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 | ||
c9da161c DB |
85 | static void bpf_map_put_uref(struct bpf_map *map) |
86 | { | |
87 | if (atomic_dec_and_test(&map->usercnt)) { | |
88 | if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) | |
89 | bpf_fd_array_map_clear(map); | |
90 | } | |
91 | } | |
92 | ||
99c55f7d AS |
93 | /* decrement map refcnt and schedule it for freeing via workqueue |
94 | * (unrelying map implementation ops->map_free() might sleep) | |
95 | */ | |
96 | void bpf_map_put(struct bpf_map *map) | |
97 | { | |
98 | if (atomic_dec_and_test(&map->refcnt)) { | |
99 | INIT_WORK(&map->work, bpf_map_free_deferred); | |
100 | schedule_work(&map->work); | |
101 | } | |
102 | } | |
103 | ||
c9da161c | 104 | void bpf_map_put_with_uref(struct bpf_map *map) |
99c55f7d | 105 | { |
c9da161c | 106 | bpf_map_put_uref(map); |
99c55f7d | 107 | bpf_map_put(map); |
c9da161c DB |
108 | } |
109 | ||
110 | static int bpf_map_release(struct inode *inode, struct file *filp) | |
111 | { | |
112 | bpf_map_put_with_uref(filp->private_data); | |
99c55f7d AS |
113 | return 0; |
114 | } | |
115 | ||
f99bf205 DB |
116 | #ifdef CONFIG_PROC_FS |
117 | static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp) | |
118 | { | |
119 | const struct bpf_map *map = filp->private_data; | |
120 | ||
121 | seq_printf(m, | |
122 | "map_type:\t%u\n" | |
123 | "key_size:\t%u\n" | |
124 | "value_size:\t%u\n" | |
125 | "max_entries:\t%u\n", | |
126 | map->map_type, | |
127 | map->key_size, | |
128 | map->value_size, | |
129 | map->max_entries); | |
130 | } | |
131 | #endif | |
132 | ||
99c55f7d | 133 | static const struct file_operations bpf_map_fops = { |
f99bf205 DB |
134 | #ifdef CONFIG_PROC_FS |
135 | .show_fdinfo = bpf_map_show_fdinfo, | |
136 | #endif | |
137 | .release = bpf_map_release, | |
99c55f7d AS |
138 | }; |
139 | ||
b2197755 | 140 | int bpf_map_new_fd(struct bpf_map *map) |
aa79781b DB |
141 | { |
142 | return anon_inode_getfd("bpf-map", &bpf_map_fops, map, | |
143 | O_RDWR | O_CLOEXEC); | |
144 | } | |
145 | ||
99c55f7d AS |
146 | /* helper macro to check that unused fields 'union bpf_attr' are zero */ |
147 | #define CHECK_ATTR(CMD) \ | |
148 | memchr_inv((void *) &attr->CMD##_LAST_FIELD + \ | |
149 | sizeof(attr->CMD##_LAST_FIELD), 0, \ | |
150 | sizeof(*attr) - \ | |
151 | offsetof(union bpf_attr, CMD##_LAST_FIELD) - \ | |
152 | sizeof(attr->CMD##_LAST_FIELD)) != NULL | |
153 | ||
154 | #define BPF_MAP_CREATE_LAST_FIELD max_entries | |
155 | /* called via syscall */ | |
156 | static int map_create(union bpf_attr *attr) | |
157 | { | |
158 | struct bpf_map *map; | |
159 | int err; | |
160 | ||
161 | err = CHECK_ATTR(BPF_MAP_CREATE); | |
162 | if (err) | |
163 | return -EINVAL; | |
164 | ||
165 | /* find map type and init map: hashtable vs rbtree vs bloom vs ... */ | |
166 | map = find_and_alloc_map(attr); | |
167 | if (IS_ERR(map)) | |
168 | return PTR_ERR(map); | |
169 | ||
170 | atomic_set(&map->refcnt, 1); | |
c9da161c | 171 | atomic_set(&map->usercnt, 1); |
99c55f7d | 172 | |
aaac3ba9 AS |
173 | err = bpf_map_charge_memlock(map); |
174 | if (err) | |
175 | goto free_map; | |
176 | ||
aa79781b | 177 | err = bpf_map_new_fd(map); |
99c55f7d AS |
178 | if (err < 0) |
179 | /* failed to allocate fd */ | |
180 | goto free_map; | |
181 | ||
182 | return err; | |
