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 u8 *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)
54 * sk_run_filter - run a filter on a socket
55 * @skb: buffer to run the filter on
56 * @filter: filter to apply
57 * @flen: length of filter
59 * Decode and apply filter instructions to the skb->data.
60 * Return length to keep, 0 for none. skb is the data we are
61 * filtering, filter is the array of filter instructions, and
62 * len is the number of filter blocks in the array.
65 int sk_run_filter(struct sk_buff *skb, struct sock_filter *filter, int flen)
67 unsigned char *data = skb->data;
68 /* len is UNSIGNED. Byte wide insns relies only on implicit
69 type casts to prevent reading arbitrary memory locations.
71 unsigned int len = skb->len-skb->data_len;
72 struct sock_filter *fentry; /* We walk down these */
73 u32 A = 0; /* Accumulator */
74 u32 X = 0; /* Index Register */
75 u32 mem[BPF_MEMWORDS]; /* Scratch Memory Store */
80 * Process array of filter instructions.
82 for (pc = 0; pc < flen; pc++) {
85 switch (fentry->code) {
86 case BPF_ALU|BPF_ADD|BPF_X:
89 case BPF_ALU|BPF_ADD|BPF_K:
92 case BPF_ALU|BPF_SUB|BPF_X:
95 case BPF_ALU|BPF_SUB|BPF_K:
98 case BPF_ALU|BPF_MUL|BPF_X:
101 case BPF_ALU|BPF_MUL|BPF_K:
104 case BPF_ALU|BPF_DIV|BPF_X:
109 case BPF_ALU|BPF_DIV|BPF_K:
114 case BPF_ALU|BPF_AND|BPF_X:
117 case BPF_ALU|BPF_AND|BPF_K:
120 case BPF_ALU|BPF_OR|BPF_X:
123 case BPF_ALU|BPF_OR|BPF_K:
126 case BPF_ALU|BPF_LSH|BPF_X:
129 case BPF_ALU|BPF_LSH|BPF_K:
132 case BPF_ALU|BPF_RSH|BPF_X:
135 case BPF_ALU|BPF_RSH|BPF_K:
138 case BPF_ALU|BPF_NEG:
144 case BPF_JMP|BPF_JGT|BPF_K:
145 pc += (A > fentry->k) ? fentry->jt : fentry->jf;
147 case BPF_JMP|BPF_JGE|BPF_K:
148 pc += (A >= fentry->k) ? fentry->jt : fentry->jf;
150 case BPF_JMP|BPF_JEQ|BPF_K:
151 pc += (A == fentry->k) ? fentry->jt : fentry->jf;
153 case BPF_JMP|BPF_JSET|BPF_K:
154 pc += (A & fentry->k) ? fentry->jt : fentry->jf;
156 case BPF_JMP|BPF_JGT|BPF_X:
157 pc += (A > X) ? fentry->jt : fentry->jf;
159 case BPF_JMP|BPF_JGE|BPF_X:
160 pc += (A >= X) ? fentry->jt : fentry->jf;
162 case BPF_JMP|BPF_JEQ|BPF_X:
163 pc += (A == X) ? fentry->jt : fentry->jf;
165 case BPF_JMP|BPF_JSET|BPF_X:
166 pc += (A & X) ? fentry->jt : fentry->jf;
168 case BPF_LD|BPF_W|BPF_ABS:
176 ptr = load_pointer(skb, k);
178 A = ntohl(*(u32*)ptr);
183 p = skb_header_pointer(skb, k, 4, &_tmp);
190 case BPF_LD|BPF_H|BPF_ABS:
198 ptr = load_pointer(skb, k);
200 A = ntohs(*(u16*)ptr);
205 p = skb_header_pointer(skb, k, 2, &_tmp);
212 case BPF_LD|BPF_B|BPF_ABS:
220 ptr = load_pointer(skb, k);
227 p = skb_header_pointer(skb, k, 1, &_tmp);
234 case BPF_LD|BPF_W|BPF_LEN:
237 case BPF_LDX|BPF_W|BPF_LEN:
240 case BPF_LD|BPF_W|BPF_IND:
243 case BPF_LD|BPF_H|BPF_IND:
246 case BPF_LD|BPF_B|BPF_IND:
249 case BPF_LDX|BPF_B|BPF_MSH:
250 if (fentry->k >= len)
252 X = (data[fentry->k] & 0xf) << 2;
257 case BPF_LDX|BPF_IMM:
263 case BPF_LDX|BPF_MEM:
266 case BPF_MISC|BPF_TAX:
269 case BPF_MISC|BPF_TXA:
273 return ((unsigned int)fentry->k);
275 return ((unsigned int)A);
283 /* Invalid instruction counts as RET */
288 * Handle ancillary data, which are impossible
289 * (or very difficult) to get parsing packet contents.
