projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
net/tls: Use RCU API to access tls_ctx->netdev
[linux-block.git] / kernel / bpf / syscall.c
CommitLineData
5b497af4 1// SPDX-License-Identifier: GPL-2.0-only
99c55f7d 2/* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
99c55f7d
AS		 3 */
4#include <linux/bpf.h>
aef2feda 5#include <linux/bpf-cgroup.h>
a67edbf4 6#include <linux/bpf_trace.h>
f4364dcf 7#include <linux/bpf_lirc.h>
4a1e7c0c 8#include <linux/bpf_verifier.h>
61df10c7 9#include <linux/bsearch.h>
f56a653c 10#include <linux/btf.h>
99c55f7d
AS		 11#include <linux/syscalls.h>
12#include <linux/slab.h>
3f07c014 13#include <linux/sched/signal.h>
d407bd25
DB		 14#include <linux/vmalloc.h>
15#include <linux/mmzone.h>
99c55f7d 16#include <linux/anon_inodes.h>
41bdc4b4 17#include <linux/fdtable.h>
db20fd2b 18#include <linux/file.h>
41bdc4b4 19#include <linux/fs.h>
09756af4
AS		 20#include <linux/license.h>
21#include <linux/filter.h>
535e7b4b 22#include <linux/kernel.h>
dc4bb0e2 23#include <linux/idr.h>
cb4d2b3f
MKL		 24#include <linux/cred.h>
25#include <linux/timekeeping.h>
26#include <linux/ctype.h>
9ef09e35 27#include <linux/nospec.h>
bae141f5 28#include <linux/audit.h>
ccfe29eb 29#include <uapi/linux/btf.h>
ca5999fd 30#include <linux/pgtable.h>
9e4e01df 31#include <linux/bpf_lsm.h>
457f4436 32#include <linux/poll.h>
4d7d7f69 33#include <linux/sort.h>
a3fd7cee 34#include <linux/bpf-netns.h>
1e6c62a8 35#include <linux/rcupdate_trace.h>
48edc1f7 36#include <linux/memcontrol.h>
0dcac272 37#include <linux/trace_events.h>
99c55f7d 38
da765a2f
DB		 39#define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
40 (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
41 (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
42#define IS_FD_PROG_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY)
14dc6f04 43#define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
da765a2f
DB		 44#define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map) || \
45 IS_FD_HASH(map))
14dc6f04 46
6e71b04a
CF		 47#define BPF_OBJ_FLAG_MASK (BPF_F_RDONLY | BPF_F_WRONLY)
48
b121d1e7 49DEFINE_PER_CPU(int, bpf_prog_active);
dc4bb0e2
MKL		 50static DEFINE_IDR(prog_idr);
51static DEFINE_SPINLOCK(prog_idr_lock);
f3f1c054
MKL		 52static DEFINE_IDR(map_idr);
53static DEFINE_SPINLOCK(map_idr_lock);
a3b80e10
AN		 54static DEFINE_IDR(link_idr);
55static DEFINE_SPINLOCK(link_idr_lock);
b121d1e7 56
08389d88
DB		 57int sysctl_unprivileged_bpf_disabled __read_mostly =
58 IS_BUILTIN(CONFIG_BPF_UNPRIV_DEFAULT_OFF) ? 2 : 0;
1be7f75d 59
40077e0c 60static const struct bpf_map_ops * const bpf_map_types[] = {
91cc1a99 61#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
40077e0c
JB		 62#define BPF_MAP_TYPE(_id, _ops) \
63 [_id] = &_ops,
f2e10bff 64#define BPF_LINK_TYPE(_id, _name)
40077e0c
JB		 65#include <linux/bpf_types.h>
66#undef BPF_PROG_TYPE
67#undef BPF_MAP_TYPE
f2e10bff 68#undef BPF_LINK_TYPE
40077e0c 69};
99c55f7d 70
752ba56f
MS		 71/*
72 * If we're handed a bigger struct than we know of, ensure all the unknown bits
73 * are 0 - i.e. new user-space does not rely on any kernel feature extensions
74 * we don't know about yet.
75 *
76 * There is a ToCToU between this function call and the following
77 * copy_from_user() call. However, this is not a concern since this function is
78 * meant to be a future-proofing of bits.
79 */
af2ac3e1 80int bpf_check_uarg_tail_zero(bpfptr_t uaddr,
dcab51f1
MKL		 81 size_t expected_size,
82 size_t actual_size)
58291a74 83{
b7e4b65f 84 int res;
58291a74 85
752ba56f
MS		 86 if (unlikely(actual_size > PAGE_SIZE)) /* silly large */
87 return -E2BIG;
88
58291a74
MS		 89 if (actual_size <= expected_size)
90 return 0;
91
af2ac3e1
AS		 92 if (uaddr.is_kernel)
93 res = memchr_inv(uaddr.kernel + expected_size, 0,
94 actual_size - expected_size) == NULL;
95 else
96 res = check_zeroed_user(uaddr.user + expected_size,
97 actual_size - expected_size);
b7e4b65f
AV		 98 if (res < 0)
99 return res;
100 return res ? 0 : -E2BIG;
58291a74
MS		 101}
102
a3884572 103const struct bpf_map_ops bpf_map_offload_ops = {
f4d05259 104 .map_meta_equal = bpf_map_meta_equal,
a3884572
JK		 105 .map_alloc = bpf_map_offload_map_alloc,
106 .map_free = bpf_map_offload_map_free,
e8d2bec0 107 .map_check_btf = map_check_no_btf,
a3884572
JK		 108};
109
99c55f7d
AS		 110static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
111{
1110f3a9 112 const struct bpf_map_ops *ops;
9ef09e35 113 u32 type = attr->map_type;
99c55f7d 114 struct bpf_map *map;
1110f3a9 115 int err;
99c55f7d 116
9ef09e35 117 if (type >= ARRAY_SIZE(bpf_map_types))
1110f3a9 118 return ERR_PTR(-EINVAL);
9ef09e35
MR		 119 type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
120 ops = bpf_map_types[type];
1110f3a9 121 if (!ops)
40077e0c 122 return ERR_PTR(-EINVAL);
99c55f7d 123
1110f3a9
JK		 124 if (ops->map_alloc_check) {
125 err = ops->map_alloc_check(attr);
126 if (err)
127 return ERR_PTR(err);
128 }
a3884572
JK		 129 if (attr->map_ifindex)
130 ops = &bpf_map_offload_ops;
1110f3a9 131 map = ops->map_alloc(attr);
40077e0c
JB		 132 if (IS_ERR(map))
133 return map;
1110f3a9 134 map->ops = ops;
9ef09e35 135 map->map_type = type;
40077e0c 136 return map;
99c55f7d
AS		 137}
138
353050be
DB		 139static void bpf_map_write_active_inc(struct bpf_map *map)
140{
141 atomic64_inc(&map->writecnt);
142}
143
144static void bpf_map_write_active_dec(struct bpf_map *map)
145{
146 atomic64_dec(&map->writecnt);
147}
148
149bool bpf_map_write_active(const struct bpf_map *map)
150{
151 return atomic64_read(&map->writecnt) != 0;
152}
153
80ee81e0 154static u32 bpf_map_value_size(const struct bpf_map *map)
15c14a3d
BV		 155{
156 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
157 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
158 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
159 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
160 return round_up(map->value_size, 8) * num_possible_cpus();
161 else if (IS_FD_MAP(map))
162 return sizeof(u32);
163 else
164 return map->value_size;
165}
166
167static void maybe_wait_bpf_programs(struct bpf_map *map)
168{
169 /* Wait for any running BPF programs to complete so that
170 * userspace, when we return to it, knows that all programs
171 * that could be running use the new map value.
172 */
173 if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
174 map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
175 synchronize_rcu();
176}
177
178static int bpf_map_update_value(struct bpf_map *map, struct fd f, void *key,
179 void *value, __u64 flags)
180{
181 int err;
182
183 /* Need to create a kthread, thus must support schedule */
184 if (bpf_map_is_dev_bound(map)) {
185 return bpf_map_offload_update_elem(map, key, value, flags);
186 } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
15c14a3d
BV		 187 map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
188 return map->ops->map_update_elem(map, key, value, flags);
13b79d3f
LB		 189 } else if (map->map_type == BPF_MAP_TYPE_SOCKHASH ||
190 map->map_type == BPF_MAP_TYPE_SOCKMAP) {
191 return sock_map_update_elem_sys(map, key, value, flags);
15c14a3d
BV		 192 } else if (IS_FD_PROG_ARRAY(map)) {
193 return bpf_fd_array_map_update_elem(map, f.file, key, value,
194 flags);
195 }
196
b6e5dae1 197 bpf_disable_instrumentation();
15c14a3d
BV		 198 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
199 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
200 err = bpf_percpu_hash_update(map, key, value, flags);
201 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
202 err = bpf_percpu_array_update(map, key, value, flags);
203 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
204 err = bpf_percpu_cgroup_storage_update(map, key, value,
205 flags);
206 } else if (IS_FD_ARRAY(map)) {
207 rcu_read_lock();
208 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
209 flags);
210 rcu_read_unlock();
211 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
212 rcu_read_lock();
213 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
214 flags);
215 rcu_read_unlock();
216 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
217 /* rcu_read_lock() is not needed */
218 err = bpf_fd_reuseport_array_update_elem(map, key, value,
219 flags);
220 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
9330986c
JK		 221 map->map_type == BPF_MAP_TYPE_STACK ||
222 map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
15c14a3d
BV		 223 err = map->ops->map_push_elem(map, value, flags);
224 } else {
225 rcu_read_lock();
226 err = map->ops->map_update_elem(map, key, value, flags);
227 rcu_read_unlock();
228 }
b6e5dae1 229 bpf_enable_instrumentation();
15c14a3d
BV		 230 maybe_wait_bpf_programs(map);
231
232 return err;
233}
234
235static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
236 __u64 flags)
237{
238 void *ptr;
239 int err;
240
cb4d03ab
BV		 241 if (bpf_map_is_dev_bound(map))
242 return bpf_map_offload_lookup_elem(map, key, value);
15c14a3d 243
b6e5dae1 244 bpf_disable_instrumentation();
15c14a3d
BV		 245 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
246 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
247 err = bpf_percpu_hash_copy(map, key, value);
248 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
249 err = bpf_percpu_array_copy(map, key, value);
250 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
251 err = bpf_percpu_cgroup_storage_copy(map, key, value);
252 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
253 err = bpf_stackmap_copy(map, key, value);
254 } else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
255 err = bpf_fd_array_map_lookup_elem(map, key, value);
256 } else if (IS_FD_HASH(map)) {
257 err = bpf_fd_htab_map_lookup_elem(map, key, value);
258 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
259 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
260 } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
9330986c
JK		 261 map->map_type == BPF_MAP_TYPE_STACK ||
262 map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
15c14a3d
BV		 263 err = map->ops->map_peek_elem(map, value);
264 } else if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
265 /* struct_ops map requires directly updating "value" */
266 err = bpf_struct_ops_map_sys_lookup_elem(map, key, value);
267 } else {
268 rcu_read_lock();
269 if (map->ops->map_lookup_elem_sys_only)
270 ptr = map->ops->map_lookup_elem_sys_only(map, key);
271 else
272 ptr = map->ops->map_lookup_elem(map, key);
273 if (IS_ERR(ptr)) {
274 err = PTR_ERR(ptr);
275 } else if (!ptr) {
276 err = -ENOENT;
277 } else {
278 err = 0;
279 if (flags & BPF_F_LOCK)
280 /* lock 'ptr' and copy everything but lock */
281 copy_map_value_locked(map, value, ptr, true);
282 else
283 copy_map_value(map, value, ptr);
68134668
AS		 284 /* mask lock and timer, since value wasn't zero inited */
285 check_and_init_map_value(map, value);
15c14a3d
BV		 286 }
287 rcu_read_unlock();
288 }
289
b6e5dae1 290 bpf_enable_instrumentation();
15c14a3d
BV		 291 maybe_wait_bpf_programs(map);
292
293 return err;
294}
295
d5299b67
RG		 296/* Please, do not use this function outside from the map creation path
297 * (e.g. in map update path) without taking care of setting the active
298 * memory cgroup (see at bpf_map_kmalloc_node() for example).
299 */
196e8ca7 300static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
d407bd25 301{
f01a7dbe
MP		 302 /* We really just want to fail instead of triggering OOM killer
303 * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
304 * which is used for lower order allocation requests.
305 *
306 * It has been observed that higher order allocation requests done by
307 * vmalloc with __GFP_NORETRY being set might fail due to not trying
308 * to reclaim memory from the page cache, thus we set
309 * __GFP_RETRY_MAYFAIL to avoid such situations.
d407bd25 310 */
f01a7dbe 311
d5299b67 312 const gfp_t gfp = __GFP_NOWARN | __GFP_ZERO | __GFP_ACCOUNT;
041de93f
CH		 313 unsigned int flags = 0;
314 unsigned long align = 1;
d407bd25
DB		 315 void *area;
316
196e8ca7
DB		 317 if (size >= SIZE_MAX)
318 return NULL;
319
fc970227 320 /* kmalloc()'ed memory can't be mmap()'ed */
041de93f
CH		 321 if (mmapable) {
322 BUG_ON(!PAGE_ALIGNED(size));
323 align = SHMLBA;
324 flags = VM_USERMAP;
325 } else if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
326 area = kmalloc_node(size, gfp | GFP_USER | __GFP_NORETRY,
f01a7dbe 327 numa_node);
d407bd25
DB		 328 if (area != NULL)
329 return area;
330 }
041de93f
CH		 331
332 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
333 gfp | GFP_KERNEL | __GFP_RETRY_MAYFAIL, PAGE_KERNEL,
334 flags, numa_node, __builtin_return_address(0));
d407bd25
DB		 335}
336
196e8ca7 337void *bpf_map_area_alloc(u64 size, int numa_node)
fc970227
AN		 338{
339 return __bpf_map_area_alloc(size, numa_node, false);
340}
341
196e8ca7 342void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
fc970227
AN		 343{
344 return __bpf_map_area_alloc(size, numa_node, true);
345}
346
d407bd25
DB		 347void bpf_map_area_free(void *area)
348{
349 kvfree(area);
350}
351
be70bcd5
DB		 352static u32 bpf_map_flags_retain_permanent(u32 flags)
353{
354 /* Some map creation flags are not tied to the map object but
355 * rather to the map fd instead, so they have no meaning upon
356 * map object inspection since multiple file descriptors with
357 * different (access) properties can exist here. Thus, given
358 * this has zero meaning for the map itself, lets clear these
359 * from here.
360 */
361 return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
362}
363
bd475643
JK		 364void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
365{
366 map->map_type = attr->map_type;
367 map->key_size = attr->key_size;
368 map->value_size = attr->value_size;
369 map->max_entries = attr->max_entries;
be70bcd5 370 map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
bd475643 371 map->numa_node = bpf_map_attr_numa_node(attr);
9330986c 372 map->map_extra = attr->map_extra;
bd475643
JK		 373}
374
f3f1c054
MKL		 375static int bpf_map_alloc_id(struct bpf_map *map)
376{
377 int id;
378
b76354cd 379 idr_preload(GFP_KERNEL);
f3f1c054
MKL		 380 spin_lock_bh(&map_idr_lock);
381 id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
382 if (id > 0)
383 map->id = id;
384 spin_unlock_bh(&map_idr_lock);
b76354cd 385 idr_preload_end();
f3f1c054
MKL		 386
387 if (WARN_ON_ONCE(!id))
388 return -ENOSPC;
389
390 return id > 0 ? 0 : id;
391}
392
a3884572 393void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
f3f1c054 394{
930651a7
ED		 395 unsigned long flags;
396
a3884572
JK		 397 /* Offloaded maps are removed from the IDR store when their device
398 * disappears - even if someone holds an fd to them they are unusable,
399 * the memory is gone, all ops will fail; they are simply waiting for
400 * refcnt to drop to be freed.
401 */
402 if (!map->id)
403 return;
404
bd5f5f4e 405 if (do_idr_lock)
930651a7 406 spin_lock_irqsave(&map_idr_lock, flags);
bd5f5f4e
MKL		 407 else
408 __acquire(&map_idr_lock);
409
f3f1c054 410 idr_remove(&map_idr, map->id);
a3884572 411 map->id = 0;
bd5f5f4e
MKL		 412
413 if (do_idr_lock)
930651a7 414 spin_unlock_irqrestore(&map_idr_lock, flags);
bd5f5f4e
MKL		 415 else
416 __release(&map_idr_lock);
f3f1c054
MKL		 417}
418
48edc1f7
RG		 419#ifdef CONFIG_MEMCG_KMEM
420static void bpf_map_save_memcg(struct bpf_map *map)
421{
4201d9ab
RG		 422 /* Currently if a map is created by a process belonging to the root
423 * memory cgroup, get_obj_cgroup_from_current() will return NULL.
424 * So we have to check map->objcg for being NULL each time it's
425 * being used.
426 */
427 map->objcg = get_obj_cgroup_from_current();
48edc1f7
RG		 428}
429
430static void bpf_map_release_memcg(struct bpf_map *map)
431{
4201d9ab
RG		 432 if (map->objcg)
433 obj_cgroup_put(map->objcg);
434}
435
436static struct mem_cgroup *bpf_map_get_memcg(const struct bpf_map *map)
437{
438 if (map->objcg)
439 return get_mem_cgroup_from_objcg(map->objcg);
440
441 return root_mem_cgroup;
48edc1f7
RG		 442}
443
444void *bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags,
445 int node)
446{
4201d9ab 447 struct mem_cgroup *memcg, *old_memcg;
48edc1f7
RG		 448 void *ptr;
449
4201d9ab
RG		 450 memcg = bpf_map_get_memcg(map);
451 old_memcg = set_active_memcg(memcg);
48edc1f7
RG		 452 ptr = kmalloc_node(size, flags | __GFP_ACCOUNT, node);
453 set_active_memcg(old_memcg);
4201d9ab 454 mem_cgroup_put(memcg);
48edc1f7
RG		 455
456 return ptr;
457}
458
459void *bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags)
460{
4201d9ab 461 struct mem_cgroup *memcg, *old_memcg;
48edc1f7
RG		 462 void *ptr;
463
4201d9ab
RG		 464 memcg = bpf_map_get_memcg(map);
465 old_memcg = set_active_memcg(memcg);
48edc1f7
RG		 466 ptr = kzalloc(size, flags | __GFP_ACCOUNT);
467 set_active_memcg(old_memcg);
4201d9ab 468 mem_cgroup_put(memcg);
48edc1f7
RG		 469
470 return ptr;
471}
472
473void __percpu *bpf_map_alloc_percpu(const struct bpf_map *map, size_t size,
474 size_t align, gfp_t flags)
475{
4201d9ab 476 struct mem_cgroup *memcg, *old_memcg;
48edc1f7
RG		 477 void __percpu *ptr;
478
4201d9ab
RG		 479 memcg = bpf_map_get_memcg(map);
480 old_memcg = set_active_memcg(memcg);
48edc1f7
RG		 481 ptr = __alloc_percpu_gfp(size, align, flags | __GFP_ACCOUNT);
482 set_active_memcg(old_memcg);
4201d9ab 483 mem_cgroup_put(memcg);
48edc1f7
RG		 484
485 return ptr;
486}
487
488#else
489static void bpf_map_save_memcg(struct bpf_map *map)
490{
491}
492
493static void bpf_map_release_memcg(struct bpf_map *map)
494{
495}
496#endif
497
61df10c7
KKD		 498static int bpf_map_kptr_off_cmp(const void *a, const void *b)
499{
500 const struct bpf_map_value_off_desc *off_desc1 = a, *off_desc2 = b;
501
502 if (off_desc1->offset < off_desc2->offset)
503 return -1;
504 else if (off_desc1->offset > off_desc2->offset)
505 return 1;
506 return 0;
507}
508
509struct bpf_map_value_off_desc *bpf_map_kptr_off_contains(struct bpf_map *map, u32 offset)
510{
511 /* Since members are iterated in btf_find_field in increasing order,
512 * offsets appended to kptr_off_tab are in increasing order, so we can
513 * do bsearch to find exact match.
514 */
515 struct bpf_map_value_off *tab;
516
517 if (!map_value_has_kptrs(map))
518 return NULL;
519 tab = map->kptr_off_tab;
520 return bsearch(&offset, tab->off, tab->nr_off, sizeof(tab->off[0]), bpf_map_kptr_off_cmp);
521}
522
523void bpf_map_free_kptr_off_tab(struct bpf_map *map)
524{
525 struct bpf_map_value_off *tab = map->kptr_off_tab;
526 int i;
527
528 if (!map_value_has_kptrs(map))
529 return;
14a324f6
KKD		 530 for (i = 0; i < tab->nr_off; i++) {
531 if (tab->off[i].kptr.module)
532 module_put(tab->off[i].kptr.module);
61df10c7 533 btf_put(tab->off[i].kptr.btf);
14a324f6 534 }
61df10c7
KKD		 535 kfree(tab);
536 map->kptr_off_tab = NULL;
537}
538
539struct bpf_map_value_off *bpf_map_copy_kptr_off_tab(const struct bpf_map *map)
540{
541 struct bpf_map_value_off *tab = map->kptr_off_tab, *new_tab;
542 int size, i;
543
544 if (!map_value_has_kptrs(map))
545 return ERR_PTR(-ENOENT);
546 size = offsetof(struct bpf_map_value_off, off[tab->nr_off]);
547 new_tab = kmemdup(tab, size, GFP_KERNEL | __GFP_NOWARN);
548 if (!new_tab)
549 return ERR_PTR(-ENOMEM);
550 /* Do a deep copy of the kptr_off_tab */
14a324f6 551 for (i = 0; i < tab->nr_off; i++) {
61df10c7 552 btf_get(tab->off[i].kptr.btf);
14a324f6
KKD		 553 if (tab->off[i].kptr.module && !try_module_get(tab->off[i].kptr.module)) {
554 while (i--) {
555 if (tab->off[i].kptr.module)
556 module_put(tab->off[i].kptr.module);
557 btf_put(tab->off[i].kptr.btf);
558 }
559 kfree(new_tab);
560 return ERR_PTR(-ENXIO);
561 }
562 }
61df10c7
KKD		 563 return new_tab;
564}
565
566bool bpf_map_equal_kptr_off_tab(const struct bpf_map *map_a, const struct bpf_map *map_b)
567{
568 struct bpf_map_value_off *tab_a = map_a->kptr_off_tab, *tab_b = map_b->kptr_off_tab;
569 bool a_has_kptr = map_value_has_kptrs(map_a), b_has_kptr = map_value_has_kptrs(map_b);
570 int size;
571
572 if (!a_has_kptr && !b_has_kptr)
573 return true;
574 if (a_has_kptr != b_has_kptr)
575 return false;
576 if (tab_a->nr_off != tab_b->nr_off)
577 return false;
578 size = offsetof(struct bpf_map_value_off, off[tab_a->nr_off]);
579 return !memcmp(tab_a, tab_b, size);
580}
581
14a324f6
KKD		 582/* Caller must ensure map_value_has_kptrs is true. Note that this function can
583 * be called on a map value while the map_value is visible to BPF programs, as
584 * it ensures the correct synchronization, and we already enforce the same using
585 * the bpf_kptr_xchg helper on the BPF program side for referenced kptrs.
586 */
587void bpf_map_free_kptrs(struct bpf_map *map, void *map_value)
588{
589 struct bpf_map_value_off *tab = map->kptr_off_tab;
590 unsigned long *btf_id_ptr;
591 int i;
592
593 for (i = 0; i < tab->nr_off; i++) {
594 struct bpf_map_value_off_desc *off_desc = &tab->off[i];
595 unsigned long old_ptr;
596
597 btf_id_ptr = map_value + off_desc->offset;
598 if (off_desc->type == BPF_KPTR_UNREF) {
599 u64 *p = (u64 *)btf_id_ptr;
600
601 WRITE_ONCE(p, 0);
602 continue;
603 }
604 old_ptr = xchg(btf_id_ptr, 0);
605 off_desc->kptr.dtor((void *)old_ptr);
606 }
607}
608
99c55f7d
AS		 609/* called from workqueue */
610static void bpf_map_free_deferred(struct work_struct *work)
611{
612 struct bpf_map *map = container_of(work, struct bpf_map, work);
613
afdb09c7 614 security_bpf_map_free(map);
4d7d7f69 615 kfree(map->off_arr);
48edc1f7 616 bpf_map_release_memcg(map);
14a324f6
KKD		 617 /* implementation dependent freeing, map_free callback also does
618 * bpf_map_free_kptr_off_tab, if needed.
619 */
99c55f7d
AS		 620 map->ops->map_free(map);
621}
622
c9da161c
DB		 623static void bpf_map_put_uref(struct bpf_map *map)
624{
1e0bd5a0 625 if (atomic64_dec_and_test(&map->usercnt)) {
ba6b8de4
JF		 626 if (map->ops->map_release_uref)
627 map->ops->map_release_uref(map);
c9da161c
DB		 628 }
629}
630
99c55f7d
AS		 631/* decrement map refcnt and schedule it for freeing via workqueue
632 * (unrelying map implementation ops->map_free() might sleep)
633 */
bd5f5f4e 634static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
99c55f7d 635{
1e0bd5a0 636 if (atomic64_dec_and_test(&map->refcnt)) {
34ad5580 637 /* bpf_map_free_id() must be called first */
bd5f5f4e 638 bpf_map_free_id(map, do_idr_lock);
78958fca 639 btf_put(map->btf);
99c55f7d
AS		 640 INIT_WORK(&map->work, bpf_map_free_deferred);
641 schedule_work(&map->work);
642 }
643}
644
bd5f5f4e
MKL		 645void bpf_map_put(struct bpf_map *map)
646{
647 __bpf_map_put(map, true);
648}
630a4d38 649EXPORT_SYMBOL_GPL(bpf_map_put);
bd5f5f4e 650
c9da161c 651void bpf_map_put_with_uref(struct bpf_map *map)
99c55f7d 652{
c9da161c 653 bpf_map_put_uref(map);
99c55f7d 654 bpf_map_put(map);
c9da161c
DB		 655}
656
657static int bpf_map_release(struct inode *inode, struct file *filp)
658{
61d1b6a4
DB		 659 struct bpf_map *map = filp->private_data;
660
661 if (map->ops->map_release)
662 map->ops->map_release(map, filp);
663
664 bpf_map_put_with_uref(map);
99c55f7d
AS		 665 return 0;
666}
667
87df15de
DB		 668static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
669{
670 fmode_t mode = f.file->f_mode;
671
672 /* Our file permissions may have been overridden by global
673 * map permissions facing syscall side.
