]>
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> | |
a67edbf4 | 13 | #include <linux/bpf_trace.h> |
99c55f7d AS |
14 | #include <linux/syscalls.h> |
15 | #include <linux/slab.h> | |
3f07c014 | 16 | #include <linux/sched/signal.h> |
d407bd25 DB |
17 | #include <linux/vmalloc.h> |
18 | #include <linux/mmzone.h> | |
99c55f7d | 19 | #include <linux/anon_inodes.h> |
db20fd2b | 20 | #include <linux/file.h> |
09756af4 AS |
21 | #include <linux/license.h> |
22 | #include <linux/filter.h> | |
2541517c | 23 | #include <linux/version.h> |
535e7b4b | 24 | #include <linux/kernel.h> |
dc4bb0e2 | 25 | #include <linux/idr.h> |
99c55f7d | 26 | |
14dc6f04 MKL |
27 | #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY || \ |
28 | (map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \ | |
29 | (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \ | |
30 | (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS) | |
31 | #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) | |
32 | #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_HASH(map)) | |
33 | ||
b121d1e7 | 34 | DEFINE_PER_CPU(int, bpf_prog_active); |
dc4bb0e2 MKL |
35 | static DEFINE_IDR(prog_idr); |
36 | static DEFINE_SPINLOCK(prog_idr_lock); | |
f3f1c054 MKL |
37 | static DEFINE_IDR(map_idr); |
38 | static DEFINE_SPINLOCK(map_idr_lock); | |
b121d1e7 | 39 | |
1be7f75d AS |
40 | int sysctl_unprivileged_bpf_disabled __read_mostly; |
41 | ||
40077e0c JB |
42 | static const struct bpf_map_ops * const bpf_map_types[] = { |
43 | #define BPF_PROG_TYPE(_id, _ops) | |
44 | #define BPF_MAP_TYPE(_id, _ops) \ | |
45 | [_id] = &_ops, | |
46 | #include <linux/bpf_types.h> | |
47 | #undef BPF_PROG_TYPE | |
48 | #undef BPF_MAP_TYPE | |
49 | }; | |
99c55f7d AS |
50 | |
51 | static struct bpf_map *find_and_alloc_map(union bpf_attr *attr) | |
52 | { | |
99c55f7d AS |
53 | struct bpf_map *map; |
54 | ||
40077e0c JB |
55 | if (attr->map_type >= ARRAY_SIZE(bpf_map_types) || |
56 | !bpf_map_types[attr->map_type]) | |
57 | return ERR_PTR(-EINVAL); | |
99c55f7d | 58 | |
40077e0c JB |
59 | map = bpf_map_types[attr->map_type]->map_alloc(attr); |
60 | if (IS_ERR(map)) | |
61 | return map; | |
62 | map->ops = bpf_map_types[attr->map_type]; | |
63 | map->map_type = attr->map_type; | |
64 | return map; | |
99c55f7d AS |
65 | } |
66 | ||
d407bd25 DB |
67 | void *bpf_map_area_alloc(size_t size) |
68 | { | |
69 | /* We definitely need __GFP_NORETRY, so OOM killer doesn't | |
70 | * trigger under memory pressure as we really just want to | |
71 | * fail instead. | |
72 | */ | |
73 | const gfp_t flags = __GFP_NOWARN | __GFP_NORETRY | __GFP_ZERO; | |
74 | void *area; | |
75 | ||
76 | if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) { | |
77 | area = kmalloc(size, GFP_USER | flags); | |
78 | if (area != NULL) | |
79 | return area; | |
80 | } | |
81 | ||
19809c2d | 82 | return __vmalloc(size, GFP_KERNEL | flags, PAGE_KERNEL); |
d407bd25 DB |
83 | } |
84 | ||
85 | void bpf_map_area_free(void *area) | |
86 | { | |
87 | kvfree(area); | |
88 | } | |
89 | ||
6c905981 AS |
90 | int bpf_map_precharge_memlock(u32 pages) |
91 | { | |
92 | struct user_struct *user = get_current_user(); | |
93 | unsigned long memlock_limit, cur; | |
94 | ||
95 | memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
96 | cur = atomic_long_read(&user->locked_vm); | |
97 | free_uid(user); | |
98 | if (cur + pages > memlock_limit) | |
99 | return -EPERM; | |
100 | return 0; | |
101 | } | |
102 | ||
aaac3ba9 AS |
103 | static int bpf_map_charge_memlock(struct bpf_map *map) |
104 | { | |
105 | struct user_struct *user = get_current_user(); | |
106 | unsigned long memlock_limit; | |
107 | ||
108 | memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
109 | ||
110 | atomic_long_add(map->pages, &user->locked_vm); | |
111 | ||
112 | if (atomic_long_read(&user->locked_vm) > memlock_limit) { | |
113 | atomic_long_sub(map->pages, &user->locked_vm); | |
114 | free_uid(user); | |
115 | return -EPERM; | |
116 | } | |
117 | map->user = user; | |
118 | return 0; | |
119 | } | |
120 | ||
121 | static void bpf_map_uncharge_memlock(struct bpf_map *map) | |
122 | { | |
123 | struct user_struct *user = map->user; | |
124 | ||
125 | atomic_long_sub(map->pages, &user->locked_vm); | |
126 | free_uid(user); | |
127 | } | |
128 | ||
f3f1c054 MKL |
129 | static int bpf_map_alloc_id(struct bpf_map *map) |
130 | { | |
131 | int id; | |
132 | ||
133 | spin_lock_bh(&map_idr_lock); | |
134 | id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC); | |
135 | if (id > 0) | |
136 | map->id = id; | |
137 | spin_unlock_bh(&map_idr_lock); | |
138 | ||
139 | if (WARN_ON_ONCE(!id)) | |
140 | return -ENOSPC; | |
141 | ||
142 | return id > 0 ? 0 : id; | |
143 | } | |
144 | ||
bd5f5f4e | 145 | static void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock) |
f3f1c054 | 146 | { |
bd5f5f4e MKL |
147 | if (do_idr_lock) |
148 | spin_lock_bh(&map_idr_lock); | |
149 | else | |
150 | __acquire(&map_idr_lock); | |
151 | ||
f3f1c054 | 152 | idr_remove(&map_idr, map->id); |
bd5f5f4e MKL |
153 | |
154 | if (do_idr_lock) | |
155 | spin_unlock_bh(&map_idr_lock); | |
156 | else | |
157 | __release(&map_idr_lock); | |
f3f1c054 MKL |
158 | } |
159 | ||
99c55f7d AS |
160 | /* called from workqueue */ |
161 | static void bpf_map_free_deferred(struct work_struct *work) | |
162 | { | |
163 | struct bpf_map *map = container_of(work, struct bpf_map, work); | |
164 | ||
aaac3ba9 | 165 | bpf_map_uncharge_memlock(map); |
99c55f7d AS |
166 | /* implementation dependent freeing */ |
167 | map->ops->map_free(map); | |
168 | } | |
169 | ||
c9da161c DB |
170 | static void bpf_map_put_uref(struct bpf_map *map) |
171 | { | |
172 | if (atomic_dec_and_test(&map->usercnt)) { | |
173 | if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) | |
174 | bpf_fd_array_map_clear(map); | |
175 | } | |
176 | } | |
177 | ||
99c55f7d AS |
178 | /* decrement map refcnt and schedule it for freeing via workqueue |
179 | * (unrelying map implementation ops->map_free() might sleep) | |
180 | */ | |
bd5f5f4e | 181 | static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock) |
99c55f7d AS |
182 | { |
183 | if (atomic_dec_and_test(&map->refcnt)) { | |
34ad5580 | 184 | /* bpf_map_free_id() must be called first */ |
