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 | 50 | |
752ba56f MS |
51 | /* |
52 | * If we're handed a bigger struct than we know of, ensure all the unknown bits | |
53 | * are 0 - i.e. new user-space does not rely on any kernel feature extensions | |
54 | * we don't know about yet. | |
55 | * | |
56 | * There is a ToCToU between this function call and the following | |
57 | * copy_from_user() call. However, this is not a concern since this function is | |
58 | * meant to be a future-proofing of bits. | |
59 | */ | |
58291a74 MS |
60 | static int check_uarg_tail_zero(void __user *uaddr, |
61 | size_t expected_size, | |
62 | size_t actual_size) | |
63 | { | |
64 | unsigned char __user *addr; | |
65 | unsigned char __user *end; | |
66 | unsigned char val; | |
67 | int err; | |
68 | ||
752ba56f MS |
69 | if (unlikely(actual_size > PAGE_SIZE)) /* silly large */ |
70 | return -E2BIG; | |
71 | ||
72 | if (unlikely(!access_ok(VERIFY_READ, uaddr, actual_size))) | |
73 | return -EFAULT; | |
74 | ||
58291a74 MS |
75 | if (actual_size <= expected_size) |
76 | return 0; | |
77 | ||
78 | addr = uaddr + expected_size; | |
79 | end = uaddr + actual_size; | |
80 | ||
81 | for (; addr < end; addr++) { | |
82 | err = get_user(val, addr); | |
83 | if (err) | |
84 | return err; | |
85 | if (val) | |
86 | return -E2BIG; | |
87 | } | |
88 | ||
89 | return 0; | |
90 | } | |
91 | ||
99c55f7d AS |
92 | static struct bpf_map *find_and_alloc_map(union bpf_attr *attr) |
93 | { | |
99c55f7d AS |
94 | struct bpf_map *map; |
95 | ||
40077e0c JB |
96 | if (attr->map_type >= ARRAY_SIZE(bpf_map_types) || |
97 | !bpf_map_types[attr->map_type]) | |
98 | return ERR_PTR(-EINVAL); | |
99c55f7d | 99 | |
40077e0c JB |
100 | map = bpf_map_types[attr->map_type]->map_alloc(attr); |
101 | if (IS_ERR(map)) | |
102 | return map; | |
103 | map->ops = bpf_map_types[attr->map_type]; | |
104 | map->map_type = attr->map_type; | |
105 | return map; | |
99c55f7d AS |
106 | } |
107 | ||
96eabe7a | 108 | void *bpf_map_area_alloc(size_t size, int numa_node) |
d407bd25 DB |
109 | { |
110 | /* We definitely need __GFP_NORETRY, so OOM killer doesn't | |
111 | * trigger under memory pressure as we really just want to | |
112 | * fail instead. | |
113 | */ | |
114 | const gfp_t flags = __GFP_NOWARN | __GFP_NORETRY | __GFP_ZERO; | |
115 | void *area; | |
116 | ||
117 | if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) { | |
96eabe7a | 118 | area = kmalloc_node(size, GFP_USER | flags, numa_node); |
d407bd25 DB |
119 | if (area != NULL) |
120 | return area; | |
121 | } | |
122 | ||
96eabe7a MKL |
123 | return __vmalloc_node_flags_caller(size, numa_node, GFP_KERNEL | flags, |
124 | __builtin_return_address(0)); | |
d407bd25 DB |
125 | } |
126 | ||
127 | void bpf_map_area_free(void *area) | |
128 | { | |
129 | kvfree(area); | |
130 | } | |
131 | ||
6c905981 AS |
132 | int bpf_map_precharge_memlock(u32 pages) |
133 | { | |
134 | struct user_struct *user = get_current_user(); | |
135 | unsigned long memlock_limit, cur; | |
136 | ||
137 | memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
138 | cur = atomic_long_read(&user->locked_vm); | |
139 | free_uid(user); | |
140 | if (cur + pages > memlock_limit) | |
141 | return -EPERM; | |
142 | return 0; | |
143 | } | |
144 | ||
aaac3ba9 AS |
145 | static int bpf_map_charge_memlock(struct bpf_map *map) |
146 | { | |
147 | struct user_struct *user = get_current_user(); | |
148 | unsigned long memlock_limit; | |
149 | ||
150 | memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
151 | ||
152 | atomic_long_add(map->pages, &user->locked_vm); | |
153 | ||
154 | if (atomic_long_read(&user->locked_vm) > memlock_limit) { | |
155 | atomic_long_sub(map->pages, &user->locked_vm); | |
156 | free_uid(user); | |
157 | return -EPERM; | |
158 | } | |
159 | map->user = user; | |
160 | return 0; | |
161 | } | |
162 | ||
163 | static void bpf_map_uncharge_memlock(struct bpf_map *map) | |
164 | { | |
165 | struct user_struct *user = map->user; | |
166 | ||
167 | atomic_long_sub(map->pages, &user->locked_vm); | |
168 | free_uid(user); | |
169 | } | |
170 | ||
f3f1c054 MKL |
171 | static int bpf_map_alloc_id(struct bpf_map *map) |
172 | { | |
173 | int id; | |
174 | ||
175 | spin_lock_bh(&map_idr_lock); | |
176 | id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC); | |
177 | if (id > 0) | |
178 | map->id = id; | |
179 | spin_unlock_bh(&map_idr_lock); | |
180 | ||
181 | if (WARN_ON_ONCE(!id)) | |
182 | return -ENOSPC; | |
183 | ||
184 | return id > 0 ? 0 : id; | |
185 | } | |
186 | ||
bd5f5f4e | 187 | static void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock) |
f3f1c054 | 188 | { |
930651a7 ED |
189 | unsigned long flags; |
190 | ||
bd5f5f4e | 191 | if (do_idr_lock) |
930651a7 | 192 | spin_lock_irqsave(&map_idr_lock, flags); |
bd5f5f4e MKL |
193 | else |
194 | __acquire(&map_idr_lock); | |
195 | ||
f3f1c054 | 196 | idr_remove(&map_idr, map->id); |
bd5f5f4e MKL |
197 | |
198 | if (do_idr_lock) | |
930651a7 | 199 | spin_unlock_irqrestore(&map_idr_lock, flags); |
bd5f5f4e MKL |
200 | else |
201 | __release(&map_idr_lock); | |
f3f1c054 MKL |
202 | } |
203 | ||
99c55f7d AS |
204 | /* called from workqueue */ |
205 | static void bpf_map_free_deferred(struct work_struct *work) | |
206 | { | |
207 | struct bpf_map *map = container_of(work, struct bpf_map, work); | |
208 | ||
aaac3ba9 | 209 | bpf_map_uncharge_memlock(map); |
99c55f7d AS |
210 | /* implementation dependent freeing */ |
211 | map->ops->map_free(map); | |
212 | } | |
213 | ||
c9da161c DB |
214 | static void bpf_map_put_uref(struct bpf_map *map) |
215 | { | |
216 | if (atomic_dec_and_test(&map->usercnt)) { | |
217 | if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) | |
218 | bpf_fd_array_map_clear(map); | |
219 | } | |
220 | } | |
221 | ||
99c55f7d AS |
222 | /* decrement map refcnt and schedule it for freeing via workqueue |
223 | * (unrelying map implementation ops->map_free() might sleep) | |
224 | */ | |
bd5f5f4e | 225 | static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock) |
99c55f7d AS |
226 | { |
227 | if (atomic_dec_and_test(&map->refcnt)) { | |
34ad5580 | 228 | /* bpf_map_free_id() must be called first */ |
bd5f5f4e | 229 | bpf_map_free_id(map, do_idr_lock); |
99c55f7d AS |
230 | INIT_WORK(&map->work, bpf_map_free_deferred); |
231 | schedule_work(&map->work); | |
232 | } | |