183 | ||
184 | free_map: | |
185 | map->ops->map_free(map); | |
186 | return err; | |
187 | } | |
188 | ||
db20fd2b AS |
189 | /* if error is returned, fd is released. |
190 | * On success caller should complete fd access with matching fdput() | |
191 | */ | |
c2101297 | 192 | struct bpf_map *__bpf_map_get(struct fd f) |
db20fd2b | 193 | { |
db20fd2b AS |
194 | if (!f.file) |
195 | return ERR_PTR(-EBADF); | |
db20fd2b AS |
196 | if (f.file->f_op != &bpf_map_fops) { |
197 | fdput(f); | |
198 | return ERR_PTR(-EINVAL); | |
199 | } | |
200 | ||
c2101297 DB |
201 | return f.file->private_data; |
202 | } | |
203 | ||
c9da161c DB |
204 | void bpf_map_inc(struct bpf_map *map, bool uref) |
205 | { | |
206 | atomic_inc(&map->refcnt); | |
207 | if (uref) | |
208 | atomic_inc(&map->usercnt); | |
209 | } | |
210 | ||
211 | struct bpf_map *bpf_map_get_with_uref(u32 ufd) | |
c2101297 DB |
212 | { |
213 | struct fd f = fdget(ufd); | |
214 | struct bpf_map *map; | |
215 | ||
216 | map = __bpf_map_get(f); | |
217 | if (IS_ERR(map)) | |
218 | return map; | |
219 | ||
c9da161c | 220 | bpf_map_inc(map, true); |
c2101297 | 221 | fdput(f); |
db20fd2b AS |
222 | |
223 | return map; | |
224 | } | |
225 | ||
226 | /* helper to convert user pointers passed inside __aligned_u64 fields */ | |
227 | static void __user *u64_to_ptr(__u64 val) | |
228 | { | |
229 | return (void __user *) (unsigned long) val; | |
230 | } | |
231 | ||
232 | /* last field in 'union bpf_attr' used by this command */ | |
233 | #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value | |
234 | ||
235 | static int map_lookup_elem(union bpf_attr *attr) | |
236 | { | |
237 | void __user *ukey = u64_to_ptr(attr->key); | |
238 | void __user *uvalue = u64_to_ptr(attr->value); | |
239 | int ufd = attr->map_fd; | |
db20fd2b | 240 | struct bpf_map *map; |
8ebe667c | 241 | void *key, *value, *ptr; |
592867bf | 242 | struct fd f; |
db20fd2b AS |
243 | int err; |
244 | ||
245 | if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM)) | |
246 | return -EINVAL; | |
247 | ||
592867bf | 248 | f = fdget(ufd); |
c2101297 | 249 | map = __bpf_map_get(f); |
db20fd2b AS |
250 | if (IS_ERR(map)) |
251 | return PTR_ERR(map); | |
252 | ||
253 | err = -ENOMEM; | |
254 | key = kmalloc(map->key_size, GFP_USER); | |
255 | if (!key) | |
256 | goto err_put; | |
257 | ||
258 | err = -EFAULT; | |
259 | if (copy_from_user(key, ukey, map->key_size) != 0) | |
260 | goto free_key; | |
261 | ||
8ebe667c | 262 | err = -ENOMEM; |
01b3f521 | 263 | value = kmalloc(map->value_size, GFP_USER | __GFP_NOWARN); |
db20fd2b | 264 | if (!value) |
8ebe667c AS |
265 | goto free_key; |
266 | ||
267 | rcu_read_lock(); | |
268 | ptr = map->ops->map_lookup_elem(map, key); | |
269 | if (ptr) | |
270 | memcpy(value, ptr, map->value_size); | |
271 | rcu_read_unlock(); | |
272 | ||
273 | err = -ENOENT; | |
274 | if (!ptr) | |
275 | goto free_value; | |
db20fd2b AS |
276 | |
277 | err = -EFAULT; | |
278 | if (copy_to_user(uvalue, value, map->value_size) != 0) | |
8ebe667c | 279 | goto free_value; |
db20fd2b AS |
280 | |
281 | err = 0; | |
282 | ||
8ebe667c AS |
283 | free_value: |
284 | kfree(value); | |
db20fd2b AS |
285 | free_key: |
286 | kfree(key); | |
287 | err_put: | |
288 | fdput(f); | |
289 | return err; | |
290 | } | |
291 | ||