bd5f5f4e | 185 | bpf_map_free_id(map, do_idr_lock); |
99c55f7d AS |
186 | INIT_WORK(&map->work, bpf_map_free_deferred); |
187 | schedule_work(&map->work); | |
188 | } | |
189 | } | |
190 | ||
bd5f5f4e MKL |
191 | void bpf_map_put(struct bpf_map *map) |
192 | { | |
193 | __bpf_map_put(map, true); | |
194 | } | |
195 | ||
c9da161c | 196 | void bpf_map_put_with_uref(struct bpf_map *map) |
99c55f7d | 197 | { |
c9da161c | 198 | bpf_map_put_uref(map); |
99c55f7d | 199 | bpf_map_put(map); |
c9da161c DB |
200 | } |
201 | ||
202 | static int bpf_map_release(struct inode *inode, struct file *filp) | |
203 | { | |
61d1b6a4 DB |
204 | struct bpf_map *map = filp->private_data; |
205 | ||
206 | if (map->ops->map_release) | |
207 | map->ops->map_release(map, filp); | |
208 | ||
209 | bpf_map_put_with_uref(map); | |
99c55f7d AS |
210 | return 0; |
211 | } | |
212 | ||
f99bf205 DB |
213 | #ifdef CONFIG_PROC_FS |
214 | static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp) | |
215 | { | |
216 | const struct bpf_map *map = filp->private_data; | |
21116b70 DB |
217 | const struct bpf_array *array; |
218 | u32 owner_prog_type = 0; | |
9780c0ab | 219 | u32 owner_jited = 0; |
21116b70 DB |
220 | |
221 | if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) { | |
222 | array = container_of(map, struct bpf_array, map); | |
223 | owner_prog_type = array->owner_prog_type; | |
9780c0ab | 224 | owner_jited = array->owner_jited; |
21116b70 | 225 | } |
f99bf205 DB |
226 | |
227 | seq_printf(m, | |
228 | "map_type:\t%u\n" | |
229 | "key_size:\t%u\n" | |
230 | "value_size:\t%u\n" | |
322cea2f | 231 | "max_entries:\t%u\n" |
21116b70 DB |
232 | "map_flags:\t%#x\n" |
233 | "memlock:\t%llu\n", | |
f99bf205 DB |
234 | map->map_type, |
235 | map->key_size, | |
236 | map->value_size, | |
322cea2f | 237 | map->max_entries, |
21116b70 DB |
238 | map->map_flags, |
239 | map->pages * 1ULL << PAGE_SHIFT); | |
240 | ||
9780c0ab | 241 | if (owner_prog_type) { |
21116b70 DB |
242 | seq_printf(m, "owner_prog_type:\t%u\n", |
243 | owner_prog_type); | |
9780c0ab DB |
244 | seq_printf(m, "owner_jited:\t%u\n", |
245 | owner_jited); | |
246 | } | |
f99bf205 DB |
247 | } |
248 | #endif | |
249 | ||
99c55f7d | 250 | static const struct file_operations bpf_map_fops = { |
f99bf205 DB |
251 | #ifdef CONFIG_PROC_FS |
252 | .show_fdinfo = bpf_map_show_fdinfo, | |
253 | #endif | |
254 | .release = bpf_map_release, | |
99c55f7d AS |
255 | }; |
256 | ||
b2197755 | 257 | int bpf_map_new_fd(struct bpf_map *map) |
aa79781b DB |
258 | { |
259 | return anon_inode_getfd("bpf-map", &bpf_map_fops, map, | |
260 | O_RDWR | O_CLOEXEC); | |
261 | } | |
262 | ||
99c55f7d AS |
263 | /* helper macro to check that unused fields 'union bpf_attr' are zero */ |
264 | #define CHECK_ATTR(CMD) \ | |
265 | memchr_inv((void *) &attr->CMD##_LAST_FIELD + \ | |
266 | sizeof(attr->CMD##_LAST_FIELD), 0, \ | |
267 | sizeof(*attr) - \ | |
268 | offsetof(union bpf_attr, CMD##_LAST_FIELD) - \ | |
269 | sizeof(attr->CMD##_LAST_FIELD)) != NULL | |
270 | ||
56f668df | 271 | #define BPF_MAP_CREATE_LAST_FIELD inner_map_fd |
99c55f7d AS |
272 | /* called via syscall */ |
273 | static int map_create(union bpf_attr *attr) | |
274 | { | |
275 | struct bpf_map *map; | |
276 | int err; | |
277 | ||
278 | err = CHECK_ATTR(BPF_MAP_CREATE); | |
279 | if (err) | |
280 | return -EINVAL; | |
281 | ||
282 | /* find map type and init map: hashtable vs rbtree vs bloom vs ... */ | |
283 | map = find_and_alloc_map(attr); | |
284 | if (IS_ERR(map)) | |
285 | return PTR_ERR(map); | |
286 | ||
287 | atomic_set(&map->refcnt, 1); | |
c9da161c | 288 | atomic_set(&map->usercnt, 1); |
99c55f7d | 289 | |
aaac3ba9 AS |
290 | err = bpf_map_charge_memlock(map); |
291 | if (err) | |
20b2b24f | 292 | goto free_map_nouncharge; |
aaac3ba9 | 293 | |
f3f1c054 MKL |
294 | err = bpf_map_alloc_id(map); |
295 | if (err) | |
296 | goto free_map; | |
297 | ||
aa79781b | 298 | err = bpf_map_new_fd(map); |
bd5f5f4e MKL |
299 | if (err < 0) { |
300 | /* failed to allocate fd. | |
301 | * bpf_map_put() is needed because the above | |
302 | * bpf_map_alloc_id() has published the map | |
303 | * to the userspace and the userspace may | |
304 | * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID. | |
305 | */ | |
306 | bpf_map_put(map); | |
307 | return err; | |
308 | } | |
99c55f7d | 309 | |
a67edbf4 | 310 | trace_bpf_map_create(map, err); |
99c55f7d AS |
311 | return err; |
312 | ||
313 | free_map: | |
20b2b24f DB |
314 | bpf_map_uncharge_memlock(map); |
315 | free_map_nouncharge: | |
99c55f7d AS |
316 | map->ops->map_free(map); |
317 | return err; | |
318 | } | |
319 | ||
db20fd2b AS |
320 | /* if error is returned, fd is released. |
321 | * On success caller should complete fd access with matching fdput() | |
322 | */ | |
c2101297 | 323 | struct bpf_map *__bpf_map_get(struct fd f) |
db20fd2b | 324 | { |
db20fd2b AS |
325 | if (!f.file) |
326 | return ERR_PTR(-EBADF); | |
db20fd2b AS |
327 | if (f.file->f_op != &bpf_map_fops) { |
328 | fdput(f); | |
329 | return ERR_PTR(-EINVAL); | |
330 | } | |
331 | ||
c2101297 DB |
332 | return f.file->private_data; |
333 | } | |
334 | ||
92117d84 AS |
335 | /* prog's and map's refcnt limit */ |
336 | #define BPF_MAX_REFCNT 32768 | |
337 | ||
338 | struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref) | |
c9da161c | 339 | { |
92117d84 AS |
340 | if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) { |
341 | atomic_dec(&map->refcnt); | |
342 | return ERR_PTR(-EBUSY); | |
343 | } | |
c9da161c DB |
344 | if (uref) |
345 | atomic_inc(&map->usercnt); | |
92117d84 | 346 | return map; |
c9da161c DB |
347 | } |
348 | ||
349 | struct bpf_map *bpf_map_get_with_uref(u32 ufd) | |
c2101297 DB |
350 | { |
351 | struct fd f = fdget(ufd); | |
352 | struct bpf_map *map; | |
353 | ||
354 | map = __bpf_map_get(f); | |
355 | if (IS_ERR(map)) | |
356 | return map; | |
357 | ||
92117d84 | 358 | map = bpf_map_inc(map, true); |
c2101297 | 359 | fdput(f); |
db20fd2b AS |
360 | |
361 | return map; | |
362 | } | |
363 | ||
bd5f5f4e MKL |
364 | /* map_idr_lock should have been held */ |
365 | static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map, | |
366 | bool uref) | |
367 | { | |
368 | int refold; | |
369 | ||
370 | refold = __atomic_add_unless(&map->refcnt, 1, 0); | |
371 | ||
