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 sk_chk_filter()
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>
48 /* No hurry in this branch
50 * Exported for the bpf jit load helper.
52 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, int k, unsigned int size)
57 ptr = skb_network_header(skb) + k - SKF_NET_OFF;
58 else if (k >= SKF_LL_OFF)
59 ptr = skb_mac_header(skb) + k - SKF_LL_OFF;
61 if (ptr >= skb->head && ptr + size <= skb_tail_pointer(skb))
66 static inline void *load_pointer(const struct sk_buff *skb, int k,
67 unsigned int size, void *buffer)
70 return skb_header_pointer(skb, k, size, buffer);
71 return bpf_internal_load_pointer_neg_helper(skb, k, size);
75 * sk_filter - run a packet through a socket filter
76 * @sk: sock associated with &sk_buff
77 * @skb: buffer to filter
79 * Run the filter code and then cut skb->data to correct size returned by
80 * sk_run_filter. If pkt_len is 0 we toss packet. If skb->len is smaller
81 * than pkt_len we keep whole skb->data. This is the socket level
82 * wrapper to sk_run_filter. It returns 0 if the packet should
83 * be accepted or -EPERM if the packet should be tossed.
86 int sk_filter(struct sock *sk, struct sk_buff *skb)
89 struct sk_filter *filter;
92 * If the skb was allocated from pfmemalloc reserves, only
93 * allow SOCK_MEMALLOC sockets to use it as this socket is
96 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
99 err = security_sock_rcv_skb(sk, skb);
104 filter = rcu_dereference(sk->sk_filter);
106 unsigned int pkt_len = SK_RUN_FILTER(filter, skb);
108 err = pkt_len ? pskb_trim(skb, pkt_len) : -EPERM;
114 EXPORT_SYMBOL(sk_filter);
116 /* Base function for offset calculation. Needs to go into .text section,
117 * therefore keeping it non-static as well; will also be used by JITs
118 * anyway later on, so do not let the compiler omit it.
120 noinline u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
126 * __sk_run_filter - run a filter on a given context
127 * @ctx: buffer to run the filter on
128 * @insn: filter to apply
130 * Decode and apply filter instructions to the skb->data. Return length to
131 * keep, 0 for none. @ctx is the data we are operating on, @insn is the
132 * array of filter instructions.
134 unsigned int __sk_run_filter(void *ctx, const struct sock_filter_int *insn)
136 u64 stack[MAX_BPF_STACK / sizeof(u64)];
137 u64 regs[MAX_BPF_REG], tmp;
142 #define A regs[insn->a_reg]
143 #define X regs[insn->x_reg]
146 #define CONT ({insn++; goto select_insn; })
147 #define CONT_JMP ({insn++; goto select_insn; })
149 static const void *jumptable[256] = {
150 [0 ... 255] = &&default_label,
151 /* Now overwrite non-defaults ... */
152 #define DL(A, B, C) [A|B|C] = &&A##_##B##_##C
153 DL(BPF_ALU, BPF_ADD, BPF_X),
154 DL(BPF_ALU, BPF_ADD, BPF_K),
155 DL(BPF_ALU, BPF_SUB, BPF_X),
156 DL(BPF_ALU, BPF_SUB, BPF_K),
157 DL(BPF_ALU, BPF_AND, BPF_X),
158 DL(BPF_ALU, BPF_AND, BPF_K),
159 DL(BPF_ALU, BPF_OR, BPF_X),
160 DL(BPF_ALU, BPF_OR, BPF_K),
161 DL(BPF_ALU, BPF_LSH, BPF_X),
162 DL(BPF_ALU, BPF_LSH, BPF_K),
163 DL(BPF_ALU, BPF_RSH, BPF_X),
164 DL(BPF_ALU, BPF_RSH, BPF_K),
165 DL(BPF_ALU, BPF_XOR, BPF_X),
166 DL(BPF_ALU, BPF_XOR, BPF_K),
167 DL(BPF_ALU, BPF_MUL, BPF_X),
168 DL(BPF_ALU, BPF_MUL, BPF_K),
169 DL(BPF_ALU, BPF_MOV, BPF_X),
170 DL(BPF_ALU, BPF_MOV, BPF_K),
171 DL(BPF_ALU, BPF_DIV, BPF_X),
172 DL(BPF_ALU, BPF_DIV, BPF_K),
173 DL(BPF_ALU, BPF_MOD, BPF_X),
174 DL(BPF_ALU, BPF_MOD, BPF_K),
175 DL(BPF_ALU, BPF_NEG, 0),
176 DL(BPF_ALU, BPF_END, BPF_TO_BE),
177 DL(BPF_ALU, BPF_END, BPF_TO_LE),
178 DL(BPF_ALU64, BPF_ADD, BPF_X),
179 DL(BPF_ALU64, BPF_ADD, BPF_K),
180 DL(BPF_ALU64, BPF_SUB, BPF_X),
181 DL(BPF_ALU64, BPF_SUB, BPF_K),
182 DL(BPF_ALU64, BPF_AND, BPF_X),
183 DL(BPF_ALU64, BPF_AND, BPF_K),
184 DL(BPF_ALU64, BPF_OR, BPF_X),
185 DL(BPF_ALU64, BPF_OR, BPF_K),
186 DL(BPF_ALU64, BPF_LSH, BPF_X),
187 DL(BPF_ALU64, BPF_LSH, BPF_K),
188 DL(BPF_ALU64, BPF_RSH, BPF_X),
189 DL(BPF_ALU64, BPF_RSH, BPF_K),
190 DL(BPF_ALU64, BPF_XOR, BPF_X),
191 DL(BPF_ALU64, BPF_XOR, BPF_K),
192 DL(BPF_ALU64, BPF_MUL, BPF_X),
193 DL(BPF_ALU64, BPF_MUL, BPF_K),
194 DL(BPF_ALU64, BPF_MOV, BPF_X),
195 DL(BPF_ALU64, BPF_MOV, BPF_K),
196 DL(BPF_ALU64, BPF_ARSH, BPF_X),
197 DL(BPF_ALU64, BPF_ARSH, BPF_K),
198 DL(BPF_ALU64, BPF_DIV, BPF_X),
199 DL(BPF_ALU64, BPF_DIV, BPF_K),