674 */
675 if (READ_ONCE(map->frozen))
676 mode &= ~FMODE_CAN_WRITE;
677 return mode;
678}
679
f99bf205 680#ifdef CONFIG_PROC_FS
80ee81e0
RG		 681/* Provides an approximation of the map's memory footprint.
682 * Used only to provide a backward compatibility and display
683 * a reasonable "memlock" info.
684 */
685static unsigned long bpf_map_memory_footprint(const struct bpf_map *map)
686{
687 unsigned long size;
688
689 size = round_up(map->key_size + bpf_map_value_size(map), 8);
690
691 return round_up(map->max_entries * size, PAGE_SIZE);
692}
693
f99bf205
DB		 694static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
695{
f45d5b6c 696 struct bpf_map *map = filp->private_data;
2beee5f5 697 u32 type = 0, jited = 0;
21116b70 698
f45d5b6c
THJ		 699 if (map_type_contains_progs(map)) {
700 spin_lock(&map->owner.lock);
701 type = map->owner.type;
702 jited = map->owner.jited;
703 spin_unlock(&map->owner.lock);
21116b70 704 }
f99bf205
DB		 705
706 seq_printf(m,
707 "map_type:\t%u\n"
708 "key_size:\t%u\n"
709 "value_size:\t%u\n"
322cea2f 710 "max_entries:\t%u\n"
21116b70 711 "map_flags:\t%#x\n"
9330986c 712 "map_extra:\t%#llx\n"
80ee81e0 713 "memlock:\t%lu\n"
87df15de
DB		 714 "map_id:\t%u\n"
715 "frozen:\t%u\n",
f99bf205
DB		 716 map->map_type,
717 map->key_size,
718 map->value_size,
322cea2f 719 map->max_entries,
21116b70 720 map->map_flags,
9330986c 721 (unsigned long long)map->map_extra,
80ee81e0 722 bpf_map_memory_footprint(map),
87df15de
DB		 723 map->id,
724 READ_ONCE(map->frozen));
2beee5f5
DB		 725 if (type) {
726 seq_printf(m, "owner_prog_type:\t%u\n", type);
727 seq_printf(m, "owner_jited:\t%u\n", jited);
9780c0ab 728 }
f99bf205
DB		 729}
730#endif
731
6e71b04a
CF		 732static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
733 loff_t *ppos)
734{
735 /* We need this handler such that alloc_file() enables
736 * f_mode with FMODE_CAN_READ.
737 */
738 return -EINVAL;
739}
740
741static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
742 size_t siz, loff_t *ppos)
743{
744 /* We need this handler such that alloc_file() enables
745 * f_mode with FMODE_CAN_WRITE.
746 */
747 return -EINVAL;
748}
749
fc970227
AN		 750/* called for any extra memory-mapped regions (except initial) */
751static void bpf_map_mmap_open(struct vm_area_struct *vma)
752{
753 struct bpf_map *map = vma->vm_file->private_data;
754
353050be
DB		 755 if (vma->vm_flags & VM_MAYWRITE)
756 bpf_map_write_active_inc(map);
fc970227
AN		 757}
758
759/* called for all unmapped memory region (including initial) */
760static void bpf_map_mmap_close(struct vm_area_struct *vma)
761{
762 struct bpf_map *map = vma->vm_file->private_data;
763
353050be
DB		 764 if (vma->vm_flags & VM_MAYWRITE)
765 bpf_map_write_active_dec(map);
fc970227
AN		 766}
767
768static const struct vm_operations_struct bpf_map_default_vmops = {
769 .open = bpf_map_mmap_open,
770 .close = bpf_map_mmap_close,
771};
772
773static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma)
774{
775 struct bpf_map *map = filp->private_data;
776 int err;
777
68134668 778 if (!map->ops->map_mmap || map_value_has_spin_lock(map) ||
61df10c7 779 map_value_has_timer(map) || map_value_has_kptrs(map))
fc970227
AN		 780 return -ENOTSUPP;
781
782 if (!(vma->vm_flags & VM_SHARED))
783 return -EINVAL;
784
785 mutex_lock(&map->freeze_mutex);
786
dfeb376d
AN		 787 if (vma->vm_flags & VM_WRITE) {
788 if (map->frozen) {
789 err = -EPERM;
790 goto out;
791 }
792 /* map is meant to be read-only, so do not allow mapping as
793 * writable, because it's possible to leak a writable page
794 * reference and allows user-space to still modify it after
795 * freezing, while verifier will assume contents do not change
796 */
797 if (map->map_flags & BPF_F_RDONLY_PROG) {
798 err = -EACCES;
799 goto out;
800 }
fc970227
AN		 801 }
802
803 /* set default open/close callbacks */
804 vma->vm_ops = &bpf_map_default_vmops;
805 vma->vm_private_data = map;
1f6cb19b
AN		 806 vma->vm_flags &= ~VM_MAYEXEC;
807 if (!(vma->vm_flags & VM_WRITE))
808 /* disallow re-mapping with PROT_WRITE */
809 vma->vm_flags &= ~VM_MAYWRITE;
fc970227
AN		 810
811 err = map->ops->map_mmap(map, vma);
812 if (err)
813 goto out;
814
1f6cb19b 815 if (vma->vm_flags & VM_MAYWRITE)
353050be 816 bpf_map_write_active_inc(map);
fc970227
AN		 817out:
818 mutex_unlock(&map->freeze_mutex);
819 return err;
820}
821
457f4436
AN		 822static __poll_t bpf_map_poll(struct file *filp, struct poll_table_struct *pts)
823{
824 struct bpf_map *map = filp->private_data;
825
826 if (map->ops->map_poll)
827 return map->ops->map_poll(map, filp, pts);
828
829 return EPOLLERR;
830}
831
f66e448c 832const struct file_operations bpf_map_fops = {
f99bf205
DB		 833#ifdef CONFIG_PROC_FS
834 .show_fdinfo = bpf_map_show_fdinfo,
835#endif
836 .release = bpf_map_release,
6e71b04a
CF		 837 .read = bpf_dummy_read,
838 .write = bpf_dummy_write,
fc970227 839 .mmap = bpf_map_mmap,
457f4436 840 .poll = bpf_map_poll,
99c55f7d
AS		 841};
842
6e71b04a 843int bpf_map_new_fd(struct bpf_map *map, int flags)
aa79781b 844{
afdb09c7
CF		 845 int ret;
846
847 ret = security_bpf_map(map, OPEN_FMODE(flags));
848 if (ret < 0)
849 return ret;
850
aa79781b 851 return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
6e71b04a
CF		 852 flags | O_CLOEXEC);
853}
854
855int bpf_get_file_flag(int flags)
856{
857 if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
858 return -EINVAL;
859 if (flags & BPF_F_RDONLY)
860 return O_RDONLY;
861 if (flags & BPF_F_WRONLY)
862 return O_WRONLY;
863 return O_RDWR;
aa79781b
DB		 864}
865
99c55f7d
AS		 866/* helper macro to check that unused fields 'union bpf_attr' are zero */
867#define CHECK_ATTR(CMD) \
868 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
869 sizeof(attr->CMD##_LAST_FIELD), 0, \
870 sizeof(*attr) - \
871 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
872 sizeof(attr->CMD##_LAST_FIELD)) != NULL
873
8e7ae251
MKL		 874/* dst and src must have at least "size" number of bytes.
875 * Return strlen on success and < 0 on error.
cb4d2b3f 876 */
8e7ae251 877int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size)
cb4d2b3f 878{
8e7ae251
MKL		 879 const char *end = src + size;
880 const char *orig_src = src;
cb4d2b3f 881
8e7ae251 882 memset(dst, 0, size);
3e0ddc4f 883 /* Copy all isalnum(), '_' and '.' chars. */
cb4d2b3f 884 while (src < end && *src) {
3e0ddc4f
DB		 885 if (!isalnum(*src) &&
886 *src != '_' && *src != '.')
cb4d2b3f
MKL		 887 return -EINVAL;
888 *dst++ = *src++;
889 }
890
8e7ae251 891 /* No '\0' found in "size" number of bytes */
cb4d2b3f
MKL		 892 if (src == end)
893 return -EINVAL;
894
8e7ae251 895 return src - orig_src;
cb4d2b3f
MKL		 896}
897
e8d2bec0 898int map_check_no_btf(const struct bpf_map *map,
1b2b234b 899 const struct btf *btf,
e8d2bec0
DB		 900 const struct btf_type *key_type,
901 const struct btf_type *value_type)
902{
903 return -ENOTSUPP;
904}
905
4d7d7f69
KKD		 906static int map_off_arr_cmp(const void *_a, const void *_b, const void *priv)
907{
908 const u32 a = *(const u32 *)_a;
909 const u32 b = *(const u32 *)_b;
910
911 if (a < b)
912 return -1;
913 else if (a > b)
914 return 1;
915 return 0;
916}
917
918static void map_off_arr_swap(void *_a, void *_b, int size, const void *priv)
919{
920 struct bpf_map *map = (struct bpf_map *)priv;
921 u32 *off_base = map->off_arr->field_off;
922 u32 *a = _a, *b = _b;
923 u8 *sz_a, *sz_b;
924
925 sz_a = map->off_arr->field_sz + (a - off_base);
926 sz_b = map->off_arr->field_sz + (b - off_base);
927
928 swap(*a, *b);
929 swap(*sz_a, *sz_b);
930}
931
932static int bpf_map_alloc_off_arr(struct bpf_map *map)
933{
934 bool has_spin_lock = map_value_has_spin_lock(map);
935 bool has_timer = map_value_has_timer(map);
936 bool has_kptrs = map_value_has_kptrs(map);
937 struct bpf_map_off_arr *off_arr;
938 u32 i;
939
940 if (!has_spin_lock && !has_timer && !has_kptrs) {
941 map->off_arr = NULL;
942 return 0;
943 }
944
945 off_arr = kmalloc(sizeof(*map->off_arr), GFP_KERNEL | __GFP_NOWARN);
946 if (!off_arr)
947 return -ENOMEM;
948 map->off_arr = off_arr;
949
950 off_arr->cnt = 0;
951 if (has_spin_lock) {
952 i = off_arr->cnt;
953
954 off_arr->field_off[i] = map->spin_lock_off;
955 off_arr->field_sz[i] = sizeof(struct bpf_spin_lock);
956 off_arr->cnt++;
957 }
958 if (has_timer) {
959 i = off_arr->cnt;
960
961 off_arr->field_off[i] = map->timer_off;
962 off_arr->field_sz[i] = sizeof(struct bpf_timer);
963 off_arr->cnt++;
964 }
965 if (has_kptrs) {
966 struct bpf_map_value_off *tab = map->kptr_off_tab;
967 u32 *off = &off_arr->field_off[off_arr->cnt];
968 u8 *sz = &off_arr->field_sz[off_arr->cnt];
969
970 for (i = 0; i < tab->nr_off; i++) {
971 *off++ = tab->off[i].offset;
972 *sz++ = sizeof(u64);
973 }
974 off_arr->cnt += tab->nr_off;
975 }
976
977 if (off_arr->cnt == 1)
978 return 0;
979 sort_r(off_arr->field_off, off_arr->cnt, sizeof(off_arr->field_off[0]),
980 map_off_arr_cmp, map_off_arr_swap, map);
981 return 0;
982}
983
d83525ca 984static int map_check_btf(struct bpf_map *map, const struct btf *btf,
e8d2bec0
DB		 985 u32 btf_key_id, u32 btf_value_id)
986{
987 const struct btf_type *key_type, *value_type;
988 u32 key_size, value_size;
989 int ret = 0;
990
2824ecb7
DB		 991 /* Some maps allow key to be unspecified. */
992 if (btf_key_id) {
993 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
994 if (!key_type || key_size != map->key_size)
995 return -EINVAL;
996 } else {
997 key_type = btf_type_by_id(btf, 0);
998 if (!map->ops->map_check_btf)
999 return -EINVAL;
1000 }
e8d2bec0
DB		 1001
1002 value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
1003 if (!value_type || value_size != map->value_size)
1004 return -EINVAL;
1005
d83525ca
AS		 1006 map->spin_lock_off = btf_find_spin_lock(btf, value_type);
1007
1008 if (map_value_has_spin_lock(map)) {
591fe988
DB		 1009 if (map->map_flags & BPF_F_RDONLY_PROG)
1010 return -EACCES;
d83525ca 1011 if (map->map_type != BPF_MAP_TYPE_HASH &&
e16d2f1a 1012 map->map_type != BPF_MAP_TYPE_ARRAY &&
6ac99e8f 1013 map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
8ea63684 1014 map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
4cf1bc1f
KS		 1015 map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
1016 map->map_type != BPF_MAP_TYPE_TASK_STORAGE)
d83525ca
AS		 1017 return -ENOTSUPP;
1018 if (map->spin_lock_off + sizeof(struct bpf_spin_lock) >
1019 map->value_size) {
1020 WARN_ONCE(1,
1021 "verifier bug spin_lock_off %d value_size %d\n",
1022 map->spin_lock_off, map->value_size);
1023 return -EFAULT;
1024 }
1025 }
1026
68134668
AS		 1027 map->timer_off = btf_find_timer(btf, value_type);
1028 if (map_value_has_timer(map)) {
1029 if (map->map_flags & BPF_F_RDONLY_PROG)
1030 return -EACCES;
1031 if (map->map_type != BPF_MAP_TYPE_HASH &&
1032 map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1033 map->map_type != BPF_MAP_TYPE_ARRAY)
1034 return -EOPNOTSUPP;
1035 }
1036
61df10c7
KKD		 1037 map->kptr_off_tab = btf_parse_kptrs(btf, value_type);
1038 if (map_value_has_kptrs(map)) {
1039 if (!bpf_capable()) {
1040 ret = -EPERM;
1041 goto free_map_tab;
1042 }
1043 if (map->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG)) {
1044 ret = -EACCES;
1045 goto free_map_tab;
1046 }
1047 if (map->map_type != BPF_MAP_TYPE_HASH &&
1048 map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1049 map->map_type != BPF_MAP_TYPE_ARRAY) {
1050 ret = -EOPNOTSUPP;
1051 goto free_map_tab;
1052 }
1053 }
1054
1055 if (map->ops->map_check_btf) {
1b2b234b 1056 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
61df10c7
KKD		 1057 if (ret < 0)
1058 goto free_map_tab;
1059 }
e8d2bec0 1060
61df10c7
KKD		 1061 return ret;
1062free_map_tab:
1063 bpf_map_free_kptr_off_tab(map);
e8d2bec0
DB		 1064 return ret;
1065}
1066
9330986c 1067#define BPF_MAP_CREATE_LAST_FIELD map_extra
99c55f7d
AS		 1068/* called via syscall */
1069static int map_create(union bpf_attr *attr)
1070{
96eabe7a 1071 int numa_node = bpf_map_attr_numa_node(attr);
99c55f7d 1072 struct bpf_map *map;
6e71b04a 1073 int f_flags;
99c55f7d
AS		 1074 int err;
1075
1076 err = CHECK_ATTR(BPF_MAP_CREATE);
1077 if (err)
1078 return -EINVAL;
1079
85d33df3
MKL		 1080 if (attr->btf_vmlinux_value_type_id) {
1081 if (attr->map_type != BPF_MAP_TYPE_STRUCT_OPS ||
1082 attr->btf_key_type_id || attr->btf_value_type_id)
1083 return -EINVAL;
1084 } else if (attr->btf_key_type_id && !attr->btf_value_type_id) {
1085 return -EINVAL;
1086 }
1087
9330986c
JK		 1088 if (attr->map_type != BPF_MAP_TYPE_BLOOM_FILTER &&
1089 attr->map_extra != 0)
1090 return -EINVAL;
1091
6e71b04a
CF		 1092 f_flags = bpf_get_file_flag(attr->map_flags);
1093 if (f_flags < 0)
1094 return f_flags;
1095
96eabe7a 1096 if (numa_node != NUMA_NO_NODE &&
96e5ae4e
ED		 1097 ((unsigned int)numa_node >= nr_node_ids ||
1098 !node_online(numa_node)))
96eabe7a
MKL		 1099 return -EINVAL;
1100
99c55f7d
AS		 1101 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
1102 map = find_and_alloc_map(attr);
1103 if (IS_ERR(map))
1104 return PTR_ERR(map);
1105
8e7ae251
MKL		 1106 err = bpf_obj_name_cpy(map->name, attr->map_name,
1107 sizeof(attr->map_name));
1108 if (err < 0)
b936ca64 1109 goto free_map;
ad5b177b 1110
1e0bd5a0
AN		 1111 atomic64_set(&map->refcnt, 1);
1112 atomic64_set(&map->usercnt, 1);
fc970227 1113 mutex_init(&map->freeze_mutex);
f45d5b6c 1114 spin_lock_init(&map->owner.lock);
99c55f7d 1115
85d33df3 1116 map->spin_lock_off = -EINVAL;
68134668 1117 map->timer_off = -EINVAL;
85d33df3
MKL		 1118 if (attr->btf_key_type_id || attr->btf_value_type_id ||
1119 /* Even the map's value is a kernel's struct,
1120 * the bpf_prog.o must have BTF to begin with
1121 * to figure out the corresponding kernel's
1122 * counter part. Thus, attr->btf_fd has
1123 * to be valid also.
1124 */
1125 attr->btf_vmlinux_value_type_id) {
a26ca7c9
MKL		 1126 struct btf *btf;
1127
a26ca7c9
MKL		 1128 btf = btf_get_by_fd(attr->btf_fd);
1129 if (IS_ERR(btf)) {
1130 err = PTR_ERR(btf);
b936ca64 1131 goto free_map;
a26ca7c9 1132 }
350a5c4d
AS		 1133 if (btf_is_kernel(btf)) {
1134 btf_put(btf);
1135 err = -EACCES;
1136 goto free_map;
1137 }
85d33df3 1138 map->btf = btf;
a26ca7c9 1139
85d33df3
MKL		 1140 if (attr->btf_value_type_id) {
1141 err = map_check_btf(map, btf, attr->btf_key_type_id,
1142 attr->btf_value_type_id);
1143 if (err)
1144 goto free_map;
a26ca7c9
MKL		 1145 }
1146
9b2cf328
MKL		 1147 map->btf_key_type_id = attr->btf_key_type_id;
1148 map->btf_value_type_id = attr->btf_value_type_id;
85d33df3
MKL		 1149 map->btf_vmlinux_value_type_id =
1150 attr->btf_vmlinux_value_type_id;
a26ca7c9
MKL		 1151 }
1152
4d7d7f69 1153 err = bpf_map_alloc_off_arr(map);
aaac3ba9 1154 if (err)
b936ca64 1155 goto free_map;
afdb09c7 1156
4d7d7f69
KKD		 1157 err = security_bpf_map_alloc(map);
1158 if (err)
1159 goto free_map_off_arr;
1160
f3f1c054
MKL		 1161 err = bpf_map_alloc_id(map);
1162 if (err)
b936ca64 1163 goto free_map_sec;
f3f1c054 1164
48edc1f7
RG		 1165 bpf_map_save_memcg(map);
1166
6e71b04a 1167 err = bpf_map_new_fd(map, f_flags);
bd5f5f4e
MKL		 1168 if (err < 0) {
1169 /* failed to allocate fd.
352d20d6 1170 * bpf_map_put_with_uref() is needed because the above
bd5f5f4e
MKL		 1171 * bpf_map_alloc_id() has published the map
1172 * to the userspace and the userspace may
1173 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
1174 */
352d20d6 1175 bpf_map_put_with_uref(map);
bd5f5f4e
MKL		 1176 return err;
1177 }
99c55f7d
AS		 1178
1179 return err;
1180
afdb09c7
CF		 1181free_map_sec:
1182 security_bpf_map_free(map);
4d7d7f69
KKD		 1183free_map_off_arr:
1184 kfree(map->off_arr);
b936ca64 1185free_map:
a26ca7c9 1186 btf_put(map->btf);
99c55f7d
AS		 1187 map->ops->map_free(map);
1188 return err;
1189}
1190
db20fd2b
AS		 1191/* if error is returned, fd is released.
