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>
55 * sk_filter - run a packet through a socket filter
56 * @sk: sock associated with &sk_buff
57 * @skb: buffer to filter
59 * Run the eBPF program and then cut skb->data to correct size returned by
60 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
61 * than pkt_len we keep whole skb->data. This is the socket level
62 * wrapper to BPF_PROG_RUN. It returns 0 if the packet should
63 * be accepted or -EPERM if the packet should be tossed.
66 int sk_filter(struct sock *sk, struct sk_buff *skb)
69 struct sk_filter *filter;
72 * If the skb was allocated from pfmemalloc reserves, only
73 * allow SOCK_MEMALLOC sockets to use it as this socket is
76 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
79 err = security_sock_rcv_skb(sk, skb);
84 filter = rcu_dereference(sk->sk_filter);
86 unsigned int pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
88 err = pkt_len ? pskb_trim(skb, pkt_len) : -EPERM;
94 EXPORT_SYMBOL(sk_filter);
96 static u64 __skb_get_pay_offset(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
98 return skb_get_poff((struct sk_buff *)(unsigned long) ctx);
101 static u64 __skb_get_nlattr(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
103 struct sk_buff *skb = (struct sk_buff *)(unsigned long) ctx;
106 if (skb_is_nonlinear(skb))
109 if (skb->len < sizeof(struct nlattr))
112 if (a > skb->len - sizeof(struct nlattr))
115 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
117 return (void *) nla - (void *) skb->data;
122 static u64 __skb_get_nlattr_nest(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
124 struct sk_buff *skb = (struct sk_buff *)(unsigned long) ctx;
127 if (skb_is_nonlinear(skb))
130 if (skb->len < sizeof(struct nlattr))
133 if (a > skb->len - sizeof(struct nlattr))
136 nla = (struct nlattr *) &skb->data[a];
137 if (nla->nla_len > skb->len - a)
140 nla = nla_find_nested(nla, x);
142 return (void *) nla - (void *) skb->data;
147 static u64 __get_raw_cpu_id(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
149 return raw_smp_processor_id();
152 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
153 struct bpf_insn *insn_buf)
155 struct bpf_insn *insn = insn_buf;
159 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
161 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
162 offsetof(struct sk_buff, mark));
166 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
167 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
168 #ifdef __BIG_ENDIAN_BITFIELD
169 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
174 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2);
176 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
177 offsetof(struct sk_buff, queue_mapping));
180 case SKF_AD_VLAN_TAG:
181 case SKF_AD_VLAN_TAG_PRESENT:
182 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2);
183 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000);
185 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
186 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
187 offsetof(struct sk_buff, vlan_tci));
188 if (skb_field == SKF_AD_VLAN_TAG) {
189 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg,
193 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 12);
195 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
200 return insn - insn_buf;
203 static bool convert_bpf_extensions(struct sock_filter *fp,
204 struct bpf_insn **insnp)
206 struct bpf_insn *insn = *insnp;
210 case SKF_AD_OFF + SKF_AD_PROTOCOL:
211 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
213 /* A = *(u16 *) (CTX + offsetof(protocol)) */
214 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
215 offsetof(struct sk_buff, protocol));
216 /* A = ntohs(A) [emitting a nop or swap16] */
217 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
220 case SKF_AD_OFF + SKF_AD_PKTTYPE:
221 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
225 case SKF_AD_OFF + SKF_AD_IFINDEX:
226 case SKF_AD_OFF + SKF_AD_HATYPE:
227 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
228 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, type) != 2);
229 BUILD_BUG_ON(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, dev)) < 0);
231 *insn++ = BPF_LDX_MEM(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, dev)),
232 BPF_REG_TMP, BPF_REG_CTX,
233 offsetof(struct sk_buff, dev));
234 /* if (tmp != 0) goto pc + 1 */
235 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
236 *insn++ = BPF_EXIT_INSN();
237 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
238 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
239 offsetof(struct net_device, ifindex));
241 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
242 offsetof(struct net_device, type));
245 case SKF_AD_OFF + SKF_AD_MARK:
246 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
250 case SKF_AD_OFF + SKF_AD_RXHASH:
251 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
253 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
254 offsetof(struct sk_buff, hash));
257 case SKF_AD_OFF + SKF_AD_QUEUE:
258 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
262 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
263 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
264 BPF_REG_A, BPF_REG_CTX, insn);
268 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
269 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
270 BPF_REG_A, BPF_REG_CTX, insn);
274 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
275 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2);
277 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
278 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
279 offsetof(struct sk_buff, vlan_proto));
280 /* A = ntohs(A) [emitting a nop or swap16] */
281 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
284 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
285 case SKF_AD_OFF + SKF_AD_NLATTR:
286 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
287 case SKF_AD_OFF + SKF_AD_CPU:
288 case SKF_AD_OFF + SKF_AD_RANDOM:
290 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
292 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
294 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
295 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
297 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
298 *insn = BPF_EMIT_CALL(__skb_get_pay_offset);
300 case SKF_AD_OFF + SKF_AD_NLATTR:
301 *insn = BPF_EMIT_CALL(__skb_get_nlattr);
303 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
304 *insn = BPF_EMIT_CALL(__skb_get_nlattr_nest);
306 case SKF_AD_OFF + SKF_AD_CPU:
307 *insn = BPF_EMIT_CALL(__get_raw_cpu_id);
309 case SKF_AD_OFF + SKF_AD_RANDOM:
310 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
311 bpf_user_rnd_init_once();
316 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
318 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
322 /* This is just a dummy call to avoid letting the compiler
323 * evict __bpf_call_base() as an optimization. Placed here
324 * where no-one bothers.
326 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
335 * bpf_convert_filter - convert filter program
336 * @prog: the user passed filter program
337 * @len: the length of the user passed filter program
338 * @new_prog: buffer where converted program will be stored
339 * @new_len: pointer to store length of converted program
341 * Remap 'sock_filter' style BPF instruction set to 'sock_filter_ext' style.
342 * Conversion workflow:
344 * 1) First pass for calculating the new program length:
345 * bpf_convert_filter(old_prog, old_len, NULL, &new_len)
347 * 2) 2nd pass to remap in two passes: 1st pass finds new
348 * jump offsets, 2nd pass remapping:
349 * new_prog = kmalloc(sizeof(struct bpf_insn) * new_len);
350 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len);
352 * User BPF's register A is mapped to our BPF register 6, user BPF
353 * register X is mapped to BPF register 7; frame pointer is always
354 * register 10; Context 'void *ctx' is stored in register 1, that is,
355 * for socket filters: ctx == 'struct sk_buff *', for seccomp:
356 * ctx == 'struct seccomp_data *'.
