projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
[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 static 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),
375 GFP_KERNEL | __GFP_NOWARN);
376 if (!addrs)
377 return -ENOMEM;
378 }
379
380 do_pass:
381 new_insn = new_prog;
382 fp = prog;
383
384 if (new_insn)
385 *new_insn = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
386 new_insn++;
387
388 for (i = 0; i < len; fp++, i++) {
389 struct bpf_insn tmp_insns[6] = { };
390 struct bpf_insn *insn = tmp_insns;
391
392 if (addrs)
393 addrs[i] = new_insn - new_prog;
394
395 switch (fp->code) {
396 /* All arithmetic insns and skb loads map as-is. */
397 case BPF_ALU | BPF_ADD | BPF_X:
398 case BPF_ALU | BPF_ADD | BPF_K:
399 case BPF_ALU | BPF_SUB | BPF_X:
400 case BPF_ALU | BPF_SUB | BPF_K:
401 case BPF_ALU | BPF_AND | BPF_X:
402 case BPF_ALU | BPF_AND | BPF_K:
403 case BPF_ALU | BPF_OR | BPF_X:
404 case BPF_ALU | BPF_OR | BPF_K:
405 case BPF_ALU | BPF_LSH | BPF_X:
406 case BPF_ALU | BPF_LSH | BPF_K:
407 case BPF_ALU | BPF_RSH | BPF_X:
408 case BPF_ALU | BPF_RSH | BPF_K:
409 case BPF_ALU | BPF_XOR | BPF_X:
410 case BPF_ALU | BPF_XOR | BPF_K:
411 case BPF_ALU | BPF_MUL | BPF_X:
412 case BPF_ALU | BPF_MUL | BPF_K:
413 case BPF_ALU | BPF_DIV | BPF_X:
414 case BPF_ALU | BPF_DIV | BPF_K:
415 case BPF_ALU | BPF_MOD | BPF_X:
416 case BPF_ALU | BPF_MOD | BPF_K:
417 case BPF_ALU | BPF_NEG:
418 case BPF_LD | BPF_ABS | BPF_W:
419 case BPF_LD | BPF_ABS | BPF_H:
420 case BPF_LD | BPF_ABS | BPF_B:
421 case BPF_LD | BPF_IND | BPF_W:
422 case BPF_LD | BPF_IND | BPF_H:
423 case BPF_LD | BPF_IND | BPF_B:
424 /* Check for overloaded BPF extension and
425 * directly convert it if found, otherwise
426 * just move on with mapping.
427 */
428 if (BPF_CLASS(fp->code) == BPF_LD &&
429 BPF_MODE(fp->code) == BPF_ABS &&
430 convert_bpf_extensions(fp, &insn))
431 break;
432
433 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
434 break;
435
436 /* Jump transformation cannot use BPF block macros
437 * everywhere as offset calculation and target updates
438 * require a bit more work than the rest, i.e. jump
439 * opcodes map as-is, but offsets need adjustment.
440 */
441
442 #define BPF_EMIT_JMP \
443 do { \
444 if (target >= len || target < 0) \
445 goto err; \
446 insn->off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
447 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
448 insn->off -= insn - tmp_insns; \
449 } while (0)
450
451 case BPF_JMP | BPF_JA:
452 target = i + fp->k + 1;
453 insn->code = fp->code;
454 BPF_EMIT_JMP;
455 break;
456
457 case BPF_JMP | BPF_JEQ | BPF_K:
458 case BPF_JMP | BPF_JEQ | BPF_X:
459 case BPF_JMP | BPF_JSET | BPF_K:
460 case BPF_JMP | BPF_JSET | BPF_X:
461 case BPF_JMP | BPF_JGT | BPF_K:
462 case BPF_JMP | BPF_JGT | BPF_X:
463 case BPF_JMP | BPF_JGE | BPF_K:
464 case BPF_JMP | BPF_JGE | BPF_X:
465 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
466 /* BPF immediates are signed, zero extend
467 * immediate into tmp register and use it
468 * in compare insn.
469 */
470 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
471
472 insn->dst_reg = BPF_REG_A;
473 insn->src_reg = BPF_REG_TMP;
474 bpf_src = BPF_X;
475 } else {
476 insn->dst_reg = BPF_REG_A;
477 insn->src_reg = BPF_REG_X;
478 insn->imm = fp->k;
479 bpf_src = BPF_SRC(fp->code);
480 }
481
482 /* Common case where 'jump_false' is next insn. */
483 if (fp->jf == 0) {
484 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
485 target = i + fp->jt + 1;
486 BPF_EMIT_JMP;
487 break;
488 }
489
490 /* Convert JEQ into JNE when 'jump_true' is next insn. */
491 if (fp->jt == 0 && BPF_OP(fp->code) == BPF_JEQ) {
492 insn->code = BPF_JMP | BPF_JNE | bpf_src;
493 target = i + fp->jf + 1;
494 BPF_EMIT_JMP;
495 break;
496 }
497
498 /* Other jumps are mapped into two insns: Jxx and JA. */
499 target = i + fp->jt + 1;
500 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
501 BPF_EMIT_JMP;
502 insn++;
503
504 insn->code = BPF_JMP | BPF_JA;
505 target = i + fp->jf + 1;
506 BPF_EMIT_JMP;
507 break;
508
509 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
510 case BPF_LDX | BPF_MSH | BPF_B:
511 /* tmp = A */
512 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_A);
513 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
514 *insn++ = BPF_LD_ABS(BPF_B, fp->k);
515 /* A &= 0xf */
516 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
517 /* A <<= 2 */
518 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
519 /* X = A */
520 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
521 /* A = tmp */
522 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
523 break;
524
525 /* RET_K, RET_A are remaped into 2 insns. */
526 case BPF_RET | BPF_A:
527 case BPF_RET | BPF_K:
528 *insn++ = BPF_MOV32_RAW(BPF_RVAL(fp->code) == BPF_K ?
