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>
30 #include <linux/inet.h>
31 #include <linux/netdevice.h>
32 #include <linux/if_packet.h>
33 #include <linux/gfp.h>
35 #include <net/protocol.h>
36 #include <net/netlink.h>
37 #include <linux/skbuff.h>
39 #include <net/flow_dissector.h>
40 #include <linux/errno.h>
41 #include <linux/timer.h>
42 #include <asm/uaccess.h>
43 #include <asm/unaligned.h>
44 #include <linux/filter.h>
45 #include <linux/ratelimit.h>
46 #include <linux/seccomp.h>
47 #include <linux/if_vlan.h>
48 #include <linux/bpf.h>
49 #include <net/sch_generic.h>
50 #include <net/cls_cgroup.h>
51 #include <net/dst_metadata.h>
54 * sk_filter - run a packet through a socket filter
55 * @sk: sock associated with &sk_buff
56 * @skb: buffer to filter
58 * Run the filter code and then cut skb->data to correct size returned by
59 * SK_RUN_FILTER. If pkt_len is 0 we toss packet. If skb->len is smaller
60 * than pkt_len we keep whole skb->data. This is the socket level
61 * wrapper to SK_RUN_FILTER. It returns 0 if the packet should
62 * be accepted or -EPERM if the packet should be tossed.
65 int sk_filter(struct sock *sk, struct sk_buff *skb)
68 struct sk_filter *filter;
71 * If the skb was allocated from pfmemalloc reserves, only
72 * allow SOCK_MEMALLOC sockets to use it as this socket is
75 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
78 err = security_sock_rcv_skb(sk, skb);
83 filter = rcu_dereference(sk->sk_filter);
85 unsigned int pkt_len = SK_RUN_FILTER(filter, skb);
87 err = pkt_len ? pskb_trim(skb, pkt_len) : -EPERM;
93 EXPORT_SYMBOL(sk_filter);
95 static u64 __skb_get_pay_offset(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
97 return skb_get_poff((struct sk_buff *)(unsigned long) ctx);
100 static u64 __skb_get_nlattr(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
102 struct sk_buff *skb = (struct sk_buff *)(unsigned long) ctx;
105 if (skb_is_nonlinear(skb))
108 if (skb->len < sizeof(struct nlattr))
111 if (a > skb->len - sizeof(struct nlattr))
114 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
116 return (void *) nla - (void *) skb->data;
121 static u64 __skb_get_nlattr_nest(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
123 struct sk_buff *skb = (struct sk_buff *)(unsigned long) ctx;
126 if (skb_is_nonlinear(skb))
129 if (skb->len < sizeof(struct nlattr))
132 if (a > skb->len - sizeof(struct nlattr))
135 nla = (struct nlattr *) &skb->data[a];
136 if (nla->nla_len > skb->len - a)
139 nla = nla_find_nested(nla, x);
141 return (void *) nla - (void *) skb->data;
146 static u64 __get_raw_cpu_id(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
148 return raw_smp_processor_id();
151 /* note that this only generates 32-bit random numbers */
152 static u64 __get_random_u32(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
154 return prandom_u32();
157 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
158 struct bpf_insn *insn_buf)
160 struct bpf_insn *insn = insn_buf;
164 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
166 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
167 offsetof(struct sk_buff, mark));
171 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
172 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
173 #ifdef __BIG_ENDIAN_BITFIELD
174 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
179 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2);
181 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
182 offsetof(struct sk_buff, queue_mapping));
185 case SKF_AD_VLAN_TAG:
186 case SKF_AD_VLAN_TAG_PRESENT:
187 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2);
188 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000);
190 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
191 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
192 offsetof(struct sk_buff, vlan_tci));
193 if (skb_field == SKF_AD_VLAN_TAG) {
194 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg,
198 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 12);
200 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
205 return insn - insn_buf;
208 static bool convert_bpf_extensions(struct sock_filter *fp,
209 struct bpf_insn **insnp)
211 struct bpf_insn *insn = *insnp;
215 case SKF_AD_OFF + SKF_AD_PROTOCOL:
216 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
218 /* A = *(u16 *) (CTX + offsetof(protocol)) */
219 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
220 offsetof(struct sk_buff, protocol));
221 /* A = ntohs(A) [emitting a nop or swap16] */
222 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
225 case SKF_AD_OFF + SKF_AD_PKTTYPE:
226 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
230 case SKF_AD_OFF + SKF_AD_IFINDEX:
231 case SKF_AD_OFF + SKF_AD_HATYPE:
232 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
233 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, type) != 2);
234 BUILD_BUG_ON(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, dev)) < 0);
236 *insn++ = BPF_LDX_MEM(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, dev)),
237 BPF_REG_TMP, BPF_REG_CTX,
238 offsetof(struct sk_buff, dev));
239 /* if (tmp != 0) goto pc + 1 */
240 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
241 *insn++ = BPF_EXIT_INSN();
242 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
243 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
244 offsetof(struct net_device, ifindex));
246 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
247 offsetof(struct net_device, type));
250 case SKF_AD_OFF + SKF_AD_MARK:
251 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
255 case SKF_AD_OFF + SKF_AD_RXHASH:
256 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
258 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
259 offsetof(struct sk_buff, hash));
262 case SKF_AD_OFF + SKF_AD_QUEUE:
263 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
267 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
268 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
269 BPF_REG_A, BPF_REG_CTX, insn);
273 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
274 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
275 BPF_REG_A, BPF_REG_CTX, insn);
279 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
280 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2);
282 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
283 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
284 offsetof(struct sk_buff, vlan_proto));
285 /* A = ntohs(A) [emitting a nop or swap16] */
286 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
289 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
290 case SKF_AD_OFF + SKF_AD_NLATTR:
291 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
292 case SKF_AD_OFF + SKF_AD_CPU:
293 case SKF_AD_OFF + SKF_AD_RANDOM:
295 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
297 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
299 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
300 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
302 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
303 *insn = BPF_EMIT_CALL(__skb_get_pay_offset);
305 case SKF_AD_OFF + SKF_AD_NLATTR:
306 *insn = BPF_EMIT_CALL(__skb_get_nlattr);
308 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
309 *insn = BPF_EMIT_CALL(__skb_get_nlattr_nest);
311 case SKF_AD_OFF + SKF_AD_CPU:
312 *insn = BPF_EMIT_CALL(__get_raw_cpu_id);
314 case SKF_AD_OFF + SKF_AD_RANDOM:
315 *insn = BPF_EMIT_CALL(__get_random_u32);
320 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
322 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
326 /* This is just a dummy call to avoid letting the compiler
327 * evict __bpf_call_base() as an optimization. Placed here
328 * where no-one bothers.
