]>
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 | { |
0441bb4a ED |
147 | unsigned long flags; |
148 | ||
bd5f5f4e | 149 | if (do_idr_lock) |
0441bb4a | 150 | spin_lock_irqsave(&map_idr_lock, flags); |
bd5f5f4e MKL |
151 | else |
152 | __acquire(&map_idr_lock); | |
153 | ||
f3f1c054 | 154 | idr_remove(&map_idr, map->id); |
bd5f5f4e MKL |
155 | |
156 | if (do_idr_lock) | |
0441bb4a | 157 | spin_unlock_irqrestore(&map_idr_lock, flags); |
bd5f5f4e MKL |
158 | else |
159 | __release(&map_idr_lock); | |
f3f1c054 MKL |
160 | } |
161 | ||
99c55f7d AS |
162 | /* called from workqueue */ |
163 | static void bpf_map_free_deferred(struct work_struct *work) | |
164 | { | |
165 | struct bpf_map *map = container_of(work, struct bpf_map, work); | |
166 | ||
aaac3ba9 | 167 | bpf_map_uncharge_memlock(map); |
99c55f7d AS |
168 | /* implementation dependent freeing */ |
169 | map->ops->map_free(map); | |
170 | } | |
171 | ||
c9da161c DB |
172 | static void bpf_map_put_uref(struct bpf_map *map) |
173 | { | |
174 | if (atomic_dec_and_test(&map->usercnt)) { | |
175 | if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) | |
176 | bpf_fd_array_map_clear(map); | |
177 | } | |
178 | } | |
179 | ||
99c55f7d AS |
180 | /* decrement map refcnt and schedule it for freeing via workqueue |
181 | * (unrelying map implementation ops->map_free() might sleep) | |
182 | */ | |
bd5f5f4e | 183 | static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock) |
99c55f7d AS |
184 | { |
185 | if (atomic_dec_and_test(&map->refcnt)) { | |
34ad5580 | 186 | /* bpf_map_free_id() must be called first */ |
bd5f5f4e | 187 | bpf_map_free_id(map, do_idr_lock); |
99c55f7d AS |
188 | INIT_WORK(&map->work, bpf_map_free_deferred); |
189 | schedule_work(&map->work); | |
190 | } | |
191 | } | |
192 | ||
bd5f5f4e MKL |
193 | void bpf_map_put(struct bpf_map *map) |
194 | { | |
195 | __bpf_map_put(map, true); | |
196 | } | |
197 | ||
c9da161c | 198 | void bpf_map_put_with_uref(struct bpf_map *map) |
99c55f7d | 199 | { |
c9da161c | 200 | bpf_map_put_uref(map); |
99c55f7d | 201 | bpf_map_put(map); |
c9da161c DB |
202 | } |
203 | ||
204 | static int bpf_map_release(struct inode *inode, struct file *filp) | |
205 | { | |
61d1b6a4 DB |
206 | struct bpf_map *map = filp->private_data; |
207 | ||
208 | if (map->ops->map_release) | |
209 | map->ops->map_release(map, filp); | |
210 | ||
211 | bpf_map_put_with_uref(map); | |
99c55f7d AS |
212 | return 0; |
213 | } | |
214 | ||
f99bf205 DB |
215 | #ifdef CONFIG_PROC_FS |
216 | static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp) | |
217 | { | |
218 | const struct bpf_map *map = filp->private_data; | |
21116b70 DB |
219 | const struct bpf_array *array; |
220 | u32 owner_prog_type = 0; | |
9780c0ab | 221 | u32 owner_jited = 0; |
21116b70 DB |
222 | |
223 | if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) { | |
224 | array = container_of(map, struct bpf_array, map); | |
225 | owner_prog_type = array->owner_prog_type; | |
9780c0ab | 226 | owner_jited = array->owner_jited; |
21116b70 | 227 | } |
f99bf205 DB |
228 | |
229 | seq_printf(m, | |
230 | "map_type:\t%u\n" | |
231 | "key_size:\t%u\n" | |
232 | "value_size:\t%u\n" | |
322cea2f | 233 | "max_entries:\t%u\n" |
21116b70 DB |
234 | "map_flags:\t%#x\n" |
235 | "memlock:\t%llu\n", | |
f99bf205 DB |
236 | map->map_type, |
237 | map->key_size, | |
238 | map->value_size, | |
322cea2f | 239 | map->max_entries, |
21116b70 DB |
240 | map->map_flags, |
241 | map->pages * 1ULL << PAGE_SHIFT); | |
242 | ||
9780c0ab | 243 | if (owner_prog_type) { |
21116b70 DB |
244 | seq_printf(m, "owner_prog_type:\t%u\n", |
245 | owner_prog_type); | |
9780c0ab DB |
246 | seq_printf(m, "owner_jited:\t%u\n", |
247 | owner_jited); | |
248 | } | |
f99bf205 DB |
249 | } |
250 | #endif | |
251 | ||
99c55f7d | 252 | static const struct file_operations bpf_map_fops = { |
f99bf205 DB |
253 | #ifdef CONFIG_PROC_FS |
254 | .show_fdinfo = bpf_map_show_fdinfo, | |
255 | #endif | |
256 | .release = bpf_map_release, | |
99c55f7d AS |
257 | }; |
258 | ||
b2197755 | 259 | int bpf_map_new_fd(struct bpf_map *map) |
aa79781b DB |
260 | { |
261 | return anon_inode_getfd("bpf-map", &bpf_map_fops, map, | |
262 | O_RDWR | O_CLOEXEC); | |
263 | } | |
264 | ||
99c55f7d AS |
265 | /* helper macro to check that unused fields 'union bpf_attr' are zero */ |
266 | #define CHECK_ATTR(CMD) \ | |
267 | memchr_inv((void *) &attr->CMD##_LAST_FIELD + \ | |
268 | sizeof(attr->CMD##_LAST_FIELD), 0, \ | |
269 | sizeof(*attr) - \ | |
270 | offsetof(union bpf_attr, CMD##_LAST_FIELD) - \ | |
271 | sizeof(attr->CMD##_LAST_FIELD)) != NULL | |
272 | ||
56f668df | 273 | #define BPF_MAP_CREATE_LAST_FIELD inner_map_fd |
99c55f7d AS |
274 | /* called via syscall */ |
275 | static int map_create(union bpf_attr *attr) | |
276 | { | |
277 | struct bpf_map *map; | |
278 | int err; | |
279 | ||
280 | err = CHECK_ATTR(BPF_MAP_CREATE); | |
281 | if (err) | |
282 | return -EINVAL; | |
283 | ||
284 | /* find map type and init map: hashtable vs rbtree vs bloom vs ... */ | |
285 | map = find_and_alloc_map(attr); | |
286 | if (IS_ERR(map)) | |
287 | return PTR_ERR(map); | |
288 | ||
289 | atomic_set(&map->refcnt, 1); | |
c9da161c | 290 | atomic_set(&map->usercnt, 1); |
99c55f7d | 291 | |
aaac3ba9 AS |
292 | err = bpf_map_charge_memlock(map); |
293 | if (err) | |
20b2b24f | 294 | goto free_map_nouncharge; |
aaac3ba9 | 295 | |
f3f1c054 MKL |
296 | err = bpf_map_alloc_id(map); |
297 | if (err) | |
298 | goto free_map; | |
299 | ||
aa79781b | 300 | err = bpf_map_new_fd(map); |
bd5f5f4e MKL |
301 | if (err < 0) { |
302 | /* failed to allocate fd. | |
303 | * bpf_map_put() is needed because the above | |
