]>
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> |
535e7b4b | 20 | #include <linux/kernel.h> |
99c55f7d | 21 | |
b121d1e7 AS |
22 | DEFINE_PER_CPU(int, bpf_prog_active); |
23 | ||
1be7f75d AS |
24 | int sysctl_unprivileged_bpf_disabled __read_mostly; |
25 | ||
99c55f7d AS |
26 | static LIST_HEAD(bpf_map_types); |
27 | ||
28 | static struct bpf_map *find_and_alloc_map(union bpf_attr *attr) | |
29 | { | |
30 | struct bpf_map_type_list *tl; | |
31 | struct bpf_map *map; | |
32 | ||
33 | list_for_each_entry(tl, &bpf_map_types, list_node) { | |
34 | if (tl->type == attr->map_type) { | |
35 | map = tl->ops->map_alloc(attr); | |
36 | if (IS_ERR(map)) | |
37 | return map; | |
38 | map->ops = tl->ops; | |
39 | map->map_type = attr->map_type; | |
40 | return map; | |
41 | } | |
42 | } | |
43 | return ERR_PTR(-EINVAL); | |
44 | } | |
45 | ||
46 | /* boot time registration of different map implementations */ | |
47 | void bpf_register_map_type(struct bpf_map_type_list *tl) | |
48 | { | |
49 | list_add(&tl->list_node, &bpf_map_types); | |
50 | } | |
51 | ||
6c905981 AS |
52 | int bpf_map_precharge_memlock(u32 pages) |
53 | { | |
54 | struct user_struct *user = get_current_user(); | |
55 | unsigned long memlock_limit, cur; | |
56 | ||
57 | memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
58 | cur = atomic_long_read(&user->locked_vm); | |
59 | free_uid(user); | |
60 | if (cur + pages > memlock_limit) | |
61 | return -EPERM; | |
62 | return 0; | |
63 | } | |
64 | ||
aaac3ba9 AS |
65 | static int bpf_map_charge_memlock(struct bpf_map *map) |
66 | { | |
67 | struct user_struct *user = get_current_user(); | |
68 | unsigned long memlock_limit; | |
69 | ||
70 | memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
71 | ||
72 | atomic_long_add(map->pages, &user->locked_vm); | |
73 | ||
74 | if (atomic_long_read(&user->locked_vm) > memlock_limit) { | |
75 | atomic_long_sub(map->pages, &user->locked_vm); | |
76 | free_uid(user); | |
77 | return -EPERM; | |
78 | } | |
79 | map->user = user; | |
80 | return 0; | |
81 | } | |
82 | ||
83 | static void bpf_map_uncharge_memlock(struct bpf_map *map) | |
84 | { | |
85 | struct user_struct *user = map->user; | |
86 | ||
87 | atomic_long_sub(map->pages, &user->locked_vm); | |
88 | free_uid(user); | |
89 | } | |
90 | ||
99c55f7d AS |
91 | /* called from workqueue */ |
92 | static void bpf_map_free_deferred(struct work_struct *work) | |
93 | { | |
94 | struct bpf_map *map = container_of(work, struct bpf_map, work); | |
95 | ||
aaac3ba9 | 96 | bpf_map_uncharge_memlock(map); |
99c55f7d AS |
97 | /* implementation dependent freeing */ |
98 | map->ops->map_free(map); | |
99 | } | |
100 | ||
c9da161c DB |
101 | static void bpf_map_put_uref(struct bpf_map *map) |
102 | { | |
103 | if (atomic_dec_and_test(&map->usercnt)) { | |
104 | if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) | |
105 | bpf_fd_array_map_clear(map); | |
106 | } | |
107 | } | |
108 | ||
99c55f7d AS |
109 | /* decrement map refcnt and schedule it for freeing via workqueue |
110 | * (unrelying map implementation ops->map_free() might sleep) | |
111 | */ | |
112 | void bpf_map_put(struct bpf_map *map) | |
113 | { | |
114 | if (atomic_dec_and_test(&map->refcnt)) { | |
115 | INIT_WORK(&map->work, bpf_map_free_deferred); | |
116 | schedule_work(&map->work); | |
117 | } | |
118 | } | |
119 | ||
c9da161c | 120 | void bpf_map_put_with_uref(struct bpf_map *map) |
99c55f7d | 121 | { |
c9da161c | 122 | bpf_map_put_uref(map); |
99c55f7d | 123 | bpf_map_put(map); |
c9da161c DB |
124 | } |
125 | ||
126 | static int bpf_map_release(struct inode *inode, struct file *filp) | |
127 | { | |
61d1b6a4 DB |
128 | struct bpf_map *map = filp->private_data; |
129 | ||
130 | if (map->ops->map_release) | |
131 | map->ops->map_release(map, filp); | |
132 | ||
