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" | |
322cea2f DB |
140 | "max_entries:\t%u\n" |
141 | "map_flags:\t%#x\n", | |
f99bf205 DB |
142 | map->map_type, |
143 | map->key_size, | |
144 | map->value_size, | |
322cea2f DB |
145 | map->max_entries, |
146 | map->map_flags); | |
f99bf205 DB |
147 | } |
148 | #endif | |
149 | ||
99c55f7d | 150 | static const struct file_operations bpf_map_fops = { |
f99bf205 DB |
151 | #ifdef CONFIG_PROC_FS |
152 | .show_fdinfo = bpf_map_show_fdinfo, | |
153 | #endif | |
154 | .release = bpf_map_release, | |
99c55f7d AS |
155 | }; |
156 | ||
b2197755 | 157 | int bpf_map_new_fd(struct bpf_map *map) |
aa79781b DB |
158 | { |
159 | return anon_inode_getfd("bpf-map", &bpf_map_fops, map, | |
160 | O_RDWR | O_CLOEXEC); | |
161 | } | |
162 | ||
99c55f7d AS |
163 | /* helper macro to check that unused fields 'union bpf_attr' are zero */ |
164 | #define CHECK_ATTR(CMD) \ | |
165 | memchr_inv((void *) &attr->CMD##_LAST_FIELD + \ | |
166 | sizeof(attr->CMD##_LAST_FIELD), 0, \ | |
167 | sizeof(*attr) - \ | |
168 | offsetof(union bpf_attr, CMD##_LAST_FIELD) - \ | |
169 | sizeof(attr->CMD##_LAST_FIELD)) != NULL | |
170 | ||
6c905981 | 171 | #define BPF_MAP_CREATE_LAST_FIELD map_flags |
99c55f7d AS |
172 | /* called via syscall */ |
173 | static int map_create(union bpf_attr *attr) | |
174 | { | |
175 | struct bpf_map *map; | |
176 | int err; | |
177 | ||
178 | err = CHECK_ATTR(BPF_MAP_CREATE); | |
179 | if (err) | |
180 | return -EINVAL; | |
181 | ||
182 | /* find map type and init map: hashtable vs rbtree vs bloom vs ... */ | |
183 | map = find_and_alloc_map(attr); | |
184 | if (IS_ERR(map)) | |
185 | return PTR_ERR(map); | |
186 | ||
187 | atomic_set(&map->refcnt, 1); | |
c9da161c | 188 | atomic_set(&map->usercnt, 1); |
99c55f7d | 189 | |
aaac3ba9 AS |
190 | err = bpf_map_charge_memlock(map); |
191 | if (err) | |
192 | goto free_map; | |
193 | ||
aa79781b | 194 | err = bpf_map_new_fd(map); |
99c55f7d AS |
195 | if (err < 0) |
196 | /* failed to allocate fd */ | |
197 | goto free_map; | |
198 | ||
199 | return err; | |
200 | ||
201 | free_map: | |
202 | map->ops->map_free(map); | |
203 | return err; | |
204 | } | |
205 | ||
db20fd2b AS |
206 | /* if error is returned, fd is released. |
207 | * On success caller should complete fd access with matching fdput() | |
208 | */ | |
c2101297 | 209 | struct bpf_map *__bpf_map_get(struct fd f) |
db20fd2b | 210 | { |
db20fd2b AS |
211 | if (!f.file) |
212 | return ERR_PTR(-EBADF); | |
db20fd2b AS |
213 | if (f.file->f_op != &bpf_map_fops) { |
214 | fdput(f); | |
215 | return ERR_PTR(-EINVAL); | |
216 | } | |
217 | ||
c2101297 DB |
218 | return f.file->private_data; |
219 | } | |
220 | ||
92117d84 AS |
221 | /* prog's and map's refcnt limit */ |
222 | #define BPF_MAX_REFCNT 32768 | |
223 | ||
224 | struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref) | |
c9da161c | 225 | { |
92117d84 AS |
226 | if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) { |
227 | atomic_dec(&map->refcnt); | |
228 | return ERR_PTR(-EBUSY); | |
