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)
125 /* Register mappings for user programs. */
133 * __sk_run_filter - run a filter on a given context
134 * @ctx: buffer to run the filter on
135 * @insn: filter to apply
137 * Decode and apply filter instructions to the skb->data. Return length to
138 * keep, 0 for none. @ctx is the data we are operating on, @insn is the
139 * array of filter instructions.
141 unsigned int __sk_run_filter(void *ctx, const struct sock_filter_int *insn)
143 u64 stack[MAX_BPF_STACK / sizeof(u64)];
144 u64 regs[MAX_BPF_REG], tmp;
149 #define A regs[insn->a_reg]
150 #define X regs[insn->x_reg]
153 #define CONT ({insn++; goto select_insn; })
154 #define CONT_JMP ({insn++; goto select_insn; })
156 static const void *jumptable[256] = {
157 [0 ... 255] = &&default_label,
158 /* Now overwrite non-defaults ... */
159 #define DL(A, B, C) [A|B|C] = &&A##_##B##_##C
160 DL(BPF_ALU, BPF_ADD, BPF_X),
161 DL(BPF_ALU, BPF_ADD, BPF_K),
162 DL(BPF_ALU, BPF_SUB, BPF_X),
163 DL(BPF_ALU, BPF_SUB, BPF_K),
164 DL(BPF_ALU, BPF_AND, BPF_X),
165 DL(BPF_ALU, BPF_AND, BPF_K),
166 DL(BPF_ALU, BPF_OR, BPF_X),
167 DL(BPF_ALU, BPF_OR, BPF_K),
168 DL(BPF_ALU, BPF_LSH, BPF_X),
169 DL(BPF_ALU, BPF_LSH, BPF_K),
170 DL(BPF_ALU, BPF_RSH, BPF_X),
171 DL(BPF_ALU, BPF_RSH, BPF_K),
172 DL(BPF_ALU, BPF_XOR, BPF_X),
173 DL(BPF_ALU, BPF_XOR, BPF_K),
174 DL(BPF_ALU, BPF_MUL, BPF_X),
175 DL(BPF_ALU, BPF_MUL, BPF_K),
176 DL(BPF_ALU, BPF_MOV, BPF_X),
177 DL(BPF_ALU, BPF_MOV, BPF_K),
178 DL(BPF_ALU, BPF_DIV, BPF_X),
179 DL(BPF_ALU, BPF_DIV, BPF_K),
180 DL(BPF_ALU, BPF_MOD, BPF_X),
181 DL(BPF_ALU, BPF_MOD, BPF_K),
182 DL(BPF_ALU, BPF_NEG, 0),
183 DL(BPF_ALU, BPF_END, BPF_TO_BE),
184 DL(BPF_ALU, BPF_END, BPF_TO_LE),
185 DL(BPF_ALU64, BPF_ADD, BPF_X),
186 DL(BPF_ALU64, BPF_ADD, BPF_K),
187 DL(BPF_ALU64, BPF_SUB, BPF_X),
188 DL(BPF_ALU64, BPF_SUB, BPF_K),
189 DL(BPF_ALU64, BPF_AND, BPF_X),
190 DL(BPF_ALU64, BPF_AND, BPF_K),
191 DL(BPF_ALU64, BPF_OR, BPF_X),
192 DL(BPF_ALU64, BPF_OR, BPF_K),
193 DL(BPF_ALU64, BPF_LSH, BPF_X),
194 DL(BPF_ALU64, BPF_LSH, BPF_K),
195 DL(BPF_ALU64, BPF_RSH, BPF_X),
196 DL(BPF_ALU64, BPF_RSH, BPF_K),
197 DL(BPF_ALU64, BPF_XOR, BPF_X),
198 DL(BPF_ALU64, BPF_XOR, BPF_K),
199 DL(BPF_ALU64, BPF_MUL, BPF_X),
200 DL(BPF_ALU64, BPF_MUL, BPF_K),
201 DL(BPF_ALU64, BPF_MOV, BPF_X),
202 DL(BPF_ALU64, BPF_MOV, BPF_K),
203 DL(BPF_ALU64, BPF_ARSH, BPF_X),
204 DL(BPF_ALU64, BPF_ARSH, BPF_K),
205 DL(BPF_ALU64, BPF_DIV, BPF_X),
206 DL(BPF_ALU64, BPF_DIV, BPF_K),
207 DL(BPF_ALU64, BPF_MOD, BPF_X),
208 DL(BPF_ALU64, BPF_MOD, BPF_K),
209 DL(BPF_ALU64, BPF_NEG, 0),
210 DL(BPF_JMP, BPF_CALL, 0),
211 DL(BPF_JMP, BPF_JA, 0),
212 DL(BPF_JMP, BPF_JEQ, BPF_X),
213 DL(BPF_JMP, BPF_JEQ, BPF_K),
214 DL(BPF_JMP, BPF_JNE, BPF_X),
215 DL(BPF_JMP, BPF_JNE, BPF_K),
216 DL(BPF_JMP, BPF_JGT, BPF_X),
217 DL(BPF_JMP, BPF_JGT, BPF_K),
218 DL(BPF_JMP, BPF_JGE, BPF_X),
219 DL(BPF_JMP, BPF_JGE, BPF_K),
220 DL(BPF_JMP, BPF_JSGT, BPF_X),
221 DL(BPF_JMP, BPF_JSGT, BPF_K),
222 DL(BPF_JMP, BPF_JSGE, BPF_X),
223 DL(BPF_JMP, BPF_JSGE, BPF_K),
224 DL(BPF_JMP, BPF_JSET, BPF_X),
225 DL(BPF_JMP, BPF_JSET, BPF_K),
226 DL(BPF_JMP, BPF_EXIT, 0),
227 DL(BPF_STX, BPF_MEM, BPF_B),
228 DL(BPF_STX, BPF_MEM, BPF_H),
229 DL(BPF_STX, BPF_MEM, BPF_W),
230 DL(BPF_STX, BPF_MEM, BPF_DW),
231 DL(BPF_STX, BPF_XADD, BPF_W),
232 DL(BPF_STX, BPF_XADD, BPF_DW),
233 DL(BPF_ST, BPF_MEM, BPF_B),
234 DL(BPF_ST, BPF_MEM, BPF_H),
235 DL(BPF_ST, BPF_MEM, BPF_W),
236 DL(BPF_ST, BPF_MEM, BPF_DW),
237 DL(BPF_LDX, BPF_MEM, BPF_B),
238 DL(BPF_LDX, BPF_MEM, BPF_H),
239 DL(BPF_LDX, BPF_MEM, BPF_W),
240 DL(BPF_LDX, BPF_MEM, BPF_DW),
241 DL(BPF_LD, BPF_ABS, BPF_W),
242 DL(BPF_LD, BPF_ABS, BPF_H),
243 DL(BPF_LD, BPF_ABS, BPF_B),
244 DL(BPF_LD, BPF_IND, BPF_W),
245 DL(BPF_LD, BPF_IND, BPF_H),
246 DL(BPF_LD, BPF_IND, BPF_B),
250 regs[FP_REG] = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)];
251 regs[ARG1_REG] = (u64) (unsigned long) ctx;
256 goto *jumptable[insn->code];
259 #define ALU(OPCODE, OP) \
260 BPF_ALU64_##OPCODE##_BPF_X: \
263 BPF_ALU_##OPCODE##_BPF_X: \
264 A = (u32) A OP (u32) X; \
266 BPF_ALU64_##OPCODE##_BPF_K: \