133 | bpf_map_put_with_uref(map); | |
99c55f7d AS |
134 | return 0; |
135 | } | |
136 | ||
f99bf205 DB |
137 | #ifdef CONFIG_PROC_FS |
138 | static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp) | |
139 | { | |
140 | const struct bpf_map *map = filp->private_data; | |
141 | ||
142 | seq_printf(m, | |
143 | "map_type:\t%u\n" | |
144 | "key_size:\t%u\n" | |
145 | "value_size:\t%u\n" | |
322cea2f DB |
146 | "max_entries:\t%u\n" |
147 | "map_flags:\t%#x\n", | |
f99bf205 DB |
148 | map->map_type, |
149 | map->key_size, | |
150 | map->value_size, | |
322cea2f DB |
151 | map->max_entries, |
152 | map->map_flags); | |
f99bf205 DB |
153 | } |
154 | #endif | |
155 | ||
99c55f7d | 156 | static const struct file_operations bpf_map_fops = { |
f99bf205 DB |
157 | #ifdef CONFIG_PROC_FS |
158 | .show_fdinfo = bpf_map_show_fdinfo, | |
159 | #endif | |
160 | .release = bpf_map_release, | |
99c55f7d AS |
161 | }; |
162 | ||
b2197755 | 163 | int bpf_map_new_fd(struct bpf_map *map) |
aa79781b DB |
164 | { |
165 | return anon_inode_getfd("bpf-map", &bpf_map_fops, map, | |
166 | O_RDWR | O_CLOEXEC); | |
167 | } | |
168 | ||
99c55f7d AS |
169 | /* helper macro to check that unused fields 'union bpf_attr' are zero */ |
170 | #define CHECK_ATTR(CMD) \ | |
171 | memchr_inv((void *) &attr->CMD##_LAST_FIELD + \ | |
172 | sizeof(attr->CMD##_LAST_FIELD), 0, \ | |
173 | sizeof(*attr) - \ | |
174 | offsetof(union bpf_attr, CMD##_LAST_FIELD) - \ | |
175 | sizeof(attr->CMD##_LAST_FIELD)) != NULL | |
176 | ||
6c905981 | 177 | #define BPF_MAP_CREATE_LAST_FIELD map_flags |
99c55f7d AS |
178 | /* called via syscall */ |
179 | static int map_create(union bpf_attr *attr) | |
180 | { | |
181 | struct bpf_map *map; | |
182 | int err; | |
183 | ||
184 | err = CHECK_ATTR(BPF_MAP_CREATE); | |
185 | if (err) | |
186 | return -EINVAL; | |
187 | ||
188 | /* find map type and init map: hashtable vs rbtree vs bloom vs ... */ | |
189 | map = find_and_alloc_map(attr); | |
190 | if (IS_ERR(map)) | |
191 | return PTR_ERR(map); | |
192 | ||
193 | atomic_set(&map->refcnt, 1); | |
c9da161c | 194 | atomic_set(&map->usercnt, 1); |
99c55f7d | 195 | |
aaac3ba9 AS |
196 | err = bpf_map_charge_memlock(map); |
197 | if (err) | |
20b2b24f | 198 | goto free_map_nouncharge; |
aaac3ba9 | 199 | |
aa79781b | 200 | err = bpf_map_new_fd(map); |
99c55f7d AS |
201 | if (err < 0) |
202 | /* failed to allocate fd */ | |
203 | goto free_map; | |
204 | ||
205 | return err; | |
206 | ||
207 | free_map: | |
20b2b24f DB |
208 | bpf_map_uncharge_memlock(map); |
209 | free_map_nouncharge: | |
99c55f7d AS |
210 | map->ops->map_free(map); |
211 | return err; | |
212 | } | |
213 | ||
db20fd2b AS |
214 | /* if error is returned, fd is released. |
215 | * On success caller should complete fd access with matching fdput() | |
216 | */ | |
c2101297 | 217 | struct bpf_map *__bpf_map_get(struct fd f) |
db20fd2b | 218 | { |
db20fd2b AS |
219 | if (!f.file) |
220 | return ERR_PTR(-EBADF); | |
db20fd2b AS |
221 | if (f.file->f_op != &bpf_map_fops) { |
222 | fdput(f); | |
223 | return ERR_PTR(-EINVAL); | |
224 | } | |
225 | ||
c2101297 DB |
226 | return f.file->private_data; |
227 | } | |
228 | ||
92117d84 AS |
229 | /* prog's and map's refcnt limit */ |
230 | #define BPF_MAX_REFCNT 32768 | |
231 | ||
232 | struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref) | |
c9da161c | 233 | { |
92117d84 AS |
234 | if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) { |
235 | atomic_dec(&map->refcnt); | |
236 | return ERR_PTR(-EBUSY); | |
237 | } | |
c9da161c DB |
238 | if (uref) |
239 | atomic_inc(&map->usercnt); | |
92117d84 | 240 | return map; |
c9da161c DB |
241 | } |
242 | ||
243 | struct bpf_map *bpf_map_get_with_uref(u32 ufd) | |
c2101297 DB |
244 | { |
245 | struct fd f = fdget(ufd); | |
246 | struct bpf_map *map; | |