229 | } | |
c9da161c DB |
230 | if (uref) |
231 | atomic_inc(&map->usercnt); | |
92117d84 | 232 | return map; |
c9da161c DB |
233 | } |
234 | ||
235 | struct bpf_map *bpf_map_get_with_uref(u32 ufd) | |
c2101297 DB |
236 | { |
237 | struct fd f = fdget(ufd); | |
238 | struct bpf_map *map; | |
239 | ||
240 | map = __bpf_map_get(f); | |
241 | if (IS_ERR(map)) | |
242 | return map; | |
243 | ||
92117d84 | 244 | map = bpf_map_inc(map, true); |
c2101297 | 245 | fdput(f); |
db20fd2b AS |
246 | |
247 | return map; | |
248 | } | |
249 | ||
250 | /* helper to convert user pointers passed inside __aligned_u64 fields */ | |
251 | static void __user *u64_to_ptr(__u64 val) | |
252 | { | |
253 | return (void __user *) (unsigned long) val; | |
254 | } | |
255 | ||
b8cdc051 AS |
256 | int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value) |
257 | { | |
258 | return -ENOTSUPP; | |
259 | } | |
260 | ||
db20fd2b AS |
261 | /* last field in 'union bpf_attr' used by this command */ |
262 | #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value | |
263 | ||
264 | static int map_lookup_elem(union bpf_attr *attr) | |
265 | { | |
266 | void __user *ukey = u64_to_ptr(attr->key); | |
267 | void __user *uvalue = u64_to_ptr(attr->value); | |
268 | int ufd = attr->map_fd; | |
db20fd2b | 269 | struct bpf_map *map; |
8ebe667c | 270 | void *key, *value, *ptr; |
15a07b33 | 271 | u32 value_size; |
592867bf | 272 | struct fd f; |
db20fd2b AS |
273 | int err; |
274 | ||
275 | if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM)) | |
276 | return -EINVAL; | |
277 | ||
592867bf | 278 | f = fdget(ufd); |
c2101297 | 279 | map = __bpf_map_get(f); |
db20fd2b AS |
280 | if (IS_ERR(map)) |
281 | return PTR_ERR(map); | |
282 | ||
283 | err = -ENOMEM; | |
284 | key = kmalloc(map->key_size, GFP_USER); | |
285 | if (!key) | |
286 | goto err_put; | |
287 | ||
288 | err = -EFAULT; | |
289 | if (copy_from_user(key, ukey, map->key_size) != 0) | |
290 | goto free_key; | |
291 | ||
15a07b33 AS |
292 | if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || |
293 | map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) | |
294 | value_size = round_up(map->value_size, 8) * num_possible_cpus(); | |
295 | else | |
296 | value_size = map->value_size; | |
297 | ||
8ebe667c | 298 | err = -ENOMEM; |
15a07b33 | 299 | value = kmalloc(value_size, GFP_USER | __GFP_NOWARN); |
db20fd2b | 300 | if (!value) |
8ebe667c AS |
301 | goto free_key; |
302 | ||
15a07b33 AS |
303 | if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH) { |
304 | err = bpf_percpu_hash_copy(map, key, value); | |
305 | } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) { | |
306 | err = bpf_percpu_array_copy(map, key, value); | |
557c0c6e AS |
307 | } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) { |
308 | err = bpf_stackmap_copy(map, key, value); | |
15a07b33 AS |
309 | } else { |
310 | rcu_read_lock(); | |
311 | ptr = map->ops->map_lookup_elem(map, key); | |
312 | if (ptr) | |
313 | memcpy(value, ptr, value_size); | |
314 | rcu_read_unlock(); | |
