sk || !sk_fullsock(sk))
3060 return sock_cgroup_classid(&sk->sk_cgrp_data);
3063 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3064 .func = bpf_skb_cgroup_classid,
3066 .ret_type = RET_INTEGER,
3067 .arg1_type = ARG_PTR_TO_CTX,
3071 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3073 return task_get_classid(skb);
3076 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3077 .func = bpf_get_cgroup_classid,
3079 .ret_type = RET_INTEGER,
3080 .arg1_type = ARG_PTR_TO_CTX,
3083 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3085 return dst_tclassid(skb);
3088 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3089 .func = bpf_get_route_realm,
3091 .ret_type = RET_INTEGER,
3092 .arg1_type = ARG_PTR_TO_CTX,
3095 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3097 /* If skb_clear_hash() was called due to mangling, we can
3098 * trigger SW recalculation here. Later access to hash
3099 * can then use the inline skb->hash via context directly
3100 * instead of calling this helper again.
3102 return skb_get_hash(skb);
3105 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3106 .func = bpf_get_hash_recalc,
3108 .ret_type = RET_INTEGER,
3109 .arg1_type = ARG_PTR_TO_CTX,
3112 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3114 /* After all direct packet write, this can be used once for
3115 * triggering a lazy recalc on next skb_get_hash() invocation.
3117 skb_clear_hash(skb);
3121 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3122 .func = bpf_set_hash_invalid,
3124 .ret_type = RET_INTEGER,
3125 .arg1_type = ARG_PTR_TO_CTX,
3128 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3130 /* Set user specified hash as L4(+), so that it gets returned
3131 * on skb_get_hash() call unless BPF prog later on triggers a
3134 __skb_set_sw_hash(skb, hash, true);
3138 static const struct bpf_func_proto bpf_set_hash_proto = {
3139 .func = bpf_set_hash,
3141 .ret_type = RET_INTEGER,
3142 .arg1_type = ARG_PTR_TO_CTX,
3143 .arg2_type = ARG_ANYTHING,
3146 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3151 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3152 vlan_proto != htons(ETH_P_8021AD)))
3153 vlan_proto = htons(ETH_P_8021Q);
3155 bpf_push_mac_rcsum(skb);
3156 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3157 bpf_pull_mac_rcsum(skb);
3159 bpf_compute_data_pointers(skb);
3163 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3164 .func = bpf_skb_vlan_push,
3166 .ret_type = RET_INTEGER,
3167 .arg1_type = ARG_PTR_TO_CTX,
3168 .arg2_type = ARG_ANYTHING,
3169 .arg3_type = ARG_ANYTHING,
3172 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3176 bpf_push_mac_rcsum(skb);
3177 ret = skb_vlan_pop(skb);
3178 bpf_pull_mac_rcsum(skb);
3180 bpf_compute_data_pointers(skb);
3184 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3185 .func = bpf_skb_vlan_pop,
3187 .ret_type = RET_INTEGER,
3188 .arg1_type = ARG_PTR_TO_CTX,
3191 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3193 /* Caller already did skb_cow() with len as headroom,
3194 * so no need to do it here.
3197 memmove(skb->data, skb->data + len, off);
3198 memset(skb->data + off, 0, len);
3200 /* No skb_postpush_rcsum(skb, skb->data + off, len)
3201 * needed here as it does not change the skb->csum
3202 * result for checksum complete when summing over
3208 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3212 /* skb_ensure_writable() is not needed here, as we're
3213 * already working on an uncloned skb.
3215 if (unlikely(!pskb_may_pull(skb, off + len)))
3218 old_data = skb->data;
3219 __skb_pull(skb, len);
3220 skb_postpull_rcsum(skb, old_data + off, len);
3221 memmove(skb->data, old_data, off);
3226 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3228 bool trans_same = skb->transport_header == skb->network_header;
3231 /* There's no need for __skb_push()/__skb_pull() pair to
3232 * get to the start of the mac header as we're guaranteed
3233 * to always start from here under eBPF.
3235 ret = bpf_skb_generic_push(skb, off, len);
3237 skb->mac_header -= len;
3238 skb->network_header -= len;
3240 skb->transport_header = skb->network_header;
3246 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3248 bool trans_same = skb->transport_header == skb->network_header;
3251 /* Same here, __skb_push()/__skb_pull() pair not needed. */
3252 ret = bpf_skb_generic_pop(skb, off, len);
3254 skb->mac_header += len;
3255 skb->network_header += len;
3257 skb->transport_header = skb->network_header;
3263 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3265 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3266 u32 off = skb_mac_header_len(skb);
3269 ret = skb_cow(skb, len_diff);
3270 if (unlikely(ret < 0))
3273 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3274 if (unlikely(ret < 0))
3277 if (skb_is_gso(skb)) {
3278 struct skb_shared_info *shinfo = skb_shinfo(skb);
3280 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */
3281 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3282 shinfo->gso_type &= ~SKB_GSO_TCPV4;
3283 shinfo->gso_type |= SKB_GSO_TCPV6;
3287 skb->protocol = htons(ETH_P_IPV6);
3288 skb_clear_hash(skb);
3293 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3295 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3296 u32 off = skb_mac_header_len(skb);
3299 ret = skb_unclone(skb, GFP_ATOMIC);
3300 if (unlikely(ret < 0))
3303 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3304 if (unlikely(ret < 0))
3307 if (skb_is_gso(skb)) {
3308 struct skb_shared_info *shinfo = skb_shinfo(skb);
3310 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */
3311 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3312 shinfo->gso_type &= ~SKB_GSO_TCPV6;
3313 shinfo->gso_type |= SKB_GSO_TCPV4;
3317 skb->protocol = htons(ETH_P_IP);
3318 skb_clear_hash(skb);
3323 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3325 __be16 from_proto = skb->protocol;
3327 if (from_proto == htons(ETH_P_IP) &&
3328 to_proto == htons(ETH_P_IPV6))
3329 return bpf_skb_proto_4_to_6(skb);
3331 if (from_proto == htons(ETH_P_IPV6) &&
3332 to_proto == htons(ETH_P_IP))
3333 return bpf_skb_proto_6_to_4(skb);
3338 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3343 if (unlikely(flags))
3346 /* General idea is that this helper does the basic groundwork
3347 * needed for changing the protocol, and eBPF program fills the
3348 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3349 * and other helpers, rather than passing a raw buffer here.
3351 * The rationale is to keep this minimal and without a need to
3352 * deal with raw packet data. F.e. even if we would pass buffers
3353 * here, the program still needs to call the bpf_lX_csum_replace()
3354 * helpers anyway. Plus, this way we keep also separation of
3355 * concerns, since f.e. bpf_skb_store_bytes() should only take
3358 * Currently, additional options and extension header space are
3359 * not supported, but flags register is reserved so we can adapt
3360 * that. For offloads, we mark packet as dodgy, so that headers
3361 * need to be verified first.
3363 ret = bpf_skb_proto_xlat(skb, proto);
3364 bpf_compute_data_pointers(skb);
3368 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3369 .func = bpf_skb_change_proto,
3371 .ret_type = RET_INTEGER,
3372 .arg1_type = ARG_PTR_TO_CTX,
3373 .arg2_type = ARG_ANYTHING,
3374 .arg3_type = ARG_ANYTHING,
3377 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3379 /* We only allow a restricted subset to be changed for now. */
3380 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3381 !skb_pkt_type_ok(pkt_type)))
3384 skb->pkt_type = pkt_type;
3388 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3389 .func = bpf_skb_change_type,
3391 .ret_type = RET_INTEGER,
3392 .arg1_type = ARG_PTR_TO_CTX,
3393 .arg2_type = ARG_ANYTHING,
3396 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3398 switch (skb->protocol) {
3399 case htons(ETH_P_IP):
3400 return sizeof(struct iphdr);
3401 case htons(ETH_P_IPV6):
3402 return sizeof(struct ipv6hdr);
3408 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3409 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3411 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \
3412 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3413 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3414 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3415 BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \
3416 BPF_F_ADJ_ROOM_ENCAP_L2( \
3417 BPF_ADJ_ROOM_ENCAP_L2_MASK))
3419 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3422 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3423 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3424 u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3425 unsigned int gso_type = SKB_GSO_DODGY;
3428 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3429 /* udp gso_size delineates datagrams, only allow if fixed */
3430 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3431 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3435 ret = skb_cow_head(skb, len_diff);
3436 if (unlikely(ret < 0))
3440 if (skb->protocol != htons(ETH_P_IP) &&
3441 skb->protocol != htons(ETH_P_IPV6))
3444 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3445 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3448 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3449 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3452 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH &&
3453 inner_mac_len < ETH_HLEN)
3456 if (skb->encapsulation)
3459 mac_len = skb->network_header - skb->mac_header;
3460 inner_net = skb->network_header;
3461 if (inner_mac_len > len_diff)
3463 inner_trans = skb->transport_header;
3466 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3467 if (unlikely(ret < 0))
3471 skb->inner_mac_header = inner_net - inner_mac_len;
3472 skb->inner_network_header = inner_net;
3473 skb->inner_transport_header = inner_trans;
3475 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH)
3476 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
3478 skb_set_inner_protocol(skb, skb->protocol);
3480 skb->encapsulation = 1;
3481 skb_set_network_header(skb, mac_len);
3483 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3484 gso_type |= SKB_GSO_UDP_TUNNEL;
3485 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3486 gso_type |= SKB_GSO_GRE;
3487 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3488 gso_type |= SKB_GSO_IPXIP6;
3489 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3490 gso_type |= SKB_GSO_IPXIP4;
3492 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3493 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3494 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3495 sizeof(struct ipv6hdr) :
3496 sizeof(struct iphdr);
3498 skb_set_transport_header(skb, mac_len + nh_len);
3501 /* Match skb->protocol to new outer l3 protocol */
3502 if (skb->protocol == htons(ETH_P_IP) &&
3503 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3504 skb->protocol = htons(ETH_P_IPV6);
3505 else if (skb->protocol == htons(ETH_P_IPV6) &&
3506 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3507 skb->protocol = htons(ETH_P_IP);
3510 if (skb_is_gso(skb)) {
3511 struct skb_shared_info *shinfo = skb_shinfo(skb);
3513 /* Due to header grow, MSS needs to be downgraded. */
3514 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3515 skb_decrease_gso_size(shinfo, len_diff);
3517 /* Header must be checked, and gso_segs recomputed. */
3518 shinfo->gso_type |= gso_type;
3519 shinfo->gso_segs = 0;
3525 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3530 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3531 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3534 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3535 /* udp gso_size delineates datagrams, only allow if fixed */
3536 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3537 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3541 ret = skb_unclone(skb, GFP_ATOMIC);
3542 if (unlikely(ret < 0))
3545 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3546 if (unlikely(ret < 0))
3549 if (skb_is_gso(skb)) {
3550 struct skb_shared_info *shinfo = skb_shinfo(skb);
3552 /* Due to header shrink, MSS can be upgraded. */
3553 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3554 skb_increase_gso_size(shinfo, len_diff);
3556 /* Header must be checked, and gso_segs recomputed. */
3557 shinfo->gso_type |= SKB_GSO_DODGY;
3558 shinfo->gso_segs = 0;
3564 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3566 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3567 u32, mode, u64, flags)
3569 u32 len_diff_abs = abs(len_diff);
3570 bool shrink = len_diff < 0;
3573 if (unlikely(flags || mode))
3575 if (unlikely(len_diff_abs > 0xfffU))
3579 ret = skb_cow(skb, len_diff);
3580 if (unlikely(ret < 0))
3582 __skb_push(skb, len_diff_abs);
3583 memset(skb->data, 0, len_diff_abs);
3585 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3587 __skb_pull(skb, len_diff_abs);
3589 if (tls_sw_has_ctx_rx(skb->sk)) {
3590 struct strp_msg *rxm = strp_msg(skb);
3592 rxm->full_len += len_diff;
3597 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3598 .func = sk_skb_adjust_room,
3600 .ret_type = RET_INTEGER,
3601 .arg1_type = ARG_PTR_TO_CTX,
3602 .arg2_type = ARG_ANYTHING,
3603 .arg3_type = ARG_ANYTHING,
3604 .arg4_type = ARG_ANYTHING,
3607 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3608 u32, mode, u64, flags)
3610 u32 len_cur, len_diff_abs = abs(len_diff);
3611 u32 len_min = bpf_skb_net_base_len(skb);
3612 u32 len_max = BPF_SKB_MAX_LEN;
3613 __be16 proto = skb->protocol;
3614 bool shrink = len_diff < 0;
3618 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3619 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3621 if (unlikely(len_diff_abs > 0xfffU))
3623 if (unlikely(proto != htons(ETH_P_IP) &&
3624 proto != htons(ETH_P_IPV6)))
3627 off = skb_mac_header_len(skb);
3629 case BPF_ADJ_ROOM_NET:
3630 off += bpf_skb_net_base_len(skb);
3632 case BPF_ADJ_ROOM_MAC:
3638 len_cur = skb->len - skb_network_offset(skb);
3639 if ((shrink && (len_diff_abs >= len_cur ||
3640 len_cur - len_diff_abs < len_min)) ||
3641 (!shrink && (skb->len + len_diff_abs > len_max &&
3645 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3646 bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3647 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3648 __skb_reset_checksum_unnecessary(skb);
3650 bpf_compute_data_pointers(skb);
3654 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3655 .func = bpf_skb_adjust_room,
3657 .ret_type = RET_INTEGER,
3658 .arg1_type = ARG_PTR_TO_CTX,
3659 .arg2_type = ARG_ANYTHING,
3660 .arg3_type = ARG_ANYTHING,
3661 .arg4_type = ARG_ANYTHING,
3664 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3666 u32 min_len = skb_network_offset(skb);
3668 if (skb_transport_header_was_set(skb))
3669 min_len = skb_transport_offset(skb);
3670 if (skb->ip_summed == CHECKSUM_PARTIAL)
3671 min_len = skb_checksum_start_offset(skb) +
3672 skb->csum_offset + sizeof(__sum16);
3676 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3678 unsigned int old_len = skb->len;
3681 ret = __skb_grow_rcsum(skb, new_len);
3683 memset(skb->data + old_len, 0, new_len - old_len);
3687 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3689 return __skb_trim_rcsum(skb, new_len);
3692 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3695 u32 max_len = BPF_SKB_MAX_LEN;
3696 u32 min_len = __bpf_skb_min_len(skb);
3699 if (unlikely(flags || new_len > max_len || new_len < min_len))
3701 if (skb->encapsulation)
3704 /* The basic idea of this helper is that it's performing the
3705 * needed work to either grow or trim an skb, and eBPF program
3706 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3707 * bpf_lX_csum_replace() and others rather than passing a raw
3708 * buffer here. This one is a slow path helper and intended
3709 * for replies with control messages.
3711 * Like in bpf_skb_change_proto(), we want to keep this rather
3712 * minimal and without protocol specifics so that we are able
3713 * to separate concerns as in bpf_skb_store_bytes() should only
3714 * be the one responsible for writing buffers.
3716 * It's really expected to be a slow path operation here for
3717 * control message replies, so we're implicitly linearizing,
3718 * uncloning and drop offloads from the skb by this.
3720 ret = __bpf_try_make_writable(skb, skb->len);
3722 if (new_len > skb->len)
3723 ret = bpf_skb_grow_rcsum(skb, new_len);
3724 else if (new_len < skb->len)
3725 ret = bpf_skb_trim_rcsum(skb, new_len);
3726 if (!ret && skb_is_gso(skb))
3732 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3735 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3737 bpf_compute_data_pointers(skb);
3741 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3742 .func = bpf_skb_change_tail,
3744 .ret_type = RET_INTEGER,
3745 .arg1_type = ARG_PTR_TO_CTX,
3746 .arg2_type = ARG_ANYTHING,
3747 .arg3_type = ARG_ANYTHING,
3750 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3753 return __bpf_skb_change_tail(skb, new_len, flags);
3756 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3757 .func = sk_skb_change_tail,
3759 .ret_type = RET_INTEGER,
3760 .arg1_type = ARG_PTR_TO_CTX,
3761 .arg2_type = ARG_ANYTHING,
3762 .arg3_type = ARG_ANYTHING,
3765 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3768 u32 max_len = BPF_SKB_MAX_LEN;
3769 u32 new_len = skb->len + head_room;
3772 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3773 new_len < skb->len))
3776 ret = skb_cow(skb, head_room);
3778 /* Idea for this helper is that we currently only
3779 * allow to expand on mac header. This means that
3780 * skb->protocol network header, etc, stay as is.
3781 * Compared to bpf_skb_change_tail(), we're more
3782 * flexible due to not needing to linearize or
3783 * reset GSO. Intention for this helper is to be
3784 * used by an L3 skb that needs to push mac header
3785 * for redirection into L2 device.
3787 __skb_push(skb, head_room);
3788 memset(skb->data, 0, head_room);
3789 skb_reset_mac_header(skb);
3790 skb_reset_mac_len(skb);
3796 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3799 int ret = __bpf_skb_change_head(skb, head_room, flags);
3801 bpf_compute_data_pointers(skb);
3805 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3806 .func = bpf_skb_change_head,
3808 .ret_type = RET_INTEGER,
3809 .arg1_type = ARG_PTR_TO_CTX,
3810 .arg2_type = ARG_ANYTHING,
3811 .arg3_type = ARG_ANYTHING,
3814 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3817 return __bpf_skb_change_head(skb, head_room, flags);
3820 static const struct bpf_func_proto sk_skb_change_head_proto = {
3821 .func = sk_skb_change_head,
3823 .ret_type = RET_INTEGER,
3824 .arg1_type = ARG_PTR_TO_CTX,
3825 .arg2_type = ARG_ANYTHING,
3826 .arg3_type = ARG_ANYTHING,
3829 BPF_CALL_1(bpf_xdp_get_buff_len, struct xdp_buff*, xdp)
3831 return xdp_get_buff_len(xdp);
3834 static const struct bpf_func_proto bpf_xdp_get_buff_len_proto = {
3835 .func = bpf_xdp_get_buff_len,
3837 .ret_type = RET_INTEGER,
3838 .arg1_type = ARG_PTR_TO_CTX,
3841 BTF_ID_LIST_SINGLE(bpf_xdp_get_buff_len_bpf_ids, struct, xdp_buff)
3843 const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto = {
3844 .func = bpf_xdp_get_buff_len,
3846 .arg1_type = ARG_PTR_TO_BTF_ID,
3847 .arg1_btf_id = &bpf_xdp_get_buff_len_bpf_ids[0],
3850 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3852 return xdp_data_meta_unsupported(xdp) ? 0 :
3853 xdp->data - xdp->data_meta;
3856 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3858 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3859 unsigned long metalen = xdp_get_metalen(xdp);
3860 void *data_start = xdp_frame_end + metalen;
3861 void *data = xdp->data + offset;
3863 if (unlikely(data < data_start ||
3864 data > xdp->data_end - ETH_HLEN))
3868 memmove(xdp->data_meta + offset,
3869 xdp->data_meta, metalen);
3870 xdp->data_meta += offset;
3876 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3877 .func = bpf_xdp_adjust_head,
3879 .ret_type = RET_INTEGER,
3880 .arg1_type = ARG_PTR_TO_CTX,
3881 .arg2_type = ARG_ANYTHING,
3884 static void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off,
3885 void *buf, unsigned long len, bool flush)
3887 unsigned long ptr_len, ptr_off = 0;
3888 skb_frag_t *next_frag, *end_frag;
3889 struct skb_shared_info *sinfo;
3893 if (likely(xdp->data_end - xdp->data >= off + len)) {
3894 src = flush ? buf : xdp->data + off;
3895 dst = flush ? xdp->data + off : buf;
3896 memcpy(dst, src, len);
3900 sinfo = xdp_get_shared_info_from_buff(xdp);
3901 end_frag = &sinfo->frags[sinfo->nr_frags];
3902 next_frag = &sinfo->frags[0];
3904 ptr_len = xdp->data_end - xdp->data;
3905 ptr_buf = xdp->data;
3908 if (off < ptr_off + ptr_len) {
3909 unsigned long copy_off = off - ptr_off;
3910 unsigned long copy_len = min(len, ptr_len - copy_off);
3912 src = flush ? buf : ptr_buf + copy_off;
3913 dst = flush ? ptr_buf + copy_off : buf;
3914 memcpy(dst, src, copy_len);
3921 if (!len || next_frag == end_frag)
3925 ptr_buf = skb_frag_address(next_frag);
3926 ptr_len = skb_frag_size(next_frag);
3931 static void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len)
3933 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
3934 u32 size = xdp->data_end - xdp->data;
3935 void *addr = xdp->data;
3938 if (unlikely(offset > 0xffff || len > 0xffff))
3939 return ERR_PTR(-EFAULT);
3941 if (offset + len > xdp_get_buff_len(xdp))
3942 return ERR_PTR(-EINVAL);
3944 if (offset < size) /* linear area */
3948 for (i = 0; i < sinfo->nr_frags; i++) { /* paged area */
3949 u32 frag_size = skb_frag_size(&sinfo->frags[i]);
3951 if (offset < frag_size) {
3952 addr = skb_frag_address(&sinfo->frags[i]);
3956 offset -= frag_size;
3959 return offset + len <= size ? addr + offset : NULL;
3962 BPF_CALL_4(bpf_xdp_load_bytes, struct xdp_buff *, xdp, u32, offset,
3963 void *, buf, u32, len)
3967 ptr = bpf_xdp_pointer(xdp, offset, len);
3969 return PTR_ERR(ptr);
3972 bpf_xdp_copy_buf(xdp, offset, buf, len, false);
3974 memcpy(buf, ptr, len);
3979 static const struct bpf_func_proto bpf_xdp_load_bytes_proto = {
3980 .func = bpf_xdp_load_bytes,
3982 .ret_type = RET_INTEGER,
3983 .arg1_type = ARG_PTR_TO_CTX,
3984 .arg2_type = ARG_ANYTHING,
3985 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
3986 .arg4_type = ARG_CONST_SIZE,
3989 BPF_CALL_4(bpf_xdp_store_bytes, struct xdp_buff *, xdp, u32, offset,
3990 void *, buf, u32, len)
3994 ptr = bpf_xdp_pointer(xdp, offset, len);
3996 return PTR_ERR(ptr);
3999 bpf_xdp_copy_buf(xdp, offset, buf, len, true);
4001 memcpy(ptr, buf, len);
4006 static const struct bpf_func_proto bpf_xdp_store_bytes_proto = {
4007 .func = bpf_xdp_store_bytes,
4009 .ret_type = RET_INTEGER,
4010 .arg1_type = ARG_PTR_TO_CTX,
4011 .arg2_type = ARG_ANYTHING,
4012 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
4013 .arg4_type = ARG_CONST_SIZE,
4016 static int bpf_xdp_frags_increase_tail(struct xdp_buff *xdp, int offset)
4018 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
4019 skb_frag_t *frag = &sinfo->frags[sinfo->nr_frags - 1];
4020 struct xdp_rxq_info *rxq = xdp->rxq;
4021 unsigned int tailroom;
4023 if (!rxq->frag_size || rxq->frag_size > xdp->frame_sz)
4026 tailroom = rxq->frag_size - skb_frag_size(frag) - skb_frag_off(frag);
4027 if (unlikely(offset > tailroom))
4030 memset(skb_frag_address(frag) + skb_frag_size(frag), 0, offset);
4031 skb_frag_size_add(frag, offset);
4032 sinfo->xdp_frags_size += offset;
4037 static int bpf_xdp_frags_shrink_tail(struct xdp_buff *xdp, int offset)
4039 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
4040 int i, n_frags_free = 0, len_free = 0;
4042 if (unlikely(offset > (int)xdp_get_buff_len(xdp) - ETH_HLEN))
4045 for (i = sinfo->nr_frags - 1; i >= 0 && offset > 0; i--) {
4046 skb_frag_t *frag = &sinfo->frags[i];
4047 int shrink = min_t(int, offset, skb_frag_size(frag));
4052 if (skb_frag_size(frag) == shrink) {
4053 struct page *page = skb_frag_page(frag);
4055 __xdp_return(page_address(page), &xdp->rxq->mem,
4059 skb_frag_size_sub(frag, shrink);
4063 sinfo->nr_frags -= n_frags_free;
4064 sinfo->xdp_frags_size -= len_free;
4066 if (unlikely(!sinfo->nr_frags)) {
4067 xdp_buff_clear_frags_flag(xdp);
4068 xdp->data_end -= offset;
4074 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
4076 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
4077 void *data_end = xdp->data_end + offset;
4079 if (unlikely(xdp_buff_has_frags(xdp))) { /* non-linear xdp buff */
4081 return bpf_xdp_frags_shrink_tail(xdp, -offset);
4083 return bpf_xdp_frags_increase_tail(xdp, offset);
4086 /* Notice that xdp_data_hard_end have reserved some tailroom */
4087 if (unlikely(data_end > data_hard_end))
4090 if (unlikely(data_end < xdp->data + ETH_HLEN))
4093 /* Clear memory area on grow, can contain uninit kernel memory */
4095 memset(xdp->data_end, 0, offset);
4097 xdp->data_end = data_end;
4102 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
4103 .func = bpf_xdp_adjust_tail,
4105 .ret_type = RET_INTEGER,
4106 .arg1_type = ARG_PTR_TO_CTX,
4107 .arg2_type = ARG_ANYTHING,
4110 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
4112 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
4113 void *meta = xdp->data_meta + offset;
4114 unsigned long metalen = xdp->data - meta;
4116 if (xdp_data_meta_unsupported(xdp))
4118 if (unlikely(meta < xdp_frame_end ||
4121 if (unlikely(xdp_metalen_invalid(metalen)))
4124 xdp->data_meta = meta;
4129 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
4130 .func = bpf_xdp_adjust_meta,
4132 .ret_type = RET_INTEGER,
4133 .arg1_type = ARG_PTR_TO_CTX,
4134 .arg2_type = ARG_ANYTHING,
4137 /* XDP_REDIRECT works by a three-step process, implemented in the functions
4140 * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target
4141 * of the redirect and store it (along with some other metadata) in a per-CPU
4142 * struct bpf_redirect_info.
4144 * 2. When the program returns the XDP_REDIRECT return code, the driver will
4145 * call xdp_do_redirect() which will use the information in struct
4146 * bpf_redirect_info to actually enqueue the frame into a map type-specific
4147 * bulk queue structure.
4149 * 3. Before exiting its NAPI poll loop, the driver will call xdp_do_flush(),
4150 * which will flush all the different bulk queues, thus completing the
4153 * Pointers to the map entries will be kept around for this whole sequence of
4154 * steps, protected by RCU. However, there is no top-level rcu_read_lock() in
4155 * the core code; instead, the RCU protection relies on everything happening
4156 * inside a single NAPI poll sequence, which means it's between a pair of calls
4157 * to local_bh_disable()/local_bh_enable().
4159 * The map entries are marked as __rcu and the map code makes sure to
4160 * dereference those pointers with rcu_dereference_check() in a way that works
4161 * for both sections that to hold an rcu_read_lock() and sections that are
4162 * called from NAPI without a separate rcu_read_lock(). The code below does not
4163 * use RCU annotations, but relies on those in the map code.
4165 void xdp_do_flush(void)
4171 EXPORT_SYMBOL_GPL(xdp_do_flush);
4173 void bpf_clear_redirect_map(struct bpf_map *map)
4175 struct bpf_redirect_info *ri;
4178 for_each_possible_cpu(cpu) {
4179 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
4180 /* Avoid polluting remote cacheline due to writes if
4181 * not needed. Once we pass this test, we need the
4182 * cmpxchg() to make sure it hasn't been changed in
4183 * the meantime by remote CPU.
