allow coexistance of N build and AC build.
[tomato.git] / release / src-rt-6.x / linux / linux-2.6 / net / core / filter.c
blob9c704a72376cef076f78186eb9ab7da6100c218f
1 /*
2 * Linux Socket Filter - Kernel level socket filtering
4 * Author:
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.
16 * Kris Katterjohn - Added many additional checks in sk_chk_filter()
19 #include <linux/module.h>
20 #include <linux/types.h>
21 #include <linux/mm.h>
22 #include <linux/fcntl.h>
23 #include <linux/socket.h>
24 #include <linux/in.h>
25 #include <linux/inet.h>
26 #include <linux/netdevice.h>
27 #include <linux/if_packet.h>
28 #include <net/ip.h>
29 #include <net/protocol.h>
30 #include <linux/skbuff.h>
31 #include <net/sock.h>
32 #include <linux/errno.h>
33 #include <linux/timer.h>
34 #include <asm/system.h>
35 #include <asm/uaccess.h>
36 #include <asm/unaligned.h>
37 #include <linux/filter.h>
39 /* No hurry in this branch */
40 static void *__load_pointer(struct sk_buff *skb, int k)
42 u8 *ptr = NULL;
44 if (k >= SKF_NET_OFF)
45 ptr = skb_network_header(skb) + k - SKF_NET_OFF;
46 else if (k >= SKF_LL_OFF)
47 ptr = skb_mac_header(skb) + k - SKF_LL_OFF;
49 if (ptr >= skb->head && ptr < skb_tail_pointer(skb))
50 return ptr;
51 return NULL;
54 static inline void *load_pointer(struct sk_buff *skb, int k,
55 unsigned int size, void *buffer)
57 if (k >= 0)
58 return skb_header_pointer(skb, k, size, buffer);
59 else {
60 if (k >= SKF_AD_OFF)
61 return NULL;
62 return __load_pointer(skb, k);
66 /**
67 * sk_run_filter - run a filter on a socket
68 * @skb: buffer to run the filter on
69 * @filter: filter to apply
70 * @flen: length of filter
72 * Decode and apply filter instructions to the skb->data.
73 * Return length to keep, 0 for none. skb is the data we are
74 * filtering, filter is the array of filter instructions, and
75 * len is the number of filter blocks in the array.
77 unsigned int sk_run_filter(struct sk_buff *skb, struct sock_filter *filter, int flen)
79 struct sock_filter *fentry; /* We walk down these */
80 void *ptr;
81 u32 A = 0; /* Accumulator */
82 u32 X = 0; /* Index Register */
83 u32 mem[BPF_MEMWORDS]; /* Scratch Memory Store */
84 u32 tmp;
85 int k;
86 int pc;
89 * Process array of filter instructions.
91 for (pc = 0; pc < flen; pc++) {
92 fentry = &filter[pc];
94 switch (fentry->code) {
95 case BPF_S_ALU_ADD_X:
96 A += X;
97 continue;
98 case BPF_S_ALU_ADD_K:
99 A += fentry->k;
100 continue;
101 case BPF_S_ALU_SUB_X:
102 A -= X;
103 continue;
104 case BPF_S_ALU_SUB_K:
105 A -= fentry->k;
106 continue;
107 case BPF_S_ALU_MUL_X:
108 A *= X;
109 continue;
110 case BPF_S_ALU_MUL_K:
111 A *= fentry->k;
112 continue;
113 case BPF_S_ALU_DIV_X:
114 if (X == 0)
115 return 0;
116 A /= X;
117 continue;
118 case BPF_S_ALU_DIV_K:
119 A /= fentry->k;
120 continue;
121 case BPF_S_ALU_AND_X:
122 A &= X;
123 continue;
124 case BPF_S_ALU_AND_K:
125 A &= fentry->k;
126 continue;
127 case BPF_S_ALU_OR_X:
128 A |= X;
129 continue;
130 case BPF_S_ALU_OR_K:
131 A |= fentry->k;
132 continue;
133 case BPF_S_ALU_LSH_X:
134 A <<= X;
135 continue;
136 case BPF_S_ALU_LSH_K:
137 A <<= fentry->k;
138 continue;
139 case BPF_S_ALU_RSH_X:
140 A >>= X;
141 continue;
142 case BPF_S_ALU_RSH_K:
143 A >>= fentry->k;
144 continue;
