projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
net: filter: get rid of BPF_S_* enum
[deliverable/linux.git] / net / core / filter.c
1 /*
2 * Linux Socket Filter - Kernel level socket filtering
3 *
4 * Based on the design of the Berkeley Packet Filter. The new
5 * internal format has been designed by PLUMgrid:
6 *
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>
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.
21 * Kris Katterjohn - Added many additional checks in sk_chk_filter()
22 */
23
24 #include <linux/module.h>
25 #include <linux/types.h>
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>
33 #include <linux/gfp.h>
34 #include <net/ip.h>
35 #include <net/protocol.h>
36 #include <net/netlink.h>
37 #include <linux/skbuff.h>
38 #include <net/sock.h>
39 #include <linux/errno.h>
40 #include <linux/timer.h>
41 #include <asm/uaccess.h>
42 #include <asm/unaligned.h>
43 #include <linux/filter.h>
44 #include <linux/ratelimit.h>
45 #include <linux/seccomp.h>
46 #include <linux/if_vlan.h>
47
48 /* Registers */
49 #define BPF_R0 regs[BPF_REG_0]
50 #define BPF_R1 regs[BPF_REG_1]
51 #define BPF_R2 regs[BPF_REG_2]
52 #define BPF_R3 regs[BPF_REG_3]
53 #define BPF_R4 regs[BPF_REG_4]
54 #define BPF_R5 regs[BPF_REG_5]
55 #define BPF_R6 regs[BPF_REG_6]
56 #define BPF_R7 regs[BPF_REG_7]
57 #define BPF_R8 regs[BPF_REG_8]
58 #define BPF_R9 regs[BPF_REG_9]
59 #define BPF_R10 regs[BPF_REG_10]
60
61 /* Named registers */
62 #define A regs[insn->a_reg]
63 #define X regs[insn->x_reg]
64 #define FP regs[BPF_REG_FP]
65 #define ARG1 regs[BPF_REG_ARG1]
66 #define CTX regs[BPF_REG_CTX]
67 #define K insn->imm
68
69 /* No hurry in this branch
70 *
71 * Exported for the bpf jit load helper.
72 */
73 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, int k, unsigned int size)
74 {
75 u8 *ptr = NULL;
76
77 if (k >= SKF_NET_OFF)
78 ptr = skb_network_header(skb) + k - SKF_NET_OFF;
79 else if (k >= SKF_LL_OFF)
80 ptr = skb_mac_header(skb) + k - SKF_LL_OFF;
81 if (ptr >= skb->head && ptr + size <= skb_tail_pointer(skb))
82 return ptr;
83
84 return NULL;
85 }
86
87 static inline void *load_pointer(const struct sk_buff *skb, int k,
88 unsigned int size, void *buffer)
89 {
90 if (k >= 0)
91 return skb_header_pointer(skb, k, size, buffer);
92
93 return bpf_internal_load_pointer_neg_helper(skb, k, size);
94 }
95
96 /**
97 * sk_filter - run a packet through a socket filter
98 * @sk: sock associated with &sk_buff
99 * @skb: buffer to filter
100 *
101 * Run the filter code and then cut skb->data to correct size returned by
102 * sk_run_filter. If pkt_len is 0 we toss packet. If skb->len is smaller
103 * than pkt_len we keep whole skb->data. This is the socket level
104 * wrapper to sk_run_filter. It returns 0 if the packet should
105 * be accepted or -EPERM if the packet should be tossed.
106 *
107 */
108 int sk_filter(struct sock *sk, struct sk_buff *skb)
109 {
110 int err;
111 struct sk_filter *filter;
112
113 /*
114 * If the skb was allocated from pfmemalloc reserves, only
115 * allow SOCK_MEMALLOC sockets to use it as this socket is
116 * helping free memory
117 */
118 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
119 return -ENOMEM;
120
121 err = security_sock_rcv_skb(sk, skb);
122 if (err)
123 return err;
124
125 rcu_read_lock();
126 filter = rcu_dereference(sk->sk_filter);
127 if (filter) {
128 unsigned int pkt_len = SK_RUN_FILTER(filter, skb);
129
130 err = pkt_len ? pskb_trim(skb, pkt_len) : -EPERM;
131 }
132 rcu_read_unlock();
133
134 return err;
135 }
136 EXPORT_SYMBOL(sk_filter);
137
138 /* Base function for offset calculation. Needs to go into .text section,
139 * therefore keeping it non-static as well; will also be used by JITs
140 * anyway later on, so do not let the compiler omit it.
141 */
142 noinline u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
143 {
144 return 0;
145 }
146
147 /**
148 * __sk_run_filter - run a filter on a given context
149 * @ctx: buffer to run the filter on
150 * @insn: filter to apply
151 *
152 * Decode and apply filter instructions to the skb->data. Return length to
153 * keep, 0 for none. @ctx is the data we are operating on, @insn is the
154 * array of filter instructions.
155 */
156 static unsigned int __sk_run_filter(void *ctx, const struct sock_filter_int *insn)
157 {
158 u64 stack[MAX_BPF_STACK / sizeof(u64)];
159 u64 regs[MAX_BPF_REG], tmp;
160 static const void *jumptable[256] = {
161 [0 ... 255] = &&default_label,
162 /* Now overwrite non-defaults ... */
163 /* 32 bit ALU operations */
164 [BPF_ALU | BPF_ADD | BPF_X] = &&ALU_ADD_X,
165 [BPF_ALU | BPF_ADD | BPF_K] = &&ALU_ADD_K,
166 [BPF_ALU | BPF_SUB | BPF_X] = &&ALU_SUB_X,
167 [BPF_ALU | BPF_SUB | BPF_K] = &&ALU_SUB_K,
168 [BPF_ALU | BPF_AND | BPF_X] = &&ALU_AND_X,
169 [BPF_ALU | BPF_AND | BPF_K] = &&ALU_AND_K,
170 [BPF_ALU | BPF_OR | BPF_X] = &&ALU_OR_X,
171 [BPF_ALU | BPF_OR | BPF_K] = &&ALU_OR_K,
172 [BPF_ALU | BPF_LSH | BPF_X] = &&ALU_LSH_X,
173 [BPF_ALU | BPF_LSH | BPF_K] = &&ALU_LSH_K,
174 [BPF_ALU | BPF_RSH | BPF_X] = &&ALU_RSH_X,
175 [BPF_ALU | BPF_RSH | BPF_K] = &&ALU_RSH_K,
176 [BPF_ALU | BPF_XOR | BPF_X] = &&ALU_XOR_X,
177 [BPF_ALU | BPF_XOR | BPF_K] = &&ALU_XOR_K,
178 [BPF_ALU | BPF_MUL | BPF_X] = &&ALU_MUL_X,
179 [BPF_ALU | BPF_MUL | BPF_K] = &&ALU_MUL_K,
180 [BPF_ALU | BPF_MOV | BPF_X] = &&ALU_MOV_X,
181 [BPF_ALU | BPF_MOV | BPF_K] = &&ALU_MOV_K,
182 [BPF_ALU | BPF_DIV | BPF_X] = &&ALU_DIV_X,
183 [BPF_ALU | BPF_DIV | BPF_K] = &&ALU_DIV_K,
184 [BPF_ALU | BPF_MOD | BPF_X] = &&ALU_MOD_X,
185 [BPF_ALU | BPF_MOD | BPF_K] = &&ALU_MOD_K,
186 [BPF_ALU | BPF_NEG] = &&ALU_NEG,
187 [BPF_ALU | BPF_END | BPF_TO_BE] = &&ALU_END_TO_BE,
188 [BPF_ALU | BPF_END | BPF_TO_LE] = &&ALU_END_TO_LE,
