2 * Linux Socket Filter - Kernel level socket filtering
5 * Jay Schulist <jschlst@samba.org>
7 * Based on the design of:
8 * - The Berkeley Packet Filter
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
15 * Andi Kleen - Fix a few bad bugs and races.
18 #include <linux/module.h>
19 #include <linux/types.h>
20 #include <linux/sched.h>
22 #include <linux/fcntl.h>
23 #include <linux/socket.h>
25 #include <linux/inet.h>
26 #include <linux/netdevice.h>
27 #include <linux/if_packet.h>
29 #include <net/protocol.h>
30 #include <linux/skbuff.h>
32 #include <linux/errno.h>
33 #include <linux/timer.h>
34 #include <asm/system.h>
35 #include <asm/uaccess.h>
36 #include <linux/filter.h>
38 /* No hurry in this branch */
39 static void *__load_pointer(struct sk_buff *skb, int k)
44 ptr = skb->nh.raw + k - SKF_NET_OFF;
45 else if (k >= SKF_LL_OFF)
46 ptr = skb->mac.raw + k - SKF_LL_OFF;
48 if (ptr >= skb->head && ptr < skb->tail)
53 static inline void *load_pointer(struct sk_buff *skb, int k,
54 unsigned int size, void *buffer)
57 return skb_header_pointer(skb, k, size, buffer);
61 return __load_pointer(skb, k);
66 * sk_run_filter - run a filter on a socket
67 * @skb: buffer to run the filter on
68 * @filter: filter to apply
69 * @flen: length of filter
71 * Decode and apply filter instructions to the skb->data.
72 * Return length to keep, 0 for none. skb is the data we are
73 * filtering, filter is the array of filter instructions, and
74 * len is the number of filter blocks in the array.
77 int sk_run_filter(struct sk_buff *skb, struct sock_filter *filter, int flen)
79 /* len is UNSIGNED. Byte wide insns relies only on implicit
80 type casts to prevent reading arbitrary memory locations.
82 unsigned int len = skb->len-skb->data_len;
83 struct sock_filter *fentry; /* We walk down these */
85 u32 A = 0; /* Accumulator */
86 u32 X = 0; /* Index Register */
87 u32 mem[BPF_MEMWORDS]; /* Scratch Memory Store */
93 * Process array of filter instructions.
95 for (pc = 0; pc < flen; pc++) {
98 switch (fentry->code) {
99 case BPF_ALU|BPF_ADD|BPF_X:
102 case BPF_ALU|BPF_ADD|BPF_K:
105 case BPF_ALU|BPF_SUB|BPF_X:
108 case BPF_ALU|BPF_SUB|BPF_K:
111 case BPF_ALU|BPF_MUL|BPF_X:
114 case BPF_ALU|BPF_MUL|BPF_K:
117 case BPF_ALU|BPF_DIV|BPF_X:
122 case BPF_ALU|BPF_DIV|BPF_K:
127 case BPF_ALU|BPF_AND|BPF_X:
130 case BPF_ALU|BPF_AND|BPF_K:
133 case BPF_ALU|BPF_OR|BPF_X:
136 case BPF_ALU|BPF_OR|BPF_K:
139 case BPF_ALU|BPF_LSH|BPF_X:
142 case BPF_ALU|BPF_LSH|BPF_K:
145 case BPF_ALU|BPF_RSH|BPF_X:
148 case BPF_ALU|BPF_RSH|BPF_K:
151 case BPF_ALU|BPF_NEG:
157 case BPF_JMP|BPF_JGT|BPF_K:
158 pc += (A > fentry->k) ? fentry->jt : fentry->jf;
160 case BPF_JMP|BPF_JGE|BPF_K:
161 pc += (A >= fentry->k) ? fentry->jt : fentry->jf;
163 case BPF_JMP|BPF_JEQ|BPF_K:
164 pc += (A == fentry->k) ? fentry->jt : fentry->jf;
166 case BPF_JMP|BPF_JSET|BPF_K:
167 pc += (A & fentry->k) ? fentry->jt : fentry->jf;
169 case BPF_JMP|BPF_JGT|BPF_X:
170 pc += (A > X) ? fentry->jt : fentry->jf;
172 case BPF_JMP|BPF_JGE|BPF_X:
173 pc += (A >= X) ? fentry->jt : fentry->jf;
175 case BPF_JMP|BPF_JEQ|BPF_X:
176 pc += (A == X) ? fentry->jt : fentry->jf;
178 case BPF_JMP|BPF_JSET|BPF_X:
179 pc += (A & X) ? fentry->jt : fentry->jf;
181 case BPF_LD|BPF_W|BPF_ABS:
184 ptr = load_pointer(skb, k, 4, &tmp);
186 A = ntohl(*(u32 *)ptr);
190 case BPF_LD|BPF_H|BPF_ABS:
193 ptr = load_pointer(skb, k, 2, &tmp);
195 A = ntohs(*(u16 *)ptr);
199 case BPF_LD|BPF_B|BPF_ABS:
202 ptr = load_pointer(skb, k, 1, &tmp);
208 case BPF_LD|BPF_W|BPF_LEN:
211 case BPF_LDX|BPF_W|BPF_LEN:
214 case BPF_LD|BPF_W|BPF_IND:
217 case BPF_LD|BPF_H|BPF_IND:
220 case BPF_LD|BPF_B|BPF_IND:
223 case BPF_LDX|BPF_B|BPF_MSH:
224 ptr = load_pointer(skb, fentry->k, 1, &tmp);
226 X = (*(u8 *)ptr & 0xf) << 2;
233 case BPF_LDX|BPF_IMM:
239 case BPF_LDX|BPF_MEM:
242 case BPF_MISC|BPF_TAX:
245 case BPF_MISC|BPF_TXA:
249 return ((unsigned int)fentry->k);
251 return ((unsigned int)A);
259 /* Invalid instruction counts as RET */
264 * Handle ancillary data, which are impossible
265 * (or very difficult) to get parsing packet contents.