529 BPF_K : BPF_X, BPF_REG_0,
530 BPF_REG_A, fp->k);
531 *insn = BPF_EXIT_INSN();
532 break;
533
534 /* Store to stack. */
535 case BPF_ST:
536 case BPF_STX:
537 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
538 BPF_ST ? BPF_REG_A : BPF_REG_X,
539 -(BPF_MEMWORDS - fp->k) * 4);
540 break;
541
542 /* Load from stack. */
543 case BPF_LD | BPF_MEM:
544 case BPF_LDX | BPF_MEM:
545 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
546 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
547 -(BPF_MEMWORDS - fp->k) * 4);
548 break;
549
550 /* A = K or X = K */
551 case BPF_LD | BPF_IMM:
552 case BPF_LDX | BPF_IMM:
553 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
554 BPF_REG_A : BPF_REG_X, fp->k);
555 break;
556
557 /* X = A */
558 case BPF_MISC | BPF_TAX:
559 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
560 break;
561
562 /* A = X */
563 case BPF_MISC | BPF_TXA:
564 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
565 break;
566
567 /* A = skb->len or X = skb->len */
568 case BPF_LD | BPF_W | BPF_LEN:
569 case BPF_LDX | BPF_W | BPF_LEN:
570 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
571 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
572 offsetof(struct sk_buff, len));
573 break;
574
575 /* Access seccomp_data fields. */
576 case BPF_LDX | BPF_ABS | BPF_W:
577 /* A = *(u32 *) (ctx + K) */
578 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
579 break;
580
581 /* Unknown instruction. */
582 default:
583 goto err;
584 }
585
586 insn++;
587 if (new_prog)
588 memcpy(new_insn, tmp_insns,
589 sizeof(*insn) * (insn - tmp_insns));
590 new_insn += insn - tmp_insns;
591 }
592
593 if (!new_prog) {
594 /* Only calculating new length. */
595 *new_len = new_insn - new_prog;
596 return 0;
597 }
598
599 pass++;
600 if (new_flen != new_insn - new_prog) {
601 new_flen = new_insn - new_prog;
602 if (pass > 2)
603 goto err;
604 goto do_pass;
605 }
606
607 kfree(addrs);
608 BUG_ON(*new_len != new_flen);
609 return 0;
610 err:
611 kfree(addrs);
612 return -EINVAL;
613 }
614
615 /* Security:
616 *
617 * As we dont want to clear mem[] array for each packet going through
618 * __bpf_prog_run(), we check that filter loaded by user never try to read
619 * a cell if not previously written, and we check all branches to be sure
620 * a malicious user doesn't try to abuse us.
621 */
622 static int check_load_and_stores(const struct sock_filter *filter, int flen)
623 {
624 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
625 int pc, ret = 0;
626
627 BUILD_BUG_ON(BPF_MEMWORDS > 16);
628
629 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
630 if (!masks)
631 return -ENOMEM;
632
633 memset(masks, 0xff, flen * sizeof(*masks));
634
635 for (pc = 0; pc < flen; pc++) {
636 memvalid &= masks[pc];
637
638 switch (filter[pc].code) {
639 case BPF_ST:
640 case BPF_STX:
641 memvalid |= (1 << filter[pc].k);
642 break;
643 case BPF_LD | BPF_MEM:
644 case BPF_LDX | BPF_MEM:
645 if (!(memvalid & (1 << filter[pc].k))) {
646 ret = -EINVAL;
647 goto error;
648 }
649 break;
650 case BPF_JMP | BPF_JA:
651 /* A jump must set masks on target */
652 masks[pc + 1 + filter[pc].k] &= memvalid;
653 memvalid = ~0;
654 break;
655 case BPF_JMP | BPF_JEQ | BPF_K:
656 case BPF_JMP | BPF_JEQ | BPF_X:
657 case BPF_JMP | BPF_JGE | BPF_K:
658 case BPF_JMP | BPF_JGE | BPF_X:
659 case BPF_JMP | BPF_JGT | BPF_K:
660 case BPF_JMP | BPF_JGT | BPF_X:
661 case BPF_JMP | BPF_JSET | BPF_K:
662 case BPF_JMP | BPF_JSET | BPF_X:
663 /* A jump must set masks on targets */
664 masks[pc + 1 + filter[pc].jt] &= memvalid;
665 masks[pc + 1 + filter[pc].jf] &= memvalid;
666 memvalid = ~0;
667 break;
668 }
669 }
670 error:
671 kfree(masks);