358 static int bpf_convert_filter(struct sock_filter *prog, int len,
359 struct bpf_insn *new_prog, int *new_len)
361 int new_flen = 0, pass = 0, target, i;
362 struct bpf_insn *new_insn;
363 struct sock_filter *fp;
367 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
368 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
370 if (len <= 0 || len > BPF_MAXINSNS)
374 addrs = kcalloc(len, sizeof(*addrs),
375 GFP_KERNEL | __GFP_NOWARN);
385 *new_insn = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
388 for (i = 0; i < len; fp++, i++) {
389 struct bpf_insn tmp_insns[6] = { };
390 struct bpf_insn *insn = tmp_insns;
393 addrs[i] = new_insn - new_prog;
396 /* All arithmetic insns and skb loads map as-is. */
397 case BPF_ALU | BPF_ADD | BPF_X:
398 case BPF_ALU | BPF_ADD | BPF_K:
399 case BPF_ALU | BPF_SUB | BPF_X:
400 case BPF_ALU | BPF_SUB | BPF_K:
401 case BPF_ALU | BPF_AND | BPF_X:
402 case BPF_ALU | BPF_AND | BPF_K:
403 case BPF_ALU | BPF_OR | BPF_X:
404 case BPF_ALU | BPF_OR | BPF_K:
405 case BPF_ALU | BPF_LSH | BPF_X:
406 case BPF_ALU | BPF_LSH | BPF_K:
407 case BPF_ALU | BPF_RSH | BPF_X:
408 case BPF_ALU | BPF_RSH | BPF_K:
409 case BPF_ALU | BPF_XOR | BPF_X:
410 case BPF_ALU | BPF_XOR | BPF_K:
411 case BPF_ALU | BPF_MUL | BPF_X:
412 case BPF_ALU | BPF_MUL | BPF_K:
413 case BPF_ALU | BPF_DIV | BPF_X:
414 case BPF_ALU | BPF_DIV | BPF_K:
415 case BPF_ALU | BPF_MOD | BPF_X:
416 case BPF_ALU | BPF_MOD | BPF_K:
417 case BPF_ALU | BPF_NEG:
418 case BPF_LD | BPF_ABS | BPF_W:
419 case BPF_LD | BPF_ABS | BPF_H:
420 case BPF_LD | BPF_ABS | BPF_B:
421 case BPF_LD | BPF_IND | BPF_W:
422 case BPF_LD | BPF_IND | BPF_H:
423 case BPF_LD | BPF_IND | BPF_B:
424 /* Check for overloaded BPF extension and
425 * directly convert it if found, otherwise
426 * just move on with mapping.
428 if (BPF_CLASS(fp->code) == BPF_LD &&
429 BPF_MODE(fp->code) == BPF_ABS &&
430 convert_bpf_extensions(fp, &insn))
433 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
436 /* Jump transformation cannot use BPF block macros
437 * everywhere as offset calculation and target updates
438 * require a bit more work than the rest, i.e. jump
439 * opcodes map as-is, but offsets need adjustment.
442 #define BPF_EMIT_JMP \
444 if (target >= len || target < 0) \
446 insn->off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
447 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
448 insn->off -= insn - tmp_insns; \
451 case BPF_JMP | BPF_JA:
452 target = i + fp->k + 1;
453 insn->code = fp->code;
457 case BPF_JMP | BPF_JEQ | BPF_K:
458 case BPF_JMP | BPF_JEQ | BPF_X:
459 case BPF_JMP | BPF_JSET | BPF_K:
460 case BPF_JMP | BPF_JSET | BPF_X:
461 case BPF_JMP | BPF_JGT | BPF_K:
462 case BPF_JMP | BPF_JGT | BPF_X:
463 case BPF_JMP | BPF_JGE | BPF_K:
464 case BPF_JMP | BPF_JGE | BPF_X:
465 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
466 /* BPF immediates are signed, zero extend
467 * immediate into tmp register and use it
470 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
472 insn->dst_reg = BPF_REG_A;
473 insn->src_reg = BPF_REG_TMP;
476 insn->dst_reg = BPF_REG_A;
478 bpf_src = BPF_SRC(fp->code);
479 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
482 /* Common case where 'jump_false' is next insn. */
484 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
485 target = i + fp->jt + 1;
490 /* Convert JEQ into JNE when 'jump_true' is next insn. */
491 if (fp->jt == 0 && BPF_OP(fp->code) == BPF_JEQ) {
492 insn->code = BPF_JMP | BPF_JNE | bpf_src;
493 target = i + fp->jf + 1;
498 /* Other jumps are mapped into two insns: Jxx and JA. */
499 target = i + fp->jt + 1;
500 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
504 insn->code = BPF_JMP | BPF_JA;
505 target = i + fp->jf + 1;
509 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
510 case BPF_LDX | BPF_MSH | BPF_B:
512 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_A);
513 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
514 *insn++ = BPF_LD_ABS(BPF_B, fp->k);
516 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
518 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
520 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
522 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
525 /* RET_K, RET_A are remaped into 2 insns. */
526 case BPF_RET | BPF_A:
527 case BPF_RET | BPF_K:
528 *insn++ = BPF_MOV32_RAW(BPF_RVAL(fp->code) == BPF_K ?
529 BPF_K : BPF_X, BPF_REG_0,
531 *insn = BPF_EXIT_INSN();
534 /* Store to stack. */
537 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
538 BPF_ST ? BPF_REG_A : BPF_REG_X,
539 -(BPF_MEMWORDS - fp->k) * 4);
542 /* Load from stack. */
543 case BPF_LD | BPF_MEM:
544 case BPF_LDX | BPF_MEM:
545 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
546 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
547 -(BPF_MEMWORDS - fp->k) * 4);
551 case BPF_LD | BPF_IMM:
552 case BPF_LDX | BPF_IMM:
553 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
554 BPF_REG_A : BPF_REG_X, fp->k);
558 case BPF_MISC | BPF_TAX:
559 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
563 case BPF_MISC | BPF_TXA:
564 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
567 /* A = skb->len or X = skb->len */
568 case BPF_LD | BPF_W | BPF_LEN:
569 case BPF_LDX | BPF_W | BPF_LEN:
570 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
571 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
572 offsetof(struct sk_buff, len));
575 /* Access seccomp_data fields. */
576 case BPF_LDX | BPF_ABS | BPF_W:
577 /* A = *(u32 *) (ctx + K) */
578 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
581 /* Unknown instruction. */
588 memcpy(new_insn, tmp_insns,
589 sizeof(*insn) * (insn - tmp_insns));
590 new_insn += insn - tmp_insns;
594 /* Only calculating new length. */
595 *new_len = new_insn - new_prog;
600 if (new_flen != new_insn - new_prog) {
601 new_flen = new_insn - new_prog;
608 BUG_ON(*new_len != new_flen);
617 * As we dont want to clear mem[] array for each packet going through
618 * __bpf_prog_run(), we check that filter loaded by user never try to read
619 * a cell if not previously written, and we check all branches to be sure
620 * a malicious user doesn't try to abuse us.