3274f520 | 292 | #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags |
db20fd2b AS |
293 | |
294 | static int map_update_elem(union bpf_attr *attr) | |
295 | { | |
296 | void __user *ukey = u64_to_ptr(attr->key); | |
297 | void __user *uvalue = u64_to_ptr(attr->value); | |
298 | int ufd = attr->map_fd; | |
db20fd2b AS |
299 | struct bpf_map *map; |
300 | void *key, *value; | |
592867bf | 301 | struct fd f; |
db20fd2b AS |
302 | int err; |
303 | ||
304 | if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM)) | |
305 | return -EINVAL; | |
306 | ||
592867bf | 307 | f = fdget(ufd); |
c2101297 | 308 | map = __bpf_map_get(f); |
db20fd2b AS |
309 | if (IS_ERR(map)) |
310 | return PTR_ERR(map); | |
311 | ||
312 | err = -ENOMEM; | |
313 | key = kmalloc(map->key_size, GFP_USER); | |
314 | if (!key) | |
315 | goto err_put; | |
316 | ||
317 | err = -EFAULT; | |
318 | if (copy_from_user(key, ukey, map->key_size) != 0) | |
319 | goto free_key; | |
320 | ||
321 | err = -ENOMEM; | |
01b3f521 | 322 | value = kmalloc(map->value_size, GFP_USER | __GFP_NOWARN); |
db20fd2b AS |
323 | if (!value) |
324 | goto free_key; | |
325 | ||
326 | err = -EFAULT; | |
327 | if (copy_from_user(value, uvalue, map->value_size) != 0) | |
328 | goto free_value; | |
329 | ||
330 | /* eBPF program that use maps are running under rcu_read_lock(), | |
331 | * therefore all map accessors rely on this fact, so do the same here | |
332 | */ | |
333 | rcu_read_lock(); | |
3274f520 | 334 | err = map->ops->map_update_elem(map, key, value, attr->flags); |
db20fd2b AS |
335 | rcu_read_unlock(); |
336 | ||
337 | free_value: | |
338 | kfree(value); | |
339 | free_key: | |
340 | kfree(key); | |
341 | err_put: | |
342 | fdput(f); | |
343 | return err; | |
344 | } | |
345 | ||
346 | #define BPF_MAP_DELETE_ELEM_LAST_FIELD key | |
347 | ||
348 | static int map_delete_elem(union bpf_attr *attr) | |
349 | { | |
350 | void __user *ukey = u64_to_ptr(attr->key); | |
351 | int ufd = attr->map_fd; | |
db20fd2b | 352 | struct bpf_map *map; |
592867bf | 353 | struct fd f; |
db20fd2b AS |
354 | void *key; |
355 | int err; | |
356 | ||
357 | if (CHECK_ATTR(BPF_MAP_DELETE_ELEM)) | |
358 | return -EINVAL; | |
359 | ||
592867bf | 360 | f = fdget(ufd); |
c2101297 | 361 | map = __bpf_map_get(f); |
db20fd2b AS |
362 | if (IS_ERR(map)) |
363 | return PTR_ERR(map); | |
364 | ||
365 | err = -ENOMEM; | |
366 | key = kmalloc(map->key_size, GFP_USER); | |
367 | if (!key) | |
368 | goto err_put; | |
369 | ||
370 | err = -EFAULT; | |
371 | if (copy_from_user(key, ukey, map->key_size) != 0) | |
372 | goto free_key; | |
373 | ||
374 | rcu_read_lock(); | |
375 | err = map->ops->map_delete_elem(map, key); | |
376 | rcu_read_unlock(); | |
377 | ||
378 | free_key: | |
379 | kfree(key); | |
380 | err_put: | |
381 | fdput(f); | |
382 | return err; | |
383 | } | |
384 | ||
385 | /* last field in 'union bpf_attr' used by this command */ | |
386 | #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key | |
387 | ||
388 | static int map_get_next_key(union bpf_attr *attr) | |
389 | { | |
390 | void __user *ukey = u64_to_ptr(attr->key); | |
391 | void __user *unext_key = u64_to_ptr(attr->next_key); | |
392 | int ufd = attr->map_fd; | |
db20fd2b AS |
393 | struct bpf_map *map; |
394 | void *key, *next_key; | |
592867bf | 395 | struct fd f; |
db20fd2b AS |
396 | int err; |