4185 if (unlikely(READ_ONCE(ri->map) == map))
4186 cmpxchg(&ri->map, map, NULL);
4190 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
4191 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key);
4193 u32 xdp_master_redirect(struct xdp_buff *xdp)
4195 struct net_device *master, *slave;
4196 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4198 master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev);
4199 slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp);
4200 if (slave && slave != xdp->rxq->dev) {
4201 /* The target device is different from the receiving device, so
4202 * redirect it to the new device.
4203 * Using XDP_REDIRECT gets the correct behaviour from XDP enabled
4204 * drivers to unmap the packet from their rx ring.
4206 ri->tgt_index = slave->ifindex;
4207 ri->map_id = INT_MAX;
4208 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4209 return XDP_REDIRECT;
4213 EXPORT_SYMBOL_GPL(xdp_master_redirect);
4215 static inline int __xdp_do_redirect_xsk(struct bpf_redirect_info *ri,
4216 struct net_device *dev,
4217 struct xdp_buff *xdp,
4218 struct bpf_prog *xdp_prog)
4220 enum bpf_map_type map_type = ri->map_type;
4221 void *fwd = ri->tgt_value;
4222 u32 map_id = ri->map_id;
4225 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4226 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4228 err = __xsk_map_redirect(fwd, xdp);
4232 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4235 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4239 static __always_inline int __xdp_do_redirect_frame(struct bpf_redirect_info *ri,
4240 struct net_device *dev,
4241 struct xdp_frame *xdpf,
4242 struct bpf_prog *xdp_prog)
4244 enum bpf_map_type map_type = ri->map_type;
4245 void *fwd = ri->tgt_value;
4246 u32 map_id = ri->map_id;
4247 struct bpf_map *map;
4250 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4251 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4253 if (unlikely(!xdpf)) {
4259 case BPF_MAP_TYPE_DEVMAP:
4261 case BPF_MAP_TYPE_DEVMAP_HASH:
4262 map = READ_ONCE(ri->map);
4263 if (unlikely(map)) {
4264 WRITE_ONCE(ri->map, NULL);
4265 err = dev_map_enqueue_multi(xdpf, dev, map,
4266 ri->flags & BPF_F_EXCLUDE_INGRESS);
4268 err = dev_map_enqueue(fwd, xdpf, dev);
4271 case BPF_MAP_TYPE_CPUMAP:
4272 err = cpu_map_enqueue(fwd, xdpf, dev);
4274 case BPF_MAP_TYPE_UNSPEC:
4275 if (map_id == INT_MAX) {
4276 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4277 if (unlikely(!fwd)) {
4281 err = dev_xdp_enqueue(fwd, xdpf, dev);
4292 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4295 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4299 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
4300 struct bpf_prog *xdp_prog)
4302 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4303 enum bpf_map_type map_type = ri->map_type;
4305 /* XDP_REDIRECT is not fully supported yet for xdp frags since
4306 * not all XDP capable drivers can map non-linear xdp_frame in
4309 if (unlikely(xdp_buff_has_frags(xdp) &&
4310 map_type != BPF_MAP_TYPE_CPUMAP))
4313 if (map_type == BPF_MAP_TYPE_XSKMAP)
4314 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4316 return __xdp_do_redirect_frame(ri, dev, xdp_convert_buff_to_frame(xdp),
4319 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4321 int xdp_do_redirect_frame(struct net_device *dev, struct xdp_buff *xdp,
4322 struct xdp_frame *xdpf, struct bpf_prog *xdp_prog)
4324 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4325 enum bpf_map_type map_type = ri->map_type;
4327 if (map_type == BPF_MAP_TYPE_XSKMAP)
4328 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4330 return __xdp_do_redirect_frame(ri, dev, xdpf, xdp_prog);
4332 EXPORT_SYMBOL_GPL(xdp_do_redirect_frame);
4334 static int xdp_do_generic_redirect_map(struct net_device *dev,
4335 struct sk_buff *skb,
4336 struct xdp_buff *xdp,
4337 struct bpf_prog *xdp_prog,
4339 enum bpf_map_type map_type, u32 map_id)
4341 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4342 struct bpf_map *map;
4346 case BPF_MAP_TYPE_DEVMAP:
4348 case BPF_MAP_TYPE_DEVMAP_HASH:
4349 map = READ_ONCE(ri->map);
4350 if (unlikely(map)) {
4351 WRITE_ONCE(ri->map, NULL);
4352 err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4353 ri->flags & BPF_F_EXCLUDE_INGRESS);
4355 err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4360 case BPF_MAP_TYPE_XSKMAP:
4361 err = xsk_generic_rcv(fwd, xdp);
4366 case BPF_MAP_TYPE_CPUMAP:
4367 err = cpu_map_generic_redirect(fwd, skb);
4376 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4379 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4383 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4384 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4386 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4387 enum bpf_map_type map_type = ri->map_type;
4388 void *fwd = ri->tgt_value;
4389 u32 map_id = ri->map_id;
4392 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4393 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4395 if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4396 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4397 if (unlikely(!fwd)) {
4402 err = xdp_ok_fwd_dev(fwd, skb->len);
4407 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4408 generic_xdp_tx(skb, xdp_prog);
4412 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id);
4414 _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4418 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4420 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4422 if (unlikely(flags))
4425 /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4426 * by map_idr) is used for ifindex based XDP redirect.
4428 ri->tgt_index = ifindex;
4429 ri->map_id = INT_MAX;
4430 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4432 return XDP_REDIRECT;
4435 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4436 .func = bpf_xdp_redirect,
4438 .ret_type = RET_INTEGER,
4439 .arg1_type = ARG_ANYTHING,
4440 .arg2_type = ARG_ANYTHING,
4443 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
4446 return map->ops->map_redirect(map, ifindex, flags);
4449 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4450 .func = bpf_xdp_redirect_map,
4452 .ret_type = RET_INTEGER,
4453 .arg1_type = ARG_CONST_MAP_PTR,
4454 .arg2_type = ARG_ANYTHING,
4455 .arg3_type = ARG_ANYTHING,
4458 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4459 unsigned long off, unsigned long len)
4461 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4465 if (ptr != dst_buff)
4466 memcpy(dst_buff, ptr, len);
4471 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4472 u64, flags, void *, meta, u64, meta_size)
4474 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4476 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4478 if (unlikely(!skb || skb_size > skb->len))
4481 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4485 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4486 .func = bpf_skb_event_output,
4488 .ret_type = RET_INTEGER,
4489 .arg1_type = ARG_PTR_TO_CTX,
4490 .arg2_type = ARG_CONST_MAP_PTR,
4491 .arg3_type = ARG_ANYTHING,
4492 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4493 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4496 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4498 const struct bpf_func_proto bpf_skb_output_proto = {
4499 .func = bpf_skb_event_output,
4501 .ret_type = RET_INTEGER,
4502 .arg1_type = ARG_PTR_TO_BTF_ID,
4503 .arg1_btf_id = &bpf_skb_output_btf_ids[0],
4504 .arg2_type = ARG_CONST_MAP_PTR,
4505 .arg3_type = ARG_ANYTHING,
4506 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4507 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4510 static unsigned short bpf_tunnel_key_af(u64 flags)
4512 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4515 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4516 u32, size, u64, flags)
4518 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4519 u8 compat[sizeof(struct bpf_tunnel_key)];
4523 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6 |
4524 BPF_F_TUNINFO_FLAGS)))) {
4528 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4532 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4535 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4536 case offsetof(struct bpf_tunnel_key, tunnel_label):
4537 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4539 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4540 /* Fixup deprecated structure layouts here, so we have
4541 * a common path later on.
4543 if (ip_tunnel_info_af(info) != AF_INET)
4546 to = (struct bpf_tunnel_key *)compat;
4553 to->tunnel_id = be64_to_cpu(info->key.tun_id);
4554 to->tunnel_tos = info->key.tos;
4555 to->tunnel_ttl = info->key.ttl;
4556 if (flags & BPF_F_TUNINFO_FLAGS)
4557 to->tunnel_flags = info->key.tun_flags;
4561 if (flags & BPF_F_TUNINFO_IPV6) {
4562 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4563 sizeof(to->remote_ipv6));
4564 memcpy(to->local_ipv6, &info->key.u.ipv6.dst,
4565 sizeof(to->local_ipv6));
4566 to->tunnel_label = be32_to_cpu(info->key.label);
4568 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4569 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4570 to->local_ipv4 = be32_to_cpu(info->key.u.ipv4.dst);
4571 memset(&to->local_ipv6[1], 0, sizeof(__u32) * 3);
4572 to->tunnel_label = 0;
4575 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4576 memcpy(to_orig, to, size);
4580 memset(to_orig, 0, size);
4584 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4585 .func = bpf_skb_get_tunnel_key,
4587 .ret_type = RET_INTEGER,
4588 .arg1_type = ARG_PTR_TO_CTX,
4589 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4590 .arg3_type = ARG_CONST_SIZE,
4591 .arg4_type = ARG_ANYTHING,
4594 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4596 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4599 if (unlikely(!info ||
4600 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4604 if (unlikely(size < info->options_len)) {
4609 ip_tunnel_info_opts_get(to, info);
4610 if (size > info->options_len)
4611 memset(to + info->options_len, 0, size - info->options_len);
4613 return info->options_len;
4615 memset(to, 0, size);
4619 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4620 .func = bpf_skb_get_tunnel_opt,
4622 .ret_type = RET_INTEGER,
4623 .arg1_type = ARG_PTR_TO_CTX,
4624 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4625 .arg3_type = ARG_CONST_SIZE,
4628 static struct metadata_dst __percpu *md_dst;
4630 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4631 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4633 struct metadata_dst *md = this_cpu_ptr(md_dst);
4634 u8 compat[sizeof(struct bpf_tunnel_key)];
4635 struct ip_tunnel_info *info;
4637 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4638 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
4640 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4642 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4643 case offsetof(struct bpf_tunnel_key, tunnel_label):
4644 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4645 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4646 /* Fixup deprecated structure layouts here, so we have
4647 * a common path later on.
4649 memcpy(compat, from, size);
4650 memset(compat + size, 0, sizeof(compat) - size);
4651 from = (const struct bpf_tunnel_key *) compat;
4657 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4662 dst_hold((struct dst_entry *) md);
4663 skb_dst_set(skb, (struct dst_entry *) md);
4665 info = &md->u.tun_info;
4666 memset(info, 0, sizeof(*info));
4667 info->mode = IP_TUNNEL_INFO_TX;
4669 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4670 if (flags & BPF_F_DONT_FRAGMENT)
4671 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4672 if (flags & BPF_F_ZERO_CSUM_TX)
4673 info->key.tun_flags &= ~TUNNEL_CSUM;
4674 if (flags & BPF_F_SEQ_NUMBER)
4675 info->key.tun_flags |= TUNNEL_SEQ;
4677 info->key.tun_id = cpu_to_be64(from->tunnel_id);
4678 info->key.tos = from->tunnel_tos;
4679 info->key.ttl = from->tunnel_ttl;
4681 if (flags & BPF_F_TUNINFO_IPV6) {
4682 info->mode |= IP_TUNNEL_INFO_IPV6;
4683 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4684 sizeof(from->remote_ipv6));
4685 memcpy(&info->key.u.ipv6.src, from->local_ipv6,
4686 sizeof(from->local_ipv6));
4687 info->key.label = cpu_to_be32(from->tunnel_label) &
4688 IPV6_FLOWLABEL_MASK;
4690 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4691 info->key.u.ipv4.src = cpu_to_be32(from->local_ipv4);
4692 info->key.flow_flags = FLOWI_FLAG_ANYSRC;
4698 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4699 .func = bpf_skb_set_tunnel_key,
4701 .ret_type = RET_INTEGER,
4702 .arg1_type = ARG_PTR_TO_CTX,
4703 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4704 .arg3_type = ARG_CONST_SIZE,
4705 .arg4_type = ARG_ANYTHING,
4708 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4709 const u8 *, from, u32, size)
4711 struct ip_tunnel_info *info = skb_tunnel_info(skb);
4712 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4714 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4716 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4719 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4724 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4725 .func = bpf_skb_set_tunnel_opt,
4727 .ret_type = RET_INTEGER,
4728 .arg1_type = ARG_PTR_TO_CTX,
4729 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4730 .arg3_type = ARG_CONST_SIZE,
4733 static const struct bpf_func_proto *
4734 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4737 struct metadata_dst __percpu *tmp;
4739 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4744 if (cmpxchg(&md_dst, NULL, tmp))
4745 metadata_dst_free_percpu(tmp);
4749 case BPF_FUNC_skb_set_tunnel_key:
4750 return &bpf_skb_set_tunnel_key_proto;
4751 case BPF_FUNC_skb_set_tunnel_opt:
4752 return &bpf_skb_set_tunnel_opt_proto;
4758 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4761 struct bpf_array *array = container_of(map, struct bpf_array, map);
4762 struct cgroup *cgrp;
4765 sk = skb_to_full_sk(skb);
4766 if (!sk || !sk_fullsock(sk))
4768 if (unlikely(idx >= array->map.max_entries))
4771 cgrp = READ_ONCE(array->ptrs[idx]);
4772 if (unlikely(!cgrp))
4775 return sk_under_cgroup_hierarchy(sk, cgrp);
4778 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4779 .func = bpf_skb_under_cgroup,
4781 .ret_type = RET_INTEGER,
4782 .arg1_type = ARG_PTR_TO_CTX,
4783 .arg2_type = ARG_CONST_MAP_PTR,
4784 .arg3_type = ARG_ANYTHING,
4787 #ifdef CONFIG_SOCK_CGROUP_DATA
4788 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4790 struct cgroup *cgrp;
4792 sk = sk_to_full_sk(sk);
4793 if (!sk || !sk_fullsock(sk))
4796 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4797 return cgroup_id(cgrp);
4800 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4802 return __bpf_sk_cgroup_id(skb->sk);
4805 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4806 .func = bpf_skb_cgroup_id,
4808 .ret_type = RET_INTEGER,
4809 .arg1_type = ARG_PTR_TO_CTX,
4812 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4815 struct cgroup *ancestor;
4816 struct cgroup *cgrp;
4818 sk = sk_to_full_sk(sk);
4819 if (!sk || !sk_fullsock(sk))
4822 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4823 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4827 return cgroup_id(ancestor);
4830 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4833 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4836 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4837 .func = bpf_skb_ancestor_cgroup_id,
4839 .ret_type = RET_INTEGER,
4840 .arg1_type = ARG_PTR_TO_CTX,
4841 .arg2_type = ARG_ANYTHING,
4844 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4846 return __bpf_sk_cgroup_id(sk);
4849 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4850 .func = bpf_sk_cgroup_id,
4852 .ret_type = RET_INTEGER,
4853 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4856 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4858 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4861 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4862 .func = bpf_sk_ancestor_cgroup_id,
4864 .ret_type = RET_INTEGER,
4865 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4866 .arg2_type = ARG_ANYTHING,
4870 static unsigned long bpf_xdp_copy(void *dst, const void *ctx,
4871 unsigned long off, unsigned long len)
4873 struct xdp_buff *xdp = (struct xdp_buff *)ctx;
4875 bpf_xdp_copy_buf(xdp, off, dst, len, false);
4879 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4880 u64, flags, void *, meta, u64, meta_size)
4882 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4884 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4887 if (unlikely(!xdp || xdp_size > xdp_get_buff_len(xdp)))
4890 return bpf_event_output(map, flags, meta, meta_size, xdp,
4891 xdp_size, bpf_xdp_copy);
4894 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4895 .func = bpf_xdp_event_output,
4897 .ret_type = RET_INTEGER,
4898 .arg1_type = ARG_PTR_TO_CTX,
4899 .arg2_type = ARG_CONST_MAP_PTR,
4900 .arg3_type = ARG_ANYTHING,
4901 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4902 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4905 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4907 const struct bpf_func_proto bpf_xdp_output_proto = {
4908 .func = bpf_xdp_event_output,
4910 .ret_type = RET_INTEGER,
4911 .arg1_type = ARG_PTR_TO_BTF_ID,
4912 .arg1_btf_id = &bpf_xdp_output_btf_ids[0],
4913 .arg2_type = ARG_CONST_MAP_PTR,
4914 .arg3_type = ARG_ANYTHING,
4915 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4916 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4919 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4921 return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4924 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4925 .func = bpf_get_socket_cookie,
4927 .ret_type = RET_INTEGER,
4928 .arg1_type = ARG_PTR_TO_CTX,
4931 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4933 return __sock_gen_cookie(ctx->sk);
4936 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4937 .func = bpf_get_socket_cookie_sock_addr,
4939 .ret_type = RET_INTEGER,
4940 .arg1_type = ARG_PTR_TO_CTX,
4943 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4945 return __sock_gen_cookie(ctx);
4948 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4949 .func = bpf_get_socket_cookie_sock,
4951 .ret_type = RET_INTEGER,
4952 .arg1_type = ARG_PTR_TO_CTX,
4955 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
4957 return sk ? sock_gen_cookie(sk) : 0;
4960 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
4961 .func = bpf_get_socket_ptr_cookie,
4963 .ret_type = RET_INTEGER,
4964 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4967 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4969 return __sock_gen_cookie(ctx->sk);
4972 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4973 .func = bpf_get_socket_cookie_sock_ops,
4975 .ret_type = RET_INTEGER,
4976 .arg1_type = ARG_PTR_TO_CTX,
4979 static u64 __bpf_get_netns_cookie(struct sock *sk)
4981 const struct net *net = sk ? sock_net(sk) : &init_net;
4983 return net->net_cookie;
4986 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
4988 return __bpf_get_netns_cookie(ctx);
4991 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
4992 .func = bpf_get_netns_cookie_sock,
4994 .ret_type = RET_INTEGER,
4995 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
4998 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
5000 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5003 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
5004 .func = bpf_get_netns_cookie_sock_addr,
5006 .ret_type = RET_INTEGER,
5007 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5010 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
5012 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5015 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = {
5016 .func = bpf_get_netns_cookie_sock_ops,
5018 .ret_type = RET_INTEGER,
5019 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5022 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx)
5024 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5027 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = {
5028 .func = bpf_get_netns_cookie_sk_msg,
5030 .ret_type = RET_INTEGER,
5031 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5034 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
5036 struct sock *sk = sk_to_full_sk(skb->sk);
5039 if (!sk || !sk_fullsock(sk))
5041 kuid = sock_net_uid(sock_net(sk), sk);
5042 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
5045 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
5046 .func = bpf_get_socket_uid,
5048 .ret_type = RET_INTEGER,
5049 .arg1_type = ARG_PTR_TO_CTX,
5052 static int sol_socket_sockopt(struct sock *sk, int optname,
5053 char *optval, int *optlen,
5065 case SO_MAX_PACING_RATE:
5066 case SO_BINDTOIFINDEX:
5068 if (*optlen != sizeof(int))
5071 case SO_BINDTODEVICE:
5078 if (optname == SO_BINDTODEVICE)
5080 return sk_getsockopt(sk, SOL_SOCKET, optname,
5081 KERNEL_SOCKPTR(optval),
5082 KERNEL_SOCKPTR(optlen));
5085 return sk_setsockopt(sk, SOL_SOCKET, optname,
5086 KERNEL_SOCKPTR(optval), *optlen);
5089 static int bpf_sol_tcp_setsockopt(struct sock *sk, int optname,
5090 char *optval, int optlen)
5092 struct tcp_sock *tp = tcp_sk(sk);
5093 unsigned long timeout;
5096 if (optlen != sizeof(int))
5099 val = *(int *)optval;
5101 /* Only some options are supported */
5104 if (val <= 0 || tp->data_segs_out > tp->syn_data)
5106 tcp_snd_cwnd_set(tp, val);
5108 case TCP_BPF_SNDCWND_CLAMP:
5111 tp->snd_cwnd_clamp = val;
5112 tp->snd_ssthresh = val;
5114 case TCP_BPF_DELACK_MAX:
5115 timeout = usecs_to_jiffies(val);
5116 if (timeout > TCP_DELACK_MAX ||
5117 timeout < TCP_TIMEOUT_MIN)
5119 inet_csk(sk)->icsk_delack_max = timeout;
5121 case TCP_BPF_RTO_MIN:
5122 timeout = usecs_to_jiffies(val);
5123 if (timeout > TCP_RTO_MIN ||
5124 timeout < TCP_TIMEOUT_MIN)
5126 inet_csk(sk)->icsk_rto_min = timeout;
5135 static int sol_tcp_sockopt_congestion(struct sock *sk, char *optval,
5136 int *optlen, bool getopt)
5138 struct tcp_sock *tp;
5145 if (!inet_csk(sk)->icsk_ca_ops)
5147 /* BPF expects NULL-terminated tcp-cc string */
5148 optval[--(*optlen)] = '\0';
5149 return do_tcp_getsockopt(sk, SOL_TCP, TCP_CONGESTION,
5150 KERNEL_SOCKPTR(optval),
5151 KERNEL_SOCKPTR(optlen));
5154 /* "cdg" is the only cc that alloc a ptr
5155 * in inet_csk_ca area. The bpf-tcp-cc may
5156 * overwrite this ptr after switching to cdg.
5158 if (*optlen >= sizeof("cdg") - 1 && !strncmp("cdg", optval, *optlen))
5161 /* It stops this looping
5163 * .init => bpf_setsockopt(tcp_cc) => .init =>
5164 * bpf_setsockopt(tcp_cc)" => .init => ....
5166 * The second bpf_setsockopt(tcp_cc) is not allowed
5167 * in order to break the loop when both .init
5168 * are the same bpf prog.
5170 * This applies even the second bpf_setsockopt(tcp_cc)
5171 * does not cause a loop. This limits only the first
5172 * '.init' can call bpf_setsockopt(TCP_CONGESTION) to
5173 * pick a fallback cc (eg. peer does not support ECN)
5174 * and the second '.init' cannot fallback to
5178 if (tp->bpf_chg_cc_inprogress)
5181 tp->bpf_chg_cc_inprogress = 1;
5182 ret = do_tcp_setsockopt(sk, SOL_TCP, TCP_CONGESTION,
5183 KERNEL_SOCKPTR(optval), *optlen);
5184 tp->bpf_chg_cc_inprogress = 0;
5188 static int sol_tcp_sockopt(struct sock *sk, int optname,
5189 char *optval, int *optlen,
5192 if (sk->sk_prot->setsockopt != tcp_setsockopt)
5202 case TCP_WINDOW_CLAMP:
5203 case TCP_THIN_LINEAR_TIMEOUTS:
5204 case TCP_USER_TIMEOUT:
5205 case TCP_NOTSENT_LOWAT:
5207 if (*optlen != sizeof(int))
5210 case TCP_CONGESTION:
5211 return sol_tcp_sockopt_congestion(sk, optval, optlen, getopt);
5219 return bpf_sol_tcp_setsockopt(sk, optname, optval, *optlen);
5223 if (optname == TCP_SAVED_SYN) {
5224 struct tcp_sock *tp = tcp_sk(sk);
5226 if (!tp->saved_syn ||
5227 *optlen > tcp_saved_syn_len(tp->saved_syn))
5229 memcpy(optval, tp->saved_syn->data, *optlen);
5230 /* It cannot free tp->saved_syn here because it
5231 * does not know if the user space still needs it.
5236 return do_tcp_getsockopt(sk, SOL_TCP, optname,
5237 KERNEL_SOCKPTR(optval),
5238 KERNEL_SOCKPTR(optlen));
5241 return do_tcp_setsockopt(sk, SOL_TCP, optname,
5242 KERNEL_SOCKPTR(optval), *optlen);
5245 static int sol_ip_sockopt(struct sock *sk, int optname,
5246 char *optval, int *optlen,
5249 if (sk->sk_family != AF_INET)
5254 if (*optlen != sizeof(int))
5262 return do_ip_getsockopt(sk, SOL_IP, optname,
5263 KERNEL_SOCKPTR(optval),
5264 KERNEL_SOCKPTR(optlen));
5266 return do_ip_setsockopt(sk, SOL_IP, optname,
5267 KERNEL_SOCKPTR(optval), *optlen);
5270 static int sol_ipv6_sockopt(struct sock *sk, int optname,
5271 char *optval, int *optlen,
5274 if (sk->sk_family != AF_INET6)
5279 case IPV6_AUTOFLOWLABEL:
5280 if (*optlen != sizeof(int))
5288 return ipv6_bpf_stub->ipv6_getsockopt(sk, SOL_IPV6, optname,
5289 KERNEL_SOCKPTR(optval),
5290 KERNEL_SOCKPTR(optlen));
5292 return ipv6_bpf_stub->ipv6_setsockopt(sk, SOL_IPV6, optname,
5293 KERNEL_SOCKPTR(optval), *optlen);
5296 static int __bpf_setsockopt(struct sock *sk, int level, int optname,
5297 char *optval, int optlen)
5299 if (!sk_fullsock(sk))
5302 if (level == SOL_SOCKET)
5303 return sol_socket_sockopt(sk, optname, optval, &optlen, false);
5304 else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5305 return sol_ip_sockopt(sk, optname, optval, &optlen, false);
5306 else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5307 return sol_ipv6_sockopt(sk, optname, optval, &optlen, false);
5308 else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5309 return sol_tcp_sockopt(sk, optname, optval, &optlen, false);
5314 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
5315 char *optval, int optlen)
5317 if (sk_fullsock(sk))
5318 sock_owned_by_me(sk);
5319 return __bpf_setsockopt(sk, level, optname, optval, optlen);
5322 static int __bpf_getsockopt(struct sock *sk, int level, int optname,
5323 char *optval, int optlen)
5325 int err, saved_optlen = optlen;
5327 if (!sk_fullsock(sk)) {
5332 if (level == SOL_SOCKET)
5333 err = sol_socket_sockopt(sk, optname, optval, &optlen, true);
5334 else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5335 err = sol_tcp_sockopt(sk, optname, optval, &optlen, true);
5336 else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5337 err = sol_ip_sockopt(sk, optname, optval, &optlen, true);
5338 else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5339 err = sol_ipv6_sockopt(sk, optname, optval, &optlen, true);
5346 if (optlen < saved_optlen)
5347 memset(optval + optlen, 0, saved_optlen - optlen);
5351 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
5352 char *optval, int optlen)
5354 if (sk_fullsock(sk))
5355 sock_owned_by_me(sk);
5356 return __bpf_getsockopt(sk, level, optname, optval, optlen);
5359 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level,
5360 int, optname, char *, optval, int, optlen)
5362 return _bpf_setsockopt(sk, level, optname, optval, optlen);
5365 const struct bpf_func_proto bpf_sk_setsockopt_proto = {
5366 .func = bpf_sk_setsockopt,
5368 .ret_type = RET_INTEGER,
5369 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5370 .arg2_type = ARG_ANYTHING,
5371 .arg3_type = ARG_ANYTHING,
5372 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5373 .arg5_type = ARG_CONST_SIZE,
5376 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level,
5377 int, optname, char *, optval, int, optlen)
5379 return _bpf_getsockopt(sk, level, optname, optval, optlen);
5382 const struct bpf_func_proto bpf_sk_getsockopt_proto = {
5383 .func = bpf_sk_getsockopt,
5385 .ret_type = RET_INTEGER,
5386 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5387 .arg2_type = ARG_ANYTHING,
5388 .arg3_type = ARG_ANYTHING,
5389 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5390 .arg5_type = ARG_CONST_SIZE,
5393 BPF_CALL_5(bpf_unlocked_sk_setsockopt, struct sock *, sk, int, level,
5394 int, optname, char *, optval, int, optlen)
5396 return __bpf_setsockopt(sk, level, optname, optval, optlen);
5399 const struct bpf_func_proto bpf_unlocked_sk_setsockopt_proto = {
5400 .func = bpf_unlocked_sk_setsockopt,
5402 .ret_type = RET_INTEGER,
5403 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5404 .arg2_type = ARG_ANYTHING,
5405 .arg3_type = ARG_ANYTHING,
5406 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5407 .arg5_type = ARG_CONST_SIZE,
5410 BPF_CALL_5(bpf_unlocked_sk_getsockopt, struct sock *, sk, int, level,
5411 int, optname, char *, optval, int, optlen)
5413 return __bpf_getsockopt(sk, level, optname, optval, optlen);
5416 const struct bpf_func_proto bpf_unlocked_sk_getsockopt_proto = {
5417 .func = bpf_unlocked_sk_getsockopt,
5419 .ret_type = RET_INTEGER,
5420 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5421 .arg2_type = ARG_ANYTHING,
5422 .arg3_type = ARG_ANYTHING,
5423 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5424 .arg5_type = ARG_CONST_SIZE,
5427 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5428 int, level, int, optname, char *, optval, int, optlen)
5430 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5433 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5434 .func = bpf_sock_addr_setsockopt,
5436 .ret_type = RET_INTEGER,
5437 .arg1_type = ARG_PTR_TO_CTX,
5438 .arg2_type = ARG_ANYTHING,
5439 .arg3_type = ARG_ANYTHING,
5440 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5441 .arg5_type = ARG_CONST_SIZE,
5444 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5445 int, level, int, optname, char *, optval, int, optlen)
5447 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5450 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5451 .func = bpf_sock_addr_getsockopt,
5453 .ret_type = RET_INTEGER,
5454 .arg1_type = ARG_PTR_TO_CTX,
5455 .arg2_type = ARG_ANYTHING,
5456 .arg3_type = ARG_ANYTHING,
5457 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5458 .arg5_type = ARG_CONST_SIZE,
5461 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5462 int, level, int, optname, char *, optval, int, optlen)
5464 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5467 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5468 .func = bpf_sock_ops_setsockopt,
5470 .ret_type = RET_INTEGER,
5471 .arg1_type = ARG_PTR_TO_CTX,
5472 .arg2_type = ARG_ANYTHING,
5473 .arg3_type = ARG_ANYTHING,
5474 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5475 .arg5_type = ARG_CONST_SIZE,
5478 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5479 int optname, const u8 **start)
5481 struct sk_buff *syn_skb = bpf_sock->syn_skb;
5482 const u8 *hdr_start;
5486 /* sk is a request_sock here */
5488 if (optname == TCP_BPF_SYN) {
5489 hdr_start = syn_skb->data;
5490 ret = tcp_hdrlen(syn_skb);
5491 } else if (optname == TCP_BPF_SYN_IP) {
5492 hdr_start = skb_network_header(syn_skb);
5493 ret = skb_network_header_len(syn_skb) +
5494 tcp_hdrlen(syn_skb);
5496 /* optname == TCP_BPF_SYN_MAC */
5497 hdr_start = skb_mac_header(syn_skb);
5498 ret = skb_mac_header_len(syn_skb) +
5499 skb_network_header_len(syn_skb) +
5500 tcp_hdrlen(syn_skb);
5503 struct sock *sk = bpf_sock->sk;
5504 struct saved_syn *saved_syn;
5506 if (sk->sk_state == TCP_NEW_SYN_RECV)
5507 /* synack retransmit. bpf_sock->syn_skb will
5508 * not be available. It has to resort to
5509 * saved_syn (if it is saved).