233 | } | |
234 | ||
bd5f5f4e MKL |
235 | void bpf_map_put(struct bpf_map *map) |
236 | { | |
237 | __bpf_map_put(map, true); | |
238 | } | |
239 | ||
c9da161c | 240 | void bpf_map_put_with_uref(struct bpf_map *map) |
99c55f7d | 241 | { |
c9da161c | 242 | bpf_map_put_uref(map); |
99c55f7d | 243 | bpf_map_put(map); |
c9da161c DB |
244 | } |
245 | ||
246 | static int bpf_map_release(struct inode *inode, struct file *filp) | |
247 | { | |
61d1b6a4 DB |
248 | struct bpf_map *map = filp->private_data; |
249 | ||
250 | if (map->ops->map_release) | |
251 | map->ops->map_release(map, filp); | |
252 | ||
253 | bpf_map_put_with_uref(map); | |
99c55f7d AS |
254 | return 0; |
255 | } | |
256 | ||
f99bf205 DB |
257 | #ifdef CONFIG_PROC_FS |
258 | static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp) | |
259 | { | |
260 | const struct bpf_map *map = filp->private_data; | |
21116b70 DB |
261 | const struct bpf_array *array; |
262 | u32 owner_prog_type = 0; | |
9780c0ab | 263 | u32 owner_jited = 0; |
21116b70 DB |
264 | |
265 | if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) { | |
266 | array = container_of(map, struct bpf_array, map); | |
267 | owner_prog_type = array->owner_prog_type; | |
9780c0ab | 268 | owner_jited = array->owner_jited; |
21116b70 | 269 | } |
f99bf205 DB |
270 | |
271 | seq_printf(m, | |
272 | "map_type:\t%u\n" | |
273 | "key_size:\t%u\n" | |
274 | "value_size:\t%u\n" | |
322cea2f | 275 | "max_entries:\t%u\n" |
21116b70 DB |
276 | "map_flags:\t%#x\n" |
277 | "memlock:\t%llu\n", | |
f99bf205 DB |
278 | map->map_type, |
279 | map->key_size, | |
280 | map->value_size, | |
322cea2f | 281 | map->max_entries, |
21116b70 DB |
282 | map->map_flags, |
283 | map->pages * 1ULL << PAGE_SHIFT); | |
284 | ||
9780c0ab | 285 | if (owner_prog_type) { |
21116b70 DB |
286 | seq_printf(m, "owner_prog_type:\t%u\n", |
287 | owner_prog_type); | |
9780c0ab DB |
288 | seq_printf(m, "owner_jited:\t%u\n", |
289 | owner_jited); | |
290 | } | |
f99bf205 DB |
291 | } |
292 | #endif | |
293 | ||
99c55f7d | 294 | static const struct file_operations bpf_map_fops = { |
f99bf205 DB |
295 | #ifdef CONFIG_PROC_FS |
296 | .show_fdinfo = bpf_map_show_fdinfo, | |
297 | #endif | |
298 | .release = bpf_map_release, | |
99c55f7d AS |
299 | }; |
300 | ||
b2197755 | 301 | int bpf_map_new_fd(struct bpf_map *map) |
aa79781b DB |
302 | { |
303 | return anon_inode_getfd("bpf-map", &bpf_map_fops, map, | |
304 | O_RDWR | O_CLOEXEC); | |
305 | } | |
306 | ||
99c55f7d AS |
307 | /* helper macro to check that unused fields 'union bpf_attr' are zero */ |
308 | #define CHECK_ATTR(CMD) \ | |
309 | memchr_inv((void *) &attr->CMD##_LAST_FIELD + \ | |
310 | sizeof(attr->CMD##_LAST_FIELD), 0, \ | |
311 | sizeof(*attr) - \ | |
312 | offsetof(union bpf_attr, CMD##_LAST_FIELD) - \ | |
313 | sizeof(attr->CMD##_LAST_FIELD)) != NULL | |
314 | ||
96eabe7a | 315 | #define BPF_MAP_CREATE_LAST_FIELD numa_node |
99c55f7d AS |
316 | /* called via syscall */ |
317 | static int map_create(union bpf_attr *attr) | |
318 | { | |
96eabe7a | 319 | int numa_node = bpf_map_attr_numa_node(attr); |
99c55f7d AS |
320 | struct bpf_map *map; |
321 | int err; | |
322 | ||
323 | err = CHECK_ATTR(BPF_MAP_CREATE); | |
324 | if (err) | |
325 | return -EINVAL; | |
326 | ||
96eabe7a | 327 | if (numa_node != NUMA_NO_NODE && |
96e5ae4e ED |
328 | ((unsigned int)numa_node >= nr_node_ids || |
329 | !node_online(numa_node))) | |
96eabe7a MKL |
330 | return -EINVAL; |
331 | ||
99c55f7d AS |
332 | /* find map type and init map: hashtable vs rbtree vs bloom vs ... */ |
333 | map = find_and_alloc_map(attr); | |
334 | if (IS_ERR(map)) | |
335 | return PTR_ERR(map); | |
336 | ||
337 | atomic_set(&map->refcnt, 1); | |
c9da161c | 338 | atomic_set(&map->usercnt, 1); |
99c55f7d | 339 | |
aaac3ba9 AS |
340 | err = bpf_map_charge_memlock(map); |
341 | if (err) | |
20b2b24f | 342 | goto free_map_nouncharge; |
aaac3ba9 | 343 | |
f3f1c054 MKL |
344 | err = bpf_map_alloc_id(map); |
345 | if (err) | |
346 | goto free_map; | |
347 | ||
aa79781b | 348 | err = bpf_map_new_fd(map); |
bd5f5f4e MKL |
349 | if (err < 0) { |
350 | /* failed to allocate fd. | |
351 | * bpf_map_put() is needed because the above | |
352 | * bpf_map_alloc_id() has published the map | |
353 | * to the userspace and the userspace may | |
354 | * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID. | |
355 | */ | |
356 | bpf_map_put(map); | |
357 | return err; | |
358 | } | |
99c55f7d | 359 | |
a67edbf4 | 360 | trace_bpf_map_create(map, err); |
99c55f7d AS |
361 | return err; |
362 | ||
363 | free_map: | |
20b2b24f DB |
364 | bpf_map_uncharge_memlock(map); |
365 | free_map_nouncharge: | |
99c55f7d AS |
366 | map->ops->map_free(map); |
367 | return err; | |
368 | } | |
369 | ||
db20fd2b AS |
370 | /* if error is returned, fd is released. |
371 | * On success caller should complete fd access with matching fdput() | |
372 | */ | |
c2101297 | 373 | struct bpf_map *__bpf_map_get(struct fd f) |
db20fd2b | 374 | { |
db20fd2b AS |
375 | if (!f.file) |
376 | return ERR_PTR(-EBADF); | |
db20fd2b AS |
377 | if (f.file->f_op != &bpf_map_fops) { |
378 | fdput(f); | |
379 | return ERR_PTR(-EINVAL); | |
380 | } | |
381 | ||
c2101297 DB |
382 | return f.file->private_data; |
383 | } | |
384 | ||
92117d84 AS |
385 | /* prog's and map's refcnt limit */ |
386 | #define BPF_MAX_REFCNT 32768 | |
387 | ||
388 | struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref) | |
c9da161c | 389 | { |
92117d84 AS |
390 | if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) { |
391 | atomic_dec(&map->refcnt); | |
392 | return ERR_PTR(-EBUSY); | |
393 | } | |
c9da161c DB |
394 | if (uref) |
395 | atomic_inc(&map->usercnt); | |
92117d84 | 396 | return map; |
c9da161c DB |
397 | } |
398 | ||
399 | struct bpf_map *bpf_map_get_with_uref(u32 ufd) | |
c2101297 DB |
400 | { |
401 | struct fd f = fdget(ufd); | |
402 | struct bpf_map *map; | |
403 | ||
404 | map = __bpf_map_get(f); | |
405 | if (IS_ERR(map)) | |
406 | return map; | |
407 | ||
92117d84 | 408 | map = bpf_map_inc(map, true); |
c2101297 | 409 | fdput(f); |
db20fd2b AS |
410 | |
411 | return map; | |
412 | } | |
413 | ||
bd5f5f4e MKL |
414 | /* map_idr_lock should have been held */ |
415 | static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map, | |
416 | bool uref) | |
417 | { | |
418 | int refold; | |
419 | ||
