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 <linux/errno.h>
40 #include <linux/timer.h>
41 #include <asm/uaccess.h>
42 #include <asm/unaligned.h>
43 #include <linux/filter.h>
44 #include <linux/ratelimit.h>
45 #include <linux/seccomp.h>
46 #include <linux/if_vlan.h>
47 #include <linux/bpf.h>
50 * sk_filter - run a packet through a socket filter
51 * @sk: sock associated with &sk_buff
52 * @skb: buffer to filter
54 * Run the filter code and then cut skb->data to correct size returned by
55 * SK_RUN_FILTER. If pkt_len is 0 we toss packet. If skb->len is smaller
56 * than pkt_len we keep whole skb->data. This is the socket level
57 * wrapper to SK_RUN_FILTER. It returns 0 if the packet should
58 * be accepted or -EPERM if the packet should be tossed.
61 int sk_filter(struct sock *sk, struct sk_buff *skb)
64 struct sk_filter *filter;
67 * If the skb was allocated from pfmemalloc reserves, only
68 * allow SOCK_MEMALLOC sockets to use it as this socket is
71 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
74 err = security_sock_rcv_skb(sk, skb);
79 filter = rcu_dereference(sk->sk_filter);
81 unsigned int pkt_len = SK_RUN_FILTER(filter, skb);
83 err = pkt_len ? pskb_trim(skb, pkt_len) : -EPERM;
89 EXPORT_SYMBOL(sk_filter);
91 static u64 __skb_get_pay_offset(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
93 return skb_get_poff((struct sk_buff *)(unsigned long) ctx);
96 static u64 __skb_get_nlattr(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
98 struct sk_buff *skb = (struct sk_buff *)(unsigned long) ctx;
101 if (skb_is_nonlinear(skb))
104 if (skb->len < sizeof(struct nlattr))
107 if (a > skb->len - sizeof(struct nlattr))
110 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
112 return (void *) nla - (void *) skb->data;
117 static u64 __skb_get_nlattr_nest(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
119 struct sk_buff *skb = (struct sk_buff *)(unsigned long) ctx;
122 if (skb_is_nonlinear(skb))
125 if (skb->len < sizeof(struct nlattr))
128 if (a > skb->len - sizeof(struct nlattr))
131 nla = (struct nlattr *) &skb->data[a];
132 if (nla->nla_len > skb->len - a)
135 nla = nla_find_nested(nla, x);
137 return (void *) nla - (void *) skb->data;
142 static u64 __get_raw_cpu_id(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
144 return raw_smp_processor_id();
147 /* note that this only generates 32-bit random numbers */
148 static u64 __get_random_u32(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
150 return prandom_u32();
153 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
154 struct bpf_insn *insn_buf)
156 struct bpf_insn *insn = insn_buf;
160 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
162 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
163 offsetof(struct sk_buff, mark));
167 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
168 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
169 #ifdef __BIG_ENDIAN_BITFIELD
170 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
175 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2);
177 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
178 offsetof(struct sk_buff, queue_mapping));
181 case SKF_AD_VLAN_TAG:
182 case SKF_AD_VLAN_TAG_PRESENT:
183 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2);
184 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000);
186 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
187 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
188 offsetof(struct sk_buff, vlan_tci));
189 if (skb_field == SKF_AD_VLAN_TAG) {
190 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg,
194 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 12);
196 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
201 return insn - insn_buf;
204 static bool convert_bpf_extensions(struct sock_filter *fp,
205 struct bpf_insn **insnp)
207 struct bpf_insn *insn = *insnp;
211 case SKF_AD_OFF + SKF_AD_PROTOCOL:
212 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
214 /* A = *(u16 *) (CTX + offsetof(protocol)) */
215 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