397 | ||
398 | if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY)) | |
399 | return -EINVAL; | |
400 | ||
592867bf | 401 | f = fdget(ufd); |
c2101297 | 402 | map = __bpf_map_get(f); |
db20fd2b AS |
403 | if (IS_ERR(map)) |
404 | return PTR_ERR(map); | |
405 | ||
406 | err = -ENOMEM; | |
407 | key = kmalloc(map->key_size, GFP_USER); | |
408 | if (!key) | |
409 | goto err_put; | |
410 | ||
411 | err = -EFAULT; | |
412 | if (copy_from_user(key, ukey, map->key_size) != 0) | |
413 | goto free_key; | |
414 | ||
415 | err = -ENOMEM; | |
416 | next_key = kmalloc(map->key_size, GFP_USER); | |
417 | if (!next_key) | |
418 | goto free_key; | |
419 | ||
420 | rcu_read_lock(); | |
421 | err = map->ops->map_get_next_key(map, key, next_key); | |
422 | rcu_read_unlock(); | |
423 | if (err) | |
424 | goto free_next_key; | |
425 | ||
426 | err = -EFAULT; | |
427 | if (copy_to_user(unext_key, next_key, map->key_size) != 0) | |
428 | goto free_next_key; | |
429 | ||
430 | err = 0; | |
431 | ||
432 | free_next_key: | |
433 | kfree(next_key); | |
434 | free_key: | |
435 | kfree(key); | |
436 | err_put: | |
437 | fdput(f); | |
438 | return err; | |
439 | } | |
440 | ||
09756af4 AS |
441 | static LIST_HEAD(bpf_prog_types); |
442 | ||
443 | static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog) | |
444 | { | |
445 | struct bpf_prog_type_list *tl; | |
446 | ||
447 | list_for_each_entry(tl, &bpf_prog_types, list_node) { | |
448 | if (tl->type == type) { | |
449 | prog->aux->ops = tl->ops; | |
24701ece | 450 | prog->type = type; |
09756af4 AS |
451 | return 0; |
452 | } | |
453 | } | |
24701ece | 454 | |
09756af4 AS |
455 | return -EINVAL; |
456 | } | |
457 | ||
458 | void bpf_register_prog_type(struct bpf_prog_type_list *tl) | |
459 | { | |
460 | list_add(&tl->list_node, &bpf_prog_types); | |
461 | } | |
462 | ||
0a542a86 AS |
463 | /* fixup insn->imm field of bpf_call instructions: |
464 | * if (insn->imm == BPF_FUNC_map_lookup_elem) | |
465 | * insn->imm = bpf_map_lookup_elem - __bpf_call_base; | |
466 | * else if (insn->imm == BPF_FUNC_map_update_elem) | |
467 | * insn->imm = bpf_map_update_elem - __bpf_call_base; | |
468 | * else ... | |
469 | * | |
470 | * this function is called after eBPF program passed verification | |
471 | */ | |
472 | static void fixup_bpf_calls(struct bpf_prog *prog) | |
473 | { | |
474 | const struct bpf_func_proto *fn; | |
475 | int i; | |
476 | ||
477 | for (i = 0; i < prog->len; i++) { | |
478 | struct bpf_insn *insn = &prog->insnsi[i]; | |
479 | ||
480 | if (insn->code == (BPF_JMP | BPF_CALL)) { | |
481 | /* we reach here when program has bpf_call instructions | |
482 | * and it passed bpf_check(), means that | |
483 | * ops->get_func_proto must have been supplied, check it | |
484 | */ | |
485 | BUG_ON(!prog->aux->ops->get_func_proto); | |
486 | ||
c46646d0 DB |
487 | if (insn->imm == BPF_FUNC_get_route_realm) |
488 | prog->dst_needed = 1; | |
3ad00405 DB |
489 | if (insn->imm == BPF_FUNC_get_prandom_u32) |
490 | bpf_user_rnd_init_once(); | |
04fd61ab AS |
491 | if (insn->imm == BPF_FUNC_tail_call) { |
492 | /* mark bpf_tail_call as different opcode | |
493 | * to avoid conditional branch in | |
494 | * interpeter for every normal call | |
495 | * and to prevent accidental JITing by | |
496 | * JIT compiler that doesn't support | |