420 | refold = __atomic_add_unless(&map->refcnt, 1, 0); | |
421 | ||
422 | if (refold >= BPF_MAX_REFCNT) { | |
423 | __bpf_map_put(map, false); | |
424 | return ERR_PTR(-EBUSY); | |
425 | } | |
426 | ||
427 | if (!refold) | |
428 | return ERR_PTR(-ENOENT); | |
429 | ||
430 | if (uref) | |
431 | atomic_inc(&map->usercnt); | |
432 | ||
433 | return map; | |
434 | } | |
435 | ||
b8cdc051 AS |
436 | int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value) |
437 | { | |
438 | return -ENOTSUPP; | |
439 | } | |
440 | ||
db20fd2b AS |
441 | /* last field in 'union bpf_attr' used by this command */ |
442 | #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value | |
443 | ||
444 | static int map_lookup_elem(union bpf_attr *attr) | |
445 | { | |
535e7b4b MS |
446 | void __user *ukey = u64_to_user_ptr(attr->key); |
447 | void __user *uvalue = u64_to_user_ptr(attr->value); | |
db20fd2b | 448 | int ufd = attr->map_fd; |
db20fd2b | 449 | struct bpf_map *map; |
8ebe667c | 450 | void *key, *value, *ptr; |
15a07b33 | 451 | u32 value_size; |
592867bf | 452 | struct fd f; |
db20fd2b AS |
453 | int err; |
454 | ||
455 | if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM)) | |
456 | return -EINVAL; | |
457 | ||
592867bf | 458 | f = fdget(ufd); |
c2101297 | 459 | map = __bpf_map_get(f); |
db20fd2b AS |
460 | if (IS_ERR(map)) |
461 | return PTR_ERR(map); | |
462 | ||
e4448ed8 AV |
463 | key = memdup_user(ukey, map->key_size); |
464 | if (IS_ERR(key)) { | |
465 | err = PTR_ERR(key); | |
db20fd2b | 466 | goto err_put; |
e4448ed8 | 467 | } |
db20fd2b | 468 | |
15a07b33 | 469 | if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || |
8f844938 | 470 | map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH || |
15a07b33 AS |
471 | map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) |
472 | value_size = round_up(map->value_size, 8) * num_possible_cpus(); | |
14dc6f04 MKL |
473 | else if (IS_FD_MAP(map)) |
474 | value_size = sizeof(u32); | |
15a07b33 AS |
475 | else |
476 | value_size = map->value_size; | |
477 | ||
8ebe667c | 478 | err = -ENOMEM; |
15a07b33 | 479 | value = kmalloc(value_size, GFP_USER | __GFP_NOWARN); |
db20fd2b | 480 | if (!value) |
8ebe667c AS |
481 | goto free_key; |
482 | ||
8f844938 MKL |
483 | if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || |
484 | map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) { | |
15a07b33 AS |
485 | err = bpf_percpu_hash_copy(map, key, value); |
486 | } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) { | |
487 | err = bpf_percpu_array_copy(map, key, value); | |
557c0c6e AS |
488 | } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) { |
489 | err = bpf_stackmap_copy(map, key, value); | |
14dc6f04 MKL |
490 | } else if (IS_FD_ARRAY(map)) { |
491 | err = bpf_fd_array_map_lookup_elem(map, key, value); | |
492 | } else if (IS_FD_HASH(map)) { | |
493 | err = bpf_fd_htab_map_lookup_elem(map, key, value); | |
15a07b33 AS |
494 | } else { |
495 | rcu_read_lock(); | |
496 | ptr = map->ops->map_lookup_elem(map, key); | |
497 | if (ptr) | |
498 | memcpy(value, ptr, value_size); | |
499 | rcu_read_unlock(); | |
500 | err = ptr ? 0 : -ENOENT; | |
501 | } | |
8ebe667c | 502 | |
15a07b33 | 503 | if (err) |
8ebe667c | 504 | goto free_value; |
db20fd2b AS |
505 | |
506 | err = -EFAULT; | |
15a07b33 | 507 | if (copy_to_user(uvalue, value, value_size) != 0) |
8ebe667c | 508 | goto free_value; |
db20fd2b | 509 | |
a67edbf4 | 510 | trace_bpf_map_lookup_elem(map, ufd, key, value); |
db20fd2b AS |
511 | err = 0; |
512 | ||
8ebe667c AS |
513 | free_value: |
514 | kfree(value); | |
db20fd2b AS |
515 | free_key: |
516 | kfree(key); | |
517 | err_put: | |
518 | fdput(f); | |
519 | return err; | |
520 | } | |
521 | ||
3274f520 | 522 | #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags |
db20fd2b AS |
523 | |
524 | static int map_update_elem(union bpf_attr *attr) | |
525 | { | |
535e7b4b MS |
526 | void __user *ukey = u64_to_user_ptr(attr->key); |
527 | void __user *uvalue = u64_to_user_ptr(attr->value); | |
db20fd2b | 528 | int ufd = attr->map_fd; |
db20fd2b AS |
529 | struct bpf_map *map; |
530 | void *key, *value; | |
15a07b33 | 531 | u32 value_size; |
592867bf | 532 | struct fd f; |
db20fd2b AS |
533 | int err; |
534 | ||
535 | if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM)) | |
536 | return -EINVAL; | |
537 | ||
592867bf | 538 | f = fdget(ufd); |
c2101297 | 539 | map = __bpf_map_get(f); |
db20fd2b AS |
540 | if (IS_ERR(map)) |
541 | return PTR_ERR(map); | |
542 | ||
e4448ed8 AV |
543 | key = memdup_user(ukey, map->key_size); |
544 | if (IS_ERR(key)) { | |
545 | err = PTR_ERR(key); | |
db20fd2b | 546 | goto err_put; |
e4448ed8 | 547 | } |
db20fd2b | 548 | |
15a07b33 | 549 | if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || |
8f844938 | 550 | map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH || |
15a07b33 AS |
551 | map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) |
552 | value_size = round_up(map->value_size, 8) * num_possible_cpus(); | |
553 | else | |
554 | value_size = map->value_size; | |
555 | ||
db20fd2b | 556 | err = -ENOMEM; |
15a07b33 | 557 | value = kmalloc(value_size, GFP_USER | __GFP_NOWARN); |
db20fd2b AS |
558 | if (!value) |
559 | goto free_key; | |
560 | ||
561 | err = -EFAULT; | |
15a07b33 | 562 | if (copy_from_user(value, uvalue, value_size) != 0) |
db20fd2b AS |
563 | goto free_value; |
564 | ||
b121d1e7 AS |
565 | /* must increment bpf_prog_active to avoid kprobe+bpf triggering from |
566 | * inside bpf map update or delete otherwise deadlocks are possible | |
567 | */ | |
568 | preempt_disable(); | |
569 | __this_cpu_inc(bpf_prog_active); | |
8f844938 MKL |
570 | if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || |
571 | map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) { | |
15a07b33 AS |
572 | err = bpf_percpu_hash_update(map, key, value, attr->flags); |
573 | } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) { | |
574 | err = bpf_percpu_array_update(map, key, value, attr->flags); | |
d056a788 | 575 | } else if (map->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || |
4ed8ec52 | 576 | map->map_type == BPF_MAP_TYPE_PROG_ARRAY || |
56f668df MKL |
577 | map->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || |
578 | map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS) { | |
d056a788 DB |
579 | rcu_read_lock(); |