672 return ret;
673 }
674
675 static bool chk_code_allowed(u16 code_to_probe)
676 {
677 static const bool codes[] = {
678 /* 32 bit ALU operations */
679 [BPF_ALU | BPF_ADD | BPF_K] = true,
680 [BPF_ALU | BPF_ADD | BPF_X] = true,
681 [BPF_ALU | BPF_SUB | BPF_K] = true,
682 [BPF_ALU | BPF_SUB | BPF_X] = true,
683 [BPF_ALU | BPF_MUL | BPF_K] = true,
684 [BPF_ALU | BPF_MUL | BPF_X] = true,
685 [BPF_ALU | BPF_DIV | BPF_K] = true,
686 [BPF_ALU | BPF_DIV | BPF_X] = true,
687 [BPF_ALU | BPF_MOD | BPF_K] = true,
688 [BPF_ALU | BPF_MOD | BPF_X] = true,
689 [BPF_ALU | BPF_AND | BPF_K] = true,
690 [BPF_ALU | BPF_AND | BPF_X] = true,
691 [BPF_ALU | BPF_OR | BPF_K] = true,
692 [BPF_ALU | BPF_OR | BPF_X] = true,
693 [BPF_ALU | BPF_XOR | BPF_K] = true,
694 [BPF_ALU | BPF_XOR | BPF_X] = true,
695 [BPF_ALU | BPF_LSH | BPF_K] = true,
696 [BPF_ALU | BPF_LSH | BPF_X] = true,
697 [BPF_ALU | BPF_RSH | BPF_K] = true,
698 [BPF_ALU | BPF_RSH | BPF_X] = true,
699 [BPF_ALU | BPF_NEG] = true,
700 /* Load instructions */
701 [BPF_LD | BPF_W | BPF_ABS] = true,
702 [BPF_LD | BPF_H | BPF_ABS] = true,
703 [BPF_LD | BPF_B | BPF_ABS] = true,
704 [BPF_LD | BPF_W | BPF_LEN] = true,
705 [BPF_LD | BPF_W | BPF_IND] = true,
706 [BPF_LD | BPF_H | BPF_IND] = true,
707 [BPF_LD | BPF_B | BPF_IND] = true,
708 [BPF_LD | BPF_IMM] = true,
709 [BPF_LD | BPF_MEM] = true,
710 [BPF_LDX | BPF_W | BPF_LEN] = true,
711 [BPF_LDX | BPF_B | BPF_MSH] = true,
712 [BPF_LDX | BPF_IMM] = true,
713 [BPF_LDX | BPF_MEM] = true,
714 /* Store instructions */
715 [BPF_ST] = true,
716 [BPF_STX] = true,
717 /* Misc instructions */
718 [BPF_MISC | BPF_TAX] = true,
719 [BPF_MISC | BPF_TXA] = true,
720 /* Return instructions */
721 [BPF_RET | BPF_K] = true,
722 [BPF_RET | BPF_A] = true,
723 /* Jump instructions */
724 [BPF_JMP | BPF_JA] = true,
725 [BPF_JMP | BPF_JEQ | BPF_K] = true,
726 [BPF_JMP | BPF_JEQ | BPF_X] = true,
727 [BPF_JMP | BPF_JGE | BPF_K] = true,
728 [BPF_JMP | BPF_JGE | BPF_X] = true,
729 [BPF_JMP | BPF_JGT | BPF_K] = true,
730 [BPF_JMP | BPF_JGT | BPF_X] = true,
731 [BPF_JMP | BPF_JSET | BPF_K] = true,
732 [BPF_JMP | BPF_JSET | BPF_X] = true,
733 };
734
735 if (code_to_probe >= ARRAY_SIZE(codes))
736 return false;
737
738 return codes[code_to_probe];
739 }
740
741 /**
742 * bpf_check_classic - verify socket filter code
743 * @filter: filter to verify
744 * @flen: length of filter
745 *
746 * Check the user's filter code. If we let some ugly
747 * filter code slip through kaboom! The filter must contain
748 * no references or jumps that are out of range, no illegal
749 * instructions, and must end with a RET instruction.
750 *
751 * All jumps are forward as they are not signed.
752 *
753 * Returns 0 if the rule set is legal or -EINVAL if not.
754 */
755 static int bpf_check_classic(const struct sock_filter *filter,
756 unsigned int flen)
757 {
758 bool anc_found;
759 int pc;
760
761 if (flen == 0 || flen > BPF_MAXINSNS)
762 return -EINVAL;
763
764 /* Check the filter code now */
765 for (pc = 0; pc < flen; pc++) {
766 const struct sock_filter *ftest = &filter[pc];
767
768 /* May we actually operate on this code? */
769 if (!chk_code_allowed(ftest->code))
770 return -EINVAL;
771
772 /* Some instructions need special checks */
773 switch (ftest->code) {
774 case BPF_ALU | BPF_DIV | BPF_K:
775 case BPF_ALU | BPF_MOD | BPF_K:
776 /* Check for division by zero */
777 if (ftest->k == 0)
778 return -EINVAL;
779 break;
780 case BPF_LD | BPF_MEM:
781 case BPF_LDX | BPF_MEM:
782 case BPF_ST:
783 case BPF_STX:
784 /* Check for invalid memory addresses */
785 if (ftest->k >= BPF_MEMWORDS)
786 return -EINVAL;
787 break;
788 case BPF_JMP | BPF_JA:
789 /* Note, the large ftest->k might cause loops.