1192 * On success caller should complete fd access with matching fdput()
1193 */
c2101297 1194struct bpf_map *__bpf_map_get(struct fd f)
db20fd2b 1195{
db20fd2b
AS		 1196 if (!f.file)
1197 return ERR_PTR(-EBADF);
db20fd2b
AS		 1198 if (f.file->f_op != &bpf_map_fops) {
1199 fdput(f);
1200 return ERR_PTR(-EINVAL);
1201 }
1202
c2101297
DB		 1203 return f.file->private_data;
1204}
1205
1e0bd5a0 1206void bpf_map_inc(struct bpf_map *map)
c9da161c 1207{
1e0bd5a0 1208 atomic64_inc(&map->refcnt);
c9da161c 1209}
630a4d38 1210EXPORT_SYMBOL_GPL(bpf_map_inc);
c9da161c 1211
1e0bd5a0
AN		 1212void bpf_map_inc_with_uref(struct bpf_map *map)
1213{
1214 atomic64_inc(&map->refcnt);
1215 atomic64_inc(&map->usercnt);
1216}
1217EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
1218
1ed4d924
MKL		 1219struct bpf_map *bpf_map_get(u32 ufd)
1220{
1221 struct fd f = fdget(ufd);
1222 struct bpf_map *map;
1223
1224 map = __bpf_map_get(f);
1225 if (IS_ERR(map))
1226 return map;
1227
1228 bpf_map_inc(map);
1229 fdput(f);
1230
1231 return map;
1232}
b1d18a75 1233EXPORT_SYMBOL(bpf_map_get);
1ed4d924 1234
c9da161c 1235struct bpf_map *bpf_map_get_with_uref(u32 ufd)
c2101297
DB		 1236{
1237 struct fd f = fdget(ufd);
1238 struct bpf_map *map;
1239
1240 map = __bpf_map_get(f);
1241 if (IS_ERR(map))
1242 return map;
1243
1e0bd5a0 1244 bpf_map_inc_with_uref(map);
c2101297 1245 fdput(f);
db20fd2b
AS		 1246
1247 return map;
1248}
1249
bd5f5f4e 1250/* map_idr_lock should have been held */
1e0bd5a0 1251static struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
bd5f5f4e
MKL		 1252{
1253 int refold;
1254
1e0bd5a0 1255 refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
bd5f5f4e
MKL		 1256 if (!refold)
1257 return ERR_PTR(-ENOENT);
bd5f5f4e 1258 if (uref)
1e0bd5a0 1259 atomic64_inc(&map->usercnt);
bd5f5f4e
MKL		 1260
1261 return map;
1262}
1263
1e0bd5a0 1264struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
b0e4701c
SF		 1265{
1266 spin_lock_bh(&map_idr_lock);
1e0bd5a0 1267 map = __bpf_map_inc_not_zero(map, false);
b0e4701c
SF		 1268 spin_unlock_bh(&map_idr_lock);
1269
1270 return map;
1271}
1272EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
1273
b8cdc051
AS		 1274int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
1275{
1276 return -ENOTSUPP;
1277}
1278
c9d29f46
MV		 1279static void *__bpf_copy_key(void __user *ukey, u64 key_size)
1280{
1281 if (key_size)
44779a4b 1282 return vmemdup_user(ukey, key_size);
c9d29f46
MV		 1283
1284 if (ukey)
1285 return ERR_PTR(-EINVAL);
1286
1287 return NULL;
1288}
1289
af2ac3e1
AS		 1290static void *___bpf_copy_key(bpfptr_t ukey, u64 key_size)
1291{
1292 if (key_size)
44779a4b 1293 return kvmemdup_bpfptr(ukey, key_size);
af2ac3e1
AS		 1294
1295 if (!bpfptr_is_null(ukey))
1296 return ERR_PTR(-EINVAL);
1297
1298 return NULL;
1299}
1300
db20fd2b 1301/* last field in 'union bpf_attr' used by this command */
96049f3a 1302#define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
db20fd2b
AS		 1303
1304static int map_lookup_elem(union bpf_attr *attr)
1305{
535e7b4b
MS		 1306 void __user *ukey = u64_to_user_ptr(attr->key);
1307 void __user *uvalue = u64_to_user_ptr(attr->value);
db20fd2b 1308 int ufd = attr->map_fd;
db20fd2b 1309 struct bpf_map *map;
15c14a3d 1310 void *key, *value;
15a07b33 1311 u32 value_size;
592867bf 1312 struct fd f;
db20fd2b
AS		 1313 int err;
1314
1315 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
1316 return -EINVAL;
1317
96049f3a
AS		 1318 if (attr->flags & ~BPF_F_LOCK)
1319 return -EINVAL;
1320
592867bf 1321 f = fdget(ufd);
c2101297 1322 map = __bpf_map_get(f);
db20fd2b
AS		 1323 if (IS_ERR(map))
1324 return PTR_ERR(map);
87df15de 1325 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
6e71b04a
CF		 1326 err = -EPERM;
1327 goto err_put;
1328 }
1329
96049f3a
AS		 1330 if ((attr->flags & BPF_F_LOCK) &&
1331 !map_value_has_spin_lock(map)) {
1332 err = -EINVAL;
1333 goto err_put;
1334 }
1335
c9d29f46 1336 key = __bpf_copy_key(ukey, map->key_size);
e4448ed8
AV		 1337 if (IS_ERR(key)) {
1338 err = PTR_ERR(key);
db20fd2b 1339 goto err_put;
e4448ed8 1340 }
db20fd2b 1341
15c14a3d 1342 value_size = bpf_map_value_size(map);
15a07b33 1343
8ebe667c 1344 err = -ENOMEM;
f0dce1d9 1345 value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
db20fd2b 1346 if (!value)
8ebe667c
AS		 1347 goto free_key;
1348
9330986c
JK		 1349 if (map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
1350 if (copy_from_user(value, uvalue, value_size))
1351 err = -EFAULT;
1352 else
1353 err = bpf_map_copy_value(map, key, value, attr->flags);
1354 goto free_value;
1355 }
1356
15c14a3d 1357 err = bpf_map_copy_value(map, key, value, attr->flags);
15a07b33 1358 if (err)
8ebe667c 1359 goto free_value;
db20fd2b
AS		 1360
1361 err = -EFAULT;
15a07b33 1362 if (copy_to_user(uvalue, value, value_size) != 0)
8ebe667c 1363 goto free_value;
db20fd2b
AS		 1364
1365 err = 0;
1366
8ebe667c 1367free_value:
f0dce1d9 1368 kvfree(value);
db20fd2b 1369free_key:
44779a4b 1370 kvfree(key);
db20fd2b
AS		 1371err_put:
1372 fdput(f);
1373 return err;
1374}
1375
1ae80cf3 1376
3274f520 1377#define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
db20fd2b 1378
af2ac3e1 1379static int map_update_elem(union bpf_attr *attr, bpfptr_t uattr)
db20fd2b 1380{
af2ac3e1
AS		 1381 bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1382 bpfptr_t uvalue = make_bpfptr(attr->value, uattr.is_kernel);
db20fd2b 1383 int ufd = attr->map_fd;
db20fd2b
AS		 1384 struct bpf_map *map;
1385 void *key, *value;
15a07b33 1386 u32 value_size;
592867bf 1387 struct fd f;
db20fd2b
AS		 1388 int err;
1389
1390 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1391 return -EINVAL;
1392
592867bf 1393 f = fdget(ufd);
c2101297 1394 map = __bpf_map_get(f);
db20fd2b
AS		 1395 if (IS_ERR(map))
1396 return PTR_ERR(map);
353050be 1397 bpf_map_write_active_inc(map);
87df15de 1398 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
6e71b04a
CF		 1399 err = -EPERM;
1400 goto err_put;
1401 }
1402
96049f3a
AS		 1403 if ((attr->flags & BPF_F_LOCK) &&
1404 !map_value_has_spin_lock(map)) {
1405 err = -EINVAL;
1406 goto err_put;
1407 }
1408
af2ac3e1 1409 key = ___bpf_copy_key(ukey, map->key_size);
e4448ed8
AV		 1410 if (IS_ERR(key)) {
1411 err = PTR_ERR(key);
db20fd2b 1412 goto err_put;
e4448ed8 1413 }
db20fd2b 1414
f0dce1d9 1415 value_size = bpf_map_value_size(map);
15a07b33 1416
db20fd2b 1417 err = -ENOMEM;
f0dce1d9 1418 value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
db20fd2b
AS		 1419 if (!value)
1420 goto free_key;
1421
1422 err = -EFAULT;
af2ac3e1 1423 if (copy_from_bpfptr(value, uvalue, value_size) != 0)
db20fd2b
AS		 1424 goto free_value;
1425
15c14a3d 1426 err = bpf_map_update_value(map, f, key, value, attr->flags);
6710e112 1427
db20fd2b 1428free_value:
f0dce1d9 1429 kvfree(value);
db20fd2b 1430free_key:
44779a4b 1431 kvfree(key);
db20fd2b 1432err_put:
353050be 1433 bpf_map_write_active_dec(map);
db20fd2b
AS		 1434 fdput(f);
1435 return err;
1436}
1437
1438#define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1439
1440static int map_delete_elem(union bpf_attr *attr)
1441{
535e7b4b 1442 void __user *ukey = u64_to_user_ptr(attr->key);
db20fd2b 1443 int ufd = attr->map_fd;
db20fd2b 1444 struct bpf_map *map;
592867bf 1445 struct fd f;
db20fd2b
AS		 1446 void *key;
1447 int err;
1448
1449 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1450 return -EINVAL;
1451
592867bf 1452 f = fdget(ufd);
c2101297 1453 map = __bpf_map_get(f);
db20fd2b
AS		 1454 if (IS_ERR(map))
1455 return PTR_ERR(map);
353050be 1456 bpf_map_write_active_inc(map);
87df15de 1457 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
6e71b04a
CF		 1458 err = -EPERM;
1459 goto err_put;
1460 }
1461
c9d29f46 1462 key = __bpf_copy_key(ukey, map->key_size);
e4448ed8
AV		 1463 if (IS_ERR(key)) {
1464 err = PTR_ERR(key);
db20fd2b 1465 goto err_put;
e4448ed8 1466 }
db20fd2b 1467
a3884572
JK		 1468 if (bpf_map_is_dev_bound(map)) {
1469 err = bpf_map_offload_delete_elem(map, key);
1470 goto out;
85d33df3
MKL		 1471 } else if (IS_FD_PROG_ARRAY(map) ||
1472 map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1473 /* These maps require sleepable context */
da765a2f
DB		 1474 err = map->ops->map_delete_elem(map, key);
1475 goto out;
a3884572
JK		 1476 }
1477
b6e5dae1 1478 bpf_disable_instrumentation();
db20fd2b
AS		 1479 rcu_read_lock();
1480 err = map->ops->map_delete_elem(map, key);
1481 rcu_read_unlock();
b6e5dae1 1482 bpf_enable_instrumentation();
1ae80cf3 1483 maybe_wait_bpf_programs(map);
a3884572 1484out:
44779a4b 1485 kvfree(key);
db20fd2b 1486err_put:
353050be 1487 bpf_map_write_active_dec(map);
db20fd2b
AS		 1488 fdput(f);
1489 return err;
1490}
1491
1492/* last field in 'union bpf_attr' used by this command */
1493#define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1494
1495static int map_get_next_key(union bpf_attr *attr)
1496{
535e7b4b
MS		 1497 void __user *ukey = u64_to_user_ptr(attr->key);
1498 void __user *unext_key = u64_to_user_ptr(attr->next_key);
db20fd2b 1499 int ufd = attr->map_fd;
db20fd2b
AS		 1500 struct bpf_map *map;
1501 void *key, *next_key;
592867bf 1502 struct fd f;
db20fd2b
AS		 1503 int err;
1504
1505 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1506 return -EINVAL;
1507
592867bf 1508 f = fdget(ufd);
c2101297 1509 map = __bpf_map_get(f);
db20fd2b
AS		 1510 if (IS_ERR(map))
1511 return PTR_ERR(map);
87df15de 1512 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
6e71b04a
CF		 1513 err = -EPERM;
1514 goto err_put;
1515 }
1516
8fe45924 1517 if (ukey) {
c9d29f46 1518 key = __bpf_copy_key(ukey, map->key_size);
e4448ed8
AV		 1519 if (IS_ERR(key)) {
1520 err = PTR_ERR(key);
8fe45924 1521 goto err_put;
e4448ed8 1522 }
8fe45924
TQ		 1523 } else {
1524 key = NULL;
1525 }
db20fd2b
AS		 1526
1527 err = -ENOMEM;
44779a4b 1528 next_key = kvmalloc(map->key_size, GFP_USER);
db20fd2b
AS		 1529 if (!next_key)
1530 goto free_key;
1531
a3884572
JK		 1532 if (bpf_map_is_dev_bound(map)) {
1533 err = bpf_map_offload_get_next_key(map, key, next_key);
1534 goto out;
1535 }
1536
db20fd2b
AS		 1537 rcu_read_lock();
1538 err = map->ops->map_get_next_key(map, key, next_key);
1539 rcu_read_unlock();
a3884572 1540out:
db20fd2b
AS		 1541 if (err)
1542 goto free_next_key;
1543
1544 err = -EFAULT;
1545 if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1546 goto free_next_key;
1547
1548 err = 0;
1549
1550free_next_key:
44779a4b 1551 kvfree(next_key);
db20fd2b 1552free_key:
44779a4b 1553 kvfree(key);
db20fd2b
AS		 1554err_put:
1555 fdput(f);
1556 return err;
1557}
1558
aa2e93b8
BV		 1559int generic_map_delete_batch(struct bpf_map *map,
1560 const union bpf_attr *attr,
1561 union bpf_attr __user *uattr)
1562{
1563 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1564 u32 cp, max_count;
1565 int err = 0;
1566 void *key;
1567
1568 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1569 return -EINVAL;
1570
1571 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1572 !map_value_has_spin_lock(map)) {
1573 return -EINVAL;
1574 }
1575
1576 max_count = attr->batch.count;
1577 if (!max_count)
1578 return 0;
1579
44779a4b 1580 key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
2e3a94aa
BV		 1581 if (!key)
1582 return -ENOMEM;
1583
aa2e93b8 1584 for (cp = 0; cp < max_count; cp++) {
2e3a94aa
BV		 1585 err = -EFAULT;
1586 if (copy_from_user(key, keys + cp * map->key_size,
1587 map->key_size))
aa2e93b8 1588 break;
aa2e93b8
BV		 1589
1590 if (bpf_map_is_dev_bound(map)) {
1591 err = bpf_map_offload_delete_elem(map, key);
1592 break;
1593 }
1594
b6e5dae1 1595 bpf_disable_instrumentation();
aa2e93b8
BV		 1596 rcu_read_lock();
1597 err = map->ops->map_delete_elem(map, key);
1598 rcu_read_unlock();
b6e5dae1 1599 bpf_enable_instrumentation();
aa2e93b8
BV		 1600 if (err)
1601 break;
75134f16 1602 cond_resched();
aa2e93b8
BV		 1603 }
1604 if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1605 err = -EFAULT;
2e3a94aa 1606
44779a4b 1607 kvfree(key);
9087c6ff
ED		 1608
1609 maybe_wait_bpf_programs(map);
aa2e93b8
BV		 1610 return err;
1611}
1612
1613int generic_map_update_batch(struct bpf_map *map,
1614 const union bpf_attr *attr,
1615 union bpf_attr __user *uattr)
1616{
1617 void __user *values = u64_to_user_ptr(attr->batch.values);
1618 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1619 u32 value_size, cp, max_count;
fda7a387 1620 int ufd = attr->batch.map_fd;
aa2e93b8
BV		 1621 void *key, *value;
1622 struct fd f;
1623 int err = 0;
1624
aa2e93b8
BV		 1625 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1626 return -EINVAL;
1627
1628 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1629 !map_value_has_spin_lock(map)) {
1630 return -EINVAL;
1631 }
1632
1633 value_size = bpf_map_value_size(map);
1634
1635 max_count = attr->batch.count;
1636 if (!max_count)
1637 return 0;
1638
44779a4b 1639 key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
2e3a94aa
BV		 1640 if (!key)
1641 return -ENOMEM;
1642
f0dce1d9 1643 value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
2e3a94aa 1644 if (!value) {
44779a4b 1645 kvfree(key);
aa2e93b8 1646 return -ENOMEM;
2e3a94aa 1647 }
aa2e93b8 1648
fda7a387 1649 f = fdget(ufd); /* bpf_map_do_batch() guarantees ufd is valid */
aa2e93b8 1650 for (cp = 0; cp < max_count; cp++) {
aa2e93b8 1651 err = -EFAULT;
2e3a94aa
BV		 1652 if (copy_from_user(key, keys + cp * map->key_size,
1653 map->key_size) ||
1654 copy_from_user(value, values + cp * value_size, value_size))
aa2e93b8
BV		 1655 break;
1656
1657 err = bpf_map_update_value(map, f, key, value,
1658 attr->batch.elem_flags);
1659
1660 if (err)
1661 break;
75134f16 1662 cond_resched();
aa2e93b8
BV		 1663 }
1664
1665 if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1666 err = -EFAULT;
1667
f0dce1d9 1668 kvfree(value);
44779a4b 1669 kvfree(key);
fda7a387 1670 fdput(f);
aa2e93b8
BV		 1671 return err;
1672}
1673
cb4d03ab
BV		 1674#define MAP_LOOKUP_RETRIES 3
1675
1676int generic_map_lookup_batch(struct bpf_map *map,
1677 const union bpf_attr *attr,
1678 union bpf_attr __user *uattr)
1679{
1680 void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
1681 void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1682 void __user *values = u64_to_user_ptr(attr->batch.values);
1683 void __user *keys = u64_to_user_ptr(attr->batch.keys);
1684 void *buf, *buf_prevkey, *prev_key, *key, *value;
1685 int err, retry = MAP_LOOKUP_RETRIES;
1686 u32 value_size, cp, max_count;
cb4d03ab
BV		 1687
1688 if (attr->batch.elem_flags & ~BPF_F_LOCK)
1689 return -EINVAL;
1690
1691 if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1692 !map_value_has_spin_lock(map))
1693 return -EINVAL;
1694
1695 value_size = bpf_map_value_size(map);
1696
1697 max_count = attr->batch.count;
1698 if (!max_count)
1699 return 0;
1700
1701 if (put_user(0, &uattr->batch.count))
1702 return -EFAULT;
1703
44779a4b 1704 buf_prevkey = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
cb4d03ab
BV		 1705 if (!buf_prevkey)
1706 return -ENOMEM;
1707
f0dce1d9 1708 buf = kvmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
cb4d03ab 1709 if (!buf) {
44779a4b 1710 kvfree(buf_prevkey);
cb4d03ab
BV		 1711 return -ENOMEM;
1712 }
1713
1714 err = -EFAULT;
cb4d03ab
BV		 1715 prev_key = NULL;
1716 if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
1717 goto free_buf;
1718 key = buf;
1719 value = key + map->key_size;
1720 if (ubatch)
1721 prev_key = buf_prevkey;
1722
1723 for (cp = 0; cp < max_count;) {
1724 rcu_read_lock();
1725 err = map->ops->map_get_next_key(map, prev_key, key);
1726 rcu_read_unlock();
1727 if (err)
1728 break;
1729 err = bpf_map_copy_value(map, key, value,
1730 attr->batch.elem_flags);
1731
1732 if (err == -ENOENT) {
1733 if (retry) {
1734 retry--;
1735 continue;
1736 }
1737 err = -EINTR;
1738 break;
1739 }
1740
1741 if (err)
1742 goto free_buf;
1743
1744 if (copy_to_user(keys + cp * map->key_size, key,
1745 map->key_size)) {
1746 err = -EFAULT;
1747 goto free_buf;
1748 }
1749 if (copy_to_user(values + cp * value_size, value, value_size)) {
1750 err = -EFAULT;
1751 goto free_buf;
1752 }
1753
1754 if (!prev_key)
1755 prev_key = buf_prevkey;
1756
1757 swap(prev_key, key);
1758 retry = MAP_LOOKUP_RETRIES;
1759 cp++;
75134f16 1760 cond_resched();
cb4d03ab
BV		 1761 }
1762
1763 if (err == -EFAULT)
1764 goto free_buf;
1765
1766 if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
1767 (cp && copy_to_user(uobatch, prev_key, map->key_size))))
1768 err = -EFAULT;
1769
1770free_buf:
44779a4b 1771 kvfree(buf_prevkey);
f0dce1d9 1772 kvfree(buf);
cb4d03ab
BV		 1773 return err;
1774}
1775
3e87f192 1776#define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD flags
bd513cd0
MV		 1777
1778static int map_lookup_and_delete_elem(union bpf_attr *attr)
1779{
1780 void __user *ukey = u64_to_user_ptr(attr->key);
1781 void __user *uvalue = u64_to_user_ptr(attr->value);
1782 int ufd = attr->map_fd;
1783 struct bpf_map *map;
540fefc0 1784 void *key, *value;
bd513cd0
MV		 1785 u32 value_size;
1786 struct fd f;
1787 int err;
1788
1789 if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1790 return -EINVAL;
1791
3e87f192
DS		 1792 if (attr->flags & ~BPF_F_LOCK)
1793 return -EINVAL;
1794
bd513cd0
MV		 1795 f = fdget(ufd);
1796 map = __bpf_map_get(f);
1797 if (IS_ERR(map))
1798 return PTR_ERR(map);
353050be 1799 bpf_map_write_active_inc(map);
1ea0f912
AP		 1800 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ) ||
1801 !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
bd513cd0
MV		 1802 err = -EPERM;
1803 goto err_put;
1804 }
1805
3e87f192
DS		 1806 if (attr->flags &&
1807 (map->map_type == BPF_MAP_TYPE_QUEUE ||
1808 map->map_type == BPF_MAP_TYPE_STACK)) {
1809 err = -EINVAL;
1810 goto err_put;
1811 }
1812
1813 if ((attr->flags & BPF_F_LOCK) &&
1814 !map_value_has_spin_lock(map)) {
1815 err = -EINVAL;
1816 goto err_put;
1817 }
1818
bd513cd0
MV		 1819 key = __bpf_copy_key(ukey, map->key_size);
1820 if (IS_ERR(key)) {
1821 err = PTR_ERR(key);
1822 goto err_put;
1823 }
1824
3e87f192 1825 value_size = bpf_map_value_size(map);
bd513cd0
MV		 1826
1827 err = -ENOMEM;
f0dce1d9 1828 value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
bd513cd0
MV		 1829 if (!value)
1830 goto free_key;
1831
3e87f192 1832 err = -ENOTSUPP;
bd513cd0
MV		 1833 if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1834 map->map_type == BPF_MAP_TYPE_STACK) {
1835 err = map->ops->map_pop_elem(map, value);
3e87f192
DS		 1836 } else if (map->map_type == BPF_MAP_TYPE_HASH ||
1837 map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
1838 map->map_type == BPF_MAP_TYPE_LRU_HASH ||
1839 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
1840 if (!bpf_map_is_dev_bound(map)) {
1841 bpf_disable_instrumentation();
1842 rcu_read_lock();
1843 err = map->ops->map_lookup_and_delete_elem(map, key, value, attr->flags);
1844 rcu_read_unlock();
1845 bpf_enable_instrumentation();
1846 }
bd513cd0
MV		 1847 }
1848
1849 if (err)
1850 goto free_value;
1851
7f645462
WY		 1852 if (copy_to_user(uvalue, value, value_size) != 0) {
1853 err = -EFAULT;
bd513cd0 1854 goto free_value;
7f645462 1855 }
bd513cd0
MV		 1856
1857 err = 0;
1858
1859free_value:
f0dce1d9 1860 kvfree(value);
bd513cd0 1861free_key:
44779a4b 1862 kvfree(key);
bd513cd0 1863err_put:
353050be 1864 bpf_map_write_active_dec(map);
bd513cd0
MV		 1865 fdput(f);
1866 return err;
1867}
1868
87df15de
DB		 1869#define BPF_MAP_FREEZE_LAST_FIELD map_fd
1870
1871static int map_freeze(const union bpf_attr *attr)
1872{
1873 int err = 0, ufd = attr->map_fd;
1874 struct bpf_map *map;
1875 struct fd f;
1876
1877 if (CHECK_ATTR(BPF_MAP_FREEZE))
1878 return -EINVAL;
1879
1880 f = fdget(ufd);
1881 map = __bpf_map_get(f);
1882 if (IS_ERR(map))
1883 return PTR_ERR(map);
fc970227 1884
68134668 1885 if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS ||
61df10c7 1886 map_value_has_timer(map) || map_value_has_kptrs(map)) {
849b4d94
MKL		 1887 fdput(f);
1888 return -ENOTSUPP;
1889 }
1890
fc970227 1891 mutex_lock(&map->freeze_mutex);
353050be 1892 if (bpf_map_write_active(map)) {
fc970227
AN		 1893 err = -EBUSY;
1894 goto err_put;
1895 }
87df15de
DB		 1896 if (READ_ONCE(map->frozen)) {
1897 err = -EBUSY;
1898 goto err_put;
1899 }
2c78ee89 1900 if (!bpf_capable()) {
87df15de
DB		 1901 err = -EPERM;
1902 goto err_put;
1903 }
1904
1905 WRITE_ONCE(map->frozen, true);
1906err_put:
fc970227 1907 mutex_unlock(&map->freeze_mutex);
87df15de
DB		 1908 fdput(f);
1909 return err;
1910}
1911
7de16e3a 1912static const struct bpf_prog_ops * const bpf_prog_types[] = {
91cc1a99 1913#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
7de16e3a
JK		 1914 [_id] = & _name ## _prog_ops,
1915#define BPF_MAP_TYPE(_id, _ops)
f2e10bff 1916#define BPF_LINK_TYPE(_id, _name)
7de16e3a
JK		 1917#include <linux/bpf_types.h>
1918#undef BPF_PROG_TYPE
1919#undef BPF_MAP_TYPE
f2e10bff 1920#undef BPF_LINK_TYPE
7de16e3a
JK		 1921};
1922
09756af4
AS		 1923static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1924{
d0f1a451
DB		 1925 const struct bpf_prog_ops *ops;
1926
1927 if (type >= ARRAY_SIZE(bpf_prog_types))
1928 return -EINVAL;
1929 type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1930 ops = bpf_prog_types[type];
1931 if (!ops)
be9370a7 1932 return -EINVAL;
09756af4 1933
ab3f0063 1934 if (!bpf_prog_is_dev_bound(prog->aux))
d0f1a451 1935 prog->aux->ops = ops;
ab3f0063
JK		 1936 else
1937 prog->aux->ops = &bpf_offload_prog_ops;
be9370a7
JB		 1938 prog->type = type;
1939 return 0;
09756af4
AS		 1940}
1941
bae141f5
DB		 1942enum bpf_audit {
1943 BPF_AUDIT_LOAD,
1944 BPF_AUDIT_UNLOAD,
1945 BPF_AUDIT_MAX,
1946};
1947
1948static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
1949 [BPF_AUDIT_LOAD] = "LOAD",
1950 [BPF_AUDIT_UNLOAD] = "UNLOAD",
1951};
1952
1953static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
1954{
1955 struct audit_context *ctx = NULL;
1956 struct audit_buffer *ab;
1957
1958 if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
1959 return;
1960 if (audit_enabled == AUDIT_OFF)
1961 return;
1962 if (op == BPF_AUDIT_LOAD)
1963 ctx = audit_context();
1964 ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
1965 if (unlikely(!ab))
1966 return;
1967 audit_log_format(ab, "prog-id=%u op=%s",
1968 prog->aux->id, bpf_audit_str[op]);
1969 audit_log_end(ab);
1970}
1971
dc4bb0e2
MKL		 1972static int bpf_prog_alloc_id(struct bpf_prog *prog)
1973{
1974 int id;
1975
b76354cd 1976 idr_preload(GFP_KERNEL);
dc4bb0e2
MKL		 1977 spin_lock_bh(&prog_idr_lock);
1978 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1979 if (id > 0)
1980 prog->aux->id = id;
1981 spin_unlock_bh(&prog_idr_lock);
b76354cd 1982 idr_preload_end();
dc4bb0e2
MKL		 1983
1984 /* id is in [1, INT_MAX) */
1985 if (WARN_ON_ONCE(!id))
1986 return -ENOSPC;
1987
1988 return id > 0 ? 0 : id;
1989}
1990
ad8ad79f 1991void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
dc4bb0e2 1992{
d809e134
AS		 1993 unsigned long flags;
1994
ad8ad79f
JK		 1995 /* cBPF to eBPF migrations are currently not in the idr store.
1996 * Offloaded programs are removed from the store when their device
1997 * disappears - even if someone grabs an fd to them they are unusable,
1998 * simply waiting for refcnt to drop to be freed.
1999 */
dc4bb0e2
MKL		 2000 if (!prog->aux->id)
2001 return;
2002
b16d9aa4 2003 if (do_idr_lock)
d809e134 2004 spin_lock_irqsave(&prog_idr_lock, flags);
b16d9aa4
MKL		 2005 else
2006 __acquire(&prog_idr_lock);
2007
dc4bb0e2 2008 idr_remove(&prog_idr, prog->aux->id);
ad8ad79f 2009 prog->aux->id = 0;
b16d9aa4
MKL		 2010
2011 if (do_idr_lock)
d809e134 2012 spin_unlock_irqrestore(&prog_idr_lock, flags);
b16d9aa4
MKL		 2013 else
2014 __release(&prog_idr_lock);
dc4bb0e2
MKL		 2015}
2016
1aacde3d 2017static void __bpf_prog_put_rcu(struct rcu_head *rcu)
abf2e7d6
AS		 2018{
2019 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
2020
3b4d9eb2 2021 kvfree(aux->func_info);
8c1b6e69 2022 kfree(aux->func_info_aux);
3ac1f01b 2023 free_uid(aux->user);
afdb09c7 2024 security_bpf_prog_free(aux);
abf2e7d6
AS		 2025 bpf_prog_free(aux->prog);
2026}
2027
cd7455f1
DB		 2028static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
2029{
2030 bpf_prog_kallsyms_del_all(prog);
2031 btf_put(prog->aux->btf);
e16301fb
MKL		 2032 kvfree(prog->aux->jited_linfo);
2033 kvfree(prog->aux->linfo);
e6ac2450 2034 kfree(prog->aux->kfunc_tab);
22dc4a0f
AN		 2035 if (prog->aux->attach_btf)
2036 btf_put(prog->aux->attach_btf);
cd7455f1 2037
1e6c62a8
AS		 2038 if (deferred) {
2039 if (prog->aux->sleepable)
2040 call_rcu_tasks_trace(&prog->aux->rcu, __bpf_prog_put_rcu);
2041 else
2042 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
2043 } else {
cd7455f1 2044 __bpf_prog_put_rcu(&prog->aux->rcu);
1e6c62a8 2045 }
cd7455f1
DB		 2046}
2047
d809e134
AS		 2048static void bpf_prog_put_deferred(struct work_struct *work)
2049{
2050 struct bpf_prog_aux *aux;
2051 struct bpf_prog *prog;
2052
2053 aux = container_of(work, struct bpf_prog_aux, work);
2054 prog = aux->prog;
2055 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
2056 bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
2057 __bpf_prog_put_noref(prog, true);
2058}
2059
b16d9aa4 2060static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
09756af4 2061{
d809e134
AS		 2062 struct bpf_prog_aux *aux = prog->aux;
2063
2064 if (atomic64_dec_and_test(&aux->refcnt)) {
34ad5580 2065 /* bpf_prog_free_id() must be called first */
b16d9aa4 2066 bpf_prog_free_id(prog, do_idr_lock);
d809e134
AS		 2067
2068 if (in_irq() || irqs_disabled()) {
2069 INIT_WORK(&aux->work, bpf_prog_put_deferred);
2070 schedule_work(&aux->work);
2071 } else {
2072 bpf_prog_put_deferred(&aux->work);
2073 }
a67edbf4 2074 }
09756af4 2075}
b16d9aa4
MKL		 2076
2077void bpf_prog_put(struct bpf_prog *prog)
2078{
2079 __bpf_prog_put(prog, true);
2080}
e2e9b654 2081EXPORT_SYMBOL_GPL(bpf_prog_put);
09756af4
AS		 2082
2083static int bpf_prog_release(struct inode *inode, struct file *filp)
2084{
2085 struct bpf_prog *prog = filp->private_data;
2086
1aacde3d 2087 bpf_prog_put(prog);
09756af4
AS		 2088 return 0;
2089}
2090
61a0abae
ED		 2091struct bpf_prog_kstats {
2092 u64 nsecs;
2093 u64 cnt;
2094 u64 misses;
2095};
2096
492ecee8 2097static void bpf_prog_get_stats(const struct bpf_prog *prog,
61a0abae 2098 struct bpf_prog_kstats *stats)
492ecee8 2099{
9ed9e9ba 2100 u64 nsecs = 0, cnt = 0, misses = 0;
492ecee8
AS		 2101 int cpu;
2102
2103 for_each_possible_cpu(cpu) {
2104 const struct bpf_prog_stats *st;
2105 unsigned int start;
9ed9e9ba 2106 u64 tnsecs, tcnt, tmisses;
492ecee8 2107
700d4796 2108 st = per_cpu_ptr(prog->stats, cpu);
492ecee8
AS		 2109 do {
2110 start = u64_stats_fetch_begin_irq(&st->syncp);
61a0abae
ED		 2111 tnsecs = u64_stats_read(&st->nsecs);
2112 tcnt = u64_stats_read(&st->cnt);
2113 tmisses = u64_stats_read(&st->misses);
492ecee8
AS		 2114 } while (u64_stats_fetch_retry_irq(&st->syncp, start));
2115 nsecs += tnsecs;
2116 cnt += tcnt;
9ed9e9ba 2117 misses += tmisses;
492ecee8
AS		 2118 }
2119 stats->nsecs = nsecs;
2120 stats->cnt = cnt;
9ed9e9ba 2121 stats->misses = misses;
492ecee8
AS		 2122}
2123
7bd509e3
DB		 2124#ifdef CONFIG_PROC_FS
2125static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
2126{
2127 const struct bpf_prog *prog = filp->private_data;
f1f7714e 2128 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
61a0abae 2129 struct bpf_prog_kstats stats;
7bd509e3 2130
492ecee8 2131 bpf_prog_get_stats(prog, &stats);
f1f7714e 2132 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
7bd509e3
DB		 2133 seq_printf(m,
2134 "prog_type:\t%u\n"
2135 "prog_jited:\t%u\n"
f1f7714e 2136 "prog_tag:\t%s\n"
4316b409 2137 "memlock:\t%llu\n"
492ecee8
AS		 2138 "prog_id:\t%u\n"
2139 "run_time_ns:\t%llu\n"
9ed9e9ba 2140 "run_cnt:\t%llu\n"
aba64c7d
DM		 2141 "recursion_misses:\t%llu\n"
2142 "verified_insns:\t%u\n",
7bd509e3
DB		 2143 prog->type,
2144 prog->jited,
f1f7714e 2145 prog_tag,
4316b409 2146 prog->pages * 1ULL << PAGE_SHIFT,
492ecee8
AS		 2147 prog->aux->id,
2148 stats.nsecs,
9ed9e9ba 2149 stats.cnt,
aba64c7d
DM		 2150 stats.misses,
2151 prog->aux->verified_insns);
7bd509e3
DB		 2152}
2153#endif
2154
f66e448c 2155const struct file_operations bpf_prog_fops = {
7bd509e3
DB		 2156#ifdef CONFIG_PROC_FS
2157 .show_fdinfo = bpf_prog_show_fdinfo,
2158#endif
2159 .release = bpf_prog_release,
6e71b04a
CF		 2160 .read = bpf_dummy_read,
2161 .write = bpf_dummy_write,
09756af4
AS		 2162};
2163
b2197755 2164int bpf_prog_new_fd(struct bpf_prog *prog)
aa79781b 2165{
afdb09c7
CF		 2166 int ret;
2167
2168 ret = security_bpf_prog(prog);
2169 if (ret < 0)
2170 return ret;
2171
aa79781b
DB		 2172 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
2173 O_RDWR | O_CLOEXEC);
2174}
2175
113214be 2176static struct bpf_prog *____bpf_prog_get(struct fd f)
09756af4 2177{
09756af4
AS		 2178 if (!f.file)
2179 return ERR_PTR(-EBADF);
09756af4
AS		 2180 if (f.file->f_op != &bpf_prog_fops) {
2181 fdput(f);
2182 return ERR_PTR(-EINVAL);
2183 }
2184
c2101297 2185 return f.file->private_data;
09756af4
AS		 2186}
2187
85192dbf 2188void bpf_prog_add(struct bpf_prog *prog, int i)
92117d84 2189{
85192dbf 2190 atomic64_add(i, &prog->aux->refcnt);
92117d84 2191}
59d3656d
BB		 2192EXPORT_SYMBOL_GPL(bpf_prog_add);
2193
c540594f
DB		 2194void bpf_prog_sub(struct bpf_prog *prog, int i)
2195{
2196 /* Only to be used for undoing previous bpf_prog_add() in some
2197 * error path. We still know that another entity in our call
2198 * path holds a reference to the program, thus atomic_sub() can
2199 * be safely used in such cases!
