2 * Linux Socket Filter - Kernel level socket filtering
4 * Based on the design of the Berkeley Packet Filter. The new
5 * internal format has been designed by PLUMgrid:
7 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
11 * Jay Schulist <jschlst@samba.org>
12 * Alexei Starovoitov <ast@plumgrid.com>
13 * Daniel Borkmann <dborkman@redhat.com>
15 * This program is free software; you can redistribute it and/or
16 * modify it under the terms of the GNU General Public License
17 * as published by the Free Software Foundation; either version
18 * 2 of the License, or (at your option) any later version.
20 * Andi Kleen - Fix a few bad bugs and races.
21 * Kris Katterjohn - Added many additional checks in bpf_check_classic()
24 #include <linux/module.h>
25 #include <linux/types.h>
27 #include <linux/fcntl.h>
28 #include <linux/socket.h>
29 #include <linux/sock_diag.h>
31 #include <linux/inet.h>
32 #include <linux/netdevice.h>
33 #include <linux/if_packet.h>
34 #include <linux/if_arp.h>
35 #include <linux/gfp.h>
37 #include <net/protocol.h>
38 #include <net/netlink.h>
39 #include <linux/skbuff.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 <net/sch_generic.h>
52 #include <net/cls_cgroup.h>
53 #include <net/dst_metadata.h>
55 #include <net/sock_reuseport.h>
56 #include <net/busy_poll.h>
58 #include <linux/bpf_trace.h>
61 * sk_filter_trim_cap - run a packet through a socket filter
62 * @sk: sock associated with &sk_buff
63 * @skb: buffer to filter
64 * @cap: limit on how short the eBPF program may trim the packet
66 * Run the eBPF program and then cut skb->data to correct size returned by
67 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
68 * than pkt_len we keep whole skb->data. This is the socket level
69 * wrapper to BPF_PROG_RUN. It returns 0 if the packet should
70 * be accepted or -EPERM if the packet should be tossed.
73 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
76 struct sk_filter *filter;
79 * If the skb was allocated from pfmemalloc reserves, only
80 * allow SOCK_MEMALLOC sockets to use it as this socket is
83 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
84 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
87 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
91 err = security_sock_rcv_skb(sk, skb);
96 filter = rcu_dereference(sk->sk_filter);
98 struct sock *save_sk = skb->sk;
102 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
104 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
110 EXPORT_SYMBOL(sk_filter_trim_cap);
112 BPF_CALL_1(__skb_get_pay_offset, struct sk_buff *, skb)
114 return skb_get_poff(skb);
117 BPF_CALL_3(__skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
121 if (skb_is_nonlinear(skb))
124 if (skb->len < sizeof(struct nlattr))
127 if (a > skb->len - sizeof(struct nlattr))
130 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
132 return (void *) nla - (void *) skb->data;
137 BPF_CALL_3(__skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
141 if (skb_is_nonlinear(skb))
144 if (skb->len < sizeof(struct nlattr))
147 if (a > skb->len - sizeof(struct nlattr))
150 nla = (struct nlattr *) &skb->data[a];
151 if (nla->nla_len > skb->len - a)
154 nla = nla_find_nested(nla, x);
156 return (void *) nla - (void *) skb->data;
161 BPF_CALL_0(__get_raw_cpu_id)
163 return raw_smp_processor_id();
166 static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = {
167 .func = __get_raw_cpu_id,
169 .ret_type = RET_INTEGER,
172 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
173 struct bpf_insn *insn_buf)
175 struct bpf_insn *insn = insn_buf;
179 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
181 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
182 offsetof(struct sk_buff, mark));
186 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
187 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
188 #ifdef __BIG_ENDIAN_BITFIELD
189 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
194 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2);
196 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
197 offsetof(struct sk_buff, queue_mapping));
200 case SKF_AD_VLAN_TAG:
201 case SKF_AD_VLAN_TAG_PRESENT:
202 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2);
203 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000);
205 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
206 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
207 offsetof(struct sk_buff, vlan_tci));
208 if (skb_field == SKF_AD_VLAN_TAG) {
209 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg,
213 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 12);
215 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
220 return insn - insn_buf;
223 static bool convert_bpf_extensions(struct sock_filter *fp,
224 struct bpf_insn **insnp)
226 struct bpf_insn *insn = *insnp;
230 case SKF_AD_OFF + SKF_AD_PROTOCOL:
231 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
233 /* A = *(u16 *) (CTX + offsetof(protocol)) */
234 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
235 offsetof(struct sk_buff, protocol));
236 /* A = ntohs(A) [emitting a nop or swap16] */
237 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
240 case SKF_AD_OFF + SKF_AD_PKTTYPE:
241 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
245 case SKF_AD_OFF + SKF_AD_IFINDEX:
246 case SKF_AD_OFF + SKF_AD_HATYPE:
247 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
248 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, type) != 2);
250 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
251 BPF_REG_TMP, BPF_REG_CTX,
252 offsetof(struct sk_buff, dev));
253 /* if (tmp != 0) goto pc + 1 */
254 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
255 *insn++ = BPF_EXIT_INSN();
256 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
257 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
258 offsetof(struct net_device, ifindex));
260 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
261 offsetof(struct net_device, type));
264 case SKF_AD_OFF + SKF_AD_MARK:
265 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
269 case SKF_AD_OFF + SKF_AD_RXHASH:
270 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
272 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
273 offsetof(struct sk_buff, hash));
276 case SKF_AD_OFF + SKF_AD_QUEUE:
277 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
281 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
282 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
283 BPF_REG_A, BPF_REG_CTX, insn);
287 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
288 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
289 BPF_REG_A, BPF_REG_CTX, insn);
293 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
294 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2);
296 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
297 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
298 offsetof(struct sk_buff, vlan_proto));
299 /* A = ntohs(A) [emitting a nop or swap16] */
300 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
303 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
304 case SKF_AD_OFF + SKF_AD_NLATTR:
305 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
306 case SKF_AD_OFF + SKF_AD_CPU:
307 case SKF_AD_OFF + SKF_AD_RANDOM:
309 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
311 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
313 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
314 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
316 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
317 *insn = BPF_EMIT_CALL(__skb_get_pay_offset);
319 case SKF_AD_OFF + SKF_AD_NLATTR:
320 *insn = BPF_EMIT_CALL(__skb_get_nlattr);
322 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
323 *insn = BPF_EMIT_CALL(__skb_get_nlattr_nest);
325 case SKF_AD_OFF + SKF_AD_CPU:
326 *insn = BPF_EMIT_CALL(__get_raw_cpu_id);
328 case SKF_AD_OFF + SKF_AD_RANDOM:
329 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
330 bpf_user_rnd_init_once();
335 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
337 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
341 /* This is just a dummy call to avoid letting the compiler
342 * evict __bpf_call_base() as an optimization. Placed here
343 * where no-one bothers.
345 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
354 * bpf_convert_filter - convert filter program
355 * @prog: the user passed filter program
356 * @len: the length of the user passed filter program
357 * @new_prog: allocated 'struct bpf_prog' or NULL
358 * @new_len: pointer to store length of converted program
360 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
361 * style extended BPF (eBPF).
362 * Conversion workflow:
364 * 1) First pass for calculating the new program length:
365 * bpf_convert_filter(old_prog, old_len, NULL, &new_len)
367 * 2) 2nd pass to remap in two passes: 1st pass finds new
368 * jump offsets, 2nd pass remapping:
369 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len);
371 static int bpf_convert_filter(struct sock_filter *prog, int len,
372 struct bpf_prog *new_prog, int *new_len)
374 int new_flen = 0, pass = 0, target, i, stack_off;
375 struct bpf_insn *new_insn, *first_insn = NULL;
376 struct sock_filter *fp;
380 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
381 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
383 if (len <= 0 || len > BPF_MAXINSNS)
387 first_insn = new_prog->insnsi;
388 addrs = kcalloc(len, sizeof(*addrs),
389 GFP_KERNEL | __GFP_NOWARN);
395 new_insn = first_insn;
398 /* Classic BPF related prologue emission. */
400 /* Classic BPF expects A and X to be reset first. These need
401 * to be guaranteed to be the first two instructions.
403 *new_insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
404 *new_insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
406 /* All programs must keep CTX in callee saved BPF_REG_CTX.
407 * In eBPF case it's done by the compiler, here we need to
408 * do this ourself. Initial CTX is present in BPF_REG_ARG1.
410 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
415 for (i = 0; i < len; fp++, i++) {
416 struct bpf_insn tmp_insns[6] = { };
417 struct bpf_insn *insn = tmp_insns;
420 addrs[i] = new_insn - first_insn;
423 /* All arithmetic insns and skb loads map as-is. */
424 case BPF_ALU | BPF_ADD | BPF_X:
425 case BPF_ALU | BPF_ADD | BPF_K:
426 case BPF_ALU | BPF_SUB | BPF_X:
427 case BPF_ALU | BPF_SUB | BPF_K:
428 case BPF_ALU | BPF_AND | BPF_X:
429 case BPF_ALU | BPF_AND | BPF_K:
430 case BPF_ALU | BPF_OR | BPF_X:
431 case BPF_ALU | BPF_OR | BPF_K:
432 case BPF_ALU | BPF_LSH | BPF_X:
433 case BPF_ALU | BPF_LSH | BPF_K:
434 case BPF_ALU | BPF_RSH | BPF_X:
435 case BPF_ALU | BPF_RSH | BPF_K:
436 case BPF_ALU | BPF_XOR | BPF_X:
437 case BPF_ALU | BPF_XOR | BPF_K:
438 case BPF_ALU | BPF_MUL | BPF_X:
439 case BPF_ALU | BPF_MUL | BPF_K:
440 case BPF_ALU | BPF_DIV | BPF_X:
441 case BPF_ALU | BPF_DIV | BPF_K:
442 case BPF_ALU | BPF_MOD | BPF_X:
443 case BPF_ALU | BPF_MOD | BPF_K:
444 case BPF_ALU | BPF_NEG:
445 case BPF_LD | BPF_ABS | BPF_W:
446 case BPF_LD | BPF_ABS | BPF_H:
447 case BPF_LD | BPF_ABS | BPF_B:
448 case BPF_LD | BPF_IND | BPF_W:
449 case BPF_LD | BPF_IND | BPF_H:
450 case BPF_LD | BPF_IND | BPF_B:
451 /* Check for overloaded BPF extension and
452 * directly convert it if found, otherwise
453 * just move on with mapping.
455 if (BPF_CLASS(fp->code) == BPF_LD &&
456 BPF_MODE(fp->code) == BPF_ABS &&
457 convert_bpf_extensions(fp, &insn))
460 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
463 /* Jump transformation cannot use BPF block macros
464 * everywhere as offset calculation and target updates
465 * require a bit more work than the rest, i.e. jump
466 * opcodes map as-is, but offsets need adjustment.
469 #define BPF_EMIT_JMP \
471 if (target >= len || target < 0) \
473 insn->off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
474 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
475 insn->off -= insn - tmp_insns; \
478 case BPF_JMP | BPF_JA:
479 target = i + fp->k + 1;
480 insn->code = fp->code;
484 case BPF_JMP | BPF_JEQ | BPF_K:
485 case BPF_JMP | BPF_JEQ | BPF_X:
486 case BPF_JMP | BPF_JSET | BPF_K:
487 case BPF_JMP | BPF_JSET | BPF_X:
488 case BPF_JMP | BPF_JGT | BPF_K:
489 case BPF_JMP | BPF_JGT | BPF_X:
490 case BPF_JMP | BPF_JGE | BPF_K:
491 case BPF_JMP | BPF_JGE | BPF_X:
492 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
493 /* BPF immediates are signed, zero extend
494 * immediate into tmp register and use it
497 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
499 insn->dst_reg = BPF_REG_A;
500 insn->src_reg = BPF_REG_TMP;
503 insn->dst_reg = BPF_REG_A;
505 bpf_src = BPF_SRC(fp->code);
506 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
509 /* Common case where 'jump_false' is next insn. */
511 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
512 target = i + fp->jt + 1;
517 /* Convert some jumps when 'jump_true' is next insn. */
519 switch (BPF_OP(fp->code)) {
521 insn->code = BPF_JMP | BPF_JNE | bpf_src;
524 insn->code = BPF_JMP | BPF_JLE | bpf_src;
527 insn->code = BPF_JMP | BPF_JLT | bpf_src;
533 target = i + fp->jf + 1;
538 /* Other jumps are mapped into two insns: Jxx and JA. */
539 target = i + fp->jt + 1;
540 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
544 insn->code = BPF_JMP | BPF_JA;
545 target = i + fp->jf + 1;
549 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
550 case BPF_LDX | BPF_MSH | BPF_B:
552 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_A);
553 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
554 *insn++ = BPF_LD_ABS(BPF_B, fp->k);
556 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
558 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
560 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
562 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
565 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
566 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
568 case BPF_RET | BPF_A:
569 case BPF_RET | BPF_K:
570 if (BPF_RVAL(fp->code) == BPF_K)
571 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
573 *insn = BPF_EXIT_INSN();
576 /* Store to stack. */
579 stack_off = fp->k * 4 + 4;
580 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
581 BPF_ST ? BPF_REG_A : BPF_REG_X,
583 /* check_load_and_stores() verifies that classic BPF can
584 * load from stack only after write, so tracking
585 * stack_depth for ST|STX insns is enough
587 if (new_prog && new_prog->aux->stack_depth < stack_off)
588 new_prog->aux->stack_depth = stack_off;
591 /* Load from stack. */
592 case BPF_LD | BPF_MEM:
593 case BPF_LDX | BPF_MEM:
594 stack_off = fp->k * 4 + 4;
595 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
596 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
601 case BPF_LD | BPF_IMM:
602 case BPF_LDX | BPF_IMM:
603 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
604 BPF_REG_A : BPF_REG_X, fp->k);
608 case BPF_MISC | BPF_TAX:
609 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
613 case BPF_MISC | BPF_TXA:
614 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
617 /* A = skb->len or X = skb->len */
618 case BPF_LD | BPF_W | BPF_LEN:
619 case BPF_LDX | BPF_W | BPF_LEN:
620 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
621 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
622 offsetof(struct sk_buff, len));
625 /* Access seccomp_data fields. */
626 case BPF_LDX | BPF_ABS | BPF_W:
627 /* A = *(u32 *) (ctx + K) */
628 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
631 /* Unknown instruction. */
638 memcpy(new_insn, tmp_insns,
639 sizeof(*insn) * (insn - tmp_insns));
640 new_insn += insn - tmp_insns;
644 /* Only calculating new length. */
645 *new_len = new_insn - first_insn;
650 if (new_flen != new_insn - first_insn) {
651 new_flen = new_insn - first_insn;
658 BUG_ON(*new_len != new_flen);
667 * As we dont want to clear mem[] array for each packet going through
668 * __bpf_prog_run(), we check that filter loaded by user never try to read
669 * a cell if not previously written, and we check all branches to be sure
670 * a malicious user doesn't try to abuse us.