247 | ||
248 | map = __bpf_map_get(f); | |
249 | if (IS_ERR(map)) | |
250 | return map; | |
251 | ||
92117d84 | 252 | map = bpf_map_inc(map, true); |
c2101297 | 253 | fdput(f); |
db20fd2b AS |
254 | |
255 | return map; | |
256 | } | |
257 | ||
b8cdc051 AS |
258 | int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value) |
259 | { | |
260 | return -ENOTSUPP; | |
261 | } | |
262 | ||
db20fd2b AS |
263 | /* last field in 'union bpf_attr' used by this command */ |
264 | #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value | |
265 | ||
266 | static int map_lookup_elem(union bpf_attr *attr) | |
267 | { | |
535e7b4b MS |
268 | void __user *ukey = u64_to_user_ptr(attr->key); |
269 | void __user *uvalue = u64_to_user_ptr(attr->value); | |
db20fd2b | 270 | int ufd = attr->map_fd; |
db20fd2b | 271 | struct bpf_map *map; |
8ebe667c | 272 | void *key, *value, *ptr; |
15a07b33 | 273 | u32 value_size; |
592867bf | 274 | struct fd f; |
db20fd2b AS |
275 | int err; |
276 | ||
277 | if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM)) | |
278 | return -EINVAL; | |
279 | ||
592867bf | 280 | f = fdget(ufd); |
c2101297 | 281 | map = __bpf_map_get(f); |
db20fd2b AS |
282 | if (IS_ERR(map)) |
283 | return PTR_ERR(map); | |
284 | ||
285 | err = -ENOMEM; | |
286 | key = kmalloc(map->key_size, GFP_USER); | |
287 | if (!key) | |
288 | goto err_put; | |
289 | ||
290 | err = -EFAULT; | |
291 | if (copy_from_user(key, ukey, map->key_size) != 0) | |
292 | goto free_key; | |
293 | ||
15a07b33 | 294 | if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || |
8f844938 | 295 | map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH || |
15a07b33 AS |
296 | map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) |
297 | value_size = round_up(map->value_size, 8) * num_possible_cpus(); | |
298 | else | |
299 | value_size = map->value_size; | |
300 | ||
8ebe667c | 301 | err = -ENOMEM; |
15a07b33 | 302 | value = kmalloc(value_size, GFP_USER | __GFP_NOWARN); |
db20fd2b | 303 | if (!value) |
8ebe667c AS |
304 | goto free_key; |
305 | ||
8f844938 MKL |
306 | if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || |
307 | map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) { | |
15a07b33 AS |
308 | err = bpf_percpu_hash_copy(map, key, value); |
309 | } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) { | |
310 | err = bpf_percpu_array_copy(map, key, value); | |
557c0c6e AS |
311 | } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) { |
312 | err = bpf_stackmap_copy(map, key, value); | |
15a07b33 AS |
313 | } else { |
314 | rcu_read_lock(); | |
315 | ptr = map->ops->map_lookup_elem(map, key); | |
316 | if (ptr) | |
317 | memcpy(value, ptr, value_size); | |
318 | rcu_read_unlock(); | |
319 | err = ptr ? 0 : -ENOENT; | |
320 | } | |
8ebe667c | 321 | |
15a07b33 | 322 | if (err) |
8ebe667c | 323 | goto free_value; |
db20fd2b AS |
324 | |
325 | err = -EFAULT; | |
15a07b33 | 326 | if (copy_to_user(uvalue, value, value_size) != 0) |
8ebe667c | 327 | goto free_value; |
db20fd2b AS |
328 | |
329 | err = 0; | |
330 | ||
8ebe667c AS |
331 | free_value: |
332 | kfree(value); | |
db20fd2b AS |
333 | free_key: |
334 | kfree(key); | |
335 | err_put: | |
336 | fdput(f); | |
337 | return err; | |
338 | } | |
339 | ||
3274f520 | 340 | #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags |
db20fd2b AS |
341 | |
342 | static int map_update_elem(union bpf_attr *attr) | |
343 | { | |
535e7b4b MS |
344 | void __user *ukey = u64_to_user_ptr(attr->key); |
345 | void __user *uvalue = u64_to_user_ptr(attr->value); | |
db20fd2b | 346 | int ufd = attr->map_fd; |
db20fd2b AS |
347 | struct bpf_map *map; |
348 | void *key, *value; | |
15a07b33 | 349 | u32 value_size; |
592867bf | 350 | struct fd f; |
db20fd2b AS |
351 | int err; |
352 | ||
