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 | |
b121d1e7 AS |
21 | DEFINE_PER_CPU(int, bpf_prog_active); |
22 | ||
1be7f75d AS |
23 | int sysctl_unprivileged_bpf_disabled __read_mostly; |
24 | ||
99c55f7d AS |
25 | static LIST_HEAD(bpf_map_types); |
26 | ||
27 | static struct bpf_map *find_and_alloc_map(union bpf_attr *attr) | |
28 | { | |
29 | struct bpf_map_type_list *tl; | |
30 | struct bpf_map *map; | |
31 | ||
32 | list_for_each_entry(tl, &bpf_map_types, list_node) { | |
33 | if (tl->type == attr->map_type) { | |
34 | map = tl->ops->map_alloc(attr); | |
35 | if (IS_ERR(map)) | |
36 | return map; | |
37 | map->ops = tl->ops; | |
38 | map->map_type = attr->map_type; | |
39 | return map; | |
40 | } | |
41 | } | |
42 | return ERR_PTR(-EINVAL); | |
43 | } | |
44 | ||
45 | /* boot time registration of different map implementations */ | |
46 | void bpf_register_map_type(struct bpf_map_type_list *tl) | |
47 | { | |
48 | list_add(&tl->list_node, &bpf_map_types); | |
49 | } | |
50 | ||
6c905981 AS |
51 | int bpf_map_precharge_memlock(u32 pages) |
52 | { | |
53 | struct user_struct *user = get_current_user(); | |
54 | unsigned long memlock_limit, cur; | |
55 | ||
56 | memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
57 | cur = atomic_long_read(&user->locked_vm); | |
58 | free_uid(user); | |
59 | if (cur + pages > memlock_limit) | |
60 | return -EPERM; | |
61 | return 0; | |
62 | } | |
63 | ||
aaac3ba9 AS |
64 | static int bpf_map_charge_memlock(struct bpf_map *map) |
65 | { | |
66 | struct user_struct *user = get_current_user(); | |
67 | unsigned long memlock_limit; | |
68 | ||
69 | memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
70 | ||
71 | atomic_long_add(map->pages, &user->locked_vm); | |
72 | ||
73 | if (atomic_long_read(&user->locked_vm) > memlock_limit) { | |
74 | atomic_long_sub(map->pages, &user->locked_vm); | |
75 | free_uid(user); | |
76 | return -EPERM; | |
77 | } | |
78 | map->user = user; | |
79 | return 0; | |
80 | } | |
81 | ||
82 | static void bpf_map_uncharge_memlock(struct bpf_map *map) | |
83 | { | |
84 | struct user_struct *user = map->user; | |
85 | ||
86 | atomic_long_sub(map->pages, &user->locked_vm); | |
87 | free_uid(user); | |
88 | } | |
89 | ||
99c55f7d AS |
90 | /* called from workqueue */ |
91 | static void bpf_map_free_deferred(struct work_struct *work) | |
92 | { | |
93 | struct bpf_map *map = container_of(work, struct bpf_map, work); | |
94 | ||
aaac3ba9 | 95 | bpf_map_uncharge_memlock(map); |
99c55f7d AS |
96 | /* implementation dependent freeing */ |
97 | map->ops->map_free(map); | |
98 | } | |
99 | ||
c9da161c DB |
100 | static void bpf_map_put_uref(struct bpf_map *map) |
101 | { | |
102 | if (atomic_dec_and_test(&map->usercnt)) { | |
103 | if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) | |
104 | bpf_fd_array_map_clear(map); | |
105 | } | |
106 | } | |
107 | ||
99c55f7d AS |
108 | /* decrement map refcnt and schedule it for freeing via workqueue |
109 | * (unrelying map implementation ops->map_free() might sleep) | |
110 | */ | |
111 | void bpf_map_put(struct bpf_map *map) | |
112 | { | |
113 | if (atomic_dec_and_test(&map->refcnt)) { | |
114 | INIT_WORK(&map->work, bpf_map_free_deferred); | |
115 | schedule_work(&map->work); | |
116 | } | |
117 | } | |
118 | ||
c9da161c | 119 | void bpf_map_put_with_uref(struct bpf_map *map) |
99c55f7d | 120 | { |
c9da161c | 121 | bpf_map_put_uref(map); |
99c55f7d | 122 | bpf_map_put(map); |