200 DL(BPF_ALU64, BPF_MOD, BPF_X),
201 DL(BPF_ALU64, BPF_MOD, BPF_K),
202 DL(BPF_ALU64, BPF_NEG, 0),
203 DL(BPF_JMP, BPF_CALL, 0),
204 DL(BPF_JMP, BPF_JA, 0),
205 DL(BPF_JMP, BPF_JEQ, BPF_X),
206 DL(BPF_JMP, BPF_JEQ, BPF_K),
207 DL(BPF_JMP, BPF_JNE, BPF_X),
208 DL(BPF_JMP, BPF_JNE, BPF_K),
209 DL(BPF_JMP, BPF_JGT, BPF_X),
210 DL(BPF_JMP, BPF_JGT, BPF_K),
211 DL(BPF_JMP, BPF_JGE, BPF_X),
212 DL(BPF_JMP, BPF_JGE, BPF_K),
213 DL(BPF_JMP, BPF_JSGT, BPF_X),
214 DL(BPF_JMP, BPF_JSGT, BPF_K),
215 DL(BPF_JMP, BPF_JSGE, BPF_X),
216 DL(BPF_JMP, BPF_JSGE, BPF_K),
217 DL(BPF_JMP, BPF_JSET, BPF_X),
218 DL(BPF_JMP, BPF_JSET, BPF_K),
219 DL(BPF_JMP, BPF_EXIT, 0),
220 DL(BPF_STX, BPF_MEM, BPF_B),
221 DL(BPF_STX, BPF_MEM, BPF_H),
222 DL(BPF_STX, BPF_MEM, BPF_W),
223 DL(BPF_STX, BPF_MEM, BPF_DW),
224 DL(BPF_STX, BPF_XADD, BPF_W),
225 DL(BPF_STX, BPF_XADD, BPF_DW),
226 DL(BPF_ST, BPF_MEM, BPF_B),
227 DL(BPF_ST, BPF_MEM, BPF_H),
228 DL(BPF_ST, BPF_MEM, BPF_W),
229 DL(BPF_ST, BPF_MEM, BPF_DW),
230 DL(BPF_LDX, BPF_MEM, BPF_B),
231 DL(BPF_LDX, BPF_MEM, BPF_H),
232 DL(BPF_LDX, BPF_MEM, BPF_W),
233 DL(BPF_LDX, BPF_MEM, BPF_DW),
234 DL(BPF_LD, BPF_ABS, BPF_W),
235 DL(BPF_LD, BPF_ABS, BPF_H),
236 DL(BPF_LD, BPF_ABS, BPF_B),
237 DL(BPF_LD, BPF_IND, BPF_W),
238 DL(BPF_LD, BPF_IND, BPF_H),
239 DL(BPF_LD, BPF_IND, BPF_B),
243 regs[FP_REG] = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)];
244 regs[ARG1_REG] = (u64) (unsigned long) ctx;
247 goto *jumptable[insn->code];
250 #define ALU(OPCODE, OP) \
251 BPF_ALU64_##OPCODE##_BPF_X: \
254 BPF_ALU_##OPCODE##_BPF_X: \
255 A = (u32) A OP (u32) X; \
257 BPF_ALU64_##OPCODE##_BPF_K: \
260 BPF_ALU_##OPCODE##_BPF_K: \
261 A = (u32) A OP (u32) K; \
279 BPF_ALU_BPF_MOV_BPF_X:
282 BPF_ALU_BPF_MOV_BPF_K:
285 BPF_ALU64_BPF_MOV_BPF_X:
288 BPF_ALU64_BPF_MOV_BPF_K:
291 BPF_ALU64_BPF_ARSH_BPF_X:
294 BPF_ALU64_BPF_ARSH_BPF_K:
297 BPF_ALU64_BPF_MOD_BPF_X:
298 if (unlikely(X == 0))
303 BPF_ALU_BPF_MOD_BPF_X:
304 if (unlikely(X == 0))
307 A = do_div(tmp, (u32) X);
309 BPF_ALU64_BPF_MOD_BPF_K:
313 BPF_ALU_BPF_MOD_BPF_K:
315 A = do_div(tmp, (u32) K);
317 BPF_ALU64_BPF_DIV_BPF_X:
318 if (unlikely(X == 0))
322 BPF_ALU_BPF_DIV_BPF_X:
323 if (unlikely(X == 0))
326 do_div(tmp, (u32) X);
329 BPF_ALU64_BPF_DIV_BPF_K:
332 BPF_ALU_BPF_DIV_BPF_K:
334 do_div(tmp, (u32) K);
337 BPF_ALU_BPF_END_BPF_TO_BE:
340 A = (__force u16) cpu_to_be16(A);
343 A = (__force u32) cpu_to_be32(A);
346 A = (__force u64) cpu_to_be64(A);
350 BPF_ALU_BPF_END_BPF_TO_LE:
353 A = (__force u16) cpu_to_le16(A);
356 A = (__force u32) cpu_to_le32(A);
359 A = (__force u64) cpu_to_le64(A);
366 /* Function call scratches R1-R5 registers, preserves R6-R9,
367 * and stores return value into R0.
369 R0 = (__bpf_call_base + insn->imm)(regs[1], regs[2], regs[3],
377 BPF_JMP_BPF_JEQ_BPF_X:
383 BPF_JMP_BPF_JEQ_BPF_K:
389 BPF_JMP_BPF_JNE_BPF_X:
395 BPF_JMP_BPF_JNE_BPF_K:
401 BPF_JMP_BPF_JGT_BPF_X:
407 BPF_JMP_BPF_JGT_BPF_K:
413 BPF_JMP_BPF_JGE_BPF_X:
419 BPF_JMP_BPF_JGE_BPF_K:
425 BPF_JMP_BPF_JSGT_BPF_X:
426 if (((s64)A) > ((s64)X)) {
431 BPF_JMP_BPF_JSGT_BPF_K:
432 if (((s64)A) > ((s64)K)) {
437 BPF_JMP_BPF_JSGE_BPF_X:
438 if (((s64)A) >= ((s64)X)) {
443 BPF_JMP_BPF_JSGE_BPF_K:
444 if (((s64)A) >= ((s64)K)) {
449 BPF_JMP_BPF_JSET_BPF_X:
455 BPF_JMP_BPF_JSET_BPF_K:
464 /* STX and ST and LDX*/
465 #define LDST(SIZEOP, SIZE) \
466 BPF_STX_BPF_MEM_##SIZEOP: \
467 *(SIZE *)(unsigned long) (A + insn->off) = X; \
469 BPF_ST_BPF_MEM_##SIZEOP: \
470 *(SIZE *)(unsigned long) (A + insn->off) = K; \
472 BPF_LDX_BPF_MEM_##SIZEOP: \
473 A = *(SIZE *)(unsigned long) (X + insn->off); \
481 BPF_STX_BPF_XADD_BPF_W: /* lock xadd *(u32 *)(A + insn->off) += X */
482 atomic_add((u32) X, (atomic_t *)(unsigned long)
485 BPF_STX_BPF_XADD_BPF_DW: /* lock xadd *(u64 *)(A + insn->off) += X */
486 atomic64_add((u64) X, (atomic64_t *)(unsigned long)
489 BPF_LD_BPF_ABS_BPF_W: /* R0 = ntohl(*(u32 *) (skb->data + K)) */
492 /* BPF_LD + BPD_ABS and BPF_LD + BPF_IND insns are only
493 * appearing in the programs where ctx == skb. All programs
494 * keep 'ctx' in regs[CTX_REG] == R6, sk_convert_filter()
495 * saves it in R6, internal BPF verifier will check that
498 * BPF_ABS and BPF_IND are wrappers of function calls, so
499 * they scratch R1-R5 registers, preserve R6-R9, and store
500 * return value into R0.