580 | err = bpf_fd_array_map_update_elem(map, f.file, key, value, | |
581 | attr->flags); | |
582 | rcu_read_unlock(); | |
bcc6b1b7 MKL |
583 | } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) { |
584 | rcu_read_lock(); | |
585 | err = bpf_fd_htab_map_update_elem(map, f.file, key, value, | |
586 | attr->flags); | |
587 | rcu_read_unlock(); | |
15a07b33 AS |
588 | } else { |
589 | rcu_read_lock(); | |
590 | err = map->ops->map_update_elem(map, key, value, attr->flags); | |
591 | rcu_read_unlock(); | |
592 | } | |
b121d1e7 AS |
593 | __this_cpu_dec(bpf_prog_active); |
594 | preempt_enable(); | |
db20fd2b | 595 | |
a67edbf4 DB |
596 | if (!err) |
597 | trace_bpf_map_update_elem(map, ufd, key, value); | |
db20fd2b AS |
598 | free_value: |
599 | kfree(value); | |
600 | free_key: | |
601 | kfree(key); | |
602 | err_put: | |
603 | fdput(f); | |
604 | return err; | |
605 | } | |
606 | ||
607 | #define BPF_MAP_DELETE_ELEM_LAST_FIELD key | |
608 | ||
609 | static int map_delete_elem(union bpf_attr *attr) | |
610 | { | |
535e7b4b | 611 | void __user *ukey = u64_to_user_ptr(attr->key); |
db20fd2b | 612 | int ufd = attr->map_fd; |
db20fd2b | 613 | struct bpf_map *map; |
592867bf | 614 | struct fd f; |
db20fd2b AS |
615 | void *key; |
616 | int err; | |
617 | ||
618 | if (CHECK_ATTR(BPF_MAP_DELETE_ELEM)) | |
619 | return -EINVAL; | |
620 | ||
592867bf | 621 | f = fdget(ufd); |
c2101297 | 622 | map = __bpf_map_get(f); |
db20fd2b AS |
623 | if (IS_ERR(map)) |
624 | return PTR_ERR(map); | |
625 | ||
e4448ed8 AV |
626 | key = memdup_user(ukey, map->key_size); |
627 | if (IS_ERR(key)) { | |
628 | err = PTR_ERR(key); | |
db20fd2b | 629 | goto err_put; |
e4448ed8 | 630 | } |
db20fd2b | 631 | |
b121d1e7 AS |
632 | preempt_disable(); |
633 | __this_cpu_inc(bpf_prog_active); | |
db20fd2b AS |
634 | rcu_read_lock(); |
635 | err = map->ops->map_delete_elem(map, key); | |
636 | rcu_read_unlock(); | |
b121d1e7 AS |
637 | __this_cpu_dec(bpf_prog_active); |
638 | preempt_enable(); | |
db20fd2b | 639 | |
a67edbf4 DB |
640 | if (!err) |
641 | trace_bpf_map_delete_elem(map, ufd, key); | |
db20fd2b AS |
642 | kfree(key); |
643 | err_put: | |
644 | fdput(f); | |
645 | return err; | |
646 | } | |
647 | ||
648 | /* last field in 'union bpf_attr' used by this command */ | |
649 | #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key | |
650 | ||
651 | static int map_get_next_key(union bpf_attr *attr) | |
652 | { | |
535e7b4b MS |
653 | void __user *ukey = u64_to_user_ptr(attr->key); |
654 | void __user *unext_key = u64_to_user_ptr(attr->next_key); | |
db20fd2b | 655 | int ufd = attr->map_fd; |
db20fd2b AS |
656 | struct bpf_map *map; |
657 | void *key, *next_key; | |
592867bf | 658 | struct fd f; |
db20fd2b AS |
659 | int err; |
660 | ||
661 | if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY)) | |
662 | return -EINVAL; | |
663 | ||
592867bf | 664 | f = fdget(ufd); |
c2101297 | 665 | map = __bpf_map_get(f); |
db20fd2b AS |
666 | if (IS_ERR(map)) |
667 | return PTR_ERR(map); | |
668 | ||
8fe45924 | 669 | if (ukey) { |
e4448ed8 AV |
670 | key = memdup_user(ukey, map->key_size); |
671 | if (IS_ERR(key)) { | |
672 | err = PTR_ERR(key); | |
8fe45924 | 673 | goto err_put; |
e4448ed8 | 674 | } |
8fe45924 TQ |
675 | } else { |
676 | key = NULL; | |
677 | } | |
db20fd2b AS |
678 | |
679 | err = -ENOMEM; | |
680 | next_key = kmalloc(map->key_size, GFP_USER); | |
681 | if (!next_key) | |
682 | goto free_key; | |
683 | ||
684 | rcu_read_lock(); | |
685 | err = map->ops->map_get_next_key(map, key, next_key); | |
686 | rcu_read_unlock(); | |
687 | if (err) | |
688 | goto free_next_key; | |
689 | ||
690 | err = -EFAULT; | |
691 | if (copy_to_user(unext_key, next_key, map->key_size) != 0) | |
692 | goto free_next_key; | |
693 | ||
a67edbf4 | 694 | trace_bpf_map_next_key(map, ufd, key, next_key); |
db20fd2b AS |
695 | err = 0; |
696 | ||
697 | free_next_key: | |
698 | kfree(next_key); | |
699 | free_key: | |
700 | kfree(key); | |
701 | err_put: | |
702 | fdput(f); | |
703 | return err; | |
704 | } | |
705 | ||
be9370a7 JB |
706 | static const struct bpf_verifier_ops * const bpf_prog_types[] = { |
707 | #define BPF_PROG_TYPE(_id, _ops) \ | |
708 | [_id] = &_ops, | |
40077e0c | 709 | #define BPF_MAP_TYPE(_id, _ops) |
be9370a7 JB |
710 | #include <linux/bpf_types.h> |
711 | #undef BPF_PROG_TYPE | |
40077e0c | 712 | #undef BPF_MAP_TYPE |
be9370a7 | 713 | }; |
09756af4 AS |
714 | |
715 | static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog) | |
716 | { | |
be9370a7 JB |
717 | if (type >= ARRAY_SIZE(bpf_prog_types) || !bpf_prog_types[type]) |
718 | return -EINVAL; | |
09756af4 | 719 | |
be9370a7 JB |
720 | prog->aux->ops = bpf_prog_types[type]; |
721 | prog->type = type; | |
722 | return 0; | |
09756af4 AS |
723 | } |
724 | ||
725 | /* drop refcnt on maps used by eBPF program and free auxilary data */ | |
726 | static void free_used_maps(struct bpf_prog_aux *aux) | |
727 | { | |
728 | int i; | |
729 | ||
730 | for (i = 0; i < aux->used_map_cnt; i++) | |
731 | bpf_map_put(aux->used_maps[i]); | |
732 | ||
733 | kfree(aux->used_maps); | |
734 | } | |
735 | ||
5ccb071e DB |
736 | int __bpf_prog_charge(struct user_struct *user, u32 pages) |
737 | { | |
738 | unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
739 | unsigned long user_bufs; | |
740 | ||
741 | if (user) { | |
742 | user_bufs = atomic_long_add_return(pages, &user->locked_vm); | |
743 | if (user_bufs > memlock_limit) { | |
744 | atomic_long_sub(pages, &user->locked_vm); | |
745 | return -EPERM; | |
746 | } | |
747 | } | |
748 | ||
749 | return 0; | |
750 | } | |
751 | ||
752 | void __bpf_prog_uncharge(struct user_struct *user, u32 pages) | |
753 | { | |
754 | if (user) | |
755 | atomic_long_sub(pages, &user->locked_vm); | |
756 | } | |
757 | ||
aaac3ba9 AS |
758 | static int bpf_prog_charge_memlock(struct bpf_prog *prog) |
759 | { | |
760 | struct user_struct *user = get_current_user(); | |
5ccb071e | 761 | int ret; |
aaac3ba9 | 762 | |
5ccb071e DB |
763 | ret = __bpf_prog_charge(user, prog->pages); |
764 | if (ret) { | |
aaac3ba9 | 765 | free_uid(user); |
5ccb071e | 766 | return ret; |
aaac3ba9 | 767 | } |
5ccb071e | 768 | |
aaac3ba9 AS |
769 | prog->aux->user = user; |
770 | return 0; | |
771 | } | |
772 | ||
773 | static void bpf_prog_uncharge_memlock(struct bpf_prog *prog) | |
774 | { | |
