]>
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 | { | |
61d1b6a4 DB |
127 | struct bpf_map *map = filp->private_data; |
128 | ||
129 | if (map->ops->map_release) | |
130 | map->ops->map_release(map, filp); | |
131 | ||
132 | bpf_map_put_with_uref(map); | |
99c55f7d AS |
133 | return 0; |
134 | } | |
135 | ||
f99bf205 DB |
136 | #ifdef CONFIG_PROC_FS |
137 | static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp) | |
138 | { | |
139 | const struct bpf_map *map = filp->private_data; | |
140 | ||
141 | seq_printf(m, | |
142 | "map_type:\t%u\n" | |
143 | "key_size:\t%u\n" | |
144 | "value_size:\t%u\n" | |
322cea2f DB |
145 | "max_entries:\t%u\n" |
146 | "map_flags:\t%#x\n", | |
f99bf205 DB |
147 | map->map_type, |
148 | map->key_size, | |
149 | map->value_size, | |
322cea2f DB |
150 | map->max_entries, |
151 | map->map_flags); | |
f99bf205 DB |
152 | } |
153 | #endif | |
154 | ||
99c55f7d | 155 | static const struct file_operations bpf_map_fops = { |
f99bf205 DB |
156 | #ifdef CONFIG_PROC_FS |
157 | .show_fdinfo = bpf_map_show_fdinfo, | |
158 | #endif | |
159 | .release = bpf_map_release, | |
99c55f7d AS |
160 | }; |
161 | ||
b2197755 | 162 | int bpf_map_new_fd(struct bpf_map *map) |
aa79781b DB |
163 | { |
164 | return anon_inode_getfd("bpf-map", &bpf_map_fops, map, | |
165 | O_RDWR | O_CLOEXEC); | |
166 | } | |
167 | ||
99c55f7d AS |
168 | /* helper macro to check that unused fields 'union bpf_attr' are zero */ |
169 | #define CHECK_ATTR(CMD) \ | |
170 | memchr_inv((void *) &attr->CMD##_LAST_FIELD + \ | |
171 | sizeof(attr->CMD##_LAST_FIELD), 0, \ | |
172 | sizeof(*attr) - \ | |
173 | offsetof(union bpf_attr, CMD##_LAST_FIELD) - \ | |
174 | sizeof(attr->CMD##_LAST_FIELD)) != NULL | |
175 | ||
6c905981 | 176 | #define BPF_MAP_CREATE_LAST_FIELD map_flags |
99c55f7d AS |
177 | /* called via syscall */ |
178 | static int map_create(union bpf_attr *attr) | |
179 | { | |
180 | struct bpf_map *map; | |
181 | int err; | |
182 | ||
183 | err = CHECK_ATTR(BPF_MAP_CREATE); | |
184 | if (err) | |
185 | return -EINVAL; | |
186 | ||
187 | /* find map type and init map: hashtable vs rbtree vs bloom vs ... */ | |
188 | map = find_and_alloc_map(attr); | |
189 | if (IS_ERR(map)) | |
190 | return PTR_ERR(map); | |
191 | ||
192 | atomic_set(&map->refcnt, 1); | |
c9da161c | 193 | atomic_set(&map->usercnt, 1); |
99c55f7d | 194 | |
aaac3ba9 AS |
195 | err = bpf_map_charge_memlock(map); |
196 | if (err) | |
20b2b24f | 197 | goto free_map_nouncharge; |
aaac3ba9 | 198 | |
aa79781b | 199 | err = bpf_map_new_fd(map); |
99c55f7d AS |
200 | if (err < 0) |
201 | /* failed to allocate fd */ | |
202 | goto free_map; | |
203 | ||
204 | return err; | |
205 | ||
206 | free_map: | |
20b2b24f DB |
207 | bpf_map_uncharge_memlock(map); |
208 | free_map_nouncharge: | |
99c55f7d AS |
209 | map->ops->map_free(map); |
210 | return err; | |
211 | } | |
212 | ||
db20fd2b AS |
213 | /* if error is returned, fd is released. |
214 | * On success caller should complete fd access with matching fdput() | |
215 | */ | |
c2101297 | 216 | struct bpf_map *__bpf_map_get(struct fd f) |
db20fd2b | 217 | { |
db20fd2b AS |
218 | if (!f.file) |
219 | return ERR_PTR(-EBADF); | |
db20fd2b AS |
220 | if (f.file->f_op != &bpf_map_fops) { |
