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