315 | err = ptr ? 0 : -ENOENT; | |
316 | } | |
8ebe667c | 317 | |
15a07b33 | 318 | if (err) |
8ebe667c | 319 | goto free_value; |
db20fd2b AS |
320 | |
321 | err = -EFAULT; | |
15a07b33 | 322 | if (copy_to_user(uvalue, value, value_size) != 0) |
8ebe667c | 323 | goto free_value; |
db20fd2b AS |
324 | |
325 | err = 0; | |
326 | ||
8ebe667c AS |
327 | free_value: |
328 | kfree(value); | |
db20fd2b AS |
329 | free_key: |
330 | kfree(key); | |
331 | err_put: | |
332 | fdput(f); | |
333 | return err; | |
334 | } | |
335 | ||
3274f520 | 336 | #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags |
db20fd2b AS |
337 | |
338 | static int map_update_elem(union bpf_attr *attr) | |
339 | { | |
340 | void __user *ukey = u64_to_ptr(attr->key); | |
341 | void __user *uvalue = u64_to_ptr(attr->value); | |
342 | int ufd = attr->map_fd; | |
db20fd2b AS |
343 | struct bpf_map *map; |
344 | void *key, *value; | |
15a07b33 | 345 | u32 value_size; |
592867bf | 346 | struct fd f; |
db20fd2b AS |
347 | int err; |
348 | ||
349 | if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM)) | |
350 | return -EINVAL; | |
351 | ||
592867bf | 352 | f = fdget(ufd); |
c2101297 | 353 | map = __bpf_map_get(f); |
db20fd2b AS |
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 | ||
15a07b33 AS |
366 | if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || |
367 | map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) | |
368 | value_size = round_up(map->value_size, 8) * num_possible_cpus(); | |
369 | else | |
370 | value_size = map->value_size; | |
371 | ||
db20fd2b | 372 | err = -ENOMEM; |
15a07b33 | 373 | value = kmalloc(value_size, GFP_USER | __GFP_NOWARN); |
db20fd2b AS |
374 | if (!value) |
375 | goto free_key; | |
376 | ||
377 | err = -EFAULT; | |
15a07b33 | 378 | if (copy_from_user(value, uvalue, value_size) != 0) |
db20fd2b AS |
379 | goto free_value; |
380 | ||
b121d1e7 AS |
381 | /* must increment bpf_prog_active to avoid kprobe+bpf triggering from |
382 | * inside bpf map update or delete otherwise deadlocks are possible | |
383 | */ | |
384 | preempt_disable(); | |
385 | __this_cpu_inc(bpf_prog_active); | |
15a07b33 AS |
386 | if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH) { |
387 | err = bpf_percpu_hash_update(map, key, value, attr->flags); | |
388 | } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) { | |
389 | err = bpf_percpu_array_update(map, key, value, attr->flags); | |
390 | } else { | |
391 | rcu_read_lock(); | |
392 | err = map->ops->map_update_elem(map, key, value, attr->flags); | |
393 | rcu_read_unlock(); | |
394 | } | |
b121d1e7 AS |
395 | __this_cpu_dec(bpf_prog_active); |
396 | preempt_enable(); | |
db20fd2b AS |
397 | |
398 | free_value: | |
399 | kfree(value); | |
400 | free_key: | |
401 | kfree(key); | |
402 | err_put: | |
403 | fdput(f); | |
404 | return err; | |
405 | } | |
406 | ||
407 | #define BPF_MAP_DELETE_ELEM_LAST_FIELD key | |
408 | ||
409 | static int map_delete_elem(union bpf_attr *attr) | |
410 | { | |
411 | void __user *ukey = u64_to_ptr(attr->key); | |
