projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge ath-next from ath.git
[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 bpf_check_classic()
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 #include <linux/bpf.h>
48
49 /**
50 * sk_filter - run a packet through a socket filter
51 * @sk: sock associated with &sk_buff
52 * @skb: buffer to filter
53 *
54 * Run the filter code and then cut skb->data to correct size returned by
55 * SK_RUN_FILTER. If pkt_len is 0 we toss packet. If skb->len is smaller
56 * than pkt_len we keep whole skb->data. This is the socket level
57 * wrapper to SK_RUN_FILTER. It returns 0 if the packet should
58 * be accepted or -EPERM if the packet should be tossed.
59 *
60 */
61 int sk_filter(struct sock *sk, struct sk_buff *skb)
62 {
63 int err;
64 struct sk_filter *filter;
65
66 /*
67 * If the skb was allocated from pfmemalloc reserves, only
68 * allow SOCK_MEMALLOC sockets to use it as this socket is
69 * helping free memory
70 */
71 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
72 return -ENOMEM;
73
74 err = security_sock_rcv_skb(sk, skb);
75 if (err)
76 return err;
77
78 rcu_read_lock();
79 filter = rcu_dereference(sk->sk_filter);
80 if (filter) {
81 unsigned int pkt_len = SK_RUN_FILTER(filter, skb);
82
83 err = pkt_len ? pskb_trim(skb, pkt_len) : -EPERM;
84 }
85 rcu_read_unlock();
86
87 return err;
88 }
89 EXPORT_SYMBOL(sk_filter);
90
91 static u64 __skb_get_pay_offset(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
92 {
93 return skb_get_poff((struct sk_buff *)(unsigned long) ctx);
94 }
95
96 static u64 __skb_get_nlattr(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
97 {
98 struct sk_buff *skb = (struct sk_buff *)(unsigned long) ctx;
99 struct nlattr *nla;
100
101 if (skb_is_nonlinear(skb))
102 return 0;
103
104 if (skb->len < sizeof(struct nlattr))
105 return 0;
106
107 if (a > skb->len - sizeof(struct nlattr))
108 return 0;
109
110 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
111 if (nla)
112 return (void *) nla - (void *) skb->data;
113
114 return 0;
115 }
116
117 static u64 __skb_get_nlattr_nest(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
118 {
119 struct sk_buff *skb = (struct sk_buff *)(unsigned long) ctx;
120 struct nlattr *nla;
121
122 if (skb_is_nonlinear(skb))
123 return 0;
124
125 if (skb->len < sizeof(struct nlattr))
126 return 0;
127
128 if (a > skb->len - sizeof(struct nlattr))
129 return 0;
130
131 nla = (struct nlattr *) &skb->data[a];
132 if (nla->nla_len > skb->len - a)
133 return 0;
134
135 nla = nla_find_nested(nla, x);
136 if (nla)
137 return (void *) nla - (void *) skb->data;
138
139 return 0;
140 }
141
142 static u64 __get_raw_cpu_id(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
143 {
144 return raw_smp_processor_id();
145 }
146
147 /* note that this only generates 32-bit random numbers */
148 static u64 __get_random_u32(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
149 {
150 return prandom_u32();
151 }
152
153 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
154 struct bpf_insn *insn_buf)
155 {
156 struct bpf_insn *insn = insn_buf;
157
158 switch (skb_field) {
159 case SKF_AD_MARK:
160 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
161
162 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
163 offsetof(struct sk_buff, mark));
164 break;
165
166 case SKF_AD_PKTTYPE:
167 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
168 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
169 #ifdef __BIG_ENDIAN_BITFIELD
170 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
171 #endif
172 break;
173
174 case SKF_AD_QUEUE:
175 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2);
176
177 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
178 offsetof(struct sk_buff, queue_mapping));
179 break;
180
181 case SKF_AD_VLAN_TAG:
182 case SKF_AD_VLAN_TAG_PRESENT:
183 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2);
184 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000);
185
186 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
187 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
188 offsetof(struct sk_buff, vlan_tci));
189 if (skb_field == SKF_AD_VLAN_TAG) {
190 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg,
191 ~VLAN_TAG_PRESENT);
192 } else {
193 /* dst_reg >>= 12 */
194 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 12);
195 /* dst_reg &= 1 */
196 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
197 }
198 break;
199 }
200
201 return insn - insn_buf;
202 }
203
204 static bool convert_bpf_extensions(struct sock_filter *fp,
205 struct bpf_insn **insnp)
206 {
207 struct bpf_insn *insn = *insnp;
208 u32 cnt;
209
210 switch (fp->k) {
211 case SKF_AD_OFF + SKF_AD_PROTOCOL:
212 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
213
214 /* A = *(u16 *) (CTX + offsetof(protocol)) */
215 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
216 offsetof(struct sk_buff, protocol));
217 /* A = ntohs(A) [emitting a nop or swap16] */
218 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
219 break;
220
221 case SKF_AD_OFF + SKF_AD_PKTTYPE:
222 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
223 insn += cnt - 1;
224 break;
225
226 case SKF_AD_OFF + SKF_AD_IFINDEX:
227 case SKF_AD_OFF + SKF_AD_HATYPE:
228 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
229 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, type) != 2);
230 BUILD_BUG_ON(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, dev)) < 0);
231
232 *insn++ = BPF_LDX_MEM(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, dev)),
233 BPF_REG_TMP, BPF_REG_CTX,
234 offsetof(struct sk_buff, dev));
235 /* if (tmp != 0) goto pc + 1 */
236 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
237 *insn++ = BPF_EXIT_INSN();
238 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
239 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
240 offsetof(struct net_device, ifindex));
241 else
242 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
243 offsetof(struct net_device, type));
244 break;
245
246 case SKF_AD_OFF + SKF_AD_MARK:
247 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
248 insn += cnt - 1;
249 break;
250
251 case SKF_AD_OFF + SKF_AD_RXHASH:
252 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
253
254 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
255 offsetof(struct sk_buff, hash));
256 break;
257
258 case SKF_AD_OFF + SKF_AD_QUEUE:
259 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
260 insn += cnt - 1;
261 break;
262
263 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
264 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
265 BPF_REG_A, BPF_REG_CTX, insn);
266 insn += cnt - 1;
267 break;
268
269 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
270 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
271 BPF_REG_A, BPF_REG_CTX, insn);
272 insn += cnt - 1;
273 break;
274
275 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
276 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2);
277
278 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
279 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
280 offsetof(struct sk_buff, vlan_proto));
281 /* A = ntohs(A) [emitting a nop or swap16] */
282 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
283 break;
284
285 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
286 case SKF_AD_OFF + SKF_AD_NLATTR:
287 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
288 case SKF_AD_OFF + SKF_AD_CPU:
289 case SKF_AD_OFF + SKF_AD_RANDOM:
290 /* arg1 = CTX */
291 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
292 /* arg2 = A */
293 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
294 /* arg3 = X */
295 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
296 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
297 switch (fp->k) {
298 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
299 *insn = BPF_EMIT_CALL(__skb_get_pay_offset);
300 break;
301 case SKF_AD_OFF + SKF_AD_NLATTR:
302 *insn = BPF_EMIT_CALL(__skb_get_nlattr);
303 break;
304 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
305 *insn = BPF_EMIT_CALL(__skb_get_nlattr_nest);
306 break;
307 case SKF_AD_OFF + SKF_AD_CPU:
308 *insn = BPF_EMIT_CALL(__get_raw_cpu_id);
309 break;
310 case SKF_AD_OFF + SKF_AD_RANDOM:
311 *insn = BPF_EMIT_CALL(__get_random_u32);
312 break;
313 }
314 break;
315
316 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
317 /* A ^= X */
318 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
319 break;
320
321 default:
322 /* This is just a dummy call to avoid letting the compiler
323 * evict __bpf_call_base() as an optimization. Placed here
324 * where no-one bothers.