216 offsetof(struct sk_buff, protocol));
217 /* A = ntohs(A) [emitting a nop or swap16] */
218 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
221 case SKF_AD_OFF + SKF_AD_PKTTYPE:
222 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
226 case SKF_AD_OFF + SKF_AD_IFINDEX:
227 case SKF_AD_OFF + SKF_AD_HATYPE:
228 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
229 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, type) != 2);
230 BUILD_BUG_ON(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, dev)) < 0);
232 *insn++ = BPF_LDX_MEM(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, dev)),
233 BPF_REG_TMP, BPF_REG_CTX,
234 offsetof(struct sk_buff, dev));
235 /* if (tmp != 0) goto pc + 1 */
236 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
237 *insn++ = BPF_EXIT_INSN();
238 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
239 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
240 offsetof(struct net_device, ifindex));
242 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
243 offsetof(struct net_device, type));
246 case SKF_AD_OFF + SKF_AD_MARK:
247 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
251 case SKF_AD_OFF + SKF_AD_RXHASH:
252 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
254 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
255 offsetof(struct sk_buff, hash));
258 case SKF_AD_OFF + SKF_AD_QUEUE:
259 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
263 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
264 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
265 BPF_REG_A, BPF_REG_CTX, insn);
269 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
270 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
271 BPF_REG_A, BPF_REG_CTX, insn);
275 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
276 case SKF_AD_OFF + SKF_AD_NLATTR:
277 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
278 case SKF_AD_OFF + SKF_AD_CPU:
279 case SKF_AD_OFF + SKF_AD_RANDOM:
281 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
283 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
285 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
286 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
288 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
289 *insn = BPF_EMIT_CALL(__skb_get_pay_offset);
291 case SKF_AD_OFF + SKF_AD_NLATTR:
292 *insn = BPF_EMIT_CALL(__skb_get_nlattr);
294 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
295 *insn = BPF_EMIT_CALL(__skb_get_nlattr_nest);
297 case SKF_AD_OFF + SKF_AD_CPU:
298 *insn = BPF_EMIT_CALL(__get_raw_cpu_id);
300 case SKF_AD_OFF + SKF_AD_RANDOM:
301 *insn = BPF_EMIT_CALL(__get_random_u32);
306 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
308 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
312 /* This is just a dummy call to avoid letting the compiler
313 * evict __bpf_call_base() as an optimization. Placed here
314 * where no-one bothers.
316 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
325 * bpf_convert_filter - convert filter program
326 * @prog: the user passed filter program
327 * @len: the length of the user passed filter program
328 * @new_prog: buffer where converted program will be stored
329 * @new_len: pointer to store length of converted program
331 * Remap 'sock_filter' style BPF instruction set to 'sock_filter_ext' style.
332 * Conversion workflow:
334 * 1) First pass for calculating the new program length:
335 * bpf_convert_filter(old_prog, old_len, NULL, &new_len)
337 * 2) 2nd pass to remap in two passes: 1st pass finds new
338 * jump offsets, 2nd pass remapping:
339 * new_prog = kmalloc(sizeof(struct bpf_insn) * new_len);
340 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len);
342 * User BPF's register A is mapped to our BPF register 6, user BPF
343 * register X is mapped to BPF register 7; frame pointer is always
344 * register 10; Context 'void *ctx' is stored in register 1, that is,
345 * for socket filters: ctx == 'struct sk_buff *', for seccomp:
346 * ctx == 'struct seccomp_data *'.
