Commit | Line | Data |
---|---|---|
99c55f7d AS |
1 | /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com |
2 | * | |
3 | * This program is free software; you can redistribute it and/or | |
4 | * modify it under the terms of version 2 of the GNU General Public | |
5 | * License as published by the Free Software Foundation. | |
6 | * | |
7 | * This program is distributed in the hope that it will be useful, but | |
8 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
9 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
10 | * General Public License for more details. | |
11 | */ | |
12 | #include <linux/bpf.h> | |
13 | #include <linux/syscalls.h> | |
14 | #include <linux/slab.h> | |
15 | #include <linux/anon_inodes.h> | |
db20fd2b | 16 | #include <linux/file.h> |
09756af4 AS |
17 | #include <linux/license.h> |
18 | #include <linux/filter.h> | |
2541517c | 19 | #include <linux/version.h> |
99c55f7d | 20 | |
1be7f75d AS |
21 | int sysctl_unprivileged_bpf_disabled __read_mostly; |
22 | ||
99c55f7d AS |
23 | static LIST_HEAD(bpf_map_types); |
24 | ||
25 | static struct bpf_map *find_and_alloc_map(union bpf_attr *attr) | |
26 | { | |
27 | struct bpf_map_type_list *tl; | |
28 | struct bpf_map *map; | |
29 | ||
30 | list_for_each_entry(tl, &bpf_map_types, list_node) { | |
31 | if (tl->type == attr->map_type) { | |
32 | map = tl->ops->map_alloc(attr); | |
33 | if (IS_ERR(map)) | |
34 | return map; | |
35 | map->ops = tl->ops; | |
36 | map->map_type = attr->map_type; | |
37 | return map; | |
38 | } | |
39 | } | |
40 | return ERR_PTR(-EINVAL); | |
41 | } | |
42 | ||
43 | /* boot time registration of different map implementations */ | |
44 | void bpf_register_map_type(struct bpf_map_type_list *tl) | |
45 | { | |
46 | list_add(&tl->list_node, &bpf_map_types); | |
47 | } | |
48 | ||
aaac3ba9 AS |
49 | static int bpf_map_charge_memlock(struct bpf_map *map) |
50 | { | |
51 | struct user_struct *user = get_current_user(); | |
52 | unsigned long memlock_limit; | |
53 | ||
54 | memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
55 | ||
56 | atomic_long_add(map->pages, &user->locked_vm); | |
57 | ||
58 | if (atomic_long_read(&user->locked_vm) > memlock_limit) { | |
59 | atomic_long_sub(map->pages, &user->locked_vm); | |
60 | free_uid(user); | |
61 | return -EPERM; | |
62 | } | |
63 | map->user = user; | |
64 | return 0; | |
65 | } | |
66 | ||
67 | static void bpf_map_uncharge_memlock(struct bpf_map *map) | |
68 | { | |
69 | struct user_struct *user = map->user; | |
70 | ||
71 | atomic_long_sub(map->pages, &user->locked_vm); | |
72 | free_uid(user); | |
73 | } | |
74 | ||
99c55f7d AS |
75 | /* called from workqueue */ |
76 | static void bpf_map_free_deferred(struct work_struct *work) | |
77 | { | |
78 | struct bpf_map *map = container_of(work, struct bpf_map, work); | |
79 | ||
aaac3ba9 | 80 | bpf_map_uncharge_memlock(map); |
99c55f7d AS |
81 | /* implementation dependent freeing */ |
82 | map->ops->map_free(map); | |
83 | } | |
84 | ||
85 | /* decrement map refcnt and schedule it for freeing via workqueue | |
86 | * (unrelying map implementation ops->map_free() might sleep) | |
87 | */ | |
88 | void bpf_map_put(struct bpf_map *map) | |
89 | { | |
90 | if (atomic_dec_and_test(&map->refcnt)) { | |
91 | INIT_WORK(&map->work, bpf_map_free_deferred); | |
92 | schedule_work(&map->work); | |
93 | } | |
94 | } | |
95 | ||