412 | int ufd = attr->map_fd; | |
db20fd2b | 413 | struct bpf_map *map; |
592867bf | 414 | struct fd f; |
db20fd2b AS |
415 | void *key; |
416 | int err; | |
417 | ||
418 | if (CHECK_ATTR(BPF_MAP_DELETE_ELEM)) | |
419 | return -EINVAL; | |
420 | ||
592867bf | 421 | f = fdget(ufd); |
c2101297 | 422 | map = __bpf_map_get(f); |
db20fd2b AS |
423 | if (IS_ERR(map)) |
424 | return PTR_ERR(map); | |
425 | ||
426 | err = -ENOMEM; | |
427 | key = kmalloc(map->key_size, GFP_USER); | |
428 | if (!key) | |
429 | goto err_put; | |
430 | ||
431 | err = -EFAULT; | |
432 | if (copy_from_user(key, ukey, map->key_size) != 0) | |
433 | goto free_key; | |
434 | ||
b121d1e7 AS |
435 | preempt_disable(); |
436 | __this_cpu_inc(bpf_prog_active); | |
db20fd2b AS |
437 | rcu_read_lock(); |
438 | err = map->ops->map_delete_elem(map, key); | |
439 | rcu_read_unlock(); | |
b121d1e7 AS |
440 | __this_cpu_dec(bpf_prog_active); |
441 | preempt_enable(); | |
db20fd2b AS |
442 | |
443 | free_key: | |
444 | kfree(key); | |
445 | err_put: | |
446 | fdput(f); | |
447 | return err; | |
448 | } | |
449 | ||
450 | /* last field in 'union bpf_attr' used by this command */ | |
451 | #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key | |
452 | ||
453 | static int map_get_next_key(union bpf_attr *attr) | |
454 | { | |
455 | void __user *ukey = u64_to_ptr(attr->key); | |
456 | void __user *unext_key = u64_to_ptr(attr->next_key); | |
457 | int ufd = attr->map_fd; | |
db20fd2b AS |
458 | struct bpf_map *map; |
459 | void *key, *next_key; | |
592867bf | 460 | struct fd f; |
db20fd2b AS |
461 | int err; |
462 | ||
463 | if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY)) | |
464 | return -EINVAL; | |
465 | ||
592867bf | 466 | f = fdget(ufd); |
c2101297 | 467 | map = __bpf_map_get(f); |
db20fd2b AS |
468 | if (IS_ERR(map)) |
469 | return PTR_ERR(map); | |
470 | ||
471 | err = -ENOMEM; | |
472 | key = kmalloc(map->key_size, GFP_USER); | |
473 | if (!key) | |
474 | goto err_put; | |
475 | ||
476 | err = -EFAULT; | |
477 | if (copy_from_user(key, ukey, map->key_size) != 0) | |
478 | goto free_key; | |
479 | ||
480 | err = -ENOMEM; | |
481 | next_key = kmalloc(map->key_size, GFP_USER); | |
482 | if (!next_key) | |
483 | goto free_key; | |
484 | ||
485 | rcu_read_lock(); | |
486 | err = map->ops->map_get_next_key(map, key, next_key); | |
487 | rcu_read_unlock(); | |
488 | if (err) | |
489 | goto free_next_key; | |
490 | ||
491 | err = -EFAULT; | |
492 | if (copy_to_user(unext_key, next_key, map->key_size) != 0) | |
493 | goto free_next_key; | |
494 | ||
495 | err = 0; | |
496 | ||
497 | free_next_key: | |
498 | kfree(next_key); | |
499 | free_key: | |
500 | kfree(key); | |
501 | err_put: | |
502 | fdput(f); | |
503 | return err; | |
504 | } | |
505 | ||
09756af4 AS |
506 | static LIST_HEAD(bpf_prog_types); |
507 | ||
508 | static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog) | |
509 | { | |
510 | struct bpf_prog_type_list *tl; | |
511 | ||
512 | list_for_each_entry(tl, &bpf_prog_types, list_node) { | |
