?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ipv4: Don't do expensive useless work during inetdev destroy.
[linux-2.6/btrfs-unstable.git] / net / core / filter.c
blob6fc3893a6170517d3e41f3b4243bf76af5f55bcf
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 #include <net/sock_reuseport.h>
54
55 /**
56 * sk_filter - run a packet through a socket filter
57 * @sk: sock associated with &sk_buff
58 * @skb: buffer to filter
59 *
60 * Run the eBPF program and then cut skb->data to correct size returned by
61 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
62 * than pkt_len we keep whole skb->data. This is the socket level
63 * wrapper to BPF_PROG_RUN. It returns 0 if the packet should
64 * be accepted or -EPERM if the packet should be tossed.
65 *
66 */
67 int sk_filter(struct sock *sk, struct sk_buff *skb)
68 {
69 int err;
70 struct sk_filter *filter;
71
72 /*
73 * If the skb was allocated from pfmemalloc reserves, only
74 * allow SOCK_MEMALLOC sockets to use it as this socket is
75 * helping free memory
76 */
77 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
78 return -ENOMEM;
79
80 err = security_sock_rcv_skb(sk, skb);
81 if (err)
82 return err;
83
84 rcu_read_lock();
85 filter = rcu_dereference(sk->sk_filter);
86 if (filter) {
87 unsigned int pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
88
89 err = pkt_len ? pskb_trim(skb, pkt_len) : -EPERM;
90 }
91 rcu_read_unlock();
92
93 return err;
94 }
95 EXPORT_SYMBOL(sk_filter);
96
97 static u64 __skb_get_pay_offset(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
98 {
99 return skb_get_poff((struct sk_buff *)(unsigned long) ctx);
100 }
101
102 static u64 __skb_get_nlattr(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
103 {
104 struct sk_buff *skb = (struct sk_buff *)(unsigned long) ctx;
105 struct nlattr *nla;
106
107 if (skb_is_nonlinear(skb))
108 return 0;
109
110 if (skb->len < sizeof(struct nlattr))
111 return 0;
112
113 if (a > skb->len - sizeof(struct nlattr))
114 return 0;
115
116 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
117 if (nla)
118 return (void *) nla - (void *) skb->data;
119
120 return 0;
121 }
122
123 static u64 __skb_get_nlattr_nest(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
124 {
125 struct sk_buff *skb = (struct sk_buff *)(unsigned long) ctx;
126 struct nlattr *nla;
127
128 if (skb_is_nonlinear(skb))
129 return 0;
130
131 if (skb->len < sizeof(struct nlattr))
132 return 0;
133
134 if (a > skb->len - sizeof(struct nlattr))
135 return 0;
136
137 nla = (struct nlattr *) &skb->data[a];
138 if (nla->nla_len > skb->len - a)
139 return 0;
140
141 nla = nla_find_nested(nla, x);
142 if (nla)
143 return (void *) nla - (void *) skb->data;
144
145 return 0;
146 }
147
148 static u64 __get_raw_cpu_id(u64 ctx, u64 a, u64 x, u64 r4, u64 r5)
149 {
150 return raw_smp_processor_id();
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(bpf_user_rnd_u32);
312 bpf_user_rnd_init_once();
313 break;
314 }
315 break;
316
317 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
318 /* A ^= X */
319 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
320 break;
321
322 default:
323 /* This is just a dummy call to avoid letting the compiler
324 * evict __bpf_call_base() as an optimization. Placed here
325 * where no-one bothers.
326 */
327 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
328 return false;
329 }
330
331 *insnp = insn;
332 return true;
333 }
334
335 /**
336 * bpf_convert_filter - convert filter program
337 * @prog: the user passed filter program
338 * @len: the length of the user passed filter program
339 * @new_prog: buffer where converted program will be stored
340 * @new_len: pointer to store length of converted program
341 *
342 * Remap 'sock_filter' style BPF instruction set to 'sock_filter_ext' style.
343 * Conversion workflow:
344 *
345 * 1) First pass for calculating the new program length:
346 * bpf_convert_filter(old_prog, old_len, NULL, &new_len)
347 *
348 * 2) 2nd pass to remap in two passes: 1st pass finds new
349 * jump offsets, 2nd pass remapping:
350 * new_prog = kmalloc(sizeof(struct bpf_insn) * new_len);
351 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len);
352 */
353 static int bpf_convert_filter(struct sock_filter *prog, int len,
354 struct bpf_insn *new_prog, int *new_len)
355 {
356 int new_flen = 0, pass = 0, target, i;
357 struct bpf_insn *new_insn;
358 struct sock_filter *fp;
359 int *addrs = NULL;
360 u8 bpf_src;
361
362 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
363 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
364
365 if (len <= 0 || len > BPF_MAXINSNS)
366 return -EINVAL;
367
368 if (new_prog) {
369 addrs = kcalloc(len, sizeof(*addrs),
370 GFP_KERNEL | __GFP_NOWARN);
371 if (!addrs)
372 return -ENOMEM;
373 }
374
375 do_pass:
376 new_insn = new_prog;
377 fp = prog;
378
379 /* Classic BPF related prologue emission. */
380 if (new_insn) {
381 /* Classic BPF expects A and X to be reset first. These need
382 * to be guaranteed to be the first two instructions.
383 */
384 *new_insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
385 *new_insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
386
387 /* All programs must keep CTX in callee saved BPF_REG_CTX.
388 * In eBPF case it's done by the compiler, here we need to
389 * do this ourself. Initial CTX is present in BPF_REG_ARG1.
390 */
391 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
392 } else {
393 new_insn += 3;
394 }
395
396 for (i = 0; i < len; fp++, i++) {
397 struct bpf_insn tmp_insns[6] = { };
398 struct bpf_insn *insn = tmp_insns;
399
400 if (addrs)
401 addrs[i] = new_insn - new_prog;
402
403 switch (fp->code) {
404 /* All arithmetic insns and skb loads map as-is. */
405 case BPF_ALU | BPF_ADD | BPF_X:
406 case BPF_ALU | BPF_ADD | BPF_K:
407 case BPF_ALU | BPF_SUB | BPF_X:
408 case BPF_ALU | BPF_SUB | BPF_K:
409 case BPF_ALU | BPF_AND | BPF_X:
410 case BPF_ALU | BPF_AND | BPF_K:
411 case BPF_ALU | BPF_OR | BPF_X:
412 case BPF_ALU | BPF_OR | BPF_K:
413 case BPF_ALU | BPF_LSH | BPF_X:
414 case BPF_ALU | BPF_LSH | BPF_K:
415 case BPF_ALU | BPF_RSH | BPF_X:
416 case BPF_ALU | BPF_RSH | BPF_K:
417 case BPF_ALU | BPF_XOR | BPF_X:
418 case BPF_ALU | BPF_XOR | BPF_K:
419 case BPF_ALU | BPF_MUL | BPF_X:
420 case BPF_ALU | BPF_MUL | BPF_K:
421 case BPF_ALU | BPF_DIV | BPF_X:
422 case BPF_ALU | BPF_DIV | BPF_K:
423 case BPF_ALU | BPF_MOD | BPF_X:
424 case BPF_ALU | BPF_MOD | BPF_K:
425 case BPF_ALU | BPF_NEG:
426 case BPF_LD | BPF_ABS | BPF_W:
427 case BPF_LD | BPF_ABS | BPF_H:
428 case BPF_LD | BPF_ABS | BPF_B:
429 case BPF_LD | BPF_IND | BPF_W:
430 case BPF_LD | BPF_IND | BPF_H:
431 case BPF_LD | BPF_IND | BPF_B:
432 /* Check for overloaded BPF extension and
433 * directly convert it if found, otherwise
434 * just move on with mapping.
435 */
436 if (BPF_CLASS(fp->code) == BPF_LD &&
437 BPF_MODE(fp->code) == BPF_ABS &&
438 convert_bpf_extensions(fp, &insn))
439 break;
440
441 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
442 break;
443
444 /* Jump transformation cannot use BPF block macros
445 * everywhere as offset calculation and target updates
446 * require a bit more work than the rest, i.e. jump
447 * opcodes map as-is, but offsets need adjustment.
448 */
449
450 #define BPF_EMIT_JMP \
451 do { \
452 if (target >= len || target < 0) \
453 goto err; \
454 insn->off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
455 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
456 insn->off -= insn - tmp_insns; \
457 } while (0)
458
459 case BPF_JMP | BPF_JA:
460 target = i + fp->k + 1;
461 insn->code = fp->code;
462 BPF_EMIT_JMP;
463 break;
464
465 case BPF_JMP | BPF_JEQ | BPF_K:
466 case BPF_JMP | BPF_JEQ | BPF_X:
467 case BPF_JMP | BPF_JSET | BPF_K:
468 case BPF_JMP | BPF_JSET | BPF_X:
469 case BPF_JMP | BPF_JGT | BPF_K:
470 case BPF_JMP | BPF_JGT | BPF_X:
471 case BPF_JMP | BPF_JGE | BPF_K:
472 case BPF_JMP | BPF_JGE | BPF_X:
473 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
474 /* BPF immediates are signed, zero extend
475 * immediate into tmp register and use it
476 * in compare insn.