269 BPF_ALU_##OPCODE##_BPF_K: \
270 A = (u32) A OP (u32) K; \
288 BPF_ALU_BPF_MOV_BPF_X:
291 BPF_ALU_BPF_MOV_BPF_K:
294 BPF_ALU64_BPF_MOV_BPF_X:
297 BPF_ALU64_BPF_MOV_BPF_K:
300 BPF_ALU64_BPF_ARSH_BPF_X:
303 BPF_ALU64_BPF_ARSH_BPF_K:
306 BPF_ALU64_BPF_MOD_BPF_X:
307 if (unlikely(X == 0))
312 BPF_ALU_BPF_MOD_BPF_X:
313 if (unlikely(X == 0))
316 A = do_div(tmp, (u32) X);
318 BPF_ALU64_BPF_MOD_BPF_K:
322 BPF_ALU_BPF_MOD_BPF_K:
324 A = do_div(tmp, (u32) K);
326 BPF_ALU64_BPF_DIV_BPF_X:
327 if (unlikely(X == 0))
331 BPF_ALU_BPF_DIV_BPF_X:
332 if (unlikely(X == 0))
335 do_div(tmp, (u32) X);
338 BPF_ALU64_BPF_DIV_BPF_K:
341 BPF_ALU_BPF_DIV_BPF_K:
343 do_div(tmp, (u32) K);
346 BPF_ALU_BPF_END_BPF_TO_BE:
349 A = (__force u16) cpu_to_be16(A);
352 A = (__force u32) cpu_to_be32(A);
355 A = (__force u64) cpu_to_be64(A);
359 BPF_ALU_BPF_END_BPF_TO_LE:
362 A = (__force u16) cpu_to_le16(A);
365 A = (__force u32) cpu_to_le32(A);
368 A = (__force u64) cpu_to_le64(A);
375 /* Function call scratches R1-R5 registers, preserves R6-R9,
376 * and stores return value into R0.
378 R0 = (__bpf_call_base + insn->imm)(regs[1], regs[2], regs[3],
386 BPF_JMP_BPF_JEQ_BPF_X:
392 BPF_JMP_BPF_JEQ_BPF_K:
398 BPF_JMP_BPF_JNE_BPF_X:
404 BPF_JMP_BPF_JNE_BPF_K:
410 BPF_JMP_BPF_JGT_BPF_X:
416 BPF_JMP_BPF_JGT_BPF_K:
422 BPF_JMP_BPF_JGE_BPF_X:
428 BPF_JMP_BPF_JGE_BPF_K:
434 BPF_JMP_BPF_JSGT_BPF_X:
435 if (((s64)A) > ((s64)X)) {
440 BPF_JMP_BPF_JSGT_BPF_K:
441 if (((s64)A) > ((s64)K)) {
446 BPF_JMP_BPF_JSGE_BPF_X:
447 if (((s64)A) >= ((s64)X)) {
452 BPF_JMP_BPF_JSGE_BPF_K:
453 if (((s64)A) >= ((s64)K)) {
458 BPF_JMP_BPF_JSET_BPF_X:
464 BPF_JMP_BPF_JSET_BPF_K:
473 /* STX and ST and LDX*/
474 #define LDST(SIZEOP, SIZE) \
475 BPF_STX_BPF_MEM_##SIZEOP: \
476 *(SIZE *)(unsigned long) (A + insn->off) = X; \
478 BPF_ST_BPF_MEM_##SIZEOP: \
479 *(SIZE *)(unsigned long) (A + insn->off) = K; \
481 BPF_LDX_BPF_MEM_##SIZEOP: \
482 A = *(SIZE *)(unsigned long) (X + insn->off); \
490 BPF_STX_BPF_XADD_BPF_W: /* lock xadd *(u32 *)(A + insn->off) += X */
491 atomic_add((u32) X, (atomic_t *)(unsigned long)
494 BPF_STX_BPF_XADD_BPF_DW: /* lock xadd *(u64 *)(A + insn->off) += X */
495 atomic64_add((u64) X, (atomic64_t *)(unsigned long)
498 BPF_LD_BPF_ABS_BPF_W: /* R0 = ntohl(*(u32 *) (skb->data + K)) */
501 /* BPF_LD + BPD_ABS and BPF_LD + BPF_IND insns are only
502 * appearing in the programs where ctx == skb. All programs
503 * keep 'ctx' in regs[CTX_REG] == R6, sk_convert_filter()
504 * saves it in R6, internal BPF verifier will check that
507 * BPF_ABS and BPF_IND are wrappers of function calls, so
508 * they scratch R1-R5 registers, preserve R6-R9, and store
509 * return value into R0.
516 * K == 32-bit immediate
519 * R0 - 8/16/32-bit skb data converted to cpu endianness
521 ptr = load_pointer((struct sk_buff *) ctx, off, 4, &tmp);
522 if (likely(ptr != NULL)) {
523 R0 = get_unaligned_be32(ptr);
527 BPF_LD_BPF_ABS_BPF_H: /* R0 = ntohs(*(u16 *) (skb->data + K)) */
530 ptr = load_pointer((struct sk_buff *) ctx, off, 2, &tmp);
531 if (likely(ptr != NULL)) {
532 R0 = get_unaligned_be16(ptr);
536 BPF_LD_BPF_ABS_BPF_B: /* R0 = *(u8 *) (ctx + K) */
539 ptr = load_pointer((struct sk_buff *) ctx, off, 1, &tmp);
540 if (likely(ptr != NULL)) {
545 BPF_LD_BPF_IND_BPF_W: /* R0 = ntohl(*(u32 *) (skb->data + X + K)) */
548 BPF_LD_BPF_IND_BPF_H: /* R0 = ntohs(*(u16 *) (skb->data + X + K)) */
551 BPF_LD_BPF_IND_BPF_B: /* R0 = *(u8 *) (skb->data + X + K) */
556 /* If we ever reach this, we have a bug somewhere. */
557 WARN_RATELIMIT(1, "unknown opcode %02x\n", insn->code);
568 u32 sk_run_filter_int_seccomp(const struct seccomp_data *ctx,
569 const struct sock_filter_int *insni)
570 __attribute__ ((alias ("__sk_run_filter")));
572 u32 sk_run_filter_int_skb(const struct sk_buff *ctx,
573 const struct sock_filter_int *insni)
574 __attribute__ ((alias ("__sk_run_filter")));
575 EXPORT_SYMBOL_GPL(sk_run_filter_int_skb);
577 /* Helper to find the offset of pkt_type in sk_buff structure. We want
578 * to make sure its still a 3bit field starting at a byte boundary;
579 * taken from arch/x86/net/bpf_jit_comp.c.