2200 */
85192dbf 2201 WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
c540594f
DB		 2202}
2203EXPORT_SYMBOL_GPL(bpf_prog_sub);
2204
85192dbf 2205void bpf_prog_inc(struct bpf_prog *prog)
59d3656d 2206{
85192dbf 2207 atomic64_inc(&prog->aux->refcnt);
59d3656d 2208}
97bc402d 2209EXPORT_SYMBOL_GPL(bpf_prog_inc);
92117d84 2210
b16d9aa4 2211/* prog_idr_lock should have been held */
a6f6df69 2212struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
b16d9aa4
MKL		 2213{
2214 int refold;
2215
85192dbf 2216 refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
b16d9aa4
MKL		 2217
2218 if (!refold)
2219 return ERR_PTR(-ENOENT);
2220
2221 return prog;
2222}
a6f6df69 2223EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
b16d9aa4 2224
040ee692 2225bool bpf_prog_get_ok(struct bpf_prog *prog,
288b3de5 2226 enum bpf_prog_type *attach_type, bool attach_drv)
248f346f 2227{
288b3de5
JK		 2228 /* not an attachment, just a refcount inc, always allow */
2229 if (!attach_type)
2230 return true;
248f346f
JK		 2231
2232 if (prog->type != *attach_type)
2233 return false;
288b3de5 2234 if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
248f346f
JK		 2235 return false;
2236
2237 return true;
2238}
2239
2240static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
288b3de5 2241 bool attach_drv)
09756af4
AS		 2242{
2243 struct fd f = fdget(ufd);
2244 struct bpf_prog *prog;
2245
113214be 2246 prog = ____bpf_prog_get(f);
09756af4
AS		 2247 if (IS_ERR(prog))
2248 return prog;
288b3de5 2249 if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
113214be
DB		 2250 prog = ERR_PTR(-EINVAL);
2251 goto out;
2252 }
09756af4 2253
85192dbf 2254 bpf_prog_inc(prog);
113214be 2255out:
09756af4
AS		 2256 fdput(f);
2257 return prog;
2258}
113214be
DB		 2259
2260struct bpf_prog *bpf_prog_get(u32 ufd)
2261{
288b3de5 2262 return __bpf_prog_get(ufd, NULL, false);
113214be
DB		 2263}
2264
248f346f 2265struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
288b3de5 2266 bool attach_drv)
248f346f 2267{
4d220ed0 2268 return __bpf_prog_get(ufd, &type, attach_drv);
248f346f 2269}
6c8dfe21 2270EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
248f346f 2271
aac3fc32
AI		 2272/* Initially all BPF programs could be loaded w/o specifying
2273 * expected_attach_type. Later for some of them specifying expected_attach_type
2274 * at load time became required so that program could be validated properly.
2275 * Programs of types that are allowed to be loaded both w/ and w/o (for
2276 * backward compatibility) expected_attach_type, should have the default attach
2277 * type assigned to expected_attach_type for the latter case, so that it can be
2278 * validated later at attach time.
2279 *
2280 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
2281 * prog type requires it but has some attach types that have to be backward
2282 * compatible.
2283 */
2284static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
2285{
2286 switch (attr->prog_type) {
2287 case BPF_PROG_TYPE_CGROUP_SOCK:
2288 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
2289 * exist so checking for non-zero is the way to go here.
2290 */
2291 if (!attr->expected_attach_type)
2292 attr->expected_attach_type =
2293 BPF_CGROUP_INET_SOCK_CREATE;
2294 break;
d5e4ddae
KI		 2295 case BPF_PROG_TYPE_SK_REUSEPORT:
2296 if (!attr->expected_attach_type)
2297 attr->expected_attach_type =
2298 BPF_SK_REUSEPORT_SELECT;
2299 break;
aac3fc32
AI		 2300 }
2301}
2302
5e43f899 2303static int
ccfe29eb
AS		 2304bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
2305 enum bpf_attach_type expected_attach_type,
290248a5
AN		 2306 struct btf *attach_btf, u32 btf_id,
2307 struct bpf_prog *dst_prog)
5e43f899 2308{
27ae7997 2309 if (btf_id) {
c108e3c1
AS		 2310 if (btf_id > BTF_MAX_TYPE)
2311 return -EINVAL;
27ae7997 2312
290248a5
AN		 2313 if (!attach_btf && !dst_prog)
2314 return -EINVAL;
2315
27ae7997
MKL		 2316 switch (prog_type) {
2317 case BPF_PROG_TYPE_TRACING:
9e4e01df 2318 case BPF_PROG_TYPE_LSM:
27ae7997 2319 case BPF_PROG_TYPE_STRUCT_OPS:
be8704ff 2320 case BPF_PROG_TYPE_EXT:
27ae7997
MKL		 2321 break;
2322 default:
c108e3c1 2323 return -EINVAL;
27ae7997 2324 }
c108e3c1
AS		 2325 }
2326
290248a5
AN		 2327 if (attach_btf && (!btf_id || dst_prog))
2328 return -EINVAL;
2329
2330 if (dst_prog && prog_type != BPF_PROG_TYPE_TRACING &&
be8704ff 2331 prog_type != BPF_PROG_TYPE_EXT)
27ae7997
MKL		 2332 return -EINVAL;
2333
4fbac77d 2334 switch (prog_type) {
aac3fc32
AI		 2335 case BPF_PROG_TYPE_CGROUP_SOCK:
2336 switch (expected_attach_type) {
2337 case BPF_CGROUP_INET_SOCK_CREATE:
f5836749 2338 case BPF_CGROUP_INET_SOCK_RELEASE:
aac3fc32
AI		 2339 case BPF_CGROUP_INET4_POST_BIND:
2340 case BPF_CGROUP_INET6_POST_BIND:
2341 return 0;
2342 default:
2343 return -EINVAL;
2344 }
4fbac77d
AI		 2345 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2346 switch (expected_attach_type) {
2347 case BPF_CGROUP_INET4_BIND:
2348 case BPF_CGROUP_INET6_BIND:
d74bad4e
AI		 2349 case BPF_CGROUP_INET4_CONNECT:
2350 case BPF_CGROUP_INET6_CONNECT:
1b66d253
DB		 2351 case BPF_CGROUP_INET4_GETPEERNAME:
2352 case BPF_CGROUP_INET6_GETPEERNAME:
2353 case BPF_CGROUP_INET4_GETSOCKNAME:
2354 case BPF_CGROUP_INET6_GETSOCKNAME:
1cedee13
AI		 2355 case BPF_CGROUP_UDP4_SENDMSG:
2356 case BPF_CGROUP_UDP6_SENDMSG:
983695fa
DB		 2357 case BPF_CGROUP_UDP4_RECVMSG:
2358 case BPF_CGROUP_UDP6_RECVMSG:
4fbac77d
AI		 2359 return 0;
2360 default:
2361 return -EINVAL;
2362 }
5cf1e914 2363 case BPF_PROG_TYPE_CGROUP_SKB:
2364 switch (expected_attach_type) {
2365 case BPF_CGROUP_INET_INGRESS:
2366 case BPF_CGROUP_INET_EGRESS:
2367 return 0;
2368 default:
2369 return -EINVAL;
2370 }
0d01da6a
SF		 2371 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2372 switch (expected_attach_type) {
2373 case BPF_CGROUP_SETSOCKOPT:
2374 case BPF_CGROUP_GETSOCKOPT:
2375 return 0;
2376 default:
2377 return -EINVAL;
2378 }
e9ddbb77
JS		 2379 case BPF_PROG_TYPE_SK_LOOKUP:
2380 if (expected_attach_type == BPF_SK_LOOKUP)
2381 return 0;
2382 return -EINVAL;
d5e4ddae
KI		 2383 case BPF_PROG_TYPE_SK_REUSEPORT:
2384 switch (expected_attach_type) {
2385 case BPF_SK_REUSEPORT_SELECT:
2386 case BPF_SK_REUSEPORT_SELECT_OR_MIGRATE:
2387 return 0;
2388 default:
2389 return -EINVAL;
2390 }
79a7f8bd 2391 case BPF_PROG_TYPE_SYSCALL:
be8704ff
AS		 2392 case BPF_PROG_TYPE_EXT:
2393 if (expected_attach_type)
2394 return -EINVAL;
df561f66 2395 fallthrough;
4fbac77d
AI		 2396 default:
2397 return 0;
2398 }
5e43f899
AI		 2399}
2400
2c78ee89
AS		 2401static bool is_net_admin_prog_type(enum bpf_prog_type prog_type)
2402{
2403 switch (prog_type) {
2404 case BPF_PROG_TYPE_SCHED_CLS:
2405 case BPF_PROG_TYPE_SCHED_ACT:
2406 case BPF_PROG_TYPE_XDP:
2407 case BPF_PROG_TYPE_LWT_IN:
2408 case BPF_PROG_TYPE_LWT_OUT:
2409 case BPF_PROG_TYPE_LWT_XMIT:
2410 case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2411 case BPF_PROG_TYPE_SK_SKB:
2412 case BPF_PROG_TYPE_SK_MSG:
2413 case BPF_PROG_TYPE_LIRC_MODE2:
2414 case BPF_PROG_TYPE_FLOW_DISSECTOR:
2415 case BPF_PROG_TYPE_CGROUP_DEVICE:
2416 case BPF_PROG_TYPE_CGROUP_SOCK:
2417 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2418 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2419 case BPF_PROG_TYPE_CGROUP_SYSCTL:
2420 case BPF_PROG_TYPE_SOCK_OPS:
2421 case BPF_PROG_TYPE_EXT: /* extends any prog */
2422 return true;
2423 case BPF_PROG_TYPE_CGROUP_SKB:
2424 /* always unpriv */
2425 case BPF_PROG_TYPE_SK_REUSEPORT:
2426 /* equivalent to SOCKET_FILTER. need CAP_BPF only */
2427 default:
2428 return false;
2429 }
2430}
2431
2432static bool is_perfmon_prog_type(enum bpf_prog_type prog_type)
2433{
2434 switch (prog_type) {
2435 case BPF_PROG_TYPE_KPROBE:
2436 case BPF_PROG_TYPE_TRACEPOINT:
2437 case BPF_PROG_TYPE_PERF_EVENT:
2438 case BPF_PROG_TYPE_RAW_TRACEPOINT:
2439 case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2440 case BPF_PROG_TYPE_TRACING:
2441 case BPF_PROG_TYPE_LSM:
2442 case BPF_PROG_TYPE_STRUCT_OPS: /* has access to struct sock */
2443 case BPF_PROG_TYPE_EXT: /* extends any prog */
2444 return true;
2445 default:
2446 return false;
2447 }
2448}
2449
09756af4 2450/* last field in 'union bpf_attr' used by this command */
fbd94c7a 2451#define BPF_PROG_LOAD_LAST_FIELD core_relo_rec_size
09756af4 2452
af2ac3e1 2453static int bpf_prog_load(union bpf_attr *attr, bpfptr_t uattr)
09756af4
AS		 2454{
2455 enum bpf_prog_type type = attr->prog_type;
290248a5
AN		 2456 struct bpf_prog *prog, *dst_prog = NULL;
2457 struct btf *attach_btf = NULL;
09756af4
AS		 2458 int err;
2459 char license[128];
2460 bool is_gpl;
2461
2462 if (CHECK_ATTR(BPF_PROG_LOAD))
2463 return -EINVAL;
2464
c240eff6
JW		 2465 if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2466 BPF_F_ANY_ALIGNMENT |
10d274e8 2467 BPF_F_TEST_STATE_FREQ |
1e6c62a8 2468 BPF_F_SLEEPABLE |
c2f2cdbe
LB		 2469 BPF_F_TEST_RND_HI32 |
2470 BPF_F_XDP_HAS_FRAGS))
e07b98d9
DM		 2471 return -EINVAL;
2472
e9ee9efc
DM		 2473 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2474 (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2c78ee89 2475 !bpf_capable())
e9ee9efc
DM		 2476 return -EPERM;
2477
09756af4 2478 /* copy eBPF program license from user space */
af2ac3e1
AS		 2479 if (strncpy_from_bpfptr(license,
2480 make_bpfptr(attr->license, uattr.is_kernel),
2481 sizeof(license) - 1) < 0)
09756af4
AS		 2482 return -EFAULT;
2483 license[sizeof(license) - 1] = 0;
2484
2485 /* eBPF programs must be GPL compatible to use GPL-ed functions */
2486 is_gpl = license_is_gpl_compatible(license);
2487
c04c0d2b 2488 if (attr->insn_cnt == 0 ||
2c78ee89 2489 attr->insn_cnt > (bpf_capable() ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
ef0915ca 2490 return -E2BIG;
80b7d819
CF		 2491 if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2492 type != BPF_PROG_TYPE_CGROUP_SKB &&
2c78ee89
AS		 2493 !bpf_capable())
2494 return -EPERM;
2495
b338cb92 2496 if (is_net_admin_prog_type(type) && !capable(CAP_NET_ADMIN) && !capable(CAP_SYS_ADMIN))
2c78ee89
AS		 2497 return -EPERM;
2498 if (is_perfmon_prog_type(type) && !perfmon_capable())
1be7f75d
AS		 2499 return -EPERM;
2500
290248a5
AN		 2501 /* attach_prog_fd/attach_btf_obj_fd can specify fd of either bpf_prog
2502 * or btf, we need to check which one it is
2503 */
2504 if (attr->attach_prog_fd) {
2505 dst_prog = bpf_prog_get(attr->attach_prog_fd);
2506 if (IS_ERR(dst_prog)) {
2507 dst_prog = NULL;
2508 attach_btf = btf_get_by_fd(attr->attach_btf_obj_fd);
2509 if (IS_ERR(attach_btf))
2510 return -EINVAL;
2511 if (!btf_is_kernel(attach_btf)) {
8bdd8e27
AN		 2512 /* attaching through specifying bpf_prog's BTF
2513 * objects directly might be supported eventually
2514 */
290248a5 2515 btf_put(attach_btf);
8bdd8e27 2516 return -ENOTSUPP;
290248a5
AN		 2517 }
2518 }
2519 } else if (attr->attach_btf_id) {
2520 /* fall back to vmlinux BTF, if BTF type ID is specified */
2521 attach_btf = bpf_get_btf_vmlinux();
2522 if (IS_ERR(attach_btf))
2523 return PTR_ERR(attach_btf);
2524 if (!attach_btf)
2525 return -EINVAL;
2526 btf_get(attach_btf);
2527 }
2528
aac3fc32 2529 bpf_prog_load_fixup_attach_type(attr);
ccfe29eb 2530 if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
290248a5
AN		 2531 attach_btf, attr->attach_btf_id,
2532 dst_prog)) {
2533 if (dst_prog)
2534 bpf_prog_put(dst_prog);
2535 if (attach_btf)
2536 btf_put(attach_btf);
5e43f899 2537 return -EINVAL;
290248a5 2538 }
5e43f899 2539
09756af4
AS		 2540 /* plain bpf_prog allocation */
2541 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
290248a5
AN		 2542 if (!prog) {
2543 if (dst_prog)
2544 bpf_prog_put(dst_prog);
2545 if (attach_btf)
2546 btf_put(attach_btf);
09756af4 2547 return -ENOMEM;
290248a5 2548 }
09756af4 2549
5e43f899 2550 prog->expected_attach_type = attr->expected_attach_type;
290248a5 2551 prog->aux->attach_btf = attach_btf;
ccfe29eb 2552 prog->aux->attach_btf_id = attr->attach_btf_id;
290248a5 2553 prog->aux->dst_prog = dst_prog;
9a18eedb 2554 prog->aux->offload_requested = !!attr->prog_ifindex;
1e6c62a8 2555 prog->aux->sleepable = attr->prog_flags & BPF_F_SLEEPABLE;
c2f2cdbe 2556 prog->aux->xdp_has_frags = attr->prog_flags & BPF_F_XDP_HAS_FRAGS;
9a18eedb 2557
afdb09c7 2558 err = security_bpf_prog_alloc(prog->aux);
aaac3ba9 2559 if (err)
3ac1f01b 2560 goto free_prog;
afdb09c7 2561
3ac1f01b 2562 prog->aux->user = get_current_user();
09756af4
AS		 2563 prog->len = attr->insn_cnt;
2564
2565 err = -EFAULT;
af2ac3e1
AS		 2566 if (copy_from_bpfptr(prog->insns,
2567 make_bpfptr(attr->insns, uattr.is_kernel),
2568 bpf_prog_insn_size(prog)) != 0)
3ac1f01b 2569 goto free_prog_sec;
09756af4
AS		 2570
2571 prog->orig_prog = NULL;
a91263d5 2572 prog->jited = 0;
09756af4 2573
85192dbf 2574 atomic64_set(&prog->aux->refcnt, 1);
a91263d5 2575 prog->gpl_compatible = is_gpl ? 1 : 0;
09756af4 2576
9a18eedb 2577 if (bpf_prog_is_dev_bound(prog->aux)) {
ab3f0063
JK		 2578 err = bpf_prog_offload_init(prog, attr);
2579 if (err)
3ac1f01b 2580 goto free_prog_sec;
ab3f0063
JK		 2581 }
2582
09756af4
AS		 2583 /* find program type: socket_filter vs tracing_filter */
2584 err = find_prog_type(type, prog);
2585 if (err < 0)
3ac1f01b 2586 goto free_prog_sec;
09756af4 2587
9285ec4c 2588 prog->aux->load_time = ktime_get_boottime_ns();
8e7ae251
MKL		 2589 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name,
2590 sizeof(attr->prog_name));
2591 if (err < 0)
3ac1f01b 2592 goto free_prog_sec;
cb4d2b3f 2593
09756af4 2594 /* run eBPF verifier */
838e9690 2595 err = bpf_check(&prog, attr, uattr);
09756af4
AS		 2596 if (err < 0)
2597 goto free_used_maps;
2598
9facc336 2599 prog = bpf_prog_select_runtime(prog, &err);
04fd61ab
AS		 2600 if (err < 0)
2601 goto free_used_maps;
09756af4 2602
dc4bb0e2
MKL		 2603 err = bpf_prog_alloc_id(prog);
2604 if (err)
2605 goto free_used_maps;
2606
c751798a
DB		 2607 /* Upon success of bpf_prog_alloc_id(), the BPF prog is
2608 * effectively publicly exposed. However, retrieving via
2609 * bpf_prog_get_fd_by_id() will take another reference,
2610 * therefore it cannot be gone underneath us.
2611 *
2612 * Only for the time /after/ successful bpf_prog_new_fd()
2613 * and before returning to userspace, we might just hold
2614 * one reference and any parallel close on that fd could
2615 * rip everything out. Hence, below notifications must
2616 * happen before bpf_prog_new_fd().
2617 *
2618 * Also, any failure handling from this point onwards must
2619 * be using bpf_prog_put() given the program is exposed.
2620 */
74451e66 2621 bpf_prog_kallsyms_add(prog);
6ee52e2a 2622 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
bae141f5 2623 bpf_audit_prog(prog, BPF_AUDIT_LOAD);
c751798a
DB		 2624
2625 err = bpf_prog_new_fd(prog);
2626 if (err < 0)
2627 bpf_prog_put(prog);
09756af4
AS		 2628 return err;
2629
2630free_used_maps:
cd7455f1
DB		 2631 /* In case we have subprogs, we need to wait for a grace
2632 * period before we can tear down JIT memory since symbols
2633 * are already exposed under kallsyms.
2634 */
2635 __bpf_prog_put_noref(prog, prog->aux->func_cnt);
2636 return err;
afdb09c7 2637free_prog_sec:
3ac1f01b 2638 free_uid(prog->aux->user);
afdb09c7 2639 security_bpf_prog_free(prog->aux);
3ac1f01b 2640free_prog:
22dc4a0f
AN		 2641 if (prog->aux->attach_btf)
2642 btf_put(prog->aux->attach_btf);
09756af4
AS		 2643 bpf_prog_free(prog);
2644 return err;
2645}
2646
6e71b04a 2647#define BPF_OBJ_LAST_FIELD file_flags
b2197755
DB		 2648
2649static int bpf_obj_pin(const union bpf_attr *attr)
2650{
6e71b04a 2651 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
b2197755
DB		 2652 return -EINVAL;
2653
535e7b4b 2654 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
b2197755
DB		 2655}
2656
2657static int bpf_obj_get(const union bpf_attr *attr)
2658{
6e71b04a
CF		 2659 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
2660 attr->file_flags & ~BPF_OBJ_FLAG_MASK)
b2197755
DB		 2661 return -EINVAL;
2662
6e71b04a
CF		 2663 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
2664 attr->file_flags);
b2197755
DB		 2665}
2666
f2e10bff 2667void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
a3b80e10 2668 const struct bpf_link_ops *ops, struct bpf_prog *prog)
fec56f58 2669{
70ed506c 2670 atomic64_set(&link->refcnt, 1);
f2e10bff 2671 link->type = type;
a3b80e10 2672 link->id = 0;
70ed506c
AN		 2673 link->ops = ops;
2674 link->prog = prog;
2675}
2676
a3b80e10
AN		 2677static void bpf_link_free_id(int id)
2678{
2679 if (!id)
2680 return;
2681
2682 spin_lock_bh(&link_idr_lock);
2683 idr_remove(&link_idr, id);
2684 spin_unlock_bh(&link_idr_lock);
2685}
2686
98868668
AN		 2687/* Clean up bpf_link and corresponding anon_inode file and FD. After
2688 * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
a3b80e10
AN		 2689 * anon_inode's release() call. This helper marksbpf_link as
2690 * defunct, releases anon_inode file and puts reserved FD. bpf_prog's refcnt
2691 * is not decremented, it's the responsibility of a calling code that failed
2692 * to complete bpf_link initialization.