189 /* 64 bit ALU operations */
190 [BPF_ALU64 | BPF_ADD | BPF_X] = &&ALU64_ADD_X,
191 [BPF_ALU64 | BPF_ADD | BPF_K] = &&ALU64_ADD_K,
192 [BPF_ALU64 | BPF_SUB | BPF_X] = &&ALU64_SUB_X,
193 [BPF_ALU64 | BPF_SUB | BPF_K] = &&ALU64_SUB_K,
194 [BPF_ALU64 | BPF_AND | BPF_X] = &&ALU64_AND_X,
195 [BPF_ALU64 | BPF_AND | BPF_K] = &&ALU64_AND_K,
196 [BPF_ALU64 | BPF_OR | BPF_X] = &&ALU64_OR_X,
197 [BPF_ALU64 | BPF_OR | BPF_K] = &&ALU64_OR_K,
198 [BPF_ALU64 | BPF_LSH | BPF_X] = &&ALU64_LSH_X,
199 [BPF_ALU64 | BPF_LSH | BPF_K] = &&ALU64_LSH_K,
200 [BPF_ALU64 | BPF_RSH | BPF_X] = &&ALU64_RSH_X,
201 [BPF_ALU64 | BPF_RSH | BPF_K] = &&ALU64_RSH_K,
202 [BPF_ALU64 | BPF_XOR | BPF_X] = &&ALU64_XOR_X,
203 [BPF_ALU64 | BPF_XOR | BPF_K] = &&ALU64_XOR_K,
204 [BPF_ALU64 | BPF_MUL | BPF_X] = &&ALU64_MUL_X,
205 [BPF_ALU64 | BPF_MUL | BPF_K] = &&ALU64_MUL_K,
206 [BPF_ALU64 | BPF_MOV | BPF_X] = &&ALU64_MOV_X,
207 [BPF_ALU64 | BPF_MOV | BPF_K] = &&ALU64_MOV_K,
208 [BPF_ALU64 | BPF_ARSH | BPF_X] = &&ALU64_ARSH_X,
209 [BPF_ALU64 | BPF_ARSH | BPF_K] = &&ALU64_ARSH_K,
210 [BPF_ALU64 | BPF_DIV | BPF_X] = &&ALU64_DIV_X,
211 [BPF_ALU64 | BPF_DIV | BPF_K] = &&ALU64_DIV_K,
212 [BPF_ALU64 | BPF_MOD | BPF_X] = &&ALU64_MOD_X,
213 [BPF_ALU64 | BPF_MOD | BPF_K] = &&ALU64_MOD_K,
214 [BPF_ALU64 | BPF_NEG] = &&ALU64_NEG,
215 /* Call instruction */
216 [BPF_JMP | BPF_CALL] = &&JMP_CALL,
217 /* Jumps */
218 [BPF_JMP | BPF_JA] = &&JMP_JA,
219 [BPF_JMP | BPF_JEQ | BPF_X] = &&JMP_JEQ_X,
220 [BPF_JMP | BPF_JEQ | BPF_K] = &&JMP_JEQ_K,
221 [BPF_JMP | BPF_JNE | BPF_X] = &&JMP_JNE_X,
222 [BPF_JMP | BPF_JNE | BPF_K] = &&JMP_JNE_K,
223 [BPF_JMP | BPF_JGT | BPF_X] = &&JMP_JGT_X,
224 [BPF_JMP | BPF_JGT | BPF_K] = &&JMP_JGT_K,
225 [BPF_JMP | BPF_JGE | BPF_X] = &&JMP_JGE_X,
226 [BPF_JMP | BPF_JGE | BPF_K] = &&JMP_JGE_K,
227 [BPF_JMP | BPF_JSGT | BPF_X] = &&JMP_JSGT_X,
228 [BPF_JMP | BPF_JSGT | BPF_K] = &&JMP_JSGT_K,
229 [BPF_JMP | BPF_JSGE | BPF_X] = &&JMP_JSGE_X,
230 [BPF_JMP | BPF_JSGE | BPF_K] = &&JMP_JSGE_K,
231 [BPF_JMP | BPF_JSET | BPF_X] = &&JMP_JSET_X,
232 [BPF_JMP | BPF_JSET | BPF_K] = &&JMP_JSET_K,
233 /* Program return */
234 [BPF_JMP | BPF_EXIT] = &&JMP_EXIT,
235 /* Store instructions */
236 [BPF_STX | BPF_MEM | BPF_B] = &&STX_MEM_B,
237 [BPF_STX | BPF_MEM | BPF_H] = &&STX_MEM_H,
238 [BPF_STX | BPF_MEM | BPF_W] = &&STX_MEM_W,
239 [BPF_STX | BPF_MEM | BPF_DW] = &&STX_MEM_DW,
240 [BPF_STX | BPF_XADD | BPF_W] = &&STX_XADD_W,
241 [BPF_STX | BPF_XADD | BPF_DW] = &&STX_XADD_DW,
242 [BPF_ST | BPF_MEM | BPF_B] = &&ST_MEM_B,
243 [BPF_ST | BPF_MEM | BPF_H] = &&ST_MEM_H,
244 [BPF_ST | BPF_MEM | BPF_W] = &&ST_MEM_W,
245 [BPF_ST | BPF_MEM | BPF_DW] = &&ST_MEM_DW,
246 /* Load instructions */
247 [BPF_LDX | BPF_MEM | BPF_B] = &&LDX_MEM_B,
248 [BPF_LDX | BPF_MEM | BPF_H] = &&LDX_MEM_H,
249 [BPF_LDX | BPF_MEM | BPF_W] = &&LDX_MEM_W,
250 [BPF_LDX | BPF_MEM | BPF_DW] = &&LDX_MEM_DW,
251 [BPF_LD | BPF_ABS | BPF_W] = &&LD_ABS_W,
252 [BPF_LD | BPF_ABS | BPF_H] = &&LD_ABS_H,
253 [BPF_LD | BPF_ABS | BPF_B] = &&LD_ABS_B,
254 [BPF_LD | BPF_IND | BPF_W] = &&LD_IND_W,
255 [BPF_LD | BPF_IND | BPF_H] = &&LD_IND_H,
256 [BPF_LD | BPF_IND | BPF_B] = &&LD_IND_B,
257 };
258 void *ptr;
259 int off;
260
261 #define CONT ({ insn++; goto select_insn; })
262 #define CONT_JMP ({ insn++; goto select_insn; })
263
264 FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)];
265 ARG1 = (u64) (unsigned long) ctx;
266
267 /* Register for user BPF programs need to be reset first. */
268 regs[BPF_REG_A] = 0;
269 regs[BPF_REG_X] = 0;
270
271 select_insn:
272 goto *jumptable[insn->code];
273
274 /* ALU */
275 #define ALU(OPCODE, OP) \
276 ALU64_##OPCODE##_X: \
277 A = A OP X; \
278 CONT; \
279 ALU_##OPCODE##_X: \
280 A = (u32) A OP (u32) X; \
281 CONT; \
282 ALU64_##OPCODE##_K: \
283 A = A OP K; \
284 CONT; \
285 ALU_##OPCODE##_K: \
286 A = (u32) A OP (u32) K; \
287 CONT;
288
289 ALU(ADD, +)
290 ALU(SUB, -)
291 ALU(AND, &)
292 ALU(OR, |)
293 ALU(LSH, <<)
294 ALU(RSH, >>)
295 ALU(XOR, ^)
296 ALU(MUL, *)
297 #undef ALU
298 ALU_NEG:
299 A = (u32) -A;
300 CONT;
301 ALU64_NEG:
302 A = -A;
303 CONT;
304 ALU_MOV_X:
305 A = (u32) X;
306 CONT;
307 ALU_MOV_K:
308 A = (u32) K;
309 CONT;
310 ALU64_MOV_X:
311 A = X;
312 CONT;
313 ALU64_MOV_K:
314 A = K;
315 CONT;
316 ALU64_ARSH_X:
317 (*(s64 *) &A) >>= X;
318 CONT;
319 ALU64_ARSH_K:
320 (*(s64 *) &A) >>= K;
321 CONT;
322 ALU64_MOD_X:
323 if (unlikely(X == 0))
324 return 0;
325 tmp = A;
326 A = do_div(tmp, X);
327 CONT;
328 ALU_MOD_X:
329 if (unlikely(X == 0))
330 return 0;
331 tmp = (u32) A;
332 A = do_div(tmp, (u32) X);
333 CONT;
334 ALU64_MOD_K:
335 tmp = A;
336 A = do_div(tmp, K);
337 CONT;
338 ALU_MOD_K:
339 tmp = (u32) A;
340 A = do_div(tmp, (u32) K);
341 CONT;
342 ALU64_DIV_X:
343 if (unlikely(X == 0))
344 return 0;
345 do_div(A, X);
346 CONT;
347 ALU_DIV_X:
348 if (unlikely(X == 0))
349 return 0;
350 tmp = (u32) A;
351 do_div(tmp, (u32) X);
352 A = (u32) tmp;
353 CONT;
354 ALU64_DIV_K:
355 do_div(A, K);
356 CONT;
357 ALU_DIV_K:
358 tmp = (u32) A;
359 do_div(tmp, (u32) K);
360 A = (u32) tmp;
361 CONT;
362 ALU_END_TO_BE:
363 switch (K) {
364 case 16:
365 A = (__force u16) cpu_to_be16(A);
366 break;
367 case 32:
368 A = (__force u32) cpu_to_be32(A);
369 break;
370 case 64:
371 A = (__force u64) cpu_to_be64(A);
372 break;
373 }
374 CONT;
375 ALU_END_TO_LE:
376 switch (K) {
377 case 16:
378 A = (__force u16) cpu_to_le16(A);
379 break;
380 case 32:
381 A = (__force u32) cpu_to_le32(A);
382 break;
383 case 64:
384 A = (__force u64) cpu_to_le64(A);
385 break;
386 }
387 CONT;
388
389 /* CALL */
390 JMP_CALL:
391 /* Function call scratches BPF_R1-BPF_R5 registers,
392 * preserves BPF_R6-BPF_R9, and stores return value
393 * into BPF_R0.
