Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Linux Socket Filter - Kernel level socket filtering | |
3 | * | |
bd4cf0ed AS |
4 | * Based on the design of the Berkeley Packet Filter. The new |
5 | * internal format has been designed by PLUMgrid: | |
1da177e4 | 6 | * |
bd4cf0ed AS |
7 | * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com |
8 | * | |
9 | * Authors: | |
10 | * | |
11 | * Jay Schulist <jschlst@samba.org> | |
12 | * Alexei Starovoitov <ast@plumgrid.com> | |
13 | * Daniel Borkmann <dborkman@redhat.com> | |
1da177e4 LT |
14 | * |
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. | |
19 | * | |
20 | * Andi Kleen - Fix a few bad bugs and races. | |
93699863 | 21 | * Kris Katterjohn - Added many additional checks in sk_chk_filter() |
1da177e4 LT |
22 | */ |
23 | ||
24 | #include <linux/module.h> | |
25 | #include <linux/types.h> | |
1da177e4 LT |
26 | #include <linux/mm.h> |
27 | #include <linux/fcntl.h> | |
28 | #include <linux/socket.h> | |
29 | #include <linux/in.h> | |
30 | #include <linux/inet.h> | |
31 | #include <linux/netdevice.h> | |
32 | #include <linux/if_packet.h> | |
5a0e3ad6 | 33 | #include <linux/gfp.h> |
1da177e4 LT |
34 | #include <net/ip.h> |
35 | #include <net/protocol.h> | |
4738c1db | 36 | #include <net/netlink.h> |
1da177e4 LT |
37 | #include <linux/skbuff.h> |
38 | #include <net/sock.h> | |
39 | #include <linux/errno.h> | |
40 | #include <linux/timer.h> | |
1da177e4 | 41 | #include <asm/uaccess.h> |
40daafc8 | 42 | #include <asm/unaligned.h> |
1da177e4 | 43 | #include <linux/filter.h> |
86e4ca66 | 44 | #include <linux/ratelimit.h> |
46b325c7 | 45 | #include <linux/seccomp.h> |
f3335031 | 46 | #include <linux/if_vlan.h> |
1da177e4 | 47 | |
f03fb3f4 JS |
48 | /* No hurry in this branch |
49 | * | |
50 | * Exported for the bpf jit load helper. | |
51 | */ | |
52 | void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, int k, unsigned int size) | |
1da177e4 LT |
53 | { |
54 | u8 *ptr = NULL; | |
55 | ||
56 | if (k >= SKF_NET_OFF) | |
d56f90a7 | 57 | ptr = skb_network_header(skb) + k - SKF_NET_OFF; |
1da177e4 | 58 | else if (k >= SKF_LL_OFF) |
98e399f8 | 59 | ptr = skb_mac_header(skb) + k - SKF_LL_OFF; |
1da177e4 | 60 | |
4bc65dd8 | 61 | if (ptr >= skb->head && ptr + size <= skb_tail_pointer(skb)) |
1da177e4 LT |
62 | return ptr; |
63 | return NULL; | |
64 | } | |
65 | ||
62ab0812 | 66 | static inline void *load_pointer(const struct sk_buff *skb, int k, |
4ec93edb | 67 | unsigned int size, void *buffer) |
0b05b2a4 PM |
68 | { |
69 | if (k >= 0) | |
70 | return skb_header_pointer(skb, k, size, buffer); | |
f03fb3f4 | 71 | return bpf_internal_load_pointer_neg_helper(skb, k, size); |
0b05b2a4 PM |
72 | } |
73 | ||
43db6d65 SH |
74 | /** |
75 | * sk_filter - run a packet through a socket filter | |
76 | * @sk: sock associated with &sk_buff | |
77 | * @skb: buffer to filter | |
43db6d65 SH |
78 | * |
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. | |
84 | * | |
85 | */ | |
86 | int sk_filter(struct sock *sk, struct sk_buff *skb) | |
87 | { | |
88 | int err; | |
89 | struct sk_filter *filter; | |
90 | ||
c93bdd0e MG |
91 | /* |
92 | * If the skb was allocated from pfmemalloc reserves, only | |
93 | * allow SOCK_MEMALLOC sockets to use it as this socket is | |
94 | * helping free memory | |
95 | */ | |
96 | if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) | |
97 | return -ENOMEM; | |
98 | ||
43db6d65 SH |
99 | err = security_sock_rcv_skb(sk, skb); |
100 | if (err) | |
101 | return err; | |
102 | ||
80f8f102 ED |
103 | rcu_read_lock(); |
104 | filter = rcu_dereference(sk->sk_filter); | |
43db6d65 | 105 | if (filter) { |
0a14842f | 106 | unsigned int pkt_len = SK_RUN_FILTER(filter, skb); |
0d7da9dd | 107 | |
43db6d65 SH |
108 | err = pkt_len ? pskb_trim(skb, pkt_len) : -EPERM; |
109 | } | |
80f8f102 | 110 | rcu_read_unlock(); |
43db6d65 SH |
111 | |
112 | return err; | |
113 | } | |
114 | EXPORT_SYMBOL(sk_filter); | |
115 | ||
bd4cf0ed AS |
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. | |
119 | */ | |
120 | noinline u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) | |
121 | { | |
122 | return 0; | |
123 | } | |
124 | ||
1da177e4 | 125 | /** |
bd4cf0ed AS |
126 | * __sk_run_filter - run a filter on a given context |
127 | * @ctx: buffer to run the filter on | |
01d32f6e | 128 | * @insn: filter to apply |
1da177e4 | 129 | * |
bd4cf0ed | 130 | * Decode and apply filter instructions to the skb->data. Return length to |
01d32f6e | 131 | * keep, 0 for none. @ctx is the data we are operating on, @insn is the |
bd4cf0ed | 132 | * array of filter instructions. |
1da177e4 | 133 | */ |
bd4cf0ed | 134 | unsigned int __sk_run_filter(void *ctx, const struct sock_filter_int *insn) |
1da177e4 | 135 | { |
bd4cf0ed AS |
136 | u64 stack[MAX_BPF_STACK / sizeof(u64)]; |
137 | u64 regs[MAX_BPF_REG], tmp; | |
0b05b2a4 | 138 | void *ptr; |
bd4cf0ed AS |
139 | int off; |
140 | ||
141 | #define K insn->imm | |
142 | #define A regs[insn->a_reg] | |
143 | #define X regs[insn->x_reg] | |
144 | #define R0 regs[0] | |
145 | ||
146 | #define CONT ({insn++; goto select_insn; }) | |
147 | #define CONT_JMP ({insn++; goto select_insn; }) | |
148 | ||
149 | static const void *jumptable[256] = { | |
150 | [0 ... 255] = &&default_label, | |
151 | /* Now overwrite non-defaults ... */ | |
152 | #define DL(A, B, C) [A|B|C] = &&A##_##B##_##C | |
153 | DL(BPF_ALU, BPF_ADD, BPF_X), | |
154 | DL(BPF_ALU, BPF_ADD, BPF_K), | |
155 | DL(BPF_ALU, BPF_SUB, BPF_X), | |
156 | DL(BPF_ALU, BPF_SUB, BPF_K), | |
157 | DL(BPF_ALU, BPF_AND, BPF_X), | |
158 | DL(BPF_ALU, BPF_AND, BPF_K), | |
159 | DL(BPF_ALU, BPF_OR, BPF_X), | |
160 | DL(BPF_ALU, BPF_OR, BPF_K), | |
161 | DL(BPF_ALU, BPF_LSH, BPF_X), | |
162 | DL(BPF_ALU, BPF_LSH, BPF_K), | |
163 | DL(BPF_ALU, BPF_RSH, BPF_X), | |
164 | DL(BPF_ALU, BPF_RSH, BPF_K), | |
165 | DL(BPF_ALU, BPF_XOR, BPF_X), | |
166 | DL(BPF_ALU, BPF_XOR, BPF_K), | |
167 | DL(BPF_ALU, BPF_MUL, BPF_X), | |
168 | DL(BPF_ALU, BPF_MUL, BPF_K), | |
169 | DL(BPF_ALU, BPF_MOV, BPF_X), | |
170 | DL(BPF_ALU, BPF_MOV, BPF_K), | |
171 | DL(BPF_ALU, BPF_DIV, BPF_X), | |
172 | DL(BPF_ALU, BPF_DIV, BPF_K), | |
173 | DL(BPF_ALU, BPF_MOD, BPF_X), | |
174 | DL(BPF_ALU, BPF_MOD, BPF_K), | |
175 | DL(BPF_ALU, BPF_NEG, 0), | |
176 | DL(BPF_ALU, BPF_END, BPF_TO_BE), | |
177 | DL(BPF_ALU, BPF_END, BPF_TO_LE), | |
178 | DL(BPF_ALU64, BPF_ADD, BPF_X), | |
179 | DL(BPF_ALU64, BPF_ADD, BPF_K), | |
180 | DL(BPF_ALU64, BPF_SUB, BPF_X), | |
181 | DL(BPF_ALU64, BPF_SUB, BPF_K), | |
182 | DL(BPF_ALU64, BPF_AND, BPF_X), | |
183 | DL(BPF_ALU64, BPF_AND, BPF_K), | |
184 | DL(BPF_ALU64, BPF_OR, BPF_X), | |
185 | DL(BPF_ALU64, BPF_OR, BPF_K), | |
186 | DL(BPF_ALU64, BPF_LSH, BPF_X), | |
187 | DL(BPF_ALU64, BPF_LSH, BPF_K), | |
188 | DL(BPF_ALU64, BPF_RSH, BPF_X), | |
189 | DL(BPF_ALU64, BPF_RSH, BPF_K), | |
190 | DL(BPF_ALU64, BPF_XOR, BPF_X), | |
191 | DL(BPF_ALU64, BPF_XOR, BPF_K), | |
192 | DL(BPF_ALU64, BPF_MUL, BPF_X), | |
193 | DL(BPF_ALU64, BPF_MUL, BPF_K), | |
194 | DL(BPF_ALU64, BPF_MOV, BPF_X), | |
195 | DL(BPF_ALU64, BPF_MOV, BPF_K), | |
196 | DL(BPF_ALU64, BPF_ARSH, BPF_X), | |
197 | DL(BPF_ALU64, BPF_ARSH, BPF_K), | |
198 | DL(BPF_ALU64, BPF_DIV, BPF_X), | |
199 | DL(BPF_ALU64, BPF_DIV, BPF_K), | |
200 | DL(BPF_ALU64, BPF_MOD, BPF_X), | |
201 | DL(BPF_ALU64, BPF_MOD, BPF_K), | |
202 | DL(BPF_ALU64, BPF_NEG, 0), | |
203 | DL(BPF_JMP, BPF_CALL, 0), | |
204 | DL(BPF_JMP, BPF_JA, 0), | |
205 | DL(BPF_JMP, BPF_JEQ, BPF_X), | |
206 | DL(BPF_JMP, BPF_JEQ, BPF_K), | |
207 | DL(BPF_JMP, BPF_JNE, BPF_X), | |
208 | DL(BPF_JMP, BPF_JNE, BPF_K), | |
209 | DL(BPF_JMP, BPF_JGT, BPF_X), | |
210 | DL(BPF_JMP, BPF_JGT, BPF_K), | |
211 | DL(BPF_JMP, BPF_JGE, BPF_X), | |
212 | DL(BPF_JMP, BPF_JGE, BPF_K), | |
213 | DL(BPF_JMP, BPF_JSGT, BPF_X), | |
214 | DL(BPF_JMP, BPF_JSGT, BPF_K), | |
215 | DL(BPF_JMP, BPF_JSGE, BPF_X), | |
216 | DL(BPF_JMP, BPF_JSGE, BPF_K), | |
217 | DL(BPF_JMP, BPF_JSET, BPF_X), | |
218 | DL(BPF_JMP, BPF_JSET, BPF_K), | |
219 | DL(BPF_JMP, BPF_EXIT, 0), | |
