?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ipv4: Fix ip_queue_xmit to pass sk into ip_local_out_sk
[linux-2.6/btrfs-unstable.git] / net / core / filter.c
blobda3e5357f13866424f53d939f9ea4fb96f71d36d
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 <net/flow_dissector.h>
40 #include <linux/errno.h>
41 #include <linux/timer.h>
42 #include <asm/uaccess.h>
43 #include <asm/unaligned.h>
44 #include <linux/filter.h>
45 #include <linux/ratelimit.h>
46 #include <linux/seccomp.h>
47 #include <linux/if_vlan.h>
48 #include <linux/bpf.h>
49 #include <net/sch_generic.h>
50 #include <net/cls_cgroup.h>
51 #include <net/dst_metadata.h>
52 #include <net/dst.h>
53
54 /**
55 * sk_filter - run a packet through a socket filter
56 * @sk: sock associated with &sk_buff
57 * @skb: buffer to filter
58 *
59 * Run the filter code and then cut skb->data to correct size returned by
60 * SK_RUN_FILTER. If pkt_len is 0 we toss packet. If skb->len is smaller
61 * than pkt_len we keep whole skb->data. This is the socket level
62 * wrapper to SK_RUN_FILTER. It returns 0 if the packet should
63 * be accepted or -EPERM if the packet should be tossed.
64 *
65 */
66 int sk_filter(struct sock *sk, struct sk_buff *skb)
67 {
68 int err;
69 struct sk_filter *filter;
70
71 /*
72 * If the skb was allocated from pfmemalloc reserves, only
73 * allow SOCK_MEMALLOC sockets to use it as this socket is
74 * helping free memory
75 */
76 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
77 return -ENOMEM;
78
79 err = security_sock_rcv_skb(sk, skb);
80 if (err)
81 return err;
82
83 rcu_read_lock();
84 filter = rcu_dereference(sk->sk_filter);
85 if (filter) {
86 unsigned int pkt_len = SK_RUN_FILTER(filter, skb);
87
88 err = pkt_len ? pskb_trim(skb, pkt_len) : -EPERM;
89 }
90 rcu_read_unlock();
91
92 return err;
93 }
94 EXPORT_SYMBOL(sk_filter);
95
96 static u64 __skb_get_pay_offset(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
97 {
98 return skb_get_poff((struct sk_buff *)(unsigned long) ctx);
99 }
100
101 static u64 __skb_get_nlattr(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
102 {
103 struct sk_buff *skb = (struct sk_buff *)(unsigned long) ctx;
104 struct nlattr *nla;
105
106 if (skb_is_nonlinear(skb))
107 return 0;
108
109 if (skb->len < sizeof(struct nlattr))
110 return 0;
111
112 if (a > skb->len - sizeof(struct nlattr))
113 return 0;
114
115 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
116 if (nla)
117 return (void *) nla - (void *) skb->data;
118
119 return 0;
120 }
121
122 static u64 __skb_get_nlattr_nest(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
123 {
124 struct sk_buff *skb = (struct sk_buff *)(unsigned long) ctx;
125 struct nlattr *nla;
126
127 if (skb_is_nonlinear(skb))
128 return 0;
129
130 if (skb->len < sizeof(struct nlattr))
131 return 0;
132
133 if (a > skb->len - sizeof(struct nlattr))
134 return 0;
135
136 nla = (struct nlattr *) &skb->data[a];
137 if (nla->nla_len > skb->len - a)
138 return 0;
139
140 nla = nla_find_nested(nla, x);
141 if (nla)
142 return (void *) nla - (void *) skb->data;
143
144 return 0;
145 }
146
147 static u64 __get_raw_cpu_id(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
148 {
149 return raw_smp_processor_id();
150 }
151
152 /* note that this only generates 32-bit random numbers */
153 static u64 __get_random_u32(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
154 {
155 return prandom_u32();
156 }
157
158 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
159 struct bpf_insn *insn_buf)
160 {
161 struct bpf_insn *insn = insn_buf;
162
163 switch (skb_field) {
164 case SKF_AD_MARK:
165 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
166
167 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
168 offsetof(struct sk_buff, mark));
169 break;
170
171 case SKF_AD_PKTTYPE:
172 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
173 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
174 #ifdef __BIG_ENDIAN_BITFIELD
175 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
176 #endif
177 break;
178
179 case SKF_AD_QUEUE:
180 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2);
181
182 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
183 offsetof(struct sk_buff, queue_mapping));
184 break;
185
186 case SKF_AD_VLAN_TAG:
187 case SKF_AD_VLAN_TAG_PRESENT:
188 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2);
189 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000);
190
191 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
192 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
193 offsetof(struct sk_buff, vlan_tci));
194 if (skb_field == SKF_AD_VLAN_TAG) {
195 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg,
196 ~VLAN_TAG_PRESENT);
197 } else {
198 /* dst_reg >>= 12 */
199 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 12);
200 /* dst_reg &= 1 */
201 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
202 }
203 break;
204 }
205
206 return insn - insn_buf;
207 }
208
209 static bool convert_bpf_extensions(struct sock_filter *fp,
210 struct bpf_insn **insnp)
211 {
212 struct bpf_insn *insn = *insnp;
213 u32 cnt;
214
215 switch (fp->k) {
216 case SKF_AD_OFF + SKF_AD_PROTOCOL:
217 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
218
219 /* A = *(u16 *) (CTX + offsetof(protocol)) */
220 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
221 offsetof(struct sk_buff, protocol));
222 /* A = ntohs(A) [emitting a nop or swap16] */
223 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
224 break;
225
226 case SKF_AD_OFF + SKF_AD_PKTTYPE:
227 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
228 insn += cnt - 1;
229 break;
230
231 case SKF_AD_OFF + SKF_AD_IFINDEX:
232 case SKF_AD_OFF + SKF_AD_HATYPE:
233 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
234 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, type) != 2);
235 BUILD_BUG_ON(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, dev)) < 0);
236
237 *insn++ = BPF_LDX_MEM(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, dev)),
238 BPF_REG_TMP, BPF_REG_CTX,
239 offsetof(struct sk_buff, dev));
240 /* if (tmp != 0) goto pc + 1 */
241 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
242 *insn++ = BPF_EXIT_INSN();
243 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
244 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
245 offsetof(struct net_device, ifindex));
246 else
247 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
248 offsetof(struct net_device, type));
249 break;
250
251 case SKF_AD_OFF + SKF_AD_MARK:
252 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
253 insn += cnt - 1;
254 break;
255
256 case SKF_AD_OFF + SKF_AD_RXHASH:
257 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
258
259 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
260 offsetof(struct sk_buff, hash));
261 break;
262
263 case SKF_AD_OFF + SKF_AD_QUEUE:
264 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
265 insn += cnt - 1;
266 break;
267
268 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
269 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
270 BPF_REG_A, BPF_REG_CTX, insn);
271 insn += cnt - 1;
272 break;
273
274 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
275 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
276 BPF_REG_A, BPF_REG_CTX, insn);
277 insn += cnt - 1;
278 break;
279
280 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
281 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2);
282
283 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
284 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
285 offsetof(struct sk_buff, vlan_proto));
286 /* A = ntohs(A) [emitting a nop or swap16] */
287 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
288 break;
289
290 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
291 case SKF_AD_OFF + SKF_AD_NLATTR:
292 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
293 case SKF_AD_OFF + SKF_AD_CPU:
294 case SKF_AD_OFF + SKF_AD_RANDOM:
295 /* arg1 = CTX */
296 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
297 /* arg2 = A */
298 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
299 /* arg3 = X */
300 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
301 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
302 switch (fp->k) {
303 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
304 *insn = BPF_EMIT_CALL(__skb_get_pay_offset);
305 break;
306 case SKF_AD_OFF + SKF_AD_NLATTR:
307 *insn = BPF_EMIT_CALL(__skb_get_nlattr);
308 break;
309 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
310 *insn = BPF_EMIT_CALL(__skb_get_nlattr_nest);
311 break;
312 case SKF_AD_OFF + SKF_AD_CPU:
313 *insn = BPF_EMIT_CALL(__get_raw_cpu_id);
314 break;
315 case SKF_AD_OFF + SKF_AD_RANDOM:
316 *insn = BPF_EMIT_CALL(__get_random_u32);
317 break;
318 }
319 break;
320
321 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
322 /* A ^= X */
323 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
324 break;
325
326 default:
327 /* This is just a dummy call to avoid letting the compiler
328 * evict __bpf_call_base() as an optimization. Placed here
329 * where no-one bothers.