790 * Compare this with conditional jumps below,
791 * where offsets are limited. --ANK (981016)
792 */
793 if (ftest->k >= (unsigned int)(flen - pc - 1))
794 return -EINVAL;
795 break;
796 case BPF_JMP | BPF_JEQ | BPF_K:
797 case BPF_JMP | BPF_JEQ | BPF_X:
798 case BPF_JMP | BPF_JGE | BPF_K:
799 case BPF_JMP | BPF_JGE | BPF_X:
800 case BPF_JMP | BPF_JGT | BPF_K:
801 case BPF_JMP | BPF_JGT | BPF_X:
802 case BPF_JMP | BPF_JSET | BPF_K:
803 case BPF_JMP | BPF_JSET | BPF_X:
804 /* Both conditionals must be safe */
805 if (pc + ftest->jt + 1 >= flen ||
806 pc + ftest->jf + 1 >= flen)
807 return -EINVAL;
808 break;
809 case BPF_LD | BPF_W | BPF_ABS:
810 case BPF_LD | BPF_H | BPF_ABS:
811 case BPF_LD | BPF_B | BPF_ABS:
812 anc_found = false;
813 if (bpf_anc_helper(ftest) & BPF_ANC)
814 anc_found = true;
815 /* Ancillary operation unknown or unsupported */
816 if (anc_found == false && ftest->k >= SKF_AD_OFF)
817 return -EINVAL;
818 }
819 }
820
821 /* Last instruction must be a RET code */
822 switch (filter[flen - 1].code) {
823 case BPF_RET | BPF_K:
824 case BPF_RET | BPF_A:
825 return check_load_and_stores(filter, flen);
826 }
827
828 return -EINVAL;
829 }
830
831 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
832 const struct sock_fprog *fprog)
833 {
834 unsigned int fsize = bpf_classic_proglen(fprog);
835 struct sock_fprog_kern *fkprog;
836
837 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
838 if (!fp->orig_prog)
839 return -ENOMEM;
840
841 fkprog = fp->orig_prog;
842 fkprog->len = fprog->len;
843
844 fkprog->filter = kmemdup(fp->insns, fsize,
845 GFP_KERNEL | __GFP_NOWARN);
846 if (!fkprog->filter) {
847 kfree(fp->orig_prog);
848 return -ENOMEM;
849 }
850
851 return 0;
852 }
853
854 static void bpf_release_orig_filter(struct bpf_prog *fp)
855 {
856 struct sock_fprog_kern *fprog = fp->orig_prog;
857
858 if (fprog) {
859 kfree(fprog->filter);
860 kfree(fprog);
861 }
862 }
863
864 static void __bpf_prog_release(struct bpf_prog *prog)
865 {
866 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
867 bpf_prog_put(prog);
868 } else {
869 bpf_release_orig_filter(prog);
870 bpf_prog_free(prog);
871 }
872 }
873
874 static void __sk_filter_release(struct sk_filter *fp)
875 {
876 __bpf_prog_release(fp->prog);
877 kfree(fp);
878 }
879
880 /**
881 * sk_filter_release_rcu - Release a socket filter by rcu_head
882 * @rcu: rcu_head that contains the sk_filter to free
883 */
884 static void sk_filter_release_rcu(struct rcu_head *rcu)
885 {
886 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
887
888 __sk_filter_release(fp);
889 }
890
891 /**
892 * sk_filter_release - release a socket filter
893 * @fp: filter to remove
894 *
895 * Remove a filter from a socket and release its resources.
896 */
897 static void sk_filter_release(struct sk_filter *fp)
898 {
899 if (atomic_dec_and_test(&fp->refcnt))
900 call_rcu(&fp->rcu, sk_filter_release_rcu);
901 }
902
903 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
904 {
905 u32 filter_size = bpf_prog_size(fp->prog->len);
906
907 atomic_sub(filter_size, &sk->sk_omem_alloc);
908 sk_filter_release(fp);
909 }
910
911 /* try to charge the socket memory if there is space available
912 * return true on success
913 */
914 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
915 {
916 u32 filter_size = bpf_prog_size(fp->prog->len);
917
918 /* same check as in sock_kmalloc() */
919 if (filter_size <= sysctl_optmem_max &&
920 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
921 atomic_inc(&fp->refcnt);
922 atomic_add(filter_size, &sk->sk_omem_alloc);
923 return true;
924 }
925 return false;
926 }
927
928 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
929 {
930 struct sock_filter *old_prog;
931 struct bpf_prog *old_fp;
932 int err, new_len, old_len = fp->len;
933
934 /* We are free to overwrite insns et al right here as it
935 * won't be used at this point in time anymore internally
936 * after the migration to the internal BPF instruction
937 * representation.
938 */
939 BUILD_BUG_ON(sizeof(struct sock_filter) !=
940 sizeof(struct bpf_insn));
941
942 /* Conversion cannot happen on overlapping memory areas,
943 * so we need to keep the user BPF around until the 2nd
944 * pass. At this time, the user BPF is stored in fp->insns.
945 */
946 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
947 GFP_KERNEL | __GFP_NOWARN);
948 if (!old_prog) {
949 err = -ENOMEM;
950 goto out_err;
951 }
952
953 /* 1st pass: calculate the new program length. */
954 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len);
955 if (err)
956 goto out_err_free;
957
958 /* Expand fp for appending the new filter representation. */
959 old_fp = fp;
960 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
961 if (!fp) {
962 /* The old_fp is still around in case we couldn't
963 * allocate new memory, so uncharge on that one.
964 */
965 fp = old_fp;
966 err = -ENOMEM;
967 goto out_err_free;
968 }
969
970 fp->len = new_len;
971
972 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
973 err = bpf_convert_filter(old_prog, old_len, fp->insnsi, &new_len);
974 if (err)
975 /* 2nd bpf_convert_filter() can fail only if it fails
976 * to allocate memory, remapping must succeed. Note,
977 * that at this time old_fp has already been released
978 * by krealloc().