348 int bpf_convert_filter(struct sock_filter *prog, int len,
349 struct bpf_insn *new_prog, int *new_len)
351 int new_flen = 0, pass = 0, target, i;
352 struct bpf_insn *new_insn;
353 struct sock_filter *fp;
357 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
358 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
360 if (len <= 0 || len > BPF_MAXINSNS)
364 addrs = kcalloc(len, sizeof(*addrs), GFP_KERNEL);
374 *new_insn = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
377 for (i = 0; i < len; fp++, i++) {
378 struct bpf_insn tmp_insns[6] = { };
379 struct bpf_insn *insn = tmp_insns;
382 addrs[i] = new_insn - new_prog;
385 /* All arithmetic insns and skb loads map as-is. */
386 case BPF_ALU | BPF_ADD | BPF_X:
387 case BPF_ALU | BPF_ADD | BPF_K:
388 case BPF_ALU | BPF_SUB | BPF_X:
389 case BPF_ALU | BPF_SUB | BPF_K:
390 case BPF_ALU | BPF_AND | BPF_X:
391 case BPF_ALU | BPF_AND | BPF_K:
392 case BPF_ALU | BPF_OR | BPF_X:
393 case BPF_ALU | BPF_OR | BPF_K:
394 case BPF_ALU | BPF_LSH | BPF_X:
395 case BPF_ALU | BPF_LSH | BPF_K:
396 case BPF_ALU | BPF_RSH | BPF_X:
397 case BPF_ALU | BPF_RSH | BPF_K:
398 case BPF_ALU | BPF_XOR | BPF_X:
399 case BPF_ALU | BPF_XOR | BPF_K:
400 case BPF_ALU | BPF_MUL | BPF_X:
401 case BPF_ALU | BPF_MUL | BPF_K:
402 case BPF_ALU | BPF_DIV | BPF_X:
403 case BPF_ALU | BPF_DIV | BPF_K:
404 case BPF_ALU | BPF_MOD | BPF_X:
405 case BPF_ALU | BPF_MOD | BPF_K:
406 case BPF_ALU | BPF_NEG:
407 case BPF_LD | BPF_ABS | BPF_W:
408 case BPF_LD | BPF_ABS | BPF_H:
409 case BPF_LD | BPF_ABS | BPF_B:
410 case BPF_LD | BPF_IND | BPF_W:
411 case BPF_LD | BPF_IND | BPF_H:
412 case BPF_LD | BPF_IND | BPF_B:
413 /* Check for overloaded BPF extension and
414 * directly convert it if found, otherwise
415 * just move on with mapping.
417 if (BPF_CLASS(fp->code) == BPF_LD &&
418 BPF_MODE(fp->code) == BPF_ABS &&
419 convert_bpf_extensions(fp, &insn))
422 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
425 /* Jump transformation cannot use BPF block macros
426 * everywhere as offset calculation and target updates
427 * require a bit more work than the rest, i.e. jump
428 * opcodes map as-is, but offsets need adjustment.
431 #define BPF_EMIT_JMP \
433 if (target >= len || target < 0) \
435 insn->off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
436 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
437 insn->off -= insn - tmp_insns; \
440 case BPF_JMP | BPF_JA:
441 target = i + fp->k + 1;
442 insn->code = fp->code;
446 case BPF_JMP | BPF_JEQ | BPF_K:
447 case BPF_JMP | BPF_JEQ | BPF_X:
448 case BPF_JMP | BPF_JSET | BPF_K:
449 case BPF_JMP | BPF_JSET | BPF_X:
450 case BPF_JMP | BPF_JGT | BPF_K:
451 case BPF_JMP | BPF_JGT | BPF_X:
452 case BPF_JMP | BPF_JGE | BPF_K:
453 case BPF_JMP | BPF_JGE | BPF_X:
454 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
455 /* BPF immediates are signed, zero extend
456 * immediate into tmp register and use it
459 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
461 insn->dst_reg = BPF_REG_A;
462 insn->src_reg = BPF_REG_TMP;
465 insn->dst_reg = BPF_REG_A;
466 insn->src_reg = BPF_REG_X;
468 bpf_src = BPF_SRC(fp->code);
471 /* Common case where 'jump_false' is next insn. */
473 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
474 target = i + fp->jt + 1;
479 /* Convert JEQ into JNE when 'jump_true' is next insn. */
480 if (fp->jt == 0 && BPF_OP(fp->code) == BPF_JEQ) {
481 insn->code = BPF_JMP | BPF_JNE | bpf_src;
482 target = i + fp->jf + 1;
487 /* Other jumps are mapped into two insns: Jxx and JA. */
488 target = i + fp->jt + 1;
489 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
493 insn->code = BPF_JMP | BPF_JA;
494 target = i + fp->jf + 1;
498 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
499 case BPF_LDX | BPF_MSH | BPF_B:
501 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_A);
502 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
503 *insn++ = BPF_LD_ABS(BPF_B, fp->k);
505 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
507 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
509 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
511 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
514 /* RET_K, RET_A are remaped into 2 insns. */
515 case BPF_RET | BPF_A:
516 case BPF_RET | BPF_K:
517 *insn++ = BPF_MOV32_RAW(BPF_RVAL(fp->code) == BPF_K ?