394 */
395 BPF_R0 = (__bpf_call_base + insn->imm)(BPF_R1, BPF_R2, BPF_R3,
396 BPF_R4, BPF_R5);
397 CONT;
398
399 /* JMP */
400 JMP_JA:
401 insn += insn->off;
402 CONT;
403 JMP_JEQ_X:
404 if (A == X) {
405 insn += insn->off;
406 CONT_JMP;
407 }
408 CONT;
409 JMP_JEQ_K:
410 if (A == K) {
411 insn += insn->off;
412 CONT_JMP;
413 }
414 CONT;
415 JMP_JNE_X:
416 if (A != X) {
417 insn += insn->off;
418 CONT_JMP;
419 }
420 CONT;
421 JMP_JNE_K:
422 if (A != K) {
423 insn += insn->off;
424 CONT_JMP;
425 }
426 CONT;
427 JMP_JGT_X:
428 if (A > X) {
429 insn += insn->off;
430 CONT_JMP;
431 }
432 CONT;
433 JMP_JGT_K:
434 if (A > K) {
435 insn += insn->off;
436 CONT_JMP;
437 }
438 CONT;
439 JMP_JGE_X:
440 if (A >= X) {
441 insn += insn->off;
442 CONT_JMP;
443 }
444 CONT;
445 JMP_JGE_K:
446 if (A >= K) {
447 insn += insn->off;
448 CONT_JMP;
449 }
450 CONT;
451 JMP_JSGT_X:
452 if (((s64) A) > ((s64) X)) {
453 insn += insn->off;
454 CONT_JMP;
455 }
456 CONT;
457 JMP_JSGT_K:
458 if (((s64) A) > ((s64) K)) {
459 insn += insn->off;
460 CONT_JMP;
461 }
462 CONT;
463 JMP_JSGE_X:
464 if (((s64) A) >= ((s64) X)) {
465 insn += insn->off;
466 CONT_JMP;
467 }
468 CONT;
469 JMP_JSGE_K:
470 if (((s64) A) >= ((s64) K)) {
471 insn += insn->off;
472 CONT_JMP;
473 }
474 CONT;
475 JMP_JSET_X:
476 if (A & X) {
477 insn += insn->off;
478 CONT_JMP;
479 }
480 CONT;
481 JMP_JSET_K:
482 if (A & K) {
483 insn += insn->off;
484 CONT_JMP;
485 }
486 CONT;
487 JMP_EXIT:
488 return BPF_R0;
489
490 /* STX and ST and LDX*/
491 #define LDST(SIZEOP, SIZE) \
492 STX_MEM_##SIZEOP: \
493 *(SIZE *)(unsigned long) (A + insn->off) = X; \
494 CONT; \
495 ST_MEM_##SIZEOP: \
496 *(SIZE *)(unsigned long) (A + insn->off) = K; \
497 CONT; \
498 LDX_MEM_##SIZEOP: \
499 A = *(SIZE *)(unsigned long) (X + insn->off); \
500 CONT;
501
502 LDST(B, u8)
503 LDST(H, u16)
504 LDST(W, u32)
505 LDST(DW, u64)
506 #undef LDST
507 STX_XADD_W: /* lock xadd *(u32 *)(A + insn->off) += X */
508 atomic_add((u32) X, (atomic_t *)(unsigned long)
509 (A + insn->off));
510 CONT;
511 STX_XADD_DW: /* lock xadd *(u64 *)(A + insn->off) += X */
512 atomic64_add((u64) X, (atomic64_t *)(unsigned long)
513 (A + insn->off));
514 CONT;
515 LD_ABS_W: /* BPF_R0 = ntohl(*(u32 *) (skb->data + K)) */
516 off = K;
517 load_word:
518 /* BPF_LD + BPD_ABS and BPF_LD + BPF_IND insns are
519 * only appearing in the programs where ctx ==
520 * skb. All programs keep 'ctx' in regs[BPF_REG_CTX]
521 * == BPF_R6, sk_convert_filter() saves it in BPF_R6,
522 * internal BPF verifier will check that BPF_R6 ==
523 * ctx.
524 *
525 * BPF_ABS and BPF_IND are wrappers of function calls,
526 * so they scratch BPF_R1-BPF_R5 registers, preserve
527 * BPF_R6-BPF_R9, and store return value into BPF_R0.
528 *
529 * Implicit input:
530 * ctx
531 *
532 * Explicit input:
533 * X == any register
534 * K == 32-bit immediate
535 *
536 * Output:
537 * BPF_R0 - 8/16/32-bit skb data converted to cpu endianness
538 */
539
540 ptr = load_pointer((struct sk_buff *) ctx, off, 4, &tmp);
541 if (likely(ptr != NULL)) {
542 BPF_R0 = get_unaligned_be32(ptr);
543 CONT;
544 }
545
546 return 0;
547 LD_ABS_H: /* BPF_R0 = ntohs(*(u16 *) (skb->data + K)) */
548 off = K;
549 load_half:
550 ptr = load_pointer((struct sk_buff *) ctx, off, 2, &tmp);
551 if (likely(ptr != NULL)) {
552 BPF_R0 = get_unaligned_be16(ptr);
553 CONT;
554 }
555
556 return 0;
557 LD_ABS_B: /* BPF_R0 = *(u8 *) (ctx + K) */
558 off = K;
559 load_byte:
560 ptr = load_pointer((struct sk_buff *) ctx, off, 1, &tmp);
561 if (likely(ptr != NULL)) {
562 BPF_R0 = *(u8 *)ptr;
563 CONT;
564 }
565
566 return 0;
567 LD_IND_W: /* BPF_R0 = ntohl(*(u32 *) (skb->data + X + K)) */
568 off = K + X;
569 goto load_word;
570 LD_IND_H: /* BPF_R0 = ntohs(*(u16 *) (skb->data + X + K)) */
571 off = K + X;
572 goto load_half;
573 LD_IND_B: /* BPF_R0 = *(u8 *) (skb->data + X + K) */
574 off = K + X;
575 goto load_byte;
576
577 default_label:
578 /* If we ever reach this, we have a bug somewhere. */
579 WARN_RATELIMIT(1, "unknown opcode %02x\n", insn->code);
580 return 0;
581 }
582
583 /* Helper to find the offset of pkt_type in sk_buff structure. We want
584 * to make sure its still a 3bit field starting at a byte boundary;
585 * taken from arch/x86/net/bpf_jit_comp.c.