330 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
339 * bpf_convert_filter - convert filter program
340 * @prog: the user passed filter program
341 * @len: the length of the user passed filter program
342 * @new_prog: buffer where converted program will be stored
343 * @new_len: pointer to store length of converted program
345 * Remap 'sock_filter' style BPF instruction set to 'sock_filter_ext' style.
346 * Conversion workflow:
348 * 1) First pass for calculating the new program length:
349 * bpf_convert_filter(old_prog, old_len, NULL, &new_len)
351 * 2) 2nd pass to remap in two passes: 1st pass finds new
352 * jump offsets, 2nd pass remapping:
353 * new_prog = kmalloc(sizeof(struct bpf_insn) * new_len);
354 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len);
356 * User BPF's register A is mapped to our BPF register 6, user BPF
357 * register X is mapped to BPF register 7; frame pointer is always
358 * register 10; Context 'void *ctx' is stored in register 1, that is,
359 * for socket filters: ctx == 'struct sk_buff *', for seccomp:
360 * ctx == 'struct seccomp_data *'.
362 static int bpf_convert_filter(struct sock_filter *prog, int len,
363 struct bpf_insn *new_prog, int *new_len)
365 int new_flen = 0, pass = 0, target, i;
366 struct bpf_insn *new_insn;
367 struct sock_filter *fp;
371 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
372 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
374 if (len <= 0 || len > BPF_MAXINSNS)
378 addrs = kcalloc(len, sizeof(*addrs),
379 GFP_KERNEL | __GFP_NOWARN);
389 *new_insn = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
392 for (i = 0; i < len; fp++, i++) {
393 struct bpf_insn tmp_insns[6] = { };
394 struct bpf_insn *insn = tmp_insns;
397 addrs[i] = new_insn - new_prog;
400 /* All arithmetic insns and skb loads map as-is. */
401 case BPF_ALU | BPF_ADD | BPF_X:
402 case BPF_ALU | BPF_ADD | BPF_K:
403 case BPF_ALU | BPF_SUB | BPF_X:
404 case BPF_ALU | BPF_SUB | BPF_K:
405 case BPF_ALU | BPF_AND | BPF_X:
406 case BPF_ALU | BPF_AND | BPF_K:
407 case BPF_ALU | BPF_OR | BPF_X:
408 case BPF_ALU | BPF_OR | BPF_K:
409 case BPF_ALU | BPF_LSH | BPF_X:
410 case BPF_ALU | BPF_LSH | BPF_K:
411 case BPF_ALU | BPF_RSH | BPF_X:
412 case BPF_ALU | BPF_RSH | BPF_K:
413 case BPF_ALU | BPF_XOR | BPF_X:
414 case BPF_ALU | BPF_XOR | BPF_K:
415 case BPF_ALU | BPF_MUL | BPF_X:
416 case BPF_ALU | BPF_MUL | BPF_K:
417 case BPF_ALU | BPF_DIV | BPF_X:
418 case BPF_ALU | BPF_DIV | BPF_K:
419 case BPF_ALU | BPF_MOD | BPF_X:
420 case BPF_ALU | BPF_MOD | BPF_K:
421 case BPF_ALU | BPF_NEG:
422 case BPF_LD | BPF_ABS | BPF_W:
423 case BPF_LD | BPF_ABS | BPF_H:
424 case BPF_LD | BPF_ABS | BPF_B:
425 case BPF_LD | BPF_IND | BPF_W:
426 case BPF_LD | BPF_IND | BPF_H:
427 case BPF_LD | BPF_IND | BPF_B:
428 /* Check for overloaded BPF extension and
429 * directly convert it if found, otherwise
430 * just move on with mapping.
432 if (BPF_CLASS(fp->code) == BPF_LD &&
433 BPF_MODE(fp->code) == BPF_ABS &&
434 convert_bpf_extensions(fp, &insn))
437 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
440 /* Jump transformation cannot use BPF block macros
441 * everywhere as offset calculation and target updates
442 * require a bit more work than the rest, i.e. jump
443 * opcodes map as-is, but offsets need adjustment.
446 #define BPF_EMIT_JMP \
448 if (target >= len || target < 0) \
450 insn->off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
451 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
452 insn->off -= insn - tmp_insns; \
455 case BPF_JMP | BPF_JA:
456 target = i + fp->k + 1;
457 insn->code = fp->code;
461 case BPF_JMP | BPF_JEQ | BPF_K:
462 case BPF_JMP | BPF_JEQ | BPF_X:
463 case BPF_JMP | BPF_JSET | BPF_K:
464 case BPF_JMP | BPF_JSET | BPF_X:
465 case BPF_JMP | BPF_JGT | BPF_K:
466 case BPF_JMP | BPF_JGT | BPF_X:
467 case BPF_JMP | BPF_JGE | BPF_K:
468 case BPF_JMP | BPF_JGE | BPF_X:
469 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
470 /* BPF immediates are signed, zero extend
471 * immediate into tmp register and use it
474 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
476 insn->dst_reg = BPF_REG_A;
477 insn->src_reg = BPF_REG_TMP;
480 insn->dst_reg = BPF_REG_A;
482 bpf_src = BPF_SRC(fp->code);
483 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
486 /* Common case where 'jump_false' is next insn. */
488 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
489 target = i + fp->jt + 1;
494 /* Convert JEQ into JNE when 'jump_true' is next insn. */
495 if (fp->jt == 0 && BPF_OP(fp->code) == BPF_JEQ) {
496 insn->code = BPF_JMP | BPF_JNE | bpf_src;
497 target = i + fp->jf + 1;
502 /* Other jumps are mapped into two insns: Jxx and JA. */
503 target = i + fp->jt + 1;
504 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
508 insn->code = BPF_JMP | BPF_JA;
509 target = i + fp->jf + 1;
513 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
514 case BPF_LDX | BPF_MSH | BPF_B:
516 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_A);
517 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
518 *insn++ = BPF_LD_ABS(BPF_B, fp->k);
520 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
522 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
524 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
526 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
529 /* RET_K, RET_A are remaped into 2 insns. */
530 case BPF_RET | BPF_A:
531 case BPF_RET | BPF_K:
532 *insn++ = BPF_MOV32_RAW(BPF_RVAL(fp->code) == BPF_K ?