372 | if (refold >= BPF_MAX_REFCNT) { | |
373 | __bpf_map_put(map, false); | |
374 | return ERR_PTR(-EBUSY); | |
375 | } | |
376 | ||
377 | if (!refold) | |
378 | return ERR_PTR(-ENOENT); | |
379 | ||
380 | if (uref) | |
381 | atomic_inc(&map->usercnt); | |
382 | ||
383 | return map; | |
384 | } | |
385 | ||
b8cdc051 AS |
386 | int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value) |
387 | { | |
388 | return -ENOTSUPP; | |
389 | } | |
390 | ||
db20fd2b AS |
391 | /* last field in 'union bpf_attr' used by this command */ |
392 | #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value | |
393 | ||
394 | static int map_lookup_elem(union bpf_attr *attr) | |
395 | { | |
535e7b4b MS |
396 | void __user *ukey = u64_to_user_ptr(attr->key); |
397 | void __user *uvalue = u64_to_user_ptr(attr->value); | |
db20fd2b | 398 | int ufd = attr->map_fd; |
db20fd2b | 399 | struct bpf_map *map; |
8ebe667c | 400 | void *key, *value, *ptr; |
15a07b33 | 401 | u32 value_size; |
592867bf | 402 | struct fd f; |
db20fd2b AS |
403 | int err; |
404 | ||
405 | if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM)) | |
406 | return -EINVAL; | |
407 | ||
592867bf | 408 | f = fdget(ufd); |
c2101297 | 409 | map = __bpf_map_get(f); |
db20fd2b AS |
410 | if (IS_ERR(map)) |
411 | return PTR_ERR(map); | |
412 | ||
e4448ed8 AV |
413 | key = memdup_user(ukey, map->key_size); |
414 | if (IS_ERR(key)) { | |
415 | err = PTR_ERR(key); | |
db20fd2b | 416 | goto err_put; |
e4448ed8 | 417 | } |
db20fd2b | 418 | |
15a07b33 | 419 | if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || |
8f844938 | 420 | map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH || |
15a07b33 AS |
421 | map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) |
422 | value_size = round_up(map->value_size, 8) * num_possible_cpus(); | |
14dc6f04 MKL |
423 | else if (IS_FD_MAP(map)) |
424 | value_size = sizeof(u32); | |
15a07b33 AS |
425 | else |
426 | value_size = map->value_size; | |
427 | ||
8ebe667c | 428 | err = -ENOMEM; |
15a07b33 | 429 | value = kmalloc(value_size, GFP_USER | __GFP_NOWARN); |
db20fd2b | 430 | if (!value) |
8ebe667c AS |
431 | goto free_key; |
432 | ||
8f844938 MKL |
433 | if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || |
434 | map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) { | |
15a07b33 AS |
435 | err = bpf_percpu_hash_copy(map, key, value); |
436 | } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) { | |
437 | err = bpf_percpu_array_copy(map, key, value); | |
557c0c6e AS |
438 | } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) { |
439 | err = bpf_stackmap_copy(map, key, value); | |
14dc6f04 MKL |
440 | } else if (IS_FD_ARRAY(map)) { |
441 | err = bpf_fd_array_map_lookup_elem(map, key, value); | |
442 | } else if (IS_FD_HASH(map)) { | |
443 | err = bpf_fd_htab_map_lookup_elem(map, key, value); | |
15a07b33 AS |
444 | } else { |
445 | rcu_read_lock(); | |
446 | ptr = map->ops->map_lookup_elem(map, key); | |
447 | if (ptr) | |
448 | memcpy(value, ptr, value_size); | |
449 | rcu_read_unlock(); | |
450 | err = ptr ? 0 : -ENOENT; | |
451 | } | |
8ebe667c | 452 | |
15a07b33 | 453 | if (err) |
8ebe667c | 454 | goto free_value; |
db20fd2b AS |
455 | |
456 | err = -EFAULT; | |
15a07b33 | 457 | if (copy_to_user(uvalue, value, value_size) != 0) |
8ebe667c | 458 | goto free_value; |
db20fd2b | 459 | |
a67edbf4 | 460 | trace_bpf_map_lookup_elem(map, ufd, key, value); |
db20fd2b AS |
461 | err = 0; |
462 | ||
8ebe667c AS |
463 | free_value: |
464 | kfree(value); | |
db20fd2b AS |
465 | free_key: |
466 | kfree(key); | |
467 | err_put: | |
468 | fdput(f); | |
469 | return err; | |
470 | } | |
471 | ||
3274f520 | 472 | #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags |
db20fd2b AS |
473 | |
474 | static int map_update_elem(union bpf_attr *attr) | |
475 | { | |
535e7b4b MS |
476 | void __user *ukey = u64_to_user_ptr(attr->key); |
477 | void __user *uvalue = u64_to_user_ptr(attr->value); | |
db20fd2b | 478 | int ufd = attr->map_fd; |
db20fd2b AS |
479 | struct bpf_map *map; |
480 | void *key, *value; | |
15a07b33 | 481 | u32 value_size; |
592867bf | 482 | struct fd f; |
db20fd2b AS |
483 | int err; |
484 | ||
485 | if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM)) | |
486 | return -EINVAL; | |
487 | ||
592867bf | 488 | f = fdget(ufd); |
c2101297 | 489 | map = __bpf_map_get(f); |
db20fd2b AS |
490 | if (IS_ERR(map)) |
491 | return PTR_ERR(map); | |
492 | ||
e4448ed8 AV |
493 | key = memdup_user(ukey, map->key_size); |
494 | if (IS_ERR(key)) { | |
495 | err = PTR_ERR(key); | |
db20fd2b | 496 | goto err_put; |
e4448ed8 | 497 | } |
db20fd2b | 498 | |
15a07b33 | 499 | if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || |
8f844938 | 500 | map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH || |
15a07b33 AS |
501 | map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) |
502 | value_size = round_up(map->value_size, 8) * num_possible_cpus(); | |
503 | else | |
504 | value_size = map->value_size; | |
505 | ||
db20fd2b | 506 | err = -ENOMEM; |
15a07b33 | 507 | value = kmalloc(value_size, GFP_USER | __GFP_NOWARN); |
db20fd2b AS |
508 | if (!value) |
509 | goto free_key; | |
510 | ||
511 | err = -EFAULT; | |
15a07b33 | 512 | if (copy_from_user(value, uvalue, value_size) != 0) |
db20fd2b AS |
513 | goto free_value; |
514 | ||
b121d1e7 AS |
515 | /* must increment bpf_prog_active to avoid kprobe+bpf triggering from |
516 | * inside bpf map update or delete otherwise deadlocks are possible | |
517 | */ | |
518 | preempt_disable(); | |
519 | __this_cpu_inc(bpf_prog_active); | |
8f844938 MKL |
520 | if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || |
521 | map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) { | |
15a07b33 AS |
522 | err = bpf_percpu_hash_update(map, key, value, attr->flags); |
523 | } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) { | |
524 | err = bpf_percpu_array_update(map, key, value, attr->flags); | |
d056a788 | 525 | } else if (map->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || |
4ed8ec52 | 526 | map->map_type == BPF_MAP_TYPE_PROG_ARRAY || |
56f668df MKL |
527 | map->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || |
528 | map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS) { | |
d056a788 DB |