497 | * bpf_tail_call yet | |
498 | */ | |
499 | insn->imm = 0; | |
500 | insn->code |= BPF_X; | |
501 | continue; | |
502 | } | |
503 | ||
0a542a86 AS |
504 | fn = prog->aux->ops->get_func_proto(insn->imm); |
505 | /* all functions that have prototype and verifier allowed | |
506 | * programs to call them, must be real in-kernel functions | |
507 | */ | |
508 | BUG_ON(!fn->func); | |
509 | insn->imm = fn->func - __bpf_call_base; | |
510 | } | |
511 | } | |
512 | } | |
513 | ||
09756af4 AS |
514 | /* drop refcnt on maps used by eBPF program and free auxilary data */ |
515 | static void free_used_maps(struct bpf_prog_aux *aux) | |
516 | { | |
517 | int i; | |
518 | ||
519 | for (i = 0; i < aux->used_map_cnt; i++) | |
520 | bpf_map_put(aux->used_maps[i]); | |
521 | ||
522 | kfree(aux->used_maps); | |
523 | } | |
524 | ||
aaac3ba9 AS |
525 | static int bpf_prog_charge_memlock(struct bpf_prog *prog) |
526 | { | |
527 | struct user_struct *user = get_current_user(); | |
528 | unsigned long memlock_limit; | |
529 | ||
530 | memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
531 | ||
532 | atomic_long_add(prog->pages, &user->locked_vm); | |
533 | if (atomic_long_read(&user->locked_vm) > memlock_limit) { | |
534 | atomic_long_sub(prog->pages, &user->locked_vm); | |
535 | free_uid(user); | |
536 | return -EPERM; | |
537 | } | |
538 | prog->aux->user = user; | |
539 | return 0; | |
540 | } | |
541 | ||
542 | static void bpf_prog_uncharge_memlock(struct bpf_prog *prog) | |
543 | { | |
544 | struct user_struct *user = prog->aux->user; | |
545 | ||
546 | atomic_long_sub(prog->pages, &user->locked_vm); | |
547 | free_uid(user); | |
548 | } | |
549 | ||
e9d8afa9 | 550 | static void __prog_put_common(struct rcu_head *rcu) |
abf2e7d6 AS |
551 | { |
552 | struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu); | |
553 | ||
554 | free_used_maps(aux); | |
aaac3ba9 | 555 | bpf_prog_uncharge_memlock(aux->prog); |
abf2e7d6 AS |
556 | bpf_prog_free(aux->prog); |
557 | } | |
558 | ||
559 | /* version of bpf_prog_put() that is called after a grace period */ | |
560 | void bpf_prog_put_rcu(struct bpf_prog *prog) | |
561 | { | |
e9d8afa9 DB |
562 | if (atomic_dec_and_test(&prog->aux->refcnt)) |
563 | call_rcu(&prog->aux->rcu, __prog_put_common); | |
abf2e7d6 AS |
564 | } |
565 | ||
09756af4 AS |
566 | void bpf_prog_put(struct bpf_prog *prog) |
567 | { | |
e9d8afa9 DB |
568 | if (atomic_dec_and_test(&prog->aux->refcnt)) |
569 | __prog_put_common(&prog->aux->rcu); | |
09756af4 | 570 | } |
e2e9b654 | 571 | EXPORT_SYMBOL_GPL(bpf_prog_put); |
09756af4 AS |
572 | |
573 | static int bpf_prog_release(struct inode *inode, struct file *filp) | |
574 | { | |
575 | struct bpf_prog *prog = filp->private_data; | |
576 | ||
abf2e7d6 | 577 | bpf_prog_put_rcu(prog); |
09756af4 AS |
578 | return 0; |
579 | } | |
580 | ||
581 | static const struct file_operations bpf_prog_fops = { | |
582 | .release = bpf_prog_release, | |
583 | }; | |
584 | ||
b2197755 | 585 | int bpf_prog_new_fd(struct bpf_prog *prog) |
aa79781b DB |
586 | { |
587 | return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog, | |
588 | O_RDWR | O_CLOEXEC); | |
589 | } | |
590 | ||
c2101297 | 591 | static struct bpf_prog *__bpf_prog_get(struct fd f) |
09756af4 | 592 | { |
09756af4 AS |
593 | if (!f.file) |
594 | return ERR_PTR(-EBADF); | |