672 static int check_load_and_stores(const struct sock_filter *filter, int flen)
674 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
677 BUILD_BUG_ON(BPF_MEMWORDS > 16);
679 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
683 memset(masks, 0xff, flen * sizeof(*masks));
685 for (pc = 0; pc < flen; pc++) {
686 memvalid &= masks[pc];
688 switch (filter[pc].code) {
691 memvalid |= (1 << filter[pc].k);
693 case BPF_LD | BPF_MEM:
694 case BPF_LDX | BPF_MEM:
695 if (!(memvalid & (1 << filter[pc].k))) {
700 case BPF_JMP | BPF_JA:
701 /* A jump must set masks on target */
702 masks[pc + 1 + filter[pc].k] &= memvalid;
705 case BPF_JMP | BPF_JEQ | BPF_K:
706 case BPF_JMP | BPF_JEQ | BPF_X:
707 case BPF_JMP | BPF_JGE | BPF_K:
708 case BPF_JMP | BPF_JGE | BPF_X:
709 case BPF_JMP | BPF_JGT | BPF_K:
710 case BPF_JMP | BPF_JGT | BPF_X:
711 case BPF_JMP | BPF_JSET | BPF_K:
712 case BPF_JMP | BPF_JSET | BPF_X:
713 /* A jump must set masks on targets */
714 masks[pc + 1 + filter[pc].jt] &= memvalid;
715 masks[pc + 1 + filter[pc].jf] &= memvalid;
725 static bool chk_code_allowed(u16 code_to_probe)
727 static const bool codes[] = {
728 /* 32 bit ALU operations */
729 [BPF_ALU | BPF_ADD | BPF_K] = true,
730 [BPF_ALU | BPF_ADD | BPF_X] = true,
731 [BPF_ALU | BPF_SUB | BPF_K] = true,
732 [BPF_ALU | BPF_SUB | BPF_X] = true,
733 [BPF_ALU | BPF_MUL | BPF_K] = true,
734 [BPF_ALU | BPF_MUL | BPF_X] = true,
735 [BPF_ALU | BPF_DIV | BPF_K] = true,
736 [BPF_ALU | BPF_DIV | BPF_X] = true,
737 [BPF_ALU | BPF_MOD | BPF_K] = true,
738 [BPF_ALU | BPF_MOD | BPF_X] = true,
739 [BPF_ALU | BPF_AND | BPF_K] = true,
740 [BPF_ALU | BPF_AND | BPF_X] = true,
741 [BPF_ALU | BPF_OR | BPF_K] = true,
742 [BPF_ALU | BPF_OR | BPF_X] = true,
743 [BPF_ALU | BPF_XOR | BPF_K] = true,
744 [BPF_ALU | BPF_XOR | BPF_X] = true,
745 [BPF_ALU | BPF_LSH | BPF_K] = true,
746 [BPF_ALU | BPF_LSH | BPF_X] = true,
747 [BPF_ALU | BPF_RSH | BPF_K] = true,
748 [BPF_ALU | BPF_RSH | BPF_X] = true,
749 [BPF_ALU | BPF_NEG] = true,
750 /* Load instructions */
751 [BPF_LD | BPF_W | BPF_ABS] = true,
752 [BPF_LD | BPF_H | BPF_ABS] = true,
753 [BPF_LD | BPF_B | BPF_ABS] = true,
754 [BPF_LD | BPF_W | BPF_LEN] = true,
755 [BPF_LD | BPF_W | BPF_IND] = true,
756 [BPF_LD | BPF_H | BPF_IND] = true,
757 [BPF_LD | BPF_B | BPF_IND] = true,
758 [BPF_LD | BPF_IMM] = true,
759 [BPF_LD | BPF_MEM] = true,
760 [BPF_LDX | BPF_W | BPF_LEN] = true,
761 [BPF_LDX | BPF_B | BPF_MSH] = true,
762 [BPF_LDX | BPF_IMM] = true,
763 [BPF_LDX | BPF_MEM] = true,
764 /* Store instructions */
767 /* Misc instructions */
768 [BPF_MISC | BPF_TAX] = true,
769 [BPF_MISC | BPF_TXA] = true,
770 /* Return instructions */
771 [BPF_RET | BPF_K] = true,
772 [BPF_RET | BPF_A] = true,
773 /* Jump instructions */
774 [BPF_JMP | BPF_JA] = true,
775 [BPF_JMP | BPF_JEQ | BPF_K] = true,
776 [BPF_JMP | BPF_JEQ | BPF_X] = true,
777 [BPF_JMP | BPF_JGE | BPF_K] = true,
778 [BPF_JMP | BPF_JGE | BPF_X] = true,
779 [BPF_JMP | BPF_JGT | BPF_K] = true,
780 [BPF_JMP | BPF_JGT | BPF_X] = true,
781 [BPF_JMP | BPF_JSET | BPF_K] = true,
782 [BPF_JMP | BPF_JSET | BPF_X] = true,
785 if (code_to_probe >= ARRAY_SIZE(codes))
788 return codes[code_to_probe];
791 static bool bpf_check_basics_ok(const struct sock_filter *filter,
796 if (flen == 0 || flen > BPF_MAXINSNS)
803 * bpf_check_classic - verify socket filter code
804 * @filter: filter to verify
805 * @flen: length of filter
807 * Check the user's filter code. If we let some ugly
808 * filter code slip through kaboom! The filter must contain
809 * no references or jumps that are out of range, no illegal
810 * instructions, and must end with a RET instruction.
812 * All jumps are forward as they are not signed.
814 * Returns 0 if the rule set is legal or -EINVAL if not.
816 static int bpf_check_classic(const struct sock_filter *filter,
822 /* Check the filter code now */
823 for (pc = 0; pc < flen; pc++) {
824 const struct sock_filter *ftest = &filter[pc];
826 /* May we actually operate on this code? */
827 if (!chk_code_allowed(ftest->code))
830 /* Some instructions need special checks */
831 switch (ftest->code) {
832 case BPF_ALU | BPF_DIV | BPF_K:
833 case BPF_ALU | BPF_MOD | BPF_K:
834 /* Check for division by zero */
838 case BPF_ALU | BPF_LSH | BPF_K:
839 case BPF_ALU | BPF_RSH | BPF_K:
843 case BPF_LD | BPF_MEM:
844 case BPF_LDX | BPF_MEM:
847 /* Check for invalid memory addresses */
848 if (ftest->k >= BPF_MEMWORDS)
851 case BPF_JMP | BPF_JA:
852 /* Note, the large ftest->k might cause loops.
853 * Compare this with conditional jumps below,
854 * where offsets are limited. --ANK (981016)
856 if (ftest->k >= (unsigned int)(flen - pc - 1))
859 case BPF_JMP | BPF_JEQ | BPF_K:
860 case BPF_JMP | BPF_JEQ | BPF_X:
861 case BPF_JMP | BPF_JGE | BPF_K:
862 case BPF_JMP | BPF_JGE | BPF_X:
863 case BPF_JMP | BPF_JGT | BPF_K:
864 case BPF_JMP | BPF_JGT | BPF_X:
865 case BPF_JMP | BPF_JSET | BPF_K:
866 case BPF_JMP | BPF_JSET | BPF_X:
867 /* Both conditionals must be safe */
868 if (pc + ftest->jt + 1 >= flen ||
869 pc + ftest->jf + 1 >= flen)
872 case BPF_LD | BPF_W | BPF_ABS:
873 case BPF_LD | BPF_H | BPF_ABS:
874 case BPF_LD | BPF_B | BPF_ABS:
876 if (bpf_anc_helper(ftest) & BPF_ANC)
878 /* Ancillary operation unknown or unsupported */
879 if (anc_found == false && ftest->k >= SKF_AD_OFF)
884 /* Last instruction must be a RET code */
885 switch (filter[flen - 1].code) {
886 case BPF_RET | BPF_K:
887 case BPF_RET | BPF_A:
888 return check_load_and_stores(filter, flen);
894 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
895 const struct sock_fprog *fprog)
897 unsigned int fsize = bpf_classic_proglen(fprog);
898 struct sock_fprog_kern *fkprog;
900 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
904 fkprog = fp->orig_prog;
905 fkprog->len = fprog->len;
907 fkprog->filter = kmemdup(fp->insns, fsize,
908 GFP_KERNEL | __GFP_NOWARN);
909 if (!fkprog->filter) {
910 kfree(fp->orig_prog);
917 static void bpf_release_orig_filter(struct bpf_prog *fp)
919 struct sock_fprog_kern *fprog = fp->orig_prog;
922 kfree(fprog->filter);
927 static void __bpf_prog_release(struct bpf_prog *prog)
929 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
932 bpf_release_orig_filter(prog);
937 static void __sk_filter_release(struct sk_filter *fp)
939 __bpf_prog_release(fp->prog);
944 * sk_filter_release_rcu - Release a socket filter by rcu_head
945 * @rcu: rcu_head that contains the sk_filter to free
947 static void sk_filter_release_rcu(struct rcu_head *rcu)
949 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
951 __sk_filter_release(fp);
955 * sk_filter_release - release a socket filter
956 * @fp: filter to remove
958 * Remove a filter from a socket and release its resources.
960 static void sk_filter_release(struct sk_filter *fp)
962 if (refcount_dec_and_test(&fp->refcnt))
963 call_rcu(&fp->rcu, sk_filter_release_rcu);
966 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
968 u32 filter_size = bpf_prog_size(fp->prog->len);
970 atomic_sub(filter_size, &sk->sk_omem_alloc);
971 sk_filter_release(fp);
974 /* try to charge the socket memory if there is space available
975 * return true on success
977 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
979 u32 filter_size = bpf_prog_size(fp->prog->len);
981 /* same check as in sock_kmalloc() */
982 if (filter_size <= sysctl_optmem_max &&
983 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
984 atomic_add(filter_size, &sk->sk_omem_alloc);
990 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
992 if (!refcount_inc_not_zero(&fp->refcnt))
995 if (!__sk_filter_charge(sk, fp)) {
996 sk_filter_release(fp);
1002 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1004 struct sock_filter *old_prog;
1005 struct bpf_prog *old_fp;
1006 int err, new_len, old_len = fp->len;
1008 /* We are free to overwrite insns et al right here as it
1009 * won't be used at this point in time anymore internally
1010 * after the migration to the internal BPF instruction
1013 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1014 sizeof(struct bpf_insn));
1016 /* Conversion cannot happen on overlapping memory areas,
1017 * so we need to keep the user BPF around until the 2nd
1018 * pass. At this time, the user BPF is stored in fp->insns.
1020 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1021 GFP_KERNEL | __GFP_NOWARN);
1027 /* 1st pass: calculate the new program length. */
1028 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len);
1032 /* Expand fp for appending the new filter representation. */
1034 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1036 /* The old_fp is still around in case we couldn't
1037 * allocate new memory, so uncharge on that one.
1046 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1047 err = bpf_convert_filter(old_prog, old_len, fp, &new_len);
1049 /* 2nd bpf_convert_filter() can fail only if it fails
1050 * to allocate memory, remapping must succeed. Note,
1051 * that at this time old_fp has already been released
1056 /* We are guaranteed to never error here with cBPF to eBPF
1057 * transitions, since there's no issue with type compatibility
1058 * checks on program arrays.
1060 fp = bpf_prog_select_runtime(fp, &err);
1068 __bpf_prog_release(fp);
1069 return ERR_PTR(err);
1072 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1073 bpf_aux_classic_check_t trans)
1077 fp->bpf_func = NULL;
1080 err = bpf_check_classic(fp->insns, fp->len);
1082 __bpf_prog_release(fp);
1083 return ERR_PTR(err);
1086 /* There might be additional checks and transformations
1087 * needed on classic filters, f.e. in case of seccomp.
1090 err = trans(fp->insns, fp->len);
1092 __bpf_prog_release(fp);
1093 return ERR_PTR(err);
1097 /* Probe if we can JIT compile the filter and if so, do
1098 * the compilation of the filter.
1100 bpf_jit_compile(fp);
1102 /* JIT compiler couldn't process this filter, so do the
1103 * internal BPF translation for the optimized interpreter.
1106 fp = bpf_migrate_filter(fp);
1112 * bpf_prog_create - create an unattached filter
1113 * @pfp: the unattached filter that is created
1114 * @fprog: the filter program
1116 * Create a filter independent of any socket. We first run some
1117 * sanity checks on it to make sure it does not explode on us later.