145 case BPF_S_ALU_NEG:
146 A = -A;
147 continue;
148 case BPF_S_JMP_JA:
149 pc += fentry->k;
150 continue;
151 case BPF_S_JMP_JGT_K:
152 pc += (A > fentry->k) ? fentry->jt : fentry->jf;
153 continue;
154 case BPF_S_JMP_JGE_K:
155 pc += (A >= fentry->k) ? fentry->jt : fentry->jf;
156 continue;
157 case BPF_S_JMP_JEQ_K:
158 pc += (A == fentry->k) ? fentry->jt : fentry->jf;
159 continue;
160 case BPF_S_JMP_JSET_K:
161 pc += (A & fentry->k) ? fentry->jt : fentry->jf;
162 continue;
163 case BPF_S_JMP_JGT_X:
164 pc += (A > X) ? fentry->jt : fentry->jf;
165 continue;
166 case BPF_S_JMP_JGE_X:
167 pc += (A >= X) ? fentry->jt : fentry->jf;
168 continue;
169 case BPF_S_JMP_JEQ_X:
170 pc += (A == X) ? fentry->jt : fentry->jf;
171 continue;
172 case BPF_S_JMP_JSET_X:
173 pc += (A & X) ? fentry->jt : fentry->jf;
174 continue;
175 case BPF_S_LD_W_ABS:
176 k = fentry->k;
177 load_w:
178 ptr = load_pointer(skb, k, 4, &tmp);
179 if (ptr != NULL) {
180 A = ntohl(get_unaligned((__be32 *)ptr));
181 continue;
183 break;
184 case BPF_S_LD_H_ABS:
185 k = fentry->k;
186 load_h:
187 ptr = load_pointer(skb, k, 2, &tmp);
188 if (ptr != NULL) {
189 A = ntohs(get_unaligned((__be16 *)ptr));
190 continue;
192 break;
193 case BPF_S_LD_B_ABS:
194 k = fentry->k;
195 load_b:
196 ptr = load_pointer(skb, k, 1, &tmp);
197 if (ptr != NULL) {
198 A = *(u8 *)ptr;
199 continue;
201 break;
202 case BPF_S_LD_W_LEN:
203 A = skb->len;
204 continue;
205 case BPF_S_LDX_W_LEN:
206 X = skb->len;
207 continue;
208 case BPF_S_LD_W_IND:
209 k = X + fentry->k;
210 goto load_w;
211 case BPF_S_LD_H_IND:
212 k = X + fentry->k;
213 goto load_h;
214 case BPF_S_LD_B_IND:
215 k = X + fentry->k;
216 goto load_b;
217 case BPF_S_LDX_B_MSH:
218 ptr = load_pointer(skb, fentry->k, 1, &tmp);
219 if (ptr != NULL) {
220 X = (*(u8 *)ptr & 0xf) << 2;
221 continue;
223 return 0;
224 case BPF_S_LD_IMM:
225 A = fentry->k;
226 continue;
227 case BPF_S_LDX_IMM:
228 X = fentry->k;
229 continue;
230 case BPF_S_LD_MEM:
231 A = mem[fentry->k];
232 continue;
233 case BPF_S_LDX_MEM:
234 X = mem[fentry->k];
235 continue;
236 case BPF_S_MISC_TAX:
237 X = A;
238 continue;
239 case BPF_S_MISC_TXA:
240 A = X;
241 continue;
242 case BPF_S_RET_K:
243 return fentry->k;
244 case BPF_S_RET_A:
245 return A;
246 case BPF_S_ST:
247 mem[fentry->k] = A;
248 continue;
249 case BPF_S_STX:
250 mem[fentry->k] = X;
251 continue;
252 default:
253 WARN_ON(1);
254 return 0;
258 * Handle ancillary data, which are impossible
259 * (or very difficult) to get parsing packet contents.
261 switch (k-SKF_AD_OFF) {
262 case SKF_AD_PROTOCOL:
263 A = ntohs(skb->protocol);
264 continue;
265 case SKF_AD_PKTTYPE:
266 A = skb->pkt_type;
267 continue;
268 case SKF_AD_IFINDEX:
269 A = skb->dev->ifindex;
270 continue;
271 default:
272 return 0;
276 return 0;
280 * sk_chk_filter - verify socket filter code
281 * @filter: filter to verify
282 * @flen: length of filter
284 * Check the user's filter code. If we let some ugly
285 * filter code slip through kaboom! The filter must contain
286 * no references or jumps that are out of range, no illegal
287 * instructions, and must end with a RET instruction.
289 * All jumps are forward as they are not signed.
291 * Returns 0 if the rule set is legal or -EINVAL if not.