c9da161c DB |
123 | } |
124 | ||
125 | static int bpf_map_release(struct inode *inode, struct file *filp) | |
126 | { | |
127 | bpf_map_put_with_uref(filp->private_data); | |
99c55f7d AS |
128 | return 0; |
129 | } | |
130 | ||
f99bf205 DB |
131 | #ifdef CONFIG_PROC_FS |
132 | static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp) | |
133 | { | |
134 | const struct bpf_map *map = filp->private_data; | |
135 | ||
136 | seq_printf(m, | |
137 | "map_type:\t%u\n" | |
138 | "key_size:\t%u\n" | |
139 | "value_size:\t%u\n" | |
140 | "max_entries:\t%u\n", | |
141 | map->map_type, | |
142 | map->key_size, | |
143 | map->value_size, | |
144 | map->max_entries); | |
145 | } | |
146 | #endif | |
147 | ||
99c55f7d | 148 | static const struct file_operations bpf_map_fops = { |
f99bf205 DB |
149 | #ifdef CONFIG_PROC_FS |
150 | .show_fdinfo = bpf_map_show_fdinfo, | |
151 | #endif | |
152 | .release = bpf_map_release, | |
99c55f7d AS |
153 | }; |
154 | ||
b2197755 | 155 | int bpf_map_new_fd(struct bpf_map *map) |
aa79781b DB |
156 | { |
157 | return anon_inode_getfd("bpf-map", &bpf_map_fops, map, | |
158 | O_RDWR | O_CLOEXEC); | |
159 | } | |
160 | ||
99c55f7d AS |
161 | /* helper macro to check that unused fields 'union bpf_attr' are zero */ |
162 | #define CHECK_ATTR(CMD) \ | |
163 | memchr_inv((void *) &attr->CMD##_LAST_FIELD + \ | |
164 | sizeof(attr->CMD##_LAST_FIELD), 0, \ | |
165 | sizeof(*attr) - \ | |
166 | offsetof(union bpf_attr, CMD##_LAST_FIELD) - \ | |
167 | sizeof(attr->CMD##_LAST_FIELD)) != NULL | |
168 | ||
6c905981 | 169 | #define BPF_MAP_CREATE_LAST_FIELD map_flags |
99c55f7d AS |
170 | /* called via syscall */ |
171 | static int map_create(union bpf_attr *attr) | |
172 | { | |
173 | struct bpf_map *map; | |
174 | int err; | |
175 | ||
176 | err = CHECK_ATTR(BPF_MAP_CREATE); | |
177 | if (err) | |
178 | return -EINVAL; | |
179 | ||
180 | /* find map type and init map: hashtable vs rbtree vs bloom vs ... */ | |
181 | map = find_and_alloc_map(attr); | |
182 | if (IS_ERR(map)) | |
183 | return PTR_ERR(map); | |
184 | ||
185 | atomic_set(&map->refcnt, 1); | |
c9da161c | 186 | atomic_set(&map->usercnt, 1); |
99c55f7d | 187 | |
aaac3ba9 AS |
188 | err = bpf_map_charge_memlock(map); |
189 | if (err) | |
190 | goto free_map; | |
191 | ||
aa79781b | 192 | err = bpf_map_new_fd(map); |
99c55f7d AS |
193 | if (err < 0) |
194 | /* failed to allocate fd */ | |
195 | goto free_map; | |
196 | ||
197 | return err; | |
198 | ||
199 | free_map: | |
200 | map->ops->map_free(map); | |
201 | return err; | |
202 | } | |
203 | ||
db20fd2b AS |
204 | /* if error is returned, fd is released. |
205 | * On success caller should complete fd access with matching fdput() | |
206 | */ | |
c2101297 | 207 | struct bpf_map *__bpf_map_get(struct fd f) |
db20fd2b | 208 | { |
db20fd2b AS |
209 | if (!f.file) |
210 | return ERR_PTR(-EBADF); | |
db20fd2b AS |
211 | if (f.file->f_op != &bpf_map_fops) { |
212 | fdput(f); | |
213 | return ERR_PTR(-EINVAL); | |
214 | } | |
215 | ||
c2101297 DB |
216 | return f.file->private_data; |
217 | } | |
218 | ||
c9da161c DB |
219 | void bpf_map_inc(struct bpf_map *map, bool uref) |
220 | { | |
221 | atomic_inc(&map->refcnt); | |
222 | if (uref) | |
223 | atomic_inc(&map->usercnt); | |
224 | } | |
225 | ||
226 | struct bpf_map *bpf_map_get_with_uref(u32 ufd) | |
c2101297 DB |
227 | { |
228 | struct fd f = fdget(ufd); | |
229 | struct bpf_map *map; | |
230 | ||
231 | map = __bpf_map_get(f); | |