304 | * bpf_map_alloc_id() has published the map | |
305 | * to the userspace and the userspace may | |
306 | * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID. | |
307 | */ | |
308 | bpf_map_put(map); | |
309 | return err; | |
310 | } | |
99c55f7d | 311 | |
a67edbf4 | 312 | trace_bpf_map_create(map, err); |
99c55f7d AS |
313 | return err; |
314 | ||
315 | free_map: | |
20b2b24f DB |
316 | bpf_map_uncharge_memlock(map); |
317 | free_map_nouncharge: | |
99c55f7d AS |
318 | map->ops->map_free(map); |
319 | return err; | |
320 | } | |
321 | ||
db20fd2b AS |
322 | /* if error is returned, fd is released. |
323 | * On success caller should complete fd access with matching fdput() | |
324 | */ | |
c2101297 | 325 | struct bpf_map *__bpf_map_get(struct fd f) |
db20fd2b | 326 | { |
db20fd2b AS |
327 | if (!f.file) |
328 | return ERR_PTR(-EBADF); | |
db20fd2b AS |
329 | if (f.file->f_op != &bpf_map_fops) { |
330 | fdput(f); | |
331 | return ERR_PTR(-EINVAL); | |
332 | } | |
333 | ||
c2101297 DB |
334 | return f.file->private_data; |
335 | } | |
336 | ||
92117d84 AS |
337 | /* prog's and map's refcnt limit */ |
338 | #define BPF_MAX_REFCNT 32768 | |
339 | ||
340 | struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref) | |
c9da161c | 341 | { |
92117d84 AS |
342 | if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) { |
343 | atomic_dec(&map->refcnt); | |
344 | return ERR_PTR(-EBUSY); | |
345 | } | |
c9da161c DB |
346 | if (uref) |
347 | atomic_inc(&map->usercnt); | |
92117d84 | 348 | return map; |
c9da161c DB |
349 | } |
350 | ||
351 | struct bpf_map *bpf_map_get_with_uref(u32 ufd) | |
c2101297 DB |
352 | { |
353 | struct fd f = fdget(ufd); | |
354 | struct bpf_map *map; | |
355 | ||
356 | map = __bpf_map_get(f); | |
357 | if (IS_ERR(map)) | |
358 | return map; | |
359 | ||
92117d84 | 360 | map = bpf_map_inc(map, true); |
c2101297 | 361 | fdput(f); |
db20fd2b AS |
362 | |
363 | return map; | |
364 | } | |
365 | ||
bd5f5f4e MKL |
366 | /* map_idr_lock should have been held */ |
367 | static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map, | |
368 | bool uref) | |
369 | { | |
370 | int refold; | |
371 | ||
372 | refold = __atomic_add_unless(&map->refcnt, 1, 0); | |
373 | ||
374 | if (refold >= BPF_MAX_REFCNT) { | |
375 | __bpf_map_put(map, false); | |
376 | return ERR_PTR(-EBUSY); | |
377 | } | |
378 | ||
379 | if (!refold) | |
380 | return ERR_PTR(-ENOENT); | |
381 | ||
382 | if (uref) | |
383 | atomic_inc(&map->usercnt); | |
384 | ||
385 | return map; | |
386 | } | |
387 | ||
b8cdc051 AS |
388 | int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value) |
389 | { | |
390 | return -ENOTSUPP; | |
391 | } | |
392 | ||
db20fd2b AS |
393 | /* last field in 'union bpf_attr' used by this command */ |
394 | #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value | |
395 | ||
396 | static int map_lookup_elem(union bpf_attr *attr) | |
397 | { | |
535e7b4b MS |
398 | void __user *ukey = u64_to_user_ptr(attr->key); |
399 | void __user *uvalue = u64_to_user_ptr(attr->value); | |
db20fd2b | 400 | int ufd = attr->map_fd; |
db20fd2b | 401 | struct bpf_map *map; |
8ebe667c | 402 | void *key, *value, *ptr; |
15a07b33 | 403 | u32 value_size; |
592867bf | 404 | struct fd f; |
db20fd2b AS |
405 | int err; |
406 | ||
407 | if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM)) | |
408 | return -EINVAL; | |
409 | ||
592867bf | 410 | f = fdget(ufd); |
c2101297 | 411 | map = __bpf_map_get(f); |
db20fd2b AS |
412 | if (IS_ERR(map)) |
413 | return PTR_ERR(map); | |
414 | ||
e4448ed8 AV |
415 | key = memdup_user(ukey, map->key_size); |
416 | if (IS_ERR(key)) { | |
417 | err = PTR_ERR(key); | |
db20fd2b | 418 | goto err_put; |
e4448ed8 | 419 | } |
db20fd2b | 420 | |
15a07b33 | 421 | if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || |
8f844938 | 422 | map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH || |
15a07b33 AS |
423 | map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) |
424 | value_size = round_up(map->value_size, 8) * num_possible_cpus(); | |
14dc6f04 MKL |
425 | else if (IS_FD_MAP(map)) |
426 | value_size = sizeof(u32); | |
15a07b33 AS |
427 | else |
428 | value_size = map->value_size; | |
429 | ||
8ebe667c | 430 | err = -ENOMEM; |
15a07b33 | 431 | value = kmalloc(value_size, GFP_USER | __GFP_NOWARN); |
db20fd2b | 432 | if (!value) |
8ebe667c AS |
433 | goto free_key; |
434 | ||
8f844938 MKL |
435 | if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || |
436 | map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) { | |
15a07b33 AS |
437 | err = bpf_percpu_hash_copy(map, key, value); |
438 | } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) { | |
439 | err = bpf_percpu_array_copy(map, key, value); | |
557c0c6e AS |
440 | } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) { |
441 | err = bpf_stackmap_copy(map, key, value); | |
14dc6f04 MKL |
442 | } else if (IS_FD_ARRAY(map)) { |
443 | err = bpf_fd_array_map_lookup_elem(map, key, value); | |
444 | } else if (IS_FD_HASH(map)) { | |
445 | err = bpf_fd_htab_map_lookup_elem(map, key, value); | |
15a07b33 AS |
446 | } else { |
447 | rcu_read_lock(); | |
448 | ptr = map->ops->map_lookup_elem(map, key); | |
449 | if (ptr) | |
450 | memcpy(value, ptr, value_size); | |
451 | rcu_read_unlock(); | |
452 | err = ptr ? 0 : -ENOENT; | |
453 | } | |
8ebe667c | 454 | |
15a07b33 | 455 | if (err) |
8ebe667c | 456 | goto free_value; |
db20fd2b AS |
457 | |
458 | err = -EFAULT; | |
15a07b33 | 459 | if (copy_to_user(uvalue, value, value_size) != 0) |
8ebe667c | 460 | goto free_value; |
db20fd2b | 461 | |
a67edbf4 | 462 | trace_bpf_map_lookup_elem(map, ufd, key, value); |
db20fd2b AS |
463 | err = 0; |
464 | ||
8ebe667c AS |
465 | free_value: |