220 | DL(BPF_STX, BPF_MEM, BPF_B), | |
221 | DL(BPF_STX, BPF_MEM, BPF_H), | |
222 | DL(BPF_STX, BPF_MEM, BPF_W), | |
223 | DL(BPF_STX, BPF_MEM, BPF_DW), | |
224 | DL(BPF_STX, BPF_XADD, BPF_W), | |
225 | DL(BPF_STX, BPF_XADD, BPF_DW), | |
226 | DL(BPF_ST, BPF_MEM, BPF_B), | |
227 | DL(BPF_ST, BPF_MEM, BPF_H), | |
228 | DL(BPF_ST, BPF_MEM, BPF_W), | |
229 | DL(BPF_ST, BPF_MEM, BPF_DW), | |
230 | DL(BPF_LDX, BPF_MEM, BPF_B), | |
231 | DL(BPF_LDX, BPF_MEM, BPF_H), | |
232 | DL(BPF_LDX, BPF_MEM, BPF_W), | |
233 | DL(BPF_LDX, BPF_MEM, BPF_DW), | |
234 | DL(BPF_LD, BPF_ABS, BPF_W), | |
235 | DL(BPF_LD, BPF_ABS, BPF_H), | |
236 | DL(BPF_LD, BPF_ABS, BPF_B), | |
237 | DL(BPF_LD, BPF_IND, BPF_W), | |
238 | DL(BPF_LD, BPF_IND, BPF_H), | |
239 | DL(BPF_LD, BPF_IND, BPF_B), | |
240 | #undef DL | |
241 | }; | |
1da177e4 | 242 | |
bd4cf0ed AS |
243 | regs[FP_REG] = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)]; |
244 | regs[ARG1_REG] = (u64) (unsigned long) ctx; | |
245 | ||
246 | select_insn: | |
247 | goto *jumptable[insn->code]; | |
248 | ||
249 | /* ALU */ | |
250 | #define ALU(OPCODE, OP) \ | |
251 | BPF_ALU64_##OPCODE##_BPF_X: \ | |
252 | A = A OP X; \ | |
253 | CONT; \ | |
254 | BPF_ALU_##OPCODE##_BPF_X: \ | |
255 | A = (u32) A OP (u32) X; \ | |
256 | CONT; \ | |
257 | BPF_ALU64_##OPCODE##_BPF_K: \ | |
258 | A = A OP K; \ | |
259 | CONT; \ | |
260 | BPF_ALU_##OPCODE##_BPF_K: \ | |
261 | A = (u32) A OP (u32) K; \ | |
262 | CONT; | |
263 | ||
264 | ALU(BPF_ADD, +) | |
265 | ALU(BPF_SUB, -) | |
266 | ALU(BPF_AND, &) | |
267 | ALU(BPF_OR, |) | |
268 | ALU(BPF_LSH, <<) | |
269 | ALU(BPF_RSH, >>) | |
270 | ALU(BPF_XOR, ^) | |
271 | ALU(BPF_MUL, *) | |
272 | #undef ALU | |
273 | BPF_ALU_BPF_NEG_0: | |
274 | A = (u32) -A; | |
275 | CONT; | |
276 | BPF_ALU64_BPF_NEG_0: | |
277 | A = -A; | |
278 | CONT; | |
279 | BPF_ALU_BPF_MOV_BPF_X: | |
280 | A = (u32) X; | |
281 | CONT; | |
282 | BPF_ALU_BPF_MOV_BPF_K: | |
283 | A = (u32) K; | |
284 | CONT; | |
285 | BPF_ALU64_BPF_MOV_BPF_X: | |
286 | A = X; | |
287 | CONT; | |
288 | BPF_ALU64_BPF_MOV_BPF_K: | |
289 | A = K; | |
290 | CONT; | |
291 | BPF_ALU64_BPF_ARSH_BPF_X: | |
292 | (*(s64 *) &A) >>= X; | |
293 | CONT; | |
294 | BPF_ALU64_BPF_ARSH_BPF_K: | |
295 | (*(s64 *) &A) >>= K; | |
296 | CONT; | |
297 | BPF_ALU64_BPF_MOD_BPF_X: | |
5f9fde5f DB |
298 | if (unlikely(X == 0)) |
299 | return 0; | |
bd4cf0ed | 300 | tmp = A; |
5f9fde5f | 301 | A = do_div(tmp, X); |
bd4cf0ed AS |
302 | CONT; |
303 | BPF_ALU_BPF_MOD_BPF_X: | |
5f9fde5f DB |
304 | if (unlikely(X == 0)) |
305 | return 0; | |
bd4cf0ed | 306 | tmp = (u32) A; |
5f9fde5f | 307 | A = do_div(tmp, (u32) X); |
bd4cf0ed AS |
308 | CONT; |
309 | BPF_ALU64_BPF_MOD_BPF_K: | |
310 | tmp = A; | |
5f9fde5f | 311 | A = do_div(tmp, K); |
bd4cf0ed AS |
312 | CONT; |
313 | BPF_ALU_BPF_MOD_BPF_K: | |
314 | tmp = (u32) A; | |
5f9fde5f | 315 | A = do_div(tmp, (u32) K); |
bd4cf0ed AS |
316 | CONT; |
317 | BPF_ALU64_BPF_DIV_BPF_X: | |
5f9fde5f DB |
318 | if (unlikely(X == 0)) |
319 | return 0; | |
320 | do_div(A, X); | |
bd4cf0ed AS |
321 | CONT; |
322 | BPF_ALU_BPF_DIV_BPF_X: | |
5f9fde5f DB |
323 | if (unlikely(X == 0)) |
324 | return 0; | |
bd4cf0ed | 325 | tmp = (u32) A; |
5f9fde5f | 326 | do_div(tmp, (u32) X); |
bd4cf0ed AS |
327 | A = (u32) tmp; |
328 | CONT; | |
329 | BPF_ALU64_BPF_DIV_BPF_K: | |
5f9fde5f | 330 | do_div(A, K); |
bd4cf0ed AS |
331 | CONT; |
332 | BPF_ALU_BPF_DIV_BPF_K: | |
333 | tmp = (u32) A; | |
5f9fde5f | 334 | do_div(tmp, (u32) K); |
bd4cf0ed AS |
335 | A = (u32) tmp; |
336 | CONT; | |
337 | BPF_ALU_BPF_END_BPF_TO_BE: | |
338 | switch (K) { | |
339 | case 16: | |
340 | A = (__force u16) cpu_to_be16(A); | |
341 | break; | |
342 | case 32: | |
343 | A = (__force u32) cpu_to_be32(A); | |
344 | break; | |
345 | case 64: | |
346 | A = (__force u64) cpu_to_be64(A); | |
347 | break; | |
348 | } | |
349 | CONT; | |
350 | BPF_ALU_BPF_END_BPF_TO_LE: | |
351 | switch (K) { | |
352 | case 16: | |
353 | A = (__force u16) cpu_to_le16(A); | |
354 | break; | |
355 | case 32: | |
356 | A = (__force u32) cpu_to_le32(A); | |
357 | break; | |
358 | case 64: | |
359 | A = (__force u64) cpu_to_le64(A); | |
360 | break; | |
361 | } | |
362 | CONT; | |
363 | ||
364 | /* CALL */ | |
365 | BPF_JMP_BPF_CALL_0: | |
366 | /* Function call scratches R1-R5 registers, preserves R6-R9, | |
367 | * and stores return value into R0. | |
368 | */ | |
369 | R0 = (__bpf_call_base + insn->imm)(regs[1], regs[2], regs[3], | |
370 | regs[4], regs[5]); | |
371 | CONT; | |
372 | ||
373 | /* JMP */ | |
374 | BPF_JMP_BPF_JA_0: | |
375 | insn += insn->off; | |
376 | CONT; | |
377 | BPF_JMP_BPF_JEQ_BPF_X: | |
378 | if (A == X) { | |
379 | insn += insn->off; | |
380 | CONT_JMP; | |
381 | } | |
382 | CONT; | |
383 | BPF_JMP_BPF_JEQ_BPF_K: | |
384 | if (A == K) { | |
385 | insn += insn->off; | |
386 | CONT_JMP; | |
387 | } | |
388 | CONT; | |
389 | BPF_JMP_BPF_JNE_BPF_X: | |
390 | if (A != X) { | |
391 | insn += insn->off; | |
392 | CONT_JMP; | |
393 | } | |
394 | CONT; | |
395 | BPF_JMP_BPF_JNE_BPF_K: | |
396 | if (A != K) { | |
397 | insn += insn->off; | |
398 | CONT_JMP; | |
399 | } | |
400 | CONT; | |
401 | BPF_JMP_BPF_JGT_BPF_X: | |
402 | if (A > X) { | |
403 | insn += insn->off; | |
404 | CONT_JMP; | |
405 | } | |
406 | CONT; | |
407 | BPF_JMP_BPF_JGT_BPF_K: | |
408 | if (A > K) { | |
409 | insn += insn->off; | |
410 | CONT_JMP; | |
411 | } | |
412 | CONT; | |
413 | BPF_JMP_BPF_JGE_BPF_X: | |
414 | if (A >= X) { | |
415 | insn += insn->off; | |
416 | CONT_JMP; | |
417 | } | |
418 | CONT; | |
419 | BPF_JMP_BPF_JGE_BPF_K: | |
420 | if (A >= K) { | |
421 | insn += insn->off; | |
422 | CONT_JMP; | |
423 | } | |
424 | CONT; | |
425 | BPF_JMP_BPF_JSGT_BPF_X: | |
426 | if (((s64)A) > ((s64)X)) { | |
427 | insn += insn->off; | |
428 | CONT_JMP; | |
429 | } | |
430 | CONT; | |
431 | BPF_JMP_BPF_JSGT_BPF_K: | |
432 | if (((s64)A) > ((s64)K)) { | |
433 | insn += insn->off; | |
434 | CONT_JMP; | |
435 | } | |
436 | CONT; | |
437 | BPF_JMP_BPF_JSGE_BPF_X: | |
438 | if (((s64)A) >= ((s64)X)) { | |
439 | insn += insn->off; | |
440 | CONT_JMP; | |
441 | } | |
442 | CONT; | |
443 | BPF_JMP_BPF_JSGE_BPF_K: | |
444 | if (((s64)A) >= ((s64)K)) { | |
445 | insn += insn->off; | |
446 | CONT_JMP; | |
447 | } | |
448 | CONT; | |
449 | BPF_JMP_BPF_JSET_BPF_X: | |
450 | if (A & X) { | |
451 | insn += insn->off; | |
452 | CONT_JMP; | |
453 | } | |
454 | CONT; | |
455 | BPF_JMP_BPF_JSET_BPF_K: | |
456 | if (A & K) { | |
457 | insn += insn->off; | |
458 | CONT_JMP; | |
459 | } | |
460 | CONT; | |
461 | BPF_JMP_BPF_EXIT_0: | |
462 | return R0; | |
463 | ||
464 | /* STX and ST and LDX*/ | |
465 | #define LDST(SIZEOP, SIZE) \ | |
466 | BPF_STX_BPF_MEM_##SIZEOP: \ | |
467 | *(SIZE *)(unsigned long) (A + insn->off) = X; \ | |
468 | CONT; \ | |
469 | BPF_ST_BPF_MEM_##SIZEOP: \ | |
470 | *(SIZE *)(unsigned long) (A + insn->off) = K; \ | |
471 | CONT; \ | |
472 | BPF_LDX_BPF_MEM_##SIZEOP: \ | |
473 | A = *(SIZE *)(unsigned long) (X + insn->off); \ | |
474 | CONT; | |
475 | ||
476 | LDST(BPF_B, u8) | |
477 | LDST(BPF_H, u16) | |
478 | LDST(BPF_W, u32) | |
479 | LDST(BPF_DW, u64) | |
480 | #undef LDST | |
481 | BPF_STX_BPF_XADD_BPF_W: /* lock xadd *(u32 *)(A + insn->off) += X */ | |
482 | atomic_add((u32) X, (atomic_t *)(unsigned long) | |
483 | (A + insn->off)); | |
484 | CONT; | |
485 | BPF_STX_BPF_XADD_BPF_DW: /* lock xadd *(u64 *)(A + insn->off) += X */ | |
486 | atomic64_add((u64) X, (atomic64_t *)(unsigned long) | |
487 | (A + insn->off)); | |
488 | CONT; | |
489 | BPF_LD_BPF_ABS_BPF_W: /* R0 = ntohl(*(u32 *) (skb->data + K)) */ | |
490 | off = K; | |
491 | load_word: | |
492 | /* BPF_LD + BPD_ABS and BPF_LD + BPF_IND insns are only | |
493 | * appearing in the programs where ctx == skb. All programs | |
494 | * keep 'ctx' in regs[CTX_REG] == R6, sk_convert_filter() | |
495 | * saves it in R6, internal BPF verifier will check that | |
496 | * R6 == ctx. | |
497 | * | |
498 | * BPF_ABS and BPF_IND are wrappers of function calls, so | |
499 | * they scratch R1-R5 registers, preserve R6-R9, and store | |