622 static int check_load_and_stores(const struct sock_filter *filter, int flen)
624 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
627 BUILD_BUG_ON(BPF_MEMWORDS > 16);
629 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
633 memset(masks, 0xff, flen * sizeof(*masks));
635 for (pc = 0; pc < flen; pc++) {
636 memvalid &= masks[pc];
638 switch (filter[pc].code) {
641 memvalid |= (1 << filter[pc].k);
643 case BPF_LD | BPF_MEM:
644 case BPF_LDX | BPF_MEM:
645 if (!(memvalid & (1 << filter[pc].k))) {
650 case BPF_JMP | BPF_JA:
651 /* A jump must set masks on target */
652 masks[pc + 1 + filter[pc].k] &= memvalid;
655 case BPF_JMP | BPF_JEQ | BPF_K:
656 case BPF_JMP | BPF_JEQ | BPF_X:
657 case BPF_JMP | BPF_JGE | BPF_K:
658 case BPF_JMP | BPF_JGE | BPF_X:
659 case BPF_JMP | BPF_JGT | BPF_K:
660 case BPF_JMP | BPF_JGT | BPF_X:
661 case BPF_JMP | BPF_JSET | BPF_K:
662 case BPF_JMP | BPF_JSET | BPF_X:
663 /* A jump must set masks on targets */
664 masks[pc + 1 + filter[pc].jt] &= memvalid;
665 masks[pc + 1 + filter[pc].jf] &= memvalid;
675 static bool chk_code_allowed(u16 code_to_probe)
677 static const bool codes[] = {
678 /* 32 bit ALU operations */
679 [BPF_ALU | BPF_ADD | BPF_K] = true,
680 [BPF_ALU | BPF_ADD | BPF_X] = true,
681 [BPF_ALU | BPF_SUB | BPF_K] = true,
682 [BPF_ALU | BPF_SUB | BPF_X] = true,
683 [BPF_ALU | BPF_MUL | BPF_K] = true,
684 [BPF_ALU | BPF_MUL | BPF_X] = true,
685 [BPF_ALU | BPF_DIV | BPF_K] = true,
686 [BPF_ALU | BPF_DIV | BPF_X] = true,
687 [BPF_ALU | BPF_MOD | BPF_K] = true,
688 [BPF_ALU | BPF_MOD | BPF_X] = true,
689 [BPF_ALU | BPF_AND | BPF_K] = true,
690 [BPF_ALU | BPF_AND | BPF_X] = true,
691 [BPF_ALU | BPF_OR | BPF_K] = true,
692 [BPF_ALU | BPF_OR | BPF_X] = true,
693 [BPF_ALU | BPF_XOR | BPF_K] = true,
694 [BPF_ALU | BPF_XOR | BPF_X] = true,
695 [BPF_ALU | BPF_LSH | BPF_K] = true,
696 [BPF_ALU | BPF_LSH | BPF_X] = true,
697 [BPF_ALU | BPF_RSH | BPF_K] = true,
698 [BPF_ALU | BPF_RSH | BPF_X] = true,
699 [BPF_ALU | BPF_NEG] = true,
700 /* Load instructions */
701 [BPF_LD | BPF_W | BPF_ABS] = true,
702 [BPF_LD | BPF_H | BPF_ABS] = true,
703 [BPF_LD | BPF_B | BPF_ABS] = true,
704 [BPF_LD | BPF_W | BPF_LEN] = true,
705 [BPF_LD | BPF_W | BPF_IND] = true,
706 [BPF_LD | BPF_H | BPF_IND] = true,
707 [BPF_LD | BPF_B | BPF_IND] = true,
708 [BPF_LD | BPF_IMM] = true,
709 [BPF_LD | BPF_MEM] = true,
710 [BPF_LDX | BPF_W | BPF_LEN] = true,
711 [BPF_LDX | BPF_B | BPF_MSH] = true,
712 [BPF_LDX | BPF_IMM] = true,
713 [BPF_LDX | BPF_MEM] = true,
714 /* Store instructions */
717 /* Misc instructions */
718 [BPF_MISC | BPF_TAX] = true,
719 [BPF_MISC | BPF_TXA] = true,
720 /* Return instructions */
721 [BPF_RET | BPF_K] = true,
722 [BPF_RET | BPF_A] = true,
723 /* Jump instructions */
724 [BPF_JMP | BPF_JA] = true,
725 [BPF_JMP | BPF_JEQ | BPF_K] = true,
726 [BPF_JMP | BPF_JEQ | BPF_X] = true,
727 [BPF_JMP | BPF_JGE | BPF_K] = true,
728 [BPF_JMP | BPF_JGE | BPF_X] = true,
729 [BPF_JMP | BPF_JGT | BPF_K] = true,
730 [BPF_JMP | BPF_JGT | BPF_X] = true,
731 [BPF_JMP | BPF_JSET | BPF_K] = true,
732 [BPF_JMP | BPF_JSET | BPF_X] = true,
735 if (code_to_probe >= ARRAY_SIZE(codes))
738 return codes[code_to_probe];
742 * bpf_check_classic - verify socket filter code
743 * @filter: filter to verify
744 * @flen: length of filter
746 * Check the user's filter code. If we let some ugly
747 * filter code slip through kaboom! The filter must contain
748 * no references or jumps that are out of range, no illegal
749 * instructions, and must end with a RET instruction.
751 * All jumps are forward as they are not signed.
753 * Returns 0 if the rule set is legal or -EINVAL if not.