529 | rcu_read_lock(); |
530 | err = bpf_fd_array_map_update_elem(map, f.file, key, value, | |
531 | attr->flags); | |
532 | rcu_read_unlock(); | |
bcc6b1b7 MKL |
533 | } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) { |
534 | rcu_read_lock(); | |
535 | err = bpf_fd_htab_map_update_elem(map, f.file, key, value, | |
536 | attr->flags); | |
537 | rcu_read_unlock(); | |
15a07b33 AS |
538 | } else { |
539 | rcu_read_lock(); | |
540 | err = map->ops->map_update_elem(map, key, value, attr->flags); | |
541 | rcu_read_unlock(); | |
542 | } | |
b121d1e7 AS |
543 | __this_cpu_dec(bpf_prog_active); |
544 | preempt_enable(); | |
db20fd2b | 545 | |
a67edbf4 DB |
546 | if (!err) |
547 | trace_bpf_map_update_elem(map, ufd, key, value); | |
db20fd2b AS |
548 | free_value: |
549 | kfree(value); | |
550 | free_key: | |
551 | kfree(key); | |
552 | err_put: | |
553 | fdput(f); | |
554 | return err; | |
555 | } | |
556 | ||
557 | #define BPF_MAP_DELETE_ELEM_LAST_FIELD key | |
558 | ||
559 | static int map_delete_elem(union bpf_attr *attr) | |
560 | { | |
535e7b4b | 561 | void __user *ukey = u64_to_user_ptr(attr->key); |
db20fd2b | 562 | int ufd = attr->map_fd; |
db20fd2b | 563 | struct bpf_map *map; |
592867bf | 564 | struct fd f; |
db20fd2b AS |
565 | void *key; |
566 | int err; | |
567 | ||
568 | if (CHECK_ATTR(BPF_MAP_DELETE_ELEM)) | |
569 | return -EINVAL; | |
570 | ||
592867bf | 571 | f = fdget(ufd); |
c2101297 | 572 | map = __bpf_map_get(f); |
db20fd2b AS |
573 | if (IS_ERR(map)) |
574 | return PTR_ERR(map); | |
575 | ||
e4448ed8 AV |
576 | key = memdup_user(ukey, map->key_size); |
577 | if (IS_ERR(key)) { | |
578 | err = PTR_ERR(key); | |
db20fd2b | 579 | goto err_put; |
e4448ed8 | 580 | } |
db20fd2b | 581 | |
b121d1e7 AS |
582 | preempt_disable(); |
583 | __this_cpu_inc(bpf_prog_active); | |
db20fd2b AS |
584 | rcu_read_lock(); |
585 | err = map->ops->map_delete_elem(map, key); | |
586 | rcu_read_unlock(); | |
b121d1e7 AS |
587 | __this_cpu_dec(bpf_prog_active); |
588 | preempt_enable(); | |
db20fd2b | 589 | |
a67edbf4 DB |
590 | if (!err) |
591 | trace_bpf_map_delete_elem(map, ufd, key); | |
db20fd2b AS |
592 | kfree(key); |
593 | err_put: | |
594 | fdput(f); | |
595 | return err; | |
596 | } | |
597 | ||
598 | /* last field in 'union bpf_attr' used by this command */ | |
599 | #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key | |
600 | ||
601 | static int map_get_next_key(union bpf_attr *attr) | |
602 | { | |
535e7b4b MS |
603 | void __user *ukey = u64_to_user_ptr(attr->key); |
604 | void __user *unext_key = u64_to_user_ptr(attr->next_key); | |
db20fd2b | 605 | int ufd = attr->map_fd; |
db20fd2b AS |
606 | struct bpf_map *map; |
607 | void *key, *next_key; | |
592867bf | 608 | struct fd f; |
db20fd2b AS |
609 | int err; |
610 | ||
611 | if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY)) | |
612 | return -EINVAL; | |
613 | ||
592867bf | 614 | f = fdget(ufd); |
c2101297 | 615 | map = __bpf_map_get(f); |
db20fd2b AS |
616 | if (IS_ERR(map)) |
617 | return PTR_ERR(map); | |
618 | ||
8fe45924 | 619 | if (ukey) { |
e4448ed8 AV |
620 | key = memdup_user(ukey, map->key_size); |
621 | if (IS_ERR(key)) { | |
622 | err = PTR_ERR(key); | |
8fe45924 | 623 | goto err_put; |
e4448ed8 | 624 | } |
8fe45924 TQ |
625 | } else { |
626 | key = NULL; | |
627 | } | |
db20fd2b AS |
628 | |
629 | err = -ENOMEM; | |
630 | next_key = kmalloc(map->key_size, GFP_USER); | |
631 | if (!next_key) | |
632 | goto free_key; | |
633 | ||
634 | rcu_read_lock(); | |
635 | err = map->ops->map_get_next_key(map, key, next_key); | |
636 | rcu_read_unlock(); | |
637 | if (err) | |
638 | goto free_next_key; | |
639 | ||
640 | err = -EFAULT; | |
641 | if (copy_to_user(unext_key, next_key, map->key_size) != 0) | |
642 | goto free_next_key; | |
643 | ||
a67edbf4 | 644 | trace_bpf_map_next_key(map, ufd, key, next_key); |
db20fd2b AS |
645 | err = 0; |
646 | ||
647 | free_next_key: | |
648 | kfree(next_key); | |
649 | free_key: | |
650 | kfree(key); | |
651 | err_put: | |
652 | fdput(f); | |
653 | return err; | |
654 | } | |
655 | ||
be9370a7 JB |
656 | static const struct bpf_verifier_ops * const bpf_prog_types[] = { |
657 | #define BPF_PROG_TYPE(_id, _ops) \ | |
658 | [_id] = &_ops, | |
40077e0c | 659 | #define BPF_MAP_TYPE(_id, _ops) |
be9370a7 JB |
660 | #include <linux/bpf_types.h> |
661 | #undef BPF_PROG_TYPE | |
40077e0c | 662 | #undef BPF_MAP_TYPE |
be9370a7 | 663 | }; |
09756af4 AS |
664 | |
665 | static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog) | |
666 | { | |
be9370a7 JB |
667 | if (type >= ARRAY_SIZE(bpf_prog_types) || !bpf_prog_types[type]) |
668 | return -EINVAL; | |
09756af4 | 669 | |
be9370a7 JB |
670 | prog->aux->ops = bpf_prog_types[type]; |
671 | prog->type = type; | |
672 | return 0; | |
09756af4 AS |
673 | } |
674 | ||
675 | /* drop refcnt on maps used by eBPF program and free auxilary data */ | |
676 | static void free_used_maps(struct bpf_prog_aux *aux) | |
677 | { | |
678 | int i; | |
679 | ||
680 | for (i = 0; i < aux->used_map_cnt; i++) | |
681 | bpf_map_put(aux->used_maps[i]); | |
682 | ||
683 | kfree(aux->used_maps); | |
684 | } | |
685 | ||
5ccb071e DB |
686 | int __bpf_prog_charge(struct user_struct *user, u32 pages) |
687 | { | |
688 | unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
689 | unsigned long user_bufs; | |
690 | ||
691 | if (user) { | |
692 | user_bufs = atomic_long_add_return(pages, &user->locked_vm); | |
693 | if (user_bufs > memlock_limit) { | |
694 | atomic_long_sub(pages, &user->locked_vm); | |
695 | return -EPERM; | |
696 | } | |
697 | } | |
698 | ||
699 | return 0; | |
700 | } | |
701 | ||
702 | void __bpf_prog_uncharge(struct user_struct *user, u32 pages) | |
703 | { | |
704 | if (user) | |
705 | atomic_long_sub(pages, &user->locked_vm); | |
706 | } | |
707 | ||
aaac3ba9 AS |
708 | static int bpf_prog_charge_memlock(struct bpf_prog *prog) |
709 | { | |
710 | struct user_struct *user = get_current_user(); | |
5ccb071e | 711 | int ret; |
aaac3ba9 | 712 | |
5ccb071e DB |
713 | ret = __bpf_prog_charge(user, prog->pages); |
714 | if (ret) { | |
aaac3ba9 | 715 | free_uid(user); |
5ccb071e | 716 | return ret; |
aaac3ba9 | 717 | } |
5ccb071e | 718 | |
aaac3ba9 AS |
719 | prog->aux->user = user; |
720 | return 0; | |
721 | } | |
722 | ||