325 */
326 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
327 return false;
328 }
329
330 *insnp = insn;
331 return true;
332 }
333
334 /**
335 * bpf_convert_filter - convert filter program
336 * @prog: the user passed filter program
337 * @len: the length of the user passed filter program
338 * @new_prog: buffer where converted program will be stored
339 * @new_len: pointer to store length of converted program
340 *
341 * Remap 'sock_filter' style BPF instruction set to 'sock_filter_ext' style.
342 * Conversion workflow:
343 *
344 * 1) First pass for calculating the new program length:
345 * bpf_convert_filter(old_prog, old_len, NULL, &new_len)
346 *
347 * 2) 2nd pass to remap in two passes: 1st pass finds new
348 * jump offsets, 2nd pass remapping:
349 * new_prog = kmalloc(sizeof(struct bpf_insn) * new_len);
350 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len);
351 *
352 * User BPF's register A is mapped to our BPF register 6, user BPF
353 * register X is mapped to BPF register 7; frame pointer is always
354 * register 10; Context 'void *ctx' is stored in register 1, that is,
355 * for socket filters: ctx == 'struct sk_buff *', for seccomp:
356 * ctx == 'struct seccomp_data *'.
357 */
358 int bpf_convert_filter(struct sock_filter *prog, int len,
359 struct bpf_insn *new_prog, int *new_len)
360 {
361 int new_flen = 0, pass = 0, target, i;
362 struct bpf_insn *new_insn;
363 struct sock_filter *fp;
364 int *addrs = NULL;
365 u8 bpf_src;
366
367 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
368 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
369
370 if (len <= 0 || len > BPF_MAXINSNS)
371 return -EINVAL;
372
373 if (new_prog) {
374 addrs = kcalloc(len, sizeof(*addrs), GFP_KERNEL);
375 if (!addrs)
376 return -ENOMEM;
377 }
378
379 do_pass:
380 new_insn = new_prog;
381 fp = prog;
382
383 if (new_insn)
384 *new_insn = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
385 new_insn++;
386
387 for (i = 0; i < len; fp++, i++) {
388 struct bpf_insn tmp_insns[6] = { };
389 struct bpf_insn *insn = tmp_insns;
390
391 if (addrs)
392 addrs[i] = new_insn - new_prog;
393
394 switch (fp->code) {
395 /* All arithmetic insns and skb loads map as-is. */
396 case BPF_ALU | BPF_ADD | BPF_X:
397 case BPF_ALU | BPF_ADD | BPF_K:
398 case BPF_ALU | BPF_SUB | BPF_X:
399 case BPF_ALU | BPF_SUB | BPF_K:
400 case BPF_ALU | BPF_AND | BPF_X:
401 case BPF_ALU | BPF_AND | BPF_K:
402 case BPF_ALU | BPF_OR | BPF_X:
403 case BPF_ALU | BPF_OR | BPF_K:
404 case BPF_ALU | BPF_LSH | BPF_X:
405 case BPF_ALU | BPF_LSH | BPF_K:
406 case BPF_ALU | BPF_RSH | BPF_X:
407 case BPF_ALU | BPF_RSH | BPF_K:
408 case BPF_ALU | BPF_XOR | BPF_X:
409 case BPF_ALU | BPF_XOR | BPF_K:
410 case BPF_ALU | BPF_MUL | BPF_X:
411 case BPF_ALU | BPF_MUL | BPF_K:
412 case BPF_ALU | BPF_DIV | BPF_X:
413 case BPF_ALU | BPF_DIV | BPF_K:
414 case BPF_ALU | BPF_MOD | BPF_X:
415 case BPF_ALU | BPF_MOD | BPF_K:
416 case BPF_ALU | BPF_NEG:
417 case BPF_LD | BPF_ABS | BPF_W:
418 case BPF_LD | BPF_ABS | BPF_H:
419 case BPF_LD | BPF_ABS | BPF_B:
420 case BPF_LD | BPF_IND | BPF_W:
421 case BPF_LD | BPF_IND | BPF_H:
422 case BPF_LD | BPF_IND | BPF_B:
423 /* Check for overloaded BPF extension and
424 * directly convert it if found, otherwise
425 * just move on with mapping.
426 */
427 if (BPF_CLASS(fp->code) == BPF_LD &&
428 BPF_MODE(fp->code) == BPF_ABS &&
429 convert_bpf_extensions(fp, &insn))
430 break;
431
432 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
433 break;
434
435 /* Jump transformation cannot use BPF block macros
436 * everywhere as offset calculation and target updates
437 * require a bit more work than the rest, i.e. jump
438 * opcodes map as-is, but offsets need adjustment.