775 | struct user_struct *user = prog->aux->user; | |
776 | ||
5ccb071e | 777 | __bpf_prog_uncharge(user, prog->pages); |
aaac3ba9 AS |
778 | free_uid(user); |
779 | } | |
780 | ||
dc4bb0e2 MKL |
781 | static int bpf_prog_alloc_id(struct bpf_prog *prog) |
782 | { | |
783 | int id; | |
784 | ||
785 | spin_lock_bh(&prog_idr_lock); | |
786 | id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC); | |
787 | if (id > 0) | |
788 | prog->aux->id = id; | |
789 | spin_unlock_bh(&prog_idr_lock); | |
790 | ||
791 | /* id is in [1, INT_MAX) */ | |
792 | if (WARN_ON_ONCE(!id)) | |
793 | return -ENOSPC; | |
794 | ||
795 | return id > 0 ? 0 : id; | |
796 | } | |
797 | ||
b16d9aa4 | 798 | static void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock) |
dc4bb0e2 MKL |
799 | { |
800 | /* cBPF to eBPF migrations are currently not in the idr store. */ | |
801 | if (!prog->aux->id) | |
802 | return; | |
803 | ||
b16d9aa4 MKL |
804 | if (do_idr_lock) |
805 | spin_lock_bh(&prog_idr_lock); | |
806 | else | |
807 | __acquire(&prog_idr_lock); | |
808 | ||
dc4bb0e2 | 809 | idr_remove(&prog_idr, prog->aux->id); |
b16d9aa4 MKL |
810 | |
811 | if (do_idr_lock) | |
812 | spin_unlock_bh(&prog_idr_lock); | |
813 | else | |
814 | __release(&prog_idr_lock); | |
dc4bb0e2 MKL |
815 | } |
816 | ||
1aacde3d | 817 | static void __bpf_prog_put_rcu(struct rcu_head *rcu) |
abf2e7d6 AS |
818 | { |
819 | struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu); | |
820 | ||
821 | free_used_maps(aux); | |
aaac3ba9 | 822 | bpf_prog_uncharge_memlock(aux->prog); |
abf2e7d6 AS |
823 | bpf_prog_free(aux->prog); |
824 | } | |
825 | ||
b16d9aa4 | 826 | static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock) |
09756af4 | 827 | { |
a67edbf4 DB |
828 | if (atomic_dec_and_test(&prog->aux->refcnt)) { |
829 | trace_bpf_prog_put_rcu(prog); | |
34ad5580 | 830 | /* bpf_prog_free_id() must be called first */ |
b16d9aa4 | 831 | bpf_prog_free_id(prog, do_idr_lock); |
74451e66 | 832 | bpf_prog_kallsyms_del(prog); |
1aacde3d | 833 | call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu); |
a67edbf4 | 834 | } |
09756af4 | 835 | } |
b16d9aa4 MKL |
836 | |
837 | void bpf_prog_put(struct bpf_prog *prog) | |
838 | { | |
839 | __bpf_prog_put(prog, true); | |
840 | } | |
e2e9b654 | 841 | EXPORT_SYMBOL_GPL(bpf_prog_put); |
09756af4 AS |
842 | |
843 | static int bpf_prog_release(struct inode *inode, struct file *filp) | |
844 | { | |
845 | struct bpf_prog *prog = filp->private_data; | |
846 | ||
1aacde3d | 847 | bpf_prog_put(prog); |
09756af4 AS |
848 | return 0; |
849 | } | |
850 | ||
7bd509e3 DB |
851 | #ifdef CONFIG_PROC_FS |
852 | static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp) | |
853 | { | |
854 | const struct bpf_prog *prog = filp->private_data; | |
f1f7714e | 855 | char prog_tag[sizeof(prog->tag) * 2 + 1] = { }; |
7bd509e3 | 856 | |
f1f7714e | 857 | bin2hex(prog_tag, prog->tag, sizeof(prog->tag)); |
7bd509e3 DB |
858 | seq_printf(m, |
859 | "prog_type:\t%u\n" | |
860 | "prog_jited:\t%u\n" | |
f1f7714e | 861 | "prog_tag:\t%s\n" |
7bd509e3 DB |
862 | "memlock:\t%llu\n", |
863 | prog->type, | |
864 | prog->jited, | |
f1f7714e | 865 | prog_tag, |
7bd509e3 DB |
866 | prog->pages * 1ULL << PAGE_SHIFT); |
867 | } | |
868 | #endif | |
869 | ||
09756af4 | 870 | static const struct file_operations bpf_prog_fops = { |
7bd509e3 DB |
871 | #ifdef CONFIG_PROC_FS |
872 | .show_fdinfo = bpf_prog_show_fdinfo, | |
873 | #endif | |
874 | .release = bpf_prog_release, | |
09756af4 AS |
875 | }; |
876 | ||
b2197755 | 877 | int bpf_prog_new_fd(struct bpf_prog *prog) |
aa79781b DB |
878 | { |
879 | return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog, | |
880 | O_RDWR | O_CLOEXEC); | |
881 | } | |
882 | ||
113214be | 883 | static struct bpf_prog *____bpf_prog_get(struct fd f) |
09756af4 | 884 | { |
09756af4 AS |
885 | if (!f.file) |
886 | return ERR_PTR(-EBADF); | |
09756af4 AS |
887 | if (f.file->f_op != &bpf_prog_fops) { |
888 | fdput(f); | |
889 | return ERR_PTR(-EINVAL); | |
890 | } | |
891 | ||
c2101297 | 892 | return f.file->private_data; |
09756af4 AS |
893 | } |
894 | ||
59d3656d | 895 | struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i) |
92117d84 | 896 | { |
59d3656d BB |
897 | if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) { |
898 | atomic_sub(i, &prog->aux->refcnt); | |
92117d84 AS |
899 | return ERR_PTR(-EBUSY); |
900 | } | |
901 | return prog; | |
902 | } | |
59d3656d BB |
903 | EXPORT_SYMBOL_GPL(bpf_prog_add); |
904 | ||
c540594f DB |
905 | void bpf_prog_sub(struct bpf_prog *prog, int i) |
906 | { | |
907 | /* Only to be used for undoing previous bpf_prog_add() in some | |
908 | * error path. We still know that another entity in our call | |
909 | * path holds a reference to the program, thus atomic_sub() can | |
910 | * be safely used in such cases! | |
911 | */ | |
912 | WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0); | |
913 | } | |
914 | EXPORT_SYMBOL_GPL(bpf_prog_sub); | |
915 | ||
59d3656d BB |
916 | struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog) |
917 | { | |
918 | return bpf_prog_add(prog, 1); | |
919 | } | |
97bc402d | 920 | EXPORT_SYMBOL_GPL(bpf_prog_inc); |
92117d84 | 921 | |
b16d9aa4 | 922 | /* prog_idr_lock should have been held */ |
a6f6df69 | 923 | struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog) |
b16d9aa4 MKL |
924 | { |
925 | int refold; | |
926 | ||
927 | refold = __atomic_add_unless(&prog->aux->refcnt, 1, 0); | |
928 | ||
929 | if (refold >= BPF_MAX_REFCNT) { | |
930 | __bpf_prog_put(prog, false); | |
931 | return ERR_PTR(-EBUSY); | |
932 | } | |
933 | ||
934 | if (!refold) | |
935 | return ERR_PTR(-ENOENT); | |
936 | ||
937 | return prog; | |
938 | } | |
a6f6df69 | 939 | EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero); |
b16d9aa4 | 940 | |
113214be | 941 | static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *type) |
09756af4 AS |
942 | { |
943 | struct fd f = fdget(ufd); | |
944 | struct bpf_prog *prog; | |
945 | ||
113214be | 946 | prog = ____bpf_prog_get(f); |
09756af4 AS |
947 | if (IS_ERR(prog)) |
948 | return prog; | |
113214be DB |
949 | if (type && prog->type != *type) { |
950 | prog = ERR_PTR(-EINVAL); | |
951 | goto out; | |
952 | } | |
09756af4 | 953 | |
92117d84 | 954 | prog = bpf_prog_inc(prog); |
113214be | 955 | out: |
09756af4 AS |
956 | fdput(f); |
957 | return prog; | |
958 | } | |