1118 * If an error occurs or there is insufficient memory for the filter
1119 * a negative errno code is returned. On success the return is zero.
1121 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1123 unsigned int fsize = bpf_classic_proglen(fprog);
1124 struct bpf_prog *fp;
1126 /* Make sure new filter is there and in the right amounts. */
1127 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1130 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1134 memcpy(fp->insns, fprog->filter, fsize);
1136 fp->len = fprog->len;
1137 /* Since unattached filters are not copied back to user
1138 * space through sk_get_filter(), we do not need to hold
1139 * a copy here, and can spare us the work.
1141 fp->orig_prog = NULL;
1143 /* bpf_prepare_filter() already takes care of freeing
1144 * memory in case something goes wrong.
1146 fp = bpf_prepare_filter(fp, NULL);
1153 EXPORT_SYMBOL_GPL(bpf_prog_create);
1156 * bpf_prog_create_from_user - create an unattached filter from user buffer
1157 * @pfp: the unattached filter that is created
1158 * @fprog: the filter program
1159 * @trans: post-classic verifier transformation handler
1160 * @save_orig: save classic BPF program
1162 * This function effectively does the same as bpf_prog_create(), only
1163 * that it builds up its insns buffer from user space provided buffer.
1164 * It also allows for passing a bpf_aux_classic_check_t handler.
1166 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1167 bpf_aux_classic_check_t trans, bool save_orig)
1169 unsigned int fsize = bpf_classic_proglen(fprog);
1170 struct bpf_prog *fp;
1173 /* Make sure new filter is there and in the right amounts. */
1174 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1177 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1181 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1182 __bpf_prog_free(fp);
1186 fp->len = fprog->len;
1187 fp->orig_prog = NULL;
1190 err = bpf_prog_store_orig_filter(fp, fprog);
1192 __bpf_prog_free(fp);
1197 /* bpf_prepare_filter() already takes care of freeing
1198 * memory in case something goes wrong.
1200 fp = bpf_prepare_filter(fp, trans);
1207 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1209 void bpf_prog_destroy(struct bpf_prog *fp)
1211 __bpf_prog_release(fp);
1213 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1215 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1217 struct sk_filter *fp, *old_fp;
1219 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1225 if (!__sk_filter_charge(sk, fp)) {
1229 refcount_set(&fp->refcnt, 1);
1231 old_fp = rcu_dereference_protected(sk->sk_filter,
1232 lockdep_sock_is_held(sk));
1233 rcu_assign_pointer(sk->sk_filter, fp);
1236 sk_filter_uncharge(sk, old_fp);
1241 static int __reuseport_attach_prog(struct bpf_prog *prog, struct sock *sk)
1243 struct bpf_prog *old_prog;
1246 if (bpf_prog_size(prog->len) > sysctl_optmem_max)
1249 if (sk_unhashed(sk) && sk->sk_reuseport) {
1250 err = reuseport_alloc(sk);
1253 } else if (!rcu_access_pointer(sk->sk_reuseport_cb)) {
1254 /* The socket wasn't bound with SO_REUSEPORT */
1258 old_prog = reuseport_attach_prog(sk, prog);
1260 bpf_prog_destroy(old_prog);
1266 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1268 unsigned int fsize = bpf_classic_proglen(fprog);
1269 struct bpf_prog *prog;
1272 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1273 return ERR_PTR(-EPERM);
1275 /* Make sure new filter is there and in the right amounts. */
1276 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1277 return ERR_PTR(-EINVAL);
1279 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1281 return ERR_PTR(-ENOMEM);
1283 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1284 __bpf_prog_free(prog);
1285 return ERR_PTR(-EFAULT);
1288 prog->len = fprog->len;
1290 err = bpf_prog_store_orig_filter(prog, fprog);
1292 __bpf_prog_free(prog);
1293 return ERR_PTR(-ENOMEM);
1296 /* bpf_prepare_filter() already takes care of freeing
1297 * memory in case something goes wrong.
1299 return bpf_prepare_filter(prog, NULL);
1303 * sk_attach_filter - attach a socket filter
1304 * @fprog: the filter program
1305 * @sk: the socket to use
1307 * Attach the user's filter code. We first run some sanity checks on
1308 * it to make sure it does not explode on us later. If an error
1309 * occurs or there is insufficient memory for the filter a negative
1310 * errno code is returned. On success the return is zero.
1312 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1314 struct bpf_prog *prog = __get_filter(fprog, sk);
1318 return PTR_ERR(prog);
1320 err = __sk_attach_prog(prog, sk);
1322 __bpf_prog_release(prog);
1328 EXPORT_SYMBOL_GPL(sk_attach_filter);
1330 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1332 struct bpf_prog *prog = __get_filter(fprog, sk);
1336 return PTR_ERR(prog);
1338 err = __reuseport_attach_prog(prog, sk);
1340 __bpf_prog_release(prog);
1347 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1349 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1350 return ERR_PTR(-EPERM);
1352 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1355 int sk_attach_bpf(u32 ufd, struct sock *sk)
1357 struct bpf_prog *prog = __get_bpf(ufd, sk);
1361 return PTR_ERR(prog);
1363 err = __sk_attach_prog(prog, sk);
1372 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1374 struct bpf_prog *prog = __get_bpf(ufd, sk);
1378 return PTR_ERR(prog);
1380 err = __reuseport_attach_prog(prog, sk);
1389 struct bpf_scratchpad {
1391 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1392 u8 buff[MAX_BPF_STACK];
1396 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1398 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1399 unsigned int write_len)
1401 return skb_ensure_writable(skb, write_len);
1404 static inline int bpf_try_make_writable(struct sk_buff *skb,
1405 unsigned int write_len)
1407 int err = __bpf_try_make_writable(skb, write_len);
1409 bpf_compute_data_pointers(skb);
1413 static int bpf_try_make_head_writable(struct sk_buff *skb)
1415 return bpf_try_make_writable(skb, skb_headlen(skb));
1418 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1420 if (skb_at_tc_ingress(skb))
1421 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1424 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1426 if (skb_at_tc_ingress(skb))
1427 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1430 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1431 const void *, from, u32, len, u64, flags)
1435 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1437 if (unlikely(offset > 0xffff))
1439 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1442 ptr = skb->data + offset;
1443 if (flags & BPF_F_RECOMPUTE_CSUM)
1444 __skb_postpull_rcsum(skb, ptr, len, offset);
1446 memcpy(ptr, from, len);
1448 if (flags & BPF_F_RECOMPUTE_CSUM)
1449 __skb_postpush_rcsum(skb, ptr, len, offset);
1450 if (flags & BPF_F_INVALIDATE_HASH)
1451 skb_clear_hash(skb);
1456 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1457 .func = bpf_skb_store_bytes,
1459 .ret_type = RET_INTEGER,
1460 .arg1_type = ARG_PTR_TO_CTX,
1461 .arg2_type = ARG_ANYTHING,
1462 .arg3_type = ARG_PTR_TO_MEM,
1463 .arg4_type = ARG_CONST_SIZE,
1464 .arg5_type = ARG_ANYTHING,
1467 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1468 void *, to, u32, len)
1472 if (unlikely(offset > 0xffff))
1475 ptr = skb_header_pointer(skb, offset, len, to);
1479 memcpy(to, ptr, len);
1487 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1488 .func = bpf_skb_load_bytes,
1490 .ret_type = RET_INTEGER,
1491 .arg1_type = ARG_PTR_TO_CTX,
1492 .arg2_type = ARG_ANYTHING,
1493 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1494 .arg4_type = ARG_CONST_SIZE,
1497 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1499 /* Idea is the following: should the needed direct read/write
1500 * test fail during runtime, we can pull in more data and redo
1501 * again, since implicitly, we invalidate previous checks here.
1503 * Or, since we know how much we need to make read/writeable,
1504 * this can be done once at the program beginning for direct
1505 * access case. By this we overcome limitations of only current
1506 * headroom being accessible.
1508 return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1511 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1512 .func = bpf_skb_pull_data,
1514 .ret_type = RET_INTEGER,
1515 .arg1_type = ARG_PTR_TO_CTX,
1516 .arg2_type = ARG_ANYTHING,
1519 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1520 u64, from, u64, to, u64, flags)
1524 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1526 if (unlikely(offset > 0xffff || offset & 1))
1528 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1531 ptr = (__sum16 *)(skb->data + offset);
1532 switch (flags & BPF_F_HDR_FIELD_MASK) {
1534 if (unlikely(from != 0))
1537 csum_replace_by_diff(ptr, to);
1540 csum_replace2(ptr, from, to);
1543 csum_replace4(ptr, from, to);
1552 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1553 .func = bpf_l3_csum_replace,
1555 .ret_type = RET_INTEGER,
1556 .arg1_type = ARG_PTR_TO_CTX,
1557 .arg2_type = ARG_ANYTHING,
1558 .arg3_type = ARG_ANYTHING,
1559 .arg4_type = ARG_ANYTHING,
1560 .arg5_type = ARG_ANYTHING,
1563 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1564 u64, from, u64, to, u64, flags)
1566 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1567 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1568 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1571 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1572 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1574 if (unlikely(offset > 0xffff || offset & 1))
1576 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1579 ptr = (__sum16 *)(skb->data + offset);
1580 if (is_mmzero && !do_mforce && !*ptr)
1583 switch (flags & BPF_F_HDR_FIELD_MASK) {
1585 if (unlikely(from != 0))
1588 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1591 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1594 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1600 if (is_mmzero && !*ptr)
1601 *ptr = CSUM_MANGLED_0;
1605 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1606 .func = bpf_l4_csum_replace,
1608 .ret_type = RET_INTEGER,
1609 .arg1_type = ARG_PTR_TO_CTX,
1610 .arg2_type = ARG_ANYTHING,
1611 .arg3_type = ARG_ANYTHING,
1612 .arg4_type = ARG_ANYTHING,
1613 .arg5_type = ARG_ANYTHING,
1616 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1617 __be32 *, to, u32, to_size, __wsum, seed)
1619 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1620 u32 diff_size = from_size + to_size;
1623 /* This is quite flexible, some examples:
1625 * from_size == 0, to_size > 0, seed := csum --> pushing data
1626 * from_size > 0, to_size == 0, seed := csum --> pulling data
1627 * from_size > 0, to_size > 0, seed := 0 --> diffing data
1629 * Even for diffing, from_size and to_size don't need to be equal.
1631 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
1632 diff_size > sizeof(sp->diff)))
1635 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
1636 sp->diff[j] = ~from[i];
1637 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
1638 sp->diff[j] = to[i];
1640 return csum_partial(sp->diff, diff_size, seed);
1643 static const struct bpf_func_proto bpf_csum_diff_proto = {
1644 .func = bpf_csum_diff,
1647 .ret_type = RET_INTEGER,
1648 .arg1_type = ARG_PTR_TO_MEM,
1649 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
1650 .arg3_type = ARG_PTR_TO_MEM,
1651 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
1652 .arg5_type = ARG_ANYTHING,
1655 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
1657 /* The interface is to be used in combination with bpf_csum_diff()
1658 * for direct packet writes. csum rotation for alignment as well
1659 * as emulating csum_sub() can be done from the eBPF program.
1661 if (skb->ip_summed == CHECKSUM_COMPLETE)
1662 return (skb->csum = csum_add(skb->csum, csum));
1667 static const struct bpf_func_proto bpf_csum_update_proto = {
1668 .func = bpf_csum_update,
1670 .ret_type = RET_INTEGER,
1671 .arg1_type = ARG_PTR_TO_CTX,
1672 .arg2_type = ARG_ANYTHING,
1675 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
1677 return dev_forward_skb(dev, skb);
1680 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
1681 struct sk_buff *skb)
1683 int ret = ____dev_forward_skb(dev, skb);
1687 ret = netif_rx(skb);
1693 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
1697 if (unlikely(__this_cpu_read(xmit_recursion) > XMIT_RECURSION_LIMIT)) {
1698 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
1705 __this_cpu_inc(xmit_recursion);
1706 ret = dev_queue_xmit(skb);
1707 __this_cpu_dec(xmit_recursion);
1712 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
1715 /* skb->mac_len is not set on normal egress */
1716 unsigned int mlen = skb->network_header - skb->mac_header;
1718 __skb_pull(skb, mlen);
1720 /* At ingress, the mac header has already been pulled once.
1721 * At egress, skb_pospull_rcsum has to be done in case that
1722 * the skb is originated from ingress (i.e. a forwarded skb)
1723 * to ensure that rcsum starts at net header.
1725 if (!skb_at_tc_ingress(skb))
1726 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
1727 skb_pop_mac_header(skb);
1728 skb_reset_mac_len(skb);
1729 return flags & BPF_F_INGRESS ?
1730 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
1733 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
1736 /* Verify that a link layer header is carried */
1737 if (unlikely(skb->mac_header >= skb->network_header)) {
1742 bpf_push_mac_rcsum(skb);
1743 return flags & BPF_F_INGRESS ?