330 */
331 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
332 return false;
333 }
334
335 *insnp = insn;
336 return true;
337 }
338
339 /**
340 * bpf_convert_filter - convert filter program
341 * @prog: the user passed filter program
342 * @len: the length of the user passed filter program
343 * @new_prog: buffer where converted program will be stored
344 * @new_len: pointer to store length of converted program
345 *
346 * Remap 'sock_filter' style BPF instruction set to 'sock_filter_ext' style.
347 * Conversion workflow:
348 *
349 * 1) First pass for calculating the new program length:
350 * bpf_convert_filter(old_prog, old_len, NULL, &new_len)
351 *
352 * 2) 2nd pass to remap in two passes: 1st pass finds new
353 * jump offsets, 2nd pass remapping:
354 * new_prog = kmalloc(sizeof(struct bpf_insn) * new_len);
355 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len);
356 *
357 * User BPF's register A is mapped to our BPF register 6, user BPF
358 * register X is mapped to BPF register 7; frame pointer is always
359 * register 10; Context 'void *ctx' is stored in register 1, that is,
360 * for socket filters: ctx == 'struct sk_buff *', for seccomp:
361 * ctx == 'struct seccomp_data *'.
362 */
363 static int bpf_convert_filter(struct sock_filter *prog, int len,
364 struct bpf_insn *new_prog, int *new_len)
365 {
366 int new_flen = 0, pass = 0, target, i;
367 struct bpf_insn *new_insn;
368 struct sock_filter *fp;
369 int *addrs = NULL;
370 u8 bpf_src;
371
372 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
373 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
374
375 if (len <= 0 || len > BPF_MAXINSNS)
376 return -EINVAL;
377
378 if (new_prog) {
379 addrs = kcalloc(len, sizeof(*addrs),
380 GFP_KERNEL | __GFP_NOWARN);
381 if (!addrs)
382 return -ENOMEM;
383 }
384
385 do_pass:
386 new_insn = new_prog;
387 fp = prog;
388
389 if (new_insn)
390 *new_insn = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
391 new_insn++;
392
393 for (i = 0; i < len; fp++, i++) {
394 struct bpf_insn tmp_insns[6] = { };
395 struct bpf_insn *insn = tmp_insns;
396
397 if (addrs)
398 addrs[i] = new_insn - new_prog;
399
400 switch (fp->code) {
401 /* All arithmetic insns and skb loads map as-is. */
402 case BPF_ALU | BPF_ADD | BPF_X:
403 case BPF_ALU | BPF_ADD | BPF_K:
404 case BPF_ALU | BPF_SUB | BPF_X:
405 case BPF_ALU | BPF_SUB | BPF_K:
406 case BPF_ALU | BPF_AND | BPF_X:
407 case BPF_ALU | BPF_AND | BPF_K:
408 case BPF_ALU | BPF_OR | BPF_X:
409 case BPF_ALU | BPF_OR | BPF_K:
410 case BPF_ALU | BPF_LSH | BPF_X:
411 case BPF_ALU | BPF_LSH | BPF_K:
412 case BPF_ALU | BPF_RSH | BPF_X:
413 case BPF_ALU | BPF_RSH | BPF_K:
414 case BPF_ALU | BPF_XOR | BPF_X:
415 case BPF_ALU | BPF_XOR | BPF_K:
416 case BPF_ALU | BPF_MUL | BPF_X:
417 case BPF_ALU | BPF_MUL | BPF_K:
418 case BPF_ALU | BPF_DIV | BPF_X:
419 case BPF_ALU | BPF_DIV | BPF_K:
420 case BPF_ALU | BPF_MOD | BPF_X:
421 case BPF_ALU | BPF_MOD | BPF_K:
422 case BPF_ALU | BPF_NEG:
423 case BPF_LD | BPF_ABS | BPF_W:
424 case BPF_LD | BPF_ABS | BPF_H:
425 case BPF_LD | BPF_ABS | BPF_B:
426 case BPF_LD | BPF_IND | BPF_W:
427 case BPF_LD | BPF_IND | BPF_H:
428 case BPF_LD | BPF_IND | BPF_B:
429 /* Check for overloaded BPF extension and
430 * directly convert it if found, otherwise
431 * just move on with mapping.
432 */
433 if (BPF_CLASS(fp->code) == BPF_LD &&
434 BPF_MODE(fp->code) == BPF_ABS &&
435 convert_bpf_extensions(fp, &insn))
436 break;
437
438 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
439 break;
440
441 /* Jump transformation cannot use BPF block macros
442 * everywhere as offset calculation and target updates
443 * require a bit more work than the rest, i.e. jump
444 * opcodes map as-is, but offsets need adjustment.
445 */
446
447 #define BPF_EMIT_JMP \
448 do { \
449 if (target >= len || target < 0) \
450 goto err; \
451 insn->off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
452 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
453 insn->off -= insn - tmp_insns; \
454 } while (0)
455
456 case BPF_JMP | BPF_JA:
457 target = i + fp->k + 1;
458 insn->code = fp->code;
459 BPF_EMIT_JMP;
460 break;
461
462 case BPF_JMP | BPF_JEQ | BPF_K:
463 case BPF_JMP | BPF_JEQ | BPF_X:
464 case BPF_JMP | BPF_JSET | BPF_K:
465 case BPF_JMP | BPF_JSET | BPF_X:
466 case BPF_JMP | BPF_JGT | BPF_K:
467 case BPF_JMP | BPF_JGT | BPF_X:
468 case BPF_JMP | BPF_JGE | BPF_K:
469 case BPF_JMP | BPF_JGE | BPF_X:
470 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
471 /* BPF immediates are signed, zero extend
472 * immediate into tmp register and use it
473 * in compare insn.
474 */
475 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
476
477 insn->dst_reg = BPF_REG_A;
478 insn->src_reg = BPF_REG_TMP;
479 bpf_src = BPF_X;
480 } else {
481 insn->dst_reg = BPF_REG_A;
482 insn->imm = fp->k;
483 bpf_src = BPF_SRC(fp->code);
484 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
485 }
486
487 /* Common case where 'jump_false' is next insn. */
488 if (fp->jf == 0) {
489 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
490 target = i + fp->jt + 1;
491 BPF_EMIT_JMP;
492 break;
493 }
494
495 /* Convert JEQ into JNE when 'jump_true' is next insn. */
496 if (fp->jt == 0 && BPF_OP(fp->code) == BPF_JEQ) {
497 insn->code = BPF_JMP | BPF_JNE | bpf_src;
498 target = i + fp->jf + 1;
499 BPF_EMIT_JMP;
500 break;
501 }
502
503 /* Other jumps are mapped into two insns: Jxx and JA. */
504 target = i + fp->jt + 1;
505 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
506 BPF_EMIT_JMP;
507 insn++;
508
509 insn->code = BPF_JMP | BPF_JA;
510 target = i + fp->jf + 1;
511 BPF_EMIT_JMP;
512 break;
513
514 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
515 case BPF_LDX | BPF_MSH | BPF_B:
516 /* tmp = A */
517 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_A);
518 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
519 *insn++ = BPF_LD_ABS(BPF_B, fp->k);
520 /* A &= 0xf */
521 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
522 /* A <<= 2 */
523 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
524 /* X = A */
525 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
526 /* A = tmp */
527 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
528 break;
529
530 /* RET_K, RET_A are remaped into 2 insns. */
531 case BPF_RET | BPF_A:
532 case BPF_RET | BPF_K:
533 *insn++ = BPF_MOV32_RAW(BPF_RVAL(fp->code) == BPF_K ?