291 switch (k-SKF_AD_OFF) {
292 case SKF_AD_PROTOCOL:
293 A = htons(skb->protocol);
299 A = skb->dev->ifindex;
310 * sk_chk_filter - verify socket filter code
311 * @filter: filter to verify
312 * @flen: length of filter
314 * Check the user's filter code. If we let some ugly
315 * filter code slip through kaboom! The filter must contain
316 * no references or jumps that are out of range, no illegal instructions
317 * and no backward jumps. It must end with a RET instruction
319 * Returns 0 if the rule set is legal or a negative errno code if not.
321 int sk_chk_filter(struct sock_filter *filter, int flen)
323 struct sock_filter *ftest;
326 if (((unsigned int)flen >= (~0U / sizeof(struct sock_filter))) || flen == 0)
329 /* check the filter code now */
330 for (pc = 0; pc < flen; pc++) {
331 /* all jumps are forward as they are not signed */
333 if (BPF_CLASS(ftest->code) == BPF_JMP) {
334 /* but they mustn't jump off the end */
335 if (BPF_OP(ftest->code) == BPF_JA) {
337 * Note, the large ftest->k might cause loops.
338 * Compare this with conditional jumps below,
339 * where offsets are limited. --ANK (981016)
341 if (ftest->k >= (unsigned)(flen-pc-1))
344 /* for conditionals both must be safe */
345 if (pc + ftest->jt +1 >= flen ||
346 pc + ftest->jf +1 >= flen)
351 /* check that memory operations use valid addresses. */
352 if (ftest->k >= BPF_MEMWORDS) {
353 /* but it might not be a memory operation... */
354 switch (ftest->code) {
358 case BPF_LDX|BPF_MEM:
365 * The program must end with a return. We don't care where they
366 * jumped within the script (its always forwards) but in the end
367 * they _will_ hit this.
369 return (BPF_CLASS(filter[flen - 1].code) == BPF_RET) ? 0 : -EINVAL;
373 * sk_attach_filter - attach a socket filter
374 * @fprog: the filter program
375 * @sk: the socket to use
377 * Attach the user's filter code. We first run some sanity checks on
378 * it to make sure it does not explode on us later. If an error
379 * occurs or there is insufficient memory for the filter a negative
380 * errno code is returned. On success the return is zero.
382 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
384 struct sk_filter *fp;
385 unsigned int fsize = sizeof(struct sock_filter) * fprog->len;
388 /* Make sure new filter is there and in the right amounts. */
389 if (fprog->filter == NULL || fprog->len > BPF_MAXINSNS)
392 fp = sock_kmalloc(sk, fsize+sizeof(*fp), GFP_KERNEL);
395 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
396 sock_kfree_s(sk, fp, fsize+sizeof(*fp));
400 atomic_set(&fp->refcnt, 1);
401 fp->len = fprog->len;
403 err = sk_chk_filter(fp->insns, fp->len);
405 struct sk_filter *old_fp;
407 spin_lock_bh(&sk->sk_lock.slock);
408 old_fp = sk->sk_filter;
410 spin_unlock_bh(&sk->sk_lock.slock);
415 sk_filter_release(sk, fp);
419 EXPORT_SYMBOL(sk_chk_filter);
420 EXPORT_SYMBOL(sk_run_filter);