533 BPF_K : BPF_X, BPF_REG_0,
535 *insn = BPF_EXIT_INSN();
538 /* Store to stack. */
541 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
542 BPF_ST ? BPF_REG_A : BPF_REG_X,
543 -(BPF_MEMWORDS - fp->k) * 4);
546 /* Load from stack. */
547 case BPF_LD | BPF_MEM:
548 case BPF_LDX | BPF_MEM:
549 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
550 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
551 -(BPF_MEMWORDS - fp->k) * 4);
555 case BPF_LD | BPF_IMM:
556 case BPF_LDX | BPF_IMM:
557 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
558 BPF_REG_A : BPF_REG_X, fp->k);
562 case BPF_MISC | BPF_TAX:
563 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
567 case BPF_MISC | BPF_TXA:
568 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
571 /* A = skb->len or X = skb->len */
572 case BPF_LD | BPF_W | BPF_LEN:
573 case BPF_LDX | BPF_W | BPF_LEN:
574 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
575 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
576 offsetof(struct sk_buff, len));
579 /* Access seccomp_data fields. */
580 case BPF_LDX | BPF_ABS | BPF_W:
581 /* A = *(u32 *) (ctx + K) */
582 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
585 /* Unknown instruction. */
592 memcpy(new_insn, tmp_insns,
593 sizeof(*insn) * (insn - tmp_insns));
594 new_insn += insn - tmp_insns;
598 /* Only calculating new length. */
599 *new_len = new_insn - new_prog;
604 if (new_flen != new_insn - new_prog) {
605 new_flen = new_insn - new_prog;
612 BUG_ON(*new_len != new_flen);
621 * As we dont want to clear mem[] array for each packet going through
622 * __bpf_prog_run(), we check that filter loaded by user never try to read
623 * a cell if not previously written, and we check all branches to be sure
624 * a malicious user doesn't try to abuse us.
626 static int check_load_and_stores(const struct sock_filter *filter, int flen)
628 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
631 BUILD_BUG_ON(BPF_MEMWORDS > 16);
633 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
637 memset(masks, 0xff, flen * sizeof(*masks));
639 for (pc = 0; pc < flen; pc++) {
640 memvalid &= masks[pc];
642 switch (filter[pc].code) {
645 memvalid |= (1 << filter[pc].k);
647 case BPF_LD | BPF_MEM:
648 case BPF_LDX | BPF_MEM:
649 if (!(memvalid & (1 << filter[pc].k))) {
654 case BPF_JMP | BPF_JA:
655 /* A jump must set masks on target */
656 masks[pc + 1 + filter[pc].k] &= memvalid;
659 case BPF_JMP | BPF_JEQ | BPF_K:
660 case BPF_JMP | BPF_JEQ | BPF_X:
661 case BPF_JMP | BPF_JGE | BPF_K:
662 case BPF_JMP | BPF_JGE | BPF_X:
663 case BPF_JMP | BPF_JGT | BPF_K:
664 case BPF_JMP | BPF_JGT | BPF_X:
665 case BPF_JMP | BPF_JSET | BPF_K:
666 case BPF_JMP | BPF_JSET | BPF_X:
667 /* A jump must set masks on targets */
668 masks[pc + 1 + filter[pc].jt] &= memvalid;
669 masks[pc + 1 + filter[pc].jf] &= memvalid;
679 static bool chk_code_allowed(u16 code_to_probe)
681 static const bool codes[] = {
682 /* 32 bit ALU operations */
683 [BPF_ALU | BPF_ADD | BPF_K] = true,
684 [BPF_ALU | BPF_ADD | BPF_X] = true,
685 [BPF_ALU | BPF_SUB | BPF_K] = true,
686 [BPF_ALU | BPF_SUB | BPF_X] = true,
687 [BPF_ALU | BPF_MUL | BPF_K] = true,
688 [BPF_ALU | BPF_MUL | BPF_X] = true,
689 [BPF_ALU | BPF_DIV | BPF_K] = true,
690 [BPF_ALU | BPF_DIV | BPF_X] = true,
691 [BPF_ALU | BPF_MOD | BPF_K] = true,
692 [BPF_ALU | BPF_MOD | BPF_X] = true,
693 [BPF_ALU | BPF_AND | BPF_K] = true,
694 [BPF_ALU | BPF_AND | BPF_X] = true,
695 [BPF_ALU | BPF_OR | BPF_K] = true,
696 [BPF_ALU | BPF_OR | BPF_X] = true,
697 [BPF_ALU | BPF_XOR | BPF_K] = true,
698 [BPF_ALU | BPF_XOR | BPF_X] = true,
699 [BPF_ALU | BPF_LSH | BPF_K] = true,
700 [BPF_ALU | BPF_LSH | BPF_X] = true,
701 [BPF_ALU | BPF_RSH | BPF_K] = true,
702 [BPF_ALU | BPF_RSH | BPF_X] = true,
703 [BPF_ALU | BPF_NEG] = true,
704 /* Load instructions */
705 [BPF_LD | BPF_W | BPF_ABS] = true,
706 [BPF_LD | BPF_H | BPF_ABS] = true,
707 [BPF_LD | BPF_B | BPF_ABS] = true,
708 [BPF_LD | BPF_W | BPF_LEN] = true,
709 [BPF_LD | BPF_W | BPF_IND] = true,
710 [BPF_LD | BPF_H | BPF_IND] = true,
711 [BPF_LD | BPF_B | BPF_IND] = true,
712 [BPF_LD | BPF_IMM] = true,
713 [BPF_LD | BPF_MEM] = true,
714 [BPF_LDX | BPF_W | BPF_LEN] = true,
715 [BPF_LDX | BPF_B | BPF_MSH] = true,
716 [BPF_LDX | BPF_IMM] = true,
717 [BPF_LDX | BPF_MEM] = true,
718 /* Store instructions */
721 /* Misc instructions */
722 [BPF_MISC | BPF_TAX] = true,
723 [BPF_MISC | BPF_TXA] = true,
724 /* Return instructions */
725 [BPF_RET | BPF_K] = true,
726 [BPF_RET | BPF_A] = true,
727 /* Jump instructions */
728 [BPF_JMP | BPF_JA] = true,
729 [BPF_JMP | BPF_JEQ | BPF_K] = true,
730 [BPF_JMP | BPF_JEQ | BPF_X] = true,
731 [BPF_JMP | BPF_JGE | BPF_K] = true,
732 [BPF_JMP | BPF_JGE | BPF_X] = true,
733 [BPF_JMP | BPF_JGT | BPF_K] = true,
734 [BPF_JMP | BPF_JGT | BPF_X] = true,
735 [BPF_JMP | BPF_JSET | BPF_K] = true,
736 [BPF_JMP | BPF_JSET | BPF_X] = true,
739 if (code_to_probe >= ARRAY_SIZE(codes))
742 return codes[code_to_probe];
746 * bpf_check_classic - verify socket filter code
747 * @filter: filter to verify
748 * @flen: length of filter
750 * Check the user's filter code. If we let some ugly
751 * filter code slip through kaboom! The filter must contain
752 * no references or jumps that are out of range, no illegal
753 * instructions, and must end with a RET instruction.