1744 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
1747 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
1750 if (dev_is_mac_header_xmit(dev))
1751 return __bpf_redirect_common(skb, dev, flags);
1753 return __bpf_redirect_no_mac(skb, dev, flags);
1756 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
1758 struct net_device *dev;
1759 struct sk_buff *clone;
1762 if (unlikely(flags & ~(BPF_F_INGRESS)))
1765 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
1769 clone = skb_clone(skb, GFP_ATOMIC);
1770 if (unlikely(!clone))
1773 /* For direct write, we need to keep the invariant that the skbs
1774 * we're dealing with need to be uncloned. Should uncloning fail
1775 * here, we need to free the just generated clone to unclone once
1778 ret = bpf_try_make_head_writable(skb);
1779 if (unlikely(ret)) {
1784 return __bpf_redirect(clone, dev, flags);
1787 static const struct bpf_func_proto bpf_clone_redirect_proto = {
1788 .func = bpf_clone_redirect,
1790 .ret_type = RET_INTEGER,
1791 .arg1_type = ARG_PTR_TO_CTX,
1792 .arg2_type = ARG_ANYTHING,
1793 .arg3_type = ARG_ANYTHING,
1796 struct redirect_info {
1799 struct bpf_map *map;
1800 struct bpf_map *map_to_flush;
1801 unsigned long map_owner;
1804 static DEFINE_PER_CPU(struct redirect_info, redirect_info);
1806 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
1808 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
1810 if (unlikely(flags & ~(BPF_F_INGRESS)))
1813 ri->ifindex = ifindex;
1816 return TC_ACT_REDIRECT;
1819 int skb_do_redirect(struct sk_buff *skb)
1821 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
1822 struct net_device *dev;
1824 dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->ifindex);
1826 if (unlikely(!dev)) {
1831 return __bpf_redirect(skb, dev, ri->flags);
1834 static const struct bpf_func_proto bpf_redirect_proto = {
1835 .func = bpf_redirect,
1837 .ret_type = RET_INTEGER,
1838 .arg1_type = ARG_ANYTHING,
1839 .arg2_type = ARG_ANYTHING,
1842 BPF_CALL_3(bpf_sk_redirect_map, struct bpf_map *, map, u32, key, u64, flags)
1844 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
1846 if (unlikely(flags))
1856 struct sock *do_sk_redirect_map(void)
1858 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
1859 struct sock *sk = NULL;
1862 sk = __sock_map_lookup_elem(ri->map, ri->ifindex);
1866 /* we do not clear flags for future lookup */
1872 static const struct bpf_func_proto bpf_sk_redirect_map_proto = {
1873 .func = bpf_sk_redirect_map,
1875 .ret_type = RET_INTEGER,
1876 .arg1_type = ARG_CONST_MAP_PTR,
1877 .arg2_type = ARG_ANYTHING,
1878 .arg3_type = ARG_ANYTHING,
1881 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
1883 return task_get_classid(skb);
1886 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
1887 .func = bpf_get_cgroup_classid,
1889 .ret_type = RET_INTEGER,
1890 .arg1_type = ARG_PTR_TO_CTX,
1893 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
1895 return dst_tclassid(skb);
1898 static const struct bpf_func_proto bpf_get_route_realm_proto = {
1899 .func = bpf_get_route_realm,
1901 .ret_type = RET_INTEGER,
1902 .arg1_type = ARG_PTR_TO_CTX,
1905 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
1907 /* If skb_clear_hash() was called due to mangling, we can
1908 * trigger SW recalculation here. Later access to hash
1909 * can then use the inline skb->hash via context directly
1910 * instead of calling this helper again.
1912 return skb_get_hash(skb);
1915 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
1916 .func = bpf_get_hash_recalc,
1918 .ret_type = RET_INTEGER,
1919 .arg1_type = ARG_PTR_TO_CTX,
1922 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
1924 /* After all direct packet write, this can be used once for
1925 * triggering a lazy recalc on next skb_get_hash() invocation.
1927 skb_clear_hash(skb);
1931 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
1932 .func = bpf_set_hash_invalid,
1934 .ret_type = RET_INTEGER,
1935 .arg1_type = ARG_PTR_TO_CTX,
1938 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
1940 /* Set user specified hash as L4(+), so that it gets returned
1941 * on skb_get_hash() call unless BPF prog later on triggers a
1944 __skb_set_sw_hash(skb, hash, true);
1948 static const struct bpf_func_proto bpf_set_hash_proto = {
1949 .func = bpf_set_hash,
1951 .ret_type = RET_INTEGER,
1952 .arg1_type = ARG_PTR_TO_CTX,
1953 .arg2_type = ARG_ANYTHING,
1956 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
1961 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
1962 vlan_proto != htons(ETH_P_8021AD)))
1963 vlan_proto = htons(ETH_P_8021Q);
1965 bpf_push_mac_rcsum(skb);
1966 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
1967 bpf_pull_mac_rcsum(skb);
1969 bpf_compute_data_pointers(skb);
1973 const struct bpf_func_proto bpf_skb_vlan_push_proto = {
1974 .func = bpf_skb_vlan_push,
1976 .ret_type = RET_INTEGER,
1977 .arg1_type = ARG_PTR_TO_CTX,
1978 .arg2_type = ARG_ANYTHING,
1979 .arg3_type = ARG_ANYTHING,
1981 EXPORT_SYMBOL_GPL(bpf_skb_vlan_push_proto);
1983 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
1987 bpf_push_mac_rcsum(skb);
1988 ret = skb_vlan_pop(skb);
1989 bpf_pull_mac_rcsum(skb);
1991 bpf_compute_data_pointers(skb);
1995 const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
1996 .func = bpf_skb_vlan_pop,
1998 .ret_type = RET_INTEGER,
1999 .arg1_type = ARG_PTR_TO_CTX,
2001 EXPORT_SYMBOL_GPL(bpf_skb_vlan_pop_proto);
2003 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
2005 /* Caller already did skb_cow() with len as headroom,
2006 * so no need to do it here.
2009 memmove(skb->data, skb->data + len, off);
2010 memset(skb->data + off, 0, len);
2012 /* No skb_postpush_rcsum(skb, skb->data + off, len)
2013 * needed here as it does not change the skb->csum
2014 * result for checksum complete when summing over
2020 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
2022 /* skb_ensure_writable() is not needed here, as we're
2023 * already working on an uncloned skb.
2025 if (unlikely(!pskb_may_pull(skb, off + len)))
2028 skb_postpull_rcsum(skb, skb->data + off, len);
2029 memmove(skb->data + len, skb->data, off);
2030 __skb_pull(skb, len);
2035 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
2037 bool trans_same = skb->transport_header == skb->network_header;
2040 /* There's no need for __skb_push()/__skb_pull() pair to
2041 * get to the start of the mac header as we're guaranteed
2042 * to always start from here under eBPF.
2044 ret = bpf_skb_generic_push(skb, off, len);
2046 skb->mac_header -= len;
2047 skb->network_header -= len;
2049 skb->transport_header = skb->network_header;
2055 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
2057 bool trans_same = skb->transport_header == skb->network_header;
2060 /* Same here, __skb_push()/__skb_pull() pair not needed. */
2061 ret = bpf_skb_generic_pop(skb, off, len);
2063 skb->mac_header += len;
2064 skb->network_header += len;
2066 skb->transport_header = skb->network_header;
2072 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
2074 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2075 u32 off = skb_mac_header_len(skb);
2078 ret = skb_cow(skb, len_diff);
2079 if (unlikely(ret < 0))
2082 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2083 if (unlikely(ret < 0))
2086 if (skb_is_gso(skb)) {
2087 /* SKB_GSO_TCPV4 needs to be changed into
2090 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
2091 skb_shinfo(skb)->gso_type &= ~SKB_GSO_TCPV4;
2092 skb_shinfo(skb)->gso_type |= SKB_GSO_TCPV6;
2095 /* Due to IPv6 header, MSS needs to be downgraded. */
2096 skb_shinfo(skb)->gso_size -= len_diff;
2097 /* Header must be checked, and gso_segs recomputed. */
2098 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2099 skb_shinfo(skb)->gso_segs = 0;
2102 skb->protocol = htons(ETH_P_IPV6);
2103 skb_clear_hash(skb);
2108 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
2110 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2111 u32 off = skb_mac_header_len(skb);
2114 ret = skb_unclone(skb, GFP_ATOMIC);
2115 if (unlikely(ret < 0))
2118 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2119 if (unlikely(ret < 0))
2122 if (skb_is_gso(skb)) {
2123 /* SKB_GSO_TCPV6 needs to be changed into
2126 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) {
2127 skb_shinfo(skb)->gso_type &= ~SKB_GSO_TCPV6;
2128 skb_shinfo(skb)->gso_type |= SKB_GSO_TCPV4;
2131 /* Due to IPv4 header, MSS can be upgraded. */
2132 skb_shinfo(skb)->gso_size += len_diff;
2133 /* Header must be checked, and gso_segs recomputed. */
2134 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2135 skb_shinfo(skb)->gso_segs = 0;
2138 skb->protocol = htons(ETH_P_IP);
2139 skb_clear_hash(skb);
2144 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
2146 __be16 from_proto = skb->protocol;
2148 if (from_proto == htons(ETH_P_IP) &&
2149 to_proto == htons(ETH_P_IPV6))
2150 return bpf_skb_proto_4_to_6(skb);
2152 if (from_proto == htons(ETH_P_IPV6) &&
2153 to_proto == htons(ETH_P_IP))
2154 return bpf_skb_proto_6_to_4(skb);
2159 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
2164 if (unlikely(flags))
2167 /* General idea is that this helper does the basic groundwork
2168 * needed for changing the protocol, and eBPF program fills the
2169 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
2170 * and other helpers, rather than passing a raw buffer here.
2172 * The rationale is to keep this minimal and without a need to
2173 * deal with raw packet data. F.e. even if we would pass buffers
2174 * here, the program still needs to call the bpf_lX_csum_replace()
2175 * helpers anyway. Plus, this way we keep also separation of
2176 * concerns, since f.e. bpf_skb_store_bytes() should only take
2179 * Currently, additional options and extension header space are
2180 * not supported, but flags register is reserved so we can adapt
2181 * that. For offloads, we mark packet as dodgy, so that headers
2182 * need to be verified first.
2184 ret = bpf_skb_proto_xlat(skb, proto);
2185 bpf_compute_data_pointers(skb);
2189 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
2190 .func = bpf_skb_change_proto,
2192 .ret_type = RET_INTEGER,
2193 .arg1_type = ARG_PTR_TO_CTX,
2194 .arg2_type = ARG_ANYTHING,
2195 .arg3_type = ARG_ANYTHING,
2198 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
2200 /* We only allow a restricted subset to be changed for now. */
2201 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
2202 !skb_pkt_type_ok(pkt_type)))
2205 skb->pkt_type = pkt_type;
2209 static const struct bpf_func_proto bpf_skb_change_type_proto = {
2210 .func = bpf_skb_change_type,
2212 .ret_type = RET_INTEGER,
2213 .arg1_type = ARG_PTR_TO_CTX,
2214 .arg2_type = ARG_ANYTHING,
2217 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
2219 switch (skb->protocol) {
2220 case htons(ETH_P_IP):
2221 return sizeof(struct iphdr);
2222 case htons(ETH_P_IPV6):
2223 return sizeof(struct ipv6hdr);
2229 static int bpf_skb_net_grow(struct sk_buff *skb, u32 len_diff)
2231 u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
2234 ret = skb_cow(skb, len_diff);
2235 if (unlikely(ret < 0))
2238 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2239 if (unlikely(ret < 0))
2242 if (skb_is_gso(skb)) {
2243 /* Due to header grow, MSS needs to be downgraded. */
2244 skb_shinfo(skb)->gso_size -= len_diff;
2245 /* Header must be checked, and gso_segs recomputed. */
2246 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2247 skb_shinfo(skb)->gso_segs = 0;
2253 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 len_diff)
2255 u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
2258 ret = skb_unclone(skb, GFP_ATOMIC);
2259 if (unlikely(ret < 0))
2262 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2263 if (unlikely(ret < 0))
2266 if (skb_is_gso(skb)) {
2267 /* Due to header shrink, MSS can be upgraded. */
2268 skb_shinfo(skb)->gso_size += len_diff;
2269 /* Header must be checked, and gso_segs recomputed. */
2270 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2271 skb_shinfo(skb)->gso_segs = 0;
2277 static u32 __bpf_skb_max_len(const struct sk_buff *skb)
2279 return skb->dev->mtu + skb->dev->hard_header_len;
2282 static int bpf_skb_adjust_net(struct sk_buff *skb, s32 len_diff)
2284 bool trans_same = skb->transport_header == skb->network_header;
2285 u32 len_cur, len_diff_abs = abs(len_diff);
2286 u32 len_min = bpf_skb_net_base_len(skb);
2287 u32 len_max = __bpf_skb_max_len(skb);
2288 __be16 proto = skb->protocol;
2289 bool shrink = len_diff < 0;
2292 if (unlikely(len_diff_abs > 0xfffU))
2294 if (unlikely(proto != htons(ETH_P_IP) &&
2295 proto != htons(ETH_P_IPV6)))
2298 len_cur = skb->len - skb_network_offset(skb);
2299 if (skb_transport_header_was_set(skb) && !trans_same)
2300 len_cur = skb_network_header_len(skb);
2301 if ((shrink && (len_diff_abs >= len_cur ||
2302 len_cur - len_diff_abs < len_min)) ||
2303 (!shrink && (skb->len + len_diff_abs > len_max &&
2307 ret = shrink ? bpf_skb_net_shrink(skb, len_diff_abs) :
2308 bpf_skb_net_grow(skb, len_diff_abs);
2310 bpf_compute_data_pointers(skb);
2314 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
2315 u32, mode, u64, flags)
2317 if (unlikely(flags))
2319 if (likely(mode == BPF_ADJ_ROOM_NET))
2320 return bpf_skb_adjust_net(skb, len_diff);
2325 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
2326 .func = bpf_skb_adjust_room,
2328 .ret_type = RET_INTEGER,
2329 .arg1_type = ARG_PTR_TO_CTX,
2330 .arg2_type = ARG_ANYTHING,
2331 .arg3_type = ARG_ANYTHING,
2332 .arg4_type = ARG_ANYTHING,
2335 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
2337 u32 min_len = skb_network_offset(skb);
2339 if (skb_transport_header_was_set(skb))
2340 min_len = skb_transport_offset(skb);
2341 if (skb->ip_summed == CHECKSUM_PARTIAL)
2342 min_len = skb_checksum_start_offset(skb) +
2343 skb->csum_offset + sizeof(__sum16);
2347 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
2349 unsigned int old_len = skb->len;
2352 ret = __skb_grow_rcsum(skb, new_len);
2354 memset(skb->data + old_len, 0, new_len - old_len);
2358 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
2360 return __skb_trim_rcsum(skb, new_len);
2363 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
2366 u32 max_len = __bpf_skb_max_len(skb);
2367 u32 min_len = __bpf_skb_min_len(skb);
2370 if (unlikely(flags || new_len > max_len || new_len < min_len))
2372 if (skb->encapsulation)
2375 /* The basic idea of this helper is that it's performing the
2376 * needed work to either grow or trim an skb, and eBPF program
2377 * rewrites the rest via helpers like bpf_skb_store_bytes(),
2378 * bpf_lX_csum_replace() and others rather than passing a raw
2379 * buffer here. This one is a slow path helper and intended
2380 * for replies with control messages.