221 | fdput(f); | |
222 | return ERR_PTR(-EINVAL); | |
223 | } | |
224 | ||
c2101297 DB |
225 | return f.file->private_data; |
226 | } | |
227 | ||
92117d84 AS |
228 | /* prog's and map's refcnt limit */ |
229 | #define BPF_MAX_REFCNT 32768 | |
230 | ||
231 | struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref) | |
c9da161c | 232 | { |
92117d84 AS |
233 | if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) { |
234 | atomic_dec(&map->refcnt); | |
235 | return ERR_PTR(-EBUSY); | |
236 | } | |
c9da161c DB |
237 | if (uref) |
238 | atomic_inc(&map->usercnt); | |
92117d84 | 239 | return map; |
c9da161c DB |
240 | } |
241 | ||
242 | struct bpf_map *bpf_map_get_with_uref(u32 ufd) | |
c2101297 DB |
243 | { |
244 | struct fd f = fdget(ufd); | |
245 | struct bpf_map *map; | |
246 | ||
247 | map = __bpf_map_get(f); | |
248 | if (IS_ERR(map)) | |
249 | return map; | |
250 | ||
92117d84 | 251 | map = bpf_map_inc(map, true); |
c2101297 | 252 | fdput(f); |
db20fd2b AS |
253 | |
254 | return map; | |
255 | } | |
256 | ||
257 | /* helper to convert user pointers passed inside __aligned_u64 fields */ | |
258 | static void __user *u64_to_ptr(__u64 val) | |
259 | { | |
260 | return (void __user *) (unsigned long) val; | |
261 | } | |
262 | ||
b8cdc051 AS |
263 | int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value) |
264 | { | |
265 | return -ENOTSUPP; | |
266 | } | |
267 | ||
db20fd2b AS |
268 | /* last field in 'union bpf_attr' used by this command */ |
269 | #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value | |
270 | ||
271 | static int map_lookup_elem(union bpf_attr *attr) | |
272 | { | |
273 | void __user *ukey = u64_to_ptr(attr->key); | |
274 | void __user *uvalue = u64_to_ptr(attr->value); | |
275 | int ufd = attr->map_fd; | |
db20fd2b | 276 | struct bpf_map *map; |
8ebe667c | 277 | void *key, *value, *ptr; |
15a07b33 | 278 | u32 value_size; |
592867bf | 279 | struct fd f; |
db20fd2b AS |
280 | int err; |
281 | ||
282 | if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM)) | |
283 | return -EINVAL; | |
284 | ||
592867bf | 285 | f = fdget(ufd); |
c2101297 | 286 | map = __bpf_map_get(f); |
db20fd2b AS |
287 | if (IS_ERR(map)) |
288 | return PTR_ERR(map); | |
289 | ||
290 | err = -ENOMEM; | |
291 | key = kmalloc(map->key_size, GFP_USER); | |
292 | if (!key) | |
293 | goto err_put; | |
294 | ||
295 | err = -EFAULT; | |
296 | if (copy_from_user(key, ukey, map->key_size) != 0) | |
297 | goto free_key; | |
298 | ||
15a07b33 AS |
299 | if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || |
300 | map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) | |
301 | value_size = round_up(map->value_size, 8) * num_possible_cpus(); | |
302 | else | |
303 | value_size = map->value_size; | |
304 | ||
8ebe667c | 305 | err = -ENOMEM; |
15a07b33 | 306 | value = kmalloc(value_size, GFP_USER | __GFP_NOWARN); |
db20fd2b | 307 | if (!value) |
8ebe667c AS |
308 | goto free_key; |
309 | ||
15a07b33 AS |
310 | if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH) { |
311 | err = bpf_percpu_hash_copy(map, key, value); | |
312 | } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) { | |
313 | err = bpf_percpu_array_copy(map, key, value); | |
557c0c6e AS |