755 * All jumps are forward as they are not signed.
757 * Returns 0 if the rule set is legal or -EINVAL if not.
759 static int bpf_check_classic(const struct sock_filter *filter,
765 if (flen == 0 || flen > BPF_MAXINSNS)
768 /* Check the filter code now */
769 for (pc = 0; pc < flen; pc++) {
770 const struct sock_filter *ftest = &filter[pc];
772 /* May we actually operate on this code? */
773 if (!chk_code_allowed(ftest->code))
776 /* Some instructions need special checks */
777 switch (ftest->code) {
778 case BPF_ALU | BPF_DIV | BPF_K:
779 case BPF_ALU | BPF_MOD | BPF_K:
780 /* Check for division by zero */
784 case BPF_LD | BPF_MEM:
785 case BPF_LDX | BPF_MEM:
788 /* Check for invalid memory addresses */
789 if (ftest->k >= BPF_MEMWORDS)
792 case BPF_JMP | BPF_JA:
793 /* Note, the large ftest->k might cause loops.
794 * Compare this with conditional jumps below,
795 * where offsets are limited. --ANK (981016)
797 if (ftest->k >= (unsigned int)(flen - pc - 1))
800 case BPF_JMP | BPF_JEQ | BPF_K:
801 case BPF_JMP | BPF_JEQ | BPF_X:
802 case BPF_JMP | BPF_JGE | BPF_K:
803 case BPF_JMP | BPF_JGE | BPF_X:
804 case BPF_JMP | BPF_JGT | BPF_K:
805 case BPF_JMP | BPF_JGT | BPF_X:
806 case BPF_JMP | BPF_JSET | BPF_K:
807 case BPF_JMP | BPF_JSET | BPF_X:
808 /* Both conditionals must be safe */
809 if (pc + ftest->jt + 1 >= flen ||
810 pc + ftest->jf + 1 >= flen)
813 case BPF_LD | BPF_W | BPF_ABS:
814 case BPF_LD | BPF_H | BPF_ABS:
815 case BPF_LD | BPF_B | BPF_ABS:
817 if (bpf_anc_helper(ftest) & BPF_ANC)
819 /* Ancillary operation unknown or unsupported */
820 if (anc_found == false && ftest->k >= SKF_AD_OFF)
825 /* Last instruction must be a RET code */
826 switch (filter[flen - 1].code) {
827 case BPF_RET | BPF_K:
828 case BPF_RET | BPF_A:
829 return check_load_and_stores(filter, flen);
835 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
836 const struct sock_fprog *fprog)
838 unsigned int fsize = bpf_classic_proglen(fprog);
839 struct sock_fprog_kern *fkprog;
841 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
845 fkprog = fp->orig_prog;
846 fkprog->len = fprog->len;
848 fkprog->filter = kmemdup(fp->insns, fsize,
849 GFP_KERNEL | __GFP_NOWARN);
850 if (!fkprog->filter) {
851 kfree(fp->orig_prog);
858 static void bpf_release_orig_filter(struct bpf_prog *fp)
860 struct sock_fprog_kern *fprog = fp->orig_prog;
863 kfree(fprog->filter);
868 static void __bpf_prog_release(struct bpf_prog *prog)
870 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
873 bpf_release_orig_filter(prog);
878 static void __sk_filter_release(struct sk_filter *fp)
880 __bpf_prog_release(fp->prog);
885 * sk_filter_release_rcu - Release a socket filter by rcu_head
886 * @rcu: rcu_head that contains the sk_filter to free
888 static void sk_filter_release_rcu(struct rcu_head *rcu)
890 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
892 __sk_filter_release(fp);
896 * sk_filter_release - release a socket filter
897 * @fp: filter to remove
899 * Remove a filter from a socket and release its resources.
901 static void sk_filter_release(struct sk_filter *fp)
903 if (atomic_dec_and_test(&fp->refcnt))
904 call_rcu(&fp->rcu, sk_filter_release_rcu);
907 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
909 u32 filter_size = bpf_prog_size(fp->prog->len);
911 atomic_sub(filter_size, &sk->sk_omem_alloc);
912 sk_filter_release(fp);
915 /* try to charge the socket memory if there is space available
916 * return true on success
918 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
920 u32 filter_size = bpf_prog_size(fp->prog->len);
922 /* same check as in sock_kmalloc() */
923 if (filter_size <= sysctl_optmem_max &&
924 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
925 atomic_inc(&fp->refcnt);
926 atomic_add(filter_size, &sk->sk_omem_alloc);
932 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
934 struct sock_filter *old_prog;
935 struct bpf_prog *old_fp;
936 int err, new_len, old_len = fp->len;
938 /* We are free to overwrite insns et al right here as it
939 * won't be used at this point in time anymore internally
940 * after the migration to the internal BPF instruction
943 BUILD_BUG_ON(sizeof(struct sock_filter) !=
944 sizeof(struct bpf_insn));
946 /* Conversion cannot happen on overlapping memory areas,
947 * so we need to keep the user BPF around until the 2nd
948 * pass. At this time, the user BPF is stored in fp->insns.