723 | static void bpf_prog_uncharge_memlock(struct bpf_prog *prog) | |
724 | { | |
725 | struct user_struct *user = prog->aux->user; | |
726 | ||
5ccb071e | 727 | __bpf_prog_uncharge(user, prog->pages); |
aaac3ba9 AS |
728 | free_uid(user); |
729 | } | |
730 | ||
dc4bb0e2 MKL |
731 | static int bpf_prog_alloc_id(struct bpf_prog *prog) |
732 | { | |
733 | int id; | |
734 | ||
735 | spin_lock_bh(&prog_idr_lock); | |
736 | id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC); | |
737 | if (id > 0) | |
738 | prog->aux->id = id; | |
739 | spin_unlock_bh(&prog_idr_lock); | |
740 | ||
741 | /* id is in [1, INT_MAX) */ | |
742 | if (WARN_ON_ONCE(!id)) | |
743 | return -ENOSPC; | |
744 | ||
745 | return id > 0 ? 0 : id; | |
746 | } | |
747 | ||
b16d9aa4 | 748 | static void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock) |
dc4bb0e2 MKL |
749 | { |
750 | /* cBPF to eBPF migrations are currently not in the idr store. */ | |
751 | if (!prog->aux->id) | |
752 | return; | |
753 | ||
b16d9aa4 MKL |
754 | if (do_idr_lock) |
755 | spin_lock_bh(&prog_idr_lock); | |
756 | else | |
757 | __acquire(&prog_idr_lock); | |
758 | ||
dc4bb0e2 | 759 | idr_remove(&prog_idr, prog->aux->id); |
b16d9aa4 MKL |
760 | |
761 | if (do_idr_lock) | |
762 | spin_unlock_bh(&prog_idr_lock); | |
763 | else | |
764 | __release(&prog_idr_lock); | |
dc4bb0e2 MKL |
765 | } |
766 | ||
1aacde3d | 767 | static void __bpf_prog_put_rcu(struct rcu_head *rcu) |
abf2e7d6 AS |
768 | { |
769 | struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu); | |
770 | ||
771 | free_used_maps(aux); | |
aaac3ba9 | 772 | bpf_prog_uncharge_memlock(aux->prog); |
abf2e7d6 AS |
773 | bpf_prog_free(aux->prog); |
774 | } | |
775 | ||
b16d9aa4 | 776 | static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock) |
09756af4 | 777 | { |
a67edbf4 DB |
778 | if (atomic_dec_and_test(&prog->aux->refcnt)) { |
779 | trace_bpf_prog_put_rcu(prog); | |
34ad5580 | 780 | /* bpf_prog_free_id() must be called first */ |
b16d9aa4 | 781 | bpf_prog_free_id(prog, do_idr_lock); |
74451e66 | 782 | bpf_prog_kallsyms_del(prog); |
1aacde3d | 783 | call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu); |
a67edbf4 | 784 | } |
09756af4 | 785 | } |
b16d9aa4 MKL |
786 | |
787 | void bpf_prog_put(struct bpf_prog *prog) | |
788 | { | |
789 | __bpf_prog_put(prog, true); | |
790 | } | |
e2e9b654 | 791 | EXPORT_SYMBOL_GPL(bpf_prog_put); |
09756af4 AS |
792 | |
793 | static int bpf_prog_release(struct inode *inode, struct file *filp) | |
794 | { | |
795 | struct bpf_prog *prog = filp->private_data; | |
796 | ||
1aacde3d | 797 | bpf_prog_put(prog); |
09756af4 AS |
798 | return 0; |
799 | } | |
800 | ||
7bd509e3 DB |
801 | #ifdef CONFIG_PROC_FS |
802 | static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp) | |
803 | { | |
804 | const struct bpf_prog *prog = filp->private_data; | |
f1f7714e | 805 | char prog_tag[sizeof(prog->tag) * 2 + 1] = { }; |
7bd509e3 | 806 | |
f1f7714e | 807 | bin2hex(prog_tag, prog->tag, sizeof(prog->tag)); |
7bd509e3 DB |
808 | seq_printf(m, |
809 | "prog_type:\t%u\n" | |
810 | "prog_jited:\t%u\n" | |
f1f7714e | 811 | "prog_tag:\t%s\n" |
7bd509e3 DB |
812 | "memlock:\t%llu\n", |
813 | prog->type, | |
814 | prog->jited, | |
f1f7714e | 815 | prog_tag, |
7bd509e3 DB |
816 | prog->pages * 1ULL << PAGE_SHIFT); |
817 | } | |
818 | #endif | |
819 | ||
09756af4 | 820 | static const struct file_operations bpf_prog_fops = { |
7bd509e3 DB |
821 | #ifdef CONFIG_PROC_FS |
822 | .show_fdinfo = bpf_prog_show_fdinfo, | |
823 | #endif | |
824 | .release = bpf_prog_release, | |
09756af4 AS |
825 | }; |
826 | ||
b2197755 | 827 | int bpf_prog_new_fd(struct bpf_prog *prog) |
aa79781b DB |
828 | { |
829 | return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog, | |
830 | O_RDWR | O_CLOEXEC); | |
831 | } | |
832 | ||
113214be | 833 | static struct bpf_prog *____bpf_prog_get(struct fd f) |
09756af4 | 834 | { |
09756af4 AS |
835 | if (!f.file) |
836 | return ERR_PTR(-EBADF); | |
09756af4 AS |
837 | if (f.file->f_op != &bpf_prog_fops) { |
838 | fdput(f); | |
839 | return ERR_PTR(-EINVAL); | |
840 | } | |
841 | ||
c2101297 | 842 | return f.file->private_data; |
09756af4 AS |
843 | } |
844 | ||
59d3656d | 845 | struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i) |
92117d84 | 846 | { |
59d3656d BB |
847 | if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) { |
848 | atomic_sub(i, &prog->aux->refcnt); | |
92117d84 AS |
849 | return ERR_PTR(-EBUSY); |
850 | } | |
851 | return prog; | |
852 | } | |
59d3656d BB |
853 | EXPORT_SYMBOL_GPL(bpf_prog_add); |
854 | ||
c540594f DB |
855 | void bpf_prog_sub(struct bpf_prog *prog, int i) |
856 | { | |
857 | /* Only to be used for undoing previous bpf_prog_add() in some | |
858 | * error path. We still know that another entity in our call | |
859 | * path holds a reference to the program, thus atomic_sub() can | |
860 | * be safely used in such cases! | |
861 | */ | |
862 | WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0); | |
863 | } | |
864 | EXPORT_SYMBOL_GPL(bpf_prog_sub); | |
865 | ||
59d3656d BB |
866 | struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog) |
867 | { | |
868 | return bpf_prog_add(prog, 1); | |
869 | } | |
97bc402d | 870 | EXPORT_SYMBOL_GPL(bpf_prog_inc); |
92117d84 | 871 | |
b16d9aa4 MKL |
872 | /* prog_idr_lock should have been held */ |
873 | static struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog) | |
874 | { | |
875 | int refold; | |
876 | ||
877 | refold = __atomic_add_unless(&prog->aux->refcnt, 1, 0); | |
878 | ||
879 | if (refold >= BPF_MAX_REFCNT) { | |
880 | __bpf_prog_put(prog, false); | |
881 | return ERR_PTR(-EBUSY); | |
882 | } | |
883 | ||
884 | if (!refold) | |
885 | return ERR_PTR(-ENOENT); | |
886 | ||
887 | return prog; | |
888 | } | |
889 | ||
113214be | 890 | static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *type) |
09756af4 AS |
891 | { |
892 | struct fd f = fdget(ufd); | |
893 | struct bpf_prog *prog; | |
894 | ||
113214be | 895 | prog = ____bpf_prog_get(f); |
09756af4 AS |
896 | if (IS_ERR(prog)) |
897 | return prog; | |
113214be DB |
898 | if (type && prog->type != *type) { |
899 | prog = ERR_PTR(-EINVAL); | |
900 | goto out; | |
901 | } | |
09756af4 | 902 | |
92117d84 | 903 | prog = bpf_prog_inc(prog); |