586 */
587 #define PKT_TYPE_MAX 7
588 static unsigned int pkt_type_offset(void)
589 {
590 struct sk_buff skb_probe = { .pkt_type = ~0, };
591 u8 *ct = (u8 *) &skb_probe;
592 unsigned int off;
593
594 for (off = 0; off < sizeof(struct sk_buff); off++) {
595 if (ct[off] == PKT_TYPE_MAX)
596 return off;
597 }
598
599 pr_err_once("Please fix %s, as pkt_type couldn't be found!\n", __func__);
600 return -1;
601 }
602
603 static u64 __skb_get_pay_offset(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
604 {
605 return __skb_get_poff((struct sk_buff *)(unsigned long) ctx);
606 }
607
608 static u64 __skb_get_nlattr(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
609 {
610 struct sk_buff *skb = (struct sk_buff *)(unsigned long) ctx;
611 struct nlattr *nla;
612
613 if (skb_is_nonlinear(skb))
614 return 0;
615
616 if (skb->len < sizeof(struct nlattr))
617 return 0;
618
619 if (a > skb->len - sizeof(struct nlattr))
620 return 0;
621
622 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
623 if (nla)
624 return (void *) nla - (void *) skb->data;
625
626 return 0;
627 }
628
629 static u64 __skb_get_nlattr_nest(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
630 {
631 struct sk_buff *skb = (struct sk_buff *)(unsigned long) ctx;
632 struct nlattr *nla;
633
634 if (skb_is_nonlinear(skb))
635 return 0;
636
637 if (skb->len < sizeof(struct nlattr))
638 return 0;
639
640 if (a > skb->len - sizeof(struct nlattr))
641 return 0;
642
643 nla = (struct nlattr *) &skb->data[a];
644 if (nla->nla_len > skb->len - a)
645 return 0;
646
647 nla = nla_find_nested(nla, x);
648 if (nla)
649 return (void *) nla - (void *) skb->data;
650
651 return 0;
652 }
653
654 static u64 __get_raw_cpu_id(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
655 {
656 return raw_smp_processor_id();
657 }
658
659 /* note that this only generates 32-bit random numbers */
660 static u64 __get_random_u32(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
661 {
662 return prandom_u32();
663 }
664
665 static bool convert_bpf_extensions(struct sock_filter *fp,
666 struct sock_filter_int **insnp)
667 {
668 struct sock_filter_int *insn = *insnp;
669
670 switch (fp->k) {
671 case SKF_AD_OFF + SKF_AD_PROTOCOL:
672 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
673
674 /* A = *(u16 *) (ctx + offsetof(protocol)) */
675 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
676 offsetof(struct sk_buff, protocol));
677 insn++;
678
679 /* A = ntohs(A) [emitting a nop or swap16] */
680 insn->code = BPF_ALU | BPF_END | BPF_FROM_BE;
681 insn->a_reg = BPF_REG_A;
682 insn->imm = 16;
683 break;
684
685 case SKF_AD_OFF + SKF_AD_PKTTYPE:
686 *insn = BPF_LDX_MEM(BPF_B, BPF_REG_A, BPF_REG_CTX,
687 pkt_type_offset());
688 if (insn->off < 0)
689 return false;
690 insn++;
691
692 *insn = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, PKT_TYPE_MAX);
693 break;
694
695 case SKF_AD_OFF + SKF_AD_IFINDEX:
696 case SKF_AD_OFF + SKF_AD_HATYPE:
697 *insn = BPF_LDX_MEM(size_to_bpf(FIELD_SIZEOF(struct sk_buff, dev)),
698 BPF_REG_TMP, BPF_REG_CTX,
699 offsetof(struct sk_buff, dev));
700 insn++;
701
702 /* if (tmp != 0) goto pc+1 */
703 *insn = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
704 insn++;
705
706 *insn = BPF_EXIT_INSN();
707 insn++;
708
709 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
710 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, type) != 2);
711
712 insn->a_reg = BPF_REG_A;
713 insn->x_reg = BPF_REG_TMP;
714
715 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX) {
716 insn->code = BPF_LDX | BPF_MEM | BPF_W;
717 insn->off = offsetof(struct net_device, ifindex);
718 } else {
719 insn->code = BPF_LDX | BPF_MEM | BPF_H;
720 insn->off = offsetof(struct net_device, type);
721 }
722 break;
723
724 case SKF_AD_OFF + SKF_AD_MARK:
725 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
726
727 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
728 offsetof(struct sk_buff, mark));
729 break;
730
731 case SKF_AD_OFF + SKF_AD_RXHASH:
732 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
733
734 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
735 offsetof(struct sk_buff, hash));
736 break;
737
738 case SKF_AD_OFF + SKF_AD_QUEUE:
739 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2);
740
741 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
742 offsetof(struct sk_buff, queue_mapping));
743 break;
744
745 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
746 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
747 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2);
748
749 /* A = *(u16 *) (ctx + offsetof(vlan_tci)) */
750 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
751 offsetof(struct sk_buff, vlan_tci));
752 insn++;
753
754 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000);
755
756 if (fp->k == SKF_AD_OFF + SKF_AD_VLAN_TAG) {
757 *insn = BPF_ALU32_IMM(BPF_AND, BPF_REG_A,
758 ~VLAN_TAG_PRESENT);
759 } else {
760 /* A >>= 12 */
761 *insn = BPF_ALU32_IMM(BPF_RSH, BPF_REG_A, 12);
762 insn++;
763
764 /* A &= 1 */
765 *insn = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 1);
766 }
767 break;
768
769 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
770 case SKF_AD_OFF + SKF_AD_NLATTR:
771 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
772 case SKF_AD_OFF + SKF_AD_CPU:
773 case SKF_AD_OFF + SKF_AD_RANDOM:
774 /* arg1 = ctx */
775 *insn = BPF_ALU64_REG(BPF_MOV, BPF_REG_ARG1, BPF_REG_CTX);
776 insn++;
777
778 /* arg2 = A */
779 *insn = BPF_ALU64_REG(BPF_MOV, BPF_REG_ARG2, BPF_REG_A);
780 insn++;
781
782 /* arg3 = X */
783 *insn = BPF_ALU64_REG(BPF_MOV, BPF_REG_ARG3, BPF_REG_X);
784 insn++;
785
786 /* Emit call(ctx, arg2=A, arg3=X) */
787 insn->code = BPF_JMP | BPF_CALL;
788 switch (fp->k) {
789 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
790 insn->imm = __skb_get_pay_offset - __bpf_call_base;
791 break;
792 case SKF_AD_OFF + SKF_AD_NLATTR:
793 insn->imm = __skb_get_nlattr - __bpf_call_base;
794 break;
795 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
796 insn->imm = __skb_get_nlattr_nest - __bpf_call_base;
797 break;
798 case SKF_AD_OFF + SKF_AD_CPU:
799 insn->imm = __get_raw_cpu_id - __bpf_call_base;
800 break;
801 case SKF_AD_OFF + SKF_AD_RANDOM:
802 insn->imm = __get_random_u32 - __bpf_call_base;
803 break;
804 }
805 break;
806
807 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
808 /* A ^= X */
809 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
810 break;
811
812 default:
813 /* This is just a dummy call to avoid letting the compiler
814 * evict __bpf_call_base() as an optimization. Placed here
815 * where no-one bothers.
816 */
817 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
818 return false;
819 }
820
821 *insnp = insn;
822 return true;
823 }
824
825 /**
826 * sk_convert_filter - convert filter program
827 * @prog: the user passed filter program
828 * @len: the length of the user passed filter program
829 * @new_prog: buffer where converted program will be stored
830 * @new_len: pointer to store length of converted program
831 *
832 * Remap 'sock_filter' style BPF instruction set to 'sock_filter_ext' style.
833 * Conversion workflow:
834 *
835 * 1) First pass for calculating the new program length:
836 * sk_convert_filter(old_prog, old_len, NULL, &new_len)
837 *
838 * 2) 2nd pass to remap in two passes: 1st pass finds new
839 * jump offsets, 2nd pass remapping:
840 * new_prog = kmalloc(sizeof(struct sock_filter_int) * new_len);
841 * sk_convert_filter(old_prog, old_len, new_prog, &new_len);
842 *
843 * User BPF's register A is mapped to our BPF register 6, user BPF
844 * register X is mapped to BPF register 7; frame pointer is always
845 * register 10; Context 'void *ctx' is stored in register 1, that is,
846 * for socket filters: ctx == 'struct sk_buff *', for seccomp:
847 * ctx == 'struct seccomp_data *'.