96 | static int bpf_map_release(struct inode *inode, struct file *filp) | |
97 | { | |
98 | struct bpf_map *map = filp->private_data; | |
99 | ||
04fd61ab AS |
100 | if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) |
101 | /* prog_array stores refcnt-ed bpf_prog pointers | |
102 | * release them all when user space closes prog_array_fd | |
103 | */ | |
2a36f0b9 | 104 | bpf_fd_array_map_clear(map); |
04fd61ab | 105 | |
99c55f7d AS |
106 | bpf_map_put(map); |
107 | return 0; | |
108 | } | |
109 | ||
110 | static const struct file_operations bpf_map_fops = { | |
111 | .release = bpf_map_release, | |
112 | }; | |
113 | ||
114 | /* helper macro to check that unused fields 'union bpf_attr' are zero */ | |
115 | #define CHECK_ATTR(CMD) \ | |
116 | memchr_inv((void *) &attr->CMD##_LAST_FIELD + \ | |
117 | sizeof(attr->CMD##_LAST_FIELD), 0, \ | |
118 | sizeof(*attr) - \ | |
119 | offsetof(union bpf_attr, CMD##_LAST_FIELD) - \ | |
120 | sizeof(attr->CMD##_LAST_FIELD)) != NULL | |
121 | ||
122 | #define BPF_MAP_CREATE_LAST_FIELD max_entries | |
123 | /* called via syscall */ | |
124 | static int map_create(union bpf_attr *attr) | |
125 | { | |
126 | struct bpf_map *map; | |
127 | int err; | |
128 | ||
129 | err = CHECK_ATTR(BPF_MAP_CREATE); | |
130 | if (err) | |
131 | return -EINVAL; | |
132 | ||
133 | /* find map type and init map: hashtable vs rbtree vs bloom vs ... */ | |
134 | map = find_and_alloc_map(attr); | |
135 | if (IS_ERR(map)) | |
136 | return PTR_ERR(map); | |
137 | ||
138 | atomic_set(&map->refcnt, 1); | |
139 | ||
aaac3ba9 AS |
140 | err = bpf_map_charge_memlock(map); |
141 | if (err) | |
142 | goto free_map; | |
143 | ||
99c55f7d AS |
144 | err = anon_inode_getfd("bpf-map", &bpf_map_fops, map, O_RDWR | O_CLOEXEC); |
145 | ||
146 | if (err < 0) | |
147 | /* failed to allocate fd */ | |
148 | goto free_map; | |
149 | ||
150 | return err; | |
151 | ||
152 | free_map: | |
153 | map->ops->map_free(map); | |
154 | return err; | |
155 | } | |
156 | ||
db20fd2b AS |
157 | /* if error is returned, fd is released. |
158 | * On success caller should complete fd access with matching fdput() | |
159 | */ | |
160 | struct bpf_map *bpf_map_get(struct fd f) | |
161 | { | |
162 | struct bpf_map *map; | |
163 | ||
164 | if (!f.file) | |
165 | return ERR_PTR(-EBADF); | |
166 | ||
167 | if (f.file->f_op != &bpf_map_fops) { | |
168 | fdput(f); | |
169 | return ERR_PTR(-EINVAL); | |
170 | } | |
171 | ||
172 | map = f.file->private_data; | |
173 | ||
174 | return map; | |
175 | } | |
176 | ||
177 | /* helper to convert user pointers passed inside __aligned_u64 fields */ | |
178 | static void __user *u64_to_ptr(__u64 val) | |
179 | { | |
180 | return (void __user *) (unsigned long) val; | |
181 | } | |
182 | ||
183 | /* last field in 'union bpf_attr' used by this command */ | |
184 | #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value | |
185 | ||
186 | static int map_lookup_elem(union bpf_attr *attr) | |
187 | { | |
188 | void __user *ukey = u64_to_ptr(attr->key); | |
189 | void __user *uvalue = u64_to_ptr(attr->value); | |
190 | int ufd = attr->map_fd; | |
db20fd2b | 191 | struct bpf_map *map; |
8ebe667c | 192 | void *key, *value, *ptr; |
592867bf | 193 | struct fd f; |
db20fd2b AS |
194 | int err; |
195 | ||
196 | if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM)) | |
197 | return -EINVAL; | |
198 | ||
