Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Linux Socket Filter - Kernel level socket filtering | |
3 | * | |
4 | * Author: | |
5 | * Jay Schulist <jschlst@samba.org> | |
6 | * | |
7 | * Based on the design of: | |
8 | * - The Berkeley Packet Filter | |
9 | * | |
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. | |
14 | * | |
15 | * Andi Kleen - Fix a few bad bugs and races. | |
16 | */ | |
17 | ||
18 | #include <linux/module.h> | |
19 | #include <linux/types.h> | |
20 | #include <linux/sched.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 <linux/filter.h> | |
37 | ||
38 | /* No hurry in this branch */ | |
0b05b2a4 | 39 | static void *__load_pointer(struct sk_buff *skb, int k) |
1da177e4 LT |
40 | { |
41 | u8 *ptr = NULL; | |
42 | ||
43 | if (k >= SKF_NET_OFF) | |
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; | |
47 | ||
48 | if (ptr >= skb->head && ptr < skb->tail) | |
49 | return ptr; | |
50 | return NULL; | |
51 | } | |
52 | ||
0b05b2a4 PM |
53 | static inline void *load_pointer(struct sk_buff *skb, int k, |
54 | unsigned int size, void *buffer) | |
55 | { | |
56 | if (k >= 0) | |
57 | return skb_header_pointer(skb, k, size, buffer); | |
58 | else { | |
59 | if (k >= SKF_AD_OFF) | |
60 | return NULL; | |
61 | return __load_pointer(skb, k); | |
62 | } | |
63 | } | |
64 | ||
1da177e4 LT |
65 | /** |
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 | |
70 | * | |
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. | |
75 | */ | |
76 | ||
77 | int sk_run_filter(struct sk_buff *skb, struct sock_filter *filter, int flen) | |
78 | { | |
1da177e4 LT |
79 | /* len is UNSIGNED. Byte wide insns relies only on implicit |
80 | type casts to prevent reading arbitrary memory locations. | |
81 | */ | |
82 | unsigned int len = skb->len-skb->data_len; | |
83 | struct sock_filter *fentry; /* We walk down these */ | |
0b05b2a4 | 84 | void *ptr; |
1da177e4 LT |
85 | u32 A = 0; /* Accumulator */ |
86 | u32 X = 0; /* Index Register */ | |
87 | u32 mem[BPF_MEMWORDS]; /* Scratch Memory Store */ | |
0b05b2a4 | 88 | u32 tmp; |
1da177e4 LT |
89 | int k; |
90 | int pc; | |
91 | ||
92 | /* | |
93 | * Process array of filter instructions. | |
94 | */ | |
95 | for (pc = 0; pc < flen; pc++) { | |
96 | fentry = &filter[pc]; | |
97 | ||
98 | switch (fentry->code) { | |
99 | case BPF_ALU|BPF_ADD|BPF_X: | |
100 | A += X; | |
101 | continue; | |
102 | case BPF_ALU|BPF_ADD|BPF_K: | |
103 | A += fentry->k; | |
104 | continue; | |
105 | case BPF_ALU|BPF_SUB|BPF_X: | |
106 | A -= X; | |
107 | continue; | |
108 | case BPF_ALU|BPF_SUB|BPF_K: | |
109 | A -= fentry->k; | |
110 | continue; | |
111 | case BPF_ALU|BPF_MUL|BPF_X: | |
112 | A *= X; | |
113 | continue; | |
114 | case BPF_ALU|BPF_MUL|BPF_K: | |
115 | A *= fentry->k; | |
116 | continue; | |
117 | case BPF_ALU|BPF_DIV|BPF_X: | |
118 | if (X == 0) | |
119 | return 0; | |
120 | A /= X; | |
121 | continue; | |
122 | case BPF_ALU|BPF_DIV|BPF_K: | |
123 | if (fentry->k == 0) | |
124 | return 0; | |
125 | A /= fentry->k; | |
126 | continue; | |
127 | case BPF_ALU|BPF_AND|BPF_X: | |
128 | A &= X; | |
129 | continue; | |
130 | case BPF_ALU|BPF_AND|BPF_K: | |
131 | A &= fentry->k; | |
132 | continue; | |
133 | case BPF_ALU|BPF_OR|BPF_X: | |
134 | A |= X; | |
135 | continue; | |
136 | case BPF_ALU|BPF_OR|BPF_K: | |
137 | A |= fentry->k; | |
138 | continue; | |
139 | case BPF_ALU|BPF_LSH|BPF_X: | |
140 | A <<= X; | |
141 | continue; | |
142 | case BPF_ALU|BPF_LSH|BPF_K: | |
143 | A <<= fentry->k; | |
144 | continue; | |
145 | case BPF_ALU|BPF_RSH|BPF_X: | |
146 | A >>= X; | |