113214be | 904 | out: |
09756af4 AS |
905 | fdput(f); |
906 | return prog; | |
907 | } | |
113214be DB |
908 | |
909 | struct bpf_prog *bpf_prog_get(u32 ufd) | |
910 | { | |
911 | return __bpf_prog_get(ufd, NULL); | |
912 | } | |
913 | ||
914 | struct bpf_prog *bpf_prog_get_type(u32 ufd, enum bpf_prog_type type) | |
915 | { | |
a67edbf4 DB |
916 | struct bpf_prog *prog = __bpf_prog_get(ufd, &type); |
917 | ||
918 | if (!IS_ERR(prog)) | |
919 | trace_bpf_prog_get_type(prog); | |
920 | return prog; | |
113214be DB |
921 | } |
922 | EXPORT_SYMBOL_GPL(bpf_prog_get_type); | |
09756af4 AS |
923 | |
924 | /* last field in 'union bpf_attr' used by this command */ | |
e07b98d9 | 925 | #define BPF_PROG_LOAD_LAST_FIELD prog_flags |
09756af4 AS |
926 | |
927 | static int bpf_prog_load(union bpf_attr *attr) | |
928 | { | |
929 | enum bpf_prog_type type = attr->prog_type; | |
930 | struct bpf_prog *prog; | |
931 | int err; | |
932 | char license[128]; | |
933 | bool is_gpl; | |
934 | ||
935 | if (CHECK_ATTR(BPF_PROG_LOAD)) | |
936 | return -EINVAL; | |
937 | ||
e07b98d9 DM |
938 | if (attr->prog_flags & ~BPF_F_STRICT_ALIGNMENT) |
939 | return -EINVAL; | |
940 | ||
09756af4 | 941 | /* copy eBPF program license from user space */ |
535e7b4b | 942 | if (strncpy_from_user(license, u64_to_user_ptr(attr->license), |
09756af4 AS |
943 | sizeof(license) - 1) < 0) |
944 | return -EFAULT; | |
945 | license[sizeof(license) - 1] = 0; | |
946 | ||
947 | /* eBPF programs must be GPL compatible to use GPL-ed functions */ | |
948 | is_gpl = license_is_gpl_compatible(license); | |
949 | ||
ef0915ca DB |
950 | if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS) |
951 | return -E2BIG; | |
09756af4 | 952 | |
2541517c AS |
953 | if (type == BPF_PROG_TYPE_KPROBE && |
954 | attr->kern_version != LINUX_VERSION_CODE) | |
955 | return -EINVAL; | |
956 | ||
80b7d819 CF |
957 | if (type != BPF_PROG_TYPE_SOCKET_FILTER && |
958 | type != BPF_PROG_TYPE_CGROUP_SKB && | |
959 | !capable(CAP_SYS_ADMIN)) | |
1be7f75d AS |
960 | return -EPERM; |
961 | ||
09756af4 AS |
962 | /* plain bpf_prog allocation */ |
963 | prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER); | |
964 | if (!prog) | |
965 | return -ENOMEM; | |
966 | ||
aaac3ba9 AS |
967 | err = bpf_prog_charge_memlock(prog); |
968 | if (err) | |
969 | goto free_prog_nouncharge; | |
970 | ||
09756af4 AS |
971 | prog->len = attr->insn_cnt; |
972 | ||
973 | err = -EFAULT; | |
535e7b4b | 974 | if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns), |
aafe6ae9 | 975 | bpf_prog_insn_size(prog)) != 0) |
09756af4 AS |
976 | goto free_prog; |
977 | ||
978 | prog->orig_prog = NULL; | |
a91263d5 | 979 | prog->jited = 0; |
09756af4 AS |
980 | |
981 | atomic_set(&prog->aux->refcnt, 1); | |
a91263d5 | 982 | prog->gpl_compatible = is_gpl ? 1 : 0; |
09756af4 AS |
983 | |
984 | /* find program type: socket_filter vs tracing_filter */ | |
985 | err = find_prog_type(type, prog); | |
986 | if (err < 0) | |
987 | goto free_prog; | |
988 | ||
989 | /* run eBPF verifier */ | |
9bac3d6d | 990 | err = bpf_check(&prog, attr); |
09756af4 AS |
991 | if (err < 0) |
992 | goto free_used_maps; | |
993 | ||
994 | /* eBPF program is ready to be JITed */ | |
d1c55ab5 | 995 | prog = bpf_prog_select_runtime(prog, &err); |
04fd61ab AS |
996 | if (err < 0) |
997 | goto free_used_maps; | |
09756af4 | 998 | |
dc4bb0e2 MKL |
999 | err = bpf_prog_alloc_id(prog); |
1000 | if (err) | |
1001 | goto free_used_maps; | |
1002 | ||
aa79781b | 1003 | err = bpf_prog_new_fd(prog); |
b16d9aa4 MKL |
1004 | if (err < 0) { |
1005 | /* failed to allocate fd. | |
1006 | * bpf_prog_put() is needed because the above | |
1007 | * bpf_prog_alloc_id() has published the prog | |
1008 | * to the userspace and the userspace may | |
1009 | * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID. | |
1010 | */ | |
1011 | bpf_prog_put(prog); | |
1012 | return err; | |
1013 | } | |
09756af4 | 1014 | |
74451e66 | 1015 | bpf_prog_kallsyms_add(prog); |
a67edbf4 | 1016 | trace_bpf_prog_load(prog, err); |
09756af4 AS |
1017 | return err; |
1018 | ||
1019 | free_used_maps: | |
1020 | free_used_maps(prog->aux); | |
1021 | free_prog: | |
aaac3ba9 AS |
1022 | bpf_prog_uncharge_memlock(prog); |
1023 | free_prog_nouncharge: | |
09756af4 AS |
1024 | bpf_prog_free(prog); |
1025 | return err; | |
1026 | } | |
1027 | ||
b2197755 DB |
1028 | #define BPF_OBJ_LAST_FIELD bpf_fd |
1029 | ||
1030 | static int bpf_obj_pin(const union bpf_attr *attr) | |
1031 | { | |
1032 | if (CHECK_ATTR(BPF_OBJ)) | |
1033 | return -EINVAL; | |
1034 | ||
535e7b4b | 1035 | return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname)); |
b2197755 DB |
1036 | } |
1037 | ||
1038 | static int bpf_obj_get(const union bpf_attr *attr) | |
1039 | { | |
1040 | if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0) | |
1041 | return -EINVAL; | |
1042 | ||
535e7b4b | 1043 | return bpf_obj_get_user(u64_to_user_ptr(attr->pathname)); |
b2197755 DB |
1044 | } |
1045 | ||
f4324551 DM |
1046 | #ifdef CONFIG_CGROUP_BPF |
1047 | ||
7f677633 | 1048 | #define BPF_PROG_ATTACH_LAST_FIELD attach_flags |
f4324551 DM |
1049 | |
1050 | static int bpf_prog_attach(const union bpf_attr *attr) | |
1051 | { | |
7f677633 | 1052 | enum bpf_prog_type ptype; |
f4324551 DM |
1053 | struct bpf_prog *prog; |
1054 | struct cgroup *cgrp; | |
7f677633 | 1055 | int ret; |
f4324551 DM |
1056 | |
1057 | if (!capable(CAP_NET_ADMIN)) | |
1058 | return -EPERM; | |
1059 | ||
1060 | if (CHECK_ATTR(BPF_PROG_ATTACH)) | |
1061 | return -EINVAL; | |
1062 | ||
7f677633 AS |
1063 | if (attr->attach_flags & ~BPF_F_ALLOW_OVERRIDE) |
1064 | return -EINVAL; | |
1065 | ||
f4324551 DM |
1066 | switch (attr->attach_type) { |
1067 | case BPF_CGROUP_INET_INGRESS: | |
1068 | case BPF_CGROUP_INET_EGRESS: | |
b2cd1257 | 1069 | ptype = BPF_PROG_TYPE_CGROUP_SKB; |
f4324551 | 1070 | break; |
61023658 DA |
1071 | case BPF_CGROUP_INET_SOCK_CREATE: |
1072 | ptype = BPF_PROG_TYPE_CGROUP_SOCK; | |
1073 | break; | |
40304b2a LB |
1074 | case BPF_CGROUP_SOCK_OPS: |
1075 | ptype = BPF_PROG_TYPE_SOCK_OPS; | |
1076 | break; | |
f4324551 DM |
1077 | default: |
1078 | return -EINVAL; | |
1079 | } | |
1080 | ||
b2cd1257 DA |
1081 | prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype); |
1082 | if (IS_ERR(prog)) | |