314 | } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) { |
315 | err = bpf_stackmap_copy(map, key, value); | |
15a07b33 AS |
316 | } else { |
317 | rcu_read_lock(); | |
318 | ptr = map->ops->map_lookup_elem(map, key); | |
319 | if (ptr) | |
320 | memcpy(value, ptr, value_size); | |
321 | rcu_read_unlock(); | |
322 | err = ptr ? 0 : -ENOENT; | |
323 | } | |
8ebe667c | 324 | |
15a07b33 | 325 | if (err) |
8ebe667c | 326 | goto free_value; |
db20fd2b AS |
327 | |
328 | err = -EFAULT; | |
15a07b33 | 329 | if (copy_to_user(uvalue, value, value_size) != 0) |
8ebe667c | 330 | goto free_value; |
db20fd2b AS |
331 | |
332 | err = 0; | |
333 | ||
8ebe667c AS |
334 | free_value: |
335 | kfree(value); | |
db20fd2b AS |
336 | free_key: |
337 | kfree(key); | |
338 | err_put: | |
339 | fdput(f); | |
340 | return err; | |
341 | } | |
342 | ||
3274f520 | 343 | #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags |
db20fd2b AS |
344 | |
345 | static int map_update_elem(union bpf_attr *attr) | |
346 | { | |
347 | void __user *ukey = u64_to_ptr(attr->key); | |
348 | void __user *uvalue = u64_to_ptr(attr->value); | |
349 | int ufd = attr->map_fd; | |
db20fd2b AS |
350 | struct bpf_map *map; |
351 | void *key, *value; | |
15a07b33 | 352 | u32 value_size; |
592867bf | 353 | struct fd f; |
db20fd2b AS |
354 | int err; |
355 | ||
356 | if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM)) | |
357 | return -EINVAL; | |
358 | ||
592867bf | 359 | f = fdget(ufd); |
c2101297 | 360 | map = __bpf_map_get(f); |
db20fd2b AS |
361 | if (IS_ERR(map)) |
362 | return PTR_ERR(map); | |
363 | ||
364 | err = -ENOMEM; | |
365 | key = kmalloc(map->key_size, GFP_USER); | |
366 | if (!key) | |
367 | goto err_put; | |
368 | ||
369 | err = -EFAULT; | |
370 | if (copy_from_user(key, ukey, map->key_size) != 0) | |
371 | goto free_key; | |
372 | ||
15a07b33 AS |
373 | if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || |
374 | map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) | |
375 | value_size = round_up(map->value_size, 8) * num_possible_cpus(); | |
376 | else | |
377 | value_size = map->value_size; | |
378 | ||
db20fd2b | 379 | err = -ENOMEM; |
15a07b33 | 380 | value = kmalloc(value_size, GFP_USER | __GFP_NOWARN); |
db20fd2b AS |
381 | if (!value) |
382 | goto free_key; | |
383 | ||
384 | err = -EFAULT; | |
15a07b33 | 385 | if (copy_from_user(value, uvalue, value_size) != 0) |
db20fd2b AS |
386 | goto free_value; |
387 | ||
b121d1e7 AS |
388 | /* must increment bpf_prog_active to avoid kprobe+bpf triggering from |
389 | * inside bpf map update or delete otherwise deadlocks are possible | |
390 | */ | |
391 | preempt_disable(); | |
392 | __this_cpu_inc(bpf_prog_active); | |
15a07b33 AS |
393 | if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH) { |
394 | err = bpf_percpu_hash_update(map, key, value, attr->flags); | |
395 | } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) { | |
396 | err = bpf_percpu_array_update(map, key, value, attr->flags); | |
d056a788 | 397 | } else if (map->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || |
4ed8ec52 MKL |
398 | map->map_type == BPF_MAP_TYPE_PROG_ARRAY || |
399 | map->map_type == BPF_MAP_TYPE_CGROUP_ARRAY) { | |
d056a788 DB |
400 | rcu_read_lock(); |