848 */
849 int sk_convert_filter(struct sock_filter *prog, int len,
850 struct sock_filter_int *new_prog, int *new_len)
851 {
852 int new_flen = 0, pass = 0, target, i;
853 struct sock_filter_int *new_insn;
854 struct sock_filter *fp;
855 int *addrs = NULL;
856 u8 bpf_src;
857
858 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
859 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
860
861 if (len <= 0 || len >= BPF_MAXINSNS)
862 return -EINVAL;
863
864 if (new_prog) {
865 addrs = kzalloc(len * sizeof(*addrs), GFP_KERNEL);
866 if (!addrs)
867 return -ENOMEM;
868 }
869
870 do_pass:
871 new_insn = new_prog;
872 fp = prog;
873
874 if (new_insn) {
875 *new_insn = BPF_ALU64_REG(BPF_MOV, BPF_REG_CTX, BPF_REG_ARG1);
876 }
877 new_insn++;
878
879 for (i = 0; i < len; fp++, i++) {
880 struct sock_filter_int tmp_insns[6] = { };
881 struct sock_filter_int *insn = tmp_insns;
882
883 if (addrs)
884 addrs[i] = new_insn - new_prog;
885
886 switch (fp->code) {
887 /* All arithmetic insns and skb loads map as-is. */
888 case BPF_ALU | BPF_ADD | BPF_X:
889 case BPF_ALU | BPF_ADD | BPF_K:
890 case BPF_ALU | BPF_SUB | BPF_X:
891 case BPF_ALU | BPF_SUB | BPF_K:
892 case BPF_ALU | BPF_AND | BPF_X:
893 case BPF_ALU | BPF_AND | BPF_K:
894 case BPF_ALU | BPF_OR | BPF_X:
895 case BPF_ALU | BPF_OR | BPF_K:
896 case BPF_ALU | BPF_LSH | BPF_X:
897 case BPF_ALU | BPF_LSH | BPF_K:
898 case BPF_ALU | BPF_RSH | BPF_X:
899 case BPF_ALU | BPF_RSH | BPF_K:
900 case BPF_ALU | BPF_XOR | BPF_X:
901 case BPF_ALU | BPF_XOR | BPF_K:
902 case BPF_ALU | BPF_MUL | BPF_X:
903 case BPF_ALU | BPF_MUL | BPF_K:
904 case BPF_ALU | BPF_DIV | BPF_X:
905 case BPF_ALU | BPF_DIV | BPF_K:
906 case BPF_ALU | BPF_MOD | BPF_X:
907 case BPF_ALU | BPF_MOD | BPF_K:
908 case BPF_ALU | BPF_NEG:
909 case BPF_LD | BPF_ABS | BPF_W:
910 case BPF_LD | BPF_ABS | BPF_H:
911 case BPF_LD | BPF_ABS | BPF_B:
912 case BPF_LD | BPF_IND | BPF_W:
913 case BPF_LD | BPF_IND | BPF_H:
914 case BPF_LD | BPF_IND | BPF_B:
915 /* Check for overloaded BPF extension and
916 * directly convert it if found, otherwise
917 * just move on with mapping.
918 */
919 if (BPF_CLASS(fp->code) == BPF_LD &&
920 BPF_MODE(fp->code) == BPF_ABS &&
921 convert_bpf_extensions(fp, &insn))
922 break;
923
924 insn->code = fp->code;
925 insn->a_reg = BPF_REG_A;
926 insn->x_reg = BPF_REG_X;
927 insn->imm = fp->k;
928 break;
929
930 /* Jump opcodes map as-is, but offsets need adjustment. */
931 case BPF_JMP | BPF_JA:
932 target = i + fp->k + 1;
933 insn->code = fp->code;
934 #define EMIT_JMP \
935 do { \
936 if (target >= len || target < 0) \
937 goto err; \
938 insn->off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
939 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
940 insn->off -= insn - tmp_insns; \
941 } while (0)
942
943 EMIT_JMP;
944 break;
945
946 case BPF_JMP | BPF_JEQ | BPF_K:
947 case BPF_JMP | BPF_JEQ | BPF_X:
948 case BPF_JMP | BPF_JSET | BPF_K:
949 case BPF_JMP | BPF_JSET | BPF_X:
950 case BPF_JMP | BPF_JGT | BPF_K:
951 case BPF_JMP | BPF_JGT | BPF_X:
952 case BPF_JMP | BPF_JGE | BPF_K:
953 case BPF_JMP | BPF_JGE | BPF_X:
954 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
955 /* BPF immediates are signed, zero extend
956 * immediate into tmp register and use it
957 * in compare insn.
958 */
959 insn->code = BPF_ALU | BPF_MOV | BPF_K;
960 insn->a_reg = BPF_REG_TMP;
961 insn->imm = fp->k;
962 insn++;
963
964 insn->a_reg = BPF_REG_A;
965 insn->x_reg = BPF_REG_TMP;
966 bpf_src = BPF_X;
967 } else {
968 insn->a_reg = BPF_REG_A;
969 insn->x_reg = BPF_REG_X;
970 insn->imm = fp->k;
971 bpf_src = BPF_SRC(fp->code);
972 }
973
974 /* Common case where 'jump_false' is next insn. */
975 if (fp->jf == 0) {
976 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
977 target = i + fp->jt + 1;
978 EMIT_JMP;
979 break;
980 }
981
982 /* Convert JEQ into JNE when 'jump_true' is next insn. */
983 if (fp->jt == 0 && BPF_OP(fp->code) == BPF_JEQ) {
984 insn->code = BPF_JMP | BPF_JNE | bpf_src;
985 target = i + fp->jf + 1;
986 EMIT_JMP;
987 break;
988 }
989
990 /* Other jumps are mapped into two insns: Jxx and JA. */
991 target = i + fp->jt + 1;
992 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
993 EMIT_JMP;
994 insn++;
995
996 insn->code = BPF_JMP | BPF_JA;
997 target = i + fp->jf + 1;
998 EMIT_JMP;
999 break;
1000
1001 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
1002 case BPF_LDX | BPF_MSH | BPF_B:
1003 /* tmp = A */
1004 *insn = BPF_ALU64_REG(BPF_MOV, BPF_REG_TMP, BPF_REG_A);
1005 insn++;
1006
1007 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
1008 *insn = BPF_LD_ABS(BPF_B, fp->k);
1009 insn++;
1010
1011 /* A &= 0xf */
1012 *insn = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
1013 insn++;
1014
1015 /* A <<= 2 */
1016 *insn = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
1017 insn++;
1018
1019 /* X = A */
1020 *insn = BPF_ALU64_REG(BPF_MOV, BPF_REG_X, BPF_REG_A);
1021 insn++;
1022
1023 /* A = tmp */
1024 *insn = BPF_ALU64_REG(BPF_MOV, BPF_REG_A, BPF_REG_TMP);
1025 break;
1026
1027 /* RET_K, RET_A are remaped into 2 insns. */
1028 case BPF_RET | BPF_A:
1029 case BPF_RET | BPF_K:
1030 insn->code = BPF_ALU | BPF_MOV |
1031 (BPF_RVAL(fp->code) == BPF_K ?
1032 BPF_K : BPF_X);
1033 insn->a_reg = 0;
1034 insn->x_reg = BPF_REG_A;
1035 insn->imm = fp->k;
1036 insn++;
1037
1038 *insn = BPF_EXIT_INSN();
1039 break;
1040
1041 /* Store to stack. */
1042 case BPF_ST:
1043 case BPF_STX:
1044 insn->code = BPF_STX | BPF_MEM | BPF_W;
1045 insn->a_reg = BPF_REG_FP;
1046 insn->x_reg = fp->code == BPF_ST ?
1047 BPF_REG_A : BPF_REG_X;
1048 insn->off = -(BPF_MEMWORDS - fp->k) * 4;
1049 break;
1050
1051 /* Load from stack. */
1052 case BPF_LD | BPF_MEM:
1053 case BPF_LDX | BPF_MEM:
1054 insn->code = BPF_LDX | BPF_MEM | BPF_W;
1055 insn->a_reg = BPF_CLASS(fp->code) == BPF_LD ?
1056 BPF_REG_A : BPF_REG_X;
1057 insn->x_reg = BPF_REG_FP;
1058 insn->off = -(BPF_MEMWORDS - fp->k) * 4;
1059 break;
1060
1061 /* A = K or X = K */
1062 case BPF_LD | BPF_IMM:
1063 case BPF_LDX | BPF_IMM:
1064 insn->code = BPF_ALU | BPF_MOV | BPF_K;
1065 insn->a_reg = BPF_CLASS(fp->code) == BPF_LD ?