5511 saved_syn = inet_reqsk(sk)->saved_syn;
5513 saved_syn = tcp_sk(sk)->saved_syn;
5518 if (optname == TCP_BPF_SYN) {
5519 hdr_start = saved_syn->data +
5520 saved_syn->mac_hdrlen +
5521 saved_syn->network_hdrlen;
5522 ret = saved_syn->tcp_hdrlen;
5523 } else if (optname == TCP_BPF_SYN_IP) {
5524 hdr_start = saved_syn->data +
5525 saved_syn->mac_hdrlen;
5526 ret = saved_syn->network_hdrlen +
5527 saved_syn->tcp_hdrlen;
5529 /* optname == TCP_BPF_SYN_MAC */
5531 /* TCP_SAVE_SYN may not have saved the mac hdr */
5532 if (!saved_syn->mac_hdrlen)
5535 hdr_start = saved_syn->data;
5536 ret = saved_syn->mac_hdrlen +
5537 saved_syn->network_hdrlen +
5538 saved_syn->tcp_hdrlen;
5546 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5547 int, level, int, optname, char *, optval, int, optlen)
5549 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5550 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5551 int ret, copy_len = 0;
5554 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5557 if (optlen < copy_len) {
5562 memcpy(optval, start, copy_len);
5565 /* Zero out unused buffer at the end */
5566 memset(optval + copy_len, 0, optlen - copy_len);
5571 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5574 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5575 .func = bpf_sock_ops_getsockopt,
5577 .ret_type = RET_INTEGER,
5578 .arg1_type = ARG_PTR_TO_CTX,
5579 .arg2_type = ARG_ANYTHING,
5580 .arg3_type = ARG_ANYTHING,
5581 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5582 .arg5_type = ARG_CONST_SIZE,
5585 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5588 struct sock *sk = bpf_sock->sk;
5589 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5591 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5594 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5596 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5599 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5600 .func = bpf_sock_ops_cb_flags_set,
5602 .ret_type = RET_INTEGER,
5603 .arg1_type = ARG_PTR_TO_CTX,
5604 .arg2_type = ARG_ANYTHING,
5607 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5608 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5610 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5614 struct sock *sk = ctx->sk;
5615 u32 flags = BIND_FROM_BPF;
5619 if (addr_len < offsetofend(struct sockaddr, sa_family))
5621 if (addr->sa_family == AF_INET) {
5622 if (addr_len < sizeof(struct sockaddr_in))
5624 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5625 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5626 return __inet_bind(sk, addr, addr_len, flags);
5627 #if IS_ENABLED(CONFIG_IPV6)
5628 } else if (addr->sa_family == AF_INET6) {
5629 if (addr_len < SIN6_LEN_RFC2133)
5631 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5632 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5633 /* ipv6_bpf_stub cannot be NULL, since it's called from
5634 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5636 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5637 #endif /* CONFIG_IPV6 */
5639 #endif /* CONFIG_INET */
5641 return -EAFNOSUPPORT;
5644 static const struct bpf_func_proto bpf_bind_proto = {
5647 .ret_type = RET_INTEGER,
5648 .arg1_type = ARG_PTR_TO_CTX,
5649 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5650 .arg3_type = ARG_CONST_SIZE,
5654 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5655 struct bpf_xfrm_state *, to, u32, size, u64, flags)
5657 const struct sec_path *sp = skb_sec_path(skb);
5658 const struct xfrm_state *x;
5660 if (!sp || unlikely(index >= sp->len || flags))
5663 x = sp->xvec[index];
5665 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5668 to->reqid = x->props.reqid;
5669 to->spi = x->id.spi;
5670 to->family = x->props.family;
5673 if (to->family == AF_INET6) {
5674 memcpy(to->remote_ipv6, x->props.saddr.a6,
5675 sizeof(to->remote_ipv6));
5677 to->remote_ipv4 = x->props.saddr.a4;
5678 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5683 memset(to, 0, size);
5687 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5688 .func = bpf_skb_get_xfrm_state,
5690 .ret_type = RET_INTEGER,
5691 .arg1_type = ARG_PTR_TO_CTX,
5692 .arg2_type = ARG_ANYTHING,
5693 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
5694 .arg4_type = ARG_CONST_SIZE,
5695 .arg5_type = ARG_ANYTHING,
5699 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5700 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params, u32 mtu)
5702 params->h_vlan_TCI = 0;
5703 params->h_vlan_proto = 0;
5705 params->mtu_result = mtu; /* union with tot_len */
5711 #if IS_ENABLED(CONFIG_INET)
5712 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5713 u32 flags, bool check_mtu)
5715 struct fib_nh_common *nhc;
5716 struct in_device *in_dev;
5717 struct neighbour *neigh;
5718 struct net_device *dev;
5719 struct fib_result res;
5724 dev = dev_get_by_index_rcu(net, params->ifindex);
5728 /* verify forwarding is enabled on this interface */
5729 in_dev = __in_dev_get_rcu(dev);
5730 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5731 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5733 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5735 fl4.flowi4_oif = params->ifindex;
5737 fl4.flowi4_iif = params->ifindex;
5740 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5741 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5742 fl4.flowi4_flags = 0;
5744 fl4.flowi4_proto = params->l4_protocol;
5745 fl4.daddr = params->ipv4_dst;
5746 fl4.saddr = params->ipv4_src;
5747 fl4.fl4_sport = params->sport;
5748 fl4.fl4_dport = params->dport;
5749 fl4.flowi4_multipath_hash = 0;
5751 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5752 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5753 struct fib_table *tb;
5755 if (flags & BPF_FIB_LOOKUP_TBID) {
5756 tbid = params->tbid;
5757 /* zero out for vlan output */
5761 tb = fib_get_table(net, tbid);
5763 return BPF_FIB_LKUP_RET_NOT_FWDED;
5765 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5767 fl4.flowi4_mark = 0;
5768 fl4.flowi4_secid = 0;
5769 fl4.flowi4_tun_key.tun_id = 0;
5770 fl4.flowi4_uid = sock_net_uid(net, NULL);
5772 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5776 /* map fib lookup errors to RTN_ type */
5778 return BPF_FIB_LKUP_RET_BLACKHOLE;
5779 if (err == -EHOSTUNREACH)
5780 return BPF_FIB_LKUP_RET_UNREACHABLE;
5782 return BPF_FIB_LKUP_RET_PROHIBIT;
5784 return BPF_FIB_LKUP_RET_NOT_FWDED;
5787 if (res.type != RTN_UNICAST)
5788 return BPF_FIB_LKUP_RET_NOT_FWDED;
5790 if (fib_info_num_path(res.fi) > 1)
5791 fib_select_path(net, &res, &fl4, NULL);
5794 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5795 if (params->tot_len > mtu) {
5796 params->mtu_result = mtu; /* union with tot_len */
5797 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5803 /* do not handle lwt encaps right now */
5804 if (nhc->nhc_lwtstate)
5805 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5809 params->rt_metric = res.fi->fib_priority;
5810 params->ifindex = dev->ifindex;
5812 if (flags & BPF_FIB_LOOKUP_SRC)
5813 params->ipv4_src = fib_result_prefsrc(net, &res);
5815 /* xdp and cls_bpf programs are run in RCU-bh so
5816 * rcu_read_lock_bh is not needed here
5818 if (likely(nhc->nhc_gw_family != AF_INET6)) {
5819 if (nhc->nhc_gw_family)
5820 params->ipv4_dst = nhc->nhc_gw.ipv4;
5822 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5824 params->family = AF_INET6;
5825 *dst = nhc->nhc_gw.ipv6;
5828 if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH)
5829 goto set_fwd_params;
5831 if (likely(nhc->nhc_gw_family != AF_INET6))
5832 neigh = __ipv4_neigh_lookup_noref(dev,
5833 (__force u32)params->ipv4_dst);
5835 neigh = __ipv6_neigh_lookup_noref_stub(dev, params->ipv6_dst);
5837 if (!neigh || !(READ_ONCE(neigh->nud_state) & NUD_VALID))
5838 return BPF_FIB_LKUP_RET_NO_NEIGH;
5839 memcpy(params->dmac, neigh->ha, ETH_ALEN);
5840 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5843 return bpf_fib_set_fwd_params(params, mtu);
5847 #if IS_ENABLED(CONFIG_IPV6)
5848 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5849 u32 flags, bool check_mtu)
5851 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5852 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5853 struct fib6_result res = {};
5854 struct neighbour *neigh;
5855 struct net_device *dev;
5856 struct inet6_dev *idev;
5862 /* link local addresses are never forwarded */
5863 if (rt6_need_strict(dst) || rt6_need_strict(src))
5864 return BPF_FIB_LKUP_RET_NOT_FWDED;
5866 dev = dev_get_by_index_rcu(net, params->ifindex);
5870 idev = __in6_dev_get_safely(dev);
5871 if (unlikely(!idev || !idev->cnf.forwarding))
5872 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5874 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5876 oif = fl6.flowi6_oif = params->ifindex;
5878 oif = fl6.flowi6_iif = params->ifindex;
5880 strict = RT6_LOOKUP_F_HAS_SADDR;
5882 fl6.flowlabel = params->flowinfo;
5883 fl6.flowi6_scope = 0;
5884 fl6.flowi6_flags = 0;
5887 fl6.flowi6_proto = params->l4_protocol;
5890 fl6.fl6_sport = params->sport;
5891 fl6.fl6_dport = params->dport;
5893 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5894 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5895 struct fib6_table *tb;
5897 if (flags & BPF_FIB_LOOKUP_TBID) {
5898 tbid = params->tbid;
5899 /* zero out for vlan output */
5903 tb = ipv6_stub->fib6_get_table(net, tbid);
5905 return BPF_FIB_LKUP_RET_NOT_FWDED;
5907 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5910 fl6.flowi6_mark = 0;
5911 fl6.flowi6_secid = 0;
5912 fl6.flowi6_tun_key.tun_id = 0;
5913 fl6.flowi6_uid = sock_net_uid(net, NULL);
5915 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5918 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5919 res.f6i == net->ipv6.fib6_null_entry))
5920 return BPF_FIB_LKUP_RET_NOT_FWDED;
5922 switch (res.fib6_type) {
5923 /* only unicast is forwarded */
5927 return BPF_FIB_LKUP_RET_BLACKHOLE;
5928 case RTN_UNREACHABLE:
5929 return BPF_FIB_LKUP_RET_UNREACHABLE;
5931 return BPF_FIB_LKUP_RET_PROHIBIT;
5933 return BPF_FIB_LKUP_RET_NOT_FWDED;
5936 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5937 fl6.flowi6_oif != 0, NULL, strict);
5940 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5941 if (params->tot_len > mtu) {
5942 params->mtu_result = mtu; /* union with tot_len */
5943 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5947 if (res.nh->fib_nh_lws)
5948 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5950 if (res.nh->fib_nh_gw_family)
5951 *dst = res.nh->fib_nh_gw6;
5953 dev = res.nh->fib_nh_dev;
5954 params->rt_metric = res.f6i->fib6_metric;
5955 params->ifindex = dev->ifindex;
5957 if (flags & BPF_FIB_LOOKUP_SRC) {
5958 if (res.f6i->fib6_prefsrc.plen) {
5959 *src = res.f6i->fib6_prefsrc.addr;
5961 err = ipv6_bpf_stub->ipv6_dev_get_saddr(net, dev,
5965 return BPF_FIB_LKUP_RET_NO_SRC_ADDR;
5969 if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH)
5970 goto set_fwd_params;
5972 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
5975 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5976 if (!neigh || !(READ_ONCE(neigh->nud_state) & NUD_VALID))
5977 return BPF_FIB_LKUP_RET_NO_NEIGH;
5978 memcpy(params->dmac, neigh->ha, ETH_ALEN);
5979 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5982 return bpf_fib_set_fwd_params(params, mtu);
5986 #define BPF_FIB_LOOKUP_MASK (BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT | \
5987 BPF_FIB_LOOKUP_SKIP_NEIGH | BPF_FIB_LOOKUP_TBID | \
5990 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
5991 struct bpf_fib_lookup *, params, int, plen, u32, flags)
5993 if (plen < sizeof(*params))
5996 if (flags & ~BPF_FIB_LOOKUP_MASK)
5999 switch (params->family) {
6000 #if IS_ENABLED(CONFIG_INET)
6002 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
6005 #if IS_ENABLED(CONFIG_IPV6)
6007 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
6011 return -EAFNOSUPPORT;
6014 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
6015 .func = bpf_xdp_fib_lookup,
6017 .ret_type = RET_INTEGER,
6018 .arg1_type = ARG_PTR_TO_CTX,
6019 .arg2_type = ARG_PTR_TO_MEM,
6020 .arg3_type = ARG_CONST_SIZE,
6021 .arg4_type = ARG_ANYTHING,
6024 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
6025 struct bpf_fib_lookup *, params, int, plen, u32, flags)
6027 struct net *net = dev_net(skb->dev);
6028 int rc = -EAFNOSUPPORT;
6029 bool check_mtu = false;
6031 if (plen < sizeof(*params))
6034 if (flags & ~BPF_FIB_LOOKUP_MASK)
6037 if (params->tot_len)
6040 switch (params->family) {
6041 #if IS_ENABLED(CONFIG_INET)
6043 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
6046 #if IS_ENABLED(CONFIG_IPV6)
6048 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
6053 if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
6054 struct net_device *dev;
6056 /* When tot_len isn't provided by user, check skb
6057 * against MTU of FIB lookup resulting net_device
6059 dev = dev_get_by_index_rcu(net, params->ifindex);
6060 if (!is_skb_forwardable(dev, skb))
6061 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
6063 params->mtu_result = dev->mtu; /* union with tot_len */
6069 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
6070 .func = bpf_skb_fib_lookup,
6072 .ret_type = RET_INTEGER,
6073 .arg1_type = ARG_PTR_TO_CTX,
6074 .arg2_type = ARG_PTR_TO_MEM,
6075 .arg3_type = ARG_CONST_SIZE,
6076 .arg4_type = ARG_ANYTHING,
6079 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
6082 struct net *netns = dev_net(dev_curr);
6084 /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
6088 return dev_get_by_index_rcu(netns, ifindex);
6091 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
6092 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6094 int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6095 struct net_device *dev = skb->dev;
6096 int skb_len, dev_len;
6099 if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
6102 if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
6105 dev = __dev_via_ifindex(dev, ifindex);
6109 mtu = READ_ONCE(dev->mtu);
6111 dev_len = mtu + dev->hard_header_len;
6113 /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6114 skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
6116 skb_len += len_diff; /* minus result pass check */
6117 if (skb_len <= dev_len) {
6118 ret = BPF_MTU_CHK_RET_SUCCESS;
6121 /* At this point, skb->len exceed MTU, but as it include length of all
6122 * segments, it can still be below MTU. The SKB can possibly get
6123 * re-segmented in transmit path (see validate_xmit_skb). Thus, user
6124 * must choose if segs are to be MTU checked.
6126 if (skb_is_gso(skb)) {
6127 ret = BPF_MTU_CHK_RET_SUCCESS;
6129 if (flags & BPF_MTU_CHK_SEGS &&
6130 !skb_gso_validate_network_len(skb, mtu))
6131 ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
6134 /* BPF verifier guarantees valid pointer */
6140 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
6141 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6143 struct net_device *dev = xdp->rxq->dev;
6144 int xdp_len = xdp->data_end - xdp->data;
6145 int ret = BPF_MTU_CHK_RET_SUCCESS;
6148 /* XDP variant doesn't support multi-buffer segment check (yet) */
6149 if (unlikely(flags))
6152 dev = __dev_via_ifindex(dev, ifindex);
6156 mtu = READ_ONCE(dev->mtu);
6158 /* Add L2-header as dev MTU is L3 size */
6159 dev_len = mtu + dev->hard_header_len;
6161 /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6163 xdp_len = *mtu_len + dev->hard_header_len;
6165 xdp_len += len_diff; /* minus result pass check */
6166 if (xdp_len > dev_len)
6167 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6169 /* BPF verifier guarantees valid pointer */
6175 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
6176 .func = bpf_skb_check_mtu,
6178 .ret_type = RET_INTEGER,
6179 .arg1_type = ARG_PTR_TO_CTX,
6180 .arg2_type = ARG_ANYTHING,
6181 .arg3_type = ARG_PTR_TO_INT,
6182 .arg4_type = ARG_ANYTHING,
6183 .arg5_type = ARG_ANYTHING,
6186 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
6187 .func = bpf_xdp_check_mtu,
6189 .ret_type = RET_INTEGER,
6190 .arg1_type = ARG_PTR_TO_CTX,
6191 .arg2_type = ARG_ANYTHING,
6192 .arg3_type = ARG_PTR_TO_INT,
6193 .arg4_type = ARG_ANYTHING,
6194 .arg5_type = ARG_ANYTHING,
6197 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6198 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
6201 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
6203 if (!seg6_validate_srh(srh, len, false))
6207 case BPF_LWT_ENCAP_SEG6_INLINE:
6208 if (skb->protocol != htons(ETH_P_IPV6))
6211 err = seg6_do_srh_inline(skb, srh);
6213 case BPF_LWT_ENCAP_SEG6:
6214 skb_reset_inner_headers(skb);
6215 skb->encapsulation = 1;
6216 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
6222 bpf_compute_data_pointers(skb);
6226 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
6228 return seg6_lookup_nexthop(skb, NULL, 0);
6230 #endif /* CONFIG_IPV6_SEG6_BPF */
6232 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6233 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
6236 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
6240 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
6244 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6245 case BPF_LWT_ENCAP_SEG6:
6246 case BPF_LWT_ENCAP_SEG6_INLINE:
6247 return bpf_push_seg6_encap(skb, type, hdr, len);
6249 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6250 case BPF_LWT_ENCAP_IP:
6251 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
6258 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
6259 void *, hdr, u32, len)
6262 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6263 case BPF_LWT_ENCAP_IP:
6264 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
6271 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
6272 .func = bpf_lwt_in_push_encap,
6274 .ret_type = RET_INTEGER,
6275 .arg1_type = ARG_PTR_TO_CTX,
6276 .arg2_type = ARG_ANYTHING,
6277 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6278 .arg4_type = ARG_CONST_SIZE
6281 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
6282 .func = bpf_lwt_xmit_push_encap,
6284 .ret_type = RET_INTEGER,
6285 .arg1_type = ARG_PTR_TO_CTX,
6286 .arg2_type = ARG_ANYTHING,
6287 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6288 .arg4_type = ARG_CONST_SIZE
6291 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6292 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
6293 const void *, from, u32, len)
6295 struct seg6_bpf_srh_state *srh_state =
6296 this_cpu_ptr(&seg6_bpf_srh_states);
6297 struct ipv6_sr_hdr *srh = srh_state->srh;
6298 void *srh_tlvs, *srh_end, *ptr;
6304 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
6305 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
6307 ptr = skb->data + offset;
6308 if (ptr >= srh_tlvs && ptr + len <= srh_end)
6309 srh_state->valid = false;
6310 else if (ptr < (void *)&srh->flags ||
6311 ptr + len > (void *)&srh->segments)
6314 if (unlikely(bpf_try_make_writable(skb, offset + len)))
6316 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6318 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6320 memcpy(skb->data + offset, from, len);
6324 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
6325 .func = bpf_lwt_seg6_store_bytes,
6327 .ret_type = RET_INTEGER,
6328 .arg1_type = ARG_PTR_TO_CTX,
6329 .arg2_type = ARG_ANYTHING,
6330 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6331 .arg4_type = ARG_CONST_SIZE
6334 static void bpf_update_srh_state(struct sk_buff *skb)
6336 struct seg6_bpf_srh_state *srh_state =
6337 this_cpu_ptr(&seg6_bpf_srh_states);
6340 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
6341 srh_state->srh = NULL;
6343 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6344 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
6345 srh_state->valid = true;
6349 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
6350 u32, action, void *, param, u32, param_len)
6352 struct seg6_bpf_srh_state *srh_state =
6353 this_cpu_ptr(&seg6_bpf_srh_states);
6358 case SEG6_LOCAL_ACTION_END_X:
6359 if (!seg6_bpf_has_valid_srh(skb))
6361 if (param_len != sizeof(struct in6_addr))
6363 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
6364 case SEG6_LOCAL_ACTION_END_T:
6365 if (!seg6_bpf_has_valid_srh(skb))
6367 if (param_len != sizeof(int))
6369 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6370 case SEG6_LOCAL_ACTION_END_DT6:
6371 if (!seg6_bpf_has_valid_srh(skb))
6373 if (param_len != sizeof(int))
6376 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
6378 if (!pskb_pull(skb, hdroff))
6381 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
6382 skb_reset_network_header(skb);
6383 skb_reset_transport_header(skb);
6384 skb->encapsulation = 0;
6386 bpf_compute_data_pointers(skb);
6387 bpf_update_srh_state(skb);
6388 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6389 case SEG6_LOCAL_ACTION_END_B6:
6390 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6392 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
6395 bpf_update_srh_state(skb);
6398 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
6399 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6401 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
6404 bpf_update_srh_state(skb);
6412 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
6413 .func = bpf_lwt_seg6_action,
6415 .ret_type = RET_INTEGER,
6416 .arg1_type = ARG_PTR_TO_CTX,
6417 .arg2_type = ARG_ANYTHING,
6418 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6419 .arg4_type = ARG_CONST_SIZE
6422 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
6425 struct seg6_bpf_srh_state *srh_state =
6426 this_cpu_ptr(&seg6_bpf_srh_states);
6427 struct ipv6_sr_hdr *srh = srh_state->srh;
6428 void *srh_end, *srh_tlvs, *ptr;
6429 struct ipv6hdr *hdr;
6433 if (unlikely(srh == NULL))
6436 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
6437 ((srh->first_segment + 1) << 4));
6438 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
6440 ptr = skb->data + offset;
6442 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
6444 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
6448 ret = skb_cow_head(skb, len);
6449 if (unlikely(ret < 0))
6452 ret = bpf_skb_net_hdr_push(skb, offset, len);
6454 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6457 bpf_compute_data_pointers(skb);
6458 if (unlikely(ret < 0))
6461 hdr = (struct ipv6hdr *)skb->data;
6462 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6464 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6466 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6467 srh_state->hdrlen += len;
6468 srh_state->valid = false;
6472 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6473 .func = bpf_lwt_seg6_adjust_srh,
6475 .ret_type = RET_INTEGER,
6476 .arg1_type = ARG_PTR_TO_CTX,
6477 .arg2_type = ARG_ANYTHING,
6478 .arg3_type = ARG_ANYTHING,
6480 #endif /* CONFIG_IPV6_SEG6_BPF */
6483 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6484 int dif, int sdif, u8 family, u8 proto)
6486 struct inet_hashinfo *hinfo = net->ipv4.tcp_death_row.hashinfo;
6487 bool refcounted = false;
6488 struct sock *sk = NULL;
6490 if (family == AF_INET) {
6491 __be32 src4 = tuple->ipv4.saddr;
6492 __be32 dst4 = tuple->ipv4.daddr;
6494 if (proto == IPPROTO_TCP)
6495 sk = __inet_lookup(net, hinfo, NULL, 0,
6496 src4, tuple->ipv4.sport,
6497 dst4, tuple->ipv4.dport,
6498 dif, sdif, &refcounted);
6500 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6501 dst4, tuple->ipv4.dport,
6502 dif, sdif, &udp_table, NULL);
6503 #if IS_ENABLED(CONFIG_IPV6)
6505 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6506 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6508 if (proto == IPPROTO_TCP)
6509 sk = __inet6_lookup(net, hinfo, NULL, 0,
6510 src6, tuple->ipv6.sport,
6511 dst6, ntohs(tuple->ipv6.dport),
6512 dif, sdif, &refcounted);
6513 else if (likely(ipv6_bpf_stub))
6514 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6515 src6, tuple->ipv6.sport,
6516 dst6, tuple->ipv6.dport,
6522 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6523 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6529 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6530 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6532 static struct sock *
6533 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6534 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6535 u64 flags, int sdif)
6537 struct sock *sk = NULL;
6541 if (len == sizeof(tuple->ipv4))
6543 else if (len == sizeof(tuple->ipv6))
6548 if (unlikely(flags || !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6552 if (family == AF_INET)
6553 sdif = inet_sdif(skb);
6555 sdif = inet6_sdif(skb);
6558 if ((s32)netns_id < 0) {
6560 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6562 net = get_net_ns_by_id(caller_net, netns_id);
6565 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6573 static struct sock *
6574 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6575 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6576 u64 flags, int sdif)
6578 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6579 ifindex, proto, netns_id, flags,
6583 struct sock *sk2 = sk_to_full_sk(sk);
6585 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6586 * sock refcnt is decremented to prevent a request_sock leak.