401 | err = bpf_fd_array_map_update_elem(map, f.file, key, value, | |
402 | attr->flags); | |
403 | rcu_read_unlock(); | |
15a07b33 AS |
404 | } else { |
405 | rcu_read_lock(); | |
406 | err = map->ops->map_update_elem(map, key, value, attr->flags); | |
407 | rcu_read_unlock(); | |
408 | } | |
b121d1e7 AS |
409 | __this_cpu_dec(bpf_prog_active); |
410 | preempt_enable(); | |
db20fd2b AS |
411 | |
412 | free_value: | |
413 | kfree(value); | |
414 | free_key: | |
415 | kfree(key); | |
416 | err_put: | |
417 | fdput(f); | |
418 | return err; | |
419 | } | |
420 | ||
421 | #define BPF_MAP_DELETE_ELEM_LAST_FIELD key | |
422 | ||
423 | static int map_delete_elem(union bpf_attr *attr) | |
424 | { | |
425 | void __user *ukey = u64_to_ptr(attr->key); | |
426 | int ufd = attr->map_fd; | |
db20fd2b | 427 | struct bpf_map *map; |
592867bf | 428 | struct fd f; |
db20fd2b AS |
429 | void *key; |
430 | int err; | |
431 | ||
432 | if (CHECK_ATTR(BPF_MAP_DELETE_ELEM)) | |
433 | return -EINVAL; | |
434 | ||
592867bf | 435 | f = fdget(ufd); |
c2101297 | 436 | map = __bpf_map_get(f); |
db20fd2b AS |
437 | if (IS_ERR(map)) |
438 | return PTR_ERR(map); | |
439 | ||
440 | err = -ENOMEM; | |
441 | key = kmalloc(map->key_size, GFP_USER); | |
442 | if (!key) | |
443 | goto err_put; | |
444 | ||
445 | err = -EFAULT; | |
446 | if (copy_from_user(key, ukey, map->key_size) != 0) | |
447 | goto free_key; | |
448 | ||
b121d1e7 AS |
449 | preempt_disable(); |
450 | __this_cpu_inc(bpf_prog_active); | |
db20fd2b AS |
451 | rcu_read_lock(); |
452 | err = map->ops->map_delete_elem(map, key); | |
453 | rcu_read_unlock(); | |
b121d1e7 AS |
454 | __this_cpu_dec(bpf_prog_active); |
455 | preempt_enable(); | |
db20fd2b AS |
456 | |
457 | free_key: | |
458 | kfree(key); | |
459 | err_put: | |
460 | fdput(f); | |
461 | return err; | |
462 | } | |
463 | ||
464 | /* last field in 'union bpf_attr' used by this command */ | |
465 | #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key | |
466 | ||
467 | static int map_get_next_key(union bpf_attr *attr) | |
468 | { | |
469 | void __user *ukey = u64_to_ptr(attr->key); | |
470 | void __user *unext_key = u64_to_ptr(attr->next_key); | |
471 | int ufd = attr->map_fd; | |
db20fd2b AS |
472 | struct bpf_map *map; |
473 | void *key, *next_key; | |
592867bf | 474 | struct fd f; |
db20fd2b AS |
475 | int err; |
476 | ||
477 | if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY)) | |
478 | return -EINVAL; | |
479 | ||
592867bf | 480 | f = fdget(ufd); |
c2101297 | 481 | map = __bpf_map_get(f); |
db20fd2b AS |
482 | if (IS_ERR(map)) |
483 | return PTR_ERR(map); | |
484 | ||
485 | err = -ENOMEM; | |
486 | key = kmalloc(map->key_size, GFP_USER); | |
487 | if (!key) | |
488 | goto err_put; | |
489 | ||
490 | err = -EFAULT; | |
491 | if (copy_from_user(key, ukey, map->key_size) != 0) | |
492 | goto free_key; | |
493 | ||
494 | err = -ENOMEM; | |
495 | next_key = kmalloc(map->key_size, GFP_USER); | |
496 | if (!next_key) | |
497 | goto free_key; | |
498 | ||
499 | rcu_read_lock(); | |
500 | err = map->ops->map_get_next_key(map, key, next_key); | |
501 | rcu_read_unlock(); | |
502 | if (err) | |
503 | goto free_next_key; | |
504 | ||
505 | err = -EFAULT; | |
506 | if (copy_to_user(unext_key, next_key, map->key_size) != 0) | |