477 */
478 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
479
480 insn->dst_reg = BPF_REG_A;
481 insn->src_reg = BPF_REG_TMP;
482 bpf_src = BPF_X;
483 } else {
484 insn->dst_reg = BPF_REG_A;
485 insn->imm = fp->k;
486 bpf_src = BPF_SRC(fp->code);
487 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
488 }
489
490 /* Common case where 'jump_false' is next insn. */
491 if (fp->jf == 0) {
492 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
493 target = i + fp->jt + 1;
494 BPF_EMIT_JMP;
495 break;
496 }
497
498 /* Convert JEQ into JNE when 'jump_true' is next insn. */
499 if (fp->jt == 0 && BPF_OP(fp->code) == BPF_JEQ) {
500 insn->code = BPF_JMP | BPF_JNE | bpf_src;
501 target = i + fp->jf + 1;
502 BPF_EMIT_JMP;
503 break;
504 }
505
506 /* Other jumps are mapped into two insns: Jxx and JA. */
507 target = i + fp->jt + 1;
508 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
509 BPF_EMIT_JMP;
510 insn++;
511
512 insn->code = BPF_JMP | BPF_JA;
513 target = i + fp->jf + 1;
514 BPF_EMIT_JMP;
515 break;
516
517 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
518 case BPF_LDX | BPF_MSH | BPF_B:
519 /* tmp = A */
520 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_A);
521 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
522 *insn++ = BPF_LD_ABS(BPF_B, fp->k);
523 /* A &= 0xf */
524 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
525 /* A <<= 2 */
526 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
527 /* X = A */
528 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
529 /* A = tmp */
530 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
531 break;
532
533 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
534 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
535 */
536 case BPF_RET | BPF_A:
537 case BPF_RET | BPF_K:
538 if (BPF_RVAL(fp->code) == BPF_K)
539 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
540 0, fp->k);
541 *insn = BPF_EXIT_INSN();
542 break;
543
544 /* Store to stack. */
545 case BPF_ST:
546 case BPF_STX:
547 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
548 BPF_ST ? BPF_REG_A : BPF_REG_X,
549 -(BPF_MEMWORDS - fp->k) * 4);
550 break;
551
552 /* Load from stack. */
553 case BPF_LD | BPF_MEM:
554 case BPF_LDX | BPF_MEM:
555 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
556 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
557 -(BPF_MEMWORDS - fp->k) * 4);
558 break;
559
560 /* A = K or X = K */
561 case BPF_LD | BPF_IMM:
562 case BPF_LDX | BPF_IMM:
563 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
564 BPF_REG_A : BPF_REG_X, fp->k);
565 break;
566
567 /* X = A */
568 case BPF_MISC | BPF_TAX:
569 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
570 break;
571
572 /* A = X */
573 case BPF_MISC | BPF_TXA:
574 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
575 break;
576
577 /* A = skb->len or X = skb->len */
578 case BPF_LD | BPF_W | BPF_LEN:
579 case BPF_LDX | BPF_W | BPF_LEN:
580 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
581 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
582 offsetof(struct sk_buff, len));
583 break;
584
585 /* Access seccomp_data fields. */
586 case BPF_LDX | BPF_ABS | BPF_W:
587 /* A = *(u32 *) (ctx + K) */
588 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
589 break;
590
591 /* Unknown instruction. */
592 default:
593 goto err;
594 }
595
596 insn++;
597 if (new_prog)
598 memcpy(new_insn, tmp_insns,
599 sizeof(*insn) * (insn - tmp_insns));
600 new_insn += insn - tmp_insns;
601 }
602
603 if (!new_prog) {
604 /* Only calculating new length. */
605 *new_len = new_insn - new_prog;
606 return 0;
607 }
608
609 pass++;
610 if (new_flen != new_insn - new_prog) {
611 new_flen = new_insn - new_prog;
612 if (pass > 2)
613 goto err;
614 goto do_pass;
615 }
616
617 kfree(addrs);
618 BUG_ON(*new_len != new_flen);
619 return 0;
620 err:
621 kfree(addrs);
622 return -EINVAL;
623 }
624
625 /* Security:
626 *
627 * As we dont want to clear mem[] array for each packet going through
628 * __bpf_prog_run(), we check that filter loaded by user never try to read
629 * a cell if not previously written, and we check all branches to be sure
630 * a malicious user doesn't try to abuse us.
631 */
632 static int check_load_and_stores(const struct sock_filter *filter, int flen)
633 {
634 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
635 int pc, ret = 0;
636
637 BUILD_BUG_ON(BPF_MEMWORDS > 16);
638
639 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
640 if (!masks)
641 return -ENOMEM;
642
643 memset(masks, 0xff, flen * sizeof(*masks));
644
645 for (pc = 0; pc < flen; pc++) {
646 memvalid &= masks[pc];
647
648 switch (filter[pc].code) {
649 case BPF_ST:
650 case BPF_STX:
651 memvalid |= (1 << filter[pc].k);
652 break;
653 case BPF_LD | BPF_MEM:
654 case BPF_LDX | BPF_MEM:
655 if (!(memvalid & (1 << filter[pc].k))) {
656 ret = -EINVAL;
657 goto error;
658 }
659 break;
660 case BPF_JMP | BPF_JA:
661 /* A jump must set masks on target */
662 masks[pc + 1 + filter[pc].k] &= memvalid;
663 memvalid = ~0;
664 break;
665 case BPF_JMP | BPF_JEQ | BPF_K:
666 case BPF_JMP | BPF_JEQ | BPF_X:
667 case BPF_JMP | BPF_JGE | BPF_K:
668 case BPF_JMP | BPF_JGE | BPF_X:
669 case BPF_JMP | BPF_JGT | BPF_K:
670 case BPF_JMP | BPF_JGT | BPF_X:
671 case BPF_JMP | BPF_JSET | BPF_K:
672 case BPF_JMP | BPF_JSET | BPF_X:
673 /* A jump must set masks on targets */
674 masks[pc + 1 + filter[pc].jt] &= memvalid;
675 masks[pc + 1 + filter[pc].jf] &= memvalid;
676 memvalid = ~0;
677 break;
678 }
679 }
680 error:
681 kfree(masks);
682 return ret;
683 }
684
685 static bool chk_code_allowed(u16 code_to_probe)
686 {
687 static const bool codes[] = {
688 /* 32 bit ALU operations */
689 [BPF_ALU | BPF_ADD | BPF_K] = true,
690 [BPF_ALU | BPF_ADD | BPF_X] = true,
691 [BPF_ALU | BPF_SUB | BPF_K] = true,
692 [BPF_ALU | BPF_SUB | BPF_X] = true,
693 [BPF_ALU | BPF_MUL | BPF_K] = true,
694 [BPF_ALU | BPF_MUL | BPF_X] = true,
695 [BPF_ALU | BPF_DIV | BPF_K] = true,
696 [BPF_ALU | BPF_DIV | BPF_X] = true,
697 [BPF_ALU | BPF_MOD | BPF_K] = true,
698 [BPF_ALU | BPF_MOD | BPF_X] = true,
699 [BPF_ALU | BPF_AND | BPF_K] = true,
700 [BPF_ALU | BPF_AND | BPF_X] = true,
701 [BPF_ALU | BPF_OR | BPF_K] = true,
702 [BPF_ALU | BPF_OR | BPF_X] = true,
703 [BPF_ALU | BPF_XOR | BPF_K] = true,
704 [BPF_ALU | BPF_XOR | BPF_X] = true,
705 [BPF_ALU | BPF_LSH | BPF_K] = true,
706 [BPF_ALU | BPF_LSH | BPF_X] = true,
707 [BPF_ALU | BPF_RSH | BPF_K] = true,
708 [BPF_ALU | BPF_RSH | BPF_X] = true,
709 [BPF_ALU | BPF_NEG] = true,
710 /* Load instructions */
711 [BPF_LD | BPF_W | BPF_ABS] = true,
712 [BPF_LD | BPF_H | BPF_ABS] = true,
713 [BPF_LD | BPF_B | BPF_ABS] = true,
714 [BPF_LD | BPF_W | BPF_LEN] = true,
715 [BPF_LD | BPF_W | BPF_IND] = true,
716 [BPF_LD | BPF_H | BPF_IND] = true,
717 [BPF_LD | BPF_B | BPF_IND] = true,
718 [BPF_LD | BPF_IMM] = true,
719 [BPF_LD | BPF_MEM] = true,
720 [BPF_LDX | BPF_W | BPF_LEN] = true,
721 [BPF_LDX | BPF_B | BPF_MSH] = true,
722 [BPF_LDX | BPF_IMM] = true,
723 [BPF_LDX | BPF_MEM] = true,
724 /* Store instructions */
725 [BPF_ST] = true,
726 [BPF_STX] = true,
727 /* Misc instructions */
728 [BPF_MISC | BPF_TAX] = true,
729 [BPF_MISC | BPF_TXA] = true,
730 /* Return instructions */
731 [BPF_RET | BPF_K] = true,
732 [BPF_RET | BPF_A] = true,
733 /* Jump instructions */
734 [BPF_JMP | BPF_JA] = true,
735 [BPF_JMP | BPF_JEQ | BPF_K] = true,
736 [BPF_JMP | BPF_JEQ | BPF_X] = true,
737 [BPF_JMP | BPF_JGE | BPF_K] = true,
738 [BPF_JMP | BPF_JGE | BPF_X] = true,
739 [BPF_JMP | BPF_JGT | BPF_K] = true,
740 [BPF_JMP | BPF_JGT | BPF_X] = true,
741 [BPF_JMP | BPF_JSET | BPF_K] = true,
742 [BPF_JMP | BPF_JSET | BPF_X] = true,
743 };
744
745 if (code_to_probe >= ARRAY_SIZE(codes))
746 return false;
747
748 return codes[code_to_probe];
749 }
750
751 /**
752 * bpf_check_classic - verify socket filter code
753 * @filter: filter to verify
754 * @flen: length of filter
755 *
756 * Check the user's filter code. If we let some ugly
757 * filter code slip through kaboom! The filter must contain
758 * no references or jumps that are out of range, no illegal
759 * instructions, and must end with a RET instruction.