581 #define PKT_TYPE_MAX 7
582 static unsigned int pkt_type_offset(void)
584 struct sk_buff skb_probe = { .pkt_type = ~0, };
585 u8 *ct = (u8 *) &skb_probe;
588 for (off = 0; off < sizeof(struct sk_buff); off++) {
589 if (ct[off] == PKT_TYPE_MAX)
593 pr_err_once("Please fix %s, as pkt_type couldn't be found!\n", __func__);
597 static u64 __skb_get_pay_offset(u64 ctx, u64 A, u64 X, u64 r4, u64 r5)
599 struct sk_buff *skb = (struct sk_buff *)(long) ctx;
601 return __skb_get_poff(skb);
604 static u64 __skb_get_nlattr(u64 ctx, u64 A, u64 X, u64 r4, u64 r5)
606 struct sk_buff *skb = (struct sk_buff *)(long) ctx;
609 if (skb_is_nonlinear(skb))
612 if (skb->len < sizeof(struct nlattr))
615 if (A > skb->len - sizeof(struct nlattr))
618 nla = nla_find((struct nlattr *) &skb->data[A], skb->len - A, X);
620 return (void *) nla - (void *) skb->data;
625 static u64 __skb_get_nlattr_nest(u64 ctx, u64 A, u64 X, u64 r4, u64 r5)
627 struct sk_buff *skb = (struct sk_buff *)(long) ctx;
630 if (skb_is_nonlinear(skb))
633 if (skb->len < sizeof(struct nlattr))
636 if (A > skb->len - sizeof(struct nlattr))
639 nla = (struct nlattr *) &skb->data[A];
640 if (nla->nla_len > skb->len - A)
643 nla = nla_find_nested(nla, X);
645 return (void *) nla - (void *) skb->data;
650 static u64 __get_raw_cpu_id(u64 ctx, u64 A, u64 X, u64 r4, u64 r5)
652 return raw_smp_processor_id();
655 static bool convert_bpf_extensions(struct sock_filter *fp,
656 struct sock_filter_int **insnp)
658 struct sock_filter_int *insn = *insnp;
661 case SKF_AD_OFF + SKF_AD_PROTOCOL:
662 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
664 insn->code = BPF_LDX | BPF_MEM | BPF_H;
666 insn->x_reg = CTX_REG;
667 insn->off = offsetof(struct sk_buff, protocol);
670 /* A = ntohs(A) [emitting a nop or swap16] */
671 insn->code = BPF_ALU | BPF_END | BPF_FROM_BE;
676 case SKF_AD_OFF + SKF_AD_PKTTYPE:
677 insn->code = BPF_LDX | BPF_MEM | BPF_B;
679 insn->x_reg = CTX_REG;
680 insn->off = pkt_type_offset();
685 insn->code = BPF_ALU | BPF_AND | BPF_K;
687 insn->imm = PKT_TYPE_MAX;
690 case SKF_AD_OFF + SKF_AD_IFINDEX:
691 case SKF_AD_OFF + SKF_AD_HATYPE:
692 if (FIELD_SIZEOF(struct sk_buff, dev) == 8)
693 insn->code = BPF_LDX | BPF_MEM | BPF_DW;
695 insn->code = BPF_LDX | BPF_MEM | BPF_W;
696 insn->a_reg = TMP_REG;
697 insn->x_reg = CTX_REG;
698 insn->off = offsetof(struct sk_buff, dev);
701 insn->code = BPF_JMP | BPF_JNE | BPF_K;
702 insn->a_reg = TMP_REG;
707 insn->code = BPF_JMP | BPF_EXIT;
710 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
711 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, type) != 2);
714 insn->x_reg = TMP_REG;
716 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX) {
717 insn->code = BPF_LDX | BPF_MEM | BPF_W;
718 insn->off = offsetof(struct net_device, ifindex);
720 insn->code = BPF_LDX | BPF_MEM | BPF_H;
721 insn->off = offsetof(struct net_device, type);
725 case SKF_AD_OFF + SKF_AD_MARK:
726 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
728 insn->code = BPF_LDX | BPF_MEM | BPF_W;
730 insn->x_reg = CTX_REG;
731 insn->off = offsetof(struct sk_buff, mark);
734 case SKF_AD_OFF + SKF_AD_RXHASH:
735 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
737 insn->code = BPF_LDX | BPF_MEM | BPF_W;
739 insn->x_reg = CTX_REG;
740 insn->off = offsetof(struct sk_buff, hash);
743 case SKF_AD_OFF + SKF_AD_QUEUE:
744 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2);
746 insn->code = BPF_LDX | BPF_MEM | BPF_H;
748 insn->x_reg = CTX_REG;
749 insn->off = offsetof(struct sk_buff, queue_mapping);
752 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
753 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
754 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2);
756 insn->code = BPF_LDX | BPF_MEM | BPF_H;
758 insn->x_reg = CTX_REG;
759 insn->off = offsetof(struct sk_buff, vlan_tci);
762 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000);
764 if (fp->k == SKF_AD_OFF + SKF_AD_VLAN_TAG) {
765 insn->code = BPF_ALU | BPF_AND | BPF_K;
767 insn->imm = ~VLAN_TAG_PRESENT;
769 insn->code = BPF_ALU | BPF_RSH | BPF_K;
774 insn->code = BPF_ALU | BPF_AND | BPF_K;
780 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
781 case SKF_AD_OFF + SKF_AD_NLATTR:
782 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
783 case SKF_AD_OFF + SKF_AD_CPU:
785 insn->code = BPF_ALU64 | BPF_MOV | BPF_X;
786 insn->a_reg = ARG1_REG;
787 insn->x_reg = CTX_REG;
791 insn->code = BPF_ALU64 | BPF_MOV | BPF_X;
792 insn->a_reg = ARG2_REG;
797 insn->code = BPF_ALU64 | BPF_MOV | BPF_X;
798 insn->a_reg = ARG3_REG;
802 /* Emit call(ctx, arg2=A, arg3=X) */
803 insn->code = BPF_JMP | BPF_CALL;
805 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
806 insn->imm = __skb_get_pay_offset - __bpf_call_base;
808 case SKF_AD_OFF + SKF_AD_NLATTR:
809 insn->imm = __skb_get_nlattr - __bpf_call_base;
811 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
812 insn->imm = __skb_get_nlattr_nest - __bpf_call_base;
814 case SKF_AD_OFF + SKF_AD_CPU:
815 insn->imm = __get_raw_cpu_id - __bpf_call_base;
820 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
821 insn->code = BPF_ALU | BPF_XOR | BPF_X;
827 /* This is just a dummy call to avoid letting the compiler
828 * evict __bpf_call_base() as an optimization. Placed here
829 * where no-one bothers.