500 | * return value into R0. | |
501 | * | |
502 | * Implicit input: | |
503 | * ctx | |
504 | * | |
505 | * Explicit input: | |
506 | * X == any register | |
507 | * K == 32-bit immediate | |
508 | * | |
509 | * Output: | |
510 | * R0 - 8/16/32-bit skb data converted to cpu endianness | |
511 | */ | |
512 | ptr = load_pointer((struct sk_buff *) ctx, off, 4, &tmp); | |
513 | if (likely(ptr != NULL)) { | |
514 | R0 = get_unaligned_be32(ptr); | |
515 | CONT; | |
516 | } | |
517 | return 0; | |
518 | BPF_LD_BPF_ABS_BPF_H: /* R0 = ntohs(*(u16 *) (skb->data + K)) */ | |
519 | off = K; | |
520 | load_half: | |
521 | ptr = load_pointer((struct sk_buff *) ctx, off, 2, &tmp); | |
522 | if (likely(ptr != NULL)) { | |
523 | R0 = get_unaligned_be16(ptr); | |
524 | CONT; | |
525 | } | |
526 | return 0; | |
527 | BPF_LD_BPF_ABS_BPF_B: /* R0 = *(u8 *) (ctx + K) */ | |
528 | off = K; | |
529 | load_byte: | |
530 | ptr = load_pointer((struct sk_buff *) ctx, off, 1, &tmp); | |
531 | if (likely(ptr != NULL)) { | |
532 | R0 = *(u8 *)ptr; | |
533 | CONT; | |
534 | } | |
535 | return 0; | |
536 | BPF_LD_BPF_IND_BPF_W: /* R0 = ntohl(*(u32 *) (skb->data + X + K)) */ | |
537 | off = K + X; | |
538 | goto load_word; | |
539 | BPF_LD_BPF_IND_BPF_H: /* R0 = ntohs(*(u16 *) (skb->data + X + K)) */ | |
540 | off = K + X; | |
541 | goto load_half; | |
542 | BPF_LD_BPF_IND_BPF_B: /* R0 = *(u8 *) (skb->data + X + K) */ | |
543 | off = K + X; | |
544 | goto load_byte; | |
545 | ||
546 | default_label: | |
547 | /* If we ever reach this, we have a bug somewhere. */ | |
548 | WARN_RATELIMIT(1, "unknown opcode %02x\n", insn->code); | |
549 | return 0; | |
550 | #undef CONT_JMP | |
551 | #undef CONT | |
552 | ||
553 | #undef R0 | |
554 | #undef X | |
555 | #undef A | |
556 | #undef K | |
557 | } | |
558 | ||
559 | u32 sk_run_filter_int_seccomp(const struct seccomp_data *ctx, | |
560 | const struct sock_filter_int *insni) | |
561 | __attribute__ ((alias ("__sk_run_filter"))); | |
562 | ||
563 | u32 sk_run_filter_int_skb(const struct sk_buff *ctx, | |
564 | const struct sock_filter_int *insni) | |
565 | __attribute__ ((alias ("__sk_run_filter"))); | |
566 | EXPORT_SYMBOL_GPL(sk_run_filter_int_skb); | |
567 | ||
568 | /* Helper to find the offset of pkt_type in sk_buff structure. We want | |
569 | * to make sure its still a 3bit field starting at a byte boundary; | |
570 | * taken from arch/x86/net/bpf_jit_comp.c. | |
571 | */ | |
572 | #define PKT_TYPE_MAX 7 | |
573 | static unsigned int pkt_type_offset(void) | |
574 | { | |
575 | struct sk_buff skb_probe = { .pkt_type = ~0, }; | |
576 | u8 *ct = (u8 *) &skb_probe; | |
577 | unsigned int off; | |
578 | ||
579 | for (off = 0; off < sizeof(struct sk_buff); off++) { | |
580 | if (ct[off] == PKT_TYPE_MAX) | |
581 | return off; | |
582 | } | |
583 | ||
584 | pr_err_once("Please fix %s, as pkt_type couldn't be found!\n", __func__); | |
585 | return -1; | |
586 | } | |
587 | ||
588 | static u64 __skb_get_pay_offset(u64 ctx, u64 A, u64 X, u64 r4, u64 r5) | |
589 | { | |
590 | struct sk_buff *skb = (struct sk_buff *)(long) ctx; | |
591 | ||
592 | return __skb_get_poff(skb); | |
593 | } | |
594 | ||
595 | static u64 __skb_get_nlattr(u64 ctx, u64 A, u64 X, u64 r4, u64 r5) | |
596 | { | |
597 | struct sk_buff *skb = (struct sk_buff *)(long) ctx; | |
598 | struct nlattr *nla; | |
599 | ||
600 | if (skb_is_nonlinear(skb)) | |
601 | return 0; | |
602 | ||
05ab8f26 MK |
603 | if (skb->len < sizeof(struct nlattr)) |
604 | return 0; | |
605 | ||
bd4cf0ed AS |
606 | if (A > skb->len - sizeof(struct nlattr)) |
607 | return 0; | |
608 | ||
609 | nla = nla_find((struct nlattr *) &skb->data[A], skb->len - A, X); | |
610 | if (nla) | |
611 | return (void *) nla - (void *) skb->data; | |
612 | ||
613 | return 0; | |
614 | } | |
615 | ||
616 | static u64 __skb_get_nlattr_nest(u64 ctx, u64 A, u64 X, u64 r4, u64 r5) | |
617 | { | |
618 | struct sk_buff *skb = (struct sk_buff *)(long) ctx; | |
619 | struct nlattr *nla; | |
620 | ||
621 | if (skb_is_nonlinear(skb)) | |
622 | return 0; | |
623 | ||
05ab8f26 MK |
624 | if (skb->len < sizeof(struct nlattr)) |
625 | return 0; | |
626 | ||
bd4cf0ed AS |
627 | if (A > skb->len - sizeof(struct nlattr)) |
628 | return 0; | |
629 | ||
630 | nla = (struct nlattr *) &skb->data[A]; | |
05ab8f26 | 631 | if (nla->nla_len > skb->len - A) |
bd4cf0ed AS |
632 | return 0; |
633 | ||
634 | nla = nla_find_nested(nla, X); | |
635 | if (nla) | |
636 | return (void *) nla - (void *) skb->data; | |
637 | ||
638 | return 0; | |
639 | } | |
640 | ||
641 | static u64 __get_raw_cpu_id(u64 ctx, u64 A, u64 X, u64 r4, u64 r5) | |
642 | { | |
643 | return raw_smp_processor_id(); | |
644 | } | |
645 | ||
646 | /* Register mappings for user programs. */ | |
647 | #define A_REG 0 | |
648 | #define X_REG 7 | |
649 | #define TMP_REG 8 | |
650 | #define ARG2_REG 2 | |
651 | #define ARG3_REG 3 | |
652 | ||
653 | static bool convert_bpf_extensions(struct sock_filter *fp, | |
654 | struct sock_filter_int **insnp) | |
655 | { | |
656 | struct sock_filter_int *insn = *insnp; | |
657 | ||
658 | switch (fp->k) { | |
659 | case SKF_AD_OFF + SKF_AD_PROTOCOL: | |
660 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2); | |
661 | ||
662 | insn->code = BPF_LDX | BPF_MEM | BPF_H; | |
663 | insn->a_reg = A_REG; | |
664 | insn->x_reg = CTX_REG; | |
665 | insn->off = offsetof(struct sk_buff, protocol); | |
666 | insn++; | |
667 | ||
668 | /* A = ntohs(A) [emitting a nop or swap16] */ | |
669 | insn->code = BPF_ALU | BPF_END | BPF_FROM_BE; | |
670 | insn->a_reg = A_REG; | |
671 | insn->imm = 16; | |
672 | break; | |
673 | ||
674 | case SKF_AD_OFF + SKF_AD_PKTTYPE: | |
675 | insn->code = BPF_LDX | BPF_MEM | BPF_B; | |
676 | insn->a_reg = A_REG; | |
677 | insn->x_reg = CTX_REG; | |
678 | insn->off = pkt_type_offset(); | |
679 | if (insn->off < 0) | |
680 | return false; | |
681 | insn++; | |
682 | ||
683 | insn->code = BPF_ALU | BPF_AND | BPF_K; | |
684 | insn->a_reg = A_REG; | |
685 | insn->imm = PKT_TYPE_MAX; | |
686 | break; | |
687 | ||
688 | case SKF_AD_OFF + SKF_AD_IFINDEX: | |
689 | case SKF_AD_OFF + SKF_AD_HATYPE: | |
690 | if (FIELD_SIZEOF(struct sk_buff, dev) == 8) | |
691 | insn->code = BPF_LDX | BPF_MEM | BPF_DW; | |
692 | else | |
693 | insn->code = BPF_LDX | BPF_MEM | BPF_W; | |
694 | insn->a_reg = TMP_REG; | |
695 | insn->x_reg = CTX_REG; | |
696 | insn->off = offsetof(struct sk_buff, dev); | |
697 | insn++; | |
698 | ||
699 | insn->code = BPF_JMP | BPF_JNE | BPF_K; | |
700 | insn->a_reg = TMP_REG; | |
701 | insn->imm = 0; | |
702 | insn->off = 1; | |
703 | insn++; | |
704 | ||
705 | insn->code = BPF_JMP | BPF_EXIT; | |
706 | insn++; | |
707 | ||
708 | BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4); | |
709 | BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, type) != 2); | |
710 | ||
711 | insn->a_reg = A_REG; | |
712 | insn->x_reg = TMP_REG; | |
713 | ||
714 | if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX) { | |
715 | insn->code = BPF_LDX | BPF_MEM | BPF_W; | |
716 | insn->off = offsetof(struct net_device, ifindex); | |
717 | } else { | |
718 | insn->code = BPF_LDX | BPF_MEM | BPF_H; | |
719 | insn->off = offsetof(struct net_device, type); | |
720 | } | |
721 | break; | |
722 | ||
723 | case SKF_AD_OFF + SKF_AD_MARK: | |
724 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4); | |
725 | ||
726 | insn->code = BPF_LDX | BPF_MEM | BPF_W; | |
727 | insn->a_reg = A_REG; | |
728 | insn->x_reg = CTX_REG; | |
729 | insn->off = offsetof(struct sk_buff, mark); | |
730 | break; | |
731 | ||
732 | case SKF_AD_OFF + SKF_AD_RXHASH: | |
733 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4); | |
734 | ||
735 | insn->code = BPF_LDX | BPF_MEM | BPF_W; | |
736 | insn->a_reg = A_REG; | |
737 | insn->x_reg = CTX_REG; | |
738 | insn->off = offsetof(struct sk_buff, hash); | |
739 | break; | |
740 | ||
741 | case SKF_AD_OFF + SKF_AD_QUEUE: | |
742 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2); | |
743 | ||
744 | insn->code = BPF_LDX | BPF_MEM | BPF_H; | |