755 static int bpf_check_classic(const struct sock_filter *filter,
761 if (flen == 0 || flen > BPF_MAXINSNS)
764 /* Check the filter code now */
765 for (pc = 0; pc < flen; pc++) {
766 const struct sock_filter *ftest = &filter[pc];
768 /* May we actually operate on this code? */
769 if (!chk_code_allowed(ftest->code))
772 /* Some instructions need special checks */
773 switch (ftest->code) {
774 case BPF_ALU | BPF_DIV | BPF_K:
775 case BPF_ALU | BPF_MOD | BPF_K:
776 /* Check for division by zero */
780 case BPF_ALU | BPF_LSH | BPF_K:
781 case BPF_ALU | BPF_RSH | BPF_K:
785 case BPF_LD | BPF_MEM:
786 case BPF_LDX | BPF_MEM:
789 /* Check for invalid memory addresses */
790 if (ftest->k >= BPF_MEMWORDS)
793 case BPF_JMP | BPF_JA:
794 /* Note, the large ftest->k might cause loops.
795 * Compare this with conditional jumps below,
796 * where offsets are limited. --ANK (981016)
798 if (ftest->k >= (unsigned int)(flen - pc - 1))
801 case BPF_JMP | BPF_JEQ | BPF_K:
802 case BPF_JMP | BPF_JEQ | BPF_X:
803 case BPF_JMP | BPF_JGE | BPF_K:
804 case BPF_JMP | BPF_JGE | BPF_X:
805 case BPF_JMP | BPF_JGT | BPF_K:
806 case BPF_JMP | BPF_JGT | BPF_X:
807 case BPF_JMP | BPF_JSET | BPF_K:
808 case BPF_JMP | BPF_JSET | BPF_X:
809 /* Both conditionals must be safe */
810 if (pc + ftest->jt + 1 >= flen ||
811 pc + ftest->jf + 1 >= flen)
814 case BPF_LD | BPF_W | BPF_ABS:
815 case BPF_LD | BPF_H | BPF_ABS:
816 case BPF_LD | BPF_B | BPF_ABS:
818 if (bpf_anc_helper(ftest) & BPF_ANC)
820 /* Ancillary operation unknown or unsupported */
821 if (anc_found == false && ftest->k >= SKF_AD_OFF)
826 /* Last instruction must be a RET code */
827 switch (filter[flen - 1].code) {
828 case BPF_RET | BPF_K:
829 case BPF_RET | BPF_A:
830 return check_load_and_stores(filter, flen);
836 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
837 const struct sock_fprog *fprog)
839 unsigned int fsize = bpf_classic_proglen(fprog);
840 struct sock_fprog_kern *fkprog;
842 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
846 fkprog = fp->orig_prog;
847 fkprog->len = fprog->len;
849 fkprog->filter = kmemdup(fp->insns, fsize,
850 GFP_KERNEL | __GFP_NOWARN);
851 if (!fkprog->filter) {
852 kfree(fp->orig_prog);
859 static void bpf_release_orig_filter(struct bpf_prog *fp)
861 struct sock_fprog_kern *fprog = fp->orig_prog;
864 kfree(fprog->filter);
869 static void __bpf_prog_release(struct bpf_prog *prog)
871 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
874 bpf_release_orig_filter(prog);
879 static void __sk_filter_release(struct sk_filter *fp)
881 __bpf_prog_release(fp->prog);
886 * sk_filter_release_rcu - Release a socket filter by rcu_head
887 * @rcu: rcu_head that contains the sk_filter to free
889 static void sk_filter_release_rcu(struct rcu_head *rcu)
891 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
893 __sk_filter_release(fp);
897 * sk_filter_release - release a socket filter
898 * @fp: filter to remove
900 * Remove a filter from a socket and release its resources.
902 static void sk_filter_release(struct sk_filter *fp)
904 if (atomic_dec_and_test(&fp->refcnt))
905 call_rcu(&fp->rcu, sk_filter_release_rcu);
908 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
910 u32 filter_size = bpf_prog_size(fp->prog->len);
912 atomic_sub(filter_size, &sk->sk_omem_alloc);
913 sk_filter_release(fp);
916 /* try to charge the socket memory if there is space available
917 * return true on success
919 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
921 u32 filter_size = bpf_prog_size(fp->prog->len);
923 /* same check as in sock_kmalloc() */
924 if (filter_size <= sysctl_optmem_max &&
925 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
926 atomic_inc(&fp->refcnt);
927 atomic_add(filter_size, &sk->sk_omem_alloc);
933 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
935 struct sock_filter *old_prog;
936 struct bpf_prog *old_fp;
937 int err, new_len, old_len = fp->len;
939 /* We are free to overwrite insns et al right here as it
940 * won't be used at this point in time anymore internally
941 * after the migration to the internal BPF instruction
944 BUILD_BUG_ON(sizeof(struct sock_filter) !=
945 sizeof(struct bpf_insn));
947 /* Conversion cannot happen on overlapping memory areas,
948 * so we need to keep the user BPF around until the 2nd
949 * pass. At this time, the user BPF is stored in fp->insns.
951 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
952 GFP_KERNEL | __GFP_NOWARN);
958 /* 1st pass: calculate the new program length. */
959 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len);
963 /* Expand fp for appending the new filter representation. */
965 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
967 /* The old_fp is still around in case we couldn't
968 * allocate new memory, so uncharge on that one.
977 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
978 err = bpf_convert_filter(old_prog, old_len, fp->insnsi, &new_len);
980 /* 2nd bpf_convert_filter() can fail only if it fails
981 * to allocate memory, remapping must succeed. Note,
982 * that at this time old_fp has already been released
987 bpf_prog_select_runtime(fp);
995 __bpf_prog_release(fp);
999 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1000 bpf_aux_classic_check_t trans)
1004 fp->bpf_func = NULL;
1007 err = bpf_check_classic(fp->insns, fp->len);
1009 __bpf_prog_release(fp);
1010 return ERR_PTR(err);
1013 /* There might be additional checks and transformations
1014 * needed on classic filters, f.e. in case of seccomp.
1017 err = trans(fp->insns, fp->len);
1019 __bpf_prog_release(fp);
1020 return ERR_PTR(err);
1024 /* Probe if we can JIT compile the filter and if so, do
1025 * the compilation of the filter.
1027 bpf_jit_compile(fp);
1029 /* JIT compiler couldn't process this filter, so do the
1030 * internal BPF translation for the optimized interpreter.
1033 fp = bpf_migrate_filter(fp);
1039 * bpf_prog_create - create an unattached filter
1040 * @pfp: the unattached filter that is created
1041 * @fprog: the filter program
1043 * Create a filter independent of any socket. We first run some
1044 * sanity checks on it to make sure it does not explode on us later.
1045 * If an error occurs or there is insufficient memory for the filter
1046 * a negative errno code is returned. On success the return is zero.