507 | goto free_next_key; | |
508 | ||
509 | err = 0; | |
510 | ||
511 | free_next_key: | |
512 | kfree(next_key); | |
513 | free_key: | |
514 | kfree(key); | |
515 | err_put: | |
516 | fdput(f); | |
517 | return err; | |
518 | } | |
519 | ||
09756af4 AS |
520 | static LIST_HEAD(bpf_prog_types); |
521 | ||
522 | static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog) | |
523 | { | |
524 | struct bpf_prog_type_list *tl; | |
525 | ||
526 | list_for_each_entry(tl, &bpf_prog_types, list_node) { | |
527 | if (tl->type == type) { | |
528 | prog->aux->ops = tl->ops; | |
24701ece | 529 | prog->type = type; |
09756af4 AS |
530 | return 0; |
531 | } | |
532 | } | |
24701ece | 533 | |
09756af4 AS |
534 | return -EINVAL; |
535 | } | |
536 | ||
537 | void bpf_register_prog_type(struct bpf_prog_type_list *tl) | |
538 | { | |
539 | list_add(&tl->list_node, &bpf_prog_types); | |
540 | } | |
541 | ||
0a542a86 AS |
542 | /* fixup insn->imm field of bpf_call instructions: |
543 | * if (insn->imm == BPF_FUNC_map_lookup_elem) | |
544 | * insn->imm = bpf_map_lookup_elem - __bpf_call_base; | |
545 | * else if (insn->imm == BPF_FUNC_map_update_elem) | |
546 | * insn->imm = bpf_map_update_elem - __bpf_call_base; | |
547 | * else ... | |
548 | * | |
549 | * this function is called after eBPF program passed verification | |
550 | */ | |
551 | static void fixup_bpf_calls(struct bpf_prog *prog) | |
552 | { | |
553 | const struct bpf_func_proto *fn; | |
554 | int i; | |
555 | ||
556 | for (i = 0; i < prog->len; i++) { | |
557 | struct bpf_insn *insn = &prog->insnsi[i]; | |
558 | ||
559 | if (insn->code == (BPF_JMP | BPF_CALL)) { | |
560 | /* we reach here when program has bpf_call instructions | |
561 | * and it passed bpf_check(), means that | |
562 | * ops->get_func_proto must have been supplied, check it | |
563 | */ | |
564 | BUG_ON(!prog->aux->ops->get_func_proto); | |
565 | ||
c46646d0 DB |
566 | if (insn->imm == BPF_FUNC_get_route_realm) |
567 | prog->dst_needed = 1; | |
3ad00405 DB |
568 | if (insn->imm == BPF_FUNC_get_prandom_u32) |
569 | bpf_user_rnd_init_once(); | |
04fd61ab AS |
570 | if (insn->imm == BPF_FUNC_tail_call) { |
571 | /* mark bpf_tail_call as different opcode | |
572 | * to avoid conditional branch in | |
573 | * interpeter for every normal call | |
574 | * and to prevent accidental JITing by | |
575 | * JIT compiler that doesn't support | |
576 | * bpf_tail_call yet | |
577 | */ | |
578 | insn->imm = 0; | |
579 | insn->code |= BPF_X; | |
580 | continue; | |
581 | } | |
582 | ||
0a542a86 AS |
583 | fn = prog->aux->ops->get_func_proto(insn->imm); |
584 | /* all functions that have prototype and verifier allowed | |
585 | * programs to call them, must be real in-kernel functions | |
586 | */ | |
587 | BUG_ON(!fn->func); | |
588 | insn->imm = fn->func - __bpf_call_base; | |
589 | } | |
590 | } | |
591 | } | |
592 | ||
09756af4 AS |
593 | /* drop refcnt on maps used by eBPF program and free auxilary data */ |
594 | static void free_used_maps(struct bpf_prog_aux *aux) | |
595 | { | |
596 | int i; | |
597 | ||
598 | for (i = 0; i < aux->used_map_cnt; i++) | |
599 | bpf_map_put(aux->used_maps[i]); | |
600 | ||
601 | kfree(aux->used_maps); | |
602 | } | |
603 | ||
aaac3ba9 AS |