950 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
951 GFP_KERNEL | __GFP_NOWARN);
957 /* 1st pass: calculate the new program length. */
958 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len);
962 /* Expand fp for appending the new filter representation. */
964 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
966 /* The old_fp is still around in case we couldn't
967 * allocate new memory, so uncharge on that one.
976 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
977 err = bpf_convert_filter(old_prog, old_len, fp->insnsi, &new_len);
979 /* 2nd bpf_convert_filter() can fail only if it fails
980 * to allocate memory, remapping must succeed. Note,
981 * that at this time old_fp has already been released
986 bpf_prog_select_runtime(fp);
994 __bpf_prog_release(fp);
998 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
999 bpf_aux_classic_check_t trans)
1003 fp->bpf_func = NULL;
1006 err = bpf_check_classic(fp->insns, fp->len);
1008 __bpf_prog_release(fp);
1009 return ERR_PTR(err);
1012 /* There might be additional checks and transformations
1013 * needed on classic filters, f.e. in case of seccomp.
1016 err = trans(fp->insns, fp->len);
1018 __bpf_prog_release(fp);
1019 return ERR_PTR(err);
1023 /* Probe if we can JIT compile the filter and if so, do
1024 * the compilation of the filter.
1026 bpf_jit_compile(fp);
1028 /* JIT compiler couldn't process this filter, so do the
1029 * internal BPF translation for the optimized interpreter.
1032 fp = bpf_migrate_filter(fp);
1038 * bpf_prog_create - create an unattached filter
1039 * @pfp: the unattached filter that is created
1040 * @fprog: the filter program
1042 * Create a filter independent of any socket. We first run some
1043 * sanity checks on it to make sure it does not explode on us later.
1044 * If an error occurs or there is insufficient memory for the filter
1045 * a negative errno code is returned. On success the return is zero.
1047 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1049 unsigned int fsize = bpf_classic_proglen(fprog);
1050 struct bpf_prog *fp;
1052 /* Make sure new filter is there and in the right amounts. */
1053 if (fprog->filter == NULL)
1056 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1060 memcpy(fp->insns, fprog->filter, fsize);
1062 fp->len = fprog->len;
1063 /* Since unattached filters are not copied back to user
1064 * space through sk_get_filter(), we do not need to hold
1065 * a copy here, and can spare us the work.
1067 fp->orig_prog = NULL;
1069 /* bpf_prepare_filter() already takes care of freeing
1070 * memory in case something goes wrong.
1072 fp = bpf_prepare_filter(fp, NULL);
1079 EXPORT_SYMBOL_GPL(bpf_prog_create);
1082 * bpf_prog_create_from_user - create an unattached filter from user buffer
1083 * @pfp: the unattached filter that is created
1084 * @fprog: the filter program
1085 * @trans: post-classic verifier transformation handler
1087 * This function effectively does the same as bpf_prog_create(), only
1088 * that it builds up its insns buffer from user space provided buffer.
1089 * It also allows for passing a bpf_aux_classic_check_t handler.
1091 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1092 bpf_aux_classic_check_t trans)
1094 unsigned int fsize = bpf_classic_proglen(fprog);
1095 struct bpf_prog *fp;
1097 /* Make sure new filter is there and in the right amounts. */
1098 if (fprog->filter == NULL)
1101 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1105 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1106 __bpf_prog_free(fp);
1110 fp->len = fprog->len;
1111 /* Since unattached filters are not copied back to user
1112 * space through sk_get_filter(), we do not need to hold
1113 * a copy here, and can spare us the work.
1115 fp->orig_prog = NULL;
1117 /* bpf_prepare_filter() already takes care of freeing
1118 * memory in case something goes wrong.
1120 fp = bpf_prepare_filter(fp, trans);
1127 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1129 void bpf_prog_destroy(struct bpf_prog *fp)
1131 __bpf_prog_release(fp);
1133 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1135 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1137 struct sk_filter *fp, *old_fp;
1139 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1144 atomic_set(&fp->refcnt, 0);
1146 if (!sk_filter_charge(sk, fp)) {
1151 old_fp = rcu_dereference_protected(sk->sk_filter,
1152 sock_owned_by_user(sk));
1153 rcu_assign_pointer(sk->sk_filter, fp);
1156 sk_filter_uncharge(sk, old_fp);
1162 * sk_attach_filter - attach a socket filter
1163 * @fprog: the filter program
1164 * @sk: the socket to use
1166 * Attach the user's filter code. We first run some sanity checks on
1167 * it to make sure it does not explode on us later. If an error
1168 * occurs or there is insufficient memory for the filter a negative
1169 * errno code is returned. On success the return is zero.
1171 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1173 unsigned int fsize = bpf_classic_proglen(fprog);
1174 unsigned int bpf_fsize = bpf_prog_size(fprog->len);
1175 struct bpf_prog *prog;
1178 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1181 /* Make sure new filter is there and in the right amounts. */
1182 if (fprog->filter == NULL)
1185 prog = bpf_prog_alloc(bpf_fsize, 0);
1189 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1190 __bpf_prog_free(prog);
1194 prog->len = fprog->len;
1196 err = bpf_prog_store_orig_filter(prog, fprog);
1198 __bpf_prog_free(prog);
1202 /* bpf_prepare_filter() already takes care of freeing
1203 * memory in case something goes wrong.
