blob: 0048cb24ba7b43eba7c61b6ec07fde54fc9657a5 [file] [log] [blame]
Alexei Starovoitov99c55f72014-09-26 00:16:57 -07001/* 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>
Daniel Borkmanna67edbf2017-01-25 02:28:18 +010013#include <linux/bpf_trace.h>
Alexei Starovoitov99c55f72014-09-26 00:16:57 -070014#include <linux/syscalls.h>
15#include <linux/slab.h>
Ingo Molnar3f07c012017-02-08 18:51:30 +010016#include <linux/sched/signal.h>
Daniel Borkmannd407bd22017-01-18 15:14:17 +010017#include <linux/vmalloc.h>
18#include <linux/mmzone.h>
Alexei Starovoitov99c55f72014-09-26 00:16:57 -070019#include <linux/anon_inodes.h>
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -070020#include <linux/file.h>
Alexei Starovoitov09756af2014-09-26 00:17:00 -070021#include <linux/license.h>
22#include <linux/filter.h>
Alexei Starovoitov25415172015-03-25 12:49:20 -070023#include <linux/version.h>
Mickaël Salaün535e7b4b2016-11-13 19:44:03 +010024#include <linux/kernel.h>
Martin KaFai Laudc4bb0e2017-06-05 12:15:46 -070025#include <linux/idr.h>
Martin KaFai Laucb4d2b32017-09-27 14:37:52 -070026#include <linux/cred.h>
27#include <linux/timekeeping.h>
28#include <linux/ctype.h>
Alexei Starovoitov99c55f72014-09-26 00:16:57 -070029
Martin KaFai Lau14dc6f02017-06-27 23:08:34 -070030#define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY || \
31 (map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
32 (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
33 (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
34#define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
35#define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_HASH(map))
36
Alexei Starovoitovb121d1e2016-03-07 21:57:13 -080037DEFINE_PER_CPU(int, bpf_prog_active);
Martin KaFai Laudc4bb0e2017-06-05 12:15:46 -070038static DEFINE_IDR(prog_idr);
39static DEFINE_SPINLOCK(prog_idr_lock);
Martin KaFai Lauf3f1c052017-06-05 12:15:47 -070040static DEFINE_IDR(map_idr);
41static DEFINE_SPINLOCK(map_idr_lock);
Alexei Starovoitovb121d1e2016-03-07 21:57:13 -080042
Alexei Starovoitov1be7f752015-10-07 22:23:21 -070043int sysctl_unprivileged_bpf_disabled __read_mostly;
44
Johannes Berg40077e02017-04-11 15:34:58 +020045static const struct bpf_map_ops * const bpf_map_types[] = {
46#define BPF_PROG_TYPE(_id, _ops)
47#define BPF_MAP_TYPE(_id, _ops) \
48 [_id] = &_ops,
49#include <linux/bpf_types.h>
50#undef BPF_PROG_TYPE
51#undef BPF_MAP_TYPE
52};
Alexei Starovoitov99c55f72014-09-26 00:16:57 -070053
Mickaël Salaün752ba562017-08-07 20:45:20 +020054/*
55 * If we're handed a bigger struct than we know of, ensure all the unknown bits
56 * are 0 - i.e. new user-space does not rely on any kernel feature extensions
57 * we don't know about yet.
58 *
59 * There is a ToCToU between this function call and the following
60 * copy_from_user() call. However, this is not a concern since this function is
61 * meant to be a future-proofing of bits.
62 */
Mickaël Salaün58291a72017-08-07 20:45:19 +020063static int check_uarg_tail_zero(void __user *uaddr,
64 size_t expected_size,
65 size_t actual_size)
66{
67 unsigned char __user *addr;
68 unsigned char __user *end;
69 unsigned char val;
70 int err;
71
Mickaël Salaün752ba562017-08-07 20:45:20 +020072 if (unlikely(actual_size > PAGE_SIZE)) /* silly large */
73 return -E2BIG;
74
75 if (unlikely(!access_ok(VERIFY_READ, uaddr, actual_size)))
76 return -EFAULT;
77
Mickaël Salaün58291a72017-08-07 20:45:19 +020078 if (actual_size <= expected_size)
79 return 0;
80
81 addr = uaddr + expected_size;
82 end = uaddr + actual_size;
83
84 for (; addr < end; addr++) {
85 err = get_user(val, addr);
86 if (err)
87 return err;
88 if (val)
89 return -E2BIG;
90 }
91
92 return 0;
93}
94
Alexei Starovoitov99c55f72014-09-26 00:16:57 -070095static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
96{
Alexei Starovoitov99c55f72014-09-26 00:16:57 -070097 struct bpf_map *map;
98
Johannes Berg40077e02017-04-11 15:34:58 +020099 if (attr->map_type >= ARRAY_SIZE(bpf_map_types) ||
100 !bpf_map_types[attr->map_type])
101 return ERR_PTR(-EINVAL);
Alexei Starovoitov99c55f72014-09-26 00:16:57 -0700102
Johannes Berg40077e02017-04-11 15:34:58 +0200103 map = bpf_map_types[attr->map_type]->map_alloc(attr);
104 if (IS_ERR(map))
105 return map;
106 map->ops = bpf_map_types[attr->map_type];
107 map->map_type = attr->map_type;
108 return map;
Alexei Starovoitov99c55f72014-09-26 00:16:57 -0700109}
110
Martin KaFai Lau96eabe72017-08-18 11:28:00 -0700111void *bpf_map_area_alloc(size_t size, int numa_node)
Daniel Borkmannd407bd22017-01-18 15:14:17 +0100112{
113 /* We definitely need __GFP_NORETRY, so OOM killer doesn't
114 * trigger under memory pressure as we really just want to
115 * fail instead.
116 */
117 const gfp_t flags = __GFP_NOWARN | __GFP_NORETRY | __GFP_ZERO;
118 void *area;
119
120 if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
Martin KaFai Lau96eabe72017-08-18 11:28:00 -0700121 area = kmalloc_node(size, GFP_USER | flags, numa_node);
Daniel Borkmannd407bd22017-01-18 15:14:17 +0100122 if (area != NULL)
123 return area;
124 }
125
Martin KaFai Lau96eabe72017-08-18 11:28:00 -0700126 return __vmalloc_node_flags_caller(size, numa_node, GFP_KERNEL | flags,
127 __builtin_return_address(0));
Daniel Borkmannd407bd22017-01-18 15:14:17 +0100128}
129
130void bpf_map_area_free(void *area)
131{
132 kvfree(area);
133}
134
Alexei Starovoitov6c905982016-03-07 21:57:15 -0800135int bpf_map_precharge_memlock(u32 pages)
136{
137 struct user_struct *user = get_current_user();
138 unsigned long memlock_limit, cur;
139
140 memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
141 cur = atomic_long_read(&user->locked_vm);
142 free_uid(user);
143 if (cur + pages > memlock_limit)
144 return -EPERM;
145 return 0;
146}
147
Alexei Starovoitovaaac3ba2015-10-07 22:23:22 -0700148static int bpf_map_charge_memlock(struct bpf_map *map)
149{
150 struct user_struct *user = get_current_user();
151 unsigned long memlock_limit;
152
153 memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
154
155 atomic_long_add(map->pages, &user->locked_vm);
156
157 if (atomic_long_read(&user->locked_vm) > memlock_limit) {
158 atomic_long_sub(map->pages, &user->locked_vm);
159 free_uid(user);
160 return -EPERM;
161 }
162 map->user = user;
163 return 0;
164}
165
166static void bpf_map_uncharge_memlock(struct bpf_map *map)
167{
168 struct user_struct *user = map->user;
169
170 atomic_long_sub(map->pages, &user->locked_vm);
171 free_uid(user);
172}
173
Martin KaFai Lauf3f1c052017-06-05 12:15:47 -0700174static int bpf_map_alloc_id(struct bpf_map *map)
175{
176 int id;
177
178 spin_lock_bh(&map_idr_lock);
179 id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
180 if (id > 0)
181 map->id = id;
182 spin_unlock_bh(&map_idr_lock);
183
184 if (WARN_ON_ONCE(!id))
185 return -ENOSPC;
186
187 return id > 0 ? 0 : id;
188}
189
Martin KaFai Laubd5f5f4e2017-06-05 12:15:50 -0700190static void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
Martin KaFai Lauf3f1c052017-06-05 12:15:47 -0700191{
Eric Dumazet930651a2017-09-19 09:15:59 -0700192 unsigned long flags;
193
Martin KaFai Laubd5f5f4e2017-06-05 12:15:50 -0700194 if (do_idr_lock)
Eric Dumazet930651a2017-09-19 09:15:59 -0700195 spin_lock_irqsave(&map_idr_lock, flags);
Martin KaFai Laubd5f5f4e2017-06-05 12:15:50 -0700196 else
197 __acquire(&map_idr_lock);
198
Martin KaFai Lauf3f1c052017-06-05 12:15:47 -0700199 idr_remove(&map_idr, map->id);
Martin KaFai Laubd5f5f4e2017-06-05 12:15:50 -0700200
201 if (do_idr_lock)
Eric Dumazet930651a2017-09-19 09:15:59 -0700202 spin_unlock_irqrestore(&map_idr_lock, flags);
Martin KaFai Laubd5f5f4e2017-06-05 12:15:50 -0700203 else
204 __release(&map_idr_lock);
Martin KaFai Lauf3f1c052017-06-05 12:15:47 -0700205}
206
Alexei Starovoitov99c55f72014-09-26 00:16:57 -0700207/* called from workqueue */
208static void bpf_map_free_deferred(struct work_struct *work)