513 | if (tl->type == type) { | |
514 | prog->aux->ops = tl->ops; | |
24701ece | 515 | prog->type = type; |
09756af4 AS |
516 | return 0; |
517 | } | |
518 | } | |
24701ece | 519 | |
09756af4 AS |
520 | return -EINVAL; |
521 | } | |
522 | ||
523 | void bpf_register_prog_type(struct bpf_prog_type_list *tl) | |
524 | { | |
525 | list_add(&tl->list_node, &bpf_prog_types); | |
526 | } | |
527 | ||
0a542a86 AS |
528 | /* fixup insn->imm field of bpf_call instructions: |
529 | * if (insn->imm == BPF_FUNC_map_lookup_elem) | |
530 | * insn->imm = bpf_map_lookup_elem - __bpf_call_base; | |
531 | * else if (insn->imm == BPF_FUNC_map_update_elem) | |
532 | * insn->imm = bpf_map_update_elem - __bpf_call_base; | |
533 | * else ... | |
534 | * | |
535 | * this function is called after eBPF program passed verification | |
536 | */ | |
537 | static void fixup_bpf_calls(struct bpf_prog *prog) | |
538 | { | |
539 | const struct bpf_func_proto *fn; | |
540 | int i; | |
541 | ||
542 | for (i = 0; i < prog->len; i++) { | |
543 | struct bpf_insn *insn = &prog->insnsi[i]; | |
544 | ||
545 | if (insn->code == (BPF_JMP | BPF_CALL)) { | |
546 | /* we reach here when program has bpf_call instructions | |
547 | * and it passed bpf_check(), means that | |
548 | * ops->get_func_proto must have been supplied, check it | |
549 | */ | |
550 | BUG_ON(!prog->aux->ops->get_func_proto); | |
551 | ||
c46646d0 DB |
552 | if (insn->imm == BPF_FUNC_get_route_realm) |
553 | prog->dst_needed = 1; | |
3ad00405 DB |
554 | if (insn->imm == BPF_FUNC_get_prandom_u32) |
555 | bpf_user_rnd_init_once(); | |
04fd61ab AS |
556 | if (insn->imm == BPF_FUNC_tail_call) { |
557 | /* mark bpf_tail_call as different opcode | |
558 | * to avoid conditional branch in | |
559 | * interpeter for every normal call | |
560 | * and to prevent accidental JITing by | |
561 | * JIT compiler that doesn't support | |
562 | * bpf_tail_call yet | |
563 | */ | |
564 | insn->imm = 0; | |
565 | insn->code |= BPF_X; | |
566 | continue; | |
567 | } | |
568 | ||
0a542a86 AS |
569 | fn = prog->aux->ops->get_func_proto(insn->imm); |
570 | /* all functions that have prototype and verifier allowed | |
571 | * programs to call them, must be real in-kernel functions | |
572 | */ | |
573 | BUG_ON(!fn->func); | |
574 | insn->imm = fn->func - __bpf_call_base; | |
575 | } | |
576 | } | |
577 | } | |
578 | ||
09756af4 AS |
579 | /* drop refcnt on maps used by eBPF program and free auxilary data */ |
580 | static void free_used_maps(struct bpf_prog_aux *aux) | |
581 | { | |
582 | int i; | |
583 | ||
584 | for (i = 0; i < aux->used_map_cnt; i++) | |
585 | bpf_map_put(aux->used_maps[i]); | |
586 | ||
587 | kfree(aux->used_maps); | |
588 | } | |
589 | ||
aaac3ba9 AS |
590 | static int bpf_prog_charge_memlock(struct bpf_prog *prog) |
591 | { | |
592 | struct user_struct *user = get_current_user(); | |
593 | unsigned long memlock_limit; | |
594 | ||
595 | memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
596 | ||
597 | atomic_long_add(prog->pages, &user->locked_vm); | |