232 | if (IS_ERR(map)) | |
233 | return map; | |
234 | ||
c9da161c | 235 | bpf_map_inc(map, true); |
c2101297 | 236 | fdput(f); |
db20fd2b AS |
237 | |
238 | return map; | |
239 | } | |
240 | ||
241 | /* helper to convert user pointers passed inside __aligned_u64 fields */ | |
242 | static void __user *u64_to_ptr(__u64 val) | |
243 | { | |
244 | return (void __user *) (unsigned long) val; | |
245 | } | |
246 | ||
247 | /* last field in 'union bpf_attr' used by this command */ | |
248 | #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value | |
249 | ||
250 | static int map_lookup_elem(union bpf_attr *attr) | |
251 | { | |
252 | void __user *ukey = u64_to_ptr(attr->key); | |
253 | void __user *uvalue = u64_to_ptr(attr->value); | |
254 | int ufd = attr->map_fd; | |
db20fd2b | 255 | struct bpf_map *map; |
8ebe667c | 256 | void *key, *value, *ptr; |
15a07b33 | 257 | u32 value_size; |
592867bf | 258 | struct fd f; |
db20fd2b AS |
259 | int err; |
260 | ||
261 | if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM)) | |
262 | return -EINVAL; | |
263 | ||
592867bf | 264 | f = fdget(ufd); |
c2101297 | 265 | map = __bpf_map_get(f); |
db20fd2b AS |
266 | if (IS_ERR(map)) |
267 | return PTR_ERR(map); | |
268 | ||
269 | err = -ENOMEM; | |
270 | key = kmalloc(map->key_size, GFP_USER); | |
271 | if (!key) | |
272 | goto err_put; | |
273 | ||
274 | err = -EFAULT; | |
275 | if (copy_from_user(key, ukey, map->key_size) != 0) | |
276 | goto free_key; | |
277 | ||
15a07b33 AS |
278 | if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || |
279 | map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) | |
280 | value_size = round_up(map->value_size, 8) * num_possible_cpus(); | |
281 | else | |
282 | value_size = map->value_size; | |
283 | ||
8ebe667c | 284 | err = -ENOMEM; |
15a07b33 | 285 | value = kmalloc(value_size, GFP_USER | __GFP_NOWARN); |
db20fd2b | 286 | if (!value) |
8ebe667c AS |
287 | goto free_key; |
288 | ||
15a07b33 AS |
289 | if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH) { |
290 | err = bpf_percpu_hash_copy(map, key, value); | |
291 | } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) { | |
292 | err = bpf_percpu_array_copy(map, key, value); | |
557c0c6e AS |
293 | } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) { |
294 | err = bpf_stackmap_copy(map, key, value); | |
15a07b33 AS |
295 | } else { |
296 | rcu_read_lock(); | |
297 | ptr = map->ops->map_lookup_elem(map, key); | |
298 | if (ptr) | |
299 | memcpy(value, ptr, value_size); | |
300 | rcu_read_unlock(); | |
301 | err = ptr ? 0 : -ENOENT; | |
302 | } | |
8ebe667c | 303 | |
15a07b33 | 304 | if (err) |
8ebe667c | 305 | goto free_value; |
db20fd2b AS |
306 | |
307 | err = -EFAULT; | |
15a07b33 | 308 | if (copy_to_user(uvalue, value, value_size) != 0) |
8ebe667c | 309 | goto free_value; |
db20fd2b AS |
310 | |
311 | err = 0; | |
312 | ||
8ebe667c AS |
313 | free_value: |
314 | kfree(value); | |
db20fd2b AS |
315 | free_key: |
316 | kfree(key); | |
317 | err_put: | |
318 | fdput(f); | |
319 | return err; | |
320 | } | |
321 | ||
3274f520 | 322 | #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags |
db20fd2b AS |
323 | |
324 | static int map_update_elem(union bpf_attr *attr) | |
325 | { | |
326 | void __user *ukey = u64_to_ptr(attr->key); | |
327 | void __user *uvalue = u64_to_ptr(attr->value); | |
328 | int ufd = attr->map_fd; | |
db20fd2b AS |
329 | struct bpf_map *map; |
330 | void *key, *value; | |