1066 BPF_REG_A : BPF_REG_X;
1067 insn->imm = fp->k;
1068 break;
1069
1070 /* X = A */
1071 case BPF_MISC | BPF_TAX:
1072 *insn = BPF_ALU64_REG(BPF_MOV, BPF_REG_X, BPF_REG_A);
1073 break;
1074
1075 /* A = X */
1076 case BPF_MISC | BPF_TXA:
1077 *insn = BPF_ALU64_REG(BPF_MOV, BPF_REG_A, BPF_REG_X);
1078 break;
1079
1080 /* A = skb->len or X = skb->len */
1081 case BPF_LD | BPF_W | BPF_LEN:
1082 case BPF_LDX | BPF_W | BPF_LEN:
1083 insn->code = BPF_LDX | BPF_MEM | BPF_W;
1084 insn->a_reg = BPF_CLASS(fp->code) == BPF_LD ?
1085 BPF_REG_A : BPF_REG_X;
1086 insn->x_reg = BPF_REG_CTX;
1087 insn->off = offsetof(struct sk_buff, len);
1088 break;
1089
1090 /* access seccomp_data fields */
1091 case BPF_LDX | BPF_ABS | BPF_W:
1092 /* A = *(u32 *) (ctx + K) */
1093 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
1094 break;
1095
1096 default:
1097 goto err;
1098 }
1099
1100 insn++;
1101 if (new_prog)
1102 memcpy(new_insn, tmp_insns,
1103 sizeof(*insn) * (insn - tmp_insns));
1104
1105 new_insn += insn - tmp_insns;
1106 }
1107
1108 if (!new_prog) {
1109 /* Only calculating new length. */
1110 *new_len = new_insn - new_prog;
1111 return 0;
1112 }
1113
1114 pass++;
1115 if (new_flen != new_insn - new_prog) {
1116 new_flen = new_insn - new_prog;
1117 if (pass > 2)
1118 goto err;
1119
1120 goto do_pass;
1121 }
1122
1123 kfree(addrs);
1124 BUG_ON(*new_len != new_flen);
1125 return 0;
1126 err:
1127 kfree(addrs);
1128 return -EINVAL;
1129 }
1130
1131 /* Security:
1132 *
1133 * A BPF program is able to use 16 cells of memory to store intermediate
1134 * values (check u32 mem[BPF_MEMWORDS] in sk_run_filter()).
1135 *
1136 * As we dont want to clear mem[] array for each packet going through
1137 * sk_run_filter(), we check that filter loaded by user never try to read
1138 * a cell if not previously written, and we check all branches to be sure
1139 * a malicious user doesn't try to abuse us.
1140 */
1141 static int check_load_and_stores(struct sock_filter *filter, int flen)
1142 {
1143 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
1144 int pc, ret = 0;
1145
1146 BUILD_BUG_ON(BPF_MEMWORDS > 16);
1147
1148 masks = kmalloc(flen * sizeof(*masks), GFP_KERNEL);
1149 if (!masks)
1150 return -ENOMEM;
1151
1152 memset(masks, 0xff, flen * sizeof(*masks));
1153
1154 for (pc = 0; pc < flen; pc++) {
1155 memvalid &= masks[pc];
1156
1157 switch (filter[pc].code) {
1158 case BPF_ST:
1159 case BPF_STX:
1160 memvalid |= (1 << filter[pc].k);
1161 break;
1162 case BPF_LD | BPF_MEM:
1163 case BPF_LDX | BPF_MEM:
1164 if (!(memvalid & (1 << filter[pc].k))) {
1165 ret = -EINVAL;
1166 goto error;
1167 }
1168 break;
1169 case BPF_JMP | BPF_JA:
1170 /* A jump must set masks on target */
1171 masks[pc + 1 + filter[pc].k] &= memvalid;
1172 memvalid = ~0;
1173 break;
1174 case BPF_JMP | BPF_JEQ | BPF_K:
1175 case BPF_JMP | BPF_JEQ | BPF_X:
1176 case BPF_JMP | BPF_JGE | BPF_K:
1177 case BPF_JMP | BPF_JGE | BPF_X:
1178 case BPF_JMP | BPF_JGT | BPF_K:
1179 case BPF_JMP | BPF_JGT | BPF_X:
1180 case BPF_JMP | BPF_JSET | BPF_K:
1181 case BPF_JMP | BPF_JSET | BPF_X:
1182 /* A jump must set masks on targets */
1183 masks[pc + 1 + filter[pc].jt] &= memvalid;
1184 masks[pc + 1 + filter[pc].jf] &= memvalid;
1185 memvalid = ~0;
1186 break;
1187 }
1188 }
1189 error:
1190 kfree(masks);
1191 return ret;
1192 }
1193
1194 static bool chk_code_allowed(u16 code_to_probe)
1195 {
1196 static const bool codes[] = {
1197 /* 32 bit ALU operations */
1198 [BPF_ALU | BPF_ADD | BPF_K] = true,
1199 [BPF_ALU | BPF_ADD | BPF_X] = true,
1200 [BPF_ALU | BPF_SUB | BPF_K] = true,
1201 [BPF_ALU | BPF_SUB | BPF_X] = true,
1202 [BPF_ALU | BPF_MUL | BPF_K] = true,
1203 [BPF_ALU | BPF_MUL | BPF_X] = true,
1204 [BPF_ALU | BPF_DIV | BPF_K] = true,
1205 [BPF_ALU | BPF_DIV | BPF_X] = true,
1206 [BPF_ALU | BPF_MOD | BPF_K] = true,
1207 [BPF_ALU | BPF_MOD | BPF_X] = true,
1208 [BPF_ALU | BPF_AND | BPF_K] = true,
1209 [BPF_ALU | BPF_AND | BPF_X] = true,
1210 [BPF_ALU | BPF_OR | BPF_K] = true,
1211 [BPF_ALU | BPF_OR | BPF_X] = true,
1212 [BPF_ALU | BPF_XOR | BPF_K] = true,
1213 [BPF_ALU | BPF_XOR | BPF_X] = true,
1214 [BPF_ALU | BPF_LSH | BPF_K] = true,
1215 [BPF_ALU | BPF_LSH | BPF_X] = true,
1216 [BPF_ALU | BPF_RSH | BPF_K] = true,
1217 [BPF_ALU | BPF_RSH | BPF_X] = true,
1218 [BPF_ALU | BPF_NEG] = true,
1219 /* Load instructions */
1220 [BPF_LD | BPF_W | BPF_ABS] = true,
1221 [BPF_LD | BPF_H | BPF_ABS] = true,
1222 [BPF_LD | BPF_B | BPF_ABS] = true,
1223 [BPF_LD | BPF_W | BPF_LEN] = true,
1224 [BPF_LD | BPF_W | BPF_IND] = true,
1225 [BPF_LD | BPF_H | BPF_IND] = true,
1226 [BPF_LD | BPF_B | BPF_IND] = true,
1227 [BPF_LD | BPF_IMM] = true,
1228 [BPF_LD | BPF_MEM] = true,
1229 [BPF_LDX | BPF_W | BPF_LEN] = true,
1230 [BPF_LDX | BPF_B | BPF_MSH] = true,
1231 [BPF_LDX | BPF_IMM] = true,
1232 [BPF_LDX | BPF_MEM] = true,
1233 /* Store instructions */
1234 [BPF_ST] = true,
1235 [BPF_STX] = true,
1236 /* Misc instructions */
1237 [BPF_MISC | BPF_TAX] = true,
1238 [BPF_MISC | BPF_TXA] = true,
1239 /* Return instructions */
1240 [BPF_RET | BPF_K] = true,
1241 [BPF_RET | BPF_A] = true,
1242 /* Jump instructions */
1243 [BPF_JMP | BPF_JA] = true,
1244 [BPF_JMP | BPF_JEQ | BPF_K] = true,
1245 [BPF_JMP | BPF_JEQ | BPF_X] = true,
1246 [BPF_JMP | BPF_JGE | BPF_K] = true,
1247 [BPF_JMP | BPF_JGE | BPF_X] = true,
1248 [BPF_JMP | BPF_JGT | BPF_K] = true,
1249 [BPF_JMP | BPF_JGT | BPF_X] = true,
1250 [BPF_JMP | BPF_JSET | BPF_K] = true,
1251 [BPF_JMP | BPF_JSET | BPF_X] = true,
1252 };
1253
1254 if (code_to_probe >= ARRAY_SIZE(codes))
1255 return false;
1256
1257 return codes[code_to_probe];
1258 }
1259
1260 /**
1261 * sk_chk_filter - verify socket filter code
1262 * @filter: filter to verify
1263 * @flen: length of filter
1264 *
1265 * Check the user's filter code. If we let some ugly
1266 * filter code slip through kaboom! The filter must contain
1267 * no references or jumps that are out of range, no illegal
1268 * instructions, and must end with a RET instruction.