745 | insn->a_reg = A_REG; | |
746 | insn->x_reg = CTX_REG; | |
747 | insn->off = offsetof(struct sk_buff, queue_mapping); | |
748 | break; | |
749 | ||
750 | case SKF_AD_OFF + SKF_AD_VLAN_TAG: | |
751 | case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT: | |
752 | BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2); | |
753 | ||
754 | insn->code = BPF_LDX | BPF_MEM | BPF_H; | |
755 | insn->a_reg = A_REG; | |
756 | insn->x_reg = CTX_REG; | |
757 | insn->off = offsetof(struct sk_buff, vlan_tci); | |
758 | insn++; | |
759 | ||
760 | BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000); | |
761 | ||
762 | if (fp->k == SKF_AD_OFF + SKF_AD_VLAN_TAG) { | |
763 | insn->code = BPF_ALU | BPF_AND | BPF_K; | |
764 | insn->a_reg = A_REG; | |
765 | insn->imm = ~VLAN_TAG_PRESENT; | |
766 | } else { | |
767 | insn->code = BPF_ALU | BPF_RSH | BPF_K; | |
768 | insn->a_reg = A_REG; | |
769 | insn->imm = 12; | |
770 | insn++; | |
771 | ||
772 | insn->code = BPF_ALU | BPF_AND | BPF_K; | |
773 | insn->a_reg = A_REG; | |
774 | insn->imm = 1; | |
775 | } | |
776 | break; | |
777 | ||
778 | case SKF_AD_OFF + SKF_AD_PAY_OFFSET: | |
779 | case SKF_AD_OFF + SKF_AD_NLATTR: | |
780 | case SKF_AD_OFF + SKF_AD_NLATTR_NEST: | |
781 | case SKF_AD_OFF + SKF_AD_CPU: | |
782 | /* arg1 = ctx */ | |
783 | insn->code = BPF_ALU64 | BPF_MOV | BPF_X; | |
784 | insn->a_reg = ARG1_REG; | |
785 | insn->x_reg = CTX_REG; | |
786 | insn++; | |
787 | ||
788 | /* arg2 = A */ | |
789 | insn->code = BPF_ALU64 | BPF_MOV | BPF_X; | |
790 | insn->a_reg = ARG2_REG; | |
791 | insn->x_reg = A_REG; | |
792 | insn++; | |
793 | ||
794 | /* arg3 = X */ | |
795 | insn->code = BPF_ALU64 | BPF_MOV | BPF_X; | |
796 | insn->a_reg = ARG3_REG; | |
797 | insn->x_reg = X_REG; | |
798 | insn++; | |
799 | ||
800 | /* Emit call(ctx, arg2=A, arg3=X) */ | |
801 | insn->code = BPF_JMP | BPF_CALL; | |
802 | switch (fp->k) { | |
803 | case SKF_AD_OFF + SKF_AD_PAY_OFFSET: | |
804 | insn->imm = __skb_get_pay_offset - __bpf_call_base; | |
805 | break; | |
806 | case SKF_AD_OFF + SKF_AD_NLATTR: | |
807 | insn->imm = __skb_get_nlattr - __bpf_call_base; | |
808 | break; | |
809 | case SKF_AD_OFF + SKF_AD_NLATTR_NEST: | |
810 | insn->imm = __skb_get_nlattr_nest - __bpf_call_base; | |
811 | break; | |
812 | case SKF_AD_OFF + SKF_AD_CPU: | |
813 | insn->imm = __get_raw_cpu_id - __bpf_call_base; | |
814 | break; | |
815 | } | |
816 | break; | |
817 | ||
818 | case SKF_AD_OFF + SKF_AD_ALU_XOR_X: | |
819 | insn->code = BPF_ALU | BPF_XOR | BPF_X; | |
820 | insn->a_reg = A_REG; | |
821 | insn->x_reg = X_REG; | |
822 | break; | |
823 | ||
824 | default: | |
825 | /* This is just a dummy call to avoid letting the compiler | |
826 | * evict __bpf_call_base() as an optimization. Placed here | |
827 | * where no-one bothers. | |
828 | */ | |
829 | BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0); | |
830 | return false; | |
831 | } | |
832 | ||
833 | *insnp = insn; | |
834 | return true; | |
835 | } | |
836 | ||
837 | /** | |
838 | * sk_convert_filter - convert filter program | |
839 | * @prog: the user passed filter program | |
840 | * @len: the length of the user passed filter program | |
841 | * @new_prog: buffer where converted program will be stored | |
842 | * @new_len: pointer to store length of converted program | |
843 | * | |
844 | * Remap 'sock_filter' style BPF instruction set to 'sock_filter_ext' style. | |
845 | * Conversion workflow: | |
846 | * | |
847 | * 1) First pass for calculating the new program length: | |
848 | * sk_convert_filter(old_prog, old_len, NULL, &new_len) | |
849 | * | |
850 | * 2) 2nd pass to remap in two passes: 1st pass finds new | |
851 | * jump offsets, 2nd pass remapping: | |
852 | * new_prog = kmalloc(sizeof(struct sock_filter_int) * new_len); | |
853 | * sk_convert_filter(old_prog, old_len, new_prog, &new_len); | |
854 | * | |
855 | * User BPF's register A is mapped to our BPF register 6, user BPF | |
856 | * register X is mapped to BPF register 7; frame pointer is always | |
857 | * register 10; Context 'void *ctx' is stored in register 1, that is, | |
858 | * for socket filters: ctx == 'struct sk_buff *', for seccomp: | |
859 | * ctx == 'struct seccomp_data *'. | |
860 | */ | |
861 | int sk_convert_filter(struct sock_filter *prog, int len, | |
862 | struct sock_filter_int *new_prog, int *new_len) | |
863 | { | |
864 | int new_flen = 0, pass = 0, target, i; | |
865 | struct sock_filter_int *new_insn; | |
866 | struct sock_filter *fp; | |
867 | int *addrs = NULL; | |
868 | u8 bpf_src; | |
869 | ||
870 | BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK); | |
871 | BUILD_BUG_ON(FP_REG + 1 != MAX_BPF_REG); | |
872 | ||
873 | if (len <= 0 || len >= BPF_MAXINSNS) | |
874 | return -EINVAL; | |
875 | ||
876 | if (new_prog) { | |
877 | addrs = kzalloc(len * sizeof(*addrs), GFP_KERNEL); | |
878 | if (!addrs) | |
879 | return -ENOMEM; | |
880 | } | |
881 | ||
882 | do_pass: | |
883 | new_insn = new_prog; | |
884 | fp = prog; | |
885 | ||
886 | if (new_insn) { | |
887 | new_insn->code = BPF_ALU64 | BPF_MOV | BPF_X; | |
888 | new_insn->a_reg = CTX_REG; | |
889 | new_insn->x_reg = ARG1_REG; | |
890 | } | |
891 | new_insn++; | |
892 | ||
893 | for (i = 0; i < len; fp++, i++) { | |
894 | struct sock_filter_int tmp_insns[6] = { }; | |
895 | struct sock_filter_int *insn = tmp_insns; | |
896 | ||
897 | if (addrs) | |
898 | addrs[i] = new_insn - new_prog; | |
899 | ||
900 | switch (fp->code) { | |
901 | /* All arithmetic insns and skb loads map as-is. */ | |
902 | case BPF_ALU | BPF_ADD | BPF_X: | |
903 | case BPF_ALU | BPF_ADD | BPF_K: | |
904 | case BPF_ALU | BPF_SUB | BPF_X: | |
905 | case BPF_ALU | BPF_SUB | BPF_K: | |
906 | case BPF_ALU | BPF_AND | BPF_X: | |
907 | case BPF_ALU | BPF_AND | BPF_K: | |
908 | case BPF_ALU | BPF_OR | BPF_X: | |
909 | case BPF_ALU | BPF_OR | BPF_K: | |
910 | case BPF_ALU | BPF_LSH | BPF_X: | |
911 | case BPF_ALU | BPF_LSH | BPF_K: | |
912 | case BPF_ALU | BPF_RSH | BPF_X: | |
913 | case BPF_ALU | BPF_RSH | BPF_K: | |
914 | case BPF_ALU | BPF_XOR | BPF_X: | |
915 | case BPF_ALU | BPF_XOR | BPF_K: | |
916 | case BPF_ALU | BPF_MUL | BPF_X: | |
917 | case BPF_ALU | BPF_MUL | BPF_K: | |
918 | case BPF_ALU | BPF_DIV | BPF_X: | |
919 | case BPF_ALU | BPF_DIV | BPF_K: | |
920 | case BPF_ALU | BPF_MOD | BPF_X: | |
921 | case BPF_ALU | BPF_MOD | BPF_K: | |
922 | case BPF_ALU | BPF_NEG: | |
923 | case BPF_LD | BPF_ABS | BPF_W: | |
924 | case BPF_LD | BPF_ABS | BPF_H: | |
925 | case BPF_LD | BPF_ABS | BPF_B: | |
926 | case BPF_LD | BPF_IND | BPF_W: | |
927 | case BPF_LD | BPF_IND | BPF_H: | |
928 | case BPF_LD | BPF_IND | BPF_B: | |
929 | /* Check for overloaded BPF extension and | |
930 | * directly convert it if found, otherwise | |
931 | * just move on with mapping. | |
932 | */ | |
933 | if (BPF_CLASS(fp->code) == BPF_LD && | |
934 | BPF_MODE(fp->code) == BPF_ABS && | |
935 | convert_bpf_extensions(fp, &insn)) | |
936 | break; | |
937 | ||
938 | insn->code = fp->code; | |
939 | insn->a_reg = A_REG; | |
940 | insn->x_reg = X_REG; | |
941 | insn->imm = fp->k; | |
942 | break; | |
943 | ||
944 | /* Jump opcodes map as-is, but offsets need adjustment. */ | |
945 | case BPF_JMP | BPF_JA: | |
946 | target = i + fp->k + 1; | |
947 | insn->code = fp->code; | |
948 | #define EMIT_JMP \ | |
949 | do { \ | |
950 | if (target >= len || target < 0) \ | |
951 | goto err; \ | |
952 | insn->off = addrs ? addrs[target] - addrs[i] - 1 : 0; \ | |
953 | /* Adjust pc relative offset for 2nd or 3rd insn. */ \ | |
954 | insn->off -= insn - tmp_insns; \ | |
955 | } while (0) | |
956 | ||
957 | EMIT_JMP; | |
958 | break; | |
959 | ||
960 | case BPF_JMP | BPF_JEQ | BPF_K: | |
961 | case BPF_JMP | BPF_JEQ | BPF_X: | |
962 | case BPF_JMP | BPF_JSET | BPF_K: | |
963 | case BPF_JMP | BPF_JSET | BPF_X: | |
964 | case BPF_JMP | BPF_JGT | BPF_K: | |
965 | case BPF_JMP | BPF_JGT | BPF_X: | |
966 | case BPF_JMP | BPF_JGE | BPF_K: | |
967 | case BPF_JMP | BPF_JGE | BPF_X: | |
968 | if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) { | |