15a07b33 | 331 | u32 value_size; |
592867bf | 332 | struct fd f; |
db20fd2b AS |
333 | int err; |
334 | ||
335 | if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM)) | |
336 | return -EINVAL; | |
337 | ||
592867bf | 338 | f = fdget(ufd); |
c2101297 | 339 | map = __bpf_map_get(f); |
db20fd2b AS |
340 | if (IS_ERR(map)) |
341 | return PTR_ERR(map); | |
342 | ||
343 | err = -ENOMEM; | |
344 | key = kmalloc(map->key_size, GFP_USER); | |
345 | if (!key) | |
346 | goto err_put; | |
347 | ||
348 | err = -EFAULT; | |
349 | if (copy_from_user(key, ukey, map->key_size) != 0) | |
350 | goto free_key; | |
351 | ||
15a07b33 AS |
352 | if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || |
353 | map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) | |
354 | value_size = round_up(map->value_size, 8) * num_possible_cpus(); | |
355 | else | |
356 | value_size = map->value_size; | |
357 | ||
db20fd2b | 358 | err = -ENOMEM; |
15a07b33 | 359 | value = kmalloc(value_size, GFP_USER | __GFP_NOWARN); |
db20fd2b AS |
360 | if (!value) |
361 | goto free_key; | |
362 | ||
363 | err = -EFAULT; | |
15a07b33 | 364 | if (copy_from_user(value, uvalue, value_size) != 0) |
db20fd2b AS |
365 | goto free_value; |
366 | ||
b121d1e7 AS |
367 | /* must increment bpf_prog_active to avoid kprobe+bpf triggering from |
368 | * inside bpf map update or delete otherwise deadlocks are possible | |
369 | */ | |
370 | preempt_disable(); | |
371 | __this_cpu_inc(bpf_prog_active); | |
15a07b33 AS |
372 | if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH) { |
373 | err = bpf_percpu_hash_update(map, key, value, attr->flags); | |
374 | } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) { | |
375 | err = bpf_percpu_array_update(map, key, value, attr->flags); | |
376 | } else { | |
377 | rcu_read_lock(); | |
378 | err = map->ops->map_update_elem(map, key, value, attr->flags); | |
379 | rcu_read_unlock(); | |
380 | } | |
b121d1e7 AS |
381 | __this_cpu_dec(bpf_prog_active); |
382 | preempt_enable(); | |
db20fd2b AS |
383 | |
384 | free_value: | |
385 | kfree(value); | |
386 | free_key: | |
387 | kfree(key); | |
388 | err_put: | |
389 | fdput(f); | |
390 | return err; | |
391 | } | |
392 | ||
393 | #define BPF_MAP_DELETE_ELEM_LAST_FIELD key | |
394 | ||
395 | static int map_delete_elem(union bpf_attr *attr) | |
396 | { | |
397 | void __user *ukey = u64_to_ptr(attr->key); | |
398 | int ufd = attr->map_fd; | |
db20fd2b | 399 | struct bpf_map *map; |
592867bf | 400 | struct fd f; |
db20fd2b AS |
401 | void *key; |
402 | int err; | |
403 | ||
404 | if (CHECK_ATTR(BPF_MAP_DELETE_ELEM)) | |
405 | return -EINVAL; | |
406 | ||
592867bf | 407 | f = fdget(ufd); |
c2101297 | 408 | map = __bpf_map_get(f); |
db20fd2b AS |
409 | if (IS_ERR(map)) |
410 | return PTR_ERR(map); | |
411 | ||
412 | err = -ENOMEM; | |
413 | key = kmalloc(map->key_size, GFP_USER); | |
414 | if (!key) | |
415 | goto err_put; | |
416 | ||
417 | err = -EFAULT; | |
418 | if (copy_from_user(key, ukey, map->key_size) != 0) | |
419 | goto free_key; | |
420 | ||
b121d1e7 AS |
421 | preempt_disable(); |
422 | __this_cpu_inc(bpf_prog_active); | |
db20fd2b AS |
423 | rcu_read_lock(); |
424 | err = map->ops->map_delete_elem(map, key); | |
425 | rcu_read_unlock(); | |
b121d1e7 AS |
426 | __this_cpu_dec(bpf_prog_active); |
427 | preempt_enable(); | |
db20fd2b AS |
428 | |
429 | free_key: | |
430 | kfree(key); | |
431 | err_put: | |
432 | fdput(f); | |
433 | return err; | |