604 | static int bpf_prog_charge_memlock(struct bpf_prog *prog) |
605 | { | |
606 | struct user_struct *user = get_current_user(); | |
607 | unsigned long memlock_limit; | |
608 | ||
609 | memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
610 | ||
611 | atomic_long_add(prog->pages, &user->locked_vm); | |
612 | if (atomic_long_read(&user->locked_vm) > memlock_limit) { | |
613 | atomic_long_sub(prog->pages, &user->locked_vm); | |
614 | free_uid(user); | |
615 | return -EPERM; | |
616 | } | |
617 | prog->aux->user = user; | |
618 | return 0; | |
619 | } | |
620 | ||
621 | static void bpf_prog_uncharge_memlock(struct bpf_prog *prog) | |
622 | { | |
623 | struct user_struct *user = prog->aux->user; | |
624 | ||
625 | atomic_long_sub(prog->pages, &user->locked_vm); | |
626 | free_uid(user); | |
627 | } | |
628 | ||
1aacde3d | 629 | static void __bpf_prog_put_rcu(struct rcu_head *rcu) |
abf2e7d6 AS |
630 | { |
631 | struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu); | |
632 | ||
633 | free_used_maps(aux); | |
aaac3ba9 | 634 | bpf_prog_uncharge_memlock(aux->prog); |
abf2e7d6 AS |
635 | bpf_prog_free(aux->prog); |
636 | } | |
637 | ||
09756af4 AS |
638 | void bpf_prog_put(struct bpf_prog *prog) |
639 | { | |
e9d8afa9 | 640 | if (atomic_dec_and_test(&prog->aux->refcnt)) |
1aacde3d | 641 | call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu); |
09756af4 | 642 | } |
e2e9b654 | 643 | EXPORT_SYMBOL_GPL(bpf_prog_put); |
09756af4 AS |
644 | |
645 | static int bpf_prog_release(struct inode *inode, struct file *filp) | |
646 | { | |
647 | struct bpf_prog *prog = filp->private_data; | |
648 | ||
1aacde3d | 649 | bpf_prog_put(prog); |
09756af4 AS |
650 | return 0; |
651 | } | |
652 | ||
653 | static const struct file_operations bpf_prog_fops = { | |
654 | .release = bpf_prog_release, | |
655 | }; | |
656 | ||
b2197755 | 657 | int bpf_prog_new_fd(struct bpf_prog *prog) |
aa79781b DB |
658 | { |
659 | return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog, | |
660 | O_RDWR | O_CLOEXEC); | |
661 | } | |
662 | ||
113214be | 663 | static struct bpf_prog *____bpf_prog_get(struct fd f) |
09756af4 | 664 | { |
09756af4 AS |
665 | if (!f.file) |
666 | return ERR_PTR(-EBADF); | |
09756af4 AS |
667 | if (f.file->f_op != &bpf_prog_fops) { |
668 | fdput(f); | |
669 | return ERR_PTR(-EINVAL); | |
670 | } | |
671 | ||
c2101297 | 672 | return f.file->private_data; |
09756af4 AS |
673 | } |
674 | ||
59d3656d | 675 | struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i) |
92117d84 | 676 | { |
59d3656d BB |
677 | if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) { |
678 | atomic_sub(i, &prog->aux->refcnt); | |
92117d84 AS |
679 | return ERR_PTR(-EBUSY); |
680 | } | |
681 | return prog; | |
682 | } | |
59d3656d BB |
683 | EXPORT_SYMBOL_GPL(bpf_prog_add); |
684 | ||
685 | struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog) | |
686 | { | |
687 | return bpf_prog_add(prog, 1); | |
688 | } | |
92117d84 | 689 | |
113214be | 690 | static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *type) |
09756af4 AS |
691 | { |
692 | struct fd f = fdget(ufd); | |
693 | struct bpf_prog *prog; | |
694 | ||
113214be | 695 | prog = ____bpf_prog_get(f); |
09756af4 AS |
696 | if (IS_ERR(prog)) |
697 | return prog; | |