1048 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1050 unsigned int fsize = bpf_classic_proglen(fprog);
1051 struct bpf_prog *fp;
1053 /* Make sure new filter is there and in the right amounts. */
1054 if (fprog->filter == NULL)
1057 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1061 memcpy(fp->insns, fprog->filter, fsize);
1063 fp->len = fprog->len;
1064 /* Since unattached filters are not copied back to user
1065 * space through sk_get_filter(), we do not need to hold
1066 * a copy here, and can spare us the work.
1068 fp->orig_prog = NULL;
1070 /* bpf_prepare_filter() already takes care of freeing
1071 * memory in case something goes wrong.
1073 fp = bpf_prepare_filter(fp, NULL);
1080 EXPORT_SYMBOL_GPL(bpf_prog_create);
1083 * bpf_prog_create_from_user - create an unattached filter from user buffer
1084 * @pfp: the unattached filter that is created
1085 * @fprog: the filter program
1086 * @trans: post-classic verifier transformation handler
1087 * @save_orig: save classic BPF program
1089 * This function effectively does the same as bpf_prog_create(), only
1090 * that it builds up its insns buffer from user space provided buffer.
1091 * It also allows for passing a bpf_aux_classic_check_t handler.
1093 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1094 bpf_aux_classic_check_t trans, bool save_orig)
1096 unsigned int fsize = bpf_classic_proglen(fprog);
1097 struct bpf_prog *fp;
1100 /* Make sure new filter is there and in the right amounts. */
1101 if (fprog->filter == NULL)
1104 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1108 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1109 __bpf_prog_free(fp);
1113 fp->len = fprog->len;
1114 fp->orig_prog = NULL;
1117 err = bpf_prog_store_orig_filter(fp, fprog);
1119 __bpf_prog_free(fp);
1124 /* bpf_prepare_filter() already takes care of freeing
1125 * memory in case something goes wrong.
1127 fp = bpf_prepare_filter(fp, trans);
1134 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1136 void bpf_prog_destroy(struct bpf_prog *fp)
1138 __bpf_prog_release(fp);
1140 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1142 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk,
1145 struct sk_filter *fp, *old_fp;
1147 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1152 atomic_set(&fp->refcnt, 0);
1154 if (!sk_filter_charge(sk, fp)) {
1159 old_fp = rcu_dereference_protected(sk->sk_filter, locked);
1160 rcu_assign_pointer(sk->sk_filter, fp);
1162 sk_filter_uncharge(sk, old_fp);
1168 * sk_attach_filter - attach a socket filter
1169 * @fprog: the filter program
1170 * @sk: the socket to use
1172 * Attach the user's filter code. We first run some sanity checks on
1173 * it to make sure it does not explode on us later. If an error
1174 * occurs or there is insufficient memory for the filter a negative
1175 * errno code is returned. On success the return is zero.
1177 int __sk_attach_filter(struct sock_fprog *fprog, struct sock *sk,
1180 unsigned int fsize = bpf_classic_proglen(fprog);
1181 unsigned int bpf_fsize = bpf_prog_size(fprog->len);
1182 struct bpf_prog *prog;
1185 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1188 /* Make sure new filter is there and in the right amounts. */
1189 if (fprog->filter == NULL)
1192 prog = bpf_prog_alloc(bpf_fsize, 0);
1196 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1197 __bpf_prog_free(prog);
1201 prog->len = fprog->len;
1203 err = bpf_prog_store_orig_filter(prog, fprog);
1205 __bpf_prog_free(prog);
1209 /* bpf_prepare_filter() already takes care of freeing
1210 * memory in case something goes wrong.
1212 prog = bpf_prepare_filter(prog, NULL);
1214 return PTR_ERR(prog);
1216 err = __sk_attach_prog(prog, sk, locked);
1218 __bpf_prog_release(prog);
1224 EXPORT_SYMBOL_GPL(__sk_attach_filter);
1226 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1228 return __sk_attach_filter(fprog, sk, sock_owned_by_user(sk));
1231 int sk_attach_bpf(u32 ufd, struct sock *sk)
1233 struct bpf_prog *prog;
1236 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1239 prog = bpf_prog_get(ufd);
1241 return PTR_ERR(prog);
1243 if (prog->type != BPF_PROG_TYPE_SOCKET_FILTER) {
1248 err = __sk_attach_prog(prog, sk, sock_owned_by_user(sk));
1257 #define BPF_RECOMPUTE_CSUM(flags) ((flags) & 1)
1259 static u64 bpf_skb_store_bytes(u64 r1, u64 r2, u64 r3, u64 r4, u64 flags)
1261 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1262 int offset = (int) r2;
1263 void *from = (void *) (long) r3;
1264 unsigned int len = (unsigned int) r4;
1268 /* bpf verifier guarantees that:
1269 * 'from' pointer points to bpf program stack
1270 * 'len' bytes of it were initialized
1272 * 'skb' is a valid pointer to 'struct sk_buff'
1274 * so check for invalid 'offset' and too large 'len'
1276 if (unlikely((u32) offset > 0xffff || len > sizeof(buf)))
1279 if (unlikely(skb_cloned(skb) &&
1280 !skb_clone_writable(skb, offset + len)))
1283 ptr = skb_header_pointer(skb, offset, len, buf);
1287 if (BPF_RECOMPUTE_CSUM(flags))
1288 skb_postpull_rcsum(skb, ptr, len);
1290 memcpy(ptr, from, len);
1293 /* skb_store_bits cannot return -EFAULT here */
1294 skb_store_bits(skb, offset, ptr, len);
1296 if (BPF_RECOMPUTE_CSUM(flags) && skb->ip_summed == CHECKSUM_COMPLETE)