293 int sk_chk_filter(struct sock_filter *filter, int flen)
295 struct sock_filter *ftest;
296 int pc;
298 if (flen == 0 || flen > BPF_MAXINSNS)
299 return -EINVAL;
301 /* check the filter code now */
302 for (pc = 0; pc < flen; pc++) {
303 ftest = &filter[pc];
305 /* Only allow valid instructions */
306 switch (ftest->code) {
307 case BPF_ALU|BPF_ADD|BPF_K:
308 ftest->code = BPF_S_ALU_ADD_K;
309 break;
310 case BPF_ALU|BPF_ADD|BPF_X:
311 ftest->code = BPF_S_ALU_ADD_X;
312 break;
313 case BPF_ALU|BPF_SUB|BPF_K:
314 ftest->code = BPF_S_ALU_SUB_K;
315 break;
316 case BPF_ALU|BPF_SUB|BPF_X:
317 ftest->code = BPF_S_ALU_SUB_X;
318 break;
319 case BPF_ALU|BPF_MUL|BPF_K:
320 ftest->code = BPF_S_ALU_MUL_K;
321 break;
322 case BPF_ALU|BPF_MUL|BPF_X:
323 ftest->code = BPF_S_ALU_MUL_X;
324 break;
325 case BPF_ALU|BPF_DIV|BPF_X:
326 ftest->code = BPF_S_ALU_DIV_X;
327 break;
328 case BPF_ALU|BPF_AND|BPF_K:
329 ftest->code = BPF_S_ALU_AND_K;
330 break;
331 case BPF_ALU|BPF_AND|BPF_X:
332 ftest->code = BPF_S_ALU_AND_X;
333 break;
334 case BPF_ALU|BPF_OR|BPF_K:
335 ftest->code = BPF_S_ALU_OR_K;
336 break;
337 case BPF_ALU|BPF_OR|BPF_X:
338 ftest->code = BPF_S_ALU_OR_X;
339 break;
340 case BPF_ALU|BPF_LSH|BPF_K:
341 ftest->code = BPF_S_ALU_LSH_K;
342 break;
343 case BPF_ALU|BPF_LSH|BPF_X:
344 ftest->code = BPF_S_ALU_LSH_X;
345 break;
346 case BPF_ALU|BPF_RSH|BPF_K:
347 ftest->code = BPF_S_ALU_RSH_K;
348 break;
349 case BPF_ALU|BPF_RSH|BPF_X:
350 ftest->code = BPF_S_ALU_RSH_X;
351 break;
352 case BPF_ALU|BPF_NEG:
353 ftest->code = BPF_S_ALU_NEG;
354 break;
355 case BPF_LD|BPF_W|BPF_ABS:
356 ftest->code = BPF_S_LD_W_ABS;
357 break;
358 case BPF_LD|BPF_H|BPF_ABS:
359 ftest->code = BPF_S_LD_H_ABS;
360 break;
361 case BPF_LD|BPF_B|BPF_ABS:
362 ftest->code = BPF_S_LD_B_ABS;
363 break;
364 case BPF_LD|BPF_W|BPF_LEN:
365 ftest->code = BPF_S_LD_W_LEN;
366 break;
367 case BPF_LD|BPF_W|BPF_IND:
368 ftest->code = BPF_S_LD_W_IND;
369 break;
370 case BPF_LD|BPF_H|BPF_IND:
371 ftest->code = BPF_S_LD_H_IND;
372 break;
373 case BPF_LD|BPF_B|BPF_IND:
374 ftest->code = BPF_S_LD_B_IND;
375 break;
376 case BPF_LD|BPF_IMM:
377 ftest->code = BPF_S_LD_IMM;
378 break;
379 case BPF_LDX|BPF_W|BPF_LEN:
380 ftest->code = BPF_S_LDX_W_LEN;
381 break;
382 case BPF_LDX|BPF_B|BPF_MSH:
383 ftest->code = BPF_S_LDX_B_MSH;
384 break;
385 case BPF_LDX|BPF_IMM:
386 ftest->code = BPF_S_LDX_IMM;
387 break;
388 case BPF_MISC|BPF_TAX:
389 ftest->code = BPF_S_MISC_TAX;
390 break;
391 case BPF_MISC|BPF_TXA:
392 ftest->code = BPF_S_MISC_TXA;
393 break;
394 case BPF_RET|BPF_K:
395 ftest->code = BPF_S_RET_K;
396 break;
397 case BPF_RET|BPF_A:
398 ftest->code = BPF_S_RET_A;
399 break;
401 /* Some instructions need special checks */
403 /* check for division by zero */
404 case BPF_ALU|BPF_DIV|BPF_K:
405 if (ftest->k == 0)
406 return -EINVAL;
407 ftest->code = BPF_S_ALU_DIV_K;
408 break;
410 /* check for invalid memory addresses */
411 case BPF_LD|BPF_MEM:
412 if (ftest->k >= BPF_MEMWORDS)
413 return -EINVAL;
414 ftest->code = BPF_S_LD_MEM;
415 break;
416 case BPF_LDX|BPF_MEM:
417 if (ftest->k >= BPF_MEMWORDS)
418 return -EINVAL;
419 ftest->code = BPF_S_LDX_MEM;
420 break;
421 case BPF_ST:
422 if (ftest->k >= BPF_MEMWORDS)
423 return -EINVAL;
424 ftest->code = BPF_S_ST;
425 break;
426 case BPF_STX:
427 if (ftest->k >= BPF_MEMWORDS)
428 return -EINVAL;
429 ftest->code = BPF_S_STX;
430 break;
432 case BPF_JMP|BPF_JA:
434 * Note, the large ftest->k might cause loops.