979 */
980 goto out_err_free;
981
982 bpf_prog_select_runtime(fp);
983
984 kfree(old_prog);
985 return fp;
986
987 out_err_free:
988 kfree(old_prog);
989 out_err:
990 __bpf_prog_release(fp);
991 return ERR_PTR(err);
992 }
993
994 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
995 bpf_aux_classic_check_t trans)
996 {
997 int err;
998
999 fp->bpf_func = NULL;
1000 fp->jited = false;
1001
1002 err = bpf_check_classic(fp->insns, fp->len);
1003 if (err) {
1004 __bpf_prog_release(fp);
1005 return ERR_PTR(err);
1006 }
1007
1008 /* There might be additional checks and transformations
1009 * needed on classic filters, f.e. in case of seccomp.
1010 */
1011 if (trans) {
1012 err = trans(fp->insns, fp->len);
1013 if (err) {
1014 __bpf_prog_release(fp);
1015 return ERR_PTR(err);
1016 }
1017 }
1018
1019 /* Probe if we can JIT compile the filter and if so, do
1020 * the compilation of the filter.
1021 */
1022 bpf_jit_compile(fp);
1023
1024 /* JIT compiler couldn't process this filter, so do the
1025 * internal BPF translation for the optimized interpreter.
1026 */
1027 if (!fp->jited)
1028 fp = bpf_migrate_filter(fp);
1029
1030 return fp;
1031 }
1032
1033 /**
1034 * bpf_prog_create - create an unattached filter
1035 * @pfp: the unattached filter that is created
1036 * @fprog: the filter program
1037 *
1038 * Create a filter independent of any socket. We first run some
1039 * sanity checks on it to make sure it does not explode on us later.
1040 * If an error occurs or there is insufficient memory for the filter
1041 * a negative errno code is returned. On success the return is zero.
1042 */
1043 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1044 {
1045 unsigned int fsize = bpf_classic_proglen(fprog);
1046 struct bpf_prog *fp;
1047
1048 /* Make sure new filter is there and in the right amounts. */
1049 if (fprog->filter == NULL)
1050 return -EINVAL;
1051
1052 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1053 if (!fp)
1054 return -ENOMEM;
1055
1056 memcpy(fp->insns, fprog->filter, fsize);
1057
1058 fp->len = fprog->len;
1059 /* Since unattached filters are not copied back to user
1060 * space through sk_get_filter(), we do not need to hold
1061 * a copy here, and can spare us the work.
1062 */
1063 fp->orig_prog = NULL;
1064
1065 /* bpf_prepare_filter() already takes care of freeing
1066 * memory in case something goes wrong.
1067 */
1068 fp = bpf_prepare_filter(fp, NULL);
1069 if (IS_ERR(fp))
1070 return PTR_ERR(fp);
1071
1072 *pfp = fp;
1073 return 0;
1074 }
1075 EXPORT_SYMBOL_GPL(bpf_prog_create);
1076
1077 /**
1078 * bpf_prog_create_from_user - create an unattached filter from user buffer
1079 * @pfp: the unattached filter that is created
1080 * @fprog: the filter program
1081 * @trans: post-classic verifier transformation handler
1082 *
1083 * This function effectively does the same as bpf_prog_create(), only
1084 * that it builds up its insns buffer from user space provided buffer.
1085 * It also allows for passing a bpf_aux_classic_check_t handler.
1086 */
1087 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1088 bpf_aux_classic_check_t trans)
1089 {
1090 unsigned int fsize = bpf_classic_proglen(fprog);
1091 struct bpf_prog *fp;
1092
1093 /* Make sure new filter is there and in the right amounts. */
1094 if (fprog->filter == NULL)
1095 return -EINVAL;
1096
1097 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1098 if (!fp)
1099 return -ENOMEM;
1100
1101 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1102 __bpf_prog_free(fp);
1103 return -EFAULT;
1104 }
1105
1106 fp->len = fprog->len;
1107 /* Since unattached filters are not copied back to user
1108 * space through sk_get_filter(), we do not need to hold
1109 * a copy here, and can spare us the work.
1110 */
1111 fp->orig_prog = NULL;
1112
1113 /* bpf_prepare_filter() already takes care of freeing
1114 * memory in case something goes wrong.
1115 */
1116 fp = bpf_prepare_filter(fp, trans);
1117 if (IS_ERR(fp))
1118 return PTR_ERR(fp);
1119
1120 *pfp = fp;
1121 return 0;
1122 }
1123
1124 void bpf_prog_destroy(struct bpf_prog *fp)
1125 {
1126 __bpf_prog_release(fp);
1127 }
1128 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1129
1130 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1131 {
1132 struct sk_filter *fp, *old_fp;
1133
1134 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1135 if (!fp)
1136 return -ENOMEM;
1137
1138 fp->prog = prog;
1139 atomic_set(&fp->refcnt, 0);
1140
1141 if (!sk_filter_charge(sk, fp)) {
1142 kfree(fp);
1143 return -ENOMEM;
1144 }
1145
1146 old_fp = rcu_dereference_protected(sk->sk_filter,
1147 sock_owned_by_user(sk));
1148 rcu_assign_pointer(sk->sk_filter, fp);
1149
1150 if (old_fp)
1151 sk_filter_uncharge(sk, old_fp);
1152
1153 return 0;
1154 }
1155
1156 /**
1157 * sk_attach_filter - attach a socket filter
1158 * @fprog: the filter program
1159 * @sk: the socket to use
1160 *
1161 * Attach the user's filter code. We first run some sanity checks on
1162 * it to make sure it does not explode on us later. If an error
1163 * occurs or there is insufficient memory for the filter a negative
1164 * errno code is returned. On success the return is zero.