534 BPF_K : BPF_X, BPF_REG_0,
535 BPF_REG_A, fp->k);
536 *insn = BPF_EXIT_INSN();
537 break;
538
539 /* Store to stack. */
540 case BPF_ST:
541 case BPF_STX:
542 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
543 BPF_ST ? BPF_REG_A : BPF_REG_X,
544 -(BPF_MEMWORDS - fp->k) * 4);
545 break;
546
547 /* Load from stack. */
548 case BPF_LD | BPF_MEM:
549 case BPF_LDX | BPF_MEM:
550 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
551 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
552 -(BPF_MEMWORDS - fp->k) * 4);
553 break;
554
555 /* A = K or X = K */
556 case BPF_LD | BPF_IMM:
557 case BPF_LDX | BPF_IMM:
558 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
559 BPF_REG_A : BPF_REG_X, fp->k);
560 break;
561
562 /* X = A */
563 case BPF_MISC | BPF_TAX:
564 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
565 break;
566
567 /* A = X */
568 case BPF_MISC | BPF_TXA:
569 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
570 break;
571
572 /* A = skb->len or X = skb->len */
573 case BPF_LD | BPF_W | BPF_LEN:
574 case BPF_LDX | BPF_W | BPF_LEN:
575 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
576 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
577 offsetof(struct sk_buff, len));
578 break;
579
580 /* Access seccomp_data fields. */
581 case BPF_LDX | BPF_ABS | BPF_W:
582 /* A = *(u32 *) (ctx + K) */
583 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
584 break;
585
586 /* Unknown instruction. */
587 default:
588 goto err;
589 }
590
591 insn++;
592 if (new_prog)
593 memcpy(new_insn, tmp_insns,
594 sizeof(*insn) * (insn - tmp_insns));
595 new_insn += insn - tmp_insns;
596 }
597
598 if (!new_prog) {
599 /* Only calculating new length. */
600 *new_len = new_insn - new_prog;
601 return 0;
602 }
603
604 pass++;
605 if (new_flen != new_insn - new_prog) {
606 new_flen = new_insn - new_prog;
607 if (pass > 2)
608 goto err;
609 goto do_pass;
610 }
611
612 kfree(addrs);
613 BUG_ON(*new_len != new_flen);
614 return 0;
615 err:
616 kfree(addrs);
617 return -EINVAL;
618 }
619
620 /* Security:
621 *
622 * As we dont want to clear mem[] array for each packet going through
623 * __bpf_prog_run(), we check that filter loaded by user never try to read
624 * a cell if not previously written, and we check all branches to be sure
625 * a malicious user doesn't try to abuse us.
626 */
627 static int check_load_and_stores(const struct sock_filter *filter, int flen)
628 {
629 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
630 int pc, ret = 0;
631
632 BUILD_BUG_ON(BPF_MEMWORDS > 16);
633
634 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
635 if (!masks)
636 return -ENOMEM;
637
638 memset(masks, 0xff, flen * sizeof(*masks));
639
640 for (pc = 0; pc < flen; pc++) {
641 memvalid &= masks[pc];
642
643 switch (filter[pc].code) {
644 case BPF_ST:
645 case BPF_STX:
646 memvalid |= (1 << filter[pc].k);
647 break;
648 case BPF_LD | BPF_MEM:
649 case BPF_LDX | BPF_MEM:
650 if (!(memvalid & (1 << filter[pc].k))) {
651 ret = -EINVAL;
652 goto error;
653 }
654 break;
655 case BPF_JMP | BPF_JA:
656 /* A jump must set masks on target */
657 masks[pc + 1 + filter[pc].k] &= memvalid;
658 memvalid = ~0;
659 break;
660 case BPF_JMP | BPF_JEQ | BPF_K:
661 case BPF_JMP | BPF_JEQ | BPF_X:
662 case BPF_JMP | BPF_JGE | BPF_K:
663 case BPF_JMP | BPF_JGE | BPF_X:
664 case BPF_JMP | BPF_JGT | BPF_K:
665 case BPF_JMP | BPF_JGT | BPF_X:
666 case BPF_JMP | BPF_JSET | BPF_K:
667 case BPF_JMP | BPF_JSET | BPF_X:
668 /* A jump must set masks on targets */
669 masks[pc + 1 + filter[pc].jt] &= memvalid;
670 masks[pc + 1 + filter[pc].jf] &= memvalid;
671 memvalid = ~0;
672 break;
673 }
674 }
675 error:
676 kfree(masks);
677 return ret;
678 }
679
680 static bool chk_code_allowed(u16 code_to_probe)
681 {
682 static const bool codes[] = {
683 /* 32 bit ALU operations */
684 [BPF_ALU | BPF_ADD | BPF_K] = true,
685 [BPF_ALU | BPF_ADD | BPF_X] = true,
686 [BPF_ALU | BPF_SUB | BPF_K] = true,
687 [BPF_ALU | BPF_SUB | BPF_X] = true,
688 [BPF_ALU | BPF_MUL | BPF_K] = true,
689 [BPF_ALU | BPF_MUL | BPF_X] = true,
690 [BPF_ALU | BPF_DIV | BPF_K] = true,
691 [BPF_ALU | BPF_DIV | BPF_X] = true,
692 [BPF_ALU | BPF_MOD | BPF_K] = true,
693 [BPF_ALU | BPF_MOD | BPF_X] = true,
694 [BPF_ALU | BPF_AND | BPF_K] = true,
695 [BPF_ALU | BPF_AND | BPF_X] = true,
696 [BPF_ALU | BPF_OR | BPF_K] = true,
697 [BPF_ALU | BPF_OR | BPF_X] = true,
698 [BPF_ALU | BPF_XOR | BPF_K] = true,
699 [BPF_ALU | BPF_XOR | BPF_X] = true,
700 [BPF_ALU | BPF_LSH | BPF_K] = true,
701 [BPF_ALU | BPF_LSH | BPF_X] = true,
702 [BPF_ALU | BPF_RSH | BPF_K] = true,
703 [BPF_ALU | BPF_RSH | BPF_X] = true,
704 [BPF_ALU | BPF_NEG] = true,
705 /* Load instructions */
706 [BPF_LD | BPF_W | BPF_ABS] = true,
707 [BPF_LD | BPF_H | BPF_ABS] = true,
708 [BPF_LD | BPF_B | BPF_ABS] = true,
709 [BPF_LD | BPF_W | BPF_LEN] = true,
710 [BPF_LD | BPF_W | BPF_IND] = true,
711 [BPF_LD | BPF_H | BPF_IND] = true,
712 [BPF_LD | BPF_B | BPF_IND] = true,
713 [BPF_LD | BPF_IMM] = true,
714 [BPF_LD | BPF_MEM] = true,
715 [BPF_LDX | BPF_W | BPF_LEN] = true,
716 [BPF_LDX | BPF_B | BPF_MSH] = true,
717 [BPF_LDX | BPF_IMM] = true,
718 [BPF_LDX | BPF_MEM] = true,
719 /* Store instructions */
720 [BPF_ST] = true,
721 [BPF_STX] = true,
722 /* Misc instructions */
723 [BPF_MISC | BPF_TAX] = true,
724 [BPF_MISC | BPF_TXA] = true,
725 /* Return instructions */
726 [BPF_RET | BPF_K] = true,
727 [BPF_RET | BPF_A] = true,
728 /* Jump instructions */
729 [BPF_JMP | BPF_JA] = true,
730 [BPF_JMP | BPF_JEQ | BPF_K] = true,
731 [BPF_JMP | BPF_JEQ | BPF_X] = true,
732 [BPF_JMP | BPF_JGE | BPF_K] = true,
733 [BPF_JMP | BPF_JGE | BPF_X] = true,
734 [BPF_JMP | BPF_JGT | BPF_K] = true,
735 [BPF_JMP | BPF_JGT | BPF_X] = true,
736 [BPF_JMP | BPF_JSET | BPF_K] = true,
737 [BPF_JMP | BPF_JSET | BPF_X] = true,
738 };
739
740 if (code_to_probe >= ARRAY_SIZE(codes))
741 return false;
742
743 return codes[code_to_probe];
744 }
745
746 /**
747 * bpf_check_classic - verify socket filter code
748 * @filter: filter to verify
749 * @flen: length of filter
750 *
751 * Check the user's filter code. If we let some ugly
752 * filter code slip through kaboom! The filter must contain
753 * no references or jumps that are out of range, no illegal
754 * instructions, and must end with a RET instruction.
755 *
756 * All jumps are forward as they are not signed.
757 *
758 * Returns 0 if the rule set is legal or -EINVAL if not.