439 */
440
441 #define BPF_EMIT_JMP \
442 do { \
443 if (target >= len || target < 0) \
444 goto err; \
445 insn->off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
446 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
447 insn->off -= insn - tmp_insns; \
448 } while (0)
449
450 case BPF_JMP | BPF_JA:
451 target = i + fp->k + 1;
452 insn->code = fp->code;
453 BPF_EMIT_JMP;
454 break;
455
456 case BPF_JMP | BPF_JEQ | BPF_K:
457 case BPF_JMP | BPF_JEQ | BPF_X:
458 case BPF_JMP | BPF_JSET | BPF_K:
459 case BPF_JMP | BPF_JSET | BPF_X:
460 case BPF_JMP | BPF_JGT | BPF_K:
461 case BPF_JMP | BPF_JGT | BPF_X:
462 case BPF_JMP | BPF_JGE | BPF_K:
463 case BPF_JMP | BPF_JGE | BPF_X:
464 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
465 /* BPF immediates are signed, zero extend
466 * immediate into tmp register and use it
467 * in compare insn.
468 */
469 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
470
471 insn->dst_reg = BPF_REG_A;
472 insn->src_reg = BPF_REG_TMP;
473 bpf_src = BPF_X;
474 } else {
475 insn->dst_reg = BPF_REG_A;
476 insn->src_reg = BPF_REG_X;
477 insn->imm = fp->k;
478 bpf_src = BPF_SRC(fp->code);
479 }
480
481 /* Common case where 'jump_false' is next insn. */
482 if (fp->jf == 0) {
483 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
484 target = i + fp->jt + 1;
485 BPF_EMIT_JMP;
486 break;
487 }
488
489 /* Convert JEQ into JNE when 'jump_true' is next insn. */
490 if (fp->jt == 0 && BPF_OP(fp->code) == BPF_JEQ) {
491 insn->code = BPF_JMP | BPF_JNE | bpf_src;
492 target = i + fp->jf + 1;
493 BPF_EMIT_JMP;
494 break;
495 }
496
497 /* Other jumps are mapped into two insns: Jxx and JA. */
498 target = i + fp->jt + 1;
499 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
500 BPF_EMIT_JMP;
501 insn++;
502
503 insn->code = BPF_JMP | BPF_JA;
504 target = i + fp->jf + 1;
505 BPF_EMIT_JMP;
506 break;
507
508 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
509 case BPF_LDX | BPF_MSH | BPF_B:
510 /* tmp = A */
511 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_A);
512 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
513 *insn++ = BPF_LD_ABS(BPF_B, fp->k);
514 /* A &= 0xf */
515 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
516 /* A <<= 2 */
517 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
518 /* X = A */
519 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
520 /* A = tmp */
521 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
522 break;
523
524 /* RET_K, RET_A are remaped into 2 insns. */
525 case BPF_RET | BPF_A:
526 case BPF_RET | BPF_K:
527 *insn++ = BPF_MOV32_RAW(BPF_RVAL(fp->code) == BPF_K ?
528 BPF_K : BPF_X, BPF_REG_0,
529 BPF_REG_A, fp->k);
530 *insn = BPF_EXIT_INSN();
531 break;
532
533 /* Store to stack. */
534 case BPF_ST:
535 case BPF_STX:
536 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
537 BPF_ST ? BPF_REG_A : BPF_REG_X,
538 -(BPF_MEMWORDS - fp->k) * 4);
539 break;
540
541 /* Load from stack. */
542 case BPF_LD | BPF_MEM:
543 case BPF_LDX | BPF_MEM:
544 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
545 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
546 -(BPF_MEMWORDS - fp->k) * 4);
547 break;
548
549 /* A = K or X = K */
550 case BPF_LD | BPF_IMM:
551 case BPF_LDX | BPF_IMM:
552 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
553 BPF_REG_A : BPF_REG_X, fp->k);
554 break;
555
556 /* X = A */
557 case BPF_MISC | BPF_TAX:
558 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
559 break;
560
561 /* A = X */
562 case BPF_MISC | BPF_TXA:
563 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
564 break;
565
566 /* A = skb->len or X = skb->len */
567 case BPF_LD | BPF_W | BPF_LEN:
568 case BPF_LDX | BPF_W | BPF_LEN:
569 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
570 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
571 offsetof(struct sk_buff, len));
572 break;
573
574 /* Access seccomp_data fields. */
575 case BPF_LDX | BPF_ABS | BPF_W:
576 /* A = *(u32 *) (ctx + K) */
577 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
578 break;
579
580 /* Unknown instruction. */
581 default:
582 goto err;
583 }
584
585 insn++;
586 if (new_prog)
587 memcpy(new_insn, tmp_insns,
588 sizeof(*insn) * (insn - tmp_insns));
589 new_insn += insn - tmp_insns;
590 }
591
592 if (!new_prog) {
593 /* Only calculating new length. */
594 *new_len = new_insn - new_prog;
595 return 0;
596 }
597
598 pass++;
599 if (new_flen != new_insn - new_prog) {
600 new_flen = new_insn - new_prog;
601 if (pass > 2)
602 goto err;
603 goto do_pass;
604 }
605
606 kfree(addrs);
607 BUG_ON(*new_len != new_flen);
608 return 0;
609 err:
610 kfree(addrs);
611 return -EINVAL;
612 }
613
614 /* Security:
615 *
616 * As we dont want to clear mem[] array for each packet going through
617 * __bpf_prog_run(), we check that filter loaded by user never try to read
618 * a cell if not previously written, and we check all branches to be sure
619 * a malicious user doesn't try to abuse us.