434 | } | |
435 | ||
436 | /* last field in 'union bpf_attr' used by this command */ | |
437 | #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key | |
438 | ||
439 | static int map_get_next_key(union bpf_attr *attr) | |
440 | { | |
441 | void __user *ukey = u64_to_ptr(attr->key); | |
442 | void __user *unext_key = u64_to_ptr(attr->next_key); | |
443 | int ufd = attr->map_fd; | |
db20fd2b AS |
444 | struct bpf_map *map; |
445 | void *key, *next_key; | |
592867bf | 446 | struct fd f; |
db20fd2b AS |
447 | int err; |
448 | ||
449 | if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY)) | |
450 | return -EINVAL; | |
451 | ||
592867bf | 452 | f = fdget(ufd); |
c2101297 | 453 | map = __bpf_map_get(f); |
db20fd2b AS |
454 | if (IS_ERR(map)) |
455 | return PTR_ERR(map); | |
456 | ||
457 | err = -ENOMEM; | |
458 | key = kmalloc(map->key_size, GFP_USER); | |
459 | if (!key) | |
460 | goto err_put; | |
461 | ||
462 | err = -EFAULT; | |
463 | if (copy_from_user(key, ukey, map->key_size) != 0) | |
464 | goto free_key; | |
465 | ||
466 | err = -ENOMEM; | |
467 | next_key = kmalloc(map->key_size, GFP_USER); | |
468 | if (!next_key) | |
469 | goto free_key; | |
470 | ||
471 | rcu_read_lock(); | |
472 | err = map->ops->map_get_next_key(map, key, next_key); | |
473 | rcu_read_unlock(); | |
474 | if (err) | |
475 | goto free_next_key; | |
476 | ||
477 | err = -EFAULT; | |
478 | if (copy_to_user(unext_key, next_key, map->key_size) != 0) | |
479 | goto free_next_key; | |
480 | ||
481 | err = 0; | |
482 | ||
483 | free_next_key: | |
484 | kfree(next_key); | |
485 | free_key: | |
486 | kfree(key); | |
487 | err_put: | |
488 | fdput(f); | |
489 | return err; | |
490 | } | |
491 | ||
09756af4 AS |
492 | static LIST_HEAD(bpf_prog_types); |
493 | ||
494 | static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog) | |
495 | { | |
496 | struct bpf_prog_type_list *tl; | |
497 | ||
498 | list_for_each_entry(tl, &bpf_prog_types, list_node) { | |
499 | if (tl->type == type) { | |
500 | prog->aux->ops = tl->ops; | |
24701ece | 501 | prog->type = type; |
09756af4 AS |
502 | return 0; |
503 | } | |
504 | } | |
24701ece | 505 | |
09756af4 AS |
506 | return -EINVAL; |
507 | } | |
508 | ||
509 | void bpf_register_prog_type(struct bpf_prog_type_list *tl) | |
510 | { | |
511 | list_add(&tl->list_node, &bpf_prog_types); | |
512 | } | |
513 | ||
0a542a86 AS |
514 | /* fixup insn->imm field of bpf_call instructions: |
515 | * if (insn->imm == BPF_FUNC_map_lookup_elem) | |
516 | * insn->imm = bpf_map_lookup_elem - __bpf_call_base; | |
517 | * else if (insn->imm == BPF_FUNC_map_update_elem) | |
518 | * insn->imm = bpf_map_update_elem - __bpf_call_base; | |
519 | * else ... | |
520 | * | |
521 | * this function is called after eBPF program passed verification | |
522 | */ | |
523 | static void fixup_bpf_calls(struct bpf_prog *prog) | |
524 | { | |
525 | const struct bpf_func_proto *fn; | |
526 | int i; | |
527 | ||
528 | for (i = 0; i < prog->len; i++) { | |
529 | struct bpf_insn *insn = &prog->insnsi[i]; | |
530 | ||
531 | if (insn->code == (BPF_JMP | BPF_CALL)) { | |
532 | /* we reach here when program has bpf_call instructions | |
533 | * and it passed bpf_check(), means that | |
534 | * ops->get_func_proto must have been supplied, check it | |
535 | */ | |
536 | BUG_ON(!prog->aux->ops->get_func_proto); | |
537 | ||
c46646d0 DB |
538 | if (insn->imm == BPF_FUNC_get_route_realm) |
539 | prog->dst_needed = 1; | |