760 *
761 * All jumps are forward as they are not signed.
762 *
763 * Returns 0 if the rule set is legal or -EINVAL if not.
764 */
765 static int bpf_check_classic(const struct sock_filter *filter,
766 unsigned int flen)
767 {
768 bool anc_found;
769 int pc;
770
771 if (flen == 0 || flen > BPF_MAXINSNS)
772 return -EINVAL;
773
774 /* Check the filter code now */
775 for (pc = 0; pc < flen; pc++) {
776 const struct sock_filter *ftest = &filter[pc];
777
778 /* May we actually operate on this code? */
779 if (!chk_code_allowed(ftest->code))
780 return -EINVAL;
781
782 /* Some instructions need special checks */
783 switch (ftest->code) {
784 case BPF_ALU | BPF_DIV | BPF_K:
785 case BPF_ALU | BPF_MOD | BPF_K:
786 /* Check for division by zero */
787 if (ftest->k == 0)
788 return -EINVAL;
789 break;
790 case BPF_ALU | BPF_LSH | BPF_K:
791 case BPF_ALU | BPF_RSH | BPF_K:
792 if (ftest->k >= 32)
793 return -EINVAL;
794 break;
795 case BPF_LD | BPF_MEM:
796 case BPF_LDX | BPF_MEM:
797 case BPF_ST:
798 case BPF_STX:
799 /* Check for invalid memory addresses */
800 if (ftest->k >= BPF_MEMWORDS)
801 return -EINVAL;
802 break;
803 case BPF_JMP | BPF_JA:
804 /* Note, the large ftest->k might cause loops.
805 * Compare this with conditional jumps below,
806 * where offsets are limited. --ANK (981016)
807 */
808 if (ftest->k >= (unsigned int)(flen - pc - 1))
809 return -EINVAL;
810 break;
811 case BPF_JMP | BPF_JEQ | BPF_K:
812 case BPF_JMP | BPF_JEQ | BPF_X:
813 case BPF_JMP | BPF_JGE | BPF_K:
814 case BPF_JMP | BPF_JGE | BPF_X:
815 case BPF_JMP | BPF_JGT | BPF_K:
816 case BPF_JMP | BPF_JGT | BPF_X:
817 case BPF_JMP | BPF_JSET | BPF_K:
818 case BPF_JMP | BPF_JSET | BPF_X:
819 /* Both conditionals must be safe */
820 if (pc + ftest->jt + 1 >= flen ||
821 pc + ftest->jf + 1 >= flen)
822 return -EINVAL;
823 break;
824 case BPF_LD | BPF_W | BPF_ABS:
825 case BPF_LD | BPF_H | BPF_ABS:
826 case BPF_LD | BPF_B | BPF_ABS:
827 anc_found = false;
828 if (bpf_anc_helper(ftest) & BPF_ANC)
829 anc_found = true;
830 /* Ancillary operation unknown or unsupported */
831 if (anc_found == false && ftest->k >= SKF_AD_OFF)
832 return -EINVAL;
833 }
834 }
835
836 /* Last instruction must be a RET code */
837 switch (filter[flen - 1].code) {
838 case BPF_RET | BPF_K:
839 case BPF_RET | BPF_A:
840 return check_load_and_stores(filter, flen);
841 }
842
843 return -EINVAL;
844 }
845
846 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
847 const struct sock_fprog *fprog)
848 {
849 unsigned int fsize = bpf_classic_proglen(fprog);
850 struct sock_fprog_kern *fkprog;
851
852 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
853 if (!fp->orig_prog)
854 return -ENOMEM;
855
856 fkprog = fp->orig_prog;
857 fkprog->len = fprog->len;
858
859 fkprog->filter = kmemdup(fp->insns, fsize,
860 GFP_KERNEL | __GFP_NOWARN);
861 if (!fkprog->filter) {
862 kfree(fp->orig_prog);
863 return -ENOMEM;
864 }
865
866 return 0;
867 }
868
869 static void bpf_release_orig_filter(struct bpf_prog *fp)
870 {
871 struct sock_fprog_kern *fprog = fp->orig_prog;
872
873 if (fprog) {
874 kfree(fprog->filter);
875 kfree(fprog);
876 }
877 }
878
879 static void __bpf_prog_release(struct bpf_prog *prog)
880 {
881 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
882 bpf_prog_put(prog);
883 } else {
884 bpf_release_orig_filter(prog);
885 bpf_prog_free(prog);
886 }
887 }
888
889 static void __sk_filter_release(struct sk_filter *fp)
890 {
891 __bpf_prog_release(fp->prog);
892 kfree(fp);
893 }
894
895 /**
896 * sk_filter_release_rcu - Release a socket filter by rcu_head
897 * @rcu: rcu_head that contains the sk_filter to free
898 */
899 static void sk_filter_release_rcu(struct rcu_head *rcu)
900 {
901 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
902
903 __sk_filter_release(fp);
904 }
905
906 /**
907 * sk_filter_release - release a socket filter
908 * @fp: filter to remove
909 *
910 * Remove a filter from a socket and release its resources.
911 */
912 static void sk_filter_release(struct sk_filter *fp)
913 {
914 if (atomic_dec_and_test(&fp->refcnt))
915 call_rcu(&fp->rcu, sk_filter_release_rcu);
916 }
917
918 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
919 {
920 u32 filter_size = bpf_prog_size(fp->prog->len);
921
922 atomic_sub(filter_size, &sk->sk_omem_alloc);
923 sk_filter_release(fp);
924 }
925
926 /* try to charge the socket memory if there is space available
927 * return true on success
928 */
929 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
930 {
931 u32 filter_size = bpf_prog_size(fp->prog->len);
932
933 /* same check as in sock_kmalloc() */
934 if (filter_size <= sysctl_optmem_max &&
935 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
936 atomic_inc(&fp->refcnt);
937 atomic_add(filter_size, &sk->sk_omem_alloc);
938 return true;
939 }
940 return false;
941 }
942
943 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
944 {
945 struct sock_filter *old_prog;
946 struct bpf_prog *old_fp;
947 int err, new_len, old_len = fp->len;
948
949 /* We are free to overwrite insns et al right here as it
950 * won't be used at this point in time anymore internally
951 * after the migration to the internal BPF instruction
952 * representation.
953 */
954 BUILD_BUG_ON(sizeof(struct sock_filter) !=
955 sizeof(struct bpf_insn));
956
957 /* Conversion cannot happen on overlapping memory areas,
958 * so we need to keep the user BPF around until the 2nd
959 * pass. At this time, the user BPF is stored in fp->insns.
960 */
961 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
962 GFP_KERNEL | __GFP_NOWARN);
963 if (!old_prog) {
964 err = -ENOMEM;
965 goto out_err;
966 }
967
968 /* 1st pass: calculate the new program length. */
969 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len);
970 if (err)
971 goto out_err_free;
972
973 /* Expand fp for appending the new filter representation. */
974 old_fp = fp;
975 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
976 if (!fp) {
977 /* The old_fp is still around in case we couldn't
978 * allocate new memory, so uncharge on that one.
979 */
980 fp = old_fp;
981 err = -ENOMEM;
982 goto out_err_free;
983 }
984
985 fp->len = new_len;
986
987 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
988 err = bpf_convert_filter(old_prog, old_len, fp->insnsi, &new_len);
989 if (err)
990 /* 2nd bpf_convert_filter() can fail only if it fails
991 * to allocate memory, remapping must succeed. Note,
992 * that at this time old_fp has already been released
993 * by krealloc().
994 */
995 goto out_err_free;
996
997 bpf_prog_select_runtime(fp);
998
999 kfree(old_prog);
1000 return fp;
1001
1002 out_err_free:
1003 kfree(old_prog);
1004 out_err:
1005 __bpf_prog_release(fp);
1006 return ERR_PTR(err);
1007 }
1008
1009 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1010 bpf_aux_classic_check_t trans)
1011 {
1012 int err;
1013
1014 fp->bpf_func = NULL;
1015 fp->jited = 0;
1016
1017 err = bpf_check_classic(fp->insns, fp->len);
1018 if (err) {
1019 __bpf_prog_release(fp);
1020 return ERR_PTR(err);
1021 }
1022
1023 /* There might be additional checks and transformations
1024 * needed on classic filters, f.e. in case of seccomp.