353 | if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM)) | |
354 | return -EINVAL; | |
355 | ||
592867bf | 356 | f = fdget(ufd); |
c2101297 | 357 | map = __bpf_map_get(f); |
db20fd2b AS |
358 | if (IS_ERR(map)) |
359 | return PTR_ERR(map); | |
360 | ||
361 | err = -ENOMEM; | |
362 | key = kmalloc(map->key_size, GFP_USER); | |
363 | if (!key) | |
364 | goto err_put; | |
365 | ||
366 | err = -EFAULT; | |
367 | if (copy_from_user(key, ukey, map->key_size) != 0) | |
368 | goto free_key; | |
369 | ||
15a07b33 | 370 | if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || |
8f844938 | 371 | map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH || |
15a07b33 AS |
372 | map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) |
373 | value_size = round_up(map->value_size, 8) * num_possible_cpus(); | |
374 | else | |
375 | value_size = map->value_size; | |
376 | ||
db20fd2b | 377 | err = -ENOMEM; |
15a07b33 | 378 | value = kmalloc(value_size, GFP_USER | __GFP_NOWARN); |
db20fd2b AS |
379 | if (!value) |
380 | goto free_key; | |
381 | ||
382 | err = -EFAULT; | |
15a07b33 | 383 | if (copy_from_user(value, uvalue, value_size) != 0) |
db20fd2b AS |
384 | goto free_value; |
385 | ||
b121d1e7 AS |
386 | /* must increment bpf_prog_active to avoid kprobe+bpf triggering from |
387 | * inside bpf map update or delete otherwise deadlocks are possible | |
388 | */ | |
389 | preempt_disable(); | |
390 | __this_cpu_inc(bpf_prog_active); | |
8f844938 MKL |
391 | if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || |
392 | map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) { | |
15a07b33 AS |
393 | err = bpf_percpu_hash_update(map, key, value, attr->flags); |
394 | } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) { | |
395 | err = bpf_percpu_array_update(map, key, value, attr->flags); | |
d056a788 | 396 | } else if (map->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || |
4ed8ec52 MKL |
397 | map->map_type == BPF_MAP_TYPE_PROG_ARRAY || |
398 | map->map_type == BPF_MAP_TYPE_CGROUP_ARRAY) { | |
d056a788 DB |
399 | rcu_read_lock(); |
400 | err = bpf_fd_array_map_update_elem(map, f.file, key, value, | |
401 | attr->flags); | |
402 | rcu_read_unlock(); | |
15a07b33 AS |
403 | } else { |
404 | rcu_read_lock(); | |
405 | err = map->ops->map_update_elem(map, key, value, attr->flags); | |
406 | rcu_read_unlock(); | |
407 | } | |
b121d1e7 AS |
408 | __this_cpu_dec(bpf_prog_active); |
409 | preempt_enable(); | |
db20fd2b AS |
410 | |
411 | free_value: | |
412 | kfree(value); | |
413 | free_key: | |
414 | kfree(key); | |
415 | err_put: | |
416 | fdput(f); | |
417 | return err; | |
418 | } | |
419 | ||
420 | #define BPF_MAP_DELETE_ELEM_LAST_FIELD key | |
421 | ||
422 | static int map_delete_elem(union bpf_attr *attr) | |
423 | { | |
535e7b4b | 424 | void __user *ukey = u64_to_user_ptr(attr->key); |
db20fd2b | 425 | int ufd = attr->map_fd; |
db20fd2b | 426 | struct bpf_map *map; |
592867bf | 427 | struct fd f; |
db20fd2b AS |
428 | void *key; |
429 | int err; | |
430 | ||
431 | if (CHECK_ATTR(BPF_MAP_DELETE_ELEM)) | |
432 | return -EINVAL; | |
433 | ||
592867bf | 434 | f = fdget(ufd); |
c2101297 | 435 | map = __bpf_map_get(f); |
db20fd2b AS |
436 | if (IS_ERR(map)) |
437 | return PTR_ERR(map); | |
438 | ||
439 | err = -ENOMEM; | |
440 | key = kmalloc(map->key_size, GFP_USER); | |
441 | if (!key) | |
442 | goto err_put; | |
443 | ||
444 | err = -EFAULT; | |
445 | if (copy_from_user(key, ukey, map->key_size) != 0) | |
446 | goto free_key; | |
447 | ||
b121d1e7 AS |
448 | preempt_disable(); |
449 | __this_cpu_inc(bpf_prog_active); | |
db20fd2b AS |
450 | rcu_read_lock(); |
451 | err = map->ops->map_delete_elem(map, key); | |
452 | rcu_read_unlock(); | |
b121d1e7 AS |
453 | __this_cpu_dec(bpf_prog_active); |
454 | preempt_enable(); | |
db20fd2b AS |
455 | |