969 | /* BPF immediates are signed, zero extend | |
970 | * immediate into tmp register and use it | |
971 | * in compare insn. | |
972 | */ | |
973 | insn->code = BPF_ALU | BPF_MOV | BPF_K; | |
974 | insn->a_reg = TMP_REG; | |
975 | insn->imm = fp->k; | |
976 | insn++; | |
977 | ||
978 | insn->a_reg = A_REG; | |
979 | insn->x_reg = TMP_REG; | |
980 | bpf_src = BPF_X; | |
981 | } else { | |
982 | insn->a_reg = A_REG; | |
983 | insn->x_reg = X_REG; | |
984 | insn->imm = fp->k; | |
985 | bpf_src = BPF_SRC(fp->code); | |
1da177e4 | 986 | } |
bd4cf0ed AS |
987 | |
988 | /* Common case where 'jump_false' is next insn. */ | |
989 | if (fp->jf == 0) { | |
990 | insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src; | |
991 | target = i + fp->jt + 1; | |
992 | EMIT_JMP; | |
993 | break; | |
1da177e4 | 994 | } |
bd4cf0ed AS |
995 | |
996 | /* Convert JEQ into JNE when 'jump_true' is next insn. */ | |
997 | if (fp->jt == 0 && BPF_OP(fp->code) == BPF_JEQ) { | |
998 | insn->code = BPF_JMP | BPF_JNE | bpf_src; | |
999 | target = i + fp->jf + 1; | |
1000 | EMIT_JMP; | |
1001 | break; | |
0b05b2a4 | 1002 | } |
bd4cf0ed AS |
1003 | |
1004 | /* Other jumps are mapped into two insns: Jxx and JA. */ | |
1005 | target = i + fp->jt + 1; | |
1006 | insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src; | |
1007 | EMIT_JMP; | |
1008 | insn++; | |
1009 | ||
1010 | insn->code = BPF_JMP | BPF_JA; | |
1011 | target = i + fp->jf + 1; | |
1012 | EMIT_JMP; | |
1013 | break; | |
1014 | ||
1015 | /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */ | |
1016 | case BPF_LDX | BPF_MSH | BPF_B: | |
1017 | insn->code = BPF_ALU64 | BPF_MOV | BPF_X; | |
1018 | insn->a_reg = TMP_REG; | |
1019 | insn->x_reg = A_REG; | |
1020 | insn++; | |
1021 | ||
1022 | insn->code = BPF_LD | BPF_ABS | BPF_B; | |
1023 | insn->a_reg = A_REG; | |
1024 | insn->imm = fp->k; | |
1025 | insn++; | |
1026 | ||
1027 | insn->code = BPF_ALU | BPF_AND | BPF_K; | |
1028 | insn->a_reg = A_REG; | |
1029 | insn->imm = 0xf; | |
1030 | insn++; | |
1031 | ||
1032 | insn->code = BPF_ALU | BPF_LSH | BPF_K; | |
1033 | insn->a_reg = A_REG; | |
1034 | insn->imm = 2; | |
1035 | insn++; | |
1036 | ||
1037 | insn->code = BPF_ALU64 | BPF_MOV | BPF_X; | |
1038 | insn->a_reg = X_REG; | |
1039 | insn->x_reg = A_REG; | |
1040 | insn++; | |
1041 | ||
1042 | insn->code = BPF_ALU64 | BPF_MOV | BPF_X; | |
1043 | insn->a_reg = A_REG; | |
1044 | insn->x_reg = TMP_REG; | |
1045 | break; | |
1046 | ||
1047 | /* RET_K, RET_A are remaped into 2 insns. */ | |
1048 | case BPF_RET | BPF_A: | |
1049 | case BPF_RET | BPF_K: | |
1050 | insn->code = BPF_ALU | BPF_MOV | | |
1051 | (BPF_RVAL(fp->code) == BPF_K ? | |
1052 | BPF_K : BPF_X); | |
1053 | insn->a_reg = 0; | |
1054 | insn->x_reg = A_REG; | |
1055 | insn->imm = fp->k; | |
1056 | insn++; | |
1057 | ||
1058 | insn->code = BPF_JMP | BPF_EXIT; | |
1059 | break; | |
1060 | ||
1061 | /* Store to stack. */ | |
1062 | case BPF_ST: | |
1063 | case BPF_STX: | |
1064 | insn->code = BPF_STX | BPF_MEM | BPF_W; | |
1065 | insn->a_reg = FP_REG; | |
1066 | insn->x_reg = fp->code == BPF_ST ? A_REG : X_REG; | |
1067 | insn->off = -(BPF_MEMWORDS - fp->k) * 4; | |
1068 | break; | |
1069 | ||
1070 | /* Load from stack. */ | |
1071 | case BPF_LD | BPF_MEM: | |
1072 | case BPF_LDX | BPF_MEM: | |
1073 | insn->code = BPF_LDX | BPF_MEM | BPF_W; | |
1074 | insn->a_reg = BPF_CLASS(fp->code) == BPF_LD ? | |
1075 | A_REG : X_REG; | |
1076 | insn->x_reg = FP_REG; | |
1077 | insn->off = -(BPF_MEMWORDS - fp->k) * 4; | |
1078 | break; | |
1079 | ||
1080 | /* A = K or X = K */ | |
1081 | case BPF_LD | BPF_IMM: | |
1082 | case BPF_LDX | BPF_IMM: | |
1083 | insn->code = BPF_ALU | BPF_MOV | BPF_K; | |
1084 | insn->a_reg = BPF_CLASS(fp->code) == BPF_LD ? | |
1085 | A_REG : X_REG; | |
1086 | insn->imm = fp->k; | |
1087 | break; | |
1088 | ||
1089 | /* X = A */ | |
1090 | case BPF_MISC | BPF_TAX: | |
1091 | insn->code = BPF_ALU64 | BPF_MOV | BPF_X; | |
1092 | insn->a_reg = X_REG; | |
1093 | insn->x_reg = A_REG; | |
1094 | break; | |
1095 | ||
1096 | /* A = X */ | |
1097 | case BPF_MISC | BPF_TXA: | |
1098 | insn->code = BPF_ALU64 | BPF_MOV | BPF_X; | |
1099 | insn->a_reg = A_REG; | |
1100 | insn->x_reg = X_REG; | |
1101 | break; | |
1102 | ||
1103 | /* A = skb->len or X = skb->len */ | |
1104 | case BPF_LD | BPF_W | BPF_LEN: | |
1105 | case BPF_LDX | BPF_W | BPF_LEN: | |
1106 | insn->code = BPF_LDX | BPF_MEM | BPF_W; | |
1107 | insn->a_reg = BPF_CLASS(fp->code) == BPF_LD ? | |
1108 | A_REG : X_REG; | |
1109 | insn->x_reg = CTX_REG; | |
1110 | insn->off = offsetof(struct sk_buff, len); | |
1111 | break; | |
1112 | ||
1113 | /* access seccomp_data fields */ | |
1114 | case BPF_LDX | BPF_ABS | BPF_W: | |
1115 | insn->code = BPF_LDX | BPF_MEM | BPF_W; | |
1116 | insn->a_reg = A_REG; | |
1117 | insn->x_reg = CTX_REG; | |
1118 | insn->off = fp->k; | |
1119 | break; | |
1120 | ||
1da177e4 | 1121 | default: |
bd4cf0ed | 1122 | goto err; |
1da177e4 | 1123 | } |
bd4cf0ed AS |
1124 | |
1125 | insn++; | |
1126 | if (new_prog) | |
1127 | memcpy(new_insn, tmp_insns, | |
1128 | sizeof(*insn) * (insn - tmp_insns)); | |
1129 | ||
1130 | new_insn += insn - tmp_insns; | |
1da177e4 LT |
1131 | } |
1132 | ||
bd4cf0ed AS |
1133 | if (!new_prog) { |
1134 | /* Only calculating new length. */ | |
1135 | *new_len = new_insn - new_prog; | |
1136 | return 0; | |
1137 | } | |
1138 | ||
1139 | pass++; | |
1140 | if (new_flen != new_insn - new_prog) { | |
1141 | new_flen = new_insn - new_prog; | |
1142 | if (pass > 2) | |
1143 | goto err; | |
1144 | ||
1145 | goto do_pass; | |
1146 | } | |
1147 | ||
1148 | kfree(addrs); | |
1149 | BUG_ON(*new_len != new_flen); | |
1da177e4 | 1150 | return 0; |
bd4cf0ed AS |
1151 | err: |
1152 | kfree(addrs); | |
1153 | return -EINVAL; | |
1da177e4 LT |
1154 | } |
1155 | ||
bd4cf0ed AS |
1156 | /* Security: |
1157 | * | |
2d5311e4 | 1158 | * A BPF program is able to use 16 cells of memory to store intermediate |
bd4cf0ed AS |
1159 | * values (check u32 mem[BPF_MEMWORDS] in sk_run_filter()). |
1160 | * | |
2d5311e4 ED |
1161 | * As we dont want to clear mem[] array for each packet going through |
1162 | * sk_run_filter(), we check that filter loaded by user never try to read | |
1163 | * a cell if not previously written, and we check all branches to be sure | |
25985edc | 1164 | * a malicious user doesn't try to abuse us. |
2d5311e4 ED |
1165 | */ |
1166 | static int check_load_and_stores(struct sock_filter *filter, int flen) | |
1167 | { | |
1168 | u16 *masks, memvalid = 0; /* one bit per cell, 16 cells */ | |
1169 | int pc, ret = 0; | |
1170 | ||
1171 | BUILD_BUG_ON(BPF_MEMWORDS > 16); | |
1172 | masks = kmalloc(flen * sizeof(*masks), GFP_KERNEL); | |
1173 | if (!masks) | |
1174 | return -ENOMEM; | |
1175 | memset(masks, 0xff, flen * sizeof(*masks)); | |
1176 | ||
1177 | for (pc = 0; pc < flen; pc++) { | |
1178 | memvalid &= masks[pc]; | |
1179 | ||
1180 | switch (filter[pc].code) { | |
1181 | case BPF_S_ST: | |
1182 | case BPF_S_STX: | |
1183 | memvalid |= (1 << filter[pc].k); | |
1184 | break; | |
1185 | case BPF_S_LD_MEM: | |
1186 | case BPF_S_LDX_MEM: | |
1187 | if (!(memvalid & (1 << filter[pc].k))) { | |
1188 | ret = -EINVAL; | |
1189 | goto error; | |
1190 | } | |
1191 | break; | |
1192 | case BPF_S_JMP_JA: | |
1193 | /* a jump must set masks on target */ | |
1194 | masks[pc + 1 + filter[pc].k] &= memvalid; | |
1195 | memvalid = ~0; | |
1196 | break; | |
1197 | case BPF_S_JMP_JEQ_K: | |
1198 | case BPF_S_JMP_JEQ_X: | |
1199 | case BPF_S_JMP_JGE_K: | |
1200 | case BPF_S_JMP_JGE_X: | |
1201 | case BPF_S_JMP_JGT_K: | |
1202 | case BPF_S_JMP_JGT_X: | |
1203 | case BPF_S_JMP_JSET_X: | |
1204 | case BPF_S_JMP_JSET_K: | |
1205 | /* a jump must set masks on targets */ | |
1206 | masks[pc + 1 + filter[pc].jt] &= memvalid; | |
1207 | masks[pc + 1 + filter[pc].jf] &= memvalid; | |
1208 | memvalid = ~0; | |
1209 | break; | |
1210 | } | |
1211 | } | |
1212 | error: | |
1213 | kfree(masks); | |
1214 | return ret; | |
1215 | } | |
1216 | ||