1025 */
1026 if (trans) {
1027 err = trans(fp->insns, fp->len);
1028 if (err) {
1029 __bpf_prog_release(fp);
1030 return ERR_PTR(err);
1031 }
1032 }
1033
1034 /* Probe if we can JIT compile the filter and if so, do
1035 * the compilation of the filter.
1036 */
1037 bpf_jit_compile(fp);
1038
1039 /* JIT compiler couldn't process this filter, so do the
1040 * internal BPF translation for the optimized interpreter.
1041 */
1042 if (!fp->jited)
1043 fp = bpf_migrate_filter(fp);
1044
1045 return fp;
1046 }
1047
1048 /**
1049 * bpf_prog_create - create an unattached filter
1050 * @pfp: the unattached filter that is created
1051 * @fprog: the filter program
1052 *
1053 * Create a filter independent of any socket. We first run some
1054 * sanity checks on it to make sure it does not explode on us later.
1055 * If an error occurs or there is insufficient memory for the filter
1056 * a negative errno code is returned. On success the return is zero.
1057 */
1058 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1059 {
1060 unsigned int fsize = bpf_classic_proglen(fprog);
1061 struct bpf_prog *fp;
1062
1063 /* Make sure new filter is there and in the right amounts. */
1064 if (fprog->filter == NULL)
1065 return -EINVAL;
1066
1067 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1068 if (!fp)
1069 return -ENOMEM;
1070
1071 memcpy(fp->insns, fprog->filter, fsize);
1072
1073 fp->len = fprog->len;
1074 /* Since unattached filters are not copied back to user
1075 * space through sk_get_filter(), we do not need to hold
1076 * a copy here, and can spare us the work.
1077 */
1078 fp->orig_prog = NULL;
1079
1080 /* bpf_prepare_filter() already takes care of freeing
1081 * memory in case something goes wrong.
1082 */
1083 fp = bpf_prepare_filter(fp, NULL);
1084 if (IS_ERR(fp))
1085 return PTR_ERR(fp);
1086
1087 *pfp = fp;
1088 return 0;
1089 }
1090 EXPORT_SYMBOL_GPL(bpf_prog_create);
1091
1092 /**
1093 * bpf_prog_create_from_user - create an unattached filter from user buffer
1094 * @pfp: the unattached filter that is created
1095 * @fprog: the filter program
1096 * @trans: post-classic verifier transformation handler
1097 * @save_orig: save classic BPF program
1098 *
1099 * This function effectively does the same as bpf_prog_create(), only
1100 * that it builds up its insns buffer from user space provided buffer.
1101 * It also allows for passing a bpf_aux_classic_check_t handler.
1102 */
1103 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1104 bpf_aux_classic_check_t trans, bool save_orig)
1105 {
1106 unsigned int fsize = bpf_classic_proglen(fprog);
1107 struct bpf_prog *fp;
1108 int err;
1109
1110 /* Make sure new filter is there and in the right amounts. */
1111 if (fprog->filter == NULL)
1112 return -EINVAL;
1113
1114 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1115 if (!fp)
1116 return -ENOMEM;
1117
1118 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1119 __bpf_prog_free(fp);
1120 return -EFAULT;
1121 }
1122
1123 fp->len = fprog->len;
1124 fp->orig_prog = NULL;
1125
1126 if (save_orig) {
1127 err = bpf_prog_store_orig_filter(fp, fprog);
1128 if (err) {
1129 __bpf_prog_free(fp);
1130 return -ENOMEM;
1131 }
1132 }
1133
1134 /* bpf_prepare_filter() already takes care of freeing
1135 * memory in case something goes wrong.
1136 */
1137 fp = bpf_prepare_filter(fp, trans);
1138 if (IS_ERR(fp))
1139 return PTR_ERR(fp);
1140
1141 *pfp = fp;
1142 return 0;
1143 }
1144 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1145
1146 void bpf_prog_destroy(struct bpf_prog *fp)
1147 {
1148 __bpf_prog_release(fp);
1149 }
1150 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1151
1152 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1153 {
1154 struct sk_filter *fp, *old_fp;
1155
1156 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1157 if (!fp)
1158 return -ENOMEM;
1159
1160 fp->prog = prog;
1161 atomic_set(&fp->refcnt, 0);
1162
1163 if (!sk_filter_charge(sk, fp)) {
1164 kfree(fp);
1165 return -ENOMEM;
1166 }
1167
1168 old_fp = rcu_dereference_protected(sk->sk_filter,
1169 sock_owned_by_user(sk));
1170 rcu_assign_pointer(sk->sk_filter, fp);
1171
1172 if (old_fp)
1173 sk_filter_uncharge(sk, old_fp);
1174
1175 return 0;
1176 }
1177
1178 static int __reuseport_attach_prog(struct bpf_prog *prog, struct sock *sk)
1179 {
1180 struct bpf_prog *old_prog;
1181 int err;
1182
1183 if (bpf_prog_size(prog->len) > sysctl_optmem_max)
1184 return -ENOMEM;
1185
1186 if (sk_unhashed(sk) && sk->sk_reuseport) {
1187 err = reuseport_alloc(sk);
1188 if (err)
1189 return err;
1190 } else if (!rcu_access_pointer(sk->sk_reuseport_cb)) {
1191 /* The socket wasn't bound with SO_REUSEPORT */
1192 return -EINVAL;
1193 }
1194
1195 old_prog = reuseport_attach_prog(sk, prog);
1196 if (old_prog)
1197 bpf_prog_destroy(old_prog);
1198
1199 return 0;
1200 }
1201
1202 static
1203 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1204 {
1205 unsigned int fsize = bpf_classic_proglen(fprog);
1206 unsigned int bpf_fsize = bpf_prog_size(fprog->len);
1207 struct bpf_prog *prog;
1208 int err;
1209
1210 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1211 return ERR_PTR(-EPERM);
1212
1213 /* Make sure new filter is there and in the right amounts. */
1214 if (fprog->filter == NULL)
1215 return ERR_PTR(-EINVAL);
1216
1217 prog = bpf_prog_alloc(bpf_fsize, 0);
1218 if (!prog)
1219 return ERR_PTR(-ENOMEM);
1220
1221 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1222 __bpf_prog_free(prog);
1223 return ERR_PTR(-EFAULT);
1224 }
1225
1226 prog->len = fprog->len;
1227
1228 err = bpf_prog_store_orig_filter(prog, fprog);
1229 if (err) {
1230 __bpf_prog_free(prog);
1231 return ERR_PTR(-ENOMEM);
1232 }
1233
1234 /* bpf_prepare_filter() already takes care of freeing
1235 * memory in case something goes wrong.
1236 */
1237 return bpf_prepare_filter(prog, NULL);
1238 }
1239
1240 /**
1241 * sk_attach_filter - attach a socket filter
1242 * @fprog: the filter program
1243 * @sk: the socket to use
1244 *
1245 * Attach the user's filter code. We first run some sanity checks on
1246 * it to make sure it does not explode on us later. If an error
1247 * occurs or there is insufficient memory for the filter a negative
1248 * errno code is returned. On success the return is zero.