592867bf | 199 | f = fdget(ufd); |
db20fd2b AS |
200 | map = bpf_map_get(f); |
201 | if (IS_ERR(map)) | |
202 | return PTR_ERR(map); | |
203 | ||
204 | err = -ENOMEM; | |
205 | key = kmalloc(map->key_size, GFP_USER); | |
206 | if (!key) | |
207 | goto err_put; | |
208 | ||
209 | err = -EFAULT; | |
210 | if (copy_from_user(key, ukey, map->key_size) != 0) | |
211 | goto free_key; | |
212 | ||
8ebe667c AS |
213 | err = -ENOMEM; |
214 | value = kmalloc(map->value_size, GFP_USER); | |
db20fd2b | 215 | if (!value) |
8ebe667c AS |
216 | goto free_key; |
217 | ||
218 | rcu_read_lock(); | |
219 | ptr = map->ops->map_lookup_elem(map, key); | |
220 | if (ptr) | |
221 | memcpy(value, ptr, map->value_size); | |
222 | rcu_read_unlock(); | |
223 | ||
224 | err = -ENOENT; | |
225 | if (!ptr) | |
226 | goto free_value; | |
db20fd2b AS |
227 | |
228 | err = -EFAULT; | |
229 | if (copy_to_user(uvalue, value, map->value_size) != 0) | |
8ebe667c | 230 | goto free_value; |
db20fd2b AS |
231 | |
232 | err = 0; | |
233 | ||
8ebe667c AS |
234 | free_value: |
235 | kfree(value); | |
db20fd2b AS |
236 | free_key: |
237 | kfree(key); | |
238 | err_put: | |
239 | fdput(f); | |
240 | return err; | |
241 | } | |
242 | ||
3274f520 | 243 | #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags |
db20fd2b AS |
244 | |
245 | static int map_update_elem(union bpf_attr *attr) | |
246 | { | |
247 | void __user *ukey = u64_to_ptr(attr->key); | |
248 | void __user *uvalue = u64_to_ptr(attr->value); | |
249 | int ufd = attr->map_fd; | |
db20fd2b AS |
250 | struct bpf_map *map; |
251 | void *key, *value; | |
592867bf | 252 | struct fd f; |
db20fd2b AS |
253 | int err; |
254 | ||
255 | if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM)) | |
256 | return -EINVAL; | |
257 | ||
592867bf | 258 | f = fdget(ufd); |
db20fd2b AS |
259 | map = bpf_map_get(f); |
260 | if (IS_ERR(map)) | |
261 | return PTR_ERR(map); | |
262 | ||
263 | err = -ENOMEM; | |
264 | key = kmalloc(map->key_size, GFP_USER); | |
265 | if (!key) | |
266 | goto err_put; | |
267 | ||
268 | err = -EFAULT; | |
269 | if (copy_from_user(key, ukey, map->key_size) != 0) | |
270 | goto free_key; | |
271 | ||
272 | err = -ENOMEM; | |
273 | value = kmalloc(map->value_size, GFP_USER); | |
274 | if (!value) | |
275 | goto free_key; | |
276 | ||
277 | err = -EFAULT; | |
278 | if (copy_from_user(value, uvalue, map->value_size) != 0) | |
279 | goto free_value; | |
280 | ||
281 | /* eBPF program that use maps are running under rcu_read_lock(), | |
282 | * therefore all map accessors rely on this fact, so do the same here | |
283 | */ | |
284 | rcu_read_lock(); | |
3274f520 | 285 | err = map->ops->map_update_elem(map, key, value, attr->flags); |
db20fd2b AS |
286 | rcu_read_unlock(); |
287 | ||
288 | free_value: | |
289 | kfree(value); | |
290 | free_key: | |
291 | kfree(key); | |
292 | err_put: | |
293 | fdput(f); | |
294 | return err; | |
295 | } | |
296 | ||
297 | #define BPF_MAP_DELETE_ELEM_LAST_FIELD key | |
298 | ||
299 | static int map_delete_elem(union bpf_attr *attr) | |
300 | { | |
301 | void __user *ukey = u64_to_ptr(attr->key); | |
302 | int ufd = attr->map_fd; | |
db20fd2b | 303 | struct bpf_map *map; |
592867bf | 304 | struct fd f; |
db20fd2b AS |
305 | void *key; |
306 | int err; | |
307 | ||
308 | if (CHECK_ATTR(BPF_MAP_DELETE_ELEM)) | |
309 | return -EINVAL; | |
310 | ||