2382 * Like in bpf_skb_change_proto(), we want to keep this rather
2383 * minimal and without protocol specifics so that we are able
2384 * to separate concerns as in bpf_skb_store_bytes() should only
2385 * be the one responsible for writing buffers.
2387 * It's really expected to be a slow path operation here for
2388 * control message replies, so we're implicitly linearizing,
2389 * uncloning and drop offloads from the skb by this.
2391 ret = __bpf_try_make_writable(skb, skb->len);
2393 if (new_len > skb->len)
2394 ret = bpf_skb_grow_rcsum(skb, new_len);
2395 else if (new_len < skb->len)
2396 ret = bpf_skb_trim_rcsum(skb, new_len);
2397 if (!ret && skb_is_gso(skb))
2401 bpf_compute_data_pointers(skb);
2405 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
2406 .func = bpf_skb_change_tail,
2408 .ret_type = RET_INTEGER,
2409 .arg1_type = ARG_PTR_TO_CTX,
2410 .arg2_type = ARG_ANYTHING,
2411 .arg3_type = ARG_ANYTHING,
2414 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
2417 u32 max_len = __bpf_skb_max_len(skb);
2418 u32 new_len = skb->len + head_room;
2421 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
2422 new_len < skb->len))
2425 ret = skb_cow(skb, head_room);
2427 /* Idea for this helper is that we currently only
2428 * allow to expand on mac header. This means that
2429 * skb->protocol network header, etc, stay as is.
2430 * Compared to bpf_skb_change_tail(), we're more
2431 * flexible due to not needing to linearize or
2432 * reset GSO. Intention for this helper is to be
2433 * used by an L3 skb that needs to push mac header
2434 * for redirection into L2 device.
2436 __skb_push(skb, head_room);
2437 memset(skb->data, 0, head_room);
2438 skb_reset_mac_header(skb);
2441 bpf_compute_data_pointers(skb);
2445 static const struct bpf_func_proto bpf_skb_change_head_proto = {
2446 .func = bpf_skb_change_head,
2448 .ret_type = RET_INTEGER,
2449 .arg1_type = ARG_PTR_TO_CTX,
2450 .arg2_type = ARG_ANYTHING,
2451 .arg3_type = ARG_ANYTHING,
2454 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
2456 return xdp_data_meta_unsupported(xdp) ? 0 :
2457 xdp->data - xdp->data_meta;
2460 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
2462 unsigned long metalen = xdp_get_metalen(xdp);
2463 void *data_start = xdp->data_hard_start + metalen;
2464 void *data = xdp->data + offset;
2466 if (unlikely(data < data_start ||
2467 data > xdp->data_end - ETH_HLEN))
2471 memmove(xdp->data_meta + offset,
2472 xdp->data_meta, metalen);
2473 xdp->data_meta += offset;
2479 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
2480 .func = bpf_xdp_adjust_head,
2482 .ret_type = RET_INTEGER,
2483 .arg1_type = ARG_PTR_TO_CTX,
2484 .arg2_type = ARG_ANYTHING,
2487 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
2489 void *meta = xdp->data_meta + offset;
2490 unsigned long metalen = xdp->data - meta;
2492 if (xdp_data_meta_unsupported(xdp))
2494 if (unlikely(meta < xdp->data_hard_start ||
2497 if (unlikely((metalen & (sizeof(__u32) - 1)) ||
2501 xdp->data_meta = meta;
2506 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
2507 .func = bpf_xdp_adjust_meta,
2509 .ret_type = RET_INTEGER,
2510 .arg1_type = ARG_PTR_TO_CTX,
2511 .arg2_type = ARG_ANYTHING,
2514 static int __bpf_tx_xdp(struct net_device *dev,
2515 struct bpf_map *map,
2516 struct xdp_buff *xdp,
2521 if (!dev->netdev_ops->ndo_xdp_xmit) {
2525 err = dev->netdev_ops->ndo_xdp_xmit(dev, xdp);
2529 __dev_map_insert_ctx(map, index);
2531 dev->netdev_ops->ndo_xdp_flush(dev);
2535 void xdp_do_flush_map(void)
2537 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2538 struct bpf_map *map = ri->map_to_flush;
2540 ri->map_to_flush = NULL;
2542 __dev_map_flush(map);
2544 EXPORT_SYMBOL_GPL(xdp_do_flush_map);
2546 static inline bool xdp_map_invalid(const struct bpf_prog *xdp_prog,
2549 return (unsigned long)xdp_prog->aux != aux;
2552 static int xdp_do_redirect_map(struct net_device *dev, struct xdp_buff *xdp,
2553 struct bpf_prog *xdp_prog)
2555 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2556 unsigned long map_owner = ri->map_owner;
2557 struct bpf_map *map = ri->map;
2558 struct net_device *fwd = NULL;
2559 u32 index = ri->ifindex;
2566 if (unlikely(xdp_map_invalid(xdp_prog, map_owner))) {
2572 fwd = __dev_map_lookup_elem(map, index);
2577 if (ri->map_to_flush && ri->map_to_flush != map)
2580 err = __bpf_tx_xdp(fwd, map, xdp, index);
2584 ri->map_to_flush = map;
2585 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
2588 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
2592 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
2593 struct bpf_prog *xdp_prog)
2595 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2596 struct net_device *fwd;
2597 u32 index = ri->ifindex;
2601 return xdp_do_redirect_map(dev, xdp, xdp_prog);
2603 fwd = dev_get_by_index_rcu(dev_net(dev), index);
2605 if (unlikely(!fwd)) {
2610 err = __bpf_tx_xdp(fwd, NULL, xdp, 0);
2614 _trace_xdp_redirect(dev, xdp_prog, index);
2617 _trace_xdp_redirect_err(dev, xdp_prog, index, err);
2620 EXPORT_SYMBOL_GPL(xdp_do_redirect);
2622 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
2623 struct bpf_prog *xdp_prog)
2625 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2626 unsigned long map_owner = ri->map_owner;
2627 struct bpf_map *map = ri->map;
2628 struct net_device *fwd = NULL;
2629 u32 index = ri->ifindex;
2638 if (unlikely(xdp_map_invalid(xdp_prog, map_owner))) {
2643 fwd = __dev_map_lookup_elem(map, index);
2645 fwd = dev_get_by_index_rcu(dev_net(dev), index);
2647 if (unlikely(!fwd)) {
2652 if (unlikely(!(fwd->flags & IFF_UP))) {
2657 len = fwd->mtu + fwd->hard_header_len + VLAN_HLEN;
2658 if (skb->len > len) {
2664 map ? _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index)
2665 : _trace_xdp_redirect(dev, xdp_prog, index);
2668 map ? _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err)
2669 : _trace_xdp_redirect_err(dev, xdp_prog, index, err);
2672 EXPORT_SYMBOL_GPL(xdp_do_generic_redirect);
2674 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
2676 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2678 if (unlikely(flags))
2681 ri->ifindex = ifindex;
2686 return XDP_REDIRECT;
2689 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
2690 .func = bpf_xdp_redirect,
2692 .ret_type = RET_INTEGER,
2693 .arg1_type = ARG_ANYTHING,
2694 .arg2_type = ARG_ANYTHING,
2697 BPF_CALL_4(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex, u64, flags,
2698 unsigned long, map_owner)
2700 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
2702 if (unlikely(flags))
2705 ri->ifindex = ifindex;
2708 ri->map_owner = map_owner;
2710 return XDP_REDIRECT;
2713 /* Note, arg4 is hidden from users and populated by the verifier
2714 * with the right pointer.
2716 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
2717 .func = bpf_xdp_redirect_map,
2719 .ret_type = RET_INTEGER,
2720 .arg1_type = ARG_CONST_MAP_PTR,
2721 .arg2_type = ARG_ANYTHING,
2722 .arg3_type = ARG_ANYTHING,
2725 bool bpf_helper_changes_pkt_data(void *func)
2727 if (func == bpf_skb_vlan_push ||
2728 func == bpf_skb_vlan_pop ||
2729 func == bpf_skb_store_bytes ||
2730 func == bpf_skb_change_proto ||
2731 func == bpf_skb_change_head ||
2732 func == bpf_skb_change_tail ||
2733 func == bpf_skb_adjust_room ||
2734 func == bpf_skb_pull_data ||
2735 func == bpf_clone_redirect ||
2736 func == bpf_l3_csum_replace ||
2737 func == bpf_l4_csum_replace ||
2738 func == bpf_xdp_adjust_head ||
2739 func == bpf_xdp_adjust_meta)
2745 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
2746 unsigned long off, unsigned long len)
2748 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
2752 if (ptr != dst_buff)
2753 memcpy(dst_buff, ptr, len);
2758 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
2759 u64, flags, void *, meta, u64, meta_size)
2761 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
2763 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
2765 if (unlikely(skb_size > skb->len))
2768 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
2772 static const struct bpf_func_proto bpf_skb_event_output_proto = {
2773 .func = bpf_skb_event_output,
2775 .ret_type = RET_INTEGER,
2776 .arg1_type = ARG_PTR_TO_CTX,
2777 .arg2_type = ARG_CONST_MAP_PTR,
2778 .arg3_type = ARG_ANYTHING,
2779 .arg4_type = ARG_PTR_TO_MEM,
2780 .arg5_type = ARG_CONST_SIZE,
2783 static unsigned short bpf_tunnel_key_af(u64 flags)
2785 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
2788 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
2789 u32, size, u64, flags)
2791 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
2792 u8 compat[sizeof(struct bpf_tunnel_key)];
2796 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
2800 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
2804 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
2807 case offsetof(struct bpf_tunnel_key, tunnel_label):
2808 case offsetof(struct bpf_tunnel_key, tunnel_ext):
2810 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
2811 /* Fixup deprecated structure layouts here, so we have
2812 * a common path later on.
2814 if (ip_tunnel_info_af(info) != AF_INET)
2817 to = (struct bpf_tunnel_key *)compat;
2824 to->tunnel_id = be64_to_cpu(info->key.tun_id);
2825 to->tunnel_tos = info->key.tos;
2826 to->tunnel_ttl = info->key.ttl;
2828 if (flags & BPF_F_TUNINFO_IPV6) {
2829 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
2830 sizeof(to->remote_ipv6));
2831 to->tunnel_label = be32_to_cpu(info->key.label);
2833 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
2836 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
2837 memcpy(to_orig, to, size);
2841 memset(to_orig, 0, size);
2845 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
2846 .func = bpf_skb_get_tunnel_key,
2848 .ret_type = RET_INTEGER,
2849 .arg1_type = ARG_PTR_TO_CTX,
2850 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
2851 .arg3_type = ARG_CONST_SIZE,
2852 .arg4_type = ARG_ANYTHING,
2855 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
2857 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
2860 if (unlikely(!info ||
2861 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
2865 if (unlikely(size < info->options_len)) {
2870 ip_tunnel_info_opts_get(to, info);
2871 if (size > info->options_len)
2872 memset(to + info->options_len, 0, size - info->options_len);
2874 return info->options_len;
2876 memset(to, 0, size);
2880 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
2881 .func = bpf_skb_get_tunnel_opt,
2883 .ret_type = RET_INTEGER,
2884 .arg1_type = ARG_PTR_TO_CTX,
2885 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
2886 .arg3_type = ARG_CONST_SIZE,
2889 static struct metadata_dst __percpu *md_dst;
2891 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
2892 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
2894 struct metadata_dst *md = this_cpu_ptr(md_dst);
2895 u8 compat[sizeof(struct bpf_tunnel_key)];
2896 struct ip_tunnel_info *info;
2898 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
2899 BPF_F_DONT_FRAGMENT)))
2901 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
2903 case offsetof(struct bpf_tunnel_key, tunnel_label):
2904 case offsetof(struct bpf_tunnel_key, tunnel_ext):
2905 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
2906 /* Fixup deprecated structure layouts here, so we have
2907 * a common path later on.