3ad00405 DB |
540 | if (insn->imm == BPF_FUNC_get_prandom_u32) |
541 | bpf_user_rnd_init_once(); | |
04fd61ab AS |
542 | if (insn->imm == BPF_FUNC_tail_call) { |
543 | /* mark bpf_tail_call as different opcode | |
544 | * to avoid conditional branch in | |
545 | * interpeter for every normal call | |
546 | * and to prevent accidental JITing by | |
547 | * JIT compiler that doesn't support | |
548 | * bpf_tail_call yet | |
549 | */ | |
550 | insn->imm = 0; | |
551 | insn->code |= BPF_X; | |
552 | continue; | |
553 | } | |
554 | ||
0a542a86 AS |
555 | fn = prog->aux->ops->get_func_proto(insn->imm); |
556 | /* all functions that have prototype and verifier allowed | |
557 | * programs to call them, must be real in-kernel functions | |
558 | */ | |
559 | BUG_ON(!fn->func); | |
560 | insn->imm = fn->func - __bpf_call_base; | |
561 | } | |
562 | } | |
563 | } | |
564 | ||
09756af4 AS |
565 | /* drop refcnt on maps used by eBPF program and free auxilary data */ |
566 | static void free_used_maps(struct bpf_prog_aux *aux) | |
567 | { | |
568 | int i; | |
569 | ||
570 | for (i = 0; i < aux->used_map_cnt; i++) | |
571 | bpf_map_put(aux->used_maps[i]); | |
572 | ||
573 | kfree(aux->used_maps); | |
574 | } | |
575 | ||
aaac3ba9 AS |
576 | static int bpf_prog_charge_memlock(struct bpf_prog *prog) |
577 | { | |
578 | struct user_struct *user = get_current_user(); | |
579 | unsigned long memlock_limit; | |
580 | ||
581 | memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
582 | ||
583 | atomic_long_add(prog->pages, &user->locked_vm); | |
584 | if (atomic_long_read(&user->locked_vm) > memlock_limit) { | |
585 | atomic_long_sub(prog->pages, &user->locked_vm); | |
586 | free_uid(user); | |
587 | return -EPERM; | |
588 | } | |
589 | prog->aux->user = user; | |
590 | return 0; | |
591 | } | |
592 | ||
593 | static void bpf_prog_uncharge_memlock(struct bpf_prog *prog) | |
594 | { | |
595 | struct user_struct *user = prog->aux->user; | |
596 | ||
597 | atomic_long_sub(prog->pages, &user->locked_vm); | |
598 | free_uid(user); | |
599 | } | |
600 | ||
e9d8afa9 | 601 | static void __prog_put_common(struct rcu_head *rcu) |
abf2e7d6 AS |
602 | { |
603 | struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu); | |
604 | ||
605 | free_used_maps(aux); | |
aaac3ba9 | 606 | bpf_prog_uncharge_memlock(aux->prog); |
abf2e7d6 AS |
607 | bpf_prog_free(aux->prog); |
608 | } | |
609 | ||
610 | /* version of bpf_prog_put() that is called after a grace period */ | |
611 | void bpf_prog_put_rcu(struct bpf_prog *prog) | |
612 | { | |
e9d8afa9 DB |
613 | if (atomic_dec_and_test(&prog->aux->refcnt)) |
614 | call_rcu(&prog->aux->rcu, __prog_put_common); | |
abf2e7d6 AS |
615 | } |
616 | ||
09756af4 AS |
617 | void bpf_prog_put(struct bpf_prog *prog) |
618 | { | |
e9d8afa9 DB |
619 | if (atomic_dec_and_test(&prog->aux->refcnt)) |
620 | __prog_put_common(&prog->aux->rcu); | |
09756af4 | 621 | } |
e2e9b654 | 622 | EXPORT_SYMBOL_GPL(bpf_prog_put); |
09756af4 AS |
623 | |
624 | static int bpf_prog_release(struct inode *inode, struct file *filp) | |
625 | { | |
626 | struct bpf_prog *prog = filp->private_data; | |
627 | ||
abf2e7d6 | 628 | bpf_prog_put_rcu(prog); |
09756af4 AS |
629 | return 0; |
630 | } | |
631 | ||
632 | static const struct file_operations bpf_prog_fops = { | |
633 | .release = bpf_prog_release, | |
634 | }; | |
635 | ||
b2197755 | 636 | int bpf_prog_new_fd(struct bpf_prog *prog) |
aa79781b DB |
637 | { |