759 */
760 static int bpf_check_classic(const struct sock_filter *filter,
761 unsigned int flen)
762 {
763 bool anc_found;
764 int pc;
765
766 if (flen == 0 || flen > BPF_MAXINSNS)
767 return -EINVAL;
768
769 /* Check the filter code now */
770 for (pc = 0; pc < flen; pc++) {
771 const struct sock_filter *ftest = &filter[pc];
772
773 /* May we actually operate on this code? */
774 if (!chk_code_allowed(ftest->code))
775 return -EINVAL;
776
777 /* Some instructions need special checks */
778 switch (ftest->code) {
779 case BPF_ALU | BPF_DIV | BPF_K:
780 case BPF_ALU | BPF_MOD | BPF_K:
781 /* Check for division by zero */
782 if (ftest->k == 0)
783 return -EINVAL;
784 break;
785 case BPF_LD | BPF_MEM:
786 case BPF_LDX | BPF_MEM:
787 case BPF_ST:
788 case BPF_STX:
789 /* Check for invalid memory addresses */
790 if (ftest->k >= BPF_MEMWORDS)
791 return -EINVAL;
792 break;
793 case BPF_JMP | BPF_JA:
794 /* Note, the large ftest->k might cause loops.
795 * Compare this with conditional jumps below,
796 * where offsets are limited. --ANK (981016)
797 */
798 if (ftest->k >= (unsigned int)(flen - pc - 1))
799 return -EINVAL;
800 break;
801 case BPF_JMP | BPF_JEQ | BPF_K:
802 case BPF_JMP | BPF_JEQ | BPF_X:
803 case BPF_JMP | BPF_JGE | BPF_K:
804 case BPF_JMP | BPF_JGE | BPF_X:
805 case BPF_JMP | BPF_JGT | BPF_K:
806 case BPF_JMP | BPF_JGT | BPF_X:
807 case BPF_JMP | BPF_JSET | BPF_K:
808 case BPF_JMP | BPF_JSET | BPF_X:
809 /* Both conditionals must be safe */
810 if (pc + ftest->jt + 1 >= flen ||
811 pc + ftest->jf + 1 >= flen)
812 return -EINVAL;
813 break;
814 case BPF_LD | BPF_W | BPF_ABS:
815 case BPF_LD | BPF_H | BPF_ABS:
816 case BPF_LD | BPF_B | BPF_ABS:
817 anc_found = false;
818 if (bpf_anc_helper(ftest) & BPF_ANC)
819 anc_found = true;
820 /* Ancillary operation unknown or unsupported */
821 if (anc_found == false && ftest->k >= SKF_AD_OFF)
822 return -EINVAL;
823 }
824 }
825
826 /* Last instruction must be a RET code */
827 switch (filter[flen - 1].code) {
828 case BPF_RET | BPF_K:
829 case BPF_RET | BPF_A:
830 return check_load_and_stores(filter, flen);
831 }
832
833 return -EINVAL;
834 }
835
836 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
837 const struct sock_fprog *fprog)
838 {
839 unsigned int fsize = bpf_classic_proglen(fprog);
840 struct sock_fprog_kern *fkprog;
841
842 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
843 if (!fp->orig_prog)
844 return -ENOMEM;
845
846 fkprog = fp->orig_prog;
847 fkprog->len = fprog->len;
848
849 fkprog->filter = kmemdup(fp->insns, fsize,
850 GFP_KERNEL | __GFP_NOWARN);
851 if (!fkprog->filter) {
852 kfree(fp->orig_prog);
853 return -ENOMEM;
854 }
855
856 return 0;
857 }
858
859 static void bpf_release_orig_filter(struct bpf_prog *fp)
860 {
861 struct sock_fprog_kern *fprog = fp->orig_prog;
862
863 if (fprog) {
864 kfree(fprog->filter);
865 kfree(fprog);
866 }
867 }
868
869 static void __bpf_prog_release(struct bpf_prog *prog)
870 {
871 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
872 bpf_prog_put(prog);
873 } else {
874 bpf_release_orig_filter(prog);
875 bpf_prog_free(prog);
876 }
877 }
878
879 static void __sk_filter_release(struct sk_filter *fp)
880 {
881 __bpf_prog_release(fp->prog);
882 kfree(fp);
883 }
884
885 /**
886 * sk_filter_release_rcu - Release a socket filter by rcu_head
887 * @rcu: rcu_head that contains the sk_filter to free
888 */
889 static void sk_filter_release_rcu(struct rcu_head *rcu)
890 {
891 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
892
893 __sk_filter_release(fp);
894 }
895
896 /**
897 * sk_filter_release - release a socket filter
898 * @fp: filter to remove
899 *
900 * Remove a filter from a socket and release its resources.
901 */
902 static void sk_filter_release(struct sk_filter *fp)
903 {
904 if (atomic_dec_and_test(&fp->refcnt))
905 call_rcu(&fp->rcu, sk_filter_release_rcu);
906 }
907
908 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
909 {
910 u32 filter_size = bpf_prog_size(fp->prog->len);
911
912 atomic_sub(filter_size, &sk->sk_omem_alloc);
913 sk_filter_release(fp);
914 }
915
916 /* try to charge the socket memory if there is space available
917 * return true on success
918 */
919 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
920 {
921 u32 filter_size = bpf_prog_size(fp->prog->len);
922
923 /* same check as in sock_kmalloc() */
924 if (filter_size <= sysctl_optmem_max &&
925 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
926 atomic_inc(&fp->refcnt);
927 atomic_add(filter_size, &sk->sk_omem_alloc);
928 return true;
929 }
930 return false;
931 }
932
933 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
934 {
935 struct sock_filter *old_prog;
936 struct bpf_prog *old_fp;
937 int err, new_len, old_len = fp->len;
938
939 /* We are free to overwrite insns et al right here as it
940 * won't be used at this point in time anymore internally
941 * after the migration to the internal BPF instruction
942 * representation.
943 */
944 BUILD_BUG_ON(sizeof(struct sock_filter) !=
945 sizeof(struct bpf_insn));
946
947 /* Conversion cannot happen on overlapping memory areas,
948 * so we need to keep the user BPF around until the 2nd
949 * pass. At this time, the user BPF is stored in fp->insns.
950 */
951 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
952 GFP_KERNEL | __GFP_NOWARN);
953 if (!old_prog) {
954 err = -ENOMEM;
955 goto out_err;
956 }
957
958 /* 1st pass: calculate the new program length. */
959 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len);
960 if (err)
961 goto out_err_free;
962
963 /* Expand fp for appending the new filter representation. */
964 old_fp = fp;
965 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
966 if (!fp) {
967 /* The old_fp is still around in case we couldn't
968 * allocate new memory, so uncharge on that one.
969 */
970 fp = old_fp;
971 err = -ENOMEM;
972 goto out_err_free;
973 }
974
975 fp->len = new_len;
976
977 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
978 err = bpf_convert_filter(old_prog, old_len, fp->insnsi, &new_len);
979 if (err)
980 /* 2nd bpf_convert_filter() can fail only if it fails
981 * to allocate memory, remapping must succeed. Note,
982 * that at this time old_fp has already been released
983 * by krealloc().
984 */
985 goto out_err_free;
986
987 bpf_prog_select_runtime(fp);
988
989 kfree(old_prog);
990 return fp;
991
992 out_err_free:
993 kfree(old_prog);
994 out_err:
995 __bpf_prog_release(fp);
996 return ERR_PTR(err);
997 }
998
999 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1000 bpf_aux_classic_check_t trans)
1001 {
1002 int err;
1003
1004 fp->bpf_func = NULL;
1005 fp->jited = 0;
1006
1007 err = bpf_check_classic(fp->insns, fp->len);
1008 if (err) {
1009 __bpf_prog_release(fp);
1010 return ERR_PTR(err);
1011 }
1012
1013 /* There might be additional checks and transformations
1014 * needed on classic filters, f.e. in case of seccomp.
1015 */
1016 if (trans) {
1017 err = trans(fp->insns, fp->len);
1018 if (err) {
1019 __bpf_prog_release(fp);
1020 return ERR_PTR(err);
1021 }
1022 }
1023
1024 /* Probe if we can JIT compile the filter and if so, do
1025 * the compilation of the filter.
1026 */
1027 bpf_jit_compile(fp);
1028
1029 /* JIT compiler couldn't process this filter, so do the
1030 * internal BPF translation for the optimized interpreter.