620 */
621 static int check_load_and_stores(const struct sock_filter *filter, int flen)
622 {
623 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
624 int pc, ret = 0;
625
626 BUILD_BUG_ON(BPF_MEMWORDS > 16);
627
628 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
629 if (!masks)
630 return -ENOMEM;
631
632 memset(masks, 0xff, flen * sizeof(*masks));
633
634 for (pc = 0; pc < flen; pc++) {
635 memvalid &= masks[pc];
636
637 switch (filter[pc].code) {
638 case BPF_ST:
639 case BPF_STX:
640 memvalid |= (1 << filter[pc].k);
641 break;
642 case BPF_LD | BPF_MEM:
643 case BPF_LDX | BPF_MEM:
644 if (!(memvalid & (1 << filter[pc].k))) {
645 ret = -EINVAL;
646 goto error;
647 }
648 break;
649 case BPF_JMP | BPF_JA:
650 /* A jump must set masks on target */
651 masks[pc + 1 + filter[pc].k] &= memvalid;
652 memvalid = ~0;
653 break;
654 case BPF_JMP | BPF_JEQ | BPF_K:
655 case BPF_JMP | BPF_JEQ | BPF_X:
656 case BPF_JMP | BPF_JGE | BPF_K:
657 case BPF_JMP | BPF_JGE | BPF_X:
658 case BPF_JMP | BPF_JGT | BPF_K:
659 case BPF_JMP | BPF_JGT | BPF_X:
660 case BPF_JMP | BPF_JSET | BPF_K:
661 case BPF_JMP | BPF_JSET | BPF_X:
662 /* A jump must set masks on targets */
663 masks[pc + 1 + filter[pc].jt] &= memvalid;
664 masks[pc + 1 + filter[pc].jf] &= memvalid;
665 memvalid = ~0;
666 break;
667 }
668 }
669 error:
670 kfree(masks);
671 return ret;
672 }
673
674 static bool chk_code_allowed(u16 code_to_probe)
675 {
676 static const bool codes[] = {
677 /* 32 bit ALU operations */
678 [BPF_ALU | BPF_ADD | BPF_K] = true,
679 [BPF_ALU | BPF_ADD | BPF_X] = true,
680 [BPF_ALU | BPF_SUB | BPF_K] = true,
681 [BPF_ALU | BPF_SUB | BPF_X] = true,
682 [BPF_ALU | BPF_MUL | BPF_K] = true,
683 [BPF_ALU | BPF_MUL | BPF_X] = true,
684 [BPF_ALU | BPF_DIV | BPF_K] = true,
685 [BPF_ALU | BPF_DIV | BPF_X] = true,
686 [BPF_ALU | BPF_MOD | BPF_K] = true,
687 [BPF_ALU | BPF_MOD | BPF_X] = true,
688 [BPF_ALU | BPF_AND | BPF_K] = true,
689 [BPF_ALU | BPF_AND | BPF_X] = true,
690 [BPF_ALU | BPF_OR | BPF_K] = true,
691 [BPF_ALU | BPF_OR | BPF_X] = true,
692 [BPF_ALU | BPF_XOR | BPF_K] = true,
693 [BPF_ALU | BPF_XOR | BPF_X] = true,
694 [BPF_ALU | BPF_LSH | BPF_K] = true,
695 [BPF_ALU | BPF_LSH | BPF_X] = true,
696 [BPF_ALU | BPF_RSH | BPF_K] = true,
697 [BPF_ALU | BPF_RSH | BPF_X] = true,
698 [BPF_ALU | BPF_NEG] = true,
699 /* Load instructions */
700 [BPF_LD | BPF_W | BPF_ABS] = true,
701 [BPF_LD | BPF_H | BPF_ABS] = true,
702 [BPF_LD | BPF_B | BPF_ABS] = true,
703 [BPF_LD | BPF_W | BPF_LEN] = true,
704 [BPF_LD | BPF_W | BPF_IND] = true,
705 [BPF_LD | BPF_H | BPF_IND] = true,
706 [BPF_LD | BPF_B | BPF_IND] = true,
707 [BPF_LD | BPF_IMM] = true,
708 [BPF_LD | BPF_MEM] = true,
709 [BPF_LDX | BPF_W | BPF_LEN] = true,
710 [BPF_LDX | BPF_B | BPF_MSH] = true,
711 [BPF_LDX | BPF_IMM] = true,
712 [BPF_LDX | BPF_MEM] = true,
713 /* Store instructions */
714 [BPF_ST] = true,
715 [BPF_STX] = true,
716 /* Misc instructions */
717 [BPF_MISC | BPF_TAX] = true,
718 [BPF_MISC | BPF_TXA] = true,
719 /* Return instructions */
720 [BPF_RET | BPF_K] = true,
721 [BPF_RET | BPF_A] = true,
722 /* Jump instructions */
723 [BPF_JMP | BPF_JA] = true,
724 [BPF_JMP | BPF_JEQ | BPF_K] = true,
725 [BPF_JMP | BPF_JEQ | BPF_X] = true,
726 [BPF_JMP | BPF_JGE | BPF_K] = true,
727 [BPF_JMP | BPF_JGE | BPF_X] = true,
728 [BPF_JMP | BPF_JGT | BPF_K] = true,
729 [BPF_JMP | BPF_JGT | BPF_X] = true,
730 [BPF_JMP | BPF_JSET | BPF_K] = true,
731 [BPF_JMP | BPF_JSET | BPF_X] = true,
732 };
733
734 if (code_to_probe >= ARRAY_SIZE(codes))
735 return false;
736
737 return codes[code_to_probe];
738 }
739
740 /**
741 * bpf_check_classic - verify socket filter code
742 * @filter: filter to verify
743 * @flen: length of filter
744 *
745 * Check the user's filter code. If we let some ugly
746 * filter code slip through kaboom! The filter must contain
747 * no references or jumps that are out of range, no illegal
748 * instructions, and must end with a RET instruction.
749 *
750 * All jumps are forward as they are not signed.
751 *
752 * Returns 0 if the rule set is legal or -EINVAL if not.
753 */
754 int bpf_check_classic(const struct sock_filter *filter, unsigned int flen)
755 {
756 bool anc_found;
757 int pc;
758
759 if (flen == 0 || flen > BPF_MAXINSNS)
760 return -EINVAL;
761
762 /* Check the filter code now */
763 for (pc = 0; pc < flen; pc++) {
764 const struct sock_filter *ftest = &filter[pc];
765
766 /* May we actually operate on this code? */
767 if (!chk_code_allowed(ftest->code))
768 return -EINVAL;
769
770 /* Some instructions need special checks */
771 switch (ftest->code) {
772 case BPF_ALU | BPF_DIV | BPF_K:
773 case BPF_ALU | BPF_MOD | BPF_K:
774 /* Check for division by zero */
775 if (ftest->k == 0)
776 return -EINVAL;
777 break;
778 case BPF_LD | BPF_MEM:
779 case BPF_LDX | BPF_MEM:
780 case BPF_ST:
781 case BPF_STX:
782 /* Check for invalid memory addresses */
783 if (ftest->k >= BPF_MEMWORDS)
784 return -EINVAL;
785 break;
786 case BPF_JMP | BPF_JA:
787 /* Note, the large ftest->k might cause loops.