518 BPF_K : BPF_X, BPF_REG_0,
520 *insn = BPF_EXIT_INSN();
523 /* Store to stack. */
526 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
527 BPF_ST ? BPF_REG_A : BPF_REG_X,
528 -(BPF_MEMWORDS - fp->k) * 4);
531 /* Load from stack. */
532 case BPF_LD | BPF_MEM:
533 case BPF_LDX | BPF_MEM:
534 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
535 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
536 -(BPF_MEMWORDS - fp->k) * 4);
540 case BPF_LD | BPF_IMM:
541 case BPF_LDX | BPF_IMM:
542 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
543 BPF_REG_A : BPF_REG_X, fp->k);
547 case BPF_MISC | BPF_TAX:
548 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
552 case BPF_MISC | BPF_TXA:
553 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
556 /* A = skb->len or X = skb->len */
557 case BPF_LD | BPF_W | BPF_LEN:
558 case BPF_LDX | BPF_W | BPF_LEN:
559 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
560 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
561 offsetof(struct sk_buff, len));
564 /* Access seccomp_data fields. */
565 case BPF_LDX | BPF_ABS | BPF_W:
566 /* A = *(u32 *) (ctx + K) */
567 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
570 /* Unknown instruction. */
577 memcpy(new_insn, tmp_insns,
578 sizeof(*insn) * (insn - tmp_insns));
579 new_insn += insn - tmp_insns;
583 /* Only calculating new length. */
584 *new_len = new_insn - new_prog;
589 if (new_flen != new_insn - new_prog) {
590 new_flen = new_insn - new_prog;
597 BUG_ON(*new_len != new_flen);
606 * As we dont want to clear mem[] array for each packet going through
607 * __bpf_prog_run(), we check that filter loaded by user never try to read
608 * a cell if not previously written, and we check all branches to be sure
609 * a malicious user doesn't try to abuse us.
611 static int check_load_and_stores(const struct sock_filter *filter, int flen)
613 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
616 BUILD_BUG_ON(BPF_MEMWORDS > 16);
618 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
622 memset(masks, 0xff, flen * sizeof(*masks));
624 for (pc = 0; pc < flen; pc++) {
625 memvalid &= masks[pc];
627 switch (filter[pc].code) {
630 memvalid |= (1 << filter[pc].k);
632 case BPF_LD | BPF_MEM:
633 case BPF_LDX | BPF_MEM:
634 if (!(memvalid & (1 << filter[pc].k))) {
639 case BPF_JMP | BPF_JA:
640 /* A jump must set masks on target */
641 masks[pc + 1 + filter[pc].k] &= memvalid;
644 case BPF_JMP | BPF_JEQ | BPF_K:
645 case BPF_JMP | BPF_JEQ | BPF_X:
646 case BPF_JMP | BPF_JGE | BPF_K:
647 case BPF_JMP | BPF_JGE | BPF_X:
648 case BPF_JMP | BPF_JGT | BPF_K:
649 case BPF_JMP | BPF_JGT | BPF_X:
650 case BPF_JMP | BPF_JSET | BPF_K:
651 case BPF_JMP | BPF_JSET | BPF_X:
652 /* A jump must set masks on targets */
653 masks[pc + 1 + filter[pc].jt] &= memvalid;
654 masks[pc + 1 + filter[pc].jf] &= memvalid;
664 static bool chk_code_allowed(u16 code_to_probe)
666 static const bool codes[] = {
667 /* 32 bit ALU operations */
668 [BPF_ALU | BPF_ADD | BPF_K] = true,
669 [BPF_ALU | BPF_ADD | BPF_X] = true,
670 [BPF_ALU | BPF_SUB | BPF_K] = true,