1031 */
1032 if (!fp->jited)
1033 fp = bpf_migrate_filter(fp);
1034
1035 return fp;
1036 }
1037
1038 /**
1039 * bpf_prog_create - create an unattached filter
1040 * @pfp: the unattached filter that is created
1041 * @fprog: the filter program
1042 *
1043 * Create a filter independent of any socket. We first run some
1044 * sanity checks on it to make sure it does not explode on us later.
1045 * If an error occurs or there is insufficient memory for the filter
1046 * a negative errno code is returned. On success the return is zero.
1047 */
1048 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1049 {
1050 unsigned int fsize = bpf_classic_proglen(fprog);
1051 struct bpf_prog *fp;
1052
1053 /* Make sure new filter is there and in the right amounts. */
1054 if (fprog->filter == NULL)
1055 return -EINVAL;
1056
1057 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1058 if (!fp)
1059 return -ENOMEM;
1060
1061 memcpy(fp->insns, fprog->filter, fsize);
1062
1063 fp->len = fprog->len;
1064 /* Since unattached filters are not copied back to user
1065 * space through sk_get_filter(), we do not need to hold
1066 * a copy here, and can spare us the work.
1067 */
1068 fp->orig_prog = NULL;
1069
1070 /* bpf_prepare_filter() already takes care of freeing
1071 * memory in case something goes wrong.
1072 */
1073 fp = bpf_prepare_filter(fp, NULL);
1074 if (IS_ERR(fp))
1075 return PTR_ERR(fp);
1076
1077 *pfp = fp;
1078 return 0;
1079 }
1080 EXPORT_SYMBOL_GPL(bpf_prog_create);
1081
1082 /**
1083 * bpf_prog_create_from_user - create an unattached filter from user buffer
1084 * @pfp: the unattached filter that is created
1085 * @fprog: the filter program
1086 * @trans: post-classic verifier transformation handler
1087 * @save_orig: save classic BPF program
1088 *
1089 * This function effectively does the same as bpf_prog_create(), only
1090 * that it builds up its insns buffer from user space provided buffer.
1091 * It also allows for passing a bpf_aux_classic_check_t handler.
1092 */
1093 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1094 bpf_aux_classic_check_t trans, bool save_orig)
1095 {
1096 unsigned int fsize = bpf_classic_proglen(fprog);
1097 struct bpf_prog *fp;
1098 int err;
1099
1100 /* Make sure new filter is there and in the right amounts. */
1101 if (fprog->filter == NULL)
1102 return -EINVAL;
1103
1104 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1105 if (!fp)
1106 return -ENOMEM;
1107
1108 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1109 __bpf_prog_free(fp);
1110 return -EFAULT;
1111 }
1112
1113 fp->len = fprog->len;
1114 fp->orig_prog = NULL;
1115
1116 if (save_orig) {
1117 err = bpf_prog_store_orig_filter(fp, fprog);
1118 if (err) {
1119 __bpf_prog_free(fp);
1120 return -ENOMEM;
1121 }
1122 }
1123
1124 /* bpf_prepare_filter() already takes care of freeing
1125 * memory in case something goes wrong.
1126 */
1127 fp = bpf_prepare_filter(fp, trans);
1128 if (IS_ERR(fp))
1129 return PTR_ERR(fp);
1130
1131 *pfp = fp;
1132 return 0;
1133 }
1134 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1135
1136 void bpf_prog_destroy(struct bpf_prog *fp)
1137 {
1138 __bpf_prog_release(fp);
1139 }
1140 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1141
1142 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1143 {
1144 struct sk_filter *fp, *old_fp;
1145
1146 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1147 if (!fp)
1148 return -ENOMEM;
1149
1150 fp->prog = prog;
1151 atomic_set(&fp->refcnt, 0);
1152
1153 if (!sk_filter_charge(sk, fp)) {
1154 kfree(fp);
1155 return -ENOMEM;
1156 }
1157
1158 old_fp = rcu_dereference_protected(sk->sk_filter,
1159 sock_owned_by_user(sk));
1160 rcu_assign_pointer(sk->sk_filter, fp);
1161
1162 if (old_fp)
1163 sk_filter_uncharge(sk, old_fp);
1164
1165 return 0;
1166 }
1167
1168 /**
1169 * sk_attach_filter - attach a socket filter
1170 * @fprog: the filter program
1171 * @sk: the socket to use
1172 *
1173 * Attach the user's filter code. We first run some sanity checks on
1174 * it to make sure it does not explode on us later. If an error
1175 * occurs or there is insufficient memory for the filter a negative
1176 * errno code is returned. On success the return is zero.
1177 */
1178 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1179 {
1180 unsigned int fsize = bpf_classic_proglen(fprog);
1181 unsigned int bpf_fsize = bpf_prog_size(fprog->len);
1182 struct bpf_prog *prog;
1183 int err;
1184
1185 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1186 return -EPERM;
1187
1188 /* Make sure new filter is there and in the right amounts. */
1189 if (fprog->filter == NULL)
1190 return -EINVAL;
1191
1192 prog = bpf_prog_alloc(bpf_fsize, 0);
1193 if (!prog)
1194 return -ENOMEM;
1195
1196 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1197 __bpf_prog_free(prog);
1198 return -EFAULT;
1199 }
1200
1201 prog->len = fprog->len;
1202
1203 err = bpf_prog_store_orig_filter(prog, fprog);
1204 if (err) {
1205 __bpf_prog_free(prog);
1206 return -ENOMEM;
1207 }
1208
1209 /* bpf_prepare_filter() already takes care of freeing
1210 * memory in case something goes wrong.
1211 */
1212 prog = bpf_prepare_filter(prog, NULL);
1213 if (IS_ERR(prog))
1214 return PTR_ERR(prog);
1215
1216 err = __sk_attach_prog(prog, sk);
1217 if (err < 0) {
1218 __bpf_prog_release(prog);
1219 return err;
1220 }
1221
1222 return 0;
1223 }
1224 EXPORT_SYMBOL_GPL(sk_attach_filter);
1225
1226 int sk_attach_bpf(u32 ufd, struct sock *sk)
1227 {
1228 struct bpf_prog *prog;
1229 int err;
1230
1231 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1232 return -EPERM;
1233
1234 prog = bpf_prog_get(ufd);
1235 if (IS_ERR(prog))
1236 return PTR_ERR(prog);
1237
1238 if (prog->type != BPF_PROG_TYPE_SOCKET_FILTER) {
1239 bpf_prog_put(prog);
1240 return -EINVAL;
1241 }
1242
1243 err = __sk_attach_prog(prog, sk);
1244 if (err < 0) {
1245 bpf_prog_put(prog);
1246 return err;
1247 }
1248
1249 return 0;
1250 }
1251
1252 #define BPF_RECOMPUTE_CSUM(flags) ((flags) & 1)
1253
1254 static u64 bpf_skb_store_bytes(u64 r1, u64 r2, u64 r3, u64 r4, u64 flags)
1255 {
1256 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1257 int offset = (int) r2;
1258 void *from = (void *) (long) r3;
1259 unsigned int len = (unsigned int) r4;
1260 char buf[16];
1261 void *ptr;
1262
1263 /* bpf verifier guarantees that:
1264 * 'from' pointer points to bpf program stack
1265 * 'len' bytes of it were initialized
1266 * 'len' > 0
1267 * 'skb' is a valid pointer to 'struct sk_buff'
1268 *
1269 * so check for invalid 'offset' and too large 'len'
1270 */
1271 if (unlikely((u32) offset > 0xffff || len > sizeof(buf)))
1272 return -EFAULT;
1273