466 | kfree(value); | |
db20fd2b AS |
467 | free_key: |
468 | kfree(key); | |
469 | err_put: | |
470 | fdput(f); | |
471 | return err; | |
472 | } | |
473 | ||
3274f520 | 474 | #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags |
db20fd2b AS |
475 | |
476 | static int map_update_elem(union bpf_attr *attr) | |
477 | { | |
535e7b4b MS |
478 | void __user *ukey = u64_to_user_ptr(attr->key); |
479 | void __user *uvalue = u64_to_user_ptr(attr->value); | |
db20fd2b | 480 | int ufd = attr->map_fd; |
db20fd2b AS |
481 | struct bpf_map *map; |
482 | void *key, *value; | |
15a07b33 | 483 | u32 value_size; |
592867bf | 484 | struct fd f; |
db20fd2b AS |
485 | int err; |
486 | ||
487 | if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM)) | |
488 | return -EINVAL; | |
489 | ||
592867bf | 490 | f = fdget(ufd); |
c2101297 | 491 | map = __bpf_map_get(f); |
db20fd2b AS |
492 | if (IS_ERR(map)) |
493 | return PTR_ERR(map); | |
494 | ||
e4448ed8 AV |
495 | key = memdup_user(ukey, map->key_size); |
496 | if (IS_ERR(key)) { | |
497 | err = PTR_ERR(key); | |
db20fd2b | 498 | goto err_put; |
e4448ed8 | 499 | } |
db20fd2b | 500 | |
15a07b33 | 501 | if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || |
8f844938 | 502 | map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH || |
15a07b33 AS |
503 | map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) |
504 | value_size = round_up(map->value_size, 8) * num_possible_cpus(); | |
505 | else | |
506 | value_size = map->value_size; | |
507 | ||
db20fd2b | 508 | err = -ENOMEM; |
15a07b33 | 509 | value = kmalloc(value_size, GFP_USER | __GFP_NOWARN); |
db20fd2b AS |
510 | if (!value) |
511 | goto free_key; | |
512 | ||
513 | err = -EFAULT; | |
15a07b33 | 514 | if (copy_from_user(value, uvalue, value_size) != 0) |
db20fd2b AS |
515 | goto free_value; |
516 | ||
b121d1e7 AS |
517 | /* must increment bpf_prog_active to avoid kprobe+bpf triggering from |
518 | * inside bpf map update or delete otherwise deadlocks are possible | |
519 | */ | |
520 | preempt_disable(); | |
521 | __this_cpu_inc(bpf_prog_active); | |
8f844938 MKL |
522 | if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || |
523 | map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) { | |
15a07b33 AS |
524 | err = bpf_percpu_hash_update(map, key, value, attr->flags); |
525 | } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) { | |
526 | err = bpf_percpu_array_update(map, key, value, attr->flags); | |
d056a788 | 527 | } else if (map->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || |
4ed8ec52 | 528 | map->map_type == BPF_MAP_TYPE_PROG_ARRAY || |
56f668df MKL |
529 | map->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || |
530 | map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS) { | |
d056a788 DB |
531 | rcu_read_lock(); |
532 | err = bpf_fd_array_map_update_elem(map, f.file, key, value, | |
533 | attr->flags); | |
534 | rcu_read_unlock(); | |
bcc6b1b7 MKL |
535 | } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) { |
536 | rcu_read_lock(); | |
537 | err = bpf_fd_htab_map_update_elem(map, f.file, key, value, | |
538 | attr->flags); | |
539 | rcu_read_unlock(); | |
15a07b33 AS |
540 | } else { |
541 | rcu_read_lock(); | |
542 | err = map->ops->map_update_elem(map, key, value, attr->flags); | |
543 | rcu_read_unlock(); | |
544 | } | |
b121d1e7 AS |
545 | __this_cpu_dec(bpf_prog_active); |
546 | preempt_enable(); | |
db20fd2b | 547 | |
a67edbf4 DB |
548 | if (!err) |
549 | trace_bpf_map_update_elem(map, ufd, key, value); | |
db20fd2b AS |
550 | free_value: |
551 | kfree(value); | |
552 | free_key: | |
553 | kfree(key); | |
554 | err_put: | |
555 | fdput(f); | |
556 | return err; | |
557 | } | |
558 | ||
559 | #define BPF_MAP_DELETE_ELEM_LAST_FIELD key | |
560 | ||
561 | static int map_delete_elem(union bpf_attr *attr) | |
562 | { | |
535e7b4b | 563 | void __user *ukey = u64_to_user_ptr(attr->key); |
db20fd2b | 564 | int ufd = attr->map_fd; |
db20fd2b | 565 | struct bpf_map *map; |
592867bf | 566 | struct fd f; |
db20fd2b AS |
567 | void *key; |
568 | int err; | |
569 | ||
570 | if (CHECK_ATTR(BPF_MAP_DELETE_ELEM)) | |
571 | return -EINVAL; | |
572 | ||
592867bf | 573 | f = fdget(ufd); |
c2101297 | 574 | map = __bpf_map_get(f); |
db20fd2b AS |
575 | if (IS_ERR(map)) |
576 | return PTR_ERR(map); | |
577 | ||
e4448ed8 AV |
578 | key = memdup_user(ukey, map->key_size); |
579 | if (IS_ERR(key)) { | |
580 | err = PTR_ERR(key); | |
db20fd2b | 581 | goto err_put; |
e4448ed8 | 582 | } |
db20fd2b | 583 | |
b121d1e7 AS |
584 | preempt_disable(); |
585 | __this_cpu_inc(bpf_prog_active); | |
db20fd2b AS |
586 | rcu_read_lock(); |
587 | err = map->ops->map_delete_elem(map, key); | |
588 | rcu_read_unlock(); | |
b121d1e7 AS |
589 | __this_cpu_dec(bpf_prog_active); |
590 | preempt_enable(); | |
db20fd2b | 591 | |
a67edbf4 DB |
592 | if (!err) |
593 | trace_bpf_map_delete_elem(map, ufd, key); | |
db20fd2b AS |
594 | kfree(key); |
595 | err_put: | |
596 | fdput(f); | |
597 | return err; | |
598 | } | |
599 | ||
600 | /* last field in 'union bpf_attr' used by this command */ | |
601 | #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key | |
602 | ||
603 | static int map_get_next_key(union bpf_attr *attr) | |
604 | { | |
535e7b4b MS |
605 | void __user *ukey = u64_to_user_ptr(attr->key); |
606 | void __user *unext_key = u64_to_user_ptr(attr->next_key); | |
db20fd2b | 607 | int ufd = attr->map_fd; |
db20fd2b AS |
608 | struct bpf_map *map; |
609 | void *key, *next_key; | |
592867bf | 610 | struct fd f; |
db20fd2b AS |
611 | int err; |
612 | ||
613 | if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY)) | |
614 | return -EINVAL; | |
615 | ||
592867bf | 616 | f = fdget(ufd); |