671 [BPF_ALU | BPF_SUB | BPF_X] = true,
672 [BPF_ALU | BPF_MUL | BPF_K] = true,
673 [BPF_ALU | BPF_MUL | BPF_X] = true,
674 [BPF_ALU | BPF_DIV | BPF_K] = true,
675 [BPF_ALU | BPF_DIV | BPF_X] = true,
676 [BPF_ALU | BPF_MOD | BPF_K] = true,
677 [BPF_ALU | BPF_MOD | BPF_X] = true,
678 [BPF_ALU | BPF_AND | BPF_K] = true,
679 [BPF_ALU | BPF_AND | BPF_X] = true,
680 [BPF_ALU | BPF_OR | BPF_K] = true,
681 [BPF_ALU | BPF_OR | BPF_X] = true,
682 [BPF_ALU | BPF_XOR | BPF_K] = true,
683 [BPF_ALU | BPF_XOR | BPF_X] = true,
684 [BPF_ALU | BPF_LSH | BPF_K] = true,
685 [BPF_ALU | BPF_LSH | BPF_X] = true,
686 [BPF_ALU | BPF_RSH | BPF_K] = true,
687 [BPF_ALU | BPF_RSH | BPF_X] = true,
688 [BPF_ALU | BPF_NEG] = true,
689 /* Load instructions */
690 [BPF_LD | BPF_W | BPF_ABS] = true,
691 [BPF_LD | BPF_H | BPF_ABS] = true,
692 [BPF_LD | BPF_B | BPF_ABS] = true,
693 [BPF_LD | BPF_W | BPF_LEN] = true,
694 [BPF_LD | BPF_W | BPF_IND] = true,
695 [BPF_LD | BPF_H | BPF_IND] = true,
696 [BPF_LD | BPF_B | BPF_IND] = true,
697 [BPF_LD | BPF_IMM] = true,
698 [BPF_LD | BPF_MEM] = true,
699 [BPF_LDX | BPF_W | BPF_LEN] = true,
700 [BPF_LDX | BPF_B | BPF_MSH] = true,
701 [BPF_LDX | BPF_IMM] = true,
702 [BPF_LDX | BPF_MEM] = true,
703 /* Store instructions */
706 /* Misc instructions */
707 [BPF_MISC | BPF_TAX] = true,
708 [BPF_MISC | BPF_TXA] = true,
709 /* Return instructions */
710 [BPF_RET | BPF_K] = true,
711 [BPF_RET | BPF_A] = true,
712 /* Jump instructions */
713 [BPF_JMP | BPF_JA] = true,
714 [BPF_JMP | BPF_JEQ | BPF_K] = true,
715 [BPF_JMP | BPF_JEQ | BPF_X] = true,
716 [BPF_JMP | BPF_JGE | BPF_K] = true,
717 [BPF_JMP | BPF_JGE | BPF_X] = true,
718 [BPF_JMP | BPF_JGT | BPF_K] = true,
719 [BPF_JMP | BPF_JGT | BPF_X] = true,
720 [BPF_JMP | BPF_JSET | BPF_K] = true,
721 [BPF_JMP | BPF_JSET | BPF_X] = true,
724 if (code_to_probe >= ARRAY_SIZE(codes))
727 return codes[code_to_probe];
731 * bpf_check_classic - verify socket filter code
732 * @filter: filter to verify
733 * @flen: length of filter
735 * Check the user's filter code. If we let some ugly
736 * filter code slip through kaboom! The filter must contain
737 * no references or jumps that are out of range, no illegal
738 * instructions, and must end with a RET instruction.
740 * All jumps are forward as they are not signed.
742 * Returns 0 if the rule set is legal or -EINVAL if not.
744 int bpf_check_classic(const struct sock_filter *filter, unsigned int flen)
749 if (flen == 0 || flen > BPF_MAXINSNS)
752 /* Check the filter code now */
753 for (pc = 0; pc < flen; pc++) {
754 const struct sock_filter *ftest = &filter[pc];
756 /* May we actually operate on this code? */
757 if (!chk_code_allowed(ftest->code))
760 /* Some instructions need special checks */
761 switch (ftest->code) {
762 case BPF_ALU | BPF_DIV | BPF_K:
763 case BPF_ALU | BPF_MOD | BPF_K:
764 /* Check for division by zero */
768 case BPF_LD | BPF_MEM:
769 case BPF_LDX | BPF_MEM:
772 /* Check for invalid memory addresses */
773 if (ftest->k >= BPF_MEMWORDS)
776 case BPF_JMP | BPF_JA:
777 /* Note, the large ftest->k might cause loops.