598 | if (atomic_long_read(&user->locked_vm) > memlock_limit) { | |
599 | atomic_long_sub(prog->pages, &user->locked_vm); | |
600 | free_uid(user); | |
601 | return -EPERM; | |
602 | } | |
603 | prog->aux->user = user; | |
604 | return 0; | |
605 | } | |
606 | ||
607 | static void bpf_prog_uncharge_memlock(struct bpf_prog *prog) | |
608 | { | |
609 | struct user_struct *user = prog->aux->user; | |
610 | ||
611 | atomic_long_sub(prog->pages, &user->locked_vm); | |
612 | free_uid(user); | |
613 | } | |
614 | ||
e9d8afa9 | 615 | static void __prog_put_common(struct rcu_head *rcu) |
abf2e7d6 AS |
616 | { |
617 | struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu); | |
618 | ||
619 | free_used_maps(aux); | |
aaac3ba9 | 620 | bpf_prog_uncharge_memlock(aux->prog); |
abf2e7d6 AS |
621 | bpf_prog_free(aux->prog); |
622 | } | |
623 | ||
624 | /* version of bpf_prog_put() that is called after a grace period */ | |
625 | void bpf_prog_put_rcu(struct bpf_prog *prog) | |
626 | { | |
e9d8afa9 DB |
627 | if (atomic_dec_and_test(&prog->aux->refcnt)) |
628 | call_rcu(&prog->aux->rcu, __prog_put_common); | |
abf2e7d6 AS |
629 | } |
630 | ||
09756af4 AS |
631 | void bpf_prog_put(struct bpf_prog *prog) |
632 | { | |
e9d8afa9 DB |
633 | if (atomic_dec_and_test(&prog->aux->refcnt)) |
634 | __prog_put_common(&prog->aux->rcu); | |
09756af4 | 635 | } |
e2e9b654 | 636 | EXPORT_SYMBOL_GPL(bpf_prog_put); |
09756af4 AS |
637 | |
638 | static int bpf_prog_release(struct inode *inode, struct file *filp) | |
639 | { | |
640 | struct bpf_prog *prog = filp->private_data; | |
641 | ||
abf2e7d6 | 642 | bpf_prog_put_rcu(prog); |
09756af4 AS |
643 | return 0; |
644 | } | |
645 | ||
646 | static const struct file_operations bpf_prog_fops = { | |
647 | .release = bpf_prog_release, | |
648 | }; | |
649 | ||
b2197755 | 650 | int bpf_prog_new_fd(struct bpf_prog *prog) |
aa79781b DB |
651 | { |
652 | return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog, | |
653 | O_RDWR | O_CLOEXEC); | |
654 | } | |
655 | ||
c2101297 | 656 | static struct bpf_prog *__bpf_prog_get(struct fd f) |
09756af4 | 657 | { |
09756af4 AS |
658 | if (!f.file) |
659 | return ERR_PTR(-EBADF); | |
09756af4 AS |
660 | if (f.file->f_op != &bpf_prog_fops) { |
661 | fdput(f); | |
662 | return ERR_PTR(-EINVAL); | |
663 | } | |
664 | ||
c2101297 | 665 | return f.file->private_data; |
09756af4 AS |
666 | } |
667 | ||
92117d84 AS |
668 | struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog) |
669 | { | |
670 | if (atomic_inc_return(&prog->aux->refcnt) > BPF_MAX_REFCNT) { | |
671 | atomic_dec(&prog->aux->refcnt); | |
672 | return ERR_PTR(-EBUSY); | |
673 | } | |
674 | return prog; | |
675 | } | |
676 | ||
09756af4 AS |
677 | /* called by sockets/tracing/seccomp before attaching program to an event |
678 | * pairs with bpf_prog_put() | |
679 | */ | |
680 | struct bpf_prog *bpf_prog_get(u32 ufd) | |
681 | { | |
682 | struct fd f = fdget(ufd); | |
683 | struct bpf_prog *prog; | |
684 | ||
c2101297 | 685 | prog = __bpf_prog_get(f); |
09756af4 AS |
686 | if (IS_ERR(prog)) |