c2101297 | 617 | map = __bpf_map_get(f); |
db20fd2b AS |
618 | if (IS_ERR(map)) |
619 | return PTR_ERR(map); | |
620 | ||
8fe45924 | 621 | if (ukey) { |
e4448ed8 AV |
622 | key = memdup_user(ukey, map->key_size); |
623 | if (IS_ERR(key)) { | |
624 | err = PTR_ERR(key); | |
8fe45924 | 625 | goto err_put; |
e4448ed8 | 626 | } |
8fe45924 TQ |
627 | } else { |
628 | key = NULL; | |
629 | } | |
db20fd2b AS |
630 | |
631 | err = -ENOMEM; | |
632 | next_key = kmalloc(map->key_size, GFP_USER); | |
633 | if (!next_key) | |
634 | goto free_key; | |
635 | ||
636 | rcu_read_lock(); | |
637 | err = map->ops->map_get_next_key(map, key, next_key); | |
638 | rcu_read_unlock(); | |
639 | if (err) | |
640 | goto free_next_key; | |
641 | ||
642 | err = -EFAULT; | |
643 | if (copy_to_user(unext_key, next_key, map->key_size) != 0) | |
644 | goto free_next_key; | |
645 | ||
a67edbf4 | 646 | trace_bpf_map_next_key(map, ufd, key, next_key); |
db20fd2b AS |
647 | err = 0; |
648 | ||
649 | free_next_key: | |
650 | kfree(next_key); | |
651 | free_key: | |
652 | kfree(key); | |
653 | err_put: | |
654 | fdput(f); | |
655 | return err; | |
656 | } | |
657 | ||
be9370a7 JB |
658 | static const struct bpf_verifier_ops * const bpf_prog_types[] = { |
659 | #define BPF_PROG_TYPE(_id, _ops) \ | |
660 | [_id] = &_ops, | |
40077e0c | 661 | #define BPF_MAP_TYPE(_id, _ops) |
be9370a7 JB |
662 | #include <linux/bpf_types.h> |
663 | #undef BPF_PROG_TYPE | |
40077e0c | 664 | #undef BPF_MAP_TYPE |
be9370a7 | 665 | }; |
09756af4 AS |
666 | |
667 | static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog) | |
668 | { | |
be9370a7 JB |
669 | if (type >= ARRAY_SIZE(bpf_prog_types) || !bpf_prog_types[type]) |
670 | return -EINVAL; | |
09756af4 | 671 | |
be9370a7 JB |
672 | prog->aux->ops = bpf_prog_types[type]; |
673 | prog->type = type; | |
674 | return 0; | |
09756af4 AS |
675 | } |
676 | ||
677 | /* drop refcnt on maps used by eBPF program and free auxilary data */ | |
678 | static void free_used_maps(struct bpf_prog_aux *aux) | |
679 | { | |
680 | int i; | |
681 | ||
682 | for (i = 0; i < aux->used_map_cnt; i++) | |
683 | bpf_map_put(aux->used_maps[i]); | |
684 | ||
685 | kfree(aux->used_maps); | |
686 | } | |
687 | ||
5ccb071e DB |
688 | int __bpf_prog_charge(struct user_struct *user, u32 pages) |
689 | { | |
690 | unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
691 | unsigned long user_bufs; | |
692 | ||
693 | if (user) { | |
694 | user_bufs = atomic_long_add_return(pages, &user->locked_vm); | |
695 | if (user_bufs > memlock_limit) { | |
696 | atomic_long_sub(pages, &user->locked_vm); | |
697 | return -EPERM; | |
698 | } | |
699 | } | |
700 | ||
701 | return 0; | |
702 | } | |
703 | ||
704 | void __bpf_prog_uncharge(struct user_struct *user, u32 pages) | |
705 | { | |
706 | if (user) | |
707 | atomic_long_sub(pages, &user->locked_vm); | |
708 | } | |
709 | ||
aaac3ba9 AS |
710 | static int bpf_prog_charge_memlock(struct bpf_prog *prog) |
711 | { | |
712 | struct user_struct *user = get_current_user(); | |
5ccb071e | 713 | int ret; |
aaac3ba9 | 714 | |
5ccb071e DB |
715 | ret = __bpf_prog_charge(user, prog->pages); |
716 | if (ret) { | |
aaac3ba9 | 717 | free_uid(user); |
5ccb071e | 718 | return ret; |
aaac3ba9 | 719 | } |
5ccb071e | 720 | |
aaac3ba9 AS |
721 | prog->aux->user = user; |
722 | return 0; | |
723 | } | |
724 | ||
725 | static void bpf_prog_uncharge_memlock(struct bpf_prog *prog) | |
726 | { | |
727 | struct user_struct *user = prog->aux->user; | |
728 | ||
5ccb071e | 729 | __bpf_prog_uncharge(user, prog->pages); |
aaac3ba9 AS |
730 | free_uid(user); |
731 | } | |
732 | ||
dc4bb0e2 MKL |
733 | static int bpf_prog_alloc_id(struct bpf_prog *prog) |
734 | { | |
735 | int id; | |
736 | ||
737 | spin_lock_bh(&prog_idr_lock); | |
738 | id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC); | |
739 | if (id > 0) | |
740 | prog->aux->id = id; | |
741 | spin_unlock_bh(&prog_idr_lock); | |
742 | ||
743 | /* id is in [1, INT_MAX) */ | |
744 | if (WARN_ON_ONCE(!id)) | |
745 | return -ENOSPC; | |
746 | ||
747 | return id > 0 ? 0 : id; | |
748 | } | |
749 | ||
b16d9aa4 | 750 | static void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock) |
dc4bb0e2 MKL |
751 | { |
752 | /* cBPF to eBPF migrations are currently not in the idr store. */ | |
753 | if (!prog->aux->id) | |
754 | return; | |
755 | ||
b16d9aa4 MKL |
756 | if (do_idr_lock) |
757 | spin_lock_bh(&prog_idr_lock); | |
758 | else | |
759 | __acquire(&prog_idr_lock); | |
760 | ||
dc4bb0e2 | 761 | idr_remove(&prog_idr, prog->aux->id); |
b16d9aa4 MKL |
762 | |
763 | if (do_idr_lock) | |
764 | spin_unlock_bh(&prog_idr_lock); | |
765 | else | |
766 | __release(&prog_idr_lock); | |
dc4bb0e2 MKL |
767 | } |
768 | ||
1aacde3d | 769 | static void __bpf_prog_put_rcu(struct rcu_head *rcu) |
abf2e7d6 AS |
770 | { |
771 | struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu); | |
772 | ||
773 | free_used_maps(aux); | |
aaac3ba9 | 774 | bpf_prog_uncharge_memlock(aux->prog); |
abf2e7d6 AS |
775 | bpf_prog_free(aux->prog); |
776 | } | |
777 | ||
b16d9aa4 | 778 | static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock) |
09756af4 | 779 | { |
a67edbf4 DB |
780 | if (atomic_dec_and_test(&prog->aux->refcnt)) { |
781 | trace_bpf_prog_put_rcu(prog); | |
34ad5580 | 782 | /* bpf_prog_free_id() must be called first */ |
b16d9aa4 | 783 | bpf_prog_free_id(prog, do_idr_lock); |
74451e66 | 784 | bpf_prog_kallsyms_del(prog); |
1aacde3d | 785 | call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu); |
a67edbf4 | 786 | } |
09756af4 | 787 | } |
b16d9aa4 MKL |
788 | |
789 | void bpf_prog_put(struct bpf_prog *prog) | |
790 | { | |
791 | __bpf_prog_put(prog, true); | |