638 | return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog, | |
639 | O_RDWR | O_CLOEXEC); | |
640 | } | |
641 | ||
c2101297 | 642 | static struct bpf_prog *__bpf_prog_get(struct fd f) |
09756af4 | 643 | { |
09756af4 AS |
644 | if (!f.file) |
645 | return ERR_PTR(-EBADF); | |
09756af4 AS |
646 | if (f.file->f_op != &bpf_prog_fops) { |
647 | fdput(f); | |
648 | return ERR_PTR(-EINVAL); | |
649 | } | |
650 | ||
c2101297 | 651 | return f.file->private_data; |
09756af4 AS |
652 | } |
653 | ||
654 | /* called by sockets/tracing/seccomp before attaching program to an event | |
655 | * pairs with bpf_prog_put() | |
656 | */ | |
657 | struct bpf_prog *bpf_prog_get(u32 ufd) | |
658 | { | |
659 | struct fd f = fdget(ufd); | |
660 | struct bpf_prog *prog; | |
661 | ||
c2101297 | 662 | prog = __bpf_prog_get(f); |
09756af4 AS |
663 | if (IS_ERR(prog)) |
664 | return prog; | |
665 | ||
666 | atomic_inc(&prog->aux->refcnt); | |
667 | fdput(f); | |
c2101297 | 668 | |
09756af4 AS |
669 | return prog; |
670 | } | |
e2e9b654 | 671 | EXPORT_SYMBOL_GPL(bpf_prog_get); |
09756af4 AS |
672 | |
673 | /* last field in 'union bpf_attr' used by this command */ | |
2541517c | 674 | #define BPF_PROG_LOAD_LAST_FIELD kern_version |
09756af4 AS |
675 | |
676 | static int bpf_prog_load(union bpf_attr *attr) | |
677 | { | |
678 | enum bpf_prog_type type = attr->prog_type; | |
679 | struct bpf_prog *prog; | |
680 | int err; | |
681 | char license[128]; | |
682 | bool is_gpl; | |
683 | ||
684 | if (CHECK_ATTR(BPF_PROG_LOAD)) | |
685 | return -EINVAL; | |
686 | ||
687 | /* copy eBPF program license from user space */ | |
688 | if (strncpy_from_user(license, u64_to_ptr(attr->license), | |
689 | sizeof(license) - 1) < 0) | |
690 | return -EFAULT; | |
691 | license[sizeof(license) - 1] = 0; | |
692 | ||
693 | /* eBPF programs must be GPL compatible to use GPL-ed functions */ | |
694 | is_gpl = license_is_gpl_compatible(license); | |
695 | ||
696 | if (attr->insn_cnt >= BPF_MAXINSNS) | |
697 | return -EINVAL; | |
698 | ||
2541517c AS |
699 | if (type == BPF_PROG_TYPE_KPROBE && |
700 | attr->kern_version != LINUX_VERSION_CODE) | |
701 | return -EINVAL; | |
702 | ||
1be7f75d AS |
703 | if (type != BPF_PROG_TYPE_SOCKET_FILTER && !capable(CAP_SYS_ADMIN)) |
704 | return -EPERM; | |
705 | ||
09756af4 AS |
706 | /* plain bpf_prog allocation */ |
707 | prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER); | |
708 | if (!prog) | |
709 | return -ENOMEM; | |
710 | ||
aaac3ba9 AS |
711 | err = bpf_prog_charge_memlock(prog); |
712 | if (err) | |
713 | goto free_prog_nouncharge; | |
714 | ||
09756af4 AS |
715 | prog->len = attr->insn_cnt; |
716 | ||
717 | err = -EFAULT; | |
718 | if (copy_from_user(prog->insns, u64_to_ptr(attr->insns), | |
719 | prog->len * sizeof(struct bpf_insn)) != 0) | |
720 | goto free_prog; | |
721 | ||
722 | prog->orig_prog = NULL; | |
a91263d5 | 723 | prog->jited = 0; |
09756af4 AS |
724 | |
725 | atomic_set(&prog->aux->refcnt, 1); | |
a91263d5 | 726 | prog->gpl_compatible = is_gpl ? 1 : 0; |
09756af4 AS |
727 | |
728 | /* find program type: socket_filter vs tracing_filter */ | |
729 | err = find_prog_type(type, prog); | |
730 | if (err < 0) | |
731 | goto free_prog; | |
732 | ||
733 | /* run eBPF verifier */ | |
9bac3d6d | 734 | err = bpf_check(&prog, attr); |
09756af4 AS |
735 | if (err < 0) |
736 | goto free_used_maps; | |
737 | ||
0a542a86 AS |
738 | /* fixup BPF_CALL->imm field */ |
739 | fixup_bpf_calls(prog); | |