507 * K == 32-bit immediate
510 * R0 - 8/16/32-bit skb data converted to cpu endianness
512 ptr = load_pointer((struct sk_buff *) ctx, off, 4, &tmp);
513 if (likely(ptr != NULL)) {
514 R0 = get_unaligned_be32(ptr);
518 BPF_LD_BPF_ABS_BPF_H: /* R0 = ntohs(*(u16 *) (skb->data + K)) */
521 ptr = load_pointer((struct sk_buff *) ctx, off, 2, &tmp);
522 if (likely(ptr != NULL)) {
523 R0 = get_unaligned_be16(ptr);
527 BPF_LD_BPF_ABS_BPF_B: /* R0 = *(u8 *) (ctx + K) */
530 ptr = load_pointer((struct sk_buff *) ctx, off, 1, &tmp);
531 if (likely(ptr != NULL)) {
536 BPF_LD_BPF_IND_BPF_W: /* R0 = ntohl(*(u32 *) (skb->data + X + K)) */
539 BPF_LD_BPF_IND_BPF_H: /* R0 = ntohs(*(u16 *) (skb->data + X + K)) */
542 BPF_LD_BPF_IND_BPF_B: /* R0 = *(u8 *) (skb->data + X + K) */
547 /* If we ever reach this, we have a bug somewhere. */
548 WARN_RATELIMIT(1, "unknown opcode %02x\n", insn->code);
559 u32 sk_run_filter_int_seccomp(const struct seccomp_data *ctx,
560 const struct sock_filter_int *insni)
561 __attribute__ ((alias ("__sk_run_filter")));
563 u32 sk_run_filter_int_skb(const struct sk_buff *ctx,
564 const struct sock_filter_int *insni)
565 __attribute__ ((alias ("__sk_run_filter")));
566 EXPORT_SYMBOL_GPL(sk_run_filter_int_skb);
568 /* Helper to find the offset of pkt_type in sk_buff structure. We want
569 * to make sure its still a 3bit field starting at a byte boundary;
570 * taken from arch/x86/net/bpf_jit_comp.c.
572 #define PKT_TYPE_MAX 7
573 static unsigned int pkt_type_offset(void)
575 struct sk_buff skb_probe = { .pkt_type = ~0, };
576 u8 *ct = (u8 *) &skb_probe;
579 for (off = 0; off < sizeof(struct sk_buff); off++) {
580 if (ct[off] == PKT_TYPE_MAX)
584 pr_err_once("Please fix %s, as pkt_type couldn't be found!\n", __func__);
588 static u64 __skb_get_pay_offset(u64 ctx, u64 A, u64 X, u64 r4, u64 r5)
590 struct sk_buff *skb = (struct sk_buff *)(long) ctx;
592 return __skb_get_poff(skb);
595 static u64 __skb_get_nlattr(u64 ctx, u64 A, u64 X, u64 r4, u64 r5)
597 struct sk_buff *skb = (struct sk_buff *)(long) ctx;
600 if (skb_is_nonlinear(skb))
603 if (A > skb->len - sizeof(struct nlattr))
606 nla = nla_find((struct nlattr *) &skb->data[A], skb->len - A, X);
608 return (void *) nla - (void *) skb->data;
613 static u64 __skb_get_nlattr_nest(u64 ctx, u64 A, u64 X, u64 r4, u64 r5)
615 struct sk_buff *skb = (struct sk_buff *)(long) ctx;
618 if (skb_is_nonlinear(skb))
621 if (A > skb->len - sizeof(struct nlattr))
624 nla = (struct nlattr *) &skb->data[A];
625 if (nla->nla_len > A - skb->len)
628 nla = nla_find_nested(nla, X);
630 return (void *) nla - (void *) skb->data;
635 static u64 __get_raw_cpu_id(u64 ctx, u64 A, u64 X, u64 r4, u64 r5)
637 return raw_smp_processor_id();
640 /* Register mappings for user programs. */
647 static bool convert_bpf_extensions(struct sock_filter *fp,
648 struct sock_filter_int **insnp)
650 struct sock_filter_int *insn = *insnp;
653 case SKF_AD_OFF + SKF_AD_PROTOCOL:
654 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
656 insn->code = BPF_LDX | BPF_MEM | BPF_H;
658 insn->x_reg = CTX_REG;
659 insn->off = offsetof(struct sk_buff, protocol);
662 /* A = ntohs(A) [emitting a nop or swap16] */
663 insn->code = BPF_ALU | BPF_END | BPF_FROM_BE;
668 case SKF_AD_OFF + SKF_AD_PKTTYPE:
669 insn->code = BPF_LDX | BPF_MEM | BPF_B;
671 insn->x_reg = CTX_REG;
672 insn->off = pkt_type_offset();
677 insn->code = BPF_ALU | BPF_AND | BPF_K;
679 insn->imm = PKT_TYPE_MAX;
682 case SKF_AD_OFF + SKF_AD_IFINDEX:
683 case SKF_AD_OFF + SKF_AD_HATYPE:
684 if (FIELD_SIZEOF(struct sk_buff, dev) == 8)
685 insn->code = BPF_LDX | BPF_MEM | BPF_DW;
687 insn->code = BPF_LDX | BPF_MEM | BPF_W;
688 insn->a_reg = TMP_REG;
689 insn->x_reg = CTX_REG;
690 insn->off = offsetof(struct sk_buff, dev);
693 insn->code = BPF_JMP | BPF_JNE | BPF_K;
694 insn->a_reg = TMP_REG;
699 insn->code = BPF_JMP | BPF_EXIT;
702 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
703 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, type) != 2);
706 insn->x_reg = TMP_REG;
708 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX) {
709 insn->code = BPF_LDX | BPF_MEM | BPF_W;
710 insn->off = offsetof(struct net_device, ifindex);
712 insn->code = BPF_LDX | BPF_MEM | BPF_H;
713 insn->off = offsetof(struct net_device, type);
717 case SKF_AD_OFF + SKF_AD_MARK:
718 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
720 insn->code = BPF_LDX | BPF_MEM | BPF_W;
722 insn->x_reg = CTX_REG;
723 insn->off = offsetof(struct sk_buff, mark);
726 case SKF_AD_OFF + SKF_AD_RXHASH:
727 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
729 insn->code = BPF_LDX | BPF_MEM | BPF_W;
731 insn->x_reg = CTX_REG;
732 insn->off = offsetof(struct sk_buff, hash);
735 case SKF_AD_OFF + SKF_AD_QUEUE:
736 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2);
738 insn->code = BPF_LDX | BPF_MEM | BPF_H;
740 insn->x_reg = CTX_REG;
741 insn->off = offsetof(struct sk_buff, queue_mapping);
744 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
745 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
746 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2);
748 insn->code = BPF_LDX | BPF_MEM | BPF_H;
750 insn->x_reg = CTX_REG;
751 insn->off = offsetof(struct sk_buff, vlan_tci);
754 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000);
756 if (fp->k == SKF_AD_OFF + SKF_AD_VLAN_TAG) {
757 insn->code = BPF_ALU | BPF_AND | BPF_K;
759 insn->imm = ~VLAN_TAG_PRESENT;
761 insn->code = BPF_ALU | BPF_RSH | BPF_K;
766 insn->code = BPF_ALU | BPF_AND | BPF_K;
772 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
773 case SKF_AD_OFF + SKF_AD_NLATTR:
774 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
775 case SKF_AD_OFF + SKF_AD_CPU:
777 insn->code = BPF_ALU64 | BPF_MOV | BPF_X;
778 insn->a_reg = ARG1_REG;
779 insn->x_reg = CTX_REG;
783 insn->code = BPF_ALU64 | BPF_MOV | BPF_X;
784 insn->a_reg = ARG2_REG;
789 insn->code = BPF_ALU64 | BPF_MOV | BPF_X;
790 insn->a_reg = ARG3_REG;
794 /* Emit call(ctx, arg2=A, arg3=X) */
795 insn->code = BPF_JMP | BPF_CALL;
797 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
798 insn->imm = __skb_get_pay_offset - __bpf_call_base;
800 case SKF_AD_OFF + SKF_AD_NLATTR:
801 insn->imm = __skb_get_nlattr - __bpf_call_base;
803 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
804 insn->imm = __skb_get_nlattr_nest - __bpf_call_base;
806 case SKF_AD_OFF + SKF_AD_CPU:
807 insn->imm = __get_raw_cpu_id - __bpf_call_base;
812 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
813 insn->code = BPF_ALU | BPF_XOR | BPF_X;
819 /* This is just a dummy call to avoid letting the compiler
820 * evict __bpf_call_base() as an optimization. Placed here
821 * where no-one bothers.