1274 if (unlikely(skb_cloned(skb) &&
1275 !skb_clone_writable(skb, offset + len)))
1276 return -EFAULT;
1277
1278 ptr = skb_header_pointer(skb, offset, len, buf);
1279 if (unlikely(!ptr))
1280 return -EFAULT;
1281
1282 if (BPF_RECOMPUTE_CSUM(flags))
1283 skb_postpull_rcsum(skb, ptr, len);
1284
1285 memcpy(ptr, from, len);
1286
1287 if (ptr == buf)
1288 /* skb_store_bits cannot return -EFAULT here */
1289 skb_store_bits(skb, offset, ptr, len);
1290
1291 if (BPF_RECOMPUTE_CSUM(flags) && skb->ip_summed == CHECKSUM_COMPLETE)
1292 skb->csum = csum_add(skb->csum, csum_partial(ptr, len, 0));
1293 return 0;
1294 }
1295
1296 const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1297 .func = bpf_skb_store_bytes,
1298 .gpl_only = false,
1299 .ret_type = RET_INTEGER,
1300 .arg1_type = ARG_PTR_TO_CTX,
1301 .arg2_type = ARG_ANYTHING,
1302 .arg3_type = ARG_PTR_TO_STACK,
1303 .arg4_type = ARG_CONST_STACK_SIZE,
1304 .arg5_type = ARG_ANYTHING,
1305 };
1306
1307 #define BPF_HEADER_FIELD_SIZE(flags) ((flags) & 0x0f)
1308 #define BPF_IS_PSEUDO_HEADER(flags) ((flags) & 0x10)
1309
1310 static u64 bpf_l3_csum_replace(u64 r1, u64 r2, u64 from, u64 to, u64 flags)
1311 {
1312 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1313 int offset = (int) r2;
1314 __sum16 sum, *ptr;
1315
1316 if (unlikely((u32) offset > 0xffff))
1317 return -EFAULT;
1318
1319 if (unlikely(skb_cloned(skb) &&
1320 !skb_clone_writable(skb, offset + sizeof(sum))))
1321 return -EFAULT;
1322
1323 ptr = skb_header_pointer(skb, offset, sizeof(sum), &sum);
1324 if (unlikely(!ptr))
1325 return -EFAULT;
1326
1327 switch (BPF_HEADER_FIELD_SIZE(flags)) {
1328 case 2:
1329 csum_replace2(ptr, from, to);
1330 break;
1331 case 4:
1332 csum_replace4(ptr, from, to);
1333 break;
1334 default:
1335 return -EINVAL;
1336 }
1337
1338 if (ptr == &sum)
1339 /* skb_store_bits guaranteed to not return -EFAULT here */
1340 skb_store_bits(skb, offset, ptr, sizeof(sum));
1341
1342 return 0;
1343 }
1344
1345 const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1346 .func = bpf_l3_csum_replace,
1347 .gpl_only = false,
1348 .ret_type = RET_INTEGER,
1349 .arg1_type = ARG_PTR_TO_CTX,
1350 .arg2_type = ARG_ANYTHING,
1351 .arg3_type = ARG_ANYTHING,
1352 .arg4_type = ARG_ANYTHING,
1353 .arg5_type = ARG_ANYTHING,
1354 };
1355
1356 static u64 bpf_l4_csum_replace(u64 r1, u64 r2, u64 from, u64 to, u64 flags)
1357 {
1358 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1359 bool is_pseudo = !!BPF_IS_PSEUDO_HEADER(flags);
1360 int offset = (int) r2;
1361 __sum16 sum, *ptr;
1362
1363 if (unlikely((u32) offset > 0xffff))
1364 return -EFAULT;
1365
1366 if (unlikely(skb_cloned(skb) &&
1367 !skb_clone_writable(skb, offset + sizeof(sum))))
1368 return -EFAULT;
1369
1370 ptr = skb_header_pointer(skb, offset, sizeof(sum), &sum);
1371 if (unlikely(!ptr))
1372 return -EFAULT;
1373
1374 switch (BPF_HEADER_FIELD_SIZE(flags)) {
1375 case 2:
1376 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1377 break;
1378 case 4:
1379 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1380 break;
1381 default:
1382 return -EINVAL;
1383 }
1384
1385 if (ptr == &sum)
1386 /* skb_store_bits guaranteed to not return -EFAULT here */
1387 skb_store_bits(skb, offset, ptr, sizeof(sum));
1388
1389 return 0;
1390 }
1391
1392 const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1393 .func = bpf_l4_csum_replace,
1394 .gpl_only = false,
1395 .ret_type = RET_INTEGER,
1396 .arg1_type = ARG_PTR_TO_CTX,
1397 .arg2_type = ARG_ANYTHING,
1398 .arg3_type = ARG_ANYTHING,
1399 .arg4_type = ARG_ANYTHING,
1400 .arg5_type = ARG_ANYTHING,
1401 };
1402
1403 #define BPF_IS_REDIRECT_INGRESS(flags) ((flags) & 1)
1404
1405 static u64 bpf_clone_redirect(u64 r1, u64 ifindex, u64 flags, u64 r4, u64 r5)
1406 {
1407 struct sk_buff *skb = (struct sk_buff *) (long) r1, *skb2;
1408 struct net_device *dev;
1409
1410 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
1411 if (unlikely(!dev))
1412 return -EINVAL;
1413
1414 skb2 = skb_clone(skb, GFP_ATOMIC);
1415 if (unlikely(!skb2))
1416 return -ENOMEM;
1417
1418 if (BPF_IS_REDIRECT_INGRESS(flags))
1419 return dev_forward_skb(dev, skb2);
1420
1421 skb2->dev = dev;
1422 return dev_queue_xmit(skb2);
1423 }
1424
1425 const struct bpf_func_proto bpf_clone_redirect_proto = {
1426 .func = bpf_clone_redirect,
1427 .gpl_only = false,
1428 .ret_type = RET_INTEGER,
1429 .arg1_type = ARG_PTR_TO_CTX,
1430 .arg2_type = ARG_ANYTHING,
1431 .arg3_type = ARG_ANYTHING,
1432 };
1433
1434 struct redirect_info {
1435 u32 ifindex;
1436 u32 flags;
1437 };
1438
1439 static DEFINE_PER_CPU(struct redirect_info, redirect_info);
1440 static u64 bpf_redirect(u64 ifindex, u64 flags, u64 r3, u64 r4, u64 r5)
1441 {
1442 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
1443
1444 ri->ifindex = ifindex;
1445 ri->flags = flags;
1446 return TC_ACT_REDIRECT;
1447 }
1448
1449 int skb_do_redirect(struct sk_buff *skb)
1450 {
1451 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
1452 struct net_device *dev;
1453
1454 dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->ifindex);
1455 ri->ifindex = 0;
1456 if (unlikely(!dev)) {
1457 kfree_skb(skb);
1458 return -EINVAL;
1459 }
1460
1461 if (BPF_IS_REDIRECT_INGRESS(ri->flags))
1462 return dev_forward_skb(dev, skb);
1463
1464 skb->dev = dev;
1465 return dev_queue_xmit(skb);
1466 }
1467
1468 const struct bpf_func_proto bpf_redirect_proto = {
1469 .func = bpf_redirect,
1470 .gpl_only = false,
1471 .ret_type = RET_INTEGER,
1472 .arg1_type = ARG_ANYTHING,
1473 .arg2_type = ARG_ANYTHING,
1474 };
1475
1476 static u64 bpf_get_cgroup_classid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1477 {
1478 return task_get_classid((struct sk_buff *) (unsigned long) r1);
1479 }
1480
1481 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
1482 .func = bpf_get_cgroup_classid,
1483 .gpl_only = false,
1484 .ret_type = RET_INTEGER,
1485 .arg1_type = ARG_PTR_TO_CTX,
1486 };
1487
1488 static u64 bpf_get_route_realm(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1489 {
1490 #ifdef CONFIG_IP_ROUTE_CLASSID
1491 const struct dst_entry *dst;
1492
1493 dst = skb_dst((struct sk_buff *) (unsigned long) r1);
1494 if (dst)
1495 return dst->tclassid;
1496 #endif
1497 return 0;
1498 }
1499
1500 static const struct bpf_func_proto bpf_get_route_realm_proto = {
1501 .func = bpf_get_route_realm,
1502 .gpl_only = false,
1503 .ret_type = RET_INTEGER,
1504 .arg1_type = ARG_PTR_TO_CTX,
1505 };
1506
1507 static u64 bpf_skb_vlan_push(u64 r1, u64 r2, u64 vlan_tci, u64 r4, u64 r5)
1508 {
1509 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1510 __be16 vlan_proto = (__force __be16) r2;
1511