98868668 2693 */
a3b80e10 2694void bpf_link_cleanup(struct bpf_link_primer *primer)
babf3164 2695{
a3b80e10
AN		 2696 primer->link->prog = NULL;
2697 bpf_link_free_id(primer->id);
2698 fput(primer->file);
2699 put_unused_fd(primer->fd);
babf3164
AN		 2700}
2701
70ed506c
AN		 2702void bpf_link_inc(struct bpf_link *link)
2703{
2704 atomic64_inc(&link->refcnt);
2705}
2706
2707/* bpf_link_free is guaranteed to be called from process context */
2708static void bpf_link_free(struct bpf_link *link)
2709{
a3b80e10 2710 bpf_link_free_id(link->id);
babf3164
AN		 2711 if (link->prog) {
2712 /* detach BPF program, clean up used resources */
2713 link->ops->release(link);
2714 bpf_prog_put(link->prog);
2715 }
2716 /* free bpf_link and its containing memory */
2717 link->ops->dealloc(link);
70ed506c
AN		 2718}
2719
2720static void bpf_link_put_deferred(struct work_struct *work)
2721{
2722 struct bpf_link *link = container_of(work, struct bpf_link, work);
2723
2724 bpf_link_free(link);
2725}
2726
2727/* bpf_link_put can be called from atomic context, but ensures that resources
2728 * are freed from process context
2729 */
2730void bpf_link_put(struct bpf_link *link)
2731{
2732 if (!atomic64_dec_and_test(&link->refcnt))
2733 return;
2734
f00f2f7f
AS		 2735 if (in_atomic()) {
2736 INIT_WORK(&link->work, bpf_link_put_deferred);
2737 schedule_work(&link->work);
2738 } else {
2739 bpf_link_free(link);
2740 }
70ed506c 2741}
cb80ddc6 2742EXPORT_SYMBOL(bpf_link_put);
70ed506c
AN		 2743
2744static int bpf_link_release(struct inode *inode, struct file *filp)
2745{
2746 struct bpf_link *link = filp->private_data;
2747
2748 bpf_link_put(link);
fec56f58
AS		 2749 return 0;
2750}
2751
70ed506c 2752#ifdef CONFIG_PROC_FS
f2e10bff
AN		 2753#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
2754#define BPF_MAP_TYPE(_id, _ops)
2755#define BPF_LINK_TYPE(_id, _name) [_id] = #_name,
2756static const char *bpf_link_type_strs[] = {
2757 [BPF_LINK_TYPE_UNSPEC] = "<invalid>",
2758#include <linux/bpf_types.h>
2759};
2760#undef BPF_PROG_TYPE
2761#undef BPF_MAP_TYPE
2762#undef BPF_LINK_TYPE
70ed506c
AN		 2763
2764static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
2765{
2766 const struct bpf_link *link = filp->private_data;
2767 const struct bpf_prog *prog = link->prog;
2768 char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
70ed506c
AN		 2769
2770 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2771 seq_printf(m,
2772 "link_type:\t%s\n"
a3b80e10 2773 "link_id:\t%u\n"
70ed506c
AN		 2774 "prog_tag:\t%s\n"
2775 "prog_id:\t%u\n",
f2e10bff 2776 bpf_link_type_strs[link->type],
a3b80e10 2777 link->id,
70ed506c
AN		 2778 prog_tag,
2779 prog->aux->id);
f2e10bff
AN		 2780 if (link->ops->show_fdinfo)
2781 link->ops->show_fdinfo(link, m);
70ed506c
AN		 2782}
2783#endif
2784
6f302bfb 2785static const struct file_operations bpf_link_fops = {
70ed506c
AN		 2786#ifdef CONFIG_PROC_FS
2787 .show_fdinfo = bpf_link_show_fdinfo,
2788#endif
2789 .release = bpf_link_release,
fec56f58
AS		 2790 .read = bpf_dummy_read,
2791 .write = bpf_dummy_write,
2792};
2793
a3b80e10 2794static int bpf_link_alloc_id(struct bpf_link *link)
70ed506c 2795{
a3b80e10 2796 int id;
70ed506c 2797
a3b80e10
AN		 2798 idr_preload(GFP_KERNEL);
2799 spin_lock_bh(&link_idr_lock);
2800 id = idr_alloc_cyclic(&link_idr, link, 1, INT_MAX, GFP_ATOMIC);
2801 spin_unlock_bh(&link_idr_lock);
2802 idr_preload_end();
70ed506c 2803
a3b80e10
AN		 2804 return id;
2805}
2806
2807/* Prepare bpf_link to be exposed to user-space by allocating anon_inode file,
2808 * reserving unused FD and allocating ID from link_idr. This is to be paired
2809 * with bpf_link_settle() to install FD and ID and expose bpf_link to
2810 * user-space, if bpf_link is successfully attached. If not, bpf_link and
2811 * pre-allocated resources are to be freed with bpf_cleanup() call. All the
2812 * transient state is passed around in struct bpf_link_primer.
2813 * This is preferred way to create and initialize bpf_link, especially when
c561d110 2814 * there are complicated and expensive operations in between creating bpf_link
a3b80e10
AN		 2815 * itself and attaching it to BPF hook. By using bpf_link_prime() and
2816 * bpf_link_settle() kernel code using bpf_link doesn't have to perform
2817 * expensive (and potentially failing) roll back operations in a rare case
2818 * that file, FD, or ID can't be allocated.
babf3164 2819 */
a3b80e10 2820int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer)
babf3164
AN		 2821{
2822 struct file *file;
a3b80e10 2823 int fd, id;
babf3164
AN		 2824
2825 fd = get_unused_fd_flags(O_CLOEXEC);
2826 if (fd < 0)
a3b80e10 2827 return fd;
babf3164 2828
babf3164 2829
a3b80e10
AN		 2830 id = bpf_link_alloc_id(link);
2831 if (id < 0) {
2832 put_unused_fd(fd);
a3b80e10
AN		 2833 return id;
2834 }
babf3164
AN		 2835
2836 file = anon_inode_getfile("bpf_link", &bpf_link_fops, link, O_CLOEXEC);
2837 if (IS_ERR(file)) {
138c6767 2838 bpf_link_free_id(id);
babf3164 2839 put_unused_fd(fd);
138c6767 2840 return PTR_ERR(file);
babf3164
AN		 2841 }
2842
a3b80e10
AN		 2843 primer->link = link;
2844 primer->file = file;
2845 primer->fd = fd;
2846 primer->id = id;
2847 return 0;
2848}
2849
2850int bpf_link_settle(struct bpf_link_primer *primer)
2851{
2852 /* make bpf_link fetchable by ID */
2853 spin_lock_bh(&link_idr_lock);
2854 primer->link->id = primer->id;
2855 spin_unlock_bh(&link_idr_lock);
2856 /* make bpf_link fetchable by FD */
2857 fd_install(primer->fd, primer->file);
2858 /* pass through installed FD */
2859 return primer->fd;
2860}
2861
2862int bpf_link_new_fd(struct bpf_link *link)
2863{
2864 return anon_inode_getfd("bpf-link", &bpf_link_fops, link, O_CLOEXEC);
babf3164
AN		 2865}
2866
70ed506c
AN		 2867struct bpf_link *bpf_link_get_from_fd(u32 ufd)
2868{
2869 struct fd f = fdget(ufd);
2870 struct bpf_link *link;
2871
2872 if (!f.file)
2873 return ERR_PTR(-EBADF);
2874 if (f.file->f_op != &bpf_link_fops) {
2875 fdput(f);
2876 return ERR_PTR(-EINVAL);
2877 }
2878
2879 link = f.file->private_data;
2880 bpf_link_inc(link);
2881 fdput(f);
2882
2883 return link;
2884}
cb80ddc6 2885EXPORT_SYMBOL(bpf_link_get_from_fd);
70ed506c 2886
70ed506c 2887static void bpf_tracing_link_release(struct bpf_link *link)
babf3164 2888{
3aac1ead 2889 struct bpf_tracing_link *tr_link =
f7e0beaf 2890 container_of(link, struct bpf_tracing_link, link.link);
3aac1ead 2891
f7e0beaf 2892 WARN_ON_ONCE(bpf_trampoline_unlink_prog(&tr_link->link,
3aac1ead
THJ		 2893 tr_link->trampoline));
2894
2895 bpf_trampoline_put(tr_link->trampoline);
2896
2897 /* tgt_prog is NULL if target is a kernel function */
2898 if (tr_link->tgt_prog)
2899 bpf_prog_put(tr_link->tgt_prog);
babf3164
AN		 2900}
2901
2902static void bpf_tracing_link_dealloc(struct bpf_link *link)
70ed506c
AN		 2903{
2904 struct bpf_tracing_link *tr_link =
f7e0beaf 2905 container_of(link, struct bpf_tracing_link, link.link);
70ed506c 2906
70ed506c
AN		 2907 kfree(tr_link);
2908}
2909
f2e10bff
AN		 2910static void bpf_tracing_link_show_fdinfo(const struct bpf_link *link,
2911 struct seq_file *seq)
2912{
2913 struct bpf_tracing_link *tr_link =
f7e0beaf 2914 container_of(link, struct bpf_tracing_link, link.link);
f2e10bff
AN		 2915
2916 seq_printf(seq,
2917 "attach_type:\t%d\n",
2918 tr_link->attach_type);
2919}
2920
2921static int bpf_tracing_link_fill_link_info(const struct bpf_link *link,
2922 struct bpf_link_info *info)
2923{
2924 struct bpf_tracing_link *tr_link =
f7e0beaf 2925 container_of(link, struct bpf_tracing_link, link.link);
f2e10bff
AN		 2926
2927 info->tracing.attach_type = tr_link->attach_type;
441e8c66
THJ		 2928 bpf_trampoline_unpack_key(tr_link->trampoline->key,
2929 &info->tracing.target_obj_id,
2930 &info->tracing.target_btf_id);
f2e10bff
AN		 2931
2932 return 0;
2933}
2934
70ed506c
AN		 2935static const struct bpf_link_ops bpf_tracing_link_lops = {
2936 .release = bpf_tracing_link_release,
babf3164 2937 .dealloc = bpf_tracing_link_dealloc,
f2e10bff
AN		 2938 .show_fdinfo = bpf_tracing_link_show_fdinfo,
2939 .fill_link_info = bpf_tracing_link_fill_link_info,
70ed506c
AN		 2940};
2941
4a1e7c0c
THJ		 2942static int bpf_tracing_prog_attach(struct bpf_prog *prog,
2943 int tgt_prog_fd,
2fcc8241
KFL		 2944 u32 btf_id,
2945 u64 bpf_cookie)
fec56f58 2946{
a3b80e10 2947 struct bpf_link_primer link_primer;
3aac1ead 2948 struct bpf_prog *tgt_prog = NULL;
4a1e7c0c 2949 struct bpf_trampoline *tr = NULL;
70ed506c 2950 struct bpf_tracing_link *link;
4a1e7c0c 2951 u64 key = 0;
a3b80e10 2952 int err;
fec56f58 2953
9e4e01df
KS		 2954 switch (prog->type) {
2955 case BPF_PROG_TYPE_TRACING:
2956 if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
2957 prog->expected_attach_type != BPF_TRACE_FEXIT &&
2958 prog->expected_attach_type != BPF_MODIFY_RETURN) {
2959 err = -EINVAL;
2960 goto out_put_prog;
2961 }
2962 break;
2963 case BPF_PROG_TYPE_EXT:
2964 if (prog->expected_attach_type != 0) {
2965 err = -EINVAL;
2966 goto out_put_prog;
2967 }
2968 break;
2969 case BPF_PROG_TYPE_LSM:
2970 if (prog->expected_attach_type != BPF_LSM_MAC) {
2971 err = -EINVAL;
2972 goto out_put_prog;
2973 }
2974 break;
2975 default:
fec56f58
AS		 2976 err = -EINVAL;
2977 goto out_put_prog;
2978 }
2979
4a1e7c0c
THJ		 2980 if (!!tgt_prog_fd != !!btf_id) {
2981 err = -EINVAL;
2982 goto out_put_prog;
2983 }
2984
2985 if (tgt_prog_fd) {
2986 /* For now we only allow new targets for BPF_PROG_TYPE_EXT */
2987 if (prog->type != BPF_PROG_TYPE_EXT) {
2988 err = -EINVAL;
2989 goto out_put_prog;
2990 }
2991
2992 tgt_prog = bpf_prog_get(tgt_prog_fd);
2993 if (IS_ERR(tgt_prog)) {
2994 err = PTR_ERR(tgt_prog);
2995 tgt_prog = NULL;
2996 goto out_put_prog;
2997 }
2998
22dc4a0f 2999 key = bpf_trampoline_compute_key(tgt_prog, NULL, btf_id);
4a1e7c0c
THJ		 3000 }
3001
70ed506c
AN		 3002 link = kzalloc(sizeof(*link), GFP_USER);
3003 if (!link) {
3004 err = -ENOMEM;
3005 goto out_put_prog;
3006 }
f7e0beaf 3007 bpf_link_init(&link->link.link, BPF_LINK_TYPE_TRACING,
f2e10bff
AN		 3008 &bpf_tracing_link_lops, prog);
3009 link->attach_type = prog->expected_attach_type;
2fcc8241 3010 link->link.cookie = bpf_cookie;
70ed506c 3011
3aac1ead
THJ		 3012 mutex_lock(&prog->aux->dst_mutex);
3013
4a1e7c0c
THJ		 3014 /* There are a few possible cases here:
3015 *
3016 * - if prog->aux->dst_trampoline is set, the program was just loaded
3017 * and not yet attached to anything, so we can use the values stored
3018 * in prog->aux
3019 *
3020 * - if prog->aux->dst_trampoline is NULL, the program has already been
3021 * attached to a target and its initial target was cleared (below)
3022 *
3023 * - if tgt_prog != NULL, the caller specified tgt_prog_fd +
3024 * target_btf_id using the link_create API.
3025 *
3026 * - if tgt_prog == NULL when this function was called using the old
f3a95075
JO		 3027 * raw_tracepoint_open API, and we need a target from prog->aux
3028 *
3029 * - if prog->aux->dst_trampoline and tgt_prog is NULL, the program
3030 * was detached and is going for re-attachment.
4a1e7c0c
THJ		 3031 */
3032 if (!prog->aux->dst_trampoline && !tgt_prog) {
f3a95075
JO		 3033 /*
3034 * Allow re-attach for TRACING and LSM programs. If it's
3035 * currently linked, bpf_trampoline_link_prog will fail.
3036 * EXT programs need to specify tgt_prog_fd, so they
3037 * re-attach in separate code path.
3038 */
3039 if (prog->type != BPF_PROG_TYPE_TRACING &&
3040 prog->type != BPF_PROG_TYPE_LSM) {
3041 err = -EINVAL;
3042 goto out_unlock;
3043 }
3044 btf_id = prog->aux->attach_btf_id;
3045 key = bpf_trampoline_compute_key(NULL, prog->aux->attach_btf, btf_id);
babf3164 3046 }
4a1e7c0c
THJ		 3047
3048 if (!prog->aux->dst_trampoline ||
3049 (key && key != prog->aux->dst_trampoline->key)) {
3050 /* If there is no saved target, or the specified target is
3051 * different from the destination specified at load time, we
3052 * need a new trampoline and a check for compatibility
3053 */
3054 struct bpf_attach_target_info tgt_info = {};
3055
3056 err = bpf_check_attach_target(NULL, prog, tgt_prog, btf_id,
3057 &tgt_info);
3058 if (err)
3059 goto out_unlock;
3060
3061 tr = bpf_trampoline_get(key, &tgt_info);
3062 if (!tr) {
3063 err = -ENOMEM;
3064 goto out_unlock;
3065 }
3066 } else {
3067 /* The caller didn't specify a target, or the target was the
3068 * same as the destination supplied during program load. This
3069 * means we can reuse the trampoline and reference from program
3070 * load time, and there is no need to allocate a new one. This
3071 * can only happen once for any program, as the saved values in
3072 * prog->aux are cleared below.
3073 */
3074 tr = prog->aux->dst_trampoline;
3075 tgt_prog = prog->aux->dst_prog;
3076 }
3aac1ead 3077
f7e0beaf 3078 err = bpf_link_prime(&link->link.link, &link_primer);
3aac1ead
THJ		 3079 if (err)
3080 goto out_unlock;
fec56f58 3081
f7e0beaf 3082 err = bpf_trampoline_link_prog(&link->link, tr);
babf3164 3083 if (err) {
a3b80e10 3084 bpf_link_cleanup(&link_primer);
3aac1ead
THJ		 3085 link = NULL;
3086 goto out_unlock;
fec56f58 3087 }
babf3164 3088
3aac1ead
THJ		 3089 link->tgt_prog = tgt_prog;
3090 link->trampoline = tr;
3091
4a1e7c0c
THJ		 3092 /* Always clear the trampoline and target prog from prog->aux to make
3093 * sure the original attach destination is not kept alive after a
3094 * program is (re-)attached to another target.
3095 */
3096 if (prog->aux->dst_prog &&
3097 (tgt_prog_fd || tr != prog->aux->dst_trampoline))
3098 /* got extra prog ref from syscall, or attaching to different prog */
3099 bpf_prog_put(prog->aux->dst_prog);
3100 if (prog->aux->dst_trampoline && tr != prog->aux->dst_trampoline)
3101 /* we allocated a new trampoline, so free the old one */
3102 bpf_trampoline_put(prog->aux->dst_trampoline);
3103
3aac1ead
THJ		 3104 prog->aux->dst_prog = NULL;
3105 prog->aux->dst_trampoline = NULL;
3106 mutex_unlock(&prog->aux->dst_mutex);
3107
a3b80e10 3108 return bpf_link_settle(&link_primer);
3aac1ead 3109out_unlock:
4a1e7c0c
THJ		 3110 if (tr && tr != prog->aux->dst_trampoline)
3111 bpf_trampoline_put(tr);
3aac1ead
THJ		 3112 mutex_unlock(&prog->aux->dst_mutex);
3113 kfree(link);
fec56f58 3114out_put_prog:
4a1e7c0c
THJ		 3115 if (tgt_prog_fd && tgt_prog)
3116 bpf_prog_put(tgt_prog);
fec56f58
AS		 3117 return err;
3118}
3119
70ed506c
AN		 3120struct bpf_raw_tp_link {
3121 struct bpf_link link;
c4f6699d 3122 struct bpf_raw_event_map *btp;
c4f6699d
AS		 3123};
3124
70ed506c 3125static void bpf_raw_tp_link_release(struct bpf_link *link)
c4f6699d 3126{
70ed506c
AN		 3127 struct bpf_raw_tp_link *raw_tp =
3128 container_of(link, struct bpf_raw_tp_link, link);
c4f6699d 3129
70ed506c 3130 bpf_probe_unregister(raw_tp->btp, raw_tp->link.prog);
a38d1107 3131 bpf_put_raw_tracepoint(raw_tp->btp);
babf3164
AN		 3132}
3133
3134static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
3135{
3136 struct bpf_raw_tp_link *raw_tp =
3137 container_of(link, struct bpf_raw_tp_link, link);
3138
c4f6699d 3139 kfree(raw_tp);
c4f6699d
AS		 3140}
3141
f2e10bff
AN		 3142static void bpf_raw_tp_link_show_fdinfo(const struct bpf_link *link,
3143 struct seq_file *seq)
3144{
3145 struct bpf_raw_tp_link *raw_tp_link =
3146 container_of(link, struct bpf_raw_tp_link, link);
3147
3148 seq_printf(seq,
3149 "tp_name:\t%s\n",
3150 raw_tp_link->btp->tp->name);
3151}
3152
3153static int bpf_raw_tp_link_fill_link_info(const struct bpf_link *link,
3154 struct bpf_link_info *info)
3155{
3156 struct bpf_raw_tp_link *raw_tp_link =
3157 container_of(link, struct bpf_raw_tp_link, link);
3158 char __user *ubuf = u64_to_user_ptr(info->raw_tracepoint.tp_name);
3159 const char *tp_name = raw_tp_link->btp->tp->name;
3160 u32 ulen = info->raw_tracepoint.tp_name_len;
3161 size_t tp_len = strlen(tp_name);
3162
b474959d 3163 if (!ulen ^ !ubuf)
f2e10bff
AN		 3164 return -EINVAL;
3165
3166 info->raw_tracepoint.tp_name_len = tp_len + 1;
3167
3168 if (!ubuf)
3169 return 0;
3170
3171 if (ulen >= tp_len + 1) {
3172 if (copy_to_user(ubuf, tp_name, tp_len + 1))
3173 return -EFAULT;
3174 } else {
3175 char zero = '\0';
3176
3177 if (copy_to_user(ubuf, tp_name, ulen - 1))
3178 return -EFAULT;
3179 if (put_user(zero, ubuf + ulen - 1))
3180 return -EFAULT;
3181 return -ENOSPC;
3182 }
3183
3184 return 0;
3185}
3186
a3b80e10 3187static const struct bpf_link_ops bpf_raw_tp_link_lops = {
70ed506c 3188 .release = bpf_raw_tp_link_release,
babf3164 3189 .dealloc = bpf_raw_tp_link_dealloc,
f2e10bff
AN		 3190 .show_fdinfo = bpf_raw_tp_link_show_fdinfo,
3191 .fill_link_info = bpf_raw_tp_link_fill_link_info,
c4f6699d
AS		 3192};
3193
b89fbfbb
AN		 3194#ifdef CONFIG_PERF_EVENTS
3195struct bpf_perf_link {
3196 struct bpf_link link;
3197 struct file *perf_file;
3198};
3199
3200static void bpf_perf_link_release(struct bpf_link *link)
3201{
3202 struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3203 struct perf_event *event = perf_link->perf_file->private_data;
3204
3205 perf_event_free_bpf_prog(event);
3206 fput(perf_link->perf_file);
3207}
3208
3209static void bpf_perf_link_dealloc(struct bpf_link *link)
3210{
3211 struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3212
3213 kfree(perf_link);
3214}
3215
3216static const struct bpf_link_ops bpf_perf_link_lops = {
3217 .release = bpf_perf_link_release,
3218 .dealloc = bpf_perf_link_dealloc,
3219};
3220
3221static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3222{
3223 struct bpf_link_primer link_primer;
3224 struct bpf_perf_link *link;
3225 struct perf_event *event;
3226 struct file *perf_file;
3227 int err;
3228
3229 if (attr->link_create.flags)
3230 return -EINVAL;
3231
3232 perf_file = perf_event_get(attr->link_create.target_fd);
3233 if (IS_ERR(perf_file))
3234 return PTR_ERR(perf_file);
3235
3236 link = kzalloc(sizeof(*link), GFP_USER);
3237 if (!link) {
3238 err = -ENOMEM;
3239 goto out_put_file;
3240 }
3241 bpf_link_init(&link->link, BPF_LINK_TYPE_PERF_EVENT, &bpf_perf_link_lops, prog);
3242 link->perf_file = perf_file;
3243
3244 err = bpf_link_prime(&link->link, &link_primer);
3245 if (err) {
3246 kfree(link);
3247 goto out_put_file;
3248 }
3249
3250 event = perf_file->private_data;
82e6b1ee 3251 err = perf_event_set_bpf_prog(event, prog, attr->link_create.perf_event.bpf_cookie);
b89fbfbb
AN		 3252 if (err) {
3253 bpf_link_cleanup(&link_primer);
3254 goto out_put_file;
3255 }
3256 /* perf_event_set_bpf_prog() doesn't take its own refcnt on prog */
3257 bpf_prog_inc(prog);
3258
3259 return bpf_link_settle(&link_primer);
3260
3261out_put_file:
3262 fput(perf_file);
3263 return err;
3264}
0dcac272
JO		 3265#else
3266static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3267{
3268 return -EOPNOTSUPP;
3269}
b89fbfbb
AN		 3270#endif /* CONFIG_PERF_EVENTS */
3271
df86ca0d
AN		 3272static int bpf_raw_tp_link_attach(struct bpf_prog *prog,
3273 const char __user *user_tp_name)
c4f6699d 3274{
a3b80e10 3275 struct bpf_link_primer link_primer;
babf3164 3276 struct bpf_raw_tp_link *link;
c4f6699d 3277 struct bpf_raw_event_map *btp;
ac4414b5
AS		 3278 const char *tp_name;
3279 char buf[128];
a3b80e10 3280 int err;
c4f6699d 3281
9e4e01df
KS		 3282 switch (prog->type) {
3283 case BPF_PROG_TYPE_TRACING:
3284 case BPF_PROG_TYPE_EXT:
3285 case BPF_PROG_TYPE_LSM:
df86ca0d 3286 if (user_tp_name)
fec56f58
AS		 3287 /* The attach point for this category of programs
3288 * should be specified via btf_id during program load.
ac4414b5 3289 */
df86ca0d 3290 return -EINVAL;
9e4e01df
KS		 3291 if (prog->type == BPF_PROG_TYPE_TRACING &&
3292 prog->expected_attach_type == BPF_TRACE_RAW_TP) {
fec56f58 3293 tp_name = prog->aux->attach_func_name;
9e4e01df
KS		 3294 break;
3295 }
2fcc8241 3296 return bpf_tracing_prog_attach(prog, 0, 0, 0);
9e4e01df
KS		 3297 case BPF_PROG_TYPE_RAW_TRACEPOINT:
3298 case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
df86ca0d
AN		 3299 if (strncpy_from_user(buf, user_tp_name, sizeof(buf) - 1) < 0)
3300 return -EFAULT;
ac4414b5
AS		 3301 buf[sizeof(buf) - 1] = 0;
3302 tp_name = buf;
9e4e01df
KS		 3303 break;
3304 default:
df86ca0d 3305 return -EINVAL;
ac4414b5 3306 }
c4f6699d 3307
a38d1107 3308 btp = bpf_get_raw_tracepoint(tp_name);
df86ca0d
AN		 3309 if (!btp)
3310 return -ENOENT;
c4f6699d 3311
babf3164
AN		 3312 link = kzalloc(sizeof(*link), GFP_USER);
3313 if (!link) {
a38d1107
MM		 3314 err = -ENOMEM;
3315 goto out_put_btp;
3316 }
f2e10bff
AN		 3317 bpf_link_init(&link->link, BPF_LINK_TYPE_RAW_TRACEPOINT,
3318 &bpf_raw_tp_link_lops, prog);
babf3164 3319 link->btp = btp;
c4f6699d 3320
a3b80e10
AN		 3321 err = bpf_link_prime(&link->link, &link_primer);
3322 if (err) {
babf3164 3323 kfree(link);
babf3164
AN		 3324 goto out_put_btp;
3325 }
c4f6699d 3326
babf3164
AN		 3327 err = bpf_probe_register(link->btp, prog);
3328 if (err) {
a3b80e10 3329 bpf_link_cleanup(&link_primer);
babf3164 3330 goto out_put_btp;
c4f6699d 3331 }
babf3164 3332
a3b80e10 3333 return bpf_link_settle(&link_primer);
c4f6699d 3334
a38d1107
MM		 3335out_put_btp:
3336 bpf_put_raw_tracepoint(btp);
c4f6699d
AS		 3337 return err;
3338}
3339
df86ca0d
AN		 3340#define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
3341
3342static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
3343{
3344 struct bpf_prog *prog;
3345 int fd;
3346
3347 if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
3348 return -EINVAL;
3349
3350 prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
3351 if (IS_ERR(prog))
3352 return PTR_ERR(prog);
3353
3354 fd = bpf_raw_tp_link_attach(prog, u64_to_user_ptr(attr->raw_tracepoint.name));
3355 if (fd < 0)
3356 bpf_prog_put(prog);
3357 return fd;
3358}
3359
33491588
AR		 3360static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
3361 enum bpf_attach_type attach_type)
3362{
3363 switch (prog->type) {
3364 case BPF_PROG_TYPE_CGROUP_SOCK:
3365 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
0d01da6a 3366 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
e9ddbb77 3367 case BPF_PROG_TYPE_SK_LOOKUP:
33491588 3368 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
5cf1e914 3369 case BPF_PROG_TYPE_CGROUP_SKB:
2c78ee89
AS		 3370 if (!capable(CAP_NET_ADMIN))
3371 /* cg-skb progs can be loaded by unpriv user.