823 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
832 * sk_convert_filter - convert filter program
833 * @prog: the user passed filter program
834 * @len: the length of the user passed filter program
835 * @new_prog: buffer where converted program will be stored
836 * @new_len: pointer to store length of converted program
838 * Remap 'sock_filter' style BPF instruction set to 'sock_filter_ext' style.
839 * Conversion workflow:
841 * 1) First pass for calculating the new program length:
842 * sk_convert_filter(old_prog, old_len, NULL, &new_len)
844 * 2) 2nd pass to remap in two passes: 1st pass finds new
845 * jump offsets, 2nd pass remapping:
846 * new_prog = kmalloc(sizeof(struct sock_filter_int) * new_len);
847 * sk_convert_filter(old_prog, old_len, new_prog, &new_len);
849 * User BPF's register A is mapped to our BPF register 6, user BPF
850 * register X is mapped to BPF register 7; frame pointer is always
851 * register 10; Context 'void *ctx' is stored in register 1, that is,
852 * for socket filters: ctx == 'struct sk_buff *', for seccomp:
853 * ctx == 'struct seccomp_data *'.
855 int sk_convert_filter(struct sock_filter *prog, int len,
856 struct sock_filter_int *new_prog, int *new_len)
858 int new_flen = 0, pass = 0, target, i;
859 struct sock_filter_int *new_insn;
860 struct sock_filter *fp;
864 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
865 BUILD_BUG_ON(FP_REG + 1 != MAX_BPF_REG);
867 if (len <= 0 || len >= BPF_MAXINSNS)
871 addrs = kzalloc(len * sizeof(*addrs), GFP_KERNEL);
881 new_insn->code = BPF_ALU64 | BPF_MOV | BPF_X;
882 new_insn->a_reg = CTX_REG;
883 new_insn->x_reg = ARG1_REG;
887 for (i = 0; i < len; fp++, i++) {
888 struct sock_filter_int tmp_insns[6] = { };
889 struct sock_filter_int *insn = tmp_insns;
892 addrs[i] = new_insn - new_prog;
895 /* All arithmetic insns and skb loads map as-is. */
896 case BPF_ALU | BPF_ADD | BPF_X:
897 case BPF_ALU | BPF_ADD | BPF_K:
898 case BPF_ALU | BPF_SUB | BPF_X:
899 case BPF_ALU | BPF_SUB | BPF_K:
900 case BPF_ALU | BPF_AND | BPF_X:
901 case BPF_ALU | BPF_AND | BPF_K:
902 case BPF_ALU | BPF_OR | BPF_X:
903 case BPF_ALU | BPF_OR | BPF_K:
904 case BPF_ALU | BPF_LSH | BPF_X:
905 case BPF_ALU | BPF_LSH | BPF_K:
906 case BPF_ALU | BPF_RSH | BPF_X:
907 case BPF_ALU | BPF_RSH | BPF_K:
908 case BPF_ALU | BPF_XOR | BPF_X:
909 case BPF_ALU | BPF_XOR | BPF_K:
910 case BPF_ALU | BPF_MUL | BPF_X:
911 case BPF_ALU | BPF_MUL | BPF_K:
912 case BPF_ALU | BPF_DIV | BPF_X:
913 case BPF_ALU | BPF_DIV | BPF_K:
914 case BPF_ALU | BPF_MOD | BPF_X:
915 case BPF_ALU | BPF_MOD | BPF_K:
916 case BPF_ALU | BPF_NEG:
917 case BPF_LD | BPF_ABS | BPF_W:
918 case BPF_LD | BPF_ABS | BPF_H:
919 case BPF_LD | BPF_ABS | BPF_B:
920 case BPF_LD | BPF_IND | BPF_W:
921 case BPF_LD | BPF_IND | BPF_H:
922 case BPF_LD | BPF_IND | BPF_B:
923 /* Check for overloaded BPF extension and
924 * directly convert it if found, otherwise
925 * just move on with mapping.
927 if (BPF_CLASS(fp->code) == BPF_LD &&
928 BPF_MODE(fp->code) == BPF_ABS &&
929 convert_bpf_extensions(fp, &insn))
932 insn->code = fp->code;
938 /* Jump opcodes map as-is, but offsets need adjustment. */
939 case BPF_JMP | BPF_JA:
940 target = i + fp->k + 1;
941 insn->code = fp->code;
944 if (target >= len || target < 0) \
946 insn->off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
947 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
948 insn->off -= insn - tmp_insns; \
954 case BPF_JMP | BPF_JEQ | BPF_K:
955 case BPF_JMP | BPF_JEQ | BPF_X:
956 case BPF_JMP | BPF_JSET | BPF_K:
957 case BPF_JMP | BPF_JSET | BPF_X:
958 case BPF_JMP | BPF_JGT | BPF_K:
959 case BPF_JMP | BPF_JGT | BPF_X:
960 case BPF_JMP | BPF_JGE | BPF_K:
961 case BPF_JMP | BPF_JGE | BPF_X:
962 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
963 /* BPF immediates are signed, zero extend
964 * immediate into tmp register and use it
967 insn->code = BPF_ALU | BPF_MOV | BPF_K;
968 insn->a_reg = TMP_REG;
973 insn->x_reg = TMP_REG;
979 bpf_src = BPF_SRC(fp->code);
982 /* Common case where 'jump_false' is next insn. */
984 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
985 target = i + fp->jt + 1;
990 /* Convert JEQ into JNE when 'jump_true' is next insn. */
991 if (fp->jt == 0 && BPF_OP(fp->code) == BPF_JEQ) {
992 insn->code = BPF_JMP | BPF_JNE | bpf_src;
993 target = i + fp->jf + 1;
998 /* Other jumps are mapped into two insns: Jxx and JA. */
999 target = i + fp->jt + 1;
1000 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
1004 insn->code = BPF_JMP | BPF_JA;
1005 target = i + fp->jf + 1;
1009 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
1010 case BPF_LDX | BPF_MSH | BPF_B:
1011 insn->code = BPF_ALU64 | BPF_MOV | BPF_X;
1012 insn->a_reg = TMP_REG;
1013 insn->x_reg = A_REG;
1016 insn->code = BPF_LD | BPF_ABS | BPF_B;
1017 insn->a_reg = A_REG;
1021 insn->code = BPF_ALU | BPF_AND | BPF_K;
1022 insn->a_reg = A_REG;
1026 insn->code = BPF_ALU | BPF_LSH | BPF_K;
1027 insn->a_reg = A_REG;
1031 insn->code = BPF_ALU64 | BPF_MOV | BPF_X;
1032 insn->a_reg = X_REG;
1033 insn->x_reg = A_REG;
1036 insn->code = BPF_ALU64 | BPF_MOV | BPF_X;
1037 insn->a_reg = A_REG;
1038 insn->x_reg = TMP_REG;
1041 /* RET_K, RET_A are remaped into 2 insns. */
1042 case BPF_RET | BPF_A:
1043 case BPF_RET | BPF_K:
1044 insn->code = BPF_ALU | BPF_MOV |
1045 (BPF_RVAL(fp->code) == BPF_K ?