1512 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
1513 vlan_proto != htons(ETH_P_8021AD)))
1514 vlan_proto = htons(ETH_P_8021Q);
1515
1516 return skb_vlan_push(skb, vlan_proto, vlan_tci);
1517 }
1518
1519 const struct bpf_func_proto bpf_skb_vlan_push_proto = {
1520 .func = bpf_skb_vlan_push,
1521 .gpl_only = false,
1522 .ret_type = RET_INTEGER,
1523 .arg1_type = ARG_PTR_TO_CTX,
1524 .arg2_type = ARG_ANYTHING,
1525 .arg3_type = ARG_ANYTHING,
1526 };
1527 EXPORT_SYMBOL_GPL(bpf_skb_vlan_push_proto);
1528
1529 static u64 bpf_skb_vlan_pop(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1530 {
1531 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1532
1533 return skb_vlan_pop(skb);
1534 }
1535
1536 const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
1537 .func = bpf_skb_vlan_pop,
1538 .gpl_only = false,
1539 .ret_type = RET_INTEGER,
1540 .arg1_type = ARG_PTR_TO_CTX,
1541 };
1542 EXPORT_SYMBOL_GPL(bpf_skb_vlan_pop_proto);
1543
1544 bool bpf_helper_changes_skb_data(void *func)
1545 {
1546 if (func == bpf_skb_vlan_push)
1547 return true;
1548 if (func == bpf_skb_vlan_pop)
1549 return true;
1550 return false;
1551 }
1552
1553 static u64 bpf_skb_get_tunnel_key(u64 r1, u64 r2, u64 size, u64 flags, u64 r5)
1554 {
1555 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1556 struct bpf_tunnel_key *to = (struct bpf_tunnel_key *) (long) r2;
1557 struct ip_tunnel_info *info = skb_tunnel_info(skb);
1558
1559 if (unlikely(size != sizeof(struct bpf_tunnel_key) || flags || !info))
1560 return -EINVAL;
1561 if (ip_tunnel_info_af(info) != AF_INET)
1562 return -EINVAL;
1563
1564 to->tunnel_id = be64_to_cpu(info->key.tun_id);
1565 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
1566
1567 return 0;
1568 }
1569
1570 const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
1571 .func = bpf_skb_get_tunnel_key,
1572 .gpl_only = false,
1573 .ret_type = RET_INTEGER,
1574 .arg1_type = ARG_PTR_TO_CTX,
1575 .arg2_type = ARG_PTR_TO_STACK,
1576 .arg3_type = ARG_CONST_STACK_SIZE,
1577 .arg4_type = ARG_ANYTHING,
1578 };
1579
1580 static struct metadata_dst __percpu *md_dst;
1581
1582 static u64 bpf_skb_set_tunnel_key(u64 r1, u64 r2, u64 size, u64 flags, u64 r5)
1583 {
1584 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1585 struct bpf_tunnel_key *from = (struct bpf_tunnel_key *) (long) r2;
1586 struct metadata_dst *md = this_cpu_ptr(md_dst);
1587 struct ip_tunnel_info *info;
1588
1589 if (unlikely(size != sizeof(struct bpf_tunnel_key) || flags))
1590 return -EINVAL;
1591
1592 skb_dst_drop(skb);
1593 dst_hold((struct dst_entry *) md);
1594 skb_dst_set(skb, (struct dst_entry *) md);
1595
1596 info = &md->u.tun_info;
1597 info->mode = IP_TUNNEL_INFO_TX;
1598 info->key.tun_flags = TUNNEL_KEY;
1599 info->key.tun_id = cpu_to_be64(from->tunnel_id);
1600 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
1601
1602 return 0;
1603 }
1604
1605 const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
1606 .func = bpf_skb_set_tunnel_key,
1607 .gpl_only = false,
1608 .ret_type = RET_INTEGER,
1609 .arg1_type = ARG_PTR_TO_CTX,
1610 .arg2_type = ARG_PTR_TO_STACK,
1611 .arg3_type = ARG_CONST_STACK_SIZE,
1612 .arg4_type = ARG_ANYTHING,
1613 };
1614
1615 static const struct bpf_func_proto *bpf_get_skb_set_tunnel_key_proto(void)
1616 {
1617 if (!md_dst) {
1618 /* race is not possible, since it's called from
1619 * verifier that is holding verifier mutex
1620 */
1621 md_dst = metadata_dst_alloc_percpu(0, GFP_KERNEL);
1622 if (!md_dst)
1623 return NULL;
1624 }
1625 return &bpf_skb_set_tunnel_key_proto;
1626 }
1627
1628 static const struct bpf_func_proto *
1629 sk_filter_func_proto(enum bpf_func_id func_id)
1630 {
1631 switch (func_id) {
1632 case BPF_FUNC_map_lookup_elem:
1633 return &bpf_map_lookup_elem_proto;
1634 case BPF_FUNC_map_update_elem:
1635 return &bpf_map_update_elem_proto;
1636 case BPF_FUNC_map_delete_elem:
1637 return &bpf_map_delete_elem_proto;
1638 case BPF_FUNC_get_prandom_u32:
1639 return &bpf_get_prandom_u32_proto;
1640 case BPF_FUNC_get_smp_processor_id:
1641 return &bpf_get_smp_processor_id_proto;
1642 case BPF_FUNC_tail_call:
1643 return &bpf_tail_call_proto;
1644 case BPF_FUNC_ktime_get_ns:
1645 return &bpf_ktime_get_ns_proto;
1646 case BPF_FUNC_trace_printk:
1647 return bpf_get_trace_printk_proto();
1648 default:
1649 return NULL;
1650 }
1651 }
1652
1653 static const struct bpf_func_proto *
1654 tc_cls_act_func_proto(enum bpf_func_id func_id)
1655 {
1656 switch (func_id) {
1657 case BPF_FUNC_skb_store_bytes:
1658 return &bpf_skb_store_bytes_proto;
1659 case BPF_FUNC_l3_csum_replace:
1660 return &bpf_l3_csum_replace_proto;
1661 case BPF_FUNC_l4_csum_replace:
1662 return &bpf_l4_csum_replace_proto;
1663 case BPF_FUNC_clone_redirect:
1664 return &bpf_clone_redirect_proto;
1665 case BPF_FUNC_get_cgroup_classid:
1666 return &bpf_get_cgroup_classid_proto;
1667 case BPF_FUNC_skb_vlan_push:
1668 return &bpf_skb_vlan_push_proto;
1669 case BPF_FUNC_skb_vlan_pop:
1670 return &bpf_skb_vlan_pop_proto;
1671 case BPF_FUNC_skb_get_tunnel_key:
1672 return &bpf_skb_get_tunnel_key_proto;
1673 case BPF_FUNC_skb_set_tunnel_key:
1674 return bpf_get_skb_set_tunnel_key_proto();
1675 case BPF_FUNC_redirect:
1676 return &bpf_redirect_proto;
1677 case BPF_FUNC_get_route_realm:
1678 return &bpf_get_route_realm_proto;
1679 default:
1680 return sk_filter_func_proto(func_id);
1681 }
1682 }
1683
1684 static bool __is_valid_access(int off, int size, enum bpf_access_type type)
1685 {
1686 /* check bounds */
1687 if (off < 0 || off >= sizeof(struct __sk_buff))
1688 return false;
1689
1690 /* disallow misaligned access */
1691 if (off % size != 0)
1692 return false;
1693
1694 /* all __sk_buff fields are __u32 */
1695 if (size != 4)
1696 return false;
1697
1698 return true;
1699 }
1700
1701 static bool sk_filter_is_valid_access(int off, int size,
1702 enum bpf_access_type type)
1703 {
1704 if (off == offsetof(struct __sk_buff, tc_classid))
1705 return false;
1706
1707 if (type == BPF_WRITE) {
1708 switch (off) {
1709 case offsetof(struct __sk_buff, cb[0]) ...
1710 offsetof(struct __sk_buff, cb[4]):
1711 break;
1712 default:
1713 return false;
1714 }
1715 }
1716
1717 return __is_valid_access(off, size, type);
1718 }
1719
1720 static bool tc_cls_act_is_valid_access(int off, int size,
1721 enum bpf_access_type type)
1722 {
1723 if (off == offsetof(struct __sk_buff, tc_classid))
1724 return type == BPF_WRITE ? true : false;
1725
1726 if (type == BPF_WRITE) {
1727 switch (off) {
1728 case offsetof(struct __sk_buff, mark):
1729 case offsetof(struct __sk_buff, tc_index):
1730 case offsetof(struct __sk_buff, priority):
1731 case offsetof(struct __sk_buff, cb[0]) ...