788 * Compare this with conditional jumps below,
789 * where offsets are limited. --ANK (981016)
790 */
791 if (ftest->k >= (unsigned int)(flen - pc - 1))
792 return -EINVAL;
793 break;
794 case BPF_JMP | BPF_JEQ | BPF_K:
795 case BPF_JMP | BPF_JEQ | BPF_X:
796 case BPF_JMP | BPF_JGE | BPF_K:
797 case BPF_JMP | BPF_JGE | BPF_X:
798 case BPF_JMP | BPF_JGT | BPF_K:
799 case BPF_JMP | BPF_JGT | BPF_X:
800 case BPF_JMP | BPF_JSET | BPF_K:
801 case BPF_JMP | BPF_JSET | BPF_X:
802 /* Both conditionals must be safe */
803 if (pc + ftest->jt + 1 >= flen ||
804 pc + ftest->jf + 1 >= flen)
805 return -EINVAL;
806 break;
807 case BPF_LD | BPF_W | BPF_ABS:
808 case BPF_LD | BPF_H | BPF_ABS:
809 case BPF_LD | BPF_B | BPF_ABS:
810 anc_found = false;
811 if (bpf_anc_helper(ftest) & BPF_ANC)
812 anc_found = true;
813 /* Ancillary operation unknown or unsupported */
814 if (anc_found == false && ftest->k >= SKF_AD_OFF)
815 return -EINVAL;
816 }
817 }
818
819 /* Last instruction must be a RET code */
820 switch (filter[flen - 1].code) {
821 case BPF_RET | BPF_K:
822 case BPF_RET | BPF_A:
823 return check_load_and_stores(filter, flen);
824 }
825
826 return -EINVAL;
827 }
828 EXPORT_SYMBOL(bpf_check_classic);
829
830 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
831 const struct sock_fprog *fprog)
832 {
833 unsigned int fsize = bpf_classic_proglen(fprog);
834 struct sock_fprog_kern *fkprog;
835
836 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
837 if (!fp->orig_prog)
838 return -ENOMEM;
839
840 fkprog = fp->orig_prog;
841 fkprog->len = fprog->len;
842 fkprog->filter = kmemdup(fp->insns, fsize, GFP_KERNEL);
843 if (!fkprog->filter) {
844 kfree(fp->orig_prog);
845 return -ENOMEM;
846 }
847
848 return 0;
849 }
850
851 static void bpf_release_orig_filter(struct bpf_prog *fp)
852 {
853 struct sock_fprog_kern *fprog = fp->orig_prog;
854
855 if (fprog) {
856 kfree(fprog->filter);
857 kfree(fprog);
858 }
859 }
860
861 static void __bpf_prog_release(struct bpf_prog *prog)
862 {
863 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
864 bpf_prog_put(prog);
865 } else {
866 bpf_release_orig_filter(prog);
867 bpf_prog_free(prog);
868 }
869 }
870
871 static void __sk_filter_release(struct sk_filter *fp)
872 {
873 __bpf_prog_release(fp->prog);
874 kfree(fp);
875 }
876
877 /**
878 * sk_filter_release_rcu - Release a socket filter by rcu_head
879 * @rcu: rcu_head that contains the sk_filter to free
880 */
881 static void sk_filter_release_rcu(struct rcu_head *rcu)
882 {
883 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
884
885 __sk_filter_release(fp);
886 }
887
888 /**
889 * sk_filter_release - release a socket filter
890 * @fp: filter to remove
891 *
892 * Remove a filter from a socket and release its resources.
893 */
894 static void sk_filter_release(struct sk_filter *fp)
895 {
896 if (atomic_dec_and_test(&fp->refcnt))
897 call_rcu(&fp->rcu, sk_filter_release_rcu);
898 }
899
900 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
901 {
902 u32 filter_size = bpf_prog_size(fp->prog->len);
903
904 atomic_sub(filter_size, &sk->sk_omem_alloc);
905 sk_filter_release(fp);
906 }
907
908 /* try to charge the socket memory if there is space available
909 * return true on success
910 */
911 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
912 {
913 u32 filter_size = bpf_prog_size(fp->prog->len);
914
915 /* same check as in sock_kmalloc() */
916 if (filter_size <= sysctl_optmem_max &&
917 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
918 atomic_inc(&fp->refcnt);
919 atomic_add(filter_size, &sk->sk_omem_alloc);
920 return true;
921 }
922 return false;
923 }
924
925 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
926 {
927 struct sock_filter *old_prog;
928 struct bpf_prog *old_fp;
929 int err, new_len, old_len = fp->len;
930
931 /* We are free to overwrite insns et al right here as it
932 * won't be used at this point in time anymore internally
933 * after the migration to the internal BPF instruction
934 * representation.
935 */
936 BUILD_BUG_ON(sizeof(struct sock_filter) !=
937 sizeof(struct bpf_insn));
938
939 /* Conversion cannot happen on overlapping memory areas,
940 * so we need to keep the user BPF around until the 2nd
941 * pass. At this time, the user BPF is stored in fp->insns.
942 */
943 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
944 GFP_KERNEL);
945 if (!old_prog) {
946 err = -ENOMEM;
947 goto out_err;
948 }
949
950 /* 1st pass: calculate the new program length. */
951 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len);
952 if (err)
953 goto out_err_free;
954
955 /* Expand fp for appending the new filter representation. */
956 old_fp = fp;
957 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
958 if (!fp) {
959 /* The old_fp is still around in case we couldn't
960 * allocate new memory, so uncharge on that one.
961 */
962 fp = old_fp;
963 err = -ENOMEM;
964 goto out_err_free;
965 }
966
967 fp->len = new_len;
968
969 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
970 err = bpf_convert_filter(old_prog, old_len, fp->insnsi, &new_len);
971 if (err)
972 /* 2nd bpf_convert_filter() can fail only if it fails
973 * to allocate memory, remapping must succeed. Note,
974 * that at this time old_fp has already been released
975 * by krealloc().
976 */
977 goto out_err_free;
978
979 bpf_prog_select_runtime(fp);
980
981 kfree(old_prog);
982 return fp;
983
984 out_err_free:
985 kfree(old_prog);
986 out_err:
987 __bpf_prog_release(fp);
988 return ERR_PTR(err);
989 }
990
991 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp)
992 {
993 int err;
994
995 fp->bpf_func = NULL;
996 fp->jited = false;
997
998 err = bpf_check_classic(fp->insns, fp->len);
999 if (err) {
1000 __bpf_prog_release(fp);
1001 return ERR_PTR(err);
1002 }
1003
1004 /* Probe if we can JIT compile the filter and if so, do
1005 * the compilation of the filter.