09756af4 AS |
595 | if (f.file->f_op != &bpf_prog_fops) { |
596 | fdput(f); | |
597 | return ERR_PTR(-EINVAL); | |
598 | } | |
599 | ||
c2101297 | 600 | return f.file->private_data; |
09756af4 AS |
601 | } |
602 | ||
603 | /* called by sockets/tracing/seccomp before attaching program to an event | |
604 | * pairs with bpf_prog_put() | |
605 | */ | |
606 | struct bpf_prog *bpf_prog_get(u32 ufd) | |
607 | { | |
608 | struct fd f = fdget(ufd); | |
609 | struct bpf_prog *prog; | |
610 | ||
c2101297 | 611 | prog = __bpf_prog_get(f); |
09756af4 AS |
612 | if (IS_ERR(prog)) |
613 | return prog; | |
614 | ||
615 | atomic_inc(&prog->aux->refcnt); | |
616 | fdput(f); | |
c2101297 | 617 | |
09756af4 AS |
618 | return prog; |
619 | } | |
e2e9b654 | 620 | EXPORT_SYMBOL_GPL(bpf_prog_get); |
09756af4 AS |
621 | |
622 | /* last field in 'union bpf_attr' used by this command */ | |
2541517c | 623 | #define BPF_PROG_LOAD_LAST_FIELD kern_version |
09756af4 AS |
624 | |
625 | static int bpf_prog_load(union bpf_attr *attr) | |
626 | { | |
627 | enum bpf_prog_type type = attr->prog_type; | |
628 | struct bpf_prog *prog; | |
629 | int err; | |
630 | char license[128]; | |
631 | bool is_gpl; | |
632 | ||
633 | if (CHECK_ATTR(BPF_PROG_LOAD)) | |
634 | return -EINVAL; | |
635 | ||
636 | /* copy eBPF program license from user space */ | |
637 | if (strncpy_from_user(license, u64_to_ptr(attr->license), | |
638 | sizeof(license) - 1) < 0) | |
639 | return -EFAULT; | |
640 | license[sizeof(license) - 1] = 0; | |
641 | ||
642 | /* eBPF programs must be GPL compatible to use GPL-ed functions */ | |
643 | is_gpl = license_is_gpl_compatible(license); | |
644 | ||
645 | if (attr->insn_cnt >= BPF_MAXINSNS) | |
646 | return -EINVAL; | |
647 | ||
2541517c AS |
648 | if (type == BPF_PROG_TYPE_KPROBE && |
649 | attr->kern_version != LINUX_VERSION_CODE) | |
650 | return -EINVAL; | |
651 | ||
1be7f75d AS |
652 | if (type != BPF_PROG_TYPE_SOCKET_FILTER && !capable(CAP_SYS_ADMIN)) |
653 | return -EPERM; | |
654 | ||
09756af4 AS |
655 | /* plain bpf_prog allocation */ |
656 | prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER); | |
657 | if (!prog) | |
658 | return -ENOMEM; | |
659 | ||
aaac3ba9 AS |
660 | err = bpf_prog_charge_memlock(prog); |
661 | if (err) | |
662 | goto free_prog_nouncharge; | |
663 | ||
09756af4 AS |
664 | prog->len = attr->insn_cnt; |
665 | ||
666 | err = -EFAULT; | |
667 | if (copy_from_user(prog->insns, u64_to_ptr(attr->insns), | |
668 | prog->len * sizeof(struct bpf_insn)) != 0) | |
669 | goto free_prog; | |
670 | ||
671 | prog->orig_prog = NULL; | |
a91263d5 | 672 | prog->jited = 0; |
09756af4 AS |
673 | |
674 | atomic_set(&prog->aux->refcnt, 1); | |
a91263d5 | 675 | prog->gpl_compatible = is_gpl ? 1 : 0; |
09756af4 AS |
676 | |
677 | /* find program type: socket_filter vs tracing_filter */ | |
678 | err = find_prog_type(type, prog); | |
679 | if (err < 0) | |
680 | goto free_prog; | |
681 | ||
682 | /* run eBPF verifier */ | |
9bac3d6d | 683 | err = bpf_check(&prog, attr); |
09756af4 AS |
684 | if (err < 0) |
685 | goto free_used_maps; | |
686 | ||
0a542a86 AS |
687 | /* fixup BPF_CALL->imm field */ |
688 | fixup_bpf_calls(prog); | |
689 | ||
09756af4 | 690 | /* eBPF program is ready to be JITed */ |
04fd61ab AS |
691 | err = bpf_prog_select_runtime(prog); |