1249 */
1250 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1251 {
1252 struct bpf_prog *prog = __get_filter(fprog, sk);
1253 int err;
1254
1255 if (IS_ERR(prog))
1256 return PTR_ERR(prog);
1257
1258 err = __sk_attach_prog(prog, sk);
1259 if (err < 0) {
1260 __bpf_prog_release(prog);
1261 return err;
1262 }
1263
1264 return 0;
1265 }
1266 EXPORT_SYMBOL_GPL(sk_attach_filter);
1267
1268 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1269 {
1270 struct bpf_prog *prog = __get_filter(fprog, sk);
1271 int err;
1272
1273 if (IS_ERR(prog))
1274 return PTR_ERR(prog);
1275
1276 err = __reuseport_attach_prog(prog, sk);
1277 if (err < 0) {
1278 __bpf_prog_release(prog);
1279 return err;
1280 }
1281
1282 return 0;
1283 }
1284
1285 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1286 {
1287 struct bpf_prog *prog;
1288
1289 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1290 return ERR_PTR(-EPERM);
1291
1292 prog = bpf_prog_get(ufd);
1293 if (IS_ERR(prog))
1294 return prog;
1295
1296 if (prog->type != BPF_PROG_TYPE_SOCKET_FILTER) {
1297 bpf_prog_put(prog);
1298 return ERR_PTR(-EINVAL);
1299 }
1300
1301 return prog;
1302 }
1303
1304 int sk_attach_bpf(u32 ufd, struct sock *sk)
1305 {
1306 struct bpf_prog *prog = __get_bpf(ufd, sk);
1307 int err;
1308
1309 if (IS_ERR(prog))
1310 return PTR_ERR(prog);
1311
1312 err = __sk_attach_prog(prog, sk);
1313 if (err < 0) {
1314 bpf_prog_put(prog);
1315 return err;
1316 }
1317
1318 return 0;
1319 }
1320
1321 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1322 {
1323 struct bpf_prog *prog = __get_bpf(ufd, sk);
1324 int err;
1325
1326 if (IS_ERR(prog))
1327 return PTR_ERR(prog);
1328
1329 err = __reuseport_attach_prog(prog, sk);
1330 if (err < 0) {
1331 bpf_prog_put(prog);
1332 return err;
1333 }
1334
1335 return 0;
1336 }
1337
1338 struct bpf_scratchpad {
1339 union {
1340 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1341 u8 buff[MAX_BPF_STACK];
1342 };
1343 };
1344
1345 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1346
1347 static u64 bpf_skb_store_bytes(u64 r1, u64 r2, u64 r3, u64 r4, u64 flags)
1348 {
1349 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1350 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1351 int offset = (int) r2;
1352 void *from = (void *) (long) r3;
1353 unsigned int len = (unsigned int) r4;
1354 void *ptr;
1355
1356 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1357 return -EINVAL;
1358
1359 /* bpf verifier guarantees that:
1360 * 'from' pointer points to bpf program stack
1361 * 'len' bytes of it were initialized
1362 * 'len' > 0
1363 * 'skb' is a valid pointer to 'struct sk_buff'
1364 *
1365 * so check for invalid 'offset' and too large 'len'
1366 */
1367 if (unlikely((u32) offset > 0xffff || len > sizeof(sp->buff)))
1368 return -EFAULT;
1369 if (unlikely(skb_try_make_writable(skb, offset + len)))
1370 return -EFAULT;
1371
1372 ptr = skb_header_pointer(skb, offset, len, sp->buff);
1373 if (unlikely(!ptr))
1374 return -EFAULT;
1375
1376 if (flags & BPF_F_RECOMPUTE_CSUM)
1377 skb_postpull_rcsum(skb, ptr, len);
1378
1379 memcpy(ptr, from, len);
1380
1381 if (ptr == sp->buff)
1382 /* skb_store_bits cannot return -EFAULT here */
1383 skb_store_bits(skb, offset, ptr, len);
1384
1385 if (flags & BPF_F_RECOMPUTE_CSUM)
1386 skb_postpush_rcsum(skb, ptr, len);
1387 if (flags & BPF_F_INVALIDATE_HASH)
1388 skb_clear_hash(skb);
1389
1390 return 0;
1391 }
1392
1393 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1394 .func = bpf_skb_store_bytes,
1395 .gpl_only = false,
1396 .ret_type = RET_INTEGER,
1397 .arg1_type = ARG_PTR_TO_CTX,
1398 .arg2_type = ARG_ANYTHING,
1399 .arg3_type = ARG_PTR_TO_STACK,
1400 .arg4_type = ARG_CONST_STACK_SIZE,
1401 .arg5_type = ARG_ANYTHING,
1402 };
1403
1404 static u64 bpf_skb_load_bytes(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1405 {
1406 const struct sk_buff *skb = (const struct sk_buff *)(unsigned long) r1;
1407 int offset = (int) r2;
1408 void *to = (void *)(unsigned long) r3;
1409 unsigned int len = (unsigned int) r4;
1410 void *ptr;
1411
1412 if (unlikely((u32) offset > 0xffff || len > MAX_BPF_STACK))
1413 return -EFAULT;
1414
1415 ptr = skb_header_pointer(skb, offset, len, to);
1416 if (unlikely(!ptr))
1417 return -EFAULT;
1418 if (ptr != to)
1419 memcpy(to, ptr, len);
1420
1421 return 0;
1422 }
1423
1424 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1425 .func = bpf_skb_load_bytes,
1426 .gpl_only = false,
1427 .ret_type = RET_INTEGER,
1428 .arg1_type = ARG_PTR_TO_CTX,
1429 .arg2_type = ARG_ANYTHING,
1430 .arg3_type = ARG_PTR_TO_STACK,
1431 .arg4_type = ARG_CONST_STACK_SIZE,
1432 };
1433
1434 static u64 bpf_l3_csum_replace(u64 r1, u64 r2, u64 from, u64 to, u64 flags)
1435 {
1436 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1437 int offset = (int) r2;
1438 __sum16 sum, *ptr;
1439
1440 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1441 return -EINVAL;
1442 if (unlikely((u32) offset > 0xffff))
1443 return -EFAULT;
1444 if (unlikely(skb_try_make_writable(skb, offset + sizeof(sum))))
1445 return -EFAULT;
1446
1447 ptr = skb_header_pointer(skb, offset, sizeof(sum), &sum);
1448 if (unlikely(!ptr))
1449 return -EFAULT;
1450
1451 switch (flags & BPF_F_HDR_FIELD_MASK) {
1452 case 0:
1453 if (unlikely(from != 0))
1454 return -EINVAL;
1455
1456 csum_replace_by_diff(ptr, to);
1457 break;
1458 case 2:
1459 csum_replace2(ptr, from, to);
1460 break;
1461 case 4:
1462 csum_replace4(ptr, from, to);
1463 break;
1464 default:
1465 return -EINVAL;
1466 }
1467
1468 if (ptr == &sum)
1469 /* skb_store_bits guaranteed to not return -EFAULT here */
1470 skb_store_bits(skb, offset, ptr, sizeof(sum));
1471
1472 return 0;
1473 }
1474
1475 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1476 .func = bpf_l3_csum_replace,
1477 .gpl_only = false,
1478 .ret_type = RET_INTEGER,
1479 .arg1_type = ARG_PTR_TO_CTX,
1480 .arg2_type = ARG_ANYTHING,
1481 .arg3_type = ARG_ANYTHING,
1482 .arg4_type = ARG_ANYTHING,
1483 .arg5_type = ARG_ANYTHING,
1484 };
1485
1486 static u64 bpf_l4_csum_replace(u64 r1, u64 r2, u64 from, u64 to, u64 flags)
1487 {
1488 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1489 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1490 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1491 int offset = (int) r2;
1492 __sum16 sum, *ptr;
1493
1494 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_PSEUDO_HDR |
1495 BPF_F_HDR_FIELD_MASK)))
1496 return -EINVAL;
1497 if (unlikely((u32) offset > 0xffff))
1498 return -EFAULT;
1499 if (unlikely(skb_try_make_writable(skb, offset + sizeof(sum))))
1500 return -EFAULT;
1501
1502 ptr = skb_header_pointer(skb, offset, sizeof(sum), &sum);
1503 if (unlikely(!ptr))
1504 return -EFAULT;
1505 if (is_mmzero && !*ptr)
1506 return 0;
1507
1508 switch (flags & BPF_F_HDR_FIELD_MASK) {
1509 case 0:
1510 if (unlikely(from != 0))
1511 return -EINVAL;
1512
1513 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1514 break;
1515 case 2:
1516 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1517 break;
1518 case 4:
1519 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1520 break;
1521 default:
1522 return -EINVAL;
1523 }
1524
1525 if (is_mmzero && !*ptr)
1526 *ptr = CSUM_MANGLED_0;
1527 if (ptr == &sum)
1528 /* skb_store_bits guaranteed to not return -EFAULT here */
1529 skb_store_bits(skb, offset, ptr, sizeof(sum));
1530
1531 return 0;
1532 }
1533
1534 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1535 .func = bpf_l4_csum_replace,
1536 .gpl_only = false,
1537 .ret_type = RET_INTEGER,
1538 .arg1_type = ARG_PTR_TO_CTX,
1539 .arg2_type = ARG_ANYTHING,
1540 .arg3_type = ARG_ANYTHING,
1541 .arg4_type = ARG_ANYTHING,
1542 .arg5_type = ARG_ANYTHING,
1543 };
1544
1545 static u64 bpf_csum_diff(u64 r1, u64 from_size, u64 r3, u64 to_size, u64 seed)
1546 {
1547 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1548 u64 diff_size = from_size + to_size;
1549 __be32 *from = (__be32 *) (long) r1;
1550 __be32 *to = (__be32 *) (long) r3;
1551 int i, j = 0;
1552
1553 /* This is quite flexible, some examples:
1554 *
1555 * from_size == 0, to_size > 0, seed := csum --> pushing data
1556 * from_size > 0, to_size == 0, seed := csum --> pulling data
1557 * from_size > 0, to_size > 0, seed := 0 --> diffing data
1558 *
1559 * Even for diffing, from_size and to_size don't need to be equal.