740 | ||
09756af4 | 741 | /* eBPF program is ready to be JITed */ |
04fd61ab AS |
742 | err = bpf_prog_select_runtime(prog); |
743 | if (err < 0) | |
744 | goto free_used_maps; | |
09756af4 | 745 | |
aa79781b | 746 | err = bpf_prog_new_fd(prog); |
09756af4 AS |
747 | if (err < 0) |
748 | /* failed to allocate fd */ | |
749 | goto free_used_maps; | |
750 | ||
751 | return err; | |
752 | ||
753 | free_used_maps: | |
754 | free_used_maps(prog->aux); | |
755 | free_prog: | |
aaac3ba9 AS |
756 | bpf_prog_uncharge_memlock(prog); |
757 | free_prog_nouncharge: | |
09756af4 AS |
758 | bpf_prog_free(prog); |
759 | return err; | |
760 | } | |
761 | ||
b2197755 DB |
762 | #define BPF_OBJ_LAST_FIELD bpf_fd |
763 | ||
764 | static int bpf_obj_pin(const union bpf_attr *attr) | |
765 | { | |
766 | if (CHECK_ATTR(BPF_OBJ)) | |
767 | return -EINVAL; | |
768 | ||
769 | return bpf_obj_pin_user(attr->bpf_fd, u64_to_ptr(attr->pathname)); | |
770 | } | |
771 | ||
772 | static int bpf_obj_get(const union bpf_attr *attr) | |
773 | { | |
774 | if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0) | |
775 | return -EINVAL; | |
776 | ||
777 | return bpf_obj_get_user(u64_to_ptr(attr->pathname)); | |
778 | } | |
779 | ||
99c55f7d AS |
780 | SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size) |
781 | { | |
782 | union bpf_attr attr = {}; | |
783 | int err; | |
784 | ||
1be7f75d | 785 | if (!capable(CAP_SYS_ADMIN) && sysctl_unprivileged_bpf_disabled) |
99c55f7d AS |
786 | return -EPERM; |
787 | ||
788 | if (!access_ok(VERIFY_READ, uattr, 1)) | |
789 | return -EFAULT; | |
790 | ||
791 | if (size > PAGE_SIZE) /* silly large */ | |
792 | return -E2BIG; | |
793 | ||
794 | /* If we're handed a bigger struct than we know of, | |
795 | * ensure all the unknown bits are 0 - i.e. new | |
796 | * user-space does not rely on any kernel feature | |
797 | * extensions we dont know about yet. | |
798 | */ | |
799 | if (size > sizeof(attr)) { | |
800 | unsigned char __user *addr; | |
801 | unsigned char __user *end; | |
802 | unsigned char val; | |
803 | ||
804 | addr = (void __user *)uattr + sizeof(attr); | |
805 | end = (void __user *)uattr + size; | |
806 | ||
807 | for (; addr < end; addr++) { | |
808 | err = get_user(val, addr); | |
809 | if (err) | |
810 | return err; | |
811 | if (val) | |
812 | return -E2BIG; | |
813 | } | |
814 | size = sizeof(attr); | |
815 | } | |
816 | ||
817 | /* copy attributes from user space, may be less than sizeof(bpf_attr) */ | |
818 | if (copy_from_user(&attr, uattr, size) != 0) | |
819 | return -EFAULT; | |
820 | ||
821 | switch (cmd) { | |
822 | case BPF_MAP_CREATE: | |
823 | err = map_create(&attr); | |
824 | break; | |
db20fd2b AS |
825 | case BPF_MAP_LOOKUP_ELEM: |
826 | err = map_lookup_elem(&attr); | |
827 | break; | |
828 | case BPF_MAP_UPDATE_ELEM: | |
829 | err = map_update_elem(&attr); | |
830 | break; | |
831 | case BPF_MAP_DELETE_ELEM: | |
832 | err = map_delete_elem(&attr); | |
833 | break; | |
834 | case BPF_MAP_GET_NEXT_KEY: | |
835 | err = map_get_next_key(&attr); | |
836 | break; | |
09756af4 AS |
837 | case BPF_PROG_LOAD: |
838 | err = bpf_prog_load(&attr); | |
839 | break; | |
b2197755 DB |
840 | case BPF_OBJ_PIN: |
841 | err = bpf_obj_pin(&attr); | |
842 | break; | |
843 | case BPF_OBJ_GET: | |
844 | err = bpf_obj_get(&attr); | |
845 | break; | |
99c55f7d AS |
846 | default: |
847 | err = -EINVAL; | |
848 | break; | |
849 | } | |
850 | ||
851 | return err; | |
852 | } |