6588 if (!sk_fullsock(sk2))
6592 /* Ensure there is no need to bump sk2 refcnt */
6593 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6594 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6604 static struct sock *
6605 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6606 u8 proto, u64 netns_id, u64 flags)
6608 struct net *caller_net;
6612 caller_net = dev_net(skb->dev);
6613 ifindex = skb->dev->ifindex;
6615 caller_net = sock_net(skb->sk);
6619 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6620 netns_id, flags, -1);
6623 static struct sock *
6624 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6625 u8 proto, u64 netns_id, u64 flags)
6627 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6631 struct sock *sk2 = sk_to_full_sk(sk);
6633 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6634 * sock refcnt is decremented to prevent a request_sock leak.
6636 if (!sk_fullsock(sk2))
6640 /* Ensure there is no need to bump sk2 refcnt */
6641 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6642 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6652 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6653 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6655 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6659 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6660 .func = bpf_skc_lookup_tcp,
6663 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6664 .arg1_type = ARG_PTR_TO_CTX,
6665 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6666 .arg3_type = ARG_CONST_SIZE,
6667 .arg4_type = ARG_ANYTHING,
6668 .arg5_type = ARG_ANYTHING,
6671 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6672 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6674 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6678 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6679 .func = bpf_sk_lookup_tcp,
6682 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6683 .arg1_type = ARG_PTR_TO_CTX,
6684 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6685 .arg3_type = ARG_CONST_SIZE,
6686 .arg4_type = ARG_ANYTHING,
6687 .arg5_type = ARG_ANYTHING,
6690 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6691 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6693 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6697 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6698 .func = bpf_sk_lookup_udp,
6701 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6702 .arg1_type = ARG_PTR_TO_CTX,
6703 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6704 .arg3_type = ARG_CONST_SIZE,
6705 .arg4_type = ARG_ANYTHING,
6706 .arg5_type = ARG_ANYTHING,
6709 BPF_CALL_5(bpf_tc_skc_lookup_tcp, struct sk_buff *, skb,
6710 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6712 struct net_device *dev = skb->dev;
6713 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6714 struct net *caller_net = dev_net(dev);
6716 return (unsigned long)__bpf_skc_lookup(skb, tuple, len, caller_net,
6717 ifindex, IPPROTO_TCP, netns_id,
6721 static const struct bpf_func_proto bpf_tc_skc_lookup_tcp_proto = {
6722 .func = bpf_tc_skc_lookup_tcp,
6725 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6726 .arg1_type = ARG_PTR_TO_CTX,
6727 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6728 .arg3_type = ARG_CONST_SIZE,
6729 .arg4_type = ARG_ANYTHING,
6730 .arg5_type = ARG_ANYTHING,
6733 BPF_CALL_5(bpf_tc_sk_lookup_tcp, struct sk_buff *, skb,
6734 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6736 struct net_device *dev = skb->dev;
6737 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6738 struct net *caller_net = dev_net(dev);
6740 return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6741 ifindex, IPPROTO_TCP, netns_id,
6745 static const struct bpf_func_proto bpf_tc_sk_lookup_tcp_proto = {
6746 .func = bpf_tc_sk_lookup_tcp,
6749 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6750 .arg1_type = ARG_PTR_TO_CTX,
6751 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6752 .arg3_type = ARG_CONST_SIZE,
6753 .arg4_type = ARG_ANYTHING,
6754 .arg5_type = ARG_ANYTHING,
6757 BPF_CALL_5(bpf_tc_sk_lookup_udp, struct sk_buff *, skb,
6758 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6760 struct net_device *dev = skb->dev;
6761 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6762 struct net *caller_net = dev_net(dev);
6764 return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6765 ifindex, IPPROTO_UDP, netns_id,
6769 static const struct bpf_func_proto bpf_tc_sk_lookup_udp_proto = {
6770 .func = bpf_tc_sk_lookup_udp,
6773 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6774 .arg1_type = ARG_PTR_TO_CTX,
6775 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6776 .arg3_type = ARG_CONST_SIZE,
6777 .arg4_type = ARG_ANYTHING,
6778 .arg5_type = ARG_ANYTHING,
6781 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6783 if (sk && sk_is_refcounted(sk))
6788 static const struct bpf_func_proto bpf_sk_release_proto = {
6789 .func = bpf_sk_release,
6791 .ret_type = RET_INTEGER,
6792 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON | OBJ_RELEASE,
6795 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6796 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6798 struct net_device *dev = ctx->rxq->dev;
6799 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6800 struct net *caller_net = dev_net(dev);
6802 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6803 ifindex, IPPROTO_UDP, netns_id,
6807 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6808 .func = bpf_xdp_sk_lookup_udp,
6811 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6812 .arg1_type = ARG_PTR_TO_CTX,
6813 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6814 .arg3_type = ARG_CONST_SIZE,
6815 .arg4_type = ARG_ANYTHING,
6816 .arg5_type = ARG_ANYTHING,
6819 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6820 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6822 struct net_device *dev = ctx->rxq->dev;
6823 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6824 struct net *caller_net = dev_net(dev);
6826 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6827 ifindex, IPPROTO_TCP, netns_id,
6831 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6832 .func = bpf_xdp_skc_lookup_tcp,
6835 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6836 .arg1_type = ARG_PTR_TO_CTX,
6837 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6838 .arg3_type = ARG_CONST_SIZE,
6839 .arg4_type = ARG_ANYTHING,
6840 .arg5_type = ARG_ANYTHING,
6843 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6844 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6846 struct net_device *dev = ctx->rxq->dev;
6847 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6848 struct net *caller_net = dev_net(dev);
6850 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6851 ifindex, IPPROTO_TCP, netns_id,
6855 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6856 .func = bpf_xdp_sk_lookup_tcp,
6859 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6860 .arg1_type = ARG_PTR_TO_CTX,
6861 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6862 .arg3_type = ARG_CONST_SIZE,
6863 .arg4_type = ARG_ANYTHING,
6864 .arg5_type = ARG_ANYTHING,
6867 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6868 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6870 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6871 sock_net(ctx->sk), 0,
6872 IPPROTO_TCP, netns_id, flags,
6876 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6877 .func = bpf_sock_addr_skc_lookup_tcp,
6879 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6880 .arg1_type = ARG_PTR_TO_CTX,
6881 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6882 .arg3_type = ARG_CONST_SIZE,
6883 .arg4_type = ARG_ANYTHING,
6884 .arg5_type = ARG_ANYTHING,
6887 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6888 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6890 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6891 sock_net(ctx->sk), 0, IPPROTO_TCP,
6892 netns_id, flags, -1);
6895 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6896 .func = bpf_sock_addr_sk_lookup_tcp,
6898 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6899 .arg1_type = ARG_PTR_TO_CTX,
6900 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6901 .arg3_type = ARG_CONST_SIZE,
6902 .arg4_type = ARG_ANYTHING,
6903 .arg5_type = ARG_ANYTHING,
6906 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6907 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6909 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6910 sock_net(ctx->sk), 0, IPPROTO_UDP,
6911 netns_id, flags, -1);
6914 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6915 .func = bpf_sock_addr_sk_lookup_udp,
6917 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6918 .arg1_type = ARG_PTR_TO_CTX,
6919 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6920 .arg3_type = ARG_CONST_SIZE,
6921 .arg4_type = ARG_ANYTHING,
6922 .arg5_type = ARG_ANYTHING,
6925 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6926 struct bpf_insn_access_aux *info)
6928 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6932 if (off % size != 0)
6936 case offsetof(struct bpf_tcp_sock, bytes_received):
6937 case offsetof(struct bpf_tcp_sock, bytes_acked):
6938 return size == sizeof(__u64);
6940 return size == sizeof(__u32);
6944 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6945 const struct bpf_insn *si,
6946 struct bpf_insn *insn_buf,
6947 struct bpf_prog *prog, u32 *target_size)
6949 struct bpf_insn *insn = insn_buf;
6951 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \
6953 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \
6954 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6955 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6956 si->dst_reg, si->src_reg, \
6957 offsetof(struct tcp_sock, FIELD)); \
6960 #define BPF_INET_SOCK_GET_COMMON(FIELD) \
6962 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \
6964 sizeof_field(struct bpf_tcp_sock, FIELD)); \
6965 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
6966 struct inet_connection_sock, \
6968 si->dst_reg, si->src_reg, \
6970 struct inet_connection_sock, \
6974 if (insn > insn_buf)
6975 return insn - insn_buf;
6978 case offsetof(struct bpf_tcp_sock, rtt_min):
6979 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
6980 sizeof(struct minmax));
6981 BUILD_BUG_ON(sizeof(struct minmax) <
6982 sizeof(struct minmax_sample));
6984 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6985 offsetof(struct tcp_sock, rtt_min) +
6986 offsetof(struct minmax_sample, v));
6988 case offsetof(struct bpf_tcp_sock, snd_cwnd):
6989 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
6991 case offsetof(struct bpf_tcp_sock, srtt_us):
6992 BPF_TCP_SOCK_GET_COMMON(srtt_us);
6994 case offsetof(struct bpf_tcp_sock, snd_ssthresh):
6995 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
6997 case offsetof(struct bpf_tcp_sock, rcv_nxt):
6998 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
7000 case offsetof(struct bpf_tcp_sock, snd_nxt):
7001 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
7003 case offsetof(struct bpf_tcp_sock, snd_una):
7004 BPF_TCP_SOCK_GET_COMMON(snd_una);
7006 case offsetof(struct bpf_tcp_sock, mss_cache):
7007 BPF_TCP_SOCK_GET_COMMON(mss_cache);
7009 case offsetof(struct bpf_tcp_sock, ecn_flags):
7010 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
7012 case offsetof(struct bpf_tcp_sock, rate_delivered):
7013 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
7015 case offsetof(struct bpf_tcp_sock, rate_interval_us):
7016 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
7018 case offsetof(struct bpf_tcp_sock, packets_out):
7019 BPF_TCP_SOCK_GET_COMMON(packets_out);
7021 case offsetof(struct bpf_tcp_sock, retrans_out):
7022 BPF_TCP_SOCK_GET_COMMON(retrans_out);
7024 case offsetof(struct bpf_tcp_sock, total_retrans):
7025 BPF_TCP_SOCK_GET_COMMON(total_retrans);
7027 case offsetof(struct bpf_tcp_sock, segs_in):
7028 BPF_TCP_SOCK_GET_COMMON(segs_in);
7030 case offsetof(struct bpf_tcp_sock, data_segs_in):
7031 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
7033 case offsetof(struct bpf_tcp_sock, segs_out):
7034 BPF_TCP_SOCK_GET_COMMON(segs_out);
7036 case offsetof(struct bpf_tcp_sock, data_segs_out):
7037 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
7039 case offsetof(struct bpf_tcp_sock, lost_out):
7040 BPF_TCP_SOCK_GET_COMMON(lost_out);
7042 case offsetof(struct bpf_tcp_sock, sacked_out):
7043 BPF_TCP_SOCK_GET_COMMON(sacked_out);
7045 case offsetof(struct bpf_tcp_sock, bytes_received):
7046 BPF_TCP_SOCK_GET_COMMON(bytes_received);
7048 case offsetof(struct bpf_tcp_sock, bytes_acked):
7049 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
7051 case offsetof(struct bpf_tcp_sock, dsack_dups):
7052 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
7054 case offsetof(struct bpf_tcp_sock, delivered):
7055 BPF_TCP_SOCK_GET_COMMON(delivered);
7057 case offsetof(struct bpf_tcp_sock, delivered_ce):
7058 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
7060 case offsetof(struct bpf_tcp_sock, icsk_retransmits):
7061 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
7065 return insn - insn_buf;
7068 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
7070 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
7071 return (unsigned long)sk;
7073 return (unsigned long)NULL;
7076 const struct bpf_func_proto bpf_tcp_sock_proto = {
7077 .func = bpf_tcp_sock,
7079 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
7080 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
7083 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
7085 sk = sk_to_full_sk(sk);
7087 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
7088 return (unsigned long)sk;
7090 return (unsigned long)NULL;
7093 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
7094 .func = bpf_get_listener_sock,
7096 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
7097 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
7100 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
7102 unsigned int iphdr_len;
7104 switch (skb_protocol(skb, true)) {
7105 case cpu_to_be16(ETH_P_IP):
7106 iphdr_len = sizeof(struct iphdr);
7108 case cpu_to_be16(ETH_P_IPV6):
7109 iphdr_len = sizeof(struct ipv6hdr);
7115 if (skb_headlen(skb) < iphdr_len)
7118 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
7121 return INET_ECN_set_ce(skb);
7124 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7125 struct bpf_insn_access_aux *info)
7127 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
7130 if (off % size != 0)
7135 return size == sizeof(__u32);
7139 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
7140 const struct bpf_insn *si,
7141 struct bpf_insn *insn_buf,
7142 struct bpf_prog *prog, u32 *target_size)
7144 struct bpf_insn *insn = insn_buf;
7146 #define BPF_XDP_SOCK_GET(FIELD) \
7148 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \
7149 sizeof_field(struct bpf_xdp_sock, FIELD)); \
7150 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
7151 si->dst_reg, si->src_reg, \
7152 offsetof(struct xdp_sock, FIELD)); \
7156 case offsetof(struct bpf_xdp_sock, queue_id):
7157 BPF_XDP_SOCK_GET(queue_id);
7161 return insn - insn_buf;
7164 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
7165 .func = bpf_skb_ecn_set_ce,
7167 .ret_type = RET_INTEGER,
7168 .arg1_type = ARG_PTR_TO_CTX,
7171 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7172 struct tcphdr *, th, u32, th_len)
7174 #ifdef CONFIG_SYN_COOKIES
7178 if (unlikely(!sk || th_len < sizeof(*th)))
7181 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
7182 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7185 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7188 if (!th->ack || th->rst || th->syn)
7191 if (unlikely(iph_len < sizeof(struct iphdr)))
7194 if (tcp_synq_no_recent_overflow(sk))
7197 cookie = ntohl(th->ack_seq) - 1;
7199 /* Both struct iphdr and struct ipv6hdr have the version field at the
7200 * same offset so we can cast to the shorter header (struct iphdr).
7202 switch (((struct iphdr *)iph)->version) {
7204 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7207 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
7210 #if IS_BUILTIN(CONFIG_IPV6)
7212 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7215 if (sk->sk_family != AF_INET6)
7218 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
7220 #endif /* CONFIG_IPV6 */
7223 return -EPROTONOSUPPORT;
7235 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
7236 .func = bpf_tcp_check_syncookie,
7239 .ret_type = RET_INTEGER,
7240 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7241 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7242 .arg3_type = ARG_CONST_SIZE,
7243 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7244 .arg5_type = ARG_CONST_SIZE,
7247 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7248 struct tcphdr *, th, u32, th_len)
7250 #ifdef CONFIG_SYN_COOKIES
7254 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
7257 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7260 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7263 if (!th->syn || th->ack || th->fin || th->rst)
7266 if (unlikely(iph_len < sizeof(struct iphdr)))
7269 /* Both struct iphdr and struct ipv6hdr have the version field at the
7270 * same offset so we can cast to the shorter header (struct iphdr).
7272 switch (((struct iphdr *)iph)->version) {
7274 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7277 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
7280 #if IS_BUILTIN(CONFIG_IPV6)
7282 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7285 if (sk->sk_family != AF_INET6)
7288 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
7290 #endif /* CONFIG_IPV6 */
7293 return -EPROTONOSUPPORT;
7298 return cookie | ((u64)mss << 32);
7301 #endif /* CONFIG_SYN_COOKIES */
7304 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
7305 .func = bpf_tcp_gen_syncookie,
7306 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */
7308 .ret_type = RET_INTEGER,
7309 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7310 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7311 .arg3_type = ARG_CONST_SIZE,
7312 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7313 .arg5_type = ARG_CONST_SIZE,
7316 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
7318 if (!sk || flags != 0)
7320 if (!skb_at_tc_ingress(skb))
7322 if (unlikely(dev_net(skb->dev) != sock_net(sk)))
7323 return -ENETUNREACH;
7324 if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
7325 return -ESOCKTNOSUPPORT;
7326 if (sk_unhashed(sk))
7328 if (sk_is_refcounted(sk) &&
7329 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
7334 skb->destructor = sock_pfree;
7339 static const struct bpf_func_proto bpf_sk_assign_proto = {
7340 .func = bpf_sk_assign,
7342 .ret_type = RET_INTEGER,
7343 .arg1_type = ARG_PTR_TO_CTX,
7344 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7345 .arg3_type = ARG_ANYTHING,
7348 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
7349 u8 search_kind, const u8 *magic,
7350 u8 magic_len, bool *eol)
7356 while (op < opend) {
7359 if (kind == TCPOPT_EOL) {
7361 return ERR_PTR(-ENOMSG);
7362 } else if (kind == TCPOPT_NOP) {
7367 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
7368 /* Something is wrong in the received header.
7369 * Follow the TCP stack's tcp_parse_options()
7370 * and just bail here.
7372 return ERR_PTR(-EFAULT);
7375 if (search_kind == kind) {
7379 if (magic_len > kind_len - 2)
7380 return ERR_PTR(-ENOMSG);
7382 if (!memcmp(&op[2], magic, magic_len))
7389 return ERR_PTR(-ENOMSG);
7392 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7393 void *, search_res, u32, len, u64, flags)
7395 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
7396 const u8 *op, *opend, *magic, *search = search_res;
7397 u8 search_kind, search_len, copy_len, magic_len;
7400 /* 2 byte is the minimal option len except TCPOPT_NOP and
7401 * TCPOPT_EOL which are useless for the bpf prog to learn
7402 * and this helper disallow loading them also.
7404 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
7407 search_kind = search[0];
7408 search_len = search[1];
7410 if (search_len > len || search_kind == TCPOPT_NOP ||
7411 search_kind == TCPOPT_EOL)
7414 if (search_kind == TCPOPT_EXP || search_kind == 253) {
7415 /* 16 or 32 bit magic. +2 for kind and kind length */
7416 if (search_len != 4 && search_len != 6)
7419 magic_len = search_len - 2;
7428 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
7433 op += sizeof(struct tcphdr);
7435 if (!bpf_sock->skb ||
7436 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7437 /* This bpf_sock->op cannot call this helper */
7440 opend = bpf_sock->skb_data_end;
7441 op = bpf_sock->skb->data + sizeof(struct tcphdr);
7444 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
7451 if (copy_len > len) {
7456 memcpy(search_res, op, copy_len);
7460 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
7461 .func = bpf_sock_ops_load_hdr_opt,
7463 .ret_type = RET_INTEGER,
7464 .arg1_type = ARG_PTR_TO_CTX,
7465 .arg2_type = ARG_PTR_TO_MEM,
7466 .arg3_type = ARG_CONST_SIZE,
7467 .arg4_type = ARG_ANYTHING,
7470 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7471 const void *, from, u32, len, u64, flags)
7473 u8 new_kind, new_kind_len, magic_len = 0, *opend;
7474 const u8 *op, *new_op, *magic = NULL;
7475 struct sk_buff *skb;
7478 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
7481 if (len < 2 || flags)
7485 new_kind = new_op[0];
7486 new_kind_len = new_op[1];
7488 if (new_kind_len > len || new_kind == TCPOPT_NOP ||
7489 new_kind == TCPOPT_EOL)
7492 if (new_kind_len > bpf_sock->remaining_opt_len)
7495 /* 253 is another experimental kind */
7496 if (new_kind == TCPOPT_EXP || new_kind == 253) {
7497 if (new_kind_len < 4)
7499 /* Match for the 2 byte magic also.
7500 * RFC 6994: the magic could be 2 or 4 bytes.
7501 * Hence, matching by 2 byte only is on the
7502 * conservative side but it is the right
7503 * thing to do for the 'search-for-duplication'
7510 /* Check for duplication */
7511 skb = bpf_sock->skb;
7512 op = skb->data + sizeof(struct tcphdr);
7513 opend = bpf_sock->skb_data_end;
7515 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
7520 if (PTR_ERR(op) != -ENOMSG)
7524 /* The option has been ended. Treat it as no more
7525 * header option can be written.