147 | continue; | |
148 | case BPF_ALU|BPF_RSH|BPF_K: | |
149 | A >>= fentry->k; | |
150 | continue; | |
151 | case BPF_ALU|BPF_NEG: | |
152 | A = -A; | |
153 | continue; | |
154 | case BPF_JMP|BPF_JA: | |
155 | pc += fentry->k; | |
156 | continue; | |
157 | case BPF_JMP|BPF_JGT|BPF_K: | |
158 | pc += (A > fentry->k) ? fentry->jt : fentry->jf; | |
159 | continue; | |
160 | case BPF_JMP|BPF_JGE|BPF_K: | |
161 | pc += (A >= fentry->k) ? fentry->jt : fentry->jf; | |
162 | continue; | |
163 | case BPF_JMP|BPF_JEQ|BPF_K: | |
164 | pc += (A == fentry->k) ? fentry->jt : fentry->jf; | |
165 | continue; | |
166 | case BPF_JMP|BPF_JSET|BPF_K: | |
167 | pc += (A & fentry->k) ? fentry->jt : fentry->jf; | |
168 | continue; | |
169 | case BPF_JMP|BPF_JGT|BPF_X: | |
170 | pc += (A > X) ? fentry->jt : fentry->jf; | |
171 | continue; | |
172 | case BPF_JMP|BPF_JGE|BPF_X: | |
173 | pc += (A >= X) ? fentry->jt : fentry->jf; | |
174 | continue; | |
175 | case BPF_JMP|BPF_JEQ|BPF_X: | |
176 | pc += (A == X) ? fentry->jt : fentry->jf; | |
177 | continue; | |
178 | case BPF_JMP|BPF_JSET|BPF_X: | |
179 | pc += (A & X) ? fentry->jt : fentry->jf; | |
180 | continue; | |
181 | case BPF_LD|BPF_W|BPF_ABS: | |
182 | k = fentry->k; | |
183 | load_w: | |
0b05b2a4 PM |
184 | ptr = load_pointer(skb, k, 4, &tmp); |
185 | if (ptr != NULL) { | |
186 | A = ntohl(*(u32 *)ptr); | |
187 | continue; | |
1da177e4 LT |
188 | } |
189 | return 0; | |
190 | case BPF_LD|BPF_H|BPF_ABS: | |
191 | k = fentry->k; | |
192 | load_h: | |
0b05b2a4 PM |
193 | ptr = load_pointer(skb, k, 2, &tmp); |
194 | if (ptr != NULL) { | |
195 | A = ntohs(*(u16 *)ptr); | |
196 | continue; | |
1da177e4 LT |
197 | } |
198 | return 0; | |
199 | case BPF_LD|BPF_B|BPF_ABS: | |
200 | k = fentry->k; | |
201 | load_b: | |
0b05b2a4 PM |
202 | ptr = load_pointer(skb, k, 1, &tmp); |
203 | if (ptr != NULL) { | |
204 | A = *(u8 *)ptr; | |
205 | continue; | |
1da177e4 LT |
206 | } |
207 | return 0; | |
208 | case BPF_LD|BPF_W|BPF_LEN: | |
209 | A = len; | |
210 | continue; | |
211 | case BPF_LDX|BPF_W|BPF_LEN: | |
212 | X = len; | |
213 | continue; | |
214 | case BPF_LD|BPF_W|BPF_IND: | |
215 | k = X + fentry->k; | |
216 | goto load_w; | |
217 | case BPF_LD|BPF_H|BPF_IND: | |
218 | k = X + fentry->k; | |
219 | goto load_h; | |
220 | case BPF_LD|BPF_B|BPF_IND: | |
221 | k = X + fentry->k; | |
222 | goto load_b; | |
223 | case BPF_LDX|BPF_B|BPF_MSH: | |
0b05b2a4 PM |
224 | ptr = load_pointer(skb, fentry->k, 1, &tmp); |
225 | if (ptr != NULL) { | |
226 | X = (*(u8 *)ptr & 0xf) << 2; | |
227 | continue; | |
228 | } | |
229 | return 0; | |
1da177e4 LT |
230 | case BPF_LD|BPF_IMM: |
231 | A = fentry->k; | |
232 | continue; | |
233 | case BPF_LDX|BPF_IMM: | |
234 | X = fentry->k; | |
235 | continue; | |
236 | case BPF_LD|BPF_MEM: | |
237 | A = mem[fentry->k]; | |
238 | continue; | |
239 | case BPF_LDX|BPF_MEM: | |
240 | X = mem[fentry->k]; | |
241 | continue; | |
242 | case BPF_MISC|BPF_TAX: | |
243 | X = A; | |
244 | continue; | |
245 | case BPF_MISC|BPF_TXA: | |
246 | A = X; | |
247 | continue; | |
248 | case BPF_RET|BPF_K: | |
249 | return ((unsigned int)fentry->k); | |
250 | case BPF_RET|BPF_A: | |
251 | return ((unsigned int)A); | |
252 | case BPF_ST: | |
253 | mem[fentry->k] = A; | |
254 | continue; | |
255 | case BPF_STX: | |
256 | mem[fentry->k] = X; | |
257 | continue; | |
258 | default: | |
259 | /* Invalid instruction counts as RET */ | |
260 | return 0; | |
261 | } | |
262 | ||
263 | /* | |
264 | * Handle ancillary data, which are impossible | |
265 | * (or very difficult) to get parsing packet contents. | |