1297 skb->csum = csum_add(skb->csum, csum_partial(ptr, len, 0));
1301 const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1302 .func = bpf_skb_store_bytes,
1304 .ret_type = RET_INTEGER,
1305 .arg1_type = ARG_PTR_TO_CTX,
1306 .arg2_type = ARG_ANYTHING,
1307 .arg3_type = ARG_PTR_TO_STACK,
1308 .arg4_type = ARG_CONST_STACK_SIZE,
1309 .arg5_type = ARG_ANYTHING,
1312 #define BPF_HEADER_FIELD_SIZE(flags) ((flags) & 0x0f)
1313 #define BPF_IS_PSEUDO_HEADER(flags) ((flags) & 0x10)
1315 static u64 bpf_l3_csum_replace(u64 r1, u64 r2, u64 from, u64 to, u64 flags)
1317 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1318 int offset = (int) r2;
1321 if (unlikely((u32) offset > 0xffff))
1324 if (unlikely(skb_cloned(skb) &&
1325 !skb_clone_writable(skb, offset + sizeof(sum))))
1328 ptr = skb_header_pointer(skb, offset, sizeof(sum), &sum);
1332 switch (BPF_HEADER_FIELD_SIZE(flags)) {
1334 csum_replace2(ptr, from, to);
1337 csum_replace4(ptr, from, to);
1344 /* skb_store_bits guaranteed to not return -EFAULT here */
1345 skb_store_bits(skb, offset, ptr, sizeof(sum));
1350 const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1351 .func = bpf_l3_csum_replace,
1353 .ret_type = RET_INTEGER,
1354 .arg1_type = ARG_PTR_TO_CTX,
1355 .arg2_type = ARG_ANYTHING,
1356 .arg3_type = ARG_ANYTHING,
1357 .arg4_type = ARG_ANYTHING,
1358 .arg5_type = ARG_ANYTHING,
1361 static u64 bpf_l4_csum_replace(u64 r1, u64 r2, u64 from, u64 to, u64 flags)
1363 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1364 bool is_pseudo = !!BPF_IS_PSEUDO_HEADER(flags);
1365 int offset = (int) r2;
1368 if (unlikely((u32) offset > 0xffff))
1371 if (unlikely(skb_cloned(skb) &&
1372 !skb_clone_writable(skb, offset + sizeof(sum))))
1375 ptr = skb_header_pointer(skb, offset, sizeof(sum), &sum);
1379 switch (BPF_HEADER_FIELD_SIZE(flags)) {
1381 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1384 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1391 /* skb_store_bits guaranteed to not return -EFAULT here */
1392 skb_store_bits(skb, offset, ptr, sizeof(sum));
1397 const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1398 .func = bpf_l4_csum_replace,
1400 .ret_type = RET_INTEGER,
1401 .arg1_type = ARG_PTR_TO_CTX,
1402 .arg2_type = ARG_ANYTHING,
1403 .arg3_type = ARG_ANYTHING,
1404 .arg4_type = ARG_ANYTHING,
1405 .arg5_type = ARG_ANYTHING,
1408 #define BPF_IS_REDIRECT_INGRESS(flags) ((flags) & 1)
1410 static u64 bpf_clone_redirect(u64 r1, u64 ifindex, u64 flags, u64 r4, u64 r5)
1412 struct sk_buff *skb = (struct sk_buff *) (long) r1, *skb2;
1413 struct net_device *dev;
1415 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
1419 skb2 = skb_clone(skb, GFP_ATOMIC);
1420 if (unlikely(!skb2))
1423 if (BPF_IS_REDIRECT_INGRESS(flags))
1424 return dev_forward_skb(dev, skb2);
1427 skb_sender_cpu_clear(skb2);
1428 return dev_queue_xmit(skb2);
1431 const struct bpf_func_proto bpf_clone_redirect_proto = {
1432 .func = bpf_clone_redirect,
1434 .ret_type = RET_INTEGER,
1435 .arg1_type = ARG_PTR_TO_CTX,
1436 .arg2_type = ARG_ANYTHING,
1437 .arg3_type = ARG_ANYTHING,
1440 struct redirect_info {
1445 static DEFINE_PER_CPU(struct redirect_info, redirect_info);
1446 static u64 bpf_redirect(u64 ifindex, u64 flags, u64 r3, u64 r4, u64 r5)
1448 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
1450 ri->ifindex = ifindex;
1452 return TC_ACT_REDIRECT;
1455 int skb_do_redirect(struct sk_buff *skb)
1457 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
1458 struct net_device *dev;
1460 dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->ifindex);
1462 if (unlikely(!dev)) {
1467 if (BPF_IS_REDIRECT_INGRESS(ri->flags))
1468 return dev_forward_skb(dev, skb);
1471 skb_sender_cpu_clear(skb);
1472 return dev_queue_xmit(skb);
1475 const struct bpf_func_proto bpf_redirect_proto = {
1476 .func = bpf_redirect,
1478 .ret_type = RET_INTEGER,
1479 .arg1_type = ARG_ANYTHING,
1480 .arg2_type = ARG_ANYTHING,
1483 static u64 bpf_get_cgroup_classid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1485 return task_get_classid((struct sk_buff *) (unsigned long) r1);
1488 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
1489 .func = bpf_get_cgroup_classid,
1491 .ret_type = RET_INTEGER,
1492 .arg1_type = ARG_PTR_TO_CTX,
1495 static u64 bpf_get_route_realm(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1497 #ifdef CONFIG_IP_ROUTE_CLASSID
1498 const struct dst_entry *dst;
1500 dst = skb_dst((struct sk_buff *) (unsigned long) r1);
1502 return dst->tclassid;
1507 static const struct bpf_func_proto bpf_get_route_realm_proto = {
1508 .func = bpf_get_route_realm,
1510 .ret_type = RET_INTEGER,
1511 .arg1_type = ARG_PTR_TO_CTX,
1514 static u64 bpf_skb_vlan_push(u64 r1, u64 r2, u64 vlan_tci, u64 r4, u64 r5)
1516 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1517 __be16 vlan_proto = (__force __be16) r2;
1519 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
1520 vlan_proto != htons(ETH_P_8021AD)))
1521 vlan_proto = htons(ETH_P_8021Q);
1523 return skb_vlan_push(skb, vlan_proto, vlan_tci);
1526 const struct bpf_func_proto bpf_skb_vlan_push_proto = {
1527 .func = bpf_skb_vlan_push,
1529 .ret_type = RET_INTEGER,
1530 .arg1_type = ARG_PTR_TO_CTX,
1531 .arg2_type = ARG_ANYTHING,
1532 .arg3_type = ARG_ANYTHING,