1165 */
1166 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1167 {
1168 unsigned int fsize = bpf_classic_proglen(fprog);
1169 unsigned int bpf_fsize = bpf_prog_size(fprog->len);
1170 struct bpf_prog *prog;
1171 int err;
1172
1173 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1174 return -EPERM;
1175
1176 /* Make sure new filter is there and in the right amounts. */
1177 if (fprog->filter == NULL)
1178 return -EINVAL;
1179
1180 prog = bpf_prog_alloc(bpf_fsize, 0);
1181 if (!prog)
1182 return -ENOMEM;
1183
1184 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1185 __bpf_prog_free(prog);
1186 return -EFAULT;
1187 }
1188
1189 prog->len = fprog->len;
1190
1191 err = bpf_prog_store_orig_filter(prog, fprog);
1192 if (err) {
1193 __bpf_prog_free(prog);
1194 return -ENOMEM;
1195 }
1196
1197 /* bpf_prepare_filter() already takes care of freeing
1198 * memory in case something goes wrong.
1199 */
1200 prog = bpf_prepare_filter(prog, NULL);
1201 if (IS_ERR(prog))
1202 return PTR_ERR(prog);
1203
1204 err = __sk_attach_prog(prog, sk);
1205 if (err < 0) {
1206 __bpf_prog_release(prog);
1207 return err;
1208 }
1209
1210 return 0;
1211 }
1212 EXPORT_SYMBOL_GPL(sk_attach_filter);
1213
1214 int sk_attach_bpf(u32 ufd, struct sock *sk)
1215 {
1216 struct bpf_prog *prog;
1217 int err;
1218
1219 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1220 return -EPERM;
1221
1222 prog = bpf_prog_get(ufd);
1223 if (IS_ERR(prog))
1224 return PTR_ERR(prog);
1225
1226 if (prog->type != BPF_PROG_TYPE_SOCKET_FILTER) {
1227 bpf_prog_put(prog);
1228 return -EINVAL;
1229 }
1230
1231 err = __sk_attach_prog(prog, sk);
1232 if (err < 0) {
1233 bpf_prog_put(prog);
1234 return err;
1235 }
1236
1237 return 0;
1238 }
1239
1240 /**
1241 * bpf_skb_clone_not_writable - is the header of a clone not writable
1242 * @skb: buffer to check
1243 * @len: length up to which to write, can be negative
1244 *
1245 * Returns true if modifying the header part of the cloned buffer
1246 * does require the data to be copied. I.e. this version works with
1247 * negative lengths needed for eBPF case!
1248 */
1249 static bool bpf_skb_clone_unwritable(const struct sk_buff *skb, int len)
1250 {
1251 return skb_header_cloned(skb) ||
1252 (int) skb_headroom(skb) + len > skb->hdr_len;
1253 }
1254
1255 #define BPF_RECOMPUTE_CSUM(flags) ((flags) & 1)
1256
1257 static u64 bpf_skb_store_bytes(u64 r1, u64 r2, u64 r3, u64 r4, u64 flags)
1258 {
1259 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1260 int offset = (int) r2;
1261 void *from = (void *) (long) r3;
1262 unsigned int len = (unsigned int) r4;
1263 char buf[16];
1264 void *ptr;
1265
1266 /* bpf verifier guarantees that:
1267 * 'from' pointer points to bpf program stack
1268 * 'len' bytes of it were initialized
1269 * 'len' > 0
1270 * 'skb' is a valid pointer to 'struct sk_buff'
1271 *
1272 * so check for invalid 'offset' and too large 'len'
1273 */
1274 if (unlikely((u32) offset > 0xffff || len > sizeof(buf)))
1275 return -EFAULT;
1276
1277 offset -= skb->data - skb_mac_header(skb);
1278 if (unlikely(skb_cloned(skb) &&
1279 bpf_skb_clone_unwritable(skb, offset + len)))
1280 return -EFAULT;
1281
1282 ptr = skb_header_pointer(skb, offset, len, buf);
1283 if (unlikely(!ptr))
1284 return -EFAULT;
1285
1286 if (BPF_RECOMPUTE_CSUM(flags))
1287 skb_postpull_rcsum(skb, ptr, len);
1288
1289 memcpy(ptr, from, len);
1290
1291 if (ptr == buf)
1292 /* skb_store_bits cannot return -EFAULT here */
1293 skb_store_bits(skb, offset, ptr, len);
1294
1295 if (BPF_RECOMPUTE_CSUM(flags) && skb->ip_summed == CHECKSUM_COMPLETE)
1296 skb->csum = csum_add(skb->csum, csum_partial(ptr, len, 0));
1297 return 0;
1298 }
1299
1300 const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1301 .func = bpf_skb_store_bytes,
1302 .gpl_only = false,
1303 .ret_type = RET_INTEGER,
1304 .arg1_type = ARG_PTR_TO_CTX,
1305 .arg2_type = ARG_ANYTHING,
1306 .arg3_type = ARG_PTR_TO_STACK,
1307 .arg4_type = ARG_CONST_STACK_SIZE,
1308 .arg5_type = ARG_ANYTHING,
1309 };
1310
1311 #define BPF_HEADER_FIELD_SIZE(flags) ((flags) & 0x0f)
1312 #define BPF_IS_PSEUDO_HEADER(flags) ((flags) & 0x10)
1313
1314 static u64 bpf_l3_csum_replace(u64 r1, u64 r2, u64 from, u64 to, u64 flags)
1315 {
1316 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1317 int offset = (int) r2;