435 * Compare this with conditional jumps below,
436 * where offsets are limited. --ANK (981016)
438 if (ftest->k >= (unsigned)(flen-pc-1))
439 return -EINVAL;
440 ftest->code = BPF_S_JMP_JA;
441 break;
443 case BPF_JMP|BPF_JEQ|BPF_K:
444 ftest->code = BPF_S_JMP_JEQ_K;
445 break;
446 case BPF_JMP|BPF_JEQ|BPF_X:
447 ftest->code = BPF_S_JMP_JEQ_X;
448 break;
449 case BPF_JMP|BPF_JGE|BPF_K:
450 ftest->code = BPF_S_JMP_JGE_K;
451 break;
452 case BPF_JMP|BPF_JGE|BPF_X:
453 ftest->code = BPF_S_JMP_JGE_X;
454 break;
455 case BPF_JMP|BPF_JGT|BPF_K:
456 ftest->code = BPF_S_JMP_JGT_K;
457 break;
458 case BPF_JMP|BPF_JGT|BPF_X:
459 ftest->code = BPF_S_JMP_JGT_X;
460 break;
461 case BPF_JMP|BPF_JSET|BPF_K:
462 ftest->code = BPF_S_JMP_JSET_K;
463 break;
464 case BPF_JMP|BPF_JSET|BPF_X:
465 ftest->code = BPF_S_JMP_JSET_X;
466 break;
468 default:
469 return -EINVAL;
472 /* for conditionals both must be safe */
473 switch (ftest->code) {
474 case BPF_S_JMP_JEQ_K:
475 case BPF_S_JMP_JEQ_X:
476 case BPF_S_JMP_JGE_K:
477 case BPF_S_JMP_JGE_X:
478 case BPF_S_JMP_JGT_K:
479 case BPF_S_JMP_JGT_X:
480 case BPF_S_JMP_JSET_X:
481 case BPF_S_JMP_JSET_K:
482 if (pc + ftest->jt + 1 >= flen ||
483 pc + ftest->jf + 1 >= flen)
484 return -EINVAL;
488 /* last instruction must be a RET code */
489 switch (filter[flen - 1].code) {
490 case BPF_S_RET_K:
491 case BPF_S_RET_A:
492 return 0;
493 break;
494 default:
495 return -EINVAL;
500 * sk_attach_filter - attach a socket filter
501 * @fprog: the filter program
502 * @sk: the socket to use
504 * Attach the user's filter code. We first run some sanity checks on
505 * it to make sure it does not explode on us later. If an error
506 * occurs or there is insufficient memory for the filter a negative
507 * errno code is returned. On success the return is zero.
509 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
511 struct sk_filter *fp;
512 unsigned int fsize = sizeof(struct sock_filter) * fprog->len;
513 int err;
515 /* Make sure new filter is there and in the right amounts. */
516 if (fprog->filter == NULL)
517 return -EINVAL;
519 fp = sock_kmalloc(sk, fsize+sizeof(*fp), GFP_KERNEL);
520 if (!fp)
521 return -ENOMEM;
522 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
523 sock_kfree_s(sk, fp, fsize+sizeof(*fp));
524 return -EFAULT;
527 atomic_set(&fp->refcnt, 1);
528 fp->len = fprog->len;
530 err = sk_chk_filter(fp->insns, fp->len);
531 if (!err) {
532 struct sk_filter *old_fp;
534 rcu_read_lock_bh();
535 old_fp = rcu_dereference(sk->sk_filter);
536 rcu_assign_pointer(sk->sk_filter, fp);
537 rcu_read_unlock_bh();
538 fp = old_fp;
541 if (fp)
542 sk_filter_release(sk, fp);
543 return err;
546 EXPORT_SYMBOL(sk_chk_filter);
547 EXPORT_SYMBOL(sk_run_filter);