3372 * check permissions at attach time.
3373 */
3374 return -EPERM;
5cf1e914 3375 return prog->enforce_expected_attach_type &&
3376 prog->expected_attach_type != attach_type ?
3377 -EINVAL : 0;
33491588
AR		 3378 default:
3379 return 0;
3380 }
3381}
3382
e28784e3
AN		 3383static enum bpf_prog_type
3384attach_type_to_prog_type(enum bpf_attach_type attach_type)
f4324551 3385{
e28784e3 3386 switch (attach_type) {
f4324551
DM		 3387 case BPF_CGROUP_INET_INGRESS:
3388 case BPF_CGROUP_INET_EGRESS:
e28784e3 3389 return BPF_PROG_TYPE_CGROUP_SKB;
61023658 3390 case BPF_CGROUP_INET_SOCK_CREATE:
f5836749 3391 case BPF_CGROUP_INET_SOCK_RELEASE:
aac3fc32
AI		 3392 case BPF_CGROUP_INET4_POST_BIND:
3393 case BPF_CGROUP_INET6_POST_BIND:
e28784e3 3394 return BPF_PROG_TYPE_CGROUP_SOCK;
4fbac77d
AI		 3395 case BPF_CGROUP_INET4_BIND:
3396 case BPF_CGROUP_INET6_BIND:
d74bad4e
AI		 3397 case BPF_CGROUP_INET4_CONNECT:
3398 case BPF_CGROUP_INET6_CONNECT:
1b66d253
DB		 3399 case BPF_CGROUP_INET4_GETPEERNAME:
3400 case BPF_CGROUP_INET6_GETPEERNAME:
3401 case BPF_CGROUP_INET4_GETSOCKNAME:
3402 case BPF_CGROUP_INET6_GETSOCKNAME:
1cedee13
AI		 3403 case BPF_CGROUP_UDP4_SENDMSG:
3404 case BPF_CGROUP_UDP6_SENDMSG:
983695fa
DB		 3405 case BPF_CGROUP_UDP4_RECVMSG:
3406 case BPF_CGROUP_UDP6_RECVMSG:
e28784e3 3407 return BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
40304b2a 3408 case BPF_CGROUP_SOCK_OPS:
e28784e3 3409 return BPF_PROG_TYPE_SOCK_OPS;
ebc614f6 3410 case BPF_CGROUP_DEVICE:
e28784e3 3411 return BPF_PROG_TYPE_CGROUP_DEVICE;
4f738adb 3412 case BPF_SK_MSG_VERDICT:
e28784e3 3413 return BPF_PROG_TYPE_SK_MSG;
464bc0fd
JF		 3414 case BPF_SK_SKB_STREAM_PARSER:
3415 case BPF_SK_SKB_STREAM_VERDICT:
a7ba4558 3416 case BPF_SK_SKB_VERDICT:
e28784e3 3417 return BPF_PROG_TYPE_SK_SKB;
f4364dcf 3418 case BPF_LIRC_MODE2:
e28784e3 3419 return BPF_PROG_TYPE_LIRC_MODE2;
d58e468b 3420 case BPF_FLOW_DISSECTOR:
e28784e3 3421 return BPF_PROG_TYPE_FLOW_DISSECTOR;
7b146ceb 3422 case BPF_CGROUP_SYSCTL:
e28784e3 3423 return BPF_PROG_TYPE_CGROUP_SYSCTL;
0d01da6a
SF		 3424 case BPF_CGROUP_GETSOCKOPT:
3425 case BPF_CGROUP_SETSOCKOPT:
e28784e3 3426 return BPF_PROG_TYPE_CGROUP_SOCKOPT;
de4e05ca 3427 case BPF_TRACE_ITER:
df86ca0d
AN		 3428 case BPF_TRACE_RAW_TP:
3429 case BPF_TRACE_FENTRY:
3430 case BPF_TRACE_FEXIT:
3431 case BPF_MODIFY_RETURN:
de4e05ca 3432 return BPF_PROG_TYPE_TRACING;
df86ca0d
AN		 3433 case BPF_LSM_MAC:
3434 return BPF_PROG_TYPE_LSM;
e9ddbb77
JS		 3435 case BPF_SK_LOOKUP:
3436 return BPF_PROG_TYPE_SK_LOOKUP;
aa8d3a71
AN		 3437 case BPF_XDP:
3438 return BPF_PROG_TYPE_XDP;
69fd337a
SF		 3439 case BPF_LSM_CGROUP:
3440 return BPF_PROG_TYPE_LSM;
f4324551 3441 default:
e28784e3 3442 return BPF_PROG_TYPE_UNSPEC;
f4324551 3443 }
e28784e3
AN		 3444}
3445
3446#define BPF_PROG_ATTACH_LAST_FIELD replace_bpf_fd
3447
3448#define BPF_F_ATTACH_MASK \
3449 (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI | BPF_F_REPLACE)
3450
3451static int bpf_prog_attach(const union bpf_attr *attr)
3452{
3453 enum bpf_prog_type ptype;
3454 struct bpf_prog *prog;
3455 int ret;
3456
e28784e3
AN		 3457 if (CHECK_ATTR(BPF_PROG_ATTACH))
3458 return -EINVAL;
3459
3460 if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
3461 return -EINVAL;
3462
3463 ptype = attach_type_to_prog_type(attr->attach_type);
3464 if (ptype == BPF_PROG_TYPE_UNSPEC)
3465 return -EINVAL;
f4324551 3466
b2cd1257
DA		 3467 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
3468 if (IS_ERR(prog))
3469 return PTR_ERR(prog);
3470
5e43f899
AI		 3471 if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
3472 bpf_prog_put(prog);
3473 return -EINVAL;
3474 }
3475
fdb5c453
SY		 3476 switch (ptype) {
3477 case BPF_PROG_TYPE_SK_SKB:
3478 case BPF_PROG_TYPE_SK_MSG:
604326b4 3479 ret = sock_map_get_from_fd(attr, prog);
fdb5c453
SY		 3480 break;
3481 case BPF_PROG_TYPE_LIRC_MODE2:
3482 ret = lirc_prog_attach(attr, prog);
3483 break;
d58e468b 3484 case BPF_PROG_TYPE_FLOW_DISSECTOR:
a3fd7cee 3485 ret = netns_bpf_prog_attach(attr, prog);
d58e468b 3486 break;
e28784e3
AN		 3487 case BPF_PROG_TYPE_CGROUP_DEVICE:
3488 case BPF_PROG_TYPE_CGROUP_SKB:
3489 case BPF_PROG_TYPE_CGROUP_SOCK:
3490 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3491 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3492 case BPF_PROG_TYPE_CGROUP_SYSCTL:
3493 case BPF_PROG_TYPE_SOCK_OPS:
69fd337a
SF		 3494 case BPF_PROG_TYPE_LSM:
3495 if (ptype == BPF_PROG_TYPE_LSM &&
3496 prog->expected_attach_type != BPF_LSM_CGROUP)
3497 return -EINVAL;
3498
fdb5c453 3499 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
e28784e3
AN		 3500 break;
3501 default:
3502 ret = -EINVAL;
b2cd1257
DA		 3503 }
3504
7f677633
AS		 3505 if (ret)
3506 bpf_prog_put(prog);
7f677633 3507 return ret;
f4324551
DM		 3508}
3509
3510#define BPF_PROG_DETACH_LAST_FIELD attach_type
3511
3512static int bpf_prog_detach(const union bpf_attr *attr)
3513{
324bda9e 3514 enum bpf_prog_type ptype;
f4324551 3515
f4324551
DM		 3516 if (CHECK_ATTR(BPF_PROG_DETACH))
3517 return -EINVAL;
3518
e28784e3
AN		 3519 ptype = attach_type_to_prog_type(attr->attach_type);
3520
3521 switch (ptype) {
3522 case BPF_PROG_TYPE_SK_MSG:
3523 case BPF_PROG_TYPE_SK_SKB:
bb0de313 3524 return sock_map_prog_detach(attr, ptype);
e28784e3 3525 case BPF_PROG_TYPE_LIRC_MODE2:
f4364dcf 3526 return lirc_prog_detach(attr);
e28784e3 3527 case BPF_PROG_TYPE_FLOW_DISSECTOR:
4ac2add6 3528 return netns_bpf_prog_detach(attr, ptype);
e28784e3
AN		 3529 case BPF_PROG_TYPE_CGROUP_DEVICE:
3530 case BPF_PROG_TYPE_CGROUP_SKB:
3531 case BPF_PROG_TYPE_CGROUP_SOCK:
3532 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3533 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3534 case BPF_PROG_TYPE_CGROUP_SYSCTL:
3535 case BPF_PROG_TYPE_SOCK_OPS:
69fd337a 3536 case BPF_PROG_TYPE_LSM:
e28784e3 3537 return cgroup_bpf_prog_detach(attr, ptype);
f4324551
DM		 3538 default:
3539 return -EINVAL;
3540 }
f4324551 3541}
40304b2a 3542
b79c9fc9 3543#define BPF_PROG_QUERY_LAST_FIELD query.prog_attach_flags
468e2f64
AS		 3544
3545static int bpf_prog_query(const union bpf_attr *attr,
3546 union bpf_attr __user *uattr)
3547{
468e2f64
AS		 3548 if (!capable(CAP_NET_ADMIN))
3549 return -EPERM;
3550 if (CHECK_ATTR(BPF_PROG_QUERY))
3551 return -EINVAL;
3552 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
3553 return -EINVAL;
3554
3555 switch (attr->query.attach_type) {
3556 case BPF_CGROUP_INET_INGRESS:
3557 case BPF_CGROUP_INET_EGRESS:
3558 case BPF_CGROUP_INET_SOCK_CREATE:
f5836749 3559 case BPF_CGROUP_INET_SOCK_RELEASE:
4fbac77d
AI		 3560 case BPF_CGROUP_INET4_BIND:
3561 case BPF_CGROUP_INET6_BIND:
aac3fc32
AI		 3562 case BPF_CGROUP_INET4_POST_BIND:
3563 case BPF_CGROUP_INET6_POST_BIND:
d74bad4e
AI		 3564 case BPF_CGROUP_INET4_CONNECT:
3565 case BPF_CGROUP_INET6_CONNECT:
1b66d253
DB		 3566 case BPF_CGROUP_INET4_GETPEERNAME:
3567 case BPF_CGROUP_INET6_GETPEERNAME:
3568 case BPF_CGROUP_INET4_GETSOCKNAME:
3569 case BPF_CGROUP_INET6_GETSOCKNAME:
1cedee13
AI		 3570 case BPF_CGROUP_UDP4_SENDMSG:
3571 case BPF_CGROUP_UDP6_SENDMSG:
983695fa
DB		 3572 case BPF_CGROUP_UDP4_RECVMSG:
3573 case BPF_CGROUP_UDP6_RECVMSG:
468e2f64 3574 case BPF_CGROUP_SOCK_OPS:
ebc614f6 3575 case BPF_CGROUP_DEVICE:
7b146ceb 3576 case BPF_CGROUP_SYSCTL:
0d01da6a
SF		 3577 case BPF_CGROUP_GETSOCKOPT:
3578 case BPF_CGROUP_SETSOCKOPT:
b79c9fc9 3579 case BPF_LSM_CGROUP:
e28784e3 3580 return cgroup_bpf_prog_query(attr, uattr);
f4364dcf
SY		 3581 case BPF_LIRC_MODE2:
3582 return lirc_prog_query(attr, uattr);
118c8e9a 3583 case BPF_FLOW_DISSECTOR:
e9ddbb77 3584 case BPF_SK_LOOKUP:
a3fd7cee 3585 return netns_bpf_prog_query(attr, uattr);
748cd572
DZ		 3586 case BPF_SK_SKB_STREAM_PARSER:
3587 case BPF_SK_SKB_STREAM_VERDICT:
3588 case BPF_SK_MSG_VERDICT:
3589 case BPF_SK_SKB_VERDICT:
3590 return sock_map_bpf_prog_query(attr, uattr);
468e2f64
AS		 3591 default:
3592 return -EINVAL;
3593 }
468e2f64 3594}
f4324551 3595
b530e9e1 3596#define BPF_PROG_TEST_RUN_LAST_FIELD test.batch_size
1cf1cae9
AS		 3597
3598static int bpf_prog_test_run(const union bpf_attr *attr,
3599 union bpf_attr __user *uattr)
3600{
3601 struct bpf_prog *prog;
3602 int ret = -ENOTSUPP;
3603
3604 if (CHECK_ATTR(BPF_PROG_TEST_RUN))
3605 return -EINVAL;
3606
b0b9395d
SF		 3607 if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
3608 (!attr->test.ctx_size_in && attr->test.ctx_in))
3609 return -EINVAL;
3610
3611 if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
3612 (!attr->test.ctx_size_out && attr->test.ctx_out))
3613 return -EINVAL;
3614
1cf1cae9
AS		 3615 prog = bpf_prog_get(attr->test.prog_fd);
3616 if (IS_ERR(prog))
3617 return PTR_ERR(prog);
3618
3619 if (prog->aux->ops->test_run)
3620 ret = prog->aux->ops->test_run(prog, attr, uattr);
3621
3622 bpf_prog_put(prog);
3623 return ret;
3624}
3625
34ad5580
MKL		 3626#define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
3627
3628static int bpf_obj_get_next_id(const union bpf_attr *attr,
3629 union bpf_attr __user *uattr,
3630 struct idr *idr,
3631 spinlock_t *lock)
3632{
3633 u32 next_id = attr->start_id;
3634 int err = 0;
3635
3636 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
3637 return -EINVAL;
3638
3639 if (!capable(CAP_SYS_ADMIN))
3640 return -EPERM;
3641
3642 next_id++;
3643 spin_lock_bh(lock);
3644 if (!idr_get_next(idr, &next_id))
3645 err = -ENOENT;
3646 spin_unlock_bh(lock);
3647
3648 if (!err)
3649 err = put_user(next_id, &uattr->next_id);
3650
3651 return err;
3652}
3653
6086d29d
YS		 3654struct bpf_map *bpf_map_get_curr_or_next(u32 *id)
3655{
3656 struct bpf_map *map;
3657
3658 spin_lock_bh(&map_idr_lock);
3659again:
3660 map = idr_get_next(&map_idr, id);
3661 if (map) {
3662 map = __bpf_map_inc_not_zero(map, false);
3663 if (IS_ERR(map)) {
3664 (*id)++;
3665 goto again;
3666 }
3667 }
3668 spin_unlock_bh(&map_idr_lock);
3669
3670 return map;
3671}
3672
a228a64f
AS		 3673struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id)
3674{
3675 struct bpf_prog *prog;
3676
3677 spin_lock_bh(&prog_idr_lock);
3678again:
3679 prog = idr_get_next(&prog_idr, id);
3680 if (prog) {
3681 prog = bpf_prog_inc_not_zero(prog);
3682 if (IS_ERR(prog)) {
3683 (*id)++;
3684 goto again;
3685 }
3686 }
3687 spin_unlock_bh(&prog_idr_lock);
3688
3689 return prog;
3690}
3691
b16d9aa4
MKL		 3692#define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
3693
7e6897f9 3694struct bpf_prog *bpf_prog_by_id(u32 id)
b16d9aa4
MKL		 3695{
3696 struct bpf_prog *prog;
b16d9aa4 3697
7e6897f9
BT		 3698 if (!id)
3699 return ERR_PTR(-ENOENT);
b16d9aa4
MKL		 3700
3701 spin_lock_bh(&prog_idr_lock);
3702 prog = idr_find(&prog_idr, id);
3703 if (prog)
3704 prog = bpf_prog_inc_not_zero(prog);
3705 else
3706 prog = ERR_PTR(-ENOENT);
3707 spin_unlock_bh(&prog_idr_lock);
7e6897f9
BT		 3708 return prog;
3709}
3710
3711static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
3712{
3713 struct bpf_prog *prog;
3714 u32 id = attr->prog_id;
3715 int fd;
3716
3717 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
3718 return -EINVAL;
3719
3720 if (!capable(CAP_SYS_ADMIN))
3721 return -EPERM;
b16d9aa4 3722
7e6897f9 3723 prog = bpf_prog_by_id(id);
b16d9aa4
MKL		 3724 if (IS_ERR(prog))
3725 return PTR_ERR(prog);
3726
3727 fd = bpf_prog_new_fd(prog);
3728 if (fd < 0)
3729 bpf_prog_put(prog);
3730
3731 return fd;
3732}
3733
6e71b04a 3734#define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
bd5f5f4e
MKL		 3735
3736static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
3737{
3738 struct bpf_map *map;
3739 u32 id = attr->map_id;
6e71b04a 3740 int f_flags;
bd5f5f4e
MKL		 3741 int fd;
3742
6e71b04a
CF		 3743 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
3744 attr->open_flags & ~BPF_OBJ_FLAG_MASK)
bd5f5f4e
MKL		 3745 return -EINVAL;
3746
3747 if (!capable(CAP_SYS_ADMIN))
3748 return -EPERM;
3749
6e71b04a
CF		 3750 f_flags = bpf_get_file_flag(attr->open_flags);
3751 if (f_flags < 0)
3752 return f_flags;
3753
bd5f5f4e
MKL		 3754 spin_lock_bh(&map_idr_lock);
3755 map = idr_find(&map_idr, id);
3756 if (map)
b0e4701c 3757 map = __bpf_map_inc_not_zero(map, true);
bd5f5f4e
MKL		 3758 else
3759 map = ERR_PTR(-ENOENT);
3760 spin_unlock_bh(&map_idr_lock);
3761
3762 if (IS_ERR(map))
3763 return PTR_ERR(map);
3764
6e71b04a 3765 fd = bpf_map_new_fd(map, f_flags);
bd5f5f4e 3766 if (fd < 0)
781e6282 3767 bpf_map_put_with_uref(map);
bd5f5f4e
MKL		 3768
3769 return fd;
3770}
3771
7105e828 3772static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
d8eca5bb
DB		 3773 unsigned long addr, u32 *off,
3774 u32 *type)
7105e828 3775{
d8eca5bb 3776 const struct bpf_map *map;
7105e828
DB		 3777 int i;
3778
984fe94f 3779 mutex_lock(&prog->aux->used_maps_mutex);
d8eca5bb
DB		 3780 for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
3781 map = prog->aux->used_maps[i];
3782 if (map == (void *)addr) {
3783 *type = BPF_PSEUDO_MAP_FD;
984fe94f 3784 goto out;
d8eca5bb
DB		 3785 }
3786 if (!map->ops->map_direct_value_meta)
3787 continue;
3788 if (!map->ops->map_direct_value_meta(map, addr, off)) {
3789 *type = BPF_PSEUDO_MAP_VALUE;
984fe94f 3790 goto out;
d8eca5bb
DB		 3791 }
3792 }
984fe94f 3793 map = NULL;
d8eca5bb 3794
984fe94f
YZ		 3795out:
3796 mutex_unlock(&prog->aux->used_maps_mutex);
3797 return map;
7105e828
DB		 3798}
3799
63960260
KC		 3800static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
3801 const struct cred *f_cred)
7105e828
DB		 3802{
3803 const struct bpf_map *map;
3804 struct bpf_insn *insns;
d8eca5bb 3805 u32 off, type;
7105e828 3806 u64 imm;
29fcb05b 3807 u8 code;
7105e828
DB		 3808 int i;
3809
3810 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
3811 GFP_USER);
3812 if (!insns)
3813 return insns;
3814
3815 for (i = 0; i < prog->len; i++) {
29fcb05b
AN		 3816 code = insns[i].code;
3817
3818 if (code == (BPF_JMP | BPF_TAIL_CALL)) {
7105e828
DB		 3819 insns[i].code = BPF_JMP | BPF_CALL;
3820 insns[i].imm = BPF_FUNC_tail_call;
3821 /* fall-through */
3822 }
29fcb05b
AN		 3823 if (code == (BPF_JMP | BPF_CALL) ||
3824 code == (BPF_JMP | BPF_CALL_ARGS)) {
3825 if (code == (BPF_JMP | BPF_CALL_ARGS))
7105e828 3826 insns[i].code = BPF_JMP | BPF_CALL;
63960260 3827 if (!bpf_dump_raw_ok(f_cred))
7105e828
DB		 3828 insns[i].imm = 0;
3829 continue;
3830 }
29fcb05b
AN		 3831 if (BPF_CLASS(code) == BPF_LDX && BPF_MODE(code) == BPF_PROBE_MEM) {
3832 insns[i].code = BPF_LDX | BPF_SIZE(code) | BPF_MEM;
3833 continue;
3834 }
7105e828 3835
29fcb05b 3836 if (code != (BPF_LD | BPF_IMM | BPF_DW))
7105e828
DB		 3837 continue;
3838
3839 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
d8eca5bb 3840 map = bpf_map_from_imm(prog, imm, &off, &type);
7105e828 3841 if (map) {
d8eca5bb 3842 insns[i].src_reg = type;
7105e828 3843 insns[i].imm = map->id;
d8eca5bb 3844 insns[i + 1].imm = off;
7105e828
DB		 3845 continue;
3846 }
7105e828
DB		 3847 }
3848
3849 return insns;
3850}
3851
c454a46b
MKL		 3852static int set_info_rec_size(struct bpf_prog_info *info)
3853{
3854 /*
3855 * Ensure info.*_rec_size is the same as kernel expected size
3856 *
3857 * or
3858 *
3859 * Only allow zero *_rec_size if both _rec_size and _cnt are
3860 * zero. In this case, the kernel will set the expected
3861 * _rec_size back to the info.
3862 */
3863
11d8b82d 3864 if ((info->nr_func_info || info->func_info_rec_size) &&
c454a46b
MKL		 3865 info->func_info_rec_size != sizeof(struct bpf_func_info))
3866 return -EINVAL;
3867
11d8b82d 3868 if ((info->nr_line_info || info->line_info_rec_size) &&
c454a46b
MKL		 3869 info->line_info_rec_size != sizeof(struct bpf_line_info))
3870 return -EINVAL;
3871
11d8b82d 3872 if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
c454a46b
MKL		 3873 info->jited_line_info_rec_size != sizeof(__u64))
3874 return -EINVAL;
3875
3876 info->func_info_rec_size = sizeof(struct bpf_func_info);
3877 info->line_info_rec_size = sizeof(struct bpf_line_info);
3878 info->jited_line_info_rec_size = sizeof(__u64);
3879
3880 return 0;
3881}
3882
63960260
KC		 3883static int bpf_prog_get_info_by_fd(struct file *file,
3884 struct bpf_prog *prog,
1e270976
MKL		 3885 const union bpf_attr *attr,
3886 union bpf_attr __user *uattr)
3887{
3888 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
6644aabb 3889 struct btf *attach_btf = bpf_prog_get_target_btf(prog);
5c6f2588 3890 struct bpf_prog_info info;
1e270976 3891 u32 info_len = attr->info.info_len;
61a0abae 3892 struct bpf_prog_kstats stats;
1e270976
MKL		 3893 char __user *uinsns;
3894 u32 ulen;
3895 int err;
3896
af2ac3e1 3897 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
1e270976
MKL		 3898 if (err)
3899 return err;
3900 info_len = min_t(u32, sizeof(info), info_len);
3901
5c6f2588 3902 memset(&info, 0, sizeof(info));
1e270976 3903 if (copy_from_user(&info, uinfo, info_len))
89b09689 3904 return -EFAULT;
1e270976
MKL		 3905
3906 info.type = prog->type;
3907 info.id = prog->aux->id;
cb4d2b3f
MKL		 3908 info.load_time = prog->aux->load_time;
3909 info.created_by_uid = from_kuid_munged(current_user_ns(),
3910 prog->aux->user->uid);
b85fab0e 3911 info.gpl_compatible = prog->gpl_compatible;
1e270976
MKL		 3912
3913 memcpy(info.tag, prog->tag, sizeof(prog->tag));
cb4d2b3f
MKL		 3914 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
3915
984fe94f 3916 mutex_lock(&prog->aux->used_maps_mutex);
cb4d2b3f
MKL		 3917 ulen = info.nr_map_ids;
3918 info.nr_map_ids = prog->aux->used_map_cnt;
3919 ulen = min_t(u32, info.nr_map_ids, ulen);
3920 if (ulen) {
721e08da 3921 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
cb4d2b3f
MKL		 3922 u32 i;
3923
3924 for (i = 0; i < ulen; i++)
3925 if (put_user(prog->aux->used_maps[i]->id,
984fe94f
YZ		 3926 &user_map_ids[i])) {
3927 mutex_unlock(&prog->aux->used_maps_mutex);
cb4d2b3f 3928 return -EFAULT;
984fe94f 3929 }
cb4d2b3f 3930 }
984fe94f 3931 mutex_unlock(&prog->aux->used_maps_mutex);
1e270976 3932
c454a46b
MKL		 3933 err = set_info_rec_size(&info);
3934 if (err)
3935 return err;
7337224f 3936
5f8f8b93
AS		 3937 bpf_prog_get_stats(prog, &stats);
3938 info.run_time_ns = stats.nsecs;
3939 info.run_cnt = stats.cnt;
9ed9e9ba 3940 info.recursion_misses = stats.misses;
5f8f8b93 3941
aba64c7d
DM		 3942 info.verified_insns = prog->aux->verified_insns;
3943
2c78ee89 3944 if (!bpf_capable()) {
1e270976
MKL		 3945 info.jited_prog_len = 0;
3946 info.xlated_prog_len = 0;
dbecd738 3947 info.nr_jited_ksyms = 0;
28c2fae7 3948 info.nr_jited_func_lens = 0;
11d8b82d
YS		 3949 info.nr_func_info = 0;
3950 info.nr_line_info = 0;
3951 info.nr_jited_line_info = 0;
1e270976
MKL		 3952 goto done;
3953 }
3954
1e270976 3955 ulen = info.xlated_prog_len;
9975a54b 3956 info.xlated_prog_len = bpf_prog_insn_size(prog);
1e270976 3957 if (info.xlated_prog_len && ulen) {
7105e828
DB		 3958 struct bpf_insn *insns_sanitized;
3959 bool fault;
3960
63960260 3961 if (prog->blinded && !bpf_dump_raw_ok(file->f_cred)) {
7105e828
DB		 3962 info.xlated_prog_insns = 0;
3963 goto done;
3964 }
63960260 3965 insns_sanitized = bpf_insn_prepare_dump(prog, file->f_cred);
7105e828
DB		 3966 if (!insns_sanitized)
3967 return -ENOMEM;
1e270976
MKL		 3968 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
3969 ulen = min_t(u32, info.xlated_prog_len, ulen);
7105e828
DB		 3970 fault = copy_to_user(uinsns, insns_sanitized, ulen);
3971 kfree(insns_sanitized);
3972 if (fault)
1e270976
MKL		 3973 return -EFAULT;
3974 }
3975
675fc275
JK		 3976 if (bpf_prog_is_dev_bound(prog->aux)) {
3977 err = bpf_prog_offload_info_fill(&info, prog);
3978 if (err)
3979 return err;
fcfb126d
JW		 3980 goto done;
3981 }
3982
3983 /* NOTE: the following code is supposed to be skipped for offload.