1205 prog = bpf_prepare_filter(prog, NULL);
1207 return PTR_ERR(prog);
1209 err = __sk_attach_prog(prog, sk);
1211 __bpf_prog_release(prog);
1217 EXPORT_SYMBOL_GPL(sk_attach_filter);
1219 int sk_attach_bpf(u32 ufd, struct sock *sk)
1221 struct bpf_prog *prog;
1224 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1227 prog = bpf_prog_get(ufd);
1229 return PTR_ERR(prog);
1231 if (prog->type != BPF_PROG_TYPE_SOCKET_FILTER) {
1236 err = __sk_attach_prog(prog, sk);
1245 #define BPF_RECOMPUTE_CSUM(flags) ((flags) & 1)
1247 static u64 bpf_skb_store_bytes(u64 r1, u64 r2, u64 r3, u64 r4, u64 flags)
1249 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1250 int offset = (int) r2;
1251 void *from = (void *) (long) r3;
1252 unsigned int len = (unsigned int) r4;
1256 /* bpf verifier guarantees that:
1257 * 'from' pointer points to bpf program stack
1258 * 'len' bytes of it were initialized
1260 * 'skb' is a valid pointer to 'struct sk_buff'
1262 * so check for invalid 'offset' and too large 'len'
1264 if (unlikely((u32) offset > 0xffff || len > sizeof(buf)))
1267 if (unlikely(skb_cloned(skb) &&
1268 !skb_clone_writable(skb, offset + len)))
1271 ptr = skb_header_pointer(skb, offset, len, buf);
1275 if (BPF_RECOMPUTE_CSUM(flags))
1276 skb_postpull_rcsum(skb, ptr, len);
1278 memcpy(ptr, from, len);
1281 /* skb_store_bits cannot return -EFAULT here */
1282 skb_store_bits(skb, offset, ptr, len);
1284 if (BPF_RECOMPUTE_CSUM(flags) && skb->ip_summed == CHECKSUM_COMPLETE)
1285 skb->csum = csum_add(skb->csum, csum_partial(ptr, len, 0));
1289 const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1290 .func = bpf_skb_store_bytes,
1292 .ret_type = RET_INTEGER,
1293 .arg1_type = ARG_PTR_TO_CTX,
1294 .arg2_type = ARG_ANYTHING,
1295 .arg3_type = ARG_PTR_TO_STACK,
1296 .arg4_type = ARG_CONST_STACK_SIZE,
1297 .arg5_type = ARG_ANYTHING,
1300 #define BPF_HEADER_FIELD_SIZE(flags) ((flags) & 0x0f)
1301 #define BPF_IS_PSEUDO_HEADER(flags) ((flags) & 0x10)
1303 static u64 bpf_l3_csum_replace(u64 r1, u64 r2, u64 from, u64 to, u64 flags)
1305 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1306 int offset = (int) r2;
1309 if (unlikely((u32) offset > 0xffff))
1312 if (unlikely(skb_cloned(skb) &&
1313 !skb_clone_writable(skb, offset + sizeof(sum))))
1316 ptr = skb_header_pointer(skb, offset, sizeof(sum), &sum);
1320 switch (BPF_HEADER_FIELD_SIZE(flags)) {
1322 csum_replace2(ptr, from, to);
1325 csum_replace4(ptr, from, to);
1332 /* skb_store_bits guaranteed to not return -EFAULT here */
1333 skb_store_bits(skb, offset, ptr, sizeof(sum));
1338 const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1339 .func = bpf_l3_csum_replace,
1341 .ret_type = RET_INTEGER,
1342 .arg1_type = ARG_PTR_TO_CTX,
1343 .arg2_type = ARG_ANYTHING,
1344 .arg3_type = ARG_ANYTHING,
1345 .arg4_type = ARG_ANYTHING,
1346 .arg5_type = ARG_ANYTHING,
1349 static u64 bpf_l4_csum_replace(u64 r1, u64 r2, u64 from, u64 to, u64 flags)
1351 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1352 bool is_pseudo = !!BPF_IS_PSEUDO_HEADER(flags);
1353 int offset = (int) r2;
1356 if (unlikely((u32) offset > 0xffff))
1359 if (unlikely(skb_cloned(skb) &&
1360 !skb_clone_writable(skb, offset + sizeof(sum))))
1363 ptr = skb_header_pointer(skb, offset, sizeof(sum), &sum);
1367 switch (BPF_HEADER_FIELD_SIZE(flags)) {
1369 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1372 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1379 /* skb_store_bits guaranteed to not return -EFAULT here */
1380 skb_store_bits(skb, offset, ptr, sizeof(sum));
1385 const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1386 .func = bpf_l4_csum_replace,
1388 .ret_type = RET_INTEGER,
1389 .arg1_type = ARG_PTR_TO_CTX,
1390 .arg2_type = ARG_ANYTHING,
1391 .arg3_type = ARG_ANYTHING,
1392 .arg4_type = ARG_ANYTHING,
1393 .arg5_type = ARG_ANYTHING,
1396 #define BPF_IS_REDIRECT_INGRESS(flags) ((flags) & 1)
1398 static u64 bpf_clone_redirect(u64 r1, u64 ifindex, u64 flags, u64 r4, u64 r5)
1400 struct sk_buff *skb = (struct sk_buff *) (long) r1, *skb2;
1401 struct net_device *dev;
1403 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
1407 if (unlikely(!(dev->flags & IFF_UP)))
1410 skb2 = skb_clone(skb, GFP_ATOMIC);
1411 if (unlikely(!skb2))
1414 if (BPF_IS_REDIRECT_INGRESS(flags))
1415 return dev_forward_skb(dev, skb2);
1418 skb_sender_cpu_clear(skb2);
1419 return dev_queue_xmit(skb2);
1422 const struct bpf_func_proto bpf_clone_redirect_proto = {
1423 .func = bpf_clone_redirect,
1425 .ret_type = RET_INTEGER,
1426 .arg1_type = ARG_PTR_TO_CTX,
1427 .arg2_type = ARG_ANYTHING,
1428 .arg3_type = ARG_ANYTHING,
1431 static u64 bpf_get_cgroup_classid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1433 return task_get_classid((struct sk_buff *) (unsigned long) r1);
1436 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
1437 .func = bpf_get_cgroup_classid,
1439 .ret_type = RET_INTEGER,
1440 .arg1_type = ARG_PTR_TO_CTX,
1443 static u64 bpf_skb_vlan_push(u64 r1, u64 r2, u64 vlan_tci, u64 r4, u64 r5)
1445 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1446 __be16 vlan_proto = (__force __be16) r2;
1448 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
1449 vlan_proto != htons(ETH_P_8021AD)))