687 | return prog; | |
688 | ||
92117d84 | 689 | prog = bpf_prog_inc(prog); |
09756af4 | 690 | fdput(f); |
c2101297 | 691 | |
09756af4 AS |
692 | return prog; |
693 | } | |
e2e9b654 | 694 | EXPORT_SYMBOL_GPL(bpf_prog_get); |
09756af4 AS |
695 | |
696 | /* last field in 'union bpf_attr' used by this command */ | |
2541517c | 697 | #define BPF_PROG_LOAD_LAST_FIELD kern_version |
09756af4 AS |
698 | |
699 | static int bpf_prog_load(union bpf_attr *attr) | |
700 | { | |
701 | enum bpf_prog_type type = attr->prog_type; | |
702 | struct bpf_prog *prog; | |
703 | int err; | |
704 | char license[128]; | |
705 | bool is_gpl; | |
706 | ||
707 | if (CHECK_ATTR(BPF_PROG_LOAD)) | |
708 | return -EINVAL; | |
709 | ||
710 | /* copy eBPF program license from user space */ | |
711 | if (strncpy_from_user(license, u64_to_ptr(attr->license), | |
712 | sizeof(license) - 1) < 0) | |
713 | return -EFAULT; | |
714 | license[sizeof(license) - 1] = 0; | |
715 | ||
716 | /* eBPF programs must be GPL compatible to use GPL-ed functions */ | |
717 | is_gpl = license_is_gpl_compatible(license); | |
718 | ||
719 | if (attr->insn_cnt >= BPF_MAXINSNS) | |
720 | return -EINVAL; | |
721 | ||
2541517c AS |
722 | if (type == BPF_PROG_TYPE_KPROBE && |
723 | attr->kern_version != LINUX_VERSION_CODE) | |
724 | return -EINVAL; | |
725 | ||
1be7f75d AS |
726 | if (type != BPF_PROG_TYPE_SOCKET_FILTER && !capable(CAP_SYS_ADMIN)) |
727 | return -EPERM; | |
728 | ||
09756af4 AS |
729 | /* plain bpf_prog allocation */ |
730 | prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER); | |
731 | if (!prog) | |
732 | return -ENOMEM; | |
733 | ||
aaac3ba9 AS |
734 | err = bpf_prog_charge_memlock(prog); |
735 | if (err) | |
736 | goto free_prog_nouncharge; | |
737 | ||
09756af4 AS |
738 | prog->len = attr->insn_cnt; |
739 | ||
740 | err = -EFAULT; | |
741 | if (copy_from_user(prog->insns, u64_to_ptr(attr->insns), | |
742 | prog->len * sizeof(struct bpf_insn)) != 0) | |
743 | goto free_prog; | |
744 | ||
745 | prog->orig_prog = NULL; | |
a91263d5 | 746 | prog->jited = 0; |
09756af4 AS |
747 | |
748 | atomic_set(&prog->aux->refcnt, 1); | |
a91263d5 | 749 | prog->gpl_compatible = is_gpl ? 1 : 0; |
09756af4 AS |
750 | |
751 | /* find program type: socket_filter vs tracing_filter */ | |
752 | err = find_prog_type(type, prog); | |
753 | if (err < 0) | |
754 | goto free_prog; | |
755 | ||
756 | /* run eBPF verifier */ | |
9bac3d6d | 757 | err = bpf_check(&prog, attr); |
09756af4 AS |
758 | if (err < 0) |
759 | goto free_used_maps; | |
760 | ||
0a542a86 AS |
761 | /* fixup BPF_CALL->imm field */ |
762 | fixup_bpf_calls(prog); | |
763 | ||
09756af4 | 764 | /* eBPF program is ready to be JITed */ |
d1c55ab5 | 765 | prog = bpf_prog_select_runtime(prog, &err); |
04fd61ab AS |
766 | if (err < 0) |
767 | goto free_used_maps; | |
09756af4 | 768 | |
aa79781b | 769 | err = bpf_prog_new_fd(prog); |
09756af4 AS |
770 | if (err < 0) |
771 | /* failed to allocate fd */ | |
772 | goto free_used_maps; | |
773 | ||
774 | return err; | |
775 | ||
776 | free_used_maps: | |