1560 */
1561 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
1562 diff_size > sizeof(sp->diff)))
1563 return -EINVAL;
1564
1565 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
1566 sp->diff[j] = ~from[i];
1567 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
1568 sp->diff[j] = to[i];
1569
1570 return csum_partial(sp->diff, diff_size, seed);
1571 }
1572
1573 static const struct bpf_func_proto bpf_csum_diff_proto = {
1574 .func = bpf_csum_diff,
1575 .gpl_only = false,
1576 .ret_type = RET_INTEGER,
1577 .arg1_type = ARG_PTR_TO_STACK,
1578 .arg2_type = ARG_CONST_STACK_SIZE_OR_ZERO,
1579 .arg3_type = ARG_PTR_TO_STACK,
1580 .arg4_type = ARG_CONST_STACK_SIZE_OR_ZERO,
1581 .arg5_type = ARG_ANYTHING,
1582 };
1583
1584 static u64 bpf_clone_redirect(u64 r1, u64 ifindex, u64 flags, u64 r4, u64 r5)
1585 {
1586 struct sk_buff *skb = (struct sk_buff *) (long) r1, *skb2;
1587 struct net_device *dev;
1588
1589 if (unlikely(flags & ~(BPF_F_INGRESS)))
1590 return -EINVAL;
1591
1592 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
1593 if (unlikely(!dev))
1594 return -EINVAL;
1595
1596 skb2 = skb_clone(skb, GFP_ATOMIC);
1597 if (unlikely(!skb2))
1598 return -ENOMEM;
1599
1600 if (flags & BPF_F_INGRESS) {
1601 if (skb_at_tc_ingress(skb2))
1602 skb_postpush_rcsum(skb2, skb_mac_header(skb2),
1603 skb2->mac_len);
1604 return dev_forward_skb(dev, skb2);
1605 }
1606
1607 skb2->dev = dev;
1608 return dev_queue_xmit(skb2);
1609 }
1610
1611 static const struct bpf_func_proto bpf_clone_redirect_proto = {
1612 .func = bpf_clone_redirect,
1613 .gpl_only = false,
1614 .ret_type = RET_INTEGER,
1615 .arg1_type = ARG_PTR_TO_CTX,
1616 .arg2_type = ARG_ANYTHING,
1617 .arg3_type = ARG_ANYTHING,
1618 };
1619
1620 struct redirect_info {
1621 u32 ifindex;
1622 u32 flags;
1623 };
1624
1625 static DEFINE_PER_CPU(struct redirect_info, redirect_info);
1626
1627 static u64 bpf_redirect(u64 ifindex, u64 flags, u64 r3, u64 r4, u64 r5)
1628 {
1629 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
1630
1631 if (unlikely(flags & ~(BPF_F_INGRESS)))
1632 return TC_ACT_SHOT;
1633
1634 ri->ifindex = ifindex;
1635 ri->flags = flags;
1636
1637 return TC_ACT_REDIRECT;
1638 }
1639
1640 int skb_do_redirect(struct sk_buff *skb)
1641 {
1642 struct redirect_info *ri = this_cpu_ptr(&redirect_info);
1643 struct net_device *dev;
1644
1645 dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->ifindex);
1646 ri->ifindex = 0;
1647 if (unlikely(!dev)) {
1648 kfree_skb(skb);
1649 return -EINVAL;
1650 }
1651
1652 if (ri->flags & BPF_F_INGRESS) {
1653 if (skb_at_tc_ingress(skb))
1654 skb_postpush_rcsum(skb, skb_mac_header(skb),
1655 skb->mac_len);
1656 return dev_forward_skb(dev, skb);
1657 }
1658
1659 skb->dev = dev;
1660 return dev_queue_xmit(skb);
1661 }
1662
1663 static const struct bpf_func_proto bpf_redirect_proto = {
1664 .func = bpf_redirect,
1665 .gpl_only = false,
1666 .ret_type = RET_INTEGER,
1667 .arg1_type = ARG_ANYTHING,
1668 .arg2_type = ARG_ANYTHING,
1669 };
1670
1671 static u64 bpf_get_cgroup_classid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1672 {
1673 return task_get_classid((struct sk_buff *) (unsigned long) r1);
1674 }
1675
1676 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
1677 .func = bpf_get_cgroup_classid,
1678 .gpl_only = false,
1679 .ret_type = RET_INTEGER,
1680 .arg1_type = ARG_PTR_TO_CTX,
1681 };
1682
1683 static u64 bpf_get_route_realm(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1684 {
1685 #ifdef CONFIG_IP_ROUTE_CLASSID
1686 const struct dst_entry *dst;
1687
1688 dst = skb_dst((struct sk_buff *) (unsigned long) r1);
1689 if (dst)
1690 return dst->tclassid;
1691 #endif
1692 return 0;
1693 }
1694
1695 static const struct bpf_func_proto bpf_get_route_realm_proto = {
1696 .func = bpf_get_route_realm,
1697 .gpl_only = false,
1698 .ret_type = RET_INTEGER,
1699 .arg1_type = ARG_PTR_TO_CTX,
1700 };
1701
1702 static u64 bpf_skb_vlan_push(u64 r1, u64 r2, u64 vlan_tci, u64 r4, u64 r5)
1703 {
1704 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1705 __be16 vlan_proto = (__force __be16) r2;
1706
1707 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
1708 vlan_proto != htons(ETH_P_8021AD)))
1709 vlan_proto = htons(ETH_P_8021Q);
1710
1711 return skb_vlan_push(skb, vlan_proto, vlan_tci);
1712 }
1713
1714 const struct bpf_func_proto bpf_skb_vlan_push_proto = {
1715 .func = bpf_skb_vlan_push,
1716 .gpl_only = false,
1717 .ret_type = RET_INTEGER,
1718 .arg1_type = ARG_PTR_TO_CTX,
1719 .arg2_type = ARG_ANYTHING,
1720 .arg3_type = ARG_ANYTHING,
1721 };
1722 EXPORT_SYMBOL_GPL(bpf_skb_vlan_push_proto);
1723
1724 static u64 bpf_skb_vlan_pop(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
1725 {
1726 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1727
1728 return skb_vlan_pop(skb);
1729 }
1730
1731 const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
1732 .func = bpf_skb_vlan_pop,
1733 .gpl_only = false,
1734 .ret_type = RET_INTEGER,
1735 .arg1_type = ARG_PTR_TO_CTX,
1736 };
1737 EXPORT_SYMBOL_GPL(bpf_skb_vlan_pop_proto);
1738
1739 bool bpf_helper_changes_skb_data(void *func)
1740 {
1741 if (func == bpf_skb_vlan_push)
1742 return true;
1743 if (func == bpf_skb_vlan_pop)
1744 return true;
1745 if (func == bpf_skb_store_bytes)
1746 return true;
1747 if (func == bpf_l3_csum_replace)
1748 return true;
1749 if (func == bpf_l4_csum_replace)
1750 return true;
1751
1752 return false;
1753 }
1754
1755 static unsigned short bpf_tunnel_key_af(u64 flags)
1756 {
1757 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
1758 }
1759
1760 static u64 bpf_skb_get_tunnel_key(u64 r1, u64 r2, u64 size, u64 flags, u64 r5)
1761 {
1762 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1763 struct bpf_tunnel_key *to = (struct bpf_tunnel_key *) (long) r2;
1764 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
1765 u8 compat[sizeof(struct bpf_tunnel_key)];
1766
1767 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6))))
1768 return -EINVAL;
1769 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags))
1770 return -EPROTO;
1771 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
1772 switch (size) {
1773 case offsetof(struct bpf_tunnel_key, tunnel_label):
1774 goto set_compat;
1775 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
1776 /* Fixup deprecated structure layouts here, so we have
1777 * a common path later on.
1778 */
1779 if (ip_tunnel_info_af(info) != AF_INET)
1780 return -EINVAL;
1781 set_compat:
1782 to = (struct bpf_tunnel_key *)compat;
1783 break;
1784 default:
1785 return -EINVAL;
1786 }
1787 }
1788
1789 to->tunnel_id = be64_to_cpu(info->key.tun_id);
1790 to->tunnel_tos = info->key.tos;
1791 to->tunnel_ttl = info->key.ttl;
1792
1793 if (flags & BPF_F_TUNINFO_IPV6) {
1794 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
1795 sizeof(to->remote_ipv6));
1796 to->tunnel_label = be32_to_cpu(info->key.label);
1797 } else {
1798 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
1799 }
1800
1801 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
1802 memcpy((void *)(long) r2, to, size);
1803
1804 return 0;
1805 }
1806
1807 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
1808 .func = bpf_skb_get_tunnel_key,
1809 .gpl_only = false,
1810 .ret_type = RET_INTEGER,
1811 .arg1_type = ARG_PTR_TO_CTX,
1812 .arg2_type = ARG_PTR_TO_STACK,
1813 .arg3_type = ARG_CONST_STACK_SIZE,
1814 .arg4_type = ARG_ANYTHING,
1815 };
1816
1817 static u64 bpf_skb_get_tunnel_opt(u64 r1, u64 r2, u64 size, u64 r4, u64 r5)
1818 {
1819 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1820 u8 *to = (u8 *) (long) r2;
1821 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
1822
1823 if (unlikely(!info ||
1824 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT)))
1825 return -ENOENT;
1826 if (unlikely(size < info->options_len))
1827 return -ENOMEM;
1828
1829 ip_tunnel_info_opts_get(to, info);
1830
1831 return info->options_len;
1832 }
1833
1834 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
1835 .func = bpf_skb_get_tunnel_opt,
1836 .gpl_only = false,
1837 .ret_type = RET_INTEGER,
1838 .arg1_type = ARG_PTR_TO_CTX,
1839 .arg2_type = ARG_PTR_TO_STACK,
1840 .arg3_type = ARG_CONST_STACK_SIZE,
1841 };
1842
1843 static struct metadata_dst __percpu *md_dst;
1844
1845 static u64 bpf_skb_set_tunnel_key(u64 r1, u64 r2, u64 size, u64 flags, u64 r5)
1846 {
1847 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1848 struct bpf_tunnel_key *from = (struct bpf_tunnel_key *) (long) r2;
1849 struct metadata_dst *md = this_cpu_ptr(md_dst);
1850 u8 compat[sizeof(struct bpf_tunnel_key)];
1851 struct ip_tunnel_info *info;
1852
1853 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
1854 BPF_F_DONT_FRAGMENT)))
1855 return -EINVAL;
1856 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
1857 switch (size) {
1858 case offsetof(struct bpf_tunnel_key, tunnel_label):
1859 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
1860 /* Fixup deprecated structure layouts here, so we have
1861 * a common path later on.