113214be DB |
959 | |
960 | struct bpf_prog *bpf_prog_get(u32 ufd) | |
961 | { | |
962 | return __bpf_prog_get(ufd, NULL); | |
963 | } | |
964 | ||
965 | struct bpf_prog *bpf_prog_get_type(u32 ufd, enum bpf_prog_type type) | |
966 | { | |
a67edbf4 DB |
967 | struct bpf_prog *prog = __bpf_prog_get(ufd, &type); |
968 | ||
969 | if (!IS_ERR(prog)) | |
970 | trace_bpf_prog_get_type(prog); | |
971 | return prog; | |
113214be DB |
972 | } |
973 | EXPORT_SYMBOL_GPL(bpf_prog_get_type); | |
09756af4 AS |
974 | |
975 | /* last field in 'union bpf_attr' used by this command */ | |
e07b98d9 | 976 | #define BPF_PROG_LOAD_LAST_FIELD prog_flags |
09756af4 AS |
977 | |
978 | static int bpf_prog_load(union bpf_attr *attr) | |
979 | { | |
980 | enum bpf_prog_type type = attr->prog_type; | |
981 | struct bpf_prog *prog; | |
982 | int err; | |
983 | char license[128]; | |
984 | bool is_gpl; | |
985 | ||
986 | if (CHECK_ATTR(BPF_PROG_LOAD)) | |
987 | return -EINVAL; | |
988 | ||
e07b98d9 DM |
989 | if (attr->prog_flags & ~BPF_F_STRICT_ALIGNMENT) |
990 | return -EINVAL; | |
991 | ||
09756af4 | 992 | /* copy eBPF program license from user space */ |
535e7b4b | 993 | if (strncpy_from_user(license, u64_to_user_ptr(attr->license), |
09756af4 AS |
994 | sizeof(license) - 1) < 0) |
995 | return -EFAULT; | |
996 | license[sizeof(license) - 1] = 0; | |
997 | ||
998 | /* eBPF programs must be GPL compatible to use GPL-ed functions */ | |
999 | is_gpl = license_is_gpl_compatible(license); | |
1000 | ||
ef0915ca DB |
1001 | if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS) |
1002 | return -E2BIG; | |
09756af4 | 1003 | |
2541517c AS |
1004 | if (type == BPF_PROG_TYPE_KPROBE && |
1005 | attr->kern_version != LINUX_VERSION_CODE) | |
1006 | return -EINVAL; | |
1007 | ||
80b7d819 CF |
1008 | if (type != BPF_PROG_TYPE_SOCKET_FILTER && |
1009 | type != BPF_PROG_TYPE_CGROUP_SKB && | |
1010 | !capable(CAP_SYS_ADMIN)) | |
1be7f75d AS |
1011 | return -EPERM; |
1012 | ||
09756af4 AS |
1013 | /* plain bpf_prog allocation */ |
1014 | prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER); | |
1015 | if (!prog) | |
1016 | return -ENOMEM; | |
1017 | ||
aaac3ba9 AS |
1018 | err = bpf_prog_charge_memlock(prog); |
1019 | if (err) | |
1020 | goto free_prog_nouncharge; | |
1021 | ||
09756af4 AS |
1022 | prog->len = attr->insn_cnt; |
1023 | ||
1024 | err = -EFAULT; | |
535e7b4b | 1025 | if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns), |
aafe6ae9 | 1026 | bpf_prog_insn_size(prog)) != 0) |
09756af4 AS |
1027 | goto free_prog; |
1028 | ||
1029 | prog->orig_prog = NULL; | |
a91263d5 | 1030 | prog->jited = 0; |
09756af4 AS |
1031 | |
1032 | atomic_set(&prog->aux->refcnt, 1); | |
a91263d5 | 1033 | prog->gpl_compatible = is_gpl ? 1 : 0; |
09756af4 AS |
1034 | |
1035 | /* find program type: socket_filter vs tracing_filter */ | |
1036 | err = find_prog_type(type, prog); | |
1037 | if (err < 0) | |
1038 | goto free_prog; | |
1039 | ||
1040 | /* run eBPF verifier */ | |
9bac3d6d | 1041 | err = bpf_check(&prog, attr); |
09756af4 AS |
1042 | if (err < 0) |
1043 | goto free_used_maps; | |
1044 | ||
1045 | /* eBPF program is ready to be JITed */ | |
d1c55ab5 | 1046 | prog = bpf_prog_select_runtime(prog, &err); |
04fd61ab AS |
1047 | if (err < 0) |
1048 | goto free_used_maps; | |
09756af4 | 1049 | |
dc4bb0e2 MKL |
1050 | err = bpf_prog_alloc_id(prog); |
1051 | if (err) | |
1052 | goto free_used_maps; | |
1053 | ||
aa79781b | 1054 | err = bpf_prog_new_fd(prog); |
b16d9aa4 MKL |
1055 | if (err < 0) { |
1056 | /* failed to allocate fd. | |
1057 | * bpf_prog_put() is needed because the above | |
1058 | * bpf_prog_alloc_id() has published the prog | |
1059 | * to the userspace and the userspace may | |
1060 | * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID. | |
1061 | */ | |
1062 | bpf_prog_put(prog); | |
1063 | return err; | |
1064 | } | |
09756af4 | 1065 | |
74451e66 | 1066 | bpf_prog_kallsyms_add(prog); |
a67edbf4 | 1067 | trace_bpf_prog_load(prog, err); |
09756af4 AS |
1068 | return err; |
1069 | ||
1070 | free_used_maps: | |
1071 | free_used_maps(prog->aux); | |
1072 | free_prog: | |
aaac3ba9 AS |
1073 | bpf_prog_uncharge_memlock(prog); |
1074 | free_prog_nouncharge: | |
09756af4 AS |
1075 | bpf_prog_free(prog); |
1076 | return err; | |
1077 | } | |
1078 | ||
b2197755 DB |
1079 | #define BPF_OBJ_LAST_FIELD bpf_fd |
1080 | ||
1081 | static int bpf_obj_pin(const union bpf_attr *attr) | |
1082 | { | |
1083 | if (CHECK_ATTR(BPF_OBJ)) | |
1084 | return -EINVAL; | |
1085 | ||
535e7b4b | 1086 | return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname)); |
b2197755 DB |
1087 | } |
1088 | ||
1089 | static int bpf_obj_get(const union bpf_attr *attr) | |
1090 | { | |
1091 | if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0) | |
1092 | return -EINVAL; | |
1093 | ||
535e7b4b | 1094 | return bpf_obj_get_user(u64_to_user_ptr(attr->pathname)); |
b2197755 DB |
1095 | } |
1096 | ||
f4324551 DM |
1097 | #ifdef CONFIG_CGROUP_BPF |
1098 | ||
464bc0fd | 1099 | #define BPF_PROG_ATTACH_LAST_FIELD attach_flags |
174a79ff | 1100 | |
5a67da2a | 1101 | static int sockmap_get_from_fd(const union bpf_attr *attr, bool attach) |
174a79ff | 1102 | { |
5a67da2a | 1103 | struct bpf_prog *prog = NULL; |
174a79ff JF |
1104 | int ufd = attr->target_fd; |
1105 | struct bpf_map *map; | |
1106 | struct fd f; | |
1107 | int err; | |
1108 | ||
1109 | f = fdget(ufd); | |
1110 | map = __bpf_map_get(f); | |
1111 | if (IS_ERR(map)) | |
1112 | return PTR_ERR(map); | |
1113 | ||
5a67da2a JF |
1114 | if (attach) { |
1115 | prog = bpf_prog_get_type(attr->attach_bpf_fd, | |
1116 | BPF_PROG_TYPE_SK_SKB); | |
1117 | if (IS_ERR(prog)) { | |
1118 | fdput(f); | |
1119 | return PTR_ERR(prog); | |
1120 | } | |
174a79ff JF |
1121 | } |
1122 | ||
5a67da2a | 1123 | err = sock_map_prog(map, prog, attr->attach_type); |
174a79ff JF |
1124 | if (err) { |
1125 | fdput(f); | |
5a67da2a JF |
1126 | if (prog) |
1127 | bpf_prog_put(prog); | |
ae2b27b8 | 1128 | return err; |
174a79ff JF |
1129 | } |
1130 | ||
1131 | fdput(f); | |
ae2b27b8 | 1132 | return 0; |
174a79ff | 1133 | } |
f4324551 DM |
1134 | |
1135 | static int bpf_prog_attach(const union bpf_attr *attr) | |
1136 | { | |
7f677633 | 1137 | enum bpf_prog_type ptype; |
f4324551 DM |
1138 | struct bpf_prog *prog; |
1139 | struct cgroup *cgrp; | |
7f677633 | 1140 | int ret; |
f4324551 DM |