456 | free_key: | |
457 | kfree(key); | |
458 | err_put: | |
459 | fdput(f); | |
460 | return err; | |
461 | } | |
462 | ||
463 | /* last field in 'union bpf_attr' used by this command */ | |
464 | #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key | |
465 | ||
466 | static int map_get_next_key(union bpf_attr *attr) | |
467 | { | |
535e7b4b MS |
468 | void __user *ukey = u64_to_user_ptr(attr->key); |
469 | void __user *unext_key = u64_to_user_ptr(attr->next_key); | |
db20fd2b | 470 | int ufd = attr->map_fd; |
db20fd2b AS |
471 | struct bpf_map *map; |
472 | void *key, *next_key; | |
592867bf | 473 | struct fd f; |
db20fd2b AS |
474 | int err; |
475 | ||
476 | if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY)) | |
477 | return -EINVAL; | |
478 | ||
592867bf | 479 | f = fdget(ufd); |
c2101297 | 480 | map = __bpf_map_get(f); |
db20fd2b AS |
481 | if (IS_ERR(map)) |
482 | return PTR_ERR(map); | |
483 | ||
484 | err = -ENOMEM; | |
485 | key = kmalloc(map->key_size, GFP_USER); | |
486 | if (!key) | |
487 | goto err_put; | |
488 | ||
489 | err = -EFAULT; | |
490 | if (copy_from_user(key, ukey, map->key_size) != 0) | |
491 | goto free_key; | |
492 | ||
493 | err = -ENOMEM; | |
494 | next_key = kmalloc(map->key_size, GFP_USER); | |
495 | if (!next_key) | |
496 | goto free_key; | |
497 | ||
498 | rcu_read_lock(); | |
499 | err = map->ops->map_get_next_key(map, key, next_key); | |
500 | rcu_read_unlock(); | |
501 | if (err) | |
502 | goto free_next_key; | |
503 | ||
504 | err = -EFAULT; | |
505 | if (copy_to_user(unext_key, next_key, map->key_size) != 0) | |
506 | goto free_next_key; | |
507 | ||
508 | err = 0; | |
509 | ||
510 | free_next_key: | |
511 | kfree(next_key); | |
512 | free_key: | |
513 | kfree(key); | |
514 | err_put: | |
515 | fdput(f); | |
516 | return err; | |
517 | } | |
518 | ||
09756af4 AS |
519 | static LIST_HEAD(bpf_prog_types); |
520 | ||
521 | static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog) | |
522 | { | |
523 | struct bpf_prog_type_list *tl; | |
524 | ||
525 | list_for_each_entry(tl, &bpf_prog_types, list_node) { | |
526 | if (tl->type == type) { | |
527 | prog->aux->ops = tl->ops; | |
24701ece | 528 | prog->type = type; |
09756af4 AS |
529 | return 0; |
530 | } | |
531 | } | |
24701ece | 532 | |
09756af4 AS |
533 | return -EINVAL; |
534 | } | |
535 | ||
536 | void bpf_register_prog_type(struct bpf_prog_type_list *tl) | |
537 | { | |
538 | list_add(&tl->list_node, &bpf_prog_types); | |
539 | } | |
540 | ||
0a542a86 AS |
541 | /* fixup insn->imm field of bpf_call instructions: |
542 | * if (insn->imm == BPF_FUNC_map_lookup_elem) | |
543 | * insn->imm = bpf_map_lookup_elem - __bpf_call_base; | |
544 | * else if (insn->imm == BPF_FUNC_map_update_elem) | |
545 | * insn->imm = bpf_map_update_elem - __bpf_call_base; | |
546 | * else ... | |
547 | * | |
548 | * this function is called after eBPF program passed verification | |
549 | */ | |
550 | static void fixup_bpf_calls(struct bpf_prog *prog) | |
551 | { | |
552 | const struct bpf_func_proto *fn; | |
553 | int i; | |
554 | ||
555 | for (i = 0; i < prog->len; i++) { | |
556 | struct bpf_insn *insn = &prog->insnsi[i]; | |
557 | ||
558 | if (insn->code == (BPF_JMP | BPF_CALL)) { | |
559 | /* we reach here when program has bpf_call instructions | |
560 | * and it passed bpf_check(), means that | |
561 | * ops->get_func_proto must have been supplied, check it | |
562 | */ | |
563 | BUG_ON(!prog->aux->ops->get_func_proto); | |
564 | ||
c46646d0 DB |
565 | if (insn->imm == BPF_FUNC_get_route_realm) |
566 | prog->dst_needed = 1; | |
3ad00405 DB |
567 | if (insn->imm == BPF_FUNC_get_prandom_u32) |
568 | bpf_user_rnd_init_once(); | |
04fd61ab AS |
569 | if (insn->imm == BPF_FUNC_tail_call) { |
570 | /* mark bpf_tail_call as different opcode | |