1006 */
1007 bpf_jit_compile(fp);
1008
1009 /* JIT compiler couldn't process this filter, so do the
1010 * internal BPF translation for the optimized interpreter.
1011 */
1012 if (!fp->jited)
1013 fp = bpf_migrate_filter(fp);
1014
1015 return fp;
1016 }
1017
1018 /**
1019 * bpf_prog_create - create an unattached filter
1020 * @pfp: the unattached filter that is created
1021 * @fprog: the filter program
1022 *
1023 * Create a filter independent of any socket. We first run some
1024 * sanity checks on it to make sure it does not explode on us later.
1025 * If an error occurs or there is insufficient memory for the filter
1026 * a negative errno code is returned. On success the return is zero.
1027 */
1028 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1029 {
1030 unsigned int fsize = bpf_classic_proglen(fprog);
1031 struct bpf_prog *fp;
1032
1033 /* Make sure new filter is there and in the right amounts. */
1034 if (fprog->filter == NULL)
1035 return -EINVAL;
1036
1037 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1038 if (!fp)
1039 return -ENOMEM;
1040
1041 memcpy(fp->insns, fprog->filter, fsize);
1042
1043 fp->len = fprog->len;
1044 /* Since unattached filters are not copied back to user
1045 * space through sk_get_filter(), we do not need to hold
1046 * a copy here, and can spare us the work.
1047 */
1048 fp->orig_prog = NULL;
1049
1050 /* bpf_prepare_filter() already takes care of freeing
1051 * memory in case something goes wrong.
1052 */
1053 fp = bpf_prepare_filter(fp);
1054 if (IS_ERR(fp))
1055 return PTR_ERR(fp);
1056
1057 *pfp = fp;
1058 return 0;
1059 }
1060 EXPORT_SYMBOL_GPL(bpf_prog_create);
1061
1062 void bpf_prog_destroy(struct bpf_prog *fp)
1063 {
1064 __bpf_prog_release(fp);
1065 }
1066 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1067
1068 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1069 {
1070 struct sk_filter *fp, *old_fp;
1071
1072 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1073 if (!fp)
1074 return -ENOMEM;
1075
1076 fp->prog = prog;
1077 atomic_set(&fp->refcnt, 0);
1078
1079 if (!sk_filter_charge(sk, fp)) {
1080 kfree(fp);
1081 return -ENOMEM;
1082 }
1083
1084 old_fp = rcu_dereference_protected(sk->sk_filter,
1085 sock_owned_by_user(sk));
1086 rcu_assign_pointer(sk->sk_filter, fp);
1087
1088 if (old_fp)
1089 sk_filter_uncharge(sk, old_fp);
1090
1091 return 0;
1092 }
1093
1094 /**
1095 * sk_attach_filter - attach a socket filter
1096 * @fprog: the filter program
1097 * @sk: the socket to use
1098 *
1099 * Attach the user's filter code. We first run some sanity checks on
1100 * it to make sure it does not explode on us later. If an error
1101 * occurs or there is insufficient memory for the filter a negative
1102 * errno code is returned. On success the return is zero.
1103 */
1104 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1105 {
1106 unsigned int fsize = bpf_classic_proglen(fprog);
1107 unsigned int bpf_fsize = bpf_prog_size(fprog->len);
1108 struct bpf_prog *prog;
1109 int err;
1110
1111 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1112 return -EPERM;
1113
1114 /* Make sure new filter is there and in the right amounts. */
1115 if (fprog->filter == NULL)
1116 return -EINVAL;
1117
1118 prog = bpf_prog_alloc(bpf_fsize, 0);
1119 if (!prog)
1120 return -ENOMEM;
1121
1122 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1123 __bpf_prog_free(prog);
1124 return -EFAULT;
1125 }
1126
1127 prog->len = fprog->len;
1128
1129 err = bpf_prog_store_orig_filter(prog, fprog);
1130 if (err) {
1131 __bpf_prog_free(prog);
1132 return -ENOMEM;
1133 }
1134
1135 /* bpf_prepare_filter() already takes care of freeing
1136 * memory in case something goes wrong.
1137 */
1138 prog = bpf_prepare_filter(prog);
1139 if (IS_ERR(prog))
1140 return PTR_ERR(prog);
1141
1142 err = __sk_attach_prog(prog, sk);
1143 if (err < 0) {
1144 __bpf_prog_release(prog);
1145 return err;
1146 }
1147
1148 return 0;
1149 }
1150 EXPORT_SYMBOL_GPL(sk_attach_filter);
1151
1152 int sk_attach_bpf(u32 ufd, struct sock *sk)
1153 {
1154 struct bpf_prog *prog;
1155 int err;
1156
1157 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1158 return -EPERM;
1159
1160 prog = bpf_prog_get(ufd);
1161 if (IS_ERR(prog))
1162 return PTR_ERR(prog);
1163
1164 if (prog->type != BPF_PROG_TYPE_SOCKET_FILTER) {
1165 bpf_prog_put(prog);
1166 return -EINVAL;
1167 }
1168
1169 err = __sk_attach_prog(prog, sk);
1170 if (err < 0) {
1171 bpf_prog_put(prog);
1172 return err;
1173 }
1174
1175 return 0;
1176 }
1177
1178 /**
1179 * bpf_skb_clone_not_writable - is the header of a clone not writable
1180 * @skb: buffer to check
1181 * @len: length up to which to write, can be negative
1182 *
1183 * Returns true if modifying the header part of the cloned buffer
1184 * does require the data to be copied. I.e. this version works with
1185 * negative lengths needed for eBPF case!