792 | } | |
e2e9b654 | 793 | EXPORT_SYMBOL_GPL(bpf_prog_put); |
09756af4 AS |
794 | |
795 | static int bpf_prog_release(struct inode *inode, struct file *filp) | |
796 | { | |
797 | struct bpf_prog *prog = filp->private_data; | |
798 | ||
1aacde3d | 799 | bpf_prog_put(prog); |
09756af4 AS |
800 | return 0; |
801 | } | |
802 | ||
7bd509e3 DB |
803 | #ifdef CONFIG_PROC_FS |
804 | static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp) | |
805 | { | |
806 | const struct bpf_prog *prog = filp->private_data; | |
f1f7714e | 807 | char prog_tag[sizeof(prog->tag) * 2 + 1] = { }; |
7bd509e3 | 808 | |
f1f7714e | 809 | bin2hex(prog_tag, prog->tag, sizeof(prog->tag)); |
7bd509e3 DB |
810 | seq_printf(m, |
811 | "prog_type:\t%u\n" | |
812 | "prog_jited:\t%u\n" | |
f1f7714e | 813 | "prog_tag:\t%s\n" |
7bd509e3 DB |
814 | "memlock:\t%llu\n", |
815 | prog->type, | |
816 | prog->jited, | |
f1f7714e | 817 | prog_tag, |
7bd509e3 DB |
818 | prog->pages * 1ULL << PAGE_SHIFT); |
819 | } | |
820 | #endif | |
821 | ||
09756af4 | 822 | static const struct file_operations bpf_prog_fops = { |
7bd509e3 DB |
823 | #ifdef CONFIG_PROC_FS |
824 | .show_fdinfo = bpf_prog_show_fdinfo, | |
825 | #endif | |
826 | .release = bpf_prog_release, | |
09756af4 AS |
827 | }; |
828 | ||
b2197755 | 829 | int bpf_prog_new_fd(struct bpf_prog *prog) |
aa79781b DB |
830 | { |
831 | return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog, | |
832 | O_RDWR | O_CLOEXEC); | |
833 | } | |
834 | ||
113214be | 835 | static struct bpf_prog *____bpf_prog_get(struct fd f) |
09756af4 | 836 | { |
09756af4 AS |
837 | if (!f.file) |
838 | return ERR_PTR(-EBADF); | |
09756af4 AS |
839 | if (f.file->f_op != &bpf_prog_fops) { |
840 | fdput(f); | |
841 | return ERR_PTR(-EINVAL); | |
842 | } | |
843 | ||
c2101297 | 844 | return f.file->private_data; |
09756af4 AS |
845 | } |
846 | ||
59d3656d | 847 | struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i) |
92117d84 | 848 | { |
59d3656d BB |
849 | if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) { |
850 | atomic_sub(i, &prog->aux->refcnt); | |
92117d84 AS |
851 | return ERR_PTR(-EBUSY); |
852 | } | |
853 | return prog; | |
854 | } | |
59d3656d BB |
855 | EXPORT_SYMBOL_GPL(bpf_prog_add); |
856 | ||
c540594f DB |
857 | void bpf_prog_sub(struct bpf_prog *prog, int i) |
858 | { | |
859 | /* Only to be used for undoing previous bpf_prog_add() in some | |
860 | * error path. We still know that another entity in our call | |
861 | * path holds a reference to the program, thus atomic_sub() can | |
862 | * be safely used in such cases! | |
863 | */ | |
864 | WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0); | |
865 | } | |
866 | EXPORT_SYMBOL_GPL(bpf_prog_sub); | |
867 | ||
59d3656d BB |
868 | struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog) |
869 | { | |
870 | return bpf_prog_add(prog, 1); | |
871 | } | |
97bc402d | 872 | EXPORT_SYMBOL_GPL(bpf_prog_inc); |
92117d84 | 873 | |
b16d9aa4 MKL |
874 | /* prog_idr_lock should have been held */ |
875 | static struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog) | |
876 | { | |
877 | int refold; | |
878 | ||
879 | refold = __atomic_add_unless(&prog->aux->refcnt, 1, 0); | |
880 | ||
881 | if (refold >= BPF_MAX_REFCNT) { | |
882 | __bpf_prog_put(prog, false); | |
883 | return ERR_PTR(-EBUSY); | |
884 | } | |
885 | ||
886 | if (!refold) | |
887 | return ERR_PTR(-ENOENT); | |
888 | ||
889 | return prog; | |
890 | } | |
891 | ||
113214be | 892 | static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *type) |
09756af4 AS |
893 | { |
894 | struct fd f = fdget(ufd); | |
895 | struct bpf_prog *prog; | |
896 | ||
113214be | 897 | prog = ____bpf_prog_get(f); |
09756af4 AS |
898 | if (IS_ERR(prog)) |
899 | return prog; | |
113214be DB |
900 | if (type && prog->type != *type) { |
901 | prog = ERR_PTR(-EINVAL); | |
902 | goto out; | |
903 | } | |
09756af4 | 904 | |
92117d84 | 905 | prog = bpf_prog_inc(prog); |
113214be | 906 | out: |
09756af4 AS |
907 | fdput(f); |
908 | return prog; | |
909 | } | |
113214be DB |
910 | |
911 | struct bpf_prog *bpf_prog_get(u32 ufd) | |
912 | { | |
913 | return __bpf_prog_get(ufd, NULL); | |
914 | } | |
915 | ||
916 | struct bpf_prog *bpf_prog_get_type(u32 ufd, enum bpf_prog_type type) | |
917 | { | |
a67edbf4 DB |
918 | struct bpf_prog *prog = __bpf_prog_get(ufd, &type); |
919 | ||
920 | if (!IS_ERR(prog)) | |
921 | trace_bpf_prog_get_type(prog); | |
922 | return prog; | |
113214be DB |
923 | } |
924 | EXPORT_SYMBOL_GPL(bpf_prog_get_type); | |
09756af4 AS |
925 | |
926 | /* last field in 'union bpf_attr' used by this command */ | |
e07b98d9 | 927 | #define BPF_PROG_LOAD_LAST_FIELD prog_flags |
09756af4 AS |
928 | |
929 | static int bpf_prog_load(union bpf_attr *attr) | |
930 | { | |
931 | enum bpf_prog_type type = attr->prog_type; | |
932 | struct bpf_prog *prog; | |
933 | int err; | |
934 | char license[128]; | |
935 | bool is_gpl; | |
936 | ||
937 | if (CHECK_ATTR(BPF_PROG_LOAD)) | |
938 | return -EINVAL; | |
939 | ||
e07b98d9 DM |
940 | if (attr->prog_flags & ~BPF_F_STRICT_ALIGNMENT) |
941 | return -EINVAL; | |
942 | ||
09756af4 | 943 | /* copy eBPF program license from user space */ |
535e7b4b | 944 | if (strncpy_from_user(license, u64_to_user_ptr(attr->license), |
09756af4 AS |
945 | sizeof(license) - 1) < 0) |
946 | return -EFAULT; | |
947 | license[sizeof(license) - 1] = 0; | |
948 | ||
949 | /* eBPF programs must be GPL compatible to use GPL-ed functions */ | |
950 | is_gpl = license_is_gpl_compatible(license); | |
951 | ||
ef0915ca DB |
952 | if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS) |
953 | return -E2BIG; | |
09756af4 | 954 | |
2541517c AS |
955 | if (type == BPF_PROG_TYPE_KPROBE && |