113214be DB |
698 | if (type && prog->type != *type) { |
699 | prog = ERR_PTR(-EINVAL); | |
700 | goto out; | |
701 | } | |
09756af4 | 702 | |
92117d84 | 703 | prog = bpf_prog_inc(prog); |
113214be | 704 | out: |
09756af4 AS |
705 | fdput(f); |
706 | return prog; | |
707 | } | |
113214be DB |
708 | |
709 | struct bpf_prog *bpf_prog_get(u32 ufd) | |
710 | { | |
711 | return __bpf_prog_get(ufd, NULL); | |
712 | } | |
713 | ||
714 | struct bpf_prog *bpf_prog_get_type(u32 ufd, enum bpf_prog_type type) | |
715 | { | |
716 | return __bpf_prog_get(ufd, &type); | |
717 | } | |
718 | EXPORT_SYMBOL_GPL(bpf_prog_get_type); | |
09756af4 AS |
719 | |
720 | /* last field in 'union bpf_attr' used by this command */ | |
2541517c | 721 | #define BPF_PROG_LOAD_LAST_FIELD kern_version |
09756af4 AS |
722 | |
723 | static int bpf_prog_load(union bpf_attr *attr) | |
724 | { | |
725 | enum bpf_prog_type type = attr->prog_type; | |
726 | struct bpf_prog *prog; | |
727 | int err; | |
728 | char license[128]; | |
729 | bool is_gpl; | |
730 | ||
731 | if (CHECK_ATTR(BPF_PROG_LOAD)) | |
732 | return -EINVAL; | |
733 | ||
734 | /* copy eBPF program license from user space */ | |
735 | if (strncpy_from_user(license, u64_to_ptr(attr->license), | |
736 | sizeof(license) - 1) < 0) | |
737 | return -EFAULT; | |
738 | license[sizeof(license) - 1] = 0; | |
739 | ||
740 | /* eBPF programs must be GPL compatible to use GPL-ed functions */ | |
741 | is_gpl = license_is_gpl_compatible(license); | |
742 | ||
743 | if (attr->insn_cnt >= BPF_MAXINSNS) | |
744 | return -EINVAL; | |
745 | ||
2541517c AS |
746 | if (type == BPF_PROG_TYPE_KPROBE && |
747 | attr->kern_version != LINUX_VERSION_CODE) | |
748 | return -EINVAL; | |
749 | ||
1be7f75d AS |
750 | if (type != BPF_PROG_TYPE_SOCKET_FILTER && !capable(CAP_SYS_ADMIN)) |
751 | return -EPERM; | |
752 | ||
09756af4 AS |
753 | /* plain bpf_prog allocation */ |
754 | prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER); | |
755 | if (!prog) | |
756 | return -ENOMEM; | |
757 | ||
aaac3ba9 AS |
758 | err = bpf_prog_charge_memlock(prog); |
759 | if (err) | |
760 | goto free_prog_nouncharge; | |
761 | ||
09756af4 AS |
762 | prog->len = attr->insn_cnt; |
763 | ||
764 | err = -EFAULT; | |
765 | if (copy_from_user(prog->insns, u64_to_ptr(attr->insns), | |
766 | prog->len * sizeof(struct bpf_insn)) != 0) | |
767 | goto free_prog; | |
768 | ||
769 | prog->orig_prog = NULL; | |
a91263d5 | 770 | prog->jited = 0; |
09756af4 AS |
771 | |
772 | atomic_set(&prog->aux->refcnt, 1); | |
a91263d5 | 773 | prog->gpl_compatible = is_gpl ? 1 : 0; |
09756af4 AS |
774 | |
775 | /* find program type: socket_filter vs tracing_filter */ | |
776 | err = find_prog_type(type, prog); | |
777 | if (err < 0) | |
778 | goto free_prog; | |
779 | ||
780 | /* run eBPF verifier */ | |
9bac3d6d | 781 | err = bpf_check(&prog, attr); |
09756af4 AS |
782 | if (err < 0) |
783 | goto free_used_maps; | |
784 | ||
0a542a86 AS |
785 | /* fixup BPF_CALL->imm field */ |
786 | fixup_bpf_calls(prog); | |
787 | ||
09756af4 | 788 | /* eBPF program is ready to be JITed */ |
d1c55ab5 | 789 | prog = bpf_prog_select_runtime(prog, &err); |
04fd61ab AS |
790 | if (err < 0) |
791 | goto free_used_maps; | |
09756af4 | 792 | |