267 switch (k-SKF_AD_OFF) {
268 case SKF_AD_PROTOCOL:
269 A = htons(skb->protocol);
275 A = skb->dev->ifindex;
286 * sk_chk_filter - verify socket filter code
287 * @filter: filter to verify
288 * @flen: length of filter
290 * Check the user's filter code. If we let some ugly
291 * filter code slip through kaboom! The filter must contain
292 * no references or jumps that are out of range, no illegal instructions
293 * and no backward jumps. It must end with a RET instruction
295 * Returns 0 if the rule set is legal or a negative errno code if not.
297 int sk_chk_filter(struct sock_filter *filter, int flen)
299 struct sock_filter *ftest;
302 if (((unsigned int)flen >= (~0U / sizeof(struct sock_filter))) || flen == 0)
305 /* check the filter code now */
306 for (pc = 0; pc < flen; pc++) {
307 /* all jumps are forward as they are not signed */
309 if (BPF_CLASS(ftest->code) == BPF_JMP) {
310 /* but they mustn't jump off the end */
311 if (BPF_OP(ftest->code) == BPF_JA) {
313 * Note, the large ftest->k might cause loops.
314 * Compare this with conditional jumps below,
315 * where offsets are limited. --ANK (981016)
317 if (ftest->k >= (unsigned)(flen-pc-1))
320 /* for conditionals both must be safe */
321 if (pc + ftest->jt +1 >= flen ||
322 pc + ftest->jf +1 >= flen)
327 /* check that memory operations use valid addresses. */
328 if (ftest->k >= BPF_MEMWORDS) {
329 /* but it might not be a memory operation... */
330 switch (ftest->code) {
334 case BPF_LDX|BPF_MEM:
341 * The program must end with a return. We don't care where they
342 * jumped within the script (its always forwards) but in the end
343 * they _will_ hit this.
345 return (BPF_CLASS(filter[flen - 1].code) == BPF_RET) ? 0 : -EINVAL;
349 * sk_attach_filter - attach a socket filter
350 * @fprog: the filter program
351 * @sk: the socket to use
353 * Attach the user's filter code. We first run some sanity checks on
354 * it to make sure it does not explode on us later. If an error
355 * occurs or there is insufficient memory for the filter a negative
356 * errno code is returned. On success the return is zero.
358 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
360 struct sk_filter *fp;
361 unsigned int fsize = sizeof(struct sock_filter) * fprog->len;
364 /* Make sure new filter is there and in the right amounts. */
365 if (fprog->filter == NULL || fprog->len > BPF_MAXINSNS)
368 fp = sock_kmalloc(sk, fsize+sizeof(*fp), GFP_KERNEL);
371 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
372 sock_kfree_s(sk, fp, fsize+sizeof(*fp));
376 atomic_set(&fp->refcnt, 1);
377 fp->len = fprog->len;
379 err = sk_chk_filter(fp->insns, fp->len);
381 struct sk_filter *old_fp;
383 spin_lock_bh(&sk->sk_lock.slock);
384 old_fp = sk->sk_filter;
386 spin_unlock_bh(&sk->sk_lock.slock);
391 sk_filter_release(sk, fp);
395 EXPORT_SYMBOL(sk_chk_filter);
396 EXPORT_SYMBOL(sk_run_filter);