1269 *
1270 * All jumps are forward as they are not signed.
1271 *
1272 * Returns 0 if the rule set is legal or -EINVAL if not.
1273 */
1274 int sk_chk_filter(struct sock_filter *filter, unsigned int flen)
1275 {
1276 bool anc_found;
1277 int pc;
1278
1279 if (flen == 0 || flen > BPF_MAXINSNS)
1280 return -EINVAL;
1281
1282 /* Check the filter code now */
1283 for (pc = 0; pc < flen; pc++) {
1284 struct sock_filter *ftest = &filter[pc];
1285
1286 /* May we actually operate on this code? */
1287 if (!chk_code_allowed(ftest->code))
1288 return -EINVAL;
1289
1290 /* Some instructions need special checks */
1291 switch (ftest->code) {
1292 case BPF_ALU | BPF_DIV | BPF_K:
1293 case BPF_ALU | BPF_MOD | BPF_K:
1294 /* Check for division by zero */
1295 if (ftest->k == 0)
1296 return -EINVAL;
1297 break;
1298 case BPF_LD | BPF_MEM:
1299 case BPF_LDX | BPF_MEM:
1300 case BPF_ST:
1301 case BPF_STX:
1302 /* Check for invalid memory addresses */
1303 if (ftest->k >= BPF_MEMWORDS)
1304 return -EINVAL;
1305 break;
1306 case BPF_JMP | BPF_JA:
1307 /* Note, the large ftest->k might cause loops.
1308 * Compare this with conditional jumps below,
1309 * where offsets are limited. --ANK (981016)
1310 */
1311 if (ftest->k >= (unsigned int)(flen - pc - 1))
1312 return -EINVAL;
1313 break;
1314 case BPF_JMP | BPF_JEQ | BPF_K:
1315 case BPF_JMP | BPF_JEQ | BPF_X:
1316 case BPF_JMP | BPF_JGE | BPF_K:
1317 case BPF_JMP | BPF_JGE | BPF_X:
1318 case BPF_JMP | BPF_JGT | BPF_K:
1319 case BPF_JMP | BPF_JGT | BPF_X:
1320 case BPF_JMP | BPF_JSET | BPF_K:
1321 case BPF_JMP | BPF_JSET | BPF_X:
1322 /* Both conditionals must be safe */
1323 if (pc + ftest->jt + 1 >= flen ||
1324 pc + ftest->jf + 1 >= flen)
1325 return -EINVAL;
1326 break;
1327 case BPF_LD | BPF_W | BPF_ABS:
1328 case BPF_LD | BPF_H | BPF_ABS:
1329 case BPF_LD | BPF_B | BPF_ABS:
1330 anc_found = false;
1331 if (bpf_anc_helper(ftest) & BPF_ANC)
1332 anc_found = true;
1333 /* Ancillary operation unknown or unsupported */
1334 if (anc_found == false && ftest->k >= SKF_AD_OFF)
1335 return -EINVAL;
1336 }
1337 }
1338
1339 /* Last instruction must be a RET code */
1340 switch (filter[flen - 1].code) {
1341 case BPF_RET | BPF_K:
1342 case BPF_RET | BPF_A:
1343 return check_load_and_stores(filter, flen);
1344 }
1345
1346 return -EINVAL;
1347 }
1348 EXPORT_SYMBOL(sk_chk_filter);
1349
1350 static int sk_store_orig_filter(struct sk_filter *fp,
1351 const struct sock_fprog *fprog)
1352 {
1353 unsigned int fsize = sk_filter_proglen(fprog);
1354 struct sock_fprog_kern *fkprog;
1355
1356 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1357 if (!fp->orig_prog)
1358 return -ENOMEM;
1359
1360 fkprog = fp->orig_prog;
1361 fkprog->len = fprog->len;
1362 fkprog->filter = kmemdup(fp->insns, fsize, GFP_KERNEL);
1363 if (!fkprog->filter) {
1364 kfree(fp->orig_prog);
1365 return -ENOMEM;
1366 }
1367
1368 return 0;
1369 }
1370
1371 static void sk_release_orig_filter(struct sk_filter *fp)
1372 {
1373 struct sock_fprog_kern *fprog = fp->orig_prog;
1374
1375 if (fprog) {
1376 kfree(fprog->filter);
1377 kfree(fprog);
1378 }
1379 }
1380
1381 /**
1382 * sk_filter_release_rcu - Release a socket filter by rcu_head
1383 * @rcu: rcu_head that contains the sk_filter to free
1384 */
1385 static void sk_filter_release_rcu(struct rcu_head *rcu)
1386 {
1387 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1388
1389 sk_release_orig_filter(fp);
1390 sk_filter_free(fp);
1391 }
1392
1393 /**
1394 * sk_filter_release - release a socket filter
1395 * @fp: filter to remove
1396 *
1397 * Remove a filter from a socket and release its resources.
1398 */
1399 static void sk_filter_release(struct sk_filter *fp)
1400 {
1401 if (atomic_dec_and_test(&fp->refcnt))
1402 call_rcu(&fp->rcu, sk_filter_release_rcu);
1403 }
1404
1405 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1406 {
1407 atomic_sub(sk_filter_size(fp->len), &sk->sk_omem_alloc);
1408 sk_filter_release(fp);
1409 }
1410
1411 void sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1412 {
1413 atomic_inc(&fp->refcnt);
1414 atomic_add(sk_filter_size(fp->len), &sk->sk_omem_alloc);
1415 }
1416
1417 static struct sk_filter *__sk_migrate_realloc(struct sk_filter *fp,
1418 struct sock *sk,
1419 unsigned int len)
1420 {
1421 struct sk_filter *fp_new;
1422
1423 if (sk == NULL)
1424 return krealloc(fp, len, GFP_KERNEL);
1425
1426 fp_new = sock_kmalloc(sk, len, GFP_KERNEL);
1427 if (fp_new) {
1428 *fp_new = *fp;
1429 /* As we're kepping orig_prog in fp_new along,
1430 * we need to make sure we're not evicting it
1431 * from the old fp.
1432 */
1433 fp->orig_prog = NULL;
1434 sk_filter_uncharge(sk, fp);
1435 }
1436
1437 return fp_new;
1438 }
1439
1440 static struct sk_filter *__sk_migrate_filter(struct sk_filter *fp,
1441 struct sock *sk)
1442 {
1443 struct sock_filter *old_prog;
1444 struct sk_filter *old_fp;
1445 int err, new_len, old_len = fp->len;
1446
1447 /* We are free to overwrite insns et al right here as it
1448 * won't be used at this point in time anymore internally
1449 * after the migration to the internal BPF instruction
1450 * representation.
1451 */
1452 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1453 sizeof(struct sock_filter_int));
1454
1455 /* Conversion cannot happen on overlapping memory areas,
1456 * so we need to keep the user BPF around until the 2nd
1457 * pass. At this time, the user BPF is stored in fp->insns.
1458 */
1459 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1460 GFP_KERNEL);
1461 if (!old_prog) {
1462 err = -ENOMEM;
1463 goto out_err;
1464 }
1465
1466 /* 1st pass: calculate the new program length. */
1467 err = sk_convert_filter(old_prog, old_len, NULL, &new_len);
1468 if (err)
1469 goto out_err_free;
1470
1471 /* Expand fp for appending the new filter representation. */
1472 old_fp = fp;
1473 fp = __sk_migrate_realloc(old_fp, sk, sk_filter_size(new_len));
1474 if (!fp) {
1475 /* The old_fp is still around in case we couldn't
1476 * allocate new memory, so uncharge on that one.
1477 */
1478 fp = old_fp;
1479 err = -ENOMEM;
1480 goto out_err_free;
1481 }
1482
1483 fp->len = new_len;
1484
1485 /* 2nd pass: remap sock_filter insns into sock_filter_int insns. */
1486 err = sk_convert_filter(old_prog, old_len, fp->insnsi, &new_len);
1487 if (err)
1488 /* 2nd sk_convert_filter() can fail only if it fails
1489 * to allocate memory, remapping must succeed. Note,
1490 * that at this time old_fp has already been released
1491 * by __sk_migrate_realloc().