956 | attr->kern_version != LINUX_VERSION_CODE) | |
957 | return -EINVAL; | |
958 | ||
80b7d819 CF |
959 | if (type != BPF_PROG_TYPE_SOCKET_FILTER && |
960 | type != BPF_PROG_TYPE_CGROUP_SKB && | |
961 | !capable(CAP_SYS_ADMIN)) | |
1be7f75d AS |
962 | return -EPERM; |
963 | ||
09756af4 AS |
964 | /* plain bpf_prog allocation */ |
965 | prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER); | |
966 | if (!prog) | |
967 | return -ENOMEM; | |
968 | ||
aaac3ba9 AS |
969 | err = bpf_prog_charge_memlock(prog); |
970 | if (err) | |
971 | goto free_prog_nouncharge; | |
972 | ||
09756af4 AS |
973 | prog->len = attr->insn_cnt; |
974 | ||
975 | err = -EFAULT; | |
535e7b4b | 976 | if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns), |
aafe6ae9 | 977 | bpf_prog_insn_size(prog)) != 0) |
09756af4 AS |
978 | goto free_prog; |
979 | ||
980 | prog->orig_prog = NULL; | |
a91263d5 | 981 | prog->jited = 0; |
09756af4 AS |
982 | |
983 | atomic_set(&prog->aux->refcnt, 1); | |
a91263d5 | 984 | prog->gpl_compatible = is_gpl ? 1 : 0; |
09756af4 AS |
985 | |
986 | /* find program type: socket_filter vs tracing_filter */ | |
987 | err = find_prog_type(type, prog); | |
988 | if (err < 0) | |
989 | goto free_prog; | |
990 | ||
991 | /* run eBPF verifier */ | |
9bac3d6d | 992 | err = bpf_check(&prog, attr); |
09756af4 AS |
993 | if (err < 0) |
994 | goto free_used_maps; | |
995 | ||
996 | /* eBPF program is ready to be JITed */ | |
d1c55ab5 | 997 | prog = bpf_prog_select_runtime(prog, &err); |
04fd61ab AS |
998 | if (err < 0) |
999 | goto free_used_maps; | |
09756af4 | 1000 | |
dc4bb0e2 MKL |
1001 | err = bpf_prog_alloc_id(prog); |
1002 | if (err) | |
1003 | goto free_used_maps; | |
1004 | ||
aa79781b | 1005 | err = bpf_prog_new_fd(prog); |
b16d9aa4 MKL |
1006 | if (err < 0) { |
1007 | /* failed to allocate fd. | |
1008 | * bpf_prog_put() is needed because the above | |
1009 | * bpf_prog_alloc_id() has published the prog | |
1010 | * to the userspace and the userspace may | |
1011 | * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID. | |
1012 | */ | |
1013 | bpf_prog_put(prog); | |
1014 | return err; | |
1015 | } | |
09756af4 | 1016 | |
74451e66 | 1017 | bpf_prog_kallsyms_add(prog); |
a67edbf4 | 1018 | trace_bpf_prog_load(prog, err); |
09756af4 AS |
1019 | return err; |
1020 | ||
1021 | free_used_maps: | |
1022 | free_used_maps(prog->aux); | |
1023 | free_prog: | |
aaac3ba9 AS |
1024 | bpf_prog_uncharge_memlock(prog); |
1025 | free_prog_nouncharge: | |
09756af4 AS |
1026 | bpf_prog_free(prog); |
1027 | return err; | |
1028 | } | |
1029 | ||
b2197755 DB |
1030 | #define BPF_OBJ_LAST_FIELD bpf_fd |
1031 | ||
1032 | static int bpf_obj_pin(const union bpf_attr *attr) | |
1033 | { | |
1034 | if (CHECK_ATTR(BPF_OBJ)) | |
1035 | return -EINVAL; | |
1036 | ||
535e7b4b | 1037 | return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname)); |
b2197755 DB |
1038 | } |
1039 | ||
1040 | static int bpf_obj_get(const union bpf_attr *attr) | |
1041 | { | |
1042 | if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0) | |
1043 | return -EINVAL; | |
1044 | ||
535e7b4b | 1045 | return bpf_obj_get_user(u64_to_user_ptr(attr->pathname)); |
b2197755 DB |
1046 | } |
1047 | ||
f4324551 DM |
1048 | #ifdef CONFIG_CGROUP_BPF |
1049 | ||
7f677633 | 1050 | #define BPF_PROG_ATTACH_LAST_FIELD attach_flags |
f4324551 DM |
1051 | |
1052 | static int bpf_prog_attach(const union bpf_attr *attr) | |
1053 | { | |
7f677633 | 1054 | enum bpf_prog_type ptype; |
f4324551 DM |
1055 | struct bpf_prog *prog; |
1056 | struct cgroup *cgrp; | |
7f677633 | 1057 | int ret; |
f4324551 DM |
1058 | |
1059 | if (!capable(CAP_NET_ADMIN)) | |
1060 | return -EPERM; | |
1061 | ||
1062 | if (CHECK_ATTR(BPF_PROG_ATTACH)) | |
1063 | return -EINVAL; | |
1064 | ||
7f677633 AS |
1065 | if (attr->attach_flags & ~BPF_F_ALLOW_OVERRIDE) |
1066 | return -EINVAL; | |
1067 | ||
f4324551 DM |
1068 | switch (attr->attach_type) { |
1069 | case BPF_CGROUP_INET_INGRESS: | |
1070 | case BPF_CGROUP_INET_EGRESS: | |
b2cd1257 | 1071 | ptype = BPF_PROG_TYPE_CGROUP_SKB; |
f4324551 | 1072 | break; |
61023658 DA |
1073 | case BPF_CGROUP_INET_SOCK_CREATE: |
1074 | ptype = BPF_PROG_TYPE_CGROUP_SOCK; | |
1075 | break; | |
40304b2a LB |
1076 | case BPF_CGROUP_SOCK_OPS: |
1077 | ptype = BPF_PROG_TYPE_SOCK_OPS; | |
1078 | break; | |
f4324551 DM |
1079 | default: |
1080 | return -EINVAL; | |
1081 | } | |
1082 | ||
b2cd1257 DA |
1083 | prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype); |
1084 | if (IS_ERR(prog)) | |
1085 | return PTR_ERR(prog); | |
1086 | ||
1087 | cgrp = cgroup_get_from_fd(attr->target_fd); | |
1088 | if (IS_ERR(cgrp)) { | |
1089 | bpf_prog_put(prog); | |
1090 | return PTR_ERR(cgrp); | |
1091 | } | |
1092 | ||
7f677633 AS |
1093 | ret = cgroup_bpf_update(cgrp, prog, attr->attach_type, |
1094 | attr->attach_flags & BPF_F_ALLOW_OVERRIDE); | |
1095 | if (ret) | |
1096 | bpf_prog_put(prog); | |
b2cd1257 DA |
1097 | cgroup_put(cgrp); |
1098 | ||
7f677633 | 1099 | return ret; |
f4324551 DM |
1100 | } |
1101 | ||
1102 | #define BPF_PROG_DETACH_LAST_FIELD attach_type | |
1103 | ||
1104 | static int bpf_prog_detach(const union bpf_attr *attr) | |
1105 | { | |
1106 | struct cgroup *cgrp; | |
7f677633 | 1107 | int ret; |
f4324551 DM |
1108 | |
1109 | if (!capable(CAP_NET_ADMIN)) | |
1110 | return -EPERM; | |
1111 | ||
1112 | if (CHECK_ATTR(BPF_PROG_DETACH)) | |
1113 | return -EINVAL; | |
1114 | ||
1115 | switch (attr->attach_type) { | |
1116 | case BPF_CGROUP_INET_INGRESS: | |
1117 | case BPF_CGROUP_INET_EGRESS: | |
61023658 | 1118 | case BPF_CGROUP_INET_SOCK_CREATE: |
40304b2a | 1119 | case BPF_CGROUP_SOCK_OPS: |
f4324551 DM |
1120 | cgrp = cgroup_get_from_fd(attr->target_fd); |