592867bf | 311 | f = fdget(ufd); |
db20fd2b AS |
312 | map = bpf_map_get(f); |
313 | if (IS_ERR(map)) | |
314 | return PTR_ERR(map); | |
315 | ||
316 | err = -ENOMEM; | |
317 | key = kmalloc(map->key_size, GFP_USER); | |
318 | if (!key) | |
319 | goto err_put; | |
320 | ||
321 | err = -EFAULT; | |
322 | if (copy_from_user(key, ukey, map->key_size) != 0) | |
323 | goto free_key; | |
324 | ||
325 | rcu_read_lock(); | |
326 | err = map->ops->map_delete_elem(map, key); | |
327 | rcu_read_unlock(); | |
328 | ||
329 | free_key: | |
330 | kfree(key); | |
331 | err_put: | |
332 | fdput(f); | |
333 | return err; | |
334 | } | |
335 | ||
336 | /* last field in 'union bpf_attr' used by this command */ | |
337 | #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key | |
338 | ||
339 | static int map_get_next_key(union bpf_attr *attr) | |
340 | { | |
341 | void __user *ukey = u64_to_ptr(attr->key); | |
342 | void __user *unext_key = u64_to_ptr(attr->next_key); | |
343 | int ufd = attr->map_fd; | |
db20fd2b AS |
344 | struct bpf_map *map; |
345 | void *key, *next_key; | |
592867bf | 346 | struct fd f; |
db20fd2b AS |
347 | int err; |
348 | ||
349 | if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY)) | |
350 | return -EINVAL; | |
351 | ||
592867bf | 352 | f = fdget(ufd); |
db20fd2b AS |
353 | map = bpf_map_get(f); |
354 | if (IS_ERR(map)) | |
355 | return PTR_ERR(map); | |
356 | ||
357 | err = -ENOMEM; | |
358 | key = kmalloc(map->key_size, GFP_USER); | |
359 | if (!key) | |
360 | goto err_put; | |
361 | ||
362 | err = -EFAULT; | |
363 | if (copy_from_user(key, ukey, map->key_size) != 0) | |
364 | goto free_key; | |
365 | ||
366 | err = -ENOMEM; | |
367 | next_key = kmalloc(map->key_size, GFP_USER); | |
368 | if (!next_key) | |
369 | goto free_key; | |
370 | ||
371 | rcu_read_lock(); | |
372 | err = map->ops->map_get_next_key(map, key, next_key); | |
373 | rcu_read_unlock(); | |
374 | if (err) | |
375 | goto free_next_key; | |
376 | ||
377 | err = -EFAULT; | |
378 | if (copy_to_user(unext_key, next_key, map->key_size) != 0) | |
379 | goto free_next_key; | |
380 | ||
381 | err = 0; | |
382 | ||
383 | free_next_key: | |
384 | kfree(next_key); | |
385 | free_key: | |
386 | kfree(key); | |
387 | err_put: | |
388 | fdput(f); | |
389 | return err; | |
390 | } | |
391 | ||
09756af4 AS |
392 | static LIST_HEAD(bpf_prog_types); |
393 | ||
394 | static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog) | |
395 | { | |
396 | struct bpf_prog_type_list *tl; | |
397 | ||
398 | list_for_each_entry(tl, &bpf_prog_types, list_node) { | |
399 | if (tl->type == type) { | |
400 | prog->aux->ops = tl->ops; | |
24701ece | 401 | prog->type = type; |
09756af4 AS |
402 | return 0; |
403 | } | |
404 | } | |
24701ece | 405 | |
09756af4 AS |
406 | return -EINVAL; |
407 | } | |
408 | ||
409 | void bpf_register_prog_type(struct bpf_prog_type_list *tl) | |
410 | { | |
411 | list_add(&tl->list_node, &bpf_prog_types); | |
412 | } | |
413 | ||
0a542a86 AS |
414 | /* fixup insn->imm field of bpf_call instructions: |
415 | * if (insn->imm == BPF_FUNC_map_lookup_elem) | |
416 | * insn->imm = bpf_map_lookup_elem - __bpf_call_base; | |
417 | * else if (insn->imm == BPF_FUNC_map_update_elem) | |
418 | * insn->imm = bpf_map_update_elem - __bpf_call_base; | |
419 | * else ... | |
420 | * | |
421 | * this function is called after eBPF program passed verification | |