1da177e4 LT |
1217 | /** |
1218 | * sk_chk_filter - verify socket filter code | |
1219 | * @filter: filter to verify | |
1220 | * @flen: length of filter | |
1221 | * | |
1222 | * Check the user's filter code. If we let some ugly | |
1223 | * filter code slip through kaboom! The filter must contain | |
93699863 KK |
1224 | * no references or jumps that are out of range, no illegal |
1225 | * instructions, and must end with a RET instruction. | |
1da177e4 | 1226 | * |
7b11f69f KK |
1227 | * All jumps are forward as they are not signed. |
1228 | * | |
1229 | * Returns 0 if the rule set is legal or -EINVAL if not. | |
1da177e4 | 1230 | */ |
4f25af27 | 1231 | int sk_chk_filter(struct sock_filter *filter, unsigned int flen) |
1da177e4 | 1232 | { |
cba328fc TH |
1233 | /* |
1234 | * Valid instructions are initialized to non-0. | |
1235 | * Invalid instructions are initialized to 0. | |
1236 | */ | |
1237 | static const u8 codes[] = { | |
8c1592d6 ED |
1238 | [BPF_ALU|BPF_ADD|BPF_K] = BPF_S_ALU_ADD_K, |
1239 | [BPF_ALU|BPF_ADD|BPF_X] = BPF_S_ALU_ADD_X, | |
1240 | [BPF_ALU|BPF_SUB|BPF_K] = BPF_S_ALU_SUB_K, | |
1241 | [BPF_ALU|BPF_SUB|BPF_X] = BPF_S_ALU_SUB_X, | |
1242 | [BPF_ALU|BPF_MUL|BPF_K] = BPF_S_ALU_MUL_K, | |
1243 | [BPF_ALU|BPF_MUL|BPF_X] = BPF_S_ALU_MUL_X, | |
1244 | [BPF_ALU|BPF_DIV|BPF_X] = BPF_S_ALU_DIV_X, | |
b6069a95 ED |
1245 | [BPF_ALU|BPF_MOD|BPF_K] = BPF_S_ALU_MOD_K, |
1246 | [BPF_ALU|BPF_MOD|BPF_X] = BPF_S_ALU_MOD_X, | |
8c1592d6 ED |
1247 | [BPF_ALU|BPF_AND|BPF_K] = BPF_S_ALU_AND_K, |
1248 | [BPF_ALU|BPF_AND|BPF_X] = BPF_S_ALU_AND_X, | |
1249 | [BPF_ALU|BPF_OR|BPF_K] = BPF_S_ALU_OR_K, | |
1250 | [BPF_ALU|BPF_OR|BPF_X] = BPF_S_ALU_OR_X, | |
9e49e889 DB |
1251 | [BPF_ALU|BPF_XOR|BPF_K] = BPF_S_ALU_XOR_K, |
1252 | [BPF_ALU|BPF_XOR|BPF_X] = BPF_S_ALU_XOR_X, | |
8c1592d6 ED |
1253 | [BPF_ALU|BPF_LSH|BPF_K] = BPF_S_ALU_LSH_K, |
1254 | [BPF_ALU|BPF_LSH|BPF_X] = BPF_S_ALU_LSH_X, | |
1255 | [BPF_ALU|BPF_RSH|BPF_K] = BPF_S_ALU_RSH_K, | |
1256 | [BPF_ALU|BPF_RSH|BPF_X] = BPF_S_ALU_RSH_X, | |
1257 | [BPF_ALU|BPF_NEG] = BPF_S_ALU_NEG, | |
1258 | [BPF_LD|BPF_W|BPF_ABS] = BPF_S_LD_W_ABS, | |
1259 | [BPF_LD|BPF_H|BPF_ABS] = BPF_S_LD_H_ABS, | |
1260 | [BPF_LD|BPF_B|BPF_ABS] = BPF_S_LD_B_ABS, | |
1261 | [BPF_LD|BPF_W|BPF_LEN] = BPF_S_LD_W_LEN, | |
1262 | [BPF_LD|BPF_W|BPF_IND] = BPF_S_LD_W_IND, | |
1263 | [BPF_LD|BPF_H|BPF_IND] = BPF_S_LD_H_IND, | |
1264 | [BPF_LD|BPF_B|BPF_IND] = BPF_S_LD_B_IND, | |
1265 | [BPF_LD|BPF_IMM] = BPF_S_LD_IMM, | |
1266 | [BPF_LDX|BPF_W|BPF_LEN] = BPF_S_LDX_W_LEN, | |
1267 | [BPF_LDX|BPF_B|BPF_MSH] = BPF_S_LDX_B_MSH, | |
1268 | [BPF_LDX|BPF_IMM] = BPF_S_LDX_IMM, | |
1269 | [BPF_MISC|BPF_TAX] = BPF_S_MISC_TAX, | |
1270 | [BPF_MISC|BPF_TXA] = BPF_S_MISC_TXA, | |
1271 | [BPF_RET|BPF_K] = BPF_S_RET_K, | |
1272 | [BPF_RET|BPF_A] = BPF_S_RET_A, | |
1273 | [BPF_ALU|BPF_DIV|BPF_K] = BPF_S_ALU_DIV_K, | |
1274 | [BPF_LD|BPF_MEM] = BPF_S_LD_MEM, | |
1275 | [BPF_LDX|BPF_MEM] = BPF_S_LDX_MEM, | |
1276 | [BPF_ST] = BPF_S_ST, | |
1277 | [BPF_STX] = BPF_S_STX, | |
1278 | [BPF_JMP|BPF_JA] = BPF_S_JMP_JA, | |
1279 | [BPF_JMP|BPF_JEQ|BPF_K] = BPF_S_JMP_JEQ_K, | |
1280 | [BPF_JMP|BPF_JEQ|BPF_X] = BPF_S_JMP_JEQ_X, | |
1281 | [BPF_JMP|BPF_JGE|BPF_K] = BPF_S_JMP_JGE_K, | |
1282 | [BPF_JMP|BPF_JGE|BPF_X] = BPF_S_JMP_JGE_X, | |
1283 | [BPF_JMP|BPF_JGT|BPF_K] = BPF_S_JMP_JGT_K, | |
1284 | [BPF_JMP|BPF_JGT|BPF_X] = BPF_S_JMP_JGT_X, | |
1285 | [BPF_JMP|BPF_JSET|BPF_K] = BPF_S_JMP_JSET_K, | |
1286 | [BPF_JMP|BPF_JSET|BPF_X] = BPF_S_JMP_JSET_X, | |
cba328fc | 1287 | }; |
1da177e4 | 1288 | int pc; |
aa1113d9 | 1289 | bool anc_found; |
1da177e4 | 1290 | |
1b93ae64 | 1291 | if (flen == 0 || flen > BPF_MAXINSNS) |
1da177e4 LT |
1292 | return -EINVAL; |
1293 | ||
1294 | /* check the filter code now */ | |
1295 | for (pc = 0; pc < flen; pc++) { | |
cba328fc TH |
1296 | struct sock_filter *ftest = &filter[pc]; |
1297 | u16 code = ftest->code; | |
93699863 | 1298 | |
cba328fc TH |
1299 | if (code >= ARRAY_SIZE(codes)) |
1300 | return -EINVAL; | |
1301 | code = codes[code]; | |
8c1592d6 | 1302 | if (!code) |
cba328fc | 1303 | return -EINVAL; |
93699863 | 1304 | /* Some instructions need special checks */ |
cba328fc TH |
1305 | switch (code) { |
1306 | case BPF_S_ALU_DIV_K: | |
b6069a95 ED |
1307 | case BPF_S_ALU_MOD_K: |
1308 | /* check for division by zero */ | |
1309 | if (ftest->k == 0) | |
1310 | return -EINVAL; | |
1311 | break; | |
cba328fc TH |
1312 | case BPF_S_LD_MEM: |
1313 | case BPF_S_LDX_MEM: | |
1314 | case BPF_S_ST: | |
1315 | case BPF_S_STX: | |
1316 | /* check for invalid memory addresses */ | |
93699863 KK |
1317 | if (ftest->k >= BPF_MEMWORDS) |
1318 | return -EINVAL; | |
1319 | break; | |
cba328fc | 1320 | case BPF_S_JMP_JA: |
93699863 KK |
1321 | /* |
1322 | * Note, the large ftest->k might cause loops. | |
1323 | * Compare this with conditional jumps below, | |
1324 | * where offsets are limited. --ANK (981016) | |
1325 | */ | |
95c96174 | 1326 | if (ftest->k >= (unsigned int)(flen-pc-1)) |
93699863 | 1327 | return -EINVAL; |
01f2f3f6 | 1328 | break; |
01f2f3f6 HPP |
1329 | case BPF_S_JMP_JEQ_K: |
1330 | case BPF_S_JMP_JEQ_X: | |
1331 | case BPF_S_JMP_JGE_K: | |
1332 | case BPF_S_JMP_JGE_X: | |
1333 | case BPF_S_JMP_JGT_K: | |
1334 | case BPF_S_JMP_JGT_X: | |
1335 | case BPF_S_JMP_JSET_X: | |
1336 | case BPF_S_JMP_JSET_K: | |
cba328fc | 1337 | /* for conditionals both must be safe */ |
e35bedf3 | 1338 | if (pc + ftest->jt + 1 >= flen || |
93699863 KK |
1339 | pc + ftest->jf + 1 >= flen) |
1340 | return -EINVAL; | |
cba328fc | 1341 | break; |
12b16dad ED |
1342 | case BPF_S_LD_W_ABS: |
1343 | case BPF_S_LD_H_ABS: | |
1344 | case BPF_S_LD_B_ABS: | |
aa1113d9 | 1345 | anc_found = false; |
12b16dad ED |
1346 | #define ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \ |
1347 | code = BPF_S_ANC_##CODE; \ | |
aa1113d9 | 1348 | anc_found = true; \ |
12b16dad ED |
1349 | break |
1350 | switch (ftest->k) { | |
1351 | ANCILLARY(PROTOCOL); | |
1352 | ANCILLARY(PKTTYPE); | |
1353 | ANCILLARY(IFINDEX); | |
1354 | ANCILLARY(NLATTR); | |
1355 | ANCILLARY(NLATTR_NEST); | |
1356 | ANCILLARY(MARK); | |
1357 | ANCILLARY(QUEUE); | |
1358 | ANCILLARY(HATYPE); | |
1359 | ANCILLARY(RXHASH); | |
1360 | ANCILLARY(CPU); | |
ffe06c17 | 1361 | ANCILLARY(ALU_XOR_X); |
f3335031 ED |
1362 | ANCILLARY(VLAN_TAG); |
1363 | ANCILLARY(VLAN_TAG_PRESENT); | |
3e5289d5 | 1364 | ANCILLARY(PAY_OFFSET); |
12b16dad | 1365 | } |
aa1113d9 DB |
1366 | |
1367 | /* ancillary operation unknown or unsupported */ | |
1368 | if (anc_found == false && ftest->k >= SKF_AD_OFF) | |
1369 | return -EINVAL; | |
01f2f3f6 | 1370 | } |
cba328fc | 1371 | ftest->code = code; |
01f2f3f6 | 1372 | } |
93699863 | 1373 | |
01f2f3f6 HPP |
1374 | /* last instruction must be a RET code */ |
1375 | switch (filter[flen - 1].code) { | |
1376 | case BPF_S_RET_K: | |
1377 | case BPF_S_RET_A: | |
2d5311e4 | 1378 | return check_load_and_stores(filter, flen); |
cba328fc TH |
1379 | } |
1380 | return -EINVAL; | |
1da177e4 | 1381 | } |
b715631f | 1382 | EXPORT_SYMBOL(sk_chk_filter); |
1da177e4 | 1383 | |
a3ea269b DB |
1384 | static int sk_store_orig_filter(struct sk_filter *fp, |
1385 | const struct sock_fprog *fprog) | |
1386 | { | |
1387 | unsigned int fsize = sk_filter_proglen(fprog); | |
1388 | struct sock_fprog_kern *fkprog; | |
1389 | ||
1390 | fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL); | |
1391 | if (!fp->orig_prog) | |
1392 | return -ENOMEM; | |
1393 | ||
1394 | fkprog = fp->orig_prog; | |
1395 | fkprog->len = fprog->len; | |
1396 | fkprog->filter = kmemdup(fp->insns, fsize, GFP_KERNEL); | |
1397 | if (!fkprog->filter) { | |
1398 | kfree(fp->orig_prog); | |
1399 | return -ENOMEM; | |
1400 | } | |
1401 | ||
1402 | return 0; | |
1403 | } | |
1404 | ||
1405 | static void sk_release_orig_filter(struct sk_filter *fp) | |
1406 | { | |
1407 | struct sock_fprog_kern *fprog = fp->orig_prog; | |
1408 | ||
1409 | if (fprog) { | |
1410 | kfree(fprog->filter); | |
1411 | kfree(fprog); | |
1412 | } | |
1413 | } | |
1414 | ||
47e958ea | 1415 | /** |