1862 */
1863 memcpy(compat, from, size);
1864 memset(compat + size, 0, sizeof(compat) - size);
1865 from = (struct bpf_tunnel_key *)compat;
1866 break;
1867 default:
1868 return -EINVAL;
1869 }
1870 }
1871 if (unlikely(!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label))
1872 return -EINVAL;
1873
1874 skb_dst_drop(skb);
1875 dst_hold((struct dst_entry *) md);
1876 skb_dst_set(skb, (struct dst_entry *) md);
1877
1878 info = &md->u.tun_info;
1879 info->mode = IP_TUNNEL_INFO_TX;
1880
1881 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
1882 if (flags & BPF_F_DONT_FRAGMENT)
1883 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
1884
1885 info->key.tun_id = cpu_to_be64(from->tunnel_id);
1886 info->key.tos = from->tunnel_tos;
1887 info->key.ttl = from->tunnel_ttl;
1888
1889 if (flags & BPF_F_TUNINFO_IPV6) {
1890 info->mode |= IP_TUNNEL_INFO_IPV6;
1891 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
1892 sizeof(from->remote_ipv6));
1893 info->key.label = cpu_to_be32(from->tunnel_label) &
1894 IPV6_FLOWLABEL_MASK;
1895 } else {
1896 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
1897 if (flags & BPF_F_ZERO_CSUM_TX)
1898 info->key.tun_flags &= ~TUNNEL_CSUM;
1899 }
1900
1901 return 0;
1902 }
1903
1904 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
1905 .func = bpf_skb_set_tunnel_key,
1906 .gpl_only = false,
1907 .ret_type = RET_INTEGER,
1908 .arg1_type = ARG_PTR_TO_CTX,
1909 .arg2_type = ARG_PTR_TO_STACK,
1910 .arg3_type = ARG_CONST_STACK_SIZE,
1911 .arg4_type = ARG_ANYTHING,
1912 };
1913
1914 #define BPF_TUNLEN_MAX 255
1915
1916 static u64 bpf_skb_set_tunnel_opt(u64 r1, u64 r2, u64 size, u64 r4, u64 r5)
1917 {
1918 struct sk_buff *skb = (struct sk_buff *) (long) r1;
1919 u8 *from = (u8 *) (long) r2;
1920 struct ip_tunnel_info *info = skb_tunnel_info(skb);
1921 const struct metadata_dst *md = this_cpu_ptr(md_dst);
1922
1923 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
1924 return -EINVAL;
1925 if (unlikely(size > BPF_TUNLEN_MAX))
1926 return -ENOMEM;
1927
1928 ip_tunnel_info_opts_set(info, from, size);
1929
1930 return 0;
1931 }
1932
1933 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
1934 .func = bpf_skb_set_tunnel_opt,
1935 .gpl_only = false,
1936 .ret_type = RET_INTEGER,
1937 .arg1_type = ARG_PTR_TO_CTX,
1938 .arg2_type = ARG_PTR_TO_STACK,
1939 .arg3_type = ARG_CONST_STACK_SIZE,
1940 };
1941
1942 static const struct bpf_func_proto *
1943 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
1944 {
1945 if (!md_dst) {
1946 BUILD_BUG_ON(FIELD_SIZEOF(struct ip_tunnel_info,
1947 options_len) != 1);
1948
1949 /* Race is not possible, since it's called from verifier
1950 * that is holding verifier mutex.
1951 */
1952 md_dst = metadata_dst_alloc_percpu(BPF_TUNLEN_MAX,
1953 GFP_KERNEL);
1954 if (!md_dst)
1955 return NULL;
1956 }
1957
1958 switch (which) {
1959 case BPF_FUNC_skb_set_tunnel_key:
1960 return &bpf_skb_set_tunnel_key_proto;
1961 case BPF_FUNC_skb_set_tunnel_opt:
1962 return &bpf_skb_set_tunnel_opt_proto;
1963 default:
1964 return NULL;
1965 }
1966 }
1967
1968 static const struct bpf_func_proto *
1969 sk_filter_func_proto(enum bpf_func_id func_id)
1970 {
1971 switch (func_id) {
1972 case BPF_FUNC_map_lookup_elem:
1973 return &bpf_map_lookup_elem_proto;
1974 case BPF_FUNC_map_update_elem:
1975 return &bpf_map_update_elem_proto;
1976 case BPF_FUNC_map_delete_elem:
1977 return &bpf_map_delete_elem_proto;
1978 case BPF_FUNC_get_prandom_u32:
1979 return &bpf_get_prandom_u32_proto;
1980 case BPF_FUNC_get_smp_processor_id:
1981 return &bpf_get_smp_processor_id_proto;
1982 case BPF_FUNC_tail_call:
1983 return &bpf_tail_call_proto;
1984 case BPF_FUNC_ktime_get_ns:
1985 return &bpf_ktime_get_ns_proto;
1986 case BPF_FUNC_trace_printk:
1987 if (capable(CAP_SYS_ADMIN))
1988 return bpf_get_trace_printk_proto();
1989 default:
1990 return NULL;
1991 }
1992 }
1993
1994 static const struct bpf_func_proto *
1995 tc_cls_act_func_proto(enum bpf_func_id func_id)
1996 {
1997 switch (func_id) {
1998 case BPF_FUNC_skb_store_bytes:
1999 return &bpf_skb_store_bytes_proto;
2000 case BPF_FUNC_skb_load_bytes:
2001 return &bpf_skb_load_bytes_proto;
2002 case BPF_FUNC_csum_diff:
2003 return &bpf_csum_diff_proto;
2004 case BPF_FUNC_l3_csum_replace:
2005 return &bpf_l3_csum_replace_proto;
2006 case BPF_FUNC_l4_csum_replace:
2007 return &bpf_l4_csum_replace_proto;
2008 case BPF_FUNC_clone_redirect:
2009 return &bpf_clone_redirect_proto;
2010 case BPF_FUNC_get_cgroup_classid:
2011 return &bpf_get_cgroup_classid_proto;
2012 case BPF_FUNC_skb_vlan_push:
2013 return &bpf_skb_vlan_push_proto;
2014 case BPF_FUNC_skb_vlan_pop:
2015 return &bpf_skb_vlan_pop_proto;
2016 case BPF_FUNC_skb_get_tunnel_key:
2017 return &bpf_skb_get_tunnel_key_proto;
2018 case BPF_FUNC_skb_set_tunnel_key:
2019 return bpf_get_skb_set_tunnel_proto(func_id);
2020 case BPF_FUNC_skb_get_tunnel_opt:
2021 return &bpf_skb_get_tunnel_opt_proto;
2022 case BPF_FUNC_skb_set_tunnel_opt:
2023 return bpf_get_skb_set_tunnel_proto(func_id);
2024 case BPF_FUNC_redirect:
2025 return &bpf_redirect_proto;
2026 case BPF_FUNC_get_route_realm:
2027 return &bpf_get_route_realm_proto;
2028 default:
2029 return sk_filter_func_proto(func_id);
2030 }
2031 }
2032
2033 static bool __is_valid_access(int off, int size, enum bpf_access_type type)
2034 {
2035 /* check bounds */
2036 if (off < 0 || off >= sizeof(struct __sk_buff))
2037 return false;
2038
2039 /* disallow misaligned access */
2040 if (off % size != 0)
2041 return false;
2042
2043 /* all __sk_buff fields are __u32 */
2044 if (size != 4)
2045 return false;
2046
2047 return true;
2048 }
2049
2050 static bool sk_filter_is_valid_access(int off, int size,
2051 enum bpf_access_type type)
2052 {
2053 if (off == offsetof(struct __sk_buff, tc_classid))
2054 return false;
2055
2056 if (type == BPF_WRITE) {
2057 switch (off) {
2058 case offsetof(struct __sk_buff, cb[0]) ...