7529 /* No duplication found. Store the header option. */
7530 memcpy(opend, from, new_kind_len);
7532 bpf_sock->remaining_opt_len -= new_kind_len;
7533 bpf_sock->skb_data_end += new_kind_len;
7538 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
7539 .func = bpf_sock_ops_store_hdr_opt,
7541 .ret_type = RET_INTEGER,
7542 .arg1_type = ARG_PTR_TO_CTX,
7543 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7544 .arg3_type = ARG_CONST_SIZE,
7545 .arg4_type = ARG_ANYTHING,
7548 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7549 u32, len, u64, flags)
7551 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7554 if (flags || len < 2)
7557 if (len > bpf_sock->remaining_opt_len)
7560 bpf_sock->remaining_opt_len -= len;
7565 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7566 .func = bpf_sock_ops_reserve_hdr_opt,
7568 .ret_type = RET_INTEGER,
7569 .arg1_type = ARG_PTR_TO_CTX,
7570 .arg2_type = ARG_ANYTHING,
7571 .arg3_type = ARG_ANYTHING,
7574 BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb,
7575 u64, tstamp, u32, tstamp_type)
7577 /* skb_clear_delivery_time() is done for inet protocol */
7578 if (skb->protocol != htons(ETH_P_IP) &&
7579 skb->protocol != htons(ETH_P_IPV6))
7582 switch (tstamp_type) {
7583 case BPF_SKB_TSTAMP_DELIVERY_MONO:
7586 skb->tstamp = tstamp;
7587 skb->mono_delivery_time = 1;
7589 case BPF_SKB_TSTAMP_UNSPEC:
7593 skb->mono_delivery_time = 0;
7602 static const struct bpf_func_proto bpf_skb_set_tstamp_proto = {
7603 .func = bpf_skb_set_tstamp,
7605 .ret_type = RET_INTEGER,
7606 .arg1_type = ARG_PTR_TO_CTX,
7607 .arg2_type = ARG_ANYTHING,
7608 .arg3_type = ARG_ANYTHING,
7611 #ifdef CONFIG_SYN_COOKIES
7612 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv4, struct iphdr *, iph,
7613 struct tcphdr *, th, u32, th_len)
7618 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7621 mss = tcp_parse_mss_option(th, 0) ?: TCP_MSS_DEFAULT;
7622 cookie = __cookie_v4_init_sequence(iph, th, &mss);
7624 return cookie | ((u64)mss << 32);
7627 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv4_proto = {
7628 .func = bpf_tcp_raw_gen_syncookie_ipv4,
7629 .gpl_only = true, /* __cookie_v4_init_sequence() is GPL */
7631 .ret_type = RET_INTEGER,
7632 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7633 .arg1_size = sizeof(struct iphdr),
7634 .arg2_type = ARG_PTR_TO_MEM,
7635 .arg3_type = ARG_CONST_SIZE,
7638 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv6, struct ipv6hdr *, iph,
7639 struct tcphdr *, th, u32, th_len)
7641 #if IS_BUILTIN(CONFIG_IPV6)
7642 const u16 mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
7643 sizeof(struct ipv6hdr);
7647 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7650 mss = tcp_parse_mss_option(th, 0) ?: mss_clamp;
7651 cookie = __cookie_v6_init_sequence(iph, th, &mss);
7653 return cookie | ((u64)mss << 32);
7655 return -EPROTONOSUPPORT;
7659 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv6_proto = {
7660 .func = bpf_tcp_raw_gen_syncookie_ipv6,
7661 .gpl_only = true, /* __cookie_v6_init_sequence() is GPL */
7663 .ret_type = RET_INTEGER,
7664 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7665 .arg1_size = sizeof(struct ipv6hdr),
7666 .arg2_type = ARG_PTR_TO_MEM,
7667 .arg3_type = ARG_CONST_SIZE,
7670 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv4, struct iphdr *, iph,
7671 struct tcphdr *, th)
7673 u32 cookie = ntohl(th->ack_seq) - 1;
7675 if (__cookie_v4_check(iph, th, cookie) > 0)
7681 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv4_proto = {
7682 .func = bpf_tcp_raw_check_syncookie_ipv4,
7683 .gpl_only = true, /* __cookie_v4_check is GPL */
7685 .ret_type = RET_INTEGER,
7686 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7687 .arg1_size = sizeof(struct iphdr),
7688 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7689 .arg2_size = sizeof(struct tcphdr),
7692 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv6, struct ipv6hdr *, iph,
7693 struct tcphdr *, th)
7695 #if IS_BUILTIN(CONFIG_IPV6)
7696 u32 cookie = ntohl(th->ack_seq) - 1;
7698 if (__cookie_v6_check(iph, th, cookie) > 0)
7703 return -EPROTONOSUPPORT;
7707 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv6_proto = {
7708 .func = bpf_tcp_raw_check_syncookie_ipv6,
7709 .gpl_only = true, /* __cookie_v6_check is GPL */
7711 .ret_type = RET_INTEGER,
7712 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7713 .arg1_size = sizeof(struct ipv6hdr),
7714 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7715 .arg2_size = sizeof(struct tcphdr),
7717 #endif /* CONFIG_SYN_COOKIES */
7719 #endif /* CONFIG_INET */
7721 bool bpf_helper_changes_pkt_data(void *func)
7723 if (func == bpf_skb_vlan_push ||
7724 func == bpf_skb_vlan_pop ||
7725 func == bpf_skb_store_bytes ||
7726 func == bpf_skb_change_proto ||
7727 func == bpf_skb_change_head ||
7728 func == sk_skb_change_head ||
7729 func == bpf_skb_change_tail ||
7730 func == sk_skb_change_tail ||
7731 func == bpf_skb_adjust_room ||
7732 func == sk_skb_adjust_room ||
7733 func == bpf_skb_pull_data ||
7734 func == sk_skb_pull_data ||
7735 func == bpf_clone_redirect ||
7736 func == bpf_l3_csum_replace ||
7737 func == bpf_l4_csum_replace ||
7738 func == bpf_xdp_adjust_head ||
7739 func == bpf_xdp_adjust_meta ||
7740 func == bpf_msg_pull_data ||
7741 func == bpf_msg_push_data ||
7742 func == bpf_msg_pop_data ||
7743 func == bpf_xdp_adjust_tail ||
7744 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7745 func == bpf_lwt_seg6_store_bytes ||
7746 func == bpf_lwt_seg6_adjust_srh ||
7747 func == bpf_lwt_seg6_action ||
7750 func == bpf_sock_ops_store_hdr_opt ||
7752 func == bpf_lwt_in_push_encap ||
7753 func == bpf_lwt_xmit_push_encap)
7759 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7760 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7762 static const struct bpf_func_proto *
7763 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7765 const struct bpf_func_proto *func_proto;
7767 func_proto = cgroup_common_func_proto(func_id, prog);
7771 func_proto = cgroup_current_func_proto(func_id, prog);
7776 case BPF_FUNC_get_socket_cookie:
7777 return &bpf_get_socket_cookie_sock_proto;
7778 case BPF_FUNC_get_netns_cookie:
7779 return &bpf_get_netns_cookie_sock_proto;
7780 case BPF_FUNC_perf_event_output:
7781 return &bpf_event_output_data_proto;
7782 case BPF_FUNC_sk_storage_get:
7783 return &bpf_sk_storage_get_cg_sock_proto;
7784 case BPF_FUNC_ktime_get_coarse_ns:
7785 return &bpf_ktime_get_coarse_ns_proto;
7787 return bpf_base_func_proto(func_id);
7791 static const struct bpf_func_proto *
7792 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7794 const struct bpf_func_proto *func_proto;
7796 func_proto = cgroup_common_func_proto(func_id, prog);
7800 func_proto = cgroup_current_func_proto(func_id, prog);
7806 switch (prog->expected_attach_type) {
7807 case BPF_CGROUP_INET4_CONNECT:
7808 case BPF_CGROUP_INET6_CONNECT:
7809 return &bpf_bind_proto;
7813 case BPF_FUNC_get_socket_cookie:
7814 return &bpf_get_socket_cookie_sock_addr_proto;
7815 case BPF_FUNC_get_netns_cookie:
7816 return &bpf_get_netns_cookie_sock_addr_proto;
7817 case BPF_FUNC_perf_event_output:
7818 return &bpf_event_output_data_proto;
7820 case BPF_FUNC_sk_lookup_tcp:
7821 return &bpf_sock_addr_sk_lookup_tcp_proto;
7822 case BPF_FUNC_sk_lookup_udp:
7823 return &bpf_sock_addr_sk_lookup_udp_proto;
7824 case BPF_FUNC_sk_release:
7825 return &bpf_sk_release_proto;
7826 case BPF_FUNC_skc_lookup_tcp:
7827 return &bpf_sock_addr_skc_lookup_tcp_proto;
7828 #endif /* CONFIG_INET */
7829 case BPF_FUNC_sk_storage_get:
7830 return &bpf_sk_storage_get_proto;
7831 case BPF_FUNC_sk_storage_delete:
7832 return &bpf_sk_storage_delete_proto;
7833 case BPF_FUNC_setsockopt:
7834 switch (prog->expected_attach_type) {
7835 case BPF_CGROUP_INET4_BIND:
7836 case BPF_CGROUP_INET6_BIND:
7837 case BPF_CGROUP_INET4_CONNECT:
7838 case BPF_CGROUP_INET6_CONNECT:
7839 case BPF_CGROUP_UDP4_RECVMSG:
7840 case BPF_CGROUP_UDP6_RECVMSG:
7841 case BPF_CGROUP_UDP4_SENDMSG:
7842 case BPF_CGROUP_UDP6_SENDMSG:
7843 case BPF_CGROUP_INET4_GETPEERNAME:
7844 case BPF_CGROUP_INET6_GETPEERNAME:
7845 case BPF_CGROUP_INET4_GETSOCKNAME:
7846 case BPF_CGROUP_INET6_GETSOCKNAME:
7847 return &bpf_sock_addr_setsockopt_proto;
7851 case BPF_FUNC_getsockopt:
7852 switch (prog->expected_attach_type) {
7853 case BPF_CGROUP_INET4_BIND:
7854 case BPF_CGROUP_INET6_BIND:
7855 case BPF_CGROUP_INET4_CONNECT:
7856 case BPF_CGROUP_INET6_CONNECT:
7857 case BPF_CGROUP_UDP4_RECVMSG:
7858 case BPF_CGROUP_UDP6_RECVMSG:
7859 case BPF_CGROUP_UDP4_SENDMSG:
7860 case BPF_CGROUP_UDP6_SENDMSG:
7861 case BPF_CGROUP_INET4_GETPEERNAME:
7862 case BPF_CGROUP_INET6_GETPEERNAME:
7863 case BPF_CGROUP_INET4_GETSOCKNAME:
7864 case BPF_CGROUP_INET6_GETSOCKNAME:
7865 return &bpf_sock_addr_getsockopt_proto;
7870 return bpf_sk_base_func_proto(func_id);
7874 static const struct bpf_func_proto *
7875 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7878 case BPF_FUNC_skb_load_bytes:
7879 return &bpf_skb_load_bytes_proto;
7880 case BPF_FUNC_skb_load_bytes_relative:
7881 return &bpf_skb_load_bytes_relative_proto;
7882 case BPF_FUNC_get_socket_cookie:
7883 return &bpf_get_socket_cookie_proto;
7884 case BPF_FUNC_get_socket_uid:
7885 return &bpf_get_socket_uid_proto;
7886 case BPF_FUNC_perf_event_output:
7887 return &bpf_skb_event_output_proto;
7889 return bpf_sk_base_func_proto(func_id);
7893 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7894 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7896 static const struct bpf_func_proto *
7897 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7899 const struct bpf_func_proto *func_proto;
7901 func_proto = cgroup_common_func_proto(func_id, prog);
7906 case BPF_FUNC_sk_fullsock:
7907 return &bpf_sk_fullsock_proto;
7908 case BPF_FUNC_sk_storage_get:
7909 return &bpf_sk_storage_get_proto;
7910 case BPF_FUNC_sk_storage_delete:
7911 return &bpf_sk_storage_delete_proto;
7912 case BPF_FUNC_perf_event_output:
7913 return &bpf_skb_event_output_proto;
7914 #ifdef CONFIG_SOCK_CGROUP_DATA
7915 case BPF_FUNC_skb_cgroup_id:
7916 return &bpf_skb_cgroup_id_proto;
7917 case BPF_FUNC_skb_ancestor_cgroup_id:
7918 return &bpf_skb_ancestor_cgroup_id_proto;
7919 case BPF_FUNC_sk_cgroup_id:
7920 return &bpf_sk_cgroup_id_proto;
7921 case BPF_FUNC_sk_ancestor_cgroup_id:
7922 return &bpf_sk_ancestor_cgroup_id_proto;
7925 case BPF_FUNC_sk_lookup_tcp:
7926 return &bpf_sk_lookup_tcp_proto;
7927 case BPF_FUNC_sk_lookup_udp:
7928 return &bpf_sk_lookup_udp_proto;
7929 case BPF_FUNC_sk_release:
7930 return &bpf_sk_release_proto;
7931 case BPF_FUNC_skc_lookup_tcp:
7932 return &bpf_skc_lookup_tcp_proto;
7933 case BPF_FUNC_tcp_sock:
7934 return &bpf_tcp_sock_proto;
7935 case BPF_FUNC_get_listener_sock:
7936 return &bpf_get_listener_sock_proto;
7937 case BPF_FUNC_skb_ecn_set_ce:
7938 return &bpf_skb_ecn_set_ce_proto;
7941 return sk_filter_func_proto(func_id, prog);
7945 static const struct bpf_func_proto *
7946 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7949 case BPF_FUNC_skb_store_bytes:
7950 return &bpf_skb_store_bytes_proto;
7951 case BPF_FUNC_skb_load_bytes:
7952 return &bpf_skb_load_bytes_proto;
7953 case BPF_FUNC_skb_load_bytes_relative:
7954 return &bpf_skb_load_bytes_relative_proto;
7955 case BPF_FUNC_skb_pull_data:
7956 return &bpf_skb_pull_data_proto;
7957 case BPF_FUNC_csum_diff:
7958 return &bpf_csum_diff_proto;
7959 case BPF_FUNC_csum_update:
7960 return &bpf_csum_update_proto;
7961 case BPF_FUNC_csum_level:
7962 return &bpf_csum_level_proto;
7963 case BPF_FUNC_l3_csum_replace:
7964 return &bpf_l3_csum_replace_proto;
7965 case BPF_FUNC_l4_csum_replace:
7966 return &bpf_l4_csum_replace_proto;
7967 case BPF_FUNC_clone_redirect:
7968 return &bpf_clone_redirect_proto;
7969 case BPF_FUNC_get_cgroup_classid:
7970 return &bpf_get_cgroup_classid_proto;
7971 case BPF_FUNC_skb_vlan_push:
7972 return &bpf_skb_vlan_push_proto;
7973 case BPF_FUNC_skb_vlan_pop:
7974 return &bpf_skb_vlan_pop_proto;
7975 case BPF_FUNC_skb_change_proto:
7976 return &bpf_skb_change_proto_proto;
7977 case BPF_FUNC_skb_change_type:
7978 return &bpf_skb_change_type_proto;
7979 case BPF_FUNC_skb_adjust_room:
7980 return &bpf_skb_adjust_room_proto;
7981 case BPF_FUNC_skb_change_tail:
7982 return &bpf_skb_change_tail_proto;
7983 case BPF_FUNC_skb_change_head:
7984 return &bpf_skb_change_head_proto;
7985 case BPF_FUNC_skb_get_tunnel_key:
7986 return &bpf_skb_get_tunnel_key_proto;
7987 case BPF_FUNC_skb_set_tunnel_key:
7988 return bpf_get_skb_set_tunnel_proto(func_id);
7989 case BPF_FUNC_skb_get_tunnel_opt:
7990 return &bpf_skb_get_tunnel_opt_proto;
7991 case BPF_FUNC_skb_set_tunnel_opt:
7992 return bpf_get_skb_set_tunnel_proto(func_id);
7993 case BPF_FUNC_redirect:
7994 return &bpf_redirect_proto;
7995 case BPF_FUNC_redirect_neigh:
7996 return &bpf_redirect_neigh_proto;
7997 case BPF_FUNC_redirect_peer:
7998 return &bpf_redirect_peer_proto;
7999 case BPF_FUNC_get_route_realm:
8000 return &bpf_get_route_realm_proto;
8001 case BPF_FUNC_get_hash_recalc:
8002 return &bpf_get_hash_recalc_proto;
8003 case BPF_FUNC_set_hash_invalid:
8004 return &bpf_set_hash_invalid_proto;
8005 case BPF_FUNC_set_hash:
8006 return &bpf_set_hash_proto;
8007 case BPF_FUNC_perf_event_output:
8008 return &bpf_skb_event_output_proto;
8009 case BPF_FUNC_get_smp_processor_id:
8010 return &bpf_get_smp_processor_id_proto;
8011 case BPF_FUNC_skb_under_cgroup:
8012 return &bpf_skb_under_cgroup_proto;
8013 case BPF_FUNC_get_socket_cookie:
8014 return &bpf_get_socket_cookie_proto;
8015 case BPF_FUNC_get_socket_uid:
8016 return &bpf_get_socket_uid_proto;
8017 case BPF_FUNC_fib_lookup:
8018 return &bpf_skb_fib_lookup_proto;
8019 case BPF_FUNC_check_mtu:
8020 return &bpf_skb_check_mtu_proto;
8021 case BPF_FUNC_sk_fullsock:
8022 return &bpf_sk_fullsock_proto;
8023 case BPF_FUNC_sk_storage_get:
8024 return &bpf_sk_storage_get_proto;
8025 case BPF_FUNC_sk_storage_delete:
8026 return &bpf_sk_storage_delete_proto;
8028 case BPF_FUNC_skb_get_xfrm_state:
8029 return &bpf_skb_get_xfrm_state_proto;
8031 #ifdef CONFIG_CGROUP_NET_CLASSID
8032 case BPF_FUNC_skb_cgroup_classid:
8033 return &bpf_skb_cgroup_classid_proto;
8035 #ifdef CONFIG_SOCK_CGROUP_DATA
8036 case BPF_FUNC_skb_cgroup_id:
8037 return &bpf_skb_cgroup_id_proto;
8038 case BPF_FUNC_skb_ancestor_cgroup_id:
8039 return &bpf_skb_ancestor_cgroup_id_proto;
8042 case BPF_FUNC_sk_lookup_tcp:
8043 return &bpf_tc_sk_lookup_tcp_proto;
8044 case BPF_FUNC_sk_lookup_udp:
8045 return &bpf_tc_sk_lookup_udp_proto;
8046 case BPF_FUNC_sk_release:
8047 return &bpf_sk_release_proto;
8048 case BPF_FUNC_tcp_sock:
8049 return &bpf_tcp_sock_proto;
8050 case BPF_FUNC_get_listener_sock:
8051 return &bpf_get_listener_sock_proto;
8052 case BPF_FUNC_skc_lookup_tcp:
8053 return &bpf_tc_skc_lookup_tcp_proto;
8054 case BPF_FUNC_tcp_check_syncookie:
8055 return &bpf_tcp_check_syncookie_proto;
8056 case BPF_FUNC_skb_ecn_set_ce:
8057 return &bpf_skb_ecn_set_ce_proto;
8058 case BPF_FUNC_tcp_gen_syncookie:
8059 return &bpf_tcp_gen_syncookie_proto;
8060 case BPF_FUNC_sk_assign:
8061 return &bpf_sk_assign_proto;
8062 case BPF_FUNC_skb_set_tstamp:
8063 return &bpf_skb_set_tstamp_proto;
8064 #ifdef CONFIG_SYN_COOKIES
8065 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8066 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8067 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8068 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8069 case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8070 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8071 case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8072 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8076 return bpf_sk_base_func_proto(func_id);
8080 static const struct bpf_func_proto *
8081 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8084 case BPF_FUNC_perf_event_output:
8085 return &bpf_xdp_event_output_proto;
8086 case BPF_FUNC_get_smp_processor_id:
8087 return &bpf_get_smp_processor_id_proto;
8088 case BPF_FUNC_csum_diff:
8089 return &bpf_csum_diff_proto;
8090 case BPF_FUNC_xdp_adjust_head:
8091 return &bpf_xdp_adjust_head_proto;
8092 case BPF_FUNC_xdp_adjust_meta:
8093 return &bpf_xdp_adjust_meta_proto;
8094 case BPF_FUNC_redirect:
8095 return &bpf_xdp_redirect_proto;
8096 case BPF_FUNC_redirect_map:
8097 return &bpf_xdp_redirect_map_proto;
8098 case BPF_FUNC_xdp_adjust_tail:
8099 return &bpf_xdp_adjust_tail_proto;
8100 case BPF_FUNC_xdp_get_buff_len:
8101 return &bpf_xdp_get_buff_len_proto;
8102 case BPF_FUNC_xdp_load_bytes:
8103 return &bpf_xdp_load_bytes_proto;
8104 case BPF_FUNC_xdp_store_bytes:
8105 return &bpf_xdp_store_bytes_proto;
8106 case BPF_FUNC_fib_lookup:
8107 return &bpf_xdp_fib_lookup_proto;
8108 case BPF_FUNC_check_mtu:
8109 return &bpf_xdp_check_mtu_proto;
8111 case BPF_FUNC_sk_lookup_udp:
8112 return &bpf_xdp_sk_lookup_udp_proto;
8113 case BPF_FUNC_sk_lookup_tcp:
8114 return &bpf_xdp_sk_lookup_tcp_proto;
8115 case BPF_FUNC_sk_release:
8116 return &bpf_sk_release_proto;
8117 case BPF_FUNC_skc_lookup_tcp:
8118 return &bpf_xdp_skc_lookup_tcp_proto;
8119 case BPF_FUNC_tcp_check_syncookie:
8120 return &bpf_tcp_check_syncookie_proto;
8121 case BPF_FUNC_tcp_gen_syncookie:
8122 return &bpf_tcp_gen_syncookie_proto;
8123 #ifdef CONFIG_SYN_COOKIES
8124 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8125 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8126 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8127 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8128 case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8129 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8130 case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8131 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8135 return bpf_sk_base_func_proto(func_id);
8138 #if IS_MODULE(CONFIG_NF_CONNTRACK) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES)
8139 /* The nf_conn___init type is used in the NF_CONNTRACK kfuncs. The
8140 * kfuncs are defined in two different modules, and we want to be able
8141 * to use them interchangably with the same BTF type ID. Because modules
8142 * can't de-duplicate BTF IDs between each other, we need the type to be
8143 * referenced in the vmlinux BTF or the verifier will get confused about
8144 * the different types. So we add this dummy type reference which will
8145 * be included in vmlinux BTF, allowing both modules to refer to the
8148 BTF_TYPE_EMIT(struct nf_conn___init);
8152 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
8153 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
8155 static const struct bpf_func_proto *
8156 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8158 const struct bpf_func_proto *func_proto;
8160 func_proto = cgroup_common_func_proto(func_id, prog);
8165 case BPF_FUNC_setsockopt:
8166 return &bpf_sock_ops_setsockopt_proto;
8167 case BPF_FUNC_getsockopt:
8168 return &bpf_sock_ops_getsockopt_proto;
8169 case BPF_FUNC_sock_ops_cb_flags_set:
8170 return &bpf_sock_ops_cb_flags_set_proto;
8171 case BPF_FUNC_sock_map_update:
8172 return &bpf_sock_map_update_proto;
8173 case BPF_FUNC_sock_hash_update:
8174 return &bpf_sock_hash_update_proto;
8175 case BPF_FUNC_get_socket_cookie:
8176 return &bpf_get_socket_cookie_sock_ops_proto;
8177 case BPF_FUNC_perf_event_output:
8178 return &bpf_event_output_data_proto;
8179 case BPF_FUNC_sk_storage_get:
8180 return &bpf_sk_storage_get_proto;
8181 case BPF_FUNC_sk_storage_delete:
8182 return &bpf_sk_storage_delete_proto;
8183 case BPF_FUNC_get_netns_cookie:
8184 return &bpf_get_netns_cookie_sock_ops_proto;
8186 case BPF_FUNC_load_hdr_opt:
8187 return &bpf_sock_ops_load_hdr_opt_proto;
8188 case BPF_FUNC_store_hdr_opt:
8189 return &bpf_sock_ops_store_hdr_opt_proto;
8190 case BPF_FUNC_reserve_hdr_opt:
8191 return &bpf_sock_ops_reserve_hdr_opt_proto;
8192 case BPF_FUNC_tcp_sock:
8193 return &bpf_tcp_sock_proto;
8194 #endif /* CONFIG_INET */
8196 return bpf_sk_base_func_proto(func_id);
8200 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
8201 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
8203 static const struct bpf_func_proto *
8204 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8207 case BPF_FUNC_msg_redirect_map:
8208 return &bpf_msg_redirect_map_proto;
8209 case BPF_FUNC_msg_redirect_hash:
8210 return &bpf_msg_redirect_hash_proto;
8211 case BPF_FUNC_msg_apply_bytes:
8212 return &bpf_msg_apply_bytes_proto;
8213 case BPF_FUNC_msg_cork_bytes:
8214 return &bpf_msg_cork_bytes_proto;
8215 case BPF_FUNC_msg_pull_data:
8216 return &bpf_msg_pull_data_proto;
8217 case BPF_FUNC_msg_push_data:
8218 return &bpf_msg_push_data_proto;
8219 case BPF_FUNC_msg_pop_data:
8220 return &bpf_msg_pop_data_proto;
8221 case BPF_FUNC_perf_event_output:
8222 return &bpf_event_output_data_proto;
8223 case BPF_FUNC_get_current_uid_gid:
8224 return &bpf_get_current_uid_gid_proto;
8225 case BPF_FUNC_get_current_pid_tgid:
8226 return &bpf_get_current_pid_tgid_proto;
8227 case BPF_FUNC_sk_storage_get:
8228 return &bpf_sk_storage_get_proto;
8229 case BPF_FUNC_sk_storage_delete:
8230 return &bpf_sk_storage_delete_proto;
8231 case BPF_FUNC_get_netns_cookie:
8232 return &bpf_get_netns_cookie_sk_msg_proto;
8233 #ifdef CONFIG_CGROUPS
8234 case BPF_FUNC_get_current_cgroup_id:
8235 return &bpf_get_current_cgroup_id_proto;
8236 case BPF_FUNC_get_current_ancestor_cgroup_id:
8237 return &bpf_get_current_ancestor_cgroup_id_proto;
8239 #ifdef CONFIG_CGROUP_NET_CLASSID
8240 case BPF_FUNC_get_cgroup_classid:
8241 return &bpf_get_cgroup_classid_curr_proto;
8244 return bpf_sk_base_func_proto(func_id);
8248 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
8249 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
8251 static const struct bpf_func_proto *
8252 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8255 case BPF_FUNC_skb_store_bytes:
8256 return &bpf_skb_store_bytes_proto;
8257 case BPF_FUNC_skb_load_bytes:
8258 return &bpf_skb_load_bytes_proto;
8259 case BPF_FUNC_skb_pull_data:
8260 return &sk_skb_pull_data_proto;
8261 case BPF_FUNC_skb_change_tail:
8262 return &sk_skb_change_tail_proto;
8263 case BPF_FUNC_skb_change_head:
8264 return &sk_skb_change_head_proto;
8265 case BPF_FUNC_skb_adjust_room:
8266 return &sk_skb_adjust_room_proto;
8267 case BPF_FUNC_get_socket_cookie:
8268 return &bpf_get_socket_cookie_proto;
8269 case BPF_FUNC_get_socket_uid:
8270 return &bpf_get_socket_uid_proto;
8271 case BPF_FUNC_sk_redirect_map:
8272 return &bpf_sk_redirect_map_proto;
8273 case BPF_FUNC_sk_redirect_hash:
8274 return &bpf_sk_redirect_hash_proto;
8275 case BPF_FUNC_perf_event_output:
8276 return &bpf_skb_event_output_proto;
8278 case BPF_FUNC_sk_lookup_tcp:
8279 return &bpf_sk_lookup_tcp_proto;
8280 case BPF_FUNC_sk_lookup_udp:
8281 return &bpf_sk_lookup_udp_proto;
8282 case BPF_FUNC_sk_release:
8283 return &bpf_sk_release_proto;
8284 case BPF_FUNC_skc_lookup_tcp:
8285 return &bpf_skc_lookup_tcp_proto;
8288 return bpf_sk_base_func_proto(func_id);
8292 static const struct bpf_func_proto *
8293 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8296 case BPF_FUNC_skb_load_bytes:
8297 return &bpf_flow_dissector_load_bytes_proto;
8299 return bpf_sk_base_func_proto(func_id);
8303 static const struct bpf_func_proto *
8304 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8307 case BPF_FUNC_skb_load_bytes:
8308 return &bpf_skb_load_bytes_proto;
8309 case BPF_FUNC_skb_pull_data:
8310 return &bpf_skb_pull_data_proto;
8311 case BPF_FUNC_csum_diff:
8312 return &bpf_csum_diff_proto;
8313 case BPF_FUNC_get_cgroup_classid:
8314 return &bpf_get_cgroup_classid_proto;
8315 case BPF_FUNC_get_route_realm:
8316 return &bpf_get_route_realm_proto;
8317 case BPF_FUNC_get_hash_recalc:
8318 return &bpf_get_hash_recalc_proto;
8319 case BPF_FUNC_perf_event_output:
8320 return &bpf_skb_event_output_proto;
8321 case BPF_FUNC_get_smp_processor_id:
8322 return &bpf_get_smp_processor_id_proto;
8323 case BPF_FUNC_skb_under_cgroup:
8324 return &bpf_skb_under_cgroup_proto;
8326 return bpf_sk_base_func_proto(func_id);
8330 static const struct bpf_func_proto *
8331 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8334 case BPF_FUNC_lwt_push_encap:
8335 return &bpf_lwt_in_push_encap_proto;
8337 return lwt_out_func_proto(func_id, prog);
8341 static const struct bpf_func_proto *
8342 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8345 case BPF_FUNC_skb_get_tunnel_key:
8346 return &bpf_skb_get_tunnel_key_proto;
8347 case BPF_FUNC_skb_set_tunnel_key:
8348 return bpf_get_skb_set_tunnel_proto(func_id);
8349 case BPF_FUNC_skb_get_tunnel_opt:
8350 return &bpf_skb_get_tunnel_opt_proto;
8351 case BPF_FUNC_skb_set_tunnel_opt:
8352 return bpf_get_skb_set_tunnel_proto(func_id);
8353 case BPF_FUNC_redirect:
8354 return &bpf_redirect_proto;
8355 case BPF_FUNC_clone_redirect:
8356 return &bpf_clone_redirect_proto;
8357 case BPF_FUNC_skb_change_tail:
8358 return &bpf_skb_change_tail_proto;
8359 case BPF_FUNC_skb_change_head:
8360 return &bpf_skb_change_head_proto;
8361 case BPF_FUNC_skb_store_bytes:
8362 return &bpf_skb_store_bytes_proto;
8363 case BPF_FUNC_csum_update:
8364 return &bpf_csum_update_proto;
8365 case BPF_FUNC_csum_level:
8366 return &bpf_csum_level_proto;
8367 case BPF_FUNC_l3_csum_replace:
8368 return &bpf_l3_csum_replace_proto;
8369 case BPF_FUNC_l4_csum_replace:
8370 return &bpf_l4_csum_replace_proto;
8371 case BPF_FUNC_set_hash_invalid:
8372 return &bpf_set_hash_invalid_proto;
8373 case BPF_FUNC_lwt_push_encap:
8374 return &bpf_lwt_xmit_push_encap_proto;
8376 return lwt_out_func_proto(func_id, prog);
8380 static const struct bpf_func_proto *
8381 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8384 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
8385 case BPF_FUNC_lwt_seg6_store_bytes:
8386 return &bpf_lwt_seg6_store_bytes_proto;
8387 case BPF_FUNC_lwt_seg6_action:
8388 return &bpf_lwt_seg6_action_proto;
8389 case BPF_FUNC_lwt_seg6_adjust_srh:
8390 return &bpf_lwt_seg6_adjust_srh_proto;
8393 return lwt_out_func_proto(func_id, prog);
8397 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
8398 const struct bpf_prog *prog,
8399 struct bpf_insn_access_aux *info)
8401 const int size_default = sizeof(__u32);
8403 if (off < 0 || off >= sizeof(struct __sk_buff))
8406 /* The verifier guarantees that size > 0. */
8407 if (off % size != 0)
8411 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8412 if (off + size > offsetofend(struct __sk_buff, cb[4]))
8415 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
8416 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
8417 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
8418 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
8419 case bpf_ctx_range(struct __sk_buff, data):
8420 case bpf_ctx_range(struct __sk_buff, data_meta):
8421 case bpf_ctx_range(struct __sk_buff, data_end):
8422 if (size != size_default)
8425 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8427 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8428 if (type == BPF_WRITE || size != sizeof(__u64))
8431 case bpf_ctx_range(struct __sk_buff, tstamp):
8432 if (size != sizeof(__u64))
8435 case offsetof(struct __sk_buff, sk):
8436 if (type == BPF_WRITE || size != sizeof(__u64))
8438 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
8440 case offsetof(struct __sk_buff, tstamp_type):
8442 case offsetofend(struct __sk_buff, tstamp_type) ... offsetof(struct __sk_buff, hwtstamp) - 1:
8443 /* Explicitly prohibit access to padding in __sk_buff. */
8446 /* Only narrow read access allowed for now. */
8447 if (type == BPF_WRITE) {
8448 if (size != size_default)
8451 bpf_ctx_record_field_size(info, size_default);
8452 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8460 static bool sk_filter_is_valid_access(int off, int size,
8461 enum bpf_access_type type,
8462 const struct bpf_prog *prog,
8463 struct bpf_insn_access_aux *info)
8466 case bpf_ctx_range(struct __sk_buff, tc_classid):
8467 case bpf_ctx_range(struct __sk_buff, data):
8468 case bpf_ctx_range(struct __sk_buff, data_meta):
8469 case bpf_ctx_range(struct __sk_buff, data_end):
8470 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8471 case bpf_ctx_range(struct __sk_buff, tstamp):
8472 case bpf_ctx_range(struct __sk_buff, wire_len):
8473 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8477 if (type == BPF_WRITE) {
8479 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8486 return bpf_skb_is_valid_access(off, size, type, prog, info);
8489 static bool cg_skb_is_valid_access(int off, int size,
8490 enum bpf_access_type type,
8491 const struct bpf_prog *prog,
8492 struct bpf_insn_access_aux *info)
8495 case bpf_ctx_range(struct __sk_buff, tc_classid):
8496 case bpf_ctx_range(struct __sk_buff, data_meta):
8497 case bpf_ctx_range(struct __sk_buff, wire_len):
8499 case bpf_ctx_range(struct __sk_buff, data):
8500 case bpf_ctx_range(struct __sk_buff, data_end):
8506 if (type == BPF_WRITE) {
8508 case bpf_ctx_range(struct __sk_buff, mark):
8509 case bpf_ctx_range(struct __sk_buff, priority):
8510 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8512 case bpf_ctx_range(struct __sk_buff, tstamp):
8522 case bpf_ctx_range(struct __sk_buff, data):
8523 info->reg_type = PTR_TO_PACKET;
8525 case bpf_ctx_range(struct __sk_buff, data_end):
8526 info->reg_type = PTR_TO_PACKET_END;
8530 return bpf_skb_is_valid_access(off, size, type, prog, info);
8533 static bool lwt_is_valid_access(int off, int size,
8534 enum bpf_access_type type,
8535 const struct bpf_prog *prog,
8536 struct bpf_insn_access_aux *info)
8539 case bpf_ctx_range(struct __sk_buff, tc_classid):
8540 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8541 case bpf_ctx_range(struct __sk_buff, data_meta):
8542 case bpf_ctx_range(struct __sk_buff, tstamp):
8543 case bpf_ctx_range(struct __sk_buff, wire_len):
8544 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8548 if (type == BPF_WRITE) {
8550 case bpf_ctx_range(struct __sk_buff, mark):
8551 case bpf_ctx_range(struct __sk_buff, priority):
8552 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8560 case bpf_ctx_range(struct __sk_buff, data):
8561 info->reg_type = PTR_TO_PACKET;
8563 case bpf_ctx_range(struct __sk_buff, data_end):
8564 info->reg_type = PTR_TO_PACKET_END;
8568 return bpf_skb_is_valid_access(off, size, type, prog, info);
8571 /* Attach type specific accesses */
8572 static bool __sock_filter_check_attach_type(int off,
8573 enum bpf_access_type access_type,
8574 enum bpf_attach_type attach_type)
8577 case offsetof(struct bpf_sock, bound_dev_if):
8578 case offsetof(struct bpf_sock, mark):
8579 case offsetof(struct bpf_sock, priority):
8580 switch (attach_type) {
8581 case BPF_CGROUP_INET_SOCK_CREATE:
8582 case BPF_CGROUP_INET_SOCK_RELEASE:
8587 case bpf_ctx_range(struct bpf_sock, src_ip4):
8588 switch (attach_type) {
8589 case BPF_CGROUP_INET4_POST_BIND:
8594 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8595 switch (attach_type) {
8596 case BPF_CGROUP_INET6_POST_BIND:
8601 case bpf_ctx_range(struct bpf_sock, src_port):
8602 switch (attach_type) {
8603 case BPF_CGROUP_INET4_POST_BIND:
8604 case BPF_CGROUP_INET6_POST_BIND:
8611 return access_type == BPF_READ;
8616 bool bpf_sock_common_is_valid_access(int off, int size,
8617 enum bpf_access_type type,
8618 struct bpf_insn_access_aux *info)
8621 case bpf_ctx_range_till(struct bpf_sock, type, priority):
8624 return bpf_sock_is_valid_access(off, size, type, info);
8628 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
8629 struct bpf_insn_access_aux *info)
8631 const int size_default = sizeof(__u32);
8634 if (off < 0 || off >= sizeof(struct bpf_sock))
8636 if (off % size != 0)
8640 case offsetof(struct bpf_sock, state):
8641 case offsetof(struct bpf_sock, family):
8642 case offsetof(struct bpf_sock, type):
8643 case offsetof(struct bpf_sock, protocol):
8644 case offsetof(struct bpf_sock, src_port):
8645 case offsetof(struct bpf_sock, rx_queue_mapping):
8646 case bpf_ctx_range(struct bpf_sock, src_ip4):
8647 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8648 case bpf_ctx_range(struct bpf_sock, dst_ip4):
8649 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8650 bpf_ctx_record_field_size(info, size_default);
8651 return bpf_ctx_narrow_access_ok(off, size, size_default);
8652 case bpf_ctx_range(struct bpf_sock, dst_port):
8653 field_size = size == size_default ?