1048 insn->x_reg = A_REG;
1052 insn->code = BPF_JMP | BPF_EXIT;
1055 /* Store to stack. */
1058 insn->code = BPF_STX | BPF_MEM | BPF_W;
1059 insn->a_reg = FP_REG;
1060 insn->x_reg = fp->code == BPF_ST ? A_REG : X_REG;
1061 insn->off = -(BPF_MEMWORDS - fp->k) * 4;
1064 /* Load from stack. */
1065 case BPF_LD | BPF_MEM:
1066 case BPF_LDX | BPF_MEM:
1067 insn->code = BPF_LDX | BPF_MEM | BPF_W;
1068 insn->a_reg = BPF_CLASS(fp->code) == BPF_LD ?
1070 insn->x_reg = FP_REG;
1071 insn->off = -(BPF_MEMWORDS - fp->k) * 4;
1074 /* A = K or X = K */
1075 case BPF_LD | BPF_IMM:
1076 case BPF_LDX | BPF_IMM:
1077 insn->code = BPF_ALU | BPF_MOV | BPF_K;
1078 insn->a_reg = BPF_CLASS(fp->code) == BPF_LD ?
1084 case BPF_MISC | BPF_TAX:
1085 insn->code = BPF_ALU64 | BPF_MOV | BPF_X;
1086 insn->a_reg = X_REG;
1087 insn->x_reg = A_REG;
1091 case BPF_MISC | BPF_TXA:
1092 insn->code = BPF_ALU64 | BPF_MOV | BPF_X;
1093 insn->a_reg = A_REG;
1094 insn->x_reg = X_REG;
1097 /* A = skb->len or X = skb->len */
1098 case BPF_LD | BPF_W | BPF_LEN:
1099 case BPF_LDX | BPF_W | BPF_LEN:
1100 insn->code = BPF_LDX | BPF_MEM | BPF_W;
1101 insn->a_reg = BPF_CLASS(fp->code) == BPF_LD ?
1103 insn->x_reg = CTX_REG;
1104 insn->off = offsetof(struct sk_buff, len);
1107 /* access seccomp_data fields */
1108 case BPF_LDX | BPF_ABS | BPF_W:
1109 insn->code = BPF_LDX | BPF_MEM | BPF_W;
1110 insn->a_reg = A_REG;
1111 insn->x_reg = CTX_REG;
1121 memcpy(new_insn, tmp_insns,
1122 sizeof(*insn) * (insn - tmp_insns));
1124 new_insn += insn - tmp_insns;
1128 /* Only calculating new length. */
1129 *new_len = new_insn - new_prog;
1134 if (new_flen != new_insn - new_prog) {
1135 new_flen = new_insn - new_prog;
1143 BUG_ON(*new_len != new_flen);
1152 * A BPF program is able to use 16 cells of memory to store intermediate
1153 * values (check u32 mem[BPF_MEMWORDS] in sk_run_filter()).
1155 * As we dont want to clear mem[] array for each packet going through
1156 * sk_run_filter(), we check that filter loaded by user never try to read
1157 * a cell if not previously written, and we check all branches to be sure
1158 * a malicious user doesn't try to abuse us.
1160 static int check_load_and_stores(struct sock_filter *filter, int flen)
1162 u16 *masks, memvalid = 0; /* one bit per cell, 16 cells */
1165 BUILD_BUG_ON(BPF_MEMWORDS > 16);
1166 masks = kmalloc(flen * sizeof(*masks), GFP_KERNEL);
1169 memset(masks, 0xff, flen * sizeof(*masks));
1171 for (pc = 0; pc < flen; pc++) {
1172 memvalid &= masks[pc];
1174 switch (filter[pc].code) {
1177 memvalid |= (1 << filter[pc].k);
1181 if (!(memvalid & (1 << filter[pc].k))) {
1187 /* a jump must set masks on target */
1188 masks[pc + 1 + filter[pc].k] &= memvalid;
1191 case BPF_S_JMP_JEQ_K:
1192 case BPF_S_JMP_JEQ_X:
1193 case BPF_S_JMP_JGE_K:
1194 case BPF_S_JMP_JGE_X:
1195 case BPF_S_JMP_JGT_K:
1196 case BPF_S_JMP_JGT_X:
1197 case BPF_S_JMP_JSET_X:
1198 case BPF_S_JMP_JSET_K:
1199 /* a jump must set masks on targets */
1200 masks[pc + 1 + filter[pc].jt] &= memvalid;
1201 masks[pc + 1 + filter[pc].jf] &= memvalid;
1212 * sk_chk_filter - verify socket filter code
1213 * @filter: filter to verify
1214 * @flen: length of filter
1216 * Check the user's filter code. If we let some ugly
1217 * filter code slip through kaboom! The filter must contain
1218 * no references or jumps that are out of range, no illegal
1219 * instructions, and must end with a RET instruction.
1221 * All jumps are forward as they are not signed.
1223 * Returns 0 if the rule set is legal or -EINVAL if not.
1225 int sk_chk_filter(struct sock_filter *filter, unsigned int flen)
1228 * Valid instructions are initialized to non-0.
1229 * Invalid instructions are initialized to 0.