209{
210 struct bpf_map *map = container_of(work, struct bpf_map, work);
211
Alexei Starovoitovaaac3ba2015-10-07 22:23:22 -0700212 bpf_map_uncharge_memlock(map);
Alexei Starovoitov99c55f72014-09-26 00:16:57 -0700213 /* implementation dependent freeing */
214 map->ops->map_free(map);
215}
216
Daniel Borkmannc9da1612015-11-24 21:28:15 +0100217static void bpf_map_put_uref(struct bpf_map *map)
218{
219 if (atomic_dec_and_test(&map->usercnt)) {
220 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY)
221 bpf_fd_array_map_clear(map);
222 }
223}
224
Alexei Starovoitov99c55f72014-09-26 00:16:57 -0700225/* decrement map refcnt and schedule it for freeing via workqueue
226 * (unrelying map implementation ops->map_free() might sleep)
227 */
Martin KaFai Laubd5f5f4e2017-06-05 12:15:50 -0700228static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
Alexei Starovoitov99c55f72014-09-26 00:16:57 -0700229{
230 if (atomic_dec_and_test(&map->refcnt)) {
Martin KaFai Lau34ad5582017-06-05 12:15:48 -0700231 /* bpf_map_free_id() must be called first */
Martin KaFai Laubd5f5f4e2017-06-05 12:15:50 -0700232 bpf_map_free_id(map, do_idr_lock);
Alexei Starovoitov99c55f72014-09-26 00:16:57 -0700233 INIT_WORK(&map->work, bpf_map_free_deferred);
234 schedule_work(&map->work);
235 }
236}
237
Martin KaFai Laubd5f5f4e2017-06-05 12:15:50 -0700238void bpf_map_put(struct bpf_map *map)
239{
240 __bpf_map_put(map, true);
241}
242
Daniel Borkmannc9da1612015-11-24 21:28:15 +0100243void bpf_map_put_with_uref(struct bpf_map *map)
244{
245 bpf_map_put_uref(map);
246 bpf_map_put(map);
247}
248
Alexei Starovoitov99c55f72014-09-26 00:16:57 -0700249static int bpf_map_release(struct inode *inode, struct file *filp)
250{
Daniel Borkmann61d1b6a2016-06-15 22:47:12 +0200251 struct bpf_map *map = filp->private_data;
252
253 if (map->ops->map_release)
254 map->ops->map_release(map, filp);
255
256 bpf_map_put_with_uref(map);
Alexei Starovoitov99c55f72014-09-26 00:16:57 -0700257 return 0;
258}
259
Daniel Borkmannf99bf202015-11-19 11:56:22 +0100260#ifdef CONFIG_PROC_FS
261static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
262{
263 const struct bpf_map *map = filp->private_data;
Daniel Borkmann21116b72016-11-26 01:28:07 +0100264 const struct bpf_array *array;
265 u32 owner_prog_type = 0;
Daniel Borkmann9780c0a2017-07-02 02:13:28 +0200266 u32 owner_jited = 0;
Daniel Borkmann21116b72016-11-26 01:28:07 +0100267
268 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
269 array = container_of(map, struct bpf_array, map);
270 owner_prog_type = array->owner_prog_type;
Daniel Borkmann9780c0a2017-07-02 02:13:28 +0200271 owner_jited = array->owner_jited;
Daniel Borkmann21116b72016-11-26 01:28:07 +0100272 }
Daniel Borkmannf99bf202015-11-19 11:56:22 +0100273
274 seq_printf(m,
275 "map_type:\t%u\n"
276 "key_size:\t%u\n"
277 "value_size:\t%u\n"
Daniel Borkmann322cea22016-03-25 00:30:25 +0100278 "max_entries:\t%u\n"
Daniel Borkmann21116b72016-11-26 01:28:07 +0100279 "map_flags:\t%#x\n"
280 "memlock:\t%llu\n",
Daniel Borkmannf99bf202015-11-19 11:56:22 +0100281 map->map_type,
282 map->key_size,
283 map->value_size,
Daniel Borkmann322cea22016-03-25 00:30:25 +0100284 map->max_entries,
Daniel Borkmann21116b72016-11-26 01:28:07 +0100285 map->map_flags,
286 map->pages * 1ULL << PAGE_SHIFT);
287
Daniel Borkmann9780c0a2017-07-02 02:13:28 +0200288 if (owner_prog_type) {
Daniel Borkmann21116b72016-11-26 01:28:07 +0100289 seq_printf(m, "owner_prog_type:\t%u\n",
290 owner_prog_type);
Daniel Borkmann9780c0a2017-07-02 02:13:28 +0200291 seq_printf(m, "owner_jited:\t%u\n",
292 owner_jited);
293 }
Daniel Borkmannf99bf202015-11-19 11:56:22 +0100294}
295#endif
296
Alexei Starovoitov99c55f72014-09-26 00:16:57 -0700297static const struct file_operations bpf_map_fops = {
Daniel Borkmannf99bf202015-11-19 11:56:22 +0100298#ifdef CONFIG_PROC_FS
299 .show_fdinfo = bpf_map_show_fdinfo,
300#endif
301 .release = bpf_map_release,
Alexei Starovoitov99c55f72014-09-26 00:16:57 -0700302};
303
Daniel Borkmannb2197752015-10-29 14:58:09 +0100304int bpf_map_new_fd(struct bpf_map *map)
Daniel Borkmannaa797812015-10-29 14:58:06 +0100305{
306 return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
307 O_RDWR | O_CLOEXEC);
308}
309
Alexei Starovoitov99c55f72014-09-26 00:16:57 -0700310/* helper macro to check that unused fields 'union bpf_attr' are zero */
311#define CHECK_ATTR(CMD) \
312 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
313 sizeof(attr->CMD##_LAST_FIELD), 0, \
314 sizeof(*attr) - \
315 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
316 sizeof(attr->CMD##_LAST_FIELD)) != NULL
317
Martin KaFai Laucb4d2b32017-09-27 14:37:52 -0700318/* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
319 * Return 0 on success and < 0 on error.
320 */
321static int bpf_obj_name_cpy(char *dst, const char *src)
322{
323 const char *end = src + BPF_OBJ_NAME_LEN;
324
325 /* Copy all isalnum() and '_' char */
326 while (src < end && *src) {
327 if (!isalnum(*src) && *src != '_')
328 return -EINVAL;
329 *dst++ = *src++;
330 }
331
332 /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
333 if (src == end)
334 return -EINVAL;
335
336 /* '\0' terminates dst */
337 *dst = 0;
338
339 return 0;
340}
341
Martin KaFai Lauad5b1772017-09-27 14:37:53 -0700342#define BPF_MAP_CREATE_LAST_FIELD map_name
Alexei Starovoitov99c55f72014-09-26 00:16:57 -0700343/* called via syscall */
344static int map_create(union bpf_attr *attr)
345{
Martin KaFai Lau96eabe72017-08-18 11:28:00 -0700346 int numa_node = bpf_map_attr_numa_node(attr);
Alexei Starovoitov99c55f72014-09-26 00:16:57 -0700347 struct bpf_map *map;
348 int err;
349
350 err = CHECK_ATTR(BPF_MAP_CREATE);
351 if (err)
352 return -EINVAL;
353
Martin KaFai Lau96eabe72017-08-18 11:28:00 -0700354 if (numa_node != NUMA_NO_NODE &&
Eric Dumazet96e5ae42017-09-04 22:41:02 -0700355 ((unsigned int)numa_node >= nr_node_ids ||
356 !node_online(numa_node)))
Martin KaFai Lau96eabe72017-08-18 11:28:00 -0700357 return -EINVAL;
358
Alexei Starovoitov99c55f72014-09-26 00:16:57 -0700359 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
360 map = find_and_alloc_map(attr);
361 if (IS_ERR(map))
362 return PTR_ERR(map);
363
Martin KaFai Lauad5b1772017-09-27 14:37:53 -0700364 err = bpf_obj_name_cpy(map->name, attr->map_name);
365 if (err)
366 goto free_map_nouncharge;
367
Alexei Starovoitov99c55f72014-09-26 00:16:57 -0700368 atomic_set(&map->refcnt, 1);
Daniel Borkmannc9da1612015-11-24 21:28:15 +0100369 atomic_set(&map->usercnt, 1);
Alexei Starovoitov99c55f72014-09-26 00:16:57 -0700370
Alexei Starovoitovaaac3ba2015-10-07 22:23:22 -0700371 err = bpf_map_charge_memlock(map);
372 if (err)
Daniel Borkmann20b2b242016-11-04 00:56:31 +0100373 goto free_map_nouncharge;
Alexei Starovoitovaaac3ba2015-10-07 22:23:22 -0700374
Martin KaFai Lauf3f1c052017-06-05 12:15:47 -0700375 err = bpf_map_alloc_id(map);
376 if (err)
377 goto free_map;
378
Daniel Borkmannaa797812015-10-29 14:58:06 +0100379 err = bpf_map_new_fd(map);
Martin KaFai Laubd5f5f4e2017-06-05 12:15:50 -0700380 if (err < 0) {
381 /* failed to allocate fd.
382 * bpf_map_put() is needed because the above
383 * bpf_map_alloc_id() has published the map
384 * to the userspace and the userspace may
385 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
386 */
387 bpf_map_put(map);
388 return err;
389 }
Alexei Starovoitov99c55f72014-09-26 00:16:57 -0700390
Daniel Borkmanna67edbf2017-01-25 02:28:18 +0100391 trace_bpf_map_create(map, err);
Alexei Starovoitov99c55f72014-09-26 00:16:57 -0700392 return err;
393
394free_map:
Daniel Borkmann20b2b242016-11-04 00:56:31 +0100395 bpf_map_uncharge_memlock(map);
396free_map_nouncharge:
Alexei Starovoitov99c55f72014-09-26 00:16:57 -0700397 map->ops->map_free(map);
398 return err;
399}
400
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700401/* if error is returned, fd is released.