692 | if (err < 0) | |
693 | goto free_used_maps; | |
09756af4 | 694 | |
aa79781b | 695 | err = bpf_prog_new_fd(prog); |
09756af4 AS |
696 | if (err < 0) |
697 | /* failed to allocate fd */ | |
698 | goto free_used_maps; | |
699 | ||
700 | return err; | |
701 | ||
702 | free_used_maps: | |
703 | free_used_maps(prog->aux); | |
704 | free_prog: | |
aaac3ba9 AS |
705 | bpf_prog_uncharge_memlock(prog); |
706 | free_prog_nouncharge: | |
09756af4 AS |
707 | bpf_prog_free(prog); |
708 | return err; | |
709 | } | |
710 | ||
b2197755 DB |
711 | #define BPF_OBJ_LAST_FIELD bpf_fd |
712 | ||
713 | static int bpf_obj_pin(const union bpf_attr *attr) | |
714 | { | |
715 | if (CHECK_ATTR(BPF_OBJ)) | |
716 | return -EINVAL; | |
717 | ||
718 | return bpf_obj_pin_user(attr->bpf_fd, u64_to_ptr(attr->pathname)); | |
719 | } | |
720 | ||
721 | static int bpf_obj_get(const union bpf_attr *attr) | |
722 | { | |
723 | if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0) | |
724 | return -EINVAL; | |
725 | ||
726 | return bpf_obj_get_user(u64_to_ptr(attr->pathname)); | |
727 | } | |
728 | ||
99c55f7d AS |
729 | SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size) |
730 | { | |
731 | union bpf_attr attr = {}; | |
732 | int err; | |
733 | ||
1be7f75d | 734 | if (!capable(CAP_SYS_ADMIN) && sysctl_unprivileged_bpf_disabled) |
99c55f7d AS |
735 | return -EPERM; |
736 | ||
737 | if (!access_ok(VERIFY_READ, uattr, 1)) | |
738 | return -EFAULT; | |
739 | ||
740 | if (size > PAGE_SIZE) /* silly large */ | |
741 | return -E2BIG; | |
742 | ||
743 | /* If we're handed a bigger struct than we know of, | |
744 | * ensure all the unknown bits are 0 - i.e. new | |
745 | * user-space does not rely on any kernel feature | |
746 | * extensions we dont know about yet. | |
747 | */ | |
748 | if (size > sizeof(attr)) { | |
749 | unsigned char __user *addr; | |
750 | unsigned char __user *end; | |
751 | unsigned char val; | |
752 | ||
753 | addr = (void __user *)uattr + sizeof(attr); | |
754 | end = (void __user *)uattr + size; | |
755 | ||
756 | for (; addr < end; addr++) { | |
757 | err = get_user(val, addr); | |
758 | if (err) | |
759 | return err; | |
760 | if (val) | |
761 | return -E2BIG; | |
762 | } | |
763 | size = sizeof(attr); | |
764 | } | |
765 | ||
766 | /* copy attributes from user space, may be less than sizeof(bpf_attr) */ | |
767 | if (copy_from_user(&attr, uattr, size) != 0) | |
768 | return -EFAULT; | |
769 | ||
770 | switch (cmd) { | |
771 | case BPF_MAP_CREATE: | |
772 | err = map_create(&attr); | |
773 | break; | |
db20fd2b AS |
774 | case BPF_MAP_LOOKUP_ELEM: |
775 | err = map_lookup_elem(&attr); | |
776 | break; | |
777 | case BPF_MAP_UPDATE_ELEM: | |
778 | err = map_update_elem(&attr); | |
779 | break; | |
780 | case BPF_MAP_DELETE_ELEM: | |
781 | err = map_delete_elem(&attr); | |
782 | break; | |
783 | case BPF_MAP_GET_NEXT_KEY: | |
784 | err = map_get_next_key(&attr); | |
785 | break; | |
09756af4 AS |
786 | case BPF_PROG_LOAD: |
787 | err = bpf_prog_load(&attr); | |
788 | break; | |
b2197755 DB |
789 | case BPF_OBJ_PIN: |
790 | err = bpf_obj_pin(&attr); | |
791 | break; | |
792 | case BPF_OBJ_GET: | |
793 | err = bpf_obj_get(&attr); | |
794 | break; | |
99c55f7d AS |
795 | default: |
796 | err = -EINVAL; | |
797 | break; | |
798 | } | |
799 | ||
800 | return err; | |
801 | } |