2909 memcpy(compat, from, size);
2910 memset(compat + size, 0, sizeof(compat) - size);
2911 from = (const struct bpf_tunnel_key *) compat;
2917 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
2922 dst_hold((struct dst_entry *) md);
2923 skb_dst_set(skb, (struct dst_entry *) md);
2925 info = &md->u.tun_info;
2926 info->mode = IP_TUNNEL_INFO_TX;
2928 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
2929 if (flags & BPF_F_DONT_FRAGMENT)
2930 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
2932 info->key.tun_id = cpu_to_be64(from->tunnel_id);
2933 info->key.tos = from->tunnel_tos;
2934 info->key.ttl = from->tunnel_ttl;
2936 if (flags & BPF_F_TUNINFO_IPV6) {
2937 info->mode |= IP_TUNNEL_INFO_IPV6;
2938 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
2939 sizeof(from->remote_ipv6));
2940 info->key.label = cpu_to_be32(from->tunnel_label) &
2941 IPV6_FLOWLABEL_MASK;
2943 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
2944 if (flags & BPF_F_ZERO_CSUM_TX)
2945 info->key.tun_flags &= ~TUNNEL_CSUM;
2951 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
2952 .func = bpf_skb_set_tunnel_key,
2954 .ret_type = RET_INTEGER,
2955 .arg1_type = ARG_PTR_TO_CTX,
2956 .arg2_type = ARG_PTR_TO_MEM,
2957 .arg3_type = ARG_CONST_SIZE,
2958 .arg4_type = ARG_ANYTHING,
2961 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
2962 const u8 *, from, u32, size)
2964 struct ip_tunnel_info *info = skb_tunnel_info(skb);
2965 const struct metadata_dst *md = this_cpu_ptr(md_dst);
2967 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
2969 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
2972 ip_tunnel_info_opts_set(info, from, size);
2977 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
2978 .func = bpf_skb_set_tunnel_opt,
2980 .ret_type = RET_INTEGER,
2981 .arg1_type = ARG_PTR_TO_CTX,
2982 .arg2_type = ARG_PTR_TO_MEM,
2983 .arg3_type = ARG_CONST_SIZE,
2986 static const struct bpf_func_proto *
2987 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
2990 /* Race is not possible, since it's called from verifier
2991 * that is holding verifier mutex.
2993 md_dst = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
3001 case BPF_FUNC_skb_set_tunnel_key:
3002 return &bpf_skb_set_tunnel_key_proto;
3003 case BPF_FUNC_skb_set_tunnel_opt:
3004 return &bpf_skb_set_tunnel_opt_proto;
3010 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
3013 struct bpf_array *array = container_of(map, struct bpf_array, map);
3014 struct cgroup *cgrp;
3017 sk = skb_to_full_sk(skb);
3018 if (!sk || !sk_fullsock(sk))
3020 if (unlikely(idx >= array->map.max_entries))
3023 cgrp = READ_ONCE(array->ptrs[idx]);
3024 if (unlikely(!cgrp))
3027 return sk_under_cgroup_hierarchy(sk, cgrp);
3030 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
3031 .func = bpf_skb_under_cgroup,
3033 .ret_type = RET_INTEGER,
3034 .arg1_type = ARG_PTR_TO_CTX,
3035 .arg2_type = ARG_CONST_MAP_PTR,
3036 .arg3_type = ARG_ANYTHING,
3039 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
3040 unsigned long off, unsigned long len)
3042 memcpy(dst_buff, src_buff + off, len);
3046 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
3047 u64, flags, void *, meta, u64, meta_size)
3049 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
3051 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
3053 if (unlikely(xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
3056 return bpf_event_output(map, flags, meta, meta_size, xdp->data,
3057 xdp_size, bpf_xdp_copy);
3060 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
3061 .func = bpf_xdp_event_output,
3063 .ret_type = RET_INTEGER,
3064 .arg1_type = ARG_PTR_TO_CTX,
3065 .arg2_type = ARG_CONST_MAP_PTR,
3066 .arg3_type = ARG_ANYTHING,
3067 .arg4_type = ARG_PTR_TO_MEM,
3068 .arg5_type = ARG_CONST_SIZE,
3071 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
3073 return skb->sk ? sock_gen_cookie(skb->sk) : 0;
3076 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
3077 .func = bpf_get_socket_cookie,
3079 .ret_type = RET_INTEGER,
3080 .arg1_type = ARG_PTR_TO_CTX,
3083 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
3085 struct sock *sk = sk_to_full_sk(skb->sk);
3088 if (!sk || !sk_fullsock(sk))
3090 kuid = sock_net_uid(sock_net(sk), sk);
3091 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
3094 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
3095 .func = bpf_get_socket_uid,
3097 .ret_type = RET_INTEGER,
3098 .arg1_type = ARG_PTR_TO_CTX,
3101 BPF_CALL_5(bpf_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
3102 int, level, int, optname, char *, optval, int, optlen)
3104 struct sock *sk = bpf_sock->sk;
3108 if (!sk_fullsock(sk))
3111 if (level == SOL_SOCKET) {
3112 if (optlen != sizeof(int))
3114 val = *((int *)optval);
3116 /* Only some socketops are supported */
3119 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
3120 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
3123 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
3124 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
3126 case SO_MAX_PACING_RATE:
3127 sk->sk_max_pacing_rate = val;
3128 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
3129 sk->sk_max_pacing_rate);
3132 sk->sk_priority = val;
3137 sk->sk_rcvlowat = val ? : 1;
3146 } else if (level == SOL_TCP &&
3147 sk->sk_prot->setsockopt == tcp_setsockopt) {
3148 if (optname == TCP_CONGESTION) {
3149 char name[TCP_CA_NAME_MAX];
3150 bool reinit = bpf_sock->op > BPF_SOCK_OPS_NEEDS_ECN;
3152 strncpy(name, optval, min_t(long, optlen,
3153 TCP_CA_NAME_MAX-1));
3154 name[TCP_CA_NAME_MAX-1] = 0;
3155 ret = tcp_set_congestion_control(sk, name, false, reinit);
3157 struct tcp_sock *tp = tcp_sk(sk);
3159 if (optlen != sizeof(int))
3162 val = *((int *)optval);
3163 /* Only some options are supported */
3166 if (val <= 0 || tp->data_segs_out > 0)
3171 case TCP_BPF_SNDCWND_CLAMP:
3175 tp->snd_cwnd_clamp = val;
3176 tp->snd_ssthresh = val;
3190 static const struct bpf_func_proto bpf_setsockopt_proto = {
3191 .func = bpf_setsockopt,
3193 .ret_type = RET_INTEGER,
3194 .arg1_type = ARG_PTR_TO_CTX,
3195 .arg2_type = ARG_ANYTHING,
3196 .arg3_type = ARG_ANYTHING,
3197 .arg4_type = ARG_PTR_TO_MEM,
3198 .arg5_type = ARG_CONST_SIZE,
3201 static const struct bpf_func_proto *
3202 bpf_base_func_proto(enum bpf_func_id func_id)
3205 case BPF_FUNC_map_lookup_elem:
3206 return &bpf_map_lookup_elem_proto;
3207 case BPF_FUNC_map_update_elem:
3208 return &bpf_map_update_elem_proto;
3209 case BPF_FUNC_map_delete_elem:
3210 return &bpf_map_delete_elem_proto;
3211 case BPF_FUNC_get_prandom_u32:
3212 return &bpf_get_prandom_u32_proto;
3213 case BPF_FUNC_get_smp_processor_id:
3214 return &bpf_get_raw_smp_processor_id_proto;
3215 case BPF_FUNC_get_numa_node_id:
3216 return &bpf_get_numa_node_id_proto;
3217 case BPF_FUNC_tail_call:
3218 return &bpf_tail_call_proto;
3219 case BPF_FUNC_ktime_get_ns:
3220 return &bpf_ktime_get_ns_proto;
3221 case BPF_FUNC_trace_printk:
3222 if (capable(CAP_SYS_ADMIN))
3223 return bpf_get_trace_printk_proto();
3229 static const struct bpf_func_proto *
3230 sock_filter_func_proto(enum bpf_func_id func_id)
3233 /* inet and inet6 sockets are created in a process
3234 * context so there is always a valid uid/gid
3236 case BPF_FUNC_get_current_uid_gid:
3237 return &bpf_get_current_uid_gid_proto;
3239 return bpf_base_func_proto(func_id);
3243 static const struct bpf_func_proto *
3244 sk_filter_func_proto(enum bpf_func_id func_id)
3247 case BPF_FUNC_skb_load_bytes:
3248 return &bpf_skb_load_bytes_proto;
3249 case BPF_FUNC_get_socket_cookie:
3250 return &bpf_get_socket_cookie_proto;
3251 case BPF_FUNC_get_socket_uid:
3252 return &bpf_get_socket_uid_proto;
3254 return bpf_base_func_proto(func_id);
3258 static const struct bpf_func_proto *
3259 tc_cls_act_func_proto(enum bpf_func_id func_id)
3262 case BPF_FUNC_skb_store_bytes:
3263 return &bpf_skb_store_bytes_proto;
3264 case BPF_FUNC_skb_load_bytes:
3265 return &bpf_skb_load_bytes_proto;
3266 case BPF_FUNC_skb_pull_data:
3267 return &bpf_skb_pull_data_proto;
3268 case BPF_FUNC_csum_diff:
3269 return &bpf_csum_diff_proto;
3270 case BPF_FUNC_csum_update:
3271 return &bpf_csum_update_proto;
3272 case BPF_FUNC_l3_csum_replace:
3273 return &bpf_l3_csum_replace_proto;
3274 case BPF_FUNC_l4_csum_replace:
3275 return &bpf_l4_csum_replace_proto;
3276 case BPF_FUNC_clone_redirect:
3277 return &bpf_clone_redirect_proto;
3278 case BPF_FUNC_get_cgroup_classid:
3279 return &bpf_get_cgroup_classid_proto;
3280 case BPF_FUNC_skb_vlan_push:
3281 return &bpf_skb_vlan_push_proto;
3282 case BPF_FUNC_skb_vlan_pop:
3283 return &bpf_skb_vlan_pop_proto;
3284 case BPF_FUNC_skb_change_proto:
3285 return &bpf_skb_change_proto_proto;
3286 case BPF_FUNC_skb_change_type:
3287 return &bpf_skb_change_type_proto;
3288 case BPF_FUNC_skb_adjust_room:
3289 return &bpf_skb_adjust_room_proto;
3290 case BPF_FUNC_skb_change_tail:
3291 return &bpf_skb_change_tail_proto;
3292 case BPF_FUNC_skb_get_tunnel_key:
3293 return &bpf_skb_get_tunnel_key_proto;
3294 case BPF_FUNC_skb_set_tunnel_key:
3295 return bpf_get_skb_set_tunnel_proto(func_id);
3296 case BPF_FUNC_skb_get_tunnel_opt:
3297 return &bpf_skb_get_tunnel_opt_proto;
3298 case BPF_FUNC_skb_set_tunnel_opt:
3299 return bpf_get_skb_set_tunnel_proto(func_id);
3300 case BPF_FUNC_redirect:
3301 return &bpf_redirect_proto;
3302 case BPF_FUNC_get_route_realm:
3303 return &bpf_get_route_realm_proto;
3304 case BPF_FUNC_get_hash_recalc:
3305 return &bpf_get_hash_recalc_proto;
3306 case BPF_FUNC_set_hash_invalid:
3307 return &bpf_set_hash_invalid_proto;
3308 case BPF_FUNC_set_hash:
3309 return &bpf_set_hash_proto;
3310 case BPF_FUNC_perf_event_output:
3311 return &bpf_skb_event_output_proto;
3312 case BPF_FUNC_get_smp_processor_id:
3313 return &bpf_get_smp_processor_id_proto;
3314 case BPF_FUNC_skb_under_cgroup:
3315 return &bpf_skb_under_cgroup_proto;
3316 case BPF_FUNC_get_socket_cookie:
3317 return &bpf_get_socket_cookie_proto;
3318 case BPF_FUNC_get_socket_uid:
3319 return &bpf_get_socket_uid_proto;
3321 return bpf_base_func_proto(func_id);
3325 static const struct bpf_func_proto *
3326 xdp_func_proto(enum bpf_func_id func_id)
3329 case BPF_FUNC_perf_event_output:
3330 return &bpf_xdp_event_output_proto;
3331 case BPF_FUNC_get_smp_processor_id:
3332 return &bpf_get_smp_processor_id_proto;
3333 case BPF_FUNC_xdp_adjust_head:
3334 return &bpf_xdp_adjust_head_proto;
3335 case BPF_FUNC_xdp_adjust_meta:
3336 return &bpf_xdp_adjust_meta_proto;
3337 case BPF_FUNC_redirect:
3338 return &bpf_xdp_redirect_proto;
3339 case BPF_FUNC_redirect_map:
3340 return &bpf_xdp_redirect_map_proto;
3342 return bpf_base_func_proto(func_id);
3346 static const struct bpf_func_proto *
3347 lwt_inout_func_proto(enum bpf_func_id func_id)
3350 case BPF_FUNC_skb_load_bytes:
3351 return &bpf_skb_load_bytes_proto;
3352 case BPF_FUNC_skb_pull_data:
3353 return &bpf_skb_pull_data_proto;
3354 case BPF_FUNC_csum_diff:
3355 return &bpf_csum_diff_proto;
3356 case BPF_FUNC_get_cgroup_classid:
3357 return &bpf_get_cgroup_classid_proto;
3358 case BPF_FUNC_get_route_realm:
3359 return &bpf_get_route_realm_proto;
3360 case BPF_FUNC_get_hash_recalc:
3361 return &bpf_get_hash_recalc_proto;
3362 case BPF_FUNC_perf_event_output:
3363 return &bpf_skb_event_output_proto;
3364 case BPF_FUNC_get_smp_processor_id:
3365 return &bpf_get_smp_processor_id_proto;
3366 case BPF_FUNC_skb_under_cgroup:
3367 return &bpf_skb_under_cgroup_proto;
3369 return bpf_base_func_proto(func_id);
3373 static const struct bpf_func_proto *
3374 sock_ops_func_proto(enum bpf_func_id func_id)
3377 case BPF_FUNC_setsockopt:
3378 return &bpf_setsockopt_proto;
3379 case BPF_FUNC_sock_map_update:
3380 return &bpf_sock_map_update_proto;
3382 return bpf_base_func_proto(func_id);