778 * Compare this with conditional jumps below,
779 * where offsets are limited. --ANK (981016)
781 if (ftest->k >= (unsigned int)(flen - pc - 1))
784 case BPF_JMP | BPF_JEQ | BPF_K:
785 case BPF_JMP | BPF_JEQ | BPF_X:
786 case BPF_JMP | BPF_JGE | BPF_K:
787 case BPF_JMP | BPF_JGE | BPF_X:
788 case BPF_JMP | BPF_JGT | BPF_K:
789 case BPF_JMP | BPF_JGT | BPF_X:
790 case BPF_JMP | BPF_JSET | BPF_K:
791 case BPF_JMP | BPF_JSET | BPF_X:
792 /* Both conditionals must be safe */
793 if (pc + ftest->jt + 1 >= flen ||
794 pc + ftest->jf + 1 >= flen)
797 case BPF_LD | BPF_W | BPF_ABS:
798 case BPF_LD | BPF_H | BPF_ABS:
799 case BPF_LD | BPF_B | BPF_ABS:
801 if (bpf_anc_helper(ftest) & BPF_ANC)
803 /* Ancillary operation unknown or unsupported */
804 if (anc_found == false && ftest->k >= SKF_AD_OFF)
809 /* Last instruction must be a RET code */
810 switch (filter[flen - 1].code) {
811 case BPF_RET | BPF_K:
812 case BPF_RET | BPF_A:
813 return check_load_and_stores(filter, flen);
818 EXPORT_SYMBOL(bpf_check_classic);
820 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
821 const struct sock_fprog *fprog)
823 unsigned int fsize = bpf_classic_proglen(fprog);
824 struct sock_fprog_kern *fkprog;
826 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
830 fkprog = fp->orig_prog;
831 fkprog->len = fprog->len;
832 fkprog->filter = kmemdup(fp->insns, fsize, GFP_KERNEL);
833 if (!fkprog->filter) {
834 kfree(fp->orig_prog);
841 static void bpf_release_orig_filter(struct bpf_prog *fp)
843 struct sock_fprog_kern *fprog = fp->orig_prog;
846 kfree(fprog->filter);
851 static void __bpf_prog_release(struct bpf_prog *prog)
853 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
856 bpf_release_orig_filter(prog);
861 static void __sk_filter_release(struct sk_filter *fp)
863 __bpf_prog_release(fp->prog);
868 * sk_filter_release_rcu - Release a socket filter by rcu_head
869 * @rcu: rcu_head that contains the sk_filter to free
871 static void sk_filter_release_rcu(struct rcu_head *rcu)
873 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
875 __sk_filter_release(fp);
879 * sk_filter_release - release a socket filter
880 * @fp: filter to remove
882 * Remove a filter from a socket and release its resources.
884 static void sk_filter_release(struct sk_filter *fp)
886 if (atomic_dec_and_test(&fp->refcnt))
887 call_rcu(&fp->rcu, sk_filter_release_rcu);
890 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
892 u32 filter_size = bpf_prog_size(fp->prog->len);
894 atomic_sub(filter_size, &sk->sk_omem_alloc);
895 sk_filter_release(fp);
898 /* try to charge the socket memory if there is space available
899 * return true on success
901 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
903 u32 filter_size = bpf_prog_size(fp->prog->len);
905 /* same check as in sock_kmalloc() */
906 if (filter_size <= sysctl_optmem_max &&
907 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
908 atomic_inc(&fp->refcnt);
909 atomic_add(filter_size, &sk->sk_omem_alloc);
915 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
917 struct sock_filter *old_prog;
918 struct bpf_prog *old_fp;
919 int err, new_len, old_len = fp->len;
921 /* We are free to overwrite insns et al right here as it
922 * won't be used at this point in time anymore internally
923 * after the migration to the internal BPF instruction
926 BUILD_BUG_ON(sizeof(struct sock_filter) !=
927 sizeof(struct bpf_insn));
929 /* Conversion cannot happen on overlapping memory areas,
930 * so we need to keep the user BPF around until the 2nd
931 * pass. At this time, the user BPF is stored in fp->insns.
933 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
940 /* 1st pass: calculate the new program length. */
941 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len);
945 /* Expand fp for appending the new filter representation. */
947 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
949 /* The old_fp is still around in case we couldn't
950 * allocate new memory, so uncharge on that one.