571 | * to avoid conditional branch in | |
572 | * interpeter for every normal call | |
573 | * and to prevent accidental JITing by | |
574 | * JIT compiler that doesn't support | |
575 | * bpf_tail_call yet | |
576 | */ | |
577 | insn->imm = 0; | |
578 | insn->code |= BPF_X; | |
579 | continue; | |
580 | } | |
581 | ||
0a542a86 AS |
582 | fn = prog->aux->ops->get_func_proto(insn->imm); |
583 | /* all functions that have prototype and verifier allowed | |
584 | * programs to call them, must be real in-kernel functions | |
585 | */ | |
586 | BUG_ON(!fn->func); | |
587 | insn->imm = fn->func - __bpf_call_base; | |
588 | } | |
589 | } | |
590 | } | |
591 | ||
09756af4 AS |
592 | /* drop refcnt on maps used by eBPF program and free auxilary data */ |
593 | static void free_used_maps(struct bpf_prog_aux *aux) | |
594 | { | |
595 | int i; | |
596 | ||
597 | for (i = 0; i < aux->used_map_cnt; i++) | |
598 | bpf_map_put(aux->used_maps[i]); | |
599 | ||
600 | kfree(aux->used_maps); | |
601 | } | |
602 | ||
aaac3ba9 AS |
603 | static int bpf_prog_charge_memlock(struct bpf_prog *prog) |
604 | { | |
605 | struct user_struct *user = get_current_user(); | |
606 | unsigned long memlock_limit; | |
607 | ||
608 | memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
609 | ||
610 | atomic_long_add(prog->pages, &user->locked_vm); | |
611 | if (atomic_long_read(&user->locked_vm) > memlock_limit) { | |
612 | atomic_long_sub(prog->pages, &user->locked_vm); | |
613 | free_uid(user); | |
614 | return -EPERM; | |
615 | } | |
616 | prog->aux->user = user; | |
617 | return 0; | |
618 | } | |
619 | ||
620 | static void bpf_prog_uncharge_memlock(struct bpf_prog *prog) | |
621 | { | |
622 | struct user_struct *user = prog->aux->user; | |
623 | ||
624 | atomic_long_sub(prog->pages, &user->locked_vm); | |
625 | free_uid(user); | |
626 | } | |
627 | ||
1aacde3d | 628 | static void __bpf_prog_put_rcu(struct rcu_head *rcu) |
abf2e7d6 AS |
629 | { |
630 | struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu); | |
631 | ||
632 | free_used_maps(aux); | |
aaac3ba9 | 633 | bpf_prog_uncharge_memlock(aux->prog); |
abf2e7d6 AS |
634 | bpf_prog_free(aux->prog); |
635 | } | |
636 | ||
09756af4 AS |
637 | void bpf_prog_put(struct bpf_prog *prog) |
638 | { | |
e9d8afa9 | 639 | if (atomic_dec_and_test(&prog->aux->refcnt)) |
1aacde3d | 640 | call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu); |
09756af4 | 641 | } |
e2e9b654 | 642 | EXPORT_SYMBOL_GPL(bpf_prog_put); |
09756af4 AS |
643 | |
644 | static int bpf_prog_release(struct inode *inode, struct file *filp) | |
645 | { | |
646 | struct bpf_prog *prog = filp->private_data; | |
647 | ||
1aacde3d | 648 | bpf_prog_put(prog); |
09756af4 AS |
649 | return 0; |
650 | } | |
651 | ||
652 | static const struct file_operations bpf_prog_fops = { | |
653 | .release = bpf_prog_release, | |
654 | }; | |
655 | ||
b2197755 | 656 | int bpf_prog_new_fd(struct bpf_prog *prog) |
aa79781b DB |
657 | { |
658 | return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog, | |
659 | O_RDWR | O_CLOEXEC); | |
660 | } | |
661 | ||
113214be | 662 | static struct bpf_prog *____bpf_prog_get(struct fd f) |
09756af4 | 663 | { |
09756af4 AS |
664 | if (!f.file) |
665 | return ERR_PTR(-EBADF); | |
09756af4 AS |
666 | if (f.file->f_op != &bpf_prog_fops) { |
667 | fdput(f); | |
668 | return ERR_PTR(-EINVAL); | |
669 | } | |
670 | ||
c2101297 | 671 | return f.file->private_data; |
09756af4 AS |
672 | } |
673 | ||
59d3656d | 674 | struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i) |
92117d84 | 675 | { |
59d3656d BB |
676 | if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) { |
677 | atomic_sub(i, &prog->aux->refcnt); | |
92117d84 AS |
678 | return ERR_PTR(-EBUSY); |
679 | } | |
680 | return prog; | |
681 | } | |
59d3656d BB |
682 | EXPORT_SYMBOL_GPL(bpf_prog_add); |