831 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
840 * sk_convert_filter - convert filter program
841 * @prog: the user passed filter program
842 * @len: the length of the user passed filter program
843 * @new_prog: buffer where converted program will be stored
844 * @new_len: pointer to store length of converted program
846 * Remap 'sock_filter' style BPF instruction set to 'sock_filter_ext' style.
847 * Conversion workflow:
849 * 1) First pass for calculating the new program length:
850 * sk_convert_filter(old_prog, old_len, NULL, &new_len)
852 * 2) 2nd pass to remap in two passes: 1st pass finds new
853 * jump offsets, 2nd pass remapping:
854 * new_prog = kmalloc(sizeof(struct sock_filter_int) * new_len);
855 * sk_convert_filter(old_prog, old_len, new_prog, &new_len);
857 * User BPF's register A is mapped to our BPF register 6, user BPF
858 * register X is mapped to BPF register 7; frame pointer is always
859 * register 10; Context 'void *ctx' is stored in register 1, that is,
860 * for socket filters: ctx == 'struct sk_buff *', for seccomp:
861 * ctx == 'struct seccomp_data *'.
863 int sk_convert_filter(struct sock_filter *prog, int len,
864 struct sock_filter_int *new_prog, int *new_len)
866 int new_flen = 0, pass = 0, target, i;
867 struct sock_filter_int *new_insn;
868 struct sock_filter *fp;
872 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
873 BUILD_BUG_ON(FP_REG + 1 != MAX_BPF_REG);
875 if (len <= 0 || len >= BPF_MAXINSNS)
879 addrs = kzalloc(len * sizeof(*addrs), GFP_KERNEL);
889 new_insn->code = BPF_ALU64 | BPF_MOV | BPF_X;
890 new_insn->a_reg = CTX_REG;
891 new_insn->x_reg = ARG1_REG;
895 for (i = 0; i < len; fp++, i++) {
896 struct sock_filter_int tmp_insns[6] = { };
897 struct sock_filter_int *insn = tmp_insns;
900 addrs[i] = new_insn - new_prog;
903 /* All arithmetic insns and skb loads map as-is. */
904 case BPF_ALU | BPF_ADD | BPF_X:
905 case BPF_ALU | BPF_ADD | BPF_K:
906 case BPF_ALU | BPF_SUB | BPF_X:
907 case BPF_ALU | BPF_SUB | BPF_K:
908 case BPF_ALU | BPF_AND | BPF_X:
909 case BPF_ALU | BPF_AND | BPF_K:
910 case BPF_ALU | BPF_OR | BPF_X:
911 case BPF_ALU | BPF_OR | BPF_K:
912 case BPF_ALU | BPF_LSH | BPF_X:
913 case BPF_ALU | BPF_LSH | BPF_K:
914 case BPF_ALU | BPF_RSH | BPF_X:
915 case BPF_ALU | BPF_RSH | BPF_K:
916 case BPF_ALU | BPF_XOR | BPF_X:
917 case BPF_ALU | BPF_XOR | BPF_K:
918 case BPF_ALU | BPF_MUL | BPF_X:
919 case BPF_ALU | BPF_MUL | BPF_K:
920 case BPF_ALU | BPF_DIV | BPF_X:
921 case BPF_ALU | BPF_DIV | BPF_K:
922 case BPF_ALU | BPF_MOD | BPF_X:
923 case BPF_ALU | BPF_MOD | BPF_K:
924 case BPF_ALU | BPF_NEG:
925 case BPF_LD | BPF_ABS | BPF_W:
926 case BPF_LD | BPF_ABS | BPF_H:
927 case BPF_LD | BPF_ABS | BPF_B:
928 case BPF_LD | BPF_IND | BPF_W:
929 case BPF_LD | BPF_IND | BPF_H:
930 case BPF_LD | BPF_IND | BPF_B:
931 /* Check for overloaded BPF extension and
932 * directly convert it if found, otherwise
933 * just move on with mapping.
935 if (BPF_CLASS(fp->code) == BPF_LD &&
936 BPF_MODE(fp->code) == BPF_ABS &&
937 convert_bpf_extensions(fp, &insn))
940 insn->code = fp->code;
946 /* Jump opcodes map as-is, but offsets need adjustment. */
947 case BPF_JMP | BPF_JA:
948 target = i + fp->k + 1;
949 insn->code = fp->code;
952 if (target >= len || target < 0) \
954 insn->off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
955 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
956 insn->off -= insn - tmp_insns; \
962 case BPF_JMP | BPF_JEQ | BPF_K:
963 case BPF_JMP | BPF_JEQ | BPF_X:
964 case BPF_JMP | BPF_JSET | BPF_K:
965 case BPF_JMP | BPF_JSET | BPF_X:
966 case BPF_JMP | BPF_JGT | BPF_K:
967 case BPF_JMP | BPF_JGT | BPF_X:
968 case BPF_JMP | BPF_JGE | BPF_K:
969 case BPF_JMP | BPF_JGE | BPF_X:
970 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
971 /* BPF immediates are signed, zero extend
972 * immediate into tmp register and use it
975 insn->code = BPF_ALU | BPF_MOV | BPF_K;
976 insn->a_reg = TMP_REG;
981 insn->x_reg = TMP_REG;
987 bpf_src = BPF_SRC(fp->code);
990 /* Common case where 'jump_false' is next insn. */
992 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
993 target = i + fp->jt + 1;
998 /* Convert JEQ into JNE when 'jump_true' is next insn. */
999 if (fp->jt == 0 && BPF_OP(fp->code) == BPF_JEQ) {
1000 insn->code = BPF_JMP | BPF_JNE | bpf_src;
1001 target = i + fp->jf + 1;
1006 /* Other jumps are mapped into two insns: Jxx and JA. */
1007 target = i + fp->jt + 1;
1008 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
1012 insn->code = BPF_JMP | BPF_JA;
1013 target = i + fp->jf + 1;
1017 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
1018 case BPF_LDX | BPF_MSH | BPF_B:
1019 insn->code = BPF_ALU64 | BPF_MOV | BPF_X;
1020 insn->a_reg = TMP_REG;
1021 insn->x_reg = A_REG;
1024 insn->code = BPF_LD | BPF_ABS | BPF_B;
1025 insn->a_reg = A_REG;
1029 insn->code = BPF_ALU | BPF_AND | BPF_K;
1030 insn->a_reg = A_REG;
1034 insn->code = BPF_ALU | BPF_LSH | BPF_K;
1035 insn->a_reg = A_REG;
1039 insn->code = BPF_ALU64 | BPF_MOV | BPF_X;
1040 insn->a_reg = X_REG;
1041 insn->x_reg = A_REG;
1044 insn->code = BPF_ALU64 | BPF_MOV | BPF_X;
1045 insn->a_reg = A_REG;
1046 insn->x_reg = TMP_REG;
1049 /* RET_K, RET_A are remaped into 2 insns. */
1050 case BPF_RET | BPF_A:
1051 case BPF_RET | BPF_K:
1052 insn->code = BPF_ALU | BPF_MOV |
1053 (BPF_RVAL(fp->code) == BPF_K ?