1083 | return PTR_ERR(prog); | |
1084 | ||
1085 | cgrp = cgroup_get_from_fd(attr->target_fd); | |
1086 | if (IS_ERR(cgrp)) { | |
1087 | bpf_prog_put(prog); | |
1088 | return PTR_ERR(cgrp); | |
1089 | } | |
1090 | ||
7f677633 AS |
1091 | ret = cgroup_bpf_update(cgrp, prog, attr->attach_type, |
1092 | attr->attach_flags & BPF_F_ALLOW_OVERRIDE); | |
1093 | if (ret) | |
1094 | bpf_prog_put(prog); | |
b2cd1257 DA |
1095 | cgroup_put(cgrp); |
1096 | ||
7f677633 | 1097 | return ret; |
f4324551 DM |
1098 | } |
1099 | ||
1100 | #define BPF_PROG_DETACH_LAST_FIELD attach_type | |
1101 | ||
1102 | static int bpf_prog_detach(const union bpf_attr *attr) | |
1103 | { | |
1104 | struct cgroup *cgrp; | |
7f677633 | 1105 | int ret; |
f4324551 DM |
1106 | |
1107 | if (!capable(CAP_NET_ADMIN)) | |
1108 | return -EPERM; | |
1109 | ||
1110 | if (CHECK_ATTR(BPF_PROG_DETACH)) | |
1111 | return -EINVAL; | |
1112 | ||
1113 | switch (attr->attach_type) { | |
1114 | case BPF_CGROUP_INET_INGRESS: | |
1115 | case BPF_CGROUP_INET_EGRESS: | |
61023658 | 1116 | case BPF_CGROUP_INET_SOCK_CREATE: |
40304b2a | 1117 | case BPF_CGROUP_SOCK_OPS: |
f4324551 DM |
1118 | cgrp = cgroup_get_from_fd(attr->target_fd); |
1119 | if (IS_ERR(cgrp)) | |
1120 | return PTR_ERR(cgrp); | |
1121 | ||
7f677633 | 1122 | ret = cgroup_bpf_update(cgrp, NULL, attr->attach_type, false); |
f4324551 DM |
1123 | cgroup_put(cgrp); |
1124 | break; | |
1125 | ||
1126 | default: | |
1127 | return -EINVAL; | |
1128 | } | |
1129 | ||
7f677633 | 1130 | return ret; |
f4324551 | 1131 | } |
40304b2a | 1132 | |
f4324551 DM |
1133 | #endif /* CONFIG_CGROUP_BPF */ |
1134 | ||
1cf1cae9 AS |
1135 | #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration |
1136 | ||
1137 | static int bpf_prog_test_run(const union bpf_attr *attr, | |
1138 | union bpf_attr __user *uattr) | |
1139 | { | |
1140 | struct bpf_prog *prog; | |
1141 | int ret = -ENOTSUPP; | |
1142 | ||
1143 | if (CHECK_ATTR(BPF_PROG_TEST_RUN)) | |
1144 | return -EINVAL; | |
1145 | ||
1146 | prog = bpf_prog_get(attr->test.prog_fd); | |
1147 | if (IS_ERR(prog)) | |
1148 | return PTR_ERR(prog); | |
1149 | ||
1150 | if (prog->aux->ops->test_run) | |
1151 | ret = prog->aux->ops->test_run(prog, attr, uattr); | |
1152 | ||
1153 | bpf_prog_put(prog); | |
1154 | return ret; | |
1155 | } | |
1156 | ||
34ad5580 MKL |
1157 | #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id |
1158 | ||
1159 | static int bpf_obj_get_next_id(const union bpf_attr *attr, | |
1160 | union bpf_attr __user *uattr, | |
1161 | struct idr *idr, | |
1162 | spinlock_t *lock) | |
1163 | { | |
1164 | u32 next_id = attr->start_id; | |
1165 | int err = 0; | |
1166 | ||
1167 | if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX) | |
1168 | return -EINVAL; | |
1169 | ||
1170 | if (!capable(CAP_SYS_ADMIN)) | |
1171 | return -EPERM; | |
1172 | ||
1173 | next_id++; | |
1174 | spin_lock_bh(lock); | |
1175 | if (!idr_get_next(idr, &next_id)) | |
1176 | err = -ENOENT; | |
1177 | spin_unlock_bh(lock); | |
1178 | ||
1179 | if (!err) | |
1180 | err = put_user(next_id, &uattr->next_id); | |
1181 | ||
1182 | return err; | |
1183 | } | |
1184 | ||
b16d9aa4 MKL |
1185 | #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id |
1186 | ||
1187 | static int bpf_prog_get_fd_by_id(const union bpf_attr *attr) | |
1188 | { | |
1189 | struct bpf_prog *prog; | |
1190 | u32 id = attr->prog_id; | |
1191 | int fd; | |
1192 | ||
1193 | if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID)) | |
1194 | return -EINVAL; | |
1195 | ||
1196 | if (!capable(CAP_SYS_ADMIN)) | |
1197 | return -EPERM; | |
1198 | ||
1199 | spin_lock_bh(&prog_idr_lock); | |
1200 | prog = idr_find(&prog_idr, id); | |
1201 | if (prog) | |
1202 | prog = bpf_prog_inc_not_zero(prog); | |
1203 | else | |
1204 | prog = ERR_PTR(-ENOENT); | |
1205 | spin_unlock_bh(&prog_idr_lock); | |
1206 | ||
1207 | if (IS_ERR(prog)) | |
1208 | return PTR_ERR(prog); | |
1209 | ||
1210 | fd = bpf_prog_new_fd(prog); | |
1211 | if (fd < 0) | |
1212 | bpf_prog_put(prog); | |
1213 | ||
1214 | return fd; | |
1215 | } | |
1216 | ||
bd5f5f4e MKL |
1217 | #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD map_id |
1218 | ||
1219 | static int bpf_map_get_fd_by_id(const union bpf_attr *attr) | |
1220 | { | |
1221 | struct bpf_map *map; | |
1222 | u32 id = attr->map_id; | |
1223 | int fd; | |
1224 | ||
1225 | if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID)) | |
1226 | return -EINVAL; | |
1227 | ||
1228 | if (!capable(CAP_SYS_ADMIN)) | |
1229 | return -EPERM; | |
1230 | ||
1231 | spin_lock_bh(&map_idr_lock); | |
1232 | map = idr_find(&map_idr, id); | |
1233 | if (map) | |
1234 | map = bpf_map_inc_not_zero(map, true); | |
1235 | else | |
1236 | map = ERR_PTR(-ENOENT); | |
1237 | spin_unlock_bh(&map_idr_lock); | |
1238 | ||
1239 | if (IS_ERR(map)) | |
1240 | return PTR_ERR(map); | |
1241 | ||
1242 | fd = bpf_map_new_fd(map); | |
1243 | if (fd < 0) | |
1244 | bpf_map_put(map); | |
1245 | ||
1246 | return fd; | |
1247 | } | |
1248 | ||
1e270976 MKL |
1249 | static int check_uarg_tail_zero(void __user *uaddr, |
1250 | size_t expected_size, | |
1251 | size_t actual_size) | |
1252 | { | |
1253 | unsigned char __user *addr; | |
1254 | unsigned char __user *end; | |
1255 | unsigned char val; | |
1256 | int err; | |
1257 | ||
1258 | if (actual_size <= expected_size) | |
1259 | return 0; | |
1260 | ||
1261 | addr = uaddr + expected_size; | |
1262 | end = uaddr + actual_size; | |
1263 | ||
1264 | for (; addr < end; addr++) { | |
1265 | err = get_user(val, addr); | |
1266 | if (err) | |
1267 | return err; | |
1268 | if (val) | |
1269 | return -E2BIG; | |
1270 | } | |
1271 | ||
1272 | return 0; | |
1273 | } | |
1274 | ||
1275 | static int bpf_prog_get_info_by_fd(struct bpf_prog *prog, | |
1276 | const union bpf_attr *attr, | |
1277 | union bpf_attr __user *uattr) | |
1278 | { | |
1279 | struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info); | |
1280 | struct bpf_prog_info info = {}; | |
1281 | u32 info_len = attr->info.info_len; | |
1282 | char __user *uinsns; | |
1283 | u32 ulen; | |
1284 | int err; | |
1285 | ||
1286 | err = check_uarg_tail_zero(uinfo, sizeof(info), info_len); | |
1287 | if (err) | |
1288 | return err; | |
1289 | info_len = min_t(u32, sizeof(info), info_len); | |
1290 | ||