8654 size_default : sizeof_field(struct bpf_sock, dst_port);
8655 bpf_ctx_record_field_size(info, field_size);
8656 return bpf_ctx_narrow_access_ok(off, size, field_size);
8657 case offsetofend(struct bpf_sock, dst_port) ...
8658 offsetof(struct bpf_sock, dst_ip4) - 1:
8662 return size == size_default;
8665 static bool sock_filter_is_valid_access(int off, int size,
8666 enum bpf_access_type type,
8667 const struct bpf_prog *prog,
8668 struct bpf_insn_access_aux *info)
8670 if (!bpf_sock_is_valid_access(off, size, type, info))
8672 return __sock_filter_check_attach_type(off, type,
8673 prog->expected_attach_type);
8676 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
8677 const struct bpf_prog *prog)
8679 /* Neither direct read nor direct write requires any preliminary
8685 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
8686 const struct bpf_prog *prog, int drop_verdict)
8688 struct bpf_insn *insn = insn_buf;
8693 /* if (!skb->cloned)
8696 * (Fast-path, otherwise approximation that we might be
8697 * a clone, do the rest in helper.)
8699 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET);
8700 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
8701 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
8703 /* ret = bpf_skb_pull_data(skb, 0); */
8704 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
8705 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
8706 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
8707 BPF_FUNC_skb_pull_data);
8710 * return TC_ACT_SHOT;
8712 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
8713 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
8714 *insn++ = BPF_EXIT_INSN();
8717 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
8719 *insn++ = prog->insnsi[0];
8721 return insn - insn_buf;
8724 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
8725 struct bpf_insn *insn_buf)
8727 bool indirect = BPF_MODE(orig->code) == BPF_IND;
8728 struct bpf_insn *insn = insn_buf;
8731 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
8733 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
8735 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
8737 /* We're guaranteed here that CTX is in R6. */
8738 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
8740 switch (BPF_SIZE(orig->code)) {
8742 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
8745 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
8748 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
8752 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
8753 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
8754 *insn++ = BPF_EXIT_INSN();
8756 return insn - insn_buf;
8759 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8760 const struct bpf_prog *prog)
8762 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8765 static bool tc_cls_act_is_valid_access(int off, int size,
8766 enum bpf_access_type type,
8767 const struct bpf_prog *prog,
8768 struct bpf_insn_access_aux *info)
8770 if (type == BPF_WRITE) {
8772 case bpf_ctx_range(struct __sk_buff, mark):
8773 case bpf_ctx_range(struct __sk_buff, tc_index):
8774 case bpf_ctx_range(struct __sk_buff, priority):
8775 case bpf_ctx_range(struct __sk_buff, tc_classid):
8776 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8777 case bpf_ctx_range(struct __sk_buff, tstamp):
8778 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8786 case bpf_ctx_range(struct __sk_buff, data):
8787 info->reg_type = PTR_TO_PACKET;
8789 case bpf_ctx_range(struct __sk_buff, data_meta):
8790 info->reg_type = PTR_TO_PACKET_META;
8792 case bpf_ctx_range(struct __sk_buff, data_end):
8793 info->reg_type = PTR_TO_PACKET_END;
8795 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8797 case offsetof(struct __sk_buff, tstamp_type):
8798 /* The convert_ctx_access() on reading and writing
8799 * __sk_buff->tstamp depends on whether the bpf prog
8800 * has used __sk_buff->tstamp_type or not.
8801 * Thus, we need to set prog->tstamp_type_access
8802 * earlier during is_valid_access() here.
8804 ((struct bpf_prog *)prog)->tstamp_type_access = 1;
8805 return size == sizeof(__u8);
8808 return bpf_skb_is_valid_access(off, size, type, prog, info);
8811 DEFINE_MUTEX(nf_conn_btf_access_lock);
8812 EXPORT_SYMBOL_GPL(nf_conn_btf_access_lock);
8814 int (*nfct_btf_struct_access)(struct bpf_verifier_log *log, const struct btf *btf,
8815 const struct btf_type *t, int off, int size,
8816 enum bpf_access_type atype, u32 *next_btf_id,
8817 enum bpf_type_flag *flag);
8818 EXPORT_SYMBOL_GPL(nfct_btf_struct_access);
8820 static int tc_cls_act_btf_struct_access(struct bpf_verifier_log *log,
8821 const struct btf *btf,
8822 const struct btf_type *t, int off,
8823 int size, enum bpf_access_type atype,
8825 enum bpf_type_flag *flag)
8829 if (atype == BPF_READ)
8830 return btf_struct_access(log, btf, t, off, size, atype, next_btf_id,
8833 mutex_lock(&nf_conn_btf_access_lock);
8834 if (nfct_btf_struct_access)
8835 ret = nfct_btf_struct_access(log, btf, t, off, size, atype, next_btf_id, flag);
8836 mutex_unlock(&nf_conn_btf_access_lock);
8841 static bool __is_valid_xdp_access(int off, int size)
8843 if (off < 0 || off >= sizeof(struct xdp_md))
8845 if (off % size != 0)
8847 if (size != sizeof(__u32))
8853 static bool xdp_is_valid_access(int off, int size,
8854 enum bpf_access_type type,
8855 const struct bpf_prog *prog,
8856 struct bpf_insn_access_aux *info)
8858 if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8860 case offsetof(struct xdp_md, egress_ifindex):
8865 if (type == BPF_WRITE) {
8866 if (bpf_prog_is_dev_bound(prog->aux)) {
8868 case offsetof(struct xdp_md, rx_queue_index):
8869 return __is_valid_xdp_access(off, size);
8876 case offsetof(struct xdp_md, data):
8877 info->reg_type = PTR_TO_PACKET;
8879 case offsetof(struct xdp_md, data_meta):
8880 info->reg_type = PTR_TO_PACKET_META;
8882 case offsetof(struct xdp_md, data_end):
8883 info->reg_type = PTR_TO_PACKET_END;
8887 return __is_valid_xdp_access(off, size);
8890 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act)
8892 const u32 act_max = XDP_REDIRECT;
8894 pr_warn_once("%s XDP return value %u on prog %s (id %d) dev %s, expect packet loss!\n",
8895 act > act_max ? "Illegal" : "Driver unsupported",
8896 act, prog->aux->name, prog->aux->id, dev ? dev->name : "N/A");
8898 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
8900 static int xdp_btf_struct_access(struct bpf_verifier_log *log,
8901 const struct btf *btf,
8902 const struct btf_type *t, int off,
8903 int size, enum bpf_access_type atype,
8905 enum bpf_type_flag *flag)
8909 if (atype == BPF_READ)
8910 return btf_struct_access(log, btf, t, off, size, atype, next_btf_id,
8913 mutex_lock(&nf_conn_btf_access_lock);
8914 if (nfct_btf_struct_access)
8915 ret = nfct_btf_struct_access(log, btf, t, off, size, atype, next_btf_id, flag);
8916 mutex_unlock(&nf_conn_btf_access_lock);
8921 static bool sock_addr_is_valid_access(int off, int size,
8922 enum bpf_access_type type,
8923 const struct bpf_prog *prog,
8924 struct bpf_insn_access_aux *info)
8926 const int size_default = sizeof(__u32);
8928 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
8930 if (off % size != 0)
8933 /* Disallow access to IPv6 fields from IPv4 contex and vise
8937 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8938 switch (prog->expected_attach_type) {
8939 case BPF_CGROUP_INET4_BIND:
8940 case BPF_CGROUP_INET4_CONNECT:
8941 case BPF_CGROUP_INET4_GETPEERNAME:
8942 case BPF_CGROUP_INET4_GETSOCKNAME:
8943 case BPF_CGROUP_UDP4_SENDMSG:
8944 case BPF_CGROUP_UDP4_RECVMSG:
8950 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8951 switch (prog->expected_attach_type) {
8952 case BPF_CGROUP_INET6_BIND:
8953 case BPF_CGROUP_INET6_CONNECT:
8954 case BPF_CGROUP_INET6_GETPEERNAME:
8955 case BPF_CGROUP_INET6_GETSOCKNAME:
8956 case BPF_CGROUP_UDP6_SENDMSG:
8957 case BPF_CGROUP_UDP6_RECVMSG:
8963 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8964 switch (prog->expected_attach_type) {
8965 case BPF_CGROUP_UDP4_SENDMSG:
8971 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8973 switch (prog->expected_attach_type) {
8974 case BPF_CGROUP_UDP6_SENDMSG:
8983 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8984 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8985 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8986 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8988 case bpf_ctx_range(struct bpf_sock_addr, user_port):
8989 if (type == BPF_READ) {
8990 bpf_ctx_record_field_size(info, size_default);
8992 if (bpf_ctx_wide_access_ok(off, size,
8993 struct bpf_sock_addr,
8997 if (bpf_ctx_wide_access_ok(off, size,
8998 struct bpf_sock_addr,
9002 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
9005 if (bpf_ctx_wide_access_ok(off, size,
9006 struct bpf_sock_addr,
9010 if (bpf_ctx_wide_access_ok(off, size,
9011 struct bpf_sock_addr,
9015 if (size != size_default)
9019 case offsetof(struct bpf_sock_addr, sk):
9020 if (type != BPF_READ)
9022 if (size != sizeof(__u64))
9024 info->reg_type = PTR_TO_SOCKET;
9027 if (type == BPF_READ) {
9028 if (size != size_default)
9038 static bool sock_ops_is_valid_access(int off, int size,
9039 enum bpf_access_type type,
9040 const struct bpf_prog *prog,
9041 struct bpf_insn_access_aux *info)
9043 const int size_default = sizeof(__u32);
9045 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
9048 /* The verifier guarantees that size > 0. */
9049 if (off % size != 0)
9052 if (type == BPF_WRITE) {
9054 case offsetof(struct bpf_sock_ops, reply):
9055 case offsetof(struct bpf_sock_ops, sk_txhash):
9056 if (size != size_default)
9064 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
9066 if (size != sizeof(__u64))
9069 case offsetof(struct bpf_sock_ops, sk):
9070 if (size != sizeof(__u64))
9072 info->reg_type = PTR_TO_SOCKET_OR_NULL;
9074 case offsetof(struct bpf_sock_ops, skb_data):
9075 if (size != sizeof(__u64))
9077 info->reg_type = PTR_TO_PACKET;
9079 case offsetof(struct bpf_sock_ops, skb_data_end):
9080 if (size != sizeof(__u64))
9082 info->reg_type = PTR_TO_PACKET_END;
9084 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9085 bpf_ctx_record_field_size(info, size_default);
9086 return bpf_ctx_narrow_access_ok(off, size,
9089 if (size != size_default)
9098 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
9099 const struct bpf_prog *prog)
9101 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
9104 static bool sk_skb_is_valid_access(int off, int size,
9105 enum bpf_access_type type,
9106 const struct bpf_prog *prog,
9107 struct bpf_insn_access_aux *info)
9110 case bpf_ctx_range(struct __sk_buff, tc_classid):
9111 case bpf_ctx_range(struct __sk_buff, data_meta):
9112 case bpf_ctx_range(struct __sk_buff, tstamp):
9113 case bpf_ctx_range(struct __sk_buff, wire_len):
9114 case bpf_ctx_range(struct __sk_buff, hwtstamp):
9118 if (type == BPF_WRITE) {
9120 case bpf_ctx_range(struct __sk_buff, tc_index):
9121 case bpf_ctx_range(struct __sk_buff, priority):
9129 case bpf_ctx_range(struct __sk_buff, mark):
9131 case bpf_ctx_range(struct __sk_buff, data):
9132 info->reg_type = PTR_TO_PACKET;
9134 case bpf_ctx_range(struct __sk_buff, data_end):
9135 info->reg_type = PTR_TO_PACKET_END;
9139 return bpf_skb_is_valid_access(off, size, type, prog, info);
9142 static bool sk_msg_is_valid_access(int off, int size,
9143 enum bpf_access_type type,
9144 const struct bpf_prog *prog,
9145 struct bpf_insn_access_aux *info)
9147 if (type == BPF_WRITE)
9150 if (off % size != 0)
9154 case offsetof(struct sk_msg_md, data):
9155 info->reg_type = PTR_TO_PACKET;
9156 if (size != sizeof(__u64))
9159 case offsetof(struct sk_msg_md, data_end):
9160 info->reg_type = PTR_TO_PACKET_END;
9161 if (size != sizeof(__u64))
9164 case offsetof(struct sk_msg_md, sk):
9165 if (size != sizeof(__u64))
9167 info->reg_type = PTR_TO_SOCKET;
9169 case bpf_ctx_range(struct sk_msg_md, family):
9170 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
9171 case bpf_ctx_range(struct sk_msg_md, local_ip4):
9172 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
9173 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
9174 case bpf_ctx_range(struct sk_msg_md, remote_port):
9175 case bpf_ctx_range(struct sk_msg_md, local_port):
9176 case bpf_ctx_range(struct sk_msg_md, size):
9177 if (size != sizeof(__u32))
9186 static bool flow_dissector_is_valid_access(int off, int size,
9187 enum bpf_access_type type,
9188 const struct bpf_prog *prog,
9189 struct bpf_insn_access_aux *info)
9191 const int size_default = sizeof(__u32);
9193 if (off < 0 || off >= sizeof(struct __sk_buff))
9196 if (type == BPF_WRITE)
9200 case bpf_ctx_range(struct __sk_buff, data):
9201 if (size != size_default)
9203 info->reg_type = PTR_TO_PACKET;
9205 case bpf_ctx_range(struct __sk_buff, data_end):
9206 if (size != size_default)
9208 info->reg_type = PTR_TO_PACKET_END;
9210 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
9211 if (size != sizeof(__u64))
9213 info->reg_type = PTR_TO_FLOW_KEYS;
9220 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
9221 const struct bpf_insn *si,
9222 struct bpf_insn *insn_buf,
9223 struct bpf_prog *prog,
9227 struct bpf_insn *insn = insn_buf;
9230 case offsetof(struct __sk_buff, data):
9231 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
9232 si->dst_reg, si->src_reg,
9233 offsetof(struct bpf_flow_dissector, data));
9236 case offsetof(struct __sk_buff, data_end):
9237 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
9238 si->dst_reg, si->src_reg,
9239 offsetof(struct bpf_flow_dissector, data_end));
9242 case offsetof(struct __sk_buff, flow_keys):
9243 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
9244 si->dst_reg, si->src_reg,
9245 offsetof(struct bpf_flow_dissector, flow_keys));
9249 return insn - insn_buf;
9252 static struct bpf_insn *bpf_convert_tstamp_type_read(const struct bpf_insn *si,
9253 struct bpf_insn *insn)
9255 __u8 value_reg = si->dst_reg;
9256 __u8 skb_reg = si->src_reg;
9257 /* AX is needed because src_reg and dst_reg could be the same */
9258 __u8 tmp_reg = BPF_REG_AX;
9260 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg,
9261 PKT_VLAN_PRESENT_OFFSET);
9262 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg,
9263 SKB_MONO_DELIVERY_TIME_MASK, 2);
9264 *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_UNSPEC);
9265 *insn++ = BPF_JMP_A(1);
9266 *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_DELIVERY_MONO);
9271 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si,
9272 struct bpf_insn *insn)
9274 /* si->dst_reg = skb_shinfo(SKB); */
9275 #ifdef NET_SKBUFF_DATA_USES_OFFSET
9276 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9277 BPF_REG_AX, si->src_reg,
9278 offsetof(struct sk_buff, end));
9279 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
9280 si->dst_reg, si->src_reg,
9281 offsetof(struct sk_buff, head));
9282 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
9284 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9285 si->dst_reg, si->src_reg,
9286 offsetof(struct sk_buff, end));
9292 static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog,
9293 const struct bpf_insn *si,
9294 struct bpf_insn *insn)
9296 __u8 value_reg = si->dst_reg;
9297 __u8 skb_reg = si->src_reg;
9299 #ifdef CONFIG_NET_CLS_ACT
9300 /* If the tstamp_type is read,
9301 * the bpf prog is aware the tstamp could have delivery time.
9302 * Thus, read skb->tstamp as is if tstamp_type_access is true.
9304 if (!prog->tstamp_type_access) {
9305 /* AX is needed because src_reg and dst_reg could be the same */
9306 __u8 tmp_reg = BPF_REG_AX;
9308 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, PKT_VLAN_PRESENT_OFFSET);
9309 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg,
9310 TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK);
9311 *insn++ = BPF_JMP32_IMM(BPF_JNE, tmp_reg,
9312 TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK, 2);
9313 /* skb->tc_at_ingress && skb->mono_delivery_time,
9314 * read 0 as the (rcv) timestamp.
9316 *insn++ = BPF_MOV64_IMM(value_reg, 0);
9317 *insn++ = BPF_JMP_A(1);
9321 *insn++ = BPF_LDX_MEM(BPF_DW, value_reg, skb_reg,
9322 offsetof(struct sk_buff, tstamp));
9326 static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog,
9327 const struct bpf_insn *si,
9328 struct bpf_insn *insn)
9330 __u8 value_reg = si->src_reg;
9331 __u8 skb_reg = si->dst_reg;
9333 #ifdef CONFIG_NET_CLS_ACT
9334 /* If the tstamp_type is read,
9335 * the bpf prog is aware the tstamp could have delivery time.
9336 * Thus, write skb->tstamp as is if tstamp_type_access is true.
9337 * Otherwise, writing at ingress will have to clear the
9338 * mono_delivery_time bit also.
9340 if (!prog->tstamp_type_access) {
9341 __u8 tmp_reg = BPF_REG_AX;
9343 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, PKT_VLAN_PRESENT_OFFSET);
9344 /* Writing __sk_buff->tstamp as ingress, goto <clear> */
9345 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1);
9347 *insn++ = BPF_JMP_A(2);
9348 /* <clear>: mono_delivery_time */
9349 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_MONO_DELIVERY_TIME_MASK);
9350 *insn++ = BPF_STX_MEM(BPF_B, skb_reg, tmp_reg, PKT_VLAN_PRESENT_OFFSET);
9354 /* <store>: skb->tstamp = tstamp */
9355 *insn++ = BPF_STX_MEM(BPF_DW, skb_reg, value_reg,
9356 offsetof(struct sk_buff, tstamp));
9360 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
9361 const struct bpf_insn *si,
9362 struct bpf_insn *insn_buf,
9363 struct bpf_prog *prog, u32 *target_size)
9365 struct bpf_insn *insn = insn_buf;
9369 case offsetof(struct __sk_buff, len):
9370 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9371 bpf_target_off(struct sk_buff, len, 4,
9375 case offsetof(struct __sk_buff, protocol):
9376 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9377 bpf_target_off(struct sk_buff, protocol, 2,
9381 case offsetof(struct __sk_buff, vlan_proto):
9382 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9383 bpf_target_off(struct sk_buff, vlan_proto, 2,
9387 case offsetof(struct __sk_buff, priority):
9388 if (type == BPF_WRITE)
9389 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9390 bpf_target_off(struct sk_buff, priority, 4,
9393 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9394 bpf_target_off(struct sk_buff, priority, 4,
9398 case offsetof(struct __sk_buff, ingress_ifindex):
9399 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9400 bpf_target_off(struct sk_buff, skb_iif, 4,
9404 case offsetof(struct __sk_buff, ifindex):
9405 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9406 si->dst_reg, si->src_reg,
9407 offsetof(struct sk_buff, dev));
9408 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9409 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9410 bpf_target_off(struct net_device, ifindex, 4,
9414 case offsetof(struct __sk_buff, hash):
9415 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9416 bpf_target_off(struct sk_buff, hash, 4,
9420 case offsetof(struct __sk_buff, mark):
9421 if (type == BPF_WRITE)
9422 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9423 bpf_target_off(struct sk_buff, mark, 4,
9426 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9427 bpf_target_off(struct sk_buff, mark, 4,
9431 case offsetof(struct __sk_buff, pkt_type):
9433 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
9435 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
9436 #ifdef __BIG_ENDIAN_BITFIELD
9437 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
9441 case offsetof(struct __sk_buff, queue_mapping):
9442 if (type == BPF_WRITE) {
9443 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
9444 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
9445 bpf_target_off(struct sk_buff,
9449 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9450 bpf_target_off(struct sk_buff,
9456 case offsetof(struct __sk_buff, vlan_present):
9458 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
9459 PKT_VLAN_PRESENT_OFFSET);
9460 if (PKT_VLAN_PRESENT_BIT)
9461 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
9462 if (PKT_VLAN_PRESENT_BIT < 7)
9463 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
9466 case offsetof(struct __sk_buff, vlan_tci):
9467 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9468 bpf_target_off(struct sk_buff, vlan_tci, 2,
9472 case offsetof(struct __sk_buff, cb[0]) ...
9473 offsetofend(struct __sk_buff, cb[4]) - 1:
9474 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
9475 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
9476 offsetof(struct qdisc_skb_cb, data)) %
9479 prog->cb_access = 1;
9481 off -= offsetof(struct __sk_buff, cb[0]);
9482 off += offsetof(struct sk_buff, cb);
9483 off += offsetof(struct qdisc_skb_cb, data);
9484 if (type == BPF_WRITE)
9485 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
9488 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
9492 case offsetof(struct __sk_buff, tc_classid):
9493 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
9496 off -= offsetof(struct __sk_buff, tc_classid);
9497 off += offsetof(struct sk_buff, cb);
9498 off += offsetof(struct qdisc_skb_cb, tc_classid);
9500 if (type == BPF_WRITE)
9501 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
9504 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
9508 case offsetof(struct __sk_buff, data):
9509 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9510 si->dst_reg, si->src_reg,
9511 offsetof(struct sk_buff, data));
9514 case offsetof(struct __sk_buff, data_meta):
9516 off -= offsetof(struct __sk_buff, data_meta);
9517 off += offsetof(struct sk_buff, cb);
9518 off += offsetof(struct bpf_skb_data_end, data_meta);
9519 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9523 case offsetof(struct __sk_buff, data_end):
9525 off -= offsetof(struct __sk_buff, data_end);
9526 off += offsetof(struct sk_buff, cb);
9527 off += offsetof(struct bpf_skb_data_end, data_end);
9528 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9532 case offsetof(struct __sk_buff, tc_index):
9533 #ifdef CONFIG_NET_SCHED
9534 if (type == BPF_WRITE)
9535 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
9536 bpf_target_off(struct sk_buff, tc_index, 2,
9539 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9540 bpf_target_off(struct sk_buff, tc_index, 2,
9544 if (type == BPF_WRITE)
9545 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
9547 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9551 case offsetof(struct __sk_buff, napi_id):
9552 #if defined(CONFIG_NET_RX_BUSY_POLL)
9553 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9554 bpf_target_off(struct sk_buff, napi_id, 4,
9556 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
9557 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9560 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9563 case offsetof(struct __sk_buff, family):
9564 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9566 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9567 si->dst_reg, si->src_reg,
9568 offsetof(struct sk_buff, sk));
9569 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9570 bpf_target_off(struct sock_common,
9574 case offsetof(struct __sk_buff, remote_ip4):
9575 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9577 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9578 si->dst_reg, si->src_reg,
9579 offsetof(struct sk_buff, sk));
9580 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9581 bpf_target_off(struct sock_common,
9585 case offsetof(struct __sk_buff, local_ip4):
9586 BUILD_BUG_ON(sizeof_field(struct sock_common,
9587 skc_rcv_saddr) != 4);
9589 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9590 si->dst_reg, si->src_reg,
9591 offsetof(struct sk_buff, sk));
9592 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9593 bpf_target_off(struct sock_common,
9597 case offsetof(struct __sk_buff, remote_ip6[0]) ...