683 | ||
c540594f DB |
684 | void bpf_prog_sub(struct bpf_prog *prog, int i) |
685 | { | |
686 | /* Only to be used for undoing previous bpf_prog_add() in some | |
687 | * error path. We still know that another entity in our call | |
688 | * path holds a reference to the program, thus atomic_sub() can | |
689 | * be safely used in such cases! | |
690 | */ | |
691 | WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0); | |
692 | } | |
693 | EXPORT_SYMBOL_GPL(bpf_prog_sub); | |
694 | ||
59d3656d BB |
695 | struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog) |
696 | { | |
697 | return bpf_prog_add(prog, 1); | |
698 | } | |
97bc402d | 699 | EXPORT_SYMBOL_GPL(bpf_prog_inc); |
92117d84 | 700 | |
113214be | 701 | static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *type) |
09756af4 AS |
702 | { |
703 | struct fd f = fdget(ufd); | |
704 | struct bpf_prog *prog; | |
705 | ||
113214be | 706 | prog = ____bpf_prog_get(f); |
09756af4 AS |
707 | if (IS_ERR(prog)) |
708 | return prog; | |
113214be DB |
709 | if (type && prog->type != *type) { |
710 | prog = ERR_PTR(-EINVAL); | |
711 | goto out; | |
712 | } | |
09756af4 | 713 | |
92117d84 | 714 | prog = bpf_prog_inc(prog); |
113214be | 715 | out: |
09756af4 AS |
716 | fdput(f); |
717 | return prog; | |
718 | } | |
113214be DB |
719 | |
720 | struct bpf_prog *bpf_prog_get(u32 ufd) | |
721 | { | |
722 | return __bpf_prog_get(ufd, NULL); | |
723 | } | |
724 | ||
725 | struct bpf_prog *bpf_prog_get_type(u32 ufd, enum bpf_prog_type type) | |
726 | { | |
727 | return __bpf_prog_get(ufd, &type); | |
728 | } | |
729 | EXPORT_SYMBOL_GPL(bpf_prog_get_type); | |
09756af4 AS |
730 | |
731 | /* last field in 'union bpf_attr' used by this command */ | |
2541517c | 732 | #define BPF_PROG_LOAD_LAST_FIELD kern_version |
09756af4 AS |
733 | |
734 | static int bpf_prog_load(union bpf_attr *attr) | |
735 | { | |
736 | enum bpf_prog_type type = attr->prog_type; | |
737 | struct bpf_prog *prog; | |
738 | int err; | |
739 | char license[128]; | |
740 | bool is_gpl; | |
741 | ||
742 | if (CHECK_ATTR(BPF_PROG_LOAD)) | |
743 | return -EINVAL; | |
744 | ||
745 | /* copy eBPF program license from user space */ | |
535e7b4b | 746 | if (strncpy_from_user(license, u64_to_user_ptr(attr->license), |
09756af4 AS |
747 | sizeof(license) - 1) < 0) |
748 | return -EFAULT; | |
749 | license[sizeof(license) - 1] = 0; | |
750 | ||
751 | /* eBPF programs must be GPL compatible to use GPL-ed functions */ | |
752 | is_gpl = license_is_gpl_compatible(license); | |
753 | ||
754 | if (attr->insn_cnt >= BPF_MAXINSNS) | |
755 | return -EINVAL; | |
756 | ||
2541517c AS |
757 | if (type == BPF_PROG_TYPE_KPROBE && |
758 | attr->kern_version != LINUX_VERSION_CODE) | |
759 | return -EINVAL; | |
760 | ||
1be7f75d AS |
761 | if (type != BPF_PROG_TYPE_SOCKET_FILTER && !capable(CAP_SYS_ADMIN)) |
762 | return -EPERM; | |
763 | ||
09756af4 AS |
764 | /* plain bpf_prog allocation */ |
765 | prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER); | |
766 | if (!prog) | |
767 | return -ENOMEM; | |
768 | ||
aaac3ba9 AS |
769 | err = bpf_prog_charge_memlock(prog); |
770 | if (err) | |
771 | goto free_prog_nouncharge; | |
772 | ||
09756af4 AS |
773 | prog->len = attr->insn_cnt; |
774 | ||
775 | err = -EFAULT; | |
535e7b4b | 776 | if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns), |
09756af4 AS |
777 | prog->len * sizeof(struct bpf_insn)) != 0) |
778 | goto free_prog; | |
779 | ||
780 | prog->orig_prog = NULL; | |
a91263d5 | 781 | prog->jited = 0; |
09756af4 AS |
782 | |
783 | atomic_set(&prog->aux->refcnt, 1); | |
a91263d5 | 784 | prog->gpl_compatible = is_gpl ? 1 : 0; |
09756af4 AS |
785 | |
786 | /* find program type: socket_filter vs tracing_filter */ | |
787 | err = find_prog_type(type, prog); | |
788 | if (err < 0) | |