3386 static const struct bpf_func_proto *sk_skb_func_proto(enum bpf_func_id func_id)
3389 case BPF_FUNC_skb_store_bytes:
3390 return &bpf_skb_store_bytes_proto;
3391 case BPF_FUNC_skb_load_bytes:
3392 return &bpf_skb_load_bytes_proto;
3393 case BPF_FUNC_skb_pull_data:
3394 return &bpf_skb_pull_data_proto;
3395 case BPF_FUNC_skb_change_tail:
3396 return &bpf_skb_change_tail_proto;
3397 case BPF_FUNC_skb_change_head:
3398 return &bpf_skb_change_head_proto;
3399 case BPF_FUNC_get_socket_cookie:
3400 return &bpf_get_socket_cookie_proto;
3401 case BPF_FUNC_get_socket_uid:
3402 return &bpf_get_socket_uid_proto;
3403 case BPF_FUNC_sk_redirect_map:
3404 return &bpf_sk_redirect_map_proto;
3406 return bpf_base_func_proto(func_id);
3410 static const struct bpf_func_proto *
3411 lwt_xmit_func_proto(enum bpf_func_id func_id)
3414 case BPF_FUNC_skb_get_tunnel_key:
3415 return &bpf_skb_get_tunnel_key_proto;
3416 case BPF_FUNC_skb_set_tunnel_key:
3417 return bpf_get_skb_set_tunnel_proto(func_id);
3418 case BPF_FUNC_skb_get_tunnel_opt:
3419 return &bpf_skb_get_tunnel_opt_proto;
3420 case BPF_FUNC_skb_set_tunnel_opt:
3421 return bpf_get_skb_set_tunnel_proto(func_id);
3422 case BPF_FUNC_redirect:
3423 return &bpf_redirect_proto;
3424 case BPF_FUNC_clone_redirect:
3425 return &bpf_clone_redirect_proto;
3426 case BPF_FUNC_skb_change_tail:
3427 return &bpf_skb_change_tail_proto;
3428 case BPF_FUNC_skb_change_head:
3429 return &bpf_skb_change_head_proto;
3430 case BPF_FUNC_skb_store_bytes:
3431 return &bpf_skb_store_bytes_proto;
3432 case BPF_FUNC_csum_update:
3433 return &bpf_csum_update_proto;
3434 case BPF_FUNC_l3_csum_replace:
3435 return &bpf_l3_csum_replace_proto;
3436 case BPF_FUNC_l4_csum_replace:
3437 return &bpf_l4_csum_replace_proto;
3438 case BPF_FUNC_set_hash_invalid:
3439 return &bpf_set_hash_invalid_proto;
3441 return lwt_inout_func_proto(func_id);
3445 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
3446 struct bpf_insn_access_aux *info)
3448 const int size_default = sizeof(__u32);
3450 if (off < 0 || off >= sizeof(struct __sk_buff))
3453 /* The verifier guarantees that size > 0. */
3454 if (off % size != 0)
3458 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
3459 if (off + size > offsetofend(struct __sk_buff, cb[4]))
3462 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
3463 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
3464 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
3465 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
3466 case bpf_ctx_range(struct __sk_buff, data):
3467 case bpf_ctx_range(struct __sk_buff, data_meta):
3468 case bpf_ctx_range(struct __sk_buff, data_end):
3469 if (size != size_default)
3473 /* Only narrow read access allowed for now. */
3474 if (type == BPF_WRITE) {
3475 if (size != size_default)
3478 bpf_ctx_record_field_size(info, size_default);
3479 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
3487 static bool sk_filter_is_valid_access(int off, int size,
3488 enum bpf_access_type type,
3489 struct bpf_insn_access_aux *info)
3492 case bpf_ctx_range(struct __sk_buff, tc_classid):
3493 case bpf_ctx_range(struct __sk_buff, data):
3494 case bpf_ctx_range(struct __sk_buff, data_meta):
3495 case bpf_ctx_range(struct __sk_buff, data_end):
3496 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
3500 if (type == BPF_WRITE) {
3502 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
3509 return bpf_skb_is_valid_access(off, size, type, info);
3512 static bool lwt_is_valid_access(int off, int size,
3513 enum bpf_access_type type,
3514 struct bpf_insn_access_aux *info)
3517 case bpf_ctx_range(struct __sk_buff, tc_classid):
3518 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
3519 case bpf_ctx_range(struct __sk_buff, data_meta):
3523 if (type == BPF_WRITE) {
3525 case bpf_ctx_range(struct __sk_buff, mark):
3526 case bpf_ctx_range(struct __sk_buff, priority):
3527 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
3535 case bpf_ctx_range(struct __sk_buff, data):
3536 info->reg_type = PTR_TO_PACKET;
3538 case bpf_ctx_range(struct __sk_buff, data_end):
3539 info->reg_type = PTR_TO_PACKET_END;
3543 return bpf_skb_is_valid_access(off, size, type, info);
3546 static bool sock_filter_is_valid_access(int off, int size,
3547 enum bpf_access_type type,
3548 struct bpf_insn_access_aux *info)
3550 if (type == BPF_WRITE) {
3552 case offsetof(struct bpf_sock, bound_dev_if):
3553 case offsetof(struct bpf_sock, mark):
3554 case offsetof(struct bpf_sock, priority):
3561 if (off < 0 || off + size > sizeof(struct bpf_sock))
3563 /* The verifier guarantees that size > 0. */
3564 if (off % size != 0)
3566 if (size != sizeof(__u32))
3572 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
3573 const struct bpf_prog *prog, int drop_verdict)
3575 struct bpf_insn *insn = insn_buf;
3580 /* if (!skb->cloned)
3583 * (Fast-path, otherwise approximation that we might be
3584 * a clone, do the rest in helper.)
3586 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
3587 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
3588 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
3590 /* ret = bpf_skb_pull_data(skb, 0); */
3591 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
3592 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
3593 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
3594 BPF_FUNC_skb_pull_data);
3597 * return TC_ACT_SHOT;
3599 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
3600 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
3601 *insn++ = BPF_EXIT_INSN();
3604 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
3606 *insn++ = prog->insnsi[0];
3608 return insn - insn_buf;
3611 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
3612 const struct bpf_prog *prog)
3614 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
3617 static bool tc_cls_act_is_valid_access(int off, int size,
3618 enum bpf_access_type type,
3619 struct bpf_insn_access_aux *info)
3621 if (type == BPF_WRITE) {
3623 case bpf_ctx_range(struct __sk_buff, mark):
3624 case bpf_ctx_range(struct __sk_buff, tc_index):
3625 case bpf_ctx_range(struct __sk_buff, priority):
3626 case bpf_ctx_range(struct __sk_buff, tc_classid):
3627 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
3635 case bpf_ctx_range(struct __sk_buff, data):
3636 info->reg_type = PTR_TO_PACKET;
3638 case bpf_ctx_range(struct __sk_buff, data_meta):
3639 info->reg_type = PTR_TO_PACKET_META;
3641 case bpf_ctx_range(struct __sk_buff, data_end):
3642 info->reg_type = PTR_TO_PACKET_END;
3644 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
3648 return bpf_skb_is_valid_access(off, size, type, info);
3651 static bool __is_valid_xdp_access(int off, int size)
3653 if (off < 0 || off >= sizeof(struct xdp_md))
3655 if (off % size != 0)
3657 if (size != sizeof(__u32))
3663 static bool xdp_is_valid_access(int off, int size,
3664 enum bpf_access_type type,
3665 struct bpf_insn_access_aux *info)
3667 if (type == BPF_WRITE)
3671 case offsetof(struct xdp_md, data):
3672 info->reg_type = PTR_TO_PACKET;
3674 case offsetof(struct xdp_md, data_meta):
3675 info->reg_type = PTR_TO_PACKET_META;
3677 case offsetof(struct xdp_md, data_end):
3678 info->reg_type = PTR_TO_PACKET_END;
3682 return __is_valid_xdp_access(off, size);
3685 void bpf_warn_invalid_xdp_action(u32 act)
3687 const u32 act_max = XDP_REDIRECT;
3689 WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
3690 act > act_max ? "Illegal" : "Driver unsupported",
3693 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
3695 static bool __is_valid_sock_ops_access(int off, int size)
3697 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
3699 /* The verifier guarantees that size > 0. */
3700 if (off % size != 0)
3702 if (size != sizeof(__u32))
3708 static bool sock_ops_is_valid_access(int off, int size,
3709 enum bpf_access_type type,
3710 struct bpf_insn_access_aux *info)
3712 if (type == BPF_WRITE) {
3714 case offsetof(struct bpf_sock_ops, op) ...
3715 offsetof(struct bpf_sock_ops, replylong[3]):
3722 return __is_valid_sock_ops_access(off, size);
3725 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
3726 const struct bpf_prog *prog)
3728 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
3731 static bool sk_skb_is_valid_access(int off, int size,
3732 enum bpf_access_type type,
3733 struct bpf_insn_access_aux *info)
3736 case bpf_ctx_range(struct __sk_buff, tc_classid):
3737 case bpf_ctx_range(struct __sk_buff, data_meta):
3741 if (type == BPF_WRITE) {
3743 case bpf_ctx_range(struct __sk_buff, mark):
3744 case bpf_ctx_range(struct __sk_buff, tc_index):
3745 case bpf_ctx_range(struct __sk_buff, priority):
3753 case bpf_ctx_range(struct __sk_buff, data):
3754 info->reg_type = PTR_TO_PACKET;
3756 case bpf_ctx_range(struct __sk_buff, data_end):
3757 info->reg_type = PTR_TO_PACKET_END;
3761 return bpf_skb_is_valid_access(off, size, type, info);
3764 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
3765 const struct bpf_insn *si,
3766 struct bpf_insn *insn_buf,
3767 struct bpf_prog *prog, u32 *target_size)
3769 struct bpf_insn *insn = insn_buf;
3773 case offsetof(struct __sk_buff, len):
3774 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
3775 bpf_target_off(struct sk_buff, len, 4,
3779 case offsetof(struct __sk_buff, protocol):
3780 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
3781 bpf_target_off(struct sk_buff, protocol, 2,
3785 case offsetof(struct __sk_buff, vlan_proto):
3786 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
3787 bpf_target_off(struct sk_buff, vlan_proto, 2,
3791 case offsetof(struct __sk_buff, priority):
3792 if (type == BPF_WRITE)
3793 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
3794 bpf_target_off(struct sk_buff, priority, 4,
3797 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
3798 bpf_target_off(struct sk_buff, priority, 4,
3802 case offsetof(struct __sk_buff, ingress_ifindex):
3803 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
3804 bpf_target_off(struct sk_buff, skb_iif, 4,
3808 case offsetof(struct __sk_buff, ifindex):
3809 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
3810 si->dst_reg, si->src_reg,
3811 offsetof(struct sk_buff, dev));
3812 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
3813 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
3814 bpf_target_off(struct net_device, ifindex, 4,
3818 case offsetof(struct __sk_buff, hash):
3819 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
3820 bpf_target_off(struct sk_buff, hash, 4,
3824 case offsetof(struct __sk_buff, mark):
3825 if (type == BPF_WRITE)
3826 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
3827 bpf_target_off(struct sk_buff, mark, 4,
3830 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
3831 bpf_target_off(struct sk_buff, mark, 4,
3835 case offsetof(struct __sk_buff, pkt_type):
3837 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
3839 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
3840 #ifdef __BIG_ENDIAN_BITFIELD
3841 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
3845 case offsetof(struct __sk_buff, queue_mapping):
3846 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
3847 bpf_target_off(struct sk_buff, queue_mapping, 2,
3851 case offsetof(struct __sk_buff, vlan_present):
3852 case offsetof(struct __sk_buff, vlan_tci):
3853 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000);
3855 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
3856 bpf_target_off(struct sk_buff, vlan_tci, 2,
3858 if (si->off == offsetof(struct __sk_buff, vlan_tci)) {
3859 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg,
3862 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 12);
3863 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
3867 case offsetof(struct __sk_buff, cb[0]) ...