1231 static const u8 codes[] = {
1232 [BPF_ALU|BPF_ADD|BPF_K] = BPF_S_ALU_ADD_K,
1233 [BPF_ALU|BPF_ADD|BPF_X] = BPF_S_ALU_ADD_X,
1234 [BPF_ALU|BPF_SUB|BPF_K] = BPF_S_ALU_SUB_K,
1235 [BPF_ALU|BPF_SUB|BPF_X] = BPF_S_ALU_SUB_X,
1236 [BPF_ALU|BPF_MUL|BPF_K] = BPF_S_ALU_MUL_K,
1237 [BPF_ALU|BPF_MUL|BPF_X] = BPF_S_ALU_MUL_X,
1238 [BPF_ALU|BPF_DIV|BPF_X] = BPF_S_ALU_DIV_X,
1239 [BPF_ALU|BPF_MOD|BPF_K] = BPF_S_ALU_MOD_K,
1240 [BPF_ALU|BPF_MOD|BPF_X] = BPF_S_ALU_MOD_X,
1241 [BPF_ALU|BPF_AND|BPF_K] = BPF_S_ALU_AND_K,
1242 [BPF_ALU|BPF_AND|BPF_X] = BPF_S_ALU_AND_X,
1243 [BPF_ALU|BPF_OR|BPF_K] = BPF_S_ALU_OR_K,
1244 [BPF_ALU|BPF_OR|BPF_X] = BPF_S_ALU_OR_X,
1245 [BPF_ALU|BPF_XOR|BPF_K] = BPF_S_ALU_XOR_K,
1246 [BPF_ALU|BPF_XOR|BPF_X] = BPF_S_ALU_XOR_X,
1247 [BPF_ALU|BPF_LSH|BPF_K] = BPF_S_ALU_LSH_K,
1248 [BPF_ALU|BPF_LSH|BPF_X] = BPF_S_ALU_LSH_X,
1249 [BPF_ALU|BPF_RSH|BPF_K] = BPF_S_ALU_RSH_K,
1250 [BPF_ALU|BPF_RSH|BPF_X] = BPF_S_ALU_RSH_X,
1251 [BPF_ALU|BPF_NEG] = BPF_S_ALU_NEG,
1252 [BPF_LD|BPF_W|BPF_ABS] = BPF_S_LD_W_ABS,
1253 [BPF_LD|BPF_H|BPF_ABS] = BPF_S_LD_H_ABS,
1254 [BPF_LD|BPF_B|BPF_ABS] = BPF_S_LD_B_ABS,
1255 [BPF_LD|BPF_W|BPF_LEN] = BPF_S_LD_W_LEN,
1256 [BPF_LD|BPF_W|BPF_IND] = BPF_S_LD_W_IND,
1257 [BPF_LD|BPF_H|BPF_IND] = BPF_S_LD_H_IND,
1258 [BPF_LD|BPF_B|BPF_IND] = BPF_S_LD_B_IND,
1259 [BPF_LD|BPF_IMM] = BPF_S_LD_IMM,
1260 [BPF_LDX|BPF_W|BPF_LEN] = BPF_S_LDX_W_LEN,
1261 [BPF_LDX|BPF_B|BPF_MSH] = BPF_S_LDX_B_MSH,
1262 [BPF_LDX|BPF_IMM] = BPF_S_LDX_IMM,
1263 [BPF_MISC|BPF_TAX] = BPF_S_MISC_TAX,
1264 [BPF_MISC|BPF_TXA] = BPF_S_MISC_TXA,
1265 [BPF_RET|BPF_K] = BPF_S_RET_K,
1266 [BPF_RET|BPF_A] = BPF_S_RET_A,
1267 [BPF_ALU|BPF_DIV|BPF_K] = BPF_S_ALU_DIV_K,
1268 [BPF_LD|BPF_MEM] = BPF_S_LD_MEM,
1269 [BPF_LDX|BPF_MEM] = BPF_S_LDX_MEM,
1270 [BPF_ST] = BPF_S_ST,
1271 [BPF_STX] = BPF_S_STX,
1272 [BPF_JMP|BPF_JA] = BPF_S_JMP_JA,
1273 [BPF_JMP|BPF_JEQ|BPF_K] = BPF_S_JMP_JEQ_K,
1274 [BPF_JMP|BPF_JEQ|BPF_X] = BPF_S_JMP_JEQ_X,
1275 [BPF_JMP|BPF_JGE|BPF_K] = BPF_S_JMP_JGE_K,
1276 [BPF_JMP|BPF_JGE|BPF_X] = BPF_S_JMP_JGE_X,
1277 [BPF_JMP|BPF_JGT|BPF_K] = BPF_S_JMP_JGT_K,
1278 [BPF_JMP|BPF_JGT|BPF_X] = BPF_S_JMP_JGT_X,
1279 [BPF_JMP|BPF_JSET|BPF_K] = BPF_S_JMP_JSET_K,
1280 [BPF_JMP|BPF_JSET|BPF_X] = BPF_S_JMP_JSET_X,
1285 if (flen == 0 || flen > BPF_MAXINSNS)
1288 /* check the filter code now */
1289 for (pc = 0; pc < flen; pc++) {
1290 struct sock_filter *ftest = &filter[pc];
1291 u16 code = ftest->code;
1293 if (code >= ARRAY_SIZE(codes))
1298 /* Some instructions need special checks */
1300 case BPF_S_ALU_DIV_K:
1301 case BPF_S_ALU_MOD_K:
1302 /* check for division by zero */
1310 /* check for invalid memory addresses */
1311 if (ftest->k >= BPF_MEMWORDS)
1316 * Note, the large ftest->k might cause loops.
1317 * Compare this with conditional jumps below,
1318 * where offsets are limited. --ANK (981016)
1320 if (ftest->k >= (unsigned int)(flen-pc-1))
1323 case BPF_S_JMP_JEQ_K:
1324 case BPF_S_JMP_JEQ_X:
1325 case BPF_S_JMP_JGE_K:
1326 case BPF_S_JMP_JGE_X:
1327 case BPF_S_JMP_JGT_K:
1328 case BPF_S_JMP_JGT_X:
1329 case BPF_S_JMP_JSET_X:
1330 case BPF_S_JMP_JSET_K:
1331 /* for conditionals both must be safe */
1332 if (pc + ftest->jt + 1 >= flen ||
1333 pc + ftest->jf + 1 >= flen)
1336 case BPF_S_LD_W_ABS:
1337 case BPF_S_LD_H_ABS:
1338 case BPF_S_LD_B_ABS:
1340 #define ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
1341 code = BPF_S_ANC_##CODE; \
1345 ANCILLARY(PROTOCOL);
1349 ANCILLARY(NLATTR_NEST);
1355 ANCILLARY(ALU_XOR_X);
1356 ANCILLARY(VLAN_TAG);
1357 ANCILLARY(VLAN_TAG_PRESENT);
1358 ANCILLARY(PAY_OFFSET);
1361 /* ancillary operation unknown or unsupported */
1362 if (anc_found == false && ftest->k >= SKF_AD_OFF)
1368 /* last instruction must be a RET code */
1369 switch (filter[flen - 1].code) {
1372 return check_load_and_stores(filter, flen);
1376 EXPORT_SYMBOL(sk_chk_filter);
1378 static int sk_store_orig_filter(struct sk_filter *fp,
1379 const struct sock_fprog *fprog)
1381 unsigned int fsize = sk_filter_proglen(fprog);
1382 struct sock_fprog_kern *fkprog;
1384 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1388 fkprog = fp->orig_prog;
1389 fkprog->len = fprog->len;
1390 fkprog->filter = kmemdup(fp->insns, fsize, GFP_KERNEL);
1391 if (!fkprog->filter) {
1392 kfree(fp->orig_prog);
1399 static void sk_release_orig_filter(struct sk_filter *fp)
1401 struct sock_fprog_kern *fprog = fp->orig_prog;
1404 kfree(fprog->filter);
1410 * sk_filter_release_rcu - Release a socket filter by rcu_head
1411 * @rcu: rcu_head that contains the sk_filter to free
1413 static void sk_filter_release_rcu(struct rcu_head *rcu)
1415 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1417 sk_release_orig_filter(fp);
1422 * sk_filter_release - release a socket filter
1423 * @fp: filter to remove
1425 * Remove a filter from a socket and release its resources.