2059 offsetof(struct __sk_buff, cb[4]):
2060 break;
2061 default:
2062 return false;
2063 }
2064 }
2065
2066 return __is_valid_access(off, size, type);
2067 }
2068
2069 static bool tc_cls_act_is_valid_access(int off, int size,
2070 enum bpf_access_type type)
2071 {
2072 if (off == offsetof(struct __sk_buff, tc_classid))
2073 return type == BPF_WRITE ? true : false;
2074
2075 if (type == BPF_WRITE) {
2076 switch (off) {
2077 case offsetof(struct __sk_buff, mark):
2078 case offsetof(struct __sk_buff, tc_index):
2079 case offsetof(struct __sk_buff, priority):
2080 case offsetof(struct __sk_buff, cb[0]) ...
2081 offsetof(struct __sk_buff, cb[4]):
2082 break;
2083 default:
2084 return false;
2085 }
2086 }
2087 return __is_valid_access(off, size, type);
2088 }
2089
2090 static u32 bpf_net_convert_ctx_access(enum bpf_access_type type, int dst_reg,
2091 int src_reg, int ctx_off,
2092 struct bpf_insn *insn_buf,
2093 struct bpf_prog *prog)
2094 {
2095 struct bpf_insn *insn = insn_buf;
2096
2097 switch (ctx_off) {
2098 case offsetof(struct __sk_buff, len):
2099 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, len) != 4);
2100
2101 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
2102 offsetof(struct sk_buff, len));
2103 break;
2104
2105 case offsetof(struct __sk_buff, protocol):
2106 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
2107
2108 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
2109 offsetof(struct sk_buff, protocol));
2110 break;
2111
2112 case offsetof(struct __sk_buff, vlan_proto):
2113 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2);
2114
2115 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
2116 offsetof(struct sk_buff, vlan_proto));
2117 break;
2118
2119 case offsetof(struct __sk_buff, priority):
2120 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, priority) != 4);
2121
2122 if (type == BPF_WRITE)
2123 *insn++ = BPF_STX_MEM(BPF_W, dst_reg, src_reg,
2124 offsetof(struct sk_buff, priority));
2125 else
2126 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
2127 offsetof(struct sk_buff, priority));
2128 break;
2129
2130 case offsetof(struct __sk_buff, ingress_ifindex):
2131 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, skb_iif) != 4);
2132
2133 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
2134 offsetof(struct sk_buff, skb_iif));
2135 break;
2136
2137 case offsetof(struct __sk_buff, ifindex):
2138 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
2139
2140 *insn++ = BPF_LDX_MEM(bytes_to_bpf_size(FIELD_SIZEOF(struct sk_buff, dev)),
2141 dst_reg, src_reg,
2142 offsetof(struct sk_buff, dev));
2143 *insn++ = BPF_JMP_IMM(BPF_JEQ, dst_reg, 0, 1);
2144 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, dst_reg,
2145 offsetof(struct net_device, ifindex));
2146 break;
2147
2148 case offsetof(struct __sk_buff, hash):
2149 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
2150
2151 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
2152 offsetof(struct sk_buff, hash));
2153 break;
2154
2155 case offsetof(struct __sk_buff, mark):
2156 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
2157
2158 if (type == BPF_WRITE)
2159 *insn++ = BPF_STX_MEM(BPF_W, dst_reg, src_reg,
2160 offsetof(struct sk_buff, mark));
2161 else
2162 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
2163 offsetof(struct sk_buff, mark));
2164 break;
2165
2166 case offsetof(struct __sk_buff, pkt_type):
2167 return convert_skb_access(SKF_AD_PKTTYPE, dst_reg, src_reg, insn);
2168
2169 case offsetof(struct __sk_buff, queue_mapping):
2170 return convert_skb_access(SKF_AD_QUEUE, dst_reg, src_reg, insn);
2171
2172 case offsetof(struct __sk_buff, vlan_present):
2173 return convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
2174 dst_reg, src_reg, insn);
2175
2176 case offsetof(struct __sk_buff, vlan_tci):
2177 return convert_skb_access(SKF_AD_VLAN_TAG,
2178 dst_reg, src_reg, insn);
2179
2180 case offsetof(struct __sk_buff, cb[0]) ...
2181 offsetof(struct __sk_buff, cb[4]):
2182 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20);
2183
2184 prog->cb_access = 1;
2185 ctx_off -= offsetof(struct __sk_buff, cb[0]);
2186 ctx_off += offsetof(struct sk_buff, cb);
2187 ctx_off += offsetof(struct qdisc_skb_cb, data);
2188 if (type == BPF_WRITE)
2189 *insn++ = BPF_STX_MEM(BPF_W, dst_reg, src_reg, ctx_off);
2190 else
2191 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, ctx_off);
2192 break;
2193
2194 case offsetof(struct __sk_buff, tc_classid):
2195 ctx_off -= offsetof(struct __sk_buff, tc_classid);
2196 ctx_off += offsetof(struct sk_buff, cb);
2197 ctx_off += offsetof(struct qdisc_skb_cb, tc_classid);
2198 WARN_ON(type != BPF_WRITE);
2199 *insn++ = BPF_STX_MEM(BPF_H, dst_reg, src_reg, ctx_off);
2200 break;
2201
2202 case offsetof(struct __sk_buff, tc_index):
2203 #ifdef CONFIG_NET_SCHED
2204 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, tc_index) != 2);
2205
2206 if (type == BPF_WRITE)
2207 *insn++ = BPF_STX_MEM(BPF_H, dst_reg, src_reg,
2208 offsetof(struct sk_buff, tc_index));
2209 else
2210 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
2211 offsetof(struct sk_buff, tc_index));
2212 break;
2213 #else
2214 if (type == BPF_WRITE)
2215 *insn++ = BPF_MOV64_REG(dst_reg, dst_reg);
2216 else
2217 *insn++ = BPF_MOV64_IMM(dst_reg, 0);
2218 break;
2219 #endif
2220 }
2221
2222 return insn - insn_buf;
2223 }
2224
2225 static const struct bpf_verifier_ops sk_filter_ops = {
2226 .get_func_proto = sk_filter_func_proto,
2227 .is_valid_access = sk_filter_is_valid_access,
2228 .convert_ctx_access = bpf_net_convert_ctx_access,
2229 };
2230
2231 static const struct bpf_verifier_ops tc_cls_act_ops = {
2232 .get_func_proto = tc_cls_act_func_proto,
2233 .is_valid_access = tc_cls_act_is_valid_access,
2234 .convert_ctx_access = bpf_net_convert_ctx_access,
2235 };
2236
2237 static struct bpf_prog_type_list sk_filter_type __read_mostly = {
2238 .ops = &sk_filter_ops,
2239 .type = BPF_PROG_TYPE_SOCKET_FILTER,
2240 };
2241
2242 static struct bpf_prog_type_list sched_cls_type __read_mostly = {
2243 .ops = &tc_cls_act_ops,
2244 .type = BPF_PROG_TYPE_SCHED_CLS,
2245 };
2246
2247 static struct bpf_prog_type_list sched_act_type __read_mostly = {
2248 .ops = &tc_cls_act_ops,
2249 .type = BPF_PROG_TYPE_SCHED_ACT,
2250 };
2251
2252 static int __init register_sk_filter_ops(void)
2253 {
2254 bpf_register_prog_type(&sk_filter_type);
2255 bpf_register_prog_type(&sched_cls_type);
2256 bpf_register_prog_type(&sched_act_type);
2257
2258 return 0;
2259 }
2260 late_initcall(register_sk_filter_ops);
2261
2262 int sk_detach_filter(struct sock *sk)
2263 {
2264 int ret = -ENOENT;
2265 struct sk_filter *filter;
2266
2267 if (sock_flag(sk, SOCK_FILTER_LOCKED))
2268 return -EPERM;
2269
2270 filter = rcu_dereference_protected(sk->sk_filter,
2271 sock_owned_by_user(sk));
2272 if (filter) {
2273 RCU_INIT_POINTER(sk->sk_filter, NULL);
2274 sk_filter_uncharge(sk, filter);
2275 ret = 0;
2276 }
2277
2278 return ret;
2279 }
2280 EXPORT_SYMBOL_GPL(sk_detach_filter);
2281
2282 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
2283 unsigned int len)
2284 {
2285 struct sock_fprog_kern *fprog;
2286 struct sk_filter *filter;
2287 int ret = 0;
2288
2289 lock_sock(sk);
2290 filter = rcu_dereference_protected(sk->sk_filter,
2291 sock_owned_by_user(sk));
2292 if (!filter)
2293 goto out;
2294
2295 /* We're copying the filter that has been originally attached,
2296 * so no conversion/decode needed anymore. eBPF programs that
2297 * have no original program cannot be dumped through this.
2298 */
2299 ret = -EACCES;
2300 fprog = filter->prog->orig_prog;
2301 if (!fprog)
2302 goto out;
2303
2304 ret = fprog->len;
2305 if (!len)
2306 /* User space only enquires number of filter blocks. */
2307 goto out;
2308
2309 ret = -EINVAL;
2310 if (len < fprog->len)
2311 goto out;
2312
2313 ret = -EFAULT;
2314 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
2315 goto out;
2316
2317 /* Instead of bytes, the API requests to return the number
2318 * of filter blocks.
2319 */
2320 ret = fprog->len;
2321 out:
2322 release_sock(sk);
2323 return ret;
2324 }
(deprecated) Btrfs devel tree