1732 offsetof(struct __sk_buff, cb[4]):
1733 break;
1734 default:
1735 return false;
1736 }
1737 }
1738 return __is_valid_access(off, size, type);
1739 }
1740
1741 static u32 bpf_net_convert_ctx_access(enum bpf_access_type type, int dst_reg,
1742 int src_reg, int ctx_off,
1743 struct bpf_insn *insn_buf)
1744 {
1745 struct bpf_insn *insn = insn_buf;
1746
1747 switch (ctx_off) {
1748 case offsetof(struct __sk_buff, len):
1749 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, len) != 4);
1750
1751 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
1752 offsetof(struct sk_buff, len));
1753 break;
1754
1755 case offsetof(struct __sk_buff, protocol):
1756 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
1757
1758 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
1759 offsetof(struct sk_buff, protocol));
1760 break;
1761
1762 case offsetof(struct __sk_buff, vlan_proto):
1763 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2);
1764
1765 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
1766 offsetof(struct sk_buff, vlan_proto));
1767 break;
1768
1769 case offsetof(struct __sk_buff, priority):
1770 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, priority) != 4);
1771
1772 if (type == BPF_WRITE)
1773 *insn++ = BPF_STX_MEM(BPF_W, dst_reg, src_reg,
1774 offsetof(struct sk_buff, priority));
1775 else
1776 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
1777 offsetof(struct sk_buff, priority));
1778 break;
1779
1780 case offsetof(struct __sk_buff, ingress_ifindex):
1781 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, skb_iif) != 4);
1782
1783 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
1784 offsetof(struct sk_buff, skb_iif));
1785 break;
1786
1787 case offsetof(struct __sk_buff, ifindex):
1788 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
1789
1790 *insn++ = BPF_LDX_MEM(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, dev)),
1791 dst_reg, src_reg,
1792 offsetof(struct sk_buff, dev));
1793 *insn++ = BPF_JMP_IMM(BPF_JEQ, dst_reg, 0, 1);
1794 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, dst_reg,
1795 offsetof(struct net_device, ifindex));
1796 break;
1797
1798 case offsetof(struct __sk_buff, hash):
1799 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
1800
1801 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
1802 offsetof(struct sk_buff, hash));
1803 break;
1804
1805 case offsetof(struct __sk_buff, mark):
1806 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
1807
1808 if (type == BPF_WRITE)
1809 *insn++ = BPF_STX_MEM(BPF_W, dst_reg, src_reg,
1810 offsetof(struct sk_buff, mark));
1811 else
1812 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
1813 offsetof(struct sk_buff, mark));
1814 break;
1815
1816 case offsetof(struct __sk_buff, pkt_type):
1817 return convert_skb_access(SKF_AD_PKTTYPE, dst_reg, src_reg, insn);
1818
1819 case offsetof(struct __sk_buff, queue_mapping):
1820 return convert_skb_access(SKF_AD_QUEUE, dst_reg, src_reg, insn);
1821
1822 case offsetof(struct __sk_buff, vlan_present):
1823 return convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
1824 dst_reg, src_reg, insn);
1825
1826 case offsetof(struct __sk_buff, vlan_tci):
1827 return convert_skb_access(SKF_AD_VLAN_TAG,
1828 dst_reg, src_reg, insn);
1829
1830 case offsetof(struct __sk_buff, cb[0]) ...
1831 offsetof(struct __sk_buff, cb[4]):
1832 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20);
1833
1834 ctx_off -= offsetof(struct __sk_buff, cb[0]);
1835 ctx_off += offsetof(struct sk_buff, cb);
1836 ctx_off += offsetof(struct qdisc_skb_cb, data);
1837 if (type == BPF_WRITE)
1838 *insn++ = BPF_STX_MEM(BPF_W, dst_reg, src_reg, ctx_off);
1839 else
1840 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, ctx_off);
1841 break;
1842
1843 case offsetof(struct __sk_buff, tc_classid):
1844 ctx_off -= offsetof(struct __sk_buff, tc_classid);
1845 ctx_off += offsetof(struct sk_buff, cb);
1846 ctx_off += offsetof(struct qdisc_skb_cb, tc_classid);
1847 WARN_ON(type != BPF_WRITE);
1848 *insn++ = BPF_STX_MEM(BPF_H, dst_reg, src_reg, ctx_off);
1849 break;
1850
1851 case offsetof(struct __sk_buff, tc_index):
1852 #ifdef CONFIG_NET_SCHED
1853 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, tc_index) != 2);
1854
1855 if (type == BPF_WRITE)
1856 *insn++ = BPF_STX_MEM(BPF_H, dst_reg, src_reg,
1857 offsetof(struct sk_buff, tc_index));
1858 else
1859 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
1860 offsetof(struct sk_buff, tc_index));
1861 break;
1862 #else
1863 if (type == BPF_WRITE)
1864 *insn++ = BPF_MOV64_REG(dst_reg, dst_reg);
1865 else
1866 *insn++ = BPF_MOV64_IMM(dst_reg, 0);
1867 break;
1868 #endif
1869 }
1870
1871 return insn - insn_buf;
1872 }
1873
1874 static const struct bpf_verifier_ops sk_filter_ops = {
1875 .get_func_proto = sk_filter_func_proto,
1876 .is_valid_access = sk_filter_is_valid_access,
1877 .convert_ctx_access = bpf_net_convert_ctx_access,
1878 };
1879
1880 static const struct bpf_verifier_ops tc_cls_act_ops = {
1881 .get_func_proto = tc_cls_act_func_proto,
1882 .is_valid_access = tc_cls_act_is_valid_access,
1883 .convert_ctx_access = bpf_net_convert_ctx_access,
1884 };
1885
1886 static struct bpf_prog_type_list sk_filter_type __read_mostly = {
1887 .ops = &sk_filter_ops,
1888 .type = BPF_PROG_TYPE_SOCKET_FILTER,
1889 };
1890
1891 static struct bpf_prog_type_list sched_cls_type __read_mostly = {
1892 .ops = &tc_cls_act_ops,
1893 .type = BPF_PROG_TYPE_SCHED_CLS,
1894 };
1895
1896 static struct bpf_prog_type_list sched_act_type __read_mostly = {
1897 .ops = &tc_cls_act_ops,
1898 .type = BPF_PROG_TYPE_SCHED_ACT,
1899 };
1900
1901 static int __init register_sk_filter_ops(void)
1902 {
1903 bpf_register_prog_type(&sk_filter_type);
1904 bpf_register_prog_type(&sched_cls_type);
1905 bpf_register_prog_type(&sched_act_type);
1906
1907 return 0;
1908 }
1909 late_initcall(register_sk_filter_ops);
1910
1911 int sk_detach_filter(struct sock *sk)
1912 {
1913 int ret = -ENOENT;
1914 struct sk_filter *filter;
1915
1916 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1917 return -EPERM;
1918
1919 filter = rcu_dereference_protected(sk->sk_filter,
1920 sock_owned_by_user(sk));
1921 if (filter) {
1922 RCU_INIT_POINTER(sk->sk_filter, NULL);
1923 sk_filter_uncharge(sk, filter);
1924 ret = 0;
1925 }
1926
1927 return ret;
1928 }
1929 EXPORT_SYMBOL_GPL(sk_detach_filter);
1930
1931 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
1932 unsigned int len)
1933 {
1934 struct sock_fprog_kern *fprog;
1935 struct sk_filter *filter;
1936 int ret = 0;
1937
1938 lock_sock(sk);
1939 filter = rcu_dereference_protected(sk->sk_filter,
1940 sock_owned_by_user(sk));
1941 if (!filter)
1942 goto out;
1943
1944 /* We're copying the filter that has been originally attached,
1945 * so no conversion/decode needed anymore.
1946 */
1947 fprog = filter->prog->orig_prog;
1948
1949 ret = fprog->len;
1950 if (!len)
1951 /* User space only enquires number of filter blocks. */
1952 goto out;
1953
1954 ret = -EINVAL;
1955 if (len < fprog->len)
1956 goto out;
1957
1958 ret = -EFAULT;
1959 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
1960 goto out;
1961
1962 /* Instead of bytes, the API requests to return the number
1963 * of filter blocks.
1964 */
1965 ret = fprog->len;
1966 out:
1967 release_sock(sk);
1968 return ret;
1969 }
(deprecated) Btrfs devel tree