1427 static void sk_filter_release(struct sk_filter *fp)
1429 if (atomic_dec_and_test(&fp->refcnt))
1430 call_rcu(&fp->rcu, sk_filter_release_rcu);
1433 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1435 atomic_sub(sk_filter_size(fp->len), &sk->sk_omem_alloc);
1436 sk_filter_release(fp);
1439 void sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1441 atomic_inc(&fp->refcnt);
1442 atomic_add(sk_filter_size(fp->len), &sk->sk_omem_alloc);
1445 static struct sk_filter *__sk_migrate_realloc(struct sk_filter *fp,
1449 struct sk_filter *fp_new;
1452 return krealloc(fp, len, GFP_KERNEL);
1454 fp_new = sock_kmalloc(sk, len, GFP_KERNEL);
1456 memcpy(fp_new, fp, sizeof(struct sk_filter));
1457 /* As we're kepping orig_prog in fp_new along,
1458 * we need to make sure we're not evicting it
1461 fp->orig_prog = NULL;
1462 sk_filter_uncharge(sk, fp);
1468 static struct sk_filter *__sk_migrate_filter(struct sk_filter *fp,
1471 struct sock_filter *old_prog;
1472 struct sk_filter *old_fp;
1473 int i, err, new_len, old_len = fp->len;
1475 /* We are free to overwrite insns et al right here as it
1476 * won't be used at this point in time anymore internally
1477 * after the migration to the internal BPF instruction
1480 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1481 sizeof(struct sock_filter_int));
1483 /* For now, we need to unfiddle BPF_S_* identifiers in place.
1484 * This can sooner or later on be subject to removal, e.g. when
1485 * JITs have been converted.
1487 for (i = 0; i < fp->len; i++)
1488 sk_decode_filter(&fp->insns[i], &fp->insns[i]);
1490 /* Conversion cannot happen on overlapping memory areas,
1491 * so we need to keep the user BPF around until the 2nd
1492 * pass. At this time, the user BPF is stored in fp->insns.
1494 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1501 /* 1st pass: calculate the new program length. */
1502 err = sk_convert_filter(old_prog, old_len, NULL, &new_len);
1506 /* Expand fp for appending the new filter representation. */
1508 fp = __sk_migrate_realloc(old_fp, sk, sk_filter_size(new_len));
1510 /* The old_fp is still around in case we couldn't
1511 * allocate new memory, so uncharge on that one.
1518 fp->bpf_func = sk_run_filter_int_skb;
1521 /* 2nd pass: remap sock_filter insns into sock_filter_int insns. */
1522 err = sk_convert_filter(old_prog, old_len, fp->insnsi, &new_len);
1524 /* 2nd sk_convert_filter() can fail only if it fails
1525 * to allocate memory, remapping must succeed. Note,
1526 * that at this time old_fp has already been released
1527 * by __sk_migrate_realloc().
1537 /* Rollback filter setup. */
1539 sk_filter_uncharge(sk, fp);
1542 return ERR_PTR(err);
1545 static struct sk_filter *__sk_prepare_filter(struct sk_filter *fp,
1550 fp->bpf_func = NULL;
1553 err = sk_chk_filter(fp->insns, fp->len);
1555 return ERR_PTR(err);
1557 /* Probe if we can JIT compile the filter and if so, do
1558 * the compilation of the filter.
1560 bpf_jit_compile(fp);
1562 /* JIT compiler couldn't process this filter, so do the
1563 * internal BPF translation for the optimized interpreter.
1566 fp = __sk_migrate_filter(fp, sk);
1572 * sk_unattached_filter_create - create an unattached filter
1573 * @fprog: the filter program
1574 * @pfp: the unattached filter that is created
1576 * Create a filter independent of any socket. We first run some
1577 * sanity checks on it to make sure it does not explode on us later.
1578 * If an error occurs or there is insufficient memory for the filter
1579 * a negative errno code is returned. On success the return is zero.
1581 int sk_unattached_filter_create(struct sk_filter **pfp,
1582 struct sock_fprog *fprog)
1584 unsigned int fsize = sk_filter_proglen(fprog);
1585 struct sk_filter *fp;
1587 /* Make sure new filter is there and in the right amounts. */
1588 if (fprog->filter == NULL)
1591 fp = kmalloc(sk_filter_size(fprog->len), GFP_KERNEL);
1595 memcpy(fp->insns, fprog->filter, fsize);
1597 atomic_set(&fp->refcnt, 1);
1598 fp->len = fprog->len;
1599 /* Since unattached filters are not copied back to user
1600 * space through sk_get_filter(), we do not need to hold
1601 * a copy here, and can spare us the work.
1603 fp->orig_prog = NULL;
1605 /* __sk_prepare_filter() already takes care of uncharging
1606 * memory in case something goes wrong.
1608 fp = __sk_prepare_filter(fp, NULL);
1615 EXPORT_SYMBOL_GPL(sk_unattached_filter_create);
1617 void sk_unattached_filter_destroy(struct sk_filter *fp)
1619 sk_filter_release(fp);
1621 EXPORT_SYMBOL_GPL(sk_unattached_filter_destroy);
1624 * sk_attach_filter - attach a socket filter
1625 * @fprog: the filter program
1626 * @sk: the socket to use
1628 * Attach the user's filter code. We first run some sanity checks on
1629 * it to make sure it does not explode on us later. If an error
1630 * occurs or there is insufficient memory for the filter a negative
1631 * errno code is returned. On success the return is zero.
1633 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1635 struct sk_filter *fp, *old_fp;
1636 unsigned int fsize = sk_filter_proglen(fprog);
1637 unsigned int sk_fsize = sk_filter_size(fprog->len);
1640 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1643 /* Make sure new filter is there and in the right amounts. */
1644 if (fprog->filter == NULL)
1647 fp = sock_kmalloc(sk, sk_fsize, GFP_KERNEL);
1651 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1652 sock_kfree_s(sk, fp, sk_fsize);
1656 atomic_set(&fp->refcnt, 1);
1657 fp->len = fprog->len;
1659 err = sk_store_orig_filter(fp, fprog);
1661 sk_filter_uncharge(sk, fp);
1665 /* __sk_prepare_filter() already takes care of uncharging
1666 * memory in case something goes wrong.