1056 insn->x_reg = A_REG;
1060 insn->code = BPF_JMP | BPF_EXIT;
1063 /* Store to stack. */
1066 insn->code = BPF_STX | BPF_MEM | BPF_W;
1067 insn->a_reg = FP_REG;
1068 insn->x_reg = fp->code == BPF_ST ? A_REG : X_REG;
1069 insn->off = -(BPF_MEMWORDS - fp->k) * 4;
1072 /* Load from stack. */
1073 case BPF_LD | BPF_MEM:
1074 case BPF_LDX | BPF_MEM:
1075 insn->code = BPF_LDX | BPF_MEM | BPF_W;
1076 insn->a_reg = BPF_CLASS(fp->code) == BPF_LD ?
1078 insn->x_reg = FP_REG;
1079 insn->off = -(BPF_MEMWORDS - fp->k) * 4;
1082 /* A = K or X = K */
1083 case BPF_LD | BPF_IMM:
1084 case BPF_LDX | BPF_IMM:
1085 insn->code = BPF_ALU | BPF_MOV | BPF_K;
1086 insn->a_reg = BPF_CLASS(fp->code) == BPF_LD ?
1092 case BPF_MISC | BPF_TAX:
1093 insn->code = BPF_ALU64 | BPF_MOV | BPF_X;
1094 insn->a_reg = X_REG;
1095 insn->x_reg = A_REG;
1099 case BPF_MISC | BPF_TXA:
1100 insn->code = BPF_ALU64 | BPF_MOV | BPF_X;
1101 insn->a_reg = A_REG;
1102 insn->x_reg = X_REG;
1105 /* A = skb->len or X = skb->len */
1106 case BPF_LD | BPF_W | BPF_LEN:
1107 case BPF_LDX | BPF_W | BPF_LEN:
1108 insn->code = BPF_LDX | BPF_MEM | BPF_W;
1109 insn->a_reg = BPF_CLASS(fp->code) == BPF_LD ?
1111 insn->x_reg = CTX_REG;
1112 insn->off = offsetof(struct sk_buff, len);
1115 /* access seccomp_data fields */
1116 case BPF_LDX | BPF_ABS | BPF_W:
1117 insn->code = BPF_LDX | BPF_MEM | BPF_W;
1118 insn->a_reg = A_REG;
1119 insn->x_reg = CTX_REG;
1129 memcpy(new_insn, tmp_insns,
1130 sizeof(*insn) * (insn - tmp_insns));
1132 new_insn += insn - tmp_insns;
1136 /* Only calculating new length. */
1137 *new_len = new_insn - new_prog;
1142 if (new_flen != new_insn - new_prog) {
1143 new_flen = new_insn - new_prog;
1151 BUG_ON(*new_len != new_flen);
1160 * A BPF program is able to use 16 cells of memory to store intermediate
1161 * values (check u32 mem[BPF_MEMWORDS] in sk_run_filter()).
1163 * As we dont want to clear mem[] array for each packet going through
1164 * sk_run_filter(), we check that filter loaded by user never try to read
1165 * a cell if not previously written, and we check all branches to be sure
1166 * a malicious user doesn't try to abuse us.
1168 static int check_load_and_stores(struct sock_filter *filter, int flen)
1170 u16 *masks, memvalid = 0; /* one bit per cell, 16 cells */
1173 BUILD_BUG_ON(BPF_MEMWORDS > 16);
1174 masks = kmalloc(flen * sizeof(*masks), GFP_KERNEL);
1177 memset(masks, 0xff, flen * sizeof(*masks));
1179 for (pc = 0; pc < flen; pc++) {
1180 memvalid &= masks[pc];
1182 switch (filter[pc].code) {
1185 memvalid |= (1 << filter[pc].k);
1189 if (!(memvalid & (1 << filter[pc].k))) {
1195 /* a jump must set masks on target */
1196 masks[pc + 1 + filter[pc].k] &= memvalid;
1199 case BPF_S_JMP_JEQ_K:
1200 case BPF_S_JMP_JEQ_X:
1201 case BPF_S_JMP_JGE_K:
1202 case BPF_S_JMP_JGE_X:
1203 case BPF_S_JMP_JGT_K:
1204 case BPF_S_JMP_JGT_X:
1205 case BPF_S_JMP_JSET_X:
1206 case BPF_S_JMP_JSET_K:
1207 /* a jump must set masks on targets */
1208 masks[pc + 1 + filter[pc].jt] &= memvalid;
1209 masks[pc + 1 + filter[pc].jf] &= memvalid;
1220 * sk_chk_filter - verify socket filter code
1221 * @filter: filter to verify
1222 * @flen: length of filter
1224 * Check the user's filter code. If we let some ugly
1225 * filter code slip through kaboom! The filter must contain
1226 * no references or jumps that are out of range, no illegal
1227 * instructions, and must end with a RET instruction.
1229 * All jumps are forward as they are not signed.
1231 * Returns 0 if the rule set is legal or -EINVAL if not.
1233 int sk_chk_filter(struct sock_filter *filter, unsigned int flen)
1236 * Valid instructions are initialized to non-0.
1237 * Invalid instructions are initialized to 0.