789 | goto free_prog; | |
790 | ||
791 | /* run eBPF verifier */ | |
9bac3d6d | 792 | err = bpf_check(&prog, attr); |
09756af4 AS |
793 | if (err < 0) |
794 | goto free_used_maps; | |
795 | ||
0a542a86 AS |
796 | /* fixup BPF_CALL->imm field */ |
797 | fixup_bpf_calls(prog); | |
798 | ||
09756af4 | 799 | /* eBPF program is ready to be JITed */ |
d1c55ab5 | 800 | prog = bpf_prog_select_runtime(prog, &err); |
04fd61ab AS |
801 | if (err < 0) |
802 | goto free_used_maps; | |
09756af4 | 803 | |
aa79781b | 804 | err = bpf_prog_new_fd(prog); |
09756af4 AS |
805 | if (err < 0) |
806 | /* failed to allocate fd */ | |
807 | goto free_used_maps; | |
808 | ||
809 | return err; | |
810 | ||
811 | free_used_maps: | |
812 | free_used_maps(prog->aux); | |
813 | free_prog: | |
aaac3ba9 AS |
814 | bpf_prog_uncharge_memlock(prog); |
815 | free_prog_nouncharge: | |
09756af4 AS |
816 | bpf_prog_free(prog); |
817 | return err; | |
818 | } | |
819 | ||
b2197755 DB |
820 | #define BPF_OBJ_LAST_FIELD bpf_fd |
821 | ||
822 | static int bpf_obj_pin(const union bpf_attr *attr) | |
823 | { | |
824 | if (CHECK_ATTR(BPF_OBJ)) | |
825 | return -EINVAL; | |
826 | ||
535e7b4b | 827 | return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname)); |
b2197755 DB |
828 | } |
829 | ||
830 | static int bpf_obj_get(const union bpf_attr *attr) | |
831 | { | |
832 | if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0) | |
833 | return -EINVAL; | |
834 | ||
535e7b4b | 835 | return bpf_obj_get_user(u64_to_user_ptr(attr->pathname)); |
b2197755 DB |
836 | } |
837 | ||
99c55f7d AS |
838 | SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size) |
839 | { | |
840 | union bpf_attr attr = {}; | |
841 | int err; | |
842 | ||
1be7f75d | 843 | if (!capable(CAP_SYS_ADMIN) && sysctl_unprivileged_bpf_disabled) |
99c55f7d AS |
844 | return -EPERM; |
845 | ||
846 | if (!access_ok(VERIFY_READ, uattr, 1)) | |
847 | return -EFAULT; | |
848 | ||
849 | if (size > PAGE_SIZE) /* silly large */ | |
850 | return -E2BIG; | |
851 | ||
852 | /* If we're handed a bigger struct than we know of, | |
853 | * ensure all the unknown bits are 0 - i.e. new | |
854 | * user-space does not rely on any kernel feature | |
855 | * extensions we dont know about yet. | |
856 | */ | |
857 | if (size > sizeof(attr)) { | |
858 | unsigned char __user *addr; | |
859 | unsigned char __user *end; | |
860 | unsigned char val; | |
861 | ||
862 | addr = (void __user *)uattr + sizeof(attr); | |
863 | end = (void __user *)uattr + size; | |
864 | ||
865 | for (; addr < end; addr++) { | |
866 | err = get_user(val, addr); | |
867 | if (err) | |
868 | return err; | |
869 | if (val) | |
870 | return -E2BIG; | |
871 | } | |
872 | size = sizeof(attr); | |
873 | } | |
874 | ||
875 | /* copy attributes from user space, may be less than sizeof(bpf_attr) */ | |
876 | if (copy_from_user(&attr, uattr, size) != 0) | |
877 | return -EFAULT; | |
878 | ||
879 | switch (cmd) { | |
880 | case BPF_MAP_CREATE: | |
881 | err = map_create(&attr); | |
882 | break; | |
db20fd2b AS |
883 | case BPF_MAP_LOOKUP_ELEM: |
884 | err = map_lookup_elem(&attr); | |
885 | break; | |
886 | case BPF_MAP_UPDATE_ELEM: | |
887 | err = map_update_elem(&attr); | |
888 | break; | |
889 | case BPF_MAP_DELETE_ELEM: | |
890 | err = map_delete_elem(&attr); | |
891 | break; | |
892 | case BPF_MAP_GET_NEXT_KEY: | |
893 | err = map_get_next_key(&attr); | |
894 | break; | |
09756af4 AS |
895 | case BPF_PROG_LOAD: |
896 | err = bpf_prog_load(&attr); | |
897 | break; | |
b2197755 DB |
898 | case BPF_OBJ_PIN: |
899 | err = bpf_obj_pin(&attr); | |
900 | break; | |
901 | case BPF_OBJ_GET: | |
902 | err = bpf_obj_get(&attr); | |
903 | break; | |
99c55f7d AS |
904 | default: |
905 | err = -EINVAL; | |
906 | break; | |
907 | } | |
908 | ||
909 | return err; | |
910 | } |