1318 __sum16 sum, *ptr;
1319
1320 if (unlikely((u32) offset > 0xffff))
1321 return -EFAULT;
1322
1323 offset -= skb->data - skb_mac_header(skb);
1324 if (unlikely(skb_cloned(skb) &&
1325 bpf_skb_clone_unwritable(skb, offset + sizeof(sum))))
1326 return -EFAULT;
1327
1328 ptr = skb_header_pointer(skb, offset, sizeof(sum), &sum);
1329 if (unlikely(!ptr))
1330 return -EFAULT;
1331
1332 switch (BPF_HEADER_FIELD_SIZE(flags)) {
1333 case 2:
1334 csum_replace2(ptr, from, to);
1335 break;
1336 case 4:
1337 csum_replace4(ptr, from, to);
1338 break;
1339 default:
1340 return -EINVAL;
1341 }
1342
1343 if (ptr == &sum)
1344 /* skb_store_bits guaranteed to not return -EFAULT here */
1345 skb_store_bits(skb, offset, ptr, sizeof(sum));
1346
1347 return 0;
1348 }
1349
1350 const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1351 .func = bpf_l3_csum_replace,
1352 .gpl_only = false,
1353 .ret_type = RET_INTEGER,
1354 .arg1_type = ARG_PTR_TO_CTX,
1355 .arg2_type = ARG_ANYTHING,
1356 .arg3_type = ARG_ANYTHING,
1357 .arg4_type = ARG_ANYTHING,
1358 .arg5_type = ARG_ANYTHING,
1359 };
1360
1361 static u64 bpf_l4_csum_replace(u64 r1, u64 r2, u64 from, u64 to, u64 flags)
1362 {
1363 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1364 u32 is_pseudo = BPF_IS_PSEUDO_HEADER(flags);
1365 int offset = (int) r2;
1366 __sum16 sum, *ptr;
1367
1368 if (unlikely((u32) offset > 0xffff))
1369 return -EFAULT;
1370
1371 offset -= skb->data - skb_mac_header(skb);
1372 if (unlikely(skb_cloned(skb) &&
1373 bpf_skb_clone_unwritable(skb, offset + sizeof(sum))))
1374 return -EFAULT;
1375
1376 ptr = skb_header_pointer(skb, offset, sizeof(sum), &sum);
1377 if (unlikely(!ptr))
1378 return -EFAULT;
1379
1380 switch (BPF_HEADER_FIELD_SIZE(flags)) {
1381 case 2:
1382 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1383 break;
1384 case 4:
1385 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1386 break;
1387 default:
1388 return -EINVAL;
1389 }
1390
1391 if (ptr == &sum)
1392 /* skb_store_bits guaranteed to not return -EFAULT here */
1393 skb_store_bits(skb, offset, ptr, sizeof(sum));
1394
1395 return 0;
1396 }
1397
1398 const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1399 .func = bpf_l4_csum_replace,
1400 .gpl_only = false,
1401 .ret_type = RET_INTEGER,
1402 .arg1_type = ARG_PTR_TO_CTX,
1403 .arg2_type = ARG_ANYTHING,
1404 .arg3_type = ARG_ANYTHING,
1405 .arg4_type = ARG_ANYTHING,
1406 .arg5_type = ARG_ANYTHING,
1407 };
1408
1409 static const struct bpf_func_proto *
1410 sk_filter_func_proto(enum bpf_func_id func_id)
1411 {
1412 switch (func_id) {
1413 case BPF_FUNC_map_lookup_elem:
1414 return &bpf_map_lookup_elem_proto;
1415 case BPF_FUNC_map_update_elem:
1416 return &bpf_map_update_elem_proto;
1417 case BPF_FUNC_map_delete_elem:
1418 return &bpf_map_delete_elem_proto;
1419 case BPF_FUNC_get_prandom_u32:
1420 return &bpf_get_prandom_u32_proto;
1421 case BPF_FUNC_get_smp_processor_id:
1422 return &bpf_get_smp_processor_id_proto;
1423 default:
1424 return NULL;
1425 }
1426 }
1427
1428 static const struct bpf_func_proto *
1429 tc_cls_act_func_proto(enum bpf_func_id func_id)
1430 {
1431 switch (func_id) {
1432 case BPF_FUNC_skb_store_bytes:
1433 return &bpf_skb_store_bytes_proto;
1434 case BPF_FUNC_l3_csum_replace:
1435 return &bpf_l3_csum_replace_proto;
1436 case BPF_FUNC_l4_csum_replace:
1437 return &bpf_l4_csum_replace_proto;
1438 default:
1439 return sk_filter_func_proto(func_id);
1440 }
1441 }
1442
1443 static bool sk_filter_is_valid_access(int off, int size,
1444 enum bpf_access_type type)
1445 {
1446 /* only read is allowed */
1447 if (type != BPF_READ)
1448 return false;
1449
1450 /* check bounds */
1451 if (off < 0 || off >= sizeof(struct __sk_buff))
1452 return false;
1453
1454 /* disallow misaligned access */
1455 if (off % size != 0)
1456 return false;
1457
1458 /* all __sk_buff fields are __u32 */
1459 if (size != 4)
1460 return false;
1461
1462 return true;
1463 }
1464
1465 static u32 sk_filter_convert_ctx_access(int dst_reg, int src_reg, int ctx_off,
1466 struct bpf_insn *insn_buf)
1467 {
1468 struct bpf_insn *insn = insn_buf;
1469
1470 switch (ctx_off) {
1471 case offsetof(struct __sk_buff, len):
1472 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, len) != 4);