1668 fp = __sk_prepare_filter(fp, sk);
1672 old_fp = rcu_dereference_protected(sk->sk_filter,
1673 sock_owned_by_user(sk));
1674 rcu_assign_pointer(sk->sk_filter, fp);
1677 sk_filter_uncharge(sk, old_fp);
1681 EXPORT_SYMBOL_GPL(sk_attach_filter);
1683 int sk_detach_filter(struct sock *sk)
1686 struct sk_filter *filter;
1688 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1691 filter = rcu_dereference_protected(sk->sk_filter,
1692 sock_owned_by_user(sk));
1694 RCU_INIT_POINTER(sk->sk_filter, NULL);
1695 sk_filter_uncharge(sk, filter);
1701 EXPORT_SYMBOL_GPL(sk_detach_filter);
1703 void sk_decode_filter(struct sock_filter *filt, struct sock_filter *to)
1705 static const u16 decodes[] = {
1706 [BPF_S_ALU_ADD_K] = BPF_ALU|BPF_ADD|BPF_K,
1707 [BPF_S_ALU_ADD_X] = BPF_ALU|BPF_ADD|BPF_X,
1708 [BPF_S_ALU_SUB_K] = BPF_ALU|BPF_SUB|BPF_K,
1709 [BPF_S_ALU_SUB_X] = BPF_ALU|BPF_SUB|BPF_X,
1710 [BPF_S_ALU_MUL_K] = BPF_ALU|BPF_MUL|BPF_K,
1711 [BPF_S_ALU_MUL_X] = BPF_ALU|BPF_MUL|BPF_X,
1712 [BPF_S_ALU_DIV_X] = BPF_ALU|BPF_DIV|BPF_X,
1713 [BPF_S_ALU_MOD_K] = BPF_ALU|BPF_MOD|BPF_K,
1714 [BPF_S_ALU_MOD_X] = BPF_ALU|BPF_MOD|BPF_X,
1715 [BPF_S_ALU_AND_K] = BPF_ALU|BPF_AND|BPF_K,
1716 [BPF_S_ALU_AND_X] = BPF_ALU|BPF_AND|BPF_X,
1717 [BPF_S_ALU_OR_K] = BPF_ALU|BPF_OR|BPF_K,
1718 [BPF_S_ALU_OR_X] = BPF_ALU|BPF_OR|BPF_X,
1719 [BPF_S_ALU_XOR_K] = BPF_ALU|BPF_XOR|BPF_K,
1720 [BPF_S_ALU_XOR_X] = BPF_ALU|BPF_XOR|BPF_X,
1721 [BPF_S_ALU_LSH_K] = BPF_ALU|BPF_LSH|BPF_K,
1722 [BPF_S_ALU_LSH_X] = BPF_ALU|BPF_LSH|BPF_X,
1723 [BPF_S_ALU_RSH_K] = BPF_ALU|BPF_RSH|BPF_K,
1724 [BPF_S_ALU_RSH_X] = BPF_ALU|BPF_RSH|BPF_X,
1725 [BPF_S_ALU_NEG] = BPF_ALU|BPF_NEG,
1726 [BPF_S_LD_W_ABS] = BPF_LD|BPF_W|BPF_ABS,
1727 [BPF_S_LD_H_ABS] = BPF_LD|BPF_H|BPF_ABS,
1728 [BPF_S_LD_B_ABS] = BPF_LD|BPF_B|BPF_ABS,
1729 [BPF_S_ANC_PROTOCOL] = BPF_LD|BPF_B|BPF_ABS,
1730 [BPF_S_ANC_PKTTYPE] = BPF_LD|BPF_B|BPF_ABS,
1731 [BPF_S_ANC_IFINDEX] = BPF_LD|BPF_B|BPF_ABS,
1732 [BPF_S_ANC_NLATTR] = BPF_LD|BPF_B|BPF_ABS,
1733 [BPF_S_ANC_NLATTR_NEST] = BPF_LD|BPF_B|BPF_ABS,
1734 [BPF_S_ANC_MARK] = BPF_LD|BPF_B|BPF_ABS,
1735 [BPF_S_ANC_QUEUE] = BPF_LD|BPF_B|BPF_ABS,
1736 [BPF_S_ANC_HATYPE] = BPF_LD|BPF_B|BPF_ABS,
1737 [BPF_S_ANC_RXHASH] = BPF_LD|BPF_B|BPF_ABS,
1738 [BPF_S_ANC_CPU] = BPF_LD|BPF_B|BPF_ABS,
1739 [BPF_S_ANC_ALU_XOR_X] = BPF_LD|BPF_B|BPF_ABS,
1740 [BPF_S_ANC_SECCOMP_LD_W] = BPF_LD|BPF_B|BPF_ABS,
1741 [BPF_S_ANC_VLAN_TAG] = BPF_LD|BPF_B|BPF_ABS,
1742 [BPF_S_ANC_VLAN_TAG_PRESENT] = BPF_LD|BPF_B|BPF_ABS,
1743 [BPF_S_ANC_PAY_OFFSET] = BPF_LD|BPF_B|BPF_ABS,
1744 [BPF_S_LD_W_LEN] = BPF_LD|BPF_W|BPF_LEN,
1745 [BPF_S_LD_W_IND] = BPF_LD|BPF_W|BPF_IND,
1746 [BPF_S_LD_H_IND] = BPF_LD|BPF_H|BPF_IND,
1747 [BPF_S_LD_B_IND] = BPF_LD|BPF_B|BPF_IND,
1748 [BPF_S_LD_IMM] = BPF_LD|BPF_IMM,
1749 [BPF_S_LDX_W_LEN] = BPF_LDX|BPF_W|BPF_LEN,
1750 [BPF_S_LDX_B_MSH] = BPF_LDX|BPF_B|BPF_MSH,
1751 [BPF_S_LDX_IMM] = BPF_LDX|BPF_IMM,
1752 [BPF_S_MISC_TAX] = BPF_MISC|BPF_TAX,
1753 [BPF_S_MISC_TXA] = BPF_MISC|BPF_TXA,
1754 [BPF_S_RET_K] = BPF_RET|BPF_K,
1755 [BPF_S_RET_A] = BPF_RET|BPF_A,
1756 [BPF_S_ALU_DIV_K] = BPF_ALU|BPF_DIV|BPF_K,
1757 [BPF_S_LD_MEM] = BPF_LD|BPF_MEM,
1758 [BPF_S_LDX_MEM] = BPF_LDX|BPF_MEM,
1759 [BPF_S_ST] = BPF_ST,
1760 [BPF_S_STX] = BPF_STX,
1761 [BPF_S_JMP_JA] = BPF_JMP|BPF_JA,
1762 [BPF_S_JMP_JEQ_K] = BPF_JMP|BPF_JEQ|BPF_K,
1763 [BPF_S_JMP_JEQ_X] = BPF_JMP|BPF_JEQ|BPF_X,
1764 [BPF_S_JMP_JGE_K] = BPF_JMP|BPF_JGE|BPF_K,
1765 [BPF_S_JMP_JGE_X] = BPF_JMP|BPF_JGE|BPF_X,
1766 [BPF_S_JMP_JGT_K] = BPF_JMP|BPF_JGT|BPF_K,
1767 [BPF_S_JMP_JGT_X] = BPF_JMP|BPF_JGT|BPF_X,
1768 [BPF_S_JMP_JSET_K] = BPF_JMP|BPF_JSET|BPF_K,
1769 [BPF_S_JMP_JSET_X] = BPF_JMP|BPF_JSET|BPF_X,
1775 to->code = decodes[code];
1781 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
1784 struct sock_fprog_kern *fprog;
1785 struct sk_filter *filter;
1789 filter = rcu_dereference_protected(sk->sk_filter,
1790 sock_owned_by_user(sk));
1794 /* We're copying the filter that has been originally attached,
1795 * so no conversion/decode needed anymore.
1797 fprog = filter->orig_prog;
1801 /* User space only enquires number of filter blocks. */
1805 if (len < fprog->len)
1809 if (copy_to_user(ubuf, fprog->filter, sk_filter_proglen(fprog)))
1812 /* Instead of bytes, the API requests to return the number