46bcf14f | 1416 | * sk_filter_release_rcu - Release a socket filter by rcu_head |
47e958ea PE |
1417 | * @rcu: rcu_head that contains the sk_filter to free |
1418 | */ | |
fbc907f0 | 1419 | static void sk_filter_release_rcu(struct rcu_head *rcu) |
47e958ea PE |
1420 | { |
1421 | struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu); | |
1422 | ||
a3ea269b | 1423 | sk_release_orig_filter(fp); |
0a14842f | 1424 | bpf_jit_free(fp); |
47e958ea | 1425 | } |
fbc907f0 DB |
1426 | |
1427 | /** | |
1428 | * sk_filter_release - release a socket filter | |
1429 | * @fp: filter to remove | |
1430 | * | |
1431 | * Remove a filter from a socket and release its resources. | |
1432 | */ | |
1433 | static void sk_filter_release(struct sk_filter *fp) | |
1434 | { | |
1435 | if (atomic_dec_and_test(&fp->refcnt)) | |
1436 | call_rcu(&fp->rcu, sk_filter_release_rcu); | |
1437 | } | |
1438 | ||
1439 | void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp) | |
1440 | { | |
1441 | atomic_sub(sk_filter_size(fp->len), &sk->sk_omem_alloc); | |
1442 | sk_filter_release(fp); | |
1443 | } | |
1444 | ||
1445 | void sk_filter_charge(struct sock *sk, struct sk_filter *fp) | |
1446 | { | |
1447 | atomic_inc(&fp->refcnt); | |
1448 | atomic_add(sk_filter_size(fp->len), &sk->sk_omem_alloc); | |
1449 | } | |
47e958ea | 1450 | |
bd4cf0ed AS |
1451 | static struct sk_filter *__sk_migrate_realloc(struct sk_filter *fp, |
1452 | struct sock *sk, | |
1453 | unsigned int len) | |
1454 | { | |
1455 | struct sk_filter *fp_new; | |
1456 | ||
1457 | if (sk == NULL) | |
1458 | return krealloc(fp, len, GFP_KERNEL); | |
1459 | ||
1460 | fp_new = sock_kmalloc(sk, len, GFP_KERNEL); | |
1461 | if (fp_new) { | |
1462 | memcpy(fp_new, fp, sizeof(struct sk_filter)); | |
1463 | /* As we're kepping orig_prog in fp_new along, | |
1464 | * we need to make sure we're not evicting it | |
1465 | * from the old fp. | |
1466 | */ | |
1467 | fp->orig_prog = NULL; | |
1468 | sk_filter_uncharge(sk, fp); | |
1469 | } | |
1470 | ||
1471 | return fp_new; | |
1472 | } | |
1473 | ||
1474 | static struct sk_filter *__sk_migrate_filter(struct sk_filter *fp, | |
1475 | struct sock *sk) | |
1476 | { | |
1477 | struct sock_filter *old_prog; | |
1478 | struct sk_filter *old_fp; | |
1479 | int i, err, new_len, old_len = fp->len; | |
1480 | ||
1481 | /* We are free to overwrite insns et al right here as it | |
1482 | * won't be used at this point in time anymore internally | |
1483 | * after the migration to the internal BPF instruction | |
1484 | * representation. | |
1485 | */ | |
1486 | BUILD_BUG_ON(sizeof(struct sock_filter) != | |
1487 | sizeof(struct sock_filter_int)); | |
1488 | ||
1489 | /* For now, we need to unfiddle BPF_S_* identifiers in place. | |
1490 | * This can sooner or later on be subject to removal, e.g. when | |
1491 | * JITs have been converted. | |
1492 | */ | |
1493 | for (i = 0; i < fp->len; i++) | |
1494 | sk_decode_filter(&fp->insns[i], &fp->insns[i]); | |
1495 | ||
1496 | /* Conversion cannot happen on overlapping memory areas, | |
1497 | * so we need to keep the user BPF around until the 2nd | |
1498 | * pass. At this time, the user BPF is stored in fp->insns. | |
1499 | */ | |
1500 | old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter), | |
1501 | GFP_KERNEL); | |
1502 | if (!old_prog) { | |
1503 | err = -ENOMEM; | |
1504 | goto out_err; | |
1505 | } | |
1506 | ||
1507 | /* 1st pass: calculate the new program length. */ | |
1508 | err = sk_convert_filter(old_prog, old_len, NULL, &new_len); | |
1509 | if (err) | |
1510 | goto out_err_free; | |
1511 | ||
1512 | /* Expand fp for appending the new filter representation. */ | |
1513 | old_fp = fp; | |
1514 | fp = __sk_migrate_realloc(old_fp, sk, sk_filter_size(new_len)); | |
1515 | if (!fp) { | |
1516 | /* The old_fp is still around in case we couldn't | |
1517 | * allocate new memory, so uncharge on that one. | |
1518 | */ | |
1519 | fp = old_fp; | |
1520 | err = -ENOMEM; | |
1521 | goto out_err_free; | |
1522 | } | |
1523 | ||
1524 | fp->bpf_func = sk_run_filter_int_skb; | |
1525 | fp->len = new_len; | |
1526 | ||
1527 | /* 2nd pass: remap sock_filter insns into sock_filter_int insns. */ | |
1528 | err = sk_convert_filter(old_prog, old_len, fp->insnsi, &new_len); | |
1529 | if (err) | |
1530 | /* 2nd sk_convert_filter() can fail only if it fails | |
1531 | * to allocate memory, remapping must succeed. Note, | |
1532 | * that at this time old_fp has already been released | |
1533 | * by __sk_migrate_realloc(). | |
1534 | */ | |
1535 | goto out_err_free; | |
1536 | ||
1537 | kfree(old_prog); | |
1538 | return fp; | |
1539 | ||
1540 | out_err_free: | |
1541 | kfree(old_prog); | |
1542 | out_err: | |
1543 | /* Rollback filter setup. */ | |
1544 | if (sk != NULL) | |
1545 | sk_filter_uncharge(sk, fp); | |
1546 | else | |
1547 | kfree(fp); | |
1548 | return ERR_PTR(err); | |
1549 | } | |
1550 | ||
1551 | static struct sk_filter *__sk_prepare_filter(struct sk_filter *fp, | |
1552 | struct sock *sk) | |
302d6637 JP |
1553 | { |
1554 | int err; | |
1555 | ||
bd4cf0ed | 1556 | fp->bpf_func = NULL; |
f8bbbfc3 | 1557 | fp->jited = 0; |
302d6637 JP |
1558 | |
1559 | err = sk_chk_filter(fp->insns, fp->len); | |
1560 | if (err) | |
bd4cf0ed | 1561 | return ERR_PTR(err); |
302d6637 | 1562 | |
bd4cf0ed AS |
1563 | /* Probe if we can JIT compile the filter and if so, do |
1564 | * the compilation of the filter. | |
1565 | */ | |
302d6637 | 1566 | bpf_jit_compile(fp); |
bd4cf0ed AS |
1567 | |
1568 | /* JIT compiler couldn't process this filter, so do the | |
1569 | * internal BPF translation for the optimized interpreter. | |
1570 | */ | |
1571 | if (!fp->jited) | |
1572 | fp = __sk_migrate_filter(fp, sk); | |
1573 | ||
1574 | return fp; | |
302d6637 JP |
1575 | } |
1576 | ||
1577 | /** | |
1578 | * sk_unattached_filter_create - create an unattached filter | |
1579 | * @fprog: the filter program | |
c6c4b97c | 1580 | * @pfp: the unattached filter that is created |
302d6637 | 1581 | * |
c6c4b97c | 1582 | * Create a filter independent of any socket. We first run some |
302d6637 JP |
1583 | * sanity checks on it to make sure it does not explode on us later. |
1584 | * If an error occurs or there is insufficient memory for the filter | |
1585 | * a negative errno code is returned. On success the return is zero. | |
1586 | */ | |
1587 | int sk_unattached_filter_create(struct sk_filter **pfp, | |
1588 | struct sock_fprog *fprog) | |
1589 | { | |
a3ea269b | 1590 | unsigned int fsize = sk_filter_proglen(fprog); |
302d6637 | 1591 | struct sk_filter *fp; |
302d6637 JP |
1592 | |
1593 | /* Make sure new filter is there and in the right amounts. */ | |
1594 | if (fprog->filter == NULL) | |
1595 | return -EINVAL; | |
1596 | ||
d45ed4a4 | 1597 | fp = kmalloc(sk_filter_size(fprog->len), GFP_KERNEL); |
302d6637 JP |
1598 | if (!fp) |
1599 | return -ENOMEM; | |
a3ea269b | 1600 | |
302d6637 JP |
1601 | memcpy(fp->insns, fprog->filter, fsize); |
1602 | ||
1603 | atomic_set(&fp->refcnt, 1); | |
1604 | fp->len = fprog->len; | |
a3ea269b DB |
1605 | /* Since unattached filters are not copied back to user |
1606 | * space through sk_get_filter(), we do not need to hold | |
1607 | * a copy here, and can spare us the work. | |
1608 | */ | |
1609 | fp->orig_prog = NULL; | |
302d6637 | 1610 | |
bd4cf0ed AS |
1611 | /* __sk_prepare_filter() already takes care of uncharging |
1612 | * memory in case something goes wrong. | |
1613 | */ | |
1614 | fp = __sk_prepare_filter(fp, NULL); | |
1615 | if (IS_ERR(fp)) | |
1616 | return PTR_ERR(fp); | |
302d6637 JP |
1617 | |
1618 | *pfp = fp; | |
1619 | return 0; | |
302d6637 JP |
1620 | } |
1621 | EXPORT_SYMBOL_GPL(sk_unattached_filter_create); | |
1622 | ||
1623 | void sk_unattached_filter_destroy(struct sk_filter *fp) | |
1624 | { | |
1625 | sk_filter_release(fp); | |
1626 | } | |
1627 | EXPORT_SYMBOL_GPL(sk_unattached_filter_destroy); | |
1628 | ||
1da177e4 LT |
1629 | /** |
1630 | * sk_attach_filter - attach a socket filter | |
1631 | * @fprog: the filter program | |