1141 | |
1142 | if (!capable(CAP_NET_ADMIN)) | |
1143 | return -EPERM; | |
1144 | ||
1145 | if (CHECK_ATTR(BPF_PROG_ATTACH)) | |
1146 | return -EINVAL; | |
1147 | ||
7f677633 AS |
1148 | if (attr->attach_flags & ~BPF_F_ALLOW_OVERRIDE) |
1149 | return -EINVAL; | |
1150 | ||
f4324551 DM |
1151 | switch (attr->attach_type) { |
1152 | case BPF_CGROUP_INET_INGRESS: | |
1153 | case BPF_CGROUP_INET_EGRESS: | |
b2cd1257 | 1154 | ptype = BPF_PROG_TYPE_CGROUP_SKB; |
f4324551 | 1155 | break; |
61023658 DA |
1156 | case BPF_CGROUP_INET_SOCK_CREATE: |
1157 | ptype = BPF_PROG_TYPE_CGROUP_SOCK; | |
1158 | break; | |
40304b2a LB |
1159 | case BPF_CGROUP_SOCK_OPS: |
1160 | ptype = BPF_PROG_TYPE_SOCK_OPS; | |
1161 | break; | |
464bc0fd JF |
1162 | case BPF_SK_SKB_STREAM_PARSER: |
1163 | case BPF_SK_SKB_STREAM_VERDICT: | |
5a67da2a | 1164 | return sockmap_get_from_fd(attr, true); |
f4324551 DM |
1165 | default: |
1166 | return -EINVAL; | |
1167 | } | |
1168 | ||
b2cd1257 DA |
1169 | prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype); |
1170 | if (IS_ERR(prog)) | |
1171 | return PTR_ERR(prog); | |
1172 | ||
1173 | cgrp = cgroup_get_from_fd(attr->target_fd); | |
1174 | if (IS_ERR(cgrp)) { | |
1175 | bpf_prog_put(prog); | |
1176 | return PTR_ERR(cgrp); | |
1177 | } | |
1178 | ||
7f677633 AS |
1179 | ret = cgroup_bpf_update(cgrp, prog, attr->attach_type, |
1180 | attr->attach_flags & BPF_F_ALLOW_OVERRIDE); | |
1181 | if (ret) | |
1182 | bpf_prog_put(prog); | |
b2cd1257 DA |
1183 | cgroup_put(cgrp); |
1184 | ||
7f677633 | 1185 | return ret; |
f4324551 DM |
1186 | } |
1187 | ||
1188 | #define BPF_PROG_DETACH_LAST_FIELD attach_type | |
1189 | ||
1190 | static int bpf_prog_detach(const union bpf_attr *attr) | |
1191 | { | |
1192 | struct cgroup *cgrp; | |
7f677633 | 1193 | int ret; |
f4324551 DM |
1194 | |
1195 | if (!capable(CAP_NET_ADMIN)) | |
1196 | return -EPERM; | |
1197 | ||
1198 | if (CHECK_ATTR(BPF_PROG_DETACH)) | |
1199 | return -EINVAL; | |
1200 | ||
1201 | switch (attr->attach_type) { | |
1202 | case BPF_CGROUP_INET_INGRESS: | |
1203 | case BPF_CGROUP_INET_EGRESS: | |
61023658 | 1204 | case BPF_CGROUP_INET_SOCK_CREATE: |
40304b2a | 1205 | case BPF_CGROUP_SOCK_OPS: |
f4324551 DM |
1206 | cgrp = cgroup_get_from_fd(attr->target_fd); |
1207 | if (IS_ERR(cgrp)) | |
1208 | return PTR_ERR(cgrp); | |
1209 | ||
7f677633 | 1210 | ret = cgroup_bpf_update(cgrp, NULL, attr->attach_type, false); |
f4324551 DM |
1211 | cgroup_put(cgrp); |
1212 | break; | |
5a67da2a JF |
1213 | case BPF_SK_SKB_STREAM_PARSER: |
1214 | case BPF_SK_SKB_STREAM_VERDICT: | |
1215 | ret = sockmap_get_from_fd(attr, false); | |
1216 | break; | |
f4324551 DM |
1217 | default: |
1218 | return -EINVAL; | |
1219 | } | |
1220 | ||
7f677633 | 1221 | return ret; |
f4324551 | 1222 | } |
40304b2a | 1223 | |
f4324551 DM |
1224 | #endif /* CONFIG_CGROUP_BPF */ |
1225 | ||
1cf1cae9 AS |
1226 | #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration |
1227 | ||
1228 | static int bpf_prog_test_run(const union bpf_attr *attr, | |
1229 | union bpf_attr __user *uattr) | |
1230 | { | |
1231 | struct bpf_prog *prog; | |
1232 | int ret = -ENOTSUPP; | |
1233 | ||
1234 | if (CHECK_ATTR(BPF_PROG_TEST_RUN)) | |
1235 | return -EINVAL; | |
1236 | ||
1237 | prog = bpf_prog_get(attr->test.prog_fd); | |
1238 | if (IS_ERR(prog)) | |
1239 | return PTR_ERR(prog); | |
1240 | ||
1241 | if (prog->aux->ops->test_run) | |
1242 | ret = prog->aux->ops->test_run(prog, attr, uattr); | |
1243 | ||
1244 | bpf_prog_put(prog); | |
1245 | return ret; | |
1246 | } | |
1247 | ||
34ad5580 MKL |
1248 | #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id |
1249 | ||
1250 | static int bpf_obj_get_next_id(const union bpf_attr *attr, | |
1251 | union bpf_attr __user *uattr, | |
1252 | struct idr *idr, | |
1253 | spinlock_t *lock) | |
1254 | { | |
1255 | u32 next_id = attr->start_id; | |
1256 | int err = 0; | |
1257 | ||
1258 | if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX) | |
1259 | return -EINVAL; | |
1260 | ||
1261 | if (!capable(CAP_SYS_ADMIN)) | |
1262 | return -EPERM; | |
1263 | ||
1264 | next_id++; | |
1265 | spin_lock_bh(lock); | |
1266 | if (!idr_get_next(idr, &next_id)) | |
1267 | err = -ENOENT; | |
1268 | spin_unlock_bh(lock); | |
1269 | ||
1270 | if (!err) | |
1271 | err = put_user(next_id, &uattr->next_id); | |
1272 | ||
1273 | return err; | |
1274 | } | |
1275 | ||
b16d9aa4 MKL |
1276 | #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id |
1277 | ||
1278 | static int bpf_prog_get_fd_by_id(const union bpf_attr *attr) | |
1279 | { | |
1280 | struct bpf_prog *prog; | |
1281 | u32 id = attr->prog_id; | |
1282 | int fd; | |
1283 | ||
1284 | if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID)) | |
1285 | return -EINVAL; | |
1286 | ||
1287 | if (!capable(CAP_SYS_ADMIN)) | |
1288 | return -EPERM; | |
1289 | ||
1290 | spin_lock_bh(&prog_idr_lock); | |
1291 | prog = idr_find(&prog_idr, id); | |
1292 | if (prog) | |
1293 | prog = bpf_prog_inc_not_zero(prog); | |
1294 | else | |
1295 | prog = ERR_PTR(-ENOENT); | |
1296 | spin_unlock_bh(&prog_idr_lock); | |
1297 | ||
1298 | if (IS_ERR(prog)) | |
1299 | return PTR_ERR(prog); | |
1300 | ||
1301 | fd = bpf_prog_new_fd(prog); | |
1302 | if (fd < 0) | |
1303 | bpf_prog_put(prog); | |
1304 | ||
1305 | return fd; | |
1306 | } | |
1307 | ||
bd5f5f4e MKL |
1308 | #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD map_id |
1309 | ||
1310 | static int bpf_map_get_fd_by_id(const union bpf_attr *attr) | |
1311 | { | |
1312 | struct bpf_map *map; | |
1313 | u32 id = attr->map_id; | |
1314 | int fd; | |
1315 | ||
1316 | if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID)) | |
1317 | return -EINVAL; | |
1318 | ||
1319 | if (!capable(CAP_SYS_ADMIN)) | |
1320 | return -EPERM; | |
1321 | ||
1322 | spin_lock_bh(&map_idr_lock); | |
1323 | map = idr_find(&map_idr, id); | |
1324 | if (map) | |
1325 | map = bpf_map_inc_not_zero(map, true); | |
1326 | else | |
1327 | map = ERR_PTR(-ENOENT); | |
1328 | spin_unlock_bh(&map_idr_lock); | |
1329 | ||
1330 | if (IS_ERR(map)) | |
1331 | return PTR_ERR(map); | |
1332 | ||
1333 | fd = bpf_map_new_fd(map); | |
1334 | if (fd < 0) | |
1335 | bpf_map_put(map); | |
1336 | ||
1337 | return fd; | |
1338 | } | |
1339 | ||
1e270976 MKL |
1340 | static int bpf_prog_get_info_by_fd(struct bpf_prog *prog, |
1341 | const union bpf_attr *attr, | |