1450 vlan_proto = htons(ETH_P_8021Q);
1452 return skb_vlan_push(skb, vlan_proto, vlan_tci);
1455 const struct bpf_func_proto bpf_skb_vlan_push_proto = {
1456 .func = bpf_skb_vlan_push,
1458 .ret_type = RET_INTEGER,
1459 .arg1_type = ARG_PTR_TO_CTX,
1460 .arg2_type = ARG_ANYTHING,
1461 .arg3_type = ARG_ANYTHING,
1463 EXPORT_SYMBOL_GPL(bpf_skb_vlan_push_proto);
1465 static u64 bpf_skb_vlan_pop(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1467 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1469 return skb_vlan_pop(skb);
1472 const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
1473 .func = bpf_skb_vlan_pop,
1475 .ret_type = RET_INTEGER,
1476 .arg1_type = ARG_PTR_TO_CTX,
1478 EXPORT_SYMBOL_GPL(bpf_skb_vlan_pop_proto);
1480 bool bpf_helper_changes_skb_data(void *func)
1482 if (func == bpf_skb_vlan_push)
1484 if (func == bpf_skb_vlan_pop)
1489 static u64 bpf_skb_get_tunnel_key(u64 r1, u64 r2, u64 size, u64 flags, u64 r5)
1491 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1492 struct bpf_tunnel_key *to = (struct bpf_tunnel_key *) (long) r2;
1493 struct ip_tunnel_info *info = skb_tunnel_info(skb);
1495 if (unlikely(size != sizeof(struct bpf_tunnel_key) || flags || !info))
1497 if (ip_tunnel_info_af(info) != AF_INET)
1500 to->tunnel_id = be64_to_cpu(info->key.tun_id);
1501 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
1506 const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
1507 .func = bpf_skb_get_tunnel_key,
1509 .ret_type = RET_INTEGER,
1510 .arg1_type = ARG_PTR_TO_CTX,
1511 .arg2_type = ARG_PTR_TO_STACK,
1512 .arg3_type = ARG_CONST_STACK_SIZE,
1513 .arg4_type = ARG_ANYTHING,
1516 static struct metadata_dst __percpu *md_dst;
1518 static u64 bpf_skb_set_tunnel_key(u64 r1, u64 r2, u64 size, u64 flags, u64 r5)
1520 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1521 struct bpf_tunnel_key *from = (struct bpf_tunnel_key *) (long) r2;
1522 struct metadata_dst *md = this_cpu_ptr(md_dst);
1523 struct ip_tunnel_info *info;
1525 if (unlikely(size != sizeof(struct bpf_tunnel_key) || flags))
1529 dst_hold((struct dst_entry *) md);
1530 skb_dst_set(skb, (struct dst_entry *) md);
1532 info = &md->u.tun_info;
1533 info->mode = IP_TUNNEL_INFO_TX;
1534 info->key.tun_flags = TUNNEL_KEY;
1535 info->key.tun_id = cpu_to_be64(from->tunnel_id);
1536 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
1541 const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
1542 .func = bpf_skb_set_tunnel_key,
1544 .ret_type = RET_INTEGER,
1545 .arg1_type = ARG_PTR_TO_CTX,
1546 .arg2_type = ARG_PTR_TO_STACK,
1547 .arg3_type = ARG_CONST_STACK_SIZE,
1548 .arg4_type = ARG_ANYTHING,
1551 static const struct bpf_func_proto *bpf_get_skb_set_tunnel_key_proto(void)
1554 /* race is not possible, since it's called from
1555 * verifier that is holding verifier mutex
1557 md_dst = metadata_dst_alloc_percpu(0, GFP_KERNEL);
1561 return &bpf_skb_set_tunnel_key_proto;
1564 static const struct bpf_func_proto *
1565 sk_filter_func_proto(enum bpf_func_id func_id)
1568 case BPF_FUNC_map_lookup_elem:
1569 return &bpf_map_lookup_elem_proto;
1570 case BPF_FUNC_map_update_elem:
1571 return &bpf_map_update_elem_proto;
1572 case BPF_FUNC_map_delete_elem:
1573 return &bpf_map_delete_elem_proto;
1574 case BPF_FUNC_get_prandom_u32:
1575 return &bpf_get_prandom_u32_proto;
1576 case BPF_FUNC_get_smp_processor_id:
1577 return &bpf_get_smp_processor_id_proto;
1578 case BPF_FUNC_tail_call:
1579 return &bpf_tail_call_proto;
1580 case BPF_FUNC_ktime_get_ns:
1581 return &bpf_ktime_get_ns_proto;
1582 case BPF_FUNC_trace_printk:
1583 return bpf_get_trace_printk_proto();
1589 static const struct bpf_func_proto *
1590 tc_cls_act_func_proto(enum bpf_func_id func_id)
1593 case BPF_FUNC_skb_store_bytes:
1594 return &bpf_skb_store_bytes_proto;
1595 case BPF_FUNC_l3_csum_replace:
1596 return &bpf_l3_csum_replace_proto;
1597 case BPF_FUNC_l4_csum_replace:
1598 return &bpf_l4_csum_replace_proto;
1599 case BPF_FUNC_clone_redirect:
1600 return &bpf_clone_redirect_proto;
1601 case BPF_FUNC_get_cgroup_classid:
1602 return &bpf_get_cgroup_classid_proto;
1603 case BPF_FUNC_skb_vlan_push:
1604 return &bpf_skb_vlan_push_proto;
1605 case BPF_FUNC_skb_vlan_pop:
1606 return &bpf_skb_vlan_pop_proto;
1607 case BPF_FUNC_skb_get_tunnel_key:
1608 return &bpf_skb_get_tunnel_key_proto;
1609 case BPF_FUNC_skb_set_tunnel_key:
1610 return bpf_get_skb_set_tunnel_key_proto();
1612 return sk_filter_func_proto(func_id);
1616 static bool __is_valid_access(int off, int size, enum bpf_access_type type)
1619 if (off < 0 || off >= sizeof(struct __sk_buff))
1622 /* disallow misaligned access */
1623 if (off % size != 0)
1626 /* all __sk_buff fields are __u32 */
1633 static bool sk_filter_is_valid_access(int off, int size,
1634 enum bpf_access_type type)
1636 if (type == BPF_WRITE) {
1638 case offsetof(struct __sk_buff, cb[0]) ...
1639 offsetof(struct __sk_buff, cb[4]):
1646 return __is_valid_access(off, size, type);
1649 static bool tc_cls_act_is_valid_access(int off, int size,
1650 enum bpf_access_type type)
1652 if (type == BPF_WRITE) {
1654 case offsetof(struct __sk_buff, mark):
1655 case offsetof(struct __sk_buff, tc_index):
1656 case offsetof(struct __sk_buff, cb[0]) ...