1632 | * @sk: the socket to use | |
1633 | * | |
1634 | * Attach the user's filter code. We first run some sanity checks on | |
1635 | * it to make sure it does not explode on us later. If an error | |
1636 | * occurs or there is insufficient memory for the filter a negative | |
1637 | * errno code is returned. On success the return is zero. | |
1638 | */ | |
1639 | int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk) | |
1640 | { | |
d3904b73 | 1641 | struct sk_filter *fp, *old_fp; |
a3ea269b | 1642 | unsigned int fsize = sk_filter_proglen(fprog); |
d45ed4a4 | 1643 | unsigned int sk_fsize = sk_filter_size(fprog->len); |
1da177e4 LT |
1644 | int err; |
1645 | ||
d59577b6 VB |
1646 | if (sock_flag(sk, SOCK_FILTER_LOCKED)) |
1647 | return -EPERM; | |
1648 | ||
1da177e4 | 1649 | /* Make sure new filter is there and in the right amounts. */ |
e35bedf3 KK |
1650 | if (fprog->filter == NULL) |
1651 | return -EINVAL; | |
1da177e4 | 1652 | |
d45ed4a4 | 1653 | fp = sock_kmalloc(sk, sk_fsize, GFP_KERNEL); |
1da177e4 LT |
1654 | if (!fp) |
1655 | return -ENOMEM; | |
a3ea269b | 1656 | |
1da177e4 | 1657 | if (copy_from_user(fp->insns, fprog->filter, fsize)) { |
d45ed4a4 | 1658 | sock_kfree_s(sk, fp, sk_fsize); |
1da177e4 LT |
1659 | return -EFAULT; |
1660 | } | |
1661 | ||
1662 | atomic_set(&fp->refcnt, 1); | |
1663 | fp->len = fprog->len; | |
1664 | ||
a3ea269b DB |
1665 | err = sk_store_orig_filter(fp, fprog); |
1666 | if (err) { | |
1667 | sk_filter_uncharge(sk, fp); | |
1668 | return -ENOMEM; | |
1669 | } | |
1670 | ||
bd4cf0ed AS |
1671 | /* __sk_prepare_filter() already takes care of uncharging |
1672 | * memory in case something goes wrong. | |
1673 | */ | |
1674 | fp = __sk_prepare_filter(fp, sk); | |
1675 | if (IS_ERR(fp)) | |
1676 | return PTR_ERR(fp); | |
1da177e4 | 1677 | |
f91ff5b9 ED |
1678 | old_fp = rcu_dereference_protected(sk->sk_filter, |
1679 | sock_owned_by_user(sk)); | |
d3904b73 | 1680 | rcu_assign_pointer(sk->sk_filter, fp); |
d3904b73 | 1681 | |
9b013e05 | 1682 | if (old_fp) |
46bcf14f | 1683 | sk_filter_uncharge(sk, old_fp); |
a3ea269b | 1684 | |
d3904b73 | 1685 | return 0; |
1da177e4 | 1686 | } |
5ff3f073 | 1687 | EXPORT_SYMBOL_GPL(sk_attach_filter); |
1da177e4 | 1688 | |
55b33325 PE |
1689 | int sk_detach_filter(struct sock *sk) |
1690 | { | |
1691 | int ret = -ENOENT; | |
1692 | struct sk_filter *filter; | |
1693 | ||
d59577b6 VB |
1694 | if (sock_flag(sk, SOCK_FILTER_LOCKED)) |
1695 | return -EPERM; | |
1696 | ||
f91ff5b9 ED |
1697 | filter = rcu_dereference_protected(sk->sk_filter, |
1698 | sock_owned_by_user(sk)); | |
55b33325 | 1699 | if (filter) { |
a9b3cd7f | 1700 | RCU_INIT_POINTER(sk->sk_filter, NULL); |
46bcf14f | 1701 | sk_filter_uncharge(sk, filter); |
55b33325 PE |
1702 | ret = 0; |
1703 | } | |
a3ea269b | 1704 | |
55b33325 PE |
1705 | return ret; |
1706 | } | |
5ff3f073 | 1707 | EXPORT_SYMBOL_GPL(sk_detach_filter); |
a8fc9277 | 1708 | |
ed13998c | 1709 | void sk_decode_filter(struct sock_filter *filt, struct sock_filter *to) |
a8fc9277 PE |
1710 | { |
1711 | static const u16 decodes[] = { | |
1712 | [BPF_S_ALU_ADD_K] = BPF_ALU|BPF_ADD|BPF_K, | |
1713 | [BPF_S_ALU_ADD_X] = BPF_ALU|BPF_ADD|BPF_X, | |
1714 | [BPF_S_ALU_SUB_K] = BPF_ALU|BPF_SUB|BPF_K, | |
1715 | [BPF_S_ALU_SUB_X] = BPF_ALU|BPF_SUB|BPF_X, | |
1716 | [BPF_S_ALU_MUL_K] = BPF_ALU|BPF_MUL|BPF_K, | |
1717 | [BPF_S_ALU_MUL_X] = BPF_ALU|BPF_MUL|BPF_X, | |
1718 | [BPF_S_ALU_DIV_X] = BPF_ALU|BPF_DIV|BPF_X, | |
1719 | [BPF_S_ALU_MOD_K] = BPF_ALU|BPF_MOD|BPF_K, | |
1720 | [BPF_S_ALU_MOD_X] = BPF_ALU|BPF_MOD|BPF_X, | |
1721 | [BPF_S_ALU_AND_K] = BPF_ALU|BPF_AND|BPF_K, | |
1722 | [BPF_S_ALU_AND_X] = BPF_ALU|BPF_AND|BPF_X, | |
1723 | [BPF_S_ALU_OR_K] = BPF_ALU|BPF_OR|BPF_K, | |
1724 | [BPF_S_ALU_OR_X] = BPF_ALU|BPF_OR|BPF_X, | |
1725 | [BPF_S_ALU_XOR_K] = BPF_ALU|BPF_XOR|BPF_K, | |
1726 | [BPF_S_ALU_XOR_X] = BPF_ALU|BPF_XOR|BPF_X, | |
1727 | [BPF_S_ALU_LSH_K] = BPF_ALU|BPF_LSH|BPF_K, | |
1728 | [BPF_S_ALU_LSH_X] = BPF_ALU|BPF_LSH|BPF_X, | |
1729 | [BPF_S_ALU_RSH_K] = BPF_ALU|BPF_RSH|BPF_K, | |
1730 | [BPF_S_ALU_RSH_X] = BPF_ALU|BPF_RSH|BPF_X, | |
1731 | [BPF_S_ALU_NEG] = BPF_ALU|BPF_NEG, | |
1732 | [BPF_S_LD_W_ABS] = BPF_LD|BPF_W|BPF_ABS, | |
1733 | [BPF_S_LD_H_ABS] = BPF_LD|BPF_H|BPF_ABS, | |
1734 | [BPF_S_LD_B_ABS] = BPF_LD|BPF_B|BPF_ABS, | |
1735 | [BPF_S_ANC_PROTOCOL] = BPF_LD|BPF_B|BPF_ABS, | |
1736 | [BPF_S_ANC_PKTTYPE] = BPF_LD|BPF_B|BPF_ABS, | |
1737 | [BPF_S_ANC_IFINDEX] = BPF_LD|BPF_B|BPF_ABS, | |
1738 | [BPF_S_ANC_NLATTR] = BPF_LD|BPF_B|BPF_ABS, | |
1739 | [BPF_S_ANC_NLATTR_NEST] = BPF_LD|BPF_B|BPF_ABS, | |
1740 | [BPF_S_ANC_MARK] = BPF_LD|BPF_B|BPF_ABS, | |
1741 | [BPF_S_ANC_QUEUE] = BPF_LD|BPF_B|BPF_ABS, | |
1742 | [BPF_S_ANC_HATYPE] = BPF_LD|BPF_B|BPF_ABS, | |
1743 | [BPF_S_ANC_RXHASH] = BPF_LD|BPF_B|BPF_ABS, | |
1744 | [BPF_S_ANC_CPU] = BPF_LD|BPF_B|BPF_ABS, | |
1745 | [BPF_S_ANC_ALU_XOR_X] = BPF_LD|BPF_B|BPF_ABS, | |
a8fc9277 PE |
1746 | [BPF_S_ANC_VLAN_TAG] = BPF_LD|BPF_B|BPF_ABS, |
1747 | [BPF_S_ANC_VLAN_TAG_PRESENT] = BPF_LD|BPF_B|BPF_ABS, | |
3e5289d5 | 1748 | [BPF_S_ANC_PAY_OFFSET] = BPF_LD|BPF_B|BPF_ABS, |
a8fc9277 PE |
1749 | [BPF_S_LD_W_LEN] = BPF_LD|BPF_W|BPF_LEN, |
1750 | [BPF_S_LD_W_IND] = BPF_LD|BPF_W|BPF_IND, | |
1751 | [BPF_S_LD_H_IND] = BPF_LD|BPF_H|BPF_IND, | |
1752 | [BPF_S_LD_B_IND] = BPF_LD|BPF_B|BPF_IND, | |
1753 | [BPF_S_LD_IMM] = BPF_LD|BPF_IMM, | |
1754 | [BPF_S_LDX_W_LEN] = BPF_LDX|BPF_W|BPF_LEN, | |
1755 | [BPF_S_LDX_B_MSH] = BPF_LDX|BPF_B|BPF_MSH, | |
1756 | [BPF_S_LDX_IMM] = BPF_LDX|BPF_IMM, | |
1757 | [BPF_S_MISC_TAX] = BPF_MISC|BPF_TAX, | |
1758 | [BPF_S_MISC_TXA] = BPF_MISC|BPF_TXA, | |
1759 | [BPF_S_RET_K] = BPF_RET|BPF_K, | |
1760 | [BPF_S_RET_A] = BPF_RET|BPF_A, | |
1761 | [BPF_S_ALU_DIV_K] = BPF_ALU|BPF_DIV|BPF_K, | |
1762 | [BPF_S_LD_MEM] = BPF_LD|BPF_MEM, | |
1763 | [BPF_S_LDX_MEM] = BPF_LDX|BPF_MEM, | |
1764 | [BPF_S_ST] = BPF_ST, | |
1765 | [BPF_S_STX] = BPF_STX, | |
1766 | [BPF_S_JMP_JA] = BPF_JMP|BPF_JA, | |
1767 | [BPF_S_JMP_JEQ_K] = BPF_JMP|BPF_JEQ|BPF_K, | |
1768 | [BPF_S_JMP_JEQ_X] = BPF_JMP|BPF_JEQ|BPF_X, | |
1769 | [BPF_S_JMP_JGE_K] = BPF_JMP|BPF_JGE|BPF_K, | |
1770 | [BPF_S_JMP_JGE_X] = BPF_JMP|BPF_JGE|BPF_X, | |
1771 | [BPF_S_JMP_JGT_K] = BPF_JMP|BPF_JGT|BPF_K, | |
1772 | [BPF_S_JMP_JGT_X] = BPF_JMP|BPF_JGT|BPF_X, | |
1773 | [BPF_S_JMP_JSET_K] = BPF_JMP|BPF_JSET|BPF_K, | |
1774 | [BPF_S_JMP_JSET_X] = BPF_JMP|BPF_JSET|BPF_X, | |
1775 | }; | |
1776 | u16 code; | |
1777 | ||
1778 | code = filt->code; | |
1779 | ||
1780 | to->code = decodes[code]; | |
1781 | to->jt = filt->jt; | |
1782 | to->jf = filt->jf; | |
aee636c4 | 1783 | to->k = filt->k; |
a8fc9277 PE |
1784 | } |
1785 | ||
a3ea269b DB |
1786 | int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf, |
1787 | unsigned int len) | |
a8fc9277 | 1788 | { |
a3ea269b | 1789 | struct sock_fprog_kern *fprog; |
a8fc9277 | 1790 | struct sk_filter *filter; |
a3ea269b | 1791 | int ret = 0; |
a8fc9277 PE |
1792 | |
1793 | lock_sock(sk); | |
1794 | filter = rcu_dereference_protected(sk->sk_filter, | |
a3ea269b | 1795 | sock_owned_by_user(sk)); |
a8fc9277 PE |
1796 | if (!filter) |
1797 | goto out; | |
a3ea269b DB |
1798 | |
1799 | /* We're copying the filter that has been originally attached, | |
1800 | * so no conversion/decode needed anymore. | |
1801 | */ | |
1802 | fprog = filter->orig_prog; | |
1803 | ||
1804 | ret = fprog->len; | |
a8fc9277 | 1805 | if (!len) |
a3ea269b | 1806 | /* User space only enquires number of filter blocks. */ |
a8fc9277 | 1807 | goto out; |
a3ea269b | 1808 | |
a8fc9277 | 1809 | ret = -EINVAL; |
a3ea269b | 1810 | if (len < fprog->len) |
a8fc9277 PE |
1811 | goto out; |
1812 | ||
1813 | ret = -EFAULT; | |
a3ea269b DB |
1814 | if (copy_to_user(ubuf, fprog->filter, sk_filter_proglen(fprog))) |
1815 | goto out; | |
a8fc9277 | 1816 | |
a3ea269b DB |
1817 | /* Instead of bytes, the API requests to return the number |
1818 | * of filter blocks. | |
1819 | */ | |
1820 | ret = fprog->len; | |
a8fc9277 PE |
1821 | out: |
1822 | release_sock(sk); | |
1823 | return ret; | |
1824 | } |