1534 EXPORT_SYMBOL_GPL(bpf_skb_vlan_push_proto);
1536 static u64 bpf_skb_vlan_pop(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1538 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1540 return skb_vlan_pop(skb);
1543 const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
1544 .func = bpf_skb_vlan_pop,
1546 .ret_type = RET_INTEGER,
1547 .arg1_type = ARG_PTR_TO_CTX,
1549 EXPORT_SYMBOL_GPL(bpf_skb_vlan_pop_proto);
1551 bool bpf_helper_changes_skb_data(void *func)
1553 if (func == bpf_skb_vlan_push)
1555 if (func == bpf_skb_vlan_pop)
1560 static u64 bpf_skb_get_tunnel_key(u64 r1, u64 r2, u64 size, u64 flags, u64 r5)
1562 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1563 struct bpf_tunnel_key *to = (struct bpf_tunnel_key *) (long) r2;
1564 struct ip_tunnel_info *info = skb_tunnel_info(skb);
1566 if (unlikely(size != sizeof(struct bpf_tunnel_key) || flags || !info))
1568 if (ip_tunnel_info_af(info) != AF_INET)
1571 to->tunnel_id = be64_to_cpu(info->key.tun_id);
1572 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
1577 const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
1578 .func = bpf_skb_get_tunnel_key,
1580 .ret_type = RET_INTEGER,
1581 .arg1_type = ARG_PTR_TO_CTX,
1582 .arg2_type = ARG_PTR_TO_STACK,
1583 .arg3_type = ARG_CONST_STACK_SIZE,
1584 .arg4_type = ARG_ANYTHING,
1587 static struct metadata_dst __percpu *md_dst;
1589 static u64 bpf_skb_set_tunnel_key(u64 r1, u64 r2, u64 size, u64 flags, u64 r5)
1591 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1592 struct bpf_tunnel_key *from = (struct bpf_tunnel_key *) (long) r2;
1593 struct metadata_dst *md = this_cpu_ptr(md_dst);
1594 struct ip_tunnel_info *info;
1596 if (unlikely(size != sizeof(struct bpf_tunnel_key) || flags))
1600 dst_hold((struct dst_entry *) md);
1601 skb_dst_set(skb, (struct dst_entry *) md);
1603 info = &md->u.tun_info;
1604 info->mode = IP_TUNNEL_INFO_TX;
1605 info->key.tun_flags = TUNNEL_KEY;
1606 info->key.tun_id = cpu_to_be64(from->tunnel_id);
1607 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
1612 const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
1613 .func = bpf_skb_set_tunnel_key,
1615 .ret_type = RET_INTEGER,
1616 .arg1_type = ARG_PTR_TO_CTX,
1617 .arg2_type = ARG_PTR_TO_STACK,
1618 .arg3_type = ARG_CONST_STACK_SIZE,
1619 .arg4_type = ARG_ANYTHING,
1622 static const struct bpf_func_proto *bpf_get_skb_set_tunnel_key_proto(void)
1625 /* race is not possible, since it's called from
1626 * verifier that is holding verifier mutex
1628 md_dst = metadata_dst_alloc_percpu(0, GFP_KERNEL);
1632 return &bpf_skb_set_tunnel_key_proto;
1635 static const struct bpf_func_proto *
1636 sk_filter_func_proto(enum bpf_func_id func_id)
1639 case BPF_FUNC_map_lookup_elem:
1640 return &bpf_map_lookup_elem_proto;
1641 case BPF_FUNC_map_update_elem:
1642 return &bpf_map_update_elem_proto;
1643 case BPF_FUNC_map_delete_elem:
1644 return &bpf_map_delete_elem_proto;
1645 case BPF_FUNC_get_prandom_u32:
1646 return &bpf_get_prandom_u32_proto;
1647 case BPF_FUNC_get_smp_processor_id:
1648 return &bpf_get_smp_processor_id_proto;
1649 case BPF_FUNC_tail_call:
1650 return &bpf_tail_call_proto;
1651 case BPF_FUNC_ktime_get_ns:
1652 return &bpf_ktime_get_ns_proto;
1653 case BPF_FUNC_trace_printk:
1654 if (capable(CAP_SYS_ADMIN))
1655 return bpf_get_trace_printk_proto();
1661 static const struct bpf_func_proto *
1662 tc_cls_act_func_proto(enum bpf_func_id func_id)
1665 case BPF_FUNC_skb_store_bytes:
1666 return &bpf_skb_store_bytes_proto;
1667 case BPF_FUNC_l3_csum_replace:
1668 return &bpf_l3_csum_replace_proto;
1669 case BPF_FUNC_l4_csum_replace:
1670 return &bpf_l4_csum_replace_proto;
1671 case BPF_FUNC_clone_redirect:
1672 return &bpf_clone_redirect_proto;
1673 case BPF_FUNC_get_cgroup_classid:
1674 return &bpf_get_cgroup_classid_proto;
1675 case BPF_FUNC_skb_vlan_push:
1676 return &bpf_skb_vlan_push_proto;
1677 case BPF_FUNC_skb_vlan_pop:
1678 return &bpf_skb_vlan_pop_proto;
1679 case BPF_FUNC_skb_get_tunnel_key:
1680 return &bpf_skb_get_tunnel_key_proto;
1681 case BPF_FUNC_skb_set_tunnel_key:
1682 return bpf_get_skb_set_tunnel_key_proto();
1683 case BPF_FUNC_redirect:
1684 return &bpf_redirect_proto;
1685 case BPF_FUNC_get_route_realm:
1686 return &bpf_get_route_realm_proto;
1688 return sk_filter_func_proto(func_id);
1692 static bool __is_valid_access(int off, int size, enum bpf_access_type type)
1695 if (off < 0 || off >= sizeof(struct __sk_buff))
1698 /* disallow misaligned access */
1699 if (off % size != 0)
1702 /* all __sk_buff fields are __u32 */
1709 static bool sk_filter_is_valid_access(int off, int size,
1710 enum bpf_access_type type)
1712 if (off == offsetof(struct __sk_buff, tc_classid))
1715 if (type == BPF_WRITE) {
1717 case offsetof(struct __sk_buff, cb[0]) ...
1718 offsetof(struct __sk_buff, cb[4]):
1725 return __is_valid_access(off, size, type);
1728 static bool tc_cls_act_is_valid_access(int off, int size,
1729 enum bpf_access_type type)
1731 if (off == offsetof(struct __sk_buff, tc_classid))
1732 return type == BPF_WRITE ? true : false;
1734 if (type == BPF_WRITE) {
1736 case offsetof(struct __sk_buff, mark):
1737 case offsetof(struct __sk_buff, tc_index):
1738 case offsetof(struct __sk_buff, priority):
1739 case offsetof(struct __sk_buff, cb[0]) ...