1239 static const u8 codes[] = {
1240 [BPF_ALU|BPF_ADD|BPF_K] = BPF_S_ALU_ADD_K,
1241 [BPF_ALU|BPF_ADD|BPF_X] = BPF_S_ALU_ADD_X,
1242 [BPF_ALU|BPF_SUB|BPF_K] = BPF_S_ALU_SUB_K,
1243 [BPF_ALU|BPF_SUB|BPF_X] = BPF_S_ALU_SUB_X,
1244 [BPF_ALU|BPF_MUL|BPF_K] = BPF_S_ALU_MUL_K,
1245 [BPF_ALU|BPF_MUL|BPF_X] = BPF_S_ALU_MUL_X,
1246 [BPF_ALU|BPF_DIV|BPF_X] = BPF_S_ALU_DIV_X,
1247 [BPF_ALU|BPF_MOD|BPF_K] = BPF_S_ALU_MOD_K,
1248 [BPF_ALU|BPF_MOD|BPF_X] = BPF_S_ALU_MOD_X,
1249 [BPF_ALU|BPF_AND|BPF_K] = BPF_S_ALU_AND_K,
1250 [BPF_ALU|BPF_AND|BPF_X] = BPF_S_ALU_AND_X,
1251 [BPF_ALU|BPF_OR|BPF_K] = BPF_S_ALU_OR_K,
1252 [BPF_ALU|BPF_OR|BPF_X] = BPF_S_ALU_OR_X,
1253 [BPF_ALU|BPF_XOR|BPF_K] = BPF_S_ALU_XOR_K,
1254 [BPF_ALU|BPF_XOR|BPF_X] = BPF_S_ALU_XOR_X,
1255 [BPF_ALU|BPF_LSH|BPF_K] = BPF_S_ALU_LSH_K,
1256 [BPF_ALU|BPF_LSH|BPF_X] = BPF_S_ALU_LSH_X,
1257 [BPF_ALU|BPF_RSH|BPF_K] = BPF_S_ALU_RSH_K,
1258 [BPF_ALU|BPF_RSH|BPF_X] = BPF_S_ALU_RSH_X,
1259 [BPF_ALU|BPF_NEG] = BPF_S_ALU_NEG,
1260 [BPF_LD|BPF_W|BPF_ABS] = BPF_S_LD_W_ABS,
1261 [BPF_LD|BPF_H|BPF_ABS] = BPF_S_LD_H_ABS,
1262 [BPF_LD|BPF_B|BPF_ABS] = BPF_S_LD_B_ABS,
1263 [BPF_LD|BPF_W|BPF_LEN] = BPF_S_LD_W_LEN,
1264 [BPF_LD|BPF_W|BPF_IND] = BPF_S_LD_W_IND,
1265 [BPF_LD|BPF_H|BPF_IND] = BPF_S_LD_H_IND,
1266 [BPF_LD|BPF_B|BPF_IND] = BPF_S_LD_B_IND,
1267 [BPF_LD|BPF_IMM] = BPF_S_LD_IMM,
1268 [BPF_LDX|BPF_W|BPF_LEN] = BPF_S_LDX_W_LEN,
1269 [BPF_LDX|BPF_B|BPF_MSH] = BPF_S_LDX_B_MSH,
1270 [BPF_LDX|BPF_IMM] = BPF_S_LDX_IMM,
1271 [BPF_MISC|BPF_TAX] = BPF_S_MISC_TAX,
1272 [BPF_MISC|BPF_TXA] = BPF_S_MISC_TXA,
1273 [BPF_RET|BPF_K] = BPF_S_RET_K,
1274 [BPF_RET|BPF_A] = BPF_S_RET_A,
1275 [BPF_ALU|BPF_DIV|BPF_K] = BPF_S_ALU_DIV_K,
1276 [BPF_LD|BPF_MEM] = BPF_S_LD_MEM,
1277 [BPF_LDX|BPF_MEM] = BPF_S_LDX_MEM,
1278 [BPF_ST] = BPF_S_ST,
1279 [BPF_STX] = BPF_S_STX,
1280 [BPF_JMP|BPF_JA] = BPF_S_JMP_JA,
1281 [BPF_JMP|BPF_JEQ|BPF_K] = BPF_S_JMP_JEQ_K,
1282 [BPF_JMP|BPF_JEQ|BPF_X] = BPF_S_JMP_JEQ_X,
1283 [BPF_JMP|BPF_JGE|BPF_K] = BPF_S_JMP_JGE_K,
1284 [BPF_JMP|BPF_JGE|BPF_X] = BPF_S_JMP_JGE_X,
1285 [BPF_JMP|BPF_JGT|BPF_K] = BPF_S_JMP_JGT_K,
1286 [BPF_JMP|BPF_JGT|BPF_X] = BPF_S_JMP_JGT_X,
1287 [BPF_JMP|BPF_JSET|BPF_K] = BPF_S_JMP_JSET_K,
1288 [BPF_JMP|BPF_JSET|BPF_X] = BPF_S_JMP_JSET_X,
1293 if (flen == 0 || flen > BPF_MAXINSNS)
1296 /* check the filter code now */
1297 for (pc = 0; pc < flen; pc++) {
1298 struct sock_filter *ftest = &filter[pc];
1299 u16 code = ftest->code;
1301 if (code >= ARRAY_SIZE(codes))
1306 /* Some instructions need special checks */
1308 case BPF_S_ALU_DIV_K:
1309 case BPF_S_ALU_MOD_K:
1310 /* check for division by zero */
1318 /* check for invalid memory addresses */
1319 if (ftest->k >= BPF_MEMWORDS)
1324 * Note, the large ftest->k might cause loops.
1325 * Compare this with conditional jumps below,
1326 * where offsets are limited. --ANK (981016)
1328 if (ftest->k >= (unsigned int)(flen-pc-1))
1331 case BPF_S_JMP_JEQ_K:
1332 case BPF_S_JMP_JEQ_X:
1333 case BPF_S_JMP_JGE_K:
1334 case BPF_S_JMP_JGE_X:
1335 case BPF_S_JMP_JGT_K:
1336 case BPF_S_JMP_JGT_X:
1337 case BPF_S_JMP_JSET_X:
1338 case BPF_S_JMP_JSET_K:
1339 /* for conditionals both must be safe */
1340 if (pc + ftest->jt + 1 >= flen ||
1341 pc + ftest->jf + 1 >= flen)
1344 case BPF_S_LD_W_ABS:
1345 case BPF_S_LD_H_ABS:
1346 case BPF_S_LD_B_ABS:
1348 #define ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
1349 code = BPF_S_ANC_##CODE; \
1353 ANCILLARY(PROTOCOL);
1357 ANCILLARY(NLATTR_NEST);
1363 ANCILLARY(ALU_XOR_X);
1364 ANCILLARY(VLAN_TAG);
1365 ANCILLARY(VLAN_TAG_PRESENT);
1366 ANCILLARY(PAY_OFFSET);
1369 /* ancillary operation unknown or unsupported */
1370 if (anc_found == false && ftest->k >= SKF_AD_OFF)
1376 /* last instruction must be a RET code */
1377 switch (filter[flen - 1].code) {
1380 return check_load_and_stores(filter, flen);
1384 EXPORT_SYMBOL(sk_chk_filter);
1386 static int sk_store_orig_filter(struct sk_filter *fp,
1387 const struct sock_fprog *fprog)
1389 unsigned int fsize = sk_filter_proglen(fprog);
1390 struct sock_fprog_kern *fkprog;
1392 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1396 fkprog = fp->orig_prog;
1397 fkprog->len = fprog->len;
1398 fkprog->filter = kmemdup(fp->insns, fsize, GFP_KERNEL);
1399 if (!fkprog->filter) {
1400 kfree(fp->orig_prog);
1407 static void sk_release_orig_filter(struct sk_filter *fp)
1409 struct sock_fprog_kern *fprog = fp->orig_prog;
1412 kfree(fprog->filter);
1418 * sk_filter_release_rcu - Release a socket filter by rcu_head
1419 * @rcu: rcu_head that contains the sk_filter to free
1421 static void sk_filter_release_rcu(struct rcu_head *rcu)
1423 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1425 sk_release_orig_filter(fp);
1430 * sk_filter_release - release a socket filter
1431 * @fp: filter to remove
1433 * Remove a filter from a socket and release its resources.