402 * On success caller should complete fd access with matching fdput()
403 */
Daniel Borkmannc2101292015-10-29 14:58:07 +0100404struct bpf_map *__bpf_map_get(struct fd f)
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700405{
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700406 if (!f.file)
407 return ERR_PTR(-EBADF);
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700408 if (f.file->f_op != &bpf_map_fops) {
409 fdput(f);
410 return ERR_PTR(-EINVAL);
411 }
412
Daniel Borkmannc2101292015-10-29 14:58:07 +0100413 return f.file->private_data;
414}
415
Alexei Starovoitov92117d82016-04-27 18:56:20 -0700416/* prog's and map's refcnt limit */
417#define BPF_MAX_REFCNT 32768
418
419struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
Daniel Borkmannc9da1612015-11-24 21:28:15 +0100420{
Alexei Starovoitov92117d82016-04-27 18:56:20 -0700421 if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
422 atomic_dec(&map->refcnt);
423 return ERR_PTR(-EBUSY);
424 }
Daniel Borkmannc9da1612015-11-24 21:28:15 +0100425 if (uref)
426 atomic_inc(&map->usercnt);
Alexei Starovoitov92117d82016-04-27 18:56:20 -0700427 return map;
Daniel Borkmannc9da1612015-11-24 21:28:15 +0100428}
429
430struct bpf_map *bpf_map_get_with_uref(u32 ufd)
Daniel Borkmannc2101292015-10-29 14:58:07 +0100431{
432 struct fd f = fdget(ufd);
433 struct bpf_map *map;
434
435 map = __bpf_map_get(f);
436 if (IS_ERR(map))
437 return map;
438
Alexei Starovoitov92117d82016-04-27 18:56:20 -0700439 map = bpf_map_inc(map, true);
Daniel Borkmannc2101292015-10-29 14:58:07 +0100440 fdput(f);
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700441
442 return map;
443}
444
Martin KaFai Laubd5f5f4e2017-06-05 12:15:50 -0700445/* map_idr_lock should have been held */
446static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
447 bool uref)
448{
449 int refold;
450
451 refold = __atomic_add_unless(&map->refcnt, 1, 0);
452
453 if (refold >= BPF_MAX_REFCNT) {
454 __bpf_map_put(map, false);
455 return ERR_PTR(-EBUSY);
456 }
457
458 if (!refold)
459 return ERR_PTR(-ENOENT);
460
461 if (uref)
462 atomic_inc(&map->usercnt);
463
464 return map;
465}
466
Alexei Starovoitovb8cdc052016-03-09 18:56:49 -0800467int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
468{
469 return -ENOTSUPP;
470}
471
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700472/* last field in 'union bpf_attr' used by this command */
473#define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
474
475static int map_lookup_elem(union bpf_attr *attr)
476{
Mickaël Salaün535e7b4b2016-11-13 19:44:03 +0100477 void __user *ukey = u64_to_user_ptr(attr->key);
478 void __user *uvalue = u64_to_user_ptr(attr->value);
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700479 int ufd = attr->map_fd;
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700480 struct bpf_map *map;
Alexei Starovoitov8ebe6672015-01-22 17:11:08 -0800481 void *key, *value, *ptr;
Alexei Starovoitov15a07b32016-02-01 22:39:55 -0800482 u32 value_size;
Daniel Borkmann592867b2015-09-08 18:00:09 +0200483 struct fd f;
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700484 int err;
485
486 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
487 return -EINVAL;
488
Daniel Borkmann592867b2015-09-08 18:00:09 +0200489 f = fdget(ufd);
Daniel Borkmannc2101292015-10-29 14:58:07 +0100490 map = __bpf_map_get(f);
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700491 if (IS_ERR(map))
492 return PTR_ERR(map);
493
Al Viroe4448ed2017-05-13 18:43:00 -0400494 key = memdup_user(ukey, map->key_size);
495 if (IS_ERR(key)) {
496 err = PTR_ERR(key);
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700497 goto err_put;
Al Viroe4448ed2017-05-13 18:43:00 -0400498 }
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700499
Alexei Starovoitov15a07b32016-02-01 22:39:55 -0800500 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
Martin KaFai Lau8f844932016-11-11 10:55:10 -0800501 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
Alexei Starovoitov15a07b32016-02-01 22:39:55 -0800502 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
503 value_size = round_up(map->value_size, 8) * num_possible_cpus();
Martin KaFai Lau14dc6f02017-06-27 23:08:34 -0700504 else if (IS_FD_MAP(map))
505 value_size = sizeof(u32);
Alexei Starovoitov15a07b32016-02-01 22:39:55 -0800506 else
507 value_size = map->value_size;
508
Alexei Starovoitov8ebe6672015-01-22 17:11:08 -0800509 err = -ENOMEM;
Alexei Starovoitov15a07b32016-02-01 22:39:55 -0800510 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700511 if (!value)
Alexei Starovoitov8ebe6672015-01-22 17:11:08 -0800512 goto free_key;
513
Martin KaFai Lau8f844932016-11-11 10:55:10 -0800514 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
515 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
Alexei Starovoitov15a07b32016-02-01 22:39:55 -0800516 err = bpf_percpu_hash_copy(map, key, value);
517 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
518 err = bpf_percpu_array_copy(map, key, value);
Alexei Starovoitov557c0c62016-03-07 21:57:17 -0800519 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
520 err = bpf_stackmap_copy(map, key, value);
Martin KaFai Lau14dc6f02017-06-27 23:08:34 -0700521 } else if (IS_FD_ARRAY(map)) {
522 err = bpf_fd_array_map_lookup_elem(map, key, value);
523 } else if (IS_FD_HASH(map)) {
524 err = bpf_fd_htab_map_lookup_elem(map, key, value);
Alexei Starovoitov15a07b32016-02-01 22:39:55 -0800525 } else {
526 rcu_read_lock();
527 ptr = map->ops->map_lookup_elem(map, key);
528 if (ptr)
529 memcpy(value, ptr, value_size);
530 rcu_read_unlock();
531 err = ptr ? 0 : -ENOENT;
532 }
Alexei Starovoitov8ebe6672015-01-22 17:11:08 -0800533
Alexei Starovoitov15a07b32016-02-01 22:39:55 -0800534 if (err)
Alexei Starovoitov8ebe6672015-01-22 17:11:08 -0800535 goto free_value;
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700536
537 err = -EFAULT;
Alexei Starovoitov15a07b32016-02-01 22:39:55 -0800538 if (copy_to_user(uvalue, value, value_size) != 0)
Alexei Starovoitov8ebe6672015-01-22 17:11:08 -0800539 goto free_value;
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700540
Daniel Borkmanna67edbf2017-01-25 02:28:18 +0100541 trace_bpf_map_lookup_elem(map, ufd, key, value);
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700542 err = 0;
543
Alexei Starovoitov8ebe6672015-01-22 17:11:08 -0800544free_value:
545 kfree(value);
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700546free_key:
547 kfree(key);
548err_put:
549 fdput(f);
550 return err;
551}
552
Alexei Starovoitov3274f522014-11-13 17:36:44 -0800553#define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700554
555static int map_update_elem(union bpf_attr *attr)
556{
Mickaël Salaün535e7b4b2016-11-13 19:44:03 +0100557 void __user *ukey = u64_to_user_ptr(attr->key);
558 void __user *uvalue = u64_to_user_ptr(attr->value);
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700559 int ufd = attr->map_fd;
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700560 struct bpf_map *map;
561 void *key, *value;
Alexei Starovoitov15a07b32016-02-01 22:39:55 -0800562 u32 value_size;
Daniel Borkmann592867b2015-09-08 18:00:09 +0200563 struct fd f;
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700564 int err;
565
566 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
567 return -EINVAL;
568
Daniel Borkmann592867b2015-09-08 18:00:09 +0200569 f = fdget(ufd);
Daniel Borkmannc2101292015-10-29 14:58:07 +0100570 map = __bpf_map_get(f);
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700571 if (IS_ERR(map))
572 return PTR_ERR(map);
573
Al Viroe4448ed2017-05-13 18:43:00 -0400574 key = memdup_user(ukey, map->key_size);
575 if (IS_ERR(key)) {
576 err = PTR_ERR(key);
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700577 goto err_put;
Al Viroe4448ed2017-05-13 18:43:00 -0400578 }
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700579
Alexei Starovoitov15a07b32016-02-01 22:39:55 -0800580 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