1492 */
1493 goto out_err_free;
1494
1495 sk_filter_select_runtime(fp);
1496
1497 kfree(old_prog);
1498 return fp;
1499
1500 out_err_free:
1501 kfree(old_prog);
1502 out_err:
1503 /* Rollback filter setup. */
1504 if (sk != NULL)
1505 sk_filter_uncharge(sk, fp);
1506 else
1507 kfree(fp);
1508 return ERR_PTR(err);
1509 }
1510
1511 void __weak bpf_int_jit_compile(struct sk_filter *prog)
1512 {
1513 }
1514
1515 /**
1516 * sk_filter_select_runtime - select execution runtime for BPF program
1517 * @fp: sk_filter populated with internal BPF program
1518 *
1519 * try to JIT internal BPF program, if JIT is not available select interpreter
1520 * BPF program will be executed via SK_RUN_FILTER() macro
1521 */
1522 void sk_filter_select_runtime(struct sk_filter *fp)
1523 {
1524 fp->bpf_func = (void *) __sk_run_filter;
1525
1526 /* Probe if internal BPF can be JITed */
1527 bpf_int_jit_compile(fp);
1528 }
1529 EXPORT_SYMBOL_GPL(sk_filter_select_runtime);
1530
1531 /* free internal BPF program */
1532 void sk_filter_free(struct sk_filter *fp)
1533 {
1534 bpf_jit_free(fp);
1535 }
1536 EXPORT_SYMBOL_GPL(sk_filter_free);
1537
1538 static struct sk_filter *__sk_prepare_filter(struct sk_filter *fp,
1539 struct sock *sk)
1540 {
1541 int err;
1542
1543 fp->bpf_func = NULL;
1544 fp->jited = 0;
1545
1546 err = sk_chk_filter(fp->insns, fp->len);
1547 if (err)
1548 return ERR_PTR(err);
1549
1550 /* Probe if we can JIT compile the filter and if so, do
1551 * the compilation of the filter.
1552 */
1553 bpf_jit_compile(fp);
1554
1555 /* JIT compiler couldn't process this filter, so do the
1556 * internal BPF translation for the optimized interpreter.
1557 */
1558 if (!fp->jited)
1559 fp = __sk_migrate_filter(fp, sk);
1560
1561 return fp;
1562 }
1563
1564 /**
1565 * sk_unattached_filter_create - create an unattached filter
1566 * @fprog: the filter program
1567 * @pfp: the unattached filter that is created
1568 *
1569 * Create a filter independent of any socket. We first run some
1570 * sanity checks on it to make sure it does not explode on us later.
1571 * If an error occurs or there is insufficient memory for the filter
1572 * a negative errno code is returned. On success the return is zero.
1573 */
1574 int sk_unattached_filter_create(struct sk_filter **pfp,
1575 struct sock_fprog_kern *fprog)
1576 {
1577 unsigned int fsize = sk_filter_proglen(fprog);
1578 struct sk_filter *fp;
1579
1580 /* Make sure new filter is there and in the right amounts. */
1581 if (fprog->filter == NULL)
1582 return -EINVAL;
1583
1584 fp = kmalloc(sk_filter_size(fprog->len), GFP_KERNEL);
1585 if (!fp)
1586 return -ENOMEM;
1587
1588 memcpy(fp->insns, fprog->filter, fsize);
1589
1590 atomic_set(&fp->refcnt, 1);
1591 fp->len = fprog->len;
1592 /* Since unattached filters are not copied back to user
1593 * space through sk_get_filter(), we do not need to hold
1594 * a copy here, and can spare us the work.
1595 */
1596 fp->orig_prog = NULL;
1597
1598 /* __sk_prepare_filter() already takes care of uncharging
1599 * memory in case something goes wrong.
1600 */
1601 fp = __sk_prepare_filter(fp, NULL);
1602 if (IS_ERR(fp))
1603 return PTR_ERR(fp);
1604
1605 *pfp = fp;
1606 return 0;
1607 }
1608 EXPORT_SYMBOL_GPL(sk_unattached_filter_create);
1609
1610 void sk_unattached_filter_destroy(struct sk_filter *fp)
1611 {
1612 sk_filter_release(fp);
1613 }
1614 EXPORT_SYMBOL_GPL(sk_unattached_filter_destroy);
1615
1616 /**
1617 * sk_attach_filter - attach a socket filter
1618 * @fprog: the filter program
1619 * @sk: the socket to use
1620 *
1621 * Attach the user's filter code. We first run some sanity checks on
1622 * it to make sure it does not explode on us later. If an error
1623 * occurs or there is insufficient memory for the filter a negative
1624 * errno code is returned. On success the return is zero.
1625 */
1626 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1627 {
1628 struct sk_filter *fp, *old_fp;
1629 unsigned int fsize = sk_filter_proglen(fprog);
1630 unsigned int sk_fsize = sk_filter_size(fprog->len);
1631 int err;
1632
1633 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1634 return -EPERM;
1635
1636 /* Make sure new filter is there and in the right amounts. */
1637 if (fprog->filter == NULL)
1638 return -EINVAL;
1639
1640 fp = sock_kmalloc(sk, sk_fsize, GFP_KERNEL);
1641 if (!fp)
1642 return -ENOMEM;
1643
1644 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1645 sock_kfree_s(sk, fp, sk_fsize);
1646 return -EFAULT;
1647 }
1648
1649 atomic_set(&fp->refcnt, 1);
1650 fp->len = fprog->len;
1651
1652 err = sk_store_orig_filter(fp, fprog);
1653 if (err) {
1654 sk_filter_uncharge(sk, fp);
1655 return -ENOMEM;
1656 }
1657
1658 /* __sk_prepare_filter() already takes care of uncharging
1659 * memory in case something goes wrong.
1660 */
1661 fp = __sk_prepare_filter(fp, sk);
1662 if (IS_ERR(fp))
1663 return PTR_ERR(fp);
1664
1665 old_fp = rcu_dereference_protected(sk->sk_filter,
1666 sock_owned_by_user(sk));
1667 rcu_assign_pointer(sk->sk_filter, fp);
1668
1669 if (old_fp)
1670 sk_filter_uncharge(sk, old_fp);
1671
1672 return 0;
1673 }
1674 EXPORT_SYMBOL_GPL(sk_attach_filter);
1675
1676 int sk_detach_filter(struct sock *sk)
1677 {
1678 int ret = -ENOENT;
1679 struct sk_filter *filter;
1680
1681 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1682 return -EPERM;
1683
1684 filter = rcu_dereference_protected(sk->sk_filter,
1685 sock_owned_by_user(sk));
1686 if (filter) {
1687 RCU_INIT_POINTER(sk->sk_filter, NULL);
1688 sk_filter_uncharge(sk, filter);
1689 ret = 0;
1690 }
1691
1692 return ret;
1693 }
1694 EXPORT_SYMBOL_GPL(sk_detach_filter);
1695
1696 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
1697 unsigned int len)
1698 {
1699 struct sock_fprog_kern *fprog;
1700 struct sk_filter *filter;
1701 int ret = 0;
1702
1703 lock_sock(sk);
1704 filter = rcu_dereference_protected(sk->sk_filter,
1705 sock_owned_by_user(sk));
1706 if (!filter)
1707 goto out;
1708
1709 /* We're copying the filter that has been originally attached,
1710 * so no conversion/decode needed anymore.
1711 */
1712 fprog = filter->orig_prog;
1713
1714 ret = fprog->len;
1715 if (!len)
1716 /* User space only enquires number of filter blocks. */
1717 goto out;
1718
1719 ret = -EINVAL;
1720 if (len < fprog->len)
1721 goto out;
1722
1723 ret = -EFAULT;
1724 if (copy_to_user(ubuf, fprog->filter, sk_filter_proglen(fprog)))
1725 goto out;
1726
1727 /* Instead of bytes, the API requests to return the number
1728 * of filter blocks.
1729 */
1730 ret = fprog->len;
1731 out:
1732 release_sock(sk);
1733 return ret;
1734 }
Linux kernel - Deliverables
This page took 0.067229 seconds and 5 git commands to generate.