1186 */
1187 static bool bpf_skb_clone_unwritable(const struct sk_buff *skb, int len)
1188 {
1189 return skb_header_cloned(skb) ||
1190 (int) skb_headroom(skb) + len > skb->hdr_len;
1191 }
1192
1193 #define BPF_RECOMPUTE_CSUM(flags) ((flags) & 1)
1194
1195 static u64 bpf_skb_store_bytes(u64 r1, u64 r2, u64 r3, u64 r4, u64 flags)
1196 {
1197 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1198 int offset = (int) r2;
1199 void *from = (void *) (long) r3;
1200 unsigned int len = (unsigned int) r4;
1201 char buf[16];
1202 void *ptr;
1203
1204 /* bpf verifier guarantees that:
1205 * 'from' pointer points to bpf program stack
1206 * 'len' bytes of it were initialized
1207 * 'len' > 0
1208 * 'skb' is a valid pointer to 'struct sk_buff'
1209 *
1210 * so check for invalid 'offset' and too large 'len'
1211 */
1212 if (unlikely((u32) offset > 0xffff || len > sizeof(buf)))
1213 return -EFAULT;
1214
1215 offset -= skb->data - skb_mac_header(skb);
1216 if (unlikely(skb_cloned(skb) &&
1217 bpf_skb_clone_unwritable(skb, offset + len)))
1218 return -EFAULT;
1219
1220 ptr = skb_header_pointer(skb, offset, len, buf);
1221 if (unlikely(!ptr))
1222 return -EFAULT;
1223
1224 if (BPF_RECOMPUTE_CSUM(flags))
1225 skb_postpull_rcsum(skb, ptr, len);
1226
1227 memcpy(ptr, from, len);
1228
1229 if (ptr == buf)
1230 /* skb_store_bits cannot return -EFAULT here */
1231 skb_store_bits(skb, offset, ptr, len);
1232
1233 if (BPF_RECOMPUTE_CSUM(flags) && skb->ip_summed == CHECKSUM_COMPLETE)
1234 skb->csum = csum_add(skb->csum, csum_partial(ptr, len, 0));
1235 return 0;
1236 }
1237
1238 const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1239 .func = bpf_skb_store_bytes,
1240 .gpl_only = false,
1241 .ret_type = RET_INTEGER,
1242 .arg1_type = ARG_PTR_TO_CTX,
1243 .arg2_type = ARG_ANYTHING,
1244 .arg3_type = ARG_PTR_TO_STACK,
1245 .arg4_type = ARG_CONST_STACK_SIZE,
1246 .arg5_type = ARG_ANYTHING,
1247 };
1248
1249 #define BPF_HEADER_FIELD_SIZE(flags) ((flags) & 0x0f)
1250 #define BPF_IS_PSEUDO_HEADER(flags) ((flags) & 0x10)
1251
1252 static u64 bpf_l3_csum_replace(u64 r1, u64 r2, u64 from, u64 to, u64 flags)
1253 {
1254 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1255 int offset = (int) r2;
1256 __sum16 sum, *ptr;
1257
1258 if (unlikely((u32) offset > 0xffff))
1259 return -EFAULT;
1260
1261 offset -= skb->data - skb_mac_header(skb);
1262 if (unlikely(skb_cloned(skb) &&
1263 bpf_skb_clone_unwritable(skb, offset + sizeof(sum))))
1264 return -EFAULT;
1265
1266 ptr = skb_header_pointer(skb, offset, sizeof(sum), &sum);
1267 if (unlikely(!ptr))
1268 return -EFAULT;
1269
1270 switch (BPF_HEADER_FIELD_SIZE(flags)) {
1271 case 2:
1272 csum_replace2(ptr, from, to);
1273 break;
1274 case 4:
1275 csum_replace4(ptr, from, to);
1276 break;
1277 default:
1278 return -EINVAL;
1279 }
1280
1281 if (ptr == &sum)
1282 /* skb_store_bits guaranteed to not return -EFAULT here */
1283 skb_store_bits(skb, offset, ptr, sizeof(sum));
1284
1285 return 0;
1286 }
1287
1288 const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1289 .func = bpf_l3_csum_replace,
1290 .gpl_only = false,
1291 .ret_type = RET_INTEGER,
1292 .arg1_type = ARG_PTR_TO_CTX,
1293 .arg2_type = ARG_ANYTHING,
1294 .arg3_type = ARG_ANYTHING,
1295 .arg4_type = ARG_ANYTHING,
1296 .arg5_type = ARG_ANYTHING,
1297 };
1298
1299 static u64 bpf_l4_csum_replace(u64 r1, u64 r2, u64 from, u64 to, u64 flags)
1300 {
1301 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1302 u32 is_pseudo = BPF_IS_PSEUDO_HEADER(flags);
1303 int offset = (int) r2;
1304 __sum16 sum, *ptr;
1305
1306 if (unlikely((u32) offset > 0xffff))
1307 return -EFAULT;
1308
1309 offset -= skb->data - skb_mac_header(skb);
1310 if (unlikely(skb_cloned(skb) &&
1311 bpf_skb_clone_unwritable(skb, offset + sizeof(sum))))
1312 return -EFAULT;
1313
1314 ptr = skb_header_pointer(skb, offset, sizeof(sum), &sum);
1315 if (unlikely(!ptr))
1316 return -EFAULT;
1317
1318 switch (BPF_HEADER_FIELD_SIZE(flags)) {
1319 case 2:
1320 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1321 break;
1322 case 4:
1323 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1324 break;
1325 default:
1326 return -EINVAL;
1327 }
1328
1329 if (ptr == &sum)
1330 /* skb_store_bits guaranteed to not return -EFAULT here */
1331 skb_store_bits(skb, offset, ptr, sizeof(sum));
1332
1333 return 0;
1334 }
1335
1336 const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1337 .func = bpf_l4_csum_replace,
1338 .gpl_only = false,
1339 .ret_type = RET_INTEGER,
1340 .arg1_type = ARG_PTR_TO_CTX,
1341 .arg2_type = ARG_ANYTHING,
1342 .arg3_type = ARG_ANYTHING,
1343 .arg4_type = ARG_ANYTHING,
1344 .arg5_type = ARG_ANYTHING,
1345 };
1346
1347 static const struct bpf_func_proto *
1348 sk_filter_func_proto(enum bpf_func_id func_id)
1349 {
1350 switch (func_id) {
1351 case BPF_FUNC_map_lookup_elem:
1352 return &bpf_map_lookup_elem_proto;
1353 case BPF_FUNC_map_update_elem:
1354 return &bpf_map_update_elem_proto;
1355 case BPF_FUNC_map_delete_elem:
1356 return &bpf_map_delete_elem_proto;
1357 case BPF_FUNC_get_prandom_u32:
1358 return &bpf_get_prandom_u32_proto;
1359 case BPF_FUNC_get_smp_processor_id:
1360 return &bpf_get_smp_processor_id_proto;
1361 default:
1362 return NULL;
1363 }
1364 }
1365
1366 static const struct bpf_func_proto *
1367 tc_cls_act_func_proto(enum bpf_func_id func_id)
1368 {
1369 switch (func_id) {