1121 | if (IS_ERR(cgrp)) | |
1122 | return PTR_ERR(cgrp); | |
1123 | ||
7f677633 | 1124 | ret = cgroup_bpf_update(cgrp, NULL, attr->attach_type, false); |
f4324551 DM |
1125 | cgroup_put(cgrp); |
1126 | break; | |
1127 | ||
1128 | default: | |
1129 | return -EINVAL; | |
1130 | } | |
1131 | ||
7f677633 | 1132 | return ret; |
f4324551 | 1133 | } |
40304b2a | 1134 | |
f4324551 DM |
1135 | #endif /* CONFIG_CGROUP_BPF */ |
1136 | ||
1cf1cae9 AS |
1137 | #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration |
1138 | ||
1139 | static int bpf_prog_test_run(const union bpf_attr *attr, | |
1140 | union bpf_attr __user *uattr) | |
1141 | { | |
1142 | struct bpf_prog *prog; | |
1143 | int ret = -ENOTSUPP; | |
1144 | ||
1145 | if (CHECK_ATTR(BPF_PROG_TEST_RUN)) | |
1146 | return -EINVAL; | |
1147 | ||
1148 | prog = bpf_prog_get(attr->test.prog_fd); | |
1149 | if (IS_ERR(prog)) | |
1150 | return PTR_ERR(prog); | |
1151 | ||
1152 | if (prog->aux->ops->test_run) | |
1153 | ret = prog->aux->ops->test_run(prog, attr, uattr); | |
1154 | ||
1155 | bpf_prog_put(prog); | |
1156 | return ret; | |
1157 | } | |
1158 | ||
34ad5580 MKL |
1159 | #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id |
1160 | ||
1161 | static int bpf_obj_get_next_id(const union bpf_attr *attr, | |
1162 | union bpf_attr __user *uattr, | |
1163 | struct idr *idr, | |
1164 | spinlock_t *lock) | |
1165 | { | |
1166 | u32 next_id = attr->start_id; | |
1167 | int err = 0; | |
1168 | ||
1169 | if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX) | |
1170 | return -EINVAL; | |
1171 | ||
1172 | if (!capable(CAP_SYS_ADMIN)) | |
1173 | return -EPERM; | |
1174 | ||
1175 | next_id++; | |
1176 | spin_lock_bh(lock); | |
1177 | if (!idr_get_next(idr, &next_id)) | |
1178 | err = -ENOENT; | |
1179 | spin_unlock_bh(lock); | |
1180 | ||
1181 | if (!err) | |
1182 | err = put_user(next_id, &uattr->next_id); | |
1183 | ||
1184 | return err; | |
1185 | } | |
1186 | ||
b16d9aa4 MKL |
1187 | #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id |
1188 | ||
1189 | static int bpf_prog_get_fd_by_id(const union bpf_attr *attr) | |
1190 | { | |
1191 | struct bpf_prog *prog; | |
1192 | u32 id = attr->prog_id; | |
1193 | int fd; | |
1194 | ||
1195 | if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID)) | |
1196 | return -EINVAL; | |
1197 | ||
1198 | if (!capable(CAP_SYS_ADMIN)) | |
1199 | return -EPERM; | |
1200 | ||
1201 | spin_lock_bh(&prog_idr_lock); | |
1202 | prog = idr_find(&prog_idr, id); | |
1203 | if (prog) | |
1204 | prog = bpf_prog_inc_not_zero(prog); | |
1205 | else | |
1206 | prog = ERR_PTR(-ENOENT); | |
1207 | spin_unlock_bh(&prog_idr_lock); | |
1208 | ||
1209 | if (IS_ERR(prog)) | |
1210 | return PTR_ERR(prog); | |
1211 | ||
1212 | fd = bpf_prog_new_fd(prog); | |
1213 | if (fd < 0) | |
1214 | bpf_prog_put(prog); | |
1215 | ||
1216 | return fd; | |
1217 | } | |
1218 | ||
bd5f5f4e MKL |
1219 | #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD map_id |
1220 | ||
1221 | static int bpf_map_get_fd_by_id(const union bpf_attr *attr) | |
1222 | { | |
1223 | struct bpf_map *map; | |
1224 | u32 id = attr->map_id; | |
1225 | int fd; | |
1226 | ||
1227 | if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID)) | |
1228 | return -EINVAL; | |
1229 | ||
1230 | if (!capable(CAP_SYS_ADMIN)) | |
1231 | return -EPERM; | |
1232 | ||
1233 | spin_lock_bh(&map_idr_lock); | |
1234 | map = idr_find(&map_idr, id); | |
1235 | if (map) | |
1236 | map = bpf_map_inc_not_zero(map, true); | |
1237 | else | |
1238 | map = ERR_PTR(-ENOENT); | |
1239 | spin_unlock_bh(&map_idr_lock); | |
1240 | ||
1241 | if (IS_ERR(map)) | |
1242 | return PTR_ERR(map); | |
1243 | ||
1244 | fd = bpf_map_new_fd(map); | |
1245 | if (fd < 0) | |
1246 | bpf_map_put(map); | |
1247 | ||
1248 | return fd; | |
1249 | } | |
1250 | ||
1e270976 MKL |
1251 | static int check_uarg_tail_zero(void __user *uaddr, |
1252 | size_t expected_size, | |
1253 | size_t actual_size) | |
1254 | { | |
1255 | unsigned char __user *addr; | |
1256 | unsigned char __user *end; | |
1257 | unsigned char val; | |
1258 | int err; | |
1259 | ||
1260 | if (actual_size <= expected_size) | |
1261 | return 0; | |
1262 | ||
1263 | addr = uaddr + expected_size; | |
1264 | end = uaddr + actual_size; | |
1265 | ||
1266 | for (; addr < end; addr++) { | |
1267 | err = get_user(val, addr); | |
1268 | if (err) | |
1269 | return err; | |
1270 | if (val) | |
1271 | return -E2BIG; | |
1272 | } | |
1273 | ||
1274 | return 0; | |
1275 | } | |
1276 | ||
1277 | static int bpf_prog_get_info_by_fd(struct bpf_prog *prog, | |
1278 | const union bpf_attr *attr, | |
1279 | union bpf_attr __user *uattr) | |
1280 | { | |
1281 | struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info); | |
1282 | struct bpf_prog_info info = {}; | |
1283 | u32 info_len = attr->info.info_len; | |
1284 | char __user *uinsns; | |
1285 | u32 ulen; | |
1286 | int err; | |
1287 | ||
1288 | err = check_uarg_tail_zero(uinfo, sizeof(info), info_len); | |
1289 | if (err) | |
1290 | return err; | |
1291 | info_len = min_t(u32, sizeof(info), info_len); | |
1292 | ||
1293 | if (copy_from_user(&info, uinfo, info_len)) | |
89b09689 | 1294 | return -EFAULT; |
1e270976 MKL |
1295 | |
1296 | info.type = prog->type; | |
1297 | info.id = prog->aux->id; | |
1298 | ||
1299 | memcpy(info.tag, prog->tag, sizeof(prog->tag)); | |
1300 | ||
1301 | if (!capable(CAP_SYS_ADMIN)) { | |
1302 | info.jited_prog_len = 0; | |
1303 | info.xlated_prog_len = 0; | |
1304 | goto done; | |
1305 | } | |
1306 | ||
1307 | ulen = info.jited_prog_len; | |
1308 | info.jited_prog_len = prog->jited_len; | |
1309 | if (info.jited_prog_len && ulen) { | |
1310 | uinsns = u64_to_user_ptr(info.jited_prog_insns); | |