422 | */ | |
423 | static void fixup_bpf_calls(struct bpf_prog *prog) | |
424 | { | |
425 | const struct bpf_func_proto *fn; | |
426 | int i; | |
427 | ||
428 | for (i = 0; i < prog->len; i++) { | |
429 | struct bpf_insn *insn = &prog->insnsi[i]; | |
430 | ||
431 | if (insn->code == (BPF_JMP | BPF_CALL)) { | |
432 | /* we reach here when program has bpf_call instructions | |
433 | * and it passed bpf_check(), means that | |
434 | * ops->get_func_proto must have been supplied, check it | |
435 | */ | |
436 | BUG_ON(!prog->aux->ops->get_func_proto); | |
437 | ||
c46646d0 DB |
438 | if (insn->imm == BPF_FUNC_get_route_realm) |
439 | prog->dst_needed = 1; | |
3ad00405 DB |
440 | if (insn->imm == BPF_FUNC_get_prandom_u32) |
441 | bpf_user_rnd_init_once(); | |
04fd61ab AS |
442 | if (insn->imm == BPF_FUNC_tail_call) { |
443 | /* mark bpf_tail_call as different opcode | |
444 | * to avoid conditional branch in | |
445 | * interpeter for every normal call | |
446 | * and to prevent accidental JITing by | |
447 | * JIT compiler that doesn't support | |
448 | * bpf_tail_call yet | |
449 | */ | |
450 | insn->imm = 0; | |
451 | insn->code |= BPF_X; | |
452 | continue; | |
453 | } | |
454 | ||
0a542a86 AS |
455 | fn = prog->aux->ops->get_func_proto(insn->imm); |
456 | /* all functions that have prototype and verifier allowed | |
457 | * programs to call them, must be real in-kernel functions | |
458 | */ | |
459 | BUG_ON(!fn->func); | |
460 | insn->imm = fn->func - __bpf_call_base; | |
461 | } | |
462 | } | |
463 | } | |
464 | ||
09756af4 AS |
465 | /* drop refcnt on maps used by eBPF program and free auxilary data */ |
466 | static void free_used_maps(struct bpf_prog_aux *aux) | |
467 | { | |
468 | int i; | |
469 | ||
470 | for (i = 0; i < aux->used_map_cnt; i++) | |
471 | bpf_map_put(aux->used_maps[i]); | |
472 | ||
473 | kfree(aux->used_maps); | |
474 | } | |
475 | ||
aaac3ba9 AS |
476 | static int bpf_prog_charge_memlock(struct bpf_prog *prog) |
477 | { | |
478 | struct user_struct *user = get_current_user(); | |
479 | unsigned long memlock_limit; | |
480 | ||
481 | memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
482 | ||
483 | atomic_long_add(prog->pages, &user->locked_vm); | |
484 | if (atomic_long_read(&user->locked_vm) > memlock_limit) { | |
485 | atomic_long_sub(prog->pages, &user->locked_vm); | |
486 | free_uid(user); | |
487 | return -EPERM; | |
488 | } | |
489 | prog->aux->user = user; | |
490 | return 0; | |
491 | } | |
492 | ||
493 | static void bpf_prog_uncharge_memlock(struct bpf_prog *prog) | |
494 | { | |
495 | struct user_struct *user = prog->aux->user; | |
496 | ||
497 | atomic_long_sub(prog->pages, &user->locked_vm); | |
498 | free_uid(user); | |
499 | } | |
500 | ||
abf2e7d6 AS |
501 | static void __prog_put_rcu(struct rcu_head *rcu) |
502 | { | |
503 | struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu); | |
504 | ||
505 | free_used_maps(aux); | |
aaac3ba9 | 506 | bpf_prog_uncharge_memlock(aux->prog); |
abf2e7d6 AS |
507 | bpf_prog_free(aux->prog); |
508 | } | |
509 | ||
510 | /* version of bpf_prog_put() that is called after a grace period */ | |
511 | void bpf_prog_put_rcu(struct bpf_prog *prog) | |
512 | { | |
513 | if (atomic_dec_and_test(&prog->aux->refcnt)) { | |
514 | prog->aux->prog = prog; | |
515 | call_rcu(&prog->aux->rcu, __prog_put_rcu); | |
516 | } | |
517 | } | |
518 | ||