1473
1474 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
1475 offsetof(struct sk_buff, len));
1476 break;
1477
1478 case offsetof(struct __sk_buff, protocol):
1479 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
1480
1481 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
1482 offsetof(struct sk_buff, protocol));
1483 break;
1484
1485 case offsetof(struct __sk_buff, vlan_proto):
1486 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2);
1487
1488 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
1489 offsetof(struct sk_buff, vlan_proto));
1490 break;
1491
1492 case offsetof(struct __sk_buff, priority):
1493 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, priority) != 4);
1494
1495 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
1496 offsetof(struct sk_buff, priority));
1497 break;
1498
1499 case offsetof(struct __sk_buff, mark):
1500 return convert_skb_access(SKF_AD_MARK, dst_reg, src_reg, insn);
1501
1502 case offsetof(struct __sk_buff, pkt_type):
1503 return convert_skb_access(SKF_AD_PKTTYPE, dst_reg, src_reg, insn);
1504
1505 case offsetof(struct __sk_buff, queue_mapping):
1506 return convert_skb_access(SKF_AD_QUEUE, dst_reg, src_reg, insn);
1507
1508 case offsetof(struct __sk_buff, vlan_present):
1509 return convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
1510 dst_reg, src_reg, insn);
1511
1512 case offsetof(struct __sk_buff, vlan_tci):
1513 return convert_skb_access(SKF_AD_VLAN_TAG,
1514 dst_reg, src_reg, insn);
1515 }
1516
1517 return insn - insn_buf;
1518 }
1519
1520 static const struct bpf_verifier_ops sk_filter_ops = {
1521 .get_func_proto = sk_filter_func_proto,
1522 .is_valid_access = sk_filter_is_valid_access,
1523 .convert_ctx_access = sk_filter_convert_ctx_access,
1524 };
1525
1526 static const struct bpf_verifier_ops tc_cls_act_ops = {
1527 .get_func_proto = tc_cls_act_func_proto,
1528 .is_valid_access = sk_filter_is_valid_access,
1529 .convert_ctx_access = sk_filter_convert_ctx_access,
1530 };
1531
1532 static struct bpf_prog_type_list sk_filter_type __read_mostly = {
1533 .ops = &sk_filter_ops,
1534 .type = BPF_PROG_TYPE_SOCKET_FILTER,
1535 };
1536
1537 static struct bpf_prog_type_list sched_cls_type __read_mostly = {
1538 .ops = &tc_cls_act_ops,
1539 .type = BPF_PROG_TYPE_SCHED_CLS,
1540 };
1541
1542 static struct bpf_prog_type_list sched_act_type __read_mostly = {
1543 .ops = &tc_cls_act_ops,
1544 .type = BPF_PROG_TYPE_SCHED_ACT,
1545 };
1546
1547 static int __init register_sk_filter_ops(void)
1548 {
1549 bpf_register_prog_type(&sk_filter_type);
1550 bpf_register_prog_type(&sched_cls_type);
1551 bpf_register_prog_type(&sched_act_type);
1552
1553 return 0;
1554 }
1555 late_initcall(register_sk_filter_ops);
1556
1557 int sk_detach_filter(struct sock *sk)
1558 {
1559 int ret = -ENOENT;
1560 struct sk_filter *filter;
1561
1562 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1563 return -EPERM;
1564
1565 filter = rcu_dereference_protected(sk->sk_filter,
1566 sock_owned_by_user(sk));
1567 if (filter) {
1568 RCU_INIT_POINTER(sk->sk_filter, NULL);
1569 sk_filter_uncharge(sk, filter);
1570 ret = 0;
1571 }
1572
1573 return ret;
1574 }
1575 EXPORT_SYMBOL_GPL(sk_detach_filter);
1576
1577 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
1578 unsigned int len)
1579 {
1580 struct sock_fprog_kern *fprog;
1581 struct sk_filter *filter;
1582 int ret = 0;
1583
1584 lock_sock(sk);
1585 filter = rcu_dereference_protected(sk->sk_filter,
1586 sock_owned_by_user(sk));
1587 if (!filter)
1588 goto out;
1589
1590 /* We're copying the filter that has been originally attached,
1591 * so no conversion/decode needed anymore.
1592 */
1593 fprog = filter->prog->orig_prog;
1594
1595 ret = fprog->len;
1596 if (!len)
1597 /* User space only enquires number of filter blocks. */
1598 goto out;
1599
1600 ret = -EINVAL;
1601 if (len < fprog->len)
1602 goto out;
1603
1604 ret = -EFAULT;
1605 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
1606 goto out;
1607
1608 /* Instead of bytes, the API requests to return the number
1609 * of filter blocks.
1610 */
1611 ret = fprog->len;
1612 out:
1613 release_sock(sk);
1614 return ret;
1615 }
Linux kernel - Deliverables
This page took 0.077179 seconds and 6 git commands to generate.