1311 | ulen = min_t(u32, info.jited_prog_len, ulen); | |
1312 | if (copy_to_user(uinsns, prog->bpf_func, ulen)) | |
1313 | return -EFAULT; | |
1314 | } | |
1315 | ||
1316 | ulen = info.xlated_prog_len; | |
9975a54b | 1317 | info.xlated_prog_len = bpf_prog_insn_size(prog); |
1e270976 MKL |
1318 | if (info.xlated_prog_len && ulen) { |
1319 | uinsns = u64_to_user_ptr(info.xlated_prog_insns); | |
1320 | ulen = min_t(u32, info.xlated_prog_len, ulen); | |
1321 | if (copy_to_user(uinsns, prog->insnsi, ulen)) | |
1322 | return -EFAULT; | |
1323 | } | |
1324 | ||
1325 | done: | |
1326 | if (copy_to_user(uinfo, &info, info_len) || | |
1327 | put_user(info_len, &uattr->info.info_len)) | |
1328 | return -EFAULT; | |
1329 | ||
1330 | return 0; | |
1331 | } | |
1332 | ||
1333 | static int bpf_map_get_info_by_fd(struct bpf_map *map, | |
1334 | const union bpf_attr *attr, | |
1335 | union bpf_attr __user *uattr) | |
1336 | { | |
1337 | struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info); | |
1338 | struct bpf_map_info info = {}; | |
1339 | u32 info_len = attr->info.info_len; | |
1340 | int err; | |
1341 | ||
1342 | err = check_uarg_tail_zero(uinfo, sizeof(info), info_len); | |
1343 | if (err) | |
1344 | return err; | |
1345 | info_len = min_t(u32, sizeof(info), info_len); | |
1346 | ||
1347 | info.type = map->map_type; | |
1348 | info.id = map->id; | |
1349 | info.key_size = map->key_size; | |
1350 | info.value_size = map->value_size; | |
1351 | info.max_entries = map->max_entries; | |
1352 | info.map_flags = map->map_flags; | |
1353 | ||
1354 | if (copy_to_user(uinfo, &info, info_len) || | |
1355 | put_user(info_len, &uattr->info.info_len)) | |
1356 | return -EFAULT; | |
1357 | ||
1358 | return 0; | |
1359 | } | |
1360 | ||
1361 | #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info | |
1362 | ||
1363 | static int bpf_obj_get_info_by_fd(const union bpf_attr *attr, | |
1364 | union bpf_attr __user *uattr) | |
1365 | { | |
1366 | int ufd = attr->info.bpf_fd; | |
1367 | struct fd f; | |
1368 | int err; | |
1369 | ||
1370 | if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD)) | |
1371 | return -EINVAL; | |
1372 | ||
1373 | f = fdget(ufd); | |
1374 | if (!f.file) | |
1375 | return -EBADFD; | |
1376 | ||
1377 | if (f.file->f_op == &bpf_prog_fops) | |
1378 | err = bpf_prog_get_info_by_fd(f.file->private_data, attr, | |
1379 | uattr); | |
1380 | else if (f.file->f_op == &bpf_map_fops) | |
1381 | err = bpf_map_get_info_by_fd(f.file->private_data, attr, | |
1382 | uattr); | |
1383 | else | |
1384 | err = -EINVAL; | |
1385 | ||
1386 | fdput(f); | |
1387 | return err; | |
1388 | } | |
1389 | ||
99c55f7d AS |
1390 | SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size) |
1391 | { | |
1392 | union bpf_attr attr = {}; | |
1393 | int err; | |
1394 | ||
1be7f75d | 1395 | if (!capable(CAP_SYS_ADMIN) && sysctl_unprivileged_bpf_disabled) |
99c55f7d AS |
1396 | return -EPERM; |
1397 | ||
1398 | if (!access_ok(VERIFY_READ, uattr, 1)) | |
1399 | return -EFAULT; | |
1400 | ||
1401 | if (size > PAGE_SIZE) /* silly large */ | |
1402 | return -E2BIG; | |
1403 | ||
1404 | /* If we're handed a bigger struct than we know of, | |
1405 | * ensure all the unknown bits are 0 - i.e. new | |
1406 | * user-space does not rely on any kernel feature | |
1407 | * extensions we dont know about yet. | |
1408 | */ | |
1e270976 MKL |
1409 | err = check_uarg_tail_zero(uattr, sizeof(attr), size); |
1410 | if (err) | |
1411 | return err; | |
1412 | size = min_t(u32, size, sizeof(attr)); | |
99c55f7d AS |
1413 | |
1414 | /* copy attributes from user space, may be less than sizeof(bpf_attr) */ | |
1415 | if (copy_from_user(&attr, uattr, size) != 0) | |
1416 | return -EFAULT; | |
1417 | ||
1418 | switch (cmd) { | |
1419 | case BPF_MAP_CREATE: | |
1420 | err = map_create(&attr); | |
1421 | break; | |
db20fd2b AS |
1422 | case BPF_MAP_LOOKUP_ELEM: |
1423 | err = map_lookup_elem(&attr); | |
1424 | break; | |
1425 | case BPF_MAP_UPDATE_ELEM: | |
1426 | err = map_update_elem(&attr); | |
1427 | break; | |
1428 | case BPF_MAP_DELETE_ELEM: | |
1429 | err = map_delete_elem(&attr); | |
1430 | break; | |
1431 | case BPF_MAP_GET_NEXT_KEY: | |
1432 | err = map_get_next_key(&attr); | |
1433 | break; | |
09756af4 AS |
1434 | case BPF_PROG_LOAD: |
1435 | err = bpf_prog_load(&attr); | |
1436 | break; | |
b2197755 DB |
1437 | case BPF_OBJ_PIN: |
1438 | err = bpf_obj_pin(&attr); | |
1439 | break; | |
1440 | case BPF_OBJ_GET: | |
1441 | err = bpf_obj_get(&attr); | |
1442 | break; | |
f4324551 DM |
1443 | #ifdef CONFIG_CGROUP_BPF |
1444 | case BPF_PROG_ATTACH: | |
1445 | err = bpf_prog_attach(&attr); | |
1446 | break; | |
1447 | case BPF_PROG_DETACH: | |
1448 | err = bpf_prog_detach(&attr); | |
1449 | break; | |
1450 | #endif | |
1cf1cae9 AS |
1451 | case BPF_PROG_TEST_RUN: |
1452 | err = bpf_prog_test_run(&attr, uattr); | |
1453 | break; | |
34ad5580 MKL |
1454 | case BPF_PROG_GET_NEXT_ID: |
1455 | err = bpf_obj_get_next_id(&attr, uattr, | |
1456 | &prog_idr, &prog_idr_lock); | |
1457 | break; | |
1458 | case BPF_MAP_GET_NEXT_ID: | |
1459 | err = bpf_obj_get_next_id(&attr, uattr, | |
1460 | &map_idr, &map_idr_lock); | |
1461 | break; | |
b16d9aa4 MKL |
1462 | case BPF_PROG_GET_FD_BY_ID: |
1463 | err = bpf_prog_get_fd_by_id(&attr); | |
1464 | break; | |
bd5f5f4e MKL |
1465 | case BPF_MAP_GET_FD_BY_ID: |
1466 | err = bpf_map_get_fd_by_id(&attr); | |
1467 | break; | |
1e270976 MKL |
1468 | case BPF_OBJ_GET_INFO_BY_FD: |
1469 | err = bpf_obj_get_info_by_fd(&attr, uattr); | |
1470 | break; | |
99c55f7d AS |
1471 | default: |
1472 | err = -EINVAL; | |
1473 | break; | |
1474 | } | |
1475 | ||
1476 | return err; | |
1477 | } |