1435 static void sk_filter_release(struct sk_filter *fp)
1437 if (atomic_dec_and_test(&fp->refcnt))
1438 call_rcu(&fp->rcu, sk_filter_release_rcu);
1441 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1443 atomic_sub(sk_filter_size(fp->len), &sk->sk_omem_alloc);
1444 sk_filter_release(fp);
1447 void sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1449 atomic_inc(&fp->refcnt);
1450 atomic_add(sk_filter_size(fp->len), &sk->sk_omem_alloc);
1453 static struct sk_filter *__sk_migrate_realloc(struct sk_filter *fp,
1457 struct sk_filter *fp_new;
1460 return krealloc(fp, len, GFP_KERNEL);
1462 fp_new = sock_kmalloc(sk, len, GFP_KERNEL);
1464 memcpy(fp_new, fp, sizeof(struct sk_filter));
1465 /* As we're kepping orig_prog in fp_new along,
1466 * we need to make sure we're not evicting it
1469 fp->orig_prog = NULL;
1470 sk_filter_uncharge(sk, fp);
1476 static struct sk_filter *__sk_migrate_filter(struct sk_filter *fp,
1479 struct sock_filter *old_prog;
1480 struct sk_filter *old_fp;
1481 int i, err, new_len, old_len = fp->len;
1483 /* We are free to overwrite insns et al right here as it
1484 * won't be used at this point in time anymore internally
1485 * after the migration to the internal BPF instruction
1488 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1489 sizeof(struct sock_filter_int));
1491 /* For now, we need to unfiddle BPF_S_* identifiers in place.
1492 * This can sooner or later on be subject to removal, e.g. when
1493 * JITs have been converted.
1495 for (i = 0; i < fp->len; i++)
1496 sk_decode_filter(&fp->insns[i], &fp->insns[i]);
1498 /* Conversion cannot happen on overlapping memory areas,
1499 * so we need to keep the user BPF around until the 2nd
1500 * pass. At this time, the user BPF is stored in fp->insns.
1502 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1509 /* 1st pass: calculate the new program length. */
1510 err = sk_convert_filter(old_prog, old_len, NULL, &new_len);
1514 /* Expand fp for appending the new filter representation. */
1516 fp = __sk_migrate_realloc(old_fp, sk, sk_filter_size(new_len));
1518 /* The old_fp is still around in case we couldn't
1519 * allocate new memory, so uncharge on that one.
1526 fp->bpf_func = sk_run_filter_int_skb;
1529 /* 2nd pass: remap sock_filter insns into sock_filter_int insns. */
1530 err = sk_convert_filter(old_prog, old_len, fp->insnsi, &new_len);
1532 /* 2nd sk_convert_filter() can fail only if it fails
1533 * to allocate memory, remapping must succeed. Note,
1534 * that at this time old_fp has already been released
1535 * by __sk_migrate_realloc().
1545 /* Rollback filter setup. */
1547 sk_filter_uncharge(sk, fp);
1550 return ERR_PTR(err);
1553 static struct sk_filter *__sk_prepare_filter(struct sk_filter *fp,
1558 fp->bpf_func = NULL;
1561 err = sk_chk_filter(fp->insns, fp->len);
1563 return ERR_PTR(err);
1565 /* Probe if we can JIT compile the filter and if so, do
1566 * the compilation of the filter.
1568 bpf_jit_compile(fp);
1570 /* JIT compiler couldn't process this filter, so do the
1571 * internal BPF translation for the optimized interpreter.
1574 fp = __sk_migrate_filter(fp, sk);
1580 * sk_unattached_filter_create - create an unattached filter
1581 * @fprog: the filter program
1582 * @pfp: the unattached filter that is created
1584 * Create a filter independent of any socket. We first run some
1585 * sanity checks on it to make sure it does not explode on us later.
1586 * If an error occurs or there is insufficient memory for the filter
1587 * a negative errno code is returned. On success the return is zero.
1589 int sk_unattached_filter_create(struct sk_filter **pfp,
1590 struct sock_fprog *fprog)
1592 unsigned int fsize = sk_filter_proglen(fprog);
1593 struct sk_filter *fp;
1595 /* Make sure new filter is there and in the right amounts. */
1596 if (fprog->filter == NULL)
1599 fp = kmalloc(sk_filter_size(fprog->len), GFP_KERNEL);
1603 memcpy(fp->insns, fprog->filter, fsize);
1605 atomic_set(&fp->refcnt, 1);
1606 fp->len = fprog->len;
1607 /* Since unattached filters are not copied back to user
1608 * space through sk_get_filter(), we do not need to hold
1609 * a copy here, and can spare us the work.
1611 fp->orig_prog = NULL;
1613 /* __sk_prepare_filter() already takes care of uncharging
1614 * memory in case something goes wrong.
1616 fp = __sk_prepare_filter(fp, NULL);
1623 EXPORT_SYMBOL_GPL(sk_unattached_filter_create);
1625 void sk_unattached_filter_destroy(struct sk_filter *fp)
1627 sk_filter_release(fp);
1629 EXPORT_SYMBOL_GPL(sk_unattached_filter_destroy);
1632 * sk_attach_filter - attach a socket filter
1633 * @fprog: the filter program
1634 * @sk: the socket to use
1636 * Attach the user's filter code. We first run some sanity checks on
1637 * it to make sure it does not explode on us later. If an error
1638 * occurs or there is insufficient memory for the filter a negative
1639 * errno code is returned. On success the return is zero.
1641 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1643 struct sk_filter *fp, *old_fp;
1644 unsigned int fsize = sk_filter_proglen(fprog);
1645 unsigned int sk_fsize = sk_filter_size(fprog->len);
1648 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1651 /* Make sure new filter is there and in the right amounts. */
1652 if (fprog->filter == NULL)
1655 fp = sock_kmalloc(sk, sk_fsize, GFP_KERNEL);
1659 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1660 sock_kfree_s(sk, fp, sk_fsize);
1664 atomic_set(&fp->refcnt, 1);
1665 fp->len = fprog->len;
1667 err = sk_store_orig_filter(fp, fprog);
1669 sk_filter_uncharge(sk, fp);
1673 /* __sk_prepare_filter() already takes care of uncharging
1674 * memory in case something goes wrong.