1740 offsetof(struct __sk_buff, cb[4]):
1746 return __is_valid_access(off, size, type);
1749 static u32 bpf_net_convert_ctx_access(enum bpf_access_type type, int dst_reg,
1750 int src_reg, int ctx_off,
1751 struct bpf_insn *insn_buf,
1752 struct bpf_prog *prog)
1754 struct bpf_insn *insn = insn_buf;
1757 case offsetof(struct __sk_buff, len):
1758 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, len) != 4);
1760 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
1761 offsetof(struct sk_buff, len));
1764 case offsetof(struct __sk_buff, protocol):
1765 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
1767 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
1768 offsetof(struct sk_buff, protocol));
1771 case offsetof(struct __sk_buff, vlan_proto):
1772 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2);
1774 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
1775 offsetof(struct sk_buff, vlan_proto));
1778 case offsetof(struct __sk_buff, priority):
1779 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, priority) != 4);
1781 if (type == BPF_WRITE)
1782 *insn++ = BPF_STX_MEM(BPF_W, dst_reg, src_reg,
1783 offsetof(struct sk_buff, priority));
1785 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
1786 offsetof(struct sk_buff, priority));
1789 case offsetof(struct __sk_buff, ingress_ifindex):
1790 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, skb_iif) != 4);
1792 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
1793 offsetof(struct sk_buff, skb_iif));
1796 case offsetof(struct __sk_buff, ifindex):
1797 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
1799 *insn++ = BPF_LDX_MEM(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, dev)),
1801 offsetof(struct sk_buff, dev));
1802 *insn++ = BPF_JMP_IMM(BPF_JEQ, dst_reg, 0, 1);
1803 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, dst_reg,
1804 offsetof(struct net_device, ifindex));
1807 case offsetof(struct __sk_buff, hash):
1808 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
1810 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
1811 offsetof(struct sk_buff, hash));
1814 case offsetof(struct __sk_buff, mark):
1815 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
1817 if (type == BPF_WRITE)
1818 *insn++ = BPF_STX_MEM(BPF_W, dst_reg, src_reg,
1819 offsetof(struct sk_buff, mark));
1821 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
1822 offsetof(struct sk_buff, mark));
1825 case offsetof(struct __sk_buff, pkt_type):
1826 return convert_skb_access(SKF_AD_PKTTYPE, dst_reg, src_reg, insn);
1828 case offsetof(struct __sk_buff, queue_mapping):
1829 return convert_skb_access(SKF_AD_QUEUE, dst_reg, src_reg, insn);
1831 case offsetof(struct __sk_buff, vlan_present):
1832 return convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
1833 dst_reg, src_reg, insn);
1835 case offsetof(struct __sk_buff, vlan_tci):
1836 return convert_skb_access(SKF_AD_VLAN_TAG,
1837 dst_reg, src_reg, insn);
1839 case offsetof(struct __sk_buff, cb[0]) ...
1840 offsetof(struct __sk_buff, cb[4]):
1841 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20);
1843 prog->cb_access = 1;
1844 ctx_off -= offsetof(struct __sk_buff, cb[0]);
1845 ctx_off += offsetof(struct sk_buff, cb);
1846 ctx_off += offsetof(struct qdisc_skb_cb, data);
1847 if (type == BPF_WRITE)
1848 *insn++ = BPF_STX_MEM(BPF_W, dst_reg, src_reg, ctx_off);
1850 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, ctx_off);
1853 case offsetof(struct __sk_buff, tc_classid):
1854 ctx_off -= offsetof(struct __sk_buff, tc_classid);
1855 ctx_off += offsetof(struct sk_buff, cb);
1856 ctx_off += offsetof(struct qdisc_skb_cb, tc_classid);
1857 WARN_ON(type != BPF_WRITE);
1858 *insn++ = BPF_STX_MEM(BPF_H, dst_reg, src_reg, ctx_off);
1861 case offsetof(struct __sk_buff, tc_index):
1862 #ifdef CONFIG_NET_SCHED
1863 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, tc_index) != 2);
1865 if (type == BPF_WRITE)
1866 *insn++ = BPF_STX_MEM(BPF_H, dst_reg, src_reg,
1867 offsetof(struct sk_buff, tc_index));
1869 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
1870 offsetof(struct sk_buff, tc_index));
1873 if (type == BPF_WRITE)
1874 *insn++ = BPF_MOV64_REG(dst_reg, dst_reg);
1876 *insn++ = BPF_MOV64_IMM(dst_reg, 0);
1881 return insn - insn_buf;
1884 static const struct bpf_verifier_ops sk_filter_ops = {
1885 .get_func_proto = sk_filter_func_proto,
1886 .is_valid_access = sk_filter_is_valid_access,
1887 .convert_ctx_access = bpf_net_convert_ctx_access,
1890 static const struct bpf_verifier_ops tc_cls_act_ops = {
1891 .get_func_proto = tc_cls_act_func_proto,
1892 .is_valid_access = tc_cls_act_is_valid_access,
1893 .convert_ctx_access = bpf_net_convert_ctx_access,
1896 static struct bpf_prog_type_list sk_filter_type __read_mostly = {
1897 .ops = &sk_filter_ops,
1898 .type = BPF_PROG_TYPE_SOCKET_FILTER,
1901 static struct bpf_prog_type_list sched_cls_type __read_mostly = {
1902 .ops = &tc_cls_act_ops,
1903 .type = BPF_PROG_TYPE_SCHED_CLS,
1906 static struct bpf_prog_type_list sched_act_type __read_mostly = {
1907 .ops = &tc_cls_act_ops,
1908 .type = BPF_PROG_TYPE_SCHED_ACT,
1911 static int __init register_sk_filter_ops(void)
1913 bpf_register_prog_type(&sk_filter_type);
1914 bpf_register_prog_type(&sched_cls_type);
1915 bpf_register_prog_type(&sched_act_type);
1919 late_initcall(register_sk_filter_ops);
1921 int __sk_detach_filter(struct sock *sk, bool locked)
1924 struct sk_filter *filter;
1926 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1929 filter = rcu_dereference_protected(sk->sk_filter, locked);
1931 RCU_INIT_POINTER(sk->sk_filter, NULL);
1932 sk_filter_uncharge(sk, filter);
1938 EXPORT_SYMBOL_GPL(__sk_detach_filter);
1940 int sk_detach_filter(struct sock *sk)
1942 return __sk_detach_filter(sk, sock_owned_by_user(sk));
1945 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
1948 struct sock_fprog_kern *fprog;
1949 struct sk_filter *filter;
1953 filter = rcu_dereference_protected(sk->sk_filter,
1954 sock_owned_by_user(sk));
1958 /* We're copying the filter that has been originally attached,
1959 * so no conversion/decode needed anymore. eBPF programs that
1960 * have no original program cannot be dumped through this.
1963 fprog = filter->prog->orig_prog;
1969 /* User space only enquires number of filter blocks. */
1973 if (len < fprog->len)
1977 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
1980 /* Instead of bytes, the API requests to return the number