3868 offsetofend(struct __sk_buff, cb[4]) - 1:
3869 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20);
3870 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
3871 offsetof(struct qdisc_skb_cb, data)) %
3874 prog->cb_access = 1;
3876 off -= offsetof(struct __sk_buff, cb[0]);
3877 off += offsetof(struct sk_buff, cb);
3878 off += offsetof(struct qdisc_skb_cb, data);
3879 if (type == BPF_WRITE)
3880 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
3883 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
3887 case offsetof(struct __sk_buff, tc_classid):
3888 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, tc_classid) != 2);
3891 off -= offsetof(struct __sk_buff, tc_classid);
3892 off += offsetof(struct sk_buff, cb);
3893 off += offsetof(struct qdisc_skb_cb, tc_classid);
3895 if (type == BPF_WRITE)
3896 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
3899 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
3903 case offsetof(struct __sk_buff, data):
3904 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
3905 si->dst_reg, si->src_reg,
3906 offsetof(struct sk_buff, data));
3909 case offsetof(struct __sk_buff, data_meta):
3911 off -= offsetof(struct __sk_buff, data_meta);
3912 off += offsetof(struct sk_buff, cb);
3913 off += offsetof(struct bpf_skb_data_end, data_meta);
3914 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
3918 case offsetof(struct __sk_buff, data_end):
3920 off -= offsetof(struct __sk_buff, data_end);
3921 off += offsetof(struct sk_buff, cb);
3922 off += offsetof(struct bpf_skb_data_end, data_end);
3923 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
3927 case offsetof(struct __sk_buff, tc_index):
3928 #ifdef CONFIG_NET_SCHED
3929 if (type == BPF_WRITE)
3930 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
3931 bpf_target_off(struct sk_buff, tc_index, 2,
3934 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
3935 bpf_target_off(struct sk_buff, tc_index, 2,
3939 if (type == BPF_WRITE)
3940 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
3942 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
3946 case offsetof(struct __sk_buff, napi_id):
3947 #if defined(CONFIG_NET_RX_BUSY_POLL)
3948 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
3949 bpf_target_off(struct sk_buff, napi_id, 4,
3951 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
3952 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
3955 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
3958 case offsetof(struct __sk_buff, family):
3959 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
3961 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
3962 si->dst_reg, si->src_reg,
3963 offsetof(struct sk_buff, sk));
3964 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
3965 bpf_target_off(struct sock_common,
3969 case offsetof(struct __sk_buff, remote_ip4):
3970 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
3972 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
3973 si->dst_reg, si->src_reg,
3974 offsetof(struct sk_buff, sk));
3975 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
3976 bpf_target_off(struct sock_common,
3980 case offsetof(struct __sk_buff, local_ip4):
3981 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
3982 skc_rcv_saddr) != 4);
3984 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
3985 si->dst_reg, si->src_reg,
3986 offsetof(struct sk_buff, sk));
3987 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
3988 bpf_target_off(struct sock_common,
3992 case offsetof(struct __sk_buff, remote_ip6[0]) ...
3993 offsetof(struct __sk_buff, remote_ip6[3]):
3994 #if IS_ENABLED(CONFIG_IPV6)
3995 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
3996 skc_v6_daddr.s6_addr32[0]) != 4);
3999 off -= offsetof(struct __sk_buff, remote_ip6[0]);
4001 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
4002 si->dst_reg, si->src_reg,
4003 offsetof(struct sk_buff, sk));
4004 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4005 offsetof(struct sock_common,
4006 skc_v6_daddr.s6_addr32[0]) +
4009 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
4012 case offsetof(struct __sk_buff, local_ip6[0]) ...
4013 offsetof(struct __sk_buff, local_ip6[3]):
4014 #if IS_ENABLED(CONFIG_IPV6)
4015 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
4016 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
4019 off -= offsetof(struct __sk_buff, local_ip6[0]);
4021 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
4022 si->dst_reg, si->src_reg,
4023 offsetof(struct sk_buff, sk));
4024 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4025 offsetof(struct sock_common,
4026 skc_v6_rcv_saddr.s6_addr32[0]) +
4029 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
4033 case offsetof(struct __sk_buff, remote_port):
4034 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
4036 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
4037 si->dst_reg, si->src_reg,
4038 offsetof(struct sk_buff, sk));
4039 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
4040 bpf_target_off(struct sock_common,
4043 #ifndef __BIG_ENDIAN_BITFIELD
4044 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
4048 case offsetof(struct __sk_buff, local_port):
4049 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
4051 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
4052 si->dst_reg, si->src_reg,
4053 offsetof(struct sk_buff, sk));
4054 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
4055 bpf_target_off(struct sock_common,
4056 skc_num, 2, target_size));
4060 return insn - insn_buf;
4063 static u32 sock_filter_convert_ctx_access(enum bpf_access_type type,
4064 const struct bpf_insn *si,
4065 struct bpf_insn *insn_buf,
4066 struct bpf_prog *prog, u32 *target_size)
4068 struct bpf_insn *insn = insn_buf;
4071 case offsetof(struct bpf_sock, bound_dev_if):
4072 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_bound_dev_if) != 4);
4074 if (type == BPF_WRITE)
4075 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
4076 offsetof(struct sock, sk_bound_dev_if));
4078 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4079 offsetof(struct sock, sk_bound_dev_if));
4082 case offsetof(struct bpf_sock, mark):
4083 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_mark) != 4);
4085 if (type == BPF_WRITE)
4086 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
4087 offsetof(struct sock, sk_mark));
4089 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4090 offsetof(struct sock, sk_mark));
4093 case offsetof(struct bpf_sock, priority):
4094 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_priority) != 4);
4096 if (type == BPF_WRITE)
4097 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
4098 offsetof(struct sock, sk_priority));
4100 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4101 offsetof(struct sock, sk_priority));
4104 case offsetof(struct bpf_sock, family):
4105 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_family) != 2);
4107 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
4108 offsetof(struct sock, sk_family));
4111 case offsetof(struct bpf_sock, type):
4112 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4113 offsetof(struct sock, __sk_flags_offset));
4114 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
4115 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
4118 case offsetof(struct bpf_sock, protocol):
4119 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4120 offsetof(struct sock, __sk_flags_offset));
4121 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
4122 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_PROTO_SHIFT);
4126 return insn - insn_buf;
4129 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
4130 const struct bpf_insn *si,
4131 struct bpf_insn *insn_buf,
4132 struct bpf_prog *prog, u32 *target_size)
4134 struct bpf_insn *insn = insn_buf;
4137 case offsetof(struct __sk_buff, ifindex):
4138 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
4139 si->dst_reg, si->src_reg,
4140 offsetof(struct sk_buff, dev));
4141 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4142 bpf_target_off(struct net_device, ifindex, 4,
4146 return bpf_convert_ctx_access(type, si, insn_buf, prog,
4150 return insn - insn_buf;
4153 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
4154 const struct bpf_insn *si,
4155 struct bpf_insn *insn_buf,
4156 struct bpf_prog *prog, u32 *target_size)
4158 struct bpf_insn *insn = insn_buf;
4161 case offsetof(struct xdp_md, data):
4162 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
4163 si->dst_reg, si->src_reg,
4164 offsetof(struct xdp_buff, data));
4166 case offsetof(struct xdp_md, data_meta):
4167 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
4168 si->dst_reg, si->src_reg,
4169 offsetof(struct xdp_buff, data_meta));
4171 case offsetof(struct xdp_md, data_end):
4172 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
4173 si->dst_reg, si->src_reg,
4174 offsetof(struct xdp_buff, data_end));
4178 return insn - insn_buf;
4181 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
4182 const struct bpf_insn *si,
4183 struct bpf_insn *insn_buf,
4184 struct bpf_prog *prog,
4187 struct bpf_insn *insn = insn_buf;
4191 case offsetof(struct bpf_sock_ops, op) ...
4192 offsetof(struct bpf_sock_ops, replylong[3]):
4193 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, op) !=
4194 FIELD_SIZEOF(struct bpf_sock_ops_kern, op));
4195 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, reply) !=
4196 FIELD_SIZEOF(struct bpf_sock_ops_kern, reply));
4197 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, replylong) !=
4198 FIELD_SIZEOF(struct bpf_sock_ops_kern, replylong));
4200 off -= offsetof(struct bpf_sock_ops, op);
4201 off += offsetof(struct bpf_sock_ops_kern, op);
4202 if (type == BPF_WRITE)
4203 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
4206 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
4210 case offsetof(struct bpf_sock_ops, family):
4211 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
4213 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4214 struct bpf_sock_ops_kern, sk),
4215 si->dst_reg, si->src_reg,
4216 offsetof(struct bpf_sock_ops_kern, sk));
4217 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
4218 offsetof(struct sock_common, skc_family));
4221 case offsetof(struct bpf_sock_ops, remote_ip4):
4222 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
4224 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4225 struct bpf_sock_ops_kern, sk),
4226 si->dst_reg, si->src_reg,
4227 offsetof(struct bpf_sock_ops_kern, sk));
4228 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4229 offsetof(struct sock_common, skc_daddr));
4232 case offsetof(struct bpf_sock_ops, local_ip4):
4233 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_rcv_saddr) != 4);
4235 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4236 struct bpf_sock_ops_kern, sk),
4237 si->dst_reg, si->src_reg,
4238 offsetof(struct bpf_sock_ops_kern, sk));
4239 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4240 offsetof(struct sock_common,
4244 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
4245 offsetof(struct bpf_sock_ops, remote_ip6[3]):
4246 #if IS_ENABLED(CONFIG_IPV6)
4247 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
4248 skc_v6_daddr.s6_addr32[0]) != 4);
4251 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
4252 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4253 struct bpf_sock_ops_kern, sk),
4254 si->dst_reg, si->src_reg,
4255 offsetof(struct bpf_sock_ops_kern, sk));
4256 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4257 offsetof(struct sock_common,
4258 skc_v6_daddr.s6_addr32[0]) +
4261 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
4265 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
4266 offsetof(struct bpf_sock_ops, local_ip6[3]):
4267 #if IS_ENABLED(CONFIG_IPV6)
4268 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
4269 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
4272 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
4273 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4274 struct bpf_sock_ops_kern, sk),
4275 si->dst_reg, si->src_reg,
4276 offsetof(struct bpf_sock_ops_kern, sk));
4277 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
4278 offsetof(struct sock_common,
4279 skc_v6_rcv_saddr.s6_addr32[0]) +
4282 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
4286 case offsetof(struct bpf_sock_ops, remote_port):
4287 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
4289 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4290 struct bpf_sock_ops_kern, sk),
4291 si->dst_reg, si->src_reg,
4292 offsetof(struct bpf_sock_ops_kern, sk));
4293 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
4294 offsetof(struct sock_common, skc_dport));
4295 #ifndef __BIG_ENDIAN_BITFIELD
4296 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
4300 case offsetof(struct bpf_sock_ops, local_port):
4301 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
4303 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
4304 struct bpf_sock_ops_kern, sk),
4305 si->dst_reg, si->src_reg,
4306 offsetof(struct bpf_sock_ops_kern, sk));
4307 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
4308 offsetof(struct sock_common, skc_num));
4311 return insn - insn_buf;
4314 const struct bpf_verifier_ops sk_filter_prog_ops = {
4315 .get_func_proto = sk_filter_func_proto,
4316 .is_valid_access = sk_filter_is_valid_access,
4317 .convert_ctx_access = bpf_convert_ctx_access,
4320 const struct bpf_verifier_ops tc_cls_act_prog_ops = {
4321 .get_func_proto = tc_cls_act_func_proto,
4322 .is_valid_access = tc_cls_act_is_valid_access,
4323 .convert_ctx_access = tc_cls_act_convert_ctx_access,
4324 .gen_prologue = tc_cls_act_prologue,
4325 .test_run = bpf_prog_test_run_skb,
4328 const struct bpf_verifier_ops xdp_prog_ops = {
4329 .get_func_proto = xdp_func_proto,
4330 .is_valid_access = xdp_is_valid_access,
4331 .convert_ctx_access = xdp_convert_ctx_access,
4332 .test_run = bpf_prog_test_run_xdp,
4335 const struct bpf_verifier_ops cg_skb_prog_ops = {
4336 .get_func_proto = sk_filter_func_proto,
4337 .is_valid_access = sk_filter_is_valid_access,
4338 .convert_ctx_access = bpf_convert_ctx_access,
4339 .test_run = bpf_prog_test_run_skb,
4342 const struct bpf_verifier_ops lwt_inout_prog_ops = {
4343 .get_func_proto = lwt_inout_func_proto,
4344 .is_valid_access = lwt_is_valid_access,
4345 .convert_ctx_access = bpf_convert_ctx_access,
4346 .test_run = bpf_prog_test_run_skb,
4349 const struct bpf_verifier_ops lwt_xmit_prog_ops = {
4350 .get_func_proto = lwt_xmit_func_proto,
4351 .is_valid_access = lwt_is_valid_access,
4352 .convert_ctx_access = bpf_convert_ctx_access,
4353 .gen_prologue = tc_cls_act_prologue,
4354 .test_run = bpf_prog_test_run_skb,
4357 const struct bpf_verifier_ops cg_sock_prog_ops = {
4358 .get_func_proto = sock_filter_func_proto,
4359 .is_valid_access = sock_filter_is_valid_access,
4360 .convert_ctx_access = sock_filter_convert_ctx_access,
4363 const struct bpf_verifier_ops sock_ops_prog_ops = {
4364 .get_func_proto = sock_ops_func_proto,
4365 .is_valid_access = sock_ops_is_valid_access,
4366 .convert_ctx_access = sock_ops_convert_ctx_access,
4369 const struct bpf_verifier_ops sk_skb_prog_ops = {
4370 .get_func_proto = sk_skb_func_proto,
4371 .is_valid_access = sk_skb_is_valid_access,
4372 .convert_ctx_access = bpf_convert_ctx_access,
4373 .gen_prologue = sk_skb_prologue,
4376 int sk_detach_filter(struct sock *sk)
4379 struct sk_filter *filter;
4381 if (sock_flag(sk, SOCK_FILTER_LOCKED))
4384 filter = rcu_dereference_protected(sk->sk_filter,
4385 lockdep_sock_is_held(sk));
4387 RCU_INIT_POINTER(sk->sk_filter, NULL);
4388 sk_filter_uncharge(sk, filter);
4394 EXPORT_SYMBOL_GPL(sk_detach_filter);
4396 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
4399 struct sock_fprog_kern *fprog;
4400 struct sk_filter *filter;
4404 filter = rcu_dereference_protected(sk->sk_filter,
4405 lockdep_sock_is_held(sk));
4409 /* We're copying the filter that has been originally attached,
4410 * so no conversion/decode needed anymore. eBPF programs that
4411 * have no original program cannot be dumped through this.
4414 fprog = filter->prog->orig_prog;
4420 /* User space only enquires number of filter blocks. */
4424 if (len < fprog->len)
4428 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
4431 /* Instead of bytes, the API requests to return the number