959 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
960 err = bpf_convert_filter(old_prog, old_len, fp->insnsi, &new_len);
962 /* 2nd bpf_convert_filter() can fail only if it fails
963 * to allocate memory, remapping must succeed. Note,
964 * that at this time old_fp has already been released
969 bpf_prog_select_runtime(fp);
977 __bpf_prog_release(fp);
981 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp)
988 err = bpf_check_classic(fp->insns, fp->len);
990 __bpf_prog_release(fp);
994 /* Probe if we can JIT compile the filter and if so, do
995 * the compilation of the filter.
999 /* JIT compiler couldn't process this filter, so do the
1000 * internal BPF translation for the optimized interpreter.
1003 fp = bpf_migrate_filter(fp);
1009 * bpf_prog_create - create an unattached filter
1010 * @pfp: the unattached filter that is created
1011 * @fprog: the filter program
1013 * Create a filter independent of any socket. We first run some
1014 * sanity checks on it to make sure it does not explode on us later.
1015 * If an error occurs or there is insufficient memory for the filter
1016 * a negative errno code is returned. On success the return is zero.
1018 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1020 unsigned int fsize = bpf_classic_proglen(fprog);
1021 struct bpf_prog *fp;
1023 /* Make sure new filter is there and in the right amounts. */
1024 if (fprog->filter == NULL)
1027 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1031 memcpy(fp->insns, fprog->filter, fsize);
1033 fp->len = fprog->len;
1034 /* Since unattached filters are not copied back to user
1035 * space through sk_get_filter(), we do not need to hold
1036 * a copy here, and can spare us the work.
1038 fp->orig_prog = NULL;
1040 /* bpf_prepare_filter() already takes care of freeing
1041 * memory in case something goes wrong.
1043 fp = bpf_prepare_filter(fp);
1050 EXPORT_SYMBOL_GPL(bpf_prog_create);
1052 void bpf_prog_destroy(struct bpf_prog *fp)
1054 __bpf_prog_release(fp);
1056 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1058 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1060 struct sk_filter *fp, *old_fp;
1062 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1067 atomic_set(&fp->refcnt, 0);
1069 if (!sk_filter_charge(sk, fp)) {
1074 old_fp = rcu_dereference_protected(sk->sk_filter,
1075 sock_owned_by_user(sk));
1076 rcu_assign_pointer(sk->sk_filter, fp);
1079 sk_filter_uncharge(sk, old_fp);
1085 * sk_attach_filter - attach a socket filter
1086 * @fprog: the filter program
1087 * @sk: the socket to use
1089 * Attach the user's filter code. We first run some sanity checks on
1090 * it to make sure it does not explode on us later. If an error
1091 * occurs or there is insufficient memory for the filter a negative
1092 * errno code is returned. On success the return is zero.
1094 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1096 unsigned int fsize = bpf_classic_proglen(fprog);
1097 unsigned int bpf_fsize = bpf_prog_size(fprog->len);
1098 struct bpf_prog *prog;
1101 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1104 /* Make sure new filter is there and in the right amounts. */
1105 if (fprog->filter == NULL)
1108 prog = bpf_prog_alloc(bpf_fsize, 0);
1112 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1113 __bpf_prog_free(prog);
1117 prog->len = fprog->len;
1119 err = bpf_prog_store_orig_filter(prog, fprog);
1121 __bpf_prog_free(prog);
1125 /* bpf_prepare_filter() already takes care of freeing
1126 * memory in case something goes wrong.