266 | */ | |
267 | switch (k-SKF_AD_OFF) { | |
268 | case SKF_AD_PROTOCOL: | |
269 | A = htons(skb->protocol); | |
270 | continue; | |
271 | case SKF_AD_PKTTYPE: | |
272 | A = skb->pkt_type; | |
273 | continue; | |
274 | case SKF_AD_IFINDEX: | |
275 | A = skb->dev->ifindex; | |
276 | continue; | |
277 | default: | |
278 | return 0; | |
279 | } | |
280 | } | |
281 | ||
282 | return 0; | |
283 | } | |
284 | ||
285 | /** | |
286 | * sk_chk_filter - verify socket filter code | |
287 | * @filter: filter to verify | |
288 | * @flen: length of filter | |
289 | * | |
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 | |
294 | * | |
295 | * Returns 0 if the rule set is legal or a negative errno code if not. | |
296 | */ | |
297 | int sk_chk_filter(struct sock_filter *filter, int flen) | |
298 | { | |
299 | struct sock_filter *ftest; | |
300 | int pc; | |
301 | ||
302 | if (((unsigned int)flen >= (~0U / sizeof(struct sock_filter))) || flen == 0) | |
303 | return -EINVAL; | |
304 | ||
305 | /* check the filter code now */ | |
306 | for (pc = 0; pc < flen; pc++) { | |
307 | /* all jumps are forward as they are not signed */ | |
308 | ftest = &filter[pc]; | |
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) { | |
312 | /* | |
313 | * Note, the large ftest->k might cause loops. | |
314 | * Compare this with conditional jumps below, | |
315 | * where offsets are limited. --ANK (981016) | |
316 | */ | |
317 | if (ftest->k >= (unsigned)(flen-pc-1)) | |
318 | return -EINVAL; | |
319 | } else { | |
320 | /* for conditionals both must be safe */ | |
321 | if (pc + ftest->jt +1 >= flen || | |
322 | pc + ftest->jf +1 >= flen) | |
323 | return -EINVAL; | |
324 | } | |
325 | } | |
326 | ||
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) { | |
331 | case BPF_ST: | |
332 | case BPF_STX: | |
333 | case BPF_LD|BPF_MEM: | |
334 | case BPF_LDX|BPF_MEM: | |
335 | return -EINVAL; | |
336 | } | |
337 | } | |
338 | } | |
339 | ||
340 | /* | |
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. | |
344 | */ | |
345 | return (BPF_CLASS(filter[flen - 1].code) == BPF_RET) ? 0 : -EINVAL; | |
346 | } | |
347 | ||
348 | /** | |
349 | * sk_attach_filter - attach a socket filter | |
350 | * @fprog: the filter program | |
351 | * @sk: the socket to use | |
352 | * | |
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. | |
357 | */ | |
358 | int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk) | |
359 | { | |
360 | struct sk_filter *fp; | |
361 | unsigned int fsize = sizeof(struct sock_filter) * fprog->len; | |
362 | int err; | |
363 | ||
364 | /* Make sure new filter is there and in the right amounts. */ | |
365 | if (fprog->filter == NULL || fprog->len > BPF_MAXINSNS) | |
366 | return -EINVAL; | |
367 | ||
368 | fp = sock_kmalloc(sk, fsize+sizeof(*fp), GFP_KERNEL); | |
369 | if (!fp) | |
370 | return -ENOMEM; | |
371 | if (copy_from_user(fp->insns, fprog->filter, fsize)) { | |
372 | sock_kfree_s(sk, fp, fsize+sizeof(*fp)); | |
373 | return -EFAULT; | |
374 | } | |
375 | ||
376 | atomic_set(&fp->refcnt, 1); | |
377 | fp->len = fprog->len; | |
378 | ||
379 | err = sk_chk_filter(fp->insns, fp->len); | |
380 | if (!err) { | |
381 | struct sk_filter *old_fp; | |
382 | ||
383 | spin_lock_bh(&sk->sk_lock.slock); | |
384 | old_fp = sk->sk_filter; | |
385 | sk->sk_filter = fp; | |
386 | spin_unlock_bh(&sk->sk_lock.slock); | |
387 | fp = old_fp; | |
388 | } | |
389 | ||
390 | if (fp) | |
391 | sk_filter_release(sk, fp); | |
392 | return err; | |
393 | } | |
394 | ||
395 | EXPORT_SYMBOL(sk_chk_filter); | |
396 | EXPORT_SYMBOL(sk_run_filter); |