777 | free_used_maps(prog->aux); | |
778 | free_prog: | |
aaac3ba9 AS |
779 | bpf_prog_uncharge_memlock(prog); |
780 | free_prog_nouncharge: | |
09756af4 AS |
781 | bpf_prog_free(prog); |
782 | return err; | |
783 | } | |
784 | ||
b2197755 DB |
785 | #define BPF_OBJ_LAST_FIELD bpf_fd |
786 | ||
787 | static int bpf_obj_pin(const union bpf_attr *attr) | |
788 | { | |
789 | if (CHECK_ATTR(BPF_OBJ)) | |
790 | return -EINVAL; | |
791 | ||
792 | return bpf_obj_pin_user(attr->bpf_fd, u64_to_ptr(attr->pathname)); | |
793 | } | |
794 | ||
795 | static int bpf_obj_get(const union bpf_attr *attr) | |
796 | { | |
797 | if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0) | |
798 | return -EINVAL; | |
799 | ||
800 | return bpf_obj_get_user(u64_to_ptr(attr->pathname)); | |
801 | } | |
802 | ||
99c55f7d AS |
803 | SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size) |
804 | { | |
805 | union bpf_attr attr = {}; | |
806 | int err; | |
807 | ||
1be7f75d | 808 | if (!capable(CAP_SYS_ADMIN) && sysctl_unprivileged_bpf_disabled) |
99c55f7d AS |
809 | return -EPERM; |
810 | ||
811 | if (!access_ok(VERIFY_READ, uattr, 1)) | |
812 | return -EFAULT; | |
813 | ||
814 | if (size > PAGE_SIZE) /* silly large */ | |
815 | return -E2BIG; | |
816 | ||
817 | /* If we're handed a bigger struct than we know of, | |
818 | * ensure all the unknown bits are 0 - i.e. new | |
819 | * user-space does not rely on any kernel feature | |
820 | * extensions we dont know about yet. | |
821 | */ | |
822 | if (size > sizeof(attr)) { | |
823 | unsigned char __user *addr; | |
824 | unsigned char __user *end; | |
825 | unsigned char val; | |
826 | ||
827 | addr = (void __user *)uattr + sizeof(attr); | |
828 | end = (void __user *)uattr + size; | |
829 | ||
830 | for (; addr < end; addr++) { | |
831 | err = get_user(val, addr); | |
832 | if (err) | |
833 | return err; | |
834 | if (val) | |
835 | return -E2BIG; | |
836 | } | |
837 | size = sizeof(attr); | |
838 | } | |
839 | ||
840 | /* copy attributes from user space, may be less than sizeof(bpf_attr) */ | |
841 | if (copy_from_user(&attr, uattr, size) != 0) | |
842 | return -EFAULT; | |
843 | ||
844 | switch (cmd) { | |
845 | case BPF_MAP_CREATE: | |
846 | err = map_create(&attr); | |
847 | break; | |
db20fd2b AS |
848 | case BPF_MAP_LOOKUP_ELEM: |
849 | err = map_lookup_elem(&attr); | |
850 | break; | |
851 | case BPF_MAP_UPDATE_ELEM: | |
852 | err = map_update_elem(&attr); | |
853 | break; | |
854 | case BPF_MAP_DELETE_ELEM: | |
855 | err = map_delete_elem(&attr); | |
856 | break; | |
857 | case BPF_MAP_GET_NEXT_KEY: | |
858 | err = map_get_next_key(&attr); | |
859 | break; | |
09756af4 AS |
860 | case BPF_PROG_LOAD: |
861 | err = bpf_prog_load(&attr); | |
862 | break; | |
b2197755 DB |
863 | case BPF_OBJ_PIN: |
864 | err = bpf_obj_pin(&attr); | |
865 | break; | |
866 | case BPF_OBJ_GET: | |
867 | err = bpf_obj_get(&attr); | |
868 | break; | |
99c55f7d AS |
869 | default: |
870 | err = -EINVAL; | |
871 | break; | |
872 | } | |
873 | ||
874 | return err; | |
875 | } |