9598 offsetof(struct __sk_buff, remote_ip6[3]):
9599 #if IS_ENABLED(CONFIG_IPV6)
9600 BUILD_BUG_ON(sizeof_field(struct sock_common,
9601 skc_v6_daddr.s6_addr32[0]) != 4);
9604 off -= offsetof(struct __sk_buff, remote_ip6[0]);
9606 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9607 si->dst_reg, si->src_reg,
9608 offsetof(struct sk_buff, sk));
9609 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9610 offsetof(struct sock_common,
9611 skc_v6_daddr.s6_addr32[0]) +
9614 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9617 case offsetof(struct __sk_buff, local_ip6[0]) ...
9618 offsetof(struct __sk_buff, local_ip6[3]):
9619 #if IS_ENABLED(CONFIG_IPV6)
9620 BUILD_BUG_ON(sizeof_field(struct sock_common,
9621 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9624 off -= offsetof(struct __sk_buff, local_ip6[0]);
9626 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9627 si->dst_reg, si->src_reg,
9628 offsetof(struct sk_buff, sk));
9629 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9630 offsetof(struct sock_common,
9631 skc_v6_rcv_saddr.s6_addr32[0]) +
9634 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9638 case offsetof(struct __sk_buff, remote_port):
9639 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9641 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9642 si->dst_reg, si->src_reg,
9643 offsetof(struct sk_buff, sk));
9644 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9645 bpf_target_off(struct sock_common,
9648 #ifndef __BIG_ENDIAN_BITFIELD
9649 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9653 case offsetof(struct __sk_buff, local_port):
9654 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9656 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9657 si->dst_reg, si->src_reg,
9658 offsetof(struct sk_buff, sk));
9659 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9660 bpf_target_off(struct sock_common,
9661 skc_num, 2, target_size));
9664 case offsetof(struct __sk_buff, tstamp):
9665 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
9667 if (type == BPF_WRITE)
9668 insn = bpf_convert_tstamp_write(prog, si, insn);
9670 insn = bpf_convert_tstamp_read(prog, si, insn);
9673 case offsetof(struct __sk_buff, tstamp_type):
9674 insn = bpf_convert_tstamp_type_read(si, insn);
9677 case offsetof(struct __sk_buff, gso_segs):
9678 insn = bpf_convert_shinfo_access(si, insn);
9679 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
9680 si->dst_reg, si->dst_reg,
9681 bpf_target_off(struct skb_shared_info,
9685 case offsetof(struct __sk_buff, gso_size):
9686 insn = bpf_convert_shinfo_access(si, insn);
9687 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
9688 si->dst_reg, si->dst_reg,
9689 bpf_target_off(struct skb_shared_info,
9693 case offsetof(struct __sk_buff, wire_len):
9694 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
9697 off -= offsetof(struct __sk_buff, wire_len);
9698 off += offsetof(struct sk_buff, cb);
9699 off += offsetof(struct qdisc_skb_cb, pkt_len);
9701 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
9704 case offsetof(struct __sk_buff, sk):
9705 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9706 si->dst_reg, si->src_reg,
9707 offsetof(struct sk_buff, sk));
9709 case offsetof(struct __sk_buff, hwtstamp):
9710 BUILD_BUG_ON(sizeof_field(struct skb_shared_hwtstamps, hwtstamp) != 8);
9711 BUILD_BUG_ON(offsetof(struct skb_shared_hwtstamps, hwtstamp) != 0);
9713 insn = bpf_convert_shinfo_access(si, insn);
9714 *insn++ = BPF_LDX_MEM(BPF_DW,
9715 si->dst_reg, si->dst_reg,
9716 bpf_target_off(struct skb_shared_info,
9722 return insn - insn_buf;
9725 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
9726 const struct bpf_insn *si,
9727 struct bpf_insn *insn_buf,
9728 struct bpf_prog *prog, u32 *target_size)
9730 struct bpf_insn *insn = insn_buf;
9734 case offsetof(struct bpf_sock, bound_dev_if):
9735 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
9737 if (type == BPF_WRITE)
9738 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9739 offsetof(struct sock, sk_bound_dev_if));
9741 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9742 offsetof(struct sock, sk_bound_dev_if));
9745 case offsetof(struct bpf_sock, mark):
9746 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
9748 if (type == BPF_WRITE)
9749 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9750 offsetof(struct sock, sk_mark));
9752 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9753 offsetof(struct sock, sk_mark));
9756 case offsetof(struct bpf_sock, priority):
9757 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
9759 if (type == BPF_WRITE)
9760 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9761 offsetof(struct sock, sk_priority));
9763 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9764 offsetof(struct sock, sk_priority));
9767 case offsetof(struct bpf_sock, family):
9768 *insn++ = BPF_LDX_MEM(
9769 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
9770 si->dst_reg, si->src_reg,
9771 bpf_target_off(struct sock_common,
9773 sizeof_field(struct sock_common,
9778 case offsetof(struct bpf_sock, type):
9779 *insn++ = BPF_LDX_MEM(
9780 BPF_FIELD_SIZEOF(struct sock, sk_type),
9781 si->dst_reg, si->src_reg,
9782 bpf_target_off(struct sock, sk_type,
9783 sizeof_field(struct sock, sk_type),
9787 case offsetof(struct bpf_sock, protocol):
9788 *insn++ = BPF_LDX_MEM(
9789 BPF_FIELD_SIZEOF(struct sock, sk_protocol),
9790 si->dst_reg, si->src_reg,
9791 bpf_target_off(struct sock, sk_protocol,
9792 sizeof_field(struct sock, sk_protocol),
9796 case offsetof(struct bpf_sock, src_ip4):
9797 *insn++ = BPF_LDX_MEM(
9798 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9799 bpf_target_off(struct sock_common, skc_rcv_saddr,
9800 sizeof_field(struct sock_common,
9805 case offsetof(struct bpf_sock, dst_ip4):
9806 *insn++ = BPF_LDX_MEM(
9807 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9808 bpf_target_off(struct sock_common, skc_daddr,
9809 sizeof_field(struct sock_common,
9814 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
9815 #if IS_ENABLED(CONFIG_IPV6)
9817 off -= offsetof(struct bpf_sock, src_ip6[0]);
9818 *insn++ = BPF_LDX_MEM(
9819 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9822 skc_v6_rcv_saddr.s6_addr32[0],
9823 sizeof_field(struct sock_common,
9824 skc_v6_rcv_saddr.s6_addr32[0]),
9825 target_size) + off);
9828 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9832 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
9833 #if IS_ENABLED(CONFIG_IPV6)
9835 off -= offsetof(struct bpf_sock, dst_ip6[0]);
9836 *insn++ = BPF_LDX_MEM(
9837 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9838 bpf_target_off(struct sock_common,
9839 skc_v6_daddr.s6_addr32[0],
9840 sizeof_field(struct sock_common,
9841 skc_v6_daddr.s6_addr32[0]),
9842 target_size) + off);
9844 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9849 case offsetof(struct bpf_sock, src_port):
9850 *insn++ = BPF_LDX_MEM(
9851 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
9852 si->dst_reg, si->src_reg,
9853 bpf_target_off(struct sock_common, skc_num,
9854 sizeof_field(struct sock_common,
9859 case offsetof(struct bpf_sock, dst_port):
9860 *insn++ = BPF_LDX_MEM(
9861 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
9862 si->dst_reg, si->src_reg,
9863 bpf_target_off(struct sock_common, skc_dport,
9864 sizeof_field(struct sock_common,
9869 case offsetof(struct bpf_sock, state):
9870 *insn++ = BPF_LDX_MEM(
9871 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
9872 si->dst_reg, si->src_reg,
9873 bpf_target_off(struct sock_common, skc_state,
9874 sizeof_field(struct sock_common,
9878 case offsetof(struct bpf_sock, rx_queue_mapping):
9879 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
9880 *insn++ = BPF_LDX_MEM(
9881 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
9882 si->dst_reg, si->src_reg,
9883 bpf_target_off(struct sock, sk_rx_queue_mapping,
9884 sizeof_field(struct sock,
9885 sk_rx_queue_mapping),
9887 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
9889 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9891 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9897 return insn - insn_buf;
9900 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
9901 const struct bpf_insn *si,
9902 struct bpf_insn *insn_buf,
9903 struct bpf_prog *prog, u32 *target_size)
9905 struct bpf_insn *insn = insn_buf;
9908 case offsetof(struct __sk_buff, ifindex):
9909 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9910 si->dst_reg, si->src_reg,
9911 offsetof(struct sk_buff, dev));
9912 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9913 bpf_target_off(struct net_device, ifindex, 4,
9917 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9921 return insn - insn_buf;
9924 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
9925 const struct bpf_insn *si,
9926 struct bpf_insn *insn_buf,
9927 struct bpf_prog *prog, u32 *target_size)
9929 struct bpf_insn *insn = insn_buf;
9932 case offsetof(struct xdp_md, data):
9933 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
9934 si->dst_reg, si->src_reg,
9935 offsetof(struct xdp_buff, data));
9937 case offsetof(struct xdp_md, data_meta):
9938 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
9939 si->dst_reg, si->src_reg,
9940 offsetof(struct xdp_buff, data_meta));
9942 case offsetof(struct xdp_md, data_end):
9943 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
9944 si->dst_reg, si->src_reg,
9945 offsetof(struct xdp_buff, data_end));
9947 case offsetof(struct xdp_md, ingress_ifindex):
9948 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9949 si->dst_reg, si->src_reg,
9950 offsetof(struct xdp_buff, rxq));
9951 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
9952 si->dst_reg, si->dst_reg,
9953 offsetof(struct xdp_rxq_info, dev));
9954 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9955 offsetof(struct net_device, ifindex));
9957 case offsetof(struct xdp_md, rx_queue_index):
9958 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9959 si->dst_reg, si->src_reg,
9960 offsetof(struct xdp_buff, rxq));
9961 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9962 offsetof(struct xdp_rxq_info,
9965 case offsetof(struct xdp_md, egress_ifindex):
9966 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
9967 si->dst_reg, si->src_reg,
9968 offsetof(struct xdp_buff, txq));
9969 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
9970 si->dst_reg, si->dst_reg,
9971 offsetof(struct xdp_txq_info, dev));
9972 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9973 offsetof(struct net_device, ifindex));
9977 return insn - insn_buf;
9980 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
9981 * context Structure, F is Field in context structure that contains a pointer
9982 * to Nested Structure of type NS that has the field NF.
9984 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
9985 * sure that SIZE is not greater than actual size of S.F.NF.
9987 * If offset OFF is provided, the load happens from that offset relative to
9990 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
9992 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
9993 si->src_reg, offsetof(S, F)); \
9994 *insn++ = BPF_LDX_MEM( \
9995 SIZE, si->dst_reg, si->dst_reg, \
9996 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
10001 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
10002 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
10003 BPF_FIELD_SIZEOF(NS, NF), 0)
10005 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
10006 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
10008 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
10009 * "register" since two registers available in convert_ctx_access are not
10010 * enough: we can't override neither SRC, since it contains value to store, nor
10011 * DST since it contains pointer to context that may be used by later
10012 * instructions. But we need a temporary place to save pointer to nested
10013 * structure whose field we want to store to.
10015 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
10017 int tmp_reg = BPF_REG_9; \
10018 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
10020 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
10022 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
10023 offsetof(S, TF)); \
10024 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
10025 si->dst_reg, offsetof(S, F)); \
10026 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \
10027 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
10030 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
10031 offsetof(S, TF)); \
10034 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
10037 if (type == BPF_WRITE) { \
10038 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
10041 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
10042 S, NS, F, NF, SIZE, OFF); \
10046 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
10047 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
10048 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
10050 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
10051 const struct bpf_insn *si,
10052 struct bpf_insn *insn_buf,
10053 struct bpf_prog *prog, u32 *target_size)
10055 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
10056 struct bpf_insn *insn = insn_buf;
10059 case offsetof(struct bpf_sock_addr, user_family):
10060 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10061 struct sockaddr, uaddr, sa_family);
10064 case offsetof(struct bpf_sock_addr, user_ip4):
10065 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10066 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
10067 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
10070 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
10072 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
10073 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10074 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
10075 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
10079 case offsetof(struct bpf_sock_addr, user_port):
10080 /* To get port we need to know sa_family first and then treat
10081 * sockaddr as either sockaddr_in or sockaddr_in6.
10082 * Though we can simplify since port field has same offset and
10083 * size in both structures.
10084 * Here we check this invariant and use just one of the
10085 * structures if it's true.
10087 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
10088 offsetof(struct sockaddr_in6, sin6_port));
10089 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
10090 sizeof_field(struct sockaddr_in6, sin6_port));
10091 /* Account for sin6_port being smaller than user_port. */
10092 port_size = min(port_size, BPF_LDST_BYTES(si));
10093 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10094 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
10095 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
10098 case offsetof(struct bpf_sock_addr, family):
10099 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10100 struct sock, sk, sk_family);
10103 case offsetof(struct bpf_sock_addr, type):
10104 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10105 struct sock, sk, sk_type);
10108 case offsetof(struct bpf_sock_addr, protocol):
10109 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10110 struct sock, sk, sk_protocol);
10113 case offsetof(struct bpf_sock_addr, msg_src_ip4):
10114 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
10115 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10116 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
10117 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
10120 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
10123 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
10124 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
10125 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10126 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
10127 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
10129 case offsetof(struct bpf_sock_addr, sk):
10130 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
10131 si->dst_reg, si->src_reg,
10132 offsetof(struct bpf_sock_addr_kern, sk));
10136 return insn - insn_buf;
10139 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
10140 const struct bpf_insn *si,
10141 struct bpf_insn *insn_buf,
10142 struct bpf_prog *prog,
10145 struct bpf_insn *insn = insn_buf;
10148 /* Helper macro for adding read access to tcp_sock or sock fields. */
10149 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
10151 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \
10152 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
10153 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
10154 if (si->dst_reg == reg || si->src_reg == reg) \
10156 if (si->dst_reg == reg || si->src_reg == reg) \
10158 if (si->dst_reg == si->src_reg) { \
10159 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
10160 offsetof(struct bpf_sock_ops_kern, \
10162 fullsock_reg = reg; \
10165 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10166 struct bpf_sock_ops_kern, \
10168 fullsock_reg, si->src_reg, \
10169 offsetof(struct bpf_sock_ops_kern, \
10171 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
10172 if (si->dst_reg == si->src_reg) \
10173 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10174 offsetof(struct bpf_sock_ops_kern, \
10176 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10177 struct bpf_sock_ops_kern, sk),\
10178 si->dst_reg, si->src_reg, \
10179 offsetof(struct bpf_sock_ops_kern, sk));\
10180 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
10182 si->dst_reg, si->dst_reg, \
10183 offsetof(OBJ, OBJ_FIELD)); \
10184 if (si->dst_reg == si->src_reg) { \
10185 *insn++ = BPF_JMP_A(1); \
10186 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10187 offsetof(struct bpf_sock_ops_kern, \
10192 #define SOCK_OPS_GET_SK() \
10194 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \
10195 if (si->dst_reg == reg || si->src_reg == reg) \
10197 if (si->dst_reg == reg || si->src_reg == reg) \
10199 if (si->dst_reg == si->src_reg) { \
10200 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
10201 offsetof(struct bpf_sock_ops_kern, \
10203 fullsock_reg = reg; \
10206 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10207 struct bpf_sock_ops_kern, \
10209 fullsock_reg, si->src_reg, \
10210 offsetof(struct bpf_sock_ops_kern, \
10212 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
10213 if (si->dst_reg == si->src_reg) \
10214 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10215 offsetof(struct bpf_sock_ops_kern, \
10217 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10218 struct bpf_sock_ops_kern, sk),\
10219 si->dst_reg, si->src_reg, \
10220 offsetof(struct bpf_sock_ops_kern, sk));\
10221 if (si->dst_reg == si->src_reg) { \
10222 *insn++ = BPF_JMP_A(1); \
10223 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10224 offsetof(struct bpf_sock_ops_kern, \
10229 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
10230 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
10232 /* Helper macro for adding write access to tcp_sock or sock fields.
10233 * The macro is called with two registers, dst_reg which contains a pointer
10234 * to ctx (context) and src_reg which contains the value that should be
10235 * stored. However, we need an additional register since we cannot overwrite
10236 * dst_reg because it may be used later in the program.
10237 * Instead we "borrow" one of the other register. We first save its value
10238 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
10239 * it at the end of the macro.
10241 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
10243 int reg = BPF_REG_9; \
10244 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
10245 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
10246 if (si->dst_reg == reg || si->src_reg == reg) \
10248 if (si->dst_reg == reg || si->src_reg == reg) \
10250 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
10251 offsetof(struct bpf_sock_ops_kern, \
10253 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10254 struct bpf_sock_ops_kern, \
10256 reg, si->dst_reg, \
10257 offsetof(struct bpf_sock_ops_kern, \
10259 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
10260 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10261 struct bpf_sock_ops_kern, sk),\
10262 reg, si->dst_reg, \
10263 offsetof(struct bpf_sock_ops_kern, sk));\
10264 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \
10265 reg, si->src_reg, \
10266 offsetof(OBJ, OBJ_FIELD)); \
10267 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
10268 offsetof(struct bpf_sock_ops_kern, \
10272 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
10274 if (TYPE == BPF_WRITE) \
10275 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
10277 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
10280 if (insn > insn_buf)
10281 return insn - insn_buf;
10284 case offsetof(struct bpf_sock_ops, op):
10285 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10287 si->dst_reg, si->src_reg,
10288 offsetof(struct bpf_sock_ops_kern, op));
10291 case offsetof(struct bpf_sock_ops, replylong[0]) ...
10292 offsetof(struct bpf_sock_ops, replylong[3]):
10293 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
10294 sizeof_field(struct bpf_sock_ops_kern, reply));
10295 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
10296 sizeof_field(struct bpf_sock_ops_kern, replylong));
10298 off -= offsetof(struct bpf_sock_ops, replylong[0]);
10299 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
10300 if (type == BPF_WRITE)
10301 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
10304 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10308 case offsetof(struct bpf_sock_ops, family):
10309 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10311 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10312 struct bpf_sock_ops_kern, sk),
10313 si->dst_reg, si->src_reg,
10314 offsetof(struct bpf_sock_ops_kern, sk));
10315 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10316 offsetof(struct sock_common, skc_family));
10319 case offsetof(struct bpf_sock_ops, remote_ip4):
10320 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10322 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10323 struct bpf_sock_ops_kern, sk),
10324 si->dst_reg, si->src_reg,
10325 offsetof(struct bpf_sock_ops_kern, sk));
10326 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10327 offsetof(struct sock_common, skc_daddr));
10330 case offsetof(struct bpf_sock_ops, local_ip4):
10331 BUILD_BUG_ON(sizeof_field(struct sock_common,
10332 skc_rcv_saddr) != 4);
10334 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10335 struct bpf_sock_ops_kern, sk),
10336 si->dst_reg, si->src_reg,
10337 offsetof(struct bpf_sock_ops_kern, sk));
10338 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10339 offsetof(struct sock_common,
10343 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
10344 offsetof(struct bpf_sock_ops, remote_ip6[3]):
10345 #if IS_ENABLED(CONFIG_IPV6)
10346 BUILD_BUG_ON(sizeof_field(struct sock_common,
10347 skc_v6_daddr.s6_addr32[0]) != 4);
10350 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
10351 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10352 struct bpf_sock_ops_kern, sk),
10353 si->dst_reg, si->src_reg,
10354 offsetof(struct bpf_sock_ops_kern, sk));
10355 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10356 offsetof(struct sock_common,
10357 skc_v6_daddr.s6_addr32[0]) +
10360 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10364 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
10365 offsetof(struct bpf_sock_ops, local_ip6[3]):
10366 #if IS_ENABLED(CONFIG_IPV6)
10367 BUILD_BUG_ON(sizeof_field(struct sock_common,
10368 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10371 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
10372 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10373 struct bpf_sock_ops_kern, sk),
10374 si->dst_reg, si->src_reg,
10375 offsetof(struct bpf_sock_ops_kern, sk));
10376 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10377 offsetof(struct sock_common,
10378 skc_v6_rcv_saddr.s6_addr32[0]) +
10381 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10385 case offsetof(struct bpf_sock_ops, remote_port):
10386 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10388 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10389 struct bpf_sock_ops_kern, sk),
10390 si->dst_reg, si->src_reg,
10391 offsetof(struct bpf_sock_ops_kern, sk));
10392 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10393 offsetof(struct sock_common, skc_dport));
10394 #ifndef __BIG_ENDIAN_BITFIELD
10395 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10399 case offsetof(struct bpf_sock_ops, local_port):
10400 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10402 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10403 struct bpf_sock_ops_kern, sk),
10404 si->dst_reg, si->src_reg,
10405 offsetof(struct bpf_sock_ops_kern, sk));
10406 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10407 offsetof(struct sock_common, skc_num));
10410 case offsetof(struct bpf_sock_ops, is_fullsock):
10411 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10412 struct bpf_sock_ops_kern,
10414 si->dst_reg, si->src_reg,
10415 offsetof(struct bpf_sock_ops_kern,
10419 case offsetof(struct bpf_sock_ops, state):
10420 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
10422 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10423 struct bpf_sock_ops_kern, sk),
10424 si->dst_reg, si->src_reg,
10425 offsetof(struct bpf_sock_ops_kern, sk));
10426 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
10427 offsetof(struct sock_common, skc_state));
10430 case offsetof(struct bpf_sock_ops, rtt_min):
10431 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
10432 sizeof(struct minmax));
10433 BUILD_BUG_ON(sizeof(struct minmax) <
10434 sizeof(struct minmax_sample));
10436 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10437 struct bpf_sock_ops_kern, sk),
10438 si->dst_reg, si->src_reg,
10439 offsetof(struct bpf_sock_ops_kern, sk));
10440 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10441 offsetof(struct tcp_sock, rtt_min) +
10442 sizeof_field(struct minmax_sample, t));
10445 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
10446 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
10450 case offsetof(struct bpf_sock_ops, sk_txhash):
10451 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
10452 struct sock, type);
10454 case offsetof(struct bpf_sock_ops, snd_cwnd):
10455 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
10457 case offsetof(struct bpf_sock_ops, srtt_us):
10458 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
10460 case offsetof(struct bpf_sock_ops, snd_ssthresh):
10461 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
10463 case offsetof(struct bpf_sock_ops, rcv_nxt):
10464 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
10466 case offsetof(struct bpf_sock_ops, snd_nxt):
10467 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
10469 case offsetof(struct bpf_sock_ops, snd_una):
10470 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
10472 case offsetof(struct bpf_sock_ops, mss_cache):
10473 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
10475 case offsetof(struct bpf_sock_ops, ecn_flags):
10476 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
10478 case offsetof(struct bpf_sock_ops, rate_delivered):
10479 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
10481 case offsetof(struct bpf_sock_ops, rate_interval_us):
10482 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
10484 case offsetof(struct bpf_sock_ops, packets_out):
10485 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
10487 case offsetof(struct bpf_sock_ops, retrans_out):
10488 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
10490 case offsetof(struct bpf_sock_ops, total_retrans):
10491 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
10493 case offsetof(struct bpf_sock_ops, segs_in):
10494 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
10496 case offsetof(struct bpf_sock_ops, data_segs_in):
10497 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
10499 case offsetof(struct bpf_sock_ops, segs_out):
10500 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
10502 case offsetof(struct bpf_sock_ops, data_segs_out):
10503 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
10505 case offsetof(struct bpf_sock_ops, lost_out):
10506 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
10508 case offsetof(struct bpf_sock_ops, sacked_out):
10509 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
10511 case offsetof(struct bpf_sock_ops, bytes_received):
10512 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
10514 case offsetof(struct bpf_sock_ops, bytes_acked):
10515 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
10517 case offsetof(struct bpf_sock_ops, sk):
10520 case offsetof(struct bpf_sock_ops, skb_data_end):
10521 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10523 si->dst_reg, si->src_reg,
10524 offsetof(struct bpf_sock_ops_kern,
10527 case offsetof(struct bpf_sock_ops, skb_data):
10528 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10530 si->dst_reg, si->src_reg,
10531 offsetof(struct bpf_sock_ops_kern,
10533 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10534 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10535 si->dst_reg, si->dst_reg,
10536 offsetof(struct sk_buff, data));
10538 case offsetof(struct bpf_sock_ops, skb_len):
10539 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10541 si->dst_reg, si->src_reg,
10542 offsetof(struct bpf_sock_ops_kern,
10544 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10545 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10546 si->dst_reg, si->dst_reg,
10547 offsetof(struct sk_buff, len));
10549 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
10550 off = offsetof(struct sk_buff, cb);
10551 off += offsetof(struct tcp_skb_cb, tcp_flags);
10552 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
10553 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10555 si->dst_reg, si->src_reg,
10556 offsetof(struct bpf_sock_ops_kern,
10558 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10559 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
10561 si->dst_reg, si->dst_reg, off);
10564 return insn - insn_buf;
10567 /* data_end = skb->data + skb_headlen() */
10568 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
10569 struct bpf_insn *insn)
10572 int temp_reg_off = offsetof(struct sk_buff, cb) +
10573 offsetof(struct sk_skb_cb, temp_reg);
10575 if (si->src_reg == si->dst_reg) {
10576 /* We need an extra register, choose and save a register. */
10578 if (si->src_reg == reg || si->dst_reg == reg)
10580 if (si->src_reg == reg || si->dst_reg == reg)
10582 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off);
10587 /* reg = skb->data */
10588 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10590 offsetof(struct sk_buff, data));
10591 /* AX = skb->len */
10592 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10593 BPF_REG_AX, si->src_reg,
10594 offsetof(struct sk_buff, len));
10595 /* reg = skb->data + skb->len */
10596 *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX);
10597 /* AX = skb->data_len */
10598 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
10599 BPF_REG_AX, si->src_reg,
10600 offsetof(struct sk_buff, data_len));
10602 /* reg = skb->data + skb->len - skb->data_len */
10603 *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX);
10605 if (si->src_reg == si->dst_reg) {
10606 /* Restore the saved register */
10607 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg);
10608 *insn++ = BPF_MOV64_REG(si->dst_reg, reg);
10609 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off);
10615 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
10616 const struct bpf_insn *si,
10617 struct bpf_insn *insn_buf,
10618 struct bpf_prog *prog, u32 *target_size)
10620 struct bpf_insn *insn = insn_buf;
10624 case offsetof(struct __sk_buff, data_end):
10625 insn = bpf_convert_data_end_access(si, insn);
10627 case offsetof(struct __sk_buff, cb[0]) ...
10628 offsetofend(struct __sk_buff, cb[4]) - 1:
10629 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20);
10630 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
10631 offsetof(struct sk_skb_cb, data)) %
10634 prog->cb_access = 1;
10636 off -= offsetof(struct __sk_buff, cb[0]);
10637 off += offsetof(struct sk_buff, cb);
10638 off += offsetof(struct sk_skb_cb, data);
10639 if (type == BPF_WRITE)
10640 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
10643 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
10649 return bpf_convert_ctx_access(type, si, insn_buf, prog,
10653 return insn - insn_buf;
10656 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
10657 const struct bpf_insn *si,
10658 struct bpf_insn *insn_buf,
10659 struct bpf_prog *prog, u32 *target_size)
10661 struct bpf_insn *insn = insn_buf;
10662 #if IS_ENABLED(CONFIG_IPV6)
10666 /* convert ctx uses the fact sg element is first in struct */
10667 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
10670 case offsetof(struct sk_msg_md, data):
10671 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
10672 si->dst_reg, si->src_reg,
10673 offsetof(struct sk_msg, data));
10675 case offsetof(struct sk_msg_md, data_end):
10676 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
10677 si->dst_reg, si->src_reg,
10678 offsetof(struct sk_msg, data_end));
10680 case offsetof(struct sk_msg_md, family):
10681 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10683 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10684 struct sk_msg, sk),
10685 si->dst_reg, si->src_reg,
10686 offsetof(struct sk_msg, sk));
10687 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10688 offsetof(struct sock_common, skc_family));
10691 case offsetof(struct sk_msg_md, remote_ip4):
10692 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10694 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10695 struct sk_msg, sk),
10696 si->dst_reg, si->src_reg,
10697 offsetof(struct sk_msg, sk));
10698 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10699 offsetof(struct sock_common, skc_daddr));
10702 case offsetof(struct sk_msg_md, local_ip4):
10703 BUILD_BUG_ON(sizeof_field(struct sock_common,
10704 skc_rcv_saddr) != 4);
10706 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10707 struct sk_msg, sk),
10708 si->dst_reg, si->src_reg,
10709 offsetof(struct sk_msg, sk));
10710 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10711 offsetof(struct sock_common,
10715 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
10716 offsetof(struct sk_msg_md, remote_ip6[3]):
10717 #if IS_ENABLED(CONFIG_IPV6)
10718 BUILD_BUG_ON(sizeof_field(struct sock_common,
10719 skc_v6_daddr.s6_addr32[0]) != 4);
10722 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
10723 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10724 struct sk_msg, sk),
10725 si->dst_reg, si->src_reg,
10726 offsetof(struct sk_msg, sk));
10727 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10728 offsetof(struct sock_common,
10729 skc_v6_daddr.s6_addr32[0]) +
10732 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10736 case offsetof(struct sk_msg_md, local_ip6[0]) ...