1291 | if (copy_from_user(&info, uinfo, info_len)) | |
89b09689 | 1292 | return -EFAULT; |
1e270976 MKL |
1293 | |
1294 | info.type = prog->type; | |
1295 | info.id = prog->aux->id; | |
1296 | ||
1297 | memcpy(info.tag, prog->tag, sizeof(prog->tag)); | |
1298 | ||
1299 | if (!capable(CAP_SYS_ADMIN)) { | |
1300 | info.jited_prog_len = 0; | |
1301 | info.xlated_prog_len = 0; | |
1302 | goto done; | |
1303 | } | |
1304 | ||
1305 | ulen = info.jited_prog_len; | |
1306 | info.jited_prog_len = prog->jited_len; | |
1307 | if (info.jited_prog_len && ulen) { | |
1308 | uinsns = u64_to_user_ptr(info.jited_prog_insns); | |
1309 | ulen = min_t(u32, info.jited_prog_len, ulen); | |
1310 | if (copy_to_user(uinsns, prog->bpf_func, ulen)) | |
1311 | return -EFAULT; | |
1312 | } | |
1313 | ||
1314 | ulen = info.xlated_prog_len; | |
9975a54b | 1315 | info.xlated_prog_len = bpf_prog_insn_size(prog); |
1e270976 MKL |
1316 | if (info.xlated_prog_len && ulen) { |
1317 | uinsns = u64_to_user_ptr(info.xlated_prog_insns); | |
1318 | ulen = min_t(u32, info.xlated_prog_len, ulen); | |
1319 | if (copy_to_user(uinsns, prog->insnsi, ulen)) | |
1320 | return -EFAULT; | |
1321 | } | |
1322 | ||
1323 | done: | |
1324 | if (copy_to_user(uinfo, &info, info_len) || | |
1325 | put_user(info_len, &uattr->info.info_len)) | |
1326 | return -EFAULT; | |
1327 | ||
1328 | return 0; | |
1329 | } | |
1330 | ||
1331 | static int bpf_map_get_info_by_fd(struct bpf_map *map, | |
1332 | const union bpf_attr *attr, | |
1333 | union bpf_attr __user *uattr) | |
1334 | { | |
1335 | struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info); | |
1336 | struct bpf_map_info info = {}; | |
1337 | u32 info_len = attr->info.info_len; | |
1338 | int err; | |
1339 | ||
1340 | err = check_uarg_tail_zero(uinfo, sizeof(info), info_len); | |
1341 | if (err) | |
1342 | return err; | |
1343 | info_len = min_t(u32, sizeof(info), info_len); | |
1344 | ||
1345 | info.type = map->map_type; | |
1346 | info.id = map->id; | |
1347 | info.key_size = map->key_size; | |
1348 | info.value_size = map->value_size; | |
1349 | info.max_entries = map->max_entries; | |
1350 | info.map_flags = map->map_flags; | |
1351 | ||
1352 | if (copy_to_user(uinfo, &info, info_len) || | |
1353 | put_user(info_len, &uattr->info.info_len)) | |
1354 | return -EFAULT; | |
1355 | ||
1356 | return 0; | |
1357 | } | |
1358 | ||
1359 | #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info | |
1360 | ||
1361 | static int bpf_obj_get_info_by_fd(const union bpf_attr *attr, | |
1362 | union bpf_attr __user *uattr) | |
1363 | { | |
1364 | int ufd = attr->info.bpf_fd; | |
1365 | struct fd f; | |
1366 | int err; | |
1367 | ||
1368 | if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD)) | |
1369 | return -EINVAL; | |
1370 | ||
1371 | f = fdget(ufd); | |
1372 | if (!f.file) | |
1373 | return -EBADFD; | |
1374 | ||
1375 | if (f.file->f_op == &bpf_prog_fops) | |
1376 | err = bpf_prog_get_info_by_fd(f.file->private_data, attr, | |
1377 | uattr); | |
1378 | else if (f.file->f_op == &bpf_map_fops) | |
1379 | err = bpf_map_get_info_by_fd(f.file->private_data, attr, | |
1380 | uattr); | |
1381 | else | |
1382 | err = -EINVAL; | |
1383 | ||
1384 | fdput(f); | |
1385 | return err; | |
1386 | } | |
1387 | ||
99c55f7d AS |
1388 | SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size) |
1389 | { | |
1390 | union bpf_attr attr = {}; | |
1391 | int err; | |
1392 | ||
1be7f75d | 1393 | if (!capable(CAP_SYS_ADMIN) && sysctl_unprivileged_bpf_disabled) |
99c55f7d AS |
1394 | return -EPERM; |
1395 | ||
1396 | if (!access_ok(VERIFY_READ, uattr, 1)) | |
1397 | return -EFAULT; | |
1398 | ||
1399 | if (size > PAGE_SIZE) /* silly large */ | |
1400 | return -E2BIG; | |
1401 | ||
1402 | /* If we're handed a bigger struct than we know of, | |
1403 | * ensure all the unknown bits are 0 - i.e. new | |
1404 | * user-space does not rely on any kernel feature | |
1405 | * extensions we dont know about yet. | |
1406 | */ | |
1e270976 MKL |
1407 | err = check_uarg_tail_zero(uattr, sizeof(attr), size); |
1408 | if (err) | |
1409 | return err; | |
1410 | size = min_t(u32, size, sizeof(attr)); | |
99c55f7d AS |
1411 | |
1412 | /* copy attributes from user space, may be less than sizeof(bpf_attr) */ | |
1413 | if (copy_from_user(&attr, uattr, size) != 0) | |
1414 | return -EFAULT; | |
1415 | ||
1416 | switch (cmd) { | |
1417 | case BPF_MAP_CREATE: | |
1418 | err = map_create(&attr); | |
1419 | break; | |
db20fd2b AS |
1420 | case BPF_MAP_LOOKUP_ELEM: |
1421 | err = map_lookup_elem(&attr); | |
1422 | break; | |
1423 | case BPF_MAP_UPDATE_ELEM: | |
1424 | err = map_update_elem(&attr); | |
1425 | break; | |
1426 | case BPF_MAP_DELETE_ELEM: | |
1427 | err = map_delete_elem(&attr); | |
1428 | break; | |
1429 | case BPF_MAP_GET_NEXT_KEY: | |
1430 | err = map_get_next_key(&attr); | |
1431 | break; | |
09756af4 AS |
1432 | case BPF_PROG_LOAD: |
1433 | err = bpf_prog_load(&attr); | |
1434 | break; | |
b2197755 DB |
1435 | case BPF_OBJ_PIN: |
1436 | err = bpf_obj_pin(&attr); | |
1437 | break; | |
1438 | case BPF_OBJ_GET: | |
1439 | err = bpf_obj_get(&attr); | |
1440 | break; | |
f4324551 DM |
1441 | #ifdef CONFIG_CGROUP_BPF |
1442 | case BPF_PROG_ATTACH: | |
1443 | err = bpf_prog_attach(&attr); | |
1444 | break; | |
1445 | case BPF_PROG_DETACH: | |
1446 | err = bpf_prog_detach(&attr); | |
1447 | break; | |
1448 | #endif | |
1cf1cae9 AS |
1449 | case BPF_PROG_TEST_RUN: |
1450 | err = bpf_prog_test_run(&attr, uattr); | |
1451 | break; | |
34ad5580 MKL |
1452 | case BPF_PROG_GET_NEXT_ID: |
1453 | err = bpf_obj_get_next_id(&attr, uattr, | |
1454 | &prog_idr, &prog_idr_lock); | |
1455 | break; | |
1456 | case BPF_MAP_GET_NEXT_ID: | |
1457 | err = bpf_obj_get_next_id(&attr, uattr, | |
1458 | &map_idr, &map_idr_lock); | |
1459 | break; | |
b16d9aa4 MKL |
1460 | case BPF_PROG_GET_FD_BY_ID: |
1461 | err = bpf_prog_get_fd_by_id(&attr); | |
1462 | break; | |
bd5f5f4e MKL |
1463 | case BPF_MAP_GET_FD_BY_ID: |
1464 | err = bpf_map_get_fd_by_id(&attr); | |
1465 | break; | |
1e270976 MKL |
1466 | case BPF_OBJ_GET_INFO_BY_FD: |
1467 | err = bpf_obj_get_info_by_fd(&attr, uattr); | |
1468 | break; | |
99c55f7d AS |
1469 | default: |
1470 | err = -EINVAL; | |
1471 | break; | |
1472 | } | |
1473 | ||
1474 | return err; | |
1475 | } |