1676 fp = __sk_prepare_filter(fp, sk);
1680 old_fp = rcu_dereference_protected(sk->sk_filter,
1681 sock_owned_by_user(sk));
1682 rcu_assign_pointer(sk->sk_filter, fp);
1685 sk_filter_uncharge(sk, old_fp);
1689 EXPORT_SYMBOL_GPL(sk_attach_filter);
1691 int sk_detach_filter(struct sock *sk)
1694 struct sk_filter *filter;
1696 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1699 filter = rcu_dereference_protected(sk->sk_filter,
1700 sock_owned_by_user(sk));
1702 RCU_INIT_POINTER(sk->sk_filter, NULL);
1703 sk_filter_uncharge(sk, filter);
1709 EXPORT_SYMBOL_GPL(sk_detach_filter);
1711 void sk_decode_filter(struct sock_filter *filt, struct sock_filter *to)
1713 static const u16 decodes[] = {
1714 [BPF_S_ALU_ADD_K] = BPF_ALU|BPF_ADD|BPF_K,
1715 [BPF_S_ALU_ADD_X] = BPF_ALU|BPF_ADD|BPF_X,
1716 [BPF_S_ALU_SUB_K] = BPF_ALU|BPF_SUB|BPF_K,
1717 [BPF_S_ALU_SUB_X] = BPF_ALU|BPF_SUB|BPF_X,
1718 [BPF_S_ALU_MUL_K] = BPF_ALU|BPF_MUL|BPF_K,
1719 [BPF_S_ALU_MUL_X] = BPF_ALU|BPF_MUL|BPF_X,
1720 [BPF_S_ALU_DIV_X] = BPF_ALU|BPF_DIV|BPF_X,
1721 [BPF_S_ALU_MOD_K] = BPF_ALU|BPF_MOD|BPF_K,
1722 [BPF_S_ALU_MOD_X] = BPF_ALU|BPF_MOD|BPF_X,
1723 [BPF_S_ALU_AND_K] = BPF_ALU|BPF_AND|BPF_K,
1724 [BPF_S_ALU_AND_X] = BPF_ALU|BPF_AND|BPF_X,
1725 [BPF_S_ALU_OR_K] = BPF_ALU|BPF_OR|BPF_K,
1726 [BPF_S_ALU_OR_X] = BPF_ALU|BPF_OR|BPF_X,
1727 [BPF_S_ALU_XOR_K] = BPF_ALU|BPF_XOR|BPF_K,
1728 [BPF_S_ALU_XOR_X] = BPF_ALU|BPF_XOR|BPF_X,
1729 [BPF_S_ALU_LSH_K] = BPF_ALU|BPF_LSH|BPF_K,
1730 [BPF_S_ALU_LSH_X] = BPF_ALU|BPF_LSH|BPF_X,
1731 [BPF_S_ALU_RSH_K] = BPF_ALU|BPF_RSH|BPF_K,
1732 [BPF_S_ALU_RSH_X] = BPF_ALU|BPF_RSH|BPF_X,
1733 [BPF_S_ALU_NEG] = BPF_ALU|BPF_NEG,
1734 [BPF_S_LD_W_ABS] = BPF_LD|BPF_W|BPF_ABS,
1735 [BPF_S_LD_H_ABS] = BPF_LD|BPF_H|BPF_ABS,
1736 [BPF_S_LD_B_ABS] = BPF_LD|BPF_B|BPF_ABS,
1737 [BPF_S_ANC_PROTOCOL] = BPF_LD|BPF_B|BPF_ABS,
1738 [BPF_S_ANC_PKTTYPE] = BPF_LD|BPF_B|BPF_ABS,
1739 [BPF_S_ANC_IFINDEX] = BPF_LD|BPF_B|BPF_ABS,
1740 [BPF_S_ANC_NLATTR] = BPF_LD|BPF_B|BPF_ABS,
1741 [BPF_S_ANC_NLATTR_NEST] = BPF_LD|BPF_B|BPF_ABS,
1742 [BPF_S_ANC_MARK] = BPF_LD|BPF_B|BPF_ABS,
1743 [BPF_S_ANC_QUEUE] = BPF_LD|BPF_B|BPF_ABS,
1744 [BPF_S_ANC_HATYPE] = BPF_LD|BPF_B|BPF_ABS,
1745 [BPF_S_ANC_RXHASH] = BPF_LD|BPF_B|BPF_ABS,
1746 [BPF_S_ANC_CPU] = BPF_LD|BPF_B|BPF_ABS,
1747 [BPF_S_ANC_ALU_XOR_X] = BPF_LD|BPF_B|BPF_ABS,
1748 [BPF_S_ANC_VLAN_TAG] = BPF_LD|BPF_B|BPF_ABS,
1749 [BPF_S_ANC_VLAN_TAG_PRESENT] = BPF_LD|BPF_B|BPF_ABS,
1750 [BPF_S_ANC_PAY_OFFSET] = BPF_LD|BPF_B|BPF_ABS,
1751 [BPF_S_LD_W_LEN] = BPF_LD|BPF_W|BPF_LEN,
1752 [BPF_S_LD_W_IND] = BPF_LD|BPF_W|BPF_IND,
1753 [BPF_S_LD_H_IND] = BPF_LD|BPF_H|BPF_IND,
1754 [BPF_S_LD_B_IND] = BPF_LD|BPF_B|BPF_IND,
1755 [BPF_S_LD_IMM] = BPF_LD|BPF_IMM,
1756 [BPF_S_LDX_W_LEN] = BPF_LDX|BPF_W|BPF_LEN,
1757 [BPF_S_LDX_B_MSH] = BPF_LDX|BPF_B|BPF_MSH,
1758 [BPF_S_LDX_IMM] = BPF_LDX|BPF_IMM,
1759 [BPF_S_MISC_TAX] = BPF_MISC|BPF_TAX,
1760 [BPF_S_MISC_TXA] = BPF_MISC|BPF_TXA,
1761 [BPF_S_RET_K] = BPF_RET|BPF_K,
1762 [BPF_S_RET_A] = BPF_RET|BPF_A,
1763 [BPF_S_ALU_DIV_K] = BPF_ALU|BPF_DIV|BPF_K,
1764 [BPF_S_LD_MEM] = BPF_LD|BPF_MEM,
1765 [BPF_S_LDX_MEM] = BPF_LDX|BPF_MEM,
1766 [BPF_S_ST] = BPF_ST,
1767 [BPF_S_STX] = BPF_STX,
1768 [BPF_S_JMP_JA] = BPF_JMP|BPF_JA,
1769 [BPF_S_JMP_JEQ_K] = BPF_JMP|BPF_JEQ|BPF_K,
1770 [BPF_S_JMP_JEQ_X] = BPF_JMP|BPF_JEQ|BPF_X,
1771 [BPF_S_JMP_JGE_K] = BPF_JMP|BPF_JGE|BPF_K,
1772 [BPF_S_JMP_JGE_X] = BPF_JMP|BPF_JGE|BPF_X,
1773 [BPF_S_JMP_JGT_K] = BPF_JMP|BPF_JGT|BPF_K,
1774 [BPF_S_JMP_JGT_X] = BPF_JMP|BPF_JGT|BPF_X,
1775 [BPF_S_JMP_JSET_K] = BPF_JMP|BPF_JSET|BPF_K,
1776 [BPF_S_JMP_JSET_X] = BPF_JMP|BPF_JSET|BPF_X,
1782 to->code = decodes[code];
1788 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
1791 struct sock_fprog_kern *fprog;
1792 struct sk_filter *filter;
1796 filter = rcu_dereference_protected(sk->sk_filter,
1797 sock_owned_by_user(sk));
1801 /* We're copying the filter that has been originally attached,
1802 * so no conversion/decode needed anymore.
1804 fprog = filter->orig_prog;
1808 /* User space only enquires number of filter blocks. */
1812 if (len < fprog->len)
1816 if (copy_to_user(ubuf, fprog->filter, sk_filter_proglen(fprog)))
1819 /* Instead of bytes, the API requests to return the number