aa79781b | 793 | err = bpf_prog_new_fd(prog); |
09756af4 AS |
794 | if (err < 0) |
795 | /* failed to allocate fd */ | |
796 | goto free_used_maps; | |
797 | ||
798 | return err; | |
799 | ||
800 | free_used_maps: | |
801 | free_used_maps(prog->aux); | |
802 | free_prog: | |
aaac3ba9 AS |
803 | bpf_prog_uncharge_memlock(prog); |
804 | free_prog_nouncharge: | |
09756af4 AS |
805 | bpf_prog_free(prog); |
806 | return err; | |
807 | } | |
808 | ||
b2197755 DB |
809 | #define BPF_OBJ_LAST_FIELD bpf_fd |
810 | ||
811 | static int bpf_obj_pin(const union bpf_attr *attr) | |
812 | { | |
813 | if (CHECK_ATTR(BPF_OBJ)) | |
814 | return -EINVAL; | |
815 | ||
816 | return bpf_obj_pin_user(attr->bpf_fd, u64_to_ptr(attr->pathname)); | |
817 | } | |
818 | ||
819 | static int bpf_obj_get(const union bpf_attr *attr) | |
820 | { | |
821 | if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0) | |
822 | return -EINVAL; | |
823 | ||
824 | return bpf_obj_get_user(u64_to_ptr(attr->pathname)); | |
825 | } | |
826 | ||
99c55f7d AS |
827 | SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size) |
828 | { | |
829 | union bpf_attr attr = {}; | |
830 | int err; | |
831 | ||
1be7f75d | 832 | if (!capable(CAP_SYS_ADMIN) && sysctl_unprivileged_bpf_disabled) |
99c55f7d AS |
833 | return -EPERM; |
834 | ||
835 | if (!access_ok(VERIFY_READ, uattr, 1)) | |
836 | return -EFAULT; | |
837 | ||
838 | if (size > PAGE_SIZE) /* silly large */ | |
839 | return -E2BIG; | |
840 | ||
841 | /* If we're handed a bigger struct than we know of, | |
842 | * ensure all the unknown bits are 0 - i.e. new | |
843 | * user-space does not rely on any kernel feature | |
844 | * extensions we dont know about yet. | |
845 | */ | |
846 | if (size > sizeof(attr)) { | |
847 | unsigned char __user *addr; | |
848 | unsigned char __user *end; | |
849 | unsigned char val; | |
850 | ||
851 | addr = (void __user *)uattr + sizeof(attr); | |
852 | end = (void __user *)uattr + size; | |
853 | ||
854 | for (; addr < end; addr++) { | |
855 | err = get_user(val, addr); | |
856 | if (err) | |
857 | return err; | |
858 | if (val) | |
859 | return -E2BIG; | |
860 | } | |
861 | size = sizeof(attr); | |
862 | } | |
863 | ||
864 | /* copy attributes from user space, may be less than sizeof(bpf_attr) */ | |
865 | if (copy_from_user(&attr, uattr, size) != 0) | |
866 | return -EFAULT; | |
867 | ||
868 | switch (cmd) { | |
869 | case BPF_MAP_CREATE: | |
870 | err = map_create(&attr); | |
871 | break; | |
db20fd2b AS |
872 | case BPF_MAP_LOOKUP_ELEM: |
873 | err = map_lookup_elem(&attr); | |
874 | break; | |
875 | case BPF_MAP_UPDATE_ELEM: | |
876 | err = map_update_elem(&attr); | |
877 | break; | |
878 | case BPF_MAP_DELETE_ELEM: | |
879 | err = map_delete_elem(&attr); | |
880 | break; | |
881 | case BPF_MAP_GET_NEXT_KEY: | |
882 | err = map_get_next_key(&attr); | |
883 | break; | |
09756af4 AS |
884 | case BPF_PROG_LOAD: |
885 | err = bpf_prog_load(&attr); | |
886 | break; | |
b2197755 DB |
887 | case BPF_OBJ_PIN: |
888 | err = bpf_obj_pin(&attr); | |
889 | break; | |
890 | case BPF_OBJ_GET: | |
891 | err = bpf_obj_get(&attr); | |
892 | break; | |
99c55f7d AS |
893 | default: |
894 | err = -EINVAL; | |
895 | break; | |
896 | } | |
897 | ||
898 | return err; | |
899 | } |