Martin KaFai Lau8f844932016-11-11 10:55:10 -0800581 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
Alexei Starovoitov15a07b32016-02-01 22:39:55 -0800582 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
583 value_size = round_up(map->value_size, 8) * num_possible_cpus();
584 else
585 value_size = map->value_size;
586
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700587 err = -ENOMEM;
Alexei Starovoitov15a07b32016-02-01 22:39:55 -0800588 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700589 if (!value)
590 goto free_key;
591
592 err = -EFAULT;
Alexei Starovoitov15a07b32016-02-01 22:39:55 -0800593 if (copy_from_user(value, uvalue, value_size) != 0)
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700594 goto free_value;
595
Alexei Starovoitovb121d1e2016-03-07 21:57:13 -0800596 /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
597 * inside bpf map update or delete otherwise deadlocks are possible
598 */
599 preempt_disable();
600 __this_cpu_inc(bpf_prog_active);
Martin KaFai Lau8f844932016-11-11 10:55:10 -0800601 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
602 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
Alexei Starovoitov15a07b32016-02-01 22:39:55 -0800603 err = bpf_percpu_hash_update(map, key, value, attr->flags);
604 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
605 err = bpf_percpu_array_update(map, key, value, attr->flags);
Daniel Borkmannd056a782016-06-15 22:47:13 +0200606 } else if (map->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY ||
Martin KaFai Lau4ed8ec52016-06-30 10:28:43 -0700607 map->map_type == BPF_MAP_TYPE_PROG_ARRAY ||
Martin KaFai Lau56f668d2017-03-22 10:00:33 -0700608 map->map_type == BPF_MAP_TYPE_CGROUP_ARRAY ||
609 map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS) {
Daniel Borkmannd056a782016-06-15 22:47:13 +0200610 rcu_read_lock();
611 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
612 attr->flags);
613 rcu_read_unlock();
Martin KaFai Laubcc6b1b2017-03-22 10:00:34 -0700614 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
615 rcu_read_lock();
616 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
617 attr->flags);
618 rcu_read_unlock();
Alexei Starovoitov15a07b32016-02-01 22:39:55 -0800619 } else {
620 rcu_read_lock();
621 err = map->ops->map_update_elem(map, key, value, attr->flags);
622 rcu_read_unlock();
623 }
Alexei Starovoitovb121d1e2016-03-07 21:57:13 -0800624 __this_cpu_dec(bpf_prog_active);
625 preempt_enable();
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700626
Daniel Borkmanna67edbf2017-01-25 02:28:18 +0100627 if (!err)
628 trace_bpf_map_update_elem(map, ufd, key, value);
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700629free_value:
630 kfree(value);
631free_key:
632 kfree(key);
633err_put:
634 fdput(f);
635 return err;
636}
637
638#define BPF_MAP_DELETE_ELEM_LAST_FIELD key
639
640static int map_delete_elem(union bpf_attr *attr)
641{
Mickaël Salaün535e7b4b2016-11-13 19:44:03 +0100642 void __user *ukey = u64_to_user_ptr(attr->key);
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700643 int ufd = attr->map_fd;
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700644 struct bpf_map *map;
Daniel Borkmann592867b2015-09-08 18:00:09 +0200645 struct fd f;
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700646 void *key;
647 int err;
648
649 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
650 return -EINVAL;
651
Daniel Borkmann592867b2015-09-08 18:00:09 +0200652 f = fdget(ufd);
Daniel Borkmannc2101292015-10-29 14:58:07 +0100653 map = __bpf_map_get(f);
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700654 if (IS_ERR(map))
655 return PTR_ERR(map);
656
Al Viroe4448ed2017-05-13 18:43:00 -0400657 key = memdup_user(ukey, map->key_size);
658 if (IS_ERR(key)) {
659 err = PTR_ERR(key);
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700660 goto err_put;
Al Viroe4448ed2017-05-13 18:43:00 -0400661 }
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700662
Alexei Starovoitovb121d1e2016-03-07 21:57:13 -0800663 preempt_disable();
664 __this_cpu_inc(bpf_prog_active);
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700665 rcu_read_lock();
666 err = map->ops->map_delete_elem(map, key);
667 rcu_read_unlock();
Alexei Starovoitovb121d1e2016-03-07 21:57:13 -0800668 __this_cpu_dec(bpf_prog_active);
669 preempt_enable();
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700670
Daniel Borkmanna67edbf2017-01-25 02:28:18 +0100671 if (!err)
672 trace_bpf_map_delete_elem(map, ufd, key);
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700673 kfree(key);
674err_put:
675 fdput(f);
676 return err;
677}
678
679/* last field in 'union bpf_attr' used by this command */
680#define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
681
682static int map_get_next_key(union bpf_attr *attr)
683{
Mickaël Salaün535e7b4b2016-11-13 19:44:03 +0100684 void __user *ukey = u64_to_user_ptr(attr->key);
685 void __user *unext_key = u64_to_user_ptr(attr->next_key);
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700686 int ufd = attr->map_fd;
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700687 struct bpf_map *map;
688 void *key, *next_key;
Daniel Borkmann592867b2015-09-08 18:00:09 +0200689 struct fd f;
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700690 int err;
691
692 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
693 return -EINVAL;
694
Daniel Borkmann592867b2015-09-08 18:00:09 +0200695 f = fdget(ufd);
Daniel Borkmannc2101292015-10-29 14:58:07 +0100696 map = __bpf_map_get(f);
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700697 if (IS_ERR(map))
698 return PTR_ERR(map);
699
Teng Qin8fe45922017-04-24 19:00:37 -0700700 if (ukey) {
Al Viroe4448ed2017-05-13 18:43:00 -0400701 key = memdup_user(ukey, map->key_size);
702 if (IS_ERR(key)) {
703 err = PTR_ERR(key);
Teng Qin8fe45922017-04-24 19:00:37 -0700704 goto err_put;
Al Viroe4448ed2017-05-13 18:43:00 -0400705 }
Teng Qin8fe45922017-04-24 19:00:37 -0700706 } else {
707 key = NULL;
708 }
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700709
710 err = -ENOMEM;
711 next_key = kmalloc(map->key_size, GFP_USER);
712 if (!next_key)
713 goto free_key;
714
715 rcu_read_lock();
716 err = map->ops->map_get_next_key(map, key, next_key);
717 rcu_read_unlock();
718 if (err)
719 goto free_next_key;
720
721 err = -EFAULT;
722 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
723 goto free_next_key;
724
Daniel Borkmanna67edbf2017-01-25 02:28:18 +0100725 trace_bpf_map_next_key(map, ufd, key, next_key);
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -0700726 err = 0;
727
728free_next_key:
729 kfree(next_key);
730free_key:
731 kfree(key);
732err_put:
733 fdput(f);
734 return err;
735}
736
Johannes Bergbe9370a2017-04-11 15:34:57 +0200737static const struct bpf_verifier_ops * const bpf_prog_types[] = {
738#define BPF_PROG_TYPE(_id, _ops) \
739 [_id] = &_ops,
Johannes Berg40077e02017-04-11 15:34:58 +0200740#define BPF_MAP_TYPE(_id, _ops)
Johannes Bergbe9370a2017-04-11 15:34:57 +0200741#include <linux/bpf_types.h>
742#undef BPF_PROG_TYPE
Johannes Berg40077e02017-04-11 15:34:58 +0200743#undef BPF_MAP_TYPE
Johannes Bergbe9370a2017-04-11 15:34:57 +0200744};
Alexei Starovoitov09756af2014-09-26 00:17:00 -0700745
746static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
747{
Johannes Bergbe9370a2017-04-11 15:34:57 +0200748 if (type >= ARRAY_SIZE(bpf_prog_types) || !bpf_prog_types[type])
749 return -EINVAL;
Alexei Starovoitov09756af2014-09-26 00:17:00 -0700750
Johannes Bergbe9370a2017-04-11 15:34:57 +0200751 prog->aux->ops = bpf_prog_types[type];
752 prog->type = type;
753 return 0;
Alexei Starovoitov09756af2014-09-26 00:17:00 -0700754}
755
756/* drop refcnt on maps used by eBPF program and free auxilary data */
757static void free_used_maps(struct bpf_prog_aux *aux)
758{
759 int i;
760
761 for (i = 0; i < aux->used_map_cnt; i++)
762 bpf_map_put(aux->used_maps[i]);
763
764 kfree(aux->used_maps);
765}
766
Daniel Borkmann5ccb0712016-12-18 01:52:58 +0100767int __bpf_prog_charge(struct user_struct *user, u32 pages)
768{
769 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
770 unsigned long user_bufs;
771
772 if (user) {
773 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
774 if (user_bufs > memlock_limit) {
775 atomic_long_sub(pages, &user->locked_vm);
776 return -EPERM;
777 }
778 }
779
780 return 0;
781}
782
783void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
784{
785 if (user)
786 atomic_long_sub(pages, &user->locked_vm);
787}
788
Alexei Starovoitovaaac3ba2015-10-07 22:23:22 -0700789static int bpf_prog_charge_memlock(struct bpf_prog *prog)
790{
791 struct user_struct *user = get_current_user();
Daniel Borkmann5ccb0712016-12-18 01:52:58 +0100792 int ret;
Alexei Starovoitovaaac3ba2015-10-07 22:23:22 -0700793
Daniel Borkmann5ccb0712016-12-18 01:52:58 +0100794 ret = __bpf_prog_charge(user, prog->pages);
795 if (ret) {
Alexei Starovoitovaaac3ba2015-10-07 22:23:22 -0700796 free_uid(user);
Daniel Borkmann5ccb0712016-12-18 01:52:58 +0100797 return ret;
Alexei Starovoitovaaac3ba2015-10-07 22:23:22 -0700798 }
Daniel Borkmann5ccb0712016-12-18 01:52:58 +0100799
Alexei Starovoitovaaac3ba2015-10-07 22:23:22 -0700800 prog->aux->user = user;
801 return 0;
802}
803
804static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
805{
806 struct user_struct *user = prog->aux->user;
807
Daniel Borkmann5ccb0712016-12-18 01:52:58 +0100808 __bpf_prog_uncharge(user, prog->pages);
Alexei Starovoitovaaac3ba2015-10-07 22:23:22 -0700809 free_uid(user);
810}
811
Martin KaFai Laudc4bb0e2017-06-05 12:15:46 -0700812static int bpf_prog_alloc_id(struct bpf_prog *prog)
813{
814 int id;
815
816 spin_lock_bh(&prog_idr_lock);
817 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
818 if (id > 0)
819 prog->aux->id = id;
820 spin_unlock_bh(&prog_idr_lock);
821
822 /* id is in [1, INT_MAX) */
823 if (WARN_ON_ONCE(!id))
824 return -ENOSPC;
825
826 return id > 0 ? 0 : id;
827}
828
Martin KaFai Laub16d9aa2017-06-05 12:15:49 -0700829static void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
Martin KaFai Laudc4bb0e2017-06-05 12:15:46 -0700830{
831 /* cBPF to eBPF migrations are currently not in the idr store. */
832 if (!prog->aux->id)
833 return;
834
Martin KaFai Laub16d9aa2017-06-05 12:15:49 -0700835 if (do_idr_lock)
836 spin_lock_bh(&prog_idr_lock);
837 else
838 __acquire(&prog_idr_lock);
839
Martin KaFai Laudc4bb0e2017-06-05 12:15:46 -0700840 idr_remove(&prog_idr, prog->aux->id);
Martin KaFai Laub16d9aa2017-06-05 12:15:49 -0700841
842 if (do_idr_lock)
843 spin_unlock_bh(&prog_idr_lock);
844 else
845 __release(&prog_idr_lock);
Martin KaFai Laudc4bb0e2017-06-05 12:15:46 -0700846}
847
Daniel Borkmann1aacde32016-06-30 17:24:43 +0200848static void __bpf_prog_put_rcu(struct rcu_head *rcu)
Alexei Starovoitovabf2e7d2015-05-28 19:26:02 -0700849{
850 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
851
852 free_used_maps(aux);
Alexei Starovoitovaaac3ba2015-10-07 22:23:22 -0700853 bpf_prog_uncharge_memlock(aux->prog);
Alexei Starovoitovabf2e7d2015-05-28 19:26:02 -0700854 bpf_prog_free(aux->prog);
855}
856
Martin KaFai Laub16d9aa2017-06-05 12:15:49 -0700857static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
Alexei Starovoitov09756af2014-09-26 00:17:00 -0700858{
Daniel Borkmanna67edbf2017-01-25 02:28:18 +0100859 if (atomic_dec_and_test(&prog->aux->refcnt)) {
860 trace_bpf_prog_put_rcu(prog);
Martin KaFai Lau34ad5582017-06-05 12:15:48 -0700861 /* bpf_prog_free_id() must be called first */
Martin KaFai Laub16d9aa2017-06-05 12:15:49 -0700862 bpf_prog_free_id(prog, do_idr_lock);
Daniel Borkmann74451e662017-02-16 22:24:50 +0100863 bpf_prog_kallsyms_del(prog);
Daniel Borkmann1aacde32016-06-30 17:24:43 +0200864 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
Daniel Borkmanna67edbf2017-01-25 02:28:18 +0100865 }
Alexei Starovoitov09756af2014-09-26 00:17:00 -0700866}
Martin KaFai Laub16d9aa2017-06-05 12:15:49 -0700867
868void bpf_prog_put(struct bpf_prog *prog)
869{
870 __bpf_prog_put(prog, true);
871}
Daniel Borkmanne2e9b652015-03-01 12:31:48 +0100872EXPORT_SYMBOL_GPL(bpf_prog_put);
Alexei Starovoitov09756af2014-09-26 00:17:00 -0700873
874static int bpf_prog_release(struct inode *inode, struct file *filp)
875{
876 struct bpf_prog *prog = filp->private_data;
877
Daniel Borkmann1aacde32016-06-30 17:24:43 +0200878 bpf_prog_put(prog);
Alexei Starovoitov09756af2014-09-26 00:17:00 -0700879 return 0;
880}
881
Daniel Borkmann7bd509e2016-12-04 23:19:41 +0100882#ifdef CONFIG_PROC_FS
883static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
884{
885 const struct bpf_prog *prog = filp->private_data;
Daniel Borkmannf1f77142017-01-13 23:38:15 +0100886 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
Daniel Borkmann7bd509e2016-12-04 23:19:41 +0100887
Daniel Borkmannf1f77142017-01-13 23:38:15 +0100888 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
Daniel Borkmann7bd509e2016-12-04 23:19:41 +0100889 seq_printf(m,
890 "prog_type:\t%u\n"
891 "prog_jited:\t%u\n"
Daniel Borkmannf1f77142017-01-13 23:38:15 +0100892 "prog_tag:\t%s\n"
Daniel Borkmann7bd509e2016-12-04 23:19:41 +0100893 "memlock:\t%llu\n",
894 prog->type,
895 prog->jited,
Daniel Borkmannf1f77142017-01-13 23:38:15 +0100896 prog_tag,
Daniel Borkmann7bd509e2016-12-04 23:19:41 +0100897 prog->pages * 1ULL << PAGE_SHIFT);
898}
899#endif
900
Alexei Starovoitov09756af2014-09-26 00:17:00 -0700901static const struct file_operations bpf_prog_fops = {
Daniel Borkmann7bd509e2016-12-04 23:19:41 +0100902#ifdef CONFIG_PROC_FS
903 .show_fdinfo = bpf_prog_show_fdinfo,
904#endif
905 .release = bpf_prog_release,
Alexei Starovoitov09756af2014-09-26 00:17:00 -0700906};
907
Daniel Borkmannb2197752015-10-29 14:58:09 +0100908int bpf_prog_new_fd(struct bpf_prog *prog)
Daniel Borkmannaa797812015-10-29 14:58:06 +0100909{
910 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
911 O_RDWR | O_CLOEXEC);
912}
913
Daniel Borkmann113214b2016-06-30 17:24:44 +0200914static struct bpf_prog *____bpf_prog_get(struct fd f)
Alexei Starovoitov09756af2014-09-26 00:17:00 -0700915{
Alexei Starovoitov09756af2014-09-26 00:17:00 -0700916 if (!f.file)
917 return ERR_PTR(-EBADF);
Alexei Starovoitov09756af2014-09-26 00:17:00 -0700918 if (f.file->f_op != &bpf_prog_fops) {
919 fdput(f);
920 return ERR_PTR(-EINVAL);
921 }
922
Daniel Borkmannc2101292015-10-29 14:58:07 +0100923 return f.file->private_data;
Alexei Starovoitov09756af2014-09-26 00:17:00 -0700924}
925
Brenden Blanco59d36562016-07-19 12:16:46 -0700926struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
Alexei Starovoitov92117d82016-04-27 18:56:20 -0700927{
Brenden Blanco59d36562016-07-19 12:16:46 -0700928 if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
929 atomic_sub(i, &prog->aux->refcnt);
Alexei Starovoitov92117d82016-04-27 18:56:20 -0700930 return ERR_PTR(-EBUSY);
931 }
932 return prog;
933}
Brenden Blanco59d36562016-07-19 12:16:46 -0700934EXPORT_SYMBOL_GPL(bpf_prog_add);
935
Daniel Borkmannc5405942016-11-09 22:02:34 +0100936void bpf_prog_sub(struct bpf_prog *prog, int i)
937{
938 /* Only to be used for undoing previous bpf_prog_add() in some
939 * error path. We still know that another entity in our call
940 * path holds a reference to the program, thus atomic_sub() can
941 * be safely used in such cases!
942 */
943 WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
944}
945EXPORT_SYMBOL_GPL(bpf_prog_sub);
946
Brenden Blanco59d36562016-07-19 12:16:46 -0700947struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
948{
949 return bpf_prog_add(prog, 1);
950}
Daniel Borkmann97bc4022016-11-19 01:45:00 +0100951EXPORT_SYMBOL_GPL(bpf_prog_inc);
Alexei Starovoitov92117d82016-04-27 18:56:20 -0700952
Martin KaFai Laub16d9aa2017-06-05 12:15:49 -0700953/* prog_idr_lock should have been held */
John Fastabenda6f6df62017-08-15 22:32:22 -0700954struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
Martin KaFai Laub16d9aa2017-06-05 12:15:49 -0700955{
956 int refold;
957
958 refold = __atomic_add_unless(&prog->aux->refcnt, 1, 0);
959
960 if (refold >= BPF_MAX_REFCNT) {
961 __bpf_prog_put(prog, false);
962 return ERR_PTR(-EBUSY);
963 }
964
965 if (!refold)
966 return ERR_PTR(-ENOENT);
967
968 return prog;
969}
John Fastabenda6f6df62017-08-15 22:32:22 -0700970EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
Martin KaFai Laub16d9aa2017-06-05 12:15:49 -0700971
Daniel Borkmann113214b2016-06-30 17:24:44 +0200972static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *type)
Alexei Starovoitov09756af2014-09-26 00:17:00 -0700973{
974 struct fd f = fdget(ufd);
975 struct bpf_prog *prog;
976
Daniel Borkmann113214b2016-06-30 17:24:44 +0200977 prog = ____bpf_prog_get(f);
Alexei Starovoitov09756af2014-09-26 00:17:00 -0700978 if (IS_ERR(prog))
979 return prog;
Daniel Borkmann113214b2016-06-30 17:24:44 +0200980 if (type && prog->type != *type) {
981 prog = ERR_PTR(-EINVAL);
982 goto out;
983 }
Alexei Starovoitov09756af2014-09-26 00:17:00 -0700984
Alexei Starovoitov92117d82016-04-27 18:56:20 -0700985 prog = bpf_prog_inc(prog);
Daniel Borkmann113214b2016-06-30 17:24:44 +0200986out:
Alexei Starovoitov09756af2014-09-26 00:17:00 -0700987 fdput(f);
988 return prog;
989}
Daniel Borkmann113214b2016-06-30 17:24:44 +0200990
991struct bpf_prog *bpf_prog_get(u32 ufd)
992{
993 return __bpf_prog_get(ufd, NULL);
994}
995
996struct bpf_prog *bpf_prog_get_type(u32 ufd, enum bpf_prog_type type)
997{
Daniel Borkmanna67edbf2017-01-25 02:28:18 +0100998 struct bpf_prog *prog = __bpf_prog_get(ufd, &type);
999
1000 if (!IS_ERR(prog))
1001 trace_bpf_prog_get_type(prog);
1002 return prog;
Daniel Borkmann113214b2016-06-30 17:24:44 +02001003}
1004EXPORT_SYMBOL_GPL(bpf_prog_get_type);
Alexei Starovoitov09756af2014-09-26 00:17:00 -07001005
1006/* last field in 'union bpf_attr' used by this command */
Martin KaFai Laucb4d2b32017-09-27 14:37:52 -07001007#define BPF_PROG_LOAD_LAST_FIELD prog_name
Alexei Starovoitov09756af2014-09-26 00:17:00 -07001008
1009static int bpf_prog_load(union bpf_attr *attr)
1010{
1011 enum bpf_prog_type type = attr->prog_type;
1012 struct bpf_prog *prog;
1013 int err;
1014 char license[128];
1015 bool is_gpl;
1016
1017 if (CHECK_ATTR(BPF_PROG_LOAD))
1018 return -EINVAL;
1019
David S. Millere07b98d2017-05-10 11:38:07 -07001020 if (attr->prog_flags & ~BPF_F_STRICT_ALIGNMENT)
1021 return -EINVAL;
1022
Alexei Starovoitov09756af2014-09-26 00:17:00 -07001023 /* copy eBPF program license from user space */
Mickaël Salaün535e7b4b2016-11-13 19:44:03 +01001024 if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
Alexei Starovoitov09756af2014-09-26 00:17:00 -07001025 sizeof(license) - 1) < 0)
1026 return -EFAULT;
1027 license[sizeof(license) - 1] = 0;
1028
1029 /* eBPF programs must be GPL compatible to use GPL-ed functions */
1030 is_gpl = license_is_gpl_compatible(license);
1031
Daniel Borkmannef0915c2016-12-07 01:15:44 +01001032 if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
1033 return -E2BIG;
Alexei Starovoitov09756af2014-09-26 00:17:00 -07001034
Alexei Starovoitov25415172015-03-25 12:49:20 -07001035 if (type == BPF_PROG_TYPE_KPROBE &&
1036 attr->kern_version != LINUX_VERSION_CODE)
1037 return -EINVAL;
1038
Chenbo Feng80b7d812017-05-31 18:16:00 -07001039 if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1040 type != BPF_PROG_TYPE_CGROUP_SKB &&
1041 !capable(CAP_SYS_ADMIN))
Alexei Starovoitov1be7f752015-10-07 22:23:21 -07001042 return -EPERM;
1043
Alexei Starovoitov09756af2014-09-26 00:17:00 -07001044 /* plain bpf_prog allocation */
1045 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1046 if (!prog)
1047 return -ENOMEM;
1048
Alexei Starovoitovaaac3ba2015-10-07 22:23:22 -07001049 err = bpf_prog_charge_memlock(prog);
1050 if (err)
1051 goto free_prog_nouncharge;
1052
Alexei Starovoitov09756af2014-09-26 00:17:00 -07001053 prog->len = attr->insn_cnt;
1054
1055 err = -EFAULT;
Mickaël Salaün535e7b4b2016-11-13 19:44:03 +01001056 if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
Daniel Borkmannaafe6ae2016-12-18 01:52:57 +01001057 bpf_prog_insn_size(prog)) != 0)
Alexei Starovoitov09756af2014-09-26 00:17:00 -07001058 goto free_prog;
1059
1060 prog->orig_prog = NULL;
Daniel Borkmanna91263d2015-09-30 01:41:50 +02001061 prog->jited = 0;
Alexei Starovoitov09756af2014-09-26 00:17:00 -07001062
1063 atomic_set(&prog->aux->refcnt, 1);
Daniel Borkmanna91263d2015-09-30 01:41:50 +02001064 prog->gpl_compatible = is_gpl ? 1 : 0;
Alexei Starovoitov09756af2014-09-26 00:17:00 -07001065
1066 /* find program type: socket_filter vs tracing_filter */
1067 err = find_prog_type(type, prog);
1068 if (err < 0)
1069 goto free_prog;
1070
Martin KaFai Laucb4d2b32017-09-27 14:37:52 -07001071 prog->aux->load_time = ktime_get_boot_ns();
1072 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1073 if (err)
1074 goto free_prog;
1075
Alexei Starovoitov09756af2014-09-26 00:17:00 -07001076 /* run eBPF verifier */
Alexei Starovoitov9bac3d62015-03-13 11:57:42 -07001077 err = bpf_check(&prog, attr);
Alexei Starovoitov09756af2014-09-26 00:17:00 -07001078 if (err < 0)
1079 goto free_used_maps;
1080
1081 /* eBPF program is ready to be JITed */
Daniel Borkmannd1c55ab2016-05-13 19:08:31 +02001082 prog = bpf_prog_select_runtime(prog, &err);
Alexei Starovoitov04fd61ab2015-05-19 16:59:03 -07001083 if (err < 0)
1084 goto free_used_maps;
Alexei Starovoitov09756af2014-09-26 00:17:00 -07001085
Martin KaFai Laudc4bb0e2017-06-05 12:15:46 -07001086 err = bpf_prog_alloc_id(prog);
1087 if (err)
1088 goto free_used_maps;
1089
Daniel Borkmannaa797812015-10-29 14:58:06 +01001090 err = bpf_prog_new_fd(prog);
Martin KaFai Laub16d9aa2017-06-05 12:15:49 -07001091 if (err < 0) {
1092 /* failed to allocate fd.
1093 * bpf_prog_put() is needed because the above
1094 * bpf_prog_alloc_id() has published the prog
1095 * to the userspace and the userspace may
1096 * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1097 */
1098 bpf_prog_put(prog);
1099 return err;
1100 }
Alexei Starovoitov09756af2014-09-26 00:17:00 -07001101
Daniel Borkmann74451e662017-02-16 22:24:50 +01001102 bpf_prog_kallsyms_add(prog);
Daniel Borkmanna67edbf2017-01-25 02:28:18 +01001103 trace_bpf_prog_load(prog, err);
Alexei Starovoitov09756af2014-09-26 00:17:00 -07001104 return err;
1105
1106free_used_maps:
1107 free_used_maps(prog->aux);
1108free_prog:
Alexei Starovoitovaaac3ba2015-10-07 22:23:22 -07001109 bpf_prog_uncharge_memlock(prog);
1110free_prog_nouncharge:
Alexei Starovoitov09756af2014-09-26 00:17:00 -07001111 bpf_prog_free(prog);
1112 return err;
1113}
1114
Daniel Borkmannb2197752015-10-29 14:58:09 +01001115#define BPF_OBJ_LAST_FIELD bpf_fd
1116
1117static int bpf_obj_pin(const union bpf_attr *attr)
1118{
1119 if (CHECK_ATTR(BPF_OBJ))
1120 return -EINVAL;
1121
Mickaël Salaün535e7b4b2016-11-13 19:44:03 +01001122 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
Daniel Borkmannb2197752015-10-29 14:58:09 +01001123}
1124
1125static int bpf_obj_get(const union bpf_attr *attr)
1126{
1127 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0)
1128 return -EINVAL;
1129
Mickaël Salaün535e7b4b2016-11-13 19:44:03 +01001130 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname));
Daniel Borkmannb2197752015-10-29 14:58:09 +01001131}
1132
Daniel Mackf4324552016-11-23 16:52:27 +01001133#ifdef CONFIG_CGROUP_BPF
1134
John Fastabend464bc0f2017-08-28 07:10:04 -07001135#define BPF_PROG_ATTACH_LAST_FIELD attach_flags
John Fastabend174a79f2017-08-15 22:32:47 -07001136
John Fastabend5a67da22017-09-08 14:00:49 -07001137static int sockmap_get_from_fd(const union bpf_attr *attr, bool attach)
John Fastabend174a79f2017-08-15 22:32:47 -07001138{
John Fastabend5a67da22017-09-08 14:00:49 -07001139 struct bpf_prog *prog = NULL;
John Fastabend174a79f2017-08-15 22:32:47 -07001140 int ufd = attr->target_fd;
1141 struct bpf_map *map;
1142 struct fd f;
1143 int err;
1144
1145 f = fdget(ufd);
1146 map = __bpf_map_get(f);
1147 if (IS_ERR(map))
1148 return PTR_ERR(map);
1149
John Fastabend5a67da22017-09-08 14:00:49 -07001150 if (attach) {
1151 prog = bpf_prog_get_type(attr->attach_bpf_fd,
1152 BPF_PROG_TYPE_SK_SKB);
1153 if (IS_ERR(prog)) {
1154 fdput(f);
1155 return PTR_ERR(prog);
1156 }
John Fastabend174a79f2017-08-15 22:32:47 -07001157 }
1158
John Fastabend5a67da22017-09-08 14:00:49 -07001159 err = sock_map_prog(map, prog, attr->attach_type);
John Fastabend174a79f2017-08-15 22:32:47 -07001160 if (err) {
1161 fdput(f);
John Fastabend5a67da22017-09-08 14:00:49 -07001162 if (prog)
1163 bpf_prog_put(prog);
Dan Carpenterae2b27b2017-08-18 10:27:02 +03001164 return err;
John Fastabend174a79f2017-08-15 22:32:47 -07001165 }
1166
1167 fdput(f);
Dan Carpenterae2b27b2017-08-18 10:27:02 +03001168 return 0;
John Fastabend174a79f2017-08-15 22:32:47 -07001169}
Daniel Mackf4324552016-11-23 16:52:27 +01001170
Alexei Starovoitov324bda9e62017-10-02 22:50:21 -07001171#define BPF_F_ATTACH_MASK \
1172 (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1173
Daniel Mackf4324552016-11-23 16:52:27 +01001174static int bpf_prog_attach(const union bpf_attr *attr)
1175{
Alexei Starovoitov7f677632017-02-10 20:28:24 -08001176 enum bpf_prog_type ptype;
Daniel Mackf4324552016-11-23 16:52:27 +01001177 struct bpf_prog *prog;
1178 struct cgroup *cgrp;
Alexei Starovoitov7f677632017-02-10 20:28:24 -08001179 int ret;
Daniel Mackf4324552016-11-23 16:52:27 +01001180
1181 if (!capable(CAP_NET_ADMIN))
1182 return -EPERM;
1183
1184 if (CHECK_ATTR(BPF_PROG_ATTACH))
1185 return -EINVAL;
1186
Alexei Starovoitov324bda9e62017-10-02 22:50:21 -07001187 if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
Alexei Starovoitov7f677632017-02-10 20:28:24 -08001188 return -EINVAL;
1189
Daniel Mackf4324552016-11-23 16:52:27 +01001190 switch (attr->attach_type) {
1191 case BPF_CGROUP_INET_INGRESS:
1192 case BPF_CGROUP_INET_EGRESS:
David Ahernb2cd1252016-12-01 08:48:03 -08001193 ptype = BPF_PROG_TYPE_CGROUP_SKB;
Daniel Mackf4324552016-11-23 16:52:27 +01001194 break;
David Ahern610236582016-12-01 08:48:04 -08001195 case BPF_CGROUP_INET_SOCK_CREATE:
1196 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1197 break;
Lawrence Brakmo40304b22017-06-30 20:02:40 -07001198 case BPF_CGROUP_SOCK_OPS:
1199 ptype = BPF_PROG_TYPE_SOCK_OPS;
1200 break;
John Fastabend464bc0f2017-08-28 07:10:04 -07001201 case BPF_SK_SKB_STREAM_PARSER:
1202 case BPF_SK_SKB_STREAM_VERDICT:
John Fastabend5a67da22017-09-08 14:00:49 -07001203 return sockmap_get_from_fd(attr, true);
Daniel Mackf4324552016-11-23 16:52:27 +01001204 default:
1205 return -EINVAL;
1206 }
1207
David Ahernb2cd1252016-12-01 08:48:03 -08001208 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1209 if (IS_ERR(prog))
1210 return PTR_ERR(prog);
1211
1212 cgrp = cgroup_get_from_fd(attr->target_fd);
1213 if (IS_ERR(cgrp)) {
1214 bpf_prog_put(prog);
1215 return PTR_ERR(cgrp);
1216 }
1217
Alexei Starovoitov324bda9e62017-10-02 22:50:21 -07001218 ret = cgroup_bpf_attach(cgrp, prog, attr->attach_type,
1219 attr->attach_flags);
Alexei Starovoitov7f677632017-02-10 20:28:24 -08001220 if (ret)
1221 bpf_prog_put(prog);
David Ahernb2cd1252016-12-01 08:48:03 -08001222 cgroup_put(cgrp);
1223
Alexei Starovoitov7f677632017-02-10 20:28:24 -08001224 return ret;
Daniel Mackf4324552016-11-23 16:52:27 +01001225}
1226
1227#define BPF_PROG_DETACH_LAST_FIELD attach_type
1228
1229static int bpf_prog_detach(const union bpf_attr *attr)
1230{
Alexei Starovoitov324bda9e62017-10-02 22:50:21 -07001231 enum bpf_prog_type ptype;
1232 struct bpf_prog *prog;
Daniel Mackf4324552016-11-23 16:52:27 +01001233 struct cgroup *cgrp;
Alexei Starovoitov7f677632017-02-10 20:28:24 -08001234 int ret;
Daniel Mackf4324552016-11-23 16:52:27 +01001235
1236 if (!capable(CAP_NET_ADMIN))
1237 return -EPERM;
1238
1239 if (CHECK_ATTR(BPF_PROG_DETACH))
1240 return -EINVAL;
1241
1242 switch (attr->attach_type) {
1243 case BPF_CGROUP_INET_INGRESS:
1244 case BPF_CGROUP_INET_EGRESS:
Alexei Starovoitov324bda9e62017-10-02 22:50:21 -07001245 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1246 break;
David Ahern610236582016-12-01 08:48:04 -08001247 case BPF_CGROUP_INET_SOCK_CREATE:
Alexei Starovoitov324bda9e62017-10-02 22:50:21 -07001248 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1249 break;
Lawrence Brakmo40304b22017-06-30 20:02:40 -07001250 case BPF_CGROUP_SOCK_OPS:
Alexei Starovoitov324bda9e62017-10-02 22:50:21 -07001251 ptype = BPF_PROG_TYPE_SOCK_OPS;
Daniel Mackf4324552016-11-23 16:52:27 +01001252 break;
John Fastabend5a67da22017-09-08 14:00:49 -07001253 case BPF_SK_SKB_STREAM_PARSER:
1254 case BPF_SK_SKB_STREAM_VERDICT:
Alexei Starovoitov324bda9e62017-10-02 22:50:21 -07001255 return sockmap_get_from_fd(attr, false);
Daniel Mackf4324552016-11-23 16:52:27 +01001256 default:
1257 return -EINVAL;
1258 }
1259
Alexei Starovoitov324bda9e62017-10-02 22:50:21 -07001260 cgrp = cgroup_get_from_fd(attr->target_fd);
1261 if (IS_ERR(cgrp))
1262 return PTR_ERR(cgrp);
1263
1264 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1265 if (IS_ERR(prog))
1266 prog = NULL;
1267
1268 ret = cgroup_bpf_detach(cgrp, prog, attr->attach_type, 0);
1269 if (prog)
1270 bpf_prog_put(prog);
1271 cgroup_put(cgrp);
Alexei Starovoitov7f677632017-02-10 20:28:24 -08001272 return ret;
Daniel Mackf4324552016-11-23 16:52:27 +01001273}
Lawrence Brakmo40304b22017-06-30 20:02:40 -07001274
Alexei Starovoitov468e2f62017-10-02 22:50:22 -07001275#define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1276
1277static int bpf_prog_query(const union bpf_attr *attr,
1278 union bpf_attr __user *uattr)
1279{
1280 struct cgroup *cgrp;
1281 int ret;
1282
1283 if (!capable(CAP_NET_ADMIN))
1284 return -EPERM;
1285 if (CHECK_ATTR(BPF_PROG_QUERY))
1286 return -EINVAL;
1287 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
1288 return -EINVAL;
1289
1290 switch (attr->query.attach_type) {
1291 case BPF_CGROUP_INET_INGRESS:
1292 case BPF_CGROUP_INET_EGRESS:
1293 case BPF_CGROUP_INET_SOCK_CREATE:
1294 case BPF_CGROUP_SOCK_OPS:
1295 break;
1296 default:
1297 return -EINVAL;
1298 }
1299 cgrp = cgroup_get_from_fd(attr->query.target_fd);
1300 if (IS_ERR(cgrp))
1301 return PTR_ERR(cgrp);
1302 ret = cgroup_bpf_query(cgrp, attr, uattr);
1303 cgroup_put(cgrp);
1304 return ret;
1305}
Daniel Mackf4324552016-11-23 16:52:27 +01001306#endif /* CONFIG_CGROUP_BPF */
1307
Alexei Starovoitov1cf1cae2017-03-30 21:45:38 -07001308#define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
1309
1310static int bpf_prog_test_run(const union bpf_attr *attr,
1311 union bpf_attr __user *uattr)
1312{
1313 struct bpf_prog *prog;
1314 int ret = -ENOTSUPP;
1315
1316 if (CHECK_ATTR(BPF_PROG_TEST_RUN))
1317 return -EINVAL;
1318
1319 prog = bpf_prog_get(attr->test.prog_fd);
1320 if (IS_ERR(prog))
1321 return PTR_ERR(prog);
1322
1323 if (prog->aux->ops->test_run)
1324 ret = prog->aux->ops->test_run(prog, attr, uattr);
1325
1326 bpf_prog_put(prog);
1327 return ret;
1328}
1329
Martin KaFai Lau34ad5582017-06-05 12:15:48 -07001330#define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
1331
1332static int bpf_obj_get_next_id(const union bpf_attr *attr,
1333 union bpf_attr __user *uattr,
1334 struct idr *idr,
1335 spinlock_t *lock)
1336{
1337 u32 next_id = attr->start_id;
1338 int err = 0;
1339
1340 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
1341 return -EINVAL;
1342
1343 if (!capable(CAP_SYS_ADMIN))
1344 return -EPERM;
1345
1346 next_id++;
1347 spin_lock_bh(lock);
1348 if (!idr_get_next(idr, &next_id))
1349 err = -ENOENT;
1350 spin_unlock_bh(lock);
1351
1352 if (!err)
1353 err = put_user(next_id, &uattr->next_id);
1354
1355 return err;
1356}
1357
Martin KaFai Laub16d9aa2017-06-05 12:15:49 -07001358#define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
1359
1360static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
1361{
1362 struct bpf_prog *prog;
1363 u32 id = attr->prog_id;
1364 int fd;
1365
1366 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
1367 return -EINVAL;
1368
1369 if (!capable(CAP_SYS_ADMIN))
1370 return -EPERM;
1371
1372 spin_lock_bh(&prog_idr_lock);
1373 prog = idr_find(&prog_idr, id);
1374 if (prog)
1375 prog = bpf_prog_inc_not_zero(prog);
1376 else
1377 prog = ERR_PTR(-ENOENT);
1378 spin_unlock_bh(&prog_idr_lock);
1379
1380 if (IS_ERR(prog))
1381 return PTR_ERR(prog);
1382
1383 fd = bpf_prog_new_fd(prog);
1384 if (fd < 0)
1385 bpf_prog_put(prog);
1386
1387 return fd;
1388}
1389
Martin KaFai Laubd5f5f4e2017-06-05 12:15:50 -07001390#define BPF_MAP_GET_FD_BY_ID_LAST_FIELD map_id
1391
1392static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
1393{
1394 struct bpf_map *map;
1395 u32 id = attr->map_id;
1396 int fd;
1397
1398 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID))
1399 return -EINVAL;
1400
1401 if (!capable(CAP_SYS_ADMIN))
1402 return -EPERM;
1403
1404 spin_lock_bh(&map_idr_lock);
1405 map = idr_find(&map_idr, id);
1406 if (map)
1407 map = bpf_map_inc_not_zero(map, true);
1408 else
1409 map = ERR_PTR(-ENOENT);
1410 spin_unlock_bh(&map_idr_lock);
1411
1412 if (IS_ERR(map))
1413 return PTR_ERR(map);
1414
1415 fd = bpf_map_new_fd(map);
1416 if (fd < 0)
1417 bpf_map_put(map);
1418
1419 return fd;
1420}
1421
Martin KaFai Lau1e270972017-06-05 12:15:52 -07001422static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
1423 const union bpf_attr *attr,
1424 union bpf_attr __user *uattr)
1425{
1426 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1427 struct bpf_prog_info info = {};
1428 u32 info_len = attr->info.info_len;
1429 char __user *uinsns;
1430 u32 ulen;
1431 int err;
1432
1433 err = check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1434 if (err)
1435 return err;
1436 info_len = min_t(u32, sizeof(info), info_len);
1437
1438 if (copy_from_user(&info, uinfo, info_len))
Daniel Borkmann89b09682017-07-27 21:02:46 +02001439 return -EFAULT;
Martin KaFai Lau1e270972017-06-05 12:15:52 -07001440
1441 info.type = prog->type;
1442 info.id = prog->aux->id;
Martin KaFai Laucb4d2b32017-09-27 14:37:52 -07001443 info.load_time = prog->aux->load_time;
1444 info.created_by_uid = from_kuid_munged(current_user_ns(),
1445 prog->aux->user->uid);
Martin KaFai Lau1e270972017-06-05 12:15:52 -07001446
1447 memcpy(info.tag, prog->tag, sizeof(prog->tag));
Martin KaFai Laucb4d2b32017-09-27 14:37:52 -07001448 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
1449
1450 ulen = info.nr_map_ids;
1451 info.nr_map_ids = prog->aux->used_map_cnt;
1452 ulen = min_t(u32, info.nr_map_ids, ulen);
1453 if (ulen) {
Martin KaFai Lau721e08d2017-09-29 10:52:17 -07001454 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
Martin KaFai Laucb4d2b32017-09-27 14:37:52 -07001455 u32 i;
1456
1457 for (i = 0; i < ulen; i++)
1458 if (put_user(prog->aux->used_maps[i]->id,
1459 &user_map_ids[i]))
1460 return -EFAULT;
1461 }
Martin KaFai Lau1e270972017-06-05 12:15:52 -07001462
1463 if (!capable(CAP_SYS_ADMIN)) {
1464 info.jited_prog_len = 0;
1465 info.xlated_prog_len = 0;
1466 goto done;
1467 }
1468
1469 ulen = info.jited_prog_len;
1470 info.jited_prog_len = prog->jited_len;
1471 if (info.jited_prog_len && ulen) {
1472 uinsns = u64_to_user_ptr(info.jited_prog_insns);
1473 ulen = min_t(u32, info.jited_prog_len, ulen);
1474 if (copy_to_user(uinsns, prog->bpf_func, ulen))
1475 return -EFAULT;
1476 }
1477
1478 ulen = info.xlated_prog_len;
Daniel Borkmann9975a542017-07-28 17:05:25 +02001479 info.xlated_prog_len = bpf_prog_insn_size(prog);
Martin KaFai Lau1e270972017-06-05 12:15:52 -07001480 if (info.xlated_prog_len && ulen) {
1481 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
1482 ulen = min_t(u32, info.xlated_prog_len, ulen);
1483 if (copy_to_user(uinsns, prog->insnsi, ulen))
1484 return -EFAULT;
1485 }
1486
1487done:
1488 if (copy_to_user(uinfo, &info, info_len) ||
1489 put_user(info_len, &uattr->info.info_len))
1490 return -EFAULT;
1491
1492 return 0;
1493}
1494
1495static int bpf_map_get_info_by_fd(struct bpf_map *map,
1496 const union bpf_attr *attr,
1497 union bpf_attr __user *uattr)
1498{
1499 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1500 struct bpf_map_info info = {};
1501 u32 info_len = attr->info.info_len;
1502 int err;
1503
1504 err = check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1505 if (err)
1506 return err;
1507 info_len = min_t(u32, sizeof(info), info_len);
1508
1509 info.type = map->map_type;
1510 info.id = map->id;
1511 info.key_size = map->key_size;
1512 info.value_size = map->value_size;
1513 info.max_entries = map->max_entries;
1514 info.map_flags = map->map_flags;
Martin KaFai Lauad5b1772017-09-27 14:37:53 -07001515 memcpy(info.name, map->name, sizeof(map->name));
Martin KaFai Lau1e270972017-06-05 12:15:52 -07001516
1517 if (copy_to_user(uinfo, &info, info_len) ||
1518 put_user(info_len, &uattr->info.info_len))
1519 return -EFAULT;
1520
1521 return 0;
1522}
1523
1524#define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
1525
1526static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
1527 union bpf_attr __user *uattr)
1528{
1529 int ufd = attr->info.bpf_fd;
1530 struct fd f;
1531 int err;
1532
1533 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
1534 return -EINVAL;
1535
1536 f = fdget(ufd);
1537 if (!f.file)
1538 return -EBADFD;
1539
1540 if (f.file->f_op == &bpf_prog_fops)
1541 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
1542 uattr);
1543 else if (f.file->f_op == &bpf_map_fops)
1544 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
1545 uattr);
1546 else
1547 err = -EINVAL;
1548
1549 fdput(f);
1550 return err;
1551}
1552
Alexei Starovoitov99c55f72014-09-26 00:16:57 -07001553SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
1554{
1555 union bpf_attr attr = {};
1556 int err;
1557
Alexei Starovoitov1be7f752015-10-07 22:23:21 -07001558 if (!capable(CAP_SYS_ADMIN) && sysctl_unprivileged_bpf_disabled)
Alexei Starovoitov99c55f72014-09-26 00:16:57 -07001559 return -EPERM;
1560
Martin KaFai Lau1e270972017-06-05 12:15:52 -07001561 err = check_uarg_tail_zero(uattr, sizeof(attr), size);
1562 if (err)
1563 return err;
1564 size = min_t(u32, size, sizeof(attr));
Alexei Starovoitov99c55f72014-09-26 00:16:57 -07001565
1566 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
1567 if (copy_from_user(&attr, uattr, size) != 0)
1568 return -EFAULT;
1569
1570 switch (cmd) {
1571 case BPF_MAP_CREATE:
1572 err = map_create(&attr);
1573 break;
Alexei Starovoitovdb20fd22014-09-26 00:16:59 -07001574 case BPF_MAP_LOOKUP_ELEM:
1575 err = map_lookup_elem(&attr);
1576 break;
1577 case BPF_MAP_UPDATE_ELEM:
1578 err = map_update_elem(&attr);
1579 break;
1580 case BPF_MAP_DELETE_ELEM:
1581 err = map_delete_elem(&attr);
1582 break;
1583 case BPF_MAP_GET_NEXT_KEY:
1584 err = map_get_next_key(&attr);
1585 break;
Alexei Starovoitov09756af2014-09-26 00:17:00 -07001586 case BPF_PROG_LOAD:
1587 err = bpf_prog_load(&attr);
1588 break;
Daniel Borkmannb2197752015-10-29 14:58:09 +01001589 case BPF_OBJ_PIN:
1590 err = bpf_obj_pin(&attr);
1591 break;
1592 case BPF_OBJ_GET:
1593 err = bpf_obj_get(&attr);
1594 break;
Daniel Mackf4324552016-11-23 16:52:27 +01001595#ifdef CONFIG_CGROUP_BPF
1596 case BPF_PROG_ATTACH:
1597 err = bpf_prog_attach(&attr);
1598 break;
1599 case BPF_PROG_DETACH:
1600 err = bpf_prog_detach(&attr);
1601 break;
Alexei Starovoitov468e2f62017-10-02 22:50:22 -07001602 case BPF_PROG_QUERY:
1603 err = bpf_prog_query(&attr, uattr);
1604 break;
Daniel Mackf4324552016-11-23 16:52:27 +01001605#endif
Alexei Starovoitov1cf1cae2017-03-30 21:45:38 -07001606 case BPF_PROG_TEST_RUN:
1607 err = bpf_prog_test_run(&attr, uattr);
1608 break;
Martin KaFai Lau34ad5582017-06-05 12:15:48 -07001609 case BPF_PROG_GET_NEXT_ID:
1610 err = bpf_obj_get_next_id(&attr, uattr,
1611 &prog_idr, &prog_idr_lock);
1612 break;
1613 case BPF_MAP_GET_NEXT_ID:
1614 err = bpf_obj_get_next_id(&attr, uattr,
1615 &map_idr, &map_idr_lock);
1616 break;
Martin KaFai Laub16d9aa2017-06-05 12:15:49 -07001617 case BPF_PROG_GET_FD_BY_ID:
1618 err = bpf_prog_get_fd_by_id(&attr);
1619 break;
Martin KaFai Laubd5f5f4e2017-06-05 12:15:50 -07001620 case BPF_MAP_GET_FD_BY_ID:
1621 err = bpf_map_get_fd_by_id(&attr);
1622 break;
Martin KaFai Lau1e270972017-06-05 12:15:52 -07001623 case BPF_OBJ_GET_INFO_BY_FD:
1624 err = bpf_obj_get_info_by_fd(&attr, uattr);
1625 break;
Alexei Starovoitov99c55f72014-09-26 00:16:57 -07001626 default:
1627 err = -EINVAL;
1628 break;
1629 }
1630
1631 return err;
1632}