09756af4 AS |
519 | void bpf_prog_put(struct bpf_prog *prog) |
520 | { | |
521 | if (atomic_dec_and_test(&prog->aux->refcnt)) { | |
522 | free_used_maps(prog->aux); | |
523 | bpf_prog_free(prog); | |
524 | } | |
525 | } | |
e2e9b654 | 526 | EXPORT_SYMBOL_GPL(bpf_prog_put); |
09756af4 AS |
527 | |
528 | static int bpf_prog_release(struct inode *inode, struct file *filp) | |
529 | { | |
530 | struct bpf_prog *prog = filp->private_data; | |
531 | ||
abf2e7d6 | 532 | bpf_prog_put_rcu(prog); |
09756af4 AS |
533 | return 0; |
534 | } | |
535 | ||
536 | static const struct file_operations bpf_prog_fops = { | |
537 | .release = bpf_prog_release, | |
538 | }; | |
539 | ||
540 | static struct bpf_prog *get_prog(struct fd f) | |
541 | { | |
542 | struct bpf_prog *prog; | |
543 | ||
544 | if (!f.file) | |
545 | return ERR_PTR(-EBADF); | |
546 | ||
547 | if (f.file->f_op != &bpf_prog_fops) { | |
548 | fdput(f); | |
549 | return ERR_PTR(-EINVAL); | |
550 | } | |
551 | ||
552 | prog = f.file->private_data; | |
553 | ||
554 | return prog; | |
555 | } | |
556 | ||
557 | /* called by sockets/tracing/seccomp before attaching program to an event | |
558 | * pairs with bpf_prog_put() | |
559 | */ | |
560 | struct bpf_prog *bpf_prog_get(u32 ufd) | |
561 | { | |
562 | struct fd f = fdget(ufd); | |
563 | struct bpf_prog *prog; | |
564 | ||
565 | prog = get_prog(f); | |
566 | ||
567 | if (IS_ERR(prog)) | |
568 | return prog; | |
569 | ||
570 | atomic_inc(&prog->aux->refcnt); | |
571 | fdput(f); | |
572 | return prog; | |
573 | } | |
e2e9b654 | 574 | EXPORT_SYMBOL_GPL(bpf_prog_get); |
09756af4 AS |
575 | |
576 | /* last field in 'union bpf_attr' used by this command */ | |
2541517c | 577 | #define BPF_PROG_LOAD_LAST_FIELD kern_version |
09756af4 AS |
578 | |
579 | static int bpf_prog_load(union bpf_attr *attr) | |
580 | { | |
581 | enum bpf_prog_type type = attr->prog_type; | |
582 | struct bpf_prog *prog; | |
583 | int err; | |
584 | char license[128]; | |
585 | bool is_gpl; | |
586 | ||
587 | if (CHECK_ATTR(BPF_PROG_LOAD)) | |
588 | return -EINVAL; | |
589 | ||
590 | /* copy eBPF program license from user space */ | |
591 | if (strncpy_from_user(license, u64_to_ptr(attr->license), | |
592 | sizeof(license) - 1) < 0) | |
593 | return -EFAULT; | |
594 | license[sizeof(license) - 1] = 0; | |
595 | ||
596 | /* eBPF programs must be GPL compatible to use GPL-ed functions */ | |
597 | is_gpl = license_is_gpl_compatible(license); | |
598 | ||
599 | if (attr->insn_cnt >= BPF_MAXINSNS) | |
600 | return -EINVAL; | |
601 | ||
2541517c AS |
602 | if (type == BPF_PROG_TYPE_KPROBE && |
603 | attr->kern_version != LINUX_VERSION_CODE) | |
604 | return -EINVAL; | |
605 | ||
1be7f75d AS |
606 | if (type != BPF_PROG_TYPE_SOCKET_FILTER && !capable(CAP_SYS_ADMIN)) |
607 | return -EPERM; | |
608 | ||
09756af4 AS |
609 | /* plain bpf_prog allocation */ |
610 | prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER); | |
611 | if (!prog) | |
612 | return -ENOMEM; | |
613 | ||
aaac3ba9 AS |
614 | err = bpf_prog_charge_memlock(prog); |
615 | if (err) | |
616 | goto free_prog_nouncharge; | |
617 | ||
09756af4 AS |
618 | prog->len = attr->insn_cnt; |
619 | ||
620 | err = -EFAULT; | |
621 | if (copy_from_user(prog->insns, u64_to_ptr(attr->insns), | |
622 | prog->len * sizeof(struct bpf_insn)) != 0) | |
623 | goto free_prog; | |
624 | ||
625 | prog->orig_prog = NULL; | |