10737 offsetof(struct sk_msg_md, local_ip6[3]):
10738 #if IS_ENABLED(CONFIG_IPV6)
10739 BUILD_BUG_ON(sizeof_field(struct sock_common,
10740 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10743 off -= offsetof(struct sk_msg_md, local_ip6[0]);
10744 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10745 struct sk_msg, sk),
10746 si->dst_reg, si->src_reg,
10747 offsetof(struct sk_msg, sk));
10748 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10749 offsetof(struct sock_common,
10750 skc_v6_rcv_saddr.s6_addr32[0]) +
10753 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10757 case offsetof(struct sk_msg_md, remote_port):
10758 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10760 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10761 struct sk_msg, sk),
10762 si->dst_reg, si->src_reg,
10763 offsetof(struct sk_msg, sk));
10764 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10765 offsetof(struct sock_common, skc_dport));
10766 #ifndef __BIG_ENDIAN_BITFIELD
10767 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10771 case offsetof(struct sk_msg_md, local_port):
10772 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10774 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10775 struct sk_msg, sk),
10776 si->dst_reg, si->src_reg,
10777 offsetof(struct sk_msg, sk));
10778 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10779 offsetof(struct sock_common, skc_num));
10782 case offsetof(struct sk_msg_md, size):
10783 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
10784 si->dst_reg, si->src_reg,
10785 offsetof(struct sk_msg_sg, size));
10788 case offsetof(struct sk_msg_md, sk):
10789 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
10790 si->dst_reg, si->src_reg,
10791 offsetof(struct sk_msg, sk));
10795 return insn - insn_buf;
10798 const struct bpf_verifier_ops sk_filter_verifier_ops = {
10799 .get_func_proto = sk_filter_func_proto,
10800 .is_valid_access = sk_filter_is_valid_access,
10801 .convert_ctx_access = bpf_convert_ctx_access,
10802 .gen_ld_abs = bpf_gen_ld_abs,
10805 const struct bpf_prog_ops sk_filter_prog_ops = {
10806 .test_run = bpf_prog_test_run_skb,
10809 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
10810 .get_func_proto = tc_cls_act_func_proto,
10811 .is_valid_access = tc_cls_act_is_valid_access,
10812 .convert_ctx_access = tc_cls_act_convert_ctx_access,
10813 .gen_prologue = tc_cls_act_prologue,
10814 .gen_ld_abs = bpf_gen_ld_abs,
10815 .btf_struct_access = tc_cls_act_btf_struct_access,
10818 const struct bpf_prog_ops tc_cls_act_prog_ops = {
10819 .test_run = bpf_prog_test_run_skb,
10822 const struct bpf_verifier_ops xdp_verifier_ops = {
10823 .get_func_proto = xdp_func_proto,
10824 .is_valid_access = xdp_is_valid_access,
10825 .convert_ctx_access = xdp_convert_ctx_access,
10826 .gen_prologue = bpf_noop_prologue,
10827 .btf_struct_access = xdp_btf_struct_access,
10830 const struct bpf_prog_ops xdp_prog_ops = {
10831 .test_run = bpf_prog_test_run_xdp,
10834 const struct bpf_verifier_ops cg_skb_verifier_ops = {
10835 .get_func_proto = cg_skb_func_proto,
10836 .is_valid_access = cg_skb_is_valid_access,
10837 .convert_ctx_access = bpf_convert_ctx_access,
10840 const struct bpf_prog_ops cg_skb_prog_ops = {
10841 .test_run = bpf_prog_test_run_skb,
10844 const struct bpf_verifier_ops lwt_in_verifier_ops = {
10845 .get_func_proto = lwt_in_func_proto,
10846 .is_valid_access = lwt_is_valid_access,
10847 .convert_ctx_access = bpf_convert_ctx_access,
10850 const struct bpf_prog_ops lwt_in_prog_ops = {
10851 .test_run = bpf_prog_test_run_skb,
10854 const struct bpf_verifier_ops lwt_out_verifier_ops = {
10855 .get_func_proto = lwt_out_func_proto,
10856 .is_valid_access = lwt_is_valid_access,
10857 .convert_ctx_access = bpf_convert_ctx_access,
10860 const struct bpf_prog_ops lwt_out_prog_ops = {
10861 .test_run = bpf_prog_test_run_skb,
10864 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
10865 .get_func_proto = lwt_xmit_func_proto,
10866 .is_valid_access = lwt_is_valid_access,
10867 .convert_ctx_access = bpf_convert_ctx_access,
10868 .gen_prologue = tc_cls_act_prologue,
10871 const struct bpf_prog_ops lwt_xmit_prog_ops = {
10872 .test_run = bpf_prog_test_run_skb,
10875 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
10876 .get_func_proto = lwt_seg6local_func_proto,
10877 .is_valid_access = lwt_is_valid_access,
10878 .convert_ctx_access = bpf_convert_ctx_access,
10881 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
10882 .test_run = bpf_prog_test_run_skb,
10885 const struct bpf_verifier_ops cg_sock_verifier_ops = {
10886 .get_func_proto = sock_filter_func_proto,
10887 .is_valid_access = sock_filter_is_valid_access,
10888 .convert_ctx_access = bpf_sock_convert_ctx_access,
10891 const struct bpf_prog_ops cg_sock_prog_ops = {
10894 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
10895 .get_func_proto = sock_addr_func_proto,
10896 .is_valid_access = sock_addr_is_valid_access,
10897 .convert_ctx_access = sock_addr_convert_ctx_access,
10900 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
10903 const struct bpf_verifier_ops sock_ops_verifier_ops = {
10904 .get_func_proto = sock_ops_func_proto,
10905 .is_valid_access = sock_ops_is_valid_access,
10906 .convert_ctx_access = sock_ops_convert_ctx_access,
10909 const struct bpf_prog_ops sock_ops_prog_ops = {
10912 const struct bpf_verifier_ops sk_skb_verifier_ops = {
10913 .get_func_proto = sk_skb_func_proto,
10914 .is_valid_access = sk_skb_is_valid_access,
10915 .convert_ctx_access = sk_skb_convert_ctx_access,
10916 .gen_prologue = sk_skb_prologue,
10919 const struct bpf_prog_ops sk_skb_prog_ops = {
10922 const struct bpf_verifier_ops sk_msg_verifier_ops = {
10923 .get_func_proto = sk_msg_func_proto,
10924 .is_valid_access = sk_msg_is_valid_access,
10925 .convert_ctx_access = sk_msg_convert_ctx_access,
10926 .gen_prologue = bpf_noop_prologue,
10929 const struct bpf_prog_ops sk_msg_prog_ops = {
10932 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
10933 .get_func_proto = flow_dissector_func_proto,
10934 .is_valid_access = flow_dissector_is_valid_access,
10935 .convert_ctx_access = flow_dissector_convert_ctx_access,
10938 const struct bpf_prog_ops flow_dissector_prog_ops = {
10939 .test_run = bpf_prog_test_run_flow_dissector,
10942 int sk_detach_filter(struct sock *sk)
10945 struct sk_filter *filter;
10947 if (sock_flag(sk, SOCK_FILTER_LOCKED))
10950 filter = rcu_dereference_protected(sk->sk_filter,
10951 lockdep_sock_is_held(sk));
10953 RCU_INIT_POINTER(sk->sk_filter, NULL);
10954 sk_filter_uncharge(sk, filter);
10960 EXPORT_SYMBOL_GPL(sk_detach_filter);
10962 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len)
10964 struct sock_fprog_kern *fprog;
10965 struct sk_filter *filter;
10968 sockopt_lock_sock(sk);
10969 filter = rcu_dereference_protected(sk->sk_filter,
10970 lockdep_sock_is_held(sk));
10974 /* We're copying the filter that has been originally attached,
10975 * so no conversion/decode needed anymore. eBPF programs that
10976 * have no original program cannot be dumped through this.
10979 fprog = filter->prog->orig_prog;
10985 /* User space only enquires number of filter blocks. */
10989 if (len < fprog->len)
10993 if (copy_to_sockptr(optval, fprog->filter, bpf_classic_proglen(fprog)))
10996 /* Instead of bytes, the API requests to return the number
10997 * of filter blocks.
11001 sockopt_release_sock(sk);
11006 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
11007 struct sock_reuseport *reuse,
11008 struct sock *sk, struct sk_buff *skb,
11009 struct sock *migrating_sk,
11012 reuse_kern->skb = skb;
11013 reuse_kern->sk = sk;
11014 reuse_kern->selected_sk = NULL;
11015 reuse_kern->migrating_sk = migrating_sk;
11016 reuse_kern->data_end = skb->data + skb_headlen(skb);
11017 reuse_kern->hash = hash;
11018 reuse_kern->reuseport_id = reuse->reuseport_id;
11019 reuse_kern->bind_inany = reuse->bind_inany;
11022 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
11023 struct bpf_prog *prog, struct sk_buff *skb,
11024 struct sock *migrating_sk,
11027 struct sk_reuseport_kern reuse_kern;
11028 enum sk_action action;
11030 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
11031 action = bpf_prog_run(prog, &reuse_kern);
11033 if (action == SK_PASS)
11034 return reuse_kern.selected_sk;
11036 return ERR_PTR(-ECONNREFUSED);
11039 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
11040 struct bpf_map *, map, void *, key, u32, flags)
11042 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
11043 struct sock_reuseport *reuse;
11044 struct sock *selected_sk;
11046 selected_sk = map->ops->map_lookup_elem(map, key);
11050 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
11052 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
11053 if (sk_is_refcounted(selected_sk))
11054 sock_put(selected_sk);
11056 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
11057 * The only (!reuse) case here is - the sk has already been
11058 * unhashed (e.g. by close()), so treat it as -ENOENT.
11060 * Other maps (e.g. sock_map) do not provide this guarantee and
11061 * the sk may never be in the reuseport group to begin with.
11063 return is_sockarray ? -ENOENT : -EINVAL;
11066 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
11067 struct sock *sk = reuse_kern->sk;
11069 if (sk->sk_protocol != selected_sk->sk_protocol)
11070 return -EPROTOTYPE;
11071 else if (sk->sk_family != selected_sk->sk_family)
11072 return -EAFNOSUPPORT;
11074 /* Catch all. Likely bound to a different sockaddr. */
11078 reuse_kern->selected_sk = selected_sk;
11083 static const struct bpf_func_proto sk_select_reuseport_proto = {
11084 .func = sk_select_reuseport,
11086 .ret_type = RET_INTEGER,
11087 .arg1_type = ARG_PTR_TO_CTX,
11088 .arg2_type = ARG_CONST_MAP_PTR,
11089 .arg3_type = ARG_PTR_TO_MAP_KEY,
11090 .arg4_type = ARG_ANYTHING,
11093 BPF_CALL_4(sk_reuseport_load_bytes,
11094 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11095 void *, to, u32, len)
11097 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
11100 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
11101 .func = sk_reuseport_load_bytes,
11103 .ret_type = RET_INTEGER,
11104 .arg1_type = ARG_PTR_TO_CTX,
11105 .arg2_type = ARG_ANYTHING,
11106 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
11107 .arg4_type = ARG_CONST_SIZE,
11110 BPF_CALL_5(sk_reuseport_load_bytes_relative,
11111 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11112 void *, to, u32, len, u32, start_header)
11114 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
11115 len, start_header);
11118 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
11119 .func = sk_reuseport_load_bytes_relative,
11121 .ret_type = RET_INTEGER,
11122 .arg1_type = ARG_PTR_TO_CTX,
11123 .arg2_type = ARG_ANYTHING,
11124 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
11125 .arg4_type = ARG_CONST_SIZE,
11126 .arg5_type = ARG_ANYTHING,
11129 static const struct bpf_func_proto *
11130 sk_reuseport_func_proto(enum bpf_func_id func_id,
11131 const struct bpf_prog *prog)
11134 case BPF_FUNC_sk_select_reuseport:
11135 return &sk_select_reuseport_proto;
11136 case BPF_FUNC_skb_load_bytes:
11137 return &sk_reuseport_load_bytes_proto;
11138 case BPF_FUNC_skb_load_bytes_relative:
11139 return &sk_reuseport_load_bytes_relative_proto;
11140 case BPF_FUNC_get_socket_cookie:
11141 return &bpf_get_socket_ptr_cookie_proto;
11142 case BPF_FUNC_ktime_get_coarse_ns:
11143 return &bpf_ktime_get_coarse_ns_proto;
11145 return bpf_base_func_proto(func_id);
11150 sk_reuseport_is_valid_access(int off, int size,
11151 enum bpf_access_type type,
11152 const struct bpf_prog *prog,
11153 struct bpf_insn_access_aux *info)
11155 const u32 size_default = sizeof(__u32);
11157 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
11158 off % size || type != BPF_READ)
11162 case offsetof(struct sk_reuseport_md, data):
11163 info->reg_type = PTR_TO_PACKET;
11164 return size == sizeof(__u64);
11166 case offsetof(struct sk_reuseport_md, data_end):
11167 info->reg_type = PTR_TO_PACKET_END;
11168 return size == sizeof(__u64);
11170 case offsetof(struct sk_reuseport_md, hash):
11171 return size == size_default;
11173 case offsetof(struct sk_reuseport_md, sk):
11174 info->reg_type = PTR_TO_SOCKET;
11175 return size == sizeof(__u64);
11177 case offsetof(struct sk_reuseport_md, migrating_sk):
11178 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
11179 return size == sizeof(__u64);
11181 /* Fields that allow narrowing */
11182 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
11183 if (size < sizeof_field(struct sk_buff, protocol))
11186 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
11187 case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
11188 case bpf_ctx_range(struct sk_reuseport_md, len):
11189 bpf_ctx_record_field_size(info, size_default);
11190 return bpf_ctx_narrow_access_ok(off, size, size_default);
11197 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
11198 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
11199 si->dst_reg, si->src_reg, \
11200 bpf_target_off(struct sk_reuseport_kern, F, \
11201 sizeof_field(struct sk_reuseport_kern, F), \
11205 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
11206 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
11211 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \
11212 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
11217 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
11218 const struct bpf_insn *si,
11219 struct bpf_insn *insn_buf,
11220 struct bpf_prog *prog,
11223 struct bpf_insn *insn = insn_buf;
11226 case offsetof(struct sk_reuseport_md, data):
11227 SK_REUSEPORT_LOAD_SKB_FIELD(data);
11230 case offsetof(struct sk_reuseport_md, len):
11231 SK_REUSEPORT_LOAD_SKB_FIELD(len);
11234 case offsetof(struct sk_reuseport_md, eth_protocol):
11235 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
11238 case offsetof(struct sk_reuseport_md, ip_protocol):
11239 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
11242 case offsetof(struct sk_reuseport_md, data_end):
11243 SK_REUSEPORT_LOAD_FIELD(data_end);
11246 case offsetof(struct sk_reuseport_md, hash):
11247 SK_REUSEPORT_LOAD_FIELD(hash);
11250 case offsetof(struct sk_reuseport_md, bind_inany):
11251 SK_REUSEPORT_LOAD_FIELD(bind_inany);
11254 case offsetof(struct sk_reuseport_md, sk):
11255 SK_REUSEPORT_LOAD_FIELD(sk);
11258 case offsetof(struct sk_reuseport_md, migrating_sk):
11259 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
11263 return insn - insn_buf;
11266 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
11267 .get_func_proto = sk_reuseport_func_proto,
11268 .is_valid_access = sk_reuseport_is_valid_access,
11269 .convert_ctx_access = sk_reuseport_convert_ctx_access,
11272 const struct bpf_prog_ops sk_reuseport_prog_ops = {
11275 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
11276 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
11278 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
11279 struct sock *, sk, u64, flags)
11281 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
11282 BPF_SK_LOOKUP_F_NO_REUSEPORT)))
11284 if (unlikely(sk && sk_is_refcounted(sk)))
11285 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
11286 if (unlikely(sk && sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN))
11287 return -ESOCKTNOSUPPORT; /* only accept TCP socket in LISTEN */
11288 if (unlikely(sk && sk_is_udp(sk) && sk->sk_state != TCP_CLOSE))
11289 return -ESOCKTNOSUPPORT; /* only accept UDP socket in CLOSE */
11291 /* Check if socket is suitable for packet L3/L4 protocol */
11292 if (sk && sk->sk_protocol != ctx->protocol)
11293 return -EPROTOTYPE;
11294 if (sk && sk->sk_family != ctx->family &&
11295 (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
11296 return -EAFNOSUPPORT;
11298 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
11301 /* Select socket as lookup result */
11302 ctx->selected_sk = sk;
11303 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
11307 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
11308 .func = bpf_sk_lookup_assign,
11310 .ret_type = RET_INTEGER,
11311 .arg1_type = ARG_PTR_TO_CTX,
11312 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL,
11313 .arg3_type = ARG_ANYTHING,
11316 static const struct bpf_func_proto *
11317 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
11320 case BPF_FUNC_perf_event_output:
11321 return &bpf_event_output_data_proto;
11322 case BPF_FUNC_sk_assign:
11323 return &bpf_sk_lookup_assign_proto;
11324 case BPF_FUNC_sk_release:
11325 return &bpf_sk_release_proto;
11327 return bpf_sk_base_func_proto(func_id);
11331 static bool sk_lookup_is_valid_access(int off, int size,
11332 enum bpf_access_type type,
11333 const struct bpf_prog *prog,
11334 struct bpf_insn_access_aux *info)
11336 if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
11338 if (off % size != 0)
11340 if (type != BPF_READ)
11344 case offsetof(struct bpf_sk_lookup, sk):
11345 info->reg_type = PTR_TO_SOCKET_OR_NULL;
11346 return size == sizeof(__u64);
11348 case bpf_ctx_range(struct bpf_sk_lookup, family):
11349 case bpf_ctx_range(struct bpf_sk_lookup, protocol):
11350 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
11351 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
11352 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
11353 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
11354 case bpf_ctx_range(struct bpf_sk_lookup, local_port):
11355 case bpf_ctx_range(struct bpf_sk_lookup, ingress_ifindex):
11356 bpf_ctx_record_field_size(info, sizeof(__u32));
11357 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
11359 case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
11360 /* Allow 4-byte access to 2-byte field for backward compatibility */
11361 if (size == sizeof(__u32))
11363 bpf_ctx_record_field_size(info, sizeof(__be16));
11364 return bpf_ctx_narrow_access_ok(off, size, sizeof(__be16));
11366 case offsetofend(struct bpf_sk_lookup, remote_port) ...
11367 offsetof(struct bpf_sk_lookup, local_ip4) - 1:
11368 /* Allow access to zero padding for backward compatibility */
11369 bpf_ctx_record_field_size(info, sizeof(__u16));
11370 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u16));
11377 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
11378 const struct bpf_insn *si,
11379 struct bpf_insn *insn_buf,
11380 struct bpf_prog *prog,
11383 struct bpf_insn *insn = insn_buf;
11386 case offsetof(struct bpf_sk_lookup, sk):
11387 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11388 offsetof(struct bpf_sk_lookup_kern, selected_sk));
11391 case offsetof(struct bpf_sk_lookup, family):
11392 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11393 bpf_target_off(struct bpf_sk_lookup_kern,
11394 family, 2, target_size));
11397 case offsetof(struct bpf_sk_lookup, protocol):
11398 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11399 bpf_target_off(struct bpf_sk_lookup_kern,
11400 protocol, 2, target_size));
11403 case offsetof(struct bpf_sk_lookup, remote_ip4):
11404 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11405 bpf_target_off(struct bpf_sk_lookup_kern,
11406 v4.saddr, 4, target_size));
11409 case offsetof(struct bpf_sk_lookup, local_ip4):
11410 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11411 bpf_target_off(struct bpf_sk_lookup_kern,
11412 v4.daddr, 4, target_size));
11415 case bpf_ctx_range_till(struct bpf_sk_lookup,
11416 remote_ip6[0], remote_ip6[3]): {
11417 #if IS_ENABLED(CONFIG_IPV6)
11420 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
11421 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11422 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11423 offsetof(struct bpf_sk_lookup_kern, v6.saddr));
11424 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11425 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11427 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11431 case bpf_ctx_range_till(struct bpf_sk_lookup,
11432 local_ip6[0], local_ip6[3]): {
11433 #if IS_ENABLED(CONFIG_IPV6)
11436 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
11437 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11438 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11439 offsetof(struct bpf_sk_lookup_kern, v6.daddr));
11440 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11441 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11443 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11447 case offsetof(struct bpf_sk_lookup, remote_port):
11448 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11449 bpf_target_off(struct bpf_sk_lookup_kern,
11450 sport, 2, target_size));
11453 case offsetofend(struct bpf_sk_lookup, remote_port):
11455 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11458 case offsetof(struct bpf_sk_lookup, local_port):
11459 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11460 bpf_target_off(struct bpf_sk_lookup_kern,
11461 dport, 2, target_size));
11464 case offsetof(struct bpf_sk_lookup, ingress_ifindex):
11465 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11466 bpf_target_off(struct bpf_sk_lookup_kern,
11467 ingress_ifindex, 4, target_size));
11471 return insn - insn_buf;
11474 const struct bpf_prog_ops sk_lookup_prog_ops = {
11475 .test_run = bpf_prog_test_run_sk_lookup,
11478 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
11479 .get_func_proto = sk_lookup_func_proto,
11480 .is_valid_access = sk_lookup_is_valid_access,
11481 .convert_ctx_access = sk_lookup_convert_ctx_access,
11484 #endif /* CONFIG_INET */
11486 DEFINE_BPF_DISPATCHER(xdp)
11488 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
11490 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
11493 BTF_ID_LIST_GLOBAL(btf_sock_ids, MAX_BTF_SOCK_TYPE)
11494 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
11496 #undef BTF_SOCK_TYPE
11498 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
11500 /* tcp6_sock type is not generated in dwarf and hence btf,
11501 * trigger an explicit type generation here.
11503 BTF_TYPE_EMIT(struct tcp6_sock);
11504 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
11505 sk->sk_family == AF_INET6)
11506 return (unsigned long)sk;
11508 return (unsigned long)NULL;
11511 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
11512 .func = bpf_skc_to_tcp6_sock,
11514 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11515 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11516 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
11519 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
11521 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
11522 return (unsigned long)sk;
11524 return (unsigned long)NULL;
11527 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
11528 .func = bpf_skc_to_tcp_sock,
11530 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11531 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11532 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
11535 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
11537 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
11538 * generated if CONFIG_INET=n. Trigger an explicit generation here.
11540 BTF_TYPE_EMIT(struct inet_timewait_sock);
11541 BTF_TYPE_EMIT(struct tcp_timewait_sock);
11544 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
11545 return (unsigned long)sk;
11548 #if IS_BUILTIN(CONFIG_IPV6)
11549 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
11550 return (unsigned long)sk;
11553 return (unsigned long)NULL;
11556 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
11557 .func = bpf_skc_to_tcp_timewait_sock,
11559 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11560 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11561 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
11564 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
11567 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11568 return (unsigned long)sk;
11571 #if IS_BUILTIN(CONFIG_IPV6)
11572 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11573 return (unsigned long)sk;
11576 return (unsigned long)NULL;
11579 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
11580 .func = bpf_skc_to_tcp_request_sock,
11582 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11583 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11584 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
11587 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
11589 /* udp6_sock type is not generated in dwarf and hence btf,
11590 * trigger an explicit type generation here.
11592 BTF_TYPE_EMIT(struct udp6_sock);
11593 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
11594 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
11595 return (unsigned long)sk;
11597 return (unsigned long)NULL;
11600 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
11601 .func = bpf_skc_to_udp6_sock,
11603 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11604 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11605 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
11608 BPF_CALL_1(bpf_skc_to_unix_sock, struct sock *, sk)
11610 /* unix_sock type is not generated in dwarf and hence btf,
11611 * trigger an explicit type generation here.
11613 BTF_TYPE_EMIT(struct unix_sock);
11614 if (sk && sk_fullsock(sk) && sk->sk_family == AF_UNIX)
11615 return (unsigned long)sk;
11617 return (unsigned long)NULL;
11620 const struct bpf_func_proto bpf_skc_to_unix_sock_proto = {
11621 .func = bpf_skc_to_unix_sock,
11623 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11624 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11625 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UNIX],
11628 BPF_CALL_1(bpf_skc_to_mptcp_sock, struct sock *, sk)
11630 BTF_TYPE_EMIT(struct mptcp_sock);
11631 return (unsigned long)bpf_mptcp_sock_from_subflow(sk);
11634 const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto = {
11635 .func = bpf_skc_to_mptcp_sock,
11637 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11638 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
11639 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_MPTCP],
11642 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
11644 return (unsigned long)sock_from_file(file);
11647 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
11648 BTF_ID(struct, socket)
11649 BTF_ID(struct, file)
11651 const struct bpf_func_proto bpf_sock_from_file_proto = {
11652 .func = bpf_sock_from_file,
11654 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11655 .ret_btf_id = &bpf_sock_from_file_btf_ids[0],
11656 .arg1_type = ARG_PTR_TO_BTF_ID,
11657 .arg1_btf_id = &bpf_sock_from_file_btf_ids[1],
11660 static const struct bpf_func_proto *
11661 bpf_sk_base_func_proto(enum bpf_func_id func_id)
11663 const struct bpf_func_proto *func;
11666 case BPF_FUNC_skc_to_tcp6_sock:
11667 func = &bpf_skc_to_tcp6_sock_proto;
11669 case BPF_FUNC_skc_to_tcp_sock:
11670 func = &bpf_skc_to_tcp_sock_proto;
11672 case BPF_FUNC_skc_to_tcp_timewait_sock:
11673 func = &bpf_skc_to_tcp_timewait_sock_proto;
11675 case BPF_FUNC_skc_to_tcp_request_sock:
11676 func = &bpf_skc_to_tcp_request_sock_proto;
11678 case BPF_FUNC_skc_to_udp6_sock:
11679 func = &bpf_skc_to_udp6_sock_proto;
11681 case BPF_FUNC_skc_to_unix_sock:
11682 func = &bpf_skc_to_unix_sock_proto;
11684 case BPF_FUNC_skc_to_mptcp_sock:
11685 func = &bpf_skc_to_mptcp_sock_proto;
11687 case BPF_FUNC_ktime_get_coarse_ns:
11688 return &bpf_ktime_get_coarse_ns_proto;
11690 return bpf_base_func_proto(func_id);
11693 if (!perfmon_capable())