1370 case BPF_FUNC_skb_store_bytes:
1371 return &bpf_skb_store_bytes_proto;
1372 case BPF_FUNC_l3_csum_replace:
1373 return &bpf_l3_csum_replace_proto;
1374 case BPF_FUNC_l4_csum_replace:
1375 return &bpf_l4_csum_replace_proto;
1376 default:
1377 return sk_filter_func_proto(func_id);
1378 }
1379 }
1380
1381 static bool sk_filter_is_valid_access(int off, int size,
1382 enum bpf_access_type type)
1383 {
1384 /* only read is allowed */
1385 if (type != BPF_READ)
1386 return false;
1387
1388 /* check bounds */
1389 if (off < 0 || off >= sizeof(struct __sk_buff))
1390 return false;
1391
1392 /* disallow misaligned access */
1393 if (off % size != 0)
1394 return false;
1395
1396 /* all __sk_buff fields are __u32 */
1397 if (size != 4)
1398 return false;
1399
1400 return true;
1401 }
1402
1403 static u32 sk_filter_convert_ctx_access(int dst_reg, int src_reg, int ctx_off,
1404 struct bpf_insn *insn_buf)
1405 {
1406 struct bpf_insn *insn = insn_buf;
1407
1408 switch (ctx_off) {
1409 case offsetof(struct __sk_buff, len):
1410 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, len) != 4);
1411
1412 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
1413 offsetof(struct sk_buff, len));
1414 break;
1415
1416 case offsetof(struct __sk_buff, protocol):
1417 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
1418
1419 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
1420 offsetof(struct sk_buff, protocol));
1421 break;
1422
1423 case offsetof(struct __sk_buff, vlan_proto):
1424 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2);
1425
1426 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
1427 offsetof(struct sk_buff, vlan_proto));
1428 break;
1429
1430 case offsetof(struct __sk_buff, priority):
1431 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, priority) != 4);
1432
1433 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
1434 offsetof(struct sk_buff, priority));
1435 break;
1436
1437 case offsetof(struct __sk_buff, mark):
1438 return convert_skb_access(SKF_AD_MARK, dst_reg, src_reg, insn);
1439
1440 case offsetof(struct __sk_buff, pkt_type):
1441 return convert_skb_access(SKF_AD_PKTTYPE, dst_reg, src_reg, insn);
1442
1443 case offsetof(struct __sk_buff, queue_mapping):
1444 return convert_skb_access(SKF_AD_QUEUE, dst_reg, src_reg, insn);
1445
1446 case offsetof(struct __sk_buff, vlan_present):
1447 return convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
1448 dst_reg, src_reg, insn);
1449
1450 case offsetof(struct __sk_buff, vlan_tci):
1451 return convert_skb_access(SKF_AD_VLAN_TAG,
1452 dst_reg, src_reg, insn);
1453 }
1454
1455 return insn - insn_buf;
1456 }
1457
1458 static const struct bpf_verifier_ops sk_filter_ops = {
1459 .get_func_proto = sk_filter_func_proto,
1460 .is_valid_access = sk_filter_is_valid_access,
1461 .convert_ctx_access = sk_filter_convert_ctx_access,
1462 };
1463
1464 static const struct bpf_verifier_ops tc_cls_act_ops = {
1465 .get_func_proto = tc_cls_act_func_proto,
1466 .is_valid_access = sk_filter_is_valid_access,
1467 .convert_ctx_access = sk_filter_convert_ctx_access,
1468 };
1469
1470 static struct bpf_prog_type_list sk_filter_type __read_mostly = {
1471 .ops = &sk_filter_ops,
1472 .type = BPF_PROG_TYPE_SOCKET_FILTER,
1473 };
1474
1475 static struct bpf_prog_type_list sched_cls_type __read_mostly = {
1476 .ops = &tc_cls_act_ops,
1477 .type = BPF_PROG_TYPE_SCHED_CLS,
1478 };
1479
1480 static struct bpf_prog_type_list sched_act_type __read_mostly = {
1481 .ops = &tc_cls_act_ops,
1482 .type = BPF_PROG_TYPE_SCHED_ACT,
1483 };
1484
1485 static int __init register_sk_filter_ops(void)
1486 {
1487 bpf_register_prog_type(&sk_filter_type);
1488 bpf_register_prog_type(&sched_cls_type);
1489 bpf_register_prog_type(&sched_act_type);
1490
1491 return 0;
1492 }
1493 late_initcall(register_sk_filter_ops);
1494
1495 int sk_detach_filter(struct sock *sk)
1496 {
1497 int ret = -ENOENT;
1498 struct sk_filter *filter;
1499
1500 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1501 return -EPERM;
1502
1503 filter = rcu_dereference_protected(sk->sk_filter,
1504 sock_owned_by_user(sk));
1505 if (filter) {
1506 RCU_INIT_POINTER(sk->sk_filter, NULL);
1507 sk_filter_uncharge(sk, filter);
1508 ret = 0;
1509 }
1510
1511 return ret;
1512 }
1513 EXPORT_SYMBOL_GPL(sk_detach_filter);
1514
1515 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
1516 unsigned int len)
1517 {
1518 struct sock_fprog_kern *fprog;
1519 struct sk_filter *filter;
1520 int ret = 0;
1521
1522 lock_sock(sk);
1523 filter = rcu_dereference_protected(sk->sk_filter,
1524 sock_owned_by_user(sk));
1525 if (!filter)
1526 goto out;
1527
1528 /* We're copying the filter that has been originally attached,
1529 * so no conversion/decode needed anymore.
1530 */
1531 fprog = filter->prog->orig_prog;
1532
1533 ret = fprog->len;
1534 if (!len)
1535 /* User space only enquires number of filter blocks. */
1536 goto out;
1537
1538 ret = -EINVAL;
1539 if (len < fprog->len)
1540 goto out;
1541
1542 ret = -EFAULT;
1543 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
1544 goto out;
1545
1546 /* Instead of bytes, the API requests to return the number
1547 * of filter blocks.
1548 */
1549 ret = fprog->len;
1550 out:
1551 release_sock(sk);
1552 return ret;
1553 }
Linux kernel - Deliverables
This page took 0.06129 seconds and 6 git commands to generate.