a91263d5 | 626 | prog->jited = 0; |
09756af4 AS |
627 | |
628 | atomic_set(&prog->aux->refcnt, 1); | |
a91263d5 | 629 | prog->gpl_compatible = is_gpl ? 1 : 0; |
09756af4 AS |
630 | |
631 | /* find program type: socket_filter vs tracing_filter */ | |
632 | err = find_prog_type(type, prog); | |
633 | if (err < 0) | |
634 | goto free_prog; | |
635 | ||
636 | /* run eBPF verifier */ | |
9bac3d6d | 637 | err = bpf_check(&prog, attr); |
09756af4 AS |
638 | if (err < 0) |
639 | goto free_used_maps; | |
640 | ||
0a542a86 AS |
641 | /* fixup BPF_CALL->imm field */ |
642 | fixup_bpf_calls(prog); | |
643 | ||
09756af4 | 644 | /* eBPF program is ready to be JITed */ |
04fd61ab AS |
645 | err = bpf_prog_select_runtime(prog); |
646 | if (err < 0) | |
647 | goto free_used_maps; | |
09756af4 AS |
648 | |
649 | err = anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog, O_RDWR | O_CLOEXEC); | |
09756af4 AS |
650 | if (err < 0) |
651 | /* failed to allocate fd */ | |
652 | goto free_used_maps; | |
653 | ||
654 | return err; | |
655 | ||
656 | free_used_maps: | |
657 | free_used_maps(prog->aux); | |
658 | free_prog: | |
aaac3ba9 AS |
659 | bpf_prog_uncharge_memlock(prog); |
660 | free_prog_nouncharge: | |
09756af4 AS |
661 | bpf_prog_free(prog); |
662 | return err; | |
663 | } | |
664 | ||
99c55f7d AS |
665 | SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size) |
666 | { | |
667 | union bpf_attr attr = {}; | |
668 | int err; | |
669 | ||
1be7f75d | 670 | if (!capable(CAP_SYS_ADMIN) && sysctl_unprivileged_bpf_disabled) |
99c55f7d AS |
671 | return -EPERM; |
672 | ||
673 | if (!access_ok(VERIFY_READ, uattr, 1)) | |
674 | return -EFAULT; | |
675 | ||
676 | if (size > PAGE_SIZE) /* silly large */ | |
677 | return -E2BIG; | |
678 | ||
679 | /* If we're handed a bigger struct than we know of, | |
680 | * ensure all the unknown bits are 0 - i.e. new | |
681 | * user-space does not rely on any kernel feature | |
682 | * extensions we dont know about yet. | |
683 | */ | |
684 | if (size > sizeof(attr)) { | |
685 | unsigned char __user *addr; | |
686 | unsigned char __user *end; | |
687 | unsigned char val; | |
688 | ||
689 | addr = (void __user *)uattr + sizeof(attr); | |
690 | end = (void __user *)uattr + size; | |
691 | ||
692 | for (; addr < end; addr++) { | |
693 | err = get_user(val, addr); | |
694 | if (err) | |
695 | return err; | |
696 | if (val) | |
697 | return -E2BIG; | |
698 | } | |
699 | size = sizeof(attr); | |
700 | } | |
701 | ||
702 | /* copy attributes from user space, may be less than sizeof(bpf_attr) */ | |
703 | if (copy_from_user(&attr, uattr, size) != 0) | |
704 | return -EFAULT; | |
705 | ||
706 | switch (cmd) { | |
707 | case BPF_MAP_CREATE: | |
708 | err = map_create(&attr); | |
709 | break; | |
db20fd2b AS |
710 | case BPF_MAP_LOOKUP_ELEM: |
711 | err = map_lookup_elem(&attr); | |
712 | break; | |
713 | case BPF_MAP_UPDATE_ELEM: | |
714 | err = map_update_elem(&attr); | |
715 | break; | |
716 | case BPF_MAP_DELETE_ELEM: | |
717 | err = map_delete_elem(&attr); | |
718 | break; | |
719 | case BPF_MAP_GET_NEXT_KEY: | |
720 | err = map_get_next_key(&attr); | |
721 | break; | |
09756af4 AS |
722 | case BPF_PROG_LOAD: |
723 | err = bpf_prog_load(&attr); | |
724 | break; | |
99c55f7d AS |
725 | default: |
726 | err = -EINVAL; | |
727 | break; | |
728 | } | |
729 | ||
730 | return err; | |
731 | } |