1128 prog = bpf_prepare_filter(prog);
1130 return PTR_ERR(prog);
1132 err = __sk_attach_prog(prog, sk);
1134 __bpf_prog_release(prog);
1140 EXPORT_SYMBOL_GPL(sk_attach_filter);
1142 int sk_attach_bpf(u32 ufd, struct sock *sk)
1144 struct bpf_prog *prog;
1147 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1150 prog = bpf_prog_get(ufd);
1152 return PTR_ERR(prog);
1154 if (prog->type != BPF_PROG_TYPE_SOCKET_FILTER) {
1159 err = __sk_attach_prog(prog, sk);
1168 static const struct bpf_func_proto *
1169 sk_filter_func_proto(enum bpf_func_id func_id)
1172 case BPF_FUNC_map_lookup_elem:
1173 return &bpf_map_lookup_elem_proto;
1174 case BPF_FUNC_map_update_elem:
1175 return &bpf_map_update_elem_proto;
1176 case BPF_FUNC_map_delete_elem:
1177 return &bpf_map_delete_elem_proto;
1178 case BPF_FUNC_get_prandom_u32:
1179 return &bpf_get_prandom_u32_proto;
1180 case BPF_FUNC_get_smp_processor_id:
1181 return &bpf_get_smp_processor_id_proto;
1187 static bool sk_filter_is_valid_access(int off, int size,
1188 enum bpf_access_type type)
1190 /* only read is allowed */
1191 if (type != BPF_READ)
1195 if (off < 0 || off >= sizeof(struct __sk_buff))
1198 /* disallow misaligned access */
1199 if (off % size != 0)
1202 /* all __sk_buff fields are __u32 */
1209 static u32 sk_filter_convert_ctx_access(int dst_reg, int src_reg, int ctx_off,
1210 struct bpf_insn *insn_buf)
1212 struct bpf_insn *insn = insn_buf;
1215 case offsetof(struct __sk_buff, len):
1216 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, len) != 4);
1218 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
1219 offsetof(struct sk_buff, len));
1222 case offsetof(struct __sk_buff, protocol):
1223 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
1225 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
1226 offsetof(struct sk_buff, protocol));
1229 case offsetof(struct __sk_buff, mark):
1230 return convert_skb_access(SKF_AD_MARK, dst_reg, src_reg, insn);
1232 case offsetof(struct __sk_buff, pkt_type):
1233 return convert_skb_access(SKF_AD_PKTTYPE, dst_reg, src_reg, insn);
1235 case offsetof(struct __sk_buff, queue_mapping):
1236 return convert_skb_access(SKF_AD_QUEUE, dst_reg, src_reg, insn);
1238 case offsetof(struct __sk_buff, vlan_present):
1239 return convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
1240 dst_reg, src_reg, insn);
1242 case offsetof(struct __sk_buff, vlan_tci):
1243 return convert_skb_access(SKF_AD_VLAN_TAG,
1244 dst_reg, src_reg, insn);
1247 return insn - insn_buf;
1250 static const struct bpf_verifier_ops sk_filter_ops = {
1251 .get_func_proto = sk_filter_func_proto,
1252 .is_valid_access = sk_filter_is_valid_access,
1253 .convert_ctx_access = sk_filter_convert_ctx_access,
1256 static struct bpf_prog_type_list sk_filter_type __read_mostly = {
1257 .ops = &sk_filter_ops,
1258 .type = BPF_PROG_TYPE_SOCKET_FILTER,
1261 static struct bpf_prog_type_list sched_cls_type __read_mostly = {
1262 .ops = &sk_filter_ops,
1263 .type = BPF_PROG_TYPE_SCHED_CLS,
1266 static int __init register_sk_filter_ops(void)
1268 bpf_register_prog_type(&sk_filter_type);
1269 bpf_register_prog_type(&sched_cls_type);
1273 late_initcall(register_sk_filter_ops);
1275 int sk_detach_filter(struct sock *sk)
1278 struct sk_filter *filter;
1280 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1283 filter = rcu_dereference_protected(sk->sk_filter,
1284 sock_owned_by_user(sk));
1286 RCU_INIT_POINTER(sk->sk_filter, NULL);
1287 sk_filter_uncharge(sk, filter);
1293 EXPORT_SYMBOL_GPL(sk_detach_filter);
1295 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
1298 struct sock_fprog_kern *fprog;
1299 struct sk_filter *filter;
1303 filter = rcu_dereference_protected(sk->sk_filter,
1304 sock_owned_by_user(sk));
1308 /* We're copying the filter that has been originally attached,
1309 * so no conversion/decode needed anymore.
1311 fprog = filter->prog->orig_prog;
1315 /* User space only enquires number of filter blocks. */
1319 if (len < fprog->len)
1323 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
1326 /* Instead of bytes, the API requests to return the number