1657 offsetof(struct __sk_buff, cb[4]):
1663 return __is_valid_access(off, size, type);
1666 static u32 bpf_net_convert_ctx_access(enum bpf_access_type type, int dst_reg,
1667 int src_reg, int ctx_off,
1668 struct bpf_insn *insn_buf)
1670 struct bpf_insn *insn = insn_buf;
1673 case offsetof(struct __sk_buff, len):
1674 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, len) != 4);
1676 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
1677 offsetof(struct sk_buff, len));
1680 case offsetof(struct __sk_buff, protocol):
1681 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
1683 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
1684 offsetof(struct sk_buff, protocol));
1687 case offsetof(struct __sk_buff, vlan_proto):
1688 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2);
1690 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
1691 offsetof(struct sk_buff, vlan_proto));
1694 case offsetof(struct __sk_buff, priority):
1695 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, priority) != 4);
1697 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
1698 offsetof(struct sk_buff, priority));
1701 case offsetof(struct __sk_buff, ingress_ifindex):
1702 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, skb_iif) != 4);
1704 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
1705 offsetof(struct sk_buff, skb_iif));
1708 case offsetof(struct __sk_buff, ifindex):
1709 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
1711 *insn++ = BPF_LDX_MEM(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, dev)),
1713 offsetof(struct sk_buff, dev));
1714 *insn++ = BPF_JMP_IMM(BPF_JEQ, dst_reg, 0, 1);
1715 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, dst_reg,
1716 offsetof(struct net_device, ifindex));
1719 case offsetof(struct __sk_buff, hash):
1720 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
1722 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
1723 offsetof(struct sk_buff, hash));
1726 case offsetof(struct __sk_buff, mark):
1727 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
1729 if (type == BPF_WRITE)
1730 *insn++ = BPF_STX_MEM(BPF_W, dst_reg, src_reg,
1731 offsetof(struct sk_buff, mark));
1733 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
1734 offsetof(struct sk_buff, mark));
1737 case offsetof(struct __sk_buff, pkt_type):
1738 return convert_skb_access(SKF_AD_PKTTYPE, dst_reg, src_reg, insn);
1740 case offsetof(struct __sk_buff, queue_mapping):
1741 return convert_skb_access(SKF_AD_QUEUE, dst_reg, src_reg, insn);
1743 case offsetof(struct __sk_buff, vlan_present):
1744 return convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
1745 dst_reg, src_reg, insn);
1747 case offsetof(struct __sk_buff, vlan_tci):
1748 return convert_skb_access(SKF_AD_VLAN_TAG,
1749 dst_reg, src_reg, insn);
1751 case offsetof(struct __sk_buff, cb[0]) ...
1752 offsetof(struct __sk_buff, cb[4]):
1753 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20);
1755 ctx_off -= offsetof(struct __sk_buff, cb[0]);
1756 ctx_off += offsetof(struct sk_buff, cb);
1757 ctx_off += offsetof(struct qdisc_skb_cb, data);
1758 if (type == BPF_WRITE)
1759 *insn++ = BPF_STX_MEM(BPF_W, dst_reg, src_reg, ctx_off);
1761 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, ctx_off);
1764 case offsetof(struct __sk_buff, tc_index):
1765 #ifdef CONFIG_NET_SCHED
1766 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, tc_index) != 2);
1768 if (type == BPF_WRITE)
1769 *insn++ = BPF_STX_MEM(BPF_H, dst_reg, src_reg,
1770 offsetof(struct sk_buff, tc_index));
1772 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
1773 offsetof(struct sk_buff, tc_index));
1776 if (type == BPF_WRITE)
1777 *insn++ = BPF_MOV64_REG(dst_reg, dst_reg);
1779 *insn++ = BPF_MOV64_IMM(dst_reg, 0);
1784 return insn - insn_buf;
1787 static const struct bpf_verifier_ops sk_filter_ops = {
1788 .get_func_proto = sk_filter_func_proto,
1789 .is_valid_access = sk_filter_is_valid_access,
1790 .convert_ctx_access = bpf_net_convert_ctx_access,
1793 static const struct bpf_verifier_ops tc_cls_act_ops = {
1794 .get_func_proto = tc_cls_act_func_proto,
1795 .is_valid_access = tc_cls_act_is_valid_access,
1796 .convert_ctx_access = bpf_net_convert_ctx_access,
1799 static struct bpf_prog_type_list sk_filter_type __read_mostly = {
1800 .ops = &sk_filter_ops,
1801 .type = BPF_PROG_TYPE_SOCKET_FILTER,
1804 static struct bpf_prog_type_list sched_cls_type __read_mostly = {
1805 .ops = &tc_cls_act_ops,
1806 .type = BPF_PROG_TYPE_SCHED_CLS,
1809 static struct bpf_prog_type_list sched_act_type __read_mostly = {
1810 .ops = &tc_cls_act_ops,
1811 .type = BPF_PROG_TYPE_SCHED_ACT,
1814 static int __init register_sk_filter_ops(void)
1816 bpf_register_prog_type(&sk_filter_type);
1817 bpf_register_prog_type(&sched_cls_type);
1818 bpf_register_prog_type(&sched_act_type);
1822 late_initcall(register_sk_filter_ops);
1824 int sk_detach_filter(struct sock *sk)
1827 struct sk_filter *filter;
1829 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1832 filter = rcu_dereference_protected(sk->sk_filter,
1833 sock_owned_by_user(sk));
1835 RCU_INIT_POINTER(sk->sk_filter, NULL);
1836 sk_filter_uncharge(sk, filter);
1842 EXPORT_SYMBOL_GPL(sk_detach_filter);
1844 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
1847 struct sock_fprog_kern *fprog;
1848 struct sk_filter *filter;
1852 filter = rcu_dereference_protected(sk->sk_filter,
1853 sock_owned_by_user(sk));
1857 /* We're copying the filter that has been originally attached,
1858 * so no conversion/decode needed anymore. eBPF programs that
1859 * have no original program cannot be dumped through this.
1862 fprog = filter->prog->orig_prog;
1868 /* User space only enquires number of filter blocks. */
1872 if (len < fprog->len)
1876 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
1879 /* Instead of bytes, the API requests to return the number