1342 | union bpf_attr __user *uattr) | |
1343 | { | |
1344 | struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info); | |
1345 | struct bpf_prog_info info = {}; | |
1346 | u32 info_len = attr->info.info_len; | |
1347 | char __user *uinsns; | |
1348 | u32 ulen; | |
1349 | int err; | |
1350 | ||
1351 | err = check_uarg_tail_zero(uinfo, sizeof(info), info_len); | |
1352 | if (err) | |
1353 | return err; | |
1354 | info_len = min_t(u32, sizeof(info), info_len); | |
1355 | ||
1356 | if (copy_from_user(&info, uinfo, info_len)) | |
89b09689 | 1357 | return -EFAULT; |
1e270976 MKL |
1358 | |
1359 | info.type = prog->type; | |
1360 | info.id = prog->aux->id; | |
1361 | ||
1362 | memcpy(info.tag, prog->tag, sizeof(prog->tag)); | |
1363 | ||
1364 | if (!capable(CAP_SYS_ADMIN)) { | |
1365 | info.jited_prog_len = 0; | |
1366 | info.xlated_prog_len = 0; | |
1367 | goto done; | |
1368 | } | |
1369 | ||
1370 | ulen = info.jited_prog_len; | |
1371 | info.jited_prog_len = prog->jited_len; | |
1372 | if (info.jited_prog_len && ulen) { | |
1373 | uinsns = u64_to_user_ptr(info.jited_prog_insns); | |
1374 | ulen = min_t(u32, info.jited_prog_len, ulen); | |
1375 | if (copy_to_user(uinsns, prog->bpf_func, ulen)) | |
1376 | return -EFAULT; | |
1377 | } | |
1378 | ||
1379 | ulen = info.xlated_prog_len; | |
9975a54b | 1380 | info.xlated_prog_len = bpf_prog_insn_size(prog); |
1e270976 MKL |
1381 | if (info.xlated_prog_len && ulen) { |
1382 | uinsns = u64_to_user_ptr(info.xlated_prog_insns); | |
1383 | ulen = min_t(u32, info.xlated_prog_len, ulen); | |
1384 | if (copy_to_user(uinsns, prog->insnsi, ulen)) | |
1385 | return -EFAULT; | |
1386 | } | |
1387 | ||
1388 | done: | |
1389 | if (copy_to_user(uinfo, &info, info_len) || | |
1390 | put_user(info_len, &uattr->info.info_len)) | |
1391 | return -EFAULT; | |
1392 | ||
1393 | return 0; | |
1394 | } | |
1395 | ||
1396 | static int bpf_map_get_info_by_fd(struct bpf_map *map, | |
1397 | const union bpf_attr *attr, | |
1398 | union bpf_attr __user *uattr) | |
1399 | { | |
1400 | struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info); | |
1401 | struct bpf_map_info info = {}; | |
1402 | u32 info_len = attr->info.info_len; | |
1403 | int err; | |
1404 | ||
1405 | err = check_uarg_tail_zero(uinfo, sizeof(info), info_len); | |
1406 | if (err) | |
1407 | return err; | |
1408 | info_len = min_t(u32, sizeof(info), info_len); | |
1409 | ||
1410 | info.type = map->map_type; | |
1411 | info.id = map->id; | |
1412 | info.key_size = map->key_size; | |
1413 | info.value_size = map->value_size; | |
1414 | info.max_entries = map->max_entries; | |
1415 | info.map_flags = map->map_flags; | |
1416 | ||
1417 | if (copy_to_user(uinfo, &info, info_len) || | |
1418 | put_user(info_len, &uattr->info.info_len)) | |
1419 | return -EFAULT; | |
1420 | ||
1421 | return 0; | |
1422 | } | |
1423 | ||
1424 | #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info | |
1425 | ||
1426 | static int bpf_obj_get_info_by_fd(const union bpf_attr *attr, | |
1427 | union bpf_attr __user *uattr) | |
1428 | { | |
1429 | int ufd = attr->info.bpf_fd; | |
1430 | struct fd f; | |
1431 | int err; | |
1432 | ||
1433 | if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD)) | |
1434 | return -EINVAL; | |
1435 | ||
1436 | f = fdget(ufd); | |
1437 | if (!f.file) | |
1438 | return -EBADFD; | |
1439 | ||
1440 | if (f.file->f_op == &bpf_prog_fops) | |
1441 | err = bpf_prog_get_info_by_fd(f.file->private_data, attr, | |
1442 | uattr); | |
1443 | else if (f.file->f_op == &bpf_map_fops) | |
1444 | err = bpf_map_get_info_by_fd(f.file->private_data, attr, | |
1445 | uattr); | |
1446 | else | |
1447 | err = -EINVAL; | |
1448 | ||
1449 | fdput(f); | |
1450 | return err; | |
1451 | } | |
1452 | ||
99c55f7d AS |
1453 | SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size) |
1454 | { | |
1455 | union bpf_attr attr = {}; | |
1456 | int err; | |
1457 | ||
1be7f75d | 1458 | if (!capable(CAP_SYS_ADMIN) && sysctl_unprivileged_bpf_disabled) |
99c55f7d AS |
1459 | return -EPERM; |
1460 | ||
1e270976 MKL |
1461 | err = check_uarg_tail_zero(uattr, sizeof(attr), size); |
1462 | if (err) | |
1463 | return err; | |
1464 | size = min_t(u32, size, sizeof(attr)); | |
99c55f7d AS |
1465 | |
1466 | /* copy attributes from user space, may be less than sizeof(bpf_attr) */ | |
1467 | if (copy_from_user(&attr, uattr, size) != 0) | |
1468 | return -EFAULT; | |
1469 | ||
1470 | switch (cmd) { | |
1471 | case BPF_MAP_CREATE: | |
1472 | err = map_create(&attr); | |
1473 | break; | |
db20fd2b AS |
1474 | case BPF_MAP_LOOKUP_ELEM: |
1475 | err = map_lookup_elem(&attr); | |
1476 | break; | |
1477 | case BPF_MAP_UPDATE_ELEM: | |
1478 | err = map_update_elem(&attr); | |
1479 | break; | |
1480 | case BPF_MAP_DELETE_ELEM: | |
1481 | err = map_delete_elem(&attr); | |
1482 | break; | |
1483 | case BPF_MAP_GET_NEXT_KEY: | |
1484 | err = map_get_next_key(&attr); | |
1485 | break; | |
09756af4 AS |
1486 | case BPF_PROG_LOAD: |
1487 | err = bpf_prog_load(&attr); | |
1488 | break; | |
b2197755 DB |
1489 | case BPF_OBJ_PIN: |
1490 | err = bpf_obj_pin(&attr); | |
1491 | break; | |
1492 | case BPF_OBJ_GET: | |
1493 | err = bpf_obj_get(&attr); | |
1494 | break; | |
f4324551 DM |
1495 | #ifdef CONFIG_CGROUP_BPF |
1496 | case BPF_PROG_ATTACH: | |
1497 | err = bpf_prog_attach(&attr); | |
1498 | break; | |
1499 | case BPF_PROG_DETACH: | |
1500 | err = bpf_prog_detach(&attr); | |
1501 | break; | |
1502 | #endif | |
1cf1cae9 AS |
1503 | case BPF_PROG_TEST_RUN: |
1504 | err = bpf_prog_test_run(&attr, uattr); | |
1505 | break; | |
34ad5580 MKL |
1506 | case BPF_PROG_GET_NEXT_ID: |
1507 | err = bpf_obj_get_next_id(&attr, uattr, | |
1508 | &prog_idr, &prog_idr_lock); | |
1509 | break; | |
1510 | case BPF_MAP_GET_NEXT_ID: | |
1511 | err = bpf_obj_get_next_id(&attr, uattr, | |
1512 | &map_idr, &map_idr_lock); | |
1513 | break; | |
b16d9aa4 MKL |
1514 | case BPF_PROG_GET_FD_BY_ID: |
1515 | err = bpf_prog_get_fd_by_id(&attr); | |
1516 | break; | |
bd5f5f4e MKL |
1517 | case BPF_MAP_GET_FD_BY_ID: |
1518 | err = bpf_map_get_fd_by_id(&attr); | |
1519 | break; | |
1e270976 MKL |
1520 | case BPF_OBJ_GET_INFO_BY_FD: |
1521 | err = bpf_obj_get_info_by_fd(&attr, uattr); | |
1522 | break; | |
99c55f7d AS |
1523 | default: |
1524 | err = -EINVAL; | |
1525 | break; | |
1526 | } | |
1527 | ||
1528 | return err; | |
1529 | } |