3984 * bpf_prog_offload_info_fill() is the place to fill similar fields
3985 * for offload.
3986 */
3987 ulen = info.jited_prog_len;
4d56a76e
SD		 3988 if (prog->aux->func_cnt) {
3989 u32 i;
3990
3991 info.jited_prog_len = 0;
3992 for (i = 0; i < prog->aux->func_cnt; i++)
3993 info.jited_prog_len += prog->aux->func[i]->jited_len;
3994 } else {
3995 info.jited_prog_len = prog->jited_len;
3996 }
3997
fcfb126d 3998 if (info.jited_prog_len && ulen) {
63960260 3999 if (bpf_dump_raw_ok(file->f_cred)) {
fcfb126d
JW		 4000 uinsns = u64_to_user_ptr(info.jited_prog_insns);
4001 ulen = min_t(u32, info.jited_prog_len, ulen);
4d56a76e
SD		 4002
4003 /* for multi-function programs, copy the JITed
4004 * instructions for all the functions
4005 */
4006 if (prog->aux->func_cnt) {
4007 u32 len, free, i;
4008 u8 *img;
4009
4010 free = ulen;
4011 for (i = 0; i < prog->aux->func_cnt; i++) {
4012 len = prog->aux->func[i]->jited_len;
4013 len = min_t(u32, len, free);
4014 img = (u8 *) prog->aux->func[i]->bpf_func;
4015 if (copy_to_user(uinsns, img, len))
4016 return -EFAULT;
4017 uinsns += len;
4018 free -= len;
4019 if (!free)
4020 break;
4021 }
4022 } else {
4023 if (copy_to_user(uinsns, prog->bpf_func, ulen))
4024 return -EFAULT;
4025 }
fcfb126d
JW		 4026 } else {
4027 info.jited_prog_insns = 0;
4028 }
675fc275
JK		 4029 }
4030
dbecd738 4031 ulen = info.nr_jited_ksyms;
ff1889fc 4032 info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
7a5725dd 4033 if (ulen) {
63960260 4034 if (bpf_dump_raw_ok(file->f_cred)) {
ff1889fc 4035 unsigned long ksym_addr;
dbecd738 4036 u64 __user *user_ksyms;
dbecd738
SD		 4037 u32 i;
4038
4039 /* copy the address of the kernel symbol
4040 * corresponding to each function
4041 */
4042 ulen = min_t(u32, info.nr_jited_ksyms, ulen);
4043 user_ksyms = u64_to_user_ptr(info.jited_ksyms);
ff1889fc
SL		 4044 if (prog->aux->func_cnt) {
4045 for (i = 0; i < ulen; i++) {
4046 ksym_addr = (unsigned long)
4047 prog->aux->func[i]->bpf_func;
4048 if (put_user((u64) ksym_addr,
4049 &user_ksyms[i]))
4050 return -EFAULT;
4051 }
4052 } else {
4053 ksym_addr = (unsigned long) prog->bpf_func;
4054 if (put_user((u64) ksym_addr, &user_ksyms[0]))
dbecd738
SD		 4055 return -EFAULT;
4056 }
4057 } else {
4058 info.jited_ksyms = 0;
4059 }
4060 }
4061
815581c1 4062 ulen = info.nr_jited_func_lens;
ff1889fc 4063 info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
7a5725dd 4064 if (ulen) {
63960260 4065 if (bpf_dump_raw_ok(file->f_cred)) {
815581c1
SD		 4066 u32 __user *user_lens;
4067 u32 func_len, i;
4068
4069 /* copy the JITed image lengths for each function */
4070 ulen = min_t(u32, info.nr_jited_func_lens, ulen);
4071 user_lens = u64_to_user_ptr(info.jited_func_lens);
ff1889fc
SL		 4072 if (prog->aux->func_cnt) {
4073 for (i = 0; i < ulen; i++) {
4074 func_len =
4075 prog->aux->func[i]->jited_len;
4076 if (put_user(func_len, &user_lens[i]))
4077 return -EFAULT;
4078 }
4079 } else {
4080 func_len = prog->jited_len;
4081 if (put_user(func_len, &user_lens[0]))
815581c1
SD		 4082 return -EFAULT;
4083 }
4084 } else {
4085 info.jited_func_lens = 0;
4086 }
4087 }
4088
7337224f 4089 if (prog->aux->btf)
22dc4a0f 4090 info.btf_id = btf_obj_id(prog->aux->btf);
b79c9fc9 4091 info.attach_btf_id = prog->aux->attach_btf_id;
6644aabb
SF		 4092 if (attach_btf)
4093 info.attach_btf_obj_id = btf_obj_id(attach_btf);
838e9690 4094
11d8b82d
YS		 4095 ulen = info.nr_func_info;
4096 info.nr_func_info = prog->aux->func_info_cnt;
4097 if (info.nr_func_info && ulen) {
9e794163 4098 char __user *user_finfo;
7337224f 4099
9e794163
MKL		 4100 user_finfo = u64_to_user_ptr(info.func_info);
4101 ulen = min_t(u32, info.nr_func_info, ulen);
4102 if (copy_to_user(user_finfo, prog->aux->func_info,
4103 info.func_info_rec_size * ulen))
4104 return -EFAULT;
838e9690
YS		 4105 }
4106
11d8b82d
YS		 4107 ulen = info.nr_line_info;
4108 info.nr_line_info = prog->aux->nr_linfo;
4109 if (info.nr_line_info && ulen) {
9e794163 4110 __u8 __user *user_linfo;
c454a46b 4111
9e794163
MKL		 4112 user_linfo = u64_to_user_ptr(info.line_info);
4113 ulen = min_t(u32, info.nr_line_info, ulen);
4114 if (copy_to_user(user_linfo, prog->aux->linfo,
4115 info.line_info_rec_size * ulen))
4116 return -EFAULT;
c454a46b
MKL		 4117 }
4118
11d8b82d 4119 ulen = info.nr_jited_line_info;
c454a46b 4120 if (prog->aux->jited_linfo)
11d8b82d 4121 info.nr_jited_line_info = prog->aux->nr_linfo;
c454a46b 4122 else
11d8b82d
YS		 4123 info.nr_jited_line_info = 0;
4124 if (info.nr_jited_line_info && ulen) {
63960260 4125 if (bpf_dump_raw_ok(file->f_cred)) {
2cd00852 4126 unsigned long line_addr;
c454a46b
MKL		 4127 __u64 __user *user_linfo;
4128 u32 i;
4129
4130 user_linfo = u64_to_user_ptr(info.jited_line_info);
11d8b82d 4131 ulen = min_t(u32, info.nr_jited_line_info, ulen);
c454a46b 4132 for (i = 0; i < ulen; i++) {
2cd00852
PL		 4133 line_addr = (unsigned long)prog->aux->jited_linfo[i];
4134 if (put_user((__u64)line_addr, &user_linfo[i]))
c454a46b
MKL		 4135 return -EFAULT;
4136 }
4137 } else {
4138 info.jited_line_info = 0;
4139 }
4140 }
4141
c872bdb3
SL		 4142 ulen = info.nr_prog_tags;
4143 info.nr_prog_tags = prog->aux->func_cnt ? : 1;
4144 if (ulen) {
4145 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
4146 u32 i;
4147
4148 user_prog_tags = u64_to_user_ptr(info.prog_tags);
4149 ulen = min_t(u32, info.nr_prog_tags, ulen);
4150 if (prog->aux->func_cnt) {
4151 for (i = 0; i < ulen; i++) {
4152 if (copy_to_user(user_prog_tags[i],
4153 prog->aux->func[i]->tag,
4154 BPF_TAG_SIZE))
4155 return -EFAULT;
4156 }
4157 } else {
4158 if (copy_to_user(user_prog_tags[0],
4159 prog->tag, BPF_TAG_SIZE))
4160 return -EFAULT;
4161 }
4162 }
4163
1e270976
MKL		 4164done:
4165 if (copy_to_user(uinfo, &info, info_len) ||
4166 put_user(info_len, &uattr->info.info_len))
4167 return -EFAULT;
4168
4169 return 0;
4170}
4171
63960260
KC		 4172static int bpf_map_get_info_by_fd(struct file *file,
4173 struct bpf_map *map,
1e270976
MKL		 4174 const union bpf_attr *attr,
4175 union bpf_attr __user *uattr)
4176{
4177 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
5c6f2588 4178 struct bpf_map_info info;
1e270976
MKL		 4179 u32 info_len = attr->info.info_len;
4180 int err;
4181
af2ac3e1 4182 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
1e270976
MKL		 4183 if (err)
4184 return err;
4185 info_len = min_t(u32, sizeof(info), info_len);
4186
5c6f2588 4187 memset(&info, 0, sizeof(info));
1e270976
MKL		 4188 info.type = map->map_type;
4189 info.id = map->id;
4190 info.key_size = map->key_size;
4191 info.value_size = map->value_size;
4192 info.max_entries = map->max_entries;
4193 info.map_flags = map->map_flags;
9330986c 4194 info.map_extra = map->map_extra;
ad5b177b 4195 memcpy(info.name, map->name, sizeof(map->name));
1e270976 4196
78958fca 4197 if (map->btf) {
22dc4a0f 4198 info.btf_id = btf_obj_id(map->btf);
9b2cf328
MKL		 4199 info.btf_key_type_id = map->btf_key_type_id;
4200 info.btf_value_type_id = map->btf_value_type_id;
78958fca 4201 }
85d33df3 4202 info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
78958fca 4203
52775b33
JK		 4204 if (bpf_map_is_dev_bound(map)) {
4205 err = bpf_map_offload_info_fill(&info, map);
4206 if (err)
4207 return err;
4208 }
4209
1e270976
MKL		 4210 if (copy_to_user(uinfo, &info, info_len) ||
4211 put_user(info_len, &uattr->info.info_len))
4212 return -EFAULT;
4213
4214 return 0;
4215}
4216
63960260
KC		 4217static int bpf_btf_get_info_by_fd(struct file *file,
4218 struct btf *btf,
62dab84c
MKL		 4219 const union bpf_attr *attr,
4220 union bpf_attr __user *uattr)
4221{
4222 struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4223 u32 info_len = attr->info.info_len;
4224 int err;
4225
af2ac3e1 4226 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(*uinfo), info_len);
62dab84c
MKL		 4227 if (err)
4228 return err;
4229
4230 return btf_get_info_by_fd(btf, attr, uattr);
4231}
4232
63960260
KC		 4233static int bpf_link_get_info_by_fd(struct file *file,
4234 struct bpf_link *link,
f2e10bff
AN		 4235 const union bpf_attr *attr,
4236 union bpf_attr __user *uattr)
4237{
4238 struct bpf_link_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4239 struct bpf_link_info info;
4240 u32 info_len = attr->info.info_len;
4241 int err;
4242
af2ac3e1 4243 err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
f2e10bff
AN		 4244 if (err)
4245 return err;
4246 info_len = min_t(u32, sizeof(info), info_len);
4247
4248 memset(&info, 0, sizeof(info));
4249 if (copy_from_user(&info, uinfo, info_len))
4250 return -EFAULT;
4251
4252 info.type = link->type;
4253 info.id = link->id;
4254 info.prog_id = link->prog->aux->id;
4255
4256 if (link->ops->fill_link_info) {
4257 err = link->ops->fill_link_info(link, &info);
4258 if (err)
4259 return err;
4260 }
4261
4262 if (copy_to_user(uinfo, &info, info_len) ||
4263 put_user(info_len, &uattr->info.info_len))
4264 return -EFAULT;
4265
4266 return 0;
4267}
4268
4269
1e270976
MKL		 4270#define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
4271
4272static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
4273 union bpf_attr __user *uattr)
4274{
4275 int ufd = attr->info.bpf_fd;
4276 struct fd f;
4277 int err;
4278
4279 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
4280 return -EINVAL;
4281
4282 f = fdget(ufd);
4283 if (!f.file)
4284 return -EBADFD;
4285
4286 if (f.file->f_op == &bpf_prog_fops)
63960260 4287 err = bpf_prog_get_info_by_fd(f.file, f.file->private_data, attr,
1e270976
MKL		 4288 uattr);
4289 else if (f.file->f_op == &bpf_map_fops)
63960260 4290 err = bpf_map_get_info_by_fd(f.file, f.file->private_data, attr,
1e270976 4291 uattr);
60197cfb 4292 else if (f.file->f_op == &btf_fops)
63960260 4293 err = bpf_btf_get_info_by_fd(f.file, f.file->private_data, attr, uattr);
f2e10bff 4294 else if (f.file->f_op == &bpf_link_fops)
63960260 4295 err = bpf_link_get_info_by_fd(f.file, f.file->private_data,
f2e10bff 4296 attr, uattr);
1e270976
MKL		 4297 else
4298 err = -EINVAL;
4299
4300 fdput(f);
4301 return err;
4302}
4303
f56a653c
MKL		 4304#define BPF_BTF_LOAD_LAST_FIELD btf_log_level
4305
c571bd75 4306static int bpf_btf_load(const union bpf_attr *attr, bpfptr_t uattr)
f56a653c
MKL		 4307{
4308 if (CHECK_ATTR(BPF_BTF_LOAD))
4309 return -EINVAL;
4310
2c78ee89 4311 if (!bpf_capable())
f56a653c
MKL		 4312 return -EPERM;
4313
c571bd75 4314 return btf_new_fd(attr, uattr);
f56a653c
MKL		 4315}
4316
78958fca
MKL		 4317#define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
4318
4319static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
4320{
4321 if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
4322 return -EINVAL;
4323
4324 if (!capable(CAP_SYS_ADMIN))
4325 return -EPERM;
4326
4327 return btf_get_fd_by_id(attr->btf_id);
4328}
4329
41bdc4b4
YS		 4330static int bpf_task_fd_query_copy(const union bpf_attr *attr,
4331 union bpf_attr __user *uattr,
4332 u32 prog_id, u32 fd_type,
4333 const char *buf, u64 probe_offset,
4334 u64 probe_addr)
4335{
4336 char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
4337 u32 len = buf ? strlen(buf) : 0, input_len;
4338 int err = 0;
4339
4340 if (put_user(len, &uattr->task_fd_query.buf_len))
4341 return -EFAULT;
4342 input_len = attr->task_fd_query.buf_len;
4343 if (input_len && ubuf) {
4344 if (!len) {
4345 /* nothing to copy, just make ubuf NULL terminated */
4346 char zero = '\0';
4347
4348 if (put_user(zero, ubuf))
4349 return -EFAULT;
4350 } else if (input_len >= len + 1) {
4351 /* ubuf can hold the string with NULL terminator */
4352 if (copy_to_user(ubuf, buf, len + 1))
4353 return -EFAULT;
4354 } else {
4355 /* ubuf cannot hold the string with NULL terminator,
4356 * do a partial copy with NULL terminator.
4357 */
4358 char zero = '\0';
4359
4360 err = -ENOSPC;
4361 if (copy_to_user(ubuf, buf, input_len - 1))
4362 return -EFAULT;
4363 if (put_user(zero, ubuf + input_len - 1))
4364 return -EFAULT;
4365 }
4366 }
4367
4368 if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
4369 put_user(fd_type, &uattr->task_fd_query.fd_type) ||
4370 put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
4371 put_user(probe_addr, &uattr->task_fd_query.probe_addr))
4372 return -EFAULT;
4373
4374 return err;
4375}
4376
4377#define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
4378
4379static int bpf_task_fd_query(const union bpf_attr *attr,
4380 union bpf_attr __user *uattr)
4381{
4382 pid_t pid = attr->task_fd_query.pid;
4383 u32 fd = attr->task_fd_query.fd;
4384 const struct perf_event *event;
41bdc4b4
YS		 4385 struct task_struct *task;
4386 struct file *file;
4387 int err;
4388
4389 if (CHECK_ATTR(BPF_TASK_FD_QUERY))
4390 return -EINVAL;
4391
4392 if (!capable(CAP_SYS_ADMIN))
4393 return -EPERM;
4394
4395 if (attr->task_fd_query.flags != 0)
4396 return -EINVAL;
4397
4398 task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
4399 if (!task)
4400 return -ENOENT;
4401
41bdc4b4 4402 err = 0;
b48845af
EB		 4403 file = fget_task(task, fd);
4404 put_task_struct(task);
41bdc4b4 4405 if (!file)
b48845af 4406 return -EBADF;
41bdc4b4 4407
70ed506c
AN		 4408 if (file->f_op == &bpf_link_fops) {
4409 struct bpf_link *link = file->private_data;
41bdc4b4 4410
a3b80e10 4411 if (link->ops == &bpf_raw_tp_link_lops) {
70ed506c
AN		 4412 struct bpf_raw_tp_link *raw_tp =
4413 container_of(link, struct bpf_raw_tp_link, link);
4414 struct bpf_raw_event_map *btp = raw_tp->btp;
4415
4416 err = bpf_task_fd_query_copy(attr, uattr,
4417 raw_tp->link.prog->aux->id,
4418 BPF_FD_TYPE_RAW_TRACEPOINT,
4419 btp->tp->name, 0, 0);
4420 goto put_file;
4421 }
4422 goto out_not_supp;
41bdc4b4
YS		 4423 }
4424
4425 event = perf_get_event(file);
4426 if (!IS_ERR(event)) {
4427 u64 probe_offset, probe_addr;
4428 u32 prog_id, fd_type;
4429 const char *buf;
4430
4431 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
4432 &buf, &probe_offset,
4433 &probe_addr);
4434 if (!err)
4435 err = bpf_task_fd_query_copy(attr, uattr, prog_id,
4436 fd_type, buf,
4437 probe_offset,
4438 probe_addr);
4439 goto put_file;
4440 }
4441
70ed506c 4442out_not_supp:
41bdc4b4
YS		 4443 err = -ENOTSUPP;
4444put_file:
4445 fput(file);
41bdc4b4
YS		 4446 return err;
4447}
4448
cb4d03ab
BV		 4449#define BPF_MAP_BATCH_LAST_FIELD batch.flags
4450
4451#define BPF_DO_BATCH(fn) \
4452 do { \
4453 if (!fn) { \
4454 err = -ENOTSUPP; \
4455 goto err_put; \
4456 } \
4457 err = fn(map, attr, uattr); \
4458 } while (0)
4459
4460static int bpf_map_do_batch(const union bpf_attr *attr,
4461 union bpf_attr __user *uattr,
4462 int cmd)
4463{
353050be
DB		 4464 bool has_read = cmd == BPF_MAP_LOOKUP_BATCH ||
4465 cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH;
4466 bool has_write = cmd != BPF_MAP_LOOKUP_BATCH;
cb4d03ab
BV		 4467 struct bpf_map *map;
4468 int err, ufd;
4469 struct fd f;
4470
4471 if (CHECK_ATTR(BPF_MAP_BATCH))
4472 return -EINVAL;
4473
4474 ufd = attr->batch.map_fd;
4475 f = fdget(ufd);
4476 map = __bpf_map_get(f);
4477 if (IS_ERR(map))
4478 return PTR_ERR(map);
353050be
DB		 4479 if (has_write)
4480 bpf_map_write_active_inc(map);
4481 if (has_read && !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
cb4d03ab
BV		 4482 err = -EPERM;
4483 goto err_put;
4484 }
353050be 4485 if (has_write && !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
cb4d03ab
BV		 4486 err = -EPERM;
4487 goto err_put;
4488 }
4489
4490 if (cmd == BPF_MAP_LOOKUP_BATCH)
4491 BPF_DO_BATCH(map->ops->map_lookup_batch);
05799638
YS		 4492 else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
4493 BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch);
aa2e93b8
BV		 4494 else if (cmd == BPF_MAP_UPDATE_BATCH)
4495 BPF_DO_BATCH(map->ops->map_update_batch);
4496 else
4497 BPF_DO_BATCH(map->ops->map_delete_batch);
cb4d03ab 4498err_put:
353050be
DB		 4499 if (has_write)
4500 bpf_map_write_active_dec(map);
cb4d03ab
BV		 4501 fdput(f);
4502 return err;
4503}
4504
ca74823c 4505#define BPF_LINK_CREATE_LAST_FIELD link_create.kprobe_multi.cookies
af2ac3e1 4506static int link_create(union bpf_attr *attr, bpfptr_t uattr)
af6eea57
AN		 4507{
4508 enum bpf_prog_type ptype;
4509 struct bpf_prog *prog;
4510 int ret;
4511
af6eea57
AN		 4512 if (CHECK_ATTR(BPF_LINK_CREATE))
4513 return -EINVAL;
4514
4a1e7c0c 4515 prog = bpf_prog_get(attr->link_create.prog_fd);
af6eea57
AN		 4516 if (IS_ERR(prog))
4517 return PTR_ERR(prog);
4518
4519 ret = bpf_prog_attach_check_attach_type(prog,
4520 attr->link_create.attach_type);
4521 if (ret)
4a1e7c0c
THJ		 4522 goto out;
4523
b89fbfbb
AN		 4524 switch (prog->type) {
4525 case BPF_PROG_TYPE_EXT:
df86ca0d 4526 break;
b89fbfbb 4527 case BPF_PROG_TYPE_PERF_EVENT:
b89fbfbb
AN		 4528 case BPF_PROG_TYPE_TRACEPOINT:
4529 if (attr->link_create.attach_type != BPF_PERF_EVENT) {
4530 ret = -EINVAL;
4531 goto out;
4532 }
b89fbfbb 4533 break;
0dcac272
JO		 4534 case BPF_PROG_TYPE_KPROBE:
4535 if (attr->link_create.attach_type != BPF_PERF_EVENT &&
4536 attr->link_create.attach_type != BPF_TRACE_KPROBE_MULTI) {
4537 ret = -EINVAL;
4538 goto out;
4539 }
0dcac272 4540 break;
b89fbfbb
AN		 4541 default:
4542 ptype = attach_type_to_prog_type(attr->link_create.attach_type);
4543 if (ptype == BPF_PROG_TYPE_UNSPEC || ptype != prog->type) {
4544 ret = -EINVAL;
4545 goto out;
4546 }
4547 break;
4a1e7c0c 4548 }
af6eea57 4549
df86ca0d 4550 switch (prog->type) {
af6eea57
AN		 4551 case BPF_PROG_TYPE_CGROUP_SKB:
4552 case BPF_PROG_TYPE_CGROUP_SOCK:
4553 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4554 case BPF_PROG_TYPE_SOCK_OPS:
4555 case BPF_PROG_TYPE_CGROUP_DEVICE:
4556 case BPF_PROG_TYPE_CGROUP_SYSCTL:
4557 case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4558 ret = cgroup_bpf_link_attach(attr, prog);
4559 break;
df86ca0d
AN		 4560 case BPF_PROG_TYPE_EXT:
4561 ret = bpf_tracing_prog_attach(prog,
4562 attr->link_create.target_fd,
2fcc8241
KFL		 4563 attr->link_create.target_btf_id,
4564 attr->link_create.tracing.cookie);
df86ca0d
AN		 4565 break;
4566 case BPF_PROG_TYPE_LSM:
de4e05ca 4567 case BPF_PROG_TYPE_TRACING:
df86ca0d
AN		 4568 if (attr->link_create.attach_type != prog->expected_attach_type) {
4569 ret = -EINVAL;
4570 goto out;
4571 }
4572 if (prog->expected_attach_type == BPF_TRACE_RAW_TP)
4573 ret = bpf_raw_tp_link_attach(prog, NULL);
4574 else if (prog->expected_attach_type == BPF_TRACE_ITER)
4575 ret = bpf_iter_link_attach(attr, uattr, prog);
69fd337a
SF		 4576 else if (prog->expected_attach_type == BPF_LSM_CGROUP)
4577 ret = cgroup_bpf_link_attach(attr, prog);
df86ca0d
AN		 4578 else
4579 ret = bpf_tracing_prog_attach(prog,
4580 attr->link_create.target_fd,
2fcc8241
KFL		 4581 attr->link_create.target_btf_id,
4582 attr->link_create.tracing.cookie);
de4e05ca 4583 break;
7f045a49 4584 case BPF_PROG_TYPE_FLOW_DISSECTOR:
e9ddbb77 4585 case BPF_PROG_TYPE_SK_LOOKUP:
7f045a49
JS		 4586 ret = netns_bpf_link_create(attr, prog);
4587 break;
310ad797 4588#ifdef CONFIG_NET
aa8d3a71
AN		 4589 case BPF_PROG_TYPE_XDP:
4590 ret = bpf_xdp_link_attach(attr, prog);
4591 break;
b89fbfbb 4592#endif
b89fbfbb
AN		 4593 case BPF_PROG_TYPE_PERF_EVENT:
4594 case BPF_PROG_TYPE_TRACEPOINT:
b89fbfbb
AN		 4595 ret = bpf_perf_link_attach(attr, prog);
4596 break;
0dcac272
JO		 4597 case BPF_PROG_TYPE_KPROBE:
4598 if (attr->link_create.attach_type == BPF_PERF_EVENT)
4599 ret = bpf_perf_link_attach(attr, prog);
4600 else
4601 ret = bpf_kprobe_multi_link_attach(attr, prog);
4602 break;
af6eea57
AN		 4603 default:
4604 ret = -EINVAL;
4605 }
4606
4a1e7c0c 4607out:
af6eea57
AN		 4608 if (ret < 0)
4609 bpf_prog_put(prog);
4610 return ret;
4611}
4612
0c991ebc
AN		 4613#define BPF_LINK_UPDATE_LAST_FIELD link_update.old_prog_fd
4614
4615static int link_update(union bpf_attr *attr)
4616{
4617 struct bpf_prog *old_prog = NULL, *new_prog;
4618 struct bpf_link *link;
4619 u32 flags;
4620 int ret;
4621
0c991ebc
AN		 4622 if (CHECK_ATTR(BPF_LINK_UPDATE))
4623 return -EINVAL;
4624
4625 flags = attr->link_update.flags;
4626 if (flags & ~BPF_F_REPLACE)
4627 return -EINVAL;
4628
4629 link = bpf_link_get_from_fd(attr->link_update.link_fd);
4630 if (IS_ERR(link))
4631 return PTR_ERR(link);
4632
4633 new_prog = bpf_prog_get(attr->link_update.new_prog_fd);
4adb7a4a
AN		 4634 if (IS_ERR(new_prog)) {
4635 ret = PTR_ERR(new_prog);
4636 goto out_put_link;
4637 }
0c991ebc
AN		 4638
4639 if (flags & BPF_F_REPLACE) {
4640 old_prog = bpf_prog_get(attr->link_update.old_prog_fd);
4641 if (IS_ERR(old_prog)) {
4642 ret = PTR_ERR(old_prog);
4643 old_prog = NULL;
4644 goto out_put_progs;
4645 }
4adb7a4a
AN		 4646 } else if (attr->link_update.old_prog_fd) {
4647 ret = -EINVAL;
4648 goto out_put_progs;
0c991ebc
AN		 4649 }
4650
f9d04127
AN		 4651 if (link->ops->update_prog)
4652 ret = link->ops->update_prog(link, new_prog, old_prog);
4653 else
fe537393 4654 ret = -EINVAL;
0c991ebc
AN		 4655
4656out_put_progs:
4657 if (old_prog)
4658 bpf_prog_put(old_prog);
4659 if (ret)
4660 bpf_prog_put(new_prog);
4adb7a4a
AN		 4661out_put_link:
4662 bpf_link_put(link);
0c991ebc
AN		 4663 return ret;
4664}
4665
73b11c2a
AN		 4666#define BPF_LINK_DETACH_LAST_FIELD link_detach.link_fd
4667
4668static int link_detach(union bpf_attr *attr)
4669{
4670 struct bpf_link *link;
4671 int ret;
4672
4673 if (CHECK_ATTR(BPF_LINK_DETACH))
4674 return -EINVAL;
4675
4676 link = bpf_link_get_from_fd(attr->link_detach.link_fd);
4677 if (IS_ERR(link))
4678 return PTR_ERR(link);
4679
4680 if (link->ops->detach)
4681 ret = link->ops->detach(link);
4682 else
4683 ret = -EOPNOTSUPP;
4684
4685 bpf_link_put(link);
4686 return ret;
4687}
4688
005142b8 4689static struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
2d602c8c 4690{
005142b8 4691 return atomic64_fetch_add_unless(&link->refcnt, 1, 0) ? link : ERR_PTR(-ENOENT);
2d602c8c
AN		 4692}
4693
005142b8 4694struct bpf_link *bpf_link_by_id(u32 id)
2d602c8c
AN		 4695{
4696 struct bpf_link *link;
2d602c8c 4697
005142b8
AS		 4698 if (!id)
4699 return ERR_PTR(-ENOENT);
2d602c8c
AN		 4700
4701 spin_lock_bh(&link_idr_lock);
2d602c8c 4702 /* before link is "settled", ID is 0, pretend it doesn't exist yet */
005142b8 4703 link = idr_find(&link_idr, id);
2d602c8c
AN		 4704 if (link) {
4705 if (link->id)
005142b8 4706 link = bpf_link_inc_not_zero(link);
2d602c8c 4707 else
005142b8 4708 link = ERR_PTR(-EAGAIN);
2d602c8c 4709 } else {
005142b8 4710 link = ERR_PTR(-ENOENT);
2d602c8c
AN		 4711 }
4712 spin_unlock_bh(&link_idr_lock);
005142b8
AS		 4713 return link;
4714}
2d602c8c 4715
9f883612
DD		 4716struct bpf_link *bpf_link_get_curr_or_next(u32 *id)
4717{
4718 struct bpf_link *link;
4719
4720 spin_lock_bh(&link_idr_lock);
4721again:
4722 link = idr_get_next(&link_idr, id);
4723 if (link) {
4724 link = bpf_link_inc_not_zero(link);
4725 if (IS_ERR(link)) {
4726 (*id)++;
4727 goto again;
4728 }
4729 }
4730 spin_unlock_bh(&link_idr_lock);
4731
4732 return link;
4733}
4734
005142b8
AS		 4735#define BPF_LINK_GET_FD_BY_ID_LAST_FIELD link_id
4736
4737static int bpf_link_get_fd_by_id(const union bpf_attr *attr)
4738{
4739 struct bpf_link *link;
4740 u32 id = attr->link_id;
4741 int fd;
4742
4743 if (CHECK_ATTR(BPF_LINK_GET_FD_BY_ID))
4744 return -EINVAL;
4745
4746 if (!capable(CAP_SYS_ADMIN))
4747 return -EPERM;
4748
4749 link = bpf_link_by_id(id);
4750 if (IS_ERR(link))
4751 return PTR_ERR(link);
2d602c8c
AN		 4752
4753 fd = bpf_link_new_fd(link);
4754 if (fd < 0)
4755 bpf_link_put(link);
4756
4757 return fd;
4758}
4759
d46edd67
SL		 4760DEFINE_MUTEX(bpf_stats_enabled_mutex);
4761
4762static int bpf_stats_release(struct inode *inode, struct file *file)
4763{
4764 mutex_lock(&bpf_stats_enabled_mutex);
4765 static_key_slow_dec(&bpf_stats_enabled_key.key);
4766 mutex_unlock(&bpf_stats_enabled_mutex);
4767 return 0;
4768}
4769
4770static const struct file_operations bpf_stats_fops = {
4771 .release = bpf_stats_release,
4772};
4773
4774static int bpf_enable_runtime_stats(void)
4775{
4776 int fd;
4777
4778 mutex_lock(&bpf_stats_enabled_mutex);
4779
4780 /* Set a very high limit to avoid overflow */
4781 if (static_key_count(&bpf_stats_enabled_key.key) > INT_MAX / 2) {
4782 mutex_unlock(&bpf_stats_enabled_mutex);
4783 return -EBUSY;
4784 }
4785
4786 fd = anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
4787 if (fd >= 0)
4788 static_key_slow_inc(&bpf_stats_enabled_key.key);
4789
4790 mutex_unlock(&bpf_stats_enabled_mutex);
4791 return fd;
4792}
4793
4794#define BPF_ENABLE_STATS_LAST_FIELD enable_stats.type
4795
4796static int bpf_enable_stats(union bpf_attr *attr)
4797{
4798
4799 if (CHECK_ATTR(BPF_ENABLE_STATS))
4800 return -EINVAL;
4801
4802 if (!capable(CAP_SYS_ADMIN))
4803 return -EPERM;
4804
4805 switch (attr->enable_stats.type) {
4806 case BPF_STATS_RUN_TIME:
4807 return bpf_enable_runtime_stats();
4808 default:
4809 break;
4810 }
4811 return -EINVAL;
4812}
4813
ac51d99b
YS		 4814#define BPF_ITER_CREATE_LAST_FIELD iter_create.flags
4815
4816static int bpf_iter_create(union bpf_attr *attr)
4817{
4818 struct bpf_link *link;
4819 int err;
4820
4821 if (CHECK_ATTR(BPF_ITER_CREATE))
4822 return -EINVAL;
4823
4824 if (attr->iter_create.flags)
4825 return -EINVAL;
4826
4827 link = bpf_link_get_from_fd(attr->iter_create.link_fd);
4828 if (IS_ERR(link))
4829 return PTR_ERR(link);
4830
4831 err = bpf_iter_new_fd(link);
4832 bpf_link_put(link);
4833
4834 return err;
4835}
4836
ef15314a
YZ		 4837#define BPF_PROG_BIND_MAP_LAST_FIELD prog_bind_map.flags
4838
4839static int bpf_prog_bind_map(union bpf_attr *attr)
4840{
4841 struct bpf_prog *prog;
4842 struct bpf_map *map;
4843 struct bpf_map **used_maps_old, **used_maps_new;
4844 int i, ret = 0;
4845
4846 if (CHECK_ATTR(BPF_PROG_BIND_MAP))
4847 return -EINVAL;
4848
4849 if (attr->prog_bind_map.flags)
4850 return -EINVAL;
4851
4852 prog = bpf_prog_get(attr->prog_bind_map.prog_fd);
4853 if (IS_ERR(prog))
4854 return PTR_ERR(prog);
4855
4856 map = bpf_map_get(attr->prog_bind_map.map_fd);
4857 if (IS_ERR(map)) {
4858 ret = PTR_ERR(map);
4859 goto out_prog_put;
4860 }
4861
4862 mutex_lock(&prog->aux->used_maps_mutex);
4863
4864 used_maps_old = prog->aux->used_maps;
4865
4866 for (i = 0; i < prog->aux->used_map_cnt; i++)
1028ae40
SF		 4867 if (used_maps_old[i] == map) {
4868 bpf_map_put(map);
ef15314a 4869 goto out_unlock;
1028ae40 4870 }
ef15314a
YZ		 4871
4872 used_maps_new = kmalloc_array(prog->aux->used_map_cnt + 1,
4873 sizeof(used_maps_new[0]),
4874 GFP_KERNEL);
4875 if (!used_maps_new) {
4876 ret = -ENOMEM;
4877 goto out_unlock;
4878 }
4879
4880 memcpy(used_maps_new, used_maps_old,
4881 sizeof(used_maps_old[0]) * prog->aux->used_map_cnt);
4882 used_maps_new[prog->aux->used_map_cnt] = map;
4883
4884 prog->aux->used_map_cnt++;
4885 prog->aux->used_maps = used_maps_new;
4886
4887 kfree(used_maps_old);
4888
4889out_unlock:
4890 mutex_unlock(&prog->aux->used_maps_mutex);
4891
4892 if (ret)
4893 bpf_map_put(map);
4894out_prog_put:
4895 bpf_prog_put(prog);
4896 return ret;
4897}
4898
af2ac3e1 4899static int __sys_bpf(int cmd, bpfptr_t uattr, unsigned int size)
99c55f7d 4900{
8096f229 4901 union bpf_attr attr;
c8644cd0 4902 bool capable;
99c55f7d
AS		 4903 int err;
4904
c8644cd0
AM		 4905 capable = bpf_capable() || !sysctl_unprivileged_bpf_disabled;
4906
4907 /* Intent here is for unprivileged_bpf_disabled to block key object
4908 * creation commands for unprivileged users; other actions depend
4909 * of fd availability and access to bpffs, so are dependent on
4910 * object creation success. Capabilities are later verified for
4911 * operations such as load and map create, so even with unprivileged
4912 * BPF disabled, capability checks are still carried out for these
4913 * and other operations.
4914 */
4915 if (!capable &&
4916 (cmd == BPF_MAP_CREATE || cmd == BPF_PROG_LOAD))
99c55f7d
AS		 4917 return -EPERM;
4918
dcab51f1 4919 err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
1e270976
MKL		 4920 if (err)
4921 return err;
4922 size = min_t(u32, size, sizeof(attr));
99c55f7d
AS		 4923
4924 /* copy attributes from user space, may be less than sizeof(bpf_attr) */
8096f229 4925 memset(&attr, 0, sizeof(attr));
af2ac3e1 4926 if (copy_from_bpfptr(&attr, uattr, size) != 0)
99c55f7d
AS		 4927 return -EFAULT;
4928
afdb09c7
CF		 4929 err = security_bpf(cmd, &attr, size);
4930 if (err < 0)
4931 return err;
4932
99c55f7d
AS		 4933 switch (cmd) {
4934 case BPF_MAP_CREATE:
4935 err = map_create(&attr);
4936 break;
db20fd2b
AS		 4937 case BPF_MAP_LOOKUP_ELEM:
4938 err = map_lookup_elem(&attr);
4939 break;
4940 case BPF_MAP_UPDATE_ELEM:
af2ac3e1 4941 err = map_update_elem(&attr, uattr);
db20fd2b
AS		 4942 break;
4943 case BPF_MAP_DELETE_ELEM:
4944 err = map_delete_elem(&attr);
4945 break;
4946 case BPF_MAP_GET_NEXT_KEY:
4947 err = map_get_next_key(&attr);
4948 break;
87df15de
DB		 4949 case BPF_MAP_FREEZE:
4950 err = map_freeze(&attr);
4951 break;
09756af4 4952 case BPF_PROG_LOAD:
838e9690 4953 err = bpf_prog_load(&attr, uattr);
09756af4 4954 break;
b2197755
DB		 4955 case BPF_OBJ_PIN:
4956 err = bpf_obj_pin(&attr);
4957 break;
4958 case BPF_OBJ_GET:
4959 err = bpf_obj_get(&attr);
4960 break;
f4324551
DM		 4961 case BPF_PROG_ATTACH:
4962 err = bpf_prog_attach(&attr);
4963 break;
4964 case BPF_PROG_DETACH:
4965 err = bpf_prog_detach(&attr);
4966 break;
468e2f64 4967 case BPF_PROG_QUERY:
af2ac3e1 4968 err = bpf_prog_query(&attr, uattr.user);
468e2f64 4969 break;
1cf1cae9 4970 case BPF_PROG_TEST_RUN:
af2ac3e1 4971 err = bpf_prog_test_run(&attr, uattr.user);
1cf1cae9 4972 break;
34ad5580 4973 case BPF_PROG_GET_NEXT_ID:
af2ac3e1 4974 err = bpf_obj_get_next_id(&attr, uattr.user,
34ad5580
MKL		 4975 &prog_idr, &prog_idr_lock);
4976 break;
4977 case BPF_MAP_GET_NEXT_ID:
af2ac3e1 4978 err = bpf_obj_get_next_id(&attr, uattr.user,
34ad5580
MKL		 4979 &map_idr, &map_idr_lock);
4980 break;
1b9ed84e 4981 case BPF_BTF_GET_NEXT_ID:
af2ac3e1 4982 err = bpf_obj_get_next_id(&attr, uattr.user,
1b9ed84e
QM		 4983 &btf_idr, &btf_idr_lock);
4984 break;
b16d9aa4
MKL		 4985 case BPF_PROG_GET_FD_BY_ID:
4986 err = bpf_prog_get_fd_by_id(&attr);
4987 break;
bd5f5f4e
MKL		 4988 case BPF_MAP_GET_FD_BY_ID:
4989 err = bpf_map_get_fd_by_id(&attr);
4990 break;
1e270976 4991 case BPF_OBJ_GET_INFO_BY_FD:
af2ac3e1 4992 err = bpf_obj_get_info_by_fd(&attr, uattr.user);
1e270976 4993 break;
c4f6699d
AS		 4994 case BPF_RAW_TRACEPOINT_OPEN:
4995 err = bpf_raw_tracepoint_open(&attr);
4996 break;
f56a653c 4997 case BPF_BTF_LOAD:
c571bd75 4998 err = bpf_btf_load(&attr, uattr);
f56a653c 4999 break;
78958fca
MKL		 5000 case BPF_BTF_GET_FD_BY_ID:
5001 err = bpf_btf_get_fd_by_id(&attr);
5002 break;
41bdc4b4 5003 case BPF_TASK_FD_QUERY:
af2ac3e1 5004 err = bpf_task_fd_query(&attr, uattr.user);
41bdc4b4 5005 break;
bd513cd0
MV		 5006 case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
5007 err = map_lookup_and_delete_elem(&attr);
5008 break;
cb4d03ab 5009 case BPF_MAP_LOOKUP_BATCH:
af2ac3e1 5010 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_LOOKUP_BATCH);
cb4d03ab 5011 break;
05799638 5012 case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
af2ac3e1 5013 err = bpf_map_do_batch(&attr, uattr.user,
05799638
YS		 5014 BPF_MAP_LOOKUP_AND_DELETE_BATCH);
5015 break;
aa2e93b8 5016 case BPF_MAP_UPDATE_BATCH:
af2ac3e1 5017 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_UPDATE_BATCH);
aa2e93b8
BV		 5018 break;
5019 case BPF_MAP_DELETE_BATCH:
af2ac3e1 5020 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_DELETE_BATCH);
aa2e93b8 5021 break;
af6eea57 5022 case BPF_LINK_CREATE:
af2ac3e1 5023 err = link_create(&attr, uattr);
af6eea57 5024 break;
0c991ebc
AN		 5025 case BPF_LINK_UPDATE:
5026 err = link_update(&attr);
5027 break;
2d602c8c
AN		 5028 case BPF_LINK_GET_FD_BY_ID:
5029 err = bpf_link_get_fd_by_id(&attr);
5030 break;
5031 case BPF_LINK_GET_NEXT_ID:
af2ac3e1 5032 err = bpf_obj_get_next_id(&attr, uattr.user,
2d602c8c
AN		 5033 &link_idr, &link_idr_lock);
5034 break;
d46edd67
SL		 5035 case BPF_ENABLE_STATS:
5036 err = bpf_enable_stats(&attr);
5037 break;
ac51d99b
YS		 5038 case BPF_ITER_CREATE:
5039 err = bpf_iter_create(&attr);
5040 break;
73b11c2a
AN		 5041 case BPF_LINK_DETACH:
5042 err = link_detach(&attr);
5043 break;
ef15314a
YZ		 5044 case BPF_PROG_BIND_MAP:
5045 err = bpf_prog_bind_map(&attr);
5046 break;
99c55f7d
AS		 5047 default:
5048 err = -EINVAL;
5049 break;
5050 }
5051
5052 return err;
5053}
79a7f8bd 5054
af2ac3e1
AS		 5055SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
5056{
5057 return __sys_bpf(cmd, USER_BPFPTR(uattr), size);
5058}
5059
79a7f8bd
AS		 5060static bool syscall_prog_is_valid_access(int off, int size,
5061 enum bpf_access_type type,
5062 const struct bpf_prog *prog,
5063 struct bpf_insn_access_aux *info)
5064{
5065 if (off < 0 || off >= U16_MAX)
5066 return false;
5067 if (off % size != 0)
5068 return false;
5069 return true;
5070}
5071
b1d18a75 5072BPF_CALL_3(bpf_sys_bpf, int, cmd, union bpf_attr *, attr, u32, attr_size)
79a7f8bd 5073{
af2ac3e1
AS		 5074 switch (cmd) {
5075 case BPF_MAP_CREATE:
5076 case BPF_MAP_UPDATE_ELEM:
5077 case BPF_MAP_FREEZE:
5078 case BPF_PROG_LOAD:
c571bd75 5079 case BPF_BTF_LOAD:
b1d18a75
AS		 5080 case BPF_LINK_CREATE:
5081 case BPF_RAW_TRACEPOINT_OPEN:
af2ac3e1 5082 break;
86f44fce
AS		 5083 default:
5084 return -EINVAL;
5085 }
5086 return __sys_bpf(cmd, KERNEL_BPFPTR(attr), attr_size);
5087}
5088
5089int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size)
5090{
5091 struct bpf_prog * __maybe_unused prog;
5092 struct bpf_tramp_run_ctx __maybe_unused run_ctx;
5093
5094 switch (cmd) {
b1d18a75
AS		 5095#ifdef CONFIG_BPF_JIT /* __bpf_prog_enter_sleepable used by trampoline and JIT */
5096 case BPF_PROG_TEST_RUN:
5097 if (attr->test.data_in || attr->test.data_out ||
5098 attr->test.ctx_out || attr->test.duration ||
5099 attr->test.repeat || attr->test.flags)
5100 return -EINVAL;
5101
5102 prog = bpf_prog_get_type(attr->test.prog_fd, BPF_PROG_TYPE_SYSCALL);
5103 if (IS_ERR(prog))
5104 return PTR_ERR(prog);
5105
5106 if (attr->test.ctx_size_in < prog->aux->max_ctx_offset ||
5107 attr->test.ctx_size_in > U16_MAX) {
5108 bpf_prog_put(prog);
5109 return -EINVAL;
5110 }
5111
e384c7b7
KFL		 5112 run_ctx.bpf_cookie = 0;
5113 run_ctx.saved_run_ctx = NULL;
5114 if (!__bpf_prog_enter_sleepable(prog, &run_ctx)) {
b1d18a75
AS		 5115 /* recursion detected */
5116 bpf_prog_put(prog);
5117 return -EBUSY;
5118 }
5119 attr->test.retval = bpf_prog_run(prog, (void *) (long) attr->test.ctx_in);
e384c7b7 5120 __bpf_prog_exit_sleepable(prog, 0 /* bpf_prog_run does runtime stats */, &run_ctx);
b1d18a75
AS		 5121 bpf_prog_put(prog);
5122 return 0;
5123#endif
af2ac3e1 5124 default:
86f44fce 5125 return ____bpf_sys_bpf(cmd, attr, size);
af2ac3e1 5126 }
79a7f8bd 5127}
86f44fce 5128EXPORT_SYMBOL(kern_sys_bpf);
79a7f8bd 5129
3a2daa72 5130static const struct bpf_func_proto bpf_sys_bpf_proto = {
79a7f8bd
AS		 5131 .func = bpf_sys_bpf,
5132 .gpl_only = false,
5133 .ret_type = RET_INTEGER,
5134 .arg1_type = ARG_ANYTHING,
216e3cd2 5135 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
79a7f8bd
AS		 5136 .arg3_type = ARG_CONST_SIZE,
5137};
5138
5139const struct bpf_func_proto * __weak
5140tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5141{
5142 return bpf_base_func_proto(func_id);
5143}
5144
3abea089
AS		 5145BPF_CALL_1(bpf_sys_close, u32, fd)
5146{
5147 /* When bpf program calls this helper there should not be
5148 * an fdget() without matching completed fdput().
5149 * This helper is allowed in the following callchain only:
5150 * sys_bpf->prog_test_run->bpf_prog->bpf_sys_close
5151 */
5152 return close_fd(fd);
5153}
5154
3a2daa72 5155static const struct bpf_func_proto bpf_sys_close_proto = {
3abea089
AS		 5156 .func = bpf_sys_close,
5157 .gpl_only = false,
5158 .ret_type = RET_INTEGER,
5159 .arg1_type = ARG_ANYTHING,
5160};
5161
d6aef08a
KKD		 5162BPF_CALL_4(bpf_kallsyms_lookup_name, const char *, name, int, name_sz, int, flags, u64 *, res)
5163{
5164 if (flags)
5165 return -EINVAL;
5166
5167 if (name_sz <= 1 || name[name_sz - 1])
5168 return -EINVAL;
5169
5170 if (!bpf_dump_raw_ok(current_cred()))
5171 return -EPERM;
5172
5173 *res = kallsyms_lookup_name(name);
5174 return *res ? 0 : -ENOENT;
5175}
5176
dc368e1c 5177static const struct bpf_func_proto bpf_kallsyms_lookup_name_proto = {
d6aef08a
KKD		 5178 .func = bpf_kallsyms_lookup_name,
5179 .gpl_only = false,
5180 .ret_type = RET_INTEGER,
5181 .arg1_type = ARG_PTR_TO_MEM,
d4efb170 5182 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
d6aef08a
KKD		 5183 .arg3_type = ARG_ANYTHING,
5184 .arg4_type = ARG_PTR_TO_LONG,
5185};
5186
79a7f8bd
AS		 5187static const struct bpf_func_proto *
5188syscall_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5189{
5190 switch (func_id) {
5191 case BPF_FUNC_sys_bpf:
5192 return &bpf_sys_bpf_proto;
3d78417b
AS		 5193 case BPF_FUNC_btf_find_by_name_kind:
5194 return &bpf_btf_find_by_name_kind_proto;
3abea089
AS		 5195 case BPF_FUNC_sys_close:
5196 return &bpf_sys_close_proto;
d6aef08a
KKD		 5197 case BPF_FUNC_kallsyms_lookup_name:
5198 return &bpf_kallsyms_lookup_name_proto;
79a7f8bd
AS		 5199 default:
5200 return tracing_prog_func_proto(func_id, prog);
5201 }
5202}
5203
5204const struct bpf_verifier_ops bpf_syscall_verifier_ops = {
5205 .get_func_proto = syscall_prog_func_proto,
5206 .is_valid_access = syscall_prog_is_valid_access,
5207};
5208
5209const struct bpf_prog_ops bpf_syscall_prog_ops = {
5210 .test_run = bpf_prog_test_run_syscall,
5211};
2900005e
YZ		 5212
5213#ifdef CONFIG_SYSCTL
5214static int bpf_stats_handler(struct ctl_table *table, int write,
5215 void *buffer, size_t *lenp, loff_t *ppos)
5216{
5217 struct static_key *key = (struct static_key *)table->data;
5218 static int saved_val;
5219 int val, ret;
5220 struct ctl_table tmp = {
5221 .data = &val,
5222 .maxlen = sizeof(val),
5223 .mode = table->mode,
5224 .extra1 = SYSCTL_ZERO,
5225 .extra2 = SYSCTL_ONE,
5226 };
5227
5228 if (write && !capable(CAP_SYS_ADMIN))
5229 return -EPERM;
5230
5231 mutex_lock(&bpf_stats_enabled_mutex);
5232 val = saved_val;
5233 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
5234 if (write && !ret && val != saved_val) {
5235 if (val)
5236 static_key_slow_inc(key);
5237 else
5238 static_key_slow_dec(key);
5239 saved_val = val;
5240 }
5241 mutex_unlock(&bpf_stats_enabled_mutex);
5242 return ret;
5243}
5244
5245void __weak unpriv_ebpf_notify(int new_state)
5246{
5247}
5248
5249static int bpf_unpriv_handler(struct ctl_table *table, int write,
5250 void *buffer, size_t *lenp, loff_t *ppos)
5251{
5252 int ret, unpriv_enable = *(int *)table->data;
5253 bool locked_state = unpriv_enable == 1;
5254 struct ctl_table tmp = *table;
5255
5256 if (write && !capable(CAP_SYS_ADMIN))
5257 return -EPERM;
5258
5259 tmp.data = &unpriv_enable;
5260 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
5261 if (write && !ret) {
5262 if (locked_state && unpriv_enable != 1)
5263 return -EPERM;
5264 *(int *)table->data = unpriv_enable;
5265 }
5266
5267 unpriv_ebpf_notify(unpriv_enable);
5268
5269 return ret;
5270}
5271
5272static struct ctl_table bpf_syscall_table[] = {
5273 {
5274 .procname = "unprivileged_bpf_disabled",
5275 .data = &sysctl_unprivileged_bpf_disabled,
5276 .maxlen = sizeof(sysctl_unprivileged_bpf_disabled),
5277 .mode = 0644,
5278 .proc_handler = bpf_unpriv_handler,
5279 .extra1 = SYSCTL_ZERO,
5280 .extra2 = SYSCTL_TWO,
5281 },
5282 {
5283 .procname = "bpf_stats_enabled",
5284 .data = &bpf_stats_enabled_key.key,
5285 .maxlen = sizeof(bpf_stats_enabled_key),
5286 .mode = 0644,
5287 .proc_handler = bpf_stats_handler,
5288 },
5289 { }
5290};
5291
5292static int __init bpf_syscall_sysctl_init(void)
5293{
5294 register_sysctl_init("kernel", bpf_syscall_table);
5295 return 0;
5296}
5297late_initcall(bpf_syscall_sysctl_init);
5298#endif /* CONFIG_